Broadband Access Networks
Optical Networks Series Editor:
Biswanath Mukherjee University of California, Davis Davis, CA
Broadband Access Networks: Technologies and Deployments Abdallah Shami, Martin Maier, and Chadi Assi (Eds.) ISBN 978-0-387-92130-3 Traffic Grooming for Optical Networks: Foundations, Techniques, and Frontiers Rudra Dutta, Ahmed E. Kamal, and George N. Rouskas (Eds.) ISBN 978-0-387-74517-6 Optical Network Design and Planning Jane M. Simmons ISBN 978-0-387-76475-7 Quality of Service in Optical Burst Switched Networks Kee Chaing Chua, Mohan Gurusamy, Yong Liu, and Minh Hoang Phung ISBN 978-0-387-34160-6 Optical WDM Networks Biswanath Mukherjee ISBN 978-0-387-29055-3 Traffic Grooming in Optical WDM Mesh Networks Keyao Zhu, Hongyue Zhu, and Biswanath Mukherjee ISBN 978-0-387-25432-6 Survivable Optical WDM Networks Canhui (Sam) Ou and Biswanath Mukherjee ISBN 978-0-387-24498-3 Optical Burst Switched Networks Jason P. Jue and Vinod M. Vokkarane ISBN 978-0-387-23756-5
Abdallah Shami · Martin Maier · Chadi Assi Editors
Broadband Access Networks Technologies and Deployments
123
Editors Abdallah Shami Department Electrical & Computer Engineering University of Western Ontario 1151 Richmond Street North London ON N6A 5B9 Canada
[email protected]
Martin Maier Instution National de la Recherche Scientifique (INRS) Universit´e de Quebec 800 de la Gauchetiere Ouest Montreal QC H5A 1K6 Canada
[email protected]
Chadi Assi Concordia Institute for Information Systems Engineering (CIISE) Concordia University 1455 de Maisonneuve Blvd. West Montreal QC H3G 1T7 Canada
[email protected]
ISSN 1935-3839 e-ISSN 1935-3847 ISBN 978-0-387-92130-3 e-ISBN 978-0-387-92131-0 DOI 10.1007/978-0-387-92131-0 Springer Dordrecht Heidelberg London New York Library of Congress Control Number: 2009920724 Springer Science+Business Media LLC 2009 All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Springer Science+Business Media, LLC, 233 Spring Street, New York, NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden. The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights. While the advice and information in this book are believed to be true and accurate at the date of going to press, neither the authors nor the editors nor the publisher can accept any legal responsibility for any errors or omissions that may be made. The publisher makes no warranty, express or implied, with respect to the material contained herein.
c
Printed on acid-free paper Springer is part of Springer Science+Business Media (www.springer.com)
Preface
At present, there is a strong worldwide push toward bringing fiber closer to individual homes and businesses. Fiber-to-the-Home/Business (FTTH/B) or close to it networks are poised to become the next major success story for optical fiber communications. In fact, FTTH connections are currently experiencing double-digit or even higher growth rates, e.g., in the United States the annual growth rate was 112% between September 2006 and September 2007, and their presence can add value of U.S. $4,000–15,000 to the selling price of a home. FTTH networks have to unleash their economic potential and societal benefits by opening up the first/last mile bandwidth bottleneck, thereby strengthening our information society while avoiding its digital divide. FTTH networks hold great promise to enable the support of a wide range of new and emerging services and applications, such as triple play, video on demand, videoconferencing, peer-to-peer (P2P) audio/video file sharing, multichannel high-definition television (HDTV), multimedia/multiparty online gaming, telemedicine, telecommuting, and surveillance. Wireless technologies, on the other hand, have seen tremendous success over the years and wireless networks have now become increasingly popular due to their fast and inexpensive deployment and their capabilities of providing flexible and ubiquitous Internet access. In particular, currently next-generation fixed wireless broadband networks (deployed as wireless mesh networks, WMNs) are motivated by several applications including broadband home networking, community and neighborhood networking, enterprise networking, and so on. The aggregate capacity and performance of these WMNs can be increased through either the use of multiple channels or through the adoption of high capacity wireless links, e.g., WiMAX. A WMN consists of stationary wireless mesh routers, forming a wireless backbone; these routers serve as access points for mobile devices and they aggregate and forward data to gateways connected to the Internet through a wired infrastructure, which is typically realized by means of a broadband optical access network. The Broadband Access Networks: Technologies and Deployments book condenses the relentless research, design, and deployment experience of state-of-theart access networks. The material presented here is intended: (1) to consolidate and disseminate the latest developments and advances in the area of broadband
v
vi
Preface
access network technologies and architectures; (2) to combine and share the emergent technologies developed and devised in the last few years; (3) to share the many experiences and lessons learned from the deployments of field/testing trials of these technologies; (4) to model and analyze the broadband access technologies in a comprehensive and detailed manner so it can be used as a supplement textbook for a graduate course on optical networks or advanced topics on computer communications, as well as for graduate students and other researchers working in this area. The book consists of the following five parts: • • • • •
Introduction and Enabling Technologies. Copper and Wireless Access Networks. Optical Access Networks. Optical-Wireless Access Networks. Deployments.
Each part consists of 2–6 chapters dealing with the topic, the book contains a total of 17 chapters. The book shares the critical steps and details of the developments and deployment of emergent access network technologies, which is very crucial particularly as telecommunications vendors and carriers are looking for cost-effective broadband “last-mile” access solutions to stay competitive. Internationally recognized skilled researchers and key talented industrial players have contributed the 17 chapters of the book; they deserve our immense gratitude for their support and professional cooperation. In addition, it has been a pleasure working with Prof. Biswanath Mukherjee and we thank him for his valuable feedback and suggestions. Finally, it is a pleasure to acknowledge our family and students, as well as Ms. Katelyn Stanne of Springer Press, who commissioned this text. London, Ontario, Canada Montreal, Quebec, Canada Montreal, Quebec, Canada
Abdallah Shami Martin Maier Chadi Assi October 2008
Contents
Part I Introduction and Enabling Technologies 1
The Anatomy of an Access Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Nicholas Gagnon 1.1 The Evolution Path of Typical Access Networks . . . . . . . . . . . . . . . 3 1.2 The Future Is Bright for IP Video . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 1.3 Not All the Access Network Connection Are Created Equal . . . . . . 4 1.4 Broadband Copper Access Network Using ADSL2+/VDSL2 Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 1.5 Fiber-to-the-Home/Building (FTTH/B) Access Network . . . . . . . . . 7 1.5.1 Point-to-Point Ethernet FTTH (Home-Run FTTH) . . . . . . 7 1.5.2 Passive Optical Network (PON) FTTH . . . . . . . . . . . . . . . . 8 1.5.3 Wavelength Division Multiplexing (WDM) PON FTTH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.6 Hybrid Fiber Coax Running DOCSIS Protocol . . . . . . . . . . . . . . . . . 9 1.7 Wireless Access Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 1.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2
Drivers for Broadband in Europe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Evi Zouganeli, Kristin Bugge, Santiago Andres Azcoitia, Juan P. Fernandez Palacios, and Antonio J. Elizondo Armengol Elizondo 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Telecom Trends in Europe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 Decline in Fixed Voice and Growth in Broadband and Mobile Minutes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.2 Customers Becoming More Demanding . . . . . . . . . . . . . . . 2.2.3 High Rate of Technology Innovation and RollOut . . . . . . . 2.2.4 Competition Moving from Pure Price-to-Price, Service- and Segment-Specific Offers . . . . . . . . . . . . . . . . . 2.3 Drivers for Broadband Services . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1 Increasing Bandwidth Demand . . . . . . . . . . . . . . . . . . . . . .
13
13 14 15 16 16 17 17 20 vii
viii
Contents
2.4
Broadband Market in Europe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.1 Broadband Lines, Penetration, and Growth . . . . . . . . . . . . 2.4.2 Broadband by Technology . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.3 Broadband Offer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.4 Competition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 Broadband Infrastructure – Evolution and Trends . . . . . . . . . . . . . . . 2.5.1 Current Broadband Access Technologies and Architectures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5.2 Impact of FTTx Deployments on Local Loop Sharing . . . 2.6 Overview on European Regulatory Framework . . . . . . . . . . . . . . . . . 2.6.1 European Regulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.6.2 Regulatory Situation in Some European Countries . . . . . . 2.6.3 Discussion – The Impact of Regulation . . . . . . . . . . . . . . . 2.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3
Enabling Techniques for Broadband Access Networks . . . . . . . . . . . . Ton Koonen 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Fiber in the Access Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 Fiber-DSL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.2 Hybrid Fiber-Coax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.3 Fiber-Wireless . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.4 Fiber to the Home . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Basic Optical Access Network Components . . . . . . . . . . . . . . . . . . . 3.3.1 Optical Fiber . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.2 Optical Power Splitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.3 Wavelength Routing Devices . . . . . . . . . . . . . . . . . . . . . . . . 3.4 FTTH Network Topologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.1 Point-to-Point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.2 Point-to-Multipoint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.3 Cost Aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5 Multiple Access Techniques for a PON . . . . . . . . . . . . . . . . . . . . . . . 3.5.1 Time Division Multiple Access . . . . . . . . . . . . . . . . . . . . . . 3.5.2 Subcarrier Multiple Access . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.3 Optical Code Division Multiple Access . . . . . . . . . . . . . . . 3.5.4 Wavelength Division Multiple Access . . . . . . . . . . . . . . . . 3.6 Radio Over Fiber . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.7 Free-Space Optical Communication . . . . . . . . . . . . . . . . . . . . . . . . . . 3.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21 21 23 25 26 27 27 29 30 30 32 33 34 35 37 37 38 38 39 40 40 41 41 44 45 47 47 48 48 50 50 52 54 55 57 60 61 62
Contents
ix
Part II Copper and Wireless Access Networks 4
5
Vectored DSLs and the Copper PON (CuPON) . . . . . . . . . . . . . . . . . . John M. Cioffi, Sumanth Jagannathan, Mehdi Mohseni, and George Ginis 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Vectored-DSM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 A Vectoring Tutorial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.2 Differential Vectoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.3 Full-Vectored Binder Capacity . . . . . . . . . . . . . . . . . . . . . . . 4.3 A CuPON for savings on access-network purchases . . . . . . . . . . . . . 4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
65
66 67 67 69 71 75 78 78
Enabling Broadband Wireless Technologies∗ . . . . . . . . . . . . . . . . . . . . 81 Quazi M. Rahman 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 5.2 Modulation: Let’s Deal with the Available Spectrum and Make Use of the Limited Power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 5.2.1 Phase Shift Keying (PSK) Modulation . . . . . . . . . . . . . . . . 84 5.2.2 Quadrature Amplitude Modulation (QAM) . . . . . . . . . . . . 85 5.2.3 Orthogonal Frequency Division Multiplexing (OFDM) . . 87 5.3 Coding Techniques: Let’s Deal with the Channel . . . . . . . . . . . . . . . 90 5.3.1 Block Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 5.3.2 Convolutional Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 5.3.3 Turbo Coding (TC) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 5.3.4 Space–Time Coding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 5.3.5 Coded Modulation Techniques . . . . . . . . . . . . . . . . . . . . . . 99 5.4 Adaptive Modulation and Coding (AMC) . . . . . . . . . . . . . . . . . . . . . 99 5.5 Multiple Access Techniques: Let’s All Share the Same Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 5.5.1 Frequency Division Multiple Access (FDMA) . . . . . . . . . 101 5.5.2 Time Division Multiple Access (TDMA) . . . . . . . . . . . . . . 102 5.5.3 Code Division Multiple Access (CDMA) . . . . . . . . . . . . . . 103 5.5.4 Orthogonal Frequency Division Multiple Access (OFDMA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 5.5.5 Combination of OFDM and CDMA Systems . . . . . . . . . . 105 5.5.6 Carrier Sense Multiple Access (CSMA) Protocol . . . . . . . 107 5.6 Diversity Techniques: Let’s Make a Better Use of the Channel . . . . 107 5.6.1 Classifications of the Diversity Techniques . . . . . . . . . . . . 108 5.6.2 Classifications of Diversity Combiners . . . . . . . . . . . . . . . . 109 5.7 Challenges and Research Evidences . . . . . . . . . . . . . . . . . . . . . . . . . . 110 5.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
x
Contents
6
WiMAX Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 Abdou R. Ahmed, Xiaofeng Bai, and Abdallah Shami 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 6.1.1 Scope of the Standard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 6.1.2 Frequency Bands of Operation . . . . . . . . . . . . . . . . . . . . . . 120 6.1.3 Reference Model of the Standard . . . . . . . . . . . . . . . . . . . . 120 6.1.4 WirelessMAN PHY Specifications . . . . . . . . . . . . . . . . . . . 121 6.1.5 MAC Sublayers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 6.2 Point-to-MultiPoint (PMP) WiMAX Networks . . . . . . . . . . . . . . . . . 125 6.2.1 MAC Frame Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 6.2.2 Service Flow and Connection . . . . . . . . . . . . . . . . . . . . . . . 129 6.2.3 Service Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 6.2.4 Bandwidth Request and Resource Allocation . . . . . . . . . . 130 6.3 WiMAX Mesh Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 6.3.1 General Properties of WiMAX Mesh . . . . . . . . . . . . . . . . . 131 6.3.2 Mesh Network Operations . . . . . . . . . . . . . . . . . . . . . . . . . . 132 6.3.3 Routing in Mesh Network . . . . . . . . . . . . . . . . . . . . . . . . . . 137 6.3.4 QoS in Mesh Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137 6.4 Mobility in WiMAX Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 6.4.1 Handovers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 6.4.2 WiMAX Standards and Mobility . . . . . . . . . . . . . . . . . . . . . 139 6.4.3 WiMAX and Homogeneous Mobility . . . . . . . . . . . . . . . . . 142 6.4.4 WiMAX and Mobility in Heterogeneous Networks . . . . . 144 6.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
Part III Optical Access Networks 7
Dynamic Bandwidth Allocation for Ethernet Passive Optical Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 Hassan Naser and Hussein T. Mouftah 7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 7.2 Data Link Layer Protocols for PON . . . . . . . . . . . . . . . . . . . . . . . . . . 153 7.3 Multi-Point Control Protocol (MPCP) . . . . . . . . . . . . . . . . . . . . . . . . 154 7.3.1 REPORT Frame Format . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 7.3.2 GATE Frame Format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159 7.3.3 Ranging Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160 7.3.4 Automatic Device Discovery . . . . . . . . . . . . . . . . . . . . . . . . 161 7.4 Dynamic Bandwidth Allocation Algorithms . . . . . . . . . . . . . . . . . . . 162 7.4.1 Class-of-Service Oriented Packet Scheduling (COPS) Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164 7.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
Contents
xi
8
Quality of Service in Ethernet Passive Optical Networks (EPONs) . . 169 Ahmad Dhaini and Chadi Assi 8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170 8.1.1 IEEE 802.1D Support for Classes of Service . . . . . . . . . . . 171 8.1.2 CoS Support in EPON . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173 8.2 Intra-ONU Scheduling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174 8.2.1 Incoming Traffic Handling Operation . . . . . . . . . . . . . . . . . 174 8.2.2 Existing Solutions and Schemes . . . . . . . . . . . . . . . . . . . . . 174 8.2.3 Intra-ONU Queue Management . . . . . . . . . . . . . . . . . . . . . . 176 8.3 QoS-Enabled Dynamic Bandwidth Allocation Algorithms (DBAs) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177 8.3.1 Existing Solutions and Schemes . . . . . . . . . . . . . . . . . . . . . 178 8.4 Quality-of-Service Protection and Admission Control in EPON . . . 180 8.4.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180 8.4.2 Traffic Characteristics and QoS Requirements . . . . . . . . . . 183 8.4.3 Local Admission Control (LAC) . . . . . . . . . . . . . . . . . . . . . 185 8.4.4 Global Admission Control (GAC) . . . . . . . . . . . . . . . . . . . . 185 8.4.5 Issues and Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186 8.4.6 Admission Control-Enabled Dynamic Bandwidth Allocation Scheme (AC-DBA) . . . . . . . . . . . . . . . . . . . . . . 187 8.4.7 Performance Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189 8.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
9
MultiChannel EPONs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197 Michael P. McGarry and Martin Reisslein 9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197 9.2 Bandwidth Management for Multichannel EPONs . . . . . . . . . . . . . . 198 9.2.1 Grant Sizing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199 9.2.2 Grant Scheduling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200 9.3 Separate Time and Wavelength Assignment . . . . . . . . . . . . . . . . . . . 200 9.3.1 Grant Wavelength Assignment . . . . . . . . . . . . . . . . . . . . . . . 201 9.3.2 Grant Ordering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 9.4 Combined Time and Wavelength Assignment . . . . . . . . . . . . . . . . . . 202 9.4.1 A Scheduling Theoretic Approach . . . . . . . . . . . . . . . . . . . 202 9.4.2 Scheduling Frameworks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204 9.4.3 Scheduling Policies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209 9.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
10
Long-Reach Optical Access . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219 Huan Song, Byoung-Whi Kim, and Biswanath Mukherjee 10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219 10.2 Research Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221 10.2.1 Signal Power Compensation . . . . . . . . . . . . . . . . . . . . . . . . 222 10.2.2 Optical Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
xii
Contents
10.2.3 Burst-Mode Receiver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223 10.2.4 Upstream Resource Allocation . . . . . . . . . . . . . . . . . . . . . . 224 10.3 Demonstrations of LR-PON . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224 10.3.1 PLANET SuperPON . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224 10.3.2 British Telecom (BT) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225 10.3.3 University College Cork, Ireland . . . . . . . . . . . . . . . . . . . . . 227 10.3.4 Electronics and Telecommunication Research Institute (ETRI), Korea . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229 10.3.5 Other Demonstrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229 10.4 Dynamic Bandwidth Assignment (DBA) . . . . . . . . . . . . . . . . . . . . . . 230 10.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234 11
Optical Access–Metro Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237 Martin Maier 11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237 11.2 Optical Regional Access Network . . . . . . . . . . . . . . . . . . . . . . . . . . . 239 11.3 Stanford University Access Network . . . . . . . . . . . . . . . . . . . . . . . . . 240 11.4 Metro Access Ring Integrated Network . . . . . . . . . . . . . . . . . . . . . . . 242 11.5 OBS Access–Metro Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244 11.6 STARGATE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247 11.6.1 Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248 11.6.2 Discovery and Registration . . . . . . . . . . . . . . . . . . . . . . . . . 251 11.6.3 Dynamic Bandwidth Allocation . . . . . . . . . . . . . . . . . . . . . 251 11.6.4 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253 11.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
12
Signal Processing Techniques for Data Confidentiality in OCDMA Access Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261 Yue-Kai Huang, Paul Toliver, and Paul R. Prucnal 12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262 12.2 Optical Encryption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263 12.3 Optical Steganography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267 12.4 Phase Scrambling OCDM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270 12.5 Multi-code Processing Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . 275 12.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
Part IV Optical-Wireless Access Networks 13
Radio-over-Fiber (RoF) Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283 John E. Mitchell 13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283 13.2 Basic Technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285 13.3 Radio-Over-Fiber Application Areas . . . . . . . . . . . . . . . . . . . . . . . . . 291
Contents
xiii
13.4
Networking Concepts and Techniques . . . . . . . . . . . . . . . . . . . . . . . . 294 13.4.1 Media Access Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294 13.4.2 Dynamic Capacity Allocation . . . . . . . . . . . . . . . . . . . . . . . 295 13.4.3 Sharing a Remote Antenna Unit – Wideband RAU Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296 13.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296 14
Integration of EPON and WiMAX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301 Gangxiang Shen and Rodney S. Tucker 14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301 14.2 Literature Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302 14.3 Integrated Architectures for EPON and WiMAX . . . . . . . . . . . . . . . 303 14.3.1 Independent Architectures . . . . . . . . . . . . . . . . . . . . . . . . . . 303 14.3.2 Hybrid Architectures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304 14.3.3 Microwave-over-Fiber Architectures . . . . . . . . . . . . . . . . . . 306 14.3.4 Multistage EPON and WiMAX Integration . . . . . . . . . . . . 307 14.4 Design and Operation Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308 14.4.1 Optimal Passive Optical Network Deployment . . . . . . . . . 309 14.4.2 Packet Forwarding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309 14.4.3 Bandwidth Allocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311 14.4.4 QoS Support and Mapping . . . . . . . . . . . . . . . . . . . . . . . . . . 313 14.4.5 Handover Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313 14.4.6 Survivability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315 14.4.7 Cooperative Transmission for Broadcast Services . . . . . . . 316 14.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
15
Hybrid Wireless–Optical Broadband Access Network (WOBAN) . . . 321 Suman Sarkar, Pulak Chowdhury, Sudhir Dixit, and Biswanath Mukherjee 15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321 15.2 WOBAN: A Network for Future . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322 15.2.1 Why Is WOBAN a Compelling Solution? . . . . . . . . . . . . . 323 15.2.2 Flavors of Converged Architecture . . . . . . . . . . . . . . . . . . . 324 15.3 Connectivity and Routing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328 15.3.1 Delay-Aware Routing Algorithm (DARA) . . . . . . . . . . . . . 330 15.3.2 Capacity and Delay-Aware Routing (CaDAR) . . . . . . . . . . 331 15.3.3 GROW-Net Integrated Routing (GIR) . . . . . . . . . . . . . . . . . 331 15.4 Fault Tolerance and Self-Healing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333 15.4.1 Risk-and-Delay Aware Routing Algorithm (RADAR) . . . 333 15.4.2 GROW-Net’s Reconfigurable Backhaul . . . . . . . . . . . . . . . 334 15.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
xiv
Contents
Part V Deployments 16
Point-to-Point FTTx . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339 Wolfgang Fischer 16.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339 16.2 Fiber Topology vs. Transmission Scheme . . . . . . . . . . . . . . . . . . . . . 340 16.3 Architectural Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341 16.4 Deployment Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342 16.5 Operational Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344 16.5.1 Traffic Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344 16.5.2 Security . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344 16.5.3 CPE Deployments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345 16.5.4 Trouble-Shooting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345 16.5.5 Power Budget Planning . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346 16.6 Cost Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346 16.6.1 Capital Cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346 16.6.2 Operation Cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347 16.7 Open Fiber Access . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348 16.8 Transmission Technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349 16.9 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350 16.10 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
17
Broadband Access Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353 Klaus Grobe, J¨org-Peter Elbers, and Stephan Neidlinger 17.1 Broadband Drivers and Network Requirements . . . . . . . . . . . . . . . . 354 17.1.1 Broadband Drivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 354 17.1.2 Network Architecture and Requirements . . . . . . . . . . . . . . 355 17.2 Scalable Broadband Access Networks . . . . . . . . . . . . . . . . . . . . . . . . 356 17.2.1 xWDM in Access Networks . . . . . . . . . . . . . . . . . . . . . . . . . 357 17.2.2 TDM Add/Drop Multiplexing . . . . . . . . . . . . . . . . . . . . . . . 358 17.2.3 Layer-2 Packet Aggregation . . . . . . . . . . . . . . . . . . . . . . . . . 359 17.2.4 Deployment Case: Telecom Italia . . . . . . . . . . . . . . . . . . . . 359 17.3 Next-Generation Access and Backhaul . . . . . . . . . . . . . . . . . . . . . . . 361 17.3.1 Migration toward WDM-PON . . . . . . . . . . . . . . . . . . . . . . . 363 17.3.2 WDM-PON Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364 17.3.3 WDM-PON Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367 17.3.4 WDM-PON Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . 369 17.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
List of Contributors
Abdou R. Ahmed Department of Electrical and Computer Engineering, The University of Western Ontario, London, ON, Canada N6A 5B8, e-mail:
[email protected] Chadi Assi Concordia University, Montreal, QC, Canada, e-mail:
[email protected] Santiago Andres Azcoitia Telefonica I+D, Emilio Vargas 6, 28043 Madvid, Spain, e-mail:
[email protected] Xiaofeng Bai Department of Electrical and Computer Engineering, The University of Western Ontario, London, ON, Canada N6A 5B8, e-mail:
[email protected] Kristin Bugge Telenor Nordic, Snaroeyveien 30, 1331 Fornebu Norway, e-mail:
[email protected] Pulak Chowdhury University of California, Davis, CA, USA, e-mail:
[email protected] John M. Cioffi Department of Electrical Engineering, Stanford University, David Packard Electrical Engineering Building, 350 Serra Mall, MC 9515, Stanford, CA 94305-9515, USA, e-mail:
[email protected]
xv
xvi
List of Contributors
Ahmad Dhaini Department of Electrical and Computer Engineering, University of Waterloo, Waterloo, ON, Canada, e-mail:
[email protected] Sudhir Dixit Nokia Siemens Networks, Mountain View, CA, USA, e-mail:
[email protected] J¨org-Peter Elbers ADVA AG Optical Networking, Fraunhoferstr. 9a, 82152 Martinsried/Munich, Germany, e-mail:
[email protected] Antonio J. Elizondo Armengol Telefonica I+D, Emilio Vargas 6, 28043 Madvid, Spain, e-mail:
[email protected] Juan P. Fernandez Palacios Telefonica I+D, Emilio Vargas 6, 28043 Madvid, Spain, e-mail:
[email protected] Wolfgang Fischer Cisco Systems Europe, e-mail:
[email protected] Nicholas Gagnon EXFO Inc., Quebec City, Quebec, Canada, e-mail:
[email protected] George Ginis ASSIA Inc., 303 Twin Dolphin Drive, Suite 203, Redwood City, CA 94065, USA Klaus Grobe ADVA AG Optical Networking, Fraunhoferstr. 9a, 82152 Martinsried/Munich, Germany, e-mail:
[email protected] Yue-Kai Huang NEC Laboratories America, Inc., 4 Independence Way, Princeton, NJ 08540, USA, e-mail:
[email protected] Sumanth Jagannathan Department of Electrical Engineering, Stanford University, David Packard Electrical Engineering Building, 350 Serra Mall, MC 9515, Stanford, CA 94305-9515 e-mail:
[email protected] Ton Koonen Electro-Optical Communication Systems Group, COBRA Institute, Eindhoven University of Technology, Eindhoven, The Netherlands, e-mail:
[email protected]
List of Contributors
xvii
Michael P. McGarry Department of Electrical and Computer Engineering, University of Akron, Akron, OH, USA and ADTRAN, Phoenix, AZ, USA, e-mail:
[email protected] Martin Maier INRS, University of Qu´ebec, Montr´eal, QC, Canada, e-mail:
[email protected] Martin Reisslein Electrical Engineering, Arizona State University, Phoenix, AZ USA, e-mail:
[email protected] John E. Mitchell Department of Electronic and Electrical Engineering, University College London, London WC1E 7JE UK, e-mail:
[email protected] Mehdi Mohseni ASSIA Inc., 303 Twin Dolphin Drive, Suite 203, Redwood City, CA 94065, USA Hussein T. Mouftah SITE, University of Ottawa, Ottawa, ON, Canada K1N 6N5, e-mail:
[email protected] Biswanath Mukherjee Department of Computer Science University of California, Davis, CA, USA, e-mail:
[email protected] Hassan Naser Department of Software Engineering, Lakehead University, Thunder Bay, ON, Canada P7B 5E1, e-mail:
[email protected] Stephan Neidlinger ADVA AG Optical Networking, Fraunhoferstr. 9a, 82152 Martinsried/Munich, Germany, e-mail:
[email protected] Paul R. Prucnal Department of Electrical Engineering, Princeton University, Princeton, NJ 08544, USA, e-mail:
[email protected] Quazi M. Rahman Department of Electrical and Computer Engineering, The University of Western Ontario, 1151 Richmond Street, London, ON, Canada N6A 5B8, e-mail:
[email protected] Suman Sarkar University of California, Davis, CA, USA, e-mail:
[email protected]
xviii
List of Contributors
Abdallah Shami Department of Electrical and Computer Engineering, The University of Western Ontario, London, ON, Canada N6A 5B8, e-mail:
[email protected] Gangxiang Shen ARC Special Research Centre for Ultra-Broadband Information Networks (CUBIN), Department Electrical and Electronic Engineering, The University of Melbourne, Melbourne, VIC 3010 Australia, e-mail:
[email protected] Huan Song Department of Computer Science, University of California, Davis, CA USA, e-mail:
[email protected] Paul Toliver Telcordia Technologies, 331 Newman Springs Road, Red Bank, NJ 07701, USA, e-mail:
[email protected] Rodney S. Tucker ARC Special Research Centre for Ultra-Broadband Information Networks (CUBIN), Department Electrical and Electronic Engineering, The University of Melbourne, Melbourne, VIC 3010 Australia, e-mail:
[email protected] Byoung-Whi Kim Electronics and Telecommunications Research Institute, Daejeon, Korea Evi Zouganeli Telenor R&I, B5d, Snaroeyveien 30, 1331 Fornebu Norway, e-mail:
[email protected]
Part I
Introduction and Enabling Technologies
Chapter 1
The Anatomy of an Access Network Nicholas Gagnon
Abstract This chapter will review the different access network technologies that promise to support multi-stream video service deliveries. Technologies such as ADSL2+/VDSL2 copper, fiber to the home/building, hybrid fiber coax, and wireless architectures will be analyzed, pinpointing the pros and cons of each of these alternatives, in context of the increasing popularity and customer demand of highdefinition video services delivery over access network architectures. All these technologies have particularities and operators establishing their technology roadmaps and their business cases need definitely to have a look at the global picture before taking any specific decisions.
1.1 The Evolution Path of Typical Access Networks Long gone are the days where access network was built only with copper cables and was hardly allowing a data transmission at 56 Kbps (kilobits per second) in the best of the cases! Nowadays, copper cables remain present in the access networks because new technologies and compression techniques allow the operators to leverage these existing assets and provide broadband services over them. So much copper loop plants are in place that operators have second thoughts about replacing them, especially when they analyze the business case of fiber-to-thehome (FTTH) projects. Copper loop plant remains important asset for the operators. ADSL2+/VDSL2 technology can be used to transmit 24–30 Mbps (megabits per second), over a typical 1–1.5 km copper loop plant. This level of bandwidth performance is interesting for many broadband applications and certain network operators consider this sufficient to deliver broadband services.
Nicholas Gagnon EXFO Inc., Quebec City, Quebec, Canada, e-mail:
[email protected]
A. Shami et al. (eds.), Broadband Access Networks, Optical Networks, c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-92131-0 1,
3
4
Nicholas Gagnon
1.2 The Future Is Bright for IP Video Operators may become aware they have much greater opportunities than offering 24–30 Mbps of broadband services, especially if they adapt their transport network to deliver converged IP (Internet protocol) services. Converged IP services are made possible when operators choose to use IP technology to transmit high speed Internet (HSI) data services, voice service (VoIP) and IP Video, including IPTV (Internet protocol television), video on demand (VoD), and streaming video on the Internet. Offering all these IP-based services simultaneously to the subscriber may require more than 24–30 Mbps, especially when subscribers want to view different video services on more than one television sets in the same house simultaneously and particularly when the video services are in high definition television (HDTV). Table 1.1 presents the different bandwidth requirements for different IP converged services delivered to subscribers over the access network. Table 1.1 Bandwidth requirement for IP services IP services Standard definition High definition Windows Media or H.264 streams 1–1.5 Mbps 7–8 Mbps (1080p) Hulu.com (Streaming Video) 1 Mbps (480p) 2.5 Mbps (720p) MPEG-2 streams (IPTV or VoD) 3.5 Mbps 18–20 Mbps MPEG-4 AVC streams (IPTV or VoD) 1.5 Mbps 9 Mbps VoIP call (full-duplex) 128 Kbps Internet on-line gaming 10 Mbps Peer-to-peer file sharing on the Internet As much bandwidth in upload/download allowed by the ISP
The bandwidth requirement for IP video services (IPTV, VoD, or Streaming video on the Internet) depends of the video codec (compression/decompression algorithm) used. MPEG-4 AVC is now deployed massively in the industry and its bandwidth consumption advantage requires about half the bandwidth than MPEG2-encoded content.
1.3 Not All the Access Network Connection Are Created Equal Having listed the different bandwidth requirements for broadband IP services, the interesting question to ask is “How much bandwidth to the home a typical subscriber may require to satisfy his/her broadband IP services needs?” The answer to this question will vary a lot depending on the region of the world, the number of television and computer sets owned and the number of family members living in the household of this subscriber and ultimately on the discretionary budget for broadband services entertainment this subscriber may have. All the preceding factors will of course influence the number of streams that may be demanded simultaneously by all the users living in the household of the subscriber.
1 The Anatomy of an Access Network
5
All these users requiring simultaneous streams (VoIP, IPTV, VoD, streaming video), peer-to-peer file sharing and Internet on-line gaming will be using the access network connection fixed bandwidth; therefore, a limited number of streams will be available simultaneously. In a not so far future, it is very realistic to envision a critical mass of subscribers requiring access network connection allowing over 40 Mbps downstream (from the network to the home) and at least 20 Mbps upstream (from the home to the network). Figure 1.1 presents a realistic breakdown of downstream and upstream bandwidth requirement separated in services.
Fig. 1.1 Realistic access network bandwidth requirement for broadband IP connections
Other access network architecture can deliver greater bandwidth performance than what ADSL2+/VDSL2 copper access network architecture can deliver. Fiber to the home is one candidate, with a solid capital expenditure (CAPEX) tag associated with this option, particularly associated with the construction of the fiber-optic network. Hybrid fiber coax (HFC) used by multi-service operators (MSO) delivering community access television (CATV) is another option, particularly when leveraging the DOCSIS 3.0 technology. HFC networks are not perfect; important operation expenditures (OPEX) are associated with this technology. Wireless is also an option (WiMAX, long-term evolution or LTE) evolving very rapidly, but it is still to demonstrate if this option can support solid competitive performance for video services. This is not the case right now, particularly for multi-streams video transmission. There is no clear winner or champion yet among all the access network architectures; each one has its pros and cons. In the following pages, we will review all these access network architectures and try to objectively compare them against the others and analyze in greater details their main particularities.
6
Nicholas Gagnon
1.4 Broadband Copper Access Network Using ADSL2+/VDSL2 Technology The bandwidth performance delivered at the home of a subscriber by a copperbased access network using ADSL2+/VDSL2 technology depends on the length of the copper loop. Typically, it is possible to achieve 24–30 Mbps over a 1–1.5 km copper loop plant. This network architecture is commonly used by numerous network operators around the world. This option is attractive for the operators because they can re-use the existing last section of the copper loop plant already installed. Doing so, the operators are leveraging an important asset they already own. They need to push the fiber-optic technology deeper in the access network to connect a remote ADSL2+/VDSL2 digital subscriber line access multiplexer (DSLAM), but this investment is less important than if they would decide to go with the fiber optic all the way to the home. It is interesting to compare the typical 24–30 Mbps bandwidth performance (downstream) and 1Mbps (upstream) over a reasonable distance of 1–1.5 km copper loop plant using ADSL2+/VDSL2 technology with the realistic access network bandwidth requirement for broadband IP connections scenario presented in Fig. 1.1. In this previous scenario, three simultaneous MPEG-4 AVC-encoded IP video streams (IPTV or VoD) are downloaded, corresponding to a bandwidth requirement of 23 Mbps, two viewed streams, fully decoded, 9 Mbps each streams, and one stream digitally recorded on a personal video recorder (PVR) at 5 Mbps. The network will require an extra 2–3 Mbps of bandwidth for buffering, leaving very little bandwidth extra for Internet on-line gaming and peer-to-peer file sharing on the total capacity of 25–30 Mbps. If this scenario seems realistic to the reader, he must come to the conclusion that an access network architecture using ADSL2+/VDSL2 technology has very little margin for error when multi-streams high definition IP video is highly considered by the subscribers. Broadband copper access network using ADSL2+/VDSL2 technology remains a technology of choice to deliver simultaneously one high-definition IP video stream with one or two standard definition IP video streams, high speed Internet services and VoIP calls (Fig. 1.2). VDSL2 technology has been used a lot for inside the building copper-based network in Multi-dwelling units (MDU). In MDU situations, where copper loop plant distances are typically smaller than 500–600 m, VDSL2 technology can deliver over
Fig. 1.2 ADSL2+/VDSL2 broadband copper access network representation
1 The Anatomy of an Access Network
7
60 Mbps of bandwidth (40 Mbps downstream and 20 Mbps upstream). VDSL2 technology deployment inside MDU has been adopted as a mainstream trend for economic and ease of installation reasons. The capital expenditure (CAPEX) of building a broadband copper access network using ADSL2+/VDSL2 technology is definitely lower than a fiber-to-thehome project. The saving associated with the re-use of the last copper loop plant is significant, when a distance of 1–1.5 km is kept. The trade-offs for the operator are not trivial, the bandwidth performance will be limited and the operation expenditure (OPEX) to maintain the copper loop plant will be greater, compared to bandwidth performance and OPEX for the maintenance of a FTTH network.
1.5 Fiber-to-the-Home/Building (FTTH/B) Access Network It is common knowledge that fiber-optic technology has no bandwidth limit. The usual limitation is not the physical medium, it is the budget allowed to connect electro-optical equipment at both ends that will normally determine the transmission speed running on the fiber optic itself. Transmission equipment makers have developed different types of electro-optical equipments creating different types of fiber-to-the-home/building (FTTH/B) access network architectures that have been already deployed in many regions of the world. The three main FTTH/B network architectures existing today are point-to-point Ethernet FTTH (or home-run fiber to the home), passive optical network (PON) FTTH, and Wavelength Division Multiplexing (WDM) PON FTTH.
1.5.1 Point-to-Point Ethernet FTTH (Home-Run FTTH) This architecture is created by connecting a subscriber with a dedicated connection between the subscriber’s location and the central office (CO) of the operators. In some region of the world this architecture has been deployed and financed by community, the subscriber being the owner of its access network connection. One other interesting particularity of this architecture is the fact that it can support 1 Gbps (Gigabit per second) rate by installing the right transmission equipments at both ends. For residential purposes 100 Mbps rate are more popular. Transmission rates are always symmetrical in the case of point-to-point Ethernet FTTH . One downside of this network architecture is the quantity of fiber in the access network is more important than other FTTH architectures. This architecture does not use passive splitter to combine the traffic of many subscribers on a portion of the fiber plant; each subscriber has its own dedicated strand of fiber optic all the way from the CO to the subscribers’ home (home-run). The distance between the CO and the subscriber location will depend on the quality and power of the laser included in the electro-optical transmission installed at both ends of the fiber-optic link. Typical distance of 40 km can be reached with good quality 1550 nm optical transceivers, and 20 km can be reached with 1310 nm optical transceivers.
8
Nicholas Gagnon
1.5.2 Passive Optical Network (PON) FTTH On the passive optical network (PON) side, two protocols are offered to organize the service layer transmission. Ethernet is one, developed around the specifications of the IEEE 802.3ah and Gigabit Encapsulation Method, or GEM protocol, developed around the ITU-T G.984 recommendation, the higher bit rate successor of broadband PON (BPON). The PON architecture uses passive splitter components in the FTTH access network to allow multiple users, typically a group of 4, 8, 16 or 32 users, to share a distribution fiber transporting typically 2.5 Gbps of traffic for GPON. The PON FTTH architecture uses significantly less fiber miles in the access network than point-to-point Ethernet FTTH. A PON access network uses at least two wavelengths, one wavelength downstream, typically 1490 nm, and one wavelength upstream, typically 1310 nm. A second downstream lambda can also be used for RF video overlay, typically 1550 nm. Even if the RF overlay wavelength usage was more popular when BPON network was deployed, the BPON transmission rate was limited at 622 Mbps per wavelength; hence, the separation of video on a second BPON card was the norm, the RF overlay wavelength can be used on GPON system. The typical distance of a PON FTTH access network is 20 km. The typical downstream bandwidth a PON FTTH connection provides, per subscriber, is 80 Mbps for a gigabit passive optical network (GPON), or 2.5 Gbps with 32 splits. The upstream bandwidth, per subscriber, is typically 40 Mbps, or 1.25 Gbps with 32 splits. Passive optical network PON architectures require important capital expense (CAPEX) investment, particularly when fiber optic is not already installed in the access network. According to Alcatel-Lucent, construction or labor costs of all the segments of the access network, from the Central Office (CO) to the home of the subscribers, are the most important (digging, trenching, installation of ducts, laying the fiber optic in ducts or aerial installation of fiber optic using poles), and account for nearly 60% of the total estimated charges of around $1250 for Greenfield situation or $1600 for overbuild situation. Equipment CAPEX (fiber, splitter, drop cable, electro-optical transmission equipments) follows, accounting for about 40% of the total estimated charges [1] (Fig. 1.3). On the other side, PON architecture is envisioned as bringing important advantages relative to operation expense (OPEX). According to IP Business News, the operation expense (OPEX) or annual maintenance charge for a mile of FTTH PON
Fig. 1.3 Representation of a PON FTTH architecture
1 The Anatomy of an Access Network
9
plant is about one-tenth of the ones to maintain hybrid fiber coax infrastructure [2]. Verizon FiOS experience briefing supports this statement in presenting the total field dispatches and OSP-related dispatches are showing solid decline from fiber-tothe-node to fiber-to-the-home maintenance process, with 80% reduction [3]. These economies alone would be important enough to justify the investment in PON FTTH and obtain a return on the investment (ROI) in 10–15 years. Standard organizations (ITU and IEEE) and operators organizations like full service access network (FSAN) are discussing the next 10 Gbps standard for PON. Some additional works at the ITU are also made to extend the reach of GPON access network to 60 km of distance.
1.5.3 Wavelength Division Multiplexing (WDM) PON FTTH WDM PON is using the same passive optical network architecture with passive optical splitter component. The wavelength division multiplexing (WDM) difference is that this system is using several wavelengths, one wavelength per subscriber. An array wave guide (AWG) splitting each wavelength to one specific route to feed one subscriber is used as a passive splitter. The commercial WDM PON available in the market at this time are providing a typical 100 Mbps per wavelength to the subscriber, but developments are in the work to increase this transmission rate at 1 Gbps per wavelength and planning activities to bring the transmission rate to 10 Gbps per wavelength are active.
1.6 Hybrid Fiber Coax Running DOCSIS Protocol The hybrid fiber coax network architecture has proven to be flexible and rugged in the last 10 years, since the multi-service operators (MSO) have decided to provide first high speed Internet (HSI) service, followed by digital phone or voice over IP (VoIP) service and more recently digital high definition (HD) video, either broadcast or video on demand (VoD), on this network architecture previously used to radio frequency (RF) broadcast video. MSO’s networks have evolved in this period of time; higher bit rate and wavelength division multiplexing (WDM) have been installed in the transport ring of the network, typically the same improvement that all network providers offering broadband services have realized in this same segment of their network (Fig. 1.4). The hybrid fiber coax access network portion begin at the distribution hub; from this point a variable length of fiber optic will feed an optical node where the opticalelectrical conversion of the signal is realized to adapt the signal for the coaxial portion of the network. The coaxial portion of the HFC network has evolved with the installation of bi-directional and more powerful amplifiers, allowing broadband services transmission. The coaxial sections, trunk and line, can reach typical distance greater than 8 km, leveraging the use of the amplifiers. The transmission of broadband services over HFC network has been greatly eased by the development of the data over cable service interface specifications
10
Nicholas Gagnon
Fig. 1.4 Representation of a hybrid fiber coax network architecture
(DOCSIS) or CableLabs Certified Cable Modem project. Founded in 1988 by cable R operating companies, Cable Television Laboratories, Inc. (CableLabs ) is a nonprofit research and development consortium that is dedicated to pursuing new cable telecommunication technologies and to helping its cable operator members integrate those technical advancements into their business objectives.1 DOCSIS is using channels of 6 MHz of RF spectrum on the coaxial medium to carry 38 Mbps of broadband services. The first commercial application of the new DOCSIS 3.0 will allow channel bonding, upstream and downstream, to allow bi-directional transmissions up to 100 Mbps for each subscriber. This level of performance is definitely competitive with fiber-to-the-home services. The main disadvantage of the HFC network remains in the operation expense (OPEX) of maintaining these cascades of trunk and line RF amplifiers, this expenditure represents about $1,100 per mile of plant per year [2]. The HFC network architecture remains performant and flexible, even if the OPEX of the maintenance of the RF amplifiers is important. MSOs have the alternative to reduce this expenditure by investing in deeper fiber architecture to reduce the length of the coaxial loop plant in the HFC network, thereby reducing or eliminating the cascade of amplifiers and the OPEX for the maintenance of these amplifiers. Another important consideration for the MSOs is to reduce to the minimum the back-office equipment modifications (cable modem termination system, CMTS, RF video equipments) at the distribution hub, as well as the cable modems and set-top boxes at the customer locations. Vendors are proposing to the MSOs a fiber-to-thehome network architecture using active optical node and amplifiers distributed in the access plant, feeding RF network interface unit at the customer location. In this case MSOs can leverage the bandwidth of FTTH without having to change back offices and customer premises equipments.2 1
CableLabs http://www.cablelabs.com/ As an example of FTTH network architecture optimized for MSO, reader can refer to CommScope Bright-Path solution at http://bb.commscope.com/eng/solutions/fttx/index.html. 2
1 The Anatomy of an Access Network
11
1.7 Wireless Access Network Wireless access network architecture is used since many years to transmit voice over Internet protocol (VoIP) and high speed Internet (HSI) data. Pre-wiMAX technologies have demonstrated typical point-to-multipoint performance of 5–6 Mbps over a distance of 8–10 km. This network infrastructure has not demonstrated performance sufficiently reliable to be used to stream IP video, particularly for high definition signals or for multi-streams of standard definition video signals. The limitation of 5–6 Mbps says it all. Not enough to transmit HD content, too much at the limit to even transmit one stream of standard definition content. WiMAX and long-term evolution (3GPP LTE) will provide technologies that will improve the performances stated above. These technologies are not deployed in the field yet, therefore, the exact specifications associated with these new technologies are not neither precise yet. About 20 Mbps downstream and 5 Mbps upstream over reasonable distance are proposed in point to multipoint. This could position wireless access network as an alternative for delivering video signals as well. This architecture should never be a candidate for multi-streams HD transmissions, but if it can transmit two video streams, one high definition and one standard definition, along with voice and HIS data, this might create an interesting alternative, particularly for rural areas where the cost of deploying wire line architecture is prohibitive.
1.8 Summary We have seen in this chapter that many technologies can be used to connect the subscribers in the last portion of the network, the access network. All these technologies provide different levels of performance in terms of bandwidth or downstream line rate and upstream line rate. It is this bandwidth or these downstream and upstream line rates that will determine how many simultaneous services the customers will be able to consume. The main driver that has pushed the operators to invest and improve their access networks in the last years has been the capacity to transmit video most effectively. The common denominator to all the network architecture presented is the fiber-optic infrastructure, or how deep or close this fiber-optic infrastructure is to the subscribers. Without a doubt, the deeper the fiber architecture is and the more expensive will be the project, but the more competitive the services will be, particularly the number of video streams in high definition to be delivered to the subscribers. The other important parameter, greatly influenced by the proximity of fiber to the subscriber, is the upstream performance of the subscriber connection. This parameter will gain its importance the more the subscribers will generate or share content from their location, again particularly video content.
12
Nicholas Gagnon
References 1. Alcatel-Lucent (2007). Deploying Fiber-to-the-Most-Economic Point. Technology White Paper, Alcatel-Lucent. Available online: http://www1.alcatel-lucent.com/com/ en/appcontent/opgss/23168 DeployFiber wp tcm228-336491635.pdf 2. Gary Kim (2007). Colliding Business Models: IPTV at the Intersection of Media and Communications. IP Business, ChannelVision. Presentation available online: http://www.ptc.org/past events/ptc07/program/presentation/T12 GaryKim.pdf 3. Verizon FiOS (2006). Verizon Communications Inc. FiOS Briefing Session. Available online: http://investor.verizon.com/news/20060927/20060927.pdf
Chapter 2
Drivers for Broadband in Europe Evi Zouganeli, Kristin Bugge, Santiago Andres Azcoitia, Juan P. Fernandez Palacios, and Antonio J. Elizondo Armengol Elizondo
Abstract In this chapter we present the current status of broadband deployment in Europe and discuss the drivers for further deployment and the expected evolution in terms of the market, the services, the choice of technical solutions, and the main players. The main drivers for broadband are identified and discussed in the context of the European environment. We present an overview of the broadband regulatory framework in Europe as well as expected developments and discuss its impact on the evolution of broadband in this part of the world.
2.1 Introduction Datacom services have become an integral part of our everyday lives in the course of the past decade. Information, communication, and entertainment are converging, the PC and the TV are merging with our phone. The changes we are witnessing in the access network today are at least as dramatic as the changes introduced by the telephone at each home in the previous century. The access network is being literally transformed from a network of plain old telephone services at 64 kbit/s carried on Evi Zouganeli Telenor R&I, B5d, Snaroeyveien 30, 1331 Fornebu, Norway, e-mail:
[email protected] Kristin Bugge Telenor Nordic, Snaroeyveien 30, 1331 Fornebu, Norway, e-mail:
[email protected] Santiago Andres Azcoitia Telefonica I + D, Emilio Vargas 6, 28043 Madrid, Spain, e-mail:
[email protected] Juan P. Fernandez Palacios Telefonica I + D, Emilio Vargas 6, 28043 Madrid, Spain, e-mail:
[email protected] Antonio J. Elizondo Armengol Telefonica I + D, Emilio Vargas 6, 28043 Madrid, Spain, e-mail:
[email protected]
A. Shami et al. (eds.), Broadband Access Networks, Optical Networks, c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-92131-0 2,
13
14
Evi Zouganeli, Kristin Bugge, Santiago Andres, Juan Fernandez, and Antonio Elizondo
copper wires, or twice that with an ISDN line, to several Mbits with the introduction of xDSL and 100 Mbit and beyond with the deployment of fiber to the home. Considerable investments have been channeled to the rollout of broadband networks during recent years. Deployment started in Japan, followed by Korea, the USA, Hong Kong, and several other smaller scale deployments. Fiber has reached several millions of people during the last couple of years. By comparison, the evolution of fiber-based access has only been quite modest in Europe that has followed suit after the volume-led price reductions due to the deployments elsewhere around the globe. Some exceptions are the Scandinavian countries, The Netherlands, and some few others. xDSL has been the decidedly main broadband access technology in Europe followed by cable modem (hybrid fiber coax). Fiber-based access (fiber to the home, FTTH; to the building, FTTB; fiber to the x, FTTx) and fixed radio access are now entering the market. DSL has relatively limited coverage in some countries and similarly the cable TV networks and the cable modem upgrading are limited to densely populated areas. To cover the residual broadband market, technologies like WiMAX and mobile broadband are needed and indeed being deployed. Broadband facilitates social and economic welfare, economic development and prosperity. European officials are aware of the importance of broadband deployment. Regulation sets the rules of competition and is therefore one of the prime determining factors for the evolution of broadband. However, quite unlike the situation in the USA, the regulatory environment in Europe has unfortunately not encouraged broadband deployment from its incumbent operators. As a result, the bulk of fiber deployments in Europe has been initiated by newcomers in the broadband market and local initiatives have been of relatively small scale and local character. In this chapter, we present the current status of broadband deployment in Europe and discuss the drivers for further deployment and the expected evolution in terms of the market, the services, the choice of technical solutions, and the main players. The main drivers for broadband are identified and discussed in the context of the European environment. We present an overview of the broadband regulatory framework in Europe as well as expected developments and discuss its impact on the evolution of broadband in this part of the world.
2.2 Telecom Trends in Europe The telecom industry is changing rapidly and is currently characterized by • • • •
Decline in fixed voice and growth in broadband and mobile minutes Customers becoming more demanding High rate of technology innovation and rollout Competition moving from pure price to price, service and segment-specific offers
2 Drivers for Broadband in Europe
15
2.2.1 Decline in Fixed Voice and Growth in Broadband and Mobile Minutes Decline in fixed voice – Austria has the highest level of fixed mobile substitution in western Europe [1]. Figure 2.1 shows the development of voice telephony volumes in Austria in the fixed net and mobile communication networks (in millions of minutes per month).
Fig. 2.1 Development of voice telephony volumes in Austria in the fixed and mobile net. Source: Telekom Austria, “Annual Report 2007”
In Norway, fixed voice accounted for 56% of outgoing minutes by the end of 2005, 37% by the end of 2007 and is expected to fall to 29% by the end of 2008. In Spain, revenues from fixed voice traffic decreased at a rate of 10% (similar declining for traffic volume) in 2006 and 6% in 2007. On the other hand, a considerable growth in mobile minutes has been evident. Outgoing mobile minutes in Norway, for example, increased by 14% from 2005 to 2007, and similarly in Spain it experienced a yearly increment of 19.23% in 2007. At the same time there is a steady growth in broadband subscribers. The number of broadband lines in Europe is growing at a rate near 50% in 2007, as discussed in more detail later in this chapter. As a consequence, the revenue breakdown for fixed operators is changing, shifting from voice traffic-dominated revenues toward broadband access-dominated revenues. In the case of mobile operators, revenues from data traffic are still limited, though growing significantly and expected to become more important.
16
Evi Zouganeli, Kristin Bugge, Santiago Andres, Juan Fernandez, and Antonio Elizondo
2.2.2 Customers Becoming More Demanding Individualistic and demanding customers. Customers are more familiar with modern communication services and develop more segment-specific needs. They are less tolerant to poor functionality and customer service and more willing to consider switching to an alternative operator. Large corporate customers demand interEuropean services. Price consciousness. Despite the increasing quality awareness, price persists as the most important purchasing parameter. Fixed pricing is expanding from broadband into mobile. Buying of bundles is driven by convenience and price. Short-term growth is in many cases a direct consequence of an operator’s price position. Demand for access-independent services. Messaging is the most important accessindependent business application (an application that works across different accesses such as broadband, wireless, and mobile). Films, music, photos, and games are rapidly gaining importance as access-independent consumer applications. New customer interface. “One stop shopping” is of increasing importance both with consumers and businesses. Online shopping and related services are being extended from broadband to mobile. Business customers require self-provisioning of service elements. Demand for online access everywhere. Mobile broadband exploded by the end of 2007. Fifty-seven percent of all Europeans plugged into the Internet in 2007. About 75% of all Internet users were online between 5 and 7 days a week, an increase of 61% since 2004. PCs are becoming portable and personalized, moving from school and work towards entertainment, at the same time that the Internet is moving into and conquering TV.
2.2.3 High Rate of Technology Innovation and RollOut Higher access network capacities. During the last years, network access speed has increased significantly for mobile, wireless, and broadband accesses. This trend will continue over the coming years. The HSDPA technology (3.5G) enables downstream speeds of up to 7.2 Mbit/s in the mobile networks, while fiber and VDSL2 technology enable 20–100 Mbps access capacity in the fixed broadband network. IP service platforms. Service platforms will converge into IP-based platforms, enabling access-independent services. IP-based service platforms will also communicate more easily between themselves, thus opening up for innovative product bundles. Increased functionality in terminals. New terminals are entering the market with increased functionality. Mobile phones with similar picture quality as today’s digital cameras are available, mobile phones can already handle as much storage as many MP3 players and more mobile phones will support both 3G and WLAN at a low cost. Ninety percent of new laptops already have built-in WLAN, and manufacturers will soon embed HSDPA into their laptops.
2 Drivers for Broadband in Europe
17
FMC is happening now. The above trends facilitate fixed/mobile convergence (FMC) so that customers can get access to a broad range of services “from all terminals anywhere.”
2.2.4 Competition Moving from Pure Price-to-Price, Service- and Segment-Specific Offers Incumbents are still dominating - Incumbents defend their strong market shares and are fighting back with aggressive bundling (broadband, fixed voice, mobile, TV). Their infrastructure investment is heavily focused on 3G and broadband. A number of traditional and innovative challengers – incumbents are being challenged by a blend of focused players, including business focus players, flat fees players, young-at-heart players, and triple-play players. European telecoms suffered one of the worst result seasons for years in Q407, as incumbents saw broadband market share under pressure. Moreover, capex guidance was higher than expected and is expected to continue to increase. FTTH/B is already being rolled out by alts, municipalities, and utilities and Incumbents are expected to follow. New business models – portals such as Yahoo, Google, and MSN – are fighting for the position as the leading search and communication platform. Advertising is their most important income stream. Skype/Ebay/PayPal represent an innovative combination of competences.
2.3 Drivers for Broadband Services So far the number of households determines residential broadband access demand whereas in the mobile market individuals are the driver. However, there are some limits. One barrier is older people who need time to adapt new broadband technology. This will probably result in the development of easy-to-use terminals, IPTV programs, and tailored services specifically targeting the needs of this group. Another current barrier is the availability of PCs in the households. Broadband penetration of PC at homes was 79% in Q4. Figure 2.2 shows the PC penetration among western European counties based on OECD statistics [2]. The main part of the western European countries has a PC penetration larger than 70%. The yearly increase is estimated to be about 3%. Students, newly established persons, and persons with flexible home locations may prefer to use mobile broadband access instead of fixed mobile access. Mobile broadband puts further pressure on incumbents to improve the fixed broadband offering significantly with faster broadband or IPTV in order to differentiate fixed broadband from mobile broadband.
18
Evi Zouganeli, Kristin Bugge, Santiago Andres, Juan Fernandez, and Antonio Elizondo 90,0
Denmark Sweden Netherlands Luxembourg Germany
70,0
Norway United Kingdom Finland Switzerland 50,0
Austria Ireland Belgium Spain France
30,0
Italy Portugal Greece 10,0 2000
2001
2002
2003
2004
2005
2006
Fig. 2.2 Evolution of PC penetration in western Europe. Source: OECD statistics
The use of the broadband lines has changed from traditional Internet surfing and telephony to more capacity-demanding services like online gaming, downloads and streaming, TV services, Interactive TV (video-on-demand). The top five contributing services are [3] as follows: 1. 2. 3. 4. 5.
Security IP telephony Gaming online Music IPTV, “Telco TV”, or video content over broadband TV.
In terms of value-added services, Telefonica leads in Europe with tailor-made products for SMEs and SOHOs. These include services like IT security using an ASP model. VAS will be one of the main areas of broadband growth for the group going forward (16% of BB ARPU in 2006 and management targets around 30% of ARPU by 2010). Security: Security remains the single most valuable value-added service. This is not surprising, as security services are essential for every broadband user, due to the always-on connection and consequent attraction for Trojans and other malicious software. IP Telephony: IP telephony services have overtaken security to be the valueadded service that generates the greatest revenue. In some markets, such as France, IP telephony has won a significant share of the telephone market. VoIP accounted for around a quarter of all telephony traffic in France by the end of 2006. Even with tariffs that are lower than traditional PSTN charges, the growing size of the VoIP subscriber base means significant revenue. Within France, VoIP was responsible for almost a quarter of voice traffic originating from fixed phones. French regulator ARCEP reported 6.6 million VoIP subscriptions at the end of 2006. Of these, 4.3 million were subscriptions in addition to an existing PSTN subscription. But 2.3
2 Drivers for Broadband in Europe
19
million VoIP subscriptions were over a fully unbundled (or “naked”) line, with no PSTN service on that line. The growth in VoIP has stopped the decline in fixed subscriptions (caused by people abandoning fixed telephony altogether in favor of mobile). In Norway, an access-independent service provider named Telio launched VoIP as early as 2004. For almost 2 years Telio had a large price advantage compared with traditional POTS and their customer base grew rapidly. In recent quarters Telenor has moved aggressively into VoIP taking over VoIP market leadership. Telenor has eliminated the price difference between PSTN and VoIP for most consumers. “Naked” ADSL in Norway will therefore have less impact forwarding the future as a PSTN customer taking ADSL will pay the same as a VoIP customer on naked DSL. Year-on-year growth is reduced from 62% by Q206 to 11% by end 2007 and now almost all new VoIP customers are fiber- or HFC broadband customers. Gaming Online: From a revenue perspective, it is the fact that these games charge a monthly subscription that is important. Some charge a monthly fee plus additional one-off payments for expansion packs. An increasing number of games consoles (almost all Microsoft Xbox and Sony Playstation) are online enabled. Gamers can play alone or with other people. Other titles are using a different business model. Runescape has free-to-play content, and more people play this than subscribe to the fuller version of the game. Music Downloads: There are two ways of obtaining music online where music is paid for and delivered digitally, mostly over broadband. 1. Buying tracks to download to a PC hard-drive or other device and/or to burn to CD. 2. Purchasing tracks, ring tones, or videos via a mobile network. By early 2007, digital music accounted for 10% of global music sales. By early 2008, this had risen to 15%. Despite this strong growth, the biggest issue for the music and digital downloads industry is that the growth in digital consumption has failed to boost the overall value of the music market. Consumers are substituting their traditional CD spending for digital, rather than increasing it. At the end of 2007, CD sales were down by 11% in the UK. The continued use of illegal P2P networks by consumers to obtain music for free still significantly hinders market growth. Illegal downloading is also affecting mobile sales. There remains a long way to go before the penetration of digital devices and the acceptance of digital downloads by the majority of consumers shifts the revenue balance to digital. That shift will also increasingly rely on revenue from new channels (mobile) and new content (video). For players in the digital market, margins remain thin as record companies continue to pressure distributors to maintain higher prices and take a small percentage per track. But there are three positive emerging trends. 1. The sharp growth in mobile: 2007 has seen mobile music truly hit the mass market, with a number of significant product launches during the year. 2. The growth of music video.
20
Evi Zouganeli, Kristin Bugge, Santiago Andres, Juan Fernandez, and Antonio Elizondo
3. The emergence of a new subscription model, in which music is bundled with other services or devices such as a broadband subscription or a phone, with record companies and artists receiving a cut of each sale. IPTV: TV over broadband became increasingly widespread during 2006 and 2007. The theory is that IPTV will enable telcos to compete with the triple and quadruple play offerings of the cable companies. But putting this theory into practice involves overcoming many financial, regulatory, and technical obstacles. The business case for delivering video content over broadband is not straightforward. It depends on a host of local factors, including content availability and licensing, network type and regulation, as well as the local penetration and cost of cable, satellite or digital terrestrial broadcasters. Because of the high cost of content, profits from IPTV services are likely to be modest. It is the ability of these services to reduce churn to cable competitors that is most important for the moment. However, increasing deployment of fiber, falling equipment costs, and increased acceptance of IPTV by content owners means that the viability of IPTV services is continuing. The number of European people using IPTV services increased to almost 5 million IPTV subscribers by the middle of 2007. Around half of these were in France, where there has been a strong performance from the incumbent, France Telecom, in response to the early deployments of rival operators Free and Neuf. The French IPTV market is now well beyond the stage of “early deployment.” All the major European incumbents now have IPTV products up and running. But many customers that watch IPTV are doing so for free and are not therefore generating any value-added revenue at the moment. In October 2007, Sweden switched off the last region for analogue TV and we saw this helping IPTV. TLSN launched its IPTV platform in 2005 but the real marketing push has been since the start of 2007. TLSN now has 304k IPTV subscribers but is currently giving it away for 12 months. Customers can then choose a package which is free (8 channels), SEK69/month (15 channels), or SEK65/month (36 channels). Telefonica is pushing its TV offer via ADSL (Imagenio) and it plans to grow its subbase to 1.2–1.4 million by the end of 2010. At the end of Q4 it reached 511k subs. Telefonica double-and triple-play offers (Duo and Trio) have been very successful.
2.3.1 Increasing Bandwidth Demand Broadband competition in the Nordic was in the beginning mainly characterized by operators and service providers offering more capacity for the same price. The last 2 years there has been more focus on services due to new telecom trends like simultaneous use of multiple broadband terminals, interactive TV, and video-ondemand, HDTV and competition. In many countries cable TV operators have entered the broadband market in densely populated areas offering capacity up to 50 Mbits. They are offering multiple play like Internet, VoIP, TV (HDTV), VoD, PVR, and mobile services to increase their ARPU.
2 Drivers for Broadband in Europe
21
Fig. 2.3 Need for capacity in relation to services used
Figure 2.3 shows customers need for capacity based on what kind of services they use. File sharing describes the swapping of digital files from computer to computer. Copying files between machines is, of course, the whole point of the Internet and this service is increasingly used. File service is the service shown in the figure that requires the highest upstream capacity. In spite of vigorous activity by the industry and lawmakers, illegal file sharing still counts for a significant proportion of Internet traffic. P2P networks continue to flourish, with those operating “outside of the law” developing increasingly sophisticated tactics for dodging the watchful eye of the authorities and ISPs. HDTV requires high downstream capacity and is the reason why most operators need to upgrade.
2.4 Broadband Market in Europe 2.4.1 Broadband Lines, Penetration, and Growth According to Eurostat [4], the number of broadband lines shifted over 90 millions in 2007 for the EU15 although broadband growth, i.e., the rate of increase, suffers from a continuous decline in the last 4 years. The situation varies among different countries regarding broadband penetration and growth, as it is represented in Figs. 2.4 and 2.5. Lines in the growth versus penetration diagram represent mean values for the EU25, 49.77% in penetration and 43% in annual growth rate. The bubble size indicates weight of broadband market. These divide the market into four segments:
Evi Zouganeli, Kristin Bugge, Santiago Andres, Juan Fernandez, and Antonio Elizondo 90
100
80
90
70
80 70
60
60
50
50 40
40
30
Growth (%)
Millions
22
30
20
20
10
10 2007m07
2007m01
2006m07
2006m01
2005m07
2005m01
2004m10
2004m07
2004m01
2003m10
2003m07
2003m01
2002m10
2002m07
0
Fig. 2.4 Broadband lines and annual growth in the EU15. Source: Eurostat
Fig. 2.5 The European broadband market at a glance. Source: Eurostat
1. The upper left sector groups countries with limited growth and high penetration. These consolidated markets correspond with nordic countries. Broadband demand is near the maximum and market is developing toward higher capacity access technologies, for instance, FTTH connections. 2. The lower left sector groups countries with low penetration and low growth rates. These are Mediterranean countries with an important percentage of broadband lines. Market may be developed by “access offers” at competitive prices, allowing an increase in the adoption rate and trying to stimulate the slowed down broadband market.
2 Drivers for Broadband in Europe
23
3. In the upper right corner, fast developing broadband markets are situated. It is good news for the EU that Germany, which accounts for the maximum number of potential broadband lines, still grows at a rate over the mean for the EU25. 4. The lower right sector groups emerging broadband markets in Europe. These countries count with latent demand due to a low penetration and a fast growth. In these cases market may progress upward in the chart as a fast developing broadband market or to the left toward the slowest developing sector. In this group it should be remarked Greece as an important unexploited market (broadband penetration of 7% due to broadband coverage of around 19%), and Central and east European countries such as Poland (with 6 million potential lines and growing at a fast rate), the Czech Republic, and Hungary.
2.4.2 Broadband by Technology Regarding technologies currently used for broadband access, xDSL technologies are mostly used in the EU (over 80% of technology share). An emerging offer of FTTx access is being reported, which is expected to act as a broadband upgrade, especially for consolidated demand looking for additional services (Fig. 2.6). Malta Cyprus Luxemb Latvia Estonia Slovenia Slovakia Lithuania Bulgaria Ireland Greece Hungary Czech Norway Romania Finland Austria Portugal Denmark Belgium Sweden Poland Netherla Spain Italy France UK Germany
DSL FTTx %Others
2
4
6
8
10
12
14
16
18
Millions
Fig. 2.6 Broadband in the EU: breakdown by technology. Source: EC Commission Report on Broadband. OECD Broadband Report for the FTTx statistics (2007) [5]
Market share for cable operators varies from one country to another, depending on the historical presence and coverage of their networks. For instance, in The Netherlands cable operators have an important share (over 35%), whereas they have much less presence in important large countries, such as France, Italy, or Germany. In most cases this is due to the fact that they entered the market later and had severe
24
Evi Zouganeli, Kristin Bugge, Santiago Andres, Juan Fernandez, and Antonio Elizondo
investment barriers (proportional to country size) that hindered reaching a significant coverage. The presence of DSL in eastern Europe is much less significant. Broadband markets have developed rapidly in eastern Europe in the last years due to the late launch of mass-market services (for instance, first DSL services 2004 in Bulgaria and 2005 in Romania). Cable TV operators have a stronger presence in the region, especially in Romania, where RCS&RDS is the first adopter of FTTB concept. This cable operator began converting in the 2006 the cable network infrastructure to FTTB and launched fiberlink, a FTTB service upto 50/4 Mbps per user that accounted for 19% of broadband market share in 2006. By the end of 2007 there were 1 million FTTH subscribers in the EU 31 and around 4.9 million homes passed. Currently, Europe is still largely lagging behind the USA (2 million FTTH subscribers) and Japan (11 million FTTH/B subscribers). At the end of 2007, FTTH Council in Europe has identified 201 FTTx projects in Europe of which 88 are new initiatives since mid-2005, as shown in Fig. 2.7 [6].
Fig. 2.7 Some significant FTTH/B European deployments at the end of 2007 [6]
The majority of subscribers are still concentrated in five countries – Sweden, Italy, Norway, the Netherlands, and Denmark. As it is shown in Table 2.1, the FTTH adoption rate varies significantly between countries. Table 2.1 FTTx statistics Europe Country
No. FTTx lines (thousands)
FTTx lines over BB lines (%)
FTTH subscribers over homes passed (%)
Sweden Italy Denmark Norway The Netherlands
417 248 173 95 64
16.18 2.63 8.56 6.12 1.19
44.8 13.2 13.7 60.6 28.6
Source: FTTx lines from OECD, FTTx subscribers/homes passed from IDATE for FTTH Council Europe.
2 Drivers for Broadband in Europe
25
Pioneer countries like Italy and Sweden are now presenting slow growths in terms of homes passed (28 and 22%, respectively). The panorama of FTTx deployments in Europe at end 2007 shows that the FTTH market in Europe continues to grow especially in terms of homes passed (+79%) reaching around 5 million, whereas the growth of FTTx subscribers is three times lower (23%). In fact, overall adoption rate (FTTH subscribers/Homes Passed) for FTTx lines is around 20% (1 million subscriber over 5 million HP), which is still low as it is expected to substitute other broadband technologies. There have been encouraging signs in 2007 for the future of FTTH in Europe in major European countries like • France, with the involvement of four leading broadband players in deploying FTTH/B and where Government and the National Regulator, ARCEP, are playing a key role in fostering OF in the access network. • Germany, where three major city networks announced ambitious FTTH rollouts (Munich, Hamburg, Cologne) • The UK where the government may intervene to promote the deployment of FTTH in front of the reluctance of the main British players • Spain, where there is no commercial offer of FTTx services in the end of 1Q2008, though it is in the short-term scope of the incumbent’s operator agenda. Some testbeds and non-commercial deployments have been performed in order to measure market response to FTTx. Launch of the first FTTx services is expected from Telefonica during the first half of 2008, whereas the main national cable operator have announced 100 Mbps broadband by upgrading to DOCSIS 3 technology. • Norway, where FTTH subscribers have grown 250% in 18 months. FTTB market in Norway is dominated by utility companies, which have so far invested more than 300 million euros in their fiber infrastructure to 100,000 customers. Telenor will deploy fiber and offer 3-play over fiber by the end of this year. Telenor also owns a cable operator named Canal Digital that offers multiple play in areas that cover about 1/4 of the population. Canal Digital has also announced 100 Mbps broadband by upgrading to DOCIS 3.0.
2.4.3 Broadband Offer Regarding broadband offer, the OECD reports the average download speed advertised per country, as shown in Fig. 2.8. Average advertised speed may be considered an indicator of maturity of FTTx offer in the country, as it boosts average speeds. FTTx means a tenfold increase in broadband downstream capacity and a 50-fold in the case of upstream bitrate. New broadband services should take advantage not only of the increment in access capacity but also of this symmetry which is particular of FTTx deployments. France seems to have a variety of broadband offers, as it derives from the mean advertised
26
Evi Zouganeli, Kristin Bugge, Santiago Andres, Juan Fernandez, and Antonio Elizondo
100 000 90 000 80 000 70 000 60 000 50 000 40 000 30 000 20 000
It a ly nl a Po nd rt Au uga st l ra N lia Lu orw x em ay U ni bo te d ur Ki g ng G dom U erm ni te an d St y at C es an ad a Sp a G in re H ece un ga r C ze Be y ch lgi R um ep u D blic e Sw nm itz ar k N erla n Sl eth ov er d ak lan R ds ep ub Au lic st r Ic i a el an Po d la n Ire d la n M d ex ic Tu o rk ey Fi
Ja pa Fr n an ce Ko Sw rea N ew ed Ze en al an d
10 000
Fig. 2.8 Average advertised broadband download speed, by country, Mbit/s October 2007. Source: OECD Broadband Report, October 2007
speed (even higher than Korea) and the number of agents involved in FTTx deployment. Politic incentives for infrastructure sharing enable private (Illiad) and public initiatives to catalyze FTTx deployment, even when French broadband xDSL market is not exhausted as may be the case for nordic countries like Sweden and Norway. This type of policy is a far cry from the government supported FTTx deployments in Asia and the positive consequence they have had. Offers of tens of Mbps are common in Japan and Korea, which are considered the most advanced broadband markets. If compared with the USA, there is a number of countries within the EU31 with higher average downstream rates.
2.4.4 Competition Competition is considered an important driver for broadband penetration. In the case of the Netherlands, which is the European country with the highest broadband penetration, it seems that the main factor that has boosted this percentage broadband penetration is competition between DSL and the cable operators. As an indicator of competition in the markets, the incumbents’ market share in the different countries is shown in Fig. 2.9. There is no clear correlation between the incumbent’s market share and broadband penetration or broadband growth in the most mature broadband markets. For instance, Denmark or Finland have high penetration rates and their incumbent has an important market share. On the other hand, France Telecom has a market share below 50% and broadband penetration and broadband growth in France are below EU mean values. These values are nonetheless quite similar in Spain where Telefonica has a market share near 60%.
2 Drivers for Broadband in Europe
27
100% 90% 80% 70% 60% 50% 40% 30% 20% 10%
ng
Ki
Ita ly N Sp et he ain rla nd Po s la Sw nd ed Be en lg D ium en m Po ark rtu g Au al st ri C ze Fin a ch la n R d ep u H bli un c ga G ry re ec Ire e Li lan th d ua Sl nia ov a Sl kia ov en Es ia to ni Lu La a xe tv m ia bo u C rg yp r N us or w ay
U ni
te
d
G
er
m an
y do Fr m an ce
0%
Incumbents Mkt Share (no resale)
Incumbents Mkt Share (with resale)
Fig. 2.9 Incumbents’ market share. Source: EC Report on Broadband (July 2007)
2.5 Broadband Infrastructure – Evolution and Trends 2.5.1 Current Broadband Access Technologies and Architectures Currently, telco’s broadband access networks in Europe are mainly based on xDSL (VDSL2 and ADSL2+). A main drawback of these technologies is its limited reach for high capacity links. For example, while ADSL2+ can provide up to 15 Mbps over loops shorter than 1 km, VDSL2 solutions could offer up to 50 Mbps but only for loops shorter than 200 m. The maximum bandwidth of xDSL is an inverse quadratic function of transmission distance. Hence fiber nodes need to be built within some hundreds of meters from the subscriber for xDSL to deliver triple-play services and beyond. In particular, three options for taking the fiber closer to the customer are often considered by network operators: • FTTH (fiber to the home). Fiber transport between the Central Office (CO) and the customer premises equipment (CPE). • FTTB (fiber to the building). Fiber transport between the CO and a remote aggregation node (e.g., DSLAM VDSL2) located in a multi dwelling unit (MDU). • FTTN (fiber to the node). Fiber transport between the CO and a remote node located in a street cabinet. WiMAX is another emerging broadband access technology in Europe. Currently there exist several WiMAX deployments in countries such as Germany, Spain, France, UK, and Italy. In most of these deployments, WiMAX is seen as possible replacement candidate for cellular phone technologies such as GSM and CDMA or as a “backhauling” technology for remote cellular operations. Furthermore, WiMAX is also seen as a good technical candidate for rural broadband deployments. Areas of low population density and flat terrain are particularly suited to WiMAX and its range. Therefore, as depicted in Fig. 2.10, telco broadband access
28
Evi Zouganeli, Kristin Bugge, Santiago Andres, Juan Fernandez, and Antonio Elizondo
network architectures are expected to be based on a combination of GPON, Ethernet star, xDSL, and wireless technologies.
Fig. 2.10 Example of telco’s broadband access architecture
Table 2.2 summarizes some technical characteristics of the main broadband access technologies currently used in Europe. Table 2.2 Comparison of broadband access technologies Technology Spectrum Data rate* Distance Mobility
WLAN OFDM/TDM Unlicensed 15 Mbps 50 m Portable
Cellular 3G T/F/CDMA Licensed 1 Mbps 2–5 km Mobile
ADSL OFDM Dedicated 8 Mbps 1 km Fixed
VDSL2 OFDM Dedicated 50 Mbps 0.2 km Fixed
Cable FDMA Shared 15 Mbps 1 km Fixed
FTTH/GPON Digital Shared 50 Mbps 8 km Fixed
WiMAX OFDMA Licensed 15 Mbps 2–5 km Mobile
* Typical out of the different possible data rates.
At this point, it is important to highlight the importance of regulatory aspects in the evolution of broadband access infrastructures in Europe. For example, according to EU regulation, new entrants could rent the elements of the incumbents’ networks with stipulated prices (by the National Regulatory Authority in each European country). Clearly, this policy regarding the opening of access networks had a large impact. It has been called unbundled local loop (ULL), by which incumbent operators are obliged to open their copper-based line access networks. Figure 2.11 shows the strategy of new entrants in access networks, as well as options to provide local loop sharing in current ADSL network.
2 Drivers for Broadband in Europe
29
Fig. 2.11 The ladder of investment. Local loop sharing options over current ADSL networks
2.5.2 Impact of FTTx Deployments on Local Loop Sharing However, given the increasing importance of FTTx deployments, we may witness a change of the most suitable access point(s) for the promotion of competition. In FTTN and FTTB deployments, unbundling of the local loop is assumed to take place at the remote node. In that case, the entrant operator can choose between climbing up on the ladder of investment by deploying its own fiber to the street cabinet (FTTN) or to the building (FTTB), or use wholesale broadband access (WBA), as illustrated in Fig. 2.12. In the case of phasing out MDF access, the importance of LLU as a means to derive competition may decrease compared to WBA, especially if entrant operators are not able to deploy their own networks toward the remote nodes.
Fig. 2.12 Local loop sharing options over FTTN/FTTB networks
On the other hand, in FTTH deployments, the entire (old) copper loop is replaced by optical fiber, along with the MDF and street cabinets, although some of these may be worth using for the optical distribution frames (ODF) and optical splitters. As
30
Evi Zouganeli, Kristin Bugge, Santiago Andres, Juan Fernandez, and Antonio Elizondo
depicted in Fig. 2.13, new entrants could choose between fiber local loop unbundling (FLLU) and WBA.
Fig. 2.13 Local loop sharing options over FTTH networks
Fiber local loop unbundling feasibility would depend on the type of architecture chosen by the incumbent operator: • In star architectures based on dedicated point-to-point fibers per user the unbundling method would be similar to that used today for copper. • In PON architectures (e.g., GPON or EPON) the unbundling of the subscriber fiber loop could only be applied connecting the splitter and the entrant operator’s ODF.
2.6 Overview on European Regulatory Framework 2.6.1 European Regulation Broadband changes the rules of the game for the traditional telecom market, fostering innovation in technology and services, and making the current regulatory framework quite inadequate. The regulatory environment is a determining factor with respect to the introduction of broadband, the players that establish themselves in this market and their competitive strength. National Regulatory Agencies (NRA) have the power to change the market dynamics and hence regulation regarding broadband is paramount when analyzing the broadband market. NRA policies aim at preventing incumbent operators from retaining an unfair competitive advantage and domination of the market due to inherited infrastructure and market position. It has been argued that although FTTx deployments represent a new investment, they can to an extent take advantage of former access infrastructure and could thus become subject to regulatory rules in this sense. Current regulatory measures include among others:
2 Drivers for Broadband in Europe
31
• Regulation of wholesale broadband access market, which includes bitstream services that permit the transmission of broadband data in both directions and other wholesale access provided over other infrastructures. • Regulation of unbundled access (including full unbundling and shared access) to metallic loops and sub-loops for the purpose of providing broadband and voice services. The European Regulators Group (ERG) for electronic communications networks and services has been set up by the commission to provide a suitable mechanism for encouraging cooperation and coordination between national regulatory authorities and the commission, in order to promote the development of the internal market for electronic communications networks and services, and to seek to achieve consistent application, in all Member States, of the provisions set out in the Directives of the new regulatory framework. ERG launched in 2007 a public consultation on the regulatory principles of next generation access (NGA) that resulted in a document with the title “ERG Opinion on Regulatory Principles of NGA” [8]. This constitutes the ERG common position recommended to NRAs. In this document ERG recommends that NRAs develop a regulatory approach early enough to provide all market players the necessary predictability. The market definition of wholesale local access should be modified to become technology neutral rather than being restricted to copper access, whereas the definition of wholesale broadband access does not need to be modified as it applies also to increased speed. The ERG proposes that • Layer 0 products, such as physical access to ducts, are enforced to assist unbundling • Layer 1 products for FTTC deployments are introduced • Layer 2 products such as next generation access and wholesale active line access are introduced depending on the effectiveness of other measures. The European communications regulatory package has been converted into national right in 2004. The recommendation of the Commission regards 18 telecommunication markets, which are examined by the NRAs. Under the procedures set out, the NRAs are required, in consultation with industry, to analyze their national markets for electronic communications and propose appropriate regulatory measures to address market failures, as well as notify their findings and proposed measures to the Commission and other national authorities. More precisely, the NRA, in accordance with competition law principles, must define the boundaries of the relevant market, assess whether anyone or more players is dominant (or has significant market power, “SMP”) in this market, and where operators are found to be dominant, propose appropriate regulatory remedies to ensure effective competition. A final mandate is not expected to be available before 2009 at the earliest, and the NRA are expected to abide with this mandate.
32
Evi Zouganeli, Kristin Bugge, Santiago Andres, Juan Fernandez, and Antonio Elizondo
2.6.2 Regulatory Situation in Some European Countries A brief presentation of the situation in some European countries is presented in the following. In the UK, a public consultation was launched by OFCOM [8] in the last quarter of 2007 in order to gather the stakeholder’s view on the policy approach to NGA. In the UK, there is a service unbundling regime with one wholesale operator, BT Wholesale, and a number of retail operators that use the infrastructure offered by BT Wholesale to provide their services. This seems to be working well, apart from the fact that there does not seem to be a lot of FTTH focus in BT’s strategy at the moment. In France, the National Regulatory Agency ARCEP identified three major questions for FTTH deployment: France Telecom’s ducts are essential facilities and access to these must be guaranteed on non-discriminatory terms and conditions to competitors. FT communicated its offer to competitors in December 2007. Access to buildings are considered a bottleneck for all players and thus French government has announced law proposals in order to require operators to share the last part of their fiber network. Municipalities are encouraged to play a key role in NGA investment by granting right of ways, coordinating deployment works, laying extra capacity ducts, or even deploying fiber. In Germany, the incumbent Deutsche Telekom wished to deploy FTTC as an intermediate solution toward full fiber access. The Commission expressly pointed out that the bill provides for bitstream access regardless of the technology used, which means that the VDSL network is included. The Commission made it clear that access must be provided as soon as the new infrastructure is complete. Competitors of Deutsche Telekom announced that they would be willing to cover part of the investments in this new network, which is expected to cost around 3 billion euros, in accordance with their current shares of the DSL market. In September 2006 the German regulator informed the incumbent Deutsche Telekom that it must open its VDSL network to competitors, and it had 90 days to propose standard prices for access. In the Netherlands, the national regulator OPTA favored initially sub-loop unbundling. Subsequently it has modified its position and suggests that commercially favorable sub-loop unbundling schemes are implemented by the incumbent KPN. This may mean that sub-loop unbundling may take place only, e.g., at the largest exchanges or for big business customers. KPN has been encouraged to come to an agreement with local loop unbundling operators as well as to provide a high-quality layer 2 wholesale product that facilitates innovation and service differentiation. Following incentives given by OPTA, KPN has signed memoranda of understanding with several alternative operators regarding access to KPN’s physical infrastructure. In Spain, the national regulator CMT launched a document on Principles of Future NGA Regulation [9] in order to give incentives for investment, promote effective and sustained competition, and protect investment already done by alternative operators. CMT recognizes the need to identify competitive and non-competitive zones based on objective and measurable variables and states that FTTH will not be
2 Drivers for Broadband in Europe
33
regulated using unbundling remedies due to technical problems to unbundle GPON deployments and in order not to compromise investment. There will be bitstream obligations for bitstream access wholesale services. In competitive zones these obligations will be subject to a sunset clause (i.e., it will be temporary) in order to encourage investment. Sub-loop unbundling is not considered a priority. Instead, a virtual unbundling has been defined to protect operators that already invested in ULL which is a bitstream service in the switching center where the operator is already collocated. CMT announces nonetheless the obligation to share access infrastructure for operators considered with SMP or even dark fiber leasing when infrastructure sharing is not possible. In Norway, the national regulator PT considers both dark fiber and optical wavelengths as transmission capacity resources that are subject to regulatory intervention since the incumbent, Telenor, has a strong market position in the market for transmission capacity. Regarding the broadband market, PT has given formal signals that any internal use of networks and technologies for delivering broadband services to the end-user will be subject to regulation according to the rules and regulations that apply within the designated “Market for broadband access connections.” This has earlier been interpreted to regard xDSL connections/bitstreams. PT has proposed that Telenor shall have an obligation to provide access wherever asked within the whole of Norway, however, after a protest the actual implementation of this has been moderated. In practice, the obligation to provide access to other operators is now limited to bitstream access to Telenor’s copper-based access network. This may change the moment Telenor deploys FTTH.
2.6.3 Discussion – The Impact of Regulation The European Commission has now the task to devise a regulatory framework for Europe’s future FTTH access network. Many indicators have pointed in the direction that incumbents’ FTTH networks will be unbundled. There are, however, many arguments against such a development. The US example shows what an impact a decision to deregulate the market can achieve – namely rapid deployment of new advanced infrastructure on a big scale. A strict regulation that removes the incentives from resourceful players to invest is arguably an effective counter-measure against EU’s own Lisbon objective mandating that by 2010 the EU shall become the most competitive and dynamic knowledge-based economy. Regulation shall provide the basis for fair competition at the same time that it promotes investment and innovation. As OFCOM states in [8] three principles are of key importance for an effective regulation: • • • • •
contestability maximizing potential for innovation equivalence reflecting risk in investment regulatory certainty.
34
Evi Zouganeli, Kristin Bugge, Santiago Andres, Juan Fernandez, and Antonio Elizondo
Contestability means that competition ought to drive investment where the opportunity to invest is contestable by as many parties as possible. Although regulation may be disproportionate in order to safeguard fairness, it shall nonetheless not inhibit efficient and timely investment. The timing of regulatory decision, or inaction, shall not result in foreclosure of options for competition in the future. At the same time, regulation must promote competition at a sustainable level which can imply introducing it at the deepest levels of infrastructure. It must promote a favorable environment for timely investment and stimulate innovation. Regulation ought to accommodate varying solutions for different products and geographies, as seen appropriate [8]. The regulatory uncertainty in Europe has put the continent’s strongest players “on hold” as it has arguably created an environment that inhibits fair competition, contestability, and innovation. The risk of ex ante regulation makes investment unjustifiable for European incumbents. The few examples of investment in FTTx by incumbents in Europe have been at very high risk and only under the immense pressure of disproportionately favored newcomers in the broadband market. It is hoped that the European regulator will rectify this situation soon by passing a clear and fair framework.
2.7 Summary In this chapter we have presented the current status of broadband deployment in Europe and evaluated the prospects for further deployments. The main drivers for broadband have been presented and discussed. Europe is a dynamic environment with demanding consumers that are willing to pay for broadband services, a high rate of technology innovation and rollout, and competition characterized by serviceand segment-specific offers. Broadband penetration varies a great deal across the continent with some countries in western Europe far ahead than the rest. The mean penetration of broadband for EU25 is almost 50% with a mean annual growth rate of 43%, however, some countries have reached 75% penetration whereas others are closer to 25% and even well below that. In general 1, eastern Europe lags behind western Europe. In general, the broadband offering in Europe lags also in terms of the average advertised download speed. The main technology used for broadband in western Europe is xDSL with coaxial cable a good second runner. The xDSL presence in eastern Europe is much less important since broadband markets have developed fast in recent years here, and cable operators have a strong presence. Fiber deployment in Europe lags quite significantly behind Asia and the USA. By the end of 2007 there are 1 million FTTH subscribers in EU31 and around 4.9 million homes passed. However, there is a formidable growth rate of 79% in terms of homes passed, with a less impressive rate of new subscribers at 23%. FTTx deployments are primarily done by newcomers rather than the incumbents although some change appeared on the horizon recently.
2 Drivers for Broadband in Europe
35
The main reason for the incumbent reluctance to invest in FTTx and for Europe lagging behind in terms of broadband deployment and FTTx, in particular, can be attributed to an uncertain regulatory environment. The regulatory uncertainty in Europe has put the continent’s strongest players “on hold” as it has arguably created an environment that inhibits fair competition, contestability, and innovation. The risk of ex ante regulation makes investment unjustifiable for European incumbents. The few examples of investment in FTTx by incumbents in Europe have been at very high risk and only under the immense pressure of disproportionately favored newcomers in the broadband market. It is hoped that the European Regulatory Authority will rectify this situation in 2009 by passing a clear and fair framework.
References 1. 2. 3. 4. 5. 6. 7. 8. 9.
Telekom Austria “Only Bad News”. Annual Report 2007. OECD, Households with Access to Home Computer (2000–2006). // http://www.oecd.org/ Point Topic, Broadband Money Makers: The Consumer BVAS Market. Eurostat. European Information Society Statistics 2008. // http://epp.eurostat.ec.europa.eu/ Organisation for Economic Cooperation and Development (OECD). OECD Broadband Portal 2008. // http://www.oecd.org/ IDATA FTTH European Panorama, Status & Dynamics, FTTH Council Europe Annual Conference, Paris, 28 February 2008. European Regulators Group (ERG). ERG Opinion on Regulatory Principles of NGA 2007. // www.erg.eu.int/ Office of Communications (OFCOM). Future Broadband. Policy Approach to Next Generation Access – Consultation 2007. // www.ofcom.org.uk/ Comision del Mercado de las Telecomunicaciones (Telecommunications Market Commision, CMT). Principles and Main Outlines of the Future Regulation of the Next Generation Access Networks (NGA) 2008. // http://www.cmt.es/
Chapter 3
Enabling Techniques for Broadband Access Networks Ton Koonen
Abstract A number of techniques are introduced which enable to deliver broadband services to residential users. The ultimate capacity is provided by optical fiber all the way to the user, thanks to the fiber’s huge bandwidth and very low losses. Also in combination with other media (twisted copper pair cable, coaxial cable, or radio link) fiber can considerably improve the capacity of the access network. Fiber access network topologies are discussed, including their economic aspects. For shared fiber topologies, a number of multiple access techniques are introduced, deploying time slot multiplexing, frequency multiplexing, code multiplexing, wavelength multiplexing, and combinations thereof. Wireless delivery of broadband services to mobile users is efficiently enabled by radio-over-fiber techniques, which consolidate the radio signal processing functions at a central site. Next to delivery by optical beams through fibers, broadband services may also be delivered by optical beams through free space, for short and clear line-of-sight links.
3.1 Introduction Access networks need to provide ever more bandwidth in order to keep up with the rising demands from residential users. The spectrum of services has broadened significantly in the past years and will expand further in the foreseeable future. Users are requesting that services are personalized, well fit to their preferences. They want video services of high-definition quality and want these on demand at any time. They like to participate in multi-party multi-media real-time games. They want to not only receive huge amounts of information, but also transmit these themselves. In the spirit of “everybody can become a service provider”, they want to upload large amounts of video information and alike. They need ways for fast file transfer, Ton Koonen Electro-Optical Communication Systems Group, COBRA Institute, Eindhoven University of Technology, Eindhoven, The Netherlands, e-mail:
[email protected]
A. Shami et al. (eds.), Broadband Access Networks, Optical Networks, c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-92131-0 3,
37
38
Ton Koonen
thus demanding from the access network high capacity in not only the downstream direction, but also upstream.
3.2 Fiber in the Access Network Much effort is made to squeeze the ultimate capacity out of the legacy access networks based on copper. Advanced DSL techniques are deployed in the twistedpair telephony network and cable modem techniques in the coaxial cable CATV network. However, these copper-based networks are reaching their limits now, and inevitably optical fiber with its inherent huge bandwidth and low losses has to come in. The higher capacities on the copper-based networks are achieved at the expense of a shorter reach, and thus fiber is making its way toward the user’s home [1]. Fiber is replacing copper cables in the feeder lines, handing over to the copper lines in cabinets put at the street curb site. These so-called fiber-to-the-curb (FTTC) solutions are illustrated in Fig. 3.1. In the (near) future, fiber will run all the way to the user’s home, providing the ultimate capacity in the so-called fiber-to-the-home (FTTH) topologies. twisted copper pair cable fiber LEX
DSL node
AN
fiber CATV AN
(a) Fiber–DSL
mini coax node
ampl.
coaxial copper cable
tap
(b) Hybrid Fiber-Coax
fiber fiber LEX fiber LEX
AN
FWA node
(c) Fiber-FWA
AN optical power splitter/combiner
(d) Fiber-to-Home
Fig. 3.1 Basic topologies for optical access networks. (LEX: local exchange; AN: access node; FWA: fixed wireless access)
3.2.1 Fiber-DSL In offering data rates up to 6 Mbit/s, asymmetric digital subscriber line (ADSL) modems may still enable twisted copper pair line lengths up to a few kilometers, which typically is sufficient to bridge the installed lines from the local exchange to
3 Enabling Techniques for Broadband Access Networks
39
the homes. However, for the upcoming very high-capacity digital subscriber line (VDSL) modems offering data rates beyond 50 Mbit/s, the copper reach typically falls below some 300 m. This leads to the fiber-fed VDSL node access network topology shown in Fig. 3.1(a). At the curb site, a cabinet housing a number of modems is placed, where each modem supports a point-to-point connection to a home. Advanced signal modulation formats are deployed in order to realize the capacity needed on the copper lines. These lines suffer from bandwidth limitations and from the crosstalk between the copper pairs as multiple lines are mounted in a single cable. Chapter 41 comprehensively deals with DSL technologies.
3.2.2 Hybrid Fiber-Coax In CATV networks, the interactive data traffic for the users is fit into the spectral slots between the bands of the CATV channels. Due to the limited slots available, and as the users have to share the coaxial feeder cable, increasing the data capacity per user requires a reduction of the number of users per fiber-coax node, and hence a further penetration of fiber toward the user’s home. A typical example of such a hybrid fiber-coax (HFC) access network is shown in Fig. 3.1(b). The fiber feeds a mini-coax node, in which opto-electronic conversion takes place and from which a bus-shaped coaxial cable network with taps and RF amplifiers leads to the user’s homes. In the CATV headend station, the set of CATV channels can be modulated on the optical carrier by means of analog subcarrier multiplexing techniques, keeping the conventional frequency-division multiplexed (FDM) format in which the CATV channels are offered via the coaxial cable network to the homes. In the gaps between the spectral bands reserved for the distribution of the TV channels and of the FM radio channels, interactive services are positioned (see Fig. 3.2). Taking into account the highly asymmetric traffic profile of the Internet, the band reserved for the data channels in downstream direction is considerably larger than the one reserved for the upstream direction. Together with the telephony (POTS) upstream services, the latter has been positioned below the downstream bands, so the shared bandwidth for upstream services is rather limited. In Europe, it typically is 5–65 MHz, and
Upstream
Downstream
POTS +data
5
30 48 /40
86
100
300
450
600
860
Fig. 3.2 Example of spectrum allocation in HFC system (POTS: plain old telephony service; NVOD: near video on demand)
1
Chapter 4: Vectored DSLs and The Copper PON (CuPON), by J. M. Cioffi et al.
40
Ton Koonen
in the United States 5–42 MHz. The data are modulated on subcarriers in multilevel modulation formats. For upstream, subcarriers spaced at typically 2.28 MHz are used, with DQPSK modulation offering some 3 Mbit/s as it is robust against the ingress noise incurred in this low-frequency band. For downstream, typically 64– 256 QAM is used on subcarriers spaced at 8 MHz, offering some 30–50 Mbit/s per 8 MHz channel.
3.2.3 Fiber-Wireless As the major part of the access network infrastructure costs is in the individual last link to the user’s home, especially in less densely populated areas, it is interesting to apply wireless communication techniques in this link. With fixed wireless access (FWA) techniques such as IEEE802.16 WiMAX operating in the 10–66 GHz range, capacities exceeding 100 Mbit/s can be transmitted wirelessly. Depending on the directionality of the antennas and the user density, the capacity over the air is shared by a number of users. Increasing the capacity per user implies feeding less users per antenna, by moving the antenna closer to the user and hence extending the fiber network closer to his/her home; see Fig. 3.1(c). The wireless communication functions, such as generation of the microwave carriers and modulation of the data signals upon them, are typically implemented in comprehensive circuitry next to the antenna. By means of radio-over-fiber techniques, these functions may also be consolidated at a central site, from which fibers bring the modulated radio signals to simplified antenna stations. Radio over fiber will be discussed further in Section 3.6. Chapters 52 and 63 are addressing the techniques for broadband wireless access in much more detail. On the same fiber feeder network wired services can also be provided, yielding a wired-wireless converged access network. Chapter 14 provides more details on the integration of WiMAX with fiber-to-the-home networks.
3.2.4 Fiber to the Home In order to realize a connection capacity per home beyond 100 Mbit/s (even beyond 1 Gbit/s is a realistic target in the foreseeable future), bringing fiber all the way to the home is commonly accepted to be the best future-proof solution. This fiber-tothe-home (FTTH) scenario can be implemented in various topologies, which will be discussed further in Section 3.4 of this chapter. Various multiple access techniques for a shared feeder fiber FTTH network topology are dealt with in Section 3.5 of this chapter. A typical example of an FTTH topology is shown in Fig. 3.1(d). The 2 3 4
Chapter 5: Enabling broadband wireless technologies, by Q. Rahman. Chapter 6: WiMAX, by A. Shami et al. Chapter 14: Integration of EPON and WiMAX, by R.S. Tucker et al.
3 Enabling Techniques for Broadband Access Networks
41
optical signal is brought all the way to the home by means of a common long feeder fiber, an optical power splitter/combiner, and last point-to-point fiber lines to the homes. As the capacity of an optical fiber is tremendously large, and its losses are far less than those of copper media or in free-space propagation, such an infrastructure principally discloses a virtually unlimited capacity to every home. Hence it may be considered the most future-proof (or forecast-tolerant, as some people say) access network solution.
3.3 Basic Optical Access Network Components 3.3.1 Optical Fiber The key building block of an optical access network obviously is the fiber itself. It basically is a silica glass cylindrical structure, where the core has a slightly higher refractive index than the surrounding cladding. There are various types of optical fiber, in which the light propagates with different characteristics; see Fig. 3.3. Depending on the dimension of the core, and on the wavelength of the light, there may be multiple modes which can propagate through the core. For a small core size (typically 9 μm) and a wavelength beyond the so-called cut-off wavelength (typically around 1.2 μm), Maxwell’s equations dictate that the light pulses can propagate in only one mode: this is the so-called single-mode fiber. For larger core sizes (typically 50 m and more) and/or shorter
cladding
r 9 m
core
n(r) 125 m
(a) Single-mode fiber r
50 to >950 m n(r) 125 m – 1 mm
(b) Step-index multimode fiber r
50 to 200 m n(r) 125–250 m
(c) Graded-index multimode fiber Fig. 3.3 Light propagation in various types of optical fiber
42
Ton Koonen
wavelengths, the light pulses can propagate in multiple modes: this is the so-called multimode fiber. As these modes usually follow different path lengths through the core, they arrive at different time instants at the fiber output. This so-called modal dispersion causes smearing out of the optical pulses, causing them to overlap and hence limiting the repetition speed with which pulses can be sent. For multimode fibers with a core having a refractive index which is constant across the core (a so-called step-index multimode fiber), the arrival time differences are large, and hence the modal dispersion is severe. More sophisticated multimode fibers have a core refractive index profile which is graded (a so-called graded-index multimode fiber), i.e., the refractive index is gradually decreasing from the center of the core toward its edge. Thus, the effective optical path lengths of the different modes can be equalized, and the modal dispersion is strongly reduced. The bandwidth×length product of graded-index multimode fiber is therefore significantly larger than that of step-index multimode fiber. On the other hand, its refractive index profile is more complex, and hence graded-index multimode fiber is more costly than step-index multimode fiber. In single-mode fiber, by definition, modal dispersion is absent, and hence the data rates at which the pulses are sent can be much higher. However, as the light from the transmitting optical source (typically a laser diode) is not monochromatic, there will be some so-called chromatic dispersion, which is caused by the wavelengthdependent propagation speed of the mode due to the wavelength-dependent waveguiding mechanism and the wavelength-dependent refractive index of the silica glass material. At a particular wavelength, the waveguide dispersion is fully offset by the material dispersion: at this so-called zero-dispersion wavelength (typically 1310 nm) the single-mode fiber exhibits no dispersion, and hence a virtually unlimited bandwidth. The attenuation of a silica fiber depends on the purity of the material. Tremendous progress has been achieved since the first light guiding in optical fiber was demonstrated in 1966 by Kao and Hockham; the first fiber made by Corning in the 1970s had losses of 20 dB/km, whereas today’s fibers have losses as low as 0.17 dB/km (in the 1.5 μm wavelength region, where losses are the lowest). The very low loss characteristics in combination with the minimal dispersion of single-mode fiber make it the prime choice for long-haul high-capacity links such as transoceanic links, and for medium-haul links such as in metropolitan area networks. As illustrated in Fig. 3.4, such links, therefore, exploit the 1280–1630 nm wavelength range representing a frequency range of no less than 50 THz. Obviously, this tremendously wide range can only be exploited fully by using optical wavelength division multiplexing (WDM) techniques, by which a large number of wavelength channels can be fit into this range. For instance, 1000 of these channels spaced at 50 GHz (appropriate for, e.g., 40 Gbit/s per channel) can be accommodated. Lowest losses are obviously most important in long-haul applications. These preferably use the 1530–1560 nm window, where next to the lowest fiber losses also the very powerful properties of erbium-doped fiber amplifiers are available (EDFAs), which can amplify optical signals directly in the optical domain without converting these to the electrical domain.
3 Enabling Techniques for Broadband Access Networks
43
0.6
EDFA Band
applications 0.5 0.4 0.3
applications
0.2 0.1 1100
1200
1300
1400
1500
1600
1700
Fig. 3.4 The enormous bandwidth of single-mode fiber
For short-haul applications, such as in access networks and in in-building networks, the losses per unit length are obviously less important. Hence the exploitable wavelength range in such networks can be considerably larger. It may be extended to the near-infrared range around 800 nm or even into the visible range (using red light sources around 650 nm, down to blue light sources around 460 nm). However, in those regions the (originally single-mode) fiber will no longer be guiding only one mode, but will operate in the multimode regime, thus giving in on its extremely high bandwidth×length performance. For short-haul links, typically in the immediate environment of the end user such as in in-building networks, the cost of installing a fiber infrastructure is an important factor when deciding which communication infrastructure will be installed. For long-haul links, there is no real alternative other than single-mode fiber. Multimode fiber, however, has a core size which is far larger than that of single-mode fiber, and thus splicing of multimode fibers and coupling them to modules require far less accuracy. Hence the installation costs of a multimode fiber infrastructure can be considerably less than that of a single-mode fiber; therefore, multimode fiber is the most popular fiber type for in-building networks. The handling efforts are even smaller when not silica, but polymer optical fiber (POF) is used. The polymer material is ductile, hence it is easier to pull the fiber through ducts without breaks, and large core sizes can be deployed without making the fiber too rigid. The installation of large-core POF in in-building networks thus may even be done by non-professionals. Step-index PMMA POF with 1-mm core diameter is becoming available commercially, and brings cheap do-it-yourself installation within reach. However, the attenuation characteristics of PMMA are by far not as good as those of silica. As shown in Fig. 3.5, PMMA POF is only suitable for use in the visible wavelength region. On the other hand, this eases installation and maintenance, as just by visual inspection one can already see whether a link is alive or not. Not only the attenuation per unit length, but also the bandwidth×length product for large-core step-index PMMA POF are significantly worse than for silica multimode fiber, let alone than for silica single-mode fiber. Hence the application of this fiber type is limited to very short-haul applications (below a few hundreds of meters such as inside buildings, down to a few meters such as inside cars).
44
Ton Koonen 5000
2000 1000
120dB/km
85dB/km 85dB/km
145dB/km
500
200 100 50 400
450 460 nm blue
500
550
600
520 nm green
650
700
750
800
850
650 nm red
Fig. 3.5 Attenuation of PMMA polymer optical fiber
Summarizing, for hybrid fiber access networks (FTTC fiber-DSL, HFC, fiberFWA) with fiber feeder lengths of several kilometers, and for FTTH access networks where typically a distance of up to 5 km needs to be bridged from the local exchange to the user’s home, single-mode fiber is the most appropriate medium. For shorter lengths, such as occur inside buildings like office buildings, apartment buildings, and residential homes, multimode silica or even polymer optical fiber is attractive thanks to their reduced installation costs.
3.3.2 Optical Power Splitter Optical power splitting devices enable the distribution of light from its input port to multiple output ports. They are therefore vital elements of optical point-tomultipoint access networks. A readily available basic optical power splitter is a 2 × 2 fused fiber coupler, schematically shown in Fig. 3.6. It is made of two fibers, which over a certain length are fused together. In this mixing region, due to proximity effects light can couple from one fiber into the other one. Some light is also lost during this process, causing the device to have the so-called excess losses. A structure with the same functionality can be realized in planar-integrated optics, by bringing two waveguides in close proximity over a certain length. The relation between the input and output powers is Pout,1 = x · ε Pin,1 + (1 − x) · ε Pin,2 Pout,2 = (1 − x) · ε Pin,1 + x · ε Pin,2 where the excess loss factor ε = (Pout,1 + Pout,2 )/(Pin,1 + Pin,2 ) with ε ≤ 1, and the slitting ration is x/(1 − x). The splitting loss from input port 1 to output port 1 is −10 lg(Pout,1 /Pin,1 )| Pin,2 =0 = −10 lg(ε) − lg x [dB]
3 Enabling Techniques for Broadband Access Networks
Pin,1
fiber
45
Pout,1
Pout,2
Pin,2 mixing region
Fig. 3.6 Fused fiber 2 × 2 power splitter
According to a fundamental law of optics saying that the luminance in a passive optical system can never increase, the losses between each pair of ports are the same when the direction of the light is reversed. A more general functionality needed is an N × N coupler. Such coupler can be made by fusing N fibers together. Alternatively, it can be made in an integrated optics process, for instance by coupling the input waveguides into a broad multimode slab waveguide where multimode interference can yield the appropriate power splitting among the output waveguides (Fig. 3.7). Pin,1
Pout,1
Pin,1
Pout,1 maxing region Pout,N
Pin,N
Fig. 3.7 N × N power splitter
The relation between the input and output powers is N Pout, j = ε i=1 xi j Pin,i where ε · xi j represents the coupling factor from input port i to output port j, taking N into account an excess loss factor ε = Nj=1 Pout, j / i=1 Pin,i with ε ≤ 1. For example, for a point-to-multipoint FTTH network feeding N homes as depicted in Fig. 3.1(d), an 1 × N power splitter is needed with a uniform coupling factor x1 j = 1/N . Hence the splitting loss from the single input port to each output port j is −10 lg(Pout, j /Pin ) = −10 lg ε + 10 lg N
[dB]
3.3.3 Wavelength Routing Devices A wavelength multiplexer accepts optical signals inserted at its input ports and routes these signals to its output port (see Fig. 3.8(a)). At each input port, it accepts
46
Ton Koonen
only a signal which is positioned in a specific wavelength band. As the routing of each signal is dependent on its wavelength, there are no power splitting losses. Hence the insertion loss from each input port in its specific wavelength band to the output port is typically small (much smaller than the loss of a 1 × N power splitter). The wavelength-dependent routing can be accomplished by means of wavelengthdependent refraction, such as in a prism, or by means of multiple beam interference processes such as in grating structures or multilayer interference filters. A wavelength demultiplexer performs a similar operation in reverse direction. It routes the different wavelength bands of its input signal to different output ports (see Fig. 3.8(b)). For each output port, the insertion losses in its passband are typically small for the relevant wavelength band. Next to small insertion losses, a good demultiplexer also should have good isolation characteristics, i.e., a high crosstalk attenuation from other wavelength bands into this passband. Pin,1 ( 1)
Pout,1 ( 1) 1
1
Pout( 1,
Pin,i ( i) i
. .
Pin,N (
2,
…,
N)
Pout ( 1,
2,
…,
Pout,i ( i)
N)
mux
i
demux
. .
N)
Pout,N ( N
N)
N
(a) Wavelength multiplexer
(b) Wavelength demultiplexer
Fig. 3.8 Wavelength multiplexer and demultiplexer
a 1 ,
a 2 ,
a 3 ,
a 4
b 1 ,
b 2 ,
b 3 ,
b 4
c 1 ,
c 2 ,
c 3 ,
c 4
d 1 ,
d 2 ,
d 3 ,
d 4
1
1
2 3 4
a 1 ,
b 2 ,
c 3 ,
d 4
b 1 ,
c 2 ,
d 3 ,
a 4
c 1 ,
d 2 ,
a 3 ,
b 4
d 1 ,
a 2 ,
b 3 ,
c 4
2
AWG
3 4
Fig. 3.9 Cyclic wavelength routing by a 4 × 4 arrayed waveguide grating router
A wavelength router with multiple input ports and output ports can perform more comprehensive wavelength routing actions. The so-called arrayed waveguide grating router (AWG) performs a cyclic routing of the wavelengths injected at its input ports toward its output ports. Figure 3.9 shows the routing scheme for a 4 × 4 AWG. When using only one of the input ports, the AWG acts as a wavelength demultiplexer. Similarly, when using only one of its output ports, it acts as a wavelength multiplexer.
3 Enabling Techniques for Broadband Access Networks
47
3.4 FTTH Network Topologies The installation of fiber into the access network can be implemented in various topologies. Figure 3.10 shows three basic topologies for bringing fiber all the way to the user’s home [1]. Next to application for FTTH, such topologies are also applicable for fiber-DSL, hybrid fiber coax, and fiber-FWA solutions, as shown in Fig. 3.1, by not terminating the fiber at the home, but at a cabinet containing the appropriate node equipment.
LEX
AN
LEX
(a) Point-to-point
LEX
AN
active node
(b) Point-to-multipoint, active star
AN splitter/combiner
(c) Point-to-multipoint, passive star Fig. 3.10 FTTH topologies
3.4.1 Point-to-Point The most straightforward FTTH topology is the point-to-point (P2P) one, where an individual fiber runs all the way from the local exchange to the home; see Fig. 3.10(a). This enables the operator to tailor the service offerings per individual home easily. When there is a need to upgrade the services for a specific user, the line-terminating equipment for only that user may need to be adapted, not the equipment for every user. As each fiber constitutes an independent transmission path, there is no competition for capacity among the users. If at the local exchange a line-terminating transceiver fails, only one home is out of service, not many of them. Moreover, simple Ethernet point-to-point transceivers can be used, which are becoming very cheap. However, there is a lot of such line-terminating equipment needed, not only at the homes, but also in the local exchange, which requires more effort for housing and powering. Moreover, also a lot of fiber needs to be installed in
48
Ton Koonen
the field. The fibers may share a cable sheet and a duct (note that digging ducts is an expensive operation), but still need to be spliced individually. And if a cable break occurs, e.g., during construction works in a city, the repair involving many splices is a cumbersome and costly operation.
3.4.2 Point-to-Multipoint Savings on the feeder part of the fiber infrastructure can be obtained by deploying a point-to-multipoint (P2MP) topology, thus sharing among the users the major part of the infrastructure, namely the feeder line. As shown in Fig. 3.10(b), the feeder fiber may connect to active equipment placed in a cabinet in the field. In this active node, after opto-electronic conversion, the incoming traffic is sorted out (e.g., by inspecting the address fields of the data packets), and to every user only the traffic parts meant for him are sent. When the last individual link is not fiber, but twisted copper pair cable, this represents the fiber-DSL solution of Fig. 3.1(a). When it is a bus-shaped coaxial cable network, it is like the hybrid fiber-coax solution of Fig. 3.1(b). And when the last link is a wireless one, it is the fiber-FWA solution of Fig. 3.1(c). A common issue in all these solutions is that a street cabinet is needed, which needs remote powering, and is vulnerable to vandalism. Also, the street cabinet and its equipment inside needs to withstand summer heat as well as winter cold, so needs to be able to operate in a wide temperature range. On the other hand, next to simple installation the single-fiber feeder cable is also easy to repair in case of a break. And in the local exchange only one (comprehensive) line-terminating circuit is needed instead of the multiple ones in the point-to-point solution, thus saving on housing space and powering. A fully passive point-to-multipoint topology, thus avoiding the issues with having active equipment in the field, is the so-called passive optical network (PON) topology shown in Fig. 3.10(c). Like Fig. 3.10(b), it offers the advantage of sharing major parts of the infrastructure, namely the feeder fiber and the line-terminating equipment in the local exchange. Splicing sections of the feeder cable is simple, and breaks in the feeder cable are easy to repair. Thus important cost savings can be made on the installation and maintenance. As there is no active equipment in the field, there is no need for powering, for equipment which has to be able to withstand large temperature variations, and for expensive street cabinets. Only an optical power splitting device as discussed in Section 3.3.2 is needed in order to distribute the light signal from the local exchange to the homes.
3.4.3 Cost Aspects When deciding which topology to choose of the ones shown in Figure 3.10, the installation costs of the access network infrastructure are an important criterion. Figure 3.11 shows a qualitative analysis of how these costs for the various topologies
System installation costs
3 Enabling Techniques for Broadband Access Networks
49
P2P, N 2>N1 P2P, N1 P2MP, N2 P2MP, N1
C C C C
0
L0(N1)
L
Fig. 3.11 Installation costs of FTTH topologies (excluding ducting costs; C: fixed costs of line c 2006 IEEE.) terminating equipment, optical splitter; N: number of homes). (After [1].
evolve as a function of the size of the access network, so in terms of increasing the reach L of the network and of the number of users N [1]. The cost elements taken into account are the fiber-optic cable, the optical power splitter, and the electrooptical line-terminating equipment both in the local exchange and in the homes. The costs of digging and constructing the ducts have not been taken into account, as the ducts may largely be identical no matter whether they host a multi-fiber cable or a single-fiber one. In good approximation, the costs C of the line-terminating equipment in the local exchange and in the homes are not dependent on the reach L. Given the same number of homes N , these fixed costs are higher for a P2MP network than for a P2P network, as the optical line terminal (OLT) in the local exchange is more complex. It needs to support a multi-access protocol on the shared feeder fiber, whereas in a P2P network it can be just a simple Ethernet transceiver; hence when serving the same number of homes N the fixed costs C P2M P > C P2P . For the P2P network, the amount of fiber equals the number of homes N times the access network reach L. When the number of homes is increased (so for N2 > N1 ), the amount of line-terminating equipment and the amount of fiber are increased with a factor N2 /N1 . Hence C P2P,N2 > C P2P,N1 , and the slope of the P2P costs-versusL curve for N2 > N1 is larger than the slope for N1 . For the P2MP network, the complexity of the OLT is slightly higher when the number of homes is increased, as it needs to handle a slightly more comprehensive multi-access protocol to share the feeder capacity among these homes; hence C P2M P,N2 > C P2M P,N1 . The feeder fiber forms the major part of the reach L. As the feeder part of the fiber network is shared, but the short drop fibers to the homes are not, the slope of the P2MP costs-versus-L curve is smaller than the one for the P2P case. For the number of homes increasing from N1 to N2 , the P2MP slope
50
Ton Koonen
slightly increases as the amount of fiber in the drop sections increases proportionally to N while the amount of feeder fiber stays constant. The considerations above yield the qualitative trend lines for system installation costs depicted in Fig. 3.11. From these, it can be concluded that for a number of homes N1 there is a certain access network reach L 0 at which a breakeven occurs between the installation costs of a P2P network and those of a P2MP network. Hence for access networks with a short reach (i.e., L < L 0 ), the installation of a P2P network is less costly than of a P2MP network, whereas for larger access networks it is cheaper to install a P2MP network. When many homes need to be connected, so for large N , it may even be cheaper to install a P2MP network for any reach. Next to the installation cost advantages offered by the P2MP topology for a longer network reach and a higher number of homes, it also offers advantages regarding operational costs. In case of a feeder cable break, the repair of the singlefiber cable in the P2MP topology is easier than of the multi-fiber cable in the P2P topology.
3.5 Multiple Access Techniques for a PON As outlined in Section 3.4, the point-to-multipoint passive optical network (PON) topology is a cost-efficient topology for access networks with many homes and/or beyond a short reach. However, in a PON the capacity of the feeder fiber needs to be shared by multiple homes. In the downstream direction, i.e., from the local exchange to the homes, each home receives all the signals sent by the local exchange, and with an appropriate mechanism has to pick from these only the signal which is addressed to that particular home. In the upstream direction, the process is a bit more complicated. As the homes may want to send information randomly, a mechanism is needed which avoids collision of all these information streams when they arrive at the power combiner before they enter the shared feeder fiber. Hence the access network needs to deploy an appropriate multiple access control (MAC) protocol to give each home a fair share of the feeder’s capacity. In the following, four MAC protocols will be discussed: time division multiple access (TDMA), subcarrier multiple access (SCMA), optical code division multiple access (OCDMA), and wavelength division multiple access (WDMA).
3.5.1 Time Division Multiple Access In the downstream direction, the OLT at the local exchange is sending the data packets over the PON and thus broadcasts these to every home. Upon receipt by the optical network unit (ONU) at each home, the address field of the packets is inspected. When matching the address of the home, the packet is accepted and delivered to the user.
3 Enabling Techniques for Broadband Access Networks
51
For communication in the upstream direction (see Fig. 3.12), a timeslot is assigned to each home in which it is allowed to send its packets. By carefully synchronizing these timeslots, collision of the packets at the optical power combiner is avoided. In order to assign the right amount of capacity to each home, the ONU may send first a request packet to the local exchange indicating how much capacity it needs, on which the local exchange may grant one (or multiple) timeslots to the ONU. Thus the upstream capacity per ONU can be adapted to its actual traffic load. An ONU (temporarily) can send more packets than the average amount per ONU, as long as another ONU sends less. This statistical multiplexing gain yields a more efficient deployment of the system’s resources.
data 1 fiber
time demux
Rx data 2
t
data 3
c 2006 IEEE.) Fig. 3.12 Upstream TDMA. (After [1].
The synchronization needed among the upstream timeslots requires that the propagation time from each ONU up to the OLT is precisely known. These times are generally different, as each ONU is at a different length from the OLT. They can be determined by sending ranging grants from the OLT, and measuring the roundtrip time upon receipt of the ranging cell returned by each ONU. By inserting appropriate hold-off times at each ONU in order to equalize the length differences, each ONU is put at virtually the same length from the OLT, and can then easily insert its packets in the right timeslot. In general, this time-synchronization process implies that the upstream data channels from the ONUs are not independent. In addition, per ONU the upstream packets arrive at the OLT with different intensity levels, depending on the different fiber link losses. The receiver in the OLT therefore needs to operate on packets arriving in burst mode, with different intensity levels, requiring a fast clock extraction and decision level setting per packet. Although the processes to be implemented in the OLT and the ONUs for TDMA are not simple, they can be implemented largely in digital electronics and thus realized at (relatively) low costs. Hence TDMA is the most popular MAC protocol for PONs up to now and has been implemented in the standardized schemes for broadband PON (BPON, ITU-T G.983 standard series), EPON (IEEE 802.3ah), and GPON (ITU-T G.984 series). In BPON, the communication has been optimized for fixed-length ATM packets, whereas in EPON the communication has been
52
Ton Koonen
optimized for variable length Ethernet packets. In GPON, both ATM and Ethernet packets can be handled efficiently, at speeds up to 2.5 Gbit/s [2]. More details about these techniques are given in Chapters 7 and 8.5
3.5.2 Subcarrier Multiple Access Another solution to avoid collision of the packets sent upstream by the ONUs is to provide an independent communication channel to each of them. The packets may be modulated on a different carrier frequency per ONU, and these subcarriers can be modulated on optical carriers (with nominally the same optical frequency) emitted by the ONUs (see Fig. 3.13). The advantage is that these subcarrier channels are independent: their spectral bands do not overlap, and hence each channel may transport a service which does not need to be synchronized with the other channels, and can even transport signals with a different signal format. However, no statistical multiplexing like in TDMA by exchanging capacity with other channels can be obtained. Analog RF functions are needed both at the OLT and at the ONUs, which are in general more costly and more difficult to implement than the digital functions needed for TDMA. Moreover, although nominally identical, the optical sources at the ONUs may have slight wavelength differences. After the heterodyning in the photodiode at the OLT’s receiver, this may result in beat notes falling into the electrical spectrum of an SCMA data channel, and thus causing possibly severe beat noise disturbances in the data detection process. To combat this, the spectra of the optical sources at the ONUs should be spaced sufficiently in order to bring these beat noise products outside the spectra of the data channels. However, this brings additional control complexity into the system.
f1
P t
f1 f2 f3
electr. Rx electr. Rx
freq. demux
0
f
f2
P
Rx 0
t
electr. Rx
t f3
P
0
t
c 2006 IEEE.) Fig. 3.13 Upstream SCMA. (After [1]. 5 Chapter 7: Dynamic Bandwidth Allocation for Ethernet Passive Optical Networks, by H. Mouftah and H. Nasser. Chapter 8: Quality of Service in Ethernet Passive Optical Networks (EPONs), by A. Dhaini and C. Assi.
3 Enabling Techniques for Broadband Access Networks
53
The higher costs of the RF electronics and of the additional control complexity have made the SCMA solution less popular than the TDMA one. Subcarrier multiplexing (SCM) is more popular in the downstream direction, in particular for carrying broadcast services such as CATV. As shown in Fig. 3.14, the downstream signals are modulated into different frequency bands and multiplexed, and the composite electrical signal modulates the light output of a laser diode (by either direct intensity modulation or by means of an external intensity modulator). A multi-channel CATV signal is such a composite electrical signal. At the receiver, one of the channels can be selected. In order to avoid interference among the channels due to intermodulation products, the linearity of the light modulation at the transmitter and of the demodulation by the photodiode at the receiver, plus the linearity of the electronics, need to be very high. Moreover, the instantaneous amplitude of the composite signal should not exceed the amplitude modulation range of the laser (or of the external modulator), otherwise clipping and thus distortion of the signal occurs. Hybrid fiber coax (HFC) systems need to deliver multi-channel CATV signals of very high quality, so carefully linearized optical transmitters with careful control of the signal amplitudes of the individual channels have been designed and are available commercially. transmitter ch. 1
amplifier + filter
f x
LO 1 ch. 2
receiver
f x laser
fiber
photodiode
LO 2 ch. 3
f
amplifier + filter x
ch. 3
amplifier x
LO (VCO)
f
selected channel
LO 3 ch. N
f x LO N
Fig. 3.14 Downstream SCM (LO: local oscillator; VCO: voltage controlled oscillator). (After [1]. c 2006 IEEE.)
The architecture of a PON is well suited for distributing broadband services. Using wavelength multiplexing, in addition to the interactive data services transported over the PON by means of TDMA (e.g., with the BPON standard), distribution services can be transported on an other wavelength by means of the SCM
54
Ton Koonen
interactive data distributive services
BPON OLT LEX DS OLT
fiber AN WDM fiber
optical power splitter/combiner
BPON ONU
interactive data
DS ONU
distributive services
WDM
Fig. 3.15 Overlay of distributive services on a TDMA PON, using wavelength multiplexing and SCM
technique; see Fig. 3.15. The isolation requirements put on the WDM mux/demux devices may be quite high, as, e.g., the performance of an analog CATV system is quite sensitive for intrusion of other signals such as the digital BPON signals. The distributive services are preferably put in the 1550–1560 nm wavelength window, where erbium-doped fiber amplifiers have their best performance for boosting the power to the level needed for, e.g., the delivery of analog CATV signals with the required high carrier-to-noise ratio. The BPON downstream traffic is allocated to the 1480–1500 nm window, and the upstream traffic to the 1260–1360 nm window (according to the G.983.3 standard).
3.5.3 Optical Code Division Multiple Access A second solution to create independent upstream channels on the PON is by using code division multiplexing. This technique is also used in wireless communications to create independent channels in the shared free-space medium. Each channel uses a specific code with which its data are scrambled at the transmitting end. At the receiving end, by correlation with the same code these data are retrieved amidst the data from the other transmitters. The code signature can be a specific sequence of narrow pulses (time-sliced code) or a specific set of slices from the wavelength spectrum of an optical source (spectrum-sliced code). The independence of the channels brings the advantage that no synchronization among these channels is needed; on the other hand, no statistical multiplexing gain can be achieved. Moreover, unless the codes become very complex with many code elements, only a limited number of code words with sufficient orthogonality can be obtained, which limits the number of homes. Non-perfect orthogonality results in crosstalk between the channels. By using two-dimensional coding, i.e., both time splicing and spectrum slicing, more orthogonal code words can be realized (Fig. 3.16). In time-sliced OCDMA, each data bit is scrambled with the time code sequence, resulting in a symbol line rate which is a multiple of the data rate. Hence the speed requirements on the line
3 Enabling Techniques for Broadband Access Networks
RX
opt. corr.
RX
opt. corr.
RX
opt. corr.
code
55
t
0
c1 c2 c3 t
0
t 0
t
Fig. 3.16 Upstream OCDMA
terminal equipment are higher than in the case of an equivalent TDMA system. In spectrum-sliced OCDMA, the spectral range deployed by the data stream is larger than when using TDMA. Hence fiber dispersion may play a more important role in restricting the achievable data rate. Chapter 126 deals with OCDMA in more detail.
3.5.4 Wavelength Division Multiple Access A third solution for realizing independent upstream channels is by using wavelength division multiplexing. Each channel uses a specific wavelength, and in the splitter point of the PON a wavelength multiplexer/demultiplexer is used to combine/separate these wavelength channels into/from the feeder fiber; see Fig. 3.17. Again, by the independence between the data channels no synchronization between them is needed, and they can transport signals with very different formats. But no statistical multiplexing gain can be achieved. As each wavelength channel constitutes an independent path from the local exchange to a home, the WDM PON actually offers point-to-point connection functionalities on a shared point-to-multipoint physical infrastructure. Moreover, as the WDM multiplexer performs wavelength routing, the losses in the network splitter point are much less than in the power splitting PON. For example, in a 1:32 power splitter the losses are at least 15 dB, whereas in a 32 channel WDM device the insertion loss may be less than 3 dB; see Section 3.3. Hence the link power budget is considerably better, and thus the reach of a WDM PON can be larger. The WDM PON architecture is less suited for providing distributive services; this requires additional means to bypass the wavelength routing functions in the network splitter point. As each channel is operating on a specific wavelength, the line-terminating equipment in the OLT at the local exchange and in the ONU at the home needs 6
Chapter 12: Signal processing techniques for data confidentiality in OCDMA access networks, by P. Prucnal et al.
56
Ton Koonen
1
1
Rx WDM WDM
Rx demux demux Rx
2
3
WDM WDM mux
2
t 3
c 2006 IEEE.) Fig. 3.17 Upstream WDMA. (After [1].
a wavelength-specific transmitter. This implies that the network operator needs to keep a costly extensive stock of spare wavelength-specific modules. A more convenient solution is to deploy universal colorless transmitter modules. Such modules can be realized in various ways. A reflective modulator or reflective semiconductor optical amplifier (RSOA) may be used in the transmitter module at the ONU, which modulates a CW wavelength channel generated at the OLT with the upstream data. These solutions require that the reflections inside the optical fiber network are very small, so high-quality low-reflection fiber splices and connectors are needed [3]. Upstream data speeds of 1.25 Gbit/s and beyond have been obtained [4]. Or the transmitter module may deploy a Fabry–Perot laser diode on which the data are modulated, and of which the spectrum is injection-locked to a CW wavelength channel from the OLT. Such an injection-locking scheme usually limits the achievable data rate (typically to less than 622 Mbit/s [5]). Or an optical source with a very broad spectrum, such as an LED, may be used, where the WDM multiplexer cuts a specific slice out of its spectrum. This implies that only a small part of the optical output power of the source can be utilized, thus reducing the system’s power budget. In addition, a narrow slice yields a larger impact of instabilities in the source’s power spectral distribution, so generates more excess noise. Enlarging the slice width yields higher power and reduces this excess noise, but also increases the impact of fiber chromatic dispersion. Depending on the fiber length and the required bit rate, an optimum slice width can be found [6]. Most PONs nowadays deploy TDMA schemes, as these feature statistical multiplexing gain and can be realized with relatively cheap digital-integrated circuitry. WDM PON is gaining popularity, as it offers a better power budget and due to its virtual P2P connectivity a higher capacity per home. When scaling a PON in order to connect more homes, the WDM PON scheme may thus be more attractive. By combining the TDM and WDM schemes into a hybrid WDM-TDM PON, an efficient
3 Enabling Techniques for Broadband Access Networks
57
large-scale FTTH network can be obtained: by deploying multiple wavelength channels a large capacity is realized, and by using TDM within each wavelength channel statistical multiplexing gain can be achieved. Such hybrid WDM-TDM schemes are being investigated for long-reach PON systems featuring a reach of more than 100 km, and more than 1000 homes connected; thus the number of local exchanges with their maintenance costs may be reduced. Chapter 107 deals with long-reach PONs in more detail. When deploying dynamic routing of the wavelength channels, the sharing factor of the TDM capacity per wavelength channel may be adapted (by e.g., routing a wavelength channel to less homes) and thus the capacity per home may be tuned to its actual demand. Such an advanced scheme may further improve the efficiency with which the network equipment is deployed. WDM PON techniques are treated in more detail in Chapter 9.8
3.6 Radio Over Fiber Wireless broadband connectivity is becoming increasingly important, e.g., for the mobile user who wants high-speed access to the Internet by means of his portable laptop computer. But next to supporting mobility, another major benefit is that no fixed infrastructure needs to be installed. Inside many homes, Internet connectivity is implemented by broadband wireless LAN routers (although issues such as security and interference demand extra care). As mentioned in Section 3.2, also in the access network, and in particular in less densely populated areas, it is interesting to apply wireless techniques in the last link to the home. Fixed wireless access techniques can offer high bandwidths at elevated frequencies (e.g., the IEEE 802.16 WiMAX standard offers more than 100 Mbit/s with carrier frequencies in the 10–66 GHz range). However, the complexity of generating such carriers and of the data modulation schemes (OFDM, x-level QAM, etc.) becomes high, so it becomes ever more expensive to install such equipment in the field next to each antenna. Moreover, as the reach of an antenna shrinks when the carrier frequency increases, ever more antennas are needed to cover a certain area, certainly when also radio emission powers should be limited due to the (yet largely unknown) potential health hazards; see Fig. 3.18(a) and (b). Hence the concept of radio over fiber is gaining interest: the generation of the high-frequency carrier signals as well as the data modulation onto those carriers is all done at a central site, from which optical fibers bring these signals to the antenna sites; see Fig. 3.18(c). At these sites only opto-electrical conversion takes place and the resulting modulated microwave signal is radiated by the antenna, so the antenna site is considerably simplified entailing a sizable reduction in installation and maintenance costs. When upgrading to, e.g., another wireless standard, basically only things at the central site need to be changed, instead of adapting all the antenna sites in the field. 7 8
Chapter 10: Long-Reach Optical Access, by B. Mukherjee et al. Chapter 9: Multi-channel EPONs, by M. Reisslein and M. McGarry.
58
Ton Koonen
BS
CS
copper cable
BS
BS
BS
BS
CS
BS
BS
BS
copper cable
BS
BS
BS
BS
BS
(a) low-capacity wireless network
(b) high-capacity wireless network
fiber
CS BS
(c) high-capacity radio-over-fiber wireless network Fig. 3.18 Wireless communication network, evolving to higher capacity (BS: base station, containing carrier generation and modulation circuitry; CS: central station, feeding the antenna sites and connecting to the next-higher network)
The transport of the modulated microwave signals over the fiber can be done in several ways. First, the microwave signal may be intensity-modulated on the output of a laser diode, either by direct modulation of the laser or by means of an external modulator [7]. A high linearity and good high-frequency performance of the laser (or the external modulator) are required. Basically, this generates two sidebands containing the modulation next to the optical carrier wave. As these sidebands may experience different phase changes due to the dispersion in the fiber, periodically along the fiber the phases of these sidebands may be opposite, thus canceling the intensity modulation. Hence fading occurs of this double-sideband signal, and it occurs at shorter intervals when higher microwave carrier frequencies are used. This fading issue can be avoided by single-sideband modulation schemes, but this adds complexity to the transmitter and receiver equipments. A second approach is to deploy heterodyning between two optical sources. For example, as exemplified in Fig. 3.19(a), when combining the CW optical carriers of two laser diodes which are operating at a slightly different wavelength, the heterodyning process taking place during detection in the photodiode yields a harmonic electrical signal with a frequency equal to the difference between the optical frequencies of the laser diodes. For example, an optical signal with a wavelength λ = 1.5 μm has an optical frequency ν = c/λ = 200 THz where c = 3 × 108 m/s
3 Enabling Techniques for Broadband Access Networks
DFB laser 1 = c0 / 1
MUX
CW
DFB laser
fmw
single-mode fiber
59
MZM
PD fmm
V fmw /2
1·(1
PD
1
data
data 2=
fmw
single-mode fiber
+ fmw/ 1)
(a) heterodyning two narrow-line width laser diodes
Two-tone optical signal 1
(b) self-heterodyning, using double-side band suppressed-carrier modulation
Fig. 3.19 Radio over fiber using optical heterodyning
is the speed of light in vacuum, so to generate a microwave of 60 GHz a wavelength difference of only λ = λ·ν/ν = 0.45 nm is needed. By intensity-modulating the light output of one of the laser diodes, after heterodyning the modulated microwave emerges. Again, the laser needs to be sufficiently linear. The spectral linewidth of the microwave signal is equal to the sum of the linewidths of the laser diodes, so very narrow-linewidth laser diodes are required in order to obtain an acceptably pure microwave signal. Such narrow linewidth may, e.g., be obtained by injection locking the two laser diodes to an external source, but that again adds to the complexity. This external locking can also help in keeping the laser diodes at the prescribed wavelength difference. Alternatively, as shown in Fig. 3.19(b), one may use just one laser diode and generate by external modulation two sidebands while the carrier is suppressed. Such modulation can be achieved by biasing a Mach Zehnder modulator at the inflexion point of its transfer characteristic, where its transmission is minimum, and applying a modulating harmonic signal around this bias point [8]. The spacing of the sidebands then equals twice the frequency of the harmonic signal, and after heterodyning a microwave signal containing this doubled frequency (plus higher even harmonics) is obtained. A third approach is to generate many harmonics of a relatively low-frequency signal by applying a non-linear process, and at the antenna site by means of a simple bandpass filter select the harmonic with the desired frequency [9]. For instance, as shown in Fig. 3.20, when the CW light of a laser diode is phase modulated in an external modulator driven by a harmonic sweep signal, this phase modulation yields many harmonics of this sweep frequency in the optical output signal. The output signal may be fed through an optical phase-to-intensity modulation converting device, such as a Mach Zehnder interferometer or a Fabry–Perot interferometer. After detection in a photodiode, the output current contains many harmonics of the sweep frequency, from which the desired microwave frequency can be selected. Further analysis of this so-called optical frequency multiplying method shows that an extremely pure microwave carrier can thus be obtained, and also that the signal is robust against fiber dispersion (see [9] and [10]). Hence it enables to carry comprehensive data modulation formats carrying high data rates over dispersive fiber links; e.g., the transport of 120 Mbit/s data using 64-QAM over 4.4 km silica 50 μm core graded-index multimode fiber has successfully been demonstrated (see, e.g., [11]).
60
Ton Koonen
fsw
+ data
= 6.4 GHz
periodic filter
Antenna Station fmm = 2N · fsw
fiber link P D
CW LD
–
BPF
i(t)
0
- data
RF power [dBm]
Central Station
–30 –60 –90 –500 0 +500
Freq. offset from 38.4 GHz carrier [Hz]
Fig. 3.20 Optical frequency multiplying method
Radio-over-fiber technologies are dealt with in more detail in Chapter 13.9
3.7 Free-Space Optical Communication Wireless communication links may, instead of by means of radio waves, also be implemented by using free-space optical beams. Obviously, such links need good line-of-sight conditions. As illustrated in Fig. 3.21, free-space optical (FSO) communication links are popular for quickly setting up links between high-rise buildings [12]. FSO links are point-to-point links and are license-free. They are rapidly deployable, which is important for, e.g., recovering communication links in disaster situations. A directional narrow beam of light is used, which offers high security as eavesdropping and jamming are nearly impossible. On the other hand, careful
optical light beam free-space optical transmitter/receiver
Fig. 3.21 Free-space optical communication 9
Chapter 13: Radio-over-Fiber (RoF) Networks, by J. Mitchell.
3 Enabling Techniques for Broadband Access Networks
61
alignment of the transmitted beam direction with the aperture of the receiver is needed, which requires a tracking and pointing system. The coarse tracking may be done by a GPS system, whereas the fine pointing may be done by beam steering using, e.g., accurately movable mirrors, acousto-optic crystals, or optical-phased array structures. Continuous adjustment of the alignment may be needed, e.g., due to the wind causing building sways and due to thermal expansion of the building. Commercially available FSO systems offer capacities of 2.5 Gbit/s and more. In the transmitter, a laser diode or superluminescent LED with low beam divergence may be used. The beam should have a diameter large enough not to be obstructed by, e.g., a single bird. Multiple transmitters may be used to counteract the building sways and to provide resilience against birds. At the receiver, a telescope system collecting the light beam is deployed, focusing the light on a single photodetector or on a detector array. Due to eye safety regulations set in IEC 825, the maximum optical power for Class 1M equipment can be 2.5 mW at a wavelength λ = 880 nm, and 150 mW at λ = 1550 nm, increasing for Class 3R up to 500 mW for both λ = 880 nm and λ = 1550 nm. The reach of an FSO link is typically a few hundred meters, and can be severely impacted by the atmospheric conditions. The reach is mainly limited by the propagation losses, which can be up to 10–100 dB/km. Contributing to the propagation losses are scintillation effects, which consist of air refractive index changes due to temperature differences between the ground and the air, and which may cause defocusing of the beam. Other contributions may come from aerosol scattering, by rain and snow, but most importantly by fog and haze as these involve particle sizes close to the wavelength (which thus heavily generate Mie scattering). In the infrared region, most losses are caused by water vapor and carbon dioxide. For wavelengths below 200 nm, the losses are too high due to oxygen and ozone, and beyond 22 μm due to water vapor. Chapter 1510 deals with FSO in more detail.
3.8 Summary The fast growing demand of residential users for capacity requires powerful and cost-effective broadband access network solutions. Optical fiber with its extremely low losses and tremendous bandwidth is the most powerful medium for a futureproof versatile infrastructure. Next to the ultimate solution offered by bringing fiber all the way to the home, fiber can very effectively team up with legacy copper twisted-pair cable and coaxial cable last link networks. Moreover, in an alliance with broadband wireless access techniques for the last link, fiber-wireless networks offer user mobility on top of broadband connectivity. The benefits of optical communication for access are even not restricted to fiber only; also broadband access by means of wireless optical techniques is a promising new domain. 10
Chapter 15: Hybrid Wireless-Optical Broadband Access Network (WOBAN), by S. Dixit et al.
62
Ton Koonen
References 1. Koonen, T. (2006). Fiber to the home/fiber to the premises: What, where, and when? Proc. IEEE. 94(5), 911–934. 2. Angelopoulos, J.D., Leligou, H.-C., Argyriou, Th., Ringoot, E. & Van Caenegem, T. (2004). Efficient transport of packets with QoS in an FSAN-aligned GPON. IEEE Commun. Mag. 42(2), 92–98. 3. Urban, P.J., Koonen, A.M.J., Khoe, G.D. & De Waardt, H. (2008). Mitigation of reflectioninduced crosstalk in a WDM access network. Proceedings of OFC. San Diego, CA. 4. Payoux, F., Chanclou, P., Moignard, M. & Brenot, R. (2005). Gigabit optical access using WDM PON based on spectrum slicing and reflective SOA. Proceedings of ECOC. Glasgow, Scotland. 5. Jung, D.K., Shin, D.J., Shin, H.S., Park, S.B., Hwang, S., Oh, Y.J. & Shim, C.S. (2005). GWavelength-division-multiplexed passive optical network for FTTx. Proceedings of OECC. Seoul, South Korea. 6. Pendock, G.J. & Sampson, D.D. (1996). Transmission performance of high bit rate spectrumsliced WDM systems. IEEE J. Lightwave Technol. 14(10), 2141–2148. 7. Gomes, N.J., Nkansah, A. & Wake D. (2008). Radio over MMF techniques – Part I: RF to Microwave Frequency Systems. IEEE J. Lightwave Technol. 26(15), 2388–2395. 8. Griffin, R.A., Lane, P.M. & O’Reilly, J.J. (1999). Radio-over-fibre distribution using an optical millimeter-wave/DWDM overlay. Proceedings of OFC. San Diego, CA. 9. Koonen, T. & Ng’Oma, A. (2006). Integrated broadband optical fibre/wireless LAN access networks, in broadband optical access networks and fiber-to-the-home: Systems technologies and deployment strategies. John Wiley and Sons, New York, Ch. 11. 10. Koonen, A.M.J. & Garc´ıa Larrod´e, M. (2008). Radio over MMF techniques – Part II: Microwave to Millimeter-Wave Systems. IEEE J. Lightwave Technol. 26(15), 2396–2408. 11. Garc´ıa Larrod´e, M. & Koonen, A.M.J. (2008). Theoretical and experimental demonstration of OFM robustness against modal dispersion impairments in radio over multimode fiber links. IEEE J. Lightwave Technol. 26(12), 1722–1728. 12. Kedar, D. & Shlomi, A. (2004). Urban optical wireless communication networks: The main challenges and possible solutions. IEEE Commun. Mag. 42(5), S2–S7.
Part II
Copper and Wireless Access Networks
Chapter 4
Vectored DSLs and the Copper PON (CuPON) John M. Cioffi, Sumanth Jagannathan, Mehdi Mohseni, and George Ginis
Abstract This chapter investigates the limits of copper twisted-pair telephone lines in digital subscriber line (DSL) transmission, specifically focusing on the most advanced form known as the copper passive-optical network or CuPON. Speeds of several hundred Mbps to each customer on 1 km (3000’) lengths of copper telephone lines are feasible using a multi-line sharing architecture of the existing telephone cables. Such speeds thus extend the life and value of the large invested base of copper telephone lines perhaps to speeds beyond those presently envisioned and targeted for shared-fiber access systems like passive optical networks (PONs).
John M. Cioffi Department of Electrical Engineering, Stanford University, David Packard Electrical Engineering Building, 350 Serra Mall, MC 9515, Stanford, CA 94305-9515, USA e-mail:
[email protected] Sumanth Jagannathan Department of Electrical Engineering, Stanford University, David Packard Electrical Engineering Building, 350 Serra Mall, MC 9515, Stanford, CA 94305-9515, USA e-mail:
[email protected] Mehdi Mohseni ASSIA Inc., 303 Twin Dolphin Drive, Suite 203 Redwood City, CA 94065, USA George Ginis ASSIA Inc., 303 Twin Dolphin Drive, Suite 203 Redwood City, CA 94065, USA This chapter is based on “CuPON: the Copper alternative to PON 100 Gb/s DSL networks”, by J. M. Cioffi, S. Jagannathan, M. Mohseni, G. Ginis, which appeared in IEEE Communications Magazine, June 2007.
A. Shami et al. (eds.), Broadband Access Networks, Optical Networks, c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-92131-0 4,
65
66
John M. Cioffi, Sumanth Jagannathan, Mehdi Mohseni, and George Ginis
4.1 Introduction DSL transmission has consistently extended the lifetime of copper telephony beyond even the most optimistic projections of all service providers. Indeed, today, DSL brings more than 100 billion dollars in fixed-line service revenue to telephonecompany service providers worldwide, single-handedly averting the extinction of wireline service. Even the most pessimistic of market visionaries see that number growing. Yet, as this chapter explores, the limits of the copper pair are not near attained when the cable of twisted pairs is viewed from a multiple-user perspective. This chapter describes those limits and a CuPON architecture that can be used to obtain the highest presently known speeds of copper telephone lines. A key element in best use of cables of telephone wires is vectoring. Vectored DSLs synchronize the transmissions of several lines in a coordinated fashion, i.e., transmitting and/or receiving vectors of line signals. The coordination dramatically increases bandwidth if performed cogently. Vectored DSL [1–10] networks profit from reuse of this crosstalk energy in various ways, lending to a large increase in available data rates to all customers. Dynamic Spectrum Management (DSM) level 3 is another name for vectored DSLs. Standards efforts have defined such vectoring in the American DSM Report [11] and in the International Telecommunication Union G.vector standard [12]. A brief tutorial on vectoring appears in Section 4.2 of this chapter, while references [1–12] contain more information. There are various forms and architectures for vectoring that address both bundled and unbundled regulatory situations. One advanced extension of vectoring of interest is the “Copper PON” or “CuPON” for short. CuPON use of existing copper-line networks can lead to data rates that may exceed those envisioned for even the most aggressive future roadmaps for passive fiber-access products. Section 4.2’s tutorial overview of salient vectoring fundamentals shows that vectoring of enough lines can render all such lines, fundamentally and theoretically, as virtually free of man-made noises such as crosstalk, radio ingress, and impulse disturbances. Nearly noise-free copper transmission can then produce enormous DSL data rates, limited in practice only by analog receiver circuits and analog/digital conversion devices, not crosstalk nor the copper. Even more intriguing, a binder of say 200 lines actually has 400 wires and thus 400 transmission paths. Section 4.2’s full vectoring finds and exploits the energy of the longdormant other 200 paths. Section 4.3 reviews the copper-line architecture called CuPON that allows some sharing of pairs, exploiting the customers’ existing 2–4 line (and thus 4–8 wire) “drop” segments. The CuPON architecture allows a DSL 0.5–1 Gbps data rate per customer and roughly 100 Gbps of readily realizable total bandwidth for a typical 200-pair telephone-company “distribution area,” more bandwidth than telcos can dream of using today. Section 4.3 also investigates aggregate, continuous, and peak data rates, while considering range, powering, and deployment issues and costs. Section 4.4 summarizes the chapter.
4 Vectored DSLs and the Copper PON (CuPON)
67
4.2 Vectored-DSM Dynamic spectrum management1 (DSM) provides an evolutionary path toward a goal of ubiquitous single-line 500 Mbps/customer DSL service [1–3]. DSM follows a series of steps from present use in DSL to the ultimate very highest DSL speeds attainable with future vectored-DSM. These steps are reviewed in [1–4, 11] and are not repeated here. Vectored-DSM DSLs coordinate simultaneous binder transmissions to reduce and to exploit crosstalk, while also removing most other noises. Such vectored DSLs promise to best use single-line standardized VDSL systems to achieve 500 Mbps per-line data rates on lines of up to 400 m in length. A neighborhood DSL binder of 200 lines could then carry 100 Gbps of potentially shared bandwidth. Level 3 vectored systems imply line-terminal coordination of the transmitted downstream and received upstream signals within those DSLs controlled by any single service provider (each service provider can independently vector their own lines). Such vectored systems have the largest gains when all the binder’s lines are commonly vectored, but also offer significant improvements in either bundled or unbundled environments where each service provider independently vectors only their own lines. With vectoring, a common DSLAM (more than likely a common multiline card in the DSLAM or LT) synchronizes downstream DSL transmissions to a common discrete multi-tone (DMT) symbol clock. VDSL2’s digitally duplexed standardized loop-timing will then automatically provide the same common symbol clock in the upstream direction. These vectored signals can correspond to different users in different customer locations but are launched in vectored cooperation into the binder. Section 4.1 provides a brief tutorial on vectoring. Section 4.2 describes differential vectoring. Section 4.3 explores the fascinating concept of the full capacity of a binder2 of lines (which is hundreds of times the data rates of passive optical networks in use today).
4.2.1 A Vectoring Tutorial The term “vector” is used because the DSL’s individual physical-layer voltages can be viewed as a coordinated set or “vector” of voltages. This vector of coordinated signals for several lines replaces the single-user’s scalar voltage in analysis and digital signal processing. The group or vector is processed by a common signal processing device for downstream transmission and also for upstream reception. It is this 1
In an affront to evolving standards’ technology, some vendors have tried to promote “digital line management (DLM)” as an alternative term. It is equivalent to level 1 DSM. DLM can also be confused because in line maintenance this term already means “dedicated line matrix.” 2 Telephone cables are groups of telephone lines inside a sheath that travel together. A binder is a subgroup within the cable that typically has 25 telephone lines, usually surrounded by paper from other binders. Thus a “cable” is a group of binders, which in turn is a group of telephone lines. Binder is the convenient group for vectoring discussions in this chapter.
68
John M. Cioffi, Sumanth Jagannathan, Mehdi Mohseni, and George Ginis
Fig. 4.1 Simple vectoring illustration
group processing that allows cancellation or removal of crosstalk. Vectoring can also reduce non-crosstalk noise. Figure 4.1 simplifies a vectored receiver to two telephone lines, and thus a two-dimensional vector, one used by DSL and the other used or possibly carrying no DSL. Each xTU contains both a DSL transmitter and a DSL receiver, but the XTU-C’s are colocated and can coordinate their processing physically. A radio or impulse noise from a common source impinges both lines. The upstream vectored receiver in the central office could, if customer 2 were not active, sense the common noise and effectively subtract3 it from line 1. Even with customer 2’s DSL signals and data rate degraded by the radio noise, line 2’s data can be detected, and its non-radio-noise signal reconstructed. The radio signal is then the difference between the (delayed) line output and the reconstructed signal. Then, line 1’s radio noise can be cancelled by filtering line 2’s noise estimate and subtracting this filtered noise from line 1’s output, which is sometimes called “spatial noise cancellation.” Such simple vectoring illustrates a user “order” or priority – which has their noise cancelled, and which has its rate reduced by the noise (that is by “going first”)? The concept generalizes to more than two lines where the priority ordering becomes even more interesting, possibly with multiple noise sources. Figure 4.1 also shows the NEXT and FEXT signals. Since the joint vector transmitter knows both transmit signals, this transmitter can contain a signal-processing device that synthesizes the upstream NEXT in all vectored lines and removes them. FEXT is removed in a fashion similar to the radio and impulse interference by reconstruction and removal of FEXT in another line after detecting a first line. Order may also be important for FEXT cancellation, but order is often of smaller consequence for FEXT than for radio signals, and it is possible in practice to simplify the electronics to a “matrix FEXT inversion” filter. In any case, a well-known result from [5] is that all upstream NEXT and FEXT can be eliminated within a vectored group. Indeed NEXT and FEXT from other non-vectored lines (or other service provider’s vectored lines) may also be treated as alien noise and largely removed as long as the vectored group has more lines than the number of sources of noise 3
Sophisticated signal processing experts will note this requires an adaptive filter, but one that is easily implemented independently on each tone of a DSL system.
4 Vectored DSLs and the Copper PON (CuPON)
69
(at any frequency where such cancellation is useful). The processing of downstream signals is analogous, but the FEXT removal is a pre-subtraction (or pre-“FEXTmatrix inversion”). Unfortunately, the ability to remove NEXT and RF signals may be more limited downstream because of the absence of a 2nd line’s receiver, so even vectored DSL systems best employ frequency separation of upstream and downstream transmissions so that there is no NEXT to cancel. Section 4.3 reviews full vectoring [9], which is a partial solution for downstream NEXT and RF/impulse cancellation. Section 4.3’s CuPON uses full vectoring to eliminate all noises, even downstream. The complexity of vectored DSM is largely at the component level, where the digital-signal-processing engine (typically today about one-third the cost of a DSL component) for a vectored DSL is about three times more complex in a welldesigned vectored design [13]. Thus, the complexity increase at the component level is about 70–100% increase. Since DSL chip costs today are a few dollars in 90 nm technology, the vectoring cost increase in 65 nm or smaller technology is minimal in terms of total DSL equipment cost.
4.2.2 Differential Vectoring Ginis’ differential vectoring [5] uses only differential excitation of each line (thus, phantoms and split-pair circuits [11] are ignored). Differential excitation is the “normal” placement of a voltage across the two wires (“tip” and “ring”) of a telephone line. Differential vectoring is also the earliest form of vectoring to be tested and eliminates all downstream and upstream FEXT [5] within the vector group. Upstream vectoring’s ability to cancel out-of-vector-group noises is calibrated by the spatial correlation of the noise. When the correlation is high, then the out-ofgroup noise is largely canceled in upstream vectored reception. High correlation means there is a single or small number of noise sources. Low correlation means there are a large number of sources each contributing a small part of the noise. AM radio noise and impulse noise have very high spatial correlation so they can be almost entirely suppressed in the upstream by vectoring. Unbundled (other service providers’) crosstalk noise from many unbundled sources can only be partially reduced in vectoring. An intriguing derived effect for unbundled competition is that a service provider with more lines can cancel the crosstalk of another service provider with fewer lines, but not vice versa – the possibilities for motivated competition are infinite, while the choice to implement vectoring should not ever be regulated because it is internal to the equipment and otherwise satisfies any and all emission standards. For single-line downstream transmission, all noise appears spatially uncorrelated because of the absence of a common vectored receiver (whether or not the noise actually is correlated, since no cancellation can occur at the receiver4 ). 4 An exception is perfect space–time correlation where for instance a known sinusoidal signal could be subtracted, but then a sinusoid is really not noise.
70
John M. Cioffi, Sumanth Jagannathan, Mehdi Mohseni, and George Ginis
For multi-line downstream transmission, sometimes called “bonding,” noises at the customer-end may be cancelled, unlike the single-line case. Differential vectoring can make a large difference in DSL deployment situations of interest. Figure 4.2 illustrates the performance of differentially vectored VDSL for a number of situations proposed recently by North American DSL service providers in [14]. The range of interest is short DSL connections of 1 km (or 3000 ft., which is typically at the location of the “cross box” in most telephone company networks), presuming a fiber to the “curb” system (where curb essentially means within the 3000 ft. distance from the residence) where the cost of the fiber deployment can be shared over hundreds of customers. The remaining connection to the customer is a single copper line in this situation; however, a vectored lineterminal DSLAM is placed at the drop point that uses differential vectoring. Zidane et al.’s [14] suggested upstream and downstream rate goals are also illustrated (and vary with length from 150 Mbps symmetric to 50-down/30-up at longer lengths) and noticeably correspond to the lower curves. The data rates are very high again for differential vectoring and easily exceed the goals in all cases (whereas a nonvectored VDSL system will not meet the goals in practice). To examine the effect of the spatial correlation, Fig. 4.2 provides upstream data-rate curves for the two extreme values of perfect correlation (rho = 0.99), intermediate correlation (rho = 0.5), and no correlation (rho = 0) for the −125 dBm/Hz noise model suggested by the service providers. The plot shows for instance that 100 Mbps, let us say the speed of “fast Ethernet,” can be delivered at 1500 ft. range (500 m) on a single wire. Clearly this is also a substantial improvement on the old four-pair Fast Ethernet systems that deliver 100
Fig. 4.2 Fiber-to-the-curb single-line high-speed VDSL data rates with differential vectoring
4 Vectored DSLs and the Copper PON (CuPON)
71
Mbps to 100 m.5 While already exceeding the data rates offered by PON services today, this is not the limit on the speed of copper – indeed not even close to the limit yet. Full vectoring can help.
4.2.3 Full-Vectored Binder Capacity Full binder capacity exploits single-wire transfers and common-mode crosstalk within the binder of lines. This subsection reviews both effects, starting with the single-wire transfers and finishing with common-mode crosstalk. The key concept in full vectoring is to use all the wires. Full vectored binder capacity investigates a binder of U DSL lines that contains 2U wires. There are possibly as many as 2U transfer modes of energy in such a binder of wires, of which U are left dormant in differential vectoring and all previous DSLs. When all these modes are considered, the binder’s full-vectored capacity can be computed. The possible exploitation of such modes is clear in MIMO systems (such as in [6, 7] where symmetric Gigabit DSLs are observed on four bonded 300-m-plus telephone lines by using all eight wires). The extra capacity is reduced for single-line systems, but is still substantial.
4.2.3.1 Single-Wire Transfers Figure 4.3 illustrates single-wire transfers between two bonded lines or a quad of four wires in the downstream direction with three loads. The concept easily generalizes to more than two lines.
Fig. 4.3 Split-pair excitation of two lines. There are 3 = 2U-1 sources since the ground/shield is not used
5 The recent IEEE 802.3 10G-BT standard [8] specifies 10 Gbps on these same 4 lines, and indeed presumes use of differential vectoring.
72
John M. Cioffi, Sumanth Jagannathan, Mehdi Mohseni, and George Ginis
With single-wire (sometimes also called “split pair”) excitation, one wire (0) in one of the pairs is viewed as a reference for three sources: V01 , V02 , and V03 . In normal differential operation, the second subscriber’s excitation voltage is V2 = V02 −V03 and only two excitations appear. However, three sources are possible more generally and in fact, can all be assigned to a single subscriber with bonding of the pairs. When the two lines go to different customers, the load resistance placements between wire 0 and wires 2 and 3 is not physically possible. If the pairs go to the same customer, then such termination is possible. Figure 4.4 shows the data rates achieved for the situations of two-pair drops and four-pair drops. The upper curve in each case corresponds to the use of the extra single-wire modes. The increase is substantial for full vectoring over differential vectoring. In each case though, both vectored curves are beyond the most aggressive data rates per customer of any current fiber system.
4.2.3.2 Common Crosstalk Modes There is yet more useful energy in the binder of lines. Figure 4.5 alternately illustrates the possibility of common-mode transfers within a binder where the sheath has been well connected to ground at all junctions. Such grounding is the purported practice of all telephone companies. However, the grounds may sometimes not be connected in practice, so such operation is possible only if they are connected. In drop segments with no sheath, one wire in the drop can be used as the ground and the results of Figs. 4.3 and 4.4 then apply. It is rare to find only one pair in a drop segment, so the full-vectoring common modes can be exploited. The various transfer functions with sheaths can be determined by multi-conductor transmission line and elementary electromagnetic theories [9], and indeed these transfers are significantly larger than those of the differential pairs. The wise old sages of telephone plant engineering will immediately object that the common-mode transfers create large crosstalk noise (and sense large noise also) – that’s why Bell originally invented the twisting. They are right – without vectoring those high-transfer modes annihilate each other in terms of very high noise. But not with vectoring – recall vectoring cancels that crosstalk effect. The impact on data rate is enormous. Figure 4.6 (see [10]) illustrates with the two upper curves for a single line, the effect of using the common-mode transfers. The rates are compared to differential vectoring alone. This is the limit6 (these curves are not quite capacity and use a gap [3] of about 11 dB) of practical transmission on the line. It is thus possible to attain almost 250 Mbps symmetric (or up to 500 Mbps asymmetric) on a single pair at distances of 500 m (or 1500 ft.), and 100 Mbps to distances of 1 km (or 3000 ft.).
6
It is dangerous to say “this is the limit” as someone will always come along later and find some new effect. However, as we know today, the only possibilities for further data rate increase are “statistical scheduling” of the pairs (which possibly increases the peak rate but not the continuous rate, so we do not count that here) or by using higher order transverse modes of the transmission binder, an area we leave to some enterprising electromagnetics theorist to explore.
4 Vectored DSLs and the Copper PON (CuPON)
(a) Two pair (three modes)
(b) Four pair (seven modes) Fig. 4.4 Symmetric upstream and downstream data rates with and without single-wire use
73
74
John M. Cioffi, Sumanth Jagannathan, Mehdi Mohseni, and George Ginis
Fig. 4.5 Binder with grounded sheath and full common- and differential-mode excitations
Fig. 4.6 Full binder versus differential vectoring data rates
These are not what service providers see today from VDSL2 equipment, but are possible with improved vectored systems in development.
4 Vectored DSLs and the Copper PON (CuPON)
75
4.3 A CuPON for savings on access-network purchases While very high single-pair limits may have been found in Fig. 4.6, the multi-pair drop has still not yet been fully exploited. To exploit this drop, the PON architecture offers a model for copper systems to exploit, as in this section. Figure 4.7 illustrates the classic PON architecture.
1-2.5Gbps OLT
Variable length packets
splitter
splitter
ONU
ONU DSL Wifi fiber
DSL Wifi fiber
Fig. 4.7 Basic PON architecture
A single fiber (or coax) emanates from an optical line terminal (OLT) and connects to all subtended customers. The 1 Gbps (EPON, [15]) to 2.448 Gbps (GPON, [16]) data rate is shared among all users. EPONs serve 32 customers, while GPONs serve up to 128 customers, leaving PON’s continuous per-customer bandwidth in the 20–30 Mbps range (deducting overheads, where 20 Mbps = 2.4 Gbps/128 for the GPON and 30 Mbps = 1.0 Gbps/32 for the EPON). Reduction of these numbers means more fibers, PONs, and thus increasing the deployment cost of already expensive systems. A higher peak speed of 100 Mbps is allowed only when other customers are inactive, creating many PON opportunities and protocols for what is called “dynamic bandwidth allocation”, which is essentially assigning packets to allow sharing among the various sharing users. Optical splitter circuits are inserted into the fiber to drop to customer locations. These passive optical devices are expensive to insert and of course require dispatch of personnel, and digging or aerial dropping to connect to the customer along the same paths presently occupied by the copper-line connections. This is a key point – considerable investment to get 20–30 Mbps to a customer, perhaps 100 Mbps if the other customers are silent. No PONs today offer higher speeds. While multi-wavelength fiber transmission in the PON might be a research topic, the addition of such wavelengths to an existing PON requires deployment of new OLTs and additional filters on the existing LT equipment in Fig. 4.8 before the very first customer can use an additional wavelength, which differs from DSL’ replacement of transceivers at each end for just one customer with unshared copper. Thus, the CuPON comparison is against the current proposed PON systems of the next several years. NG-PON is being studied, and
76
John M. Cioffi, Sumanth Jagannathan, Mehdi Mohseni, and George Ginis
200 pair Junction 4
LT DSL Junction 4
Junction 4
Junction 4
Fig. 4.8 Basic CuPON Architecture
a 10 Gbps EPON effort was recently initiated, and would provide some improvement (and again would share this bandwidth over 100 customers or more), but these are not yet seriously considered for development because of high optic component costs. Figure 4.8 copies the PON architecture but implements it with the existing 200 copper lines already present in the exact same places where fiber would be placed for a PON. Most customer drops have 2–4 pairs (or 4–8 wires). The extra copper occurs only in the last distribution area segment – in other words, the extra wires do not pass all the way back to a CO, but instead would go to an intermediate point or junction box. At that point, some customers who wanted extra phones would be connected to the more limited number of pairs in cables serving thousands of customers back to the central office. But, those extra pairs are still there in the drop, waiting to be used. In the CuPON architecture (Fig. 4.8), these drop pairs are shared, possibly between two or more homes by connection. Section 4.2 illustrated that 500 m four-pair drops could carry 500 Mbps to 1 Gbps, possibly in this case shared by perhaps two users and at most four. This result is well known for vectoring (see [3–7, 9, 10] for examples) to extend to any number of pairs outside the bonded group of four as long as they are all terminated as shown in Fig. 4.8. Thus vectoring performed between groups of pairs allows the 500 Mbps – 1 Gbps to be achieved fundamentally on any group of two to four wires to a customer. No labor is necessary other than the junction box connection – no new fiber needs to be deployed other than the fiber to the node (the node is the LT DSLAM in Fig. 4.8, and this fiber is not shown). And, the peak speed is just as high and the continuous simultaneous speeds to all customers are hundreds of Mbps, much higher than the current PON speeds. Indeed, 200 such wires have 100 Gbps of bandwidth that could be shared in a variety of ways. The speeds of the CuPON can be higher, not because copper has wider bandwidth than fiber, but because the fibers do not use their extra bandwidth,
4 Vectored DSLs and the Copper PON (CuPON)
77
and it could require changes to use it (possibly replacement of the passive splitters with WDM components, and upgrades to CO equipment, and new ONTs for new subscribers). Table 4.1 compares the traditional PONs with the CuPON. The data rates listed are unidirectional (so can be doubled for an aggregate of both directions). The comparison is for 200 pairs (essentially the 192 commonly encountered in distribution areas) in the CuPON case; it is these 200 pairs that might be replaced by a PON. The speeds of the CuPON are clearly higher, not because copper has wider bandwidth than fiber, but because the fibers do not use their extra bandwidth, and it would require major changes to use it (possibly with replacement of the fiber itself and certainly with all the electronics already there – and unfortunately, likely replacement of all CPE equipment just so the first single customer can get higher bandwidth with some future enhanced PON). The peak speed of the CuPON could be as high as 10s of Gbps if the architecture in Fig. 4.8 were extended to have many lines dropped, so the range 100 Mbps (which is a single line with differential vectoring) to 10s Gbps is listed. The peak data rate with four drop wires per customer would be 1–2 Gbps. The range of the PON is certainly longer but most telephone companies use a distribution-area size of less than 1 km for support of at least 200 customers. The main PON range advantage really translates to the absence of active elements other than at the customer’s location or the central office. Thus, electronic failure of the fiber is less likely because the active components are in the nice climate-controlled central office (although fiber to the home does require batteries at the home while DSL does not – that is normal phone service works without batteries on copper, but not with fiber). The higher-bandwidth copper has the disadvantage that for most customers, an active line or remote terminal DSLAM is not in the central office. Thus, maintenance of copper needs very sophisticated software running in the telephone company network, notably DSM. Such software is now becoming common in use and many are now in full deployment. Such methods have proven to provide value at 100 times their cost, so the advancement of such DSM to include vectoring is very promising.
Table 4.1 Comparison of CuPON and G/E PONs (G and E) PON Copper (CuPON) Aggregate data rate 1–2.5 Gbps 50 Gbps User continuous data rate 20–100 Mbps 50–1000 Mbps User peak data rate 100 Mbps–1 Gbps 100 Mbps – 10s Gbps Range 10–20 km 1 km Power Battery at ONU (no battery if FTTH) LT needs to be powered Splitter Hard Easy Maintenance Expected to be easy Needs DSM to be cost effective Deployment cost High Much less
78
John M. Cioffi, Sumanth Jagannathan, Mehdi Mohseni, and George Ginis
4.4 Summary Copper access networks continue to have enormous dormant bandwidth that can be exploited by vectoring and advanced vectoring enhancements like the CuPON. With 50 Gbps in each direction possibly shared in a CuPON structure, a 100 Gbps local access network is possible, on the existing copper. Such an access network is beyond today’s most aggressive bandwidth projections and needs. Thus, those old copper pairs may indeed be the most precious metal in the world, deployed everywhere for all to share in an enormous information age.
References 1. Cioffi, J.M. & Mohseni, M. (2004). Dynamic Spectrum Management – A methodology for providing significantly higher broadband capacity to the users. 15th International Symposium on Services and Local Access (ISSLS). Edinburgh, Scotland. 2. Song, K.B., Chung, S.T., Ginis, G., & Cioffi, J.M. (2002). Dynamic spectrum management for next-generation DSL systems. IEEE Communi. Mag. 40(10), 101–109. 3. Starr, T., Sorbara, M., Cioffi, J.M., & Silverman, P. (2003). Dynamic spectrum management for next-generation DSL systems (Chapter 11). Upper Saddle River, NJ: Prentice-Hall, Inc. 4. Cioffi, J.M., et al. (2006). Vectored DSLs with DSM: The road to ubiquitous gigabit DSLs. World Telecommunications Conference. Budapest, Hungary. 5. Ginis, G. & Cioffi, J.M. (2002). Vectored transmission for digital subscriber line systems. IEEE J. Select. Areas. Comm. 20(5), 1085-1104. 6. Lee, B., Cioffi, J.M., Jagannathan, S., & Mohseni, M. (2007). Gigabit DSL. IEEE Trans. Comm. 55(9), 1689–1692. 7. Cioffi, J.M., Lee, B., Mohseni, M., & Leshem, A., Li, Y. (2004). GDSL (Gigabit DSL). ANSI T1E1.4 Contribution 2004-487R1. Washington, DC. 8. IEEE 802.3an-2006 (Amendment to IEEE 802.3-2005). IEEE Standard for Information technology – Telecommunications and information exchange between systems-Local and metropolitan area networks-Specific requirements Part 3: Carrier Sense Multiple Access with Collision Detection (CSMA/CD) Access Method and Physical Layer Specifications. Amendment 1: Physical Layer and Management Parameters for 10 Gb/s Operation, Type 10GBASE-T. 9. Jagannathan, S., Pourahmad, V., Seong, K., Cioffi, J.M., Ouzzif, M., & Tarafi, R. (2009). Common-mode data transmission using the binder sheath in digital subscriber lines. IEEE Trans. Comm. 57(3), 831–840. 10. Cioffi, J.M., Jagannathan, S., et al. (2006). Full binder level 3 DSM capacity and vectored DSL reinforcement. ANSI NIPP-NAI Contribution 2006-041. Las Vegas, NV. 11. Dynamic Spectrum Management Technical Report. ATIS Committee NIPP Pre-Published Document ATIS-PP-0600007. 12. ITU-T G.vector (2007). G.vector: Proposed working text. ITU-T Contribution RJ-R18R1. Redbank, N.J. 13. Isaksson, M. (2006). Vectored VDSL components. IEEE globecom. San Francisco, CA. 14. Zidane, R., et al. (2005). VDSL2 Profile 30a service configurations and band-plan proposal. ANSI NIPP-NAI Contribution 2005-184R2. Las Vegas, NV. 15. IEEE 802.3-2005 (Revision of IEEE 802.3-2002). IEEE Standard for Information technology – Telecommunications and information exchange between systems-Local and metropolitan area networks-specific requirements Part 3: Carrier sense multiple access with collision detection (CSMA/CD) access method and physical layer specifications.
4 Vectored DSLs and the Copper PON (CuPON)
79
16. ITU-T Recommendation G. 984.1 Gigabit-capable Passive Optical Networks (GPON): General characteristics, Geneva, Mar. 2003, ITU-T Recommendation G. 984.2, Gigabitcapable Passive Optical Networks (GPON): Physical Media Dependent (PMD) layer specification, Geneva, Mar. 2003, ITU-T Recommendation G. 984.3, Giga-bit-capable Passive Optical Networks (GPON): Transmission conver-gence layer specification, Geneva, Feb. 2004, and ITU-T Recommendation G. 984.4, Gigabit-capable Passive Optical Net-works (GPON): ONT management and control interface specification, Geneva, June 2004.
Chapter 5
Enabling Broadband Wireless Technologies∗ Quazi M. Rahman
Abstract The fundamental features of next generation wireless communications systems include very high-speed data transmission–reception scheme and ubiquitous interactive multimedia services both of which demand wide bandwidth system. To fulfill this demand, broadband communication system has been brought into the wireless domain, and in recent years, it has been offering a remarkable growth in the telecommunications industry. The major goal of all these technologies is to deliver very high data rate (in multimegabits per seconds) throughput to the end user by satisfying some stringent quality of service (QoS) to support variety of services, such as data, voice, and multimedia. Here we discuss a number of physical layer techniques that are enabling the broadband wireless technologies to move forward by meeting some of these QoS requirements. The discussion addresses modulation, coding, multiple access, and diversity techniques. Some challenges along with some research evidences on the broadband wireless communication systems have also been presented in this chapter.
5.1 Introduction In recent years, broadband wireless communication has brought a remarkable wireless revolution in the telecommunications industry, and currently it is signaling the arrival of the next generation wireless communications systems [1]. It is appearing to be an attractive technology because of its quick and easy deployment features in the wireless network domain, and its growth is expected to continue in the future. Broadband wireless technology is on the verge of bringing the wired broadband experience, such as fast web surfing, quicker file downloads, real-time Quazi M. Rahman Department of Electrical and Computer Engineering, The University of Western Ontario, 1151 Richmond Street, London, ON, Canada N6A 5B8, e-mail:
[email protected] ∗
Some contents of this chapter has been published in Chapter 1 of [30].
A. Shami et al. (eds.), Broadband Access Networks, Optical Networks, c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-92131-0 5,
81
82
Quazi M. Rahman
audio and video streaming, multimedia conferencing, and interactive gaming, in the wireless domain [2] for the users’ benefits and convenience. Currently, wireless services are provided both for fixed and mobile wireless broadband systems. The existing and emergent broadband technologies include Worldwide Interoperability for Microwave Access (WiMAX e.g., IEEE 802.16-2004 etc.), CDMA2000, Wireless Fidelity (Wi-Fi, e.g., IEEE 802.11 a/b/g/n), 1xEVDO (IS856), Wideband Code Division Multiple Access (W-CDMA) for Universal Mobile Telecommunications Systems (UMTS), etc. The broadcast nature of these broadband wireless technologies are assuring to offer ubiquity and immediate access for both fixed and mobile users by rendering services involving voice, video, data, and mobility. To provide broadband applications to the subscribers, the mobile operators are constantly upgrading their networks to 3G technologies such as UMTS and HSDPA (high-speed downlink packet access) for GSM-based systems, 1xEV-DO (1x evolution data optimized) and TD-SCDMA (time-division synchronous CDMA) for CDMA-based systems. Current mobile broadband applications are involving high-speed vehicles, trains, and even airplanes. All these broadband wireless schemes incorporate several advanced radio transmission technologies, such as orthogonal frequency division multiplexing (OFDM), adaptive modulation and coding, adaptive forward error correction (FEC), and other techniques and technologies to provide well-defined quality-of-service (QoS) framework including high rate transmission in the range of tens and hundreds of Mbps. Designing wireless broadband technologies by meeting these rigorous service conditions is always challenged by many factors, such as hostile wireless channel, limited available spectrum, multiplexed services with variety of QoS requirements, uninterrupted mobility requirement, robust security provision, low power and low cost requirements. There is no unique solution to any of these technical challenges and that is why broadband wireless technologies are currently generating significant interest to researchers and practitioners who are involved in the design, analysis, installation, and management of mobile broadband wireless access systems and networks. This is the major motivation for this chapter to discuss different technologies that are enabling the broadband wireless experience a smooth one for its users. This chapter reviews in particular: modulation and coding aspects, multiple access techniques, and diversity combining. In discussing these issues Table1 5.1 [2] has been referred, where different characteristic parameters of some of the already deployed broadband wireless schemes have been presented. For broadband systems, the study of wave propagation through the channel is an important task when developing a wireless scheme. Therefore, an accurate modeling of the channel is necessary for the design of a broadband wireless scheme. The discussion on the channel modeling technique is not the scope of this chapter but interested readers can check [2–4] and the references therein for information on channel modeling. The outline of this chapter is as follows. Section 5.2 presents the modulation schemes which are currently in use in the wireless broadband domain. Section 5.3 discusses some channel coding techniques currently employed for the broadband 1
http://www.3gpp2.org/Public html/Misc/AboutHome.cfm http://www.3gpp.org/specs/releases-contents.htm
5 Enabling Broadband Wireless Technologies
83
Table 5.1 Characteristic parameters for different broadband wireless technologies Schemes
Standards
Modulation tech- Mobility Range niques WiMAX IEEE 802.16e-2005 BPSK, QPSK, 16 Moderate Less than 3.5 km QAM, 64 QAM WiFi IEEE 802.11 a/g/n BPSK, QPSK, 16 Low Indoor: less that QAM, 64 QAM 30 m and outdoor: less than 300 m IS856 3GPP2 8PSK, QPSK, 16 High 1.5–5 km QAM HSDPA 3GPP Release 6 QPSK, 16 QAM High 1.5–5 km
Multiplexing TDM/OFDMA CSMA
TDM/CDMA TDM/CDMA
applications. Section 5.4 has been dedicated to adaptive modulation and coding scheme where the fundamental idea of this scheme has been presented. Different multiple access schemes along with some recently applied multiple access techniques for wideband wireless applications have been addressed in Section 5.5. Section 5.6 discusses the basic diversity schemes and relates those to the existing broadband wireless schemes. Some research evidences, which are addressing current challenges in the broadband wireless research domain, have been presented in Section 5.7. Finally conclusions are drawn.
5.2 Modulation: Let’s Deal with the Available Spectrum and Make Use of the Limited Power For a broadband wireless communications system the researchers are always in search of a smart modulation scheme which can boost up both the spectral2 and power efficiency3 of the system by taking care of the limited spectrum and low power requirement challenges mentioned earlier. Joint optimization of spectral and power efficiency parameters is a challenging task. In addressing this challenge, different modulation techniques, which are already in use in the broadband wireless system have been discussed in this section. The deployed broadband wireless communications systems use different variants of phase shift keying (PSK) and quadrature amplitude modulation (QAM) techniques for achieving high spectral and, to some extent, high power efficiencies. On the other hand, for very high data rate applications, multicarrier modulation, also known as orthogonal frequency division multiplexing (OFDM), has successfully been deployed in wireless local area networks (LAN) (IEEE 802.11 a/g/n), digital video broadcasting, WiMAX and other emerging broadband wireless systems (such 2
Spectral efficiency demonstrates the ability of a system (e.g., modulation scheme) to accommodate data within an allocated bandwidth. 3 Power efficiency represents the ability of a system to reliably transmit information at a lowest possible practical power level.
84
Quazi M. Rahman
as the Flash-OFDM,4 3G LTE5 ), and next generation cellular systems. In this section we will focus on these three types of modulation techniques.
5.2.1 Phase Shift Keying (PSK) Modulation In this type of digital modulation technique, the modulating data signals shift the phase of the constant amplitude carrier signal between M number of phase angles. The analytical expression for the mth signal waveform in PSK modulations has the general form: m = 1, 2, ..., M 0≤t ≤T (5.1) sm (t) = g(t) cos 2π f c t + θm , where, g(t) is the signal pulse shape and m = 2π(m − 1)/M, m = 1, 2, .., M, are the M (M = 2 for binary-PSK and M = 4 for quadrature-PSK) possible phase angles of the carrier frequency f c that convey the transmitted information for M = 2k possible k-bit (k being a positive integer) blocks or symbols. The mapping of k information bits is preferably done through grey encoding so that the most likely errors caused by noise will result in single-bit error in the k-bit symbol. In binary shift keying (BPSK), the modulating data signals shift the phase of the constant amplitude carrier signal between 0◦ and 180◦ as shown in the state diagram of Fig. 5.1. A more common type of PSK modulation is quadrature phase shift keying (QPSK), where the modulating data signals shift the phase of the constant amplitude carrier signal in increments of 90◦ ; for example, from 45◦ to 135◦ , −45◦ , or −135◦ (Fig. 5.1). QPSK (22 = 4 states) is more spectral-efficient type of modulation scheme than BPSK (21 = 2 sates). For greater spectral efficiency in MPSK system, we can increase the value of M (2x = M, x > 0 is an integer) to a higher number but this increase in the M-value requires more signal power (Fig. 5.2) to achieve the same bit error rate (BER). In other words, we gain spectral efficiency for the price of power efficiency with higher level (M) PSK. There are many variations in the PSK modulation format, which are in use due to better power and spectral efficiency requirements. Offset QPSK (OQPSK), differential QPSK (DQPSK), and π/4 DQPSK are few examples of these PSK modulation formats. In OQPSK, the in-phase and quadrature bit streams are offset in their relative alignment by one bit period. As a result, the signal trajectories are modified in such a way so that the carrier amplitude does not go through (or near) zero (the center of the constellation). In this case the spectral efficiency of OQPSK-based system remains the same as QPSK-based system but the reduced amplitude variations for the former allow a more power-efficient, less linear Radio Frequency (RF) power amplifier to be used. For DQPSK modulation, the information is carried out by the transitions between states. In some cases there are also restrictions on allowable transitions. For example, in π/4 DQPSK modulation, the carrier trajectory does 4 5
http://www.qualcomm.com/technology/flash-ofdm/ http://www.ericsson.com/technology/tech articles/super 3g.shtml
5 Enabling Broadband Wireless Technologies
85
s2
s1
s3
s4
s1
s2
(a)
(b)
Fig. 5.1 Phase shift keying state diagrams; (a) BSPK, (b) QPSK
not go through the origin [5]. The π/4 DQPSK modulation format uses two QPSK constellations offset by 45◦ (π/4 radians). Like OQPSK, π/4 DQPSK is a power efficient modulation method and with root cosine filtering it provides better spectral efficiency than Gaussian minimum shift keying (GMSK) [5] modulation. BPSK and QPSK modulation techniques and their different variants are used in almost all the broadband wireless schemes as shown in Table 5.1.
5.2.2 Quadrature Amplitude Modulation (QAM) QAM is simply a combination of Pulse Amplitude Modulation (PAM6 ) and PSK modulation techniques. In this scheme, two orthogonal carrier frequencies (in-phase and quadrature carriers), occupying identical frequency bands, are used to transmit data over a given physical channel. The analytical expression for the mth signal waveform in QAM technique can be expressed in a general form as: (5.2) sm (t) = Am (t)g(t) cos 2π f c t + θm , m = 1, 2, ..., M 0 ≤ t ≤ T where, Am = A2mc + A2ms and θm = tan−1 (Ams /Amc ), g(t) is the signal pulse shape, f c the carrier frequency that conveys the transmitted information, and Amc and Ams are the information-bearing signal amplitudes of the quadrature carriers. In (5.2), for constant θm , sm (t) represents the PAM signal while for constant Am , it represents PSK signal. By choosing the different amplitudes and phases, different constellations of QAM signals can be formed (Fig. 5.3). In this case the power efficiency of the communication system will vary depending on the type of signal constellation [6] used for the QAM technique. Due to the flexibility 6
In PAM the modulating data signals shift the amplitude of a constant-phase carrier signal between M number of discrete levels.
86
Quazi M. Rahman 0
10
–1
10
–2
10
64-QAM
BER
64-PSK –3
10
4-QAM/PSK –4
10
16-QAM
16-PSK
–5
10
–6
10
0
5
10
15 SNR per bit
20
25
30
Fig. 5.2 BER comparison between MQAM and MPSK techniques in the AWGN channel with optimum detection
of using different amplitudes and phases, even with high level (M > 4) QAM, the choice of decision region in QAM is not as critical as PSK. As a result MQAM based system is more power efficient than M-PSK based system [7]. Furthermore, the penalty in signal-to-noise ratio (SNR) for increasing M is much less in M-QAM than in M-PSK based system (Fig. 5.2). As demonstrated in [8– 10], among the spectrally efficient modulation schemes, M-QAM offers the best trade-off between implementation complexity and performance in different channel environments. Due to its both power and spectral efficiency, QAM technique finds its application in the broadband wireless domain. In this domain, 16 and 64 QAM techniques along with QPSK modulation are used in WiMAX; 16 QAM along with QPSK is used in high-speed packet access (HSPA) scheme; in IS856 system 16-QAM technique, along with QPSK and 8QPSK modulations, is used in the forward link to support multirate data applications; in Wi-Fi systems, 16 and 64 QAM along with BPSK and QPSK are used to support multirate data applications. Besides, QAM is used in applications including microwave digital radio, digital video broadcastingcable (DVB-C), and modems.
5 Enabling Broadband Wireless Technologies
(a)
87
(b)
Fig. 5.3 8-QAM Constellations (a) 4-4 constellations. (b) Rectangular constellations
5.2.3 Orthogonal Frequency Division Multiplexing (OFDM) OFDM, also known as multicarrier modulation, is a wideband modulation scheme, which is specifically designed to cope with the problems of multipath reception. It achieves this by transmitting a large number of narrowband digital signals over a wide bandwidth. In OFDM, at the transmitter side, the high bit-rate data stream is first split into N parallel substreams at a rate of N /T streams/sec, with symbol duration T sec. Each of these substreams is then modulated by individual carrier frequencies in each branch. All these carrier frequencies, also known as subcarriers or tones, are orthogonal to each other in such a way that the pth frequency f p results f p = f 0 + p/T with f 0 being the RF. This, essentially, is the OFDM phenomenon that results in overlaps between the spectra associated with different tones (Fig. 5.4); as long as the orthogonality between these tones is unchanged, the signals carried by each tone can be recovered successfully. The multitone signal is obtained by adding different modulated carriers. The closely spaced overlapping and the orthogonality between the carriers provide the spectral efficiency for the OFDM system. Due to its parallel transmission scheme, OFDM system allows transmitting high data rates over extremely hostile channels at a relatively low complexity. In this case, the symbol duration of the OFDM signal is increased compared to that of the high data rate input stream as shown in Fig. 5.5 and that makes the system less sensitive to large delay spread7 (channel induced delay). In Fig. 5.5, let us assume that the symbol duration for a single-carrier system is T sec. With OFDM system, it becomes 2T sec for two carriers, 4T sec for four carriers, and M T sec for M carriers. Assuming that the maximum delay spread is tmax sec the inter-symbol interference (ISI) in each channel is reduced to tmax /M T sec. This demonstrates that the OFDM signal is less sensitive to large delay spread which also reduces the effect of ISI significantly [11]. 7
Range of time delay associated with the received signal over which the power of the received signal is a nonzero value.
88
Quazi M. Rahman 1
0.8
0.6
0.4
0.2
0
-0.2
-0.4
fo
fo+1/T
fo+2/T
Fig. 5.4 Spectrum associated with OFDM signal
Since OFDM technique is based on a block transmission scheme, measures are taken to avoid or compensate for interblock interference (IBI) which contributes to the overall ISI. To take care of this interference, a guard interval is introduced between consecutive OFDM symbols as a cyclic prefix. The length of the cyclic prefix is chosen to be greater than the maximum delay spread of the channel. One of the remarkable features in OFDM system is that the OFDM signals can be easily transmitted and received using the fast Fourier transform (FFT) devices [12, 13]. In the OFDM system, coding and interleaving (discussed later) techniques across the subcarriers in the frequency domain can be used to take care of the deep fade channel condition. It can also be used as a multiple access scheme, as exploited in mobile WiMAX [2], where each tone can be assigned to each user. OFDM technique has some demerit points too. Adding a guard interval, between OFDM symbols results in a decrease of both power and spectral efficiency of the system. This can be taken care by using more subcarriers that make the OFDM symbol significantly larger than the guard interval. But with longer symbol duration, a fast time-varying channel provides a challenging scenario in ensuring perfect orthogonality between the carriers due to the Doppler spread8 , 8
Spreading of the frequency spectrum of the transmitted signal resulting from the rate of change of the mobile radio channel is known as Doppler spread
5 Enabling Broadband Wireless Technologies
89
Channel Impulse Response time 0
tmax
1 carrier T
Symbol period 2 carriers 2T
4 carriers
4T
Fig. 5.5 OFDM system is less sensitive to large delay spread
and this results in inter channel/carrier interference (ICI) due to frequency offset resulting from the Doppler spread. So there is always a trade-off between delay and Doppler spreads while working with OFDM carriers. OFDM system shows high envelope power fluctuations. With many carriers, it results in a large peakto-average power ratio (PAPR), which reduces the power efficiency and increases the cost of the power consumption of the transmitter amplifier. To remedy this scenario, the operating point in the amplifiers can be backed off, but this leads to inefficient power usage. Coding methods have been proposed in [14, 15] to reduce the PAPR. The successful use of OFDM technique began in the 1960s for high-frequency military systems (KINEOLEX, ANDEFT, KATHRYN). At the beginning of twentyfirst century, it has been adopted as a standard for new high data rate wireless local area network (WLAN) standard, such as IEEE 802.11 and HIPERLAN II, as well as the Japanese Multimedia Mobile Access Communications (MMAC) [16]. Currently OFDM has become a part and parcel for the broadband wireless communications schemes, such as WiMAX and WiFi systems [17]. Moreover, OFDM technology is chosen to be a very attractive candidate when targeting high quality and highly flexible mobile multimedia communications over satellite systems [12].
90
Quazi M. Rahman
5.3 Coding Techniques: Let’s Deal with the Channel In this section, channel coding techniques for the broadband wireless system are discussed. Channel coding is a systematic approach for the replacement of the original information symbol sequence by a sequence of coded symbols, in such a way as to permit its reconstruction. It ensures adequate transmission quality of the transmitted signals through the channel by taking care of the severe transmission conditions in mobile radio environment due to multipath fading9 . Moreover, it can help to overcome very low signal-to-noise ratio (SNR) for some communications medium, such as satellite, due to limited transmit-power in the downlink. The channel encoding process generally involves mapping of every k-bit information sequence into a unique n-bit sequence, where the latter is called a code word. This kind of channel encoding process is known as forward error correction (FEC) encoding where the idea is to include redundant bits in the transmission that will allow the receiver to detect and correct a certain percentage of the encountered errors. The amount of redundancy introduced by the encoding process is measured by the ratio k/n whose reciprocal is known as code rate. The output of the channel encoder is fed to the modulator whose output is transmitted through the channel. At the receiver end, demodulation, decoding, and detection processes are carried out to decide on the transmitted signal information. In the decision process, two different strategies are used: soft decision and hard decision. When the demodulator output consists of discrete elements 0 and 1, the demodulator is said to make firm or hard decision. On the other hand, when the demodulator output consists of a continuous alphabet or its quantized approximation (with greater than two quantization levels) the demodulator is said to make soft decisions. It is theoretically shown by Shannon [18] that coding technique in general improves the (BER) performance of a communications system. Channel coding can be classified into two major areas: waveform coding and structured sequences. The objective of the waveform coding is to provide an improved waveform set so that the detection process is less subject to errors. Examples of this coding technique include M-ary signaling, antipodal, orthogonal, bi-orthogonal, and trans-orthogonal signaling. Structured sequences deal with transforming data sequences into better sequences having ordered redundancy in bits. The redundant bits can then be used for the detection and correction of errors. Examples of structured sequence type coding include block and convolutional coding schemes. Since, the currently deployed broadband wireless technologies are mainly based on structured sequence type coding schemes, our discussion in this section will be limited to this kind of coding schemes only. These schemes include linear block codes (e.g., Hamming-Codes, BCH-codes and Reed-Solomon-Codes) and convolutional codes. Turbo codes (TC) and space–time codes will also be considered in the discussion, which are getting enormous attention in the current developments of next generation broadband wireless technologies. 9
Combined distorted effect of the received signals traveling through multiple paths is known as fading.
5 Enabling Broadband Wireless Technologies
91
In general, the coding schemes provide error-rate performance gain and power gain for the price of bandwidth efficiency of the system. To provide improvement in the power efficiency without sacrificing the bandwidth efficiency, coded modulation techniques, such as trellis-coded modulation (TCM) technique, have been proposed in [19] which have also been discussed in this section because of its popularity in the broadband wireless research domain [20].
5.3.1 Block Codes In this coding scheme, each k-bit information symbol block is converted to n-bit coded symbol block with (n − k) redundancy bits added to the k-bit symbols. These redundancy bits could be parity bits or check bits which do not carry any information. The resulting code is referred as (n, k) block code. Here the redundancy of the code is defined as the ratio between the redundant bits and k-bit symbol, i.e., (n − k)/k while the code rate is defined as k/n. In block codes 2k k-bit message sequences are uniquely mapped into 2k n-bit codes, out of possible 2n n-bit codes. Although this redundancy provides performance gain and better power efficiency, it dictates a faster rate of transmission which of course requires more bandwidth. Moreover, it increases data transmission delay due to the increased block length and decoder complexity at the receiver end where the decoder searches through 2k valid code words to find the best match with the incoming 2n possible coded blocks. So there is always a trade-off between coded and uncoded schemes. Hamming codes: It is a linear block code10 characterized by the following (n, k) structure, (5.3) (n, k) = (2m − 1, 2m − 1 − m) where m = 2, 3, ..., p. These codes are capable of correcting all single-bit errors and detecting all combinations of two or fewer bits in error which can be seen from the following example. A rate 4/7, i.e., (n, k) = (7, 4) hamming code is shown in Table 5.2 where each output data block differs from all the other blocks by at least three bits. Hence if one-or two-bit error occurs in the transmission of a block, the decoder will detect that error. In the case of single-bit error, it is also possible for the receiver to match the received block to the closest valid block and thereby correct the single-bit error. If three-bit error occurs, the original block may be transformed into a new valid block and all the errors go undetected. The difference in the number of bits between two coded blocks is known as the Hamming distance and a block code of a Hamming distance d can detect up to (d − 1) errors and correct (d − 1)/2 errors. Bose–Chaudhuri–Hocquenghem (BCH) code: These codes are generalizations of Hamming codes that allow multiple error corrections. BCH codes [21] are important because at block length of a few hundred bits, these codes outperform all other block 10 In linear block coding scheme, each of the code words is formed by modulo-2 sum (EX-OR) of two or more other code words.
92
Quazi M. Rahman
Table 5.2 Hamming code of rate 4/7 Hamming code with rate 4/7 0 0000 000 0000 1 1000 110 1000 2 0100 011 0100 3 1100 101 1100 4 0010 111 0010 5 1010 001 1010 6 0110 100 0110 7 1110 010 1110 8 0001 101 0001 9 1001 011 1001 10 0101 110 0101 11 1101 000 1101 12 0011 010 0011 13 1011 100 1011 14 0111 001 0111 15 1111 111 1111
codes with the same block length and code rate. For very high coding overhead with long block length, this coding scheme can be used where reliability of transmission is the key factor and data throughput is less important. Reed-Solomon (RS) codes: RS codes are a subclass of BCH codes that operate at the block level rather than the bit level. Here, the incoming data stream is first packaged into small blocks, and these blocks are then treated as a new set of k symbols to be packaged into a super-coded block of n symbols. As a result, the decoder is able to detect and correct complete error-blocks. This is a nonbinary code set which can achieve the largest possible code-minimum-distance for any linear code with the same encoder input and output block lengths. For nonbinary codes, the distance between two code words is defined as the number of nonbinary symbols in which the sequences differ. The code minimum distance for the RS codes is given by [22]. dmin = n − k + 1
(5.4)
These codes are capable of correcting any combination of (n − k)/2 or fewer symbol errors. Despite revolutionary developments in capacity-approaching codes in recent years, RS codes remain very relevant today, especially for high rate systems with relatively small data packets. RS codes are particularly useful for burst type error corrections and so they are very effective with the channel with memory. This coding scheme is used in WiMAX system. In the downlink of the OFDM mode in WiMAX system, where subchannelization is not used, the output of the data randomizer is first encoded using an outer systematic11 RS code and then encoded using an inner rate 1/2 binary convolutional encoder (discussed later). For channel 11 The code words with the original information data unchanged and unscrambled are known as systematic code words.
5 Enabling Broadband Wireless Technologies
93
coding, RS code along with convolutional and turbo codes (discussed later) are used for UMTS terrestrial radio access (UTRA) systems. Interleaving: The block codes work best when errors are distributed evenly and randomly between incoming blocks. This is usually the case for AWGN channels such as landline telephone link. In a mobile radio environment, however, errors often occurs in bursts as the received signal fades in and out due to the multipath propagation and the user’s motion. In order to distribute these errors more evenly between coded blocks, a process known as interleaving is used. In general, to accomplish interleaving, the encoded data blocks are read as rows into a matrix. Once the matrix is full, the data can be read out in columns, redistributing the data for transmission. At the receiver, a deinterleaving process is performed using a similar matrix filling and emptying process, reconstructing the original blocks. At the same time the burst errors are uniformly redistributed across the blocks. The number of rows or columns in the matrix is sometimes referred to as the interleaving depth. The greater the interleaving depth, the greater resistance to long fades, but also greater the latency in the decoding process as both the transmitter and receiver matrix must be full before encoding or decoding can occur. All the current generation broadband wireless systems use interleaving technique to take care of the burst type errors.
5.3.2 Convolutional Codes A convolutional code is described by three different integers, n, k, and L, where the ratio k/n has the same code rate significance that it has for block codes; however, n does not define a block or codeword length as it does for block codes. The integer L is a parameter known as the constraint length which represents the length of the encoding shift register. This code is implemented on a bit-by-bit basis from the incoming data-source stream. The encoder has memory and it executes an algorithm using a predefined number of the most recent bits to yield new coded output sequence. Convolutional codes are linear where each branch word of the output sequence is a function of the input bits and (k − 1) prior bits. Since the encoding procedure is similar to the convolution operation, the coding technique is known as convolutional coding. The decoding process is usually a serial process based on present and previously received data bits (or symbols). Figure 5.6 shows a (n, k) = (2, 1) convolutional encoder with constraint length L = 3. There are n = 2 modulo-2 adders those result a two-bit coded word for each input bit upon EX-OR operation. The output switch samples the output of each modulo2 adder thus forming the two-bit code symbol associated with the single input bit. The sampling is repeated for each input bit that results-in two-bit code word. The choice of the connections between the adders and the stages of the shift register gives rise to the characteristics of the code. The challenging problem in this case is to find out an optimal connection pattern that can provide codes with best distance properties. Convolutional codes have no particular block size; nonetheless, these are
94
Quazi M. Rahman
Shift Register
Modulo 2 Adder
Bit Input
Output coded word
Modulo 2 Adder
Fig. 5.6 Rate 1/2 convolutional encoder with constraint length of 3
often forced into a block structure by periodic truncation. This requires a number of zero bits to be added at the end of the input data sequence for clearing out the data bits from the encoding shift register. Since the added zeros carry no information, the effective code rate falls below k/n. The truncation period is generally made as long as practical to keep the code rate close to k/n. In general, both the encoder and decoder can be implemented using recursive techniques, with one of the most efficient and well known being the Viterbi convolutional decoding technique [5]. Convolutional codes at rate 1/2, 2/3, 3/4, and 5/6 are used both in the downlink and uplink of WiMAX schemes; UTRA system uses convolutional coding at rate 1/2 or turbo coding (discussed later) at rate 1/3 in the uplink; W-CDMA uses covolutional codes and turbo codes for channel coding. The standards IEEE 802.11a/g and Hiperlan II use rate 3/4 convolutional codes and IEEE 802.11b includes an optional mode that uses them. Pictorial representation of convolutional encoder: A convolutional encoder can be represented pictorially in three different ways: State diagram, Tree diagram, and Trellis diagram. State diagram: In this case, the encoder is characterized by some finite number of states. The state of a “rate 1/n” convolutional encoder is defined as the contents of the rightmost L − 1 stages (Fig. 5.6) of the shift register. The necessary and sufficient condition to determine the next output of a convolutional encoder is to have the knowledge of the current state and the next input. The state diagram for the encoder shown in Fig. 5.6 can easily be drawn as shown in Fig. 5.7. The states shown in the boxes of the diagram represent the possible contents of the rightmost L − 1 stages of the register, and the paths between the states represent the output branch words resulting from such state transitions. Table 5.3 will help to understand the state transition mechanism in Fig. 5.7. Major characteristics of the state diagram follow: • 2k.(L−1) states. • 2k branches are entering each state while same number of branches is leaving each state.
5 Enabling Broadband Wireless Technologies
95
00 10 01
00 Branch Output
11 01
11 Encoder State
01
11
10
Solid Line: Input Bit 1 Dotted Line: Input Bit 0
00
10
Fig. 5.7 State diagram for the rate 1/2 convolutional encoder with constraint length of 3 as shown in Fig. 5.6
Tree diagram: Tree diagram in the convolutional encoder incorporates the time dimension in the state transition which is not provided by the state diagram. Here the possible code sequences generated by an encoder are represented as branches of a tree. With the aid of time dimension one can easily describe the encoder as a function of a particular input sequence. For examples on tree diagram, see [7, 23, 24]. Since the number of branches in the tree increases as a function of 2 S , S being the sequence length, for a very long sequence this representation is not feasible. Characteristics of the tree diagram include: • 2k branches emanate from each node. • The whole tree repeats itself after the Lth stage. Trellis diagram: Trellis diagram is an intelligent pictorial representation of the tree diagram where the repetitive nature of the tree diagram is smartly utilized
Table 5.3 State transition mechanism for the state diagram Input bit Register content Present state (content of the rightmost K − 1 stages) 0 000 00 1 100 00 1 110 10 1 111 11 0 011 11 0 001 01 0 010 10 1 101 01
Next state (content of the leftmost K − 1 stages) 00 10 11 11 01 00 01 10
Branch output present state 00 11 00 10 01 10 11 01
at
96
Quazi M. Rahman
[7, 23, 24]. Since the diagram looks like the garden trellis so it is called trellis diagram. Major characteristics of the trellis diagram include: • 2k.(L−1) states. • 2k branches are entering each state while same number of branches is leaving each state. For examples on trellis diagram, see [7, 23, 24].
5.3.3 Turbo Coding (TC) TC is specifically a decoding technique which is developed from two older concepts: concatenated coding and iterative decoding. These codes are built from parallel concatenation of two recursive systematic blocks [5] or convolutional codes with nonuniform interleaving. The term “Turbo” is used to draw an analogy of this decoding process with a turbo engine in which a part of the output energy is fed back to the input to carryout its operation. Before we discuss the principle of TC technique, we will look at the concept of concatenated coding technique. In concatenated coding method, two or more relatively simple codes are combined to provide much more powerful coding scheme. In its operation as shown in the block diagram of Fig. 5.8, the output of the first encoder (outermost) is fed to the input of the second, and so on. In the decoder, the last (or innermost) code is decoded first, and then its output is fed to the next and so on to the outermost decoder. The principle of the decoding process of the TC technique can be explained briefly with the aid of the block diagram shown in Fig. 5.9 and in terms of codearray. In this case, the decoder first performs row decoding, which generates initial estimates of the data in the array. Here, for each data bit a tentative decision and a reliability estimate for that decision are provided. The columns in the code array are then decoded by taking both the original input and the previous decoder signals into consideration. In the current decoder, the previously decoded signal information is known as a priori information on the data. This second decoding further refines the data decision and its reliability estimate. The output of this second decoding stage is fed back to the input of the fist decoder. In this case, the information which was missed in the first row decoding are decoded. The whole procedure continues until
1st Encoder
nth Encoder
1st Decoder
nth Decoder
Fig. 5.8 Concatenated coding method
Channel
5 Enabling Broadband Wireless Technologies
97
2nd Decoder
De-interleaver
De-interleaver
1st Decoder
Received Signal Decoded Data Signal
Fig. 5.9 Turbo decoder
the data estimates are converged. Turbo code finds its application in almost all of the currently existing wireless broadband communications systems such as WiMAX (optional: convolutional turbo codes at rates 1/2, 2/3, 3/4, 5/6 for both up and down link), W-CDMA, UTRA, IS856.
5.3.4 Space–Time Coding This is basically a spatial type diversity technique (discussed in Section 5.5) where multiple transmitting antennas are used (Transmit diversity) along with either single or multiple receiving antennas. This technique is known as space–time coding since it involves redundancy by transmitting the same signal using different antennas. With multiple antennas at the transmitter end, when the receiver end uses multiple antennas, the system is known as multiple-input multiple-output (MIMO) system. This MIMO configuration or space–time coding scheme increases the system reliability by decreasing the BER, increases the system capacity by increasing the achievable data rate, increases the coverage area and decreases the required transmit power. It is important to note that each of these four advantages cannot be obtained without sacrificing one or two of the above-mentioned advantages; for example, an increase in the data rate will often result in an increase in the BER or transmission power. The capacity of a wireless link generally measured in bits per second per Hertz (b/s/Hz) can be increased in a traditional single-input single-output (SISO) wireless system by increasing the signal bandwidth, allowing a corresponding increase in the bits per second. This in turn will require higher level modulation scheme which will ask for increased transmit power for a given BER (see Fig. 5.2). This increase in the transmit power can negatively impact other communications systems operating in the adjacent spectral channels or within a given geographic area. As such, bandwidth and power for a given communications system are generally well regulated, limiting the ability of the system to support any increase in capacity or performance. This limitation can be overcome by applying space–time coding scheme which exploits the idea of spatial diversity. In this case, M transmit and N receive antennas are used for the transmission–reception scheme which in turn provides an (M×N )
98
Quazi M. Rahman
Input Symbols
Space Time Encoding Unit
Space Time Decoding Unit
Decoded output Symbols
Fig. 5.10 Block diagram of a MIMO system
MIMO channel as shown in Fig. 5.10. Here, two forms of gains are achieved due to signal transmissions between M transmit and N receive antennas: diversity gain and antenna array gain. If the transmit and receive antennas are spaced sufficiently far apart, the signals traveling between the various transmit and receive antennas through the MIMO channel will fluctuate or fade in an independent manner. Based on this reasonable assumption the diversity gain will result and this gain is a product of the statistical parameters of the MIMO channels. Antenna array gain, on the other hand, does not rely on statistical diversity between the channels, and instead it achieves its performance enhancement by coherently combining the energy received by each of the antennas. Even with the completely correlated channels, which may happen in lineof-sight (LOS) scenario, the received SNR will increase linearly with the increase in the number of receiver antennas due to the array gain. Ultimately, the space–time coding scheme operating in conjunction with the MIMO channel allows the MIMO-based system to support a significant increase in both performance and capacity over an equivalent SISO system while maintaining the same bandwidth and power. Space–time coding scheme has successfully been applied in IEEE 802.11n and WiMAX. In IEEE 802.11n with MIMO application the data rate has been increased up to 600 Mbps. Recently, for the next generation wireless broad band systems space–time coding scheme is considered to be one of the most appropriate candidates.
5 Enabling Broadband Wireless Technologies
99
5.3.5 Coded Modulation Techniques In both block and convolutional coding schemes, the coding gain12 is achieved with the price paid for the bandwidth. Since in these schemes, k-bit information signal is replaced by n-bit coded words (n > k), the required bandwidth gets increased, which is a major bottleneck for the band-limited channels. To overcome this problem, combined modulation and coding schemes are considered. In this case, the coding gain is achieved with the price paid for the decoder complexity. Here different coded modulation techniques are briefly addressed. Trellis-coded modulation (TCM): TCM is based on the trellis as used in convolutional coding. In TCM, the trellis branches, instead of being labeled with binary code sequences, are represented as constellation points from the signaling constellation. Block-coded modulation (BCM): In BCM, the incoming data are divided into different levels and in each level those data streams are block coded with equal rate. Multilevel-coded modulation (MCM): In MCM, which is a generalized form of BCM, the incoming data are split into different levels/branches (serial to parallel) and each of these data levels are block coded and/or convolutionally coded with either equal or unequal rates. Finally, the multiplexed signal results in the MCM. Turbo-coded modulation: Based on the combinations of TC and either TCM or MCM, there are many versions of Turbo-coded modulation techniques available in the research area, such as turbo trellis-coded modulation (T-TCM) [25], multilevel turbo-coded modulation (ML-TCM) [26], and so on.
5.4 Adaptive Modulation and Coding (AMC) Adaptive modulation and coding (AMC) technique is a part of an adaptive transmission scheme where transmission parameters, such as modulation, code-rate, and power, are adjusted based on the channel state information (CSI). The concept of adaptive transmission, first proposed by Hayes [27], requires a destination node to send the received signal’s CSI back to the source node via a feedback channel; the source node then controls its transmit power (or code rate and/or modulation) according to the CSI feedback to compensate for the received signal level variations. This scheme has the capability to significantly increase the throughput of the wireless communications system by increasing the average data rate, spectral efficiency, and system capacity. The technique can be performed at different timescales, depending on the system capabilities and the rate at which the CSI varies. Broadband wireless communications make use of AMC to optimize transmission rate and error performance. In the AMC scheme, when the error rate at the receiver increases due to the interfered and attenuated received signal resulting from the channel, the 12 Coding gain is defined as the improvement in the SNR in dB at a specified bit error rate performance of an error-correcting coded system over an uncoded one with identical system scenario.
100
Quazi M. Rahman
Data input
Error Correcting Encoder
Symbol constellations Mapping Unit
Power Control Unit
AMC Controller
Channel Estimator
Channel
Demodulator
Data Output Decoder
Fig. 5.11 Block diagram of an AMC scheme
receiver sends this information back to the transmitter through a feedback path. The transmitter, in turn, automatically shifts to a more robust, though less efficient, AMC technique. Figure 5.11 shows the block diagram of an AMC scheme. With AMC, the power of the transmitted signal is held constant over a frame interval, and the modulation and coding format is changed to match the currently received signal quality or channel conditions. The idea is straightforward: Highest possible data rate is sent by the transmitter when the channel is in good condition, otherwise lower data rate is sent. For WIMAX systems higher data rates are achieved using larger signal constellations (such as 64 QAM) and less robust error-correcting codes (such as rate 3/4 convolutional or turbo codes); for lower data rates QPSK with rate 1/2 convolutional or turbo codes are used. In WiMAX a total of 52 combinations of modulation and coding schemes are available for the AMC technique although only a fraction of this is used in practice [2]. For HSDPA scheme AMC technique uses QPSK, 8-PSK, and 16 and 64 QAM modulation using rate 1/2 and rate 3/4 turbo codes, and can support a maximum peak data rate of 10.8 Mbps. In practice, QPSK and 16 QAM modulation schemes are used with rate 1/2 and rate 3/4 turbo codes.
5.5 Multiple Access Techniques: Let’s All Share the Same Channel A multiple access scheme (MAS) in a communications system offers the capability to share the same spectrum resources fairly and efficiently among many users desiring access to a shared communication medium. Different MASs are either being
5 Enabling Broadband Wireless Technologies
101
in use or in the research domain both in the terrestrial and satellite areas for providing capacity improvement in the system without significantly disturbing system’s performance. This section discusses wideband CDMA (W-CDMA) and orthogonal frequency division multiple access (OFDMA) techniques, which are currently in use in the terrestrial mobile multimedia systems. Also, the discussion includes carrier sense multiple access (CSMA) protocol, which is used in the Wi-Fi schemes (802.11 standards) as a multiple access method in the media access control (MAC) layer. Along with these multiple access techniques, different MASs are briefly addressed here with their merits and demerits. Combination of OFDM and CDMA techniques is also presented to complete the discussion on wideband multiple access schemes.
5.5.1 Frequency Division Multiple Access (FDMA) It is a method of combining multiple users on a given channel bandwidth using unique frequency segments. FDMA, by nature is a narrow band MAS. Here, an available frequency band (which is generally wide) is split into some smaller nonoverlapping orthogonal bands (or channels) and different information signals from different users are transmitted through these channels (Fig. 5.12). In this case, each transmitter or receiver for each user uses separate frequency (channel) for communications. Application: Advanced mobile phone systems (AMPS).
Users/Codes
Time
Frequency (Different Channels)
Fig. 5.12 FDMA scheme
Merits: In this narrow band system, the symbol duration is large compared to the average delay spread that results in low ISI. Since it is a continuous transmission
102
Quazi M. Rahman
scheme, system overhead in terms of bits is less compared to the time division multiple access (TDMA) scheme. It is also less complex compared to the other MASs. Demerits: There are some cost-related demerits on FDMA system which varies from system to system. For example, this system has higher cell-site system costs as compared to TDMA systems. For a band limited system, FDMA can accommodate only so many users without any expensive signal processing measures. This system is also performance limited by the nonlinear effects of the power amplifier and the stability of the system clock that generates different frequencies of interest.
5.5.2 Time Division Multiple Access (TDMA) In this method, multiple users using unique time segments on a given channel bandwidth are combined. In this case, a single-carrier frequency is shared between different transmitters (users) each of which is assigned with a nonoverlapping time slot (Fig. 5.13). Application: Global systems for mobile (GSM). Merits: Since the data transmission occurs in burst, the transmitter or receiver can be turned off when it is not in use and that results in low battery consumption. Due to the discontinuous nature of transmission, handoff process is simpler in TDMA system environment. Demerits: High synchronization overhead is required for TDMA-based system. The system performance is limited by the stability of the digital clock that generates different time slots of interest.
Users
Users Users Users Users Users Users 1 2 3 1 2 3 Frequency
Fig. 5.13 TDMA scheme
Time (Different Time Slots)
5 Enabling Broadband Wireless Technologies
103
5.5.3 Code Division Multiple Access (CDMA) It is a method of combining multiple users on a given channel bandwidth using unique spreading codes, or hopping patterns to distinguish any given user. In CDMA systems, several transmitters (users) simultaneously and asynchronously access a channel by modulating and spreading their narrowband information-bearing signals with preassigned wideband spreading code. This spreading code makes the system possible to multiplex several users in the same time and frequency domain (Fig. 5.14). Wideband CDMA (W-CDMA): Wideband CDMA follows the same principle of CDMA technique [28]. It gets its name for its wide bandwidth (e.g., 5 MHz) requirement. The readers are cautioned here, not to get mixed up with ultra wideband13 (UWB) technology. Merits: Multipath fading in the CDMA (or W-CDMA)-based system can be substantially reduced because the signal is spread over a large spectrum. If the spreadspectrum BW is greater than the coherence bandwidth (BW) of the channel, the inherent frequency diversity [29] will mitigate the effects of small-scale fading. On top of that the system can support multisignaling-rate services simultaneously with frequency reuse feature. In the satellite domain, where multiple signals from different satellites are linearly combined, W-CDMA (or CDMA) with universal frequency reuse and a Rake receiver is very efficient for soft handoff application [31, 32]. Demerits: In CDMA-based system a good synchronization protocol is necessary for the spreading codes to exhibit their mutual orthogonal properties. If the mobile
Frequency
Time
Users/Codes
Fig. 5.14 CDMA (W-CDMA) scheme 13
UWB is defined by Federal Communication Commission (FCC) as a radio system having −10 dB signal bandwidth greater than 500 MHz. To read more about UWB, the interested readers can see Chapter 28 of [30] and the references therein.
104
Quazi M. Rahman
terminals are not synchronized, the orthogonality between the spreading codes is compromised and the overall performance is severely degraded. The capacity of a CDMA system is not a single constant number; it depends on the locations of the users as well as on their numbers, and it is also a function of how low a SNR is deemed to be acceptable.
5.5.4 Orthogonal Frequency Division Multiple Access (OFDMA) In principle a single OFDM symbol in OFDMA method works in the same way as FDMA method with the only difference in the composition of the carrier frequencies, which are overlapping and orthogonal to each other for the OFDMA method. In other words OFDMA method is essentially equivalent to an OFDM system. In general, in OFDMA scheme, different subcarriers are assigned to distinct users, although in practice, more subcarriers can be allocated to the same user depending on the requested data rate. So at any time, maximum number of users an OFDMA scheme can support is equal to the number of subcarriers in the scheme. OFDMA scheme can also be defined as a combination of OFDM and TDMA protocols since in this case the users can be dynamically assigned different subcarriers (OFDM) at different time slots (TDMA) as shown in Fig. 5.15. OFDMA system was originally proposed in [33] for cable TV (CATV) networks [34]. Recently, it has become a part of the emerging IEEE 802.16 standard for wireless metropolitan networks (WMANs) [35] and is currently attracting vast research attention from both academia and industry as a promising candidate for next generation broadband wireless networks [36, 37]. One significant advantage of OFDMA over OFDM is its potential to reduce the transmit power and to relax the PAPR problem as mentioned in Section 5.2.3. Lower data rates and bursty data are handled much more efficiently in OFDMA than in single-user OFDM or with TDMA or CSMA (discussed later). The orthogonality among subcarriers guarantees intrinsic protection against multiple access interference (MAI) while the adoption of a dynamic subcarrier assignment strategy provides the system with high flexibility in resource management. Moreover, OFDMA enjoys all the merits inherited from OFDM technique. Although OFDMA technique provides the means for extended flexibility and multirate transmission, it requires precise time and frequency synchronization and thus calls for very high implementation complexity. Besides, it faces all the demerits, derived from OFDM, discussed in Section 5.2.3. To read more about these merits and demerits, the interested readers can check reference [37] and the references therein. OFDMA has recently been included in the IEEE 802.11 a/g wireless local area network (WLAN) standards and is also considered in the IEEE 802.15.3 standard for wireless personal area networks (WPANS) and in the evolving IEEE 802.16x WiMAX standard for broadband wireless access [36]. Adaptive OFDMA has recently been proposed for the downlink for the Third-Generation Partnership Project Long-Term Evolution (3GPP-LTE) or Evolved UTRA standardization effort [38, 39].
5 Enabling Broadband Wireless Technologies
105
UT 2
UT 3
UT 3
UT 4
UT1
UT 1
UT2
UT2
UT2
UT2
UT2
UT2
UT 1
UT 1
UT 1
UT 1
User Terminal (UT)
TS n
TS 2 SC 1
SC 2
SC 3
SC n
TS 1
Time Slot (TS)
Subcarrier Frequency (SC)
Fig. 5.15 Block diagram of an OFDMA system
Other than the basic MASs stated above there are some other versions of MASs, such as space division multiple access (SDMA) [40] and geographic MAS [41]. Combinations of different basic MASs are also available in the research areas, some of which are being considered for practical applications. These combinations include hybrid FDMA/CDMA (FCDMA) [42], hybrid direct sequence/frequency hopped multiple access (DS/FHMA) [43], time division CDMA (TCDMA) [40], and time division frequency hopping (TDFH) [44]. Currently some remarkable research work is going on with the combination of OFDM and CDMA techniques [45]. Below, these techniques are briefly discussed.
5.5.5 Combination of OFDM and CDMA Systems The combination of OFDM signaling and CDMA scheme has one major advantage in that it can lower the symbol rate in each subcarrier so that a longer symbol duration makes it easier to quasi-synchronize the transmission. To exploit this advantage and all the advantages of OFDM and CDMA systems, discussed earlier, the combinations of these two systems have been proposed in three different ways by different researchers. These are multi-tone CDMA (MT-CDMA) [46], multicarrier CDMA (MC-CDMA) [47–49], and multicarrier-direct sequence CDMA (MC-DS CDMA) [50]. A brief overview of these different schemes is given below.
106
Quazi M. Rahman 0t )
exp( j
Serial Data Stream Data Encoder
aj (t ) Serial To exp ( j Parallel Data exp(j Converter
1t )
Multiplexer
To Channel
N-1t )
(a)
frequency f1
f2
f3
f4
(b)
Fig. 5.16 MT-CDMA scheme: (a) The transmitter and (b) Power spectrum of the transmitted signal
5.5.5.1 MT-CDMA Scheme In this system the transmitter spreads the OFDM signals and sends them through the channel. Figure 5.16a shows the MT-CDMA transmitter of the jth user using PSK scheme. The transmitter spreads the serial to parallel (S/P) converted data stream after MT operation. Since, before spreading, it is nothing but the OFDM signals, the spectrum of each subcarrier prior to spreading operation can satisfy the orthogonality condition with the minimum frequency separation. After spreading, the resulting spectrum of each subcarrier no longer satisfies the orthogonality condition. When spreading is placed on top of the MT signal as proposed in [46], the main feature of the system is that for a constant BW, the ratio between the number of chips and the number of tones has to be constant. Hence, when the number of tones increases, the number of chips per symbol does as well. Figure 5.16b shows the power spectrum of the MT-CDMA transmitted signals for a number of tones Nt = 4.
5.5.5.2 MC-CDMA Scheme In this system the transmitter spreads the input symbol first and then converts the spread symbol to OFDM signals and sends them through the channel. In other words, a fraction of the symbol corresponding to a chip of the spreading code is transmitted through a different subcarrier. In this case, MC-CDMA transmitter spreads the original data stream over different subcarriers using a given spreading code in the frequency domain [47–49]. The main feature of the MC-CDMA system is that, as MC-CDMA spreads an information bit over many subcarriers, it can make use of information contained in good subcarriers to recover the original symbol in a deep frequency selective fading channel.
5 Enabling Broadband Wireless Technologies
107
5.5.5.3 MC-DS CDMA Scheme The multicarrier DS-CDMA transmitter spreads the S/P converted data streams using a given spreading code so that the resulting spectrum of each subcarrier can satisfy the orthogonality condition with the frequency separation [50]. This scheme was originally proposed for an uplink communication channel, because the introduction of OFDM signaling into DS-CDMA scheme is effective for the establishment of a quasi-synchronous channel.
5.5.6 Carrier Sense Multiple Access (CSMA) Protocol CSMA protocol improves QoS of a packet switched system by improving the transmission–reception efficiency. This is achieved by monitoring the channel before engaging in information interchange. In this protocol each terminal in the network verifies the absence of other traffic before transmitting information through the shared channel. Here, the terminal that wants to transmit information listens to the channel information in the network. If the channel is busy, the transmission is held off until later. On the other hand, if the channel remains free for a certain period of time (called DIFS for distributed inter-frame space) the terminal can transmit its information. The terminal transmits a “Ready To Send” (RTS) message containing information on the amount of data that it wishes to send and its transmission speed. The receiver (generally an access point in 802.11 standards) responds with a “Clear To Send” (CTS) message, and then the terminal starts sending information. When all the information sent by the terminal has been received, the receiver sends an acknowledgement (ACK) signal. All nearby terminals then wait for as long as it is necessary to transmit that amount of information at the declared speed. CSMA protocol is used by all the Wi-Fi schemes.
5.6 Diversity Techniques: Let’s Make a Better Use of the Channel Diversity is a family of techniques that reduces the disruptive effects of fading channel. In these techniques, several replicas of the same information signal transmitted over independently fading channels are supplied to the receiver. In this case, the probability of all the signal components, fading simultaneously, is reduced considerably. That is, if p is the probability that any one signal will fade below some critical value then p L is the probability that all L independently fading replicas of the same signal will fade below the critical value. When all these independently fading signals are combined, the technique is known as diversity combining. Figure 5.17 shows a block diagram of a simplest form of diversity combining technique.
108
Quazi M. Rahman Path 1 Rx
Output Input
Tx
Combiner
Path 2 Rx
Fig. 5.17 A simplified diversity (spatial) configuration
In general, a diversity system can be constructed if the following criteria are met: • A copy of the same signal is received over two or more different paths. • Each path fades differently. • Some type of diversity combining on the signal replicas, received over the paths, is possible.
5.6.1 Classifications of the Diversity Techniques There are several ways by which the receiver can be provided with L independently fading replicas of the same information-bearing signal. In frequency diversity the same information-bearing signal is transmitted on L carriers, where the separation between the successive carriers equals or exceeds the coherence bandwidth14 (BWcoh ) of the channel. In time diversity the same information-bearing signal is transmitted in L different time slots, where the separation between successive time slots equals or exceeds the coherence time15 (Tcoh ) of the channel. In spatial diversity (also known as antenna diversity) technique multiple antennas are used. These antennas are commonly used in the receiver section (in transmit diversity technique multiple transmitter antennas are used). These antennas are spaced sufficiently far apart so as to obtain signals that fade independently. Space diversity technique is one of the most popular forms of diversity used in wireless systems. There is one spatial diversity technique, known as transmit diversity, which uses multiple antennas in the transmitter end. This technique is also known as space–time coding (see the coding section) since this involves redundancy by transmitting the same signal using different antennas. It is known as MIMO system too as mentioned in the coding section. Currently transmit diversity is getting notable considerations [51] in the field of research due to its application in the 3G scenario.
14
The frequency band in which all the spectral components of the transmitted signal pass through a channel with equal gain and linear phase is known as coherence bandwidth of that channel. 15 The time period, during which the channel impulse response remains invariant, is known as coherence time of the channel.
5 Enabling Broadband Wireless Technologies
109
Another method of obtaining diversity is based on the use of a signal having a bandwidth much greater than the coherence bandwidth BWcoh of the channel. Such a signal with bandwidth Bs will resolve the multipath components and, thus, provide the receiver with several independently fading signal paths. The time resolution is 1/Bs . Consequently, with a multipath spread16 of Tm seconds, there are Tm Bs resolvable signal components. Since Tm = 1/BWcoh , the number of resolvable signal components may also be expressed as Bs /BWcoh . Thus, the use of a wideband signal may be viewed as just another method for obtaining frequency diversity of the order of L ≈ Bs /BWcoh . Sometimes it is called multipath frequency diversity. There are other diversity techniques such as angle-of-arrival diversity and polarization diversity. However, these techniques (e.g., see [52, 53]) have not been as widely used as the other ones described above.
5.6.2 Classifications of Diversity Combiners There are two basic types of diversity combiners, predetection type and postdetection type, which are discussed below. 5.6.2.1 Predetection diversity combiners A predetection diversity combiner cophases, weights, and combines all the signals received on the different branches before signal detection. Here, the cophasing function is a difficult task to implement. The most common predetection combiners are the selection combiner (SC), equal-gain combiner (EGC), and the maximal-ratio combiner (MRC). Recently, a minimum mean squared error (MMSE) diversity combiner specifically designed to combat ISI due to delay spread was proposed [54]. Selection diversity combining: This is the simplest diversity combining technique, where the branch output having the maximum instantaneous SNR, among L diversity branches, is selected for further processing and detection. Maximal ratio combining: Here, the diversity branch signals are cophased (phasealigned) and weighted according to their individual signal-to-noise power ratios and then summed. Equal gain combining: EGC is based on the same principle of operation as MRC except that all the branch weights are set to unity here. For detailed information on all the above combiners, the readers can see [7], [40], [55]. 5.6.2.2 Postdetection diversity combiners Postdetection diversity combiners weight and combine all diversity branches after signal detection and do not require the difficult to implement cophasing function. 16 Multipath spread of a channel is the range of values of excess time delay, over which power delay profile is essentially nonzero [40].
110
Quazi M. Rahman
Since the combiner structure can be much simpler, postdetection diversity is more attractive for mobile radio. All the predetection combining techniques can also be used for postdetection combining.
5.7 Challenges and Research Evidences This section addresses some challenges encountered in the broadband wireless domain while taking into account different modulation, coding, multiple access, and diversity schemes. For the next generation broadband wireless access, the researchers in the academia and industries are mainly focusing on OFDM, M-QAM (with M greater than 4), OFDMA, MIMO systems, turbo coding, AMC, and the cross combinations of these techniques. Some research evidences, based on the challenges encountered from these techniques, are presented here. OFDM has recently become a part of new emerging standards for broadband wireless access. In the physical layer, designing an OFDM-based system offers most challenging task when it comes down to synchronization issue, especially in the uplink scenario. In OFDMA-based system where multiple users are accessing the multimedia services, this challenge grows drastically. Besides, in multiple-user environment with OFDM, efficiently assigning subcarriers, data rates, and power levels to each user in both uplink and downlink is a challenging task. A comprehensive survey in this field can be found in [37] and in the references therein. The combination of OFDM technique and M-QAM is a promising method for high data rate applications such as multimedia. In the fading channel environment, this integration involves a challenging signal processing scenario [56, 57]. Here one of the challenges is to come up with an optimal constellation for the QAM technique both in terms of bit error rate performance and complexity. The interested readers can read [58], where a good tutorial on different QAM constellations along with respective error rate performance has been provided. This is worth mentioning that combination of OFDM and QAM is already in the application area for the IEEE 802.11-standard WLAN. The PAPR problem encountered in OFDM technique has been addressed in [15] where turbo-coded OFDM scheme has been used. Here, a selective-mapping (SLM) scheme has been proposed, which does not require the transmission of side information and thus it can reduce the PAPR in the turbo-coded OFDM system. The basic idea of SLM technique is to generate several OFDM symbols as candidates and then select the one with the lowest PAPR for actual transmission. Conventionally, the transmission of side information is needed so that the receiver can use the side information to tell which candidate is selected in the transmission. The authors in this paper have shown that the candidates of the proposed SLM are generated by a turbo encoder using various interleavers. The waiver of side information can avoid the degradation of error rate performance which results from the incorrect recovery of side information at the receiver in the conventional SLM OFDM system.
5 Enabling Broadband Wireless Technologies
111
Coded OFDM technique, where forward error correction scheme is applied to the OFDM signal, has been proposed in [59] for broadband wireless application to mitigate the problem of ISI and frequency selective channel. Currently many OFDM versions, such as vector OFDM, wideband OFDM, F-OFDM, MIMO OFMD, are present in the application area [60]. In this case OFDM needs standardization to enable its widespread use, encourage adoption, and thereby grow the market for the next generation broadband wireless systems. Recently, in the broadband wireless domain, AMC is being considered to be a negotiating technique between power and spectral efficiency of the system when the transmitted signals are facing different channel impairments. Although AMC scheme provides throughput and capacity enhancement of the communications system, it offers a number of challenges such as its sensitivity to measurement error and delay. These challenges along with system design issues and performance evaluation for adaptive schemes have been addressed in [61]. In [62] the authors have given an overview of the challenges and promises of link adaptation (LA) in future broadband wireless networks. This article reviews the fundamentals of adaptive modulation and coding techniques for MIMO broadband systems and illustrates their potential to provide significant capacity gains under ideal assumptions. The review concludes that the implementation of optimum LA is challenging due to practical limitations, but simulated performance of a realistic broadband wireless MIMO–OFDM-based system using LA is very encouraging. Turbo-coded AMC has been examined in [63] for the next generation mobile systems to observe the throughput gain. Here a method of generating soft information for higher order modulations based on the reuse of the turbo decoding circuitry is provided. It is shown that 3G style turbo coding can provide 0.5–4 dB link gain over 256-state convolutional codes, depending on the frame size, modulation, and channel. Here the link gains from channel coding do not directly translate into throughput gain for AMC, but still be expected to improve throughput significantly. The authors in [64] present a space–time turbo (iterative) equalization method for trellis-coded modulation (TCM) signals over broadband wireless channels. The equalizer proposed here consists of a broadband beamformer which processes antenna array measurements to shorten the observed channel impulse response, followed by a conventional scalar turbo equalizer. The proposed receiver structure is simulated for two-dimensional TCM signals such as 8-PSK and 16-QAM and the results indicate that the use of antenna arrays with only two or three elements allows a large decrease in the channel signal-to-noise ratio needed to achieve an acceptable BER. Adaptive OFDMA transmission scheme has been discussed in [65]. Here the authors address the challenge of packet scheduling and adaptive radio transmission for multiple users, via multiple antennas over frequency-selective wideband channels. This paper surveys techniques that adapt the transmission to the temporal, frequency, and spatial channel properties. In this paper, the authors provide a systematic overview of the design problems, such as the dimensioning of the allocated time–frequency resources, the influence of duplexing schemes, adaptation control
112
Quazi M. Rahman
issues for downlinks and uplinks, timing issues, and their relation to the required performance of channel predictors. The resource allocation problem and associated algorithms obtain an interesting twist if mutirate transmission is employed with adaptive modulation in OFDMA and TDMA and also by spreading gain adaptation in CDMA. To read more about these issues the readers can see [36]. An overview of space–time (ST) coding, space–frequency (SF) coding,17 and space–time–frequency (STF) coding for MIMO–OFDM systems is presented in [66]. In this paper the analytical results show that STF coding can achieve the maximum diversity gain in an end-to-end MIMO–OFDM system over broadband wireless channels. Furthermore, for OFDMA, the authors propose a multiuser SF coding scheme that can achieve the maximum diversity for each user while minimizing the interference introduced from all the other users. In [67] the authors explore various physical layer research challenges in broadband MIMO–OFDM wireless system design. The challenges include physical channel measurements and modeling, analog beam forming techniques using adaptive antenna arrays, space–time techniques for MIMO–OFDM, error control coding techniques, OFDM preamble and packet design, and signal processing algorithms used for performing time and frequency synchronization, channel estimation, and channel tracking in MIMO–OFDM systems. In [68] space–time-coded OFDM for the 4G cellular network has been proposed where the individual carriers of the OFDM techniques are modulated using BPSK, QPSK, and 16 QAM with coherent detection. The channel encoder consists of a half-rate convolutional encoder. Here the channel model considers wide range of possible delay spreads. The results show that the space–time block codes provide diversity gain and enhance the BER performance. At a high data rate, carrier acquisition and tracking of the incoming signal, that makes coherent detection possible in the receiver, is extremely difficult especially in strongly frequency-selective channels and mobile environments. To deal with this case, noncoherent space–frequency-coded MIMO–OFDM and the design criteria are proposed in [69], which shows that in the noncoherent case, the same maximum diversity can be achieved potentially just as the coherent case. To reduce the code matrix as encountered by the above proposed method, the authors in [70] have proposed a noncoherent sequence detection of differential space–frequency modulation (DSFM) MIMO–OFDM. In [71] the authors have considered postdetection diversity combining for OFDM–CDMA (MT-CDMA)-based system with noncoherent detection scheme. In this paper the performance of the system is analyzed in terms of bit error rate (BER) in the Ricean fading channel. The study considers indoor environment with DQPSK modulation. The analytical results, at high SNR, show that without tracking the phase of the received signal the system performs equally well when compared with its coherent counterpart. Recently, multiuser MIMO has generated a significant research interest [73] to take simultaneous advantage of mutiluser and spatial diversity. In this case, accurate 17
Space frequency coding consists of coding across antennas and OFDM subchannels [72].
5 Enabling Broadband Wireless Technologies
113
channel information is required which may not always be possible due to challenging channel scenarios. The performance gain in this case may be obtained with huge signal processing complexity. While wideband MIMO–OFDMA system is already providing significant spatial diversity gain in the presence of multiple users, the chance for multiuser MIMO to see the light of the application domain is slim, or more precisely, questionable. Besides all the above research evidences, investigations are in progress to fulfill the goal of achieving very high data rate with less complex data processing and robust system performance for the next generation broadband wireless communication system. In fulfilling this goal the researchers in the broadband wireless domain are continuously changing the intelligence, characteristics, and models of the existing technologies and at the same time they are providing new ideas to build new technologies. This is an ongoing process and all of us need to keep ourselves up to date with these technological advancements to be a part of this never ending process.
5.8 Summary With the increased market recognition for broadband wireless technology, users in both business and home environments are accelerating their demand for the next level of broadband functionality, utility, and convenience. This huge demand makes the ever growing telecommunications research sector to explore and improve the existing techniques for the broadband wireless domain. This chapter offers an overview on some of these physical layer techniques. Here, different modulation schemes, coding techniques, multiple access methods, and diversity techniques which are currently available in the application domain of broadband wireless communications schemes have been presented through a tutorial approach. Some challenges along with the corresponding research evidences have also been addressed in this chapter.
References 1. Bing, B. (2008). Next-Generation Broadband Networks. Proceedings of the IEEE Annual Communication Networks and Services Research Conference (CNSR) Halifax, NS, Canada. 2. Andrews, J. G., Ghosh, A., & Muhamed, R. (2007). Fundamental of WiMAX, Understanding Broadband Wireless Networking. Upper Saddle River, NJ: Prentice-Hall, Inc. 3. Simon, M., & Alouini, M. (2004). Digital Communication Over Fading Channels. 2nd ed. New York: John Wiley and Sons Inc. 4. Ramachandran, I., & Roy, S. (2007). Clear channel assessment in energy constrained wideband wireless networks. IEEE Transaction on Wireless Communications. 14(3), 70–78. 5. Burr, A. (2001). Modulation and Coding for Wireless Communications. Upper Saddle River, NJ: Prentice-Hall, Inc. 6. Giltin, R. D., Hayes, J. F., & Weinstein, S.B. (1992). Data Communications Principles. New York: Plenum Press.
114
Quazi M. Rahman
7. Proakis, J. G. (1995). Digital Communications. 3rd ed. New York: McGraw Hill Series. 8. Alasady, H., Ibnkahla M., & Rahman, Q. M. (2006). Symbol error rate calculation and data pre-distortion for 16-QAM transmission over nonlinear memoryless satellite channels. Wireless Communications and Mobile Computing. 8(2), 1530–8669. 9. Park, D. C., & Jeong, T. J. (2002). Complex-bilinear recurrent neural network for equalization of a digital satellite channel. IEEE Transactions on Neural Networks. 13(3), 711–725. 10. Costa, E., & Puoilin, S. (2002). M-QAM OFDM system performance in the presence of a nonlinear amplifier and phase noise. IEEE Transactions on Communications. 40(10), 101–109. 11. Shelswell, P. (1995). The COFDM modulation system: The heart of digital audio broadcasting. Electronics and Communications Engineering Journal. 7(3), 127–136. 12. Nee, R. V., & Prasad, R. (1999). OFDM Wireless Personal Communications. Artech House. 13. Cimini, L. J. (1985). Analysis and simulation of a digital mobile channel using orthogonal frequency division multiplexing. IEEE Transactions on Communications. 33(7), 665–675. 14. Li, Z., & Xia, X. (2008). PAPR reduction for repetition space-time-frequency coded MIMOOFDM systems using chu sequences. IEEE Transactions on Wirelesss Communications. 7(4), 1195–1202. 15. Tsai, Y., Deng, S., Chen, K., & Lin, M. (2008). Turbo coded OFDM for reducing PAPR and error rates. IEEE Transactions on Wirelesss Communications. 7(1), 84–89. 16. Nee, R. V., et al. (1999). New high-rate wireless LAN standards. IEEE Communications Magazine. 37(12), 82–88. 17. Smith, C., & Meyer, J. (2005). 3G Wireless with WiMAX and Wi-FI. New York: McGraw Hill Series. 18. Shanon, C. E. (1948). A mathematical theory of communication. Bell System Technical Journal. 27, 379–423, 623–657. 19. Ungerboeck, G. (1987). Trellis coded modulation with redundant signal sets, part I: introduction. IEEE Communications Magazine. 25(2), 5–11. 20. Huang, Q., Chan, S., Ping, L., & Ko, K. (2007). Performance of hybrid ARQ using trelliscoded modulation over rayleigh fading channel. IEEE Transactions on Vehicular Technology. 56(5), 2784–2790. 21. Stenbit, J. P. (1964). Tables of generators for bose-chadhuri codes. IEEE Transaction on Information Theory. 10(4), 390–391. 22. Fallager, R. G. (1968). Information Theory and Reliable Communication. New York: John Wiley and Sons Inc. 23. Lee, L. H. C. (1997). Convolutional Coding: Fundamentals and Applications. London: Artech House. 24. Sklar, B. (2001). Digital Communications, Fundamental and Applications. Upper Saddle River, NJ: Prentice-Hall, Inc. 25. Martinez, J. L., Weerakkody, W. A. R. J., Fernando, F., & Kondoz, A. M. (2008). Turbo trellis coded modulation for transform domain distributed video coding. Electronics Letters. 44(15), 899–900. 26. Odabasioglu, N., Ucan, O. N., & Hekim, Y. (2005). Multilevel turbo coded-continuous phase frequency shift keying (MLTC-CPFSK) over satellite channels in space communication. Proceedings of 2nd IEEE International Conference on Recent Advances in Space Technologies Istanbul, Turkey. 27. Hayes, J. K. (1968). Adaptive feedback communications. IEEE Transactions on Communications. 16(1), 29–34. 28. Viterbi, A. J. (1995). CDMA principles of Spread-Spectrum Communications. Reading, MA: Addison-Wesley. 29. Adachi, F., Sawahashi, M., & Suda, A. (1998). Wideband DS-CDMA for next generation mobile communication systems. IEEE Communications Magazine. 36(9), 56–69. 30. Ibnkahla, M., ed. (2004). Signal Processing for Mobile Communications Hand Book. Boca Raton, FL: CRC Press.
5 Enabling Broadband Wireless Technologies
115
31. Viterbi, A. J. et al. (1994). Soft handoff extends CDMA cell coverage and increases reverse link capacity. IEEE Journals on the Selected Areas in Communications. 12(8), 1281–1288. 32. Evans, J. V. (1998). Satellite systems for personal communications. Proceedings of the IEEE. 86(7), 1325–1341. 33. Sari, H., & Karama, G. (1998). Orthogonal frequency-division multiple access and its application to CATV networks. European Transactions on Communications. 9(6), 507–516. 34. ETSI EN standard 301–958 (2001). Interaction Channel for Digital Terrestrial Television (RCT) Incorporating Multiple Access OFDM. European Telecommunications Standards Institute 35. IEEE P802.16a/D3-2001 (Draft Amendment to IEEE P802.16a/D3-2001, 2002). IDraft Amendment to IEEE Standard for Local and Metropolitan Area Networks, Part 16: Air Interface for Fixed Broadband Wireless Access Systems-Amendment 2: Medium Access Control Modifications and Additional Physical Layer Specifications for 2–11 GHz. 36. Koutsopoulos, I., & Tassiulas, L. (2008). The impact of space division multiplexing on resource allocation: a unified treatment of TDMA, OFDMA and CDMA. IEEE Transactions on Communications. 56(2), 260–269. 37. Morelli, M., Kuo, C. C. J., & Pun, M. (2007). Synchronization techniques for orthogonal frequency division multiple access (OFDMA): a tutorial review. Proceedings of the IEEE. 95(7), 1394–1427. 38. GPP TS 36. 300 (2007). Evolved Universal Terrestial Radio Access (E-UTRA) and Evolved Universal Terrestial Radio Access Network (E-UTRAN); Overall Description: Stage 2. Relesase 8, V 8.0.0. 39. Ekstrom, H., Furuska, A., Karlsson, J., Meyer, M., Parkvall, S., Torsner, J., & Wahlquist, M. (2006). Technical solutions for the 3G long-term evolution. IEEE Communications Magazine. 44(3), 38–45. 40. Rappaport, T. S. (2002). Wireless Communications, Principles and Practice. Upper Saddle River, NJ: Prentice-Hall, Inc. 41. Hewlett Packard Co. (1997). Digital Modulation in Communications Systems – An Introduction, Application Note 1298. 42. Eng, T., & Milstein, L. B. (1993). Capacities of Hybrid FDMA/CDMA systems in multipath fading. IEEE MILCOM Conference Records. San Diego, CA. 43. Dixon, R. C.(1994). Spread Spectrum Systems with Commercial Applications. 3rd ed. New York: John Wiley and Sons Inc. 44. Skold, B. J., & Ugland, J. K. (1992). A comparison of CDMA and TDMA systems. Proceedings of the 42nd IEEE VTC. Denver, CO. 45. Rahman, Q. M., Sesay, A. B., & Hefnawi, M. (2004). Two-stage maximum likelihood estimation (TSMLE) For MT-CDMA signals. URASIP Journal on Wireless Communications and Networking. 2004(1), 55–66. 46. Vandendorpe, L. (1993). Multitone direct sequence CDMA system in an indoor wireless environment. Proceedings of IEEE First Symposium of Communications and Vehicular Technology. Delft, The Netherlands. 47. Fazel, K., & Papke, L.(1993). On the performance of convolutionally-coded CDMA/OFDM for mobile communication system. Proceedings of IEEE PIMRC. Yokohama, Japan. 48. Yee, N., Linnartz, J. P., & Fettweis, G. (1993). Multicarrier CDMA in indoor wireless radio networks. Proceedings of IEEE PIMRC. Yokohama, Japan. 49. Chouly, A., Brajal, A., & Jourdan, S. (1993). Orthogonal multicarrier techniques applied to direct sequence spread spectrum CDMA systems. Proceedings of IEEE GLOBECOM. Houston, TX. 50. DaSilva, V., & Sousa, E. S. (1993). Performance of orthogonal CDMA codes for quasisynchronous communication systems. Proceedings of IEEE ICUPC. Ottawa, Canada. 51. Kaiser, S. (2008). Rate adaptation with CDM versus adaptive coding for next generation OFDM systems. IEEE Vehicular Technology Conference. Marina Bay, Singapore. 52. Popovic, D., & Popovic, Z. (2002). Multibeam antennas with polarization and angle diversity. IEEE Transactions on Antennas and Propagation. 50(5), 651–657.
116
Quazi M. Rahman
53. Vaughan, R. G.(1990). Polarization diversity in mobile communications. IEEE Transactions on Vehicular Technologies. 39(3), 177–186. 54. Clark, M. V., Greenstein, L. J., &Shafi, M. (1992). MMSE diversity combining for wideband digital cellular radio. IEEE Transactions on Communications. 40(6), 1128–1135. 55. Jakes, W. C.(1974). Microwave Mobile Communications. New York: John Wiley and Sons Inc. 56. Dala, U. D., Kosta, Y. P., & Dasguta, K. S. (2007). Study of issues related to adaptive channel estimation in OFDM system-MWBA aspects. Third International Conference on Wireless Communication and Sensor Networks. India. 57. Ibnkahla, M., Rahman, Q. M., et al. (2004). Satellite mobile communications: technologies and challenges. Proceedings of the IEEE. 92(2), 312–339. 58. Al-Asady, H., Rahman, Q. M., & Ibnkahla, M. (2004). Signaling constellations for transmission over nonlinear channels. Chapter VII: Signal Processing for Mobile Communications Handbook. Boca Raton: CRC Press. 59. Thenmozhi, K., & Prithiviraj, V. (2007). Suitability of coded orthogonal frequency division multiplexing (COFDM) for multimedia data transmission in wireless telemedicine applications. International Conference on Computational Intelligence and Multimedia Applications. Sivakasi, India. 60. Vaughan-Nichols, S. J. (2002). OFDM: back to the wireless future. Computer. 35(12), 19–21. 61. Sampei, S., & Harada, H. (2007). System design issues and performance evaluations for adaptive modulation in new wireless access systems. Proceedings of the IEEE. 95(2), 2456–2471. 62. Catreus, S., Erceg, V., Gesbert, D., & Heath, R. W. (2007). Adaptive modulation and MIMO coding for broadband wireless data networks. IEEE Communications Magazine. 40(6), 108–115. 63. Classon, B., et al. (2002). Channel coding for 4G systems with adaptive modulation and coding. IEEE Wireless Communications. 9(2), 8–13. 64. Koca, M., & Levy, B. C. (2004). Turbo space-time equalization of TCM for broadband wireless channels. IEEE Transactions on Wireless Communications. 3(1), 50–59. 65. Sternad, M., Svensson, T., Ottosson, T., Svensson, A., & Brunstrom, A. (2007). Towards systems beyond 3G based on adaptive OFDMA transmission. Proceedings of the IEEE. 95(12), 2432–2455. 66. Wei, Z., Xia, X., & Letaief, K. B. (2007). Space-time/frequency coding for MIMO-OFDM in next generation broadband wireless systems. IEEE Wireless Communications. 14(3), 32–43. 67. Stuber, G. L., Barry, J. R., McLaughlin, S. W., Li, Y., Ingram, M. A., & Pratt, T. G. (2004). Broadband MIMO-OFDM wireless communications. Proceedings of the IEEE. 92(2), 271–294. 68. Doufexi, A., et al. (2002). COFDM performance evaluation in outdoor MIMO channel using space/polarization-time processing techniques. Electronics Letters. 38(25), 1720–1721. 69. Borgmann, M., & Bolcskei, H. (2005). Noncoherent space-frequency coded MIMO-OFDM. IEEE Journal on Selected Areas in Communications. 23(9), 1799–1810. 70. Wang, X., Li, Y., & Wei, J. (2007). Noncoherent sequence detection of differential spacefrequency modulation MIMO-OFDM. IEEE 8th Workshop on Signal Processing Advances in Wireless Communications. Helsinki, Finland. 71. Rahman, Q. M., & Sesay, A. B. (2003). Non-coherent MT-CDMA system with post-detection diversity combining. Canadian Journal on Electrical and Computer Engineering. 23(2), 81–88. 72. Lee, K. F., & Williams, D. B. (2000). A space-frequency transmitter diversity technique for OFDM systems. Proceedings of IEEE Global Commununications Conference. San Francisco, CA. 73. Goldsmith, A., Jafar, S., Jindal, N., & Vishawnath, S. (2003). Capacity limits of MIMO channels. CIEEE Journal on Selected areas in Communications. 21(5), 684–702.
Chapter 6
WiMAX Networks Abdou R. Ahmed, Xiaofeng Bai, and Abdallah Shami
Abstract WiMAX (worldwide interoperability for microwave access) network is the wireless metropolitan area broadband access solution built in compliance with the IEEE 802.16 standard. This network supports high data rates with extended service coverage for up to several tens of kilometers. It can also fill the gap between 3G cellular network and local area wireless network such as WiFi. Multiservice-oriented quality of service (QoS) and vehicular speed mobility are fully accommodated in WiMAX standardization. The standard defines a mandatory point-to-multipoint (PMP) operation mode along with an optional mesh mode for enhanced network scalability and user mobility. In providing a technical guideline for this emerging wireless technology, this chapter proceeds by addressing several fundamental topics on WiMAX, including (1) basic knowledge, as per IEEE 802.16 specification, on the operating radio frequency bands, physical layer interfaces, and media access control (MAC) sublayers of a WiMAX system; (2) MAC services via different physical interfaces, QoS mechanisms, and resource management issues for PMP operation mode; (3) mesh mode-specific scheduling, routing, and QoS provisioning in WiMAX networks; and (4) mobility in WiMAX networks, with main focus on handover techniques between both homogeneous and heterogeneous network interfaces.
Abdou R. Ahmed Department of Electrical and Computer Engineering, The University of Western Ontario, London, ON, Canada N6A 5B8, e-mail:
[email protected] Xiaofeng Bai Department of Electrical and Computer Engineering, The University of Western Ontario, London, ON, Canada N6A 5B8, e-mail:
[email protected] Abdallah Shami Department of Electrical and Computer Engineering, The University of Western Ontario, London, ON, Canada N6A 5B8, e-mail:
[email protected]
A. Shami et al. (eds.), Broadband Access Networks, Optical Networks, c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-92131-0 6,
117
118
Abdou R. Ahmed, Xiaofeng Bai, and Abdallah Shami
6.1 Introduction In the last few years, there have been ever-increasing demands for high-speed Internet access and multimedia service which led to extensive need for last mile broadband access. Broadband wireless access is emerging as a viable solution in response to such new market trend. The IEEE 802.16 group has devised a standard set for developing interoperable broadband wireless access systems [1], known as WiMAX. The first version of this specification is 802.16-2001 [2], which was later superseded by the next IEEE standard 802.16-2004 [3] and its amendment [4]. The last version of the standard, IEEE 802.16e [5], was released in 2006 to provide specifications for mobility support. The WiMAX Forum is an industry association modeled with the purpose of promoting and certifying the interoperability of broadband wireless products based on the IEEE 802.16 specifications. Although WiMAX is not a technology, but rather a certification mark or stamp of approval given to equipment that meets certain conformity and interoperability tests for the IEEE 802.16 family of standards, its name has been adopted in popular usage to denote the technologies behind it. This is likely due to the difficulty of using terms like IEEE 802.16 in common speech and writing. In this chapter the term WiMAX and IEEE 802.16 are used interchangeably. As WiMAX operates over multiple radio frequency bands, it maximizes the technology’s adaptability to select frequencies that avoid interfering with other wireless applications. In addition, WiMAX’s transmission range and data rate vary significantly depending on the frequency bands and implementation being used. This means different frequencies can be used depending on the distance and speed required for a specific transmission. As wireless backhaul, WiMAX provides long transmission range – up to 31 miles (50 km) – with increased transmission power and the use of directional antennas which produce focused signals. To prevent serving too many customers and effectively reduce each user’s bandwidth availability, providers will want in practice to serve no more than 500 subscribers per 802.16 base station. Thus, each base station roughly serves an area within a 10-mile radius. WiMAX technology achieves high data rates in part via orthogonal frequency division multiplexing (OFDM). OFDM increases bandwidth and data capacity by splitting broad channels into multiple narrow band channels – each using a different frequency – that can then carry different parts of a message simultaneously. The channels are spaced very close together but avoid interference because neighboring channels are orthogonal to one another and thus lead to non-interfered reception [6]. Theoretically, a single channel in IEEE 802.16 can provide data rates of 32–130 Mbps depending on the channel frequency width and applied modulation technique. Providers may use multiple IEEE 802.16 channels for a single transmission to provide data rate of up to 350 Mbps. A single “sector” of a 802.16 base station – where sector is defined as a single transmit/receive radio pair at the base station – provides sufficient bandwidth to simultaneously support more than 60 businesses with T1-level connectivity and hundreds of homes with DSL-rate connectivity using 20 MHz of channel bandwidth.
6 WiMAX Networks
119
This chapter is organized as follows: in the remainder of this section we introduce the operation, frequency bands, and reference model defined by the 802.16 standard. In Section 6.2 we discuss the PHY and MAC layer specifications for PMP WiMAX networks. Section 6.3 focuses on the mesh operation mode of WiMAX. Section 6.4 discusses mobility in WiMAX networks and Section 6.5 concludes this chapter.
6.1.1 Scope of the Standard The general structure of a WiMAX network consists of a base station (BS) and one or more associated subscriber stations (SS) being serviced in the coverage area of the BS. Figure 6.1 shows a conceptual illustration of such deployment. The BS is responsible for connecting to the external core network and provides diverse access services, including bandwidth-intensive data, real-time voice, and broadcast video steams, to each served SS via the established wireless interface. The SSs may be either fixed or in mobility up to vehicular level speed. The network can be operated in point-to-multipoint (PMP) mode or in mesh mode, both of which are supported by the standard. In PMP mode, each SS communicates with the BS through a direct one-hop link to the BS, while in mesh mode traffic can be routed through other SSs and can occur directly between SSs. The standard provides a common media access control (MAC) layer specification along with multiple physical layer (PHY) air interfaces to accommodate diverse application environments. These PHY specifications include single carrier (WirelessMAN-SC) for frequency bands 10–66 GHz, single carrier (WirelessMAN-SCa) for frequency bands 2–11 GHz, OFDMbased PHY (WirelessMAN-OFDM), and orthogonal frequency division multiple access (OFDMA)-based PHY (WirelessMAN-OFDMA) for frequency bands 2–11
PSTN SS
M
ilit ob
CATV
y
ISP
PMP Mode
Residence
LAN BS Mesh Mode
Business SS SS
Fig. 6.1 Illustration of an IEEE 802.16 system
120
Abdou R. Ahmed, Xiaofeng Bai, and Abdallah Shami
GHz. In addition to specifications for the licensed bands of 2–11 GHz, the wireless high-speed unlicensed metropolitan area networks (WirelessHUMAN) term addresses specific components for the dynamic frequency selection (DFS) mechanism required for operation in license exempt bands in this frequency range.
6.1.2 Frequency Bands of Operation Among the defined PHY specifications, the 10–66 GHz bands provides a physical environment where line-of-sight (LOS) is required and multipath is negligible. This environment provides high data rate of up to 120 Mb/s and is well suited for PMP operation mode serving applications in fixed locations. These frequency bands are also useful to provide long distance backhaul services to wireless hotspots or to wirelessly connect the BS with the external core network. Frequencies below 11 GHz provide a physical environment where LOS is not necessary and multipath impairments may be prominent. The ability to support non-line-of-sight (NLOS) scenarios requires additional PHY functionality, such as the support of advanced power management techniques, interference mitigation/coexistence, and multiple antennas. PHY interfaces specified for 2–11 GHz bands therefore must all support NLOS transmissions.
6.1.3 Reference Model of the Standard Figure 6.2 illustrates the reference model used in the IEEE 802.16 standard. Only MAC and PHY specifications are defined in the standard, while higher layer protocols, as well as the network management plane, are not in the scope of this standard. As shown in the figure, the MAC contains three sublayers: the service-specific convergence sublayer (CS), the MAC common part sublayer (CPS), and the security sublayer. The CS provides any transformation or mapping of external network data, received through the CS service access point (SAP), into MAC service data units (SDUs) received by the MAC CPS through the MAC SAP. This includes classifying external network SDUs and associating them to the proper MAC service flow and connection identifiers. It may also include such functions as payload header suppression (PHS) to reduce overhead. Respectively, two CS specifications for asynchronous transfer mode (ATM) and packet-based protocols are defined in the standard. The MAC CPS provides the core MAC functionality of system access, bandwidth allocation, connection management, and quality-of-service (QoS) control. The MAC CPS receives data from various CSs, through the MAC SAP, and is classified to particular MAC connections. The security sublayer resides at the bottom of the MAC layer and provides authentication, secure key exchange, and encryption. Below the MAC security
6 WiMAX Networks
121
Scope of standard CS SAP Management Entity Service-Specific Convergence Sublayer
MAC SAP MAC Common Part Sublayer (MAC CPS) Management Entity Security Sublayer
MAC Common Part Sublayer Security Sublayer
PHY
PHY SAP PHY Layer (PHY)
Data / Control Plane
Network Management System
MAC
Service-Specific Convergence Sublayer (CS)
Management Entity PYH Management Plane
c 2004 IEEE.) Fig. 6.2 Reference model of the standard. (after [3].
sublayer and interfaced via the PHY SAP is the PHY. The multiple PHY specifications supported by the standard are detailed below.
6.1.4 WirelessMAN PHY Specifications The PHY layer operation is frame based and supports both time division duplex (TDD) and frequency division duplex (FDD) configurations. Both setups use burst transmission format where the burst profiling for each SS is adaptive and may change frame by frame. In this section, the discussion is based on TDD operation in PMP mode for different PHY specifications.
6.1.4.1 WirelessMAN-SC and WirelessMAN-SCa PHY These two PHY specifications both use single radio carrier for transmission. The WirelessMAN-SC PHY is defined for the 10–66 GHz frequency licensed bands, while the WirelessMAN-SCa PHY is defined for the 2–11 GHz frequency bands; both are licensed and license-exempt. To allow flexible spectrum usage, both TDD and FDD configurations are supported in these specifications. The downlink channel operates by TDM, with the information for each SS multiplexed onto a single
122
Abdou R. Ahmed, Xiaofeng Bai, and Abdallah Shami
stream of data and received by all SSs within the same sector. The MAC at each SS listens to the downlink and only decodes frames addressed to this SS. The uplink is based on a combination of TDMA and demand assigned multiple access (DAMA). Particularly, the uplink channel is divided into a number of time slots. The number of time slots assigned for various uses (registration, contention, guard, or user traffic) is controlled by the MAC at the BS and may vary over time for optimal performance. Each uplink burst is designed to carry variable-length MAC PDUs. The transmitter randomizes the incoming data; forward error correction (FEC) encodes it, and maps the coded bits to a QPSK, 16-QAM, or 64-QAM signal constellation, where 256QAM is also allowed for WirelessMAN-SCa PHY. Compared with WirelessMANSC, the WirelessMAN-SCa PHY intends to provide more robust communication in the NLOS environments and therefore applies more complex mechanisms for randomization, FEC encoding, and interleaving.
6.1.4.2 WirelessMAN-OFDM PHY The WirelessMAN-OFDM PHY is based on OFDM modulation and designed for NLOS operation in the frequency bands below 11 GHz. This specification uses multicarrier in the available channel bandwidth for simultaneous transmission of paralleled bits based on OFDM technique. The channel is split into multiple subcarriers that are orthogonal to each other within the duration of an OFDM symbol. This extends the symbol duration of original bit stream that is otherwise transmitted by single carrier to N times longer in the OFDM symbol. Therefore, intersymbol interference due to multipath propagation is considerably alleviated. The OFDM PHY is based on a fast fourier transform (FFT) of size 256 and hence generates 256 subcarriers. Among these subcarriers, 8 subcarriers are used for pilot tones, 55 subcarriers are guard carriers, which along with the direct current (DC) subcarrier, are not used for transmission of information. The remaining 192 subcarriers are used for data transmission.
6.1.4.3 WirelessMAN-OFDMA PHY The WirelessMAN-OFDMA PHY is based on OFDM modulation and is designed for NLOS operation in the frequency bands below 11 GHz. This specification employs a larger FFT space of 2048 and 4096 carriers, which is further divided into subchannels. Analog to the OFDM PHY, there are three types of OFDMA subcarriers: data subcarriers for data transmission, pilot subcarriers for various estimation and synchronization purposes, and the null subcarriers for guard bands. A combination of pilot and data subcarriers constitutes a subchannel. The standard supports five types of subcarrier allocation schemes to form subchannels as follows: partial usage of subchannels (PUSC) on uplink and downlink; optional partial usage of subchannels (OPUSC) on uplink; full usage of subchannels (FUSC) on downlink;
6 WiMAX Networks
123
optional full usage of subchannels (OFUSC) on downlink and adaptive modulation; and coding (AMC) on uplink and downlink.
6.1.5 MAC Sublayers 6.1.5.1 Service-Specific Convergence Sublayer The service-specific CS resides on top of the MAC common part sublayer and utilizes, via the MAC SAP, the services provided by the MAC CPS. The CS performs the following functions: 1. 2. 3. 4. 5.
accepting higher layer PDUs from the higher layer; performing classification of higher layer PDUs; processing as necessary the higher layer PDUs based on the classification; delivering CS PDUs to the appropriate MAC SAP; and receiving CS PDUs from the peer entity.
In performing the above functionalities, classification and payload header suppression (PHS) are defined at the sending entity, while reconstruction is conduced at the receiving entity. Classification provides transformation or mapping of external network data received through the CS SAP. This includes processing received upperlayer SDUs and associating them with the proper connection identified by a particular CID, as shown in Fig. 6.3. Since the QoS and category of service parameters for the connection are set at connection establishment, classification guarantees the correct handling of traffic by the MAC. A classifier is a set of matching criteria applied to each SDU received by the CS SAP. If a SDU matches the specified
Upper Layer Entity (e.g., bridge, router, host)
Upper Layer Entity (e.g., bridge, router, host)
SDU
SDU
SAP
SAP CID 1 CID 2 ......
Downlink Classifier
Reconstitution (e.g., undo PHS)
CID n {SDU, CID, ...} SAP
{SDU, CID, ...} BS
802.16 MAC CPS
c 2004 IEEE.) Fig. 6.3 Classification and CID mapping. (After [3].
SAP
SS
802.16 MAC CPS
124
Abdou R. Ahmed, Xiaofeng Bai, and Abdallah Shami
matching criteria, it is then delivered to the appropriate SAP for delivery on the connection defined by the CID in the classifier. PHS, following classification, suppresses a repetitive portion of the payload header of the higher layer in the MAC SDU. This suppression is conducted by the sending entity and restored by the receiving entity for the sake of saving bandwidth. After classification and PHS, the SDU is delivered to the corresponding MAC CPS SAP. At the receiving CS entity, on the other side, the suppressed header is reconstructed before it is handed over to the higher layer protocol. The standard provides specifications for ATM CS and Packet CS, both are optional. ATM CS: The ATM CS is specifically defined to support the convergence of PDUs generated by the ATM layer protocols of an ATM network. Packet CS: Packet CS is designed for serving all packet-based protocols such as Internet protocol (IP), point-to-point protocol (PPP), and IEEE 802.3 (Ethernet).
6.1.5.2 MAC Common Part Sublayer Although the PHY layer specifications are differentiated by the spectrum of usage, the standard is designed to evolve as a set of air interfaces based on a common structure of MAC protocol with minor PHY-specific mechanisms for different PHY interface constraints.
MAC PDU Format Figure 6.4 illustrates the general form of a MAC PDU. Each PDU begins with a fixed length generic MAC header. The generic header may be followed by the payload portion of the MAC PDU, which consists of zero or more subheaders, MAC SDU, and/or fragments of the MAC SDU. The payload information may vary in length and therefore, the MAC does not require the knowledge of formats or bit patterns of various higher layer messages. CRC is optional for a MAC PDU transmitted with SC PHY, but is mandatory for SCa, OFDM, and OFDMA PHY layers. The standard defines two MAC header formats. One is the generic MAC header that begins each MAC PDU; the other is the bandwidth request MAC header used to
MAC subheaders Generic MAC header
Payload (MAC SDUs, fragments)
variable length Fig. 6.4 Classification and CID mapping
CRC
6 WiMAX Networks
125
form a stand-alone bandwidth request MAC PDU. Moreover, five types of subheader formats are defined for various purposes: 1. Fragmentation subheader – used to indicate the fragmentation state of the payload and sequence number of the current SDU fragment. 2. Grant Management subheader – used by SS to convey bandwidth management needs to the BS. 3. Packing subheader – used to indicate the case where multiple SDUs are packed into a single MAC PDU. 4. Mesh subheader – used to indicate mesh operation mode. 5. FAST-FEEDBACK allocation subheader – used to transmit PHY-related information that requires fast response from the SS.
6.2 Point-to-MultiPoint (PMP) WiMAX Networks 6.2.1 MAC Frame Structure The IEEE 802.16 PHY operation is frame based and supports both time division duplex (TDD) and frequency division duplex (FDD) configurations. A transmission frame is defined as a fixed time duration in which both the downlink and uplink transmissions complete one round. Both setups use burst transmission format where the burst profiling for each SS is adaptive and may change frame by frame. A MAC frame structure using these operations varies from TDD mode to FDD mode and is PHY layer specific. Discussion in this section is based on TDD operation in PMP mode for different PHY specifications. Details on mesh mode operation of WiMAX networks are organized in Section 6.3.
6.2.1.1 MAC Frame of WirelessMAN-SC and WirelessMAN-SCa PHY Layers With different frequency bands of operation, the MAC frame structure for Wireless MAN-SC and WirelessMAN-SC PHY interfaces take similar formats. A frame consists of a downlink (DL) subframe followed by an uplink (UL) subframe. The DL subframe begins with a frame start preamble used for synchronization and equalization by the PHY, followed by a frame control section. The frame control section is broadcast to all SSs and contains downlink MAP (DL-MAP) and uplink MAP (UL-MAP) messages that define the transmission burst profiles, including modulation and coding schemes, as well as relevant timing information for the following DL and UL transmissions, respectively. Following the frame control section is the downlink data destined to individual SSs. The downlink data is grouped into several transmission bursts using time division multiplexing (TDM). Downlink data bursts are broadcast by the BS to all SSs. The MAC in each SS listens to the downlink channel and looks for the MAC headers indicating data for this SS. Downlink
126
Abdou R. Ahmed, Xiaofeng Bai, and Abdallah Shami
transmission bursts are differentiated by their applied downlink interval usage code (DIUC), which represents a certain set of modulation and coding scheme for transmission. Downlink bursts are ordered in descending robustness. Particularly, the burst mapped into quadrature phase-shift keying (QPSK) constellation appears first, followed by bursts mapped into 16-quadrature amplitude modulation (16-QAM), and then 64-QAM (optional) constellations, with additional 256-QAM constellation only applicable to SCa PHY. This prevents SSs subjected to worse link conditions from losing synchronization before receiving their downlink data, as these SSs are only capable of demodulating and decoding data bursts with better robustness. The UL subframe follows the DL subframe. It may, under discretion of the BS, start with contention-based initial ranging and/or bandwidth request intervals. The initial ranging interval is used for new SSs to be registered into the system, while the bandwidth request contention interval is designed for connections carrying nonreal-time applications to inform the BS of their bandwidth needs. Collisions in both contention-based intervals are resolved by binary exponential back-off algorithms. The uplink data transmission is organized in TDMA fashion where the uplink bursts are differentiated by sending SSs. Each scheduled SS transmits into the UL during its granted window using burst profile associated with the uplink interval usage code (UIUC) that is informed by the UL-MAP message in the frame control section. Figure 6.5 illustrates the MAC frame format in TDD operation mode for the WirelessMAN-SC PHY. In order for the frame structure to perform properly, some functional overheads are introduced in the standard: 1. Transmit/receive transition gap (TTG) – It allows time for the BS to switch from transmit to receive mode, and the SS to switch from receive to transmit mode. 2. Receive/transmit transition gap (RTG) – It has the reverse functionality of TTG. 3. SS transition gap (SSTG) – It separates transmissions from different SSs during the UL subframe. Each SSTG consists of a gap allowing the previous burst to ramp down, followed by a preamble allowing the BS to synchronize to the new SS. In addition to the above-mentioned functionalities, the TTG should allow the last symbol of downlink transmission to propagate from BS to the most remote SS that is scheduled for transmission in the subsequent UL subframe. Moreover, the RTG should allow the last symbol of uplink transmission to propagate from source SS Downlink Subframe Frame start preamble
TTG
Uplink Subframe
Initial Ranging Window
RTG
SSTG
Frame Control TDM with DIUC a
DL-MAP
TDM with DIUC b
Scheduled Data of SS 1
TDM with DIUC c
Bandwidth Request Contention Window
UL-MAP Access burst
Collision Frame
Fig. 6.5 TDD MAC frame format of the WirelessMAN-SC PHY
Scheduled Data of SS 2
... Gap
Scheduled Data of SS n
Preamble
6 WiMAX Networks
127 Minimum TTG
TDM Portions DIUC a
DIUC b
Minimum RTG DIUC a
DIUC c
BS
SS n
SS Data SS Data
...
SS 1
t BS Processing Time
SS k (The most remote SS)
SS Data Uplink Subframe N
Downlink Subframe N
Frame N
Frame N+1
Fig. 6.6 Illustration of TTG and RTG
to BS and the processing time for BS to generate the next frame. These points are illustrated in Fig. 6.6. 6.2.1.2 MAC Frame of WirelessMAN-OFDM PHY The OFDM PHY also performs frame-based transmission consisting of a DL subframe and an UL subframe. Slightly different from the MAC frame for SC and SCa PHY, in the MAC frame of OFDM PHY, following a long preamble, is a frame control header (FCH) burst, which is one OFDM symbol long and contains DL Frame Pr e f i x (DLFP) to specify burst profile and the length of the downlink bursts immediately following the FCH. The first DL burst following the FCH should start from broadcast messages including DL-MAP and UL-MAP, if present, followed by regular MAC PDUs. Each downlink burst in the DL subframe consists of an integer number of OFDM symbols and is transmitted with a different burst profile. To form an integer number of OFDM symbols, in the DL or UL, unused bytes in the burst payload may be padded by the bytes 0xFF. Figure 6.7 illustrates the MAC frame structure for OFDM PHY. Downlink Subframe Preamble
TTG
FCH DL burst #1
DLFP
DL burst #2
Uplink Subframe
Initial Ranging Window
SSTG
UL burst from SS #1
DL burst #3
Broadcast Messages Regular MAC PDUs
Bandwidth Request Contention Window
RTG UL burst from SS #2
MAC PDU -1
...
...
MAC PDU -n
UL burst from SS #n
Pad
Frame
Fig. 6.7 TDD MAC frame format of the WirelessMAN-OFDM PHY
6.2.1.3 MAC Frame of WirelessMAN-OFDMA PHY For OFDMA PHY, the MAC data are mapped to an OFDMA data region for downlink and uplink transmissions with the minimum data allocation unit as an OFDMA
128
Abdou R. Ahmed, Xiaofeng Bai, and Abdallah Shami
slot. A slot requires both a time and a subchannel dimension for complete definition and varies for uplink and downlink transmissions. The data mapping follows the algorithms defined as follows: For DL: 1. Segment the data into blocks sized to fit into one OFDMA slot. 2. Each slot spans one or more subchannels in the subchannel axis and two OFDMA symbols in the time axis. Map the slots such that the lowest numbered slot occupies the lowest numbered subchannel in the lowest numbered OFDMA symbol. 3. Continue the mapping such that the OFDMA symbol index is increased. When edge of data region is reached, continue the mapping from the lowest numbered OFDMA symbol in the next subchannel. For UL: the mapping is similar to the mapping in DL, except that each slot spans three OFDMA symbols in the time axis. After such data mapping, the TDD MAC frame for OFDMA PHY is illustrated in Fig. 6.8. The frame structure is built from transmissions by the BS and SSs. Each frame in the downlink begins with a preamble followed by a DL transmission period and an UL transmission period. The first two subchannels in the first data symbol of the DL carry the FCH that contains the DLFP and specifies the length of the DL-MAP message immediately following the DLFP. Since the transmissions from SSs are separated by subchannels, all SSs transmit simultaneously in the UL. OFDMA symbol number k
k+1
S S+1
k+3
k+5
k+7
FCH
k+9 k+11 k+13 DL slot
t
k+15
k+17 k+20 k+23 k+26 UL slot UL burst #1
subchannel logical number
DL burst #3
DL burst #1 DL-MAP
Preamble
UL burst #2
UL burst #3
DL burst #4
DL burst #2
UL burst #4 DL burst #5 UL burst #5 Ranging subchannel
S+L DL period
UL period TTG
RTG
frame
c 2004 IEEE.) Fig. 6.8 TDD MAC frame format of the WirelessMAN-OFDMA PHY. (After [3].
6 WiMAX Networks
129
6.2.2 Service Flow and Connection The operation of MAC services is connection oriented. A connection is defined as a unidirectional mapping between BS and SS MAC peers for the purpose of transporting a service flow’s traffic. A service flow is a unidirectional flow of MAC PDUs with predefined QoS parameters. Supported service flows (available QoS setting options) are provisioned by the BS when a SS is registered into the network. When a service flow is activated, a connection is then established and mapped to the corresponding service flow. The QoS parameters defined for the service flow is therefore implicitly provided by the connection’s unique connection identifier (CID). Note, multiple upper layer applications requiring the same QoS settings may be aggregated into one service flow and carried by a single connection. For example, it is possible for several users in the same apartment building to share a common connection for web browsing service.
6.2.3 Service Classification In order to accommodate applications with different service requirements, the standard defines five types of MAC scheduling service. Service offered by each connection is categorized into one of these five service types, depending on the property of applications carried by this connection. These scheduling services are briefly described below: 1. Unsolicited grant service (UGS) – This service type is designed to support realtime data streams consisting of fixed size data packets issued at periodic intervals, such as T1/E1 and voice over IP without silence suppression. The key QoS parameters relevant to this service type are reserved traffic rate, maximum latency, and tolerated jitter (delay variation). 2. Real-time polling service (rtPS) – This service type is designed to support realtime data streams consisting of variable-sized data packets that are issued at periodic intervals, such as moving pictures experts group (MPEG) video. The key QoS parameters relevant to this service type are minimum reserved traffic rate, maximum sustained traffic rate, and maximum latency. 3. Extended real-time polling service (ertPS) – This service type was added by the 802.16e amendment and it builds on the efficiency of both UGS and rtPS. In ertPS, the BS provides unicast grants in an unsolicited manner like in UGS. However, whereas UGS allocations are fixed in size, ertPS allocations are dynamic. The ertPS is suitable for applications that need variable bit rate assigned at fixed interval period, e.g., voice over IP (VoIP) with silence suppression. 4. Non-real-time polling service (nrtPS) – This service type is designed to support delay-tolerant data streams consisting of variable-sized data packets for which a minimum data rate is required, such as FTP applications. The key QoS
130
Abdou R. Ahmed, Xiaofeng Bai, and Abdallah Shami
parameters relevant to this service type are minimum reserved traffic rate and maximum sustained traffic rate. 5. Best Effort (BE) – This service type is designed to support data streams for which no minimum service level is required and therefore may be handled on a spaceavailable basis. The key QoS parameter relevant to this service type is only maximum sustained traffic rate. This parameter upper-bounds the received service rate of a BE connection. Finally, BE connections can request bandwidth both in contention mode and in contention-free mode.
6.2.4 Bandwidth Request and Resource Allocation Each SS can request bandwidth for one of its connections by sending a stand-along MAC header (6 bytes) or by inserting a grant management subheader (2 bytes) into a MAC PDU, i.e., piggyback request. The bandwidth can be granted by the BS on per-connection (GPC) or per-SS (GPSS) basis. For a 10–66 GHz system, GPSS is the mandatory granting mode where bandwidth is granted by the BS to a SS as an aggregate of grants in response to per-connection requests from the SS. Therefore, the SS has the right to decide the ultimate allocation of the aggregated grant assigned by the BS. The per-connection bandwidth request allows the BS to allocate resources among SSs in a QoS-aware manner. Namely, the individual QoS parameter settings of each connection can be taken into consideration by the BS when the resource allocation is performed among different SSs. For example, connections with stringent QoS requirements can be better serviced when the contention for radio resource arises by appropriately granting the corresponding SSs. The per-SS granting mechanism, on the other hand, allows the SS to further adjust the usage of the granted bandwidth, hence to compensate the inherent drawback of polling operation due to information delay. Also, the per-SS granting reduces signaling overhead in the downlink, without degrading system performance. Several proposals on this topic exist in the literature. For example, Wongthavarawat and Ganz [7] introduced a packet scheduling algorithm for QoS support in the 802.16 systems, where fixed allocation, earliest deadline first (EDF), weighted fair queuing (WFQ), and equal sharing schemes are applied to the BS side resource allocation for connections of different service types. Moreover, in [8] we have proposed a robust QoS control design for PMP mode WiMAX implementation, where both BS side resource allocation and SS side transmission scheduling are suggested in detail. The principle of this design is to (1) distribute some BS functionalities for radio resource management into each SS thereby to reduce signaling overhead in the network and (2) enhance the robustness of QoS control by applying MAC-PHY cross-layer resource allocation.
6 WiMAX Networks
131
6.3 WiMAX Mesh Mode 6.3.1 General Properties of WiMAX Mesh In mesh mode the role of the BS, connecting the mesh network to the external backhaul link, e.g., Internet, can be performed by one or several nodes. Node(s) that perform the BS role are called mesh BS while the other nodes are called mesh SS. In general, stations of the mesh network are termed nodes. Unlike in PMP mode, in mesh mode (as shown in Fig. 6.9) some SSs may have no direct communication with the BS and a SS can communicate directly with its neighbors without any collaboration from the BS. Communication among nodes, including the BS, can be done on the principle of equality in distributed bandwidth scheduling or on the principle of superiority of mesh BS in centralized bandwidth scheduling depending on which of these two protocols is used. Wired connection to backhual network
Mesh BS Mesh SS
Bidirection wireless connection
Fig. 6.9 WiMAX Mesh network
WiMAX networks inherit the advantages of wireless networks: rapid deployment, high scalability, and low maintenance costs. In addition, WiMAX mesh networks (WMN) are more scalable than PMP networks. Also, mesh networks are characterized by dynamic self-organization, self-configuration, and self-correction to enable flexible integration, easy maintenance, and reliable services. Also, mesh networks have the ability to extend range and steer clear of LOS requirements in a high frequency band. Communications in mesh networks are more reliable by providing an alternative path. Due to these advantages, WMN is believed to be a promising technology that will constitute the future generation of wireless mobile networks. Within the mesh context, downlink is defined as the traffic sent from the mesh BS and uplink as the traffic sent to the mesh BS. In a mesh network, neighborhood and extended neighborhood are defined. The neighborhood of the node is nodes with direct communication with the node or only “one hop” away from the node. The extended neighborhood contains nodes that are “one hop” or “two hops” away
132
Abdou R. Ahmed, Xiaofeng Bai, and Abdallah Shami
from the node. In other words, the extended neighborhood of the node is nodes of its neighborhood and the neighborhoods of all nodes of its neighborhood. All communications in the mesh system are link based rather than connection based as in PMP networks. One link will be established between two nodes and this will carry all the data transmission. QoS parameters are set over links on a messageby-message basis. Each message has service parameters in its header, no QoS or service parameters are assigned to the link. Since the WiMAX mesh mode MAC is connectionless based, it cannot be used to support guaranteed QoS over multiple hops.
6.3.2 Mesh Network Operations In WiMAX mesh networks, the operating parameters of the network are specified by network descriptor, which is periodically advertised through the network. The physical layer in WiMAX mesh mode is based on orthogonal frequency division multiplexing (OFDM). A frame in mesh mode consists of control subframe and data subframe as shown in Fig. 6.10. The data subframe is divided into mini-slots, which carry data from/to SSs to/from BS, or other SSs. The data subframe serves the PHY transmission bursts. The data bursts start with a long preamble (two OFDM symbols) serving synchronization, immediately followed by several MAC PDUs. Each MAC PDU is comprised of a 6-byte MAC header, a 2-byte mesh subheader with Node ID, a variable length MAC payload (0-2039 bytes), and 4-byte optional cyclic redundancy check (CRC). The control subframe is fixed in length as (MSH-CTRL-LEN x 7) OFDM symbols, where the parameter MSH-CTRL-LEN is advertised in the network descriptor and is 4 bits in length (i.e., value ranges from 0 to 15). The control subframe can be one of two types: the network control sub-frame (Fig. 6.10a) and schedule control subframe (Fig. 6.10b). The network control subframe occurs periodically with the period indicated in the network descriptor. The schedule control subframe occurs in all other frames without network control subframe. In particular, the number of frames with a schedule control subframe between two frames with network control subframe is the value of the field scheduling frame in the network descriptor multiplied by 4. For example, if scheduling frame = 5, then after a frame with network control subframe, the following 20 frames have schedule control subframe, which is again followed by the next frame with network control subframe. The Network control subframe is defined primarily to help new nodes gather information about the mesh network and thereby to join it. The network control subframe is divided into MSH-CTRL-LEN portions and each portion is seven OFDM symbols in length. The first portion of network Control subframe is the network entry component carrying the information of mesh network entry message MSH-NENT. The remaining (MSH-CTRL-LEN – 1) portions are the network configuration components carrying the information of mesh network configuration message MSH-NCFG.
6 WiMAX Networks
133
Frame n-1
Frame n
Network Control sub-frame
Network Entry
Long Preamble
Frame n+1
Data sub-frame
…
Network Config.
MSH-NENT
Frame n+2
Guard Symbol
Network Config.
Long Preamble
MSH-NCFG Guard Symbol
a. Frame n has a network control sub-frame
Frame n-1
Frame n
Schedule Control sub-frame
Central Sched.
Frame n+1
Frame n+2
Data sub-frame
Central Sched.
Long Preamble MSH-CSCF/CSCH Guard Symbol
…
Distr. Sched
Long Preamble
MSH-DSCH
Guard Symbol
b. Frame n has a schedule control sub-frame
c 2006 IEEE.) Fig. 6.10 Frame structure in WiMAX mesh mode. (After [9].
The Schedule Control subframe is defined to schedule the sharing nodes in a common medium in centralized or distributed scheme. As the mesh configuration message MSH-NCFG is sent periodically, schedule control subframe may carry the MSH-NCFG message. In schedule control subframe, a mesh centralized scheduling MSH-CSCH and/or MSH-NCFG message come first and then follow the mesh distributed scheduling messages MSH-DSCH. The number of mesh distributed scheduling messages is specified by the MSH-DSCH-NUM parameter in network descriptor. This implies that the first (MSH-CTRL-LEN – MSH-DSCHNUM) portions are allocated for transmitting the mesh centralized scheduling message MSH-CSCH and/or mesh centralized configuration message MSH-CSCF. The remaining portions are allocated for transmitting the mesh distributed scheduling message MSH-DSCH.
6.3.2.1 Mesh Network Entry Process As stated before, in WiMAX mesh mode, mesh network configuration (MSHNCFG) and mesh network entry (MSH-NENT) messages are used for advertising
134
Abdou R. Ahmed, Xiaofeng Bai, and Abdallah Shami
of mesh network parameters and to allow new nodes to synchronize and to join the mesh network. Active nodes within the mesh network periodically advertise MSHNCFG messages with network descriptor, which outlines the basic network configuration information such as BS ID number and the base channel currently used. IEEE 802.16 specifies the network entry as follows: 1. A new node that plans to join an active mesh network scans for active networks and listens to MSH-NCFG message. The new node establishes coarse synchronization and starts the network entry process based on the information given by MSH-NCFG. 2. The joining node (which is called candidate node) builds a physical neighbor list from all nodes from whom it received MSH-NCFG. From the neighbor list, the candidate node selects a potential sponsoring node to connect to; then, it sends MSH-NENT:NetEntryRequest which includes node ID of the potential sponsor. Candidate node uses temporary node ID (0x0000) as transmitter’s node ID, until it obtains a unique node ID later in the entry process. NetEntryRequest requests from a potential sponsor to open a channel through which candidate node can negotiate basic capabilities and perform authorization. 3. Potential sponsor node, upon receiving NetEntryRequest with its own ID, decides either to open the channel or to reject the request and replies with MSH-NCFG. If the potential sponsoring node accepts the request and opens the channel, this node becomes the sponsoring node and the channel will be used in further steps. 4. If the candidate node receives a reject reply, it will try with another node from its neighbor list. If it does receive neither rejection nor acceptance within the timeout period, it will resend NetEntryRequest. 5. Once the candidate node receives a positive response (NetEntryOpen), it shall perform fine time synchronization with the sponsoring node and then sends MSH-NENT:NetEntryAck message. 6. The candidate and the sponsoring nodes use the schedule indicated in the NetEntryOpen message to exchange messages for node authorization, registration, establishing IP connectivity, and transferring operational parameters. Once this process is completed, the entry process will be terminated by sending MSHNENT:NetEntryClose message from the candidate node. After completion of network entry process, the sponsoring node will become the father node through which new node will connect to the network. As stated above, a new node connects to the first node that accepts its request. This may result in long paths with high interference and poor performance. To improve the network performance, several schemes have been proposed that take interference into account when the new node tries to join the mesh networks. In [10], the authors propose an interference-aware algorithm that considers interference condition in mesh network construction. The concept of blocking metric B(k) of a given path from the mesh BS toward a SS node k is introduced to model the interference level of paths in the mesh. B(k) contains all nodes interfering with any of the intermediate nodes along the route from the root node toward the destination
6 WiMAX Networks
135
node k. In the interference-aware construction scheme, the blocking metric information is incorporated into the network descriptor of a MSH-NCFG message. When a new node is scanning for active networks during the network entry process, the new node chooses the potential sponsoring node based on the blocking metric information to reduce interference of the multihop path and hence to improve throughput. In [11], the authors propose an enhancement algorithm based on the previous algorithm to improve network construction. They note that the previous algorithm considers the minimal interference along the path only when a new node joins the network. But after a SS joins the network, the interference value on the path of other SSs in the network may change. Therefore, the entry order impacts the construction of network and therefore impacts the performance of the network. An improved method of constructing the network is to have the impacted SSs select a father node once more. Figure 6.11 shows the process of new node entering and adjustment of the network after that, where SS5 is the entry node. After SS4 joined the network, the interference value from node 6 to node 2 becomes greater than the interference value from node 6 to node 5, so the father node of SS6 is adjusted from SS2 to SS4. BS SS 1
Entry Node 4
1
2
5
BS
BS
6
4 2
2 5
5 3
1
4
3
6
3
6
Fig. 6.11 Construction and adjustment of routing tree
6.3.2.2 Bandwidth Scheduling in WiMAX Mesh Network In mesh mode, distributed bandwidth scheduling and centralized bandwidth scheduling, or a hybrid of them, can be used. In distributed bandwidth scheduling, bandwidth is allocated to SSs in a distributed manner like in ad hoc networks. In contrast, in centralized bandwidth scheduling, resource allocation is done in a centralized manner like what happens in PMP mode.
Centralized Bandwidth Scheduling In centralized bandwidth scheduling, BS’s function is almost like that of the BS in PMP mode as it performs the slot allocation for all SSs. But unlike what occurs in PMP mode, traffic from/to the BS can be relayed by other SSs through a multihop route. For resource allocation, each mesh SS estimates and sends its resource request
136
Abdou R. Ahmed, Xiaofeng Bai, and Abdallah Shami
to the mesh BS, and the mesh BS determines the amount of granted resources for each link. Two control messages are used in centralized scheduling: mesh centralized scheduling configuration (MSH-CSCF) and mesh centralized scheduling (MSH-CSCH). The MSH-CSCF message delivers the information of channel configuration and routing tree, while the MSH-CSCH message contains the information of bandwidth request, grant, and updating of routing tree. In a mesh network, each node needs to keep information about path(s) to other nodes in the network; this information is kept in the routing tree. All SSs maintain a routing tree whose root is BS and the children are SSs. Nodes obtain information to build the routing tree from MSH-CSCF messages. At first, the BS generates MSH-CSCF and broadcasts it to all of its neighbors, and all neighbors shall forward (rebroadcast) this message according to the index number specified in the message. This process is repeated until all SS nodes have broadcast the MSH-CSCF message. Once each node builds its routing tree, any change in these routing trees is broadcast periodically through MSH-CSCH messages. For resources allocation SSs send their needs for resources in a MSH-CSCH: Request message and the BS sends response to these requests in MSH-CSCH:Grant message. The transmission order of MSH-CSCH:Request messages are determined according to the hop-count for nodes in different levels and according to the transmission order listed in the routing tree for nodes with the same hop count. Each SS sends its MSH-CSCH:Request to its parent. Each non-leaf SS collects its children’s requests and incorporates them into its own MSH-CSCH:Request before transmitting to the father node. Thus, the BS gathers bandwidth requests from all the SSs and assigns spatial resource for SSs. These assignments (grants) are put in MSHCSCH:Grant message, and are broadcast by BS to its children. Then, the BS’s children nodes rebroadcast the MSH-CSCH:Grant message. This process is repeated until all the SSs have received the MSH-CSCH:Grant message. From the MSH-CSCH:Grant message, the SSs determine the actual uplink and downlink transmission times by a common algorithm which divides the frame proportionally.
Distributed Bandwidth Scheduling In distributed bandwidth scheduling, all nodes are equivalent. In other words, the BS is not a coordinator of all communications in the network; therefore, there are no clearly separate downlink and uplink transmissions. Also, BS does not manage the bandwidth allocation in the network. Each node competes for access to the control subframe of the channel. Once SS has access to the control subframe of the channel, it allocates bandwidth required for its data in the next frame. Hence, the contention process in the control subframe effects mainly on the performance of the network. The data transmission during the data subframe of the frame is determined according to previous granting and is not affected by the contention in the control subframe in the same frame. In the distributed scheduling, mesh distributed scheduling MSH-DSCH messages play a significant role in the whole scheduling process. All nodes transmit
6 WiMAX Networks
137
MSH-DSCH at a regular interval to inform their neighbors of the schedule of transmitting stations. MSH-DSCH is used to broadcast resource requests and grants to the neighbors. Further MSH-DSCH messages give information about free resources that neighbors can grant them.
6.3.3 Routing in Mesh Network In mesh network, numerous routes may exist between the mesh BS and each end user (served by a SS). Each SS in the mesh network works as a wireless router and the mesh BS functions as a gateway that connects the mesh network to the backhaul. A routing algorithm is needed to route packets from end users to the mesh BS (upstream direction) and from the mesh BS to end users (downstream direction). A routing algorithm should take two points into account: the connectivity changes due to the users’ mobility and a wireless link that goes up and down on the fly. An enduser sends a data packet to one of its neighborhood routers; then, this router injects the packet into the mesh network. The packet travels through the mesh, possibly over multiple hops, to the mesh BS and then to the rest of the Internet. The routing algorithms in the mesh network run inside each wireless router and the mesh BS. The mesh routing protocols that are already used by several companies for their commercial products or in several academic projects can generally be classified into three categories: proactive, reactive, and hybrid protocols. Proactive (or table-driven) routing protocols – i.e., delay-aware routing algorithm (DARA) [12] – collect information in advance such that it will be available when the need arises. Therefore, each node maintains a full routing table to all destinations and routing updates are used in order to maintain up-to-date information. Reactive (on demand) routing protocols – i.e., predictive wireless-routing protcol (PWRP) [13], on the other hand, look for information only when needed. For example, when a node needs to reach another node, routes are dynamically created as a result. In hybrid routing protocols, some nodes may implement a proactive routing protocol and others run a reactive routing protocol.
6.3.4 QoS in Mesh Network As stated in Section 6.2.3, for IEEE 802.16 PMP mode, the standard defines connection-based five QoS classes: unsolicited grant service (UGS), extended realtime polling service (ertPS), real-time polling service (rtPS), non-real-time polling service (nrtPS), and best effort (BE). Comparatively, for mesh mode, no similar terms or schemes have been defined. Since the IEEE 802.16 mesh mode MAC is connectionless based, it cannot be used to support guaranteed QoS over multiple hops. In [9], the authors proposed a new scheme to achieve QoS for different services types in mesh mode. This scheme proposes a simple and effective method to prioritize the diverse traffics and enable the QoS requirements. The scheme is based on the distributed resource scheduling in the mesh network.
138
Abdou R. Ahmed, Xiaofeng Bai, and Abdallah Shami
6.4 Mobility in WiMAX Networks The demand for mobile systems has grown because people need to move around or travel while still connected. To fulfill this need, a great deal of effort continues to be placed into the development of new standards. Mobility within connection-oriented speech systems like global system for mobile communications (GSM) and universal mobile telecommunications system (UMTS) are widely used. When it comes to packet-based networks, the mobility aspect is a rather new area and there is still much work needed to make services reliable and highly efficient. IEEE is trying to address this area with the development of wireless MAN air interfaces such as 802.16e [5] and 802.20 [14]. There are high expectations on WiMAX and its ability to support mobility and it is an interesting area to study. Here, we will look into the roaming world of WiMAX which naturally leads to a closer look at its different handover possibilities as well as the handover support in 802.16e. First, handovers is discussed before specific discussion of mobility in WiMAX. Then, the role of WiMAX to support mobility in homogenous and heterogeneous wireless networks is explained.
6.4.1 Handovers The handover is a process during which a mobile station (MS) immigrates from airinterfaces of its current BS to air-interfaces of adjacent BS. Handover is an important process for mobility support in wireless networks. The reason for handover could be that a cell is overloaded or that the MS moves out of the BS transmission range. The BS that served the MS before the handover is often called the serving BS while the new BS is referred to as the target BS. A handover process consists of two steps: the pre-registration step and the actual change of connection. During the pre-registration step, the MS collects information about its neighboring BSs by measuring their signals, building a list of possible target BSs, and sending a handover request message. During the actual handover, MS will terminate the connection to its serving BS and open a new connection to the target BS. Regarding handling of the connections in actual handover phase, there are two different types of handovers: hard and soft handovers. There are two other types of handovers regarding the architecture of networks between which MS transfers during the handovers. These two types are horizontal and vertical handovers. In some references, these latter types of handovers are termed as architectural handover.
6.4.1.1 Hard Handover During a hard handover the mobile station interrupts communication with the serving BS and then makes a transition to the target BS. The MS has to register with the
6 WiMAX Networks
139
target BS before receiving and transmitting to the new BS. This kind of handover is often referred to as a break-before-make handover and is the most common way to perform handovers.
6.4.1.2 Soft Handover During a soft handover the MS keeps contact with both base stations throughout the second phase of the handover, often called make-before-break. As the mobile station is always connected during the handover, this enables the handover to be more robust against fast fading, shading, and multipath propagation. However, by using two connections, the overhead for handover increases. Soft handover is more complex than hard handover, but is often preferred for its enhanced stability and upper-layer application performance.
6.4.1.3 Horizontal Handover The horizontal handover is a handover between base stations belonging to the same type of network technology (homogenous networks). The horizontal handover can be a hard or soft handover. It is reasonable to support a soft version in the horizontal handover, since the collaboration between components in the network required for soft handover can be easily done between networks of the same technology.
6.4.1.4 Vertical Handover The vertical handover is performed between base stations attached to different network technologies (heterogeneous networks). It is possible for a vertical handover to be either hard or soft. However, since soft handover is more complex and demands more collaboration between components in the network, it is not commonly used as vertical handovers. A model containing technical aspects of vertical mobility in heterogeneous networks is proposed in [15].
6.4.2 WiMAX Standards and Mobility Mobility was not considered in the WiMAX standard up to the fourth version [2]. But some works showed that the standard could lend itself for use in mobility scenarios. The mobility is considered in 802.16e [5]. This version provides requirements to support mobility at vehicular speed and seamless handover while maintaining differentiated QoS. It supports nomadic roaming and seamless handover. In Section 6.4.2.1, we show how the 802.16d standard can support mobility while mobility aspects of 802.16e are studied in Section 6.4.2.2.
140
Abdou R. Ahmed, Xiaofeng Bai, and Abdallah Shami
6.4.2.1 Mobility with IEEE 802.16d Although the 802.16d standard is devised for fixed terminal locations, it is indeed possible for 802.16d without changes to support the mobility. The two requirements for mobility support are: first, as MS moves within the coverage of the BS, the QoS of established connections between the MS and its BS should be satisfied; secondly, the network should provide some mechanism to handle MS transition from one BS to another, in other words provide a handover mechanism. Also, data loss and delay in the handover process should be as minimal as possible. Correspondingly, the feasibility of mobility support for the 802.16d standard was proved in [16]. To support mobility with the features and protocols defined in the existing 802.16d standard, it is observed that the handover functionalities are quite similar to the registration of a SS with a BS upon power up. But handover needs to tear down the existing connection with the current BS and to set up a new connection with a new BS. By ignoring the delay in setting up the new connection for a moment, the initialization process defined in the 802.16d standard can be reused to assist connection handover. After reviewing the initialization process steps and by eliminating unnecessary steps, the functionalities required for handover can be obtained. Noticeably, the handover procedure actually represents a “short” initialization process. This not only enables handover to reuse existing functionalities, but also helps keep the handover latency satisfactorily low. As handover is required when SS moves away for the serving BS, it is more appropriate to have the SS initiate the handover. In fact, if the BS initiates the handover, it may take decision of handover when the SS stays silent for a long time, even if it has good communication with the BS. When a SS realizes a need for handover, it sends a handover request (HO-REQ) to its current BS. In turn, the BS returns with a handover acknowledgment (HO-ACK) message to signify that the SS can start the handover process. After responding to the SS’s request for handover, the old BS sends the backhaul network message (BN-MSG) to inform its surrounding BSs about the service parameters associated with the SS, and builds a list of neighboring BSs which are candidates for handover. After SS received the HO-ACK message, it scans and synchronizes with a new channel of a neighboring BS. Then, it obtains the uplink transmission parameters, completes the ranging and adjustment procedure, and finally, registers and sets up provisional connections with the new BS. As the HO-REQ and HO-ACK messages are not defined in the 802.16d standard, the de-registration command (DREG-CMD), with an action code of 03, can function as the HO-REQ and HO-ACK messages. The SS sends a DREG-CMD (code = 03) message to its BS to initiate the handover. The BS returns another DREG-CMD (code = 03) to the SS. This is truly based on that the standard specifies ([3], Section 6.3.2.3.26) that “the DREG-CMD message shall be transmitted by the BS on a SS’s basic CID to force the SS to change its access state”. Upon receiving a DREGCMD, the SS shall take the action indicated by the received action code. If the action code is 03, that means no change. The standard does not specify about what the BS can do if it receives the DREG-CMD (code = 03) message from its SSs. Thus, it
6 WiMAX Networks
141
is acceptable if the BS chooses to interpret the message as a request for handover (HO-REQ). After the SS sends the first DREG-CMD (code = 03) message, the SS will interpret the returned DREG-CMD from the BS as an ACK (HO-ACK). 6.4.2.2 Mobility with IEEE 802.16e The 802.16e standard is the basis for mobility in WiMAX standards. It supports handovers through procedures and functions at BS/MS level. The standard defines the means for gathering information and performing a handover, but the decision whether to perform a handover or not is left for network designers. A detailed study of mobility in WiMAX in general, the role of 802.16e in supporting it, and effort of the WiMAX forum to complete the required specifications of mobility is found in [17]. Network Topology Acquisition The MS needs to acquire information about the network in order to select a new BS and perform handover. Network topology advertisement messages are provided to help MS for collecting required information. These messages contain information about neighboring BSs and their channels, and are broadcast by all BSs periodically. In addition, the MS may scan its radio environment for potential target BSs. The MS starts scanning at intervals under permission of its serving BS and can terminate at any time, and during this interval, its data are buffered at the BS. The buffered data is sent to the MS when it exits the scanning mode. The association procedure is an optional feature in the standard and can occur during scanning. Association level can be in one of three modes: scan/association without coordination, association with coordination, and network-assisted association. The association enables the MS to collect and store information about BSs. The gathered information is saved during a reasonable period of time to assist the MS in decisions regarding handovers. An adaptive scanning algorithm, proposed in [18], determines the duration and frequency of channel scanning in order to minimize the disruptive effects of scanning on the application traffic during handovers across multiple IEEE 802.16 networks. The proposed algorithm allocates scanning intervals for multiple MSs in a manner that maintains the quality-of-service requirements of the running applications. The algorithm consists of two main components: (1) estimating the time needed by a MS to scan a list of neighboring BSs and (2) interleaving of channel scanning and data transmission intervals. This algorithm is useful especially for mobility between different networks. Handover Process The handover process in 802.16e consists of six different stages: cell reselection, handover decision and initiation, synchronization to target BS downlink, ranging,
142
Abdou R. Ahmed, Xiaofeng Bai, and Abdallah Shami
termination of service, and handover cancelation. In cell reselection, the MS collects information about BSs in the network, either through topology advertisements or scanning intervals. The handover decision can be triggered in the MS as well as in the BS. During synchronization to the target BS downlink channel, the MS receives downlink and uplink transmission parameters from the BS. The length of this process is dependent on the information that the MS already has about the BS. Initial ranging or handover ranging is needed for the MS to synchronize to the channel in order to receive the correct transmission parameters, e.g., time offset and power level. In the last step, the serving BS will terminate all connections associated with the MS and remove all information in queues, counters, etc.
Soft Handover and Fast BS Switching During a soft handover the MS will listen/transmit to several BSs concurrently by receiving the same PDUs from all BSs and sends its data to them. The MS will perform diversity combining on the signals received from the BSs and the BSs will, in turn, perform diversity combining among them to get the uplink PDUs. In the fast BS switching, the MS has an active set of BSs and connects to only one BS (called the anchor BS) at a time. In both fast BS switching and soft handover, the MS maintains an active BS set containing all BSs that are suitable as target BSs. All the BSs in the active set need to be synchronized to each other to be able to support soft handover and fast BS switching. As in the soft handover case, the MS needs to receive the same data from all BSs at the same time. In the fast BS switching, the MS can switch from one BS to another frame by frame. If the BSs send the frames at different points in time the MS may lose/receive duplicate frames when switching. A seamless handover mechanism called last packet marking (LPM) is suggested in [19] to enable seamless handover in networks based on the 802.16e standard. This mechanism integrates MAC layer with network layer handover to decrease the handover effects on TCP service performance. LPM mainly consists of 802.16e handover and adds messages containing information about routing. The main idea of LPM is to send incoming MS packets to both serving BS and target BS from the point in time when the MS is thinking of performing a handover. The target BS will buffer incoming packets and forward them to the MS when the handover is completed.
6.4.3 WiMAX and Homogeneous Mobility In homogeneous mobility, MS moves between networks of the same technology. As only one network technology is involved, mobility support is required only within this type of network. This leads to less complicated solutions for roaming mechanisms since measurements, e.g., signal-to-noise ratio (SNR), can be utilized. It is clear that homogeneous mobility is dependent on horizontal handovers
6 WiMAX Networks
143
in 802.16e and WiMAX. Even we used IEEE 802.16 and WiMAX interchangeably up to now, here we differentiate between them. While the IEEE 802.16 supports handover between BSs, WiMAX offers protocols for handover in the higher layers in the network structure. The WiMAX architecture shall support mechanisms such as intra/inter access service network (ASN) handover, roaming between network service providers (NSPs), seamless handover at vehicular speed, and micro/macro mobility [20].
6.4.3.1 Access Service Network ASN contains at least one ASN gateway (ASN GW) which is connected to the connectivity service network (CSN) and a base station which handles the connection to the MS. The relationship between ASN GWs and BS within ASN is many to many as shown in Fig. 6.12. Connections in Fig. 6.12 are not only physical connections, but also are reference points that are introduced in [20]. It is known that the two BSs participating in handover are serving BS and target BS. For ASN GWs, the serving ASN GW and the target ASN GW are the ASN GWs associated with the serving and target BSs, respectively. The term anchoring ASN GW is used when an ASN GW relays MS data to the serving ASN GW.
ASN GW
ASN GW
BS
BS
ASN GW
BS
ASN GW
BS
BS
Fig. 6.12 ASN entity
6.4.3.2 Anchoring The anchoring ASN GW is responsible for receiving the MS’s data and forwarding them to its serving ASN GW, so the network needs only to know MS’s anchoring ASN GW and does not care where MS is located exactly. This makes the mobility of the MS transparent to the CSN and the need to change the IP address becomes less frequent. 6.4.3.3 Intra ASN Handover The intra ASN handover is performed between BSs (or sectors within one BS) belonging to the same ASN. To minimize the delay and data loss during intra ASN
144
Abdou R. Ahmed, Xiaofeng Bai, and Abdallah Shami
handover, the MS does not change the IP address after the handover since the movement of the MS is not visible from outside the ASN. The WiMAX standard does not specify what messages are sent over which interface during an intra ASN handover.
6.4.3.4 Inter ASN Handover During an inter ASN handover, MS moves between BSs belonging to different ASNs; ASN GWs in separate ASNs should coordinate their actions to make the handover smooth to the MS. The path of the data flow is updated, or not, after the handover depending on whether reanchoring or anchoring is used in the handover, respectively. Target or anchor ASN GW can decide to anchor or reanchor the data path. Only if both parties indicate that reanchor is not needed then anchor takes place, otherwise reanchor will take place.
6.4.4 WiMAX and Mobility in Heterogeneous Networks The third-generation cellular networks and fixed wireless broadband networks, such as Wi-Fi, are different with regard to coverage, data rates, QoS, and mobility. The inter-networking and mobility between these two network technologies is now an active subject. WiMAX fills the gap between the two technologies caused by differences between them. Intercommunication of WiMAX with these technologies is a promising method to extend customer services especially in light of increasing 3G devices and Wi-Fi market development. WiMax can inter-network with both 3G and fixed wireless broadband networks [17]. In this section, integration of WiMAX with UMTS as representative of 3G networks and Wi-Fi as a widely used fixed wireless technology will be discussed briefly.
6.4.4.1 UMTS and WiMAX Integration Both WiMAX and UMTS contain well-defined interfaces between user equipment, base stations, and the core infrastructure. From architectures of UMTS and WiMAX (Fig. 6.13), it is clear that ASN can be mapped to the universal terrestrial radio access network (UTRAN) and CSN can be mapped to the core network (CN). As
User
UTRAN
CN
MS
Uu
ASN
CSN
802.16 UMTS
Fig. 6.13 UMTS and WiMAX architecture components [16]
WiMAX
6 WiMAX Networks
145
both WiMAX and UMTS support management in the CN, this helps in heterogeneous mobility. The factors of diversity in data rates and end-to-end delay make integration between the two networks a not-so-easy process. WiMAX supports five classes of QoS that are similar to UMTS QoS parameters and thus, traffic flows from one network can be mapped to the other. Diverse data rates between the two networks will shrink to an acceptable limit with high speed downlink packet access (HSDPA) and high speed uplink packet access (HSUPA) versions of UMTS and the coming Super 3G networks. Security level and polices are almost the same in both WiMAX and UMTS. As WiMAX network architecture has been made to fit the 3rd Generation Partnership Project (3GPP) model of inter-networking with alternate access networks [19], and UMTS has undergone standardization to inter-network with Wi-Fi, this provides the finalized requirements and standards for WiMAX to utilize and increase interoperability. Because WiMAX does not supply IP connectivity compared to the Wi-Fi analogy, a logical CSN inter-networking unit with DHCP and DNS needs to be located inside the ASN. There are two types of coupling between WiMAX and UMTS, loose and tight, depending on integration between the networks. Loose Coupling The loose coupling is suggested in the WiMAX network architecture [20]. In this scenario, access to 3G authentication, authorization and accounting (AAA) services is granted through packet data gateway (PDG) edge routers connected to the WiMAX network or routed through the Internet. The traffic of the WiMAX network is separated from the UMTS network and hence the WiMAX portion can supply its own mechanisms for mobility, authentication, and billing. In loose coupling networks, mobility needs to be managed at a higher level by protocols such as the IEEE Std. 802.21 [21] for media independent handover or mobile IP (MIP). Tight Coupling In a tightly coupled network, the WiMAX and UMTS share the core network components, such as gateways, AAA, and infrastructure. The WiMAX network is connected to the UMTS gateway general packet radio service (GPRS) support node (GGSN) and the packet switched domain of the UMTS core network. The WiMAX network appears as a radio access network (RAN) to the UMTS network and the gateway emulates radio network controller (RNC) behavior. This requires the user equipment to run UMTS protocol stacks. 6.4.4.2 Wi-Fi and WiMAX Integration Although Wi-Fi tries to fill security holes, wireless IP technology suffers from weak security compared with the cellular industry when it comes to mobility. This might
146
Abdou R. Ahmed, Xiaofeng Bai, and Abdallah Shami
limit the number of services that can be transferred and protected when migrating from WiMAX to WLAN. Transport of bulk data and cheaper access through Wi-Fi hotspots seem to be the interesting aspects of WLAN/WiMAX mobility, where a WiMAX network serves as a backhaul for WLAN access points. Although 802.11f takes a step toward the WLAN roaming, fast mobility is impractical because handovers would be performed very frequently due to the relatively small cell area. 802.11 networks have proprietary solutions for handling roaming between access points, but the 802.11f standard will allow roaming interoperability between access points of different manufacturers. Faster mobility is something that WiMAX, with its MAN coverage, might be able to handle better with the 802.16e standard for roaming. Handovers between the different systems are required when mobile devices change IP subnets, and can be managed at a higher level in WiMAX network where layer 3 (L3) handover is not performed unless the mobile device moves between ASNs. In 802.11, this type of L3 handover is performed when a device moves between access points in different subnets. MIP is extensively used to handle this type of mobility. QoS capabilities in WLAN are limited, but with 802.11e it can offer eight user priorities, and cover only one connection. On the other hand, WiMAX can cover several connections with different QoS parameters on each connection. The integration of WLAN and WiMAX access networks aims to provide services with respect to the attached users. Since both networks are IP based, integration does not require translations in protocols when bridging the infrastructure. As in the case of UMTS and WiMAX, mobility across Wi-Fi and WiMAX networks can be integrated according to two models: loose coupling and tight coupling.
6.5 Summary WiMAX mesh networks have many advantages and are promising for future networks. Possibly, mesh networks did not get more interest because IEEE 802.16 std did not specify QoS schemes in mesh networks, but some schemes can be used to guarantee QoS and improve throughput in mesh networks. So this can be a starting point for more interesting uses of WiMAX mesh network. Mobility in WiMAX networks started with 802.16e standard and studies showed that mobility can be implemented with 802.16d standard. 802.16e networks provide handover through procedures and functions at BS/MS level. WiMAX support mechanisms such as soft handover, fast BS switching, intra/inter ASN handover, roaming between NSPs, seamless handover at vehicular speed, and micro/macro mobility. WiMAX supports mobility not only in homogenous network, but also in heterogeneous network where WiMAX integrate with cellular networks such as UMTS, or fixed wireless network, such as WiFi.
6 WiMAX Networks
147
References 1. IEEE P802.16-REVd/D5-2004 (2004). Air Interface for Fixed Broadband Wireless Access Systems. 2. IEEE Std 802.16-2001 (2001). IEEE Standard for Local and Metropolitan Area NetworksPart 16: Air Interface for Fixed Broadband Wireless Access Systems. 3. IEEE Std 802.16-2004 (2004). IEEE Standard for Local and Metropolitan Area NetworksPart 16: Air Interface for Fixed Broadband Wireless Access Systems. 4. IEEE Std 802.16a-2003 (2003). IEEE Standard for Local and Metropolitan Area NetworksPart 16: Air Interface for Fixed Broadband Wireless Access Systems-Amendment 2: Medium Access Control Modifications and Additional Physical Layer Specifications for 2–11 GHz. 5. IEEE Std 802.16e-2005 and IEEE Std 802.16-2004/Cor 1-2005 (2006). Part 16: Air Interface for Fixed and Mobile Broadband Wireless Access Systems Amendment 2: Physical and Medium Access Control Layers for Combined Fixed and Mobile Operation in Licensed Bands and Corrigendum 1. 6. Vaughan-Nichols, S. J. (2004). Achieving Wireless Broadband with WiMax. IEEE Computer. 37(6), 10–13. 7. Wongthavarawat, K. & Ganz, A. (2003). Packet Scheduling for QoS Support in IEEE 802.16 Broadband Wireless Access Systems. International Journal of Communications Systems. 16(1), 81–96. 8. Bai, X., Shami, A. & Ye, Y. (2008). Robust QoS Control for Single Carrier PMP Mode IEEE 802.16 System. IEEE Transactions on Mobile Computing. 7(4), 416–429. 9. Zhang, Y., Zhou, M., Xiao, S. & Fujise, M. (2006). An Effective QoS Scheme in WiMAX Mesh Networking for Maritime ITS. Proceedings of the 6th International Conference on ITS Telecommunications. Chegdu, China. 10. Wei, H., Ganguly, S., Izmailov, R. & Haas, Z. (2005). Interference-Aware IEEE 802.16 WiMax Mesh Networks. The 61st IEEE Vehicular Technology Conference (VTC Spring’05). Dallas, TX. 11. Tao, J., Liu, F., Zeng, Z. & Lin, Z. (2005). Throughput Enhancement in WiMax Mesh Networks Using Concurrent Transmission. Proceedings of the International Conference on Wireless Communications, Networking and Mobile Computing. Montreal, Canada. 12. Sarkar, S., Yen, H., Dixit, S. & Mukherjee, B. (2007). DARA: Delay-aware routing algorithm in a hybrid wireless-optical broadband access network (WOBAN). Proceedings of IEEE ICC. Glasgow, UK. 13. Tropos Netwoks (2008). Available [Online] http://tropos.com 14. IEEE Std 802.20. (2006). IEEE 802.20 Mission and Project Scope. [Online] http://grouper.ieee.org/groups/802/20/ 15. Ylianttila, M. (2005). Vertical Handoff and Mobility – System Architecture and Transition Analysis. PhD Thesis, Faculty of Technology, University of Oulu. 16. Leung, K.K., Mukherjee, S. & Rittenhouse, G.E. (2005). Mobility Support for IEEE 802.16d Wireless Networks. Proceedings of IEEE WCNC. New Orleans, LA. 17. Lax, M. & Dammander, D. (2006). WiMAX – A Study of Mobility and a MAC-Layer Implementation in GloMoSim. Master’s Thesis, Ume University. 18. Rouil, R. & Golmie, N. (2006). Adaptive Channel Scanning for IEEE 802.16e. Proceedings of IEEE MILCOM. Washington, DC. 19. Kim, K., Kim, C. & Kim, T. (2005). A Seamless Handover Mechanism for IEEE 802.16e Broadband Wireless Access. International Conference on Computational Science. Issyk-Kul Lake, Kyrgyzstan. 20. WiMAX Forum. (2005). WiMAX End-to-End Network Systems Architecture. Technical Report, August 15 2005. Draft. Stage 2: Architecture Tenets, Reference Model and Reference Points.
148
Abdou R. Ahmed, Xiaofeng Bai, and Abdallah Shami
21. IEEE Std 802.21. (2006). IEEE 802.21 Media Independent Handover. [Online] http://wwww.ieee802.org/21/ 22. Chu, G., Wang, D. & Mei, S. (2002). A QoS Architecture for the MAC Protocol of IEEE 802.16 BWA System. IEEE International Conference on Communications, Circuits and Systems and West Sino Expositions. 1(1), 435–439.
Part III
Optical Access Networks
Chapter 7
Dynamic Bandwidth Allocation for Ethernet Passive Optical Networks Hassan Naser and Hussein T. Mouftah
Abstract This chapter studies the rich feature set of the IEEE 802.3 family of standards for the control of information transfer and for supporting dynamic bandwidth allocation in Ethernet passive optical networks (EPONs). We will study a medium access control protocol, known as multi-point control protocol (MPCP), which arbitrates access to the upstream channel and supports bandwidth allocation in EPON. The precise format of the MPCP messages as well as their processing will be given. The concept of threshold queue reporting will be presented, which is used by the end-user devices (ONUs) to report detailed information regarding the status of the individual queues to the network. The ranging process and automatic discovery of newly connected or off-line ONU stations will also be given.
7.1 Introduction During the past decade, the small growth in the copper-wire access technology has been outpaced by the explosive growth of the Internet traffic and the unprecedented levels of network capacity offered by backbone networks. Service providers are challenged with deploying access solutions that are scalable, fast, and cost-effective to ensure their competitive positions. The success depends on two key factors: the installation of fiber to premises and the reduction in component costs by eliminating the need for electronic conversions between path ends. There is also a need for designing a single multiservice access platform that can deliver packet-based services to increasingly sophisticated and constantly changing residential and Hassan Naser Department of Software Engineering, Lakehead University, Thunder Bay, ON, Canada P7B 5E1, e-mail:
[email protected] Hussein T. Mouftah SITE, University of Ottawa, Ottawa, ON, Canada K1N 6N5, e-mail:
[email protected]
A. Shami et al. (eds.), Broadband Access Networks, Optical Networks, c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-92131-0 7,
151
152
Hassan Naser and Hussein T. Mouftah
commercial applications. Access networks serve as the lifeblood for much of the residential and commercial transactions, as Internet protocol (IP) telephony, IP video broadcasting (IPTV), video-on-demand, http, and mission critical transactions represent a significant portion of these customers’ activities. Passive optical network (PON) technology can address these needs through the use of optical fibers in place of the traditional copper wire. The PON technology provides a significant jump in access network capacity by delivering these advanced services in digital form in the optical domain. Among other advantages of using PON for local access are allowing longer distances between the operator’s central office and the customers; minimizing fiber deployment in both the local exchange and the local loop; allowing for low-cost downstream video broadcasting, because of its point-to-multipoint nature; and eliminating the necessity of installing active devices in the splitting locations, thus reducing the network component costs. A PON is a data link access technology that delivers data from an optical line terminal (OLT), located in the operator’s central office, to a number of optical network units (ONUs), each located at end-user locations or at the curb. One configuration for a PON is shown in Fig. 7.1. A single fiber trunk connects the OLT to a passive optical power splitter. The splitter divides the downstream signal from the OLT into multiple, identical signals that are broadcast to the subtending ONUs. Each ONU is responsible for determining which data are intended for it, and for ignoring all others. Sharing the fiber trunk reduces the deployment cost, while sharing the “passive” optical equipment outside the plant reduces the operational and capital expenditure.
Passive Optical Power Splitter
ONU 1
>> to users
to Internet <<
OLT
ONU 2
ONU M
>> to users
>> to users
Fig. 7.1 EPON topology
A PON is a multipoint-to-point network in the upstream direction (from ONUs to OLT). The data signal transmitted by an ONU would only reach the OLT, but not other ONUs. Thus, if two or more ONU stations happen to transmit signals on the shared trunk at the same moment, the signals will collide causing the information to be corrupted. The collision cannot easily be detected by ONUs because of the directional property of the optical splitter. A medium access control (MAC) mechanism is thus required to arbitrate access dispute between ONUs. Previous research studies have shown that conventional carrier sense multiple access (CSMA) methods
7 Dynamic Bandwidth Allocation for EPONs
153
are not suitable for arbitration in PON, because of its high bandwidth-delay product [1, 2]. Instead, “polling” technique with “request-and-grant” reservation has been shown to be the best method of arbitration for the upstream transmission. While this method facilitates bandwidth allocation in PON, special bandwidth allocation algorithms are needed to offer fair, deterministic services to end-users. Another challenge to be faced when designing a PON is the broad spectrum of applications with quality of service (QoS) requirements. Due to the success of Ethernet centric equipments, these applications are now delivered over IP-based networks. The PON must thus support IP-based QoS mechanisms in order to ensure delivery of these applications. The IP differentiated services (DiffServ) mechanism has been widely used in today’s IP networks as a principal mechanism for providing QoS [3]. DiffServ offers coarse granularity quality differentiation. Application traffic can be categorized into multiple classes (aggregates), with QoS parameters defined for each class. Some of the standardized classes defined in DiffServ include Expedited Forwarding, Assured Forwarding, and Best Effort. Expedited Forwarding provides for low loss and delay-sensitive services; Assured Forwarding provides for low loss services; and, Best Effort does not require any commitment from the network. This chapter will be primarily focusing on medium access control protocols developed for EPON to facilitate bandwidth allocation in the upstream channel, and to dynamically partition this bandwidth among different QoS classes and/or different users. We will review these architectures in detail, which will provide the opportunity to discover their strengths and limitations, and to learn how to build upon their strengths in the future.
7.2 Data Link Layer Protocols for PON With PONs, the underlying communication is time division multiplexing (TDM), which uses timing as the key to interleaving multiple transmissions onto a single channel. Other transmission strategies – based on wavelength division multiplexing (WDM) or code division multiplexing (CDM) – have also been suggested for PON, but are not currently being considered because of cost and scalability considerations [4]. The major difference between the TDM-based PON systems occurs at the data link layer (framing layer). The data link layer protocols introduced so far for PON can be broadly classified into two categories: ATM (asynchronous transfer mode) and Ethernet. ATM PON (APON) is a cell-based transport protocol recommended for use in PON by the full service access network (FSAN) coalition of 21 network operators and 30 vendors. FSAN helped develop the International Telecommunication Union-Telecommunication (ITU-T) APON standard G.983.1. APON aims to provide QoS guarantees and fairness to individual connections [5]. This allows APON to deliver high quality and fair services to customers. However, the complexity of maintaining ATM connections, and the overhead involved
154
Hassan Naser and Hussein T. Mouftah
with carrying IP-over-ATM have made APON a less attractive solution. It is generally believed today that Ethernet framing offers cost and scalability advantages over ATM. Ethernet-based PON today shakes out into two separate and competing standards: Gigabit PON (GPON) and Ethernet PON (EPON). GPON has been produced and standardized by the ITU-T/FSAN organization [2]. It was designed to be an inclusive standard, covering the carriage of TDM, data, and video services, using both ATM and packet (Ethernet) payloads, at wide variety of speeds. Being aligned with the philosophy of APON specification, GPON has adopted fixed periodic transmission cycles and allows Ethernet frames to be fragmented at the OLT and ONU systems. The periodicity of the cycles requires the network-wide synchronization of OLT and ONU systems. EPON is the latest access technology, based on the IEEE standard 802.3ah [6], utilizing an optical network architecture optimized for delivery of voice, data, and video over native Ethernet to residential, businesses and enterprise customers. The ability to deliver these services to the customers via IP on an end-to-end basis over native Ethernet fiber allows service providers to simplify their network architectures and reduce their capital expenses by exploiting low cost commodity Ethernet components. In a nutshell, the major differences between the GPON and EPON specifications are (1) EPON delivers data encapsulated in variable-size Ethernet frames, which cannot be fragmented at the OLT and ONU systems [6]. (2) EPON has adopted variable transmission cycles, which can be varied up to a maximum value or down to a minimum value depending upon current load conditions, thus tolerating traffic burstiness and yielding high bandwidth utilization. EPON has been initially standardized to operate at 1 Gb/s in 2004. Since then, the technology has advanced with the increased demand for faster speeds to support the delivery of digital video services (IPTV, video-on-demand) and the need to support next-generation wireless backhaul [7]. To address these market demands, the IEEE 802.3 working group has recently formed a 10 Gb/s EPON (10GEPON) study group, known as IEEE 802.3av working group [8]. The 10GEPON effort is focused on defining a new physical layer (in comparison to the 1 Gb/s EPON), keeping the MAC layer and all the layers above unchanged. This means that there will be architectural continuity between the 1 Gb/s EPON and 10GEPON in terms of the PON-layer operations and dynamic bandwidth allocation, which are the focus of this chapter. Therefore, we refer to both the 1 Gb/s EPON and 10 Gb/s EPON as simply EPON throughout this chapter.
7.3 Multi-Point Control Protocol (MPCP) The IEEE 802.3ah task force has developed and standardized a control protocol called multi-point control protocol (MPCP) for the control and management of information transfer in the upstream direction from ONU stations to the OLT in EPON. The complete specification of this protocol is described in reference [9].
7 Dynamic Bandwidth Allocation for EPONs
155
MPCP is an in-band signaling protocol that is based on a tree (or tree-and-branch) topology. The MPCP protocol performs the following basic functions: • Request-and-grant bandwidth reservation allows the network to organize bandwidth allocation and to arbitrate access to the medium by the ONUs. The dynamic granting capability allows fast bandwidth assignment, which provides support for existing TCP services in conjunction with delay sensitive voice and video services. • ONU device discovery allows the EPON network to discover and register a newly connected or off-line ONU, and to calculate and compensate the round trip time (RTT) to that ONU. The ONUs convey their traffic loads (queue lengths) to the OLT by using request messages. Each request stipulates the amount of transmission bandwidth an ONU currently requests in order to clear its backlogged queues. The OLT performs bandwidth allocation procedures based on information from current bandwidth requests as well as QoS information (class of service profiles, priorities, etc.) stored in the device, and subsequently sends transmission grants back to the ONUs. Each grant message is used to indicate to the ONU the start time and length of the transmission interval allocated to the ONU in response to its request. In general, the transmission intervals (time slots) are variable, and can contain multiple Ethernet frames. Hence the transmission cycles are adaptive to traffic load. In MPCP, request and grant messages are known as “REPORT” and “GATE” messages, and henceforth we shall refer to them according to the MPCP standard. With MPCP, each ONU can maintain up to eight individual queues, which can be used for prioritization of traffic or for providing differentiated services in the Internet. Each ONU is allowed to transmit data frames backlogged in its queues to the OLT as soon as its time slot starts. In addition to these data frames, the ONU must also transmit a REPORT message during its assigned time slot. When requesting a time slot, an ONU should account for additional overhead associated with a REPORT message. REPORTs must be generated periodically, even when no request for bandwidth is being made. MPCP uses global timing model, where a global clock exists in the OLT, and all ONUs must adjust their local clocks to the OLT clock. The OLT and the ONUs have 32-bit counters that increment every 16 ns. These counters provide a local timestamp. All MPCP messages are timestamped by the local reference clock of the transmitting station. The timestamp inserted by the OLT in a downstream MPCP message allows the destination ONU to set its local clock register to the value in the timestamp field. The timestamp inserted by an ONU in an upstream MPCP message allows the OLT to calculate the round trip time (RTT) from itself to that ONU, as will be described later in this chapter. In addition to GATE and REPORT messages, MPCP introduces three new MAC control messages: REGISTER REQ, REGISTER, and REGISTER ACK, which are used during the device discovery process described later in this chapter. The precise format of each of these five MAC control messages and their processing have been described in [9]. All MPCP messages are transmitted using IEEE 802.3 standard
156
Hassan Naser and Hussein T. Mouftah
packets or protocol data units (PDUs). Figure 7.2 depicts the frame format for an MPCP PDU, which starts with an 8-byte preamble and ends with a 4-byte Frame Check Sequence (FCS). All MPCP PDUs are 64 bytes in length from the Destination Address (DA) field through the FCS field. The fields in the frame format are described as follows:
8 bytes
Preamble
6
6
DA
SA
2
2
L
O
T
C
4
0 to 40
TS
MPCP Data
4 P A FCS D
SA = Source Address DA = Destination Address LT = Length/Type = 88-08 OC = Opcode TS = Timestamp FCS = Frame Check Sequence
Fig. 7.2 Generic MPCP frame format
• Preamble. In IEEE 802.3 standard, the preamble field is used to establish bit synchronization and to determine the start of the frame at the receiver. In EPON, the preamble field is also used to embed a 2-byte logical address called logical link ID (LLID) that identifies the source or destination MAC control client from which this frame is originated or to which this frame is directed. • Destination Address (DA). This is the individual (or a multicast) MAC address of the destination device in the EPON network (ONU or OLT). • Source Address (SA). This is the MAC address of the source device (ONU or OLT). • Length/Type (LT). All MPCP PDUs are type encoded and carry the MAC Control type field value 88-08. • Opcode (OC). Every MPCP PDU has a distinct 2-byte opcode that identifies the specific PDU being encapsulated. • Timestamp (TS). This conveys the local time at the transmitting station at the time of transmission of the PDU. The timestamp field is 4-byte long and increments every 16 ns (or 16 bit-times). • MPCP Data. These 40 bytes are used for the payload of the MPCP PDUs. The padding is used to ensure that the Data field is 40 bytes and hence the PDU is 64 bytes in length.
7 Dynamic Bandwidth Allocation for EPONs
157
7.3.1 REPORT Frame Format The structure and encoding of the REPORT PDU is depicted in Fig. 7.3, as defined in the IEEE 802.3ah standard. It consists of the following fields: • Opcode (OC). The REPORT message has an opcode of 00-03 in hexadecimal. • Timestamp (TS). This conveys (to the OLT) the local time at the source ONU at the time of transmission of the REPORT message. This field is used by the OLT to calculate the round trip time to the ONU. • Number of Queue Sets (NQ). A 1-byte field that specifies the number of Report Sets included in this REPORT message. • Report Sets. This indicates the ONU’s current bandwidth requirements, expressed by the number of bytes backlogged in each queue. The structure of one Report Set (RS) is shown in the bottom part of Fig. 7.3 and consists of the following fields. The structure can be repeated n times in the Report Sets field, where n is the value encoded in the NQ field of this REPORT message. – Report Bitmap (RB). An 8-bit flag register that specifies which queues are represented in this Report Set. – Queue Report n (QR n). A 2-byte field which specifies the number of bytes backlogged in queue number n. The QR n is present in this Report Set only if bit n in the RB flag register is set to 1. With MPCP, there can be several queue reports for one queue in a single REPORT message. This arrangement allows an ONU to report to the OLT information regarding the marking of frame boundaries in the packet stream stored in the queue. At least one queue report for each backlogged queue must be included in the REPORT message. Therefore, if all eight queues are currently backlogged, the
8 bytes Preamble
6 DA
6 SA
2
2
L
O
T
C
4 TS
1 N
0 to 39 Report Sets
Q
4 P A FCS D
OC = Opcode TS = Timestamp
1 0/2 0/2 0/2 0/2 0/2 0/2 0/2 0/2
NQ = Number of Queue Sets RB = Report Bitmap QR n = Queue n Report
R QR QR QR QR QR QR QR QR 1 2 3 4 5 6 7 B 0
Report Set Repeated n times as indicated by NQ field
Fig. 7.3 MPCP REPORT frame format
158
Hassan Naser and Hussein T. Mouftah
first Report Set in the Data field is initialized with the maximum backlogged bytes of queue 1, queue 2, . . ., and queue 8, respectively. This is to ensure that at least one QR will be reported for each backlogged queue. In this case, the first Report Set occupies the 17 leftmost bytes of the Data field. The remaining 22 bytes of the Data field can be used to insert a second complete Report Set (17 bytes) and a third partial Report Set (5 bytes). Another arrangement would be to insert multiple partial Report Sets in the remaining 22 bytes, but each Set can now include QRs for only a limited number of queues (perhaps the highest priority queues). Which arrangement is selected depends upon the current traffic load conditions of the queues and the queue management policy adopted by the ONU. On the other hand, if a given queue is presently backlogged and all other queues have no data to transmit; up to 13 Report Sets can be inserted in the Data field. In this case, each Report Set includes a 1-byte bitmap (with only the bit corresponding to the backlogged queue is set to 1) and a 2-byte QR for the given backlogged queue. The ability to report multiple QRs for one queue allows the ONU to mark and report frame boundaries in the packet stream to the OLT. For example, the first QR can be set with the total bytes of all frames backlogged in the queue; the second QR can be set with the total bytes of a subset of frames reported in the first QR; the third QR can be set with the total bytes of a subset of frames reported in the second QR; and so on. This way if the OLT cannot grant the maximum requested bytes to the ONU, requests with lower bytes will be processed. Note that since in EPON fragmentation is not allowed, a given frame cannot be split across multiple grants. Central to the MPCP protocol is the concept of threshold queue reporting which is used by the ONUs to mark frame boundaries in their queues. In general, an ONU can have up to 13 threshold values for each of its queues. Figure 7.4 shows an illustrative example of a queue at a particular ONU with 13 threshold values (T1 to T13 ). The figure shows the content of this queue just before the ONU sends its
Threshold T13
…
T4
T3
T1
T2
P
P P P
P P
K
K K K
K K
T 6
T T T 5 4 3
T T 2 1
M1 M2 M3 M4 bytes
Fig. 7.4 Threshold reporting concept
7 Dynamic Bandwidth Allocation for EPONs
159
REPORT to the OLT. Currently, there are six packets backlogged in the buffer. The accumulative sum of the lengths of packets within each threshold is also calculated and shown by parameters Mi in the figure. For example, the sum of the lengths of packet 1 (the head of line packet) and packet 2 is M1 bytes which is less than T1 ; the sum of the lengths of packets 1, 2, and 3 is M2 bytes which is less than T2 , . . ., and finally the sum of the lengths of all six packets is M4 bytes which is less than T4 . If the ONU is to send multiple QRs for this queue in the current REPORT message, the value of the first QR is set to M4 (to request for maximum grant for this queue), and the remaining QRs are set to M3 , M2 , and M1 , respectively. Hence, with MPCP, a QR conveys the total length of packets waiting at the head of the queue with a value less than a specified threshold. In general, the thresholds can be spaced in equal distances from each other in order to provide a uniform coverage of data frames along the entire length of the buffer (queue). However, if it is desired to cover the buffer nonuniformly, that may be achieved by varying the spacing of the thresholds. By spacing the thresholds closer to each other at any region in the buffer, more number of QRs with higher degree of resolution can be produced for that region.
7.3.2 GATE Frame Format The OLT controls an ONU’s transmission by assigning transmission grants. The transmission window of an ONU is indicated in the GATE message where start time and length are specified. An ONU will begin transmission when its local time counter matches the grant start time indicated in the GATE message. An ONU will conclude its transmission when the grant length interval has elapsed. Up to four outstanding grants may be issued to each ONU. Sufficient margin has to be provided to allow the ONU to turn on and then turn off its laser. For this reason, the laser On and Off times are included in and thus consume part of each grant’s length. Figure 7.5 depicts the format of a GATE PDU as defined in the IEEE 802.3ah standard. 8 bytes Preamble
6 DA
6 SA
OC = Opcode TS = Timestamp FL = Flags GS n = Grant n Start time GL n = Grant n Length
Fig. 7.5 MPCP GATE message format
2
2
L O T C
4 TS
1 F L
0 to 39 Grants
4 P A FCS D
0/4 0/2 0/4 0/2 0/4 0/2 0/4 0/2 GS GL GS GL GS GL GS GL 4 1 1 2 2 3 3 4
160
Hassan Naser and Hussein T. Mouftah
• Opcode (OC). The opcode for the GATE PDU is 00-02 in hexadecimal. • Timestamp (TS). This encodes the OLT clock at the time of transmission of the GATE message. It allows the ONU to adjust its local clock counter to the OLT clock. • Flags (FL). An 8-bit flag register that holds the following flags: – Number of Grants. The first three bits of the Flags field (bits 0–2) are referred to as the Number of Grants field. It is used to specify the number of grants included in the Grants field of this GATE message. Although with a 3-bit field the range of permitted grants is 0 through 7, the maximum number of grants that can be issued to each ONU is currently 4. – Discovery. Bit 3 of the Flags field is referred to as the Discovery field. When this bit is set, it indicates that the included grants would be used for the device discovery process. In this case, the GATE message is referred to as discovery GATE message. – Force REPORT. The remaining 4 bits of the Flags field (bits 4–7) are referred to as the Force REPORT flags. If flag bit n (4 ≤ n ≤ 7) is set to 1, the ONU must issue a Force REPORT at the corresponding transmission opportunity indicated in the Grant number n − 3. • Grants. Currently, up to four grants can be included in the Grants field, as shown in the bottom part of Fig. 7.5. Each Grant consists of the following fields: – Grant Length (GL). A 2-byte unsigned field that encodes the length of the signaled grant in 16 bit time increments. – Grant Start Time (GS). A 4-byte unsigned field whose value would be compared with the ONU’s local clock to calculate the start of the grant. When multiple consecutive grants are specified in the same GATE message, the following condition must be satisfied: Grant number n Start Time must be less than Grant number n + 1 Start Time.
7.3.3 Ranging Process “Ranging” refers to the process used in MPCP to adjust the local clock counter for different ONU stations communicating with the OLT. The ranging process is also used to calculate the round trip time (RTT) from the OLT to each ONU. When an ONU receives an MPCP message, it sets its local clock register to the value in the timestamp field of the message. When the OLT receives an MPCP message, it uses the received timestamp value to calculate the RTT between itself and the source ONU. The RTT is calculated as the difference between the OLT local time and the incoming timestamp. This calculation can be demonstrated through the use of the following example. Figure 7.6 gives an example of the use of MPCP GATE and REPORT messages to measure RTT between the OLT and an ONU. The OLT sends a GATE message to the ONU at its local time T0 . The timestamp in the GATE message has been set
7 Dynamic Bandwidth Allocation for EPONs
161
T0
T4 TRESPONSE OLT Local
OLT
Time Timestamp = T 0
Timestamp = T 3
ONU Local Time
ONU
T1
T2
T0
T3
T DOWN
T WAIT
before adjustment ONU Local Time T UP
after adjustment
Fig. 7.6 Round trip time (RTT) measurement
to T0 accordingly. Now assume that the GATE message carries one grant and the grant start time has been set to T3 in reference to the OLT’s local time. The ONU receives the GATE message at its local time T1 . It adjusts its local clock to show time T0 . For convenience, the figure shows the ONU’s timing diagram before and after the local time adjustment. The ONU sends a REPORT message at time T3 , which is equivalent to time T2 before time adjustment. The REPORT message shows the timestamp T3 . The OLT receives the REPORT message at its local time T4 . In the figure, TDOWN and TUP represent the propagation delay in the downstream direction and upstream direction, respectively. The calculation of the RTT follows the following logic: RTT = TDOWN + TUP = TRESPONSE − TWAIT = (T4 − T0 ) − (T2 − T1 ) = (T4 − T0 ) − (T3 − T0 ) = T4 − T3 As shown, the RTT is the difference between the OLT’s local time (T4 ) at the time of the REPORT arrival and the timestamp included in the REPORT message (T3 ).
7.3.4 Automatic Device Discovery Auto-discovery is the process by which the OLT searches, identifies, and registers newly connected or off-line ONU devices. The OLT allocates and assigns a logical address called logical link ID (LLID) to the ONU as part of the registration. The LLID identifies a port (interface) on the OLT that will be used to send or receive data to or from the ONU. The OLT also gathers basic information about the ONU device it has discovered. This information is used to determine RTT and to negotiate
162
Hassan Naser and Hussein T. Mouftah
system/optical parameters including laser turn-on/off times in order to achieve best performance. The auto-discovery process is scheduled to occur periodically, during which time the OLT polls the off-line ONUs by broadcasting a discovery GATE message. The message includes the starting time and length of the discovery time interval (discovery window). The auto-discovery in EPON is a contention access method in that all off-line ONUs attempt to send a registration request (REGISTER REQ) message to the OLT, as soon as their clocks match the starting time value indicated in the discovery GATE message. A collision will occur if two (or more) off-line ONUs attempt to transmit a REGISTER REQ message at the same time. Indeed, discovery windows are the only times where multiple ONUs can access the EPON simultaneously, causing transmission collision. In order to reduce transmission collisions, the off-line ONUs do not transmit REGISTER REQ immediately when their clocks match the starting time value. Instead, they wait (backoff) for a random time period between zero and the discovery window size. When an ONU’s backoff timer reaches zero, it is allowed to transmit a REGISTER REQ message. In the message, the ONU includes its MAC address and the maximum number of pending grants. Upon receipt of a valid REGISTER REQ message, the OLT registers the ONU by assigning a new port identifier (LLID), and bonding the LLID to the MAC control client at the OLT to which the packets will be directed. Next, the OLT transmits a REGISTER message to the newly discovered ONU, which contains the ONU’s LLID, and some timing information required for synchronization. At this point in time, the OLT is ready to schedule the ONU for access to the PON. To inform the ONU that the registration is complete, the OLT transmits a standard GATE message to the ONU requesting the ONU to acknowledge the registration and accept the negotiated terms and parameters. If accepted, the ONU will respond by transmitting a REGISTER ACK to the OLT. Upon receipt of the REGISTER ACK, the discovery process for that ONU is complete, and they can start exchanging user data. Figure 7.7 illustrates the message exchanges between the OLT and an off-line ONU during the auto-discovery period.
7.4 Dynamic Bandwidth Allocation Algorithms The MPCP protocol is not concerned with any particular bandwidth allocation algorithm in EPON; it merely facilitates the implementation of these algorithms in EPON. Hence, special bandwidth allocation algorithms must be designed to support for broad spectrum of applications with quality of service (QoS) requirements and to offer fair, deterministic services to end-users. Indeed, many dynamic bandwidth allocation (DBA) algorithms have been proposed in the literature for TDM-based EPON. These architectures are designed to allocate timeslots (bandwidth) to ONUs dynamically. In order to support QoS, these architectures have employed two independent scheduling mechanisms: inter-ONU scheduling, which arbitrates the upstream bandwidth by allocating appropriate timeslots to each ONU;
7 Dynamic Bandwidth Allocation for EPONs
163
and intra-ONU scheduling, which divides the bandwidth allocated to each ONU among the different QoS classes. Two key characteristics of these DBA architectures are the time and place that inter-ONU and intra-ONU scheduling decisions are made. With some of these architectures (such as in [12] and [15]), the OLT makes scheduling decision for each ONU instantaneously upon receiving the REPORT message from that ONU. With some others (such as [11]), the OLT makes scheduling decisions for all active ONUs together. REPORT messages from all of these ONUs must be received before the scheduling is performed. The term decision place refers to which node (OLT, ONU) in the network is responsible for the inter-ONU and intra-ONU scheduling decisions. With distributed DBA architectures [10], the OLT is excluded from the implementation of both of the scheduling mechanisms (inter-ONU and intra-ONU). Instead, all of the ONUs collectively perform timeslot assignments by integrating both scheduling mechanisms and by simultaneously executing these algorithms. With centralized DBA architectures [11] and [12], the ONUs are excluded from the implementation of timeslot assignment. Instead, both of the scheduling mechanisms are integrated at the OLT, which has global knowledge of the state of the entire network. A third alternative is proposed, we call hierarchical DBA architecture, used in [13] through [17], in which the inter-ONU scheduling (timeslot assignment) is made by the OLT, and is then signaled to each ONU. Then the ONU performs the intra-ONU scheduling by dividing the bandwidth allocated to that ONU among the different QoS classes. We leave a more detailed discussion of these DBA architectures and their QoS supports to the accompanying chapter in this book. However, for completeness of the discussion and to understand how the MPCP protocol facilitates the implementation of DBA algorithms in EPON, a brief overview of one such algorithm is given below. ONU
OLT Discovery GATE
GATE starting time Random delay REGISTER_REQ
REGISTER GATE
REGISTER_ACK
Fig. 7.7 Auto-discovery message exchange
Discovery Window
164
Hassan Naser and Hussein T. Mouftah
7.4.1 Class-of-Service Oriented Packet Scheduling (COPS) Algorithm In our recent publication [11], we have presented a new centralized DBA scheme for EPON, subject to requirements of fairness, efficiency, and cost. In this scheme, we effectively employed the concepts of credit pooling and weighted bandwidth sharing policy that enable the OLT to partition the upstream bandwidth among different classes, and to prevent ONUs from monopolizing the bandwidth. In order to meet the above requirements, a joint-ONU interval-based packet scheduling algorithm, referred to as COPS (class-of-service oriented packet scheduling), was proposed. The COPS algorithm differentiates itself from other DBA (packet scheduling) algorithms proposed for EPON by employing credit pooling bandwidth sharing, rather than priority-based bandwidth sharing employed in other proposals. It has been generally shown that priority-based bandwidth sharing (queuing) will lead to unfair distribution of bandwidth among priority queues, because poorly behaved traffic in a high-priority queue can preempt well-behaved traffic in low-priority queues. The OLT executes the COPS algorithm to determine which packets the subtending ONUs must transmit and when. This information is conveyed to each ONU via a GATE message. The COPS algorithm is executed at the beginning of every transmission cycle. The transmission cycle can have a variable length with certain upper bound, denoted by Tmax . The time to execute the scheduling algorithm (Tsch ) is a small fraction of Tmax . During its transmission slot, an ONU transmits packets stored in its buffer in the form of Ethernet frames. The ONU must also send a REPORT message to the OLT during its transmission slot. The REPORT message carries up to 13 queue reports (mandated by the MPCP) for all CoSs in the ONU. The schematic diagram of the COPS system is shown in Fig. 7.8. The OLT maintains N credit pools, where N is the number of classes of service supported in the (U1, r1)
(V1, R1) 11
ONU 1 ONU 1 Credit Pool
M1
WRR Arbiter
CoS 1 Credit Pool
Bmax
Credit Queue (VN, RN)
(UM, rM)
Credit Processing Order
1N MN
ONU M ONU M Credit Pool
Fig. 7.8 COPS system
CoS N WRR Credit Pool Arbiter
7 Dynamic Bandwidth Allocation for EPONs
165
system. Each credit pool is used to enforce a long-term average rate of class of service (CoS) traffic, transmitted from all ONUs. The credit pool for CoS n has two parameters: size (Vn ) and average rate of credit arrivals (Rn ). The OLT also maintains M credit pools; each is used to control the usage of the upstream bandwidth by an ONU. The credit pool for ONU m has two parameters: size (Um ) and average rate of credit arrivals (rm ). The Um value determines the maximum number of bytes that ONU m can transmit during any particular cycle. At the beginning of every transmission cycle, all of the credit pools are initialized with their preconfigured Vn and Um bytes for the upcoming cycle. A request from a given ONU for a given CoS is honored if there exist a sufficient amount of credits in both the ONU and CoS credit pools. A fair distribution of CoS n credits between ONUs is established through the assignment of weights μmn . The number of CoS n credits to be allocated to an ONU m is determined proportionally according to its weight μmn . To enforce these weights, a weighted round robin (WRR) arbiter is placed before each CoS credit pool to ensure that the contending ONUs for this CoS will receive a fair share of the CoS credit pool, based on their weights. In Fig. 7.8, a credit queue is shown that stores the credits calculated for all ONUs during the scheduling time. Each entry of the credit queue contains three information elements: ONU I.D., CoS I.D., and the size of the credits given to the (ONU, CoS) pair. At the end of every scheduling time, the total size of credits stored in the credit queue will not be more than the number of data bytes that can be transmitted upstream by all ONUs during a maximum transmission cycle Tmax . After the scheduling is completed, the OLT will remove the colored1 grants (for each ONU) from the credit queue and place them in a GATE message sent to that ONU. The standard MPCP GATE message indeed supports the transmission of colored grants proposed only with the COPS scheme (but not with other schemes). This is achieved through a particular interpretation of the Flags and Grants fields of the MPCP GATE message illustrated in Fig. 7.5. As we described there, the first three bits of the Flags field (called Number of Grants bits) are used to specify the number of grants included in the Grants field of this GATE message. Therefore, theoretically, up to eight different grants can be issued to each ONU during each cycle. In our interpretation, each grant is associated with one CoS, and therefore, up to eight different CoSs (colored grants) can be supported in the system. Given the mechanisms discussed thus far, the OLT uses the following algorithm (COPS) to calculate grants for each ONU, whereby, which the packets will be transmitted. The COPS algorithm includes two rounds of execution. The first round is mandatory that gives rise to the distribution of most of the available grants (credits in the credit pools) between ONUs. At the end of this round, some amount of grants may still remain in the credit pools. The second round is designed to redistribute the remaining grants between ONUs. Although this round is optional and can be eliminated, it will lead to a work conserving scheduling service. Round 1: The OLT always begins processing grants for CoS1 first. In general, the OLT will begin processing CoS n grants only after all CoS (n − 1) grants have 1 A “colored” grant here means that how many bytes exactly from a particular CoS (color) the ONU must transmit.
166
Hassan Naser and Hussein T. Mouftah
been calculated. In order to calculate CoS n grant, the OLT takes the first ONU (say ONU m) determined by the WRR arbiter n and examines the REPORT message received from this ONU in the previous cycle. The REPORT message can include 0, 1, or at most 13 queue reports for CoS n, sorted in descending order of requested grant size. The OLT chooses the first queue report (if there is any) and will fully grant this request if there exist an equivalent amount of credits in the CoS n and ONU m credit pools. If either of these credit pools does not have enough credits, the OLT will consider the next queue report (for CoS n) in the REPORT message until a match is found. The OLT then calculates a grant for the next ONU in the WRR n database using the above procedure. Round 2: At the beginning of this round, the OLT determines the total amount of unused (remaining) credits in the CoS credit pools. These unused credits can be recollected and distributed between ONUs that have received partial or no grant in Round 1. The formula for redistributing these unused credits are given in [11] and omitted here for space. Once this is done, the OLT executes the same procedure as in Round 1 to distribute these unused credits between ONUs, but this time it only considers the ONUs flagged with partial or no grant (from Round 1).
7.5 Summary This chapter studied the Multi-Point Control Protocol (MPCP), which is the IEEE 802.3’s standardized medium access control protocol for the control of information transfer and for supporting dynamic bandwidth allocation in Ethernet Passive Optical Networks (EPONs). The concept of threshold queue reporting was presented, which is used by the end-user devices (ONUs) to report detailed information regarding the status of the individual queues to the network. The ranging process and automatic discovery of newly connected or off-line ONU stations were also given. To demonstrate how the MPCP facilitates the implementation of dynamic bandwidth allocation algorithms in EPON, a brief overview of one such algorithm was also given.
References 1. G. Kramer, G. Pesavento, “Ethernet Passive Optical Network (EPON): Building a NextGeneration Optical Access Network,” IEEE Communications Magazine, 40(2), February 2002, 66–73. 2. J.D. Angelopoulos, H.-C. Leligou, T. Argyriou, S. Zontos, “Efficient Transport of Packets with QoS in an FSAN-Aligned GPON,” IEEE Communications Magazine, 42(2), February 2004, 92–98. 3. S. Blake, D. Black, M. Carlson, E. Davies, Z. Wang, W. Weiss (1998) An Architecture for Differentiated Services. RFC 2475 The Internet Engineering Task Force (IETF) Homepage [Online]. Available: http://www.ietf.org/rfc/rfc2475.txt. Accessed 09 April 2008.
7 Dynamic Bandwidth Allocation for EPONs
167
4. G. Kramer, B. Mukherjee, G. Pesavento, “Ethernet PON (ePON): Design and Analysis of an Optical Access Network,” Photonic Network Communications, 3(3), 2001, 307–319. 5. I. Van de Voorde, C. Martin, J. Vandewege, X.Z. Qiu, “The SuperPON Demonstrator: An Exploration of Possible Evolution Paths for Optical Access Networks”, IEEE Communications Magazine, 38(2), February 2000, pp. 74–82. 6. IEEE 802.3ah Task Force homepage [Online]. Available: http:// www.ieee802.org/3/efm 7. F. Effenberger, G. Kramer, B. Hesse, “Passive Optical Networking Update,” IEEE Communications Magazine, 45(3), Mar 2007, S6–S8 8. IEEE P802.3av Task Force Homepage [Online]. Available: http://grouper.ieee.org/ groups/802/3/av/index.html 9. IEEE Standard 802.3-2005, Part 3, Section 5, Clause 64, Multipoint MAC Control Specifications [Online]. Available: http://standards.ieee.org/ getieee802/802.3.html. Accessed 06 May 2008. 10. S.R. Sherif, A. Hadjiantonis, G. Ellinas, C. Assi, M.A. Ali, “A Novel Decentralized EthernetBased PON Access Architecture for Provisioning Differentiated QoS,” Journal of Lightwave Technology, 22(11), November 2004, 2483–2497. 11. H. Naser, H. T. Mouftah, “A Joint-ONU Interval-Based Dynamic Scheduling Algorithm for Ethernet Passive Optical Networks,” IEEE/ACM Transactions on Networking, 14(4), August 2006, 889–899. 12. H. Naser, H. T. Mouftah, “A Fast Class-of-Service Oriented Packet Scheduling Scheme for EPON Access Networks,” IEEE Communications Letters, 10(5), May 2006, 396–398 13. M. Ma, Y. Zhu, T. H. Cheng, “A Bandwidth Guaranteed Polling MAC Protocol for Ethernet Passive Optical Networks,” Proceedings IEEE INFOCOM’03, Vol. 1, San Fran-cisco, CA, March–April 2003, pp. 22–31. 14. G. Kramer, B. Mukherjee, G. Pesavento, “IPACT: A Dynamic Protocol for an Ethernet PON (EPON),” IEEE Communications Magazine, 40(2), February 2002, pp. 74–80. 15. G. Kramer, B. Mukherjee, S. Dixit, Y. Ye, R. Hirth, “On Supporting Differentiated Classes of Service in EPON-Based Access Networks,” Journal of Optical NetWorking, 1(8/9), August 2002, 1–20. 16. D. Nikolova, B. Van Houdt, C. Blondia, “Dynamic Bandwidth Allocation Algorithms in EPON: A Simulation Study,” Proceedings of Opticomm’03, Oct 2003, pp. 369–380. 17. C.M. Assi, Y. Ye, S. Dixit, M.A. Ali, “Dynamic Bandwidth Allocation for Quality-of-Service Over Ethernet PONs,” IEEE Journal on Selected Areas in Communications, 21(9), November 2003, 1467–1477.
Chapter 8
Quality of Service in Ethernet Passive Optical Networks (EPONs) Ahmad Dhaini and Chadi Assi
Abstract Ethernet passive optical networks (EPONs) are designed to deliver multiple services and applications, such as voice communications (VoIP), standard and high-definition video (STV and HDTV), video-conferencing (interactive video), IPTV, and data traffic access network. These differentiated services (DiffServ) carry strict bandwidth and delay requirements, as well as jitter sensitivity. Supporting these bundled services in EPONs faces many challenges that require extensive research and studies. As a result, many techniques and methods have been presented to facilitate the latter. The process of supporting these services is so-called quality-of-service (QoS) support, which refers to the ability to provide different priority to different applications, users, or data flows, or to guarantee a certain level of performance to a data flow. In this chapter, we overview the tools and techniques that were presented to date aiming to provide “fair” and “efficient” QoS support in EPONs. More specifically, we overview the various intra-ONU bandwidth scheduling schemes that have been proposed and we present our intra-ONU scheduling solution, namely modified-DWRR (M-DWRR), that is based on the deficit-weighted round robin (DWRR) scheduling mechanism. M-DWRR proved to achieve adaptive fairness among different classes of services. Next, we overview the different dynamic bandwidth allocation (DBA) schemes that enable QoS support from an OLT perspective. Here, the OLT is responsible for allocating bandwidth for each class of service and hence no intra-ONU scheduling is required. Finally, we discuss the quality-of-service protection issue in EPON; which is quite interesting and challenging due the time division multiple access (TDMA) nature of EPON. To resolve this issue, we present the first admission control framework in EPON that will allow for both quality-of-service protection and efficient bandwidth allocation and reservation. Ahmad Dhaini Department of Electrical and Computer Engineering, University of Waterloo, Waterloo, QC, Canada, e-mail:
[email protected] Chadi Assi Concordia University, Montreal, QC, Canada, e-mail:
[email protected]
A. Shami et al. (eds.), Broadband Access Networks, Optical Networks, c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-92131-0 8,
169
170
Ahmad Dhaini and Chadi Assi
8.1 Introduction Quality of service (QoS) refers to the capability of a network to provide better service to selected network traffic over various technologies, including Frame Relay, asynchronous transfer mode (ATM), Ethernet and 802.1 networks, SONET, and IProuted networks that may use any or all of these underlying technologies. The primary goal of QoS is to provide priority including dedicated bandwidth, controlled jitter1 and latency (required by some real-time and interactive traffic), and improved loss characteristics. It is also important to make sure that providing priority for one or more flows does not make other flows fail. QoS technologies provide the elemental building blocks that will be used for future business applications in campus, WAN, and service provider networks. Broadband access providers view QoS and multimedia-capable networks as an essential ingredient to offer residential customers video-on-demand, audio-ondemand, voice over IP (VoIP), and high-speed Internet access. Furthermore broadband access networks, and EPON in particular, are especially appropriate for peer-to-peer applications (P2P) (which permit files to be interchanged through the Internet). It was shown that P2P applications represent a high fraction of the upstream traffic in hybrid fiber-coax cable access network [1]. Unlike early file sharing applications (such as Napster and Gnutella), many recent P2P applications include live media broadcasting, high bandwidth content distribution, and realtime audio conferencing and require high-performance access networks in order to deliver satisfying QoS to the users [2]. Table 8.1 lists the various services that are supported in access networks. As shown, different services have different requirements that must be respected to meet the users “expectations”, that is, the service level agreement (SLA). The SLA is a formally negotiated agreement between two parties. It is a contract that exists between customers and their service provider, client or between service providers. It
Table 8.1 Services supported by access networks Service
Type
Telephony
Switched
4 kHz
4 kHz
ISDN
Switched
144 kbps
144 kbps
6 MHz or 2–6 Mbps
0
Video broadcasting Broadcast Interactive video
Switched
6 Mbps
Small
Internet access
Switched
Some Mbps
Small (initially)
6 Mbps
6 Mbps
1.5 Mbps–10 Gbps
1.5 Mbps–10 Gbps
Video-conferencing Switched Enterprise services
1
Downstream bandwidth Upstream bandwidth
Switched
Jitter: The interval between successive pulses or phase of successive cycles (in this case packets).
8 Quality of Service in EPONs
171
records the common understanding about services, priorities, responsibilities, guarantee, and such – collectively, the level of service. For example, it may specify the levels of availability, serviceability, performance, operation, or other attributes of the service like billing and even penalties in the case of violation of the SLA. Since EPON is promising to be the best wired-access cost-effective network panacea, it is also supposed to support these services. Moreover, some of these services require a large amount of bandwidth (e.g., video-conferencing); hence, an arbitration of the users is required in order to deny a user/node from monopolizing the EPON limited bandwidth (1 Gbps). In particular, the access node has a key role in providing QoS for upstream traffic. Additionally, the access node has a full knowledge of the access line characteristics for downstream flows and can therefore prevent downstream congestion of high priority traffic. A variety of techniques are used to optimize network load, application performance, and fairness among services and applications. These services are mapped into the access nodes (here, ONUs) as “classes”. Each class of service (CoS) is identified by the service’s traffic requirements and properties.
8.1.1 IEEE 802.1D Support for Classes of Service To support CoS, Ethernet networks must be able to classify traffic into CoS and provide differentiated treatment to each class. This was achieved by an introduction of two new standard extensions (1) P802.1p (a supplement to MAC bridges): traffic class expediting and dynamic multicast filtering (later merged with P802.1D) and (2) P802.1Q virtual bridged local area networks. P802.1Q defines a frame-format extension allowing Ethernet frames to carry a priority information field in their header. The standard distinguishes the following traffic classes: 1. Network control: Characterized by a “must get there” requirement to maintain and support the network infrastructure. 2. Voice: Characterized by less than 10 ms delay, and hence maximum jitter. 3. Video: Characterized by less than 100 ms delay. 4. Controlled load: Important business applications subject to some form of admission control, which can be either achieved by a preplanning of the network requirement or by bandwidth reservation per flow at the time the flow admitted to the network. 5. Excellent effort (CEO’s best effort): As called, the best-effort-type services that an information services organization would deliver to its most important customers. 6. Best effort (BE): Internet traffic as we know it today (data traffic). 7. Background: Bulk transfers and other activities that are permitted on the network without affecting existing applications nor monopolizing the shared bandwidth.
172
Ahmad Dhaini and Chadi Assi
Note: If a bridge or a switch has less than seven queues, some of the traffic classes are grouped together. A variety of techniques are used to optimize network load, application performance, and fairness among applications. As discussed in the previous chapter, EPON engages two wavelengths in the shared channel between the ONUs and the OLT. That is, traffic travels in both the upstream and the downstream directions. Hence, QoS operation should be considered for both directions.
8.1.1.1 Upstream QoS In the upstream direction, flows/streams arrive from end-users “undefined” as bursts at constant or variable bit rates. Hence, the following operations are conducted before transmitting these packets to the central office node. • Traffic classification and filtering is achieved using access control lists (ACL),2 which helps identifying and classifying the traffic flows. Furthermore, filtering helps enforcing security policies by removing fraudulent traffic that could potentially compromise the overall network behavior/performance. • DSCP3 or p-bit (re)marking to values that can be trusted by the network elements in the aggregation network when providing per-flow QoS processing. The DSCP is typically used by QoS-enabled IP service routers, whereas the p-bits can be used by QoS-enabled Ethernet switches. If the DSL modem or home gateway has performed a traffic classification process, then the access node can use the DSCP or p-bit marking to perform per-flow QoS processing as well as remarking, if required. Otherwise the access node traffic filtering process classifies the traffic into QoS sub-flows, and then (re)marks the packets accordingly. • Ingress policing enforces the traffic contracts (SLAs) that specify how much traffic users can send to the network. A policer may apply to a full access line or to a set of QoS flows matching a certain classification. Policing may take into account the drop precedence coded in the DSCP or p-bits; out-of-profile traffic can be filtered/dropped or can be remarked with a higher drop precedence. • Traffic forwarding to the right egress interface is usually based on either the destination media access control (MAC) address and/or the destination IP address. • Per-QoS class queuing and scheduling on the access node based on the DSCP or p-bits: a set of egress queues is present and each QoS class is mapped to the appropriate queue. The scheduling mechanisms determine the way these packets are selected from their queues. Different schedulers (e.g., strict priority or round robin) can provide different functionality on the same set of queues.
2 Access control list (ACL): A sequence of patterns to match packets from a traffic stream, and the associated filtering actions to be taken when a packet matches a certain pattern. Patterns could be based on fields such as the Ethernet header, IP header, or UDP/TCP port numbers. 3 Differentiated services code point (DSCP): a 6-bit encoding that indicates the type of service (TOS) in the IP packet header.
8 Quality of Service in EPONs
173
8.1.1.2 Downstream QoS In the downstream direction, frames arrive at the access node with DSCP or p-bits that are properly marked by the application servers or service routers. The latter also perform ingress policing, which implies that the access node does not need to do this. After the forwarding decision, the following steps can be performed: • Egress rate limiting: Similar to ingress policing, this is used to enforce traffic contracts that are associated with the services to which the users have subscribed. • Per-QoS class queuing and scheduling on the access lines based on the DSCP or p-bits: downstream it is crucial to use multiple QoS queues per access line and to map traffic to a specific queue based on the DSCP or p-bits. The scheduling mechanisms combined with the use of queue management features, such as (weighted) random early detection (WRED), determine the exact treatment received by packets in the various queues.
8.1.2 CoS Support in EPON Current EPONs support diverse applications, various traffic sessions are aggregated into a limited number of classes to be serviced with differentiated services (DiffServ). These services are classified as follows: best effort (BE) “data” traffic, assured forwarding (AF) traffic such as variable-bit-rate (VBR) video stream, and expedited forwarding (EF) traffic used to emulate point-to-point (P2P) connections or real-time services, such as voice over IP (VoIP). The high-priority class is EF, which is delay-sensitive and requires bandwidth guarantees. The medium-priority class is AF, which is not delay-sensitive but requires bandwidth guarantees. The low-priority class is BE, which is neither delay-sensitive nor bandwidth guaranteed. This information is encapsulated in the TOS of DSCP header. Note that ONUs can support up to eight priority queues [3]. To provide QoS in EPONs, bandwidth management on the upstream channel is essential for successful implementations of these types of networks. This is done in two ways: • Intra-ONU scheduling: The process of selecting packets that are buffered into the various priority queues to be transmitted in the next transmission window. • QoS-enabled dynamic bandwidth allocation: A QoS variant of the DBA presented in the previous chapter. Here, the OLT is responsible for allocating bandwidth for each class of service in each of the connected available ONUs. In addition, to enable QoS protection and bandwidth guaranteed, admission control seems to be a necessity for QoS support in EPONs. Admission control is the ability to judge whether the network has enough resources available to accept the connection, and then either accepts or rejects the connection request. We discuss these three subjects in detail in the following sections.
174
Ahmad Dhaini and Chadi Assi
8.2 Intra-ONU Scheduling 8.2.1 Incoming Traffic Handling Operation The ONU uses the traditional “QoS classifying and scheduling” operation, that was detailed in the previous section, to handle the users’ incoming flows. As shown in Fig. 8.1, upon receiving traffic “flows” from the registered subscribers, the ONU performs three main operations before transmission in the upstream channel. First, it classifies every newly arriving packet using a “packet-based” classifier. Next, and before placing packets in the corresponding priority queues, the ONU decides whether a packet should be admitted depending on the adopted traffic policing (admission control) mechanism (e.g., leaky bucket). Finally, it selects packets from its queues, depending on the intra-/inter-ONU scheduling algorithm [4], and sends them to the OLT as “flows” in the assigned transmission window (TW). Note that “low complexity” is a key design goal for intra-ONU schedulers in order to keep the ONUs’ cost at minimum. A scheduler comprises an output link, an arbiter, and a set of input class queues. Here, the arbiter runs a scheduling algorithm to partition link capacity and deliver pre-determined (SLA) delay and throughput requirements to the input queues.
Fig. 8.1 Intra-ONU scheduling
8.2.2 Existing Solutions and Schemes Schedulers range in complexity and performance capabilities and most notable examples include priority schemes, non-work-conserving schemes; e.g., stop-andgo, weighted/hierarchical round robin (WRR/HRR), jitter earliest-due-date; and
8 Quality of Service in EPONs
175
work-conserving approximations of ideal generalized processor sharing (GPS), e.g., weighted fair queuing (WFQ/W F Q 2 ) [5], self-clocked fair queuing (SCFQ), starttime fair queuing (SFQ) [6]. Nevertheless, most schedulers have only been studied for intra-system roles, e.g., within ATM switches or QoS-enabled Ethernet/IP platforms. The further adaptation of these schemes in delayed, distributed EPON settings mandates more careful considerations. Moreover, there are two types of intra-ONU scheduling: strict and non-strict priority scheduling algorithms. In strict priority scheduling, a lower priority queue is scheduled only if all queues with higher priority are empty. However, this may result in a starvation for low-priority traffic or as dubbed in [4], “light-load penalty”. Non-strict priority scheduling addresses this problem by allowing reported packets (regardless of their priority) to be transmitted first as long as they are transmitted in the allocated TW. In other words, here, the transmission order of different priority queues is based on their priorities. As a result, all traffic classes have access to the upstream channel while maintaining their priorities; which enables fairness in scheduling and statistical multiplexing (by allowing link sharing among all classes of service). Note that inter-ONU control messages for allocating bandwidth to different ONUs are transmitted via the MPCP (multi-point control protocol) access protocol presented in Chapter 7. An ideal scheduler should allow for statistical multiplexing but guarantee a minimal portion of the available bandwidth to each priority queues. This is achieved by enabling link sharing among different priority queues. General processor sharing (GPS) achieves these goals for relatively small packets traffic (so-called fluid traffic). However, GPS is not applicable in practical systems with limited-size packets since a packet monopolizes the shared link while in service. This shortcoming was addressed in WFQ [5]. WFQ is basically a packet approximation of GPS that is derived from the ideal case bounded by the maximum packet size. It mainly calculates the packet’s start time and respective finish time under the ideal global positioning system (GPS). Consequently, packets are selected from the PQs in ascending finish time order. SCFQ and other schemes were proposed as a modified version of WFQ aiming to simplify the time calculation. However, these schemes lacked the approximation accuracy of the ideal GPS used in WFQ. Start time fair queueing (SFQ) [6] simplifies WFQ and reduces its computational complexity by calculating the packet-in-service’s start time instead of the finish time. To cope with the light-load penalty caused by applying strict priority scheduling technique, the authors of [4] proposed two methods. The first method involves a two-stage queueing process. Here, the incoming packets after sending the REPORT message are placed in the second-stage queue. Consequently, when a new GATE is received, the second-stage queue is emptied first. This, however, results in an increased average delay for all types of traffic. In the second method, the “afterreport” incoming traffic is estimated, and thus the grant window will be large enough to accommodate the newly arriving high-priority packets. Alternatively, in [7], the intra-ONU scheduler employs priority scheduling only on the packets that arrive before sending the REPORT message. This scheme eliminates the “light-load penalty” and allows all services to access the shared medium.
176
Ahmad Dhaini and Chadi Assi
Maeda et al. [8] proposed another scheduling discipline called priority with insertion scheduling (PIS) that transmits real-time packets when their delay bound does not exceed a defined threshold (i.e., as long as the packets can be delayed without any detriment). Kramer et al. [9] lately proposed a new hierarchical scheduler that fairly divides the excessive bandwidth resulting from lightly loaded ONUs among priority queues (PQs) from different ONUs. On the other hand, the authors of [6] proposed a new intra-ONU scheduling scheme named “modified start time fair queueing” (M-SFQ) that muses the performance of VBR traffic. Here, the scheduler selects for transmission of the queue with the minimal start time, derived from the head-of-line (HOL) packet in each queue, and synchronized with a global virtual time similarly to SFQ. M-SFQ mainly improves the end-to-end packet delay for assured services at the expense of other classes of service, which is not desirable in comparison to higher priority classes (EF). Moreover, M-SFQ tends to starve the best effort traffic to provide better QoS to AF and EF classes. The authors of [10] studied the shortcomings of M-SFQ and proposed a new intra-ONU scheduling scheme based on the famous token bucket (TB) traffic regulator for QoS; namely Modified-TB (M-TB). TB can reshape the traffic and limit the flow of some greedy traffic classes of service whose load are much larger than the preset limitation. M-TB algorithm assigns the bandwidth in two stages to each queue in the grant window of each ONU at a certain cycle. The advantages of M-TB over M-SFQ are its low complexity (which is a major requirement for intra-ONU schedulers). In addition, M-TB can guarantee both the priority and the fairness of the differentiated services while M-SFQ cannot. We have also studied the shortcomings of M-SFQ and proposed in [11, 12] a new intra-ONU scheduler, namely M-DWRR (modified deficit weighted round robin) that is based on the famous DWRR scheduling mechanism. M-DWRR defines weights “αi ” for each CoS queue i; where αi is adaptively set based on either the traffic requirements or the service level agreement (SLA). Moreover, M-DWRR gives the ONU enough flexibility to whether to set its weights in each cycle time or statically, which allows the support of QoS in EPON dynamic and bendable to meet traffic requirements. In the first scheduling pass, M-DWRR offers bandwidth to each P Q i according to its αi . The remaining bandwidth from the first scheduling pass is re-distributed in the second scheduling pass according to the same or different (depending on the traffic and queues requirements) sets of weights. M-DWRR has proven to outperform the M-SFQ scheme; as it ensures more fairness to lower priority traffic (e.g., best effort). In addition, the overall network throughput is improved than M-SFQ due to the elimination of best effort traffic starvation.
8.2.3 Intra-ONU Queue Management Priority queues at the ONU are defined/set by the EPON network architect with no size restriction. However, it is important to avoid high packet loss rates in the
8 Quality of Service in EPONs
177
Internet. When a packet is dropped before it reaches its destination, all of the resources it has consumed in transit are wasted. In extreme cases, this situation can lead to congestion collapse. Therefore, improving the congestion control and queue management algorithms seems important; however it was never addressed in EPONs due to the variable queue size. Nonetheless, nowadays, network planners and architects pay attention to cost reduction when designing a network. Moreover, software solutions and protocols have always been favored over the hardware ones, due to their lower cost (price and installation). For that reason, we will suggest a known queue management scheme that is widely used in other networks, that might be adopted in EPONs, aiming at reducing more the cost of installation of this costattractive solution (EPON). We investigated the weighted random early detection (WRED) [13] as a queue management scheme to improve the overall packet delay performance in EPON. WRED states that a periodic update should be made to the average queue size (AvQS) in order to achieve a periodic packet drop. Here, although a penalty is being paid, which is dropping packets, yet a higher gain is being achieved which is reducing the packet delay by consenting the queue to perform on a stable rate less than its maximum actual buffer size. The formula of queue update is given by AvQS = OAvQS × (1 − 2−n ) + CQS × 2−n
(8.1)
where OAvQS is the old average queue size, n is an exponential weight factor, a user-configurable value, AQS is the actual queue size, and CQS is the current queue size. Note that this update is performed each time a packet is dequeued. In EPON, the ONU buffer size is fixed (e.g., 10 Mbytes). However, the AvQS can be defined as a virtual and not an actual size. In other words, the ONU will create an “illusion” to the incoming packet that its actual size is the updated one. In this case, if the AvQS is “virtually” set to full then the incoming packet will be dropped, otherwise it will be enqueued. In this case, the ONU will make sure that its buffer is not being fully utilized and hence packets will be less delayed. If the value of n gets too high, WRED will not react to congestion. That is, packets will be dropped as if WRED was not in effect. Similarly, for low values of n, the average queue size closely tracks the current queue size.
8.3 QoS-Enabled Dynamic Bandwidth Allocation Algorithms (DBAs) Dynamic bandwidth allocation (DBA) is deployed at the OLT to assign transmission bandwidths for the different ONUs sharing the EPON network. DBA uses the services offered by the MPCP protocol to communicate assigned transmission windows to their appropriate ONUs. In the conventional DBA operation, as shown in Fig. 8.2, the OLT waits until all REPORTs from all ONUs
178
Ahmad Dhaini and Chadi Assi Perform DBA cycle n –1
cycle n
EN
Data Data REP ORT N
2
GAT
E2 GAT E1
GAT
Data Data REP ORT Data 1 Data REP ORT
OLT
ONU1 ONU2
ONUN
Fig. 8.2 Conventional DBA operation
are received4 in cycle n − 1 to perform the appropriate computation. Consequently, the OLT broadcasts MPCP’s GATE messages to grant transmission windows for cycle n.
8.3.1 Existing Solutions and Schemes Many schemes have been explored in the area to fairly allocate bandwidth for different classes of services. The authors in [7] presented a new method to “evenly” distribute the remaining excessive bandwidth over HL ONUs. This scheme results in a remarkable improvement of the network performance for different classes of services and also allows for statistical multiplexing traffic into unused bandwidth units. The authors also consider the option of reporting queue size using an estimator for the occupancy of the high-priority queue. Nevertheless, due to the unpredictable behavior of HL ONUs that varies from one cycle to another where a HL ONU tends to be a “slightly” HL one; and thus the allocated bandwidth is not being fully utilized, this uniform distribution of excessive bandwidth might not be the best possible solution. On the other hand, the authors of [14] proposed a new concept of DBA, where ONUs are divided into two sets, namely bandwidth guaranteed (BG) ONUs 4
Note: REPORT messages can be either sent at the end of the data transmission or at the beginning. However, if the latter is applied, the OLT might be receiving an out-of-date information from the ONUs.
8 Quality of Service in EPONs
179
(premium subscribers according to the SLAs) and non-bandwidth guaranteed (nonBG) ONUs (subscribers with best effort service). Here, the bandwidth guaranteed polling scheme (BGP) provides guaranteed bandwidth to BG ONUs while providing best effort services to non-BG ONUs. However, the proposed BGP cannot be standardized with the MPCP arbitration mechanism proposed by the IEEE 803.2ah task force for the reason that in the future emerging PON technologies, each ONU must be capable of provisioning differentiated services for different users requirements. Alternatively, the authors of [15] proposed a new DBA to support multimedia services in EPON. Here, incoming traffic to each ONU is buffered into one of the three priority queues (high, medium, and low). The sizes of these queues are reported to the OLT using an “upgraded” REPORT message. An inter-scheduler (i.e., at the OLT) is considered where the OLT, based on the priority queues sizes, issues grants separately. In particular, the DBA satisfies requests of flows by priority preference (high first, medium second, and low last). Then, if all flows are satisfied and additional bandwidth is still available, the remaining resources are distributed among all priority flows in the same manner. However, strict priority scheduling based on the traffic classes at the PON level may result in starvation of ONUs that have only low-priority traffic. To overcome this problem, the authors of [16] present a new DBA for multiservice access, namely DBAM. DBAM applies priority queuing to enqueue the EF, AF, and BE frames and gives preference to higher priority traffic. Priority-based scheduling is exploited to schedule the buffered frames, and the schedule interval is the time between sending REPORT messages. DBAM also employs class-based traffic prediction to take the frames arriving during the waiting time into account with dynamic and diverse bandwidth requests. In particular, an estimator credit α, which is the ratio of the waiting time of the ONU over the interval length, is estimated and then incorporated into the request for bandwidth of all BF, EF, and AF traffic. Multi-service access for different end-users is realized by means of classbased traffic estimation and SLA-limited bandwidth allocation. Furthermore, the authors of [17] presented a two-layer DBA where the total available bandwidth is allocated on two phases. Here, the OLT allocates bandwidth among different classes of services first, then among ONUs. This scheme provides a higher priority to the class-level quality of service over the ONU-level bandwidth guarantee. However, since subscribers are practically considered non-cooperative entities, ONU-level bandwidth guarantee should be considered first. In addition, the authors of [18] proposed a new GRANT “pre-allocation” mode for EF traffic named grant-before-report (GBR) and the traditional grant-after-report (GAR) mode for both AF and BE traffics. Here, the OLT divides the ONU transmission cycle into two sub-cycles; namely DBA sub-cycle (DBA-CL) reserved for EF traffic, and MPCP sub-cycle (MPCP-CL) reserved for AF and EF traffic. A new GATE scheduling mechanism is also presented in [19]. This mechanism allocates GRANTs based on the traffic priority rather than the ONU classification (e.g., Round Robin). Here, all high-priority traffic (from all ONUs) are granted first (in order to minimize its sensitive delay), and then low-priority traffic second. This algorithm can be also applied with any DBA.
180
Ahmad Dhaini and Chadi Assi
Alternatively, Nasser and Muftah [20] presented another method to schedule CoS traffic in EPON; namely, Class of service oriented packet scheduling (COPS). At the OLT, COPS uses two sets of “credit pools”, where each CoS of each ONU and well as each ONU are regulated. Mainly, COPS schedules the CoS with highest priority first then moves to the one(s) of lower priority and henceforth. In the first scheduling round, each ONU with traffic of the current CoS is granted up to the number of credits stored for that CoS. To mitigate the unused bandwidth for those grants that were not fully satisfied, a threshold reporting scheme is used. At the end of the first round, the unused credits are pooled together and in the next round these unused credits are distributed to the CoS per ONU pairs that were not fully satisfied. COPS provides lower delay for most CoS as compared to other famous inter-ONU scheduling schemes. Shami et al. [21], for instance, designed a new DBA, the so-called hybrid granting protocol (HGP), that mainly aims at improving the jitter performance of EF traffic. The hybrid part of HGP is that it sizes the ONU grants based on both the REPORT messages and a queue prediction mechanism, in order to accommodate all queued traffic at the point the grant begins. Moreover, each scheduling cycle is divided into two sub-cycles. The first one is used for the transmission of EF traffic and the second one for AF and BE traffic. Furthermore, to meet this cycle division, two separate REPORTs are sent from each ONU, as well as two grants are sent from the OLT to each ONU (one for EF traffic and the other for AF/BE traffic). Consequently, the grant of EF traffic is computed “earlier”, and as a result, EF traffic is not only prioritized but also protected from the fluctuations of other types of traffic; hence optimizing its jitter performance. Nonetheless, this optimization comes at the expense of more guard times per cycle caused by the separate grants of each ONU. To mitigate this drawback, a grant for AF/BE traffic is not sent if there is no waiting traffic, eliminating the need for a guard time for AF/BE traffic. The authors of [22] have also tried to improve the jitter performance of EF traffic by proposing a new hybrid slot size/Rate algorithm (HSSR). HSSR was able to stabilize the EF packet delay variation by not only fixing the cycle length but also fixing the position of EF traffic grants (at the beginning of the grant frame).
8.4 Quality-of-Service Protection and Admission Control in EPON 8.4.1 Preliminaries In order to provide sustainable QoS in the access network, bandwidth management on the upstream channel is essential. In order to support and protect the QoS of real-time traffic streams, one needs, in addition to bandwidth allocation and service differentiation, an admission control algorithm which makes decision on whether or not to admit a real-time traffic stream based on its requirements and the upstream
8 Quality of Service in EPONs
181
channel usage condition. The problem of QoS protection is significant because the bandwidth allocated by the OLT to one ONU can only be guaranteed for one cycle. Furthermore, appropriately controlling the admission of real-time traffic will prevent malicious users from manipulating the upstream channel by sending traffic into or requesting bandwidth from the network more than their SLA. Accordingly, admission control helps in protecting the QoS of existing traffic and admit new flows only if their QoS requirements can be guaranteed. Note that admission control has never been addressed in EPONs. However, the authors of [23] have proposed a DBA that is based on the BGP presented in the previous section, where admission control is conducted on an ONU basis rather than a flow/stream basis. In other words, an ONU is either accepted as a bandwidth guaranteed (BG) ONU or as a best effort non-BG ONU. In current EPON networks, the bandwidth of the upstream channel is shared among different ONUs using a TDMA scheme; the OLT allocates a transmission bandwidth for every ONU either equals to its bandwidth request from the previous cycle, or equals to the minimum bandwidth guaranteed (Bmin ), or equals to the minimum bandwidth guaranteed plus a surplus bandwidth that may remain unused in the cycle. Clearly, the bandwidth of one ONU cannot be guaranteed and may vary from one cycle to another according to the load at other ONUs. Bandwidth reservation resolves the uncertainty in allocating enough bandwidth resulting from the load variations at different ONUs. Hence, each ONU is required to reserve bandwidth for its real-time streams in order to satisfy their QoS requirements. Once this bandwidth is reserved, the OLT can no longer allocate it to other ONUs. Every ONU is guaranteed a new minimum bandwidth (Bmin ) and could be allocated up to a maximum bandwidth (Bmax ) in order to allow other ONUs to receive their share of the channel. Best effort (BE) traffic shares a fraction of the total cycle (Tcycle , Tcycle ≤ 2 ms in EPON networks [3]), e.g., α × Tcycle where α < 1. When α = 0, all the bandwidth of the upstream channel is used to transmit bandwidth-guaranteed traffic. The new cycle ((1 − α) × Tcycle ) is used to provide services for bandwidth guaranteed traffic. This new cycle in turn is divided into two sub-cycles (T1 , T2 ); the OLT computes the minimum bandwidth guaranteed (Bmin ) for each ONU using T1 (T −N ×T )×ξ (Bmin = cycle 8×N g , where ξ is the transmission speed of the PON in Mb/s, N is the number of ONUs, and Tg is the guard time that separates the TW for every ONUn and ONUn+1 ) and the ONU has total control over this bandwidth, while the bandwidth of the second sub-cycle is under the control of the OLT (please refer to Fig. 8.3 for a graphical elaboration, with N = 4). This new system enables us to implement a two-step admission control (AC); the first is a local AC at the ONU and the second is a global AC at the OLT (as explained later). Note that, although the minimum guaranteed bandwidth is under the control of the ONU, the scheduling of various ONUs is still done centrally at the OLT in order to achieve a collision-free access to the upstream channel. The two sub-cycles are selected of equal length; however, if T1 < T2 , then the OLT will have more control over the bandwidth with less bandwidth guaranteed per ONU. Conversely, the ONU is guaranteed more bandwidth, which may be un-utilized if the load at a particular ONU
182
Ahmad Dhaini and Chadi Assi
Fig. 8.3 Proposed cycle framework
is not high. Under our assumption of equal lengths for the sub-cycles, we set the maximum bandwidth that a highly loaded ONU can be allocated, Bmax = δ× Bmin . For example, when δ = 3, a highly loaded ONU may or could be assigned a maximum of 2× Bmin from the second half cycle and hence a total of 3× Bmin per cycle. For real-time applications, QoS metrics can be predefined in a policy control unit (PCU) and various thresholds could be specified/defined. For example, if the expected drop rate or the delay requirement for a certain flow/application cannot be be respected, the flow should not be admitted. Admitting such a stream will not only experience a degraded level of service but it will also degrade the QoS of existing streams. Alternatively, best effort traffic is never rejected and is always guaranteed min ). Hence, to achieve these goals, the following two rules a minimal bandwidth (BBE should not be violated before and after admitting a new real-time flow: 1. The QoS of each real-time stream (existing or new) should be guaranteed. min 2. The BE traffic throughput (BEThroughput ) ≥ BBE In every cycle, the ONU reports (using the MPCP protocol) to the OLT the BE buffer occupancy for bandwidth allocation in the next cycle; for real-time streams that the ONU has already admitted, the OLT will schedule only their transmission since the bandwidth of each stream has already been pre-determined and reserved and it is guaranteed per cycle for the rest of the lifetime of each stream.
8 Quality of Service in EPONs
183
8.4.2 Traffic Characteristics and QoS Requirements Clearly, an admission decision for a new flow arrival should be made according to both admission policies and QoS requirements supplied often by the application layer at the end-users. The set of parameters that characterize the traffic stream vary from one traffic class to another. For example, CBR traffic is non-bursty and characterized by its mean data rate (μ), which makes it quite predictable. With respect to QoS, CBR traffic requires stringent packet delays and delay variations (jitter). Alternatively, VBR traffic is quite bursty and may be characterized by the following parameters [24]: • • • •
Mean data rate (μ) in bits per second (bps). Peak arrival data Rate (σ ) in bits per second (bps). Maximum burst size (ρ) in bits. Delay bound (θ ) which is the maximum amount of time in units of microseconds allowed to transport a traffic stream (flow) measured between the arrival of the flow to the MAC layer and the start of transmission in the network.
Finally, BE traffic is bursty and requires neither delay requirements nor bandwidth guaranteed (note that network operators may set a certain minimum bandwidth that should be guaranteed for BE traffic, e.g., by appropriately adjusting the value of α). When these parameters are specified by the end-user, the problem left for the admission control unit (ACU, which is either at the ONU or at the OLT) is simply to determine whether a new stream should be admitted and whether its QoS requirements can be guaranteed while the QoS requirements for already admitted streams can be protected. For CBR traffic, the admission decision is straightforward; if the mean data rate can be supported, then the stream is admitted. Hence, enough bandwidth per cycle should be reserved to guarantee the stream data rate. Here, the average delay of CBR traffic is guaranteed to be bounded by the length of cycle. For VBR traffic, the ACU may decide to admit a stream only if its peak rate can be supported (for the best QoS) or may admit the stream as long as the mean data rate is available [24]. The former approach ends up admitting few streams and the latter approach barely supports QoS for bursty streams. Therefore, a guaranteed bandwidth based on the traffic parameters could be derived and we use a dual-token bucket for traffic regulation; this dual-token bucket is situated at the entrance of the MAC buffer and is associated with each stream. Figure 8.4 shows the dual-token bucket where the bucket size is calculated: B = ρ × (1 − μ/σ )
(8.2)
Accordingly, one can easily determine the arrival process of the stream passing through the filter [24]: A(t, t + τ ) = min(σ τ, B + μτ )
(8.3)
184
Ahmad Dhaini and Chadi Assi
Fig. 8.4 Dual-token bucket filter framework
where A(t, t + τ ) is the cumulative number of arrivals during (t, t + τ ). The arrival rate curve could be constructed from the above equation and is shown in Fig. 8.5. Therefore, the guaranteed rate for every real-time flow i can be easily derived from Fig. 8.5 using the distance formula [24]: gi =
ρi θi +
ρi σi
(8.4)
Since CBR traffic is deterministic and its peak rate is equivalent to its mean rate, therefore its bandwidth guaranteed will be: gi = μi
(8.5)
Consequently, a conventional rate-based admission control [25] can be used to determine whether a new stream can be admitted or not. For example, if S Tj W is the
Fig. 8.5 Guarantee bandwidth derivation graph [24]
8 Quality of Service in EPONs
185
bandwidth (bps) allocated and reserved for ONU j, then a new flow i + 1 could be admitted if hj j j gi ≤ S Tj W (8.6) gi+1 + i=1
where h j is the number of real-time streams (CBR or VBR) at ONU j. Now, the difficulty stems from the fact that in EPON the bandwidth assigned per ONU is not guaranteed, as mentioned earlier. Hence, we next propose a two-step admission control scheme that will provide bandwidth guaranteed for each CoS stream.
8.4.3 Local Admission Control (LAC) As we discussed earlier, each ONU is guaranteed a minimum bandwidth per cycle, Bmin . Hence, the ONU can locally perform rate-based admission control according to the bandwidth requirement of the new arriving flow and the bandwidth availj ability. For example, if g f is the guaranteed rate for the new flow, f , arriving at ONU, j, then the bandwidth requirement (in bytes) per cycle for the new flow is j j R f = g f × Tcycle . Therefore, this new flow will be admitted according to the following condition: hj j j R fi ≤ Bmin (8.7) Rf + i=1 j
where h j is the total number of flows already admitted by the ONU; R fi is the j
j
j
bandwidth requirement for a flow f i , R fi = g fi × Tcycle , and g fi is the guaranteed rate (bps) for the flow computed according to either Equation (8.4) or (8.5). The scheme classifies the arriving flow into BE traffic or real-time traffic. If it is BE, then the traffic is admitted. Otherwise, the ONU will derive the guaranteed rate and check Equation (8.7). If (8.7) holds, then the ONU will conditionally admit the flow and monitor its QoS for a predefined number of cycles (e.g., for 20 ms). If the QoS requirements of the newly admitted flow are satisfied and the QoS of existing flows remain intact, then the flow is admitted. Otherwise, the flow is dropped.
8.4.4 Global Admission Control (GAC) When a flow f cannot be admitted locally at the ONU (due to bandwidth insufficiency), the ONU reports the arrival of a new flow to the OLT. The OLT may admit this new flow only if there is bandwidth available in the second sub-cycle (T2 ) and if the ONU sending the request has not been allocated more than Bmax . Hence, the OLT maintains a variable for every ONU designating the bandwidth allocated so far h j j j j Ri , where Ri denotes the bandwidth guaranteed for to this ONU, Balloc = i=1
186
Ahmad Dhaini and Chadi Assi
already admitted h j flows for ONU j. The OLT maintains as well another variable that indicates the bandwidth that is still available, Bavail , (i.e., not committed yet) in T2 . The new flow may be admitted if the following two conditions (8.8) and (8.9) hold simultaneously: hj j j Ri ≤ Bmax (8.8) Rf + i=1 j
R f ≤ Bavail
(8.9)
Upon admitting a new flow, the OLT will reserve additional bandwidth for ONU j j and update accordingly the total available bandwidth: Bavail = Bavail − R f . Similarly, the OLT performs the above algorithm for every admission request of a new flow at any ONU. A flow will be rejected if at least one of the above two conditions is not satisfied. If both conditions are satisfied, then the OLT will conditionally admit the new flow and monitor its QoS parameters for the subsequent n cycles in order to determine whether it finally should admit the flow. When a flow leaves the network, the ONU reports to the OLT and the latter will update the j available bandwidth accordingly: Bavail = Bavail + R f .
8.4.5 Issues and Solutions In the proposed AC scheme, every real-time stream is provided a guaranteed bandwidth that is computed based on the guaranteed rate of the flow and is reserved and fixed per cycle. The OLT then allocates a transmission window that encompasses all the guaranteed bandwidth for every ONU per cycle. A subtle issue which may arise is due to the statistical nature of real-time traffic and hence guaranteeing bandwidth per flow per cycle may ultimately waste the bandwidth. In other words, if one ONU is being reserved bandwidth for a particular flow per cycle and has no traffic from this flow to transmit, then this bandwidth is not utilized and wasted. This issue arises because the allocation became static (i.e., reservation) and not dynamic as in traditional EPON systems, where the bandwidth is allocated on demand. Moreover, if a flow had more bytes to be sent than the reserved ones (i.e., guaranteed), then our purpose on providing bandwidth guaranteed in every cycle will be unsuccessful. This is because estimating the bandwidth requirement for a flow based on its guaranteed rate does not accurately reflect the real nature of the traffic, especially with respect to the arrival of its packets in a short period of time (i.e., the short length of the cycle) and hence the inefficiency of the bandwidth prediction and reservation. To resolve the above problems, we propose a two-branch solution. In the first branch, the OLT selects a super-cycle (Tsc = λ × Tcycle , where λ is a constant) instead, and every admitted real-time flow is now guaranteed a bandwidth per Tsc . The purpose of this proposal is to mitigate the inefficiency of the bandwidth reservation caused by the short-time prediction, and thus a more accurate bandwidth
8 Quality of Service in EPONs
187
estimation will take place. Here as before, the period (1 − α) × Tsc is divided into new which two periods, T1 and T2 . Each ONU is now guaranteed a bandwidth of Bmin is computed based on T1 . The OLT controls the remaining bandwidth of the superf cycle. Upon the arrival of a new flow f at ONU j with bandwidth guaranteed Bg , the flow is either admitted/rejected locally at the ONU or globally by the OLT, as described earlier. In the second branch, we ensure that the reservation does not waste any bandwidth. Here, we apply a crediting system where each flow’s estimated bandwidth is saved as credits at the OLT. In other words, every time a flow is admitted, the OLT will be informed and it will compute/estimate a total credit (number of bytes j f available per Tsc for this flow) C fi = Bg i × Tscr , where Tscr is the period between the arrival of the flow and the end of the current super-cycle. The OLT maintains as j j well a total credit per type of traffic (CCBR for CBR and CVBR for VBR) per ONU; N j j j C fi where N j is the number of CBR flows at ONU for example, CCBR = i=1 j. Now, in every cycle, the OLT deducts the requested/allocated bandwidth of this flow from its reserved credits until the time of a new super-cycle. At this point, the credits are reset to the estimated ones. Next, we will explain how this solution will help in designing a DBA with effective reservation scheme.
8.4.6 Admission Control-Enabled Dynamic Bandwidth Allocation Scheme (AC-DBA) To apply the solutions proposed in the pervious section, we propose a new hybrid DBA that will perform both bandwidth allocation and reservation at the same time. As any conventional DBA, the ONU reports to the OLT, in every cycle, its buffer occupancy (Q CBR (n − 1), Q VBR (n − 1), and Q BE (n − 1), where n is the cycle number) and requests transmission bandwidth accordingly. However, here, the OLT will allocate bandwidth to each CoS at each ONU according to its available credit in the current super-cycle, as well as based on the requests received from other ONUs. j j j Let ACBR (n), AVBR (n), ABE (n) be the bandwidth allocated for ONU j; then we have N j j min (ACBR (n) + AVBR (n)) ≤ Bcycle − Tgtt − (N × BBE ) (8.10) j=1 N
j
min ABE (n) ≤ N × BBE
(8.11)
j=1
where Bcycle is the total bandwidth available in a Tcycle and Tgtt is the total guard min is the minimum bandwidth time (in bytes) between ONUs transmissions and BBE guaranteed (in bytes) for best effort traffic computed as follows:
188
Ahmad Dhaini and Chadi Assi sc Tcycle × α×T Tcycle × α N ×ξ = ×ξ 8 × Tsc 8× N
min BBE =
(8.12)
where ξ is the PON speed (1 Gbps). Every time the OLT allocates bandwidth to one ONU, it will adjust the availj j j able credit for every CoS accordingly: CCBR (n) = CCBR (n − 1) − ACBR (n). The credit for VBR traffic is updated similarly. If the ONU has run out of credits, then the OLT does not allocate any bandwidth for this CoS at this ONU during this super-cycle. As for the computation of the available bandwidth for each CoS, the OLT waits j j j until all requests (R(Q CBR (n − 1) + Q VBR (n − 1) + Q BE (n − 1)) are received from N j j min , all ONUs. If j=1 (Q CBR (n − 1) + Q VBR (n − 1)) ≤ Bcycle − Tgtt − N × BBE j
j
j
then ACBR (n) = min(Q CBR (n − 1), CCBR (n − 1)); similarly for VBR traffic, j j j AVBR (n) = min(Q VBR (n − 1), CVBR (n − 1)) and their credits (for both CBR and VBR) are updated accordingly. Otherwise, the OLT will compute the total guaranteed bandwidth, B j , for each ONU j as follows: B j (n − 1) =
min )) R j (n − 1) × (Bcycle − Tgtt − (N × BBE N j=1 R j (n − 1)
j
(8.13)
j
where R j (n −1) = Q CBR (n −1)+ Q VBR (n −1). Then the OLT allocates bandwidth as follows: j j j (8.14) ACBR (n) = min(Q CBR (n − 1), CCBR (n − 1)) j
j
j
AVBR (n) = min(B j (n − 1) − Q CBR (n − 1), CVBR (n − 1))
(8.15)
Next, the OLT will allocate bandwidth to BE traffic based on the requests remin , ceived from the ONUs. The total BE bandwidth per cycle is BBE = N × BBE N j j which is shared by all ONUs. Note, however, if j=1 (ACBR (n) + AVBR (n)) ≤ t min Bcycle − Tgt − N × BBE , then the total bandwidth available for BE traffic becomes ⎛ min BBE = N × BBE + ⎝ Bcycle − Tgtt −
N
⎞ (ACBR (n) + AVBR (n))⎠ j
j
(8.16)
j=1 j
j
j
min , then A (n) = Q (n − 1). Otherwise, the OLT will alloIf Q BE (n − 1) ≤ BBE BE BE min and will compute the excess bandwidth cate to the ONU requesting less than BBE from these ONUs to distribute them to other ONUs requesting more BE traffic. j min , then A j (n) = B min + χ , where χ is the Accordingly, if Q BE (n − 1) > BBE j j BE BE excess bandwidth allocated for ONU j:
χj =
rem (n) α j × BBE αt
(8.17)
8 Quality of Service in EPONs
189
N
j
rem min , α = where α j = Q BE (n − 1) − BBE t j=1 α j , and BBE (n) is the remaining bandwidth in the cycle n after allocating all ONUs bandwidth for their BE traffic such that L j rem min BBE = BBE − (N − L) × BBE − Q BE (n − 1) (8.18) j=1
Now, in order to prevent the waste of bandwidth and control the allocation of surplus to various ONUs, the excess bandwidth allocated for the BE traffic at a highly loaded ONU (χ j ) is computed as follows: χ j = min(χ j , α j )
(8.19)
8.4.7 Performance Evaluation The evaluation of the AC framework is based on the second simulation model, that is, to the best of our knowledge, the first EPON simulation model that truly models the real Internet traffic (that arrives to the ONUs as flows or streams) and that enables the support of AC in EPON. We begin by testing the behavior of our admission control by showing in Fig. 8.6the number of admitted real-time traffic streams. As shown, our system reaches saturation (i.e., no more real-time flows can be admitted in the network)
Simulations Traffic Model 300 CBR VBR BE
Number of Flows
250 200 150 100 50 0
0
1000
2000
3000
4000 5000 6000 Time (ms)
7000
8000
c 2007 IEEE.) Fig. 8.6 Traffic model used for the AC framework. (After [12].
9000 10000
190
Ahmad Dhaini and Chadi Assi
at time 7000 ms. As we continue generating real-time flows until 7500 ms, all the real-time flows arriving afterward are rejected. However, this does not mean that no flows were rejected earlier since conditions (8.7) or (8.8) and (8.9) need to be respected to admit a new arriving real-time flow otherwise a flow is rejected. Figure 8.7 shows that starting 450 ms as they arrive are rejected. Real−Time Admitted Flows vs. Time Number of Real−Time Admitted Flows
500 450
With Admission Control Without Admission Control
400 350 300 250 200 150 100 50 0 0
1
2
3
4
5
6
7
8
9
10
Time (seconds)
c 2007 IEEE.) Fig. 8.7 Admission control behavior. (After [12].
Next, we study the performance of real-time traffic by measuring the instantaneous average packet delays. To reduce the measurements complexity, we choose the sampling period T = Tsc = 500 ms. Figures 8.8 and 8.9 show these measurements, with admission control (i.e., AC-DBA) and without admission control (using M-DWRR (set 1) and strict priority (SP) schedulers). Clearly, using M-DWRR and SP schedulers, CBR traffic shows the optimal performance where its average packet delay remains under 2 ms even when the load continuously increases (i.e., as the simulation time continues to increase). This shows the advantage of M-DWRR, that is, although it divides the cycle among the CoS queues based on their assigned weights, it also provides an optimal performance for CBR traffic. This is due to the fact that the assigned weights are adaptively set based on the QoS requirements. On the other hand, using the strict priority scheduler that always selects packets from higher priority queue until satisfied (i.e., until it is empty), CBR traffic will accordingly exhibit the best performance. As for AC-DBA, it makes sure to satisfy the QoS requirements defined previously (in terms of delay and throughput) by crediting every real-time traffic the appropriate bandwidth and reserving it in every super-cycle/cycle, since a CBR flow is admitted only if its guaranteed bandwidth is assured in every cycle. Hence, AC-DBA maintains a CBR average packet delay of 2–4 ms with a noticeable slight decrease pattern that repeats every super-cycle. This is due to the fact that the credits
8 Quality of Service in EPONs
191
CBR Average Delay vs. Time
× 10–3
5 CBR Average Delay every "T" Curve Fitting
8 Packet Delay (seconds)
CBR Average Delay every "T" Curve Fitting
4.5 CBR Packet Delay (seconds)
9
× 10−3 Average Sampled Packet CBR Delay vs. Time
7 6 5 4 3 2
4 3.5 3 2.5 2 1.5 1 0.5
1 0
10
20
30 40 50 60 Time (x0.1 seconds)
70
80
90
0
0
(a) AC-DBA
CBR Average Packet Delay (seconds)
5
× 10−3
20
40
60
80
100 120 140 160 180 200
Sampled Time (seconds)
(b) NO-AC using M-DWRR
Sampled CBR Average Packet Delay vs. Time CBR Average Delay every "T" Curve Fitting
4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 0
20
40
60
80
100 120 140 160 180 200
Sampled Time (seconds)
(c) NO-AC using Strict Priority
c 2007 IEEE.) Fig. 8.8 CBR packet delay. (After [12].
assigned in one super-cycle n were consumed just before the “credit refilling” for super-cycle n + 1, which is due to the statistical multiplexing nature of CBR traffic; and thus, the delay decreases after the latter operation. As for the VBR traffic, as shown in Fig. 8.9, AC-DBA maintains its delay performance to meet the specified QoS requirements of the stream (i.e., 25–30 ms) while the delay witnesses an exponential increase under both adopted schedulers (Figs. 8.9(b) and 8.9(c)), i.e., a system that does not deploy any admission control. This behavior highlights the need for the application of admission control in EPON, because when the system reaches saturation (as described earlier) and all the arriving streams are admitted, the performance is no more maintained; more specifically, no bandwidth is guaranteed for all types of traffic and the QoS requirements are no longer met (not only for new application but for existing applications as well). On the other hand, the deployment of AC in EPON allows for a bandwidth guaranteed service with guaranteed protected QoS. We further investigate our AC framework by measuring/monitoring the throughput of one flow from each CoS (i.e., CBR, VBR) with AC (i.e., AC-DBA) and with no AC (i.e., M-DWRR and SP) in Fig. 8.10. As shown and expected, the selected
192
Ahmad Dhaini and Chadi Assi VBR Average Delay vs. Time 0.08
Average Packet VBR Delay (seconds)
Packet Delay (seconds)
Average Sampled VBR Packet Delay vs. Time
VBR Average Delay every "T" Curve Fitting
0.07 0.06 0.05 0.04 0.03 0.02 0.01
VBR Average Delay every "T" Curve Fitting
0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 −0.05
0 0
10
20
30
40
50
60
70
0
80
20
40
60
80
100
120
140
Time (x0.1 seconds)
Sampled Time (seconds)
(a) AC-DBA
(b) NO-AC using M-DWRR
160
180
VBR Average Packet Delay (seconds)
VBR Sampled Average Packet Delay vs. Time 0.4 VBR Average Delay every "T" Curve Fitting
0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 −0.05 0
20
40
60
80
100
120
140
160
180
Sample Time (seconds)
(c) NO-AC using Strict Priority
c 2007 IEEE.) Fig. 8.9 VBR packet delay (After [12]. VBR Flow Throughput vs. Time
CBR Flow Throughput vs. Time
8
AC−DBA NO AC − M−DWRR NO AC − Strict Priority
0.18
VBR Flow Throughput (Mbps)
CBR Flow Throughput (Mbps)
0.2
0.16 0.14 0.12 0.1 0.08 0.06 0.04 0.02
AC−DBA NO AC−M−DWRR NO AC− Strict Priority
7 6 5 4 3 2 1 0 −1
0 0
1
2
3
4
5
6
7
8
9
0
10
1
5
6
7
(b) VBR Flow Throughput
BE traffic Throughput vs. Time
Throughput (Mpbs)
4
(a) CBR Flow Throughput
AC−DBA NO AC − M−DWRR NO AC − Strict Priority
400 350 300 250 200 150 100 50 0 1
3
Time (seconds)
450
0
2
Time (seconds)
2
3
4
5
6
7
8
9
10
Simulation Time (seconds)
(c) BE Flow Throughput
c 2007 IEEE.) Fig. 8.10 CoS traffic throughput. (After [12].
8
9
10
8 Quality of Service in EPONs
193
CBR flow exhibits the same performance with and without AC, while the selected VBR flow shows a different behavior. Here, the VBR flow with AC maintains its derived 4 Mbps throughput throughout the simulation, even after the system reaches saturation. On the other hand, when no AC is applied, the VBR flow does not show a stable throughput behavior. Moreover, when the system reaches saturation, the throughput of the VBR flow starts decreasing. This is due to the fact that when more real-time flows are admitted and no AC is applied, the bandwidth that was guaranteed for the already admitted flows (before saturation) is now shared by more flows. Hence, the guaranteed bandwidth is no longer guaranteed for the already admitted flows and for the newly admitted ones. This again shows the need for admission control in EPON to stabilize and guarantee the throughput for all admitted flows and reject the flows that will break this theme. This, in real and practical settings, will deny all malicious users from monopolizing the bandwidth provided, and at the same time, it will allow for bandwidth protection to the bandwidth assigned for other well-behaved users. As for the BE traffic, our concern is to guarantee a minimum total throughput that meets rule (2) in the AC scheme. For that reason, we measure its total throughput rather than the per-flow throughput as we did for CBR and VBR traffic. Here, the BE throughput increases to reach a total of ≈ 400 Mbps under all schemes (i.e., with AC and with no AC) when the load is low and decreases when more flows are admitted into the network. However, when the system reaches saturation, AC-DBA makes sure to preserve the minimum predefined throughput; while with M-DWRR and strict priority schedulers, the throughput is not guaranteed and hence the predefined throughput is no longer respected. Nevertheless, M-DWRR still provides a minimum throughput (which is one of the advantages of M-DWRR) by forcing the weight policy, while it reaches a very low one (≈ 0 Mbps) with SP; a phenomenon known as BE traffic starvation.
8.5 Summary Ethernet passive optical networks (EPONs) have emerged as the best solution for the last mile bottleneck. EPONs not only provide high speed bandwidth for the emerging QoS applications, but also offer high reliability, maintenance, low cost, and most importantly an easy spatial upgrade that can meet the continuous Internet growth in terms of users and bandwidth demand. Although standardized, EPON carries many “yet-to-be-solved” problems, such as efficient bandwidth allocation (inter- and intra-ONU schedulers), fairness, and QoS protection. From this point, EPON stands as an interesting to-be-investigated technology and is still exposed to intensive research from both the industry and the academia. In this chapter, we addressed these problems and surveyed various solutions that aim to improve the overall performance in the access network. Moreover, we presented a novel decentralized bandwidth intra-ONU scheduler (M-DWRR)
194
Ahmad Dhaini and Chadi Assi
that allows for a unique ONU-gripping to QoS traffic by adaptively setting weights for the different CoS. We also addressed the QoS protection problem in EPONs and motivated the need for admission control in order to “respect” QoS requirements and enable an efficient bandwidth allocation. To implement this solution, we presented the first and complete EPON framework that supports the application of admission control (AC). This framework resolves the guaranteed bandwidth issue for the QoS applications and protects the performance of on-going admitted traffic. The AC framework implements a two-stage admission control, namely locally at the ONU and globally at the OLT, with all its rules. Moreover, we have supported this framework with the first hybrid AC-enabled DBA that performs both bandwidth allocation and reservation. We have also presented the first simulation model that is designed to test this framework. Our AC framework showed that the application of admission control in EPON is becoming crucial for providing bandwidth guaranteed for the emerging QoS applications, and their protection against the malicious users that aim at monopolizing the bandwidth provided.
References 1. M. Garcia, D. F. Garcia, V. G. Garcia, and R. Bonis, “Analysis and modeling of traffic on a hybrid fiber-coax network”, IEEE JSAC, 22(9), November 2004. 2. Jin Liang, Klara Nahrstedt, “RandPeer: Membership management for QoS sensitive peer-topeer applications”, Proceedings of IEEE INFOCOM, Barcelona, April 2006. 3. G. Kramer, B. Mukherjee, and G. Pesavento, “Ethernet PON (ePON): Design and analysis of an optical access network”, Photonic Network Communications, 3, 307–319, 2001. 4. G. Kramer, B. Mukherjee, S. Dixit, Y. Ye, and R. Hirth, “Supporting differentiated classes of service in ethernet passive optical networks”, OSA Journal of Optical Networking, 1(8/9), 280–298, August 2002. 5. A. Demers, S. Keshav, and S. Shenker, “Analysis and simulation for a fair queueing algorithm”, Internetworking: Research and Experience, 1(1), September 1990. 6. N. Ghani,A. Shami, C. Assi, and Y. Raja, “Intra-ONU bandwidth scheduling in ethernet passive optical networks”, IEEE Communication Letters, 8(11), November 2004. 7. Chadi M. Assi, Yinghua Ye, Sudhir Dixit, and Mohamed A. Ali, “Dynamic bandwidth allocation for Quality-of-Service over Ethernet PONs”, IEEE JSAC, 21, 1467–1477, 2003. 8. M. Ma,L. Liu, and H. Cheng, “Adaptive scheduling for differentiated services in the ethernet passive optical networks”, IEEE ICCS 2004, 102–106, 2004. 9. G. Kramer et al., “Fair queuing with service envelopes (FQSE): A cousin-fair hierarchical scheduler for subscriber access networks”, IEEE JSAC, 1497–1513, October 2004. 10. J. Chen, B. Chen, and S. He, “A Novel algorithm for intra-ONU bandwidth allocation in ethernet passive optical networks”, IEEE Communications Letters, 9(9), September 2005. 11. Ahmad R. Dhaini, Chadi M. Assi, Abdallah Shami, and Nasir Ghani, “Adaptive Fairness through intra-ONU scheduling in ethernet passive optical networks”, In Proceedings of IEEE International Communications Conference (ICC2006), Istanbul, June 2006. 12. Ahmad R. Dhaini, Chadi M. Assi, Martin Maier, and Abdallah Shami, “Per-stream QoS and admission control in ethernet passive optical networks (EPONs)”, IEEE Journal of Lightwave Technology (JLT), 25, 1659–1669, 2007. 13. R. Stankiewicz and A. Jajszczyk, “Analytical models for multi-RED queues serving as droppers in DiffServ networks”, IEEE GLOBECOM 2007, 2667–2671, November 2007.
8 Quality of Service in EPONs
195
14. Maode Ma, Yongqing Zhu, and Tee Hiang Cheng, “A bandwidth guaranteed polling MAC protocol for ethernet passive optical networks”, Proceedings of IEEE INFOCOM, vol. 1, pp. 22–31, San Francisco, CA, March 2003. 15. Su-Il Choi and Jae-Doo Huh., “Dynamic bandwidth allocation algorithm for multimedia services over ethernet PONs”, ETRI Journal, 24(6), 465–468, December 2002. 16. Yuanqiu Luo and Nirwan Ansari, “Bandwidth allocation for multiservice access on EPONs”, IEEE Communications Magazine, 40(2), February 2002. 17. Jing Xie, Shengming Jiang, and Yuming Jiang, “A Dynamic bandwidth allocation scheme for Differentiated Services in EPONs”, IEEE Communications Magazine, 42(8), August 2004. 18. Xiaofeng Bai, A. Shami, and C. Assi, “A Hybrid granting algorithm for QoS support in ethernet passive optical networks”, IEEE International Conference on Communications (ICC’05), Seoul, Korea, May 2005. 19. Ahmed E. Kamal and Brian F. Blietz, “A priority mechanism for the IEEE 802.3ah EPON”, IEEE ICC’05, Seoul, Korea, May 2005. 20. H. Nasser and H. Muftah, “A joint-ONU interval-based dynamic scheduling algorithm for ethernet passive optical networks”, IEEE/ACM Transactions on Networking, 14(4), 889–899, August 2006. 21. A. Shami, X. Bai, C. Assi, and N. Ghani, “Jitter performance in ethernet passive optical networks,” IEEE Journal of Lightwave Technology, 23(4), 1745–1753, April 2005. 22. F. An, Y. L. K. Lim, L. White, and L. Kazovsky, “A new dynamic bandwidth allocation with quality of service in ethernet passive pptical networks,” IEEE WOC 2003, 3, 165–169, July 2003. 23. M. Ma, L. Liu, and H. Cheng, “A systematic scheme for multiple access in ethernet passive optical access network”, IEEE/OSA JLT, 23(11), 3671–3682, November 2005. 24. Chun-Ting Chou, Sai Shankar N., and Kang G. Shin, “Achieving per-stream QoS with distributed airtime allocation and admission control in IEEE 802.11e wireless LANs”, Proceedings of IEEE INFOCOM, April 2005. 25. Sugih Jamin, Peter Danzig, Scott Shenker, and Lixia Zhang, “A measurement-based admission control algorithm for integrated services packet networks”, IEEE/ACM Transactions on Networking, 5(1), 56–70, February 1997 26. OECD – Organisation for Economic Co-Operation and Development, “OECD Communications Outlook”, June 2003. 27. F. J. Effenberger, H. Ichibangase, and H. Yamashita, “Advances in broadband passive optical networking technologies”, IEEE Communications Magazine, 39(12), 118–124, December 2001. 28. S. Hardy, “Verizon staffers find fiber-to-the-home cheaper than copper”, Lightwave, PennWell, 17(134), 1, December 2000. 29. Glen Kramer, “Ethernet Passive Optical Networks”, Addison-Wesley, 2005. 30. H. Shinohara, “Broadband access in Japan: Rapidly growing FTTH market”, IEEE Communication Magazine, 72–78, September 2005. 31. Dror Sal’ee,“EPON Ushers in a new generation of bandwidth-intensive applications”, EPON Tutorial, Passave Inc. 32. J. Zheng, H. T. Mouftah, “An adaptive MAC polling protocol for ethernet passive optical networks”, Proceedings of IEEE ICC ’05, May 2005. 33. Su-Il Choi and Jae-Doo Huh., “Dynamic bandwidth allocation algorithm for multimedia services over ethernet PONs”, ETRI Journal, 24(6):465–468, December 2002. 34. Hee-Jung Byun, Ji-Myung Nho, and Jong-Tae Lim, “Dynamic bandwidth allocation algorithm in ethernet passive optical networks”, Electronics Letters, 39(13), 1001–1002, June 2003. 35. Yongquing Zhu, Maode Ma, and Tee Hiang Cheng, “A novel multiple access scheme for Ethernet Passive Optical Networks”, GLOBECOM 2003 – IEEE Global Telecommunications Conference, 22(1), 2649–2653, December 2003.
196
Ahmad Dhaini and Chadi Assi
36. A. Shami, X. Bai, C. Assi, and N. Ghani, “New dynamic bandwidth allocation scheme in ethernet passive optical access networks”, Proceedings of IEEE ICPP 2004, pp. 371–378, August 2004. 56. “Virtual bridged local area networks”, IEEE Standard 802. Iq., 1998. 57. Chuan H. Foh, Lachlan L. Andrew, Moshe Zukerman, and Elaine Wong, “Full-RCMA: A high utilization EPON”, In Proceedings of OFC, Vol. 1, pp. 282–284, Atlanta, GA, March 2003. 39. G. Kramer,B. Mukherjee, and G. Pesavento, “Ethernet passive optical network (EPON): building a next-generation optical access network”, IEEE Communications Magazine, 40(2), 66–73, February 2002. 40. M. P. McGarry, M. Maier, and M. Reisslein, “Ethernet PONs: A survey of dynamic bandiwdth allocation (DBA) algorithms”, IEEE Communications Magazine, 42(8), S8–S15, August 2004. 41. IEEE 802.3ah, Ethernet in the First Mile Task Force, http://www.ieee802.org/3/efm/ index.html 42. R. D. Feldman, E. E. Harstead, S. Jiang, T. H. Wood, and M. Zirngibl, “An evaluation of architectures incorporating wavelength division multiplexing broad-band fiber access”, IEEE Journal of Lightwave Technology, 1546–1558, 1998. 43. M. Zirngibl, C. H. Joyner, L. Stulz, C. Dragone, H. Presby, and I. Kaminow, “LARNET: A local access router network”, IEEE Photonics Technical Letters, 215–217, 1995. 44. N. J. Frigo, et al., “RITENet: A passive optical network architecture based on the remote interrogation of terminal equipment”, http://ieeexplore.ieee.org, 1994. 45. G. Mayer, M. Martinelli, A. Pattavina, and E. Salvadori, “Design and cost performance of the multistage WDM PON access networks”, Journal of Lightwave technology, IEEE, 121–142, 2000. 46. J. D. Angelopoulos, N. Lepidas, E. Fragoulopoulos, and I. Venieris “TDMA Multiplexing of ATM cells in a residential access superPON”, IEEE JSAC, 1123–1133, 1998. 47. I. Van de Voorde, C. M. Martin, J. Vandewege, X. Z. Oiu, “The superPON demonstrator: an exploration of possible evolutionpaths for optical access networks”, IEEE Communications Magazine, 38(2), 74–82, February 2000. 48. G. Kramer, B. Mukherjee, and G. Pesavento, “IPACT: A dynamic protocol for an ethernet PON (EPON)”, IEEE Communications Magazine, 40(2), February 2002. 49. Ahmad R. Dhaini, Chadi M. Assi, Martin Maier, and Abdallah Shami, “Dynamic wavelength and bandwidth allocation in hybrid TDM/WDM ethernet passive optical networks (EPONs)”, IEEE/OSA Journal of Lightwave technology, under review. 50. Ahmad R. Dhaini, Chadi M. Assi, and Abdallah Shami, “Dynamic bandwidth allocation schemes in hybrid TDM/WDM passive optical networks”, In Proceedings of IEEE Consumer Communications and Networking Conference (CCNC2006), Las Vegas, January 2006. 51. Ahmad R. Dhaini, Chadi M. Assi, and Abdallah Shami, “Quality of service in TDM/WDM passive optical networks (EPONs)”, IEEE Symposium on Computers and Communications (ISCC’06), Sardinia, Italy, June 2006. 52. J. Zheng and H. T. Mouftah, “Media access control for ethernet passive optical networks: An overview”, IEEE Communications Magazine, 145–150, February 2005. 53. W. Leland, M. Taqqu, W. Willingler, and D. Wilson, “On the self-similar nature of ethernet traffic (Extended Version)”, IEEE/ACM Transactions on Networking, 1–15, February 1994. 54. Glen Kramer, “Synthetic Traffic Generation”, C++ source code version 1, http://wwwcsif.cs.ucdavis.edu/ kramer/research.html 55. Chuck Semeria, “Supporting differentiated service classes: Queue scheduling disciplines”, white paper, Juniper Networks, January 2002. 56. Virtual bridged local area networks, IEEE Standard 802. Iq., 1998. 57. Chuan H. Foh, Lachlan L. Andrew, Moshe Zukerman, and Elaine Wong, “Full-RCMA: A high utilization EPON”, In Proceedings of OFC, Vol. 1, pp. 282–284, Atlanta, GA, March 2003. 58. ITU-T Recommendation G.1010, “End-user multimedia QoS categories”, 2001.
Chapter 9
MultiChannel EPONs Michael P. McGarry and Martin Reisslein
Abstract In this chapter, we explore the problem of bandwidth management for multichannel EPONs. Our proposed bandwidth management methods will allow for an evolutionary upgrade from single channel to multichannel EPONs. We divide the bandwidth management problem into two subproblems: (1) grant sizing and (2) grant scheduling. We briefly discuss the problem of grant sizing in multichannel EPONs and then focus on the grant scheduling problem in multichannel EPONs for the vast majority of the chapter.
9.1 Introduction As network traffic increases, service providers will scramble to provide more bandwidth to their customers. Currently, service providers are bringing fiber closer and closer to the end user. An evolution is taking place where fiber penetration in the access network gets deeper as costs come down and subscriber demand goes up. This evolving penetration will eventually result in fiber-to-the-subscriber (FTTS), i.e., business or home. At this point, there will exist a potential for very large transmission capacity. Current technology is realizing 128 10 Gbps wavelength channels [1] on a single fiber strand. These channels exist only in a region of the optical spectrum for which we currently have effective transceivers. This number of channels is sure to increase as we can effectively tap into the other regions of the optical spectrum. We need to also be aware that there are several strands of fiber coexisting in a fiber optic cable. These fibers serve little purpose for fault tolerance because they are in close proximity to other fiber strands in the same cable. To offer effective Michael P. McGarry Department of Electrical and Computer Engineering, University of Akron, Akron, OH, USA and ADTRAN, Phoenix, AZ, USA, e-mail:
[email protected] Martin Reisslein Electrical Engineering, Arizona State University, Phoenix, AZ, USA, e-mail:
[email protected]
A. Shami et al. (eds.), Broadband Access Networks, Optical Networks, c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-92131-0 9,
197
198
Michael P. McGarry and Martin Reisslein
protection capabilities the fiber strands need to be spatially diverse. For this reason, these fiber strands can simply be viewed as media for more transmission capacity. Naturally, as the access network is currently evolving in the sense of gaining deeper fiber penetration, it will continue to evolve by activating more transmission channels. These additional channels will be realized through wavelength division multiplexing (WDM) within a fiber strand and space division multiplexing (SDM) across fiber strands. This evolution is sure to be driven by demand and cost. Current EPON standards dictate a single channel used for downstream transmission and a single channel used for upstream transmission. The need for more PON bandwidth capacity will drive the utilization of multiple upstream and downstream channels. We summarize in Table 9.1 the available bandwidth per ONU for different numbers of ONUs and different cycle lengths. As the number of ONUs increases to 64 and the cycle length or polling interval for each ONU decreases to 750 μs, the Ethernet equivalent bandwidth is reduced to 8.96 Mbps per ONU. This is not an improvement over existing copper-based infrastructure, especially since the number of subscribers is often greater than the number of ONUs (e.g., fiber to the curb). Table 9.1 Ethernet equivalent bandwidth for different numbers of ONUs and cycle lengths (Guard Time = 5 μs, one grant per ONU per cycle) Cycle length 750 μs 1.5 ms 2 ms
16 ONUs 55.84 Mbps 59.17 Mbps 60 Mbps
32 ONUs 24.58 Mbps 27.92 Mbps 28.75 Mbps
64 ONUs 8.96 Mbps 12.29 Mbps 13.125 Mbps
In an effort to provide more bandwidth capacity, we can increase the bit rate as well as utilize multiple upstream and downstream channels. The transition from increased bit rate to utilizing multiple channels for an increase in bandwidth capacity will be a function of cost. At some point the transition to multiple transmission channels to increase bandwidth capacity will occur. Besides an increase in bandwidth capacity there are additional benefits provided by utilizing multiple upstream and downstream channels. The first is the ability to allow multiple ONUs to transmit concurrently. This lowers the average queueing delay [20] experienced by the Ethernet frames queued at the ONUs. Multiple channels also offer dynamic bandwidth allocation (DBA), the ability to reserve certain channels for certain traffic types. The discovery and registration process can be kept on a single channel. This would allow transmissions on the other channels to be uninterrupted by the discovery and registration process.
9.2 Bandwidth Management for Multichannel EPONs Bandwidth management for a multichannel EPON can be broken into two subproblems: grant sizing and grant scheduling. Grant sizing determines the size
9 MultiChannel EPONs
199
of a grant to an ONU, and grant scheduling determines when and on which wavelength channel to schedule the grant. Much of the published research in this area refers to dynamic bandwidth allocation (DBA) as consisting of both the grant sizing and scheduling. Grant sizing and grant scheduling can possibly become overlapping problems. In single-channel EPONs, scheduling is greatly simplified by the grants simply being ordered in time. On a multichannel EPON, there are multiple transmission channels, making it possible for transmissions to be scheduled on several channels concurrently. Efficiently scheduling grants across the multiple channels is thereby necessary to fully utilize the network resources.
9.2.1 Grant Sizing We will first briefly investigate grant sizing methods and then delve into the topic of scheduling these grants across multiple wavelengths with constraints imposed by each ONU’s WDM architecture. Existing methods of grant sizing [14] can be employed in the multichannel case. Let G i denote the grant size to ONU i, Ri the requested grant size from ONU i, Pi the predicted amount of backlog at ONU i not accounted for in Ri , G imax the maximum grant size for ONU i, and E i be the excess bandwidth assigned to ONU i. All units are bytes. The general formula for grant sizing is G i = f (Ri + Pi , G imax ) + E i . Let E total be the total excess bandwidth available and f i be the fair share of the total excess allocated to ONU i. Then, E i = E total · f i . We will now examine each component of the grant sizing process and discuss how that component is affected by the use of multiple upstream transmission channels. The different grant sizing techniques are as follows [25]: • • • • •
Fixed, G i = G imax Gated, G i = Ri Limited, G i = min(Ri , G imax ) Excess Distribution, E i = E total · f i [3, 4] Queue Size Prediction, Pi [16, 17]
Gated grant sizing is only determined by the grant size and is therefore unaffected by the number of channels for transmission. For Fixed and Limited grant sizing, the calculation of G imax can be influenced by the number of channels. G imax is calculated as the guaranteed minimum portion of a bandwidth cycle for ONU i. This portion is influenced by the number of channels available for transmission as well as the multichannel capabilities of the ONUs. Let Tcycle be the desired polling cycle time, Tguard the guard time between ONU transmissions, N the number of ONUs, M the number of channels, R the bit rate of an EPON, and wi is the weight N assigned to ONU i according to its Service Level wi = 1. If all ONUs support all of the upstream Agreement (SLA) such that i=1 transmission channels then G imax =
(Tcycle −N ·Tguard )·R·M 8
· wi .
200
Michael P. McGarry and Martin Reisslein
If the ONUs have differing multiple channel capabilities, as would exist in an evolutionary upgrade scenario, the equation would become more complex. As an example, given an EPON with two sets of multichannel capabilities, the first is the set of ONUs that support only the first channel and the second is the set of ONUs that support all channels. A possible calculation of G imax could divide the capacity on one channel amongst the set of single channel ONUs (i.e., channel 1) and divide the capacity of the other channels (i.e., channels 2 through M) amongst the set of multiple channel ONUs. Let α be the number of single channel ONUs and β be (T −α·T )·R the number of multiple channel ONUs. G imax = cycle 8 guard · wi for the single (T
−β·T
)·R·(M−1)
guard channel ONUs, and G imax = cycle · wi for the multiple channel 8 ONUs. Excess Distribution being influenced by G imax is indirectly affected by the number of channels as discussed above. Further, the method of distributing the excess cycle bandwidth can be influenced by the number of channels. The amount of excess assigned to ONU i (i.e., E i ) can potentially be influenced by the wavelength assigned to the transmission from ONU i. One example of this is to have E total computed for each wavelength, the excess available for an ONU depends on which wavelength it is assigned. Queue size prediction attempts to predict the size of the ONU Ethernet frame queue at grant time. This prediction is independent of the number of channels used for upstream transmission. It only depends on the Ethernet frame arrival process. We leave further development of grant sizing techniques for multichannel EPONs outside the scope of this chapter. The calculation of G imax as well as methods for distributing excess bandwidth require further investigation. We now focus the rest of this chapter on grant scheduling.
9.2.2 Grant Scheduling Two fundamental approaches to grant scheduling are investigated in this chapter. • Separate time and wavelength assignment (STWA) • Combined time and wavelength assignment (CTWA) (i.e., multidimensional scheduling) In this chapter, we focus on time and wavelength for multidimensional scheduling. However, scaling the scheduling techniques to more dimensions, to include fiber strand, can be achieved with straightforward extensions to the methods discussed in this chapter.
9.3 Separate Time and Wavelength Assignment In the separate time and wavelength assignment (STWA) approach, grant scheduling is broken into two separate problems:
9 MultiChannel EPONs
201
• Grant wavelength assignment • Grant ordering
9.3.1 Grant Wavelength Assignment Wavelength assignment can be done statically or dynamically. WDM PON architectures that utilize colorless ONUs and static wavelength routing between the OLT and ONUs are restricted to static wavelength assignment (SWA). WDM PON architectures that utilize single-color ONUs are also restricted to SWA. However, WDM PON architectures that utilize colorless ONUs with dynamic wavelength routing between the OLT and ONUs, typically through the use of athermal arrayed waveguide gratings (AWGs), can support dynamic wavelength assignment (DWA) as well as SWA. WDM PON architectures utilizing multicolor ONUs can also support DWA along with SWA. The WDM PON architectures that are restricted to SWA have their wavelength assignment determined by the architecture. The WDM PON architectures that also allow for DWA can implement SWA with an algorithm for assigning the wavelength when the ONU is registered on the PON. In either case, when using SWA a single physical EPON is partitioned into several virtual EPONs separated by wavelength. Some heuristics for selecting a wavelength for static assignment when an ONU is registered on the EPON are [30] • Random – randomly select a wavelength supported by the ONU. • Least Assigned – select the wavelength that is supported by the ONU, and has the least number of ONUs already assigned to it. • Least Loaded – select the wavelength that is supported by the ONU, and has the least load assigned to it (assuming that this information is available when a new ONU is registered). • Class – select the wavelength that has been allocated for the class of the ONU. The wavelength assignment can be adjusted dynamically for those WDM PON architectures that support DWA. DWA is most efficiently managed on a grant by grant timescale. In this case the time and wavelength dimensions are scheduled concurrently.
9.3.2 Grant Ordering When using SWA, bandwidth management or dynamic bandwidth allocation (DBA) becomes three subproblems: (1) grant sizing, (2) ONU wavelength assignment, and (3) time scheduling. Subproblems 1 and 3 can be solved using single-channel DBA algorithms proposed in the research literature, e.g., [3, 5, 11, 14, 18, 22, 27–29]. In this case the SWA partitions the physical EPON into multiple virtual EPONs separated by wavelength.
202
Michael P. McGarry and Martin Reisslein
Table 9.2 Table of ONU wavelength assignment criteria [23] Node 1 2 3 4 5 6
Supported channels λ1 λ1 λ1 , λ2 λ1 , λ2 λ1 , λ2 , λ3 λ1 , λ2 , λ3
Traffic load Low Medium Low Low Low High
Suppose we have a three-wavelength EPON with six ONUs, Table 9.2 shows the wavelength support and traffic load (high, medium, low) for each of the six ONUs. Figure 9.1 shows how these ONUs could be partitioned into multiple virtual EPONs through wavelength assignment. ONUs 1 and 2 only support λ1 , so they are forced to be assigned to this wavelength. Since ONU 6 has a high traffic load, it is assigned exclusively to λ3 . This leaves λ2 for ONUs 3, 4, and 5.
OLT
λ1
ONU 1
ONU 2
OLT
λ3 ONU 6
ONU 3
OLT
λ2 ONU 4
ONU 5
Fig. 9.1 Illustration of multiple virtual EPONs separated by wavelength [23]
9.4 Combined Time and Wavelength Assignment 9.4.1 A Scheduling Theoretic Approach Combining ONU grant ordering with ONU grant wavelength assignment leads to multidimensional scheduling. This multidimensional scheduling problem can be modeled using the scheduling model notation defined in [26]. Theoretical analysis of all the scheduling models discussed in this section can be found in [26]. Scheduling theory is concerned with scheduling a set of jobs with specific processing times to be executed on a set of machines as efficiently as possible with respect to an optimization criterion. We can view each ONU as representing a job, its grant size as defining its processing time, and the channels used for upstream transmission grants
9 MultiChannel EPONs
203
as representing the machines. In scheduling model notation, a scheduling problem is defined by a triple α|β|γ , where α describes the machine environment (e.g., single machine, and parallel machines), β describes the processing characteristics and constraints, and γ describes the objective to be minimized. In the multichannel EPON grant scheduling problem there are multiple wavelengths available for facilitating upstream transmission grants. Typically, these wavelengths will have the same transmission capacity and therefore are identical. As a result, the multichannel EPON grant scheduling is done in a parallel identical machine environment, which we denote with P. To support a multichannel EPON in which the ONUs have different multichannel transmission capabilities, we need to add machine (or channel) eligibility constraints, which we denote with Mi . Specifically, Mi is the set of machines (channels) that job (ONU) i can be executed (transmitted). Let Ci denote the time at which the transmission for ONU i is complete. Using the scheduling model notation, our multichannel EPON grant scheduling problem with the objective to minimize the sum of the completion times can be expressed as P|Mi | i Ci . In the above model, the Mi processing constraint is required because each ONU has its own subset of supported channels. If all of the ONUs supported transmission on all wavelengths we could remove the machine eligibility constraint to obtain the simpler scheduling model, P|| i Ci . Our performance objective in designing a scheduler for a WDM EPON is to increase resource (i.e., channel) utilization and lower queueing delays experienced by Ethernet frames that are transmitted through the EPON. To see how these performance objectives relate to the objectives from scheduling theory we first explore in detail all of the component delays in a scheduling cycle [23, 24]. We start by defining cycle length, which we also refer to as the GT G or GATE-to-GATE delay, as the time between back-to-back grants to an ONU. All of the component delays of a scheduling cycle are visualized in Fig. 9.2. GT R is the GATE-to-REPORT delay (since we append the REPORT at the end of the transmission window, GT R is equal to the transmission time of the grant), and RT G is the REPORT-to-GATE delay, that includes the propagation delay from ONU to OLT. Using GT R and RT G we express the cycle length as GT G = GT R + RT G. Schedule ONU
propagation delay
Ci RTS
OLT
ONU
DATA
DATA
STG GT
RPT
DATA
GT
RPT
GTR
RTG Cycle Length
c Fig. 9.2 Illustration of scheduling cycle [24]. 2008 IEEE
DATA
RPT
RPT
204
Michael P. McGarry and Martin Reisslein
The GT R delay is completely dependent on the grant size and the RT G delay depends on the efficiency of grant scheduling. The ST G is the Schedule-to-GATE delay which is the time between the OLT scheduling an ONU’s next grant to the time the grant starts. The ST G includes a propagation delay from OLT to ONU, a GATE message transmission time and propagation delay from ONU to OLT. The ST G along with the grant time represents the completion time of an ONU’s transmission from the point in time it is scheduled, i.e., Ci = ST G + GT R. Since, the grant time (or size) is not determined by the scheduler, the scheduler can only work to minimize the variable portion of completion time, i.e., the ST G. Minimizing Ci will minimize ST G. The RT S is the REPORT-to-Schedule delay and is the delay from the ONU transmitting its MPCP REPORT to the OLT to when the ONUs’ REPORT is considered for scheduling by the OLT. Thus, REPORT-to-GATE (RT G) delay is composed of the RT S and ST G delays, i.e., RT G = RT S + ST G. With a GATED grant sizing mechanism (i.e., a grant sizing scheme which grants every ONUs requested transmission time) the queueing delay can be anywhere between one and two times the cycle length. An Ethernet frame that arrives at the ONU just before the REPORT message prepared for transmission is counted in that REPORT message and is transmitted at the end of the next grant (i.e., one cycle length of queueing delay). However, an Ethernet frame arriving just after the REPORT has been generated will not be accounted for until the next REPORT message (a cycle length later), it will then be transmitted a cycle length after that REPORT is sent to the OLT for a total queueing delay of two cycle lengths. Lowering the cycle length, therefore will inherently lower the queueing delay. We introduce a layered approach to grant scheduling. The first layer we refer to as a scheduling framework. The second layer we refer to as a scheduling policy. The scheduling framework is a logistical framework that determines when the OLT makes scheduling decisions, whereas the scheduling policy is a method for the OLT to produce the schedule (this policy is determined by the scheduling model). The OLT can produce a schedule with partial information about the ONU transmissions to be scheduled or after waiting to receive all of the information about the ONUs transmissions to be scheduled. Once the OLT has a set of ONU transmissions to schedule, the OLT uses a scheduling policy to create the schedule. The scheduling framework will impact the RTS delay by determining at what time after a REPORT message is received an ONU’s next grant is scheduled. On the other hand, the scheduling policy will impact the STG delay by determining when an ONU grant is actually transmitted on a channel. In Section 9.4.2 we will discuss scheduling frameworks in more detail and in Section 9.4.3 we will discuss scheduling policies for our scheduling model.
9.4.2 Scheduling Frameworks As mentioned above, the scheduling framework determines when the OLT will produce a schedule. If the OLT produces a schedule as soon as any ONU REPORT is
9 MultiChannel EPONs
205
received without waiting for REPORT messages from other ONUs, this is referred to as an online scheduling framework [23, 24]. However, if the OLT were to wait for the REPORT messages from all the ONUs to be received before making scheduling decisions, this is referred to as an offline scheduling framework. The scheduling framework can be viewed as a continuum between the extremes of online and offline scheduling. Figure 9.3 illustrates this continuum. On the online scheduling end of the continuum, the OLT only considers a single ONU REPORT in a scheduling decision. On the offline scheduling end of the continuum, the OLT considers all ONU REPORTs in a scheduling decision. Any scheduling framework that lies between online and offline on the continuum is some form of online scheduling framework because not all of the ONU REPORTs have been received. We will however reserve the term online scheduling framework to indicate the case where the OLT considers only one ONU REPORT at a time. In Section 9.4.2.1 we will explore offline scheduling in the context of multichannel EPONs and in Section 9.4.2.2 we will explore online scheduling in the context of multichannel EPONs. We will see that both schemes have very different channel utilization characteristics. control
Online
least control most efficient 1 ONU
Offline
efficiency
most control least efficient N ONUs
c Fig. 9.3 Scheduling framework continuum [24] 2008 IEEE
9.4.2.1 Cyclical Offline Scheduling In a cyclical offline scheduling framework, scheduling decisions are made with full knowledge of all the jobs to be scheduled including their processing times for a particular scheduling cycle. Specifically, for a multichannel EPON, an offline scheduling framework schedules ONU grants for transmission when the OLT has received current MPCP REPORT messages from all ONUs, allowing the OLT to take into consideration, in the scheduling (and grant sizing), the current bandwidth requirements of all ONUs. The offline scheduling framework for multichannel EPONs is illustrated in Fig. 9.4. The scheduling policy is executed after the OLT receives the end of the last ONU’s gated transmission window. As a result, all ONU grants are scheduled after Cmax = maxi (Ci ) of the preceding schedule. The RT S delay for the last ONU will be negligible, however, the RT S may not be negligible for the other ONUs. This RT S will introduce further queueing delays in the ONUs because it introduces additional delay in the cycle length (GT G) for an ONU. In the example illustration of Fig. 9.4, the worst case RT S is for ONU 2. ONU 2 is the first ONU to complete its granted transmission and send its REPORT to the
206
Michael P. McGarry and Martin Reisslein C1
C 3= C max REPORT−to−GATE delay for ONU 2 REPORT−to−Schedule delay for ONU 2
C2 λd TX
OLT λ1 RX
6400 bytes data + 64 bytes RPT
λ2 RX λ1 TX
ONU 1 λ2 TX
1280
5120
5120
6400 bytes data + 64 bytes RPT
2560
1280
5120
λd RX 1: 6400
λ1 TX ONU 2 λ2 TX λd RX
1: 5120
1280
2560
2: 1280
λ1 TX ONU 3 λ2 TX λd RX
2: 2560
5120 2: 5120
1280 2: 1280
Fig. 9.4 Illustration of offline scheduler, which introduces the REPORT-to-Schedule (RTS) delay. The illustration includes one downstream wavelength λd and two upstream wavelengths λ1 and λ2 which are supported by all three ONUs. Each ONU reports its queue occupancy in the REPORT (RPT) message, which is appended to the current upstream transmission [23]
OLT. Wavelength λ1 was available for ONU 2 right after ONU 1’s transmission and could have accommodated ONU 2’s next request a guard time after that. Since the OLT waits until all REPORTs have been received before performing all the grant scheduling for the next cycle, ONU 2’s grant scheduling is delayed past the end of ONU 1’s transmission. Specifically, ONU 2 has to wait until the REPORT is received from ONU 3 which in this case is the last ONU to send its REPORT to the OLT. The offline scheduling framework results in wasted channel capacity as the OLT waits to receive all ONU REPORT messages. This wasted channel capacity results in higher cycle lengths experienced by the ONUs and consequently higher queueing delays experienced by the Ethernet frames in transit through the EPON. However, the offline scheduling framework affords the OLT the most control in making access decisions, as the OLT will have information regarding the traffic demands of all of the ONUs when making access decisions.
9.4.2.2 Online Scheduling In an online scheduling framework [8, 15, 21], the OLT schedules ONU grants one at a time without considering the bandwidth requirements of other ONUs. Specifically, an ONU is scheduled for upstream transmission as soon as the OLT receives the REPORT message from the ONU. This scheduling framework is at the far end of the online side of the scheduling framework continuum depicted in Fig. 9.3. Figure 9.5 illustrates the online scheduling framework for a multichannel EPON with three ONUs. Notice that the 2560 byte upstream transmission from ONU 2, for
9 MultiChannel EPONs
207 REPORT−to−GATE delay for ONU 2
OLT
λd TX λ1 RX λ2 RX λ1 TX
6400 bytes data + RPT 1280
2560 + RPT
1280
5120 + RPT
5120 + RPT
5120 + RPT
6400 bytes data + RPT
ONU 1 λ2 TX
5120 + RPT
λd RX 1: 6400
λ1 TX λ ONU 2 2 TX λd RX
2: 5120 2560 + RPT 2560 + RPT
1280 2: 1280
λ1 TX λ ONU 3 2 TX λd RX
1: 2560 1280 1280
5120 + RPT 2: 6400
2: 1280
Fig. 9.5 Illustration of online scheduling with Next Available Supported Channel (NASC) policy. Upon receipt of a RPT message the OLT immediately schedules the next upstream transmission for the corresponding ONU and sends a GATE message (illustrated by the dashed message) indicating the wavelength and length (in Bytes in the illustration) of the granted transmission to the ONU [23]
its second grant, is scheduled on the earliest available supported wavelength, namely wavelength 1, and is timed by the OLT such that it is separated from the preceding transmission on wavelength 1 by ONU 1 by the guard interval. With the offline scheduling framework, ONU 2 would have to wait until the REPORT message is received from ONU 3 before it would be scheduled. The dashed line boxes show the grants for cycle 2 for the offline scheduling. We can see that all the grants for cycle 2 are pushed further out in time with the offline scheduling framework compared to the online scheduling framework. ONU 2, the first ONU to complete cycle 1, experiences the largest difference in delay between the online and offline scheduling frameworks. We can also note that the REPORT-to-Schedule delay incurred by an offline scheduling framework does not exist in an online scheduling framework since an ONU is scheduled as soon as its REPORT is received. This means REPORT-toGATE delay simply reduces to the STG (Schedule-to-GATE) delay or Ci (i.e., ONU transmission completion time).
9.4.2.3 ONU Load Status Hybrid Scheduling The load status of the ONU can determine when that ONU is scheduled [9, 10]. When using the limited grant sizing scheme with excess bandwidth distribution, the grant size of an ONU is limited to some value, G imax . If the ONU queue occupancy (i.e., Ri ) is less than G imax , it is classified as underloaded. The grant sizes for underloaded ONUs do not depend on the REPORT messages of other ONUs and therefore their grants can be scheduled online. However, if an ONU queue occupancy is larger than the limit, it is classified as overloaded. The grant size for
208
Michael P. McGarry and Martin Reisslein
overloaded ONUs depends on the excess bandwidth available in a cycle and therefore requires the information contained in all the ONU REPORT messages. For this reason, overloaded ONUs are scheduled offline. Alternatively all ONUs can be scheduled online. However, since the OLT needs to grant excess bandwidth, two grants are issued. One grant as soon as the ONU REPORT is received that grants the guaranteed minimum and one grant after all REPORTs have been received which grants the excess assigned to that ONU. There are two problems with that approach: (1) each overloaded ONU receives two grants which decreases efficiency due to more guard times and (2) the split between the two grants will most likely not occur on frame boundaries causing one frame to be unnecessarily delayed till the next cycle.
9.4.2.4 Online Just-in-Time (JIT) Scheduling We now present a new scheduling framework that is a hybrid between offline and online scheduling discussed in Sections 9.4.2.1 and 9.4.2.2, respectively. We call this new scheduling framework Online Just-in-Time (JIT) scheduling. The name indicates that scheduling is performed in just in time fashion. In our online JIT scheduling framework, ONUs are added to a scheduling pool as their MPCP REPORT messages are received by the OLT. When a wavelength becomes available, the ONUs in the pool are scheduled together according to the selected scheduling policy across all wavelengths. The ONUs that are scheduled so that their transmissions would occur shortly after the time they are scheduled are classified as “imminent” and the current schedule for these ONUs is considered firm. The OLT transmits GATE messages to these ONUs to inform them of their granted transmission window. The remaining ONUs are classified as “tentative” and can remain in the scheduling pool for the next scheduling round. Alternatively, all ONUs (i.e., both imminent and tentative) can always be firmly scheduled. We will refer to the case where all ONUs are firmly scheduled as Online JIT and the case where the tentative ONUs participate in future scheduling rounds as Online JIT Tentative. The Online JIT scheduling framework gives the OLT more opportunity to make better scheduling decisions. ONUs are scheduled at the moment right before they potentially begin transmitting. To facilitate this on an EPON, we need to ensure the GATE message is transmitted by the OLT a round-trip time (RTT) before we intend the ONU to begin transmission. Since we desire the ONU to transmit as soon as the next wavelength becomes free, we need to schedule the ONUs in the pool at least an RTT before the next wavelength free time. Figure 9.6 illustrates where the Online JIT scheduling framework lies on the scheduling framework continuum. The Online JIT scheduling framework can lie somewhere from the online scheduling framework up to a point just short of the offline scheduling framework. Let us consider the bounds of where the Online JIT scheduling framework can lie with respect to number of ONU REPORTs considered. At the lower bound, grant sizes are much smaller than an RTT. At the point in time when we need to produce a schedule, which is an RTT before the next channel becomes available, there will be
9 MultiChannel EPONs
209 control
Online
least control most efficient 1 ONU
efficiency
Online JIT
Offline
moderate control most efficient 1 to (N−M) ONUs
most control least efficient N ONUs
c Fig. 9.6 Online JIT on the scheduling framework continuum [24]2008 IEEE
no ONU REPORTS received at the OLT. In this case the OLT must wait until the next ONU REPORT is received and the OLT makes a scheduling decision based on only this one REPORT (i.e., the lower bound is 1 ONU REPORT). At the upper bound, grant sizes are larger than an RTT. Therefore, only the REPORTs for ONUs that are currently transmitting on a channel have not been received at the OLT. Let N be the number of ONUs, and M be the number of channels then N − M would be the number of ONU REPORTs received at the time the OLT needs to generate a schedule (i.e., the upper bound is N − M ONU REPORTs). So, it is possible for the Online JIT scheduling framework to get very close to emulating an offline scheduling framework, especially with a small number of channels. When using the Online JIT Tentative scheduling framework an ONU may participate in several scheduling rounds as “tentative” before it becomes firmly scheduled. It is possible that certain ONUs that are unfavorable to a particular scheduling policy can continuously be preempted by those that are more favorable. To prevent these ONUs from being starved of medium access, an aging mechanism is incorporated to keep these “less favorable” ONUs (or jobs) from being starved by the scheduler. A straightforward method to implement starvation prevention is to set a threshold at which an ONU is immediately scheduled on the next available wavelength regardless of the scheduling policy. This ensures that no ONU waits indefinitely for medium access. Having this threshold based on number of participated scheduling rounds allows it to adapt to changing cycle times.
9.4.3 Scheduling Policies We now use our scheduling model i.e.,P|Mi | i Ci to find the best scheduling policies for a multichannel EPON that supports an evolutionary migration from a single upstream channel to multiple downstream channels. The OLT uses these scheduling policies once a set of ONU grants to be scheduled has been determined by the scheduling framework.
9.4.3.1 Next Available Supported Channel A simple load balancing scheduling policy for a multichannel EPON considers one ONU at a time and schedules the upstream transmission for that ONU on the wavelength channel that is available the earliest among the channels supported by that
210
Michael P. McGarry and Martin Reisslein
ONU. We refer to this scheduling policy as the Next Available Supported Channel (NASC) policy. NASC is our variation on an algorithm proposed by Graham [12] nearly 40 years ago called the List algorithm for identical parallel machines. This algorithm schedules jobs one by one and assigns them to the next available machine. Since each of our nodes has a different set of supported channels we developed a variation on this algorithm where we only consider supported machines (i.e., channels). The List algorithm has been proven to be 2 − 1/m competitive for a load balancing optimality criterion, where m is the number of machines. With a single machine (channel) this is 1-competitive (i.e., the same as optimal), with a large number of machines (channels) this approaches 2-competitive (i.e., two times worse than optimal).
9.4.3.2 Parallel Machine Model Using our multichannel EPON scheduling model we explore scheduling policies found in the scheduling research literature. If the machine eligibility constraints can be relaxed we obtain P|| i Ci which is solved to optimality by the SPT (shortest processing time first) rule [26]. Without relaxing the machine eligibility constraints, Least Flexible Job (LFJ) first scheduling is proven optimal [26] for P|Mi , pi = 1| i Ci and P|Mi , pi = 1|Cmax ( pi is the processing time of job i) if the Mi have a special nesting structure. The special nesting structure between the machine eligibility constraints for two ONUs holds if one and only one of the following relationships holds for ONUs i and j: • • • •
Mi is equal to M j Mi is a subset of M j M j is a subset of Mi Mi and M j do not overlap
This nesting structure of the supported wavelengths is not guaranteed for all WDM upgrade scenarios of EPONs. The pi = 1 component means that bandwidth requirements of all the ONUs are equal, or that each bandwidth unit is considered as a separate job. This could produce fragmentation if the individual bandwidth units are not scheduled consecutively (which would increase the number of gaps in the schedule due to guard times required between transmission grants). If we remove the pi = 1 requirement and/or the nesting structure of Mi , then LFJ is a heuristic for the problem.
9.4.3.3 Unrelated Machine Model Another possible approach to modeling the multichannel EPON grant scheduling problem is to loosen our original model by recognizing that P|Mi | i Ci can be viewed as a special case of R|| i Ci , where R refers to an unrelated machine
9 MultiChannel EPONs
211
environment where each machine executes a job at a different speed. For machines that are in Mi we set the execution time on these machines to the processing time or grant length, for machines not in Mi , we set the execution time on these machines to ∞. We will now pursue scheduling policies for this alternative unrelated machine environment model for the multichannel EPON scheduling problem. R|| i Ci can be formulated for an optimal solution [26] as an integer program with a special structure that yields an Integer Solution under LP-relaxation (Linear Program Relaxation). A common method used to solve this problem is the weighted bipartite matching. A weighted bipartite matching problem in which the number of jobs and number of machines are equal is referred to as an Assignment Problem. The time complexity of weighted bipartite matching is O(n(m + nlogn)), where in our case, m is the number of wavelengths and n is the number of ONUs. The integer program for weighted bipartite matching is formulated as follows: minimize
m n n
kp ji x jki
(9.1)
j=1 i=1 k=1
subject to
m n
x jki = 1, ∀i
(9.2)
x jki ≤ 1, ∀ j , ∀k
(9.3)
j=1 k=1 n i=1
where k is the scheduling position, p ji the grant processing time for ONU i on channel j (either ONU grant time for supported channel j, or ∞ for non-supported channel j), x jki are binary variables representing whether or not position k on machine (channel) j is selected for job (ONU) i, m the number of machines (channels), and n the number of jobs (ONUs).
9.4.3.4 Weighted Bipartite Matching Adapted for Online JIT The standard weighted bipartite matching (WBM) scheduling formulation [26] for minimizing the sum of the completion times needs to be modified to support an online scheduling framework. In any online scheduling framework, not all machines are available for scheduling immediately (i.e., they are still processing some jobs). We introduce this in the WBM formulation by setting an additive cost to a matching that is different for each machine. This additive cost is related to when the wavelength becomes available. We refer to this cost as a ji , the availability cost of wavelength j for ONU i. Let tiRTT be the RTT delay for ONU i, tiREPORT be the time the REPORT message from ONU i is received at the OLT, Λ j be the time when wavelength j is free, and tiREADY be the time when ONU i is ready to transmit. Given
212
Michael P. McGarry and Martin Reisslein
tiREADY = tiREPORT + tiRTT then a ji =| Λ j − tiREADY |. a ji is a measure of the difference between wavelength free time and ONU REPORT time. A weight can be used to control how much this availability cost can affect the solution, we will use δ to represent this weight. We also introduce an optional cost to the WBM formulation to incorporate priorities, we refer to this cost as, ri , the priority cost for ONU i. This priority cost is a linear combination of (1) age of ONU’s request, (2) ONU’s class, and (3) ONU’s fairness metric. The priority cost ri is formulated as ri = α ∗ poolT ime + β ∗ ci + ω ∗ f air ness Metric, where i is the ONU number, ci the class of ONU i, α the pool time weight, β the traffic class weight, and ω the fairness metric weight. α, β, and ω are parameters that need to be optimized, either using a linear program or through experimentation to achieve an appropriate mix of priority and fairness. The following is the integer program that represents the WBM where k is the scheduling position, p ji the grant processing time for ONU i on channel j (either ONU grant time for supported channel j, or ∞ for non-supported channel j), a j the availability cost for channel j, ri the optional priority cost for ONU i, x jki are binary variables representing whether or not position k on machine (channel) j is selected for job (ONU) i, m the number of machines (channels), and n the number of jobs (ONUs): minimize
m n n (k · p ji + δ · a j + ri ) · x jki (1)
(9.4)
j=1 i=1 k=1
subject to
m n
x jki = 1, ∀i
(9.5)
x jki ≤ 1, ∀ j , ∀k
(9.6)
j=1 k=1 n i=1
The first constraint forces an ONU i to be assigned to only one scheduling position. The second constraint forces each scheduling position to be assigned to no more than one ONU. If a single ONU supplies traffic from multiple classes, each traffic class is treated as a separate job in the WBM formulation.
9.4.3.5 Summary Using our parallel machine model, the results from scheduling theory indicate a few dispatching rules that can provide good scheduling policies. A dispatching rule (see Sections 14.1 and 14.2 in [26]) is a defined method of ordering jobs for dispatch in first fit fashion on available machines. Some examples of general dispatching rules are least flexible job (LFJ) first, shortest processing time (SPT) first, and largest processing time (LPT) first. The best scheduling policies for our multichannel EPON grant scheduling model are the dispatching rules discussed in Section
9 MultiChannel EPONs
213
9.4.3.2; LFJ for machine eligibility restrictions and SPT for minimizing the sum of the completion times. Other potential dispatching rules for the multichannel EPON grant scheduling problem are largest processing time (LPT) first, largest number of frames (LNF) first, earliest arriving frame (EAF) first, and earliest average arrival (EAA) first [6, 19, 23, 31, 32]. LNF will favor ONUs with more Ethernet frames queued, EAF will favor ONUs that have the earliest arriving head-of-line (HOL) Ethernet frame, and EAA will favor ONUs that have the earliest average Ethernet frame arrival time. Dispatching rules can be used alone or grouped together to form composite dispatching rules. For multichannel EPON grant scheduling, LFJ can be combined with some of the other dispatching rules to create a composite dispatching rule that can provide better performance. The second dispatching rule is used to break ties from the first dispatching rule. An example of a composite dispatching rule is LFJ– LNF. In LFJ–LNF, ONUs are ordered by the number of wavelengths they support for upstream transmission, ties are then broken by the number of frames queued at each ONU. The ONUs are then dispatched for transmission in the order specified by LFJ–LNF. Rather than using the second dispatching rule for tie breaking in the first dispatching rule, a weight can be set for each dispatching rule in the composite. Therefore, the ordering is a true composite of the two rules. To compute a schedule using a dispatching rule, the OLT creates an ordered list of ONUs according to the dispatching rule. The OLT, traversing the list in order, schedules the next ONU’s transmission on the first available channel supported by that ONU (i.e., NASC). For example, if the LFJ–LNF scheduling algorithm is to be used, the OLT orders ONUs in increasing order of the number of channels they support, ONUs supporting the fewest channels would appear first. The OLT then breaks ties with the number of frames queued at the ONU. For LNF, ONUs with more frames queued would be scheduled first. Alternatively, for LPT larger grant sizes would be scheduled first, and for SPT the smaller grant sizes would be scheduled first. Once a sorted list of ONUs is produced this list is used to generate the schedule as discussed above. Essentially, application of this type of scheduling policy (i.e., a dispatching rule) is implemented by a sorting operation. The weighted bipartite matching (WBM) formulation that is proven optimal for minimizing the sum of the completion times for the unrelated machine environment scheduling model operates differently from the dispatching rules. Unlike the dispatch rules, the WBM scheduling policy results in a direct assignment of each ONU grant to a specific channel and position. There is no need to apply NASC for the scheduling. The output from the WBM scheduling policy sufficiently produces the schedule. This is in contrast to the dispatching rules that specify the order in which the ONU grants are scheduled for first fit channel assignment according to NASC.
9.5 Summary In conclusion, grant scheduling in multichannel EPONs (e.g., WDM EPONs) can be facilitated with static or dynamic wavelength assignment capabilities. With static
214
Michael P. McGarry and Martin Reisslein
wavelength assignment (SWA) an EPON cannot take advantage of statistical multiplexing across wavelengths. Access network traffic is known to be very bursty in nature, therefore statistical multiplexing provides significantly better channel utilization. Without the ability to perform statistical multiplexing, SWA suffers low channel utilization. However, SWA allows for the simplest and the most cost-effective WDM PON architectures [2, 7, 13]. DWA allows for much higher channel utilization as a result of statistical multiplexing. This higher channel utilization facilitates lower queueing delays experienced by Ethernet frames transiting through the EPON. DWA is best managed through grant by grant wavelength assignment which can be combined with grant ordering to form a multidimensional scheduling problem (e.g., time and wavelength). This multidimensional scheduling problem can be modeled using standard scheduling theory model notation and optimal solution methods can be found [26]. We have divided the grant scheduling problem into two layers. The first layer is the scheduling framework which determines when an OLT will make access decisions. The most straightforward scheduling frameworks are the online and offline scheduling frameworks. In the online scheduling framework, the OLT schedules an ONU’s next transmission as soon as its MPCP REPORT message is received. This online scheduling framework provides high channel utilization by operating in work-conserving fashion. However, access decisions are not made with the latest queue occupancy data from all ONUs. As a result, no sophisticated access decisions can be made in this framework. In the offline scheduling framework, the OLT schedules all ONUs’ next transmissions at the same time once MPCP REPORT messages from all ONUs have been received [9]. This allows the OLT to make sophisticated access decisions since it knows the queue occupancies of all ONUs. However, significant channel capacity is wasted while the OLT waits to receive the MPCP REPORT messages from all ONUs. This wasted capacity results in significantly lower maximum achievable channel utilization. As the number of channels increases, the significance of the wasted channel capacity increases. There are two scheduling frameworks that attempt to mitigate the wasted channel capacity problem of offline scheduling while still gaining better access decision making. These two are hybrids between offline and online scheduling. The first is ONU load status hybrid scheduling that is designed to work with the only grant sizing technique that requires all of the ONU queue occupancy information, the limited with excess distribution scheme. This scheduling framework schedules ONUs whose queue occupancies are less than their guaranteed grant size, called underloaded ONUs, in online fashion since their grant size is not impacted by the queue occupancies of other ONUs. On the other hand, the ONUs whose queue occupancies are greater than their guaranteed grant size, called overloaded ONUs, are scheduled in offline fashion since their grant size is impacted by the queue occupancies of other ONUs. Under this scheduling framework, those ONUs that are overloaded will have their transmission grants delayed which may further exacerbate their backlog status. As a result, when an ONU becomes overloaded it may absorb into this overloaded state. Further, as the load on the network increases and many ONUs have queue occupancies greater than their guaranteed grant size, this
9 MultiChannel EPONs
215
schemes degenerates to the offline scheduling framework with respect to channel utilization. The second hybrid framework, online Just-in-Time (JIT) scheduling, uses channel availability to drive the scheduling process and as a result provides channel utilization that is as high as is achievable with the online scheduling framework. Online JIT scheduling can however provide better access decisions because it can consider the queue occupanices of multiple ONUs concurrently when making decisions. We have found limited improvement in lowering average queueing delay at the ONUs using online JIT but have found online JIT to provide significant improvements for other scheduling objectives, such as providing differentiated treatment of ONUs [24]. In practice, the best scheduling framework will likely lie between online just-intime and offline in the scheduling framework continuum illustrated in Fig. 9.6. It may be a good practice to force some channel idleness to make better access decisions. However, resorting to a completely offline scheduling framework is very inefficient. There are two reasons for considering the latest queue occupancy data of all ONUs: (1) to facilitate grant sizing decisions and (2) to provide better grant scheduling decisions. The only grant sizing scheme that considers the queue occupancy data of all ONUs is the limited with excess distribution scheme. In this chapter, we also explored scheduling policies for the EPON multichannel grant scheduling problem. We have found, through extensive simulations, that the scheduling policies discussed provided only a minor reduction in average queueing delay [24]. Largest number of frames (LNF) first scheduling helped reduce the average queueing delay at high loads due to a frame sampling effect, where ONUs with more frames queued will have a larger impact on the queueing delay measure. The scheduling policies that had the largest impact were those that were not designed to lower average queueing delay but those that were designed to treat ONUs differently. Acknowledgments Michael McGarry would like to acknowledge his wife Yesenia McGarry and his daughter Nina McGarry for their support during his PhD studies that produced much of the content presented in this chapter.
References 1. Lucent’s LamndaXtreme Transport http://www.lucent.com 2. An FT, Kim KS, Gutierrez D, Yam S, Hu E, Shrikhande K, Kazovsky L (2004) SUCCESS: A next-generation hybrid WDM/TDM optical access network architecture. IEEE/OSA Journal of Lightwave Technology 22:11:2557–2569 3. Assi CM, Ye Y, Dixit S, Ali MA (2003) Dynamic bandwidth allocation for quality-of-service over Ethernet PONs. IEEE Journal on Selected Areas in Communications 21:9:1467–1477 4. Bai X, Shami A, Assi C (2006) On the fairness of dynamic bandwidth allocation schemes in Ethernet passive optical networks. Computer Communications 29:11:2123–2135
216
Michael P. McGarry and Martin Reisslein
5. Banerjee A, Kramer G, Mukherjee B (2006) Fair sharing using dual service-level agreements to achieve open access in a passive optical network. IEEE Journal on Selected Areas in Communications 24:8:32–44 6. Bhatia S, Bartos R (2007) IPACT with smallest available report first: A new DBA algorithm for EPON. Proceedings of IEEE ICC 2168–2173 7. Bock C, Prat J, Walker SD (2005) Hybrid WDM/TDM PON using the AWG FSR and featuring centralized light generation and dynamic bandwidth allocation. IEEE/OSA Journal of Lightwave Technology 23:12:3981–3988 8. Clarke, F, Sarkar S, Mukherjee B (2006) Simultaneous and interleaved polling: an upstream protocol for WDM-PON. Proceedings of Optical Fiber Communication Conference 9. Dhaini AR, Assi CM, Shami A (2006) Dynamic bandwidth allocation schemes in hybrid TDM/WDM passive optical networks. Proceedings of IEEE CCNC 30–34 10. Dhaini AR, Assi CM, Shami A (2006) Quality of service in TDM/WDM Ethernet passive optical networks (EPONs). Proceedings of IEEE ISCC 2006 616–621 11. Ghani N, Shami A, Assi C, Raja MYA (2004) Intra-ONU bandwidth scheduling in Ethernet passive optical networks. IEEE Communications Letters 8:11:683–685 12. Graham RL (1966) Bounds for certain multiprocessing anomalies. Bell System Technical Journal 45:1563–1581 13. Hsueh YL, Rogge MS, Yamamoto S, Kazovsky L (2005) A highly flexible and efficient passive optical network employing dynamic wavelength allocation. IEEE/OSA Journal of Lightwave Technology 23:1:277–286 14. Kramer G, Mukherjee B, Pesavento G (2002) IPACT: A dynamic protocol for an Ethernet PON (EPON). IEEE Communications Magazine 40:2:74–80 15. Kwong KH, Harle D, Andonovic I (2004) Dynamic bandwidth allocation algorithm for differentiated services over WDM EPONs. Proceedings of The Ninth International Conference on Communications Systems 116–120 16. Luo Y, Ansari N (2005) Bandwidth allocation for multiservice access on EPONs. IEEE Communications Magazine 43:2:S16–S21 17. Luo Y, Ansari N (2005) Limited sharing with traffic prediction for dynamic bandwidth allocation and QoS provisioning over Ethernet passive optical networks. OSA Journal of Optical Networking 4:9:561–572 18. Ma M, Zhu Y, Cheng T (2003) A bandwidth guaranteed polling MAC protocol for Ethernet passive optical networks. Proceedings of IEEE INFOCOM 1:22–31 19. Ma M, Liu L, Cheng TH (2004) Adaptive scheduling for differentiated services in the ethernet passive optical networks. Proceedings of The Ninth International Conference on Communications Systems 102–106 20. Marsan MA, Roffinella D (1983) Multichannel local area network protocols. IEEE Journal on Selected Areas in Communications 1:885–897 21. McGarry M (2004) An evolutionary wavelength division multiplexing upgrade for ethernet passive optical networks. Master’s thesis, Arizona State University 22. McGarry MP, Maier M, Reisslein M (2004) Ethernet PONs: A survey of dynamic bandwidth allocation (DBA) algorithms. IEEE Communications Magazine 42:8:S8–S15 23. McGarry MP, Reisslein M, Maier M, Keha A (2006) Bandwidth management for WDM EPONs. OSA Journal of Optical Networking 5:9:637–654 24. McGarry MP, Reisslein M, Colbourn CJ, Maier M, Aurzada F, Scheutzow M (2008) Justin-time scheduling for multi-channel EPONs. IEEE/OSA Journal of Lightwave Technology 26:10:1204–1216 25. McGarry MP, Reisslein M, Maier M (2008) Ethernet passive optical network architectures and dynamic bandwidth allocation algorithms. IEEE Communications Surveys and Tutorials 10:3:46–60 26. Pinedo M (2002) Scheduling: Theory, algorithms, and systems, 2nd edition. Prentice Hall, Englewood cliffs 27. Shami A, Bai X, Assi C, Ghani N (2005) Jitter performance in Ethernet passive optical networks. IEEE/OSA Journal of Lightwave Technology 23:4:1745–1753
9 MultiChannel EPONs
217
28. Shami A, Bai X, Ghani N, Assi CM, Mouftah HT (2005) QoS control schemes for two-stage Ethernet passive optical access networks. IEEE Journal on Selected Areas in Communications 23:8:1467–1478 29. Sherif SR, Hadjiantonis A, Ellinas G, Assi C, Ali MA (2004) A novel decentralized ethernetbased PON access architecture for provisioning differentiated QoS. IEEE/OSA Journal of Lightwave Technology 22:11:2483–2497 30. Zang H, Jue J, Mukherjee B (2000) A review of routing and wavelength assignment approaches for wavelength-routed Optical WDM networks. Optical Networks Magazine 1:1:47–60 31. Zheng J, Mouftah HT (2005) Media access control for Ethernet passive optical networks: an overview. IEEE Communications Magazine 43:2:145–150 32. Zheng J, Mouftah HT (2005) Adaptive scheduling algorithms for Ethernet passive optical networks. IEEE Proceedings on Communications 152:643–647
Chapter 10
Long-Reach Optical Access Huan Song, Byoung-Whi Kim, and Biswanath Mukherjee
Abstract In this chapter we present the current status of broadband deployment in Europe and discuss the drivers for further deployment and the expected evolution in terms of the market, the services, the choice of technical solutions, and the main players. The main drivers for broadband are identified and discussed in the context of the European environment. We present an overview of the broadband regulatory framework in Europe as well as expected developments and discuss its impact on the evolution of broadband in this part of the world.
10.1 Introduction Much of the R&D emphasis in recent years has been on developing high-capacity backbone networks. Backbone network operators currently provide high-capacity OC-192 (10 Gbps) links, with 40 Gbps transmission also quite mature now [1, 2]. However, today’s access network technologies such as digital subscriber line (DSL) typically provide 1.5 Mbps of downstream bandwidth and 128 kbps of upstream bandwidth. While economies of scale have successfully enabled backbone networks to grow rapidly, the cost of access technologies remains prohibitively high for the average household [3]. The access network is the bottleneck for providing broadband services such as video-on-demand, interactive games, and video conferencing to end users. Huan Song Department of Computer Science, University of California, Davis, CA, USA, e-mail:
[email protected] Byoung-Whi Kim Electronics and Telecommunications Research Institute, Daejeon, Korea, e-mail: e-mail:
[email protected] Biswanath Mukherjee Department of Computer Science, University of California, Davis, CA, USA, e-mail:
[email protected]
A. Shami et al. (eds.), Broadband Access Networks, Optical Networks, c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-92131-0 10,
219
220
Huan Song, Byoung-Whi Kim, and Biswanath Mukherjee
In addition, DSL has a limitation that any of its subscribers must be within 18,000 ft from the central office (CO) because of signal distortions. Typically, DSL providers do not offer services to customer more than 12,000 ft away. Therefore, only an estimated 60% of the residential subscriber base in the United States can use DSL even if they were willing to pay for it. Although variations of DSL such as very-high-bit-rate DSL (VDSL), which can support up to 50 Mbps of downstream bandwidth, are gradually emerging, these technologies have even more severe distance limitations. For example, the maximum distance that VDSL can be supported over is limited to 1,500 ft. Some other variants of DSL include G.SHDSL (offering 2.3 Mbps in both directions), ADSL2 (offering 12 Mbps downstream), and ADSL2+ (offering 25 Mbps downstream). Another alternative for broadband access is through cable television (CATV). CATV networks provide Internet services by dedicating some radio frequency (RF) channels in co-axial cable for data. However, CATV networks are mainly built for delivering broadcast services, so they do not fit well for distributing access bandwidth. At high load, the network’s performance is usually frustrating to end users. Moreover, it is expected that emerging applications such as Internet protocol TV (IPTV), video-on-demand (VoD), video file swapping, peer-to-peer applications, real-time network games, will demand much more bandwidth, and some of these applications may require symmetric bandwidth as well (equal and high bandwidth in both downstream and upstream directions) [4, 5]. Both DSL and CATV provide limited and asymmetric bandwidth access (lesser bandwidth in the upstream direction), with the implicit assumption that present traffic is more downstream oriented. Emerging web applications require unprecedented bandwidth, exceeding the capacity of traditional VDSL or CATV technologies. The explosive demand for bandwidth is leading to new access network architectures which are bringing the highcapacity optical fiber closer to the residential homes and small businesses [3]. The FTTx model – Fiber to the home (FTTH), fiber to the curb (FTTC), fiber to the premises (FTTP), etc. – offers the potential for unprecedented access bandwidth to end users (up to 100 Mbps per user). These technologies aim at providing fiber directly to the home, or very near the home, from where technologies such as VDSL or wireless can take over. FTTx solutions are mainly based on passive optical network (PON). Developments in PON in recent years include Ethernet PON (EPON), ATM-PON (APON), broadband-PON (BPON) (which is a another name for APON), gigabit-PON (GPON) (which is also based on ATM switching), and wavelength-division-multiplexing PON (WDM-PON) [6–9]. According to a recent review, broadband access is penetrating into residential and business users. The number of broadband subscribers in the countries forming the Organization for Economic Co-operation and Development (OECD) increased 24% from 178 million in June 2006 to 221 million subscribers in June 2007 [10]. Asian countries are advancing much faster in broadband access (e.g., fiber connections account for 36% of all Japanese broadband subscriptions and 31% in Korea) [10]. Japan plans to serve 30 million optical users by 2010.1 1
https://www.ntt-review.jp/archive/ntttechnical.php?contents=ntr200706sf1.html
10 Long-Reach Optical Access
221
The Australian Government’s “BroadbandNow” project promises to deploy in 5 years broadband access nationwide, with the objective to improve the delivery of health, education, and other essential services. A new technology, called long-reach passive optical network (LR-PON), extends the coverage span of PONs mentioned above from 20 to 100 km and beyond by exploiting optical amplifier and WDM technologies. Compared with traditional PON, LR-PON consolidates the multiple optical line terminals (OLTs) and the central offices (COs) where they are located, thus significantly reducing the corresponding Operational Expenditure (OpEx) of the network. By providing extended geographic coverage, LR-PON combines optical access and metro into an integrated system. British Telecom was one of the first network operators to envision this cost-effective and easy-to-manage future Internet by proposing the attachment of the LR-PON to backbone networks. Figure 10.1 shows an overview of telecom networks. It consists of the access network, the metropolitan-area network, and the backbone network. However, with the maturing of technologies for long-reach broadband access, the traditional metro network is getting absorbed in access. As a result, the telecom network hierarchy can be simplified with the access headend close to the backbone network. Thus, the network’s Capital Expenditure (CapEx) and Operational Expenditure (OpEx) can be significantly reduced due to the need for managing fewer control units. But this architecture also brings with it its own new research challenges, which will be outlined in this chapter.
Fig. 10.1 Telecom network overview [1, 2]
10.2 Research Challenges Figure 10.2 shows the general architecture of an LR-PON. The central office (CO) connects the core network and the access network and implements layer 2 and layer 3 functions, e.g., resource allocation, service aggregation, management, and control.
222
Huan Song, Byoung-Whi Kim, and Biswanath Mukherjee Drop (10 km)
ONU
Core Network
Centr al Offic e (CO) (OLT
Feeder (up to 100 km and beyond)
Local Exchange ONU Single channel or WDM Large number of users
ONU
Fig. 10.2 Long-reach PON (LR-PON) architecture
The local exchange resides in the local users’ area, which is close to the end customer equipment: ONU (within 10 km of drop section). The optical signal propagates across the fiber forming the feeder section (100 km and beyond) with the CO and the local exchange at its two ends; then the fiber is split and connected to a large number of ONUs. In order to compensate for the power loss due to long transmission distance and high split size, optical amplifiers are used at the OLT and the local exchange.
10.2.1 Signal Power Compensation Optical amplification is indispensable in an LR-PON. Besides amplifying the signal, the amplifiers introduce two challenges, as indicated below. 1. Optical amplifiers have a detrimental effect on system performance because of unwanted noise, known as amplified spontaneous emission (ASE), a side effect of the amplification mechanism. As the high split of LR-PON would attenuate the signal significantly, the optical signal power could be sufficiently low at the input of the amplifier. As a result, the signal-to-noise ratio (SNR) could be reduced significantly because ASE remains constant. In order to amplify the optical signal while suppressing the noise, a possible scheme, called dual-stage intermediate amplification, was introduced [11]. In this scheme, the first stage is composed of a low-noise preamplifier, which produces a high SNR by maintaining its ASE at a low level and the second stage consists of amplifiers to amplify the optical signal with enough power, in order to counter the large attenuation in the feeder section (100 km and beyond). 2. Two types of optical amplifiers are used in LR-PON: erbium-doped fiber amplifier (EDFA) and semiconductor optical amplifier (SOA). The EDFA features a low noise figure, a high power gain, and a wide working bandwidth, which enable it to be advantageous in an LR-PON employing WDM. But the relatively slow speed in adjusting its gain makes it disadvantageous due to the bursty nature
10 Long-Reach Optical Access
223
of upstream TDMA traffic in LR-PON, where the optical amplifier needs to adjust its gain fast enough when packets with different DC levels pass through it, in order to output packets with uniform signal amplitude. A possible solution is to use an auxiliary wavelength that is adjusted relative to the transmitted upstream packet so that the optical power through the EDFA remains constant. Hence, the gain of EDFA remains constant for the burst duration. On the other hand, an SOA can be adjusted faster and offers the potential for monolithic or hybrid array integration with the other optical components which makes it more cost competitive. The single-channel SOA amplifier is suitable for the “pay-to-grow” business model where only the channel for premium user is amplified [11].
10.2.2 Optical Source In order to lower the CapEx and OpEx, a standard PON chooses lower-cost uncooled transmitters in the ONU, because a major investment for an optical access network is the cost associated with installing an optical transmitter and receiver in the ONU at the customer premises [3]. However, the uncooled transmitter is temperature dependant which in turn transmits wavelength with a possible drift of 20 nm. As no component in a standard PON is wavelength critical, the performance may be unaffected. But in an LR-PON which exploits WDM to satisfy huge amount of traffic, the wavelength drift become crucial, especially for certain components such as optical filters. To counter the wavelength drift, more expensive cooled transmitters are considered to ensure a stable wavelength. The transmitters in a traditional PON are usually designed for a transmission range which is less than 20 km. Another challenge arises when applying them in an LR-PON where the signal needs to cover a range of 100 km and beyond.
10.2.3 Burst-Mode Receiver The different ONU–OLT distances mean different propagation attenuations for signals from ONUs to the OLT, which in turn results in varied DC levels of bursty packets from ONUs at the OLT. The optical amplifier increases the difference of the DC level of upstream signals from different ONUs, which requires good design of burst-mode receivers at the OLT. In current PONs, a DC-coupled receiver is used which requires the receiver to decide the correct DC power threshold on a burst-by-burst basis. Problem occurs when the PON scales up in speed (10 Gbps and beyond) and number of customers supported (up to 512 users could share the same channel). Efforts have been made by researchers [12] to develop a new 10 Gbps burst-mode receiver that uses multi-stage feed-forward architecture to reduce DC offsets.
224
Huan Song, Byoung-Whi Kim, and Biswanath Mukherjee
10.2.4 Upstream Resource Allocation In LR-PON, the end users and the central office (CO) (through which users are connected to the rest of the Internet) are separated by a significant distance, typically 100 km and beyond. Hence, control-plane delays are significant, so that various known scheduling algorithms for packet-based networks are difficult to apply directly. Therefore, efficient remote-scheduling algorithms need to be developed which overcome the large CO-user distance, which support different classes of service, and which are scalable in terms of the number of users supported as well. This delay issue is particularly important since the delay budget in an access network is approximately 1–2 ms for real-time applications, and taking into account the coastto-coast propagation delay (approximately 25 ms), which is much larger for global scales, and the persistence of hearing of the human ear and the persistence of vision of the human eye, both of which are in the range of a few tens of milliseconds.
10.3 Demonstrations of LR-PON 10.3.1 PLANET SuperPON The ACTS-PLANET (Advanced Communication Technologies and ServicesPhotonic Local Access NETwork) [11, 13–17] is an EU-funded project, which explored possible upgrades of a G.983-like APON system, named super passive optical network (SuperPON) on the level of range, splitting factor, number of supported ONUs, and bit rates. The project consisted of a consortium of telecom operators and suppliers to develop a cost-effective, full-service access network. The basic architecture is depicted in Fig. 10.3, which was installed in the first quarter of 2,000. The implemented system achieves a total splitting factor of 2,048
Feeder (90 km) Feeder repeater
Drop (10 km) Local exchange WDM
ONU
OLT WDM
OAM ONU
EDFA
OAM ONU
SOA
ONU Splitting ratio: 2048
OAM ONU
Operation and maintenance
Fig. 10.3 ACTS-PLANET architecture and transport system [11, 13–17]
10 Long-Reach Optical Access
225
and a span of 100 km. The 100-km fiber span consists of a maximum feeder length of 90 km and a drop section of 10 km. The large splitting factor is achieved through two-stage splitting at the drop section. The increased transmission range from 20 to 100 km and split size from 32 to 2,048 introduces large attenuation. To compensate for the signal power loss, optical amplifiers are located at the feeder section (feeder repeater) and at the intersection between the feeder and the drop sections (amplified splitter). Due to the broadcast property on the downstream transmission, the cost of amplifiers is shared among the end users, which makes it less cost sensitive. As a result, EDFAs are used because of their high power gain and wide working bandwidth. In the upstream direction, SOAs are used due to the bursty nature of the traffic originating from the different ONUs. SOAs are placed both at the input and at the output of the splitter to amplify the optical signal before it is attenuated by a split in order to achieve the high split size of 2,048. Otherwise, if SOAs are only placed after the splitter, the SNR cannot satisfy the system requirements (an S N R > 18.6 dB was required to achieve the PLANET target performance bit-error rate (BER) of 10−9 , using ON–OFF keying and including a 3 dB optical power margin [11]), concerning the ASE noise discussion in Section 10.2. The transport system is based on asynchronous transfer mode (ATM), which offers 2.5 Gbps and 311 Mbps for downstream and upstream transmissions. A time-division multiple access (TDMA) protocol is used to allocate the upstream bandwidth among multiple users. Besides synchronizing the ONUs’ transmission, the protocol also synchronizes the SOAs to set the correct gain to compensate for different power losses of upstream signals from different ONUs. The operation administration and maintenance ONU (OAM-ONU) is responsible for this function. It receives and interprets all downstream control information. In this way, it can calculate information on which ONUs are granted access to the upstream path at a certain instant in time. The switching transient between two subsequent upstream cells can be lower than 25 ns. Other possible architectures for PLANET are discussed in [17], which use different types of amplifiers. Detailed calculation on the effect of accumulated ASE noise of the amplifiers is included in [15].
10.3.2 British Telecom (BT) British Telecom has demonstrated its long-reach PON, which is characterized by a 1,024-way split, 100-km reach, and 10-Gbps transmission rate for upstream and downstream directions [18]. Compared with SuperPON, the splitting size has been halved. This brings a major benefit by saving the amount of optical amplifiers needed, e.g., 6 optical amplifiers are enough for upstream and downstream as opposed to 39 required by SuperPON. The 1,024-way split is made up of a cascade of two N:16 and one N:4 splitters in the drop section.
226
Huan Song, Byoung-Whi Kim, and Biswanath Mukherjee
The system includes a 90-km feeder section between the OLT and the local exchange and a 10-km drop section between the local exchange and end users. In order to boost the signal attenuated by large splitting and long-distance transmission, optical amplifiers are used at the OLT and the local exchange. In order to lower the CapEx and OpEx of LR-PON, the ONU for the large number of end users has to be simple and cost efficient. As a result, the transmitter output power in an ONU is constrained because it is a key price factor for the optical transmitter. Besides the constrained signal power, the drop section has a large power loss of 40.3 dB before the signal is amplified. Hence, a very low signal power will arrive at the local exchange. To boost the signal to enough power, a two-stage amplifier as discussed in Section 10.2 is implemented, which contains (1) a low-noise preamplifier which produces a high SNR by maintaining its ASE at a low level and (2) second-stage amplifiers to amplify the optical signal with enough power. Other technologies used include forward error correction (FEC) and optical filter. FEC is a coding technique by which transmission errors can be detected and corrected by encoding the data and including a number of parity bits. FEC can alleviate the system design requirement by allowing a relatively higher pre-FEC BER, e.g., a B E R of 10−10 required by an LR-PON only needs a pre-FEC B E R of 2.9 × 10−4 . Optical filter can also promote the SNR of the received signal by reducing the ASE noise in the signal passing through. The above demonstration [18] concerned mainly the physical layer to extend the access network, e.g., power budget. However, a later demonstration [19] showed how to integrate the higher-layer issues, e.g., control protocol into LR-PON design. This demonstration extended gigabitPON (GPON) from 20 to 135 km and integrated WDM to increase the system capacity. The experimental configuration in Fig. 10.4 shows an extended GPON with 40-channel WDM system. The OLT transmits downstream at a wavelength in the 1,490-nm region, according to GPON standard. The data rate for each downstream
Central Office
125 km
Local exchange
2.5 Gbps Bespoke transponder OLT
40-lambda WDM
ONU
OEO
1.2 Gbps
ONU
x 64 split
Fig. 10.4 Experimental configuration of GPON extended to 135 km via WDM [19]
10 Long-Reach Optical Access
227
channel is 2.488 Gbps. This downstream wavelength will be converted to a WDMcompatible wavelength of 1,552.924 nm by a transponder right after the OLT. The conversion is necessary because the long-distance transmission exploits a WDM system of 125-km standard G.652 fiber with an EDFA attached to each end. When arriving at the drop section, the wavelengths are converted back to 1,490-nm range through a bespoke transponder for compliance with the GPON standard. Then, each wavelength is split into 64 ONUs within the 10-km drop section. The demonstration includes three GPON ONUs in the upstream direction with different upstream wavelength ranges: 1,310 or 1,550 nm. These different wavelengths have the same transmission rate of 1.244 Gbps and will be converted at the bespoke transponder to a WDM-compatible wavelength of 1,559.412 nm. At the OLT side, an optical filter is placed before the burst-mode receiver to improve the SNR. Due to the incorporation of the optical filter, the temperature control of the system must be precise in order to prevent wavelength drift. The demonstration achieved a B E R better than 10-10 in both directions. It also demonstrated that, by using OEO conversion at the transponder and optical amplification, the GPON can be extended to beyond 100-km range.
10.3.3 University College Cork, Ireland The demonstration of a hybrid WDM-TDM LR-PON is reported in [20]. This work supports multiple wavelengths, and each wavelength pair (upstream and downstream) can support a PON segment with a long distance (100 km) and a large split ratio (256 users). The layout in Fig. 10.5 is divided into four notional locations: (1) customer ONU, (2) street cabinet, (3) local exchange, and (4) core exchange. The core exchange and local exchange are powered to support signal amplification on both ends of the longdistance fiber transmission (88 km). The street cabinet contains cascaded optical splitters to achieve a large split size of 256. As WDM requires precise wavelength control (50 or 100 GHz channel spacing), a WDM centralized source is placed at
Core exchange AWG
88 km SSMF
Local exchange Customer ONU
Street Cabinet
6 km
Tx Ch1
6 km
RBF
RX EAM SOA 1xN split
Tx Ch17 DFB Ch24 Rx Ch1
Rx Ch17
DWDM Centralized Source DFB Ch40
1st TDM-PON 17th TDM-PON
Fig. 10.5 Hybrid DWDM-TDM long-reach PON architecture [20]
228
Huan Song, Byoung-Whi Kim, and Biswanath Mukherjee
the local exchange, which is composed of an array of distributed feedback (DFB) laser diodes to generate upstream carrier wavelengths. These upstream carrier wavelengths are split and fed to each ONU through the street cabinet. The customer ONU is colorless which uses a semiconductor electro-absorption modulator (EAM) to modulate the upstream carrier wavelength generated in the local exchange, and two SOAs are connected to the input and output of EAM to amplify the upstream wavelength before and after the modulation. Each upstream and downstream wavelength has a transmission rate of 10 Gbps. These wavelengths are achieved by WDM channel allocation, where the C-band is split into two segments: the blue half (1,529–1,541.6 nm) carrying downstream channels and the red half (1,547.2–1,560.1 nm) carrying upstream channels. The red and blue bands are separated by a guard band of approximately 5 nm. With 100-GHz channel spacing, 17 pairs of up and down channels can be achieved to support 17 PON segments in the demonstration. The demonstration offers an LRPON with 17 PON segments, each of which supports symmetric 10-Gbps upstream and downstream channels over a 100-km transmission. The system can serve a large number of users (17 × 226 = 4352 users). Experiments show a low B E R of 10−9 in both directions. In order to scale the long-reach access network, the authors in [10] demonstrated their second-stage prototype: photonic integrated extended metro and access network (PIEMAN) in the Information Society Technologies (IST) 6 framework project. As shown in Fig. 10.6, PIEMAN consists of a 100-km transmission range with 32 channels supported, each of which operates at 10 Gbps and supports a PON segment. The split rate for each PON segment is 512, which means that the maximum number of users supported is 32 × 512 = 16,384.
Central Office
90 km
Local exchange
10 km
32x10 Gbps
ONU
x 512 split OLT
32 lambdas
32x10 Gbps
ONU
EDFA
Single wavelength EDAS
Fig. 10.6 PIEMAN hybrid WDM/TDMA architecture [10]
10 Long-Reach Optical Access
229
10.3.4 Electronics and Telecommunication Research Institute (ETRI), Korea ETRI, a Korean government-funded research institute, has developed a hybrid LR-PON solution, called WE-PON (Wdm-E-PON) [21]. Instead of the “tree and branch” topology in the above demonstrations, WE-PON deployed a ring topology, which provided two-dimensional coverage and was good for network resilience. In WE-PON, 16 wavelengths are transmitted on the ring, and they can be added and dropped to customers through optical add–drop multiplexers (OADMs) on the ring. As the spit ratio of the splitter is 1:32, the system can accommodate 512 customers. With power amplification at the OADMs, the span of WE-PON can reach 100 km and beyond.
10.3.5 Other Demonstrations Recently, more demonstrations of such networks appear in the literature [22–28]. One focuses on transmitting more wavelengths in a fiber using WDM, such as a 35-channel WDM-PON with 50-GHz channel spacing [22]. Another demonstrates possible network configurations in terms of span and capacity, such as 512-ONUs50 km, 1024-ONUs-50 km, and 512-ONUs-100 km [24], and 1:512-split-100 km reach [25]. Other works incorporate novel devices or technologies, such as SOARaman Hybrid Amplifier (SRHA) with 75-nm gain bandwidth [26], an innovative bidirectional amplification which results in a power penalty of less than 0.8 dB after 100 km standard single-mode fiber (SSMF) transmission for all channels at 10 Gbps [27], and a hybrid WDM-CDM-PON with 42 dB loss budget without optical amplifier by using code division multiplexing (CDM) [28]. Another work targets decreasing system cost, such as using inexpensive uncooled lasers at ONUs [23]. As a summary, we list the main demonstrations and their characteristics in Table 10.1.
Table 10.1 Typical prototypes for long-reach broadband access networks Project
Base type
Reach (km)
Wavelengths
Down/up (Gbps) ONUs
ACTS-PLANET B.T. WDM-TDM PIMAN WE-PON
APON GPON
100 135 100 100 100
1 40 17 32 16
2.5/0.311 2.5/1.25 10/10 10/10 2.5/2.5
(G/E)PON
2,048 2,560 4,352 16,384 512
230
Huan Song, Byoung-Whi Kim, and Biswanath Mukherjee
10.4 Dynamic Bandwidth Assignment (DBA) As multiple ONUs share the same upstream channel, DBA is necessary among ONUs. Considering the LR-PON’s benefits in CapEx and OpEx, as well as its derivation from the traditional PON, the upstream bandwidth allocation is controlled and implemented by the OLT. To support the DBA arbitration in the OLT, the proposed DBA algorithms in LR-PON are based on the multi-point control protocol (MPCP) specified in the IEEE 802.3ah standard. Before explaining the DBA algorithms, we briefly introduce MPCP. MPCP is not concerned with a particular bandwidth-allocation (or inter-ONU scheduling) scheme; rather, it is a supporting mechanism that can facilitate implementation of various bandwidth-allocation algorithms in PON. This protocol relies on two specific messages: GATE and REPORT. (Additionally, MPCP defines REGISTER REQUEST, REGISTER, and REGISTER ACK messages used for an ONU’s registration.) A GATE message is sent from the OLT to an ONU, and it is used to assign a transmission time slot (bandwidth). A REPORT message is used by an ONU to convey its local conditions (such as buffer occupancy and the like) to the OLT to help the OLT make intelligent allocation decisions. Both GATE and REPORT messages are MAC (media access control) control frames and are processed by the MAC control sub-layer. The proposed DBA algorithms work in conjunction with MPCP. The OLT has to first receive all ONU REPORT messages before it imparts GATE messages to ONUs to notify them about their allocated time slot. As a result, the upstream channel will remain idle between the last packet from the last ONU transmission in a polling cycle k and the first packet from the first ONU transmission in polling cycle k + 1. This idle time equals to the round-trip time (RTT) which is needed for the control messages to propagate along the path ONU → OLT → ONU. In a traditional PON, this idle time is negligible because its RTT is only 0.1ms with 10-km span. LR-PON increases the RTT to 1 ms with 100 km of OLT–ONU distance, which results in 10× the idle time in a traditional PON. In order to combat the detrimental effect of the increased RTT, the work in [29] proposed the Multi-Thread Polling algorithm, in which several polling processes (threads) are running in parallel, and each of the threads is compatible with the proposed DBA algorithms in traditional PON. Figure 10.7 shows an example of multi-thread polling (two threads are shown in the example) and compares it with traditional DBA algorithms (so-called one-thread polling with stop). As shown in Fig. 10.7, the idle time is eliminated because, when ONUs wait for GATE messages from OLT in the current thread which incurs idle time in one-thread polling, they can transmit their upstream packets which are scheduled in another thread simultaneously. We give a high-level description of the multi-thread polling algorithm. For simplicity of illustration, we consider a system of an OLT and 2 ONUs: ONU1 and ONU2 and two threads: thread 1 and thread 2. For more ONUs and threads, the same logic can be applied.
10 Long-Reach Optical Access
231
Sub Cycle OLT
ONU1 t
ONU2 Request
Gate
Data transmission
Fig. 10.7 An example of multi-thread polling [29]
The OLT maintains a polling table, shown in Fig. 10.8. In the table, each ONU has an entry, which records the ONU’s RTT and its most recent requests in each thread (T1 and T2 ). 1. Consider time t0 , when OLT allocates bandwidth in thread T1 , and OLT knows the requested bytes for each ONU in threads T1 and T2 from polling table. At time t0 , the OLT sends a Gate message to ONU1, allowing it to send 5,000 bytes as indicated in the polling table. 2. When ONU1 receives the Gate message from the OLT, it starts transmitting its data up to the size of the granted window, i.e., up to 5,000 bytes. As ONU1 keeps receiving data from end users during the time interval it waits for the acknowledgement from the OLT, ONU1 will generate its own Request, containing this aggregated data size till the Request is generated. This Request is piggybacked to the data transmitted to the OLT. In our example, the new Request is 4,500 bytes, as shown in Fig.10.8(a). 3. OLT knows exactly when the last bit of ONU1’s transmission will arrive even at the time it sends Gate to ONU1, because the OLT knows the RTT and granted transmission window size for ONU1. Then, based on the last bit’s arrival time from ONU1 and the RTT for ONU2, the OLT can schedule a Gate message to ONU2 such that the first bit from ONU2 will arrive with a small guard interval after the last bit from ONU1. In the example, ONU2 is granted to send 2,000 bytes, as shown in Fig.10.8(b). 4. Before the new Request from ONU1 arrives, OLT schedules the Gate message of thread T2 to ONU1, shown by a gray arrow in Fig.10.8(c). Similar to Step 3, the OLT schedules the Gate message to ONU1 such that the arrival of the last bit from ONU2 and the first bit from ONU1 is separated by a guard interval. Here, the OLT allows ONU1 to send its requested 4,800 bytes in thread T2 , which was registered in the polling table by previous Request of ONU1 in T2 . Upon receiving the grant, ONU1 transmits data up to its granted window and piggybacks its new Request to update the polling table in the OLT. Note that, if OLT still schedules the bandwidth to ONU1 in thread T1 , the upstream bandwidth cannot be fully utilized. Even if the OLT sends the Gate message as soon as the new Request of thread 1 from ONU1 arrives, e.g., 4,500 bytes, there is still a wider
232
Huan Song, Byoung-Whi Kim, and Biswanath Mukherjee
(a)
(b)
(c)
(d) Fig. 10.8 Steps of multi-thread polling [29]
10 Long-Reach Optical Access
233
interval than the guard interval between the last bit of the previous ONU2 transmission and the first bit of ONU1 transmission. 5. Then, OLT schedules Gate message to ONU2 in thread T2 , as shown in Fig.10.8(d). Similar calculation as explained in step 3 applies. The new Gate message allows ONU2 to transmit its requested 4,000 bytes in T2 . Upon receiving the Gate message, ONU2 sends 4,000 bytes and piggybacks the new Request. When OLT receives the Request (2,500 bytes), it updates the polling table. The work in [30] also proposed a two-state DBA protocol for LR-PON. Accordingly, the idle time slots will constitute virtual polling cycles, during which the ONUs can transmit data by means of a prediction method to estimate their bandwidth requirement, as shown in Fig. 10.9. Normal cycle OLT Grant
Virtual grant
Virtual cycle
Grant
Report
Virtual grant
Grant Report
ONUs
Fig. 10.9 A two-state DBA protocol for LR-PON [30]
10.5 Summary Long-reach passive optical network (LR-PON) exploits the huge transmission capacity of optical technology and is oriented toward long-range transmission and a large user base. LR-PON is anchored at a central office (CO) so that all higher-layer networking functions can now be located further upstream in this “network cloud”. The OLTs of the traditional PON (which used to sit approximately 10–20 km from the end user) can now be replaced at the local exchange by some elementary hardware, which contains a small amount of compact low-power physical-layer repeater equipment, such as optical amplifiers and optical add–drop multiplexer (OADM). As a result, the telecom network hierarchy can be simplified with the access headend closer to the backbone network. Thus, the network’s Capital Expenditure (CapEx) and Operational Expenditure (OpEx) can be significantly reduced. In this chapter, we discussed the research challenges related to LR-PON, from the physical layer, e.g., power attenuation to the upper layer, e.g., bandwidth assignment. We reviewed some existing LR-PON demonstrations: SuperPON initiated the research on LR-PON and extended the network reach based on the APON model; BT’s demonstrations increased the network capacity up to 10 Gbps per channel and incorporated WDM based on a GPON model; and PIEMAN further exploited the huge optical transmission capacity to support up to 16,384 users. WE-PON offered a
234
Huan Song, Byoung-Whi Kim, and Biswanath Mukherjee
novel ring-and-spur topology to provide a better two-dimensional geographical coverage and protection for PON traffic by using a ring. The impact of increased RTT on higher-layer control was also discussed. Dynamic bandwidth-allocation (DBA) algorithms, e.g., multi-thread polling and two-state DBA, have been proposed to remedy the impact by utilizing the idle time between transmission cycles.
References 1. B. Mukherjee, Optical WDM Networks, Springer, Feb. 2006. 2. O. Gerstel, “Optical networking: a practical perspective (tutorial),” IEEE Hot Interconnects, August, 2000. 3. NSF Workshop Report, “Residential Broadband Revisited: Research Challenges in Residential Networks, Broadband Access, and Applications,” October 2003. http://cairo.cs.uiuc.edu/nsfbroadband/ 4. M. Wegleitner, “Maximizing the Impact of Optical Technology,” Keynote Address, IEEE/OSA Optical Fiber Communication Conference (OFC 2007), Anaheim, CA, March 2007. 5. Gartner Report, “One Gigabit or Bust Initiative,” A Broadband Vision for California,” http://www.cenic.org/GB/gartner/ 6. A. Banerjee, Y. Park, F. Clarke, H. Song, S. Yang, G. Kramer, K. Kim, and B. Mukherjee, “Wavelength-Division Multiplexed Passive Optical Network (WDM-PON) Technologies for Broadband Access: A Review [Invited],” OSA Journal of Optical Networking, vol. 4, no. 11, pp. 737–758, November 2005. 7. R. Lauder, “Technology and Economics for Coarse Wavelength Multiplexing Workshop,” Proc., OFC’04, Los Angeles, CA, 2004. 8. J. George, “Designing Passive Optical Networks for Cost Effective Triple Play Support,” Proc., FTTH Conference, 2004. 9. G. Agrawal, Fiber-Optic Communication Systems, John Wiley and Sons, Inc., New York, 2002. 10. M. H. Kang, “Development of Broadband Convergence Network and Services in Korea,” Proc., IEEE/OSA Optical Fiber Communication Conference (OFC 2008), San Diego, CA, February 2008 (Invited Talk). 11. I. Van de Voorde, C. M. Martin, J. Vandewege, and X. Z. Qiu, “The SuperPON Demonstrator: An Exploration of Possible Evolution Paths for Optical Access Networks,” IEEE Communication Magazine, February 2000. 12. G. Talli, et al., “Integrated Metro and Access Network: PIEMAN,” presented at 12th European Conf. Networks and Opt. Commun., Kista, Sweden, June 2007, pp. 493–500. 13. D. J. G. Mestdagh and C. M. Martin, “The Super-PON concept and its technical challenges,” Broadband Communications, Apr. 1996. 14. C. Martin et al., “Realization of a SuperPON Demonstrator,” Proc., NOC, pp. 118–93, 1997. 15. M. O. van Deventer, et al., “Evolution Phases to an Ultra-Broadband Access Network: Results from ACTS-PLANET,” IEEE Communication Magazine, vol. 35, no. 12, pp. 72–77, December 1997. 16. I. Van de Voorde, et al., “Network topologies for SuperPON,” IEEE/OSA Optical Fiber Commu-nication Conference (OFC 1997), 1997. 17. M. O. van Deventer, et al., “Architecture for 100 km 2048 split bidirectional SuperPONs for ACTS-PLANET,” SPIE, vol. 2919, 1996. 18. D. P. Shea and J. E. Mitchell, “A 10 Gb/s 1024-Way Split 100-km Long Reach Optical Access Network,” Journal of Lightwave Technology, vol. 25, no. 3, March 2007.
10 Long-Reach Optical Access
235
19. R. P. Davey et al., “DWDM Reach Extension of a GPON to 135 km” IEEE/OSA Optical Fiber Communication Conference (OFC 2005), 2005. 20. G. Talli and P. D. Townsend, “Hybrid DWDM-TDM Long-Reach PON for Next-Generation Op-tical Access,” Journal of Lightwave Technology, vol. 24, no. 7, July 2006. 21. “Wdm-E-PON (WE-PON)” working documents, ETRI, 2007. 22. J. A. Lazaro, et al., “Hybrid Dual-fiber-Ring with Single-fiber-Trees Dense Access Network Ar-chitecture using RSOA-ONU,” IEEE/OSA Optical Fiber Communication Conference (OFC 2007), 2007. 23. R. Kjar, et al., “Bi-directional 120 km Long-Reach PON Link Based on Distributed Raman Am-plification,” Proc., Photonics in Switching, 2007. 24. M. Rasztovits-Wiech, et al., “10/2.5 Gbps Demonstration in Extra-Large PON Prototype,” Proc. ECOC, 2007. 25. H. H. Lee, et al., “A hybrid-amplified PON with 75-nm downstream bandwidth, 60 km reach, 1:64 split, and multiple video services,” IEEE/OSA Optical Fiber Communication Conference (OFC 2007), 2007. 26. M.-F. Huang, et al., “A Cost-Effective WDM-PON Configuration Employing Innovative Bi-directional Amplification,” IEEE/OSA Optical Fiber Communication Conference (OFC 2007), 2007. 27. H. Iwamura, et al., “42dB Loss Budget Hybrid DWDM-CDM-PON without Optical Amplifier,” IEEE/OSA Optical Fiber Communication Conference (OFC 2007), 2007. 28. D. P. Shea and J. E. Mitchell, “Experimental Upstream Demonstration of a Long-Reach Wave-length-Converting PON with DWDM Backhaul,” IEEE/OSA Optical Fiber Communication Conference (OFC 2007), 2007. 29. H. Song, A. Banerjee, Byoung-Whi Kim, and B. Mukherjee, “Multi-Thread Polling: A Dynamic Bandwidth Distribution Scheme in Long-Reach PON,” Proc., IEEE Globecom, 2007. 30. C.-H. Chang, N. M. Alvarez et al., “Dynamic Bandwidth assignment for Multi-service access in long-reach GPON,” Proc., ECOC, 2007.
Chapter 11
Optical Access–Metro Networks Martin Maier
Abstract The majority of previous research activities to mitigate the first/last mile bottleneck and metro gap were carried out separately from each other. After paving in FTTX networks all the way or close to the end user with optical fiber, the next evolutionary step seems to be the optical integration of fiber-based access and metro networks. This chapter provides first an up-to-date overview of recently proposed all-optical access–metro networks. We then proceed with a discussion of the social impact of providing end users with advanced broadband access and a growing body of content and applications, e.g., on-line gaming and P2P file sharing, on their everyday lives as well as the implications with respect to the architecture and services of future optical access–metro networks. An all-optically integrating overlay for Ethernet-based access and metro networks is introduced which lets low-cost PON technologies follow low-cost Ethernet technologies from access networks into metro networks. The proposed approach enables an incremental add-on service for future applications such as on-line gaming and P2P file sharing as well as provides a fallback option for legacy ONU equipment and conventional triple-play traffic.
11.1 Introduction Wide area networks (WANs) were one of the first network segments that experienced the widespread deployment of optical technologies in order to provide sufficient capacity in support of heavy longhaul traffic. By means of wavelength division multiplexing (WDM) today’s optical WANs offer large bandwidth pipes where a single fiber may carry tens or even hundreds of wavelength channels, each operating at a bit rate of 10 Gb/s or higher. Current optical WDM backbone networks provide abundant bandwidth by interconnecting WDM links with reconfigurable optical add– drop multiplexers (ROADMs). Martin Maier INRS, University of Qu´ebec, Montr´eal, QC, Canada, e-mail:
[email protected]
A. Shami et al. (eds.), Broadband Access Networks, Optical Networks, c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-92131-0 11,
237
238
Martin Maier
The high-capacity backbone networks are interconnected with high-speed local access networks, e.g., Gigabit Ethernet (GbE), through metropolitan area networks (MANs). Legacy metro networks are typically based on circuit-switched synchronous optical network/synchronous digital hierarchy (SONET/SDH) technology. Legacy SONET/SDH metro networks carry bursty data traffic relatively inefficiently, resulting in a bandwidth bottleneck at the metro level. This bandwidth bottleneck, which is also known as metro gap, prevents high-speed clients and service providers in local access networks from tapping into the vast amounts of bandwidth available in the backbone networks [1]. To address the metro gap, next-generation SONET/SDH (NG-SONET/SDH) networks have been developed in order to support bursty traffic more efficiently. The inefficiencies of legacy SONET/SDH networks were alleviated by introducing the following three data-over-SONET/SDH (DoS) (also known as packet-over-SONET/SDH [PoS]) technologies standardized by the International Telecommunication Union-Telecommunication Standardization Sector (ITU-T): (i) generic framing procedure (GFP) specified in ITU-T Recommendation G.7041, (ii) virtual concatenation specified in ITU-T Recommendation G.707, and (iii) link capacity adjustment scheme (LCAS) specified in ITU-T Recommendation G.7042. While the standardization efforts of NG-SONET/SDH are not specific to metro networks, the importance of the metro gap was reflected by specifying a new standard for packet-switched metro ring networks. The recent standard IEEE 802.17 resilient packet ring (RPR) specifies a ring-based architecture consisting of two counterdirectional optical fiber rings. RPR aims at combining SONET/SDH’s carrier-class functionalities of high availability, reliability, and profitable voice service support with Ethernet’s high bandwidth utilization, low equipment cost, and simplicity [2–4]. Besides RPR, a plethora of packet-switched network architectures and protocols have been studied to alleviate the metro gap. For detailed information about state-of-the-art optical metro networks the interested reader is referred to [5]. At present, there is a strong worldwide push toward bringing fiber closer to individual homes and businesses with the goal to alleviate the last mile or first mile bottleneck, whereby the latter term emphasizes its importance to the end user. Fiber-to-the-home/building (FTTH/B) or close to it (FTTX) networks are poised to become the next major success story for optical communications [6]. Future FTTX access networks not only have to unleash their economic potential and societal benefits by opening up the first/last mile bottleneck between bandwidth-hungry subscribers and high-speed backbone networks but also have to enable the support of a wide range of new and emerging services and applications, such as triple play, video on demand (VoD), videoconferencing, peer-to-peer (P2P) audio/video file sharing, multichannel high-definition television (HDTV), multimedia/multiparty on-line gaming, telemedicine, telecommuting, and surveillance, to get back on the road to prosperity [7]. FTTX networks come in various flavors and offer a number of benefits, e.g., increased network capacity and improved network scalability [8]. Previous activities to mitigate the metro gap and first/last mile bottleneck were mostly carried out separately. That is, the majority of research activities focused either on the metro segment or on the access segment without taking their integration
11 Optical Access–Metro Networks
239
into account. After introducing optical fiber in both metro and access segments and paving all the way to the end user with optical fiber, the next evolutionary step seems to be the optical integration of fiber-based access and metro networks, leading to optical access–metro networks. Very recently, research has begun to study various optically integrated access–metro network architectures and protocols. The remainder of this chapter provides an up-to-date overview of recently investigated all-optical access–metro networks.
11.2 Optical Regional Access Network One of the first research projects on the all-optical integration of access and metro edge ring networks is the so-called next generation Internet optical network for regional access using multiwavelength protocols (NGI ONRAMP) testbed [9, 10]. The mission of the ONRAMP testbed is to implement and demonstrate apart from optical flow switching (OFS) other features such as protection, medium access control (MAC) protocols, control, and management. Figure 11.1 depicts the network architecture of ONRAMP. It consists of a bidirectional feeder WDM ring network which connects multiple access nodes (ANs) with one another and with the backbone network. The bidirectional feeder WDM ring carries eight wavelength
Backbone Network
Distribution Tree
Access Network
AN Distribution Tree
Feeder Ring Network
AN
AN AN
Distribution Ring
AN
Distribution Bus
Distribution Ring
AN = Access Node
c 2000 IEEE.) Fig. 11.1 NGI ONRAMP network architecture. (After [9].
240
Martin Maier
channels in each direction, each wavelength operating at a data rate of up to 10 Gb/s (OC-192). ONRAMP is envisioned to accommodate 10–20 ANs and 20–100 end users on each attached distribution network. Distribution networks carry both the feeder wavelength channels and additional local distribution wavelength channels. Each AN serves as a gateway to distribution networks of variable topologies on which the end users reside. Each AN consists of an IP router and a reconfigurable optical add–drop multiplexer (ROADM), which routes data flows between the feeder ring, the associated IP router, and the attached distribution network. The role of the AN is to route optical wavelength channels and individual IP data packets inside wavelength channels toward their destinations. In ONRAMP, two basic categories of services are offered: an “IP” service and an “optical” service. The IP service involves conventional electronic routing of IP packets, i.e., IP packets undergo OEO conversion and electronic buffering and processing at each intermediate IP router. The optical service represents OFS that provides the end user with an all-optical connection from source to destination in order to allow the user to send large optical flows directly to its destination. In OFS, a large set of data are routed all-optically in order to bypass and thereby offload intermediate IP routers in the data path and their associated queues. Toward this end, a dedicated lightpath is established from the source node to the destination node, allowing the source node to transmit its large data set directly to the destination node. The setup lightpath eliminates the need for packet processing at intermediate routers, e.g., buffering, routing. In ONRAMP, OFS is accomplished with dual-attached workstations having two interfaces: a standard IP-packet interface for low-speed data and control and a high-speed interface to optically flow-switch bursts of extended duration. When a workstation has a large burst to send, it sends control messages over the low-speed interface to negotiate with the ONRAMP network control and management system for an all-optical end-to-end connection. ONRAMP is able to set up not only unidirectional but also bidirectional lightpaths between workstations.
11.3 Stanford University Access Network The design objective of the Stanford University access, or in brief SUCCESS, network architecture is to provide a smooth migration path from currently widely deployed time division multiplexing (TDM) PONs to future WDM PONs and their optical integration by means of an optical single-fiber collector feeder ring, while guaranteeing backward compatibility with existing TDM PON customer premises equipment (CPE) and providing increased capacity to users on new WDM PONs [11]. Figure 11.2 illustrates the migration path of conventional TDM PONs to future WDM-enhanced optical access networks, as proposed in SUCCESS. As shown in Fig. 11.2(a), most of today’s existing TDM PONs use their own fiber cabling to connect subscribers to the central office (CO). Typically, TDM PONs have a physical
11 Optical Access–Metro Networks
241
CO
CO C
= Passive splitter
C
Co−existing CWDM/DWDM optical access networks C
W
W W = DWDM AWG
(a)
(c)
CO
CO C
C
Old ONUs and distributor fibers are preserved C
C Scalable, protected, and efficient optical access networks W
C C
C = CWDM splitter
(b)
W 2xN
(d)
Fig. 11.2 SUCCESS network architecture: migration scenario from existing TDM PONs to WDM c 2004 IEEE.) PONs (After [11].
tree topology using a passive splitter (coupler) at the remote node. The passive splitter equally distributes optical downstream signals going to and combines optical upstream signals coming from end users attached to the branches of the tree. The bandwidth management for each TDM PON is done by a separate optical line terminal (OLT) located inside the CO. The first migration step of existing TDM PONs considered in SUCCESS is depicted in Fig. 11.2(b). In the first step, the passive splitter at the remote node of each TDM PON is replaced with a thin film add/drop coarse wavelength division multiplexing (CWDM) filter which is used to add and drop wavelengths that are shared by end users for upstream and downstream communication in the attached PON. In addition, the feeder fibers of all TDM PONs are replaced with a single-fiber collector ring that strings the remote nodes served by the CO. Subscribers attached to remote nodes on the west side of the ring communicate with the optical line terminal (OLT) on the west side of the CO, while subscribers attached to the east side of the ring communicate with the OLT on the east side of the CO. In the event of a fiber cut on the collector ring, the affected remote nodes detect a loss of signal and subsequently change their transmission direction. In doing so, the SUCCESS network architecture allows for protection and restoration of a single link or node failure on the collector ring. Note that the first migration step does not require any modifications of distribution fibers and optical network units (ONUs) attached to the leaf nodes of each tree. Figure 11.2(c) depicts the second migration step, where additional arrayed-waveguide grating (AWG)-based remote nodes are
242
Martin Maier
inserted on the collector ring. The AWG is used as a dense wavelength division multiplexing (DWDM) (de)multiplexer such that each ONU in the corresponding DWDM PON is assigned a dedicated wavelength. In Fig. 11.2(b) and (c), an OLT deploys tunable transceivers (rather than an array of fixed-tuned transceivers, one for each operating wavelength channel) in order to reduce the number of required transceivers. More specifically, tunable lasers are used to send downstream data as well as continuous wave (CW) bursts for upstream transmission by means of remote modulation at the ONU, resulting in a half-duplex communication between each ONU and OLT. To perform remote modulation of a CW signal, an ONU may deploy a semiconductor optical amplifier (SOA) as a modulator and preamplifier. (Alternatively, full-duplex communication between an ONU and OLT can be achieved by using two different wavelength channels, one for downstream transmission and the other for upstream transmission, and replacing the SOA with a stabilized laser.) Finally, Fig. 11.2(d) illustrates one possible extension of the SUCCESS network. In this example, a new PON is attached to two different remote nodes via two feeder fibers for the sake of redundancy and protection. This example also shows the good scalability of the SUCCESS network architecture, in that new PONs and subscribers can be added to the collector ring in response to growing traffic demands and number of end users.
11.4 Metro Access Ring Integrated Network The so-called metro access ring integrated network (MARIN) optically integrates hybrid TDM/WDM PONs into interconnected metro access ring networks by using optical reconfigurable and parametric wavelength conversion (PWC) devices [12]. Specifically, MARIN consists of multiple interconnected metro access DWDM rings. Each metro access ring has its own CO which schedules the distribution/ collection of traffic to/from attached PONs, coordinates with COs of other metro access rings to forward MAN traffic, and interfaces with the metro core network at the higher level of the network hierarchy. The CO deploys tunable lasers whose configuration is done according to given traffic loads in the metro access rings and attached PONs. Each PON in a given metro access DWDM ring is addressed on a separate set of wavelength channels. Two different types of network nodes are used in the metro access DWDM ring: (i) MARIN gateway and (ii) MARIN switch. A MARIN gateway drops the wavelengths carrying downstream traffic destined for the attached PON subscribers. At a MARIN switch, wavelengths that carry in-transit MAN traffic are all-optically routed using a reconfigurable wavelength-selective switch (WSS) and PWC. PWC allows for the any-to-any conversion of multiple wavelengths (wavelength set) at the same time. In the resultant all-optical MARIN network, PON resources can be shared and leveraged by metro access ring networks. More specifically, a MARIN gateway can dynamically share light sources that were originally used to serve only the attached
11 Optical Access–Metro Networks
243
metro access ring
λM add
λL drop
λL add
Metro
TX
WDM PON
λ1 drop
λ1 drop
RX
Node Control Board (MAC) RX
TX
λL drop
λL add
λM add
MARIN Gateway
Access Control Board (MAC) TT
TT
TT
TT
AWG
to WDM PON
c 2007 IEEE.) Fig. 11.3 MARIN gateway architecture (After [13].
access network, resulting in a more efficient utilization of network resources and improved network performance [13]. Figure 11.3 depicts the architecture of a MARIN gateway which is able to utilize idle or over-provisioned tunable transmitters (TTs) of the attached WDM PON through a passive AWG wavelength router. The MARIN gateway is able to add/drop metro traffic on wavelength channel λL deployed in legacy single-channel ring networks, e.g., SONET/SDH or RPR. In MARIN, the legacy wavelength channel λL may also be used as control channel for wavelength reservation in support of QoS and optical burst switching (OBS) (OBS will be discussed in greater detail in Section 11.5). In addition, the MARIN gateway is able to add metro traffic on any wavelength channel λM of the DWDM ring by using one of the available TTs and drop metro traffic on a dedicated home wavelength
244
Martin Maier
channel λ1 , whereby each MARIN gateway is assigned a different home wavelength channel for reception of metro ring traffic. A frequency-cyclic AWG in conjunction with a number of TTs are used to improve network scalability and flexibility. The frequency-cyclic AWG is used as a wavelength router to dynamically share TTs and wavelengths between the metro access ring and the attached WDM PON.
11.5 OBS Access–Metro Networks In Section 11.4, we briefly mentioned that MARIN is able to support optical burst switching (OBS). OBS is one of the recently proposed optical switching techniques which received a great deal of attention [14–16]. OBS may be viewed as a switching technique that combines the merits of optical circuit switching (OCS) and optical packet switching (OPS) while avoiding their respective shortcomings. The switching granularity at the burst rather than wavelength level allows for statistical multiplexing in OBS, which is not possible in OCS, while requiring a lower control overhead and implementational complexity than OPS. In OBS, a control packet is sent prior to each data burst transmission in order to reserve resources and configure switches at intermediate nodes. In OBS, one-way reservation control packets are sent on one or more control wavelength channels and undergo OEO conversion at each intermediate node, whereas data bursts are transmitted on a separate set of data wavelength channels that are all-optically switched at intermediate nodes. Given that data are switched all-optically at the burst level, OBS combines the transparency of OCS with the statistical multiplexing gain of OPS. A new all-optical access-metro network based on OBS was proposed and investigated in [17]. The optical access network segment consists of a hybrid WDM/TDM PON with reflective ONUs, a frequency-cyclic AWG at the remote node, and a tunable laser and tunable photodetector stack at the OLT, as shown in Fig. 11.4. The data up & control up
PDP
RSOA control down
CMUX
Splitter data up & control up
data up
FSR1 : λ1
PD 1
FSR2 : λ1 + FSR
FSR1
PD
ONU1
1xN AWG data down & control down
LD1
data down
FSR1 & FSR2 λN
data down & control down
Combiner
λN + FSR
LDL
ONUN
OLT
Fig. 11.4 WDM/TDM PON architecture with shared tunable laser/photodetector stack at the OLT c 2007 IEEE.) and reflective ONUs (After [17].
11 Optical Access–Metro Networks
245
OLT deploys L tunable laser diodes (LDs) for transmitting downstream data and control as well as for generating optical CW signals for upstream data transmission by means of remote modulation at the ONU. The LDs are tunable over two free spectral ranges (FSRs) of the AWG placed at the remote node in order to enable full-duplex data transmission, whereby FSR1 is used for upstream data transmission (and downstream control) and FSR2 is used for downstream data transmission. Furthermore, the OLT deploys P tunable photodetectors (PDs) which are isolated from the LDs via a circulator and are used for reception of upstream data and control. The AWG at the remote node is used as a 1 × N wavelength (de)multiplexer which routes a different dedicated wavelength to each of the N attached ONUs. Each ONU deploys a coarse multiplexer (CMUX) to separate FSR1 from FSR2 . For FSR1 , each ONU is equipped with a reflective semiconductor optical amplifier (RSOA). The RSOA operates in either one of the following two modes: (i) reception of downstream control and (ii) transmission of upstream data by means of remote modulation of the optical CW signal. The two modes are switched on a time basis. For FSR2 , each ONU deploys a PD for reception of downstream data. Note that the use of an RSOA not only avoids the need for a costly light source at each ONU but also allows all ONUs to be identical and operate on any wavelength, resulting in major cost savings. Wavelength-independent ONUs able to operate on any wavelength are also known as colorless ONUs. The OLT executes a centralized dynamic bandwidth allocation (DBA) algorithm that combines OBS and polling for upstream and downstream transmission of data bursts. The OLT and every ONU use different priority queues for service differentiation. Arriving packets are aggregated into bursts until a maximum aggregation time or maximum burst size is reached and are put in the corresponding priority queues. The DBA algorithm applies strict priority scheduling to dynamically assign available bandwidth and transceiver resources to the different traffic classes. Figure 11.5 illustrates how multiple hybrid WDM/TDM PONs can be transparently interconnected by using an optical burst switching multiplexer (OBS-M) which in turn all-optically interfaces with a distant metro router. The OBS-M maintains the architecture of the OLT of Fig. 11.4 except for an additional optical crossconnect (OXC) that interfaces with Q attached hybrid WDM/TDM PONs and P fixed wavelength converters (FWCs) that replace the P PDs of Fig. 11.4. The FWCs are used to convert the N upstream wavelength channels of FSR1 com ing from the AWG to P fixed wavelength channels λ1 , . . . , λ P , which are multiplexed onto a common outgoing fiber link and finally demultiplexed and received by a stack of P PDs at the OLT. The distant router is equipped with an additional stack of D fixed LDs which operate on the downstream data wavelength channels λ1 , . . . , λ D , which are converted by D FWCs (fixed input wavelength/variable output wavelength) at the OBS-M to the N wavelength channels according to the location of the destination ONU. Furthermore, the OBS-M deploys a stack of P tunable LDs to send optical CW signals to the attached ONUs for remote modulation of upstream data. For signaling, the distant router has an additional fixed PDc and fixed LDc , operating on the upstream control wavelength channel λuc and
246
PD1
Martin Maier λ’1
λí1
variable FWC1
1
1xN AWG 1
Q
1xN AWG Q
ONU1
FSR1 PDP
PDC
λ’P
λ’P
λ’uc LDC
λ’uc
LDD
LDC
Up Proc.
PD c
OXC
LD 1 MUX/ DEMUX
LD1
FWCP
MUX/ DEMUX
FSR1
λ1
LD P λ1
FWC1
λD
FSR2 λD
λ dc
Distant router
λ dc PDc
FWCD
ONUQxN
Down Proc.
OBS−M
Fig. 11.5 All-optical interface between OBS multiplexer (OBS-M) with multiple attached c 2007 IEEE.) WDM/TDM PONs and distant metro router (After [17].
downstream control wavelength channel λdc , respectively. The two control wavelength channels are used for arbitration of upstream and downstream data burst transmissions. In the proposed architecture shown in Fig. 11.5, upstream and downstream data burst transmissions are controlled separately from each other. In the upstream direction, the OBS-M uses an upstream processor (denoted as “Up Proc.” in Fig. 11.5) for scheduling upstream burst data transmissions, whereby a fixed photodetector PDc receives the requests sent by the attached ONUs. The upstream scheduler processes the received requests and sets the OXC and P FWCs for upstream data burst transmission according to the DBA algorithm in use. The upstream processor informs the distant router on upstream control wavelength channel λuc about the scheduled upstream data burst transmission. In the downstream direction, the distant router sends control packets on the downstream control wavelength channel λdc to inform the downstream processor at the OBS-M (denoted as “Down Proc.” in Fig. 11.5). The control packets contain resource reservation information to enable the downstream scheduler to set the OXC and D FWCs such that the subsequent downstream data bursts can be routed to the corresponding destination ONUs. According to [17], the all-optical OBS-based access–metro network of Fig. 11.5 is strictly nonblocking in the wavelength, time, and space domains. Moreover, it was shown in [17] that the basic architecture of a single OBS-M and distant router can be extended to all-optically interconnect multiple OBS-Ms with a distant router through a ROADM-based metro network with either a tree or a ring topology. At the downside, however, the OLT architecture of Fig. 11.4 and OBS-M architecture of Fig. 11.5 with the required tunable laser stacks, tunable photodetector stacks,
11 Optical Access–Metro Networks
247
wavelength converter banks, and OXC significantly add to the cost, power consumption, footprint, and complexity of all-optical access–metro networks.
11.6 STARGATE An alternative approach to all-optically integrate access and metro networks is the so-called STARGATE network architecture [18]. STARGATE lets low-cost PON technologies follow low-cost Ethernet technologies from access networks into metro networks, resulting in significantly reduced costs and complexity. STARGATE transparently integrates optical Ethernet PONs (EPONs) and RPR metro networks. RPR can easily bridge to Ethernet networks such as EPON and may also span into MANs and WANs, making it thus possible to perform layer 2 switching from access networks far into backbone networks [2]. STARGATE makes use of an overlay island of transparency with optical bypassing capability which is key in MANs in order to support not only various legacy but also future services in an easy and costeffective manner [19]. The rationale behind STARGATE is based on the following three principles: • Evolutionary downstream SDM upgrades: Eventually, when per-user bandwidth needs grow, incrementally upgrading existing EPON tree networks with additional fibers may prove attractive or become even mandatory. In fact, some providers are already finding this option attractive since long runs of multifiber cable are almost as economical in both material and installation costs as the same lengths of cables with one or a few fibers [7]. Interestingly, the standard IEEE 802.3ah supports not only the commonly used point-to-multipoint (P2MP) topology but also a hybrid EPON topology consisting of point-to-point (P2P) links in conjunction with P2MP links.1 STARGATE explores the merits of deploying an additional P2P or P2MP fiber link in EPON tree networks to connect the OLT with a subset of one or more ONUs in an evolutionary fashion according to given traffic demands and/or cost constraints. It is important to note that STARGATE requires an additional P2P or P2MP fiber link only in the downstream direction from OLT to ONU(s) and none in the upstream direction. Thus, STARGATE makes use of evolutionary downstream space division multiplexing (SDM) upgrades of WDM/TDM EPONs. • Optical bypassing: The problem with using SDM in EPONs is the increased electro-optic port count at the OLT. To avoid this, STARGATE makes use of optical bypassing. Specifically, all wavelengths on the aforementioned additional P2P or P2MP downstream fiber link coming from the metro edge ring are not terminated at the OLT, thus avoiding the need for OEO conversion and additional transceivers at the OLT, as explained in greater detail shortly. Note that OEO conversion usually represents the major part of today’s optical networking 1 Note that in the IEEE standard 802.3ah, the acronym P2P somewhat misleadingly stands for “point-to-point” rather than “peer-to-peer.”
248
Martin Maier
infrastructure. Due to the small-to-moderate distances of STARGATE’s access– metro networks, optical bypassing and the resultant transparency can be easily implemented, thereby avoiding OEO conversion and resulting in major cost savings [20]. • Passive optical networking: Finally, the last design principle of STARGATE is based on the idea of letting low-cost passive optical networking technologies follow low-cost Ethernet technologies from access networks into metro networks. In doing so, not only PONs but also MANs benefit from passivity, which is a powerful tool to build low-cost high-performance optical networks [21]. As we will see shortly, STARGATE makes use of an athermal (i.e., temperature-insensitive) AWG wavelength router, which eliminates the need for temperature control and monitoring the wavelength shift of the AWG and thus leads to a simplified network management and to reduced costs [22]. Note that passive optical networking in all-optical wavelength-routing WDM networks has recently begun to gain momentum within the so-called time-domain wavelength interleaved networking (TWIN) concept that enables to build costeffective and flexible optical networks using readily available components [23]. In TWIN, fast TDM switching and packet switching in the passive optical wavelengthselective WDM network core are emulated through the use of emerging fast tunable lasers at the optical network edge, thus avoiding the need for fast optical switching and optical buffering. The original TWIN did not scale well since the number of nodes N was limited by the number of available wavelengths W , i.e., N = W . Recently, the so-called TWIN with wavelength reuse (TWIN-WR) was proposed, where the number of nodes is independent of the number of wavelengths, i.e., N > W [24]. Both TWIN and TWIN-WR require the network-wide scheduling of transmissions in order to avoid channel collisions. Unlike TWIN, a source node in TWIN-WR may not be able to send traffic directly to any destination node in an optical single hop, resulting in multihopping via intermediate electrical gateways. As we will see shortly, STARGATE differs from TWIN and TWIN-WR in a number of ways. First, STARGATE supports extensive wavelength reuse while providing optical single-hop communication among all ONUs. Second, STARGATE requires only local scheduling of transmissions within each separate EPON in order to completely avoid channel collisions throughout the network, thus avoiding the need for network-wide scheduling. Third, STARGATE does not require any time-of-day synchronization. And finally, while TWIN and TWIN-WR are designed to support WANs of arbitrary topology, STARGATE targets access and metro networks, whose regular topologies (tree, ring, star) help simplify scheduling significantly.
11.6.1 Architecture The network architecture of STARGATE is shown in Fig. 11.6. STARGATE consists of an RPR metro edge ring that interconnects multiple WDM EPON tree networks
11 Optical Access–Metro Networks
249
Internet
ONU
Router
Ring network
ONU
Coupler
ΛOLT ΛAWG
OLT CO
ONU
Star subnetwork
ΛPSC
Tree network
CO
Servers
ΛAWG P x P AWG P x P PSC
ONU
ONU
ONU
P2P link
CO OLT
Coupler
CO OLT
ONU
P2MP link Coupler
ONU
Tree network
ONU
= Ring node
Fig. 11.6 STARGATE network architecture comprising P = 4 COs and Nr = 12 RPR ring nodes c 2007 IEEE.) (After [18].
among each other as well as to the Internet and server farms. The RPR network consists of P COs and Nr RPR ring nodes and its round-trip time equals τring . The CO in the upper right corner of the figure is assumed to be attached to the Internet and a number of servers via a common router. We refer to this CO as the hot-spot CO. The P COs are interconnected via a single-hop WDM star subnetwork whose hub is based on a wavelength-broadcasting P × P passive star coupler (PSC) in parallel with an athermal wavelength-routing P × P AWG. The end-to-end propagation delay of the star subnetwork equals τstar . Each CO is attached to a separate input/output port of the AWG and PSC by means of two pairs of counterdirectional fiber links. Each fiber going to and coming from the AWG carries AWG = P · R wavelength channels, where R denotes the number of used FSRs of the AWG. Each fiber going to and coming from the PSC carries PSC = 1 + H + (P − 1) wavelength channels, consisting of one control channel λc , 1 ≤ H ≤ P − 1 dedicated home channels for the hot-spot CO, and (P − 1) dedicated home channels, one for each of the remaining (P − 1) COs. The home channels are fixed assigned to the COs. Data destined for a certain CO are sent on its corresponding home channel. All COs, except the hot-spot CO, are co-located with a separate OLT of an attached WDM EPON. Let OLT denote the number of used wavelengths in each WDM EPON in both directions between the ONUs and the corresponding OLT, i.e., there is one set of OLT upstream wavelength channels and another set of OLT downstream wavelength channels in each WDM EPON. Furthermore, each WDM EPON deploys an additional P2P or P2MP downstream fiber link from the CO to a single ONU or multiple ONUs, respectively. Each downstream fiber link carries the AWG wavelength channels coming from the AWG of the star subnetwork. Figure 11.7 depicts the interconnection of a given WDM EPON and the star subnetwork in greater detail, illustrating the optical bypassing of the co-located OLT and CO. Note that in the figure OLT comprises both upstream and downstream
250
Martin Maier Optical amplifiers
ΛAWG
P2P/P2MP link
to/from AWG
ΛAWG
ΛAWG tree network
ΛOLT + ΛAWG Coupler
ΛOLT WDM coupler
CO OLT
ΛPSC
to/from PSC
c 2007 IEEE.) Fig. 11.7 Optical bypassing of co-located OLT and CO (After [18].
wavelength channels which run in opposite directions on the tree network to and from the OLT, respectively. In contrast, the AWG wavelength channels are carried on the tree network only in the upstream direction, while in the downstream direction they are carried on the separate P2P or P2MP downstream fiber link. As shown in Fig. 11.7, a WDM coupler is used on the tree network in front of the OLT to separate the AWG wavelength channels from the OLT wavelength channels and to guide them directly onward to the AWG of the star subnetwork, optically amplified if necessary. In doing so, the AWG wavelength channels are able to optically bypass the CO and OLT. Similarly, the AWG wavelength channels coming from the AWG optically bypass both CO and OLT and directly travel on the P2P or P2MP link onward to the subset of attached ONU(s). As a result, the ONU(s) as well as the hot-spot CO that send and receive on any of the AWG wavelength channels are able to communicate all-optically with each other in a single hop across the AWG of the star subnetwork. In other words, the star forms a gate for all-optically interconnecting multiple WDM EPONs. Accordingly, we term the network STARGATE. Similar to an RPR node, each CO is equipped with two fixed-tuned transceivers, one for each direction of the dual-fiber ring. In addition, each CO has one transceiver fixed-tuned to the control channel λc of the star subnetwork. For data reception on its PSC home channel, each CO (except the hot-spot CO) has a single fixed-tuned receiver. The hot-spot CO is equipped with 1 ≤ H ≤ P − 1 fixed-tuned receivers. For data transmission on the PSC, each CO (except the hot-spot CO) deploys a single transmitter which can be tuned over the (P − 1) + H home channels of the COs. The hot-spot CO deploys H tunable transmitters whose tuning range covers the home channels of the remaining (P − 1) COs as well as the AWG wavelengths. Unlike the remaining COs, the hot-spot CO is equipped with an additional multiwavelength receiver operating on AWG . In each WDM EPON, the OLT is equipped with an array of fixed-tuned transmitters and fixed-tuned receivers, operating at the OLT downstream and OLT upstream wavelength channels, respectively. STARGATE does not impose any particular WDM node structure on the ONUs except for ONUs which receive data over the AWG. Those ONUs must be equipped with a multiwavelength receiver operating on the AWG wavelengths in order to avoid receiver collisions, as explained above. Otherwise, STARGATE allows ONUs to be upgraded in an evolutionary manner. That is, ONUs can take on whatever architecture is preferred at the time when they are upgraded, possibly using transceivers
11 Optical Access–Metro Networks
251
with different tuning time and tuning range or RSOAs, similar to those shown in Fig. 11.4. In doing so, STARGATE allows these decisions to be dictated by economics, state of the art of transceiver manufacturing technology, service provider preferences, given traffic demands, and/or cost constraints. The evolutionary upgrade allows for cautious pay-as-you-grow WDM upgrades of ONUs and thus helps operators realize their survival strategy for highly cost-sensitive access networks.
11.6.2 Discovery and Registration STARGATE uses the WDM extensions to EPON’s multipoint control protocol (MPCP) REPORT and GATE messages recommended in [25] which enable each OLT to schedule transmissions to and receptions from its attached WDM-enhanced ONUs on any wavelength channels supported by both the OLT and the respective ONU. In each WDM EPON, the REGISTER REQ MPCP message with WDM extensions described in [25] is deployed for the discovery and registration of ONUs. The REGISTER REQ message is sent from each ONU to its OLT and carries the MAC address as well as detailed information about the WDM node structure of the ONU. In doing so, the OLT of each WDM EPON learns about the MAC address and WDM node structure of each of its attached ONUs. After registration, the OLTs exchange via the PSC (to be described shortly) the MAC addresses of their attached ONUs that are able to receive data over the AWG. As a result, the OLTs know not only which MAC addresses can be reached via the AWG but also to which AWG output ports the corresponding ONUs are attached and thus on which of the AWG wavelengths they can be reached from a given AWG input port.
11.6.3 Dynamic Bandwidth Allocation The IEEE 802.3ah REPORT MPCP message can carry one or more queue sets, each set comprising up to eight queues, as shown in Fig. 11.8. In STARGATE, ONUs use the first queue set to report bandwidth requirements on the OLT upstream wavelengths for sending data to the OLT, e.g., conventional triple-play traffic (data, voice, video). To report bandwidth requirements on any of the AWG wavelength channels to ONUs located in different EPONs, a given ONU uses one or more additional queue sets and writes the MAC addresses of the destination ONUs in the reserved field of the REPORT MPCP message and sends it to the OLT. Thus, the bandwidth requirements on AWG ride piggyback on those on OLT within the same REPORT message. In Section 11.6.4, we will discuss an illustrative example of how the piggyback REPORT MPCP message can be used for on-line gaming and peer-to-peer (P2P) file sharing. The WDM-extended GATE message in [25] is used to coordinate the upstream transmission on the OLT wavelengths within each WDM EPON and also to
252
Martin Maier
Repeated n times as indicated by Number of queue sets
Destination Address
Octets 6
Source Address
6
Length/Type = 88−08
2
Opcode = 00−03
2
Timestamp
4
Number of queue sets
1
Report bitmap
1
Queue #0 Report
0/2
Queue #1 Report
0/2
Queue #2 Report
0/2
Queue #3 Report
0/2
Queue #4 Report
0/2
Queue #5 Report
0/2
Queue #6 Report
0/2
Queue #7 Report
0/2
Pad/Reserved
0−39
FCS
4
LSB
OCTETS WITHIN FRAME TRANSMITTED TOP−TO−BOTTOM
MSB b0
b7 BITS WITHIN FRAME TRANSMITTED LEFT−TO−RIGHT
Fig. 11.8 REPORT MPCP message
coordinate the all-optical transmission on any of the AWG wavelengths across the star subnetwork between two ONUs residing in different WDM EPONs, provided both ONUs support the AWG wavelengths. Based on the MAC addresses of the destination ONUs carried piggyback in the REPORT message, the OLT of the source WDM EPON uses the extended GATE MPCP message, which we term STARGATE message, to grant the source ONUs a time window on the wavelengths which the AWG routes to the destinations according to the DBA algorithm in use at the OLT. Similar to EPON, the STARGATE network is not restricted to any specific DBA algorithm. However, DBA algorithms for STARGATE should be able to dynamically set up transparent all-optical circuits across the AWG at the wavelength and subwavelength granularity with predictable QoS in terms of bounded delay and guaranteed bandwidth between ONUs of different WDM EPONs. Each OLT uses its DBA module to provide gated service across the AWG-based star network, a service which we correspondingly term STARGATING. This gated service enables the dynamic setup of low-latency circuits on any of the AWG wavelengths in support of applications such as P2P file sharing and periodic game traffic. ONUs unable to send and receive data across the AWG as well as RPR ring nodes send their data on the tree, ring, and/or PSC along the shortest path in terms
11 Optical Access–Metro Networks
253
of hops. Channel access on the dual-fiber ring is governed by RPR protocols, which are described in greater detail in [2]. On the PSC, time is divided into periodically recurring frames. On the control channel λc , each frame consists of P control slots, each dedicated to a different CO. Each CO stores data packets to be forwarded on the PSC in a single FIFO queue with look-ahead capability to avoid head-of-line (HOL) blocking. For each stored data packet the CO broadcasts a control packet in its assigned control slot to all COs. A control packet consists of two fields: (i) destination address and (ii) length of the corresponding data packet. All COs receive the control packet and build a common distributed transmission schedule for the collision-free transmission of the corresponding data packet on the home channel of the destination CO at the earliest possible time. The destination CO forwards the received data packet toward the final destination node.
11.6.4 Applications It is important to note that providing end users with advanced broadband access and a growing body of content and applications has a significant impact on their everyday lives. According to the Pew Internet & American Life Project, a project that examines the social impact of the Internet, subscribers of advanced access networks increasingly use the Internet as a “destination resort,” a place to go just to have fun or spend their free time [26]. Let us focus on two applications which become increasingly popular in optical access networks among subscribers spending their free time: on-line gaming and P2P file sharing, whereby the latter one is already the predominant traffic type in today’s operational access networks [27]. In the following, we deploy a top-down approach. We first outline the traffic characteristics of both applications and discuss their impact on the architecture and services of optical access networks. We then discuss how STARGATE is well suited to support both applications in an efficient, cost-effective, and future-proof manner.
11.6.4.1 On-Line Gaming Most of the popular on-line games are based on the client–server paradigm, where the server keeps track of the global state of the game. On-line gaming traffic primarily consists of information sent periodically back and forth between all clients (players) and server. On-line gaming traffic is very different than web traffic [28]. Specifically, on-line gaming requires low-latency point-to-point upstream communication from each client to the server as well as low-latency-directed broadcast downstream communication from the server to all clients. To facilitate the synchronous game logic, packets are small since the application requires extremely low latencies which makes message aggregation and message retransmission impractical. The workload of on-line games consists of large, highly periodic bursts of very small packets with predictable long-term rates, exhibiting packet bursts every 50 or
254
Martin Maier
100 ms where the payload of almost all upstream packets is smaller than 60 bytes and that of downstream packets is spread between 0 and 300 bytes. Upstream packets have an extremely narrow length distribution centered around the mean size of 40 bytes, while downstream packets have a much wider length distribution around a significantly larger mean. The unique characteristics and requirements of on-line games pose significant challenges on current network infrastructures [28]. Game vendors increasingly deploy large server farms in a single location to control the experience of players. The server farms must be provided with a means to efficiently realize directed broadcasting. More importantly, on-line gaming introduces a significant downward shift in packet size which makes the electronic processing the bottleneck vs. the link speed. Networking devices that are designed for handling larger packets will suffer from packet loss or persistent packet delay and jitter. One solution to alleviate this bottleneck is to optically bypass electronic access network devices. The high predictability of on-line gaming traffic can be exploited for efficient access network resource management by setting up subwavelength circuits. In on-line games, low latency and good scalability are the two most important network design aspects [29]. If not handled properly, network latency leads to poor game quality, sluggish responsiveness, and inconsistency. Besides latency, scalability is another key aspect of networked on-line games. The currently predominant client–server model of commercial on-line games faces scalability issues since the central server may suffer from CPU resource and bandwidth bottlenecks. Recently, the peer-to-peer (P2P) design of scalable game architectures has begun to increasingly receive attention, where clients’ computing resources are utilized to improve the latency and scalability of networked on-line games [29].
11.6.4.2 Peer-to-Peer File Sharing The use of P2P applications for sharing large audio/video files as well as software has been growing dramatically. P2P traffic represents now the by far largest amount of data traffic in today’s operational access networks, clearly surpassing web traffic [27]. In P2P file-sharing applications, the process of obtaining a file can be divided into two phases: (i) signaling and (ii) data transfer [30]. In the signaling phase, by using a specific (proprietary) P2P protocol a host identifies one or more target hosts from which to download the file. In the data transfer phase, the requesting host downloads the file from a selected target host. According to [27], the major P2P application, which represents more than 50% of the entire access network upstream data traffic, exhibits a nearly constant traffic pattern over time, independent of the number of connected subscribers. Moreover, the vast majority of upstream traffic is generated by a small number of hosts, for both weekday and weekend. Specifically, the top 1–2% of IP addresses account for more than 50% and the top 10% of IP addresses account for more than 90% of upstream traffic. Interestingly, note that this behavior, where a few hot-spot servers with popular content originate most of the P2P upstream traffic, resembles that of
11 Optical Access–Metro Networks
255
networks which are based on the conventional client–server paradigm. Similar observations were made for P2P downstream traffic, where a few heavy hitters are responsible for a high percentage of downstream traffic. Also, heavy hitters tend to have long on-times. Finally, it is worthwhile mentioning that the most popular P2P system in terms of both number of hosts and traffic volume is able to resolve most queries locally by finding nearby peers [30]. The high volume and good stability properties of P2P traffic give rise to the use of simple yet highly effective capacity planning and traffic engineering techniques as a promising solution to manage P2P file sharing. Additionally, the fact that individual hot-spot servers and heavy hitters with long on-times generate huge traffic volumes and most of the queries can be resolved locally can be exploited at the architecture and protocol levels of future P2P-friendly optical access networks that must be designed to meet the requirements of P2P applications in a resource-efficient and cost-effective manner.
11.6.4.3 Results STARGATE is well suited to meet the aforementioned requirements of on-line gaming and P2P file sharing. The all-optical subwavelength circuits may be used to carry periodic low-latency game traffic and high volumes of stable P2P traffic. Directed broadcasting can be realized by letting the hot-spot CO transmit packets on different wavelengths of AWG , whereby each packet sent on a given wavelength is locally broadcast to all ONUs attached to the coupler of the corresponding WDM EPON network. Individual ONUs which send or receive large amounts of traffic may deploy additional transceivers. STARGATE also scales well, in that additional EPON tree networks may be attached to the Nr RPR ring nodes, which in turn might be later connected to the WDM star subnetwork, and additional FSRs of the AWG may be used to increase the number AWG of used wavelengths on the AWG, as needed. Furthermore, STARGATE provides a high degree of connectivity which not only improves the network resilience and bandwidth efficiency but also decreases the number of required hops between nearby file-sharing peers. Note that F-TCP, a light-weight TCP with asynchronous loss recovery mechanism for the transfer of large files in high-speed networks, can be used at the transport layer of STARGATE to achieve reliable P2P file sharing [31]. For illustration, we set the network parameters to the following default values: Nr = 12, P = 4, H = P − 1 = 3, OLT = 4 (the same for all three WDM EPONs), AWG = 4, and R = 1. Furthermore, each tree network is 20 km long and accommodates 16 ONUs, the ring network circumference is set to 100 km, and the star subnetwork diameter is set to 100 km/π , translating into τstar = 0.16 ms, τring = 0.5 ms, and τtree = 0.1 ms (propagation delay between ONUs and OLT). All links operate at a line rate of 1 Gb/s. Let us first consider only P2P file sharing and investigate the impact of various ONU structures. We consider three different ONU types: (1) ONUs of type 1 use a single transceiver fixed-tuned to the original TDM EPON wavelength channel within OLT , (2) ONUs of type 2 use
256
Martin Maier
a single transceiver tunable over OLT , (3) ONUs of type 3 use a transmitter tunable over OLT and AWG , a receiver tunable over OLT , and a multiwavelength receiver on AWG . Note that only ONU type 3 allows to optically bypass the OLT and CO. In each EPON, we let one ONU be a hot-spot server, one ONU be a heavy hitter, one ONU be both hot-spot server and heavy hitter simultaneously, and the remaining 13 ONUs be regular. The hot-spot CO is also both hot-spot server and heavy hitter. We assume that the seven hot-spot servers generate 90% and the remaining 42 ONUs generate 10% of the total traffic. Each hot-spot server randomly sends a locally generated P2P file to any heavy hitter with probability 0.9 and to any of the remaining 42 ONUs with probability 0.1. Similarly, the remaining 10% of the generated total traffic is equally sent to any heavy hitter with probability 0.9 and to any other ONU with probability 0.1. Assuming that any (proprietary) P2P protocol has been used for signaling, we focus on the data transfer phase. We assume Poisson P2P file (batch) traffic, where each generated P2P file (batch) consists of B 1500-bytes Ethernet frames2 and B is uniformly randomly distributed over [1,100]. Figure 11.9 depicts the mean delay vs. mean aggregate throughput performance of STARGATE with 95% confidence intervals for the three different ONU types under P2P traffic, where a P2P file is scheduled by the OLT on the earliest available wavelength of AWG and OLT via MPCP REPORT and GATE messages. We observe that using only ONUs of type 1 performs worst and replacing them with type 2 ONUs improves the throughput-delay performance only slightly. In contrast, equipping hot-spot servers and heavy hitters with ONUs of type 3 while all remaining ONUs are of type 2 improves the throughput-delay performance dramatically by optically bypassing the co-located OLTs and COs. Importantly, we observe that only little further performance gain can be achieved if all ONUs are 4
Mean Delay [ms]
3
2
1
0
ONU type 1 only ONU type 2 only ONU types 2 & 3 ONU type 3 only
0
1
2
3
4
5
6
7
8
Mean Aggregate Throughput [Gb/s]
Fig. 11.9 Mean delay vs. mean aggregate throughput performance of STARGATE for three different ONU types under P2P traffic only 2
In most current P2P systems, triggered data transfers use full-sized 1500-bytes packets.
11 Optical Access–Metro Networks
257
of type 3. Thus, for P2P traffic it is sufficient to upgrade only hot-spot servers and heavy hitters with optical bypass capable type 3 ONUs and deploy ONUs of type 2 otherwise. Next, we consider also game traffic and show the additional benefit gained from setting up subwavelength circuits on the optically bypassing AWG wavelengths. We assume that all ONUs are of type 3 and P2P traffic is generated like above in Fig. 11.9. In addition, the 13 regular ONUs in each EPON generate game traffic. Specifically, every 50 ms each regular ONU generates a burst of ten 40-bytes packets destined for the hot-spot CO. For each EPON, the hot-spot CO in turn generates a burst of 13 packets every 50 ms, whereby the length of each packet is uniformly distributed over 0, 300 bytes. Figure 11.10 depicts the mean delay vs. mean aggregate throughput of P2P and game traffic separately, where game traffic is sent either with or without using dedicated subwavelength circuits on AWG to carry periodic upstream and downstream game traffic bursts. We observe that P2P traffic remains mostly unaffected by adding game traffic, whereas game traffic without using circuits experiences a slightly increased delay and jitter, indicated by larger confidence intervals, for increasing P2P traffic. By setting up subwavelength circuits on AWG , the mean delay and jitter of game traffic can be reduced significantly. 4
P2P traffic only P2P traffic & game traffic w/o circuits P2P traffic & game traffic w/ circuits
Mean Delay [ms]
3
2
P2P traffic 1 game traffic 0
0
1
2 3 4 5 6 Mean Aggregate Throughput [Gb/s]
7
8
Fig. 11.10 Mean delay vs. mean aggregate throughput performance of STARGATE with and without subwavelength circuits on AWG under both P2P and game traffic
11.7 Summary After providing an overview of recently proposed optical access–metro networks, we have described in greater detail the P2P-friendly STARGATE architecture which all-optically interconnects WDM EPONs. STARGATE is compliant with IEEE 802.3ah MPCP and supports arbitrary ONU WDM structures and DBA algorithms. STARGATE is built of low-cost passive optical components and meets the large
258
Martin Maier
bandwidth, low latency, and scalability requirements of P2P file-sharing and on-line gaming applications by capitalizing on optical bypassing and subwavelength circuit switching. STARGATE requires only a small number of ONUs to be WDM upgraded to efficiently carry unbalanced P2P traffic and significantly improves the network performance under game traffic. It is important to note that the all-optically integrating overlay enables an incremental add-on service for future applications such as P2P file sharing and on-line gaming, while conventional applications, e.g., triple play, may continue to operate in each EPON tree network on OLT . Moreover, STARGATE provides a fallback option that allows ONUs to send and receive data on OLT undergoing OEO conversions at the COs if source ONU and/or destination ONU do not support the optically bypassing wavelengths AWG . Acknowledgments The author is grateful to Martin Herzog for his work on the simulation results.
References 1. N. Ghani, J.-Y. Pan, and X. Cheng, “Metropolitan Optical Networks,” Optical Fiber Telecommunications, vol. IVB, Chapter 8, pp. 329–402, 2002. 2. F. Davik, M. Yilmaz, S. Gjessing, and N. Uzun, “IEEE 802.17 Resilient Packet Ring Tutorial,” IEEE Communications Magazine, vol. 42, no. 3, pp. 112–118, Mar. 2004. 3. P. Yuan, V. Gambiroza, and E. Knightly, “The IEEE 802.17 Media Access Protocol for HighSpeed Metropolitan-Area Resilient Packet Rings,” IEEE Network, vol. 18, no. 3, pp. 8–15, May/June 2004. 4. S. Spadaro, J. Sole-Pareta, D. Gareglio, K. Wajda, and A. Szymanski, “Positioning of the RPR Standard in Contemporary Operator Environments,” IEEE Network, vol. 18, no. 2, pp. 35–40, Mar./Apr. 2004. 5. M. Herzog, M. Maier, and M. Reisslein, “Metropolitan Area Packet-Switched WDM Networks: A Survey on Ring Systems,” IEEE Communications Surveys and Tutorials, vol. 6, no. 2, pp. 2–20, Second Quarter 2004. 6. R. Ramaswami, “Optical Networking Technologies: What Worked and What Didn’t,” IEEE Communications Magazine, vol. 44, no. 9, pp. 132–139, Sept. 2006. 7. P. E. Green, “Fiber To The Home–The New Empowerment,” Wiley, New York, 2006. 8. T. Koonen, “Fiber to the Home/Fiber to the Premises: What, Where, and When,” Proceedings of the IEEE, vol. 94, no. 5, pp. 911–934, May 2006. 9. M. Kuznetsov, N. M. Froberg, S. R. Henion, H. G. Rao, J. Korn, K. A. Rauschenbach, E. H. Modiano, and V. W. S. Chan, “A Next-Generation Optical Regional Access Network,” IEEE Communications Magazine, vol. 38, no.1, pp. 66–72, Jan. 2000. 10. N. M. Froberg, S. R. Henion, H. G. Rao, B. K. Hazzard, S. Parikh, B. R. Romkey, and M. Kuznetsov, “The NGI ONRAMP Test Bed: Reconfigurable WDM Technology for Next Generation Regional Access Networks,” IEEE/OSA Journal of Lightwave Technology, vol. 18, no. 12, pp. 1697–1708, Dec. 2000. 11. F.-T. An, K. S. Kim, D. Gutierrez, S. Yam, S.-T. Hu, K. Shrikhande, and L. G. Kazovsky, “SUCCESS: A Next-Generation Hybrid WDM/TDM Optical Access Network Architecture,” IEEE/OSA Journal of Lightwave Technology, vol. 22, no. 11, pp. 2557–2569, Nov. 2004. 12. W.-T. Shaw, G. Kalogerakis, S.-W. Wong, Y.-L. Hsueh, N. Cheng, S.-H. Yen, M. E. Marhic, and L. G. Kazovsky, “MARIN: Metro-Access Ring Integrated Network,” in Proc., IEEE GLOBECOM, San Francisco, CA, Nov./Dec. 2006.
11 Optical Access–Metro Networks
259
13. S.-W. Wong, W.-T. Shaw, K. Balasubramaniam, N. Cheng, and L. G. Kazovsky, “MARIN: Demonstration of a Flexible and Dynamic Metro-Access Integrated Architecture,” in Proc., IEEE GLOBECOM, Washington, DC, Nov. 2007. 14. T. Battestilli and H. Perros, “An Introduction to Optical Burst Switching,” IEEE Communications Magazine, vol. 41, no. 8, pp. S10–S15, Aug. 2003. 15. T. Battestilli and H. Perros, “Optical Burst Switching for the Next Generation Internet,” IEEE Potentials, vol. 23, no. 5, pp. 40–43, Dec. 2004/Jan. 2005. 16. Y. Chen, C. Qiao, and X. Yu, “Optical Burst Switching: A New Area in Optical Networking Research,” IEEE Network, vol. 18, no. 3, pp. 16–23, May/June 2004. 17. J. Segarra, V. Sales, and J. Prat, “An All-Optical Access-Metro Interface for Hybrid WDM/TDM PON Based on OBS,” IEEE/OSA Journal of Lightwave Technology, vol. 25, no. 4, pp. 1002–1016, Apr. 2007. 18. M. Maier, M. Herzog, and M. Reisslein, “STARGATE: The Next Evolutionary Step toward Unleashing the Potential of WDM EPONs,” IEEE Communications Magazine, vol. 45, no. 5, pp. 50–56, May 2007. 19. A. A. M. Saleh and J. M. Simmons, “Architectural Principles of Optical Regional and Metropolitan Area Networks,” IEEE/OSA Journal of Lightwave Technology, vol. 17, no. 12, pp. 2431–2448, Dec. 1999. 20. L. Noirie, “The Road Towards All-Optical Networks,” in Proc., Optical Fiber Communication Conference (OFC), vol. 2, pp. 615–616, Mar. 2003. 21. P. E. Green, “Paving the Last Mile with Glass,” IEEE Spectrum, vol. 39, no. 12, pp. 13–14, Dec. 2002. 22. S.-J. Park, C.-H. Lee, K.-T. Jeong, H.-J. Park, J.-G. Ahn, and K.-H. Song, “Fiber-to-the-Home Services Based on Wavelength-Division-Multiplexing Passive Optical Network,” IEEE/OSA Journal of Lightwave Technology, vol. 22, no. 11, pp. 2582–2591, Nov. 2004. 23. I. Widjaja, I. Saniee, R. Giles, and D. Mitra, “Light Core and Intelligent Edge for a Flexible, Thin-Layered, and Cost-Effective Optical Transport Network,” IEEE Communications Magazine, vol. 41, no. 5, pp. S30–S36, May 2003. 24. C. Nuzman and I. Widjaja, “Time-Domain Wavelength Interleaved Networking with Wavelength Reuse,” in Proc., IEEE INFOCOM, Barcelona, Spain, Apr. 2006. 25. M. P. McGarry, M. Maier, and M. Reisslein, “WDM Ethernet Passive Optical Networks,” IEEE Communications Magazine, vol. 44, no. 2, pp. S18–S25, Feb. 2006. 26. D. Fallows, “Surfing for Fun,” http://www.pewinternet.org, Feb. 2006 [Online on 01 June, 2008]. 27. M. Garcia, D. F. Garcia, V. G. Garcia, and R. Bonis, “Analysis and Modeling of Traffic on a Hybrid Fiber-Coax Network,” IEEE Journal on Selected Areas in Communications, vol. 22, no. 9, pp. 1718–1730, Nov. 2004. 28. W. Feng, F. Chang, W. Feng, and J. Walpole, “A Traffic Characterization of Popular On-Line Games,” IEEE/ACM Transactions on Networking, vol. 13, no. 3, pp. 488–500, June 2005. 29. X. Jiang, F. Safaei, and P. Boustead, “Latency and Scalability: A Survey of Issues and Techniques for Supporting Networked Games,” in Proc., IEEE International Conference on Networks, Nov. 2005. 30. S. Sen and J. Wang, “Analyzing Peer-to-Peer Traffic Across Large Networks,” IEEE/ACM Transactions on Networking, vol. 12, no. 2, pp. 219–232, Apr. 2004. 31. Y. Chuh, J. Kim, Y. Song, and D. Park, “F-TCP: Light-Weight TCP for File Transfer in High Bandwidth-Delay Product Networks,” in Proc., IEEE International Conference on Parallel and Distributed Systems (ICPADS), vol. 1, pp. 502–508, July 2005.
Chapter 12
Signal Processing Techniques for Data Confidentiality in OCDMA Access Networks Yue-Kai Huang, Paul Toliver, and Paul R. Prucnal
Abstract This chapter focuses on several imminent security applications in optical CDMA networks where the strong potentials of optical signal processing could be leveraged. As one of the dominant technologies in wireless communications, the unique features of CDMA have attracted wide attention in many optical networking areas. We explored the security properties of optical CDMA networks enhanced by the aid of optical signal processing. In particular, optical encryption can be incorporated into the network through optical XOR gating. Steganography, another form of information hiding, can also be achieved through temporal pulse spreading. For a coherent spectral phase-coded OCDMA network, share code scrambling is proven to be an effective and reliable way of achieving channel confidentiality. The chapter also presents a cost-effective and robust device technology. Its small footprint and multi-code processing capability could significantly simplify the node and system architecture.
Yue-Kai Huang NEC Laboratories America, Inc., 4 Independence Way, Princeton, NJ 08540, USA, e-mail:
[email protected] Paul Toliver Telcordia Technologies, 331 Newman Springs Road, Red Bank, NJ 07701, USA, e-mail:
[email protected] Paul R. Prucnal Department of Electrical Engineering, Princeton University, Princeton, NJ 08544, USA, e-mail:
[email protected] This material is based upon the work supported by DARPA under contract number MDA972-03C-0078. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the view of DARPA.
A. Shami et al. (eds.), Broadband Access Networks, Optical Networks, c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-92131-0 12,
261
262
Yue-Kai Huang, Paul Toliver, and Paul R. Prucnal
12.1 Introduction As the bandwidth of today’s broadband wireless and wired-distribution networks increases, it is becoming increasingly difficult to implement real-time processing required for information security. For example, fiber-optic backbone networks are currently being deployed with data rates of 40 Gb/s, which exceed the capability of current electronic data encryption logic. Thus it is desirable to have the processing of information security carried out on the fly by direct implementation in optical domain. Optical code division multiple access (OCDMA) network divides the medium into multiple channels through the assignment of unique code sequence, allowing spread-spectrum communication for the users over a shared medium with flexible network utilization [1]. OCDMA can provide very high spectral utilization and flexible data rates while offering the same data routing capabilities as WDM systems. These characteristics make OCDMA a strong candidate for building future fiber-tothe-home (FTTH) networks [2]. OCDMA could also bring some benefits of wireless CDMA to optical domain, such as resistance to jamming and unauthorized detection. One of the oft-repeated arguments for OCDMA has been its potential to provide a certain degree of security in the physical layer, since spread spectrum [3] has been used extensively in military radio communications to ensure low probability of intercept by unintended receivers due to the large number of codes used in CDMA. However, having a large number of available codes is not a sufficient condition to guarantee data confidentiality. In [4], the author explored various aspects of OCDMA in depth and their implications to information security. Possible strategies of eavesdropping are presented, and the study found that early OCDMA designs, particularly the ones with on–off keying modulation, are at high risk to attacks through techniques like simple energy detection and accumulative sampling. The results indicated that it is simply not enough, based on the Kerckhoff’s principle, to ensure network security just by creating a large search space. Countermeasure must be taken, and other encryption/obscurity methods coupled with OCDMA are methods coupled with OCDMA are necessary to achieve data confidentiality at the physical layer. Through the aid of optical signal processing, the degree of difficulty to achieve optical code reconfiguration associated with these encryption/obscurity methods will be significantly reduced. This chapter is divided into several sections with the attempt to cover the ongoing research of various optical signal processing techniques for enhancing OCDMA security. Section 12.2 looks at an optical encryption approach which utilizes optical XOR to secure network connectivity. In particular, special dual-code transmitter and receiver were designed to perform optical encryption and decryption. In Section 12.3 the concept of optical steganography is presented and several temporal spreading techniques for OCDMA transmission are analyzed. Steganography can be combined with optical encryption to provide an integrated solution for key distribution. We then focus our attention to enhancing data confidentiality in coherent OCDMA systems. Section 12.4 describes how the technique of optical code division
12 Data Confidentiality in OCDMA Access Networks
263
multiplexed (OCDM) phase scrambling can drastically increase the coding space of spectral phase-encoded OCDMA and thus enhanced security on multiplexed data traffic. The scrambling operating and corresponding network structures are reviewed and some recent key progresses are presented [5]. Cost-effective and robust devices are essential for employing OCDMA in commercial and military applications. An integration platform based on holographic Bragg reflectors (HBR) is discussed in Section 12.5. Its capability to simultaneously process multiple codes significantly simplifies the node architecture and enables practical applications in OCDMA security.
12.2 Optical Encryption In a generic encryption system, the message is encrypted using a side information, or key, shared only among the sender and the receiver to maintain communication privacy. For binary digital communications, the encryption can be implemented by electronic XOR gating. However, the use of electronic logic gates may result in electromagnetic radiation of signatures which may be monitored by an adversary. Therefore it is desirable, for the purpose of minimizing radiation leakage or even achieving high-speed operation, to seek implementation using optical logics. Optical layer encryption can be applied to the OCDMA network using a dual-code transmitter/receiver pair. The specific transmitter will generate two OCDMA codes, Code 0 and Code 1, and select one of the codes for transmission depending on the XOR result of the data and the key values. In other words, the code assignment for binary data 1s and 0s can be swapped based on the value of the binary key. The code assignments are listed in Table 12.1. At the receiver, the signal can be decrypted by selecting the appropriate decoder according to the key. Table 12.1 Output of an OCDMA transmitter as the result of XOR operation Data
Key
Output
0 0 1 1
0 1 0 1
Code 0 Code 1 Code 1 Code 0
The channel isolation and data confidentiality achieved by optical encryption can be classified as a security level called one-time pad (OTP). In cryptography, onetime pad is an encryption algorithm where the plaintext is combined with a random key or “pad” that is exactly the same length as the plaintext. If the key is truly random, never reused, and kept secret, the one-time pad is proven to be unbreakable [6]. In a binary symmetric channel (BSC) where 1s and 0s have equal probability of occurrence, the channel capacity can be written as
264
Yue-Kai Huang, Paul Toliver, and Paul R. Prucnal
C B SC
= 1 − H ( p) = 1 − − p log2 p − (1 − p) log2 (1 − p)
(12.1)
H ( p) is the entropy function and can be calculated using p, the probability of correct reception. If the key is kept secret, p = 0.5 for an unauthorized detection, and the channel capacity will be minimized to 0, meaning the eavesdropper will not be able to obtain any information. The implementation of OCDMA code swapping with optical XOR can be done using electro-optical (EO) switches [7]. This approach allows us to use commercially available components to achieve optical XOR. This concept was demonstrated on a system which uses (3,11) wavelength-hopping time-spreading (WHTS) prime codes in which the signature is created using 3 wavelengths and 11 time chips. WHTS OCDMA system is a 2D optical coding approach that spreads the codes in both the time and wavelength domains with a hopping pattern that can be represented as 2D code matrices [8, 9]. A typical WHTS OCDMA network with broadcast star topology is shown in Fig. 12.1, where WHTS code sequence is formed in the transmitter by sending a multi-wavelength optical source to an encoder which does wavelength-dependent time spreading. At the receiver, an optical decoder correlates the received signal with its assigned code sequence. If the signal matches, this process will undo the time spreading, aligning the pulses to form an autocorrelation peak.The other non-matching code sequences will form the low-intensity multiple access interference (MAI) which can be removed by optical thresholding or optical gating. Transmitter 1
Receiver 1
Electronic Data
WHTS WHTS Encoder Encoder & & Modulator Modulator
WHTS Code 1
WHTS Decoder
Receiver Electronics
Multi-wavelength Pulse Source
Transmitter 2
Receiver 2
OCDMA DMA Network Receiver N
Transmitter N WHTS Code N
Fig. 12.1 Schematic of a general WHTS OCDMA network
Figure 12.2 shows the transmitter architecture which allows XOR code swapping between OCDMA encoders C1 and C2 using EO switches. The WHTS OCDMA codes are generated by spectral slicing from a supercontinuum source and then delaying the slices in time [7]. The two encoders are followed by two cascaded 2 × 2 LiNbO3 switches, which operate at cross or bar state depending on the driving voltage.The first switch was driven by the data to convert the 1.25 Gb/s electronic binary stream to OCDMA codes C1 and C2 for binaries 1s and 0s, respectively. The second switch, driven by the key on a bit-to-bit basis, will either maintain or swap the original code assignment. The output will produce the exact optical XOR relation
12 Data Confidentiality in OCDMA Access Networks Transmitter C
Optical Encoder C1
265 code
PC
λ3
D
λ2
1
800 ps
×
D
4
Encoder C1 output
λ1
2×2
λ3
D
λ2
Optical OUT
EO
EO
Switch
Switch
Optical Encoder C2 PC
2×2
1 ×
D
code ( 800 ps
4
λ1
Encoder C2 output RF Data
Swapping Key
Fig. 12.2 Schematic of OCDMA transmitter C for the secure user. The inset shows the two WHTS OCDMA code sequences generated by the optical encoders C1 and C2, respectively
as shown in Table 12.1. In the demonstration the key was a 231 -1 pseudo-random bit stream (PRBS) and not truly random, an aspect which should be kept in mind when comparing with true one-time pad data transmission. The key was then distributed using a separate electrical/fiber link. The schematic for the “dual code” receiver is shown in Fig. 12.3. Here the EO switch is used to select the desired OCDMA decoder (C1 or C2 ) for decoding by applying the key value. The latency for key distribution was kept the same as the encrypted OCDMA transmission. The recovered signal after decryption and decoding is plotted in the inset of Fig. 12.3., where the observed autocorrelation peak is actually an overlay of two different peaks formed independently by the two decoders C1 and C2 . The decoded autocorrelation peak has three times the intensity of the cross correlation, an attribute of the code weight w = 3 for (3,11) WHTS codes. Receiver C Optical Decoder C1
1
PC
TFF3 TFF2
4 2×2 EO Switch
TFF1
D
Receiver C Output 1
D D
4 Recovered Data
2×1
PD
Optical Decoder C2
1
TFF3 TFF2
4
TFF1
D
1
D D
4
Swapping Key
Fig. 12.3 Schematic of OCDMA receiver C for secure user and its recovered eye diagram
An eavesdropping node, which has an optical identical decoder to the codeswapping receiver (C1 ), was designated for studying the performance of optical encryption. Figure 12.4 shows the measured BER curve plot for the eavesdropping node. When the key value stayed the same, both the receiver and the eavesdropping nodes can obtain error-free detection. When PRBS 231 -1 was used for the key,
266
Yue-Kai Huang, Paul Toliver, and Paul R. Prucnal RxC (No swap)
10–2
RxC (Code-swapped) Rx EVE (No swap)
BER (log)
10–4 10–6 10–8 10–10 10–12 10–14 –26
–24
–22
–20
–18
–16
Receiver Power [dBm] Fig. 12.4 BER plot for code-swapped channel and the eavesdropping node
the eavesdropper will then obtain a BER around 0.5, thus not be able to obtain any information from the secure channel. On the other hand, with proper decryption and decoding, error-free operation can still be maintained for the code-swapping receiver. The experimental results confirm that the EO switch-assisted XOR encryption approach is viable in increasing the security level of WHTS OCDMA networks. In order to minimize the electromagnetic radiation from the encryption process and to avoid side channel attacks, all-optical XOR operation is required for the OCDMA code swapping. Figure 12.5 illustrates a two-stage architecture where a modified terahertz optical asymmetric demultiplexer (TOAD) [10] is followed by an optically controlled Mach–Zehnder (MZ) switch [11]. The TOAD used here is modified from the original design to include an additional control port, allocate two counter-propagating control signals: data and key in RZ format [12]. At the input port, an optical pulse train will split into both directions around the loop. If presented, the key and data pulse will alter the SOA phase experienced by the input. The offset distance of the SOA to the loop center is larger than the SOA recovery time. The output port will have destructive interference if the data and key are both on or both off. If only one is on, only one direction of the split input pulses will
Code swapped out
Data
Key Pulse XOR out Duplicator
Code 1
2x1 MZ switch w/ optical control Control
Fig. 12.5 Two-stage all-optical XOR module for OCDMA code swapping
Code 2
12 Data Confidentiality in OCDMA Access Networks
267
experience a π phase shift, creating the constructive interference. The resulting XOR output will be in RZ format after optical filtering to reject the control pulses. The XOR output pulse will then be used to control the MZ switch. The two input MZ interferometric devices allow handling of two OCDMA codes for code swapping. The switch is initially biased to switch one of the input codes to the output. With the existence of an optical control pulse, the interferometric condition at the output will alter, and another code will be switched. The switching window in this case is proportional to the SOA recovery time in the MZ. By using a pulse duplicator, the window can be lengthened to allocate longer code sequence. Figure 12.6 displays the experimental results using the above architecture with (3,11) WHTS code sequences. The output, instead of being swapped between two codes, will be switched on/off with only one code because the integrated SOAbased MZ switch only had one input port available at the time for proof-of-concept demonstration. From Fig. 12.6(a) and (b),the output patterns indicate the correct XOR-switched results. Figure 12.6(c) shows the recorded output pattern when both data and key carry 27 -1 PRBS. The clear distinction between the switched and nonswitched traces effectively verified the all-optical approach. (a)
(b)
(c)
Fig. 12.6 All-optically XOR-switched OCDMA signal when (a) data = key = “1111”, (b) data = “1010” and key = “0101”, and (c) both carry 27 -1 PRBS
12.3 Optical Steganography In general, the method of information hiding can be characterized into two categories: side information and steganography [13]. We have already discussed how cryptography, a form of side-information method, can be applied to OCDMA networks. Here we will focus on another category, steganography, and its application in optical communication to increase the level of network security. In steganography, information is camouflaged and hidden from unauthorized detection. One way to implement steganography in optical communication is by sending stealth messages in time-spread optical pulses with the same wavelength as the public channels [14]. The concept of the time-spread transmission is shown in Fig. 12.7. If the amount of temporal spreading is large enough, the signal can be hidden below the noise level. Amplified spontaneous emission (ASE) noise can be intentionally added to the transmitted output for better hiding. Time spreading can be achieved through the use of fiber dispersion or spectral phase encoding [15].
268
Yue-Kai Huang, Paul Toliver, and Paul R. Prucnal Spreading
I
ASE noise addition
I
Compression
I
t
I
t
t
t
Fig. 12.7 Illustration showing an optical pulse that underwent temporal spreading, stealth hiding with added ASE noise, and compression for detection
Since the stealth message occupies the identical spectrum as public channels, for adversaries who do not possess the necessary information to compress the spread pulse, the signal-to-noise ratio (SNR) of the key recovery will be extremely low, causing extreme difficulties for data recovery. In spectral phase-encoding approach, the optical pulses can be time spread by passing through a phase mask [15] which consists of N frequency bins accommodating the entire bandwidth W of public channel j. Each bin has a bandwidth of = W N and independent random phase φn ∈ [−π, π ]. The general expression for time-spread pulse is
1 F j (ω)G j (ω)e−iwt dω (12.2) a j (t) = 2π W G j (ω) =
N
r ect
n=1
ω +
N +1 − n eiφn 2
(12.3)
G j (ω) denotes the frequency transfer function of the phase mask, assuming each frequency bin exhibits a perfect rectangular shape, F(ω) the Fourier transform of initial pulse function, and r ect (.) a unit gate function. Increasing N can achieve large spreading while creating a larger search space for the adversaries. Though spectral phase spreading will provide better protection due to the encoded nature, it takes many bins (N > 100) to spread the signal intensity to a satisfactory level, which is still difficult with the current phase mask technologies. When spreading is implemented by fiber dispersion, the amplitude function of the spread pulse is given by 1 a j (z, t) = 2π
∞
−∞
A j (0, ω)e
i 2 2 β2 ω z−iωt
dω = √
1 1 − iS
e
−
t2 2τ02 (1−i S)
(12.4)
where β2 is the dispersion parameter, z the propagation distance, A j (0, ω) the Fourier transform of the pulse waveform before spreading (assuming Gaussian profile), τ0 is the half width (at 1/e-intensity point), and S = β22z the spreading factor. τ0
Optimally the goal is to spread the energy below the noise level of the public channel. The amount of spreading needed can be derived from the optical signal-to-noise ratio (OSNR) [16]:
12 Data Confidentiality in OCDMA Access Networks
269
√ 2O S N R Log(2) S≥√ √ π Er f ( Log(2))
(12.5)
For example, ∼70 km length of single mode fiber ( β2 = 16 ps2 /km) is required to spread 5 ps optical pulses in a 20 dB OSNR channel. The same result can be achieved by using shorter fiber with higher dispersion or discrete components such as fiber Bragg gratings. One application of optical steganography is for cryptographic key distribution. In the last section, the key signal for the OCDMA system with optical XOR encryption was distributed using an isolated and secured connection which has the same latency as the data. For practical applications, such physical separation along with the latency requirement cannot be guaranteed. Thus we present an example of using time-spreading techniques to secure key distribution for code-swapping OCDMA networks [16]. Figure 12.8 shows the scheme for stealth key distribution. First three WHTS OCDMA codes were generated with the same spectral contents. Then data modulation and the code scrambling were performed on c1 and c2 using the encryption technique described in Section 12.2. The third code, c3 , was modulated with the key information using on–off keying and then sent to a temporal spreading element. The experiment is performed on a 1.25 Gb/s OCDMA channel using (3,11) WHTS codes [7]. The pulse width for the three pulses used in the WHTS codes is ∼5 ps. A highly dispersive fiber is used to temporally spread the pulse width 40 times to ∼ 200 ps. Therefore the peak intensity of the key signal would be approximately ∼16 dB less than that of the code-swapped message, assuming all three codes carried equal power. RF Data
Optical switch Decoder 1
c1
Optical switches
EDFA
c2
OCDMA Network
Decoder 2
Recovered Data
RF Key RF Key
c3
Temporal spreading
MOD
Self-wrapped OCDMA transmitter
Temporal compression Key Data
0
1
0
C
C
1
C
C
Decoder 3
Key Processing
Self-wrapped OCDMA receiver
Fig. 12.8 Stealth key distribution for code-swapping OCDMA system
After the multiplexed OCDMA signal was received, it was first split for key and data recovery. A temporal compressor made of highly dispersive fiber was used to undo the spreading of the key signal by applying a reverse dispersive slope. After passing through the decoder for c3 , we were able to obtain the hidden key information because the peak intensity of the key signal is now much higher than that of the data due to the reverse dispersive process. The key signal was then optical sampled using an electro-absorption modulator (EAM) and photodetected. The recovered key information will be used to dynamically decode using the method described in
270
Yue-Kai Huang, Paul Toliver, and Paul R. Prucnal
(a)
(b)
Power (dBm)
–20
w key w/o key
–30
–40
1548
1550
1552
1554
Wavelength (nm)
Fig. 12.9 (a) Temporal and (b) spectral images of the multiplexed signal with stealth key
Section 12.2. The original data was retrieved after the outputs of the two decoders for c1 and c2 are combined and photodetected. Figure 12.9(a) shows the measured signal trace after the data (c1 /c2 ) and key (c3 ) were multiplexed together. As suggested by the temporal scan in Fig. 12.9(a) and the spectral scan in Fig. 12.9(b), the format is highly robust against eavesdropping because the necessary key information is hidden below noise level, while since both the message and the key are encoded within the identical spectrum so the key cannot be separated by spectral filtering. The scheme allows the combination of cryptographic and steganographic schemes in the code domain and provides practical means to implement the one-time pad (OTP) security discussed in the last section.
12.4 Phase Scrambling OCDM Optical code division systems can be found in various physical implementations, each taking a unique approach to coding in time, frequency, or a combination of both [1]. The codes may be applied to optical signals as a phase shift, amplitude modulation, or a combination of both. The wavelength-hopping time-spread (WHTS) OCDMA system described in detail in Section 12.2 is one example of leveraging both time and frequency coding for enhanced data confidentiality. In this section, we examine another implementation of an optical code division system that applies complex phase coding entirely in the frequency domain. We refer to this particular implementation of optical code division as spectral phase coding (SPC). Optical implementations of spectral phase encoding and decoding were first experimentally demonstrated by Weiner et al. [17] in 1988. Later, in 1990, a spectral phase-coded communications link was analyzed in greater detail by Salehi et al. [15]. These early wideband systems relied on ultrashort pulses (< 1 ps), creating broad spectrums (
12 Data Confidentiality in OCDMA Access Networks
271
phase-coded systems (a review of progress can be found in [18]), allowing current systems to apply spectral phase codes with a resolution of 10 GHz or less. A key advantage of such high-resolution coding is that it enables modern narrowband SPCOCDM systems [19] that approach the spectral efficiency achievable in conventional dense wavelength division multiplexed (DWDM) systems. The essential elements of an SPC-OCDM transmitter and the encoding steps involved are illustrated in Fig. 12.10. A coherent light source is provided by modelocked laser (MLL), which generates a train of short pulses in the time domain, while in the frequency domain, this corresponds to a comb of phase-locked optical frequencies spaced at the pulse repetition rate. The MLL pulse train is modulated with data streams originating from individual OCDM channels or tributaries, resulting in spectral broadening across all comb components. A variety of optical modulation formats are possible including the manipulation of amplitude (e.g., RZ-OOK), phase (e.g., DPSK, DQPSK), or both (e.g., duobinary) [1]. The narrowband spectral phase-encoding process comprises three operations: first, individual comb frequencies are demultiplexed line by line; then, each spectral line is phase shifted depending on the particular tributary’s OCDM code; and finally, these shifted frequency components are recombined to produce a coded waveform. Encoding results in temporal spreading of the input signal while leaving the set of frequency amplitudes unaltered, shifting only their relative phases. The encoded optical signal is combined passively with the other N − 1 tributaries, each assigned its own unique OCDM code. The final stage of spectral phase code scrambling illustrated in Fig. 12.10 allows for enhanced confidentiality and is described in detail later in this section.
Fig. 12.10 Block diagram of SPC-OCDM transmitter (MLL: mode-locked laser)
After all tributaries have been multiplexed at the originating node, the aggregate OCDM signal can be carried over transparent paths through a reconfigurable optical network to its final destination [20]. The decoding process performed at the receiver, which is essentially a mirror image of the transmitter, is illustrated in Fig. 12.11. After code descrambling (which applies a conjugate phase to the code scrambler), optical power is passively split between N tributary decoding elements. For each OCDM tributary, the spectral phase is realigned only for that particular channel, recreating its original MLL pulse. By using synchronous transmission and optically
272
Yue-Kai Huang, Paul Toliver, and Paul R. Prucnal
Fig. 12.11 Block diagram of SPC-OCDM receiver
orthogonal OCDM coding [19], signal energy from the other N − 1 tributaries can be forced to fall in a region outside the optically gated pulse interval. After optical time gating [1, 10], each tributary’s pulse stream is independently demodulated and detected. The combination of coherent signal processing, optically orthogonal coding, synchronized transmission, and optical time gating allows all codes to be present simultaneously (even though spectral components of all tributaries overlap entirely, unlike DWDM where each channel is given a distinct frequency), resulting in high potential link spectral efficiency. Given the ability for complex-valued spectral phase masks with spectral phase encoding, orthogonal optical codes are possible through the use of the well-known binary Walsh–Hadamard sequences: (i)
(i)
(i)
(i)
(i)
c j = e jα j = h j;N with α j ∈ {0, π } ⇒ c j ∈ {−1, 1}
(i) where h (i) j;N (i, j = 1, 2, ..., N ) represents the jth element of h N which is the ith (1)
(2)
(N )
column of the Hadamard matrix H N = h N , h N , ..., h N of order N . Although orthogonal OCDM coding allows for the different tributaries to be optically multiplexed and later separated, Walsh–Hadamard codes by themselves do not provide much confidentiality for the underlying tributaries. For a given order N of Hadamard codes, the maximum number of possible orthogonal code states at one time is also N . Therefore, an eavesdropper equipped with an adjustable decoder would have to guess only on the order of N possible code settings in order to recover any given tributary. The obscurity provided by SPE OCDM systems can be improved if a large number of available codes are used [4], forcing the eavesdropper to search a much larger code space before guessing the correct decoder settings. The first technique we describe for expanding the size of the possible code search space is code scrambling through the use of modified Walsh–Hadamard sequences [21]. With this technique, a large number of equivalently orthogonal matrices W N can be generated from H N by pre-multiplying by a diagonal matrix D N of order N , with all of the on-diagonal elements being arbitrarily chosen phase shifts eiφn (where φn ∈ 0, 2π for n = 1 to N ): W N = D N × HN
12 Data Confidentiality in OCDMA Access Networks
273
Fig. 12.12 (a) Simulation of four individual Hadamard-8 coded tributaries (1, 2, 4, 8) prior to scrambling and (b) following scrambling by a random quaternary phase mask. Also shown for reference is a rectangular window, which defines the time interval over which a receiver would optically gate the multiplexed signal in order to extract a single OCDM tributary
The application of modified Walsh–Hadamard sequences to optical code division systems was first described in [5]. Figure 12.12 illustrates an example simulation of scrambling four individual OCDM tributaries, each coded with unique Hadamard8 sequence, showing the temporal intensity variation over two bit periods. Figure 12.12(a) is the original waveforms associated with Hadamard codes 1, 2, 4, and 8 from the Hadamard-8 code set. The spiky nature of the patterns and their discrete appearance in the time domain make the codes vulnerable to detection by an eavesdropper. However, the corresponding set of scrambled Hadamard-8 codes using random 0 and π phase shifts is shown in Fig. 12.12(b). This figure clearly illustrates the ability to reduce the range of intensity variations by spreading out pulse energy more uniformly over a bit interval. This degree of signal obscuration coupled with the potentially large number of possible scrambler states and the ability to dynamically change the scrambler code setting all contribute to the overall enhanced obscurity of the composite signal. There are two important points to make about the modified Walsh–Hadamard codes. First, while there are still only N distinct orthogonal codes available for possible multiplexing at any one time, the number of possible W N code configurations is governed by the large number of different arbitrarily chosen D N matrices that can exist. With p possible phase states at each of the N diagonal elements, the number of distinct D N matrices is p N . For example, with N = 16 and p = 4, the number of distinct D N matrices is 416 = 232 , while with N = 32 and p = 8 possible phase states yield 296 distinct D N . While some of these states are degenerate in that they would yield essentially the same fields (i.e., when two matrices D N differ only by a scalar multiple), it is clear that the search space for guessing the possible code settings in order to eavesdrop on a given connection can be made far larger than the N conventional Hadamard codes. A second technique for expanding the size of the possible code search space for OCDM systems even further is through the use of monomial matrices. The number of possible code configurations can be expanded by a factor of N! by redefining W N = D N × M N × H N , where M N is a monomial matrix (only one nonzero element equal to 1 in each row and in each column) of order N . Grouping the monomial term with the Hadamard matrix effectively rearranges the usual Hadamard matrix, while still maintaining orthogonality through this transformation.
274
Yue-Kai Huang, Paul Toliver, and Paul R. Prucnal
In Fig. 12.10, the encoding element is shown as logically separate from the scrambling element. However, it is straightforward to combine the encoding and scrambling matrix into a single physical element for each transmitter tributary. Similarly, the descrambling and decoding processes can be combined into a single physical element at each receiver tributary. An important consideration in practical OCDM systems is the ability to configure the system and allow for the scrambling code to be periodically updated. This, of course, requires a programmable spectral phase coder. A number of static and dynamic spectral phase-coding technologies have been demonstrated, including diffraction gratings, imaged phased array (VIPA) filters, fiber Bragg gratings, arrayed waveguide gratings, and micro-ring resonators [18]. Of these technology options, the micro-ring resonator (MRR) solution has the combined advantages of high spectral resolution, reconfigurability phase codes, and photonic integration. Figure 12.13(a) illustrates the architecture of an N -stage spectral phase encoder/decoder based on 4-ring MRR demultiplexing, and Fig. 12.13(b) shows the extremely small dimensions possible for an example 3-ring MRR filter stage fabricated in an Si O2 /Si platform [22].
Fig. 12.13 (a) Programmable spectral phase encoder/deocoder architecture based on 4-ring MRR and (b) photo of example integrated 3-ring MRR filter stage
The development of programmable encoding/decoding technologies has enabled a number of significant OCDM system demonstrations. Phase scrambling in an OCDM system was first demonstrated experimentally using binary phase codes [23], where elements of diagonal matrix D N are set to either 0 or π . The system applied spectral phase codes on a 10 GHz grid, and N was set to 8, allowing for 28 possible scrambling codes. The temporal response for an encoded 10 GHz pulse train without data modulation is shown for four of the conventional Hadamard codes (H1 , H2 , H3 , and H4 ) in Fig. 12.14(a). Experimental waveforms corresponding to the scrambled codes are shown in Fig. 12.14(b), where the scrambling diagonal for this demonstration is the randomly selected code [−1, 1, 1, 1, −1, 1,−1, −1]. Finally, the experimental waveforms of each OCDM tributary after proper descrambling and decoding (with decoding phase mask set to H4 ) are shown in Fig. 12.14(c). The ability to set the micro-ring resonator-based coders to arbitrary spectral phase masks [22] allows for the extension to multilevel scrambling codes. Quaternary scrambling was applied to an experimental 4-channel polarization-multiplexed OCDM system, where each coded tributary was operating at 5 Gb/s [24]. With N = 8, the number of possible scrambling codes is expanded to 48 . In later experiments, the size of the possible code space was further expanded in a 2-channel
12 Data Confidentiality in OCDMA Access Networks
275
Fig. 12.14 Experimental demonstration of (a) conventional Hadamard encoded signals; (b) scrambled Hadamard codes; (c) decoding with code H4 for all received tributaries
OCDM system by combining quaternary phase scrambling with monomial matrix transformations [25]. Together, the two transformations can be viewed as a secret key which increases the search space from N to (4 N −1 N!) by both randomly permuting rows of the Hadamard matrix and adding a random scrambling phase to each spectral component. In addition to validating the essential phase scrambling operations over short fiber links, another important practical consideration is the ability to propagate scrambled OCDM signals over long distances while maintaining bandwidth efficiency. Using DPSK modulation format (for improved tolerance to inter-channel crosstalk [26]), an experimental OCDM system consisting of eight coded, phase-scrambled and polarization-multiplexed tributaries, each operating at 5 Gb/s, was used to demonstrate transmission over a 400 km dispersion-compensated link [27]. The 40 Gb/s aggregate OCDM signal was successfully recovered at FEC-correctable error rates while confined to an 80 GHz optical bandwidth, demonstrating both the capability for high spectral efficiency and long-distance transmission.
12.5 Multi-code Processing Techniques Practical OCDMA system implementations, in both commercial and military networks, will require that the code processing devices be made using robust, lightweight, and low-cost technology platforms. Here we focus on a relatively new technology platform based on holographic Bragg reflector (HBR) [28]. HBRs are computer-generated planar volume holograms fabricated in slab waveguides using deep ultraviolet photolithography. The HBR technology, relying on multi-path interference, transfers the excellent passband control in planar lightwave circuit
276
Yue-Kai Huang, Paul Toliver, and Paul R. Prucnal
environment. Integrated circuit fabrication and embossing-based replication will ultimately provide inexpensive devices in high volumes. The possibility of overlaying, stacking, and interleaving of HBRs [28] provides a pathway to build WHTS en/decoders for simultaneous multi-code processing with very small footprint. Simultaneous processing of two WHTS codes using a single HBR-based device was demonstrated. The HBR-based encoders and decoders were designed to serve two OCDMA channels at 1.25 Gb/s with (3,11) WHTS codes. The encoder was designed to simultaneously produce two codes – (1, 2, 3) and (1, 3, 5). The numbers represent the position of the time chip (11 in total) where the three wavelengths λ1 , λ2 , and λ3 were placed. Figure 12.15 is a top-view schematic of the HBR encoder. The spectral width of the input pulse, typically a supercontinuum, is required to encompass λ1 ∼ λ3 . The three HBR gratings then select the desired wavelengths λ1 , λ2 , and λ3 and time delayed them by the virtue of their ∼ 7.5 mm spacing. The thuscreated (1,2,3) code sequence leaves through output 1. A broadband HBR will reflect a fraction of the impinging (1,2,3) code back to the HBR array. As the split-off passes the HBR array for the second time the time delays between the three spectral slices are doubled, as shown in Fig. 12.16(b), creating the (1,3,5) code which exits through output 2. The optical spectrum of the two codes is depicted in Fig. 12.16(a), showing excellent fall-off characteristics and spectral isolation of the spectral bins. The shape of the spectral bins and their relative intensity are slightly distorted due to the non-uniformity of the supercontinuum spectral intensity. Double passing of the HBR array was only necessitated due to the length limitation of the 33 mm long die. Single-pass designs will be sufficient for the data rates above OC-48. HBRs λ1
λ2
λ3
Broad band HBR splitter Code (1,2,3) output Multi-λ Input Code (1,3,5) output
7 mm
7.5 mm 7.5 mm
Fig. 12.15 Schematic top view of the 1 × 2 integrated holographic OCDMA encoder
The layout of the matched 1 × 2 HBR-based decoder is identical to the encoder design except that the spatial positioning of the gratings occurs in opposite order to undo the time delays created by the encoder. The HBR-based device employs a dual-layer core architecture based on the silica-on-silicon platform [28]. Both encoder and decoder were fabricated from a laser-written reticle employing a
Intensity (dB)
12 Data Confidentiality in OCDMA Access Networks –10 –15 SC Input –20 –25 –30 –35 –40 –45 –50 –55 –60 –65 –70 1548 1549 1550
(a)
277 (b)
Code (1,2,3) λ1
λ3
λ2
Code (1,2,3)
Code (1,3,5) λ1
λ2
λ3
Code (1,3,5)
1551
1552
500 ps
1553 1554
Wavelength (nm)
Fig. 12.16 Spectra (a) and time signatures (b) of the generated WHTS codes
deep UV optical stepper, standard etching, deposition, and annealing processes. The overall die size containing both the encoder and encoder was only 5 × 33 mm2 . To study the performance of the HBR-based decoder, both codes were modulated with different data and multiplexed then sent into the HBR-based decoder. BER measurements showed error-free operation for both OCDMA channels. The experimental results clearly demonstrate the capability of HBR-based devices for multi-code processing. This capability also makes feasible a number of applications that can enhance OCDMA network security. In Section 12.4, shared code scrambling is discussed in the context of spectral phase OCDMA to utilize the large search space for providing enhanced data confidentiality. Though proposed initially in the context of spectral phase codes, shared code scrambling can also be generalized to cases when the scrambling mechanism is different (e.g., WHTS) from the coding mechanism used for multiplexing (e.g., spectral phase OCDMA). A schematic of this approach is shown in Fig. 12.17. The
Insecure Network
Secure Islands WHTS code scrambler
Fixed WHTS encoder Star coupler
Multiplexed secure island traffic
c1
To insecure network
Multi-Code Multi-Code Encoder cN Encoder Optical Switch
Fig. 12.17 Schematic of a shared code scrambling network. Nodes are represented by the shaded rectangles and the lock represents the shared code scrambler
278
Yue-Kai Huang, Paul Toliver, and Paul R. Prucnal
multiplexed data of a group of users located on a secure island are deemed secure. A coder was used for each secure island to dynamically scramble the multiplexed traffic before transmission into the insecure network. The shared code was unscrambled at the destination. A key was shared among all secure islands to determine the dynamic scrambling sequence. A multi-code (M codes) encoder configured with optical switch can be used to dynamically scramble among the M codes, as shown in the inset of Fig. 12.17. The size of the multi-code encoder and the switching speed depends on the demanded level of security. M-ary modulation can achieve higher spectral efficiency and better network performance in OCDMA networks [29]. The user is allowed a set of M codes, where each of the M codes will represent a symbol which has a bit length of log2 M in binary data. In addition, M-ary schemes also thwart simple energy detection eavesdropping as all data symbols carry optical energy. Practical implementation of the M-ary approach was limited due to the requirement of M en/decoders at each network node. With the aid of multi-code en/decoders, M-ary OCDMA can be achieved by optical switching in the transmitter and receiver at the symbol rate B/log2 M, where B is the raw data rate. M-ary modulation can be combined with dynamic code scrambling to further enhance the channel isolation and security. The optical encryption method discussed in Section 12.2 is a special case of such for M = 2. The availability of multi-code processing encoder and decoder will again simplify the transmitter/receiver architectures. Figure 12.18 demonstrates how the integrated encoder/decoder will be employed in the implementation. Key 011
Data 101
(a)
(b)
Key 011 M=2
M=2 Multi-λ optical source
Encoder cAcAcA CA 2×2 Switch Encoder cBcBcB CB
cBcBcA
cAcBcA
cBcBcA
2×2 Switch
cBcA
2×2 Switch
Switch Control: “1” bar “0” cross
RF Data Out
PD cB
cBcAcB
Dual-code transmitter
Decoder CA
101
Decoder CB
Dual-code receiver
Fig. 12.18 Schematic of (a) dual-code transmitter and (b) receiver
12.6 Summary The goal of this chapter is to locate the optical signal processing techniques to aid and enhance data security in OCDMA networks. Different techniques were presented to enhance confidentiality in OCDMA broadcast systems. Optical encryption techniques based on optical XOR logics, such as EO switches or all-optical gates, were incorporated into OCDMA nodes for code swapping and enhancing data security/privacy. The utilization of the optical steganography by time-spreading techniques can provide isolated and stealth distribution of the cryptographic key for
12 Data Confidentiality in OCDMA Access Networks
279
the code-swapping OCDMA nodes. For coherent spectral phase systems, dynamic shared code scrambling of the multiplexed OCDMA traffic will achieve search space magnitudes higher than just the codes alone to enhance data confidentiality. For compact implementation, the simultaneous multi-code processing capability of integrated en/decoders implemented using HBRs and micro-rings can boost system scalability and provide secure network functionalities with minimal performance trade-off. Other than the techniques described in this chapter, there are several signalprocessing techniques worth mentioning. Optical frequency hopping of OCDMA codes, which can be achieved by parametric four-wave mixing process, will help to evade jamming and improve network availability [30]. Survivability of access network against accidental or deliberate service interruption can be supported by OCDMA code-based protection paths [31]. As for implementing optical encryption in OCDMA networks, the speed of the XOR operation can be potentially increased by substituting the SOA-based optical XOR gating with nonlinear fiber logic [32]. Ongoing studies are carried out to analyze the performance of these approaches and test their feasibility in realistic scenarios.
References 1. P. R. Prucnal, Optical code division multiple access: Fundamentals and applications, New York: Taylor and Francis, 2006. 2. K. I. Kitayama, X. Wang, and N. Wada, “OCDMA over WDM PON – solution path to gigabitsymmetric FTTH,” J. Lightwav. Technol., vol. 24, no. 4, pp. 1654–1662, Apr. 2006. 3. A. J. Viterbi, “Spread spectrum communications – Myths and realities,” IEEE Commun. Mag., vol. 17, no. 3, pp. 11–18, May 1979. 4. T. H. Shake, “Security performance of optical CDMA against eavesdropping,” J. Light-wav. Technol., vol. 23, no. 2, pp. 655 (2005). 5. R. C. Menendez, P. Toliver, S. Galli, A. Agarwal, T. Banwell, J. Jackel, J. Young, and S. Etemad, “Network applications of cascaded passive code translation for WDM-compatible spectrally phase-encoded optical CDMA”, J. Lightwav. Technol. 23, pp. 3219–3231 (2005). 6. C.E. Shannon, Bell Sys. Tech. J., 1949, vol. 28, pp. 656–715. 7. I. Glesk, Y.-K. Huang, C.-S. Bres, and P. R. Prucnal, “Design and demonstration of a novel Optical CDMA platform for avionics applications,” Opt. Comm., vol. 271, no. 1, pp. 65–70 (2007). 8. L. Tanceski and I. Andonovic, “Wavelength Hopping/Time Spreading Code Division Multiple Access Systems,” Electron. Lett., vol. 30, no. 9, pp. 721–723, 1994. 9. C.-S. Bres, I. Glesk, and P. R. Prucnal, “Demonstration of an eight-user 115-Gchip/s incoherent OCDMA system using supercontinuum generation and optical time gating;” IEEE Photon. Technol. Lett., vol. 18, no. 7, pp. 889–891, Apr. 2006. 10. J. P. Sokoloff, P. R. Prucnal, I. Glesk, and M. Kane, “A terahertz optical asymmetric demultiplexer (TOAD),” IEEE Photon. Technol. Lett., vol. 5, pp. 787–789 (1993) 11. T. S. El-Bawab, Optical switching, New York: Springer, 2006. 12. T. Houbavlis, K. Zoiros, K. Vlachos, T. Papakyriakopoulos, H. Avramopoulos, F. Giardin, G. Guekos, R. Dall’Ara, S. Hansmann, and Burkhard, “All-optical XOR in a SOA-assisted fiber sagnac gate,” IEEE Photon. Technol. Lett., vol. 11, no. 3, pp. 334–336 (1999). 13. P. Moulin and J. A. O’Sullivan, “Information- theoretic analysis of information hiding,” IEEE Trans. Inform. Theory vol. 49, pp. 563–593 (2003).
280
Yue-Kai Huang, Paul Toliver, and Paul R. Prucnal
14. B. B. Wu, P. R. Prucnal, and E. Narimanov, “Secure transmission over an existing public WDM lightwave network,” IEEE Photon. Tech. Lett., vol. 18, pp. 1870–1872, Sept. 2006. 15. J. A. Salehi, A. M. Weiner, and J. P. Heritage, “Coherent ultrashort light pulse code-division multiple access communication systems,” J. Lightwav. Technol., vol. 8, no.3, pp. 478–491 (1990). 16. Y.-K. Huang, B. Wu, I. Glesk, E.E. Narimanov, T. Wang, and P. R. Prucnal, “Combining cryptographic and steganographic security with self-wrapped optical code division multiplexing techniques,” Electron. Lett., vol 43, no. 25, pp. 1449–1451 (2007). 17. A. M. Weiner, J. P. Heritage, and J. A. Salehi, “Encoding and decoding of femtosecond pulses,” Opt. Lett., vol. 13, pp. 300–302, 1988. 18. J. P. Heritage and A. M. Weiner, “Advances in spectral optical code-division multiple-access,” IEEE J. Select. Top. Quant. Electr., vol. 13, pp. 1351–1369, 2007. 19. S. Etemad, T. Banwell, S. Galli, J. Jackel, R. Menendez, P. Toliver, J. Young, P. Delfyett, C. Price, and T. Turpin, “Optical-CDMA incorporating phase coding of coherent frequency bins: concept, simulation, experiment,” Proc. Opt. Fiber Commun. Conf., Los Angeles, CA, 2004. 20. P. Toliver, J. Young, J. Jackel, T. Banwell, R. Menendez, S. Galli, and S. Etemad, “Optical network compatibility demonstration of O-CDMA based on hyperfine Spectral Phase Coding,” LEOS 2004, Paper WE3, Puerto Rico, 2004. 21. B. J. Wysocki, T. A. Wysocki, “Modified Walsh-Hadamard sequences for DS CDMA wireless systems,” Int. J. Adapt. Control Signal Process., vol. 16, pp. 589–602, 2002. 22. A. Agarwal, P. Toliver, R. Menendez, S. Etemad, J. Jackel, J. Young, T. Banwell, B. E. Lit-tle, S. T. Chu, C. Wei, C. Wenlu, J. Hryniewicz, F. Johnson, D. Gill, O. King, R. Davidson, K. Donovan, and P. J. Delfyett, “Fully programmable ring-resonator-based integrated photonic circuit for phase coherent applications,” J. Lightw. Technol., vol. 24, no. 1, pp. 77–87, Jan. 2006. 23. A. Agarwal, R. Menendez, P. Toliver, S. Etemad, and J. Jackel, “Code scrambling in spectral phase encoded OCDMA using reconfigurable integrated ring resonator based coders,” Optical Amplifiers and Applications/Coherent Optical Technologies and Applications (OAA/COTA), Paper CFD4, Whistler, Canada, 2006. 24. A. Agarwal, R. Menendez, P. Toliver, J. Jackel, and S. Etemad, “Enhanced confidentiality with multi-level phase scrambling in SPE-OCDMA,” Conference on Lasers and Electro-optics, paper CThBB2, Baltimore, MD, 2007. 25. A. Agarwal, R. Menendez, P. Toliver, J. Jackel, and S. Etemad, “Demonstration of modified hadamard codes for OCDM-based confidentiality,” Proc. European Conf. Opt. Commun., paper 10.5.6, Berlin, Germany, 2007. 26. X. Wang, N. Wada, T. Miyazaki, and K. Kitayama, “Coherent OCDMA system using DPSK data format with balanced detection,” IEEE Photon. Technol. Lett., vol. 18, no. 7, pp. 826–828, April 2006. 27. P. Toliver, A. Agarwal, T. Banwell, R. Menendez, J. Jackel, and S. Etemad, “40 Gb/s OCDMbased signal transmission over 400 km using integrated micro-ring resonator-based spec-tral phase encoding and quaternary code scrambling for enhanced data confidentiality,” Proc. European Conf. Opt. Commun., paper PDP3.3, Berlin, Germany, 2007. 28. D. Iazikov, C. M. Greiner, T. W. Mossberg, “Integrated holographic filters for flat-passband optical multiplexers,” Opt. Expr., vol. 14, p. 3497, 2006. 29. E. Narimanov, W. C. Kwong, G.-C. Yang, and P. R. Prucnal, “Shifted carrier-hopping prime codes for multicode keying in wavelength-time O-CDMA”, IEEE Trans. Commun., vol. 53, no. 12, pp. 2150–2156, Dec. 2005. 30. Z. Wang, A. Chowdhury, and P. R. Prucnal, “Optical CDMA code wavelength conversion using PPLN to improve transmission security,” IEEE Photon. Tech. Lett., vol. 21, no.6, pp. 383–385, Mar. 2009. 31. C.-S. Br`es and P. R. Prucnal, “Code-empowered lightwave networks,” J. Lightwav. Tech-nol., vol. 25, no. 10, pp. 2911–2921, Oct. 2007. 32. K. Kravtsov, P. R. Prucnal, and M. M. Bubnov, “Simple nonlinear interferometer-based alloptical thresholder and its applications for OCDMA,” Opt. Expr., 15(20), p. 13114, Oct. 2007.
Part IV
Optical-Wireless Access Networks
Chapter 13
Radio-over-Fiber (RoF) Networks John E. Mitchell
Abstract As the bitrates required by wireless services increase and consideration starts to shift to higher carrier frequencies, the distribution of radio signals over optical fibers becomes attractive for both antenna remoting in pico-cellular applications as well as for providing more complex radio backhaul network topologies. This chapter outlines the basic structure of a radio-over-fiber (RoF) link and details some of the technology and architectural options. It continues to look at the radio applications considering some of the issues faced as well as current solutions.
13.1 Introduction The distribution of radio signals using optical fiber technology can offer a range of advantages. These are particularly acute when a high number of small cells (perhaps pico-cells) are required, where the frequencies involved are difficult to transport using co-axial cables or where dead-zones in radio coverage exist. Some of the key application areas are demonstrated in Fig. 13.1 including in-building applications (where strong coverage is difficult) and broadband wireless access. These cover most of the current, commercially available systems which provide distribution of public mobile (cellular) telephone services in high-density environments. Wellknown examples include the 2000 Olympic Games in Sydney [1], Osaka Station in Japan and the Bluewater Shopping Centre in the UK. In a traditional radio access network, cell sites are fed with baseband data packets. These feeds may be optical fiber based, although in many cases micro-wave backhaul or copper leased lines are utilized. At the cell site the base transceiver station (BTS) provides all the processing, modulation, and amplification required before the signal is radiated by the antenna unit. It should be noted that the ultimate John E. Mitchell Department of Electronic and Electrical Engineering, University College London, London WC1E 7JE, UK, e-mail:
[email protected]
A. Shami et al. (eds.), Broadband Access Networks, Optical Networks, c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-92131-0 13,
283
284
John E. Mitchell
Fig. 13.1 Typical application of radio-over-fiber networks
cell capacity is determined by the amount of equipment placed at this site. The space and plant required varies but >1m2 cabinets or even small buildings are not uncommon in GSM or UMTS networks. Radio-over-fiber techniques look to fundamentally change this architecture. The main advantages are gained through the reduction in the remote infrastructure by allowing the BTS to be centrally located and the signal then distributed to a much smaller remote antenna unit (RAU). In addition, the use optical fiber has the advantage of being a small and light weight media, as well as offering immunity to electro-magnetic interference (EMI). Using this technique it is possible to design the remote antenna unit to be wideband and protocol independent, so that a number of radio standards can be distributed over a single fiber infrastructure. The general structure of radio-over-fiber systems are all broadly similar, although a wide range of technologies have been proposed to offer the functionalities required. Figure 13.2 shows this general structure of a radio-over-fiber link with an
Fig. 13.2 Generic schematic of a radio-over-fiber system
13 Radio-over-Fiber (RoF) Networks
285
optical transceiver located at the central office taking a feed from the output of the radio node. Here the radio signal, which may already be at the final radio frequency (RF) of transmission or may be at a lower intermediate frequency (IF) for upconversion later, is modulated onto a lightwave carrier. This allows transport across an optical network to the RAU. At this point the signal is recovered from the optical carrier and amplified (if necessary) ready for wireless transmission. The losses in the optical fiber are significantly less than that of coaxial cable and, to a first approximation, independent of the frequency of the radio signal being carried. The return path involves an identical functionality of modulation onto an optical carrier, transport, and recovery. However, it should be noted that as the main aim of this technology is reduce the cost, size, and power requirements of the RAU, a large research effort has been expended in the development of low-cost devices for this part of the network. Of particular interest are single optical devices that can both receive and transmit the optical signals with some demonstrating powerless operation often termed, passive pico-cells [2]. These were based on electro-absorption transceiver (EAT) devices which will recover optical signals within a specific wavelength band and modulate data onto a carrier within a second wavelength band [3]. More recently, devices with similar functionality based on asymmetric Fabry–Perot modulators (AFPM) [4] have demonstrated wireless transmission at 5.5 GHz [5]. The techniques described above have, in some cases, been described as antenna remoting as, in essence, all the optical network is doing is allow the final antenna site to be removed to some distance from the baseband processing modulation electronics. However, as we will discuss later in this chapter the introduction of fiber into this network and the associate co-locating of all the processing and channel assignment capabilities at a central location provide the opportunity for addition functionality. To date no commercial system exists that takes advantage of this and the possibilities for dynamic capacity allocation offered, although this is an area of very active research. This chapter begins with a discussion of the basic techniques involved before outlining some of the basic applications of radio-over-fiber technology considering the currently known radio formats of interest. It concludes by demonstrating the current research that is aiming to develop the current point-to-point technologies into fully functional networks and discusses some of the challenges faced.
13.2 Basic Technologies The most basic optical link for the transport of radio signals can be seen in Fig. 13.3. The laser is biased in the linear region of its transfer characteristics and directly modulated with the radio signals. This is directly coupled into the optical fiber for transmission and then detected using a biased photodiode. Typically direct modulation is available for frequencies < 10 GHz (although some devices do exist above this [6]). External modulators based on either electro-optical effects such as the Mach–Zehnder modulator or electro-absorption-based modulators can be used for
286
John E. Mitchell
Fig. 13.3 Basic analogue over fiber link
frequencies above this and also typically demonstrate superior link performance but at the expense of additional cost. If a reasonably low modulation depth is used at the laser to maintain operation in the linear region then reasonable performance in terms of distortion can be achieved. If increased linearity is required then techniques to linearize the devices can be applied. This has been demonstrated for both directly modulated devices [7, 8] as well as to linearize external modulators [9] which is a technique commonly used in cable TV (CATV) networks. To enable the use of low-cost sources more advanced techniques such as feed-forward linearization have also been proposed [10]. A prime parameter of concern in analogue links, although often ignored in digital systems, is the relative intensity noise (RIN) of the optical device. Direct modulation of the laser diode will introduce some unwanted intensity variations caused by effects such as mode instability or spontaneous emission. This intensity noise is usually characterized via the RIN parameter. The RIN can be theoretically evaluated by solving the laser rate equations where typically it lays in the range 130–160 dB Hz−1 for a semiconductor laser. It peaks in the vicinity of the laser relaxation oscillation frequency and is a function of the modulating signal frequency and the bias current. In the directly modulated link a trade-off exists between maintaining sufficient optical power at the photodiode to overcome the receiver thermal noise, while minimizing the effect of RIN which increases with the square of the optical power [11]. Optical power reflections back into the laser also cause an increase in RIN. To achieve good RF to RF link performance appropriate impedance matching of the laser and photodiode are required. In certain circumstances gain can be achieved [12]. An excellent review of the state of the art of analogue links can be found in [13], while those looking for a rigorous analysis of the design of such links should see [11]. Taking this as the basic structure, full duplex point-to-point links can be formed. For transmission of frequencies below around 6 GHz this is typically achieved by repeating the link shown in Fig. 13.3. However, although in terms of loss the optical fiber is approximately independent of the radio frequency, above around 10 GHz chromatic dispersion begins to have a significant influence on the transmission of an intensity modulation direct detection (IM-DD) signal. Here, we will consider standard single mode fiber with a chromatic dispersion D = 17 ps/nm.km at λ = 1550 nm. If we analyze the optical signal produced by external modulation of a CW single mode laser with an ideal Mach–Zehnder modulator (MZM) biased for linear intensity we see that the optical power spectrum can be written as
13 Radio-over-Fiber (RoF) Networks
S( f ) =
Po 2
∞
287
2 J2n+1 (m)δ f − f o − (2n + 1) f mm
(13.1)
n=−∞
where f o is the mean optical frequency, Po is the total optical power, Jn represents the Bessel function of the first kind of order n, and m is the modulation index calculated as the ratio of the input voltage range to the switching voltage of the modulator. It is seen that the radio-frequency signal is carried as a lower and an upper sideband (n = −1 and n = 1) offset by the frequency of the signal from the optical carrier (n = 0). Due to the presence of three main terms in the optical spectrum we define this to be a three-term technique. Chromatic dispersion causes a differential delay to be progressively induced between the two sidebands because of their differing propagation velocities in the fiber. When detected on a square-law photodetector this difference leads to an interference pattern between the terms generated by the beating of the upper sideband with the carrier and the lower sideband with the carrier. If these resulting terms are not in-phase this process results in a power reduction in the recovered RF signal and therefore a reduction in its carrier-to-noise ratio. For an unmodulated sinusoidal RF carrier, complete extinction of the received millimeter-wave carrier occurs in a cyclic fashion whenever the phase difference between the lower and the upper sideband is π . Propagation of a three-term signal that results from direct or linear external modulation, through dispersive fiber therefore results in repetitive, link length-dependent nulls in the detected signal power. The first null occurs at a fiber length L meters, given by L=
c 2Dλ2 f 2
(13.2)
where D is the dispersion parameter, c the vacuum velocity of light, λ the mean source wavelength, and f the modulation frequency [14]. As can be seen from Fig. 13.4, at frequencies below 10 GHz propagation over significant lengths of fiber are still possible. However, beyond this, for example propagation of higher
Fig. 13.4 Length-dependent fading of radio-frequency signals over optical fiber
288
John E. Mitchell
microwave or millimeter wave frequencies is severely limited. nm.km systems operating at 1550 nm which corresponds to a dispersion of 17 ps/nm.km in standard single-mode fiber, lengths are restricted of around than 5 km at 28 GHz and less than 2 km at 60 GHz if large power penalties are to be avoided. Hence, it is necessary to consider alternatives, which may mitigate this problem. A large body of work has explored techniques to overcome these length-dependent issues. The solution lies in generating only a two-term signal. Therefore, the resultant RF signal at the photodiode is the product of only the carrier with a single sideband and any dispersion-induced phase shift produces only a relative phase shift of the RF signal which is of no significance in this application. The techniques can be categorized into a number of types: • Two Laser Heterodyning: Here two laser sources are used, spaced by the radiofrequency to be generated. Due to the large laser linewidth compared to the signal bandwidths free-running operation is not possible without significant impairment. To overcome this, techniques to lock the two lasers together are required. These may take the form of optical frequency locked loop (OFLL) [15], optical phase locked loop (OPLL) [16], optical feed forward modulation [17], optical injection locking [18], or optical injection phase-locking [19]. • Homodyne Generation: Although when operated in its linear region the Mach– Zehnder modulator gives a three-term output, modifications to the drive or design can be used to generate two-term signals. For example, if a dual-arm modulator is used with each arm driven with signals 90◦ out of phase with each other a single sideband signal is produced [20]. Optical filtering with a Bragg grating may also be used to remove one of the sidebands of a three-term technique, although this is not very power efficient [21]. A single sideband signal may also be generated by a combination of an external modulator and phase modulator in cascade [22], and a monolithically integrated DFB laser with a multimode interferometer [23]. Another option is to bias a standard Mach–Zehnder modulator at its null point [24] to give a double sideband-suppressed carrier signal. Here, the two sidebands remain with the carrier suppressed. One advantage of this technique is that a drive of only half the required frequency is needed. It is also possible to bias the device at its maximum transmission point to give two tones at four times the frequency plus a carrier [25]. Gratings can be used to filter out either the carrier or one sideband and therefore a selection of one of two frequencies is possible [26]. Although data cannot be imposed with the same modulator as the RF tone in either of these techniques due to the non-linear operation, they do have the advantage that a number of wavelengths can be up-converted using a single high-frequency device [27, 28]. To apply these techniques to a system to deliver radio signals a number of candidate architectures are possible. The different implementations are developed out of a desire to reduce the cost of the remote antenna unit by reducing component count, or by reducing the number of high-frequency components (and in particular highfrequency optical components) needed. However, this usually comes at the expense of some flexibility.
13 Radio-over-Fiber (RoF) Networks
289
Fig. 13.5 RF downstream–RF upstream configuration
The first architecture shown in Fig. 13.5 transmits the signal at its final transmission frequency in both directions. As discussed previously, at low enough frequencies this can be achieved using intensity modulation–direct detection (IM-DD) techniques. At higher frequencies (>10 GHz) techniques such as those discussed above will need to be employed. To remove the need for an optical source at the RAU it is possible to implement a loop back scheme as shown in Fig. 13.6, where two wavelengths are produced at the central station, one carrying the downstream data and one unmodulated. It is here that optical devices capable of both receiving and modulating an optical signal become attractive [2, 3], which leads to RAU simplification but may require dispersion compensation due to the limitation of IM-DD in the upstream (RAU to Central Office) direction. It is also possible for the optical carrier from a modulated downlink signal to be filtered, reused as the optical source on the uplink [29], although this technique requires the use of separate fibers for the downstream and upstream links.
Fig. 13.6 RF downstream with loop-back – RF upstream configuration
As an alternative to RF transmission, schemes have been proposed that transport the radio signal to be transmitted in both directions as an intermediate frequency (IF) rather than at the final transmission frequency. This has typically been proposed by research systems at high microwave or millimetre wave frequencies where an IF of around fIF = 1 GHz is used (i.e., at a frequency where dispersion effects are minimal and direct modulation of lasers is possible) with an RF of perhaps 40 or 60 GHz. In the commercial LGCell system by LGC wireless this technique was used for operation in cellular bands to overcome the bandwidth limits of multimode fiber. To upconvert and downconvert the signals at the RAU a stable carrier frequency around the final transmitted frequency (±fIF ) is required to act as a local oscillator (LO).
290
John E. Mitchell
Fig. 13.7 IF downstream–IF upstream configuration
In the configuration of Fig. 13.7 this is provided locally at the remote site. This has design implication in terms of the requirement of a temperature stable source and also reduces the frequency flexibility of the node. An alternative, shown in Fig. 13.8, is to provide the LO optically from the central office, either at the required frequency or as a sub-harmonic to be used for sub-harmonic injection locking of a oscillator [30] or as a reference for a phase locked loop (PLL) RF oscillator [31]. For a single point to point link, transmitting the LO offers no benefit over the transmission of the RF. However, with multiple nodes it allows the high-frequency components that generate the LO to be shared amongst a number of nodes in the network. An example of this is the architecture developed in [32], where a single 40 GHz local oscillator is distributed around a ring to a number of base transceiver stations that are each fed with an individual optical channel carrying a 2.4 GHz IF signal. For the upstream return channel the LO is also used to downconvert the incoming radio signals, thus allowing the use of low-frequency optical devices.
Fig. 13.8 IF and LO downstream–IF upstream configuration
A hybrid, RF down–IF up approach can also be taken as demonstrated in Fig. 13.9. This uses one of the techniques, described above, for the high-frequency transmission in the downstream direction to the RAU. After reception at the RAU a component of this signal is used as an LO to downconvert the upstream transmission. RF transmission on the downlink enables some of the complexity of the RAU to be eliminated (when compared to IF transmission), while the use of IF in the return channel offers all important cost savings of high-frequency optical components at the RAU. However, the linkage between the up-and downstream frequencies used mandates increased frequency planning and may incur additional constraints.
13 Radio-over-Fiber (RoF) Networks
291
Fig. 13.9 RF downstream–IF upstream configuration
13.3 Radio-Over-Fiber Application Areas So far we have considered the radio-over-fiber system as an analogue transport link. To fully consider the application of this technology we must look at the radio technologies that are to be transported over this network. From the perspective of most radio air-interface designers the inclusion of an optical fiber link into the link budget of the interface is an additional challenge. The already tight noise and distortion budgets offer little flexibility and therefore often require high performance from the optical link. In effect, the radio-over-fiber link designer is being asked to provide a virtually transparent pipe through which a radio signal can be transported. In this section we will give a brief review of some of the radio standards that have been considered for RoF systems. Table 13.1 gives an overview of the most common radio standards that have been considered. Cellular radio systems (GSM and UMTS) have been extensively investigated with commercial offering having been available for some time. On single-mode fiber systems such as BriteCell from Andrew [33] are available, while in the multimode fiber space, primarily for in-building applications, systems available include Zinwave [34], LGCWireless [35], and ADC [36]. Research has typically concentrated on multiple channel operation or considering the extreme performance limits of the systems. For example simulation studies have considered the interaction of second-generation (GSM) and third-generation (W-CDMA) wireless systems [37] or W-CDMA and wireless LAN (WLAN) [38]. In these systems distortion due to the linearity of the optical components is a major consideration as discussed previously with solutions such as predistortion being proposed [39]. As new radio standards appear, different issues arise although most are based on the radio formats’ susceptibility to distortion. For example, ultra-wideband (UWB), an impulse radio standard that uses low power and a very wideband bandwidth (typically >500 MHz), has been developed for short reach systems (see Table 13.1). This bandwidth is significantly greater than other radio standards and introduces a number of new challenges relating to linearity and induced distortion [40, 41], with investigations considering a number of aspects including its co-existence with WLAN [42]. Additionally, optical techniques have also been investigated for the generation of the short pulses that form these signals [43].
292
John E. Mitchell
Table 13.1 Current wireless modulation formats to be support by radio-over-fiber systems Standard
Modulation formats
Bit rate (Max)
Frequencies (Approx)
GSM [46]
Gaussian minimum shift keying (GMSK)
900, 1800 and 1900 MHz
UMTS (3G) [47]
Code division multiple access (CDMA), HPSK (UL), QPSK (DL)
14.4 kbps (GSM), 171.2 kbps (GPRS) 384 kbps
802.11b [48] (WiFi) 802.11a/g (WiFi) [49, 50] 802.11n (WiFi) [51]
Single carrier, BPSK, QPSK, 16QAM, or 64QAM 52 Carrier orthogonal frequency division multiplexing, BPSK, QPSK, 16QAM, or 64QAM 52 Carrier orthogonal frequency division multiplexing, with multiple input, multiple output (MIMO) 256 Carrier OFDM with burst by burst adaptive modulation from QPSK to 64QAM
11 Mbps
2.4 GHz
54 Mbps
5/2.4 GHz
< 300 Mbps
5 GHz and/or 2.4 GHz
< 75 Mbps
2–11 GHz, although 2.3, 2.5, 3.5, and 5 GHz are common 2–11 GHz, see above
802.16-2004 (WiMAX) [52] 802.16-2005 (Mobile WiMAX) [53] UWB MBOFDM [54] LTE (Longterm evolution) [55] DVB-H
LMDS
Scalable orthogonal frequency dision multiple access with up to 2048 carriers and 2 times 2 MIMO Multiple bands of ultra-wideband pulses (> 500MHz bandwidth) OFDM with a maximum of 2048 subcarriers and MIMO
< 75 Mbps
Various 1.8–2.2 GHz
< 480 Mbps
3.1–10.6 GHz
326.4 Mbps
same as UMTS
COFDM with QPSK, 16QAM, 64QAM, broadcast
31.67 Mbps
Mostly based on DOCSIS (CATV) standards
> 100 Mbps
∼ 200 MHz, ∼ 400 MHz, 1.5 GHz 26, 29, and 31 GHz
WiFi (IEEE 802.11a/b/g etc.) systems have been researched widely in part due to the easy availability of access points for experimentation. However, due to the low cost of the access points it is hard to see fiber-based system displacing electronic systems unless it forms part of a wider converged, multi-standard network [44, 45]. As can be seen from Table 13.1, orthogonal frequency division multiplexing (OFDM), which was introduced to the mainstream of communications by the digital video broadcast (DVB) standard and to data communications by the 802.11g standard, forms the mainstay of most of the next-generation communications standards. To date a number of investigations into the transport of OFDM have been made. Theoretical studies have evaluated the performance of OFDM systems in the presence of non-linear distortion from optical components; for example, Rodrigues and O’Reilly [56] used Volterra series analysis to accurately model the effects of linearity on this complex modulation format. A number of experimental
13 Radio-over-Fiber (RoF) Networks
293
investigations have been conducted, although most focus on the transmission of OFDM-based standards such as 802.11a/g [57, 58] and WiMAX [59, 60] or both [61]. All of the studies outlined above show that the transport of radio standards to significant distance using optical fiber is possible, at least at the physical layer. However, other effects can also impact on the system performance. In most experiments a commercial vector signal generator and analyzer setup is used for performance measurements. This offers a controllable signal with high purity to allow repeatable experiments. It does, however, have the drawback of only operating at the physical layer. It must be remembered that the addition of optical fiber into the system can add significant delay into what would normally be the air-interface section of the link. In wide area and centrally controlled architectures, such as GSM or WiMAX, this may not cause significant issues. However, in ad hoc and short-reach architectures, such as 802.11, the addition of 5 μs of delay per kilometer of fiber (10 μs if a round trip is considered) may make a significant contribution to the time-outs of the 802.11 media access control (MAC). It is shown in [62] that for typical implementations a maximum distance of around 13 km of fiber is permissible before the system fails with complete loss of transmission. It should also be noted that these distances are vendor dependent as the standard only specifies a minimum time-out. As shown in Fig. 13.10, the data rate reduces progressively as fiber is added due to the increased delay in the system. We then see that when the time-out (for either acknowledge of packets in the basic access mechanism or for response to a request to send (RTS) packet in the virtual carrier sense or RTS/CTS mechanism) is exceeded due to the fiber delay, transmission falls sharply.
Fig. 13.10 Data throughput of 802.11 (WLAN) signals over optical fiber
294
John E. Mitchell
In this section we have reviewed the main, current radio standards and looked at research that has considered their transmission of optical fibers. In general it is seen that transmission is possible, although in some cases linearization either by design in the components or by system techniques is required.
13.4 Networking Concepts and Techniques As we have seen the use of point-to-point links carrying radio signals over optical fiber have been very well explored, with a wide range of research and commercial offerings available for both indoor and outdoor systems. One of the main current challenges in the area of radio-over-fiber research is to translate these efforts into fully functional and flexible networks. A number of issues are raised in this transition including the design of optimal optical network architectures to appropriate multiplexing schemes. These issues will be introduced here with the reader pointed to current avenues of research in each of the areas.
13.4.1 Media Access Schemes When multiple channels are to be transported over a single optical fiber infrastructure multiple access techniques are required to enable the channel to share the fiber. There are essentially three basic multiple access techniques that can be employed: time division multiple access (TDMA), wavelength division multiple access (WDMA), and sub-carrier multiple access (SCMA). Hybrid techniques combining two or more of the above may also be used. Traditionally in digital communications systems the designer would be at liberty to make an appropriate choice of access scheme with the usual trade-offs of complexity and scalability. In radio-over-fiber networks we have some additional restrictions due to the transparent pipe requirements of the network. It must be remembered that the radio signal being carried is already operating its own access scheme for the air interface. As the RAU has little or no knowledge of the activity of the incoming signal schemes such as TDMA become very difficult, although TDMA based on interleaving sampled version of the signals is possible. The remaining candidates are therefore SCMA and WDMA. With the increased deployment of fiber in the access networks, for example GPON, the overlay of radio links is very attractive. The GPON standards specify wavelength bands for downstream (1490 ± 10 nm), upstream (1310 ± 50 nm), and an additional band for analogue video overlay (1550–1560 nm). Also in the ITU-T G983.3 [63] standard two enhancement bands were proposed; one from 1539 to 1550 nm and a second from 1560 to 1565 nm. However, although wavelength regions are available for radio overlays, no wavelength selection takes place within the distribution networks as only passive power splitting is implemented. This means that all selection must take place at the remote node. Recently, the full service access
13 Radio-over-Fiber (RoF) Networks
295
network (FSAN) next generation access (NGA) task group has mandated the introduction of optical bandpass filters in all GPON equipment to enable the overlay of wavelengths outside the current bands without the need for system upgrades. This was driven by the desire to evolve toward higher baseband line rates but is also equally beneficial for radio-over-fiber deployments.
13.4.2 Dynamic Capacity Allocation To enable full functionality from the radio-over-fiber network the ability to flexibly and dynamically reconfigure capacity is a vital design consideration. This involves the allocation of capacity to locations in the network on the basis of shifting demand. The concentration of resources at the central office reduces the probability of blocking because it reduces the probability that a RAU is out of capacity [64]. If we consider a cellular radio network overlaid on the fiber infrastructure any one of the following modes or combinations there of could be required. • Simulcast mode: A set of RAUs are grouped together to form a single supercell. On the downlink all the RAUs transmit the same signal (albeit with some relative delays), while on the uplink the different RAUs receive and forward the same signal to the head end. The RAUs therefore resemble a distributed antenna system or space diversity system. No mobile station handover takes place as the mobile station is technically still in the same cell. The main issue present is that of the media access which needs to be carefully mapped to the radio standard. At first sight it may seem that in a network, such as 802.11, the radio MAC could also be used at the optical layer as the radio layer allows only one station transmits at any one time. This is possible as long as the RAU can detect RF power being received and shut down its laser when the station is not in the subcell so that multiple return paths and noise addition do not occur [65]. • Static Microcellular mode: The RAUs are partitioned into multiple microcells each transmitting on a separate frequency (or frequency set). Operation of the system in microcellular mode implies the use of sub-carrier multiple access (SCMA) and handovers will occur between microcells. It also implies that each RAU is equipped with electrical filters to select the required radio subcarriers. Hybrid networks employing microcellular techniques with simulcast subsets may also be constructed. • Dynamic Microcellular mode: In this mode complete flexibility is available to map each individual RAU to a required microcellular or simulcast group. Practical issues surround the control of the network transitions between modes of operation. In particular, in simulcast operation, all knowledge of user location within the super-cell is lost making the deconstruction to smaller cells difficult. Dynamic microcellular network operation has the potential of delivering true flexible capacity reconfiguration. For maximum flexibility and low blocking probability tunable optical devices need to be distributed throughout the network [66]. A number of solutions are available with devices in the distribution section and/or at
296
John E. Mitchell
the ends of the network. In this case capacity reconfiguration may also be supported at the WDMA level through optical cross-connects in the networks [67]. The operation of a dynamic microcellular network requires the network to be aware of both the radio layer and the optical layer in order to optimize the capacity availability. It will need to have knowledge of the presence of any wavelength routers in the physical optical infrastructure as well as have direct control of the routing and frequency allocation protocols of the radio access layer to be able to reconfigure individual radio links. This is not a simple optimization task, requiring tight and integrated cross-layer design.
13.4.3 Sharing a Remote Antenna Unit – Wideband RAU Design In principle it is possible to construct a wideband RAU system from 0 to 60 GHz. Photodiodes with a bandwidth of 60 GHz are commercially available, albeit expensive and exhibit a relatively low responsivity. The main limitation is the design of the wideband amplifier in the RAU, although traveling wave amplifiers are promising at high frequencies. Another limitation is the lack of a broadband antenna capable of transmitting all radio signals across multiple bands. Although huge progress has been made over recent years, with technologies such as fractal patch antennas reaching a bandwidth in excess of 30% [68] these are still far from the very broadband technologies that would be required. In the mean time one would have to deploy multiple antennas to cover all the necessary bands.
13.5 Summary The field of radio-over-fiber systems has seen significant research output in recent years, as well as industrial interest and deployment. This chapter has aimed to introduce the basic concepts that surround this technology, the application areas to which it is being applied, and the main challenges that are still to be met. In the space available it has not been possible to review all the work available in the area (an IEEExplore search for radio + fiber lists over 1500 papers), however, it was our intention to point the reader to interesting and significant work in the area from which they can explore this exciting subject that, after a number of years of mainly academic interest, finally seems to be coming of age.
References 1. Allen Telecom’s Radio-over-Fiber Technology Powers Mobile Communications at Sydney 2000 Olympics Fiber Optics Business, Nov. 15, 2000 http://findarticles.com/ p/articles/mi m0IGK/is 21 14/ai 73843087 [Online on March 25, 2008]. 2. D. Wake, D. Johansson, and D. G. Moodie, “Passive Picocell: A New Concept in Wireless Network Infrastructure,” IEE Electronics Letters, vol. 33, no. 5, pp. 404–406, Feb. 1997.
13 Radio-over-Fiber (RoF) Networks
297
3. L. Westbrook, D. G. Moodie, “Simultaneous Bi-directional Analogue Fibre-optic Transmission Using an Electroabsorption Modulator,” IEE Electronics Letters, vol. 32, no. 19, pp. 1806–1807, Sep. 1996. 4. C. P. Liu et al., “High-speed 1.56μm Multiple Quantum Well Asymmetric Fabry-Perot Modulator/Detector (AFPMD) for Radio-over-Fiber Applications,” in Proc., International Topical Meeting Microwave Photonics, 2005. 5. H. Pfrommer et al., “Full-duplex DOCSIS/WirelessDOCSIS Fiber-radio Network Employing Packaged AFPM-based Base Stations,” IEEE Photonics Technology Letters, vol. 18, no. 2, pp. 406–408, Jan. 2006. 6. A. Kaszubowska, P. Anandarajah, and L. P. Barry, “Improved Performance of a Hybrid Radio/Fiber System Using a Directly Modulated Laser Transmitter with External Injection,” IEEE Photonics Technology Letters, vol. 14, no. 2, pp. 233–235, Feb. 2002. 7. K. Asatani, “Nonlinearity and Its Compensation of Semiconductor Laser Diodes for Analog Intensity Modulation Systems,” IEEE Transactions on Communications, vol. 28, no. 2, pp. 297–300, Feb. 1980. 8. H. Lin and Y. Kao, “Nonlinear Distortion and Compensations of DFB Laser Diode in AMVSB Lightwave CATV Applications,” IEEE/OSA Journal of Lightwave Technology, vol. 14, no. 11, pp. 2567–2574, Nov. 1996. 9. G. E. Betts and F. J. O’Donnell, “Optical Analog Link Using a Linearized Modulator,” in Proc., IEEE Lasers and Electro-Optics Society Annual Meeting, vol. 2, pp. 278–279, 1994. 10. T. Ismail, C.-P. Liu, J. E. Mitchell, and A. J. Seeds, “High-Dynamic-Range Wireless-OverFiber Link Using Feedforward Linearization,” IEEE/OSA Journal of Lightwave Technology, vol. 25, no. 11, pp. 3274–3282, Nov. 2007. 11. C. H. Cox, “Analog Optical Links: Theory and Practice,” Cambridge University Press, Cambridge 2004. 12. G. Betts, L. M. Johnson, C. H. Cox, and S. D. Lowney, “High-performance Optical Analog Link Using External Modulator,” IEEE Photonics Technology Letters, vol. 1, no. 11, pp. 404–406, Nov. 1989. 13. C. H. Cox, E. I. Ackerman, G. E. Betts, and J. L. Prince, “Limits on the Performance of RF-over-Fiber Links and their Impact on Device Design,” IEEE Transactions on Microwave Theory and Techniques, vol. 54, no.2, pp. 906–920, Feb. 2006. 14. U. Gliese, S. Norskov, and T. N. Nielsen, “Chromatic Dispersion in Fibre Optic Microwave and Millimetre Wave Links,” IEEE Transactions on Microwave Theory and Techniques, vol. 44, no. 10, pp. 1716–1724, Oct. 1996. 15. S. Kawanishi et al., “Wideband Frequency Measurement of Optical Receivers Using Optical Heterodyne Detection,” IEEE/OSA Journal of Lightwave Technology, vol. 7, no. 1, pp. 1242–1243, Jan. 1989. 16. R. T. Ramos and A. J. Seeds, “Fast Heterodyne Optical Phase-lock Loop Using Double Quantum Well Laser Diodes,” IEE Electronics Letters, vol. 28, no. 1, pp. 82–83, Jan. 1992. 17. R. A. Griffin and K. Kitayama, “Optical Millimetre-wave Generation with High Spectral Purity Using Feed Forward Optical Field Modulation,” IEE Electronics Letters, vol. 34, no. 8, pp. 795–796, April 1998. 18. L. Noel, D. Marcenac, and D. Wake, “120Mbit/s QPSK Radio Fibre Transmission over 100Km of Standard Fibre at 60GHz Using a Master Slave Injection Locked DFB Laser Source,” IEE Electronics Letters, vol. 32, no. 20, pp. 1895–1897, Sep. 1996. 19. L. A. Johansson, D. Wake, and A. J. Seeds, “Millimetre-wave over Fibre Transmission Using a BPSK Reference-modulated Optical Injection Phase-lock Loop,” in Proc., Optical Fiber Communication conference, vol. 3, paper WV3-1, 2001. 20. G. H. Smith, D. Novak, and Z. Ahmed, “Overcoming Chromatic Dispersion Effects in Fiberwireless Systems Incorporating External Modulators,” IEEE Transactions on Microwave Theory and Techniques, vol. 45, no. 8, pp. 1410–1415, Aug. 1997. 21. J. Park, W. V. Sorin, and K. Y. Lau, “Elimination of the Fibre Chromatic Dispersion Penalty on 1550nm Millimetre Wave Optical Transmission,” IEE Electronics Letters, vol. 33, no. 6, pp. 512–513, March 1997.
298
John E. Mitchell
22. J. Conradi, B. Davies, M. Sieben, D. Dodds, and S. Walklin, “Optical Single Sideband (OSSB) Transmission for Dispersion Avoidance and Electrical Dispersion Compensation in Microwave Subcarrier and Baseband Digital Systems,” post deadline paper presented at the Optical Fiber Communications (OFC) conference, Dallas, TX, Feb. 1997. 23. E. Vergnol, J. F. Cadiou, A. Carenco, and C. Kazmierski, “New Modulation Scheme for Integrated Single Side Band Lightwave Source Allowing Fiber Transport up to 256 QAM over 38 GHz Carrier,” in Proc., Optical Fiber Communications (OFC) conference, vol. 4, pp. 134–136, 2000. 24. J. J. O’Reilly, P. M. Lane, R. Heidermann, and R. Hofstetter, “Optical Generation of Very Narrow Linewidth Millimetre Wave Signals,” IEE Electronics Letters, vol. 28, no. 25, pp. 2309–2311, Dec. 1992. 25. J. J. O’Reilly and P. M. Lane, “Fibre Supported Optical Generation and Delivery of 60GHz Signals,” IEE Electronics Letters, vol. 30, no. 16, pp. 1329–1330, Aug. 1994. 26. J. E. Mitchell, “Simultaneous Up-conversion of Multiple Wavelengths to 18GHz and 36GHz Using 4-f Technique and Optical Filtering,” in Proc., International Topical Meeting on Microwave Photonics, paper W.4.3, 2006. 27. M. Sauer, K. Kojucharow, H. Kaluzni, D. Sommer, and W. Nowak, “Simultaneous Electrooptical Upconversion to 60 GHz of Uncoded OFDM Signals,” in Proc., International Topical Meeting on Microwave Photonics, pp. 219–222, 2006. 28. R. A. Griffin, P. M. Lane, and J. J. O’Reilly, “Radio over Fibre Distribution Using an Optical Millimetre Wave/DWDM Overlay,” in Proc., Optical Fiber Communications (OFC) conference, paper WD6, 1999. 29. A. Nirmalathas, C. Lim, D. Novak, and R. Waterhouse, “Optical Interfaces without Light Sources for Base Station Designs in Fiber Wireless Systems Incorporating WDM,” in Proc., International Topical Meeting on Microwave Photonics, pp. 119–122, 1999. 30. G. H. Smith and D. Novak, “Broadband Millimetre Wave Fibre Radio Network Incorporating Remote Up/downconversion,” in Proc., IEEE MTT-S International Microwave Symposium Digest, vol. 3, pp. 1509–1512, 1998. 31. R. A. Griffin, H. M. Salgado, P. M. Lane, and J. J. O’Reilly, “System Capacity for Millimeter Wave Radio over Fiber Distribution Employing an Optically Supported PLL,” IEEE/OSA Journal of Lightwave Technology, vol. 17, no. 12, pp. 2480–2487, Dec. 1999. 32. T. Ismail, C.-P. Liu, J. E. Mitchell, A. J. Seeds, X. Qian, A. Wonfor, R. V. Penty, and I. H. White, “Transmission of 37.6 GHz QPSK Wireless Data Over 12.8-km Fiber with Remote Millimeter-Wave Local Oscillator Delivery Using a Bi-directional SOA in a Full-duplex System with 2.2-km CWDM Fiber Ring Architecture,” IEEE Photonics Technology Letters, vol. 17, no. 9, pp. 1989–1991, Sept. 2005. 33. http://www.andrew.com 34. http://www.zinwave.com 35. http://www.lgcwireless.com 36. http://www.adc.com 37. R. E. Schuh, “Hybrid Fiber Radio for Second and Third Generation Wireless Systems,” in Proc., International Topical Meeting on Microwave Photonics, pp. 213–216, 1999. 38. R. Yuen and X. N. Fernando, “Enhanced Wireless Hotspot Downlink Supporting IEEE 802.11 and WCDMA,” in Proc., IEEE International Symposium on Personal, Indoor Mobile Radio Communications, pp. 1–6, 2006. 39. L. Roselli, V. Borgioni, F. Zepparelli, F. Ambrosi, M. Comez, P. Faccin, and A. Casini, “Analog Laser Predistortion for Multiservice Radio-over-Fiber Systems,” IEEE/OSA Journal of Lightwave Technology, vol. 21, no. 5, pp. 1211–1223, May 2003. 40. Y. Le Guennec, M. Lourdiane, B. Cabon, G. Maury, and P. Lombard, “Technologies for UWB-Over-Fiber,” in Proc., IEEE Lasers & Electro-Optics Society Annual Meeting, pp. 518–519, 2006. 41. M. L. Yee, V. H. Pham, Y. X. Guo, L. C. Ong, and B. Luo, “Performance Evaluation of MB-OFDM Ultra-Wideband Signals over Single Mode Fiber,” in Proc., IEEE International Conference on Ultra-Wideband, pp. 674–677, 2007.
13 Radio-over-Fiber (RoF) Networks
299
42. C. K. Sim, M. L. Yee, B. Luo, L. C. Ong, and M. Y. W. Chia, “Performance Evaluation for Wireless LAN, Ethernet and UWB Coexistence on Hybrid Radio-over-Fiber Picocells,” in Proc., Optical Fiber Communication (OFC) conference, 2005. 43. C. Wang, F. Zeng, and J. Yao, “All-Fiber Ultrawideband Pulse Generation based on Spectral Shaping and Dispersion-Induced Frequency-to-Time Conversion,” IEEE Photonics Technology Letters, vol. 19, no. 3, pp. 137–139, Feb. 2007. 44. M. L. Yee, H. L. Chung, P. K. Tang, L. C. Ong, B. Luo, M. T. Zhou, Z. Shao, and M. Fujise, “Radio-over-Fiber EVM Measurements for IEEE 802.11g WLAN and Cellular Signal Distribution,” in Proc., European Microwave Conference, pp. 882–885, 2006. 45. M. J. Crisp, L. Shen, A. Wonfor, R. V. Penty, and I. H. White, “Demonstration of a Radio over Fibre Distributed Antenna Network for Combined In-building WLAN and 3G Coverage,” in Proc., Optical Fiber Communication (OFC) conference, pp. 1–3, 2007. 46. S. M. Redl, M. K. Weber, and M. W. Oliphant, “GSM and Personal Communications Handbook,” Artech House, 1998. 47. The 3rd Generation Partnership Project (3GPP), http://www.3gpp.org/ 48. IEEE Std 802.11b-1999 (R2003) Higher-Speed Physical Layer Extension in the 2.4 GHz Band, http://standards.ieee.org/getieee802/download/802.11b-1999.pdf [Online on March 25, 2008]. 49. IEEE Std 802.11a-1999 High-speed Physical Layer in the 5 GHz band, http:// standards.ieee.org/getieee802/download/802.11a-1999.pdf [Online on March 25, 2008]. 50. IEEE Std 802.11g-2003: Further Higher Data Rate Extension in the 2.4 GHz Band, http://standards.ieee.org/getieee802/download/802.11g-2003.pdf [Online on March 25, 2008]. 51. IEEE 802.11 Task Group N, Project Status Reports, http:grouper.ieee.org/groups/802/ 11/Reports/tgn update.htm [Online on March 25, 2008]. 52. IEEE Std 802.16-2004 Air Interface for Fixed Broadband Wireless Access Systems http://standards.ieee.org/getieee802/download/802.16-2004.pdf [Online on March 25, 2008]. 53. IEEE Std 802.16e-2005 Air Interface for Fixed Broadband Wireless Access Systems – Amendment 2, http://standards.ieee.org/getieee802/download/802.16e-2005.pdf [Online on March 25, 2008]. 54. WiMedia Alliance, http://www.wimedia.org [Online on March 25, 2008]. 55. UTRA-UTRAN Long Term Evolution (LTE), http://www.3gpp.org/Highlights/ LTE/LTE.htm [Online on March 25, 2008]. 56. M. R. D. Rodrigues and J. J. O’Reilly, “An Analytic Technique to Assess the Impact of Nonlinearities on the Error Probability of OFDM Signals in RoF based Wireless Networks,” in Proc., IEEE International Symposium on Information Theory, p. 316, 2001. 57. I. Kostko, M. E. M. Pasandi, M. M. Sisto, S. LaRochelle, L. A. Rusch, and D. V. Plant, “A Radio-over-Fiber Link for OFDM Transmission without RF Amplification,” in Proc., Annual Meeting of the IEEE Lasers and Electro-Optics Society, 2007. 58. M. Sauer, A. Kobyakov, and A. B. Rubin, “Radio-over-Fiber Transmission with Mitigated Stimulated Brillouin Scattering,” IEEE Photonics Technology Letters, vol. 19, no. 19, pp. 1487–1489, Oct. 2007. 59. J. E. Mitchell, “Performance of OFDM at 5.8 GHz Using Radio over Fibre Link,” IEE Electronics Letters, vol. 40, no. 21, pp. 1353–1354, Oct. 2004. 60. J. B. Song and A. H. M. R. Islam, “Distortion of OFDM Signals on Radio-over-Fiber Links Integrated with an RF Amplifier and Active/Passive Electroabsorption Modulators,” IEEE/OSA Journal of Lightwave Technology, vol. 26, no. 5, pp. 467–477, March 2008. 61. T. Ismail, C.-P. Liu, J. E. Mitchell, and A. J. Seeds, “Feed-forward LInearised Uncooled DFB Laser in a Multi-channel Broadband Wireless over Fibre Transmission at 5.8 GHz,” in Proc., International Topical Meeting on Microwave Photonics, pp. 115–118, 2005. 62. B. Kalantarisabet and J. E. Mitchell, “MAC Constraints on the Distribution of 802.11 Using Optical Fibre,” in Proc., European Conference on Wireless Technology, pp. 238–240, 2006. 63. ITU-T Recommendation G.983.3 – A broadband optical access system with increased service capability by wavelength allocation.
300
John E. Mitchell
64. A. M. J. Koonen et al., “Re-configurable Broadband Fibre Wireless Network Employing Dynamic Wavelength Allocation,” in Proc., European Conference on Optical Communication, pp. 577–578, 1998. 65. M. Sauer, A. Kobyakov, and J. George, “Radio Over Fiber for Picocellular Network Architectures,” IEEE/OSA Journal of Lightwave Technology, vol. 25, no. 11, pp. 3301–3320, Nov. 2007. 66. J. C. Attard and J. E. Mitchell, “Optical Network Architectures for Dynamic Reconfiguration of Full Duplex, Multiwavelength, Radio Over Fiber,” OSA Journal of Optical Networking, vol. 5, no. 6, pp. 435–444, June 2006. 67. J. J. Vegas Olmos, T. Kuri, and K. Kitayama, “Dynamic Reconfigurable WDM 60-GHz Millimeter-Waveband Radio-Over-Fiber Access Network: Architectural Considerations and Experiment,” IEEE/OSA Journal of Lightwave Technology, vol. 25, no. 11, pp. 3374–3380, Nov. 2007. 68. J. J. Huang, F. Q. Shan, and J. Z. She, “A Novel Multiband and Broadband Fractal Patch Antenna,” in Proc., Progress in Electromagnetics Research Symposium, 2006.
Chapter 14
Integration of EPON and WiMAX Gangxiang Shen and Rodney S. Tucker
Abstract The integration of EPON and WiMAX is a novel research topic that has received extensive interest from both industry and academia. The major motivations behind the integration of EPON and WiMAX involve the potential benefits of fixed mobile convergence (FMC), which uses a single network infrastructure to provide both wired and wireless access services, and a good match of capacity hierarchy between EPON and WiMAX by using EPON as a backhaul (or feeder) to connect multiple disperse WiMAX base stations. This chapter recaps recent progress in the area of integration of EPON and WiMAX. Three different integration architectures, including independent architectures, hybrid architectures, and microwave-over-fiber (MoF) architectures, are described. Based on these architectures, a range of planning and operational issues are discussed, including optimal passive optical network deployment to connect disperse WiMAX base stations, packet forwarding, bandwidth allocation and QoS support, handover operation for mobile users, survivability, and cooperative downstream transmission for broadcast services. We hope that this will interest readers and stimulate further investigation in the area.
14.1 Introduction The Internet today is characterized by a fast growth of bandwidth-intensive services, such as IPTV, video on demand (VoD), and peer-to-peer (P2P) services. To Gangxiang Shen ARC Special Research Centre for Ultra-Broadband Information Networks (CUBIN), Department Electrical and Electronic Engineering, The University of Melbourne, Melbourne VIC 3010, Australia, e-mail:
[email protected] Rodney S. Tucker ARC Special Research Centre for Ultra-Broadband Information Networks (CUBIN), Department Electrical and Electronic Engineering, The University of Melbourne, Melbourne VIC 3010, Australia, e-mail:
[email protected]
A. Shami et al. (eds.), Broadband Access Networks, Optical Networks, c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-92131-0 14,
301
302
Gangxiang Shen and Rodney S. Tucker
cater to the bandwidth demand of these types of services, access networks are evolving from traditional copper-based ADSL technology to more advanced fiber-based passive optical network technologies [1–3], including fiber to the home (FTTH), fiber to the node (FTTN), fiber to the curb (FTTC). Nowadays, there are two major passive optical network standards, including Ethernet passive optical networks (EPON) [1, 2] and Gigabit-capable Passive Optical Networks (GPON) [3]. EPON is based on the traditional Ethernet techniques, while GPON evolved from the traditional broadband passive optical network (BPON) technology. Meanwhile, in the wireless camp, we also see fast progresses of wireless access technologies, which are evolving from the traditional WiFi technology to more advanced technologies such as WiMAX [4–7] and long-term evolution (LTE) [8] technologies for wider cell coverage and higher access capacity. Driven by potentially significant cost reduction by converging network infrastructures and control systems of wired and wireless access networks, fixed mobile convergence (FMC) [9] is envisioned as a future generation of architecture for broadband access. Motivated by the potential benefits of fixed mobile convergence, this chapter focuses on integration of EPON and WiMAX. In this chapter, we specifically look at integration architectures and their related design and operations. Three types of integration architectures [10, 11] are considered, including (i) independent architectures, which represent the most intuitive way to interconnect an EPON optical network unit (ONU) and a WiMAX base station (BS) over a standardized interface, (ii) more advanced hybrid architectures that are based on an integrated ONU-BS device box, and (iii) microwave-over-fiber (MoF) architectures, which modulate WiMAX carrier signals onto an optical carrier and transmit the carrier signals over fibers together with the EPON baseband signal. Based on these architectures, a range of network design and operational issues are discussed, including (i) passive optical network deployment to optimally interconnect disperse WiMAX base stations to a Central Office (CO), (ii) upstream packet forwarding, bandwidth allocation, and QoS support, (iii) handover operation, (iv) survivability, and (v) downstream cooperative communication for broadcast services. Although the whole framework focuses on EPON and WiMAX, it is extensible to support other types of wired and wireless network combinations, such as GPON [3] and WiMAX, EPON and WiFi, and EPON and LTE. The rest of this chapter is organized as follows. Section 14.2 is a literature survey on integration of optical and wireless access networks. Section 14.3 introduces three integration architectures for EPON and WiMAX. Based on the integration architectures, Section 14.4 discusses related research issues from the perspectives of network design and operation. We conclude the chapter in Section 14.5.
14.2 Literature Survey The integration of EPON and WiMAX is a new topic in the literature. However, it has received extensive interest. Preliminary research on integration of EPON and WiMAX was described in [12, 13], under the banner of hybrid optical wireless
14 Integration of EPON and WiMAX
303
networks. The hybrid networks in [12, 13] employed a passive optical network as a backhaul to transmit data for a large WiMAX network, in which a single WiMAX base station collocated with an optical line terminal (OLT) functions as a central controller and coordinator to perform operations for the whole WiMAX network, including packet forwarding, bandwidth allocation, etc. To fully exploit the benefits of integration of EPON and WiMAX, more integration architectures were proposed in [10, 11]. These architectures include independent architectures, hybrid architectures, unified connection-oriented architectures, and microwave-over-fiber (MoF) architectures. Based on these architectures, a variety of design and operational issues including network design and planning, packet forwarding, bandwidth allocation and QoS support, handover operation, and network survivability were discussed in [10, 11]. Other types of optical wireless access networks were also proposed in [14, 15]. In addition to directly over an ONU as in the architectures proposed in [10, 12], a wireless BS, under the architectures in [14, 15], can also send data to gateways/ONUs over other intermediate wireless BSs by taking advantage of wireless mesh networking. For this type of optical wireless networks, research efforts were mainly concentrated on the placement of wireless BSs [14], and routing algorithms and load balancing for packet forwarding in wireless mesh networks [14, 15].
14.3 Integrated Architectures for EPON and WiMAX In [10, 11], we proposed four integration architectures for EPON and WiMAX. This chapter recaps three of them, including (i) independent architectures, (ii) hybrid architectures, and (iii) microwave-over-fiber (MoF) architectures. For the fourth type of architecture, i.e., unified connection-oriented architectures, interesting readers may refer to [10, 11].
14.3.1 Independent Architectures Independent architectures are the most intuitive case for integration of EPON and WiMAX. Figure 14.1 shows an example of this type of architecture, in which multiple ONUs are connected to a common OLT over a tree passive optical network. To in-terconnect the EPON and WiMAX network segments, the architectures connect a WiMAX base station (BS) to an EPON Optical Network Unit (ONU) over a standardized interface, e.g., an Ethernet interface. In this configuration, the WiMAX BS is a generic user of the ONU, and data from users are processed regularly in both of the devices. In the upstream direction, data from a wireless subscribed station (SS) are forwarded to the WiMAX BS and its connected ONU. The ONU further forward the data over the passive optical network to an OLT, located at a Central Office (CO). In contrast, in the downstream direction, a reverse packet forwarding
304
Gangxiang Shen and Rodney S. Tucker
c Fig. 14.1 Architectures for integration of EPON and WiMAX (adapted from [10], 2007 IEEE)
process is performed to forward data from a central OLT to an intermediate ONU, through a WiMAX BS, and eventually to a targeted SS. The independent architectures have the advantage of incorporating standardized devices and interfaces. However, the physical boundary between a WiMAX BS and an EPON ONU may make it difficult to directly exchange control information between the two devices to quickly notify network state changes such as bandwidth allocation and packet scheduling in both upstream and downstream directions. Thus, the architectures may not fully exploit the potential benefits of integration of EPON and WiMAX. Moreover, two independent devices, i.e., a WiMAX BS and an EPON ONU, can be another disadvantage for the architectures, which are likely more expensive than a single integrated device as in other architectures.
14.3.2 Hybrid Architectures To overcome the disadvantage of two device boxes in the independent architectures, as well as to provide better efficiency and flexibility of bandwidth allocation and packet scheduling in the integrated EPON and WiMAX systems, another type of integrated architecture, as shown in the bottom cell in Fig. 14.1, can be proposed, called hybrid architecture. The architectures integrate an EPON ONU and a WiMAX BS into a single system box, called ONU-BS. The new box realizes all the functionalities of both EPON ONU and WiMAX BS. Moreover, because ONU and WiMAX BS are integrated in a single device, conveniences are provided to promptly exchange network state information in both EPON and WiMAX network segments. Figure 14.2 shows the ASIC chips contained in the ONU-BS and the layout of the functional modules supported by each of the chips. In hardware as shown in the
14 Integration of EPON and WiMAX
305
c Fig. 14.2 Functional modules and components of ONU-BS (adapted from [10], 2007 IEEE)
upper part of the figure, the integrated ONU-BS composes three key ASIC chips. The first chip (i.e., ASIC-1) is in charge of the communications in the EPON segment, the third chip (i.e., ASIC-3) is in charge of the communications in the wireless segment, and the middle chip (i.e., ASIC-2) is responsible for the overall control and coordination of data communications between the EPON and WiMAX network segments. For cost saving, the three chips could be further integrated into a single large ASIC chip. The lower part of the figure shows the functional modules of the ASIC chips, which mainly include the modules for upstream data transmission. The modules of ASIC-1 and ASIC-3 are mainly related to the EPON and WiMAX network segments, respectively. The module ONU-BS central controller, which corresponds to ASIC-2, provides overall control and coordination functionalities for the whole integrated device. Hybrid architectures can provide better flexibility and efficiency in bandwidth allocation and capacity utilization than independent architectures. This is because the integrated ONU-BS understands the network state information in both EPON and WiMAX network segments, and it can therefore make the most efficient bandwidth allocation and packet scheduling for the whole integrated system. For example, for upstream data transmission because the ONU-BS understands how much upstream bandwidth has been allocated to associated wireless SSs in a WiMAX segment, though not receiving user data from the SSs yet, the ONU-BS can ask for more
306
Gangxiang Shen and Rodney S. Tucker
bandwidth for the future user data from the SSs, when making an upstream bandwidth request in the EPON segment. By doing this, a shorter average packet delay and a better system throughput can be expected.
14.3.3 Microwave-over-Fiber Architectures To better utilize the fiber spectrum (in the previous architectures, both wired and wireless user data are transmitted over the fiber baseband), the third type of integrated architectures is based on microwave-over-fiber (MoF) technology [16], called microwave-over-fiber (MoF) architectures. Under this type of architecture, one can employ either TDM PON or WDM PON [17] for the integration. Specifically, under TDM PON, carrier signals from different WiMAX base stations are modulated onto a common optical carrier together with a baseband signal that carries EPON user data. To distinguish the wireless carrier frequencies, before modulation, frequency shifts are required to ensure that each WiMAX base station corresponds to a certain unique optical subcarrier in the PON system. Further details about this type of MoF architecture (based on TDM PON) can be found in [10]. Using a WDM PON arrangement, the integration does not need to distinguish the wireless carrier frequencies modulated within the PON network, as each wireless carrier signal (for each WiMAX BS) is modulated on a different wavelength in the WDM PON. Figure 14.3 shows an example for integrated MoF architectures using WDM PON technology [11]. The WDM PON carries 16 wavelengths and each wavelength carries a baseband signal for WDM PON and an optical subcarrier signal for WiMAX. The layout of carrier signals in the integrated architecture is also illustrated in Fig. 14.3. The hardware in each remote node is made up of an ONU and a dumb antenna. The ONU is responsible for data communication of the WDM PON, and the dumb antenna relays an analog WiMAX signal from and to its associated micro-cell. The
Fig. 14.3 Microwave-over-fiber (MoF) integration architectures and carrier signal spectrum layout using WDM PON (adapted from [11])
14 Integration of EPON and WiMAX
307
user data of the WDM PON are transmitted in the baseband, which occupies 1.25 GHz bandwidth based on 1 Gb/s EPON data rate, and the user data of WiMAX are modulated on a wireless carrier frequency (e.g., 2.5 GHz). The two signals are multiplexed and modulated onto a common optical carrier frequency and transmitted to a central OLT node. The central node in the MoF architectures comprises two major modules: an array of OLTs and a central WiMAX BS as shown in Fig. 14.3. In the example, there are a total of 16 OLTs in the WDM PON and 16 WiMAX-BS units that, together with a macro-BS central controller/coordinator, make up a WiMAX macroBS. When a modulated optical signal enters the central node, the signal is first converted into an electronic format, and then the latter is demultiplexed into a baseband signal and a WiMAX signal. The baseband signal is further forwarded to its corresponding OLT for data processing, while the WiMAX signal is forwarded to its corresponding WiMAX-BS unit for data processing. Major advantages of the MoF architectures are as follows. First, it can provide efficient optical spectrum utilization as different optical spectrum frequencies are employed to transmit EPON and WiMAX user data. Second, the centralized feature of the WiMAX networks enables efficient bandwidth allocation and packet scheduling and forwarding among different WiMAX micro-cells. Third, the centralized WiMAX macro-BS can provide convenience in handover operation for mobile users. However, as shortcomings, the MoF architectures may suffer from the following disadvantages. First, despite efficiency of centralized control and operation in bandwidth utilization and handover operation, the central macro-BS can become a bottleneck of the whole system as it needs to process all the data packets and bandwidth requests from all the subscribed stations (SSs) (e.g., hundreds of SSs). In addition, the central macro-BS is critical, which can cause the whole system to be out of service if it incurs a failure. Second, with the growing maturity of the highfrequency (e.g., even up to 60 GHz [18]) CMOS technology, the MoF architectures may show disadvantages of higher system cost compared to other integrated architectures such as the hybrid architectures, which employ the high-frequency CMOS technology at the boundary of EPON and WiMAX networks.
14.3.4 Multistage EPON and WiMAX Integration Based on the previous two-stage architectures, we can further extend the integration of EPON and WiMAX. A four-stage integrated EPON and WiMAX system is shown in Fig. 14.4. Specifically, with the standardization of 10G EPON technology [19], it is possible to employ a 10G EPON to function as a backhaul to interconnect multiple EPON OLTs. Meanwhile, the WiMAX standard [4, 6] enables to establish point-to-point radio links between neighboring WiMAX BSs, thereby allowing for establishing a WiMAX star network as shown in Fig. 14.4. In the WiMAX star network, a WiMAX base station connected to an EPON ONU functions as a root node
308
Gangxiang Shen and Rodney S. Tucker
c Fig. 14.4 Multistage-integrated EPON and WiMAX systems (adapted from [11], 2007 SPIE)
to relay data from all the other neighboring WiMAX base stations. The WiMAX star network can be cost-effective for the network scenarios in rural areas, where users are disperse and aggregate user bandwidth is low. Multistage architectures can cover much larger user service areas than the previous two-stage architectures. For example, if both of the 10G EPON and 1G EPON are equipped with 1:16 optical splitters and the star WiMAX network composes seven WiMAX micro-cells as shown in Fig. 14.4. The whole multistage architecture can cover a huge area, made up of a total of 1792 micro-cells. Moreover, due to the huge service area with numerous users, the benefit of statistical traffic multiplexing can be exploited to maximally improve the capacity utilization of the integrated system. In addition, the multistage architectures also provide convenience in handover operation for mobile users. We will give more detail on the handover operation later in Section 14.4. Finally, multistage architectures allow heterogeneousness of remote nodes. In addition to a WiMAX base station, a digital subscriber line access multiplexer (DSLAM) can be connected to an EPON ONU for DSL services. Also, rather than a 1G EPON ONU, a 10G EPON ONU can be directly connected to a WiMAX base station or a wired DSLAM for even higher bandwidth as shown in the bottom wireless cell in Fig. 14.4.
14.4 Design and Operation Issues Based on the above integrated architectures for EPON and WiMAX, a range of research issues emerge as one looks for a better understanding of the technology. In the following, we discuss these research issues.
14 Integration of EPON and WiMAX
309
14.4.1 Optimal Passive Optical Network Deployment Efficient planning and deployment of the physical layout of integrated EPON and WiMAX systems is an important issue. For this, we need to consider various costs, including hardware costs and operational costs. Hardware costs include the cost of EPON deployment and the cost of WiMAX BSs. In general, the cost of EPON deployment is much higher than that of WiMAX BSs due to the dominance of labor cost for laying fibers. Thus, minimizing the total cost for EPON deployment is a key optimization problem. In the context of integrated architectures, the problem of optimal EPON deployment can be defined as given a set of disperse WiMAX BSs, minimize the total deployment cost for EPON networks that connect all the WiMAX base stations (BS) to a Central Office (CO), subject to a set of EPON system constraints, including (i) maximal transmission distance, (ii) maximal differential distance, and (iii) optical split ratio. According to the EPON standard, the typical value for the maximal transmission distance between an ONU and an OLT is 20 km [1]. The maximal differential distance, which is defined as the maximal distance difference from different ONUs to a central OLT within a common EPON, is 20 km [1]. Finally, EPON can support a range of optical split ratios such as 1:16 and 1:32. For the above research problem, we have developed optimization models and efficient heuristics to find solutions for EPON deployment without considering geographic constraints such as road maps. The heuristics are scalable to find suboptimal solutions for network planning scenarios with hundreds of, or even up to several thousand ONUs. Interested readers may refer to our papers [20, 21]. Subsequent research is required to consider optimal PON deployment taking the geographic constraints such as road maps into consideration.
14.4.2 Packet Forwarding Under the integrated EPON and WiMAX architectures, there are two operational modes for user packet forwarding. User packets can be forwarded in either IP layer or MAC layer. As shown in Fig. 14.5, the IP layer forwarding mode does not require intermediate ONU-BSs to have any packet switching capability. All the user packets are forwarded in the upstream direction to a central OLT and then the latter sends the packets to an edge access IP router. The router makes routing decisions to find where the packets should be further sent, either to the public Internet or downstream back to another local SS. Such a packet forwarding model requires a simple intermediate ONU-BS. Moreover, because each user data packet goes through an edge IP router, better network security can be expected. However, as a shortcoming, we can see that the forwarding mode suffers from a bandwidth waste due to the loopback of the packets forwarded between the SSs that are subscribed to a common ONU-BS as shown in Fig. 14.5.
310
Gangxiang Shen and Rodney S. Tucker
Fig. 14.5 IP layer packet forwarding in integrated EPON and WiMAX networks (adapted c from [10], 2007 IEEE)
The MAC layer forwarding mode is an alternative to the IP layer mode, which can avoid the bandwidth waste due to the loopback of packet forwarding. This is at the cost of an extra MAC layer switch by each intermediate ONU-BS as shown in Fig. 14.6. Specifically, user packets are first encapsulated into Ethernet frames in the WiMAX network. Then the Ethernet frames are forwarded to an intermediate ONUBS. An Ethernet switch attached to the ONU-BS then directly switches the frames based on their MAC addresses. By doing this, the packets do not experience a loopback forwarding process. Under the MAC layer switching mode, the IEEE 802.3D STP protocol [22] can be employed to establish a simple minimum spanning tree (MST) for the whole integrated system with the Ethernet switch attached to an OLT as a root node. In addition, the virtual LAN (VLAN) (IEEE 802.3Q) protocol [23] can also be deployed for better network security.
Fig. 14.6 MAC layer frame forwarding in integrated EPON and WiMAX networks (adapted c from [10], 2007 IEEE)
14 Integration of EPON and WiMAX
311
14.4.3 Bandwidth Allocation Bandwidth allocation is important for both EPON [2, 24] and WiMAX [25] networks. EPON and WiMAX show similarity in bandwidth allocation in both downstream and upstream directions. In the downstream direction, the communications in both EPON and WiMAX follow the point-to-multipoint mode; that is, user data are sent from a central point, such as an EPON OLT and a WiMAX BS, to multiple disperse subscribe stations, such as EPON ONUs and WiMAX SSs. Under this type of transmission mode, no network resource competition or transmission collision occurs because when a central node transmits data to one of its subscribed users, the downstream channel is always uniquely reserved for the transmission. In contrast, bandwidth allocation in the upstream direction is generally more complicated. This is because the transmission in the upstream direction follows the multipoint-to-point mode, under which collisions can occur when multiple subscribed users send data in the upstream direction simultaneously to the central node. In order to avoid the collisions, both EPON and WiMAX have developed a set of mechanisms specially for the upstream bandwidth allocation [1, 4]. A generic poll/request/grant process is a typical example of this type of bandwidth allocation mechanisms. For example, in EPON networks, a central OLT first sends a polling message to each of the subscribed ONUs to check if the ONU has any data to transmit in the upstream direction in the next transmission cycle. The ONUs respond with bandwidth request messages based on their currently accumulated user data stored in their local buffers. Upon receiving such bandwidth request messages, the central OLT makes an optimal time-slot allocation for the different ONUs by taking into consideration of various optimization aspects such as maximizing system capacity utilization, minimizing overall system packet delay, and ensuring the QoS requirement of each ONU user service. Then the OLT notifies each of the ONUs on when they can start transmitting data in the upstream direction in the next transmission cycle. Similar bandwidth allocation mechanisms are devised for the WiMAX networks, in which a central WiMAX BS plays a central role like an OLT in an EPON network. The WiMAX BS polls each of subscribed stations (SSs) enquiring whether they have data to transmit in the upstream direction in the next transmission cycle. The SSs respond with their bandwidth requests and then the central WiMAX BS makes an optimal bandwidth allocation among the different SSs. Moreover, in addition to the above poll/request/grant mechanisms, the WiMAX technology also supports other bandwidth allocation modes such as unsolicited bandwidth allocation [25], which does not need a bandwidth request message from an SS to a central WiMAX BS. The similarities in bandwidth allocation in the upstream direction provide convenience and opportunities of performance optimization for overall bandwidth allocation within the integrated systems. Such advantages can be fully exploited when the hybrid integrated architectures are implemented. Figures 14.7 and 14.8 show examples of upstream data transmission under the independent architectures and the hybrid architectures, respectively, which illustrate how the hybrid architectures can
312
Gangxiang Shen and Rodney S. Tucker
Fig. 14.7 Upstream bandwidth allocation in independent architectures
Fig. 14.8 Upstream bandwidth allocation in hybrid architectures
improve average packet delay in the upstream direction compared to the independent architectures. Figure 14.7 shows an example of upstream bandwidth allocation for the independent architectures. There are several sequential and parallel steps in the example. A polling message is first sent from an OLT to an ONU in Step 1. Meanwhile, a WiMAX BS connected to the ONU grants y units of bandwidth to its SS users in Step 2. Upon receiving the polling message, the ONU responds with a bandwidth request message asking for x units of bandwidth for upstream transmission in Step 3. Under the independent architectures, the ONU has no idea on how much bandwidth that the WiMAX BS has granted for its SS users. Thus, the ONU requests the x-unit bandwidth simply based on the accumulated user data in the ONU local buffer as shown in Fig. 14.7. In Step 4, the WiMAX SS users transmit data in the upstream direction to the BS, and the latter forward the data directly to the ONU over the ONU-BS interface. Upon receiving the data, the ONU stores the data into its local buffer. Now, as shown in Fig. 14.7, after Step 4 there are a total of x+y units of user data stored in the ONU local buffer. In the EPON segment, upon receiving
14 Integration of EPON and WiMAX
313
the bandwidth request message from the ONU, the OLT grants x units of bandwidth for the ONU to transmit data in the upstream direction in Step 5. The grant message triggers the ONU to send x units of data in the upstream direction in Step 6, with y units of data remaining in the buffer. In contrast, Fig. 14.8 shows an example of upstream bandwidth allocation under the hybrid architectures. Due to the integration of ONU and WiMAX BS, the ONUBS understands how much bandwidth has been granted for WiMAX SS users in Step 2. The ONU-BS can predict that y units of user data will arrive in the near future from the SS users in Step 4. Thus, when requesting for upstream bandwidth from the OLT, the ONU-BS can ask for y more units of bandwidth, i.e., a total of x+y units of bandwidth, in Step 3. Upon this request, if the OLT accordingly grants x+y units of bandwidth for upstream transmission in Step 5, then in Step 6, x+y units of user data are transmitted without any data remaining in the ONU-BS local buffer. It is clear that the hybrid architectures can transmit upstream user data faster and more efficiently, which means a shorter average packet delay and a better system throughput.
14.4.4 QoS Support and Mapping EPON and WiMAX have similar QoS support, which facilitates QoS support in integrated systems. EPON classifies user packets into up to eight priority levels (i.e., eight priority queues) [1]. Likewise, WiMAX classifies user packets into up to five different QoS levels, including unsolicited grant services (UGS), extended real-time polling services (ertPS), real-time polling services (rtPS), non-real-time polling services (nrtPS), and best effort services (BE) [25]. To enable integration, an effective mapping mechanism is required between EPON priority queues and WiMAX different levels of QoS. The mapping should decide which WiMAX flow packets should be stored in which EPON priority queues for equivalent QoS levels. In addition, because EPON classifies packets based on priority queues, it actually supports QoS in a differentiated services (DiffServ) mode [26]. In contrast, because WiMAX supports QoS based on connection-oriented service flows, it actually follows an integrated services (IntServ) mode [27]. When connecting these two types of access networks in an integrated system, it is interesting to understand how QoS-level conversion can be carried out between DiffServ services and IntServ services. Also, it is interesting to understand how end-to-end QoS can be supported in the integrated systems.
14.4.5 Handover Operation User mobility is an important feature that should be supported by integrated EPON and WiMAX systems. To enable user mobility, integrated systems should be able
314
Gangxiang Shen and Rodney S. Tucker
to handle handover operation. The multistage feature of the integrated architectures provides convenience for user handover operation. In each stage, we can always find a central coordinator that can handle and coordinate handover operation. In the context of multistage integration architectures as shown in Fig. 14.9, we next explain how user handover can be realized in the different stages.
Fig. 14.9 Handover operation in a multistage-integrated system
Under the multistage-integrated architectures as shown in Fig. 14.9, there are four types of wireless cells. The first type of cell is a micro-cell within a WiMAX star network, defined as cluster micro-cell. The second type of cell is a micro-cell covered by an ONU-BS, defined as ONU-BS micro-cell. The third type of cell is called macro-cell, which composes all the micro-cells that are associated with a common central EPON OLT. The last type of cell is the largest, called macro-macrocell, which is made up of all macro-cells associated with a common 10G EPON OLT. Based on these different types of wireless cells, there are four different types of handover. The first type of handover can occur between two cluster micro-cells. For this, a central ONU-BS can function as a controller to coordinate the handover. The second type of handover can occur between two ONU-BS micro-cells. In this case, a central OLT that connects to the two ONU-BSs can function as a controller to coordinate the handover. Handover can also occur between two macro-cells, for example, between macro-cells associated with two different EPON OLTs. In this case, a central 10G EPON OLT that connects the two EPON OLTs can function as a coordinator for the handover. Finally, the last type of handover can occur between two macro-macro-cells that correspond to two different 10G EPON OLTs. In this case, a higher level of coordinator is required for the operation.
14 Integration of EPON and WiMAX
315
14.4.6 Survivability An integrated EPON and WiMAX system may contain many WiMAX micro-cells. In the example as shown in Fig. 14.9, there are up to 1792 micro-cells associated with a single integrated system. Moreover, within each micro-cell, a large number of WiMAX subscribed users may be served. Thus, for an integrated system, a fatal failure, like a failure at a central OLT or a cut of the fiber link between an OLT and an optical splitter, can disrupt the services of all the users in the system. Survivability is therefore important to the integrated EPON and WiMAX systems. For the integrated architectures, we mainly consider the network failures due to fiber cut (because of its relatively high occurring probability). As shown in Fig. 14.10, there are two types of fiber cut. The first one is a fiber cut on the link between a central OLT and an optical splitter, and the second one is a fiber cut on the link between an optical splitter and an ONU-BS. In general, the first type of fiber cut is more serious, as it disconnects all the ONU-BSs from the central OLT, thereby disrupting the services of all the users. In contrast, the second type of fiber cut will only affect a small group of users that are associated with an ONU-BS. Different protection strategies can be adopted to recover from the above two types of fiber cut. For the first type of fiber cut, it is cost-effective to deploy an extra backup fiber along with the first fiber as shown in Fig. 14.10, such that if the first fiber gets cut, the backup fiber can take over the responsibility to carry user data. This is a type of 1+1/1:1 failure protection. Because the backup fiber is shared by all the users subscribed to the integrated system, the extra cost for this second fiber is generally acceptable for each of the subscribed users. However, for the second type of fiber cut, the deployment of 1+1/1:1 protection scheme can be expensive as only a small group of users share an extra fiber segment from an optical splitter to an ONU-BS. Under the integrated architecture, a more attractive solution can be adopted, which employs point-to-point WiMAX radio links between neighboring
c Fig. 14.10 Survivability of integrated access systems (adapted from [11], 2007 SPIE)
316
Gangxiang Shen and Rodney S. Tucker
WiMAX base stations to realize failure recovery for the second type of fiber cut. As shown in Fig. 14.10, when the fiber between an optical splitter and an ONU-BS gets cut, the ONU-BS can relay its user data to its neighboring WiMAX bases stations over the two point-to-point radio links. Then the neighboring WiMAX base stations further forward the user data to the central office. This type of failure protection technique is expected to be cheap for the second types of fiber cut.
14.4.7 Cooperative Transmission for Broadcast Services Under the multi-cellular wireless communications, interference from neighboring wireless cells is always troublesome for a user attempting to receive high-quality signals. Figure 14.11 shows an example of conventional multi-cell interfering communication. Assume that there are multiple WiMAX base stations (BS) and each WiMAX BS covers a wireless micro-cell, in which multiple subscribed users are associated with the WiMAX BS and receiving data from it. Due to independent user communications within each WiMAX micro-cell, any wireless power leakage from one WiMAX cell to another is considered as interference for the latter, which is harmful to signal to interference and noise ratio (SINR) of the users within the cell. A user within region A overlaid by three neighboring micro-cells as shown in Fig. 14.11 can suffer from strong signal interference. Specifically, if the user is associated with WiMAX BS 1 and receives a signal with power P1 from base station 1, and meanwhile, it also receives interfering signals leaked from the other two neighboring WiMAX BSs (ONU-BS 2 and ONU-BS 3) with power P2 and P3, respectively, then the user has an SINR P1/(N+P2+P3), where N is the power of noise. The SINR is poor if P1, P2, and P3 are close. As a result, only
Fig. 14.11 Downstream data transmission for broadcast services: without cooperative communication
14 Integration of EPON and WiMAX
317
a low-level modulation scheme such as BPSK can be employed for data transmission from ONU-BS 1 to the user. However, integrated EPON and WiMAX systems show good potential to overcome the above interfering situation by employing a transmission scheme called cooperative communications [28–30]. This transmission scheme is especially suitable to transmit broadcast services such as IPTV over the integrated systems. Based on the infrastructure as shown in Fig. 14.1, a single copy of user data can be forwarded from a central OLT down to all the connected ONU-BSs and then the latter duplicate the data copies to all the SSs within associated WiMAX micro-cells. Under broadcast services, each user wishes to receive the same copy of data, no matter the users are in the same micro-cell or in different micro-cells. Thus, in this case the signals leaked from neighboring WiMAX micro-cells should not be considered as interference; rather, they are useful signals. By properly manipulating these signals (e.g., based on some space–time coding technique [31]), they can be used to enhance the SINR of the signal received by each user, as multiple WiMAX BSs are transmitting multiple copies of information from different directions to the user. This is a type of transmitter macro-diversity, which is helpful to improve SINR in wireless communications. To realize the above cooperative communication, space–time coding techniques are required [31]. Specifically, each WiMAX BS is allocated with a unique orthogonal code. The BS uses the code to encode user data received from the central OLT and broadcasts the encoded signal to SSs. An SS receives multiple copies of signals (or superposed signals) [32] from different WiMAX BSs. It then decodes all the orthogonal signals to construct a stronger signal, which shows a better SINR. With a good SINR, an advanced modulation scheme such as 16 QAM can be applied for a high data transmission rate. We use an example as shown in Fig. 14.12 to illustrate the cooperative transmission mechanism. Assume that we have a data stream to broadcast to all the SSs
Fig. 14.12 Downstream data transmission for broadcast services: with cooperative communication
318
Gangxiang Shen and Rodney S. Tucker
within an integrated access network. The stream is first broadcast from the central OLT to all connected ONU-BSs. Then the ONU-BSs (e.g., ONU-BS 1, 2, and 3), of which each is allocated with a unique orthogonal code (e.g., orthogonal codes 1, 2, and 3, respectively), encode the received data stream into an orthogonal signal and broadcasts the signal to all the surrounding SSs. In particular, for an SS user located within the region overlaid by the three neighboring WiMAX micro-cells, it receives superposed orthogonal signals from the three neighboring WiMAX BSs. The SINR of the user in this case is (P1+P2+P3)/N, which is significantly enhanced over the previous SINR P1/(N+P2+P3). Based on this example, we can see that the integrated architectures are talented to realize cooperative communication, which can help to eliminate the interferences suffered by the conventional multi-cellular wireless communication systems and provide a good SINR for a high transmission data rate.
14.5 Summary We have considered integration of EPON and WiMAX. Three integration architectures have been described and elaborated on. These architectures include the simplest independent architectures, more advanced hybrid architectures, and more spectrum-efficient microwave-over-fiber (MoF) architectures. Based on these architectures, the issues ranging from optimal planning to efficient operation were discussed. These issues include optimal PON network deployment to interconnect disperse ONU-BSs, packet forwarding modes, bandwidth allocation and QoS support, handover operation for mobile users, survivability, and downstream cooperative transmission in the integrated architectures. Due to the novelty of integrated EPON and WiMAX broadband access networks, these architectures and issues are open for further investigation. We hope that this chapter has stimulated the readers interest in integrated EPON-WiMAX systems. Acknowledgments The authors would like to thank the Australia Research Council (ARC) for supporting this work.
References 1. IEEE 802.3ah Task Force: http://www.ieee802.org/3/efm/ [Online on April 4, 2008]. 2. G. Kramer, B. Mukherjee, and G. Pesavento, “IPACT: A Dynamic Protocol for an Ethernet PON (EPON),” IEEE Communications Magazine, vol. 40, no. 2, pp. 74–80, Feb. 2002. 3. ITU-T G.984.4, SG 15, “Gigabit-capable Passive Optical Networks (G-PON): Transmission Convergence Layer Specification,” July 2005. 4. IEEE 802.16-2004, “Air Interface for Fixed Broadband Wireless Access Systems,” October 2004.
14 Integration of EPON and WiMAX
319
5. WiMAX Forum, “Mobile WiMAX – Part I: A Technical Overview and Performance Evaluation,” Aug. 2006. 6. IEEE 802.16e/D12, “Air Interface for Fixed and Mobile Broadband Wireless Access Systems,” Feb. 2005. 7. IEEE 802.16 Working Group: http://www.ieee802.org/16/ [Online on April 4, 2008]. 8. H. Ekstrom et al., “Technical Solutions for the 3G Long-term Evolution,” IEEE Communications Magazine, vol. 44, no. 3, pp. 38–45, March 2006. 9. M. Vrdoljak, S. I. Vrdoljak, and G. Skugor, “Fixed-Mobile Convergence Strategy: Technologies and Market Opportunities,” IEEE Communications Magazine, vol. 38, no. 2, pp. 116–121, Feb. 2000. 10. G. Shen, R. S. Tucker, and T. Chae, “Fixed Mobile Convergence (FMC) Architectures for Broadband Access: Integration of EPON and WiMAX,” IEEE Communications Magazine, vol. 45, no. 8, pp. 44–50, Aug. 2007. 11. G. Shen and R. S. Tucker, “Fixed Mobile Convergence (FMC) Architectures for Broadband Access: Integration of EPON and WiMAX (invited),” in Proc., SPIE Network Architectures, Management, and Application V, APOC, Wuhan, China, vol. 6784, pp. 678403-1-678403-13, Nov. 2007. 12. Y. Lou et al., “Integrating Optical and Wireless Services in the Access Network,” in Proc., OFC, paper NThG1, Anaheim, CA, March 2006. 13. Y. Lou et al., “QoS-aware Scheduling over Hybrid Optical Wireless Networks,” in Proc., OFC, paper NThB1, Anaheim, CA, March 2007. 14. S. Sarkar, S. Dixit, and B. Mukherjee, “Hybrid Wireless-Optical Broadband-Access Network (WOBAN): A Review of Relevant Challenges,” IEEE Journal of Lightwave Technology, vol. 25, no. 11, pp. 3329–3340, Nov. 2007. 15. W. T. Shaw, S. W. Wong, N. Cheng, K. Balasubramanian, X. Zhu, M. Maier, and L. G. Kazovsky, “Hybrid Architecture and Integrated Routing in a Scalable Optical-Wireless Access Network,” IEEE Journal of Lightwave Technology, vol. 25, no. 11, pp. 3443–3451, Nov. 2007. 16. A. Nirmalathas, D. Novak, C. Lim, and R. Waterhouse, “Wavelength Reuse in the WDM Optical Interface of a Millimetre-wave Fibre-wireless Antenna Base Station,” IEEE Transactions on Microwave Theory, vol. 49, no. 10, pp. 2006–2012, Oct. 2001. 17. M. P. McGarry, M. Reisslein, and M. Maier, “WDM Ethernet Passive Optical Networks,” IEEE Communications Magazine, vol. 44, no. 2, pp. 15–22, Feb. 2006. 18. C. H. Doan et al., “Design Considerations for 60 GHz CMOS Radio,” IEEE Communications Magazine, vol. 42, no. 12, pp. 132–140, Dec. 2004. 19. 10G EPON IEEE Working Group: http://www.ieee802.org/3/av/index.html/ [Online on September 18, 2007]. 20. J. Li and G. Shen, “Cost Minimization Planning for Passive Optical Networks,” in Proc., OFC/NFOEC, paper NThD1, San Diego, CA, March 2008. 21. J. Li and G. Shen, “Cost Minimization Planning for Passive Optical Networks,” submitted to IEEE Journal on Selected Areas in Communications, March 2008. 22. IEEE 802.1D, “Media Access Control (MAC) Bridges,” June 2004. 23. IEEE 802.1Q, “Virtual Bridged Local Area Networks,” May 2006. 24. C. M. Assi, Y. Ye, S. Dixit, and M. A. Ali, “Dynamic Bandwidth Allocation for Quality-ofService over Ethernet PONs,” IEEE Journal on Selected Areas in Communications, vol. 21, no. 9, pp. 1467–1477, Nov. 2003. 25. G. Nair et al., “IEEE 802.16 Medium Access Control and Service Provisioning,” Intel Technology Journal, vol. 8, no. 3, pp. 213–228, Aug. 2004. 26. S. Blake et al., “An Architecture of Differentiated Services,” IETF RFC2475, Dec. 1998. 27. R. Braden et al., “Integrated Services in the Internet Architecture: An Overview,” IETF RFC1633, June 1994.
320
Gangxiang Shen and Rodney S. Tucker
28. H. Zhang and H. Dai, “Cochannel Interference Mitigation and Cooperative Processing in Downlink Multicell Multiuser MIMO Networks,” EURASIP Journal on Wireless Communications and Networking, vol. 2004, no. 2, pp. 222–235, Dec. 2004. 29. I. D. Garcia, K. Sakaguchi, and K. Araki, “Cell Planning for Cooperative Transmission,” in Proc., IEEE WCNC, Las Vegas, March/April 2008. 30. P. Ho. B. Lin, J. Tapolcai, and G. Shen, “Cooperative Service Provisioning in Integrated EPON-WiMAX Networks,” submitted to IEEE Communications Magazine, March 2008. 31. V. Tarokh, H. Jafarhani, and A. R. Calderbank, “Space-time Block Codes from Orthogonal Designs,” IEEE Transactions on Information Theory, vol. 45, no. 5, pp. 1456–1467, July 1999. 32. IEEE 802.16 Task Group m (TGm): http://www.ieee802.org/16/tgm/ [Online on April 4, 2008].
Chapter 15
Hybrid Wireless–Optical Broadband Access Network (WOBAN) Suman Sarkar, Pulak Chowdhury, Sudhir Dixit, and Biswanath Mukherjee
Abstract Hybrid wireless–optical broadband access network (WOBAN) is a promising architecture for future access networks. It combines the advantages of two diverse technologies – an optical back end drives a wireless mesh in the front end. It is a cost-effective solution compared to traditional optical access networks since fiber does not need to penetrate to the end user. WOBAN offers untethered flexibility due to its wireless mesh in the front end; yet it provides high transport capacity compared to traditional wireless solutions because of its optical back end. This chapter defines WOBAN, reviews different flavors of its architecture, and provides a comprehensive outline of various design models, coupled with efficient protocols to manage the network. It also argues why the combination of optical and wireless technologies should provide an improved solution for future network design.
15.1 Introduction The growing customer demands for bandwidth-intensive services are accelerating the need to design an efficient “last- mile” access network in a cost-effective manner. Traditional “quad-play” applications (which refers to a bundle of services with voice, video, Internet, and wireless) and premium rich-media applications (e.g., multimedia, interactive gaming, and metaverse) need to be delivered over the acSuman Sarkar University of California, Davis, CA, USA, e-mail:
[email protected] Pulak Chowdhury University of California, Davis, CA, USA, e-mail:
[email protected] Sudhir Dixit Nokia Siemens Networks, Mountain View, CA, USA, e-mail:
[email protected] Biswanath Mukherjee Department of Computer Science, University of California, Davis, CA, USA e-mail:
[email protected]
A. Shami et al. (eds.), Broadband Access Networks, Optical Networks, c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-92131-0 15,
321
322
Suman Sarkar, Pulak Chowdhury, Sudhir Dixit, and Biswanath Mukherjee
cess network to the end users in a satisfactory and economical way. Thus, besides its enormous transport capacity, today’s access infrastructure should bring operational efficiencies, namely mobility and untethered convenience to end users. In this chapter, we explore a novel hybrid network paradigm – wireless–optical broadband access network (WOBAN) – a combination technology of high-capacity optical and untethered wireless access. A WOBAN consists of a wireless mesh network at the front end, and it is supported by an optical network at the back end. Running fiber from the central office (CO) and then deploying wireless mesh near end user premises is a cost-effective solution for future access networks. How far the fiber should penetrate before wireless takes over is an interesting engineering optimization problem [1]. Although this chapter includes some basic WOBAN material that can also be found in [1], it also covers other research efforts on the WOBAN architecture. While WOBAN mainly focuses on the networking aspect of the wireless–optical converged architecture, a related technology namely, radio-over-fiber (ROF), has its root in the communication challenges of sending radio signals over fiber. The radio signals in ROF can be effectively carried over an existing optical fiber infrastructure (saving “last-mile” costs) by means of the “hybrid fiber radio” (HFR) enabling technology. However, the following challenges exist with ROF: (1) to design better transmission equipment, (2) to improve the signal’s power gain, (3) to develop sophisticated signal modulation/demodulation and up/down conversion techniques, etc. These challenges are complementary to WOBAN’s research focus. For more information on ROF as well as for other topics on optical–wireless integration, please refer to [2].
15.2 WOBAN: A Network for Future WOBAN architecture consists of an optimal combination of an optical network as backhaul [e.g., a passive optical network (PON)] and wireless access in the front end (e.g., WiFi and/or WiMAX) (see Fig. 15.1). WOBAN is designed to optimize cost and maximize utilization and performance in a broadband access network. It reduces the infrastructure cost of laying fiber to each end user by using wireless technology (viz. WiFi, WiMAX, etc.) and supports large aggregation of traffic (compared to traditional wireless access) due to its high-capacity PON infrastructure in the back haul. The reliability and robustness of wireline optical communication and the flexibility (“anytime-anywhere” approach) and cost savings of a wireless network are integrated together in the WOBAN architecture [1]. Currently, one of the dominant broadband access technologies is passive optical network (PON). A PON segment starts from the telecom central office (CO) and consists of an optical line terminal (OLT) in its head end at the CO. Each OLT can drive several optical network units (ONU) which make the tail end of the PON segments, and serve end users. In WOBAN, these ONUs will support wireless base stations (BSs) in its wireless front end. The wireless BSs that are directly connected
15 Hybrid Wireless–Optical Broadband Access Network
323
Fig. 15.1 SFNet: WOBAN envisioned for part of San Francisco
to the ONUs are known as wireless gateways. Besides these gateways, the wireless front end of a WOBAN consists of other wireless routers/BSs to efficiently manage the network connectivity. Therefore, the front end of a WOBAN is essentially a multi-hop wireless mesh network (WMN) with several wireless routers and a few gateways (to connect to the ONUs and consequently to the rest of the Internet through OLTs/CO) [1].
15.2.1 Why Is WOBAN a Compelling Solution? The advantages of a WOBAN over the wireline optical and wireless networks have made the research and deployment of this type of network more attractive. These advantages can be summarized as follows. Cost-effectiveness: A WOBAN can be very cost-effective compared to a wired network. The architecture (see Fig. 15.1) demonstrates that we do not need expensive “fiber-to-the-home” (FTTH) connectivity, because installing and maintaining the fiber all the way to each user could be quite costly. Please see [1, 3] for more information on cost estimation of WOBAN deployment. In WOBAN, a user will connect to its neighborhood ONU in a wireless fashion, possibly over multiple hops through other wireless routers. At the ONU, the wireless user’s data will be processed and sent to the OLT using the optical fiber infrastructure.
324
Suman Sarkar, Pulak Chowdhury, Sudhir Dixit, and Biswanath Mukherjee
Flexibility: The wireless part of this architecture allows users inside the WOBAN to seamlessly connect to one another. So, a WOBAN is more flexible than the optical access network. Thus, WiFi and WiMAX are convenient technologies for the front end of the WOBAN, so that we can exploit the mobility features incorporated in them. Robustness: A WOBAN should be more robust than the traditional wireline network. In a traditional PON, if a fiber connecting the splitter to an ONU breaks (see Fig. 15.1), that ONU will be down. Even worse, if a trunk from OLT to the splitter breaks, all the ONUs (along with the users served by the ONUs) will fail. But, in a WOBAN, as the users have the ability to form a multi-hop mesh topology, the wireless connectivity may be able to adapt itself so that affected users may be able to find a neighboring ONU which is alive. Then, the users can communicate with that ONU; and that ONU, in turn, will communicate with another OLT in the CO. Transport capacity: Due to its high-capacity optical trunk at the back end, the WOBAN’s transport capacity is much higher than the relatively low capacity of the wireless network. Reliability: A WOBAN will be more reliable than the wireless network. This, in turn, will help in reducing the problem of congestion and information loss in a WOBAN compared to current wireless networks. Also, a user’s ability to communicate with any other ONU in its vicinity, if its primary ONU breaks or is congested, gives the WOBAN a better load-balancing capability. Self-organization: The WOBAN is “self-organizing” because of its fault-tolerant capability and because of its robustness with respect to network connectivity and load-balancing features. Incentives: In many developing regions of the world, fiber is deeply deployed (within 20 km) even in the rural areas, but the cost to provide wireline broadband connectivity is prohibitively expensive, time consuming, and difficult to maintain. In such scenarios, the governments have decided to either build or provide incentives to the operators to deploy WOBAN-like architectures.
15.2.2 Flavors of Converged Architecture A WOBAN deployment is more challenging than only an optical or a wireless access network deployment. This is because of the design interplay between two very diverse access technologies (optical and wireless). In addition, the network designer has to ensure that both parts are well designed and neither part is over-designed (with excess resources) nor under-designed (a resource bottleneck). 15.2.2.1 WOBAN An early WOBAN proposal [4] was a multi-domain hybrid network. WOBAN model is essentially an integrated tree-mesh architecture. It assumes that an OLT
15 Hybrid Wireless–Optical Broadband Access Network
325
is placed in a telecom central office, and it feeds several ONUs. Thus, from ONU to the CO, WOBAN has a traditional fiber network; and, from ONUs, end users are wirelessly connected (in single-hop or multi-hop fashion). Figure 15.1 captures a WOBAN architecture. The optical part of WOBAN assumes a tree, while a mesh is envisioned in its front-end wireless part. Besides driving wireless gateways, the ONUs can also be shared for traditional PON users. In this multi-domain architecture, the gateways (wireless routers that are physically connected to ONUs) are primary aggregation points. Multiple wireless routers can be associated with a single gateway. The gateways and wireless routers together form the front-end mesh. In the back end, the OLT acts as the parent of the tree, with gateways as leaves and ONUs as children in between. The ONUs are higher aggregation points since multiple gateways can connect to one ONU. Consequently, the OLT is the highest aggregation points for the WOBAN before the traditional metro/core aggregation occurs for the rest of the network. SFNet – a case study envisioned for San Francisco WOBAN: Figure 15.1 envisions the WOBAN architecture in a part of the city of San Francisco, California, from approximately N 37◦ 46 43.39 , W 122◦ 26 19.22 (Golden Gate AvenueDivisadero Street intersection) to N 37◦ 46 51.78 , W 122◦ 25 13.27 (Golden Gate Avenue-Van Ness Avenue intersection) and from N 37◦ 47 32.57 , W 122◦ 26 28.90 (Divisadero Street-Pacific Avenue intersection) to N 37◦ 47 41.39 , W 122◦ 25 23.71 (Van Ness Avenue-Pacific Avenue intersection). This is approximately a 1 squaremile area in downtown San Francisco with an estimated population of around 15,000 residents, where San Francisco has an area of nearly 47 square-miles with a population of around 745, 000; so the population of the region in Fig. 15.1 is quite representative of San Francisco’s population density. The San Francisco WOBAN (called SFNet) consists of 25 point-to-multipoint WiFi routers (which form the front-end wireless mesh). Five of these 25 routers are designated as gateways to the optical back end of WOBAN and placed at the edges of SFNet.1 Gateways are connected to different ONUs (and consequently different PON groups) and a single ONU (e.g., ONU1) can drive multiple gateways. To provide network robustness, different OLTs feed separate PON groups, namely OLT1 drives the top three ONUs and OLT2 drives the bottom one (ONU4). Wildhorse – a case study envisioned for Davis, California, WOBAN: Wildhorse is a neighborhood in North Davis, California. The neighborhood has only residential homes in an area of approximately 1 square-mile. A survey was conducted to obtain the information on the types of wireless devices in these homes (based on IEEE standards), their configurations (bit rates, channels used, frequencies of operation, positions on the earth, etc.), and encryption standards. After mapping the wireless routers (310 of them) in Wildhorse, and knowing the location of the CO and the city’s fiber layout, Fig. 15.2 shows the locations of wireless routers and three ONUs (black triangles). The prediction of 70 Mbps of maximum bandwidth needed per home accounts for three ONUs to support the back-end transport capacity [5]. For 1 In Fig. 15.1, black squares (five of them) are attached to the optical part of WOBAN as gateways; others (20 of them) are routers.
326
Suman Sarkar, Pulak Chowdhury, Sudhir Dixit, and Biswanath Mukherjee
Fig. 15.2 WOBAN case study of Wildhorse
optimal cost, a minimum-spanning-tree (MST) approach is used for fiber layout from Davis CO to ONUs placed in the Wildhorse. Approaches for WOBAN deployment: The network performance largely depends on the deployment of ONUs/BSs, i.e., the gateways where the optical and wireless parts meet. Proper deployment of ONUs/BSs is critical to the cost optimization of WOBAN. The approaches which have been proposed to deploy WOBAN for various optimization metrics are deterministic, greedy, simulated annealing, and mixed integer programming. Given the location of the users, these algorithms focus on how to find the “good” placement of multiple ONUs/BSs in a cost-effective manner. Table 15.1 shows the pros and cons of each of the approaches in brief. Please see [1, 3, 4, 6] for more details.
15.2.2.2 Grid Reconfigurable Optical and Wireless Network (GROW-Net) In [7], the authors proposed another WOBAN-like architecture, called grid reconfigurable optical and wireless network (GROW-Net). It consists of a WMN in the front end and a reconfigurable optical network as the backhaul. The WMN front end is used to provide low-cost service to the user ubiquitously. The WMN deploys multiple wireless mesh routers to forward traffic to and from the wired Internet entry points, called “gateways.” The mesh routers in WMN are geographically scattered in a large area such as a city. The back end should be driven by a low-loss, highcapacity, cost-effective network. Consequently, the optical network, namely PON, is a candidate to provide such a high-capacity, point-to-multipoint backhaul to the WMN front end of GROW-Net. GROW-Net architecture is shown in Fig. 15.3.
15 Hybrid Wireless–Optical Broadband Access Network
327
Table 15.1 Pros and cons of various placement approaches in WOBAN Placement scheme Deterministic [1]
Solution quality Better
Processing time Constant
Comments (in brief)
Greedy algorithm [4]
Good
Linear (in practical cases)
• Low complexity • Divide-and-conquer heuristic • Good solution for uniform distribution of users
Simulated annealing [6]
Improved over Greedy
Depends on convergence criteria
• Combinatorial optimizer • Improved solution over greedy • May not converge for discontinuous cost model
Optimal
Very high
• Complex analytical solution • Considers several constraints • Model predicts setup costs in dollars
MIP [3]
Objective Placements of optical network units (ONU) in WOBAN
Optimum setup of ONUs and BSs
• Works well for symmetric topology • Predetermined placement • No prior optimization
Optical Tree Networks OL
Central Office
ONU/Gateway
T
A Sp
litt
er
1
B
2 3
4
Wireless Mesh Network Op
Fig. 15.3 GROW-Net architecture
tica
lR
Ne ing
two
rk
328
Suman Sarkar, Pulak Chowdhury, Sudhir Dixit, and Biswanath Mukherjee
The WMN of GROW-Net is “infrastructure” based due to its limited or no mobility, less power constraints, and enhanced computational capabilities. These routers act as relay agents which can forward traffic to/from the Internet through gateways/ONUs. Gateway routers are connected to the Internet through the PON. The upstream traffic from the client (end user) is aggregated in a nearby mesh router, which then forwards the traffic to a nearby gateway. A mesh router can forward traffic to different gateway routers based on current load condition, thereby enhancing reliability and enabling load balancing in the WMN. Existing IEEE 802.11a/b/g (WiFi) technologies can be exploited in the WMN of GROW-Net. However, these technologies are not optimized for WMN at the PHY and MAC layer of the network stack. Therefore, the authors of [7] suggested that the feasibility of proprietary wireless technologies, viz. IEEE 802.16 (WiMAX) and IEEE 802.16m (WiFi with enhanced features), needs further investigation to improve the capacity, reliability, maintenance, and mobility of WMN in GROW-Net. The optical backhaul in GROW-Net consists of a ring and multiple tree networks. The root nodes (OLTs) of the tree networks also belong to the ring. Figure 15.3 shows such a hybrid topology. The leaf nodes, e.g., ONUs of the tree networks, are connected to the gateways of WMN in GROW-Net.
15.2.2.3 Optical–Wireless Integration (OWI) The authors of [8] proposed a different flavor of the WOBAN model, referred to as the optical–wireless integration (OWI) architecture. In the OWI architecture, WiMAX technology is envisioned for the wireless front end to provide the last-mile broadband solution. Passive optical network (PON) works as the optical backhaul of the OWI architecture. OWI architecture consists of point-to-multipoint WiMAX combined with multiple input multiple output (MIMO) technology such as the V-BLAST architecture [9] to improve wireless front-end capacity. MIMO uses spatially distributed multiple element arrays to exploit the space diversity of wireless channels. It also exploits multipath propagation with no extra spectrum. An example OWI architecture with MIMO technology is shown in Fig. 15.4. OWI facilitates coordinated antenna resource allocation among WiMAX base stations with MIMO links. Therefore, the OWI model is combined with a centralized antenna assignment algorithm at the OLT which coordinates among several WiMAX base stations and exploits MIMO diversity to achieve better throughput and resilience, in case of base station failures.
15.3 Connectivity and Routing Once a WOBAN is deployed, how to create a mesh topology in the front end and how to route information (data packets) through it are important problems.
15 Hybrid Wireless–Optical Broadband Access Network
329
Metro Ring OLT
PON
PON
PON WiMAX BS
WiMAX CPE WiMAX BS
MIMO Links WiMAX Customer WiMAX Base Station (BS)
Premise Equipment (CPE)
Fig. 15.4 Optical-wireless integration (OWI) architecture
A WOBAN is envisioned primarily for residential and small-to-medium business (SMB) users. The characteristics of a WOBAN’s front-end wireless mesh is different from that of the traditional wireless mesh. In a traditional wireless mesh, the connectivity may change due to users’ mobility and a wireless link going up and down on the fly. On the other hand, since the WOBAN primarily is a network of residential and SMB users, its connectivity pattern in the wireless front end can be pre-estimated. In a typical WOBAN, an end user, e.g., a subscriber with wireless devices at individual homes (scattered over a geographic area), sends a data packet to one of its neighborhood wireless routers. This router then injects the packet into the wireless mesh of the WOBAN. The packet travels through the mesh, possibly over multiple hops, to one of the gateways (and to the ONU) and is finally sent through the optical part of the WOBAN to the OLT/CO. In the upstream direction of the wireless front end (from a wireless user to a gateway/ONU), the WOBAN is an anycast network, i.e., an end user can try to deliver its packet(s) to any one of the gateways (from which the packet will find its way to the rest of the Internet). In the optical back end, the upstream part of a WOBAN (from an ONU to an OLT/CO) is a multipoint media-access network, where ONUs are deployed in a tree network with respect to their OLT and they contend for a shared upstream resource (or bandwidth). But in the downstream direction of the wireless front end (from a gateway/ONU to a wireless user), this network is a unicast network, i.e., a gateway will send a packet to only its specific destination (or user). In the optical back end, the downstream (from an OLT/CO to an ONU) of a WOBAN is a broadcast network, where a packet, destined for a particular ONU, is broadcast to all ONUs in the tree and processed selectively only by the destination ONU (all other ONUs discard the packet), as in a standard PON. Figure 15.5 captures a WOBAN’s upstream and downstream transmit modes.
330
Suman Sarkar, Pulak Chowdhury, Sudhir Dixit, and Biswanath Mukherjee
Fig. 15.5 WOBAN’s upstream and downstream protocols
The routing in WOBAN falls broadly into two parts: (1) routing in multi-hop wireless mesh at the front end and (2) media-access control in PON at the back end. These two parts should run in coherence. The basic steps of WOBAN routing are as follows: 1. Update/predict wireless link states: Update and exchange the link condition among various routers. To reduce the explosion of link-state messages, link-state prediction technique can be used. 2. Assign link weights: Several metrics, viz. hops, delay, congestion, etc., can be chosen to assign link weights in an intelligent way. 3. Find suitable path to gateway: On the basis of link metrics, find a suitable path from a router to a gateway. 4. Aggregate packets in gateway: Each gateway aggregates packets from various routers before it reports the aggregation to its upstream ONU. 5. Send cumulative aggregation report from ONU to OLT: After knowing each gateway’s aggregation report, the parent ONU computes the cumulative aggregation report before it sends a request to its OLT. 6. Grant bandwidth to ONU: OLT, after receiving cumulative reports from its ONUs, grants bandwidth to each of them periodically. 7. Flow control: Discard packets in WMN if packets cannot be delivered properly. Next, we briefly review some specific routing algorithms proposed for WOBAN.
15.3.1 Delay-Aware Routing Algorithm (DARA) DARA [10] is proposed for the front-end wireless mesh of WOBAN. The routing in the wireless mesh of a WOBAN deals with how to deliver packets from a router to a gateway and vice versa. Routing in DARA is based on two decisions: (1) the associativity of a user to a nearby wireless router in its range and (2) the path from this (ingress) router to a suitable gateway (through the wireless mesh). DARA is a proactive routing approach which focuses on the packet delay (latency) in the front end (wireless mesh) of the WOBAN, i.e., the packet delay from the router to the gateway (attached to an ONU) and vice versa. The packet delay could be significant as the packet may travel through several routers in the mesh before finally
15 Hybrid Wireless–Optical Broadband Access Network
331
reaching the gateway (in the upstream direction) or to the user (in the downstream direction). DARA is based on the following principles: -
Wireless-link-state updates (among routers and gateways), Link-state prediction (based on weighted moving average), Link-weight assignment (based on predicted link delay metric), Path computation/selection in wireless mesh (K-shortest paths based on delay awareness), - Traffic aggregation in upstream gateways, and - Admission control. Consequently, DARA finds a path that minimizes the front-end WMN delay. It shows how choosing a path from a set of paths (whose delays are below the delay requirement) can alleviate congestion and achieve better load balancing. The details of DARA can be found in [10]. In the back end, DARA envisions the standard multipoint control protocol (MPCP) protocol among ONUs and OLTs with multiple aggregation points, namely gateway-aggregation-to-ONU and ONU-aggregationto-OLT.
15.3.2 Capacity and Delay-Aware Routing (CaDAR) A follow-up work to DARA is CaDAR [11]. It explores optimal wireless capacity allocation over the delay-aware routing proposed in DARA. In the wireless mesh part of a WOBAN, a subset of wireless links carry heavier traffic compared to the other links between the routers and gateways. Among the active wireless links, the traffic flow is also uneven. Unlike DARA, where the wireless capacity is distributed evenly among neighbors, CaDAR optimally distributes the wireless capacity only among active links based on the flows on the links. This allocates higher capacity to the links with higher flow and utilizes the limited capacity of the wireless nodes efficiently. CaDAR gathers information of the wireless links, estimates the delay on the links, and performs delay-aware routing by selecting the shortest-delay path in the wireless part of WOBAN. By performing capacity awareness on top of DARA’s delay-aware routing, CaDAR improves on the throughput and average packet delay of WOBAN. Like DARA, in the back end, CaDAR also envisions the standard multipoint media-access control, namely MPCP, ONUs, and OLTs.
15.3.3 GROW-Net Integrated Routing (GIR) GROW-Net uses an integrated routing algorithm to route packets through various routes from the central hub to each end user. GIR determines the optimum
332
Suman Sarkar, Pulak Chowdhury, Sudhir Dixit, and Biswanath Mukherjee
route based on wireless link states and average traffic rates. It selects the optimal ONU/gateway for each wireless mesh router at any given network condition and performs load balancing [7]. The upstream traffic from the end user is first collected at the adjacent mesh router and then forwarded to the selected gateway router. Gateway router selection depends on the current network load. The routes between gateways and mesh routers are calculated based on real-time network conditions. For example, as shown in Fig. 15.3, the traffic from the end users adjacent to mesh router 4 is first aggregated at mesh router 4. Then, they are forwarded to gateway B through mesh routers 3 and 2 to reach gateway B. Depending on the traffic condition, the traffic may also be forwarded to gateway A through routers 3, 2, and 1. In case of downstream traffic, the traffic is first forwarded to the specific gateway, e.g., gateway B in Fig. 15.3, and then forwarded through a selected route (in this case through mesh routers 2, 3, and 4) to reach the end user. GIR has the following steps (as shown in Fig. 15.6): - Wireless-link-state update: Wireless mesh routers disseminate link states (step 1 in Fig. 15.6). - Local-WMN route calculation: ONU (and its gateway router) computes optimal route to each wireless router (step 2). - Route-cost report and gateway association: ONU reports route cost for each mesh router in both upstream and downstream directions to CO, called central hub (step 3). Once the reports from all ONUs are received, route-assignment module associates every mesh router with the gateway router of lowest cost (step 4). - Congestion monitoring and report: ONU (or gateway router) measures average flow rates in both directions for all associated mesh routers (step 5) and sends report to the route-assignment module (step 6). It helps to reduce congestion and improve load balancing.
7
Route Assignment
9
4
OLT1
OLT2
Central Hub 6
3
8
ONU1
ONU2
Interface1
Interface2
Gateway1
Gateway2
5
2
1
Mesh Router1
Mesh Router2
Mesh Router3
Mesh Router4
Fig. 15.6 Integrated routing in GROW-Net
15 Hybrid Wireless–Optical Broadband Access Network
333
- Alternative gateway lookups/gateway reassociation: Route-assignment module tries to find out new alternative gateways for each router (step 7) and reassociate mesh routers to gateway, if a predetermined congestion threshold is exceeded (step 8). - Flow control: If congestion persists, discard flows (step 9).
15.4 Fault Tolerance and Self-Healing In this section, we discuss how WOBAN architectures exhibit fault tolerance and can restore operation from possible network failure scenarios. The failures can be of several types: - Gateway failure: Failure of wireless gateway. In this case, a wireless router needs to reassociate itself with another “live” gateway. - ONU failure: If an ONU fails, the connection from its gateways (and their associated routers downstream) should be reprovisioned to other neighbor ONUs. - OLT failure: This failure is more damaging (since OLT drives several ONUs), but less frequent (due to its protection from natural calamities and human error inside the CO). Consequently, multiple ONUs supported by that OLT will fail. In this failure, traffic from a large portion of the area needs to be rerouted. - Fiber cut: Failure due to fiber cut between upstream (ONU/gateways) and downstream (OLT) components. Hence, paths between CO and PON groups will be non-functional. Packet loss may also occur due to any combination of the failure scenarios in the WOBAN, viz., gateway failure, ONU failure, and/or OLT failure. To handle these failures, WOBAN and GROW-Net use different fault-tolerance schemes. The following subsections present two fault-tolerance schemes for the WOBAN architecture.
15.4.1 Risk-and-Delay Aware Routing Algorithm (RADAR) In [12], risk-and-delay aware routing algorithm (RADAR) (an extension of “delayaware routing algorithm” (DARA) [10]) has been proposed to handle failures. RADAR can handle multiple failure scenarios. RADAR differentiates each gateway in the WOBAN by maintaining a hierarchical risk group that shows which PON group (ONU and OLT) a gateway is connected to. Each gateway is indexed, which contains its predecessors (ONU and OLT indices as well) to maintain the tree-like hierarchy of WOBAN. ONUs and OLTs are indexed in similar fashion. In RADAR, to reduce packet loss, each router maintains a “risk list (RL)” to keep track of failures. An RL in each router contains six fields, viz., path number (PN), primary gateway group (PGG), secondary gateway group (SGG), tertiary
334
Suman Sarkar, Pulak Chowdhury, Sudhir Dixit, and Biswanath Mukherjee
gateway group (TGG), path status (PS) (“live” or “stale”), and corresponding path delay (PD). The primary gateway for a router is the gateway with the minimum delay path. PGG contains paths with the primary gateway and the gateways connected to the same ONU as with the primary gateway. SGG contains paths with gateways that are connected to different ONUs but the same OLT as with the PGG. TGG contains paths with gateways that are connected to a different OLT (and consequently a different ONU). In the no-failure situation, all the paths are marked “live.” Once a failure occurs, RL will be updated and paths that lead to the failed gateway(s) will be marked “stale.” Thus, while forwarding packets, the router will only choose a “live” path. In this way, RADAR can provide protection for multiple front-end and back-end hierarchical failure scenarios. If all links adjacent to PGG go down, and that of SGG and TGG are “live,” then routers can infer that either both gateways connected to PGG have failed simultaneously or their parent ONU has failed. Then, packets will be rerouted through SGG and TGG paths. If all links adjacent to PGG and SGG go down, then routers can infer that either all the gateways connected to PGG and SGG have failed simultaneously or their corresponding upstream ONUs have failed simultaneously, or their parent OLT has failed. Then, TGG paths should take care of the packets. Therefore, we observe that a router does not always need to recompute a new set of minimumweight paths even if a failure occurs. A router will recompute paths only if all its previously computed paths fall under PGG and the ONUs/OLT fail, or all paths fall under PGG and SGG and the corresponding OLT fails. After path recomputation, packets will be admitted in WOBAN with degraded service (i.e., increased delay). This mechanism is called “self-healing” [12].
15.4.2 GROW-Net’s Reconfigurable Backhaul In GROW-Net, bandwidth demand may vary significantly in different zones at different time periods. Furthermore, ONU failure (in one PON group) can occur occasionally which requires reprovisioning of traffic across other PON groups. A tunable optical transceiver in the ONU can reallocate bandwidth to achieve fluctuation in the bandwidth demands. By tuning the transmitting and receiving wavelengths, ONUs can be deregistered from a heavily loaded PON and reregistered to a different lightly loaded PON for load balancing [7]. The central hub of GROW-Net has a network terminal (NT) which is devised to manage multiple PON groups. There is a system-bandwidth-management (SBM) module in the NT which continually monitors the buffer depth of each OLT for the downstream traffic. When a PON becomes heavily loaded (by measuring the OLT buffer depth), the SBM module instructs the heavily loaded PON to deregister some ONUs and to reregister them to the lightly loaded PON(s). Number of ONUs transferred from one PON to another depends on the PON’s individual average traffic load, which is calculated by traffic estimators (TE). Before the ONU is deregistered, its packets in the OLT’s queue are emptied. While the ONU is deregistered, the
15 Hybrid Wireless–Optical Broadband Access Network
335
incoming traffic through that ONU is temporarily stored in the queue of the NT until the reregistration is complete [7]. The reconfiguration procedure has a drawback of large tuning time due to slow tunable transceivers. If the traffic load of a PON oscillates, the reconfiguration will be triggered more frequently, thereby reducing network efficiency. Increasing the buffer size and setting a higher reconfiguration-trigger threshold may minimize this performance degradation [7]. But packets may also suffer more delay due to longer queueing time. Therefore, further investigation is required to find alternatives for enhancing quality of service (QoS) during optical backhaul reconfiguration in GROW-Net.
15.5 Summary We reviewed an architecture and a vision for wireless–optical broadband access network (WOBAN) and articulated why the combination of wireless and optical presents a compelling solution for broadband access. This chapter reviewed a converged network paradigm for future access network and investigated its architecture, design algorithms, and network protocols. In WOBAN design, we summarized the algorithms to optimize the placement of ONUs. We also evaluated various routing algorithms for WOBAN, including its fault-tolerant characteristics, and presented some novel concepts that are better suited for such hybrid networks.
References 1. Sarkar, S., Dixit, S., Mukherjee, B.: Hybrid Wireless-Optical Broadband Access Network (WOBAN): A Review of Relevant Challenges. (Invited Paper) In: IEEE/OSA Journal of Lightwave Technology, Special Issue on Convergence of Optical Wireless Access Networks, vol. 25, no. 11, pp. 3329–3340 (Nov. 2007) 2. IEEE/OSA Journal of Lightwave Technology, Special Issue on Convergence of Optical Wireless Access Networks, (Nov. 2007) 3. Sarkar, S., Yen, H.-H., Dixit, S., Mukherjee, B.: A Mixed Integer Programming Model for Optimum Placement of Base Stations and Optical Network Units in a Hybrid Wireless-Optical Broadband Access Network (WOBAN). In: IEEE Wireless Communications and Networking Conference (WCNC), Hong Kong (March 2007) 4. Sarkar, S., Mukherjee, B., Dixit, S.: Optimum Placement of Multiple Optical Network Units (ONUs) in Optical-Wireless Hybrid Access Networks. In: IEEE/OSA Optical Fiber Communications (OFC), Anaheim, California (March 2006) 5. Yang, S., Personal Communication, ETRI, South Korea (2005) 6. Sarkar, S., Mukherjee, B., Dixit, S.: Towards Global Optization of Multiple ONUs Placement in Hybrid Optical-Wireless Broadband Access Networks. In: IEEE Conference on Optical Internet (COIN), Jeju, South Korea (July 2006) 7. Shaw, W.-T., Wong, S.-W., Cheng, N., Balasubramanian, K., Zhu, X., Maier, M., Kazovsky, L.: Hybrid Architecture and Integrated Routing in a Scalable Optical-Wireless Access
336
8.
9.
10.
11.
12.
Suman Sarkar, Pulak Chowdhury, Sudhir Dixit, and Biswanath Mukherjee Network. In: IEEE/OSA Journal of Lightwave Technology, Special Issue on Convergence of Optical Wireless Access Networks, vol. 25, no. 11, pp. 3443–3451 (Nov. 2007) Lin, P., Wang, T., Suemura, Y., Qiao, C.: Improving Access Performance with an Integrated PON and WiMAX with MIMO. In: IEEE/OSA Optical Fiber Communications (OFC), Anaheim, California (March 2007) Wolniansky P., Foschini G., Golden G., Valenzuela R.: V-BLAST: an architecture for realizing very high data rates over the rich-scattering wireless channel. In: URSI International Symposium on Signals, Systems, and Electronics, pp. 295–300, New York, NY, USA (1998) Sarkar, S., Yen, H.-H., Dixit, S., Mukherjee, B.: DARA: Delay-Aware Routing Algorithm in a Hybrid Wireless-Optical Broadband Access Network (WOBAN). In: IEEE International Conference on Communications (ICC), Glasgow, Scotland (June 2007) Reaz, A., Ramamurthi, V., Sarkar, S., Ghosal, D., Dixit, S., Mukherjee, B.: CaDAR: an Efficient Routing Algorithm for Wireless-Optical Broadband Access Network. In IEEE International Conference on Communications (ICC), Beijing, China (May 2008) Sarkar, S., Yen, H.-H., Dixit, S., Mukherjee, B.: RADAR: Risk-and-Delay Aware Routing Algorithm in a Hybrid Wireless-Optical Broadband Access Network (WOBAN). In: IEEE/OSA Optical Fiber Communications (OFC), Anaheim, California (March 2007)
Part V
Deployments
Chapter 16
Point-to-Point FTTx Wolfgang Fischer
Abstract Fiber to the home (FTTH) architectures based on point-to-point fiber deployments are mainstream in Europe and Middle East because they have been technically and economically viable for a long time, and they have a number of characteristics that make them the preferred choice when technological flexibility over the lifetime of the fiber plant is required or when open access to fiber is required either from a regulatory or business case perspective. This chapter provides an in-depth discussion of the technical challenges and opportunities of point-to-point FTTH networks.
16.1 Introduction While in many parts of the world PON has become the preferred technology for FTTH, installations in Europe and Middle East are mostly based on point-to-point fiber deployments, either for direct connections between subscribers and POPs (points of presence) in real FTTH scenarios or between building switches and POPs in FTTB scenarios. This situation is mainly due to the fact that most of the initial European deployments were made by municipalities (e.g., [1]) and utilities (e.g., [2]) who were effectively working in greenfield scenarios, i.e., they had to invest into civil works which constitutes the dominant part of the cost for deploying FTTH (see Section 16.6), while the cost of fiber is virtually negligible in this context. Furthermore, as discussed below, many of these networks have explicitly been built with the objective of open fiber access which is greatly facilitated by point-to-point fiber deployments. Learning from the experience of these early players and after evaluating the characteristics of point-to-point vs. tree topologies used for PON deployments, more and more alternative service providers (e.g., [3]) and even a number of in-
Wolfgang Fischer Cisco Systems Europe, e-mail:
[email protected]
A. Shami et al. (eds.), Broadband Access Networks, Optical Networks, c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-92131-0 16,
339
340
Wolfgang Fischer
cumbent service providers (e.g., [4, 5]) have decided in favor of point-to-point fiber deployments. Table 16.1 lists the acronyms used in the remainder of this chapter.
Table 16.1 List of acronyms Acronym
Explanation
ADSL BRI CPE DSLAM EFM EPON FTTB FTTC FTTH FTTN FTTx GPON IP IPTV ISDN MAC OAM ODF ONT OTDR PON POP RF SDH SONET TDM TV VDSL
Asymmetric digital subscriber line Basic rate interface Customer premises equipment Digital subscriber line access multiplexer Ethernet in the first mile Ethernet PON (IEEE 802.3ah) Fiber to the building Fiber to the curb Fiber to the home Fiber to the node Fiber to the x ∈ {B, C, H, N} Gigabit PON (ITU-T G.984) Internet protocol TV service over Internet protocol (IP) Integrated services digital network Medium access control Operation administration and maintenance Optical distribution frame Optical network termination Optical time domain reflectometry Passive optical network Point of presence Radio frequency Synchronous digital hierarchy Synchronous optical network Time division multiplexing Television Very high speed digital subscriber line (latest version is VDSL2) Virtual local area network Wavelength-division multiplexing
VLAN WDM
16.2 Fiber Topology vs. Transmission Scheme It is important to make a clear distinction between the topologies used for the deployment of the fibers and the technologies used to transport data over the fibers. The two most widely used topologies are (passive) trees and stars. • Fiber trees with passive optical splitters in the field are deployed in order to be operated by one of the standardized PON technologies (EPON, GPON, etc.) using TDM-based MAC protocols to control the access of multiple subscribers to the shared feeder fiber.
16 Point-to-Point FTTx
341
• Fiber stars provide dedicated point-to-point fibers between the POP and either the subscriber or a first active aggregation point in the network. As there is a dedicated transparent medium a wide choice of transport technologies is available. All existing point-to-point FTTH deployments use Ethernet (100BASE-BX10 or 1000BASE-BX10) over single-mode single fibers, but this can be mixed with other transmission schemes for business applications (e.g., fiber channel and SONET/SDH), and even with PON technologies by placing the passive splitters into the POPs. Upgrades and modifications of the transmission technology can be performed on a subscriber-by-subscriber basis in the same way as dedicated copper loops were upgraded from analog voice, over ISDN BRI, over ADSL, up to VDSL2.
16.3 Architectural Considerations On the basis of point-to-point fiber deployments various access network architectures can be implemented. The choice of the architectures depends on a large number of criteria, including subscriber densities, requirements for open access, regulatory constraints, business models, required per-customer bitrates, oversubscription factors, availability of fiber infrastructure, and available POP space. • Fiber to the Home (FTTH): Each ONT/CPE device at the subscriber premise is connected via a dedicated fiber to a port on a switch in the POP, using 100BASEBX10 or 1000BASE-BX10 transmission (see Section 16.8). • Fiber to the Building (FTTB): A switch/DSLAM in the building (typically in the basement) is connected to the POP via a single fiber or a pair of fibers, carrying the aggregated traffic of the building via Gigabit Ethernet or 10-Gigabit Ethernet. The connections between subscribers and the building switch can be fiber or copper based and also use some form of Ethernet transport suited to the medium available in the vertical cabling. In some cases building switches are not individually connected to the POP but are interconnected in a chain or ring structure in order to utilize existing fibers deployed in particular topologies and to save fibers and ports in the POP. • Fiber to the Curb (FTTC): A switch/DSLAM, typically in a street cabinet is connected to the POP via a single fiber or a pair of fibers, carrying the aggregated traffic of the neighborhood via Gigabit Ethernet or 10-Gigabit Ethernet. The connections between subscribers and the switch in the street cabinet can be fiber or copper based and use either 100BASE-BX10, 1000BASE-BX10, or VDSL2. This architecture sometimes is also called “active Ethernet” as it requires active network elements in the field. For the remainder of this chapter we will consider only FTTH deployments because in the long term they are considered the target architecture due to their virtually unlimited scalability. In an overview article related to FTTH architectures it is shown how the various FTTx concepts have evolved over the past 30 years [6].
342
Wolfgang Fischer
16.4 Deployment Considerations From a civil engineering perspective the topologies for point-to-point fiber deployments are identical to those for PON. From the POP location individual feeder fibers for each subscriber are deployed toward some distribution point in the field – typically a splice point – either in some underground enclosure or a street cabinet. From this distribution point drop fibers are laid toward each individual household. As fiber densities in the feeder and drop part are very different, often different cabling techniques are employed, depending on the specific circumstances. Cabling techniques include classical cables and blown fiber in microducts. Classical cables can either be directly buried or accommodated in ducts for greater flexibility. In the feeder part deployments can be greatly facilitated not only by existing conventional ducts but by other rights of way, like sewers, tunnels, or other available tubes. Thus, in many cases the higher number of feeder fibers does not pose any major obstacle for point-to-point installations. In the POP the fibers arriving from the outside plant are terminated on an optical distribution frame (ODF) as the fiber management solution which allows to flexibly connect any customer to any port on switches in the POP. Due to the large number of fibers to be handled in a POP the density of such a fiber management solution has to be very high in order to minimize real estate requirements. Figure 16.1 shows a high-density ODF that allows to handle more than 2300 fibers in a single rack. For illustration purposes it is positioned next to a rack with active equipment that can terminate 1152 fibers on individual ports. Take rates in FTTH projects typically take some time to ramp up and usually stay well below 100%. The fiber management allows a ramp up of the number of active ports in sync with the activation of customers. This minimizes the number of unused active network elements in the POP and enables a slow ramp up of the investments.
Fig. 16.1 High-density fiber management solution (source: Huber & Suhner)
16 Point-to-Point FTTx
343
IPTV-based video solutions provide superior features over simple broadcast solutions and have, therefore, become an indespensable part of any triple-play offering. Quite frequently, however, there is a need to provide additional RF video broadcast overlays in order to support existing TV sets in subscribers, households. In tree architectures this is typically accomplished by providing an RF video signal, compatible with cable TV solutions, over an additional wavelength at 1550 nm. In point-to-point fiber installations this can be achieved by two different approaches, depending on the possibilities for fiber installation: • In the first approach, an additional fiber per customer is deployed in a tree structure (see Fig. 16.2) and carries only an RF video signal that can be fed into the inhouse coax distribution network. In this case, the split factors (e.g., ≥ 128) exceed those typically used for PONs so that the number of additional feeder fibers is minimized.
Fig. 16.2 RF video overlay using a second fiber per subscriber, deployed in a tree structure
• In the second approach, a video signal is inserted into every point-to-point fiber at 1550 nm. Figure 16.3 shows how the RF video signal carried by a dedicated wavelength from a video-OLT is first split into multiple identical streams by an optical splitter and then fed into each point-to-point fiber by means of triplexers. Mechanically, this solution is implemented in structures similar to optical distribution frames. On the customer side the wavelengths are separated, the 1550 nm signal converted into an RF signal for coax distribution and the 1490 nm signal made available on an Ethernet port. In both cases, the CPE/ONT devices comprise two distinct parts: • a media converter which takes the RF signal at 1550 nm and converts it into an electrical signal that drives a coax interface • and an optical Ethernet interface into an Ethernet switch or router.
344
Wolfgang Fischer
Fig. 16.3 Insertion of RF video signal into point-to-point fibers
In the single-fiber case, the signals are separated by a triplexer built into the CPE, while in the dual-fiber case there are individual optical interfaces for each fiber.
16.5 Operational Considerations 16.5.1 Traffic Management For obvious reasons all communication networks are oversubscribed, starting at the uplink interface of the access network element. In shared media architectures oversubscription already occurs in the access network based on appropriate MAC protocols. Management of oversubscription in packet networks requires queueing and priority mechanisms. As a general rule, such queueing and priority implementations in switches and routers can be made significantly more sophisticated and effective than those based on MAC protocols. Using a dedicated medium like a point-to-point fiber, therefore, removes the need for a MAC protocol in the access network, and it removes also an additional point of congestion to be managed within the entire context of traffic management. Access network elements follow fast innovation life cycles and are replaced or upgraded in regular intervals which are much shorter than the lifetime of the fiber. Therefore, any bottlenecks in the active infrastructure can be removed on a regular basis, but this should not have any impact on the passive access infrastructure.
16.5.2 Security A dedicated fiber is inherently secure because the information transmitted over it can only be received at either end of the fiber, and it will not be shared,
16 Point-to-Point FTTx
345
by default, with any other subscriber in the same access network domain. This inherent security, therefore, obviates the need for encryption on the fiber which is a necessary function in any shared medium architecture. In [7], the authors discuss the encryption aspects of PON and show how cost trade-offs in the implementation of encryption schemes can lead to serious security problems. A failed ONT/CPE device connected to a dedicated fiber cannot impact the traffic of any other subscriber because any unplanned behavior can be detected by the associated switch port in the POP, and the port can be deactivated until the defect is repaired. This property also eliminates denial of service attacks which can be launched by malicious users in shared medium architectures by deliberately jamming the signals.
16.5.3 CPE Deployments For the deployment of the CPEs – either simple ONTs with integrated Ethernet switches or more sophisticated routed home access gateways – the service providers have the choice of two different scenarios: • They own and install the CPE by themselves and also test the integrity of the transmission. In this case, the subscriber has no need to touch the fiber in any way but only connects his home network to the subscriber-facing interfaces of the CPE. • They can drop-ship the CPEs to the subscribers and have them connect their CPEs via optical patch cables to optical wall outlets. This requires more confidence in the subscribers, capabilities to handle optical fibers. Eventually, this will also allow CPEs to be distributed over retail channels. While there are operational pros and cons for either solution, with dedicated fiber in no case is there any security risk involved in having the subscriber handle the fiber. Any potential problem is strictly confined to this particular subscriber access line. One of the main questions in this context is rather whether the savings from self-installation can compensate the potentially higher cost of support calls.
16.5.4 Trouble-Shooting Optical time domain reflectometry (OTDR) mechanisms are used to determine any discontinuities or reflections in the fiber plant. OTDR transmits light pulses into the fiber, and the timing and intensity of the reflections indicate the location and the nature of any problem. In a point-to-point deployment the fiber is visible on its entire length from the POP to the subscriber, which greatly facilitates trouble-shooting of the fiber plant, compared to point-to-multipoint architectures which create ambiguous results for the reflections coming from the drop fibers.
346
Wolfgang Fischer
16.5.5 Power Budget Planning The Ethernet transmission technologies typically employed for point-to-point connectivity provide sufficient margin, even at the nominal maximum distances, such that a substantial number of splices and connectors can be accommodated without exceeding the maximum channel insertion loss. This facilitates the planning of the outside plant and allows for some aging of the passive components.
16.6 Cost Considerations 16.6.1 Capital Cost FTTH deployment involves a number of different cost components that can each be individually optimized. However, it is important to understand the relative contribution of each component and, thus, the relative saving potential. Figure 16.4 shows a typical cost distribution for greenfield FTTH deployments.
Fig. 16.4 Typical initial capex distribution for greenfield FTTH deployments
This graph confirms what intuitively could be expected: civil works comprise almost 70% of the total initial capex. Obviously, this is the cost component where saving efforts have the largest effect. Therefore, usage of every potential right-of-way solution should be considered in order to reduce this component. As already mentioned in Section 16.4, this comprises existing ducts, sewers, tunnels, etc. With 6% and 2%, respectively, the fiber and other passive optical elements only contribute a very small part to the capex. Therefore, the saving potential from these components is very limited. Active network elements are the second largest component with 12% contribution. Independent of the particular technology employed, this is a component where technological progress will continue to bring down per-port cost.
16 Point-to-Point FTTx
347
In the case of Ethernet point-to-point architectures the access switches in the POPs are usually derived from systems that are deployed in very large quantities in enterprise networks. Therefore, their cost benefit very strongly from manufacturing volumes which are significantly larger than they would be in the case of serviceprovider-only equipment. Furthermore, their technological evolution can be more broadly funded which leads to rapid innovation cycles and cost reduction. Based on detailed business case analyses with European service providers some typical values for the cost differences between point-to-point and point-to-multipoint deployments have been derived [8]. In those cases where just sufficient duct space in the feeder plant was available to deploy the smaller cables for point-to-multipoint architectures, but not enough for point-to-point, the cost premium for Ethernet pointto-point could run up to 25% because of the need for additional civil works. Experience with real deployments, however, implies that this situation occurs far less frequently than situations where there is either no duct shortage, or where civil works have to be carried out because either all the ducts are occupied by other cables, or in greenfield scenarios where no infrastructure is available in the first place. In those cases, the cost premium for point-to-point deployments usually stays below 5%. This difference can be mainly attributed to the more extensive fiber management in the POPs. It can be expected that these initial small project cost differences will be overcompensated, over the lifetime of the fiber plant by the inherent benefits of point-to-point deployments (flexibility, upgradability, fiber plant maintenance, etc.).
16.6.2 Operation Cost Operation costs are a multi-faceted subject. Most of these cost items are not specific to any particular access technology, like marketing, subscriber acquisition, and subscriber management. It is very difficult, though, to quantify operational cost differences. Therefore, we try to identify qualitatively the technology and architecturedependent aspects that can impact the operation cost of an FTTH access network. Aspects which are favorable for point-to-point deployments are mainly due to the relative simplicity of the architecture, as there are • • • •
ease of traffic management no encryption ease of trouble-shooting the physical layer easy upgrade to higher speeds/new technologies on a per-customer basis.
Certainly, there are also some operational disadvantages of point-to-point deployments: • More real estate in the POP location due to the more extensive fiber management • Slightly more space requirements for the active equipment, although for typical penetration rates the difference is rather small as only active subscribers require switch ports, in contrast to point-to-multipoint architectures where the first subscriber on a tree requires the allocation of an OLT port
348
Wolfgang Fischer
• Higher power consumption in the POP as every active subscriber is connected to a dedicated port, although this aspect is mitigated by the penetration rate.
16.7 Open Fiber Access Due to the high cost of FTTH deployments open access models have become very popular in many regions of the world because they allow the sharing of the cost for the infrastructure among multiple service providers. An infrastructure can be opened at various layers. The lowest layer in this context is the duct which allows to deploy parallel infrastructures – particularly in the feeder part of the access network – by sharing investment into civil works. On the other side of the spectrum, opening a transport network at a logical level is usually done either on layer 2, by providing VLAN-based access of multiple service providers to subscribers, or on layer 3 by IP-based mechanisms. Between these options there is the concept of open access to fibers. Open fiber access allows the sharing of a common passive infrastructure by multiple network operators, each of them owning and operating their own active network elements and thus providing individual transport networks. This is particularly relevant in scenarios where the passive infrastructure is built with the involvement of public organizations, and where access to this infrastructure has to be provided on non-discriminatory conditions to multiple players (see, e.g., [1]). Some vertically integrated service providers have also realized that renting dark fiber to other players can be a viable business model. FTTH deployments by any provider in a certain area usually cover 100% of all homes in order to avoid repetitive civil works. On the other hand, take rates per provider usually are far below 100% so that a large proportion of fiber would be unused. Being able to rent this fiber to other players can help improving the business case. Providing dark fiber instead of logical wholesaling also allows competition on speeds and quality, and it has the potential to optimize the network architecture for retail rather than wholesale services. Open access to fiber is achieved in a way conceptually similar to local loop unbundling for copper loops. Figure 16.5 shows how the ODF in the POP allows to flexibly connect individual subscribers to the access switches of multiple providers. ONT
OLT
ODF
ONT
1 fiber per Subscriber
ONT POP
Fig. 16.5 Open access to fiber
ONT
16 Point-to-Point FTTx
349
Customer churn just requires patching optical connections at the ODF in the POP.
16.8 Transmission Technologies Recognizing the need for Ethernet in service provider access networks IEEE had established the IEEE 802.3ah working group already in 2001, creating a standard for “Ethernet in the First Mile (EFM).” Besides standards for OAM, Ethernet over copper and EPON, two standards for fast Ethernet and Gigabit Ethernet over singlemode single fiber were created. The EFM standard was approved and published in 2004 and was included into the base IEEE 802.3 standard in 2005. The specifications for the transmission over single-mode single fiber are called 100BASE-BX10 for fast Ethernet and 1000BASE-BX10 for Gigabit Ethernet. Both specifications are defined for a nominal maximum reach of 10 km. For the separation of the directions on the same fiber wavelength-division duplexing is employed, such that for each of the bitrate classes two specifications for transceivers are defined, one for “Upstream,” i.e., from the CPE toward the POP, and one for “Downstream,” i.e., from the POP toward the CPE. Table 16.2 provides the fundamental optical parameters of these specifications. Table 16.2 Optical parameters of EFM specifications Description 100BASE-BX10-D 100BASE-BX10-U 1000BASE-BX10-D 1000BASE-BX10-U Transmit direction Nominal transmit wavelength Minimum range Maximum channel insertion loss
Downstream
Upstream
Downstream
Upstream
1550 nm
1310 nm
1490 nm
1310 nm
0.5 m–10 km
0.5 m–10 km
0.5 m–10 km
0.5 m–10 km
5.5 dB
6.0 dB
5.5 dB
6.0 dB
In order to cope with requirements not considered in the standard the market also offers optical transceivers with non-standard characteristics. Some types can bridge significantly longer distances, e.g., for deployments in rural areas. As the nominal transmit wavelength of 100BASE-BX-D (1550 nm) is the same as the standard wavelength for video overlays in PON systems, transceivers exist which can transmit at 1490 nm, which allows to insert an additional signal at 1550 nm carrying an RF-modulated video overlay signal on the same fiber using off-the-shelf video transmission equipment (see Fig. 16.3).
350
Wolfgang Fischer
16.9 Outlook A dedicated fiber per subscriber does not pose any practical limit on the bitrates that can be provided over this medium. For Ethernet there is a clear evolution path to scale speeds to dimensions that are currently not required for broadband access. However, as these technologies mature and become viable from a cost perspective they can also be deployed in the access network providing 10 Gbit/s and more per subscriber to keep pace with ever growing bitrate requirements. There will be situations where the deployment of point-to-point fiber is not viable due to limitations in the physical infrastructure (ducts, etc.). These situations are usually considered application scenarios for PONs because PONs allow to save significant amounts of fibers in the feeder part, facilitating the installation of the (smaller) cables in existing ducts. Furthermore, some service providers plan to reduce the numbers of their existing central offices by leveraging the longer achievable loop lengths of fiber-based access networks. Depending on the specific circumstances, the number of feeder fibers that can be brought into such a central office and the associated fiber management may turn out to become challenging. Passive aggregation points in the fiber plant can help in reducing these requirements and allow for building topologies for fiber access networks which are different from those in the copper era. Combining the best of both worlds – sharing feeder fibers while still providing dedicated point-to-point connectivity – is promised by the emergence of WDM PONs. These architectures provide dedicated and transparent connectivity on a wavelength per subscriber and, thus, allow very high uncontended bitrates for each connected subscriber, and provide the same inherent security as with dedicated fiber. These architectures use wavelength filters instead of splitters in the field to map each wavelength on the feeder fiber into a dedicated drop fiber. Optical transceivers have to be kept generic instead of wavelength specific in order to facilitate wavelength management. A detailed description of these mechanisms is beyond the scope of this chapter.
16.10 Summary Fiber to the home architectures based on point-to-point fiber deployments are mainstream in Europe and Middle East because they have been technically and economically viable for a long time, and they have a number of characteristics that make them the preferred choice when technological flexibility over the lifetime of the fiber plant is required, or when open access to fiber is required either from a regulatory or business case perspective. The active access network elements are based on established and mature Ethernet technologies which are not only deployed in service provider networks, but which are used in even larger volumes in today’s enterprise networks. This fact makes the solutions particularly cost-effective and guarantees ongoing evolution.
16 Point-to-Point FTTx
351
References 1. “BBned’s open access network – based on Cisco Metro Ethernet Point-to-Point – challenges established market with three-layer model,” Cisco case study on BBnde/Amsterdam CityNet, on www.cisco.com, 2008. 2. “Carrier-Ethernet Puts Triple-Play Provider in the Lead,” Cisco case study on Lyse Tele Norway, on www.cisco.com, 2007. 3. “French ‘Triple-Play’ Service Provider Deploys Fiber to the Home,” Cisco case study on Free in France, on www.cisco.com, 2008. 4. E. Kattirdji, “Cyta’s Experience with FTTH Networks,” in Proc., FTTH Council Europe Conference, Paris, Feb. 2008. 5. M. Poga˘cnik, “FTTx in Telekom Slovenije,” in Proc., FTTH Council Europe Conference, Paris, Feb. 2008. 6. P. W. Shumate, “Fiber-to-the-Home,” IEEE/OSA Journal of Lightwave Technology, vol. 26, no. 9, pp. 1093–1103, May 2008. 7. D. Gutierrez et al., “TDM-PON Security Issues: Upstream Encryption is Needed,” in Proc., Optical Fiber Communication Conference (OFC/NFOEC), paper JWA-83, March 2007. 8. “Fiber to the Home: Technology Wars,” Cisco IBSG Economic Insight, on www.cisco.com, 2007.
Chapter 17
Broadband Access Networks Klaus Grobe, J¨org-Peter Elbers, and Stephan Neidlinger
Abstract Emerging broadband services are driving a new generation of metro access and backhaul networks. Residential bandwidths of up to 100 Mb/s and business services up to 10 Gb/s have to be supported on a broad scale. In order to reduce network operators cost, unified access/backhaul architectures which support all services, applications, and bit rates and replace application-specific and legacy solutions have to be implemented. These networks should be based on a future-proof WDM infrastructure for scalability and cost reasons. Migration can start with passive WDM technology. The WDM systems can then be complemented by ADM-like TDM functionality for GbE services and Layer-2 Ethernet switching functionality. In later migration steps, point-to-multipoint logical connectivity can be implemented on a per-wavelength basis. Optional WDM amplification and protection enhance maximum reach and service availability. When wavelength grid and per-channel bit rate of these WDM-PONs are optimized, cost points of standard GPON can be reached while at the same time improving total capacity, splitting ratio, and reach.
Klaus Grobe ADVA AG Optical Networking, Fraunhoferstr. 9a, 82152 Martinsried/Munich, Germany, e-mail:
[email protected] J¨org-Peter Elbers ADVA AG Optical Networking, Fraunhoferstr. 9a, 82152 Martinsried/Munich, Germany, e-mail:
[email protected] Stephan Neidlinger ADVA AG Optical Networking, Fraunhoferstr. 9a, 82152 Martinsried/Munich, Germany, e-mail:
[email protected]
A. Shami et al. (eds.), Broadband Access Networks, Optical Networks, c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-92131-0 17,
353
354
Klaus Grobe, J¨org-Peter Elbers, and Stephan Neidlinger
17.1 Broadband Drivers and Network Requirements 17.1.1 Broadband Drivers Broadband access networks are driven by business and residential end-user demand for voice, video, data, and mobile services. Corresponding service offerings for residential access are sometimes referred to as triple or (when including mobile services) quadruple play. Residential households demand faster Internet access. In addition, they request broadband connectivity to exchange digital images and videos, or participate in online games and other bandwidth-intensive peer-to-peer applications. Recent reports place teleworking, telemedicine, and home assistance as further applications which are driving the need for broadband connectivity [1]. The demand is forcing carriers to scale their networks for provisioning of triple/quadruple play services and for accommodating future growth. In the near future, residential service offerings for an individual household will comprise • Video (broadcast IPTV, video-on-demand): 2 × High-definition television channels (8 . . . 16 Mb/s per channel), 4 × Standard-definition television channels (1.5 . . . 5 Mb/s per channel) • High-speed Internet (HSI) access: up to 5 Mb/s • Voice services: Voice over IP (VoIP), plain old telephone service (POTS), analogue), and integrated services digital network (ISDN), in total up to 100 kb/s • High-speed upstream services: Image and video upload, file sharing/storage, and home-/teleworking (up to 100 Mb/s) In particular high-speed upstream services call for an upgrade of the existing access network infrastructure. Optical fiber technologies enable symmetric upstream/downstream data rates of 100 Mb/s and beyond and facilitate a high quality of service (QoS). Business customers demand fast and managed connectivity services. Optical fiber technology has been used for business applications for many years. Examples are • TDM leased lines (e.g., on SDH/PDH/ATM level): e.g., for connecting corporate sites on a permanent basis. These services were often E1/DS1 based and are now migrating to Ethernet (e.g., 100 Mb/s Ethernet). • Local area network (LAN) interconnection: interconnections of LANs of different corporate sites (typically with 100 Mb/s or 1 Gb/s Ethernet). • Virtual private networks (VPN): e.g., nation- or continent-wide corporate backbone networks (typically 100 Mb/s or 1 Gb/s Ethernet). • Storage area networks (SAN): e.g., interconnection of storage facilities of data centers (1 or 10 Gb/s Ethernet, 2 . . . 8G Fiber Channel, InfiniBand with 10 . . . 40 Gb/s). In the context of quadruple play, network operators are investigating options to use their fixed access network infrastructure for mobile backhaul as well. The main
17 Broadband Access Networks
355
objective is to connect base stations to the backbone network. Using the fixed access network also for mobile backhaul is cost-efficient since mobile base stations are often located on top of apartment and business buildings which are already connected to the access network.
17.1.2 Network Architecture and Requirements Metro area networks have been described in detail in [2, 3]. In general, metro core networks connect to the national backbones via at least two core PoPs (points of presence) which may also host a BRAS (broadband remote access server). The core PoPs are collocated or connected to Ethernet aggregation switches. The structure of a generic metro network is shown in Fig. 17.1. In most cases, the metro core infrastructure is based on DWDM collector rings. The network roll-out started with static rings in the late 1990s, and today there is a trend toward rings based on reconfigurable optical add/drop multiplexers (ROADMs). The rings connect the metro PoPs (Aggr. 1) which also host aggregation equipment such as Ethernet switches or legacy SONET/SDH equipment, to the primary aggregation switches (Aggr. 2) or the core PoPs . The metro PoPs are also the logical hubs for a subtended aggregation layer, which can be based on SONET/SDH or CWDM rings or other suitable access technologies. These secondary-layer rings connect local exchanges (Local X) and hub them to the metro POPs.
Fig. 17.1 Layered structure of generic metro area network including two aggregation stages (VDSL = very high-speed digital subscriber line, ADSL = asymmetric digital subscriber line). Also shown are site numbers and distances between sites which are applicable to typical ILEC networks
Real-world metro access/backhaul networks may deviate from the generic description, but in most cases a layered network with two aggregation layers is implemented, e.g., [4]. The common feature of these networks is that different access/backhaul technologies are combined, e.g., SONET/SDH, WDM, resilient packet ring, active Ethernet P2P. The mixture of technologies often creates operational inefficiencies and makes poor use of the fiber infrastructure. Different access network topologies must be supported. For example, backhauling of fixed and mobile services can be based on rings, which connect several clients to the local exchange. Using different wavelengths, these rings can provide resilient backhauling for independent wireless and wireline services and may use simple time-division multiplexing (TDM) add/drop techniques for efficient wavelength
356
Klaus Grobe, J¨org-Peter Elbers, and Stephan Neidlinger
usage. Residential and business access on the other hand is based on tree or bus architectures due to topology and/or security reasons. Emerging requirements for next-generation optical access networks can be summarized as follows: • Next-generation services require symmetric access bit rates of up to 100 Mb/s per residential end-user, and 100 Mb/s . . .10 Gb/s for business users. • Fiber needs to move closer to end customers: The first step will be fiber-to-thenode (FTTN) architectures with first-mile technologies such as xDSL, coaxial cables, power line transmission, or wireless. The second step will be fiber-to-thebuilding (FTTB) and/or fiber-to-the-home (FTTH). • Smart demarcation, management, and OAM functions are generally required on the optical and Ethernet layers for simplified and automated network operation. • FTTN networks need efficient backhaul that supports bandwidth scalability, service transparency, network resilience, improved fiber/bandwidth utilizations (i.e., WDM with TDM functionality), and optimized Ethernet efficiency (i.e., packet aggregation). • The reach of optical access technologies must increase to ∼100 km in order to enable a reduction in the number of central and local exchange offices. • A migration to a converged platform for access, metro core, and regional network layers is required in order to improve time-to-market and reduce operational expenditures (simplified logistics, spares handling, inventories, maintenance, and training of personnel). • Fixed-mobile convergence (FMC) as well as high-margin enterprise services need to be supported on the same unified wireline infrastructure as residential services. • Optical wireline should be given preference over wireless access wherever possible because of overall lower network power consumption.
17.2 Scalable Broadband Access Networks Combinations of optical and Ethernet transport lead to cost-optimized, unified metro backhaul (and access) infrastructures. WDM offers low-cost transport for all applications and services, scales easily in terms of capacity and reach, and provides fast and cost-efficient protection against fiber plant failures. Fully transparent to any bit rate and protocol, WDM is the integration layer for all services. Ethernet is established as the preferred data-link protocol for carrier networks, including next-generation infrastructure for mobile, triple/quadruple play, business services, and native packet services (E-Line and E-LAN). It provides subwavelength aggregation, customer separation, and service management, including powerful OAM and demarcation functions. The combination of both, optical and Ethernet technologies, allows integrated end-to-end service provisioning and management which can be automated through a common control plane (GELS, GMPLS-enabled Ethernet label switching). In
17 Broadband Access Networks
357
addition, a single DCN (data communications network) and a single, centralized management system (NMS, network management system) can be used.
17.2.1 xWDM in Access Networks SONET/SDH and ATM have long been used in legacy networks for metro core and backhaul applications. In the metro core, these technologies have been replaced by Dense WDM (DWDM) for reasons of service/protocol transparency and scalability. For backhaul applications, CWDM transport has proven less costly and enables larger bandwidths. By combining optical transport with Ethernet functionality, service providers can guarantee high service availability, coupled with scalability, improved operational efficiency, and future flexibility. These solutions are sometimes referred to as optical+Ethernet architectures. Figure 17.2 shows a metro area network with DWDM metro core and CWDM backhaul rings. New DSLAM (digital subscriber line access multiplexer) technology uses Ethernet with GbE uplinks. Related DSL aggregation technologies such as broadband services routers (BSR) follow this trend. Ethernet is the single most important Layer-2 protocol all the way to the customer premises. In backhaul networks, the fine granularity of SONET/SDH transport has become an expensive liability. It is more costeffective to use CWDM with GbE connectivity and processing on the wavelength level instead.
Fig. 17.2 WDM metro area network with DWDM core and CWDM backhaul rings (adapted c from [15], 2008 IEEE)
358
Klaus Grobe, J¨org-Peter Elbers, and Stephan Neidlinger
Utilizing wider wavelength spacing than DWDM, CWDM uses lower cost components to deliver up to 10 Gb/s per wavelength. The resulting systems are comparatively low in price; making them a good fit even for fiber-to-the-node (FTTN) backhaul where cost sensitivity is very high. In addition, the use of separate wavelengths allows easy access unbundling and wholesale service offerings. WDM also inherently offers continued support of older DSLAMs using SONET/SDH, ATM, and other legacy technologies.
17.2.2 TDM Add/Drop Multiplexing Adding time-division multiplexing (TDM) functionality to CWDM allows multiple DSLAMs to be connected to the service node at a metro core hub using a single wavelength on a CWDM ring. This has obvious advantages for future scaling. With SONET/SDH ceasing, part of the related functionality is still required. An example is flexible add/drop functionality for GbE services. GbE-ADM (add/drop multiplexing) functionality was first introduced in 2006 [5]. Four wire-speed GbE services are multiplexed onto a 4.3-Gb/s wavelength for CWDM or DWDM transport (see Fig. 17.3). The GbE-ADM concept allows different approaches for protecting services against network and equipment failures. Since protection against fiber cuts is the main requirement, it is most cost-efficient to perform protection at the optical layer. An ADM ring architecture allows 1+1 optical line protection by sending the 4.3-Gb/s aggregate wavelength signal both ways around the ring. Switchover in a failure event must be fast, within the SONET/SDH requirement of 60 ms (10 ms detection, 50 ms switching time), and is performed by the receiving ADM. The GbE-ADM architecture also supports drop and continue of GbE services. This has proven to be very efficient in provisioning TV broadcast services. Nodes then access the required GbE stream, while it is simultaneously passed on to the next node.
Fig. 17.3 Illustration of GbE-ADM functionality
17 Broadband Access Networks
359
CWDM access rings with GbE-ADM function offer a set of advantages: • • • • •
High transport capacity at low OpEx and CapEx Wire-speed GbE services with node pass-through, add/drop, or drop and continue Low latency, high QoS, and low packet loss Easy network overlays for different (legacy) services or different operators No need for optical link engineering, optical amplifiers, or power balancing (less operational complexity) • Fast optical-layer protection • Efficient backhaul of multiple DSLAMs per collector node In summary, the combination of CWDM with flexible TDM (GbE-ADM) allows scalable, flexible access, and backhaul infrastructures at low cost per throughput unit.
17.2.3 Layer-2 Packet Aggregation The integration of Layer-2 packet aggregation with WDM transport increases transport efficiency further, when compared to WDM with integrated TDM/ADM. This is particularly useful for aggregating partially filled GbE streams. Examples are VDSL access multiplexer line ports or residential fast Ethernet services which make use of GbE PHYs. Packet aggregation takes several partially filled GbE streams and statistically multiplexes them into a single GbE or 10GbE stream. Thus, network interfaces in backhaul transport systems are utilized to a higher degree and the total number of interfaces is reduced, yielding CapEx and OpEx savings. WDM rings with packet aggregation offer a set of advantages: • Low-cost transport for high numbers of GbE streams • Optimized transport of partially filled GbE streams by using statistical multiplexing gain • Layer-2-based aggregation into 10Gb/s, integrated in WDM transport • Multicast support • Optimized utilization and reduced number of backhaul network ports In addition, Layer-2 packet aggregation can provide extensive OAM functions on the Ethernet layer, allows customer separation on the flow or VLAN level, and can offer built-in multicast support. Integrating WDM and Layer-2 functionality helps eliminate dedicated access switches and thus purpose-built platforms in an access network. The use of Layer-2 switch cards as integrated blades of a metro WDM access ring is schematically shown in Fig. 17.4.
17.2.4 Deployment Case: Telecom Italia The Telecom Italia Group is the incumbent telecommunications provider in Italy and one of the major network operators worldwide. Telecom Italia deploys WDM
360
Klaus Grobe, J¨org-Peter Elbers, and Stephan Neidlinger
Fig. 17.4 Integrated Layer-2/WDM functionality in an access ring
technology for the metro network infrastructure in selected Italian cities. The rollout started with an eight node ring network in Rome in 2003. When Telecom Italia introduced a new generation of DSLAM technology based on Ethernet, the operator was modernizing its access network as well. The design goal was to optimize DSL backhaul by supporting GbE and 10GbE services most efficiently. Further requirements were scalability and support of legacy services for smooth migration. Fig. 17.5 shows a schematic of the Telecom Italia DSLAM backhaul network.
Fig. 17.5 Telecom Italia’s DSLAM backhaul network
The core part comprises DWDM rings. Each wavelength has a transport capacity of 10 Gb/s. The core rings are connected to BRAS via Layer-2/3 core switches. These switches serve as aggregation points in the network. On the access side, the core rings are connected to CWDM access rings. The interconnection is done via Layer-2 aggregation switches. Each CWDM wavelength is carrying four GbE signals which are time-multiplexed using ADM functionality which is integrated in the CWDM channel cards. There are several DSLAMs per site and several sites per ring. The network provides optical protection and full redundancy in the optical and the switch layer.
17 Broadband Access Networks
361
17.3 Next-Generation Access and Backhaul Next-generation access services will be based on FTTB and FTTH topologies. These use either active (Ethernet) P2P or passive optical networks technology (GPON, EPON today). Both technologies have different advantages and disadvantages, and it can be shown that WDM-PON combines the advantages of P2P and TDMA PONs while minimizing their drawbacks. From a fiber infrastructure perspective, WDM-PON is similar to GPON/EPON as it uses a tree-type fiber technology. A single fiber (pair) interconnects the OLT in the CO and a WDM filter in the remote node. The WDM filter multiplexes and demultiplexes the signals which come from and go to individual buildings or homes. Single or dual fibers can be used. The low fiber count in the feeder part of the network yields lower OpEx compared to active P2P architectures. From an individual connection perspective, WDM-PONs are similar to P2P architectures since every building/home is connected via a dedicated wavelength. Hence, scalable and transparent connectivity per customer is supported. Each customer can be upgraded independently of other users. High security and availability is supported due to logical and physical separation of the customers. Like P2P, port numbers and therefore equipment floor space and power requirements in the CO are higher compared to GPON and EPON. This effect can be mitigated if WDM-PON CO transceivers are integrated and/or WDM-PON OLTs are integrated into WDM backhaul platforms. In this case, only one integrated system is required in the CO which can significantly reduce the number of intra-office interfaces, floor space, and power consumption compared to dedicated system solutions. WDM-PONs support various service offering options: • Wavelength services, both managed and unmanaged, using colored interfaces in client equipment (e.g., switches/DSLAMs located in the basement of a building or in street cabinets) • Optical-layer managed services, using passive optical-layer performance monitoring for connecting client equipment • Ethernet/electrical-layer managed services, using network termination units to connect client equipment (e.g., Ethernet media converter) These services can be augmented by protection schemes in various flavors (e.g., dual homing, end-to-end, in access network, in feeder network). Since separate wavelengths are used for different users, WDM-PON is ideally suited for unbundling and open-access approaches. In this concept, several service providers may use one network platform operated by an infrastructure provider. The need to reduce OpEx leads to the requirement to increase the maximum reach of the access network, thus enabling the consolidation of real estates in this network area. Less buildings and sites reduce OpEx significantly. Figure 17.6 shows how fiber technologies reduce the amount of active equipment and real estate in access networks. Fiber-to-the-curb (FTTC) deployments with VDSL as last mile technology typically use Layer-2 aggregation in local offices. These L2 devices
362
Klaus Grobe, J¨org-Peter Elbers, and Stephan Neidlinger
Fig. 17.6 Telecom access solutions
are backhauled to a central office using ring architectures. In this approach, buildings are necessary at LO sites, and cabinets with active equipment at curb sites. GPON/EPON avoid active curb deployments. However, active equipment at LO sites is still necessary for backhauling. WDM-PON solutions allow maximum reach in excess of 100 km. Then, no active equipment at curb and LO sites is needed, reducing OpEx dramatically. Passive optical networks (PONs) have been discussed for various fiber-to-the-x (FTTx, where x can mean the home, curb, cabinet, or building) applications for more than a decade [6]. Many PON systems have been proposed, but only two standards of TDMA (time-division multiple access) have so far been used for mass roll-out. The ITU-T standards BPON (broadband PON, G.983) and its successor GPON (Gigabit PON, G.984 [7]) are deployed in parts of the United States and Europe. EPON (Ethernet PON, according to the Ethernet-first-mile standard IEEE 802.3ah [8]) is broadly deployed in Japan and Korea. Many agree that TDMA PONs cannot cope with the requirements of future network evolution, despite the trends to increase the bit rate to 10 Gb/s, e.g., 10GEPON [9], and to wavelength enhancement bands according to the ITUTG.984.5 standard. Future wavelengths additions are complicated by the required backward compatibility to the standard EPON/GPON channel plan. Only 4 . . . 8 CWDM (depending on actual specification) or more costly DWDM wavelengths can be added, see Fig. 17.7. In addition, power budget and splitting ratio are severely limited by the use of 1:N power splitters (for example, a 1:32 splitter imposes an insertion loss > 17 dB). Independent of first-mile technologies, future metro networks will need to converge toward a single platform for the main applications. This convergence is required to replace legacy solutions, minimize OpEx, and finally reduce the number of active sites which are used for concentration of services [10].
17 Broadband Access Networks
363
Fig. 17.7 Wavelength extension as discussed for EPON/GPON. Wavelengths blocked by the original GPON/EPON standard are marked dark
17.3.1 Migration toward WDM-PON In case of copper (VDSL), cable (HFC), and wireless (WiMax/3G/4G) access, fiberto-the-node/curb (FTTN/FTTC) architectures are required to enable sufficient data rates in the local loop. With fiber-to-the-building/home (FTTB/FTTH) employing P2P Ethernet, EPON, GPON, or next-generation PON technologies, the connection between the customers premises and the central office is all-fiber. Depending on existing infrastructure, customer density, and network topology, service providers may use a mix of different first-mile technologies. In any case, operators have to bring fiber closer to the customers in order to implement scalable solutions for the second mile (i.e., backhauling of first-mile traffic). Passive WDM is a suitable technology for consolidating backhaul traffic onto a common transport infrastructure. It enables logical point-to-point connections with 1 . . . 10 Gb/s per wavelength over ∼100 km. It minimizes active equipment in the field, reduces the number of local exchange offices, and thus also minimizes power consumption in the network. It provides efficient fiber usage in the feeder network. In combination with integrated L1/L2 multiplexing (TDM, Ethernet) and optical protection, passive WDM facilitates scalable and transparent service delivery for residential, business, and carrier applications. Augmented with functions for remote CPE management and Layer-0/1/2 demarcation, it provides a self-contained transport solution, which, by integration with metro/regional transport equipment in the central office, can also minimize the number of O/E/O interfaces. Passive WDM access is shown in Fig. 17.8.
Fig. 17.8 Metro access network based on passive WDM
364
Klaus Grobe, J¨org-Peter Elbers, and Stephan Neidlinger
17.3.2 WDM-PON Concepts In a WDM-PON, ONUs are assigned individual wavelengths. This provides high dedicated bandwidth to each ONU. Each ONU operates on the individual bit rate rather than the aggregate (WDM) bit rate. Since ONUs are separated via physical wavelengths, aspects of privacy/security, network integrity, and simple active-layer access are intrinsically accounted for. WDM-PONs can also be combined with additional TDMA similar to the techniques used for EPON and GPON. This leads to hybrid WDM-TDMA PONs and improves scalability by allowing splitting ratios of up to 1:1000. It means that the respective TDMA technology must be running on any pair of the DWDM or CWDM wavelengths, instead of the standard (GPON/EPON) downstream/upstream wavelengths. It also means that both, WDM filters and subsequent TDMA splitters are traversed. This increases splitting ratios, but because of accumulated losses, APD receivers, additional FEC, and/or additional amplifiers must be integrated. WDM amplification can also support enhanced distances > 100 km. This leads to the concept of “active PONs,” which play an important role in future metro access and backhaul convergence scenarios [10]. Many proposals for WDM-PONs have been made in the last years, for examples refer to [11–14]. These almost exclusively assume dense or ultra-dense WDMPONs. While many of the proposals are technically interesting, it is clear that cost and performance, i.e., bandwidth per user, splitting ratio, and maximum reach, are the dominant criteria for commercial success. The basic WDM-PON architecture is shown in Fig. 17.9. The United States makes use of wavelength division multiple access (WDMA). Every ONU is assigned a dedicated wavelength and hence, bandwidth. Consequently, a WDM-PON provides point-to-point connections on the wavelength level.
Fig. 17.9 Basic WDM/WDMA PON architecture
In order to support all metro access and backhaul applications flexibly, the PON topology has to be generalized. Figure 17.10 shows options for both, the physical fiber and the per-wavelength topologies [15, 16]. Now, point-to-point and point-tomultipoint topologies are supported as well as access rings which are physically hubbed in the RN and logically hubbed in the OLT. If fiber topology does not support physical rings, wavelengths can be patched through to form a collapsed ring according to Fig. 17.10D.
17 Broadband Access Networks
365
ONU
ONU
OLT
OLT
ONU
OLT
ONU
ONU
ONU
ONU
ONU
ONU
OLT
ONU
ONU
ONU
Fig. 17.10 Basic WDM-PON topologies. (A) Point-to-point on wavelength level; (B) point-tomultipoint (on a shared wavelength); (C) ring behind RN (on a shared wavelength); (D) collapsed ring behind RN (on a shared wavelength)
Since WDM-PONs support all access and backhaul applications, different protection options can be used to cope with different availability and cost requirements. Relevant protection options are summarized in Fig. 17.11. The simplest protection variant, Fig. 17.11A, consists of fiber protection for the link OLT-RN only. This link carries the optical multiplex section, hence any fiber failure here would affect the entire PON. In urban areas, fiber failures lead to the most severe drop in path availabilities. The fiber protection can be implemented very cost-efficiently, still keeping the RN fully passive and using OLT-ended 1:1 link protection switching for the DS and 1+1 link switching for the US signal. Obviously, the dedicated links between RN and the ONUs are left unprotected.
Linear OLT-RN
Linear End-to-End
Ring / collapsed ring
Dual homing
Fig. 17.11 WDM-PON protection options. (A) fiber protection for link OLT-RN; (B) end-to-end link protection; (C) ring protection; (D) fully redundant with dual homing
366
Klaus Grobe, J¨org-Peter Elbers, and Stephan Neidlinger
In case higher path availabilities are required, e.g., for broadband business access, more extensive protection options can be provided. The fiber protection OLTRN can be complemented by additional protection of the links RN-ONU, see Fig. 17.11B. This protection can be operated single-ended from the respective ONU. For the access rings already discussed (Fig. 17.10C), protection is relatively simple and straightforward, see Fig. 17.11C. The RN and the ONUs form an access ring where 1+1 protection can be provided on the link level. This can be combined with protection on TDM time slot level. The ring configuration avoids the need for a TDMA scheme for wavelength sharing between ONUs. As compared to standard ring protection, the link OLT-RN must be considered as well. This requires disjoint paths between OLT and RN, and wavelength conversion somewhere in the ring if single-fiber working (SFW) is used for the link OLT-RN. Finally, highest availability can be achieved by providing full end-to-end redundancy. When duplicating the entire OLT, this can also be combined with dual homing, see Fig. 17.11D. In addition to passive components for protection, the RN should also be allowed to accommodate certain active components. This allows fully flexible adaptation to different applications requirements. It provides electrical functionality where reasonable and can help reduce complexity of access networks. Options for the RN are summarized in Fig. 17.12. The RN may contain WDM amplification in order to achieve high link lengths or power budgets, or to compensate for the loss of added splitters used for per-wavelength TDMA, Fig. 17.12B. To a certain extent, it may be possible to keep the RN quasi-passive by remotely pumping active fibers in the RN. For full flexibility regarding reach and amplifier placement, however, lumped amplifiers in active RNs may be necessary.
Fig. 17.12 Remote node options: (A) fully passive; (B) with (locally pumped) WDM amplification; (C) with Layer-1 regenerators and/or TDM (grooming) functionality; (D) with Layer-2 switching
17 Broadband Access Networks
367
An alternative to amplification is using low-cost 3R regenerators in the RN, see Fig. 17.12C. An interesting application of such regenerators is wavelength conversion to a standard CWDM/DWDM grid to support backhauling of stacked GPONs or EPONs in the fist mile. Lean aggregation of access services can be achieved by using Layer-2 functionality on WDM line cards which are accommodated in the RN, see Fig. 17.12D. This allows service provisioning, aggregation, link extension, and demarcation with an integrated WDM/Layer-2 access system, under a single end-to-end network and service management system.
17.3.3 WDM-PON Analysis Despite significantly higher bandwidth compared to GPON/EPON, DWDM-PONs lead to increased cost because they use DWDM transmitters. Hence, CWDM must be considered as a lower cost alternative. CWDM-PONs may use 18 CWDM wavelengths to support 9 ONUs in an SFW configuration. This is the entire CWDM wavelength range defined in [17] and requires low water-peak G.652C or G.652D fibers. A mechanism to reduce WDM-PON cost is sharing wavelengths between several clients, either using (point-to-multipoint) TDMA or (ring-based) TDM. As shown in Fig. 17.13, the per-ONU-link cost (i.e., the dual-ended cost per PON client) decreases dramatically when additional TDMA (TDM) is applied. It is also shown that DWDM is significantly more expensive than CWDM. CWDM combined with successive moderate splitting ratio (1:4, 1:8) can achieve price points known from GPON. The addition of amplification to a DWDM-PON does not contribute to cost significantly, given the PON is fully loaded with wavelength channels [15]. The ONU link costs shown in Fig. 17.13 include the respective portion of common components like splitters and shelves. The costs, C, have been derived as the
Fig. 17.13 Cost per ONU link. The dotted line (100%) marks the GPON reference (adapted c from [15], 2008 IEEE)
368
Klaus Grobe, J¨org-Peter Elbers, and Stephan Neidlinger
sum of all components (including multiple components, where applicable), divided by the number of ONU links and normalized to the GPON cost, C GPON : C= C T ransceiver s + C Filter s + C Splitter s/Coupler s + C Shelves + CFPGAs/SICs + C Ampli f ier s · R S /W/C G P O N · 100, (17.1) where R S is the per-wavelength splitting ratio. For example, for a splitting ratio of 1:4, R S = 0.25. W is the number of WDM wavelengths. The calculation is based on standard components and their costs. Adding TDM/TDMA and also amplification to WDM-PONs has a strong impact on the resulting power budgets and hence maximum achievable distances. This is shown in Fig. 17.14 which compares fiber power budgets (i.e., excluding filters and splitters) for various unamplified and amplified WDM-PONs with and without succeeding TDM(A). The power budgets have been calculated as the difference of (guaranteed) transmitter output power coupled into the fiber and minimum receiver sensitivity, decreased by the sum of all component losses ( L) plus G, the gain of additional optional amplifiers (only DWDM amplification was considered here): L Component + G, (17.2) B = T xmin − Rxmin − where B is the remaining power budget that can be spent for transmission fibers, and L Component is the respective loss of each individual optical component in the transmission link. All component values can be found in the literature. For the amplified scenarios, a single bidirectional lumped amplifier in an active RN was considered together with booster and preamplifiers in the OLT. In a DWDM-PON, the latter can always be used without turning the RN into an active node. Higher power budgets are possible by cascading more amplifiers. The cost and power budget curves in Figs. 17.13 and 17.14 are based on PIN receivers for 32-channel CWDM/TDMA and xWDM/WDMA configurations at bit
Fig. 17.14 WDM-PON and WDM/TDMA-PON link power budget (i.e., power budget available c for fiber loss) (adapted from [15], 2008 IEEE)
17 Broadband Access Networks
369
rates of 1.2 and 2.5 Gb/s. All other configurations are based on APD receivers. This allows lower cost at reasonable power budgets. WDM-PONs can provide per-wavelength bit rates of 1.25, 2.5, 4.3, or 10 Gb/s. The exact bit rates depend on the application and its framing (e.g., G.709, OTN). In principle, even higher bit rates, i.e., the upcoming 40 Gb/s bit rates, could be supported given the links are dispersion-managed or the respective technology provides sufficient dispersion allowance. The resulting per-ONU bit rate only depends on optional additional TDM(A). Figure 17.15 shows per-ONU bit rates for wavelength bit rates up to 10G.
c Fig. 17.15 Average per-ONU bandwidth in Mb/s (adapted from [15], 2008 IEEE)
Wavelengths of 16 CWDM and 40 DWDM have been assumed here. The perONU bit rates for WDMA PONs are the bit rates of the respective transceivers, i.e., 1G25, 2G5, 4G3, or 10G, respectively. For WDM+TDMA PONs, the per-ONU bit rate RB,ONU is the WDM channel bit rate RB, WDM multiplied by the succeeding TDMA splitting factor RS and also multiplied by the TDMA efficiency, η: R B,O N U = R B,W D M · R S · η.
(17.3)
Note that η = 0.9 has been used here, assuming similar efficiency than GPON.
17.3.4 WDM-PON Applications WDM-PON architectures extend PON application range over what is feasible with EPON and GPON technologies. Higher bandwidths and extended power budgets enable the use of WDM-PON technologies as basis of a unified metro access and backhauling network. Three main applications are supported by a unified WDM-PON infrastructure: residential access, dedicated broadband access for large enterprises, and backhauling of copper, HFC, or radio access networks. Several options exist to adapt the infrastructure to the specific application: for high maximum reach and splitting ratio, optical amplification or a regenerative remote node can be used. When transporting
370
Klaus Grobe, J¨org-Peter Elbers, and Stephan Neidlinger
large accumulated bandwidths, additional resilience (protection) functionality can be provided. It is also useful to extend the access topologies to other (physical and logical) topologies than the tree or drop structures known from EPON/GPON. For example, ring-based ONU-connectivity can be offered on one wavelength alongside with point-to-point or point-to-multipoint connectivity on another one. The ring provides resilient backhauling of wireless and wireline services and may use simple TDM add/drop techniques. It is also possible to connect various remote nodes in a ring [12]. Residential and enterprise access services independently stay with tree/drop architectures and may use additional TDMA techniques where required. A generic example of a metro network that consists of a DWDM core and unified WDM-PON backhaul/access layer is shown in Fig. 17.16. Photonic + L2/3 Core
Core PoP Multi-Degree ROADM
Metro Core
ROADM
Residentials ONUs
Resilient Connection OLT-RN 1:N
ROADM
Metro PoP ROADM OLT Amplification
λ p,p’
λ 1,1’ λ 2,2’
ONU MSAN
ONU
RN ONU
λ i...λ k’ ONU XDSL
DSL and similar Backhaul
ONU HFC
λ n,n’ ONU
ONU
Large Business Customers
ONU
Wireless Backhaul
Fig. 17.16 Metro area networks based on ROADM core and WDM-PON. The physical topology c behind the RN can be a combination of star and ring (adapted from [15], 2008 IEEE)
For most efficient fiber utilization, the WDM-PON simultaneously supports different fiber topologies. The choice of per-wavelength bit rates allows a tailoring of network costs to the demanded services. Depending on the application, combinations of CWDM and DWDM technology with additional TDM, TDMA or active Ethernet can be used. If total bandwidth and maximum reach requirements allow the use of CWDM, significant cost savings over DWDM can be achieved. All different PON variants (DWDM, CWDM, Ethernet-specific, additional TDMA) and topologies are supported end to end by one converged platform solution and are seamlessly integrated with metro/core DWDM transport. An example of an Ethernet-centric WDM-PON access system is shown in Fig. 17.17. The combination of a WDM/Layer-2 switch card (with colored line and client interfaces) and Ethernet-ONUs can provide all relevant functions in a WDM-PON infrastructure. Layer-2 aggregation of the ONU services and remote ONU management is provided by the Layer-2 switch. The WDM/Layer-2 switch is at the
17 Broadband Access Networks
371
Fig. 17.17 WDM-PON for Ethernet access
same time part of the metro core DWDM transport solution and of the WDM-PON ONU. It hence allows direct interworking/interfacing between the WDM-PON and the DWDM metro core network without any unnecessary optical interfaces. WDMPON reach extension (by amplification) and protection are provided on the optical layer (alternatively, Ethernet 1+1 protection can be used). The ONU can perform demarcation and OAM functions, and is managed and monitored through an embedded communications channel based on the EFM (Ethernet in the first mile [8]) or VLAN standards. The complete WDM-PON system is managed through a single controller in the central office, allowing end-to-end service management and monitoring with a single management system.
17.4 Summary Next-generation access and backhaul networks will be based on WDM technology for capacity and cost reasons. On the one hand, these networks have to support broadband residential and business access services with bandwidths up to 100 Mb/s (residential) and 10 Gb/s (or multiples of this, for enterprises), respectively. On the other hand, these networks must also support all backhaul applications (wireless, wireline, including standard GPON and EPON) within a unified architecture, thus replacing purpose-built and legacy network solutions, and effectively reducing (operational) cost. Migration toward next-generation access/backhaul networks can start with passive WDM links, forming a transparent infrastructure which is scalable in total transport capacity. The solution can be complemented by active
372
Klaus Grobe, J¨org-Peter Elbers, and Stephan Neidlinger
technology to support point-to-multipoint connectivity, aggregation, optical amplification, and protection. The resulting solution is a high-capacity, long-reach (or high power budget) WDM-PON. Although significant improvements over standard GPON (and EPON) can be achieved in both, total capacity and general functionality, cost per ONU client can still be as low as in the GPON case.
References 1. PricewaterhouseCoopers, “Developing a Generic Approach for FTTH Solutions Using LCA Methodology,” in Proc., FTTH Council Europe Conference, Paris, Feb. 2008. 2. A. A. M. Saleh and J. M. Simmons, “Architectural Principles of Optical Regional and Metropolitan Access Networks,” IEEE/OSA Journal of Lightwave Technology, vol. 17, no. 12, pp. 2431–2448, Dec. 1999. 3. K. Grobe, “Optical Metro Networking,” Telekommunikation Aktuell, vol. 58, no. 5/6 and 9/10, May/June and Sept./Oct. 2004. 4. M. Reeve, “The Need For Standards in the 21st Century Network,” BT Group, http://gawain.soc.staffs.ac.uk/modules/levelm/ce00227-m/nst stoke slides/nst introduction 06 21c.ppt. 5. K. Grobe et al., “Flexible WDM Solution for DSL Backhaul,” in Proc., 7. ITG-Fachtagung Photonische Netze, Leipzig, April 2006. 6. A. A. De Albuquerque et al., “Field Trials for Fiber Access in the EC,” IEEE Communications Magazine, vol. 32, no. 12, pp. 40–48, Feb. 1994. 7. ITU-T Recommendation G.983.x, Broadband Optical Access Systems based on Passive Optical Networks (PON), G.984.x, Gigabit-capable PON. 8. IEEE Standard 802.3ah, Ethernet in the First Mile (EFM). 9. IEEE Standard 802.3av, 10GEPON. 10. R. Davey et al., “Long-Reach Access and Future Broadband Network Economics,” in Proc., European Conference on Optical Communications (ECOC), Berlin, Sept. 2007. 11. A. Banerjee et al., “Wavelength-Division Multiplexed Passive Optical Network (WDM-PON) Technologies for Broadband Access: A Review (Invited),” OSA Journal of Optical Networking, vol. 4, no. 11, pp. 737–758, Nov. 2005. 12. J. Prat et al., “Next Generation Architectures for Optical Access,” in Proc., European Conference on Optical Communications (ECOC),” Cannes, Sept. 2006. 13. CIP white paper “WDM-PON Technologies,” http://www.ciphotonics.com/New PDFs/WPON White Paper v1%200.pdf. 14. J.-P. Elbers and K. Grobe, “PON Evolution from TDMA to WDM-PON,” in Proc., OFC/NFOEC, paper NThD6, San Diego, CA, USA, Feb. 2008. 15. J.-P. Elbers and K. Grobe, “PON in Adolescence: From TDMA to WDM-PON,” IEEE Communications Magazine, vol. 46, no. 1, pp. 26–34, Jan. 2008. 16. J.-P. Elbers and K. Grobe, “WDM-PON – A Platform for Consolidated Metro Access and Backhaul,” in Proc., 9. ITG-Fachtagung Photonische Netze, Leipzig, April 2008. 17. ITU-T Recommendation G.694.2, CWDM wavelength grid.
Index
10GEPON, 362 10GbE, 359, 360 3G, 17 3G/4G, 363 3R regenerator, 367 Access Control Lists, 172 Active Ethernet, 341, 355, 361 Active PON, 364 Adaptive Modulation and Coding (AMC), 99 Add-on service, 258 ADM, 358, 360 Admission Control, 169 ADSL, 19, 20, 302, 341, 355 ADSL2, 220 ADSL2+, 6, 27 Aggregation, 359 amplified spontaneous emission, ASE, 222 Anycast, 329 APD receiver, 364, 369 Arrayed-waveguide grating (AWG), 201, 241 ASIC, 304 Assured Forwarding, 153 Asymmetric Fabry-Perot modulator (AFPM), 285 Athermal AWG, 248 ATM, 357, 358 AWG, 243, 255 Backhauling, 355 bandwidth guaranteed, 178 Bandwidth reservation, 181 Base station (BS), 322 Base transceiver station (BTS), 283 Best Effort (BE), 130 Binary Shift Keying (BPSK), 84 BPON, 302, 362
BRAS, 355, 360 Broadband PON (BPON), 362 Broadband remote access server (BRAS), 355 Broadband services router (BSR), 357 Broadcast IPTV, 354 BSR, 357 Bundling, 17 Cable modem, 14 Cable Service Interface Specifications (DOCSIS), 10 Capacity and delay aware routing (CaDAR), 331 CapEx, 359 Capital Expenditure (CAPEX), 5 CATV, 286 CDMA, 27, 103 Central office (CO), 27, 322 Chromatic dispersion, 286 Class-of-service Oriented Packet Scheduling, COPS, 164, 180 Client-server paradigm, 253, 255 Coarse wavelength division multiplexing (CWDM), 241 Coaxial cable, 34 Code Division Multiplexing, CDM, 229 Colorless ONU, 201 Connection Identifier (CID), 129 Continuous wave (CW), 245 Control, 239 Convolutional codes, 93 CPE, 341, 343–345, 349, 363 Cross-layer design, 296 Crosstalk, 67 Customer premises equipment (CPE), 27, 240 CWDM, 355, 357–360, 367 CWDM-PON, 367
373
374 Data communications networks (DCN), 357 Data-over-SONET/SDH (DoS), 238 DBA, 252 DCN, 357 Deficit Weighted Round Robin (DWRR), 169 Delay-aware routing algorithm (DARA), 330 Demand Assigned Multiple Access (DAMA), 122 Denial of service, 345 Dense wavelength division multiplexing (DWDM), 242 Differential Vectoring, 69 Differentiated services (DiffServ), 313 Differentiated Services Code Point, 172 Digital Subscriber Line (DSL), 65, 219 Digital subscriber line access multiplexer (DSLAM), 308, 357 Digital video broadcast (DVB), 292 DOCSIS, 25 DSL, 308, 357, 360 DSLAM, 27, 341, 357–360 Dual homing, 366 Dual-homing, 361 dual-token bucket, 183 DWDM, 355, 357, 360 DWDM-PON, 367 Dynamic Bandwidth Allocation, 173 Dynamic bandwidth allocation (DBA), 198, 201, 245 Dynamic capacity allocation, 285 Dynamic microcellular mode, 295 Dynamic Spectrum Management (DSM), 66, 67 Dynamic wavelength assignment (DWA), 201 E-LAN, 356 E-Line, 356 EDFA, 42 EFM, 349, 371 electro-absorption modulator, EAM, 228 Electro-absorption transceiver (EAT), 285 Electro-magnetic interference (EMI), 284 Electro-optical effect, 285 Encryption, 345 Enhancement band, 294 EPON, 30, 198, 302, 340, 349, 361–364 erbium-doped fiber amplifier, EDFA, 222 Ethernet, 341, 354, 356, 360 Ethernet in the First Mile (EFM), 349 Ethernet media converter, 361 Ethernet Passive Optical Networks (EPONs), 169 Ethernet PON (EPON), 247, 362 European regulators group (ERG), 31
Index Expedited Forwarding, 153 Extended-Real-time Polling Service (ertPS), 129 F-TCP, 255 Fallback, 258 Fast Ethernet, 349, 359 FEC, 364 fiber local loop unbundling (FLLU), 30 Fiber to the curb (FTTC), 302 Fiber to the Home (FTTH), 7 Fiber to the home (FTTH), 302 Fiber to the node (FTTN), 302 Fiber-to-the-building (FTTB), 238 Fiber-to-the-Curb (FTTC), 38 Fiber-to-the-home (FTTH), 40, 238 Fiber-to-the-subscriber (FTTS), 197 Fibre Channel, 341, 354 File sharing, 354 First/last mile bottleneck, 238 Fixed mobile convergence (FMC), 302 Fixed wavelength converter (FWC), 245 Fixed wireless access (FWA), 40 Fixed-mobile convergence (FMC), 17, 356 FMC, 356 Forward Error Correction (FEC), 122 forward error correction, FEC, 226 Fractal patch antenna, 296 Fragmentation, 210 Free spectral range (FSR), 245, 255 Free-space optical (FSO), 60 Frequency Division Duplex (FDD), 125 FTTB, 14, 24, 29, 356, 361 FTTC, 31, 32, 361, 363 FTTH, 14, 22, 33, 323, 341, 356, 361 FTTN, 27, 29, 356, 358, 363 FTTX, 238 FTTx, 14, 23, 362 Full service access network (FSAN), 295 G.SHDSL, 220 Gated service, 252 GbE, 357, 359, 360 GbE-add/drop multiplexing (GbE-ADM), 358 GbE-ADM, 358 GELS, 356 General Processor Sharing, 175 Generic framing procedure (GFP), 238 Gigabit Ethernet, 349 Gigabit Ethernet (GbE), 238 Gigabit PON (GPON), 362 GMPLS-enabled Ethernet label switching (GELS), 356 GPON, 28, 30, 33, 294, 302, 340, 361–364
Index Grant scheduling, 200 Grant sizing, 199 GROW-Net, 326, 334 GROW-Net integrated routing (GIR), 331 GSM, 27, 284, 291, 293 Handover, 138, 314 HDTV, 20 Head-of-line (HOL) blocking, 253 HFC, 19, 363, 369 High-definition Television, 354 High-definition television (HDTV), 238 High-speed Internet (HSI), 354 Home access gateway, 345 Home assistance, 354 HSDPA, 16 HSI, 354 Hybrid Fiber Coax, 14 Hybrid Fiber Coax (HFC), 5 Hybrid fiber coax (HFC), 39 Hybrid fiber radio (HFR), 322 Hybrid Granting Protocol, 180 ILEC, 355 IM-DD, 289 InfiniBand, 354 Integrated services (IntServ), 313 Integrated services digital network (ISDN), 354 Intensity modulation direct detection (IM-DD), 286 inter-ONU scheduling, 163 Interactive TV, 20 Intermediate frequency (IF), 285 intra-ONU scheduling, 163, 169 IP telephony, 18 IPTV, 17, 20, 301, 343 IPTV, VoD, 4 ISDN, 14, 354 ISDN BRI, 341 Island of transparency, 247 Jamming, 345 LAN interconnection, 354 light load penalty, 175 Lightpath, 240 Linear program relaxation, 211 Link capacity adjustment scheme (LCAS), 238 List algorithm, 210 Local exchange, 355 Logical Link ID, LLID, 161 Long Term Evolution (LTE), 5
375 Long-Reach Passive Optical Network (LR-PON), 221 Long-term evolution (LTE), 302 MAC, 239, 340 MAC protocols, 344 Mach-Zehnder modulator, 285 Management, 239 Media Access Control (MAC), 119 Media access control (MAC), 293 Media converter, 343 Metro gap, 238 Metropolitan area network (MAN), 238 Microwave over fiber (MoF), 302, 306 Minimum spanning tree (MST), 310, 326 Mobility, 138 Mode instability, 286 Modified Deficit Weighted Round Robin, 176 MPCP, 204 Multi-dwelling unit (MDU), 27 Multi-Dwelling Units (MDU), 6 multi-point control protocol, 175 Multi-Point Control Protocol, MPCP, 151, 230 multi-service access, 179 Multi-Service Operators (MSO), 9 Multi-Thread Polling, 230 Multicast, 359 Multimode fiber, 42 Multiple input multiple output (MIMO), 328 Multipoint control protocol (MPCP), 251, 331 National regulatory agency (NRA), 30 Network management system (NMS), 357 Next available supported channel (NASC), 210 Next generation access (NGA), 31, 295 NG-SONET/SDH, 238 NGA, 32 NMS, 357 Non-real-time Polling Service (nrtPS), 129 OAM, 349, 356, 359, 371 OCDMA, 54 ODF, 30, 342, 348 OECD, 17 OEO conversion, 240, 244, 247 OFDM, 87 OFDMA, 104 Offline scheduling, 205 OLT, 201, 361, 364, 368 On-line gaming, 251, 253 Online games, 354 Online gaming, 18, 19 Online just-in-time (JIT) scheduling, 208 Online scheduling, 205
376 ONT, 341, 343, 345 ONU, 201, 364, 367 Open Access models, 348 Operation Expenditures (OPEX), 5 OpEx, 359, 361 Optical Add-Drop Multiplexer, OADM, 229 Optical burst switching (OBS), 243 Optical burst switching multiplexer (OBS-M), 245 Optical bypassing, 240, 247, 249, 254 Optical circuit switching (OCS), 244 Optical cross-connect, 296 Optical cross-connect (OXC), 245 Optical distribution frame (ODF), 29, 342 Optical fiber, 41 Optical flow switching (OFS), 239 Optical Line Terminal (OLT), 75 Optical line terminal (OLT), 241, 303, 322 Optical Network for Regional Access using Multiwavelength Protocols (NGI ONRAMP), 239 Optical network unit (ONU), 241, 302, 303, 322 Optical packet switching (OPS), 244 Optical time domain reflectometry (OTDR), 345 Optical+Ethernet architecture, 357 Optical-wireless integration (OWI), 328 Orthogonal Frequency Division Multiplexing (OFDM), 118 Orthogonal frequency division multiplexing (OFDM), 292 orward error correction (FEC), 82 OTDR, 345 OTN, 369 Oversubscription, 344 P2P, 21 P2P file sharing, 253 Packet-over-SONET/SDH (PoS), 238 Parametric wavelength conversion (PWC), 242 Passive Optical Network (PON), 8 Passive optical network (PON), 322, 362 Passive star coupler (PSC), 249 PC penetration, 17 Peer-to-peer, 354 Peer-to-peer (P2P), 238, 251, 254, 301 phase shift keying (PSK), 83 Pico-cell, 285 Piggybacking, 251 PIN receiver, 368 Plain old telephone service (POTS), 354 PMP WiMAX, 119 Point of Presence (PoP), 339
Index Point of presence (POP), 355 Point-to-multipoint (P2MP), 48 Point-to-Multipoint (PMP), 119 Point-to-point Ethernet FTTH, 7 Polling, 245 PON, 198, 326, 362, 363 POP, 341, 345, 348, 349, 355 PoP, 339 POTS, 354 Power line transmission, 356 priority scheduling, 175 Priority with Insertion Scheduling, 176 Privacy, 364 Protection, 239, 358, 361, 365 PSC, 253 PSK, 84 PSTN, 18 QAM, 85 QoS, 252, 302, 354 QoS protection, 181 Quadrature Amplitude Modulation (QAM), 83 Quadruple play, 354, 356 Quality of Service, 170 Quality of Service (QoS), 354 Radio frequency (RF), 285 Radio over fiber, 57 Radio-over-fiber (ROF), 322 Ranging, 160 rate-based admission control, 185 RAU, 289, 294 Real-time Polling Service (rtPS), 129 Reconfigurable optical add-drop multiplexer (ROADM), 237, 240 Reconfigurable optical add/drop multiplexer (ROADM), 355 Reflective semiconductor optical amplifier (RSOA), 245 Relative intensity noise (RIN), 286 Remote antenna unit (RAU), 284 Remote modulation, 242, 245 Resilience, 255, 356 Resilient packet ring, 355 Resilient packet ring (RPR), 238 Risk-and-delay aware routing algorithm (RADAR), 333 ROADM, 246, 355, 370 RPR, 243, 253 RTS/CTS mechanism, 293 Scalability, 254, 255, 356, 357, 360 Scheduling framework, 204 Scheduling model, 203
Index Scheduling policy, 204 SCMA, 294, 295 Security, 18, 345, 361, 364 Semiconductor optical amplifier (SOA), 242 semiconductor optical amplifier, SOA, 222 Separate time and wavelength assignment (STWA), 200 Service Level Agreement, 170 Service level agreement (SLA), 199 SFNet, 323, 325 Signal to interference noise ratio (SINR), 316 Simulcast mode, 295 Single-mode fiber, 42 Small-to-medium business (SMB), 329 SME, 18 SOA-Raman Hybrid Amplifier, SRHA, 229 SOHO, 18 SONET/SDH, 238, 243, 341, 355, 358 Space division multiplexing (SDM), 198, 247 Space-time coding, 317 Spontaneous emission, 286 Start-time fair queueing, 175 Static microcellular mode, 295 Static wavelength assignment (SWA), 201 Statistical multiplexing, 359 Storage, 354 Subscriber Station (SS), 119 SuperPON, 224 System bandwidth management (SBM), 334 TDM, 340, 355, 358 TDM leased lines, 354 TDM PON, 240, 306 TDMA, 102, 294, 362 TDMA PON, 362 Telemedicine, 354 Teleworking, 354 Thin film filter, 241 Time Division Duplex (TDD), 125 Time Division Multiple Access (TDMA), 169 Time division multiple access (TDMA), 362 Time-Division Multiple Access, TDMA, 225 Time-domain wavelength interleaved networking (TWIN), 248 Traffic engineering, 255 Transparency, 356, 357 Triple play, 238, 258, 354, 356 Triplexer, 343, 344 Tunable transmitter, 243 Turbo coding (TC), 96 TWIN with wavelength reuse (TWIN-WR), 248 UMTS, 284, 291
377 Unbundled local loop (ULL), 28 Universal Mobile Telecommunications Systems (UMTS), 82 Unsolicited Grant Service (UGS), 129 UWB, 291 V-BLAST, 328 VDSL, 220, 355, 359, 361, 363 VDSL2, 6, 27, 341 Vectored DSLs, 66 Very high capacity digital subscriber line (VDSL), 39 Video broadcast overlay, 343 Video on demand (VoD), 238 Video overlay, 294 Video-OLT, 343 Video-on-demand, 18, 20, 354 Video-on-demand (VoD), 301 Virtual concatenation, 238 Virtual LAN (VLAN), 310 Virtual private network (VPN), 354 VLAN, 348, 359, 371 Voice over IP (VoIP), 354 VoIP, 18, 354 VPN, 354 W-CDMA, 291 Wavelength conversion, 367 Wavelength division multiple access (WDMA), 364 Wavelength division multiplexing (WDM), 42, 198 Wavelength-selective switch (WSS), 242 WBA, 30 WDM, 55, 356 WDM coupler, 250 WDM PON, 9, 201, 240, 306 WDM-PON, 361 WDM/TDM EPON, 247 WDMA, 294, 364 WDMA-TDMA PON, 364 Weighted bipartite matching, 211 Wholesale broadband access (WBA), 29 Wholesale service offering, 358 Wide area network (WAN), 237 WiFi, 117, 292, 302, 322, 328 WiMAX, 11, 14, 27, 117, 293, 302, 322, 328, 363 WiMAX base station (BS), 303 WiMAX mesh networks (WMN), 131 Wireless gateway, 323 Wireless mesh network, 303 Wireless mesh network (WMN), 323 Wireless router, 323
378 WirelessMAN-SC, 121 WLAN, 16 WMN, 326
Index WOBAN, 322 xDSL, 14, 23, 33, 34, 356