No title

February 2011 Wireless Cover 1 IEEE 2/7/11 12:20 PM Page 1 W IRELESS C February 2011, Vol. 18, No. 1 OMMUNICATIONS...

Author: A Publication of the IEEE Communications Society

20 downloads 770 Views 7MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

February 2011 Wireless Cover 1

IEEE

2/7/11

12:20 PM

Page 1

W IRELESS C February 2011, Vol. 18, No. 1

OMMUNICATIONS

IMS EMERGENCY SERVICES: A PRELIMINARY STUDY MIXING NETWORK CODING AND COOPERATION FOR RELIABLE WIRELESS COMMUNICATIONS • SPECTRUM SENSING FOR COGNITIVE RADIO SYSTEMS • COOPERATIVE COMMUNICATION IN MULTIHOP COGNITIVE RADIO NETWORKS BASED ON MULTICARRIER MODULATION • TOPOLOGICAL-BASED ARCHITECTURES FOR WIRELESS MESH NETWORK • •

IMS E MERGENCY S ERVICES

®

A Publication of the IEEE Communications Society In cooperation with IEEE Computer and VehicularTechnology Societies

LYT-TOC-Feb

2/7/11

11:41 AM

Page 1

F E B R U A RY 2011/V O L . 18, N O . 1

IEEE

W IRELESS C OMMUNICATIONS

ACCEPTED FROM OPEN CALL

46

6

CHALLENGES, OPPORTUNITIES, AND SOLUTIONS FOR CONVERGED SATELLITE AND TERRESTRIAL NETWORKS

IMS EMERGENCY SERVICES: A PRELIMINARY STUDY YI-BING LIN, MENG-HSUN TSAI, AND YUAN-KUANG TU

15 MIXING NETWORK CODING AND COOPERATION FOR RELIABLE WIRELESS COMMUNICATIONS FRANCESCO ROSSETTO AND MICHELE ZORZI

22 SECURING UNDERWATER WIRELESS COMMUNICATION NETWORKS MARI CARMEN DOMINGO

TARIK TALEB, YASSINE HADJADJ-AOUL, AND TOUFIK AHMED

54 INTERFERENCE COORDINATION FOR OFDM-BASED MULTIHOP LTE-ADVANCED NETWORKS KAN ZHENG, BIN FAN, JIANHUA LIU, YICHENG LIN, AND WENBO WANG

64 DISTRIBUTED AUTOMATED INCIDENT DETECTION WITH VGRID BEHROOZ KHORASHADI, FRED LIU, DIPAK GHOSAL, MICHAEL ZHANG, AND CHEN-NEE CHUAH

30 SPECTRUM SENSING FOR COGNITIVE RADIO SYSTEMS: TECHNICAL ASPECTS AND STANDARDIZATION ACTIVITIES OF THE IEEE P1900.6 WORKING GROUP KLAUS MOESSNER, HIROSHI HARADA, CHEN SUN, YOHANNES D. ALEMSEGED, HA NGUYEN TRAN, DOMINIQUE NOGUET, RYO SAWAI, AND NAOTAKA SATO

38 MULTICARRIER MODULATION AND COOPERATIVE COMMUNICATION IN MULTIHOP COGNITIVE RADIO NETWORKS TAO LUO, FEI LIN, TAO JIANG, MOHSEN GUIZANI, AND WEN CHEN

74 TOPOLOGICAL-BASED ARCHITECTURES FOR WIRELESS MESH NETWORKS AMIR ESMAILPOUR, NIDAL NASSER, AND TARIK TALEB

82 SYNCHRONIZATION OF MULTIHOP WIRELESS SENSOR NETWORKS AT THE APPLICATION LAYER ÁLVARO MARCO, ROBERTO CASAS, JOSÉ LUIS SEVILLANO RAMOS, VICTORIAN COARASA, ÁNGEL ASENSIO, AND MOHAMMAD S. OBAIDAT

MESSAGE

FROM THE

EDITOR-IN-CHIEF — 2

SCANNING THE LITERATURE — 4 Cover image: Getty Images

IEEE Wireless Communications • February 2011

1

LYT-EDIT NOTE-February

2/7/11

10:39 AM

Page 2

M E S S A G E F R O M T H E E D I T O R - I N -C H I E F

CALL FOR MORE WIRELESS INNOVATIONS

F

munications and networking, and decision makirst of all, I am happy to report that my ing and control. The second technology on the term as the Editor-in-Chief has been top 11 list is social networking, which has close extended for one more year. I also want ties to wireless technologies as most users have to report that we have been working on already been doing their social networking with moving into Manuscript Central, which will smartphones or mobile devices, leading to the be a great step toward a more efficient ever more popular mobile social networking. As paper handling process. As a new year we can see, wireless technologies have already starts, we also look forward to more excitrevolutionized the way we live and have transing news on wireless communications, netformed our society into a completely different working, and their applications. You may kind for everything imaginable. We, the engihave already read the first issue of IEEE neers and scientists, are at center stage to utiSpectrum this year. Among the top 11 idenlize the technologies we have created to make a tified technologies (although it is not clear significantly different society, good or bad, and why it is 11 rather than the more commonly it is up to us to reshape it and our living enviused 10, perhaps to avoid a copyright lawronments for better quality of life. We need you suit from David Letterman?), smartphones YUGUANG MICHAEL FANG to invent new technologies, including wireless are ranked on top. Unfortunately, the coltechnologies, and we need new ideas and new umn does not spell out the drivers of future innovations from you. Our magazine offers you the platform smartphone technologies: the deep integration technologies of for you to spread the word. We are searching for articles on smartphones with smart environments and human interactions new innovations in the wireless area and solicit new special for ubiquitous data collection, processing, data mining, com-

IEEE Director of Magazines Andrzej Jajszczyk, AGH U. of Sci. & Tech. Poland Editor-in-Chief Yuguang Michael Fang, Univ. of Florida, USA Associate Editor-in-Chief Dilip Krishnaswamy, Qualcomm Research Center Senior Advisors Hamid Ahmadi, AT&T Labs, USA Abbas Jamalipour, University of Sydney, Australia Thomas F. La Porta, Pennsylvania State Univ., USA Mahmoud Naghshineh, IBM, USA Michele Zorzi, University of Padova, Italy Advisory Board Donald Cox, Stanford University, USA David Goodman, Polytechnic University, USA Tero Ojanperä, Nokia, Finland Kaveh Pahlavan, Worcester Polytech. Inst., USA Mahadev Satyanarayanan, CMU, USA IEEE Vehicular Technology Liaison Theodore Rappaport, Univ. of Texas, Austin, USA IEEE Computer Society Liaison Mike Liu, Ohio State University, USA Technical Editors Abouzeid Alhussein, Rensselaer Polytechnic Inst., USA Benny Bing, Georgia Tech, USA Azzedine Boukerche, Univ. of Ottawa, Canada Jyh-Cheng Chen, Natl. Chiao Tung Univ., Taiwan Carla-Fabiana Chiasserini, Politecnico di Torino, Italy Sunghyun Choi, National Seoul University, Korea Mischa Dohler, France Telecom R&D, France Ekram Hossain, University of Manitoba, Canada Thomas Hou, Virginia Tech., USA Nei Kato, Tohoku University, Japan Pascal Lorenz, U. of Haute Alsace, France Giacomo Morabito, U. di Catania, Italy Zhisheng Niu, Tsinghua University, China Symeon Papavassiliou, Natl. Tech. Univ. Athens, Greece Vincent Poor, Princeton Univ., USA Kui Ren, Illinois Institute of Tech., USA Apostolis Salkintzis, Motorola, Greece John Shea, University of Florida, USA Sherman Shen, Univ. of Waterloo, Canada Richard Wolff, Montana State Univ., USA Junshan Zhang, Arizona State Univ. Qian Zhang, Hong Kong Univ. Science & Tech., Hong Kong

2

W IRELESS C OMMUNICATIONS

Department Editors Industrial Perspectives Benny Bing, Georgia Tech, USA Scanning the Literature Yanchao Zhang, Arizona State Univ., USA Spectrum Policy and Reg. Issues Michael Marcus, Marcus Spectrum Solns., USA 2011 Communications Society Board of Governors Officers Byeong Gi Lee, President Mark Karol, VP–Technical Activities Khaled B. Letaief, VP–Conferences Sergio Benedetto, VP–Member Relations Leonard Cimini, VP–Publications Doug Zuckerman, Past President Stan Moyer, Treasurer John M. Howell, Secretary Members-at-Large Class of 2011 Robert Fish • Joseph Evans Nelson Fonseca • Michele Zorzi Class of 2012 Stefano Bregni • V. Chan Iwao Sasase • Sarah K. Wilson Class of 2013 Gerhard Fettweis • Stefano Galli Robert Shapiro • Moe Win 2011 IEEE Officers Moshe Kam, President Gordon W. Day, President-Elect Roger D. Pollard, Secretary Harold L. Flescher, Treasurer Pedro A. Ray, Past-President E. James Prendergast, Executive Director Nim Cheung, Director, Division III Joseph Milizzo, Assistant Publisher Eric Levine, Associate Publisher Susan Lange, Online Production Manager Jennifer Porcello, Production Specialist Catherine Kemelmacher, Associate Editor

IEEE Wireless Communications (ISSN 1536-1284) is published bimonthly by The Institute of Electrical and Electronics Engineers, Inc. Headquarters address: IEEE, 3 Park Avenue, 17th Floor, New York, NY 10016-5997; tel: 212-705-8900; fax: 212-705-8999; e-mail: [email protected]. Responsibility for the contents rests upon authors of signed articles and not the IEEE or its members. Unless otherwise specified, the IEEE neither endorses nor sanctions any positions or actions espoused in IEEE Wireless Communications. Annual subscription: Member subscription: $40 per year; Non-member subscription: $250 per year. Single copy: $50. Editorial correspondence: Manuscripts for consideration may be submitted to the Editor-in-Chief: Yuguang Michael Fang, University of Florida, 435 New Engineering Building, P.O. Box 116130, Gainesville, FL 32611. Electronic submissions may be sent in postscript to: [email protected]. Copyright and reprint permissions: Abstracting is permitted with credit to the source. Libraries permitted to photocopy beyond limits of U.S. Copyright law for private use of patrons: those post-1977 articles that carry a code on the bottom of first page provided the per copy fee indicated in the code is paid through the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923. For other copying, reprint, or republication permission, write to Director, Publishing Services, at IEEE Headquarters. All rights reserved. Copyright © 2011 by The Institute of Electrical and Electronics Engineers, Inc. Postmaster: Send address changes to IEEE Wireless Communications, IEEE, 445 Hoes Lane, Piscataway, NJ 088551331; or email to [email protected]. Printed in USA. Periodicals postage paid at New York, NY and at additional mailing offices. Canadian GST #40030962. Return undeliverable Canadian addresses to: Frontier, PO Box 1051, 1031 Helena Street, Fort Eire, ON L2A 6C7. Subscriptions: Send orders, address changes to: IEEE Service Center, 445 Hoes Lane, Piscataway, NJ 08855-1331; tel.: 908981-0060. Advertising: Advertising is accepted at the discretion of the publisher. Address correspondence to: Advertising Manager, IEEE Wireless Communications, 3 Park Avenue, 17th Floor, New York, NY 10016.

®

IEEE Wireless Communications • February 2011

LYT-EDIT NOTE-February

2/7/11

10:39 AM

Page 3

M E S S A G E F R O M T H E E D I T O R - I N -C H I E F issues on hot topics that are of great interest to our readership. Some of the topics we are particularly interested in are wireless technologies in cyber-physical systems, social networks, smart grids, healthcare and medical systems, urban sensing and public safety, and cloud computing systems. With the upcoming electronic paper handling process, I hope we can make the publication process much faster. For those who are engaging in hot wireless innovations, I encourage you to organize special issues to share your excitement with our general audiences. Due to the unforeseen delay in getting articles for the planned special issue, we have decided to accommodate articles from our open call to cut the long queue of accepted papers. We have selected 10 articles in this issue; their brief summaries follow. “IMS Emergency Services: A Preliminary Study,” by YiBing Lin, Meng-Hsun Tsai and Yuan-Kuang Tu, presents a study on how to support emergency call and walkie-talkie services over IP Multimedia Subsystem in wireless cellular systems. “Mixing Network Coding and Cooperation for Reliable Wireless Communications,” by Francesco Rossetto and Michele Zorzi, gives an overview of how to take advantage of both cooperation and network coding to improve the performance and error correction capabilities of radio networks and highlights the main challenges for future research. “Securing Underwater Wireless Communication Networks,” by Mari Carmen Domingo, reviews some important security design issues specific to underwater wireless communication networks and discusses possible research challenges. “Spectrum Sensing for Cognitive Radio Systems: Technical Aspects and Standardization Activities of IEEE 1900.6 Working Group,” by Klaus Moessner, Hiroshi Harada, Chen Sun, Yohannes D. Alemseged, Ha Nguyen Tran, Dominique Noguet, Ryo Sawai, and Naotaka Sato, overviews the technical issues on spectrum sensing for cognitive radio systems and the related IEEE standardization activities for sensing information exchange, particularly focused on activities from the IEEE P1900.6 working group. “Cooperative Communication in Multihop Cognitive Radio Networks Based on Multicarrier Modulation,” by Tao Luo, Fei Lin, Tao Jiang, and Mohsen Guizani, studies the multicarrier modulation schemes for multihop cognitive radio networks and shows that filtered multitone modulation performs better than orthogonal frequency-division multiplexing (OFDM) in terms of mutual interference elimination, synchronization, and transmission efficiency. Moreover, the authors combine cognitive radio capability with cooperative diversity and come up with three efficient cooperative diversity cognitive models. “Challenges, Opportunities, and Solutions for Converged Satellite and Terrestrial Networks,” by Tarik Taleb, Yassine Hadjadj-Aoul and Toufik Ahmed, investigates some important design issues related to interworking operations between the satellite and terrestrial domains in order to support a wide variety of services for users with a variety of roles (consumer, producer, or manager of communication and media), and suggests some possible solutions and their potential. “Interference Coordination in OFDM-Based Multihop Cellular Networks toward LTE-Advanced,” by Kan Zheng,

IEEE Wireless Communications • February 2011

Bin Fan, Yicheng Lin, and Wenbo Wang, presents an overview of the interference coordination strategies for OFDM-based multihop cellular networks and proposes several static or semi-static interference coordination schemes based on the framework of Long Term Evolution (LTE)Advanced networks with multihop relaying to improve coverage and increase the data rate over cell edge areas. “Distributed Automated Incident Detection with Vgrid,” by Behrooz Khorashadi, Fred Liu, Dipak Ghosal, Chen-Nee Chuah, and Michael Zhang, studies an ad hoc distributed automated incident detection algorithm for highway traffic using vehicles that are equipped with wireless communications, processing, and storage capabilities. By requesting vehicles with such capability to broadcast beacon information containing their speed, location, and lane information, the detection algorithm can make better decisions and yield better performance. “Topological-Based Architectures for Wireless Mesh Network,” by Amir Esmailpour, Nidal Nasser, and Tarik Taleb, provides an overview on architectural design for wireless mesh networks, summarizes the state-of-the-art research findings, and calls for further research on this topic. “Synchronization of Multihop Wireless Sensor Networks at the Application Layer,” by Álvaro Marco, Roberto Casas, José Luis Sevillano, Victorián Coarasa, Ángel Asensio, and Mohammad S. Obaidat, proposes a method for accurate synchronization of large multihop networks, which operates at the application layer while minimizing message exchange. I hope you enjoy reading these articles. I also wish you a productive 2011!

BIOGRAPHY YUGUANG MICHAEL FANG [F’08] ([email protected]) received a Ph.D. degree in systems engineering from Case Western Reserve University in January 1994 and a Ph.D. degree in electrical engineering from Boston University in May 1997. He was an assistant professor in the Department of Electrical and Computer Engineering at New Jersey Institute of Technology from July 1998 to May 2000. He then joined the Department of Electrical and Computer Engineering at the University of Florida in May 2000 as an assistant professor, got an early promotion to associate professor with tenure in August 2003, and to full professor in August 2005. He held a University of Florida Research Foundation (UFRF) Professorship from 2006 to 2009, a Changjiang Scholar Chair Professorship with Xidian University, Xi’an, China, from 2008 to 2011, and a Guest Chair Professorship with Tsinghua University, China, from 2009 to 2012. He has published over 250 papers in refereed professional journals and conferences. He received the National Science Foundation Faculty Early Career Award in 2001 and the Office of Naval Research Young Investigator Award in 2002, and is the recipient of the Best Paper Award from the IEEE International Conference on Network Protocols (ICNP) in 2006 and the recipient of the IEEE TCGN Best Paper Award at the IEEE High-Speed Networks Symposium, IEEE GLOBECOM in 2002. He is also active in professional activities. He is a member of ACM. He is currently serving as the Editor-in-Chief for IEEE Wireless Communications (2009–present) and serves/has served on several editorial boards of technical journals including IEEE Transactions on Mobile Computing (2003–2008, 2011–present), IEEE Transactions on Communications (2000–present), IEEE Transactions on Wireless Communications (2002–2009), IEEE Journal on Selected Areas in Communications (1999–2001), IEEE Wireless Communications (2003–2009), and ACM Wireless Networks (2001-present). He served on the Steering Committee for IEEE Transactions on Mobile Computing (2008–2010). He has been actively participating in professional conference organizations such as serving as Steering Committee Co-Chair for QShine (2004-2008), Technical Program Vice-Chair for IEEE INFOCOM’2005, Technical Program Area Chair for IEEE INFOCOM (2009–2012), Technical Program Symposium Co-Chair for IEEE GLOBECOM 2004, and member of the Technical Program Committee for IEEE INFOCOM (1998, 2000, 2003–2008).

3

LYT-SCANNING-February

2/7/11

10:40 AM

Page 4

S C A N N I N G T H E L I T E R AT U R E EDITED BY YANCHAO ZHANG Using Classification to Protect the Integrity of Spectrum Measurements in White Space Networks O. Fatemieh, A. Farhadi, R. Chandra, and C. Gunter, in the 18th Annual Network & Distributed System Security Symposium (NDSS), San Diego, CA, February 2010 The emerging paradigm for using the wireless spectrum more efficiently is based on enabling secondary users to exploit white space frequencies that are not occupied by primary users. A key enabling technology for forming networks over white spaces is distributed spectrum measurements to identify and assess the quality of unused channels. This spectrum availability data is often aggregated at a central base station or database to govern the usage of spectrum. This process is vulnerable to integrity violations if the devices are malicious and misreport spectrum sensing results. This paper presents CUSP, a new technique based on machine learning that uses a trusted initial set of signal propagation data in a region as input to build a classifier using support vector machines. The classifier is subsequently used to detect integrity violations. Using classification eliminates the need for arbitrary assumptions about signal propagation models and parameters or thresholds in favor of direct training data. Extensive evaluations using TV transmitter data from the FCC, terrain data from NASA, and house density data from the U.S. Census Bureau for areas in Illinois and Pennsylvania show that CUSP is effective against attackers of varying sophistication, while accommodating regional terrain and shadowing diversity.

Privacy-Preserving Regression Modeling of Participatory Sensing Data H. Ahmadi, N. Pham, R. Ganti, T. Abdelzaher, S. Nath, and J. Han, in the 8th ACM Conference on Embedded Networked Sensor Systems (SenSys), Zurich, Switzeland, November 2010 Many participatory sensing applications use data collected by participants to construct a public model of a system or phenomenon. For example, a health application might compute a model relating exercise and diet to amount of weight loss. While the ultimately computed model could be public, the individual

4

input and output data traces used to construct it may be private data of participants (e.g., their individual food intake, lifestyle choices, and resulting weight). This paper proposes and experimentally studies a technique that attempts to keep such input and output data traces private, while allowing accurate model construction. This is significantly different from perturbation-based techniques in that no noise is added. The main contribution of the paper is to show a certain data transformation at the client side that helps keeping the client data private while not introducing any additional error to model construction. The authors particularly focus on linear regression models which are widely used in participatory sensing applications. They use the data set from a map-based participatory sensing service to evaluate their scheme. The service in question is a green navigation service that constructs regression models from participant data to predict the fuel consumption of vehicles on road segments. They evaluate the proposed mechanism by providing empirical evidence that: i) an individual data trace is generally hard to reconstruct with any reasonable accuracy, and ii) the regression model constructed using the transformed traces has a much smaller error than one based on additive data-perturbation schemes.

Reliable Clinical Monitoring Using Wireless Sensor Networks: Experiences In A Step-Down Hospital Unit O. Chipara1, C. Lu, T. Bailey, and G. Roman, in the 8th ACM Conference on Embedded Networked Sensor Systems (SenSys), Zurich, Switzeland, November 2010 This paper presents the design, deployment, and empirical study of a wireless clinical monitoring system that collects pulse and oxygen saturation readings from patients. The primary contribution of this paper is an in-depth clinical trial that assesses the feasibility of wireless sensor networks for patient monitoring in general hospital units. The authors present a detailed analysis of the system reliability from a long-term hospital deployment over seven months involving 41 patients in a step-down cardiology unit. The network achieved high reliability (median 99.68 percent, range 95.21–100 percent). The overall reliability of the system was dominated by sens-

ing reliability of the pulse oximeters (median 80.85 percent, range 0.46–97.69 percent). Sensing failures usually occurred in short bursts, although longer periods were also present due to sensor disconnections. The authors show that the sensing reliability could be significantly improved through oversampling and by implementing a disconnection alarm system that incurs minimal intervention cost. A retrospective data analysis indicated that the system provided sufficient temporal resolution to support the detection of clinical deterioration in three patients who suffered from significant clinical events including transfer to intensive care units.

CodeOn: Cooperative Popular Content Distribution for Vehicular Networks Using Symbol Level Network Coding M. Li, Z. Yang, and W. Lou, IEEE Journal on Selected Areas in Communications (JSAC), special issue on Vehicular Communication Networks, vol. 29, no. 1, January 2011 Driven by both safety concerns and commercial interests, one of the key services offered by vehicular networks is popular content distribution (PCD). The fundamental challenges to achieve high speed content downloading come from the highly dynamic topology of vehicular ad hoc network (VANET) and the lossy nature of the vehicular wireless communications. This paper introduces CodeOn, a novel push-based PCD scheme where contents are actively broadcasted to vehicles from road side access points and further distributed among vehicles using a cooperative VANET. In CodeOn, we employ a recent technique, symbol level network coding (SLNC), to combat the lossy wireless transmissions. Through exploiting symbol level diversity, SLNC is robust to transmission errors and encourages more aggressive concurrent transmissions. In order to fully enjoy the benefits of SLNC, we propose a suite of techniques to maximize the downloading rate, including a prioritized and localized relay selection mechanism where the selection criteria are based on the usefulness of vehicle-possessed contents, and a lightweight medium access protocol that naturally exploits the abundant concurrent transmission opportunities. We also propose additional mechanisms

IEEE Wireless Communications • February 2011

LYT-SCANNING-February

2/7/11

10:40 AM

Page 5

S C A N N I N G T H E L I T E R AT U R E to reduce the protocol overhead without sacrificing the performance.

Free Side Channel: Bits over Interference K. Wu, H. Tan, Y. Liu, J. Zhang, Q. Zhang, and L. Ni, in the 16th Annual International Conference on Mobile Computing and Networking (MobiCom), Chicago, Illinois, September 2010 Interference is a critical issue in wireless communications. In a typical multiple-user environment, different users may severely interfere with each other. Coordination among users therefore is an indispensable part of interference management in wireless networks. It is known that coordination among multiple nodes is a costly operation, taking a significant amount of valuable communication resources. In this paper, the authors have an interesting observation that by generating intended patterns, some simultaneous transmissions (i.e., “interference”) can be successfully decoded without degrading the effec-

tive throughput in the original transmission. As such, an extra and “free” coordination channel can be built. Based on this idea, the authors propose a DC-MAC to leverage this “free” channel for efficient medium access in a multiple-user wireless network. They theoretically analyze the capacity of this channel under different environments with various modulation schemes.

Enabling High-Bandwidth Vehicular Content Distribution U. Shevade, Y. Chen, L. Qiu, Y. Zhang, V. Chandar, M. Han, H. Song, and Y. Seung, in the 6th International Conference on emerging Networking EXperiments and Technologies (CoNEXT), Philadelphia, PA, November 2010 This paper presents VCD, a novel system for enabling high-bandwidth content distribution in vehicular networks. In VCD, a vehicle opportunistically communicates with nearby access points (APs) to download the content of interest. To fully take advantage of such

transient contact with APs, the authors proactively push content to the APs the vehicles are likely to visit in the near future. In this way, vehicles can enjoy the full wireless capacity instead of being bottlenecked by Internet connectivity, which is either slow or even unavailable. The authors develop a new algorithm for predicting the APs that will soon be visited by the vehicles. They then develop a replication scheme that leverages the synergy among (i) Internet connectivity (which is persistent but has limited coverage and low bandwidth), (ii) local wireless connectivity (which has high bandwidth but transient duration), (iii) vehicular relay connectivity (which has high bandwidth but high delay), and (iv) mesh connectivity among APs (which has high bandwidth but low coverage). The authors demonstrate the effectiveness of the VCD system using trace-driven simulation and Emulab emulation based on real taxi traces. They further deploy VCD in two vehicular networks, one using 802.11b and the other using 802.11n, to demonstrate its effectiveness.

Open Call from the IEEE Communications Society

A

re you enthusiastic? Have you performed quality reviews for technical periodicals? Demonstrated solid technical accomplishments? Have a reputation for high ethical standards and for reliability?

www .com

soc.

org/

edit

or

You may be ready ... The IEEE Communications Society is looking for volunteers who are interested in becoming part of a prestigious Communications Society editorial board. Duties include: A commitment to handle at least two manuscripts per month; arrange for three reviews or more in a timely fashion; and the ability to make ﬁrm and fair decisions. Qualiﬁcations: Subject matter expertise, editing experience, technical program committee experience; references, representative papers.

Apply at: www.comsoc.org/editor The decision to appoint an editor rests with the Editor-in-Chief of the journal/magazine. Please note that it will not be possible to send individual acknowledgments to all applicants.

IEEE Wireless Communications • February 2011

5

LIN LAYOUT

2/7/11

12:04 PM

Page 6

ACCEPTED FROM OPEN CALL

IMS EMERGENCY SERVICES: A PRELIMINARY STUDY YI-BING LIN, NATIONAL CHIAO TUNG UNIVERSITY MENG-HSUN TSAI, NATIONAL CHENG KUNG UNIVERSITY YUAN-KUANG TU, CHUNGHWA TELECOM

ABSTRACT Fixed network

(3) MGC GW

Emergency call and walkie-talkie are two services utilized in emergency situations. During Typhoon Morakot in 2009, we experienced the deficiency of emergency call service that cannot continuously track callers in real time and walkie-talkie communications where a speaker may not be granted the permission to talk. These issues can be resolved by the emergency call and push-to-talk over cellular services in IP Multimedia Subsystem. This article conducts a preliminary study on how these two services can be effectively exercised in IMS.

INTRODUCTION Deficiencies in communications during emergencies can be resolved by the emergency call and the Push-to-talk over Cellular services in IP Multimedia Subsystem network. The authors conduct a preliminary study on how these two services can be effectively exercised in IMS.

6

IP Multimedia Subsystem (IMS) supports IPbased multimedia services. IMS was originally designed by the Third Generation Partnership Project (3GPP) to deliver Internet services over general packet radio service (GPRS) in 3G networks such as Universal Mobile Telecommunications System (UMTS). IMS was later updated to support other access networks including wireless LAN, CDMA2000, and fixed line. For the purpose of this article, Fig. 1 illustrates a simplified IMS network architecture (the reader is referred to [1] for detailed descriptions of IMS). The IMS (Fig. 1b) connects to both mobile and fixed telecommunications networks (Fig. 1a) for fixed mobile convergence (FMC). IMS is not intended to standardize applications or services. Instead, it provides a standard approach for voice/multimedia application access from user equipment (UE; 1, Fig. 1) in wireline and wireless networks. This goal is achieved by having horizontal control that isolates the access networks from the service and application networks (Fig. 1c). In the IMS, the transport of user data is separated from that for control signals, where IETF protocols such as Session Initiation Protocol (SIP) [2] are used to ease the integration with the Internet. For example, the call session control function (CSCF, 5, Fig. 1) is a SIP server, which is responsible for call control. The media gateway control function (MGCF, 3, Fig. 1) con-

1536-1284/11/$25.00 © 2011 IEEE

trols the connection for media channels in a media gateway (MGW, 4, Fig. 1). The MGW connects toward the legacy fixed and mobile networks to provide user data transport. The home subscriber server (HSS, 2, Fig. 1) is the master database containing all user-related mobile subscription and location information. In 2004 Chunghwa Telecom deployed the first commercial IMS in Taiwan to provision commercial telecommunications services such as voice, video, and Internet-based multimedia services. The initial capacity was 125,000 subscribers. Current IMS capacity can accommodate about 500,000 subscribers in daily commercial operation. During Typhoon Morakot in August 2009, Taiwan experienced serious damage from flooding and mudslides (Fig. 2, left), and the rescue missions solely relied on GSM and satellite communications that offer basic services (Fig. 2, right) such as emergency call without location tracking. From Typhoon Morakot, we learned that it is desirable to accommodate emergency services in IMS with a 3G network including emergency call and push-to-talk over cellular (PoC). A GSM user in Taiwan can make an emergency call by dialing 110, 112, or 119. However, the existing GSM emergency call service only identifies the location of the caller at the time of call setup and does not track the user’s location during the call. In Typhoon Morakot many people waiting for rescue could not be accurately located through their phone calls, which increased the difficulty of the rescue missions. PoC is a walkie-talkie-like service defined in Open Mobile Alliance (OMA) specifications [3, 4]. In 2004 Chunghwa Telecom first launched this service in Asia using 2.5G technology. Our experience with 2.5G PoC included long PoC call setup time (13 s) and handoff time (2.5–3 s for reconnecting a PoC client when it moved from one base station to another). These problems have been resolved by IMS-based PoC established on 3G networks. For example, the 3G PoC call setup time is less than 6 s. Although it is not clear if PoC will be a successful residential service, it has proven useful for business corporations and government organizations such as the National Security Bureau in Taiwan.

IEEE Wireless Communications • February 2011

LIN LAYOUT

2/7/11

12:04 PM

Page 7

In the past three years, we have studied emergency call [5] and PoC [6]. During Typhoon Morakot in 2009, emergency calls and walkietalkies were important means of communications in rescue missions. Clearly, it is desirable to support emergency services in IMS for disastrous events. Therefore, this article conducts a preliminary investigation on IMS emergency call and PoC.

(a)

LOCATION TRACKING FOR EMERGENCY CALL

(b)

An important feature of emergency call is that the system can track the location of a calling UE (1, Fig. 1) during the conversation. To support IMS emergency call, three network nodes are deployed. When the UE originates an emergency call, the call is established by a special CSCF called an emergency-CSCF (E-CSCF, 5, Fig. 1), which dispatches the call to the nearest public safety answering point (PSAP) based on the location information of the UE. The PSAP (7, Fig. 1) is an IMS application server that processes emergency calls according to the types of emergency events. For example, in a fire event the PSAP connects the UE (the caller) to the fire department (8, Fig. 1). The PSAP interacts with the E-CSCF by using SIP, and the voice conversation path is set up through the MGW (4, Fig. 1) to the UE by using Real-Time Transport Protocol (RTP) [7]. The gateway mobile location center (GMLC, 6, Fig. 1) supports a location service (LCS) [8]. Through Signaling System Number 7 (SS7) Mobile Application Part (MAP) [1], the GMLC interacts with the HSS and mobile network to obtain the accurate location of UE. The GMLC provides the location information to the PSAP and E-CSCF. The LCS merits further discussion. This service utilizes one or more positioning methods between the mobile network (Fig. 1a) and UE to determine the location of the UE [9]. The cellID-based positioning method determines the UE’s position based on the coverage of service areas (SAs). An SA includes one or more cells (base stations). The observed time difference of arrival (OTDOA) and uplink time difference of arrival (U-TDOA) positioning methods utilize trilateration to determine the UE’s position based on the time differences between downlink and uplink signal arrivals, respectively. The Assisted Global Positioning System (A-GPS) method speeds up GPS positioning by downloading GPS information through the mobile network. In Chunghwa Telecom A-GPS is utilized for location-based services. Without loss of generality, we consider the cell-ID-based method. After emergency call setup, the PSAP may need to monitor the UE’s location in real time. In the 3GPP 23.167 specification [10], the UE’s location is tracked through polling, where the PSAP periodically queries the UE’s location. For description purposes, we refer to the 3GPP 23.167 approach as the location polling scheme. In this scheme, if the UE does not change its location between two queries, the second query is wasted (this is called redundant polling). On the other hand, if the UE has visited several locations between two location queries, the PSAP may lose track of the UE in this time period (this is called mistracking). To

IEEE Wireless Communications • February 2011

(1) UE Fixed network

Mobile network

(3) MGCF

(2) HSS

(4) MGW

(5) CSCF

(c) (8) Fire department (9) PoC server

Police department (7) PSAP

(6) GMLC

Figure 1. A simplified IMS network architecture (dashed lines: control signaling; solid lines: user data/control signaling): a) fixed and mobile telecom networks; b) IP Multimedia Subsystem; c) service and application networks.

resolve these issues, the active location reporting scheme was proposed in [5]. This scheme reports the UE’s location upon change of its SA. This section describes location polling and active location reporting, and comments on their performance.

EMERGENCY CALL SETUP Figure 3 illustrates IMS emergency call setup message flow with the following steps [10]: Step A.1 The UE establishes IP connectivity to the IMS through the mobile network [1]. Step A.2 The UE sends a SIP INVITE message to the E-CSCF. This message includes the supported positioning methods of the UE (cell-ID-based in our example). Step A.3 The E-CSCF uses the received information to select a GMLC and sends the Emergency Location Request message to the GMLC. Steps A.4 and A.5 The GMLC exchanges the SS7 MAP_SEND_ROUTING_INFO_FOR_LCS and acknowledgment message pair with the HSS to identify the mobile network node responsible for connection to the UE. In UMTS this node is a serving GPRS support node (SGSN). Step A.6 The GMLC sends the SS7 MAP_PROVIDE_SUBSCRIBER_LOCATION message. Step A.7 The mobile network and UE exercise the cell-ID-based positioning procedure to obtain the location estimate information of the UE (i.e., the SA identity of the UE). Step A.8 The mobile network returns the SA identity to the GMLC through SS7 MAP_PROVIDE_SUBSCRIBER_LOCATION_ack message. Step A.9 The GMLC selects a suitable PSAP according to the SA of the UE and replies the Emergency Location Response message (with the selected PSAP address) to the E-CSCF.

7

LIN LAYOUT

2/7/11

12:04 PM

Redundant polling creates extra network traffic without providing useful location information. Furthermore, mis-tracking may result in wrong positioning in case of emergency situations. These issues are resolved by Active Location Reporting.

Page 8

Figure 2. Telecommunications services in Typhoon Morakot: (left) deploying temporary cables in flooded areas; (right) GSM/satellite communications through a vehicular base station.

Steps A.10–A.12 The E-CSCF forwards the SIP INVITE to the PSAP to set up the call. The PSAP and the UE exchange the 200 OK and the SIP ACK messages through the E-CSCF. After the PSAP has received the SIP ACK message, the emergency call is established. Step A.13 The GMLC sends the location information obtained at step A.8 to the PSAP after the call has been established, where the PSAP address is resolved by the GMLC at step A.9.

LOCATION POLLING UE may move during an emergency call, and the PSAP needs to monitor the UE’s location in real time. In location polling, the PSAP periodically queries the UE’s location. In each query, the following steps are executed (Fig. 4) [8]: Step B.1 The PSAP sends the Location Information Request message to the GMLC. Steps B.2–B.6 These steps retrieve the UE’s SA identity, which are similar to steps A.4–A.8 in Fig. 3. Step B.7 The GMLC returns the SA identity of the UE to the PSAP. Steps B.8–B.10 When the emergency call is terminated, the E-CSCF exchanges the Emergency Location Release and Response message pair with the GMLC to terminate location tracking.

ACTIVE LOCATION REPORTING To resolve redundant polling and mistracking issues in location polling, the active location reporting scheme was proposed in [5], which reports the UE’s location upon change of its SA. This scheme introduces a new locationEstimateType initiateActiveReport (to trigger active location reporting) in the MAP_PROVIDE_ SUBSCRIBER_LOCATION message (at step A.6). Since the IP connectivity exists during the IMS emergency call, the UE is in the cell-connected state and is tracked by the mobile network at the cell level [1]. Therefore, the mobile network can detect when the UE moves from one base station to another, and report the new SA identity to the GMLC. In this approach the GMLC maintains a UE-PSAP mapping table to store the (UE, PSAP) pair at step A.9. The GMLC does not

8

need to query the HSS to identify the mobile network node responsible for connection to the UE (i.e., steps B.2 and B.3 are eliminated). The active location reporting scheme is illustrated in Fig. 5 with the following steps: Step C.1 When the UE moves to a new SA, the mobile network detects this movement at the cell tracking mode and then triggers the positioning procedure. Step C.2 After the positioning procedure is executed, the UE’s SA identity is obtained. Step C.3 The mobile network sends the SS7 MAP_SUBSCRIBER_LOCATION_REPORT message with the SA identity to the GMLC. Step C.4 From the UE-PSAP mapping table, the GMLC retrieves the PSAP address of the UE stored at step A.9 and then sends the updated location information to the PSAP. When the emergency call is terminated, the following steps are executed: Step C.5 When the IMS call is released, the UE moves from the cell-connected mode to the idle mode, and the mobile network no longer tracks the movement of the UE. Step C.6 The E-CSCF sends the Emergency Location Release message to the GMLC to terminate location tracking. Step C.7 The GMLC returns the Emergency Location Response message to the E-CSCF and then deletes the (UE, PSAP) mapping from the UE-PSAP table. Note that steps C.1 and C.2 in active location reporting automatically detects the movement of UE, which is different from steps B.1–B.5 in location polling.

PRELIMINARY PERFORMANCE EVALUATION It is clear that redundant polling creates extra network traffic without providing useful location information. Furthermore, mistracking may result in wrong positioning in case of emergency situations. These issues are resolved by active location reporting. However, it is desirable to evaluate the performance of location polling to justify the modifications to the existing location tracking procedure in active location reporting. Suppose that the SA residence time has a Gamma distribution with mean 1/μ and variance

IEEE Wireless Communications • February 2011

LIN LAYOUT

2/7/11

12:04 PM

Page 9

Mobile network PSAP

GMLC

E-CSCF

HSS

UE A.1. IMS connection establishment A.2. SIP INVITE

A.3. Emergency Location Request Report initial location A.4. MAP_SEND_ROUTING_INFO_FOR_LCS A.5. MAP_SEND_ROUTING_INFO_FOR_LCS_ack

In the PoC service several predefined group members participate in one PoC session. Since PoC utilizes half-duplex communications, only one PoC member speaks at a time, and others listen.

A.6. MAP_PROVIDE-SUBSCRIBER_LOCATION A.7. Positioning A.8. MAP_PROVIDE-SUBSCRIBER_LOCATION_ack

A.9. Emergency location response Establish emergency call A.10. SIP INVITE A.11. 200 OK A.12. SIP ACK

A.11. 200 OK A.12. SIP ACK

A.13. Location information

Figure 3. IMS emergency call setup.

V m (other distributions have shown similar results [5]). The inter-query interval is a fixed period 1/λ in location polling. Several output measures are studied. Let α be the mistracking probability. An SA crossing is mistracked if there is no query between this and the next SA crossings (i.e., the system does not know that the user has visited this SA). Let β be the probability that redundant queries exist between two SA crossings. The larger the α or β values, the worse the performance of location polling. Figure 6a shows intuitive results that as the polling frequency λ increases, α decreases and β increases. We describe two effects of Vm: • Effect 1: When the SA residence time intervals become more irregular (i.e., V m increases), more SA residence time intervals without any query are observed. • Effect 2: When Vm increases, if a query arrives in an SA residence time interval, more than one query will tend to arrive in this interval. Effect 1 implies that as V m increases, more SA crossings are mistracked, and we have a nontrivial observation that α increases as Vm increas-

IEEE Wireless Communications • February 2011

es. Similarly, when λ is large (e.g., λ ≥ 5 μ), effect 1 is significant, and β is a decreasing function of Vm. The impact of Vm on β is more subtle when λ = μ. In this case β increases and then decreases as V m increases, which implies that when Vm is small, effect 2 is more significant. On the other hand, when Vm is large, effect 1 dominates. An important observation is that when 0.1/μ 2 < V m < 10/μ 2 , both α and β values are non-negligibly large, and poor performance of location polling cannot be ignored. Besides mistracking and redundant polling probabilities, we would also like to investigate the following output measures: • T i : The expected period in which the PSAP cannot correctly track the UE’s location (i.e., T i is the period between an SA crossing and when the next query arrives). In this period the system does not know the user’s correct location. • N R : The expected number of redundant queries between two SA crossings (i.e., NR is the number of queries issued within two consecutive SA crossings).

9

LIN LAYOUT

2/7/11

12:04 PM

In emergency situations, important messages may not be delivered in walkietalkie like communications if the message sender cannot obtain the permission to talk. Therefore, a mechanism is desirable to guarantee that a PoC client has a fair chance to talk. This issue can be resolved by PoC with queuing option.

Page 10

Mobile network PSAP

GMLC

E-CSCF

HSS

UE

Call setup

Report current location B.1. Location Information Request B.2. MAP_SEND_ROUTING_INFO_FOR_LCS B.3. MAP_SEND_ROUTING_INFO_FOR_LCS_ack B.4. MAP_PROVIDE_SUBSCRIBER_LOCATION B.5. Positioning B.6. MAP_PROVIDE_SUBSCRIBER_LOCATION_ack B.7. Location information

B.8. Emergency call release B.9. Emergency location release B.10. Emergency location response

Figure 4. Location polling. The larger the above output values, the worse the performance of location tracking. Figure 6b plots T i against λ and V m (the solid curves), where T i is measured by 1/μ. T i is a decreasing function of λ and is not affected by V m . We notice that for λ < 10 μ, the PSAP will mistrack UE more than 5 percent of the time, which may not be acceptable in emergency situations. It is clear that when λ > μ, N R ⋅=⋅ (λ/μ) – 1. Let NR* be the number of redundant queries between two SA crossings under the condition that there is at least one redundant query in this interval. Figure 6b plots NR* (the dashed curves). Clearly, NR* is an increasing function of λ. When Vm < 1/μ2, NR* ⋅=⋅ NR. When Vm > 1/μ2, NR* significantly increases as Vm increases. This phenomenon implies that when the UE’s movement pattern becomes irregular, the PSAP will receive many redundant (useless) location reports after it has received a correct location update.

PUSH-TO-TALK OVER CELLULAR In the PoC service several predefined group members participate in one PoC session. Since PoC utilizes half-duplex communications, only one PoC member speaks at a time, and others listen. When a PoC member attempts to speak,

10

he/she presses the push-to-talk button of his/her UE (Fig. 1 (1)) to ask for permission. This UE with the PoC application installed is called the PoC client. The PoC server (9, Fig. 1) is responsible for handling PoC session management (create or delete a PoC session). It arbitrates speak permission through the talk burst control mechanism [4]. SIP and Session Description Protocol (SDP) [11] are utilized for session establishment. After the PoC session is established, each of the PoC clients has built an RTP session with the PoC server through the MGW (Fig. 1 (4)). The talk burst control messages between the PoC clients and the PoC server are carried out by Real-Time Transmission Control Protocol (RTCP) packets [7]. In SIP a new parameter, tb_grant, is added in SDP’s attribute field so that the PoC server can arbitrate the speak permission during a session. If tb_grant = 1, the PoC client is granted the permission to speak. Otherwise (i.e., tb_grant = 0), the PoC client is not permitted to talk. In emergency situations, important messages may not be delivered via walkie-talkie-like communications if the message sender cannot obtain permission to talk. Therefore, a mechanism is desirable to guarantee that a PoC client has a fair chance to talk. This issue can be resolved by PoC with queueing option.

IEEE Wireless Communications • February 2011

LIN LAYOUT

2/7/11

12:04 PM

Page 11

Mobile network PSAP

GMLC

E-CSCF

UE

Call setup

Report current location C.1. Change of dervice area C.2. Positioning C.3. MAP_SUBSCRIBER_LOCATION_REPORT

A queueing option is provided in the talk burst control mechanism. If this option is selected, then the ungranted requests are buffered in the queue at the PoC server instead of being denied.

C.4. Location information

C.5. Emergency call release C.6. Emergency location release C.7. Emergency location response

Figure 5. Active location reporting.

TALK BURST CONTROL MECHANISM The talk burst control mechanism is implemented by finite state machines (FSMs) on both the server and client sides. Figure 7 illustrates simplified talk burst control FSMs for PoC server (called FSM G ) and PoC client (called FSM U ). For every PoC session, there is one FSMG in the PoC server and an FSM U in each of the PoC clients. Two timers are defined in FSMG. Timer T2 is used to determine whether the PoC client speaks too long. If a PoC client talks longer than the T2 period, he/she is asked to release the permission. Timer T3 is used to gracefully terminate the talk burst. After the PoC client gives up permission, the transient RTP packets it generated in T2 are continuously forwarded by the PoC server in the T3 period. When T3 expires, the PoC server allows the next PoC client to talk.

Session Initiation — For the PoC clients and PoC server involved in a PoC session, their FSMs are initialized at the start-stop state. The session initiator (a PoC client) starts a PoC session by sending a SIP INVITE message to the PoC server. The PoC server then broadcasts SIP INVITE messages with tb_grant = 0 to the invited PoC clients (i.e., PoC group members other than the session initiator). Each of the invited PoC clients answers with a SIP 200 OK message, and its FSMU enters has no permission (transition 2, Fig. 7b). This PoC client is not permitted to speak. After receiving the first SIP 200 OK message from each of the invited PoC clients, the PoC server replies a SIP 200 OK with tb_grant = 1 to

IEEE Wireless Communications • February 2011

the session initiator, and FSMG enters TB_Taken (transition 1, Fig. 7a). This state means that some PoC client (the session initiator in this case) has obtained the permission. FSMU of the session initiator enters has permission (transition 1, Fig. 7b), and the PoC client is allowed to speak. The session initiator becomes the permitted PoC client (i.e., the PoC client allowed to speak), and the invited PoC clients become listening PoC clients (i.e., the PoC clients not permitted to speak). At this moment, the session initiator speaks, and all invited PoC clients listen.

Permission Releasing — After finishing the talk, the permitted PoC client X releases the permission by sending the TB_Release message to the PoC server, and its FSM U enters pending TB_Release (Transition 6 in Fig. 7b). In this state, client X stops sending media packets and waits for the response from the PoC server. FSMG enters pending TB_Release after the PoC server has received the TB_Release message (transition 2, Fig. 7a). In this state the PoC server keeps forwarding the transient media packets issued from client X before the TB_Release message. When the last transient media packet has been processed or T3 expires, FSMG enters TB_Idle (transition 3, Fig. 7a). In this state no PoC client is granted permission to speak. The PoC server broadcasts the TB_Idle message to all PoC clients. FSMU of client X enters has no permission upon receipt of the TB_Idle message (transition 7, Fig. 7b). A listening PoC client remains in has no permission when it receives the TB_Idle message (Transition 14 in

11

LIN LAYOUT

2/7/11

12:04 PM

Page 12

103

1.0

102

0.8

0.6

0.4

101

Solid: α; Dashed: β λ=μ λ=5μ λ=10μ

100

10-1

0.2

0.0 10-3

Solid: Ti; Dashed: N*R λ=μ λ=5μ λ=10μ

10-2

10-1

100

101

102

103

10-2 10-3

Vm (unit: 1/ 2) (a)

10-2

10-1

100

101

102

103

Vm (unit: 1/μ2) (b)

Figure 6. Emergency performance measures: a) α and β; b) Ti (unit: 1/μ) and N*R. Fig. 7b). At this point, all PoC clients can compete for permission to speak.

Permission Requesting — To obtain permission, a listening PoC client X sends the TB_Request message to the PoC server. Client X becomes a requesting PoC client (for speak permission), and its FSMU enters pending TB_Request (transition 3, Fig. 7b). This state indicates that client X is waiting for arbitration from the PoC server. If some other PoC client has been granted permission, the PoC server sends the TB_Deny message to client X, and FSMU of client X moves back to has no permission (transition 4, Fig. 7b). If the PoC server grants permission to client X, it sends the TB_Granted message to client X and the TB_Taken message to other PoC clients. FSMG enters TB_Taken (transition 4, Fig. 7a), and FSMU of client X enters has permission (transition 5, Fig. 7b). When a listening PoC client receives TB_Taken, its FSMU remains at has no permission, and it is not allowed to request permission. After client X has become the permitted PoC client, T2 at the PoC server is started. This timer is used to monitor if this client X speaks too long (and therefore should be revoked). Permission Revoking — If permitted client X speaks longer than the T2 period, the PoC server will send the TB_Revoke message to reclaim the permission and start the T3 timer. Upon receipt of the TB_Revoke message, FSMU of client X enters pending TB_Revoke (transition 8, Fig. 7b). FSM G enters pending TB_Revoke (transition 5, Fig. 7a). In this state the PoC server keeps forwarding transient media packets of client X until T3 expires. Then FSMG enters TB_Idle (transition 6, Fig. 7a). The PoC server sends the TB_Idle message to all PoC clients. FSM U of client X enters has no permission (transition 9, Fig. 7b) upon receipt of the TB_Idle message.

12

When a listening PoC client receives TB_Idle, its FSM U remains at has no permission. At this point, all PoC clients can compete for permission to speak.

Permission Queuing — A queueing option is provided in the talk burst control mechanism. If this option is selected, the ungranted requests are buffered in the queue at the PoC server instead of being denied. In this option, after the permitted PoC client finishes talking, the PoC server grants the next request from the queue. The state queued in FSMU (dark oval, Fig. 7b) indicates that a request of the client is buffered in the PoC server and will be granted later. After PoC client X has obtained permission, the PoC server may receive the TB_Request message from another requesting PoC client Y. With the queueing option, FSM G is in the TB_Taken state, and FSMU of client Y is in the pending TB_Request state. The PoC server buffers the TB_Request message in the queue and replies with the TB_Queued message to client Y. FSMG stays in TB_Taken, and FSMU of client Y enters queued (transition 10, Fig. 7b). Client Y is called a queued PoC client. If a queued PoC client is not patient, it will send the TB_Release message to the PoC server to cancel the request, and its FSM U moves back to has no permission (transition 11, Fig. 7b). In this case, the PoC server removes the corresponding request from the queue. The period between when a PoC client enters the queued state and when it enters the has no permission state is called the patient time. After the permission is released (or revoked), FSMG will enter TB_Idle, and FSMU of the permitted PoC client will enter pending TB_Release (transition 6, Fig. 7b) or pending TB_Revoke (Transition 8, Fig. 7b). If the queue is not empty, the PoC server grants permission to the next queued request instead of sending the

IEEE Wireless Communications • February 2011

LIN LAYOUT

2/7/11

12:04 PM

Page 13

Start-stop Start-stop

7

2

Any State 1

1

4

pending TB_Release

5

14

9

3

pending TB_Request 10

4

3

pending TB_Revoke 6

TB_Idle (a)

Any State

has no permission

TB_Taken 2

13

7 11

pending TB_Revoke

5 queued

pending TB_Release

12 6

8

has permission (b)

Figure 7. Talk burst control finite state machines for: a) PoC server (FSMG); b) PoC client (FSMU).

TB_Idle message. The next queued PoC client will receive the TB_Granted message from the PoC server, and its FSMU enters has permission (transition 12, Fig. 7b). This queued PoC client becomes the next permitted PoC client. At the same time, other PoC clients receive the TB_Taken message from the PoC server. Client X becomes a listening PoC client, and its FSMU enters has no permission (transition 7 or 9, Fig. 7b). FSMU of every listening PoC client remains has no permission, and FSM U of every queued PoC client remains queued.

Session Termination — When a PoC client leaves the PoC session, its FSMU moves back to startstop (transition 13, Fig. 7b). The PoC session remains active for other PoC clients. After all PoC clients have left the PoC session, the session is implicitly terminated. FSMG moves back to start-stop (transition 7, Fig. 7a).

PERFORMANCE EVALUATION Based on [4], we investigate the performance of the PoC talk burst control mechanism with queueing (approach Q) and without queueing (approach NQ). Define the revoking timer as TR = T2 + T3. Let the permission request arrivals of a PoC client form a Poisson process with rate λ. The speak time is a random variable with mean 1/μ. The patient time is a random variable with mean 1/ω. Both speak and patient times are assumed to be Exponentially distributed. Other distributions show similar results, and the reader is referred to [6] for details. Let M be the number of PoC clients in a PoC group. In Chunghwa Telecom’s commercial PoC service, M is limited to 20 (i.e., at most 20 members can participate in a PoC session). In this study we assume that 5 ≤ M ≤ 40. Two output measures are considered: • PD: The denied probability that the request of a PoC client is not granted because this client is not patient in approach Q or the PoC server rejects the request in approach NQ. • W: The expected waiting time between when a PoC client requests permission to speak and when it is granted permission under the condi-

IEEE Wireless Communications • February 2011

Our study indicates that PoC with queueing can comfortably accommodate twice as many participants as PoC without queueing, where the waiting time for a PoC participant to be granted permission is reasonable short.

tion that the PoC client is not immediately granted permission. The W measure excludes the waiting times of immediately granted requests (which are 0) so that we can answer the question “If you have to wait, how long will you wait?” Under the conditions that λ = 0.01μ and TR = 3/μ, Fig. 8a indicates that for the same PD performance, approach Q can support at least twice as many clients as approach NQ. For example, to maintain PD = 0.038, approach NQ can only support M = 5, while approach Q can support M = 10 (for ω = μ) and M = 40 (for ω = 0.1μ). Figure 8b illustrates the expected waiting time performance (WQ for approach Q and WNQ for approach NQ) before a PoC client is granted permission to speak (if he/she is not immediately accepted by the PoC server). To make a fair comparison, we set ω = 0 for approach Q so that W Q will not be shortened by impatience. For approach NQ, when a PoC client fails to obtain permission, it keeps requesting until permission is granted. Therefore, WNQ is the period between its first try and when it is granted permission. Figure 8b plots WQ and WNQ (measured by 1/μ) against M. The figure indicates that WQ is much shorter than WNQ. Furthermore, even if the number of PoC participants is large (e.g., M = 40), WQ is reasonably short (less than 2.5/μ).

CONCLUSIONS This article has conducted a preliminary investigation of two emergency services for IMS: emergency call and push-to-talk over cellular (PoC). In the IMS emergency call study we observe that the existing location tracking mechanism (called location polling) might mistrack the caller and cause unnecessary signaling overhead. We describe a modified mechanism called active location reporting, which may potentially enhance the performance of location polling. Based on this conclusion, most location-based services in Chunghwa Telecom have utilized an active-location-reporting-like mechanism. In the future this mechanism will be provided in IMS.

13

LIN LAYOUT

2/7/11

0.30

12:04 PM

Page 14

Solid: Q; Dashed: NQ ω=μ ω=0.1μ

102

Solid: WQ; Dashed: WNQ λ=0.02μ λ=0.01μ λ=0.005μ

0.25

0.20

101

0.15

0.10

100

0.05

10-1

0.00 5

10

15

20

25

30

35

40

5

10

15

20

25

30

35

40

M

M

(b)

(a)

Figure 8. Effects on PD, WQ, and WNQ (TR = 3/μ): a) PD (λ = 0.01μ); b) WQ and WNQ (unit: 1/μ), (ω = 0). For PoC, to guarantee that everyone can always obtain permission to talk in emergency situations, the queueing option of PoC should be selected. Our study indicates that PoC with queueing can comfortably accommodate twice as many participants as PoC without queueing, where the waiting time for a PoC participant to be granted permission is reasonably short. In the future, we will also consider the trade-off between priority and fairness in the queueing mechanism. In summary, our preliminary study provides guidelines to deploy emergency call and PoC in a commercial IMS network.

ACKNOWLEDGMENT Y.-B. Lin’s work was supported in part by NSC 97-2221-E-009-143-MY3, NSC 098001228630, NSC 97-2219-E-009-016, Intel, Chunghwa Telecom, ITRI, NCTU joint research center, and MoE ATU plan. M.-H. Tsai’s work was supported by NSC 100-2218-E-006-015-MY2.

REFERENCES [1] Y.-B. Lin and A.-C. Pang, Wireless and Mobile All-IP Networks, Wiley, 2005. [2] IETF RFC 3261, “SIP: Session Initiation Protocol,” 2002. [3] OMA, “Push to Talk over Cellular (PoC) — Architecture,” OMA-AD-PoC-V2 0 1-20080226-C Candidate Version 2.0, Feb. 26, 2008. [4] OMA, “Push to Talk over Cellular (PoC) — User Plane,” OMA-TS-PoC-V2 0 1-20080226-C Candidate Version 2.0, Feb. 26, 2008. [5] M.-H. Tsai, Y.-B. Lin, and H.-H.Wang, “Active Location Reporting for Emergency Call in UMTS IP Multimedia Subsystem,” IEEE Trans. Wireless Commun., vol. 8, no. 12, 2009, pp. 5837–43. [6] M.-H. Tsai and Y.-B. Lin, “Talk Burst Control for Push-toTalk over Cellular,” IEEE Trans. Wireless Commun., vol. 7, no. 7, 2008, pp. 2612–18. [7] IETF RFC 3550, “RTP: A Transport Protocol for Real-Time Applications,” 2003.

14

[8] 3GPP TS 23.271, “Functional Stage 2 Description of Location Services (LCS),” v. 7.9.0, 2007. [9] 3GPP TS 25.305, “Stage 2 Functional Specification of User Equipment (UE) Positioning in UTRAN,” v. 7.4.0, 2007. [10] 3GPP TS 23.167, “Internet Protocol (IP) based IP Multimedia Subsystem (IMS) Emergency Sessions,” v. 7.11.0, 2008. [11] IETF RFC 4566, “SDP: Session Description Protocol,” 2006.

BIOGRAPHIES YI-BING LIN [M‘95, SM‘95, F‘03] ([email protected]) is the dean and chair professor of the College of Computer Science, National Chiao Tung University, Taiwan. He is also an adjunct research fellow of the Institute of Information Science, Academia Sinica, Nankang, Taipei, Taiwan, and a consultant professor of Beijing Jiaotong University, China. His current research interests include wireless communications and mobile computing. He has published over 240 journal articles and more than 200 conference papers. He is a co-author of the books Wireless and Mobile Network Architecture with Imrich Chlamtac (Wiley), Wireless and Mobile All-IP Networks with Ai-Chun Pang (Wiley), and Charging for Mobile All-IP Telecommunications with Sok-Ian Sou (Wiley). He is an ACM Fellow, an AAAS Fellow, and an IET (IEE) Fellow. M ENG -H SUN T SAI [S‘04, M‘10] ([email protected]) received B.S., M.S., and Ph.D. degrees from NCTU, Hsinchu in 2002, 2004, and 2009, respectively. He joined the Department of Computer Science and Information Engineering, National Cheng Kung University, Tainan, Taiwan, as an assistant professor in 2010. His current research interests include design and analysis of personal communications services networks, mobile computing, and performance modeling. YUAN-KUANG TU [M‘90] joined Chunghwa Telecom in 1981, working in a variety of R&D and management positions in the Telecommunication Laboratories, and served as Vice President in 2006. In 2007 he was assigned senior managing director of the Corporate Planning Department in Headquarters, in charge of strategic planning and new business development. In May 2009 he was promoted to president of Chunghwa Telecommunication Laboratories. He has published over 60 journal papers and holds 20 patents. His professional fields include optical communications, broadband networks, and mobile networks.

IEEE Wireless Communications • February 2011

ROSSETTO LAYOUT

2/7/11

10:35 AM

Page 15

ACCEPTED FROM OPEN CALL

MIXING NETWORK CODING AND COOPERATION FOR RELIABLE WIRELESS COMMUNICATIONS FRANCESCO ROSSETTO, DLR (GERMAN AEROSPACE CENTER) MICHELE ZORZI, UNIVERSITY OF PADOVA

ABSTRACT R

Traditional routing

R

Cooperation and Network Coding are two powerful techniques for wireless networks. During the past four years a surge of research activity has tried to combine the best of these two philosophies to improve the performance and error correction capabilities of radio networks.

Cooperation and network coding are two powerful techniques for wireless networks. During the past few years a surge of research activity has tried to combine the best of these two philosophies to improve the performance and error correction capabilities of radio networks. This article gives an overview of the literature in the field, summarizes the main achievements and design guidelines found so far, and finally highlights the main challenges yet to be solved.

INTRODUCTION In the past decade network coding (NC) [1] and cooperation [2] have emerged as two techniques that may deeply impact wireless networks. They tackle different issues and show complementary problems. In the former, packets are allowed to be combined together at intermediate nodes so as to reduce the redundancy needed to reliably deliver them. In other words, nodes collaborate to reduce the overhead. In the latter family of strategies, nodes cooperate by relaying the same data from different terminals so as to protect information with spatial diversity and improve the reliability and overall performance. They both aim to deliver information as efficiently as possible, but by different means. We remark that in this article cooperation is considered in the domain of physical layer relaying, in the sense of [2], with focus on the improvement of error correction capabilities. Hence, although in Network Coding nodes also “cooperate” (i.e., they share information to reach a common goal), the definition of cooperation we adopt in this article does not include Network Coding as a special case. It comes at no surprise that there has been a recent surge in the efforts to bring these two approaches together. However, three main philosophies stand out in this pursuit. In the first one, network coding is tightly coupled with channel coding (given the ties of channel coding to both the discussed systems). In the second The first author performed the work while at the University of Padova, Italy.

IEEE Wireless Communications • February 2011

approach (spurred by the groundbreaking concept of physical layer network coding [PNC] [3, 14]), packets are linearly combined on the wireless channel by means of collisions. Finally, in the third branch, NC is made more robust by means of techniques drawn from multiple-input multiple-output (MIMO). The purpose of this article is to summarize the main discoveries in this field and point out the first design guidelines that are emerging. We outline the results for channel/network coding, PNC, and MIMO_NC, respectively. The concluding remarks are then drawn.

HYBRID CHANNEL/NETWORK CODING Before any discussion, it is insightful to introduce the reference topology of this article, because virtually all interesting aspects and issues arise in its analysis. Such a scenario is depicted in Fig. 1: a set of P sources R1, …, SP send their data to a set of destinations D1, …, DK. They are supported by M relays S1, …, RM, who send redundancy based on the signals received from S 1 , …, S P . We remark that these three groups of nodes (sources, relays and destinations) are not necessarily disjoint. For instance, some sources may be destinations for other sources. In this model, P sources generate one new frame each and a total of N = P + M transmissions are performed. Note that NC is known to bring substantial benefits when different flows must be routed through the same terminals, as is indeed the case in Fig. 1, since the relays are shared by all sources. Such a scenario is rather general and applies to a variety of contexts. For instance, it is representative of the uplink of a cellular system, where there is only K =1 destination D1 (i.e., the base station). Hybrid channel/NC has chiefly been applied in this environment, whose underlying goal is to improve the efficiency of conventional cooperative protocols so that fewer relays are needed to achieve a certain degree of reliability; conversely, higher performance can be attained with the same amount of redundancy. Instead, in physical layer NC the topology is another instance of Fig. 1 with multiple destinations, but again the flows are constrained to go through a common set of relays. Channel coding is intimately related to coop-

1536-1284/11/$25.00 © 2011 IEEE

15

ROSSETTO LAYOUT

2/7/11

10:35 AM

Page 16

R1 RM

SP DK

S1 D1

Figure 1. The reference scenario for hybrid channel network coding.

Reference

P

N

rate

Diversity order

Notes

[5] (JNCC)

2

3

1/3

2

Third node is a relay

[6] (ANCC)

P

2P

1/2

≤P

[7]

2

3

1/3

2

Third node is a relay

[8]

2

2

1/3

2

No relays

Table 1. Comparison of cooperative channel/network coding schemes.

eration, especially coded cooperation. On the other hand, NC is akin to fountain codes, although their applications are often very different. It is then very natural that channel coding plays a crucial role in bringing together these worlds [4–8]. In the first and most widely analyzed approach, the transmissions are structured into two rounds [5, 6]. In the first part, each source sends its own information unit (separately encoded from the others). The destination and relays try to decode these frames, and the cooperators concatenate the decoded information units to generate additional redundancy, which is transmitted in the second round. Hence, a larger channel code that takes as input all decoded information units is created, thus obtaining spatial diversity. This effectively turns NC into a coded cooperative protocol based on Type II hybrid automatic repeat request (HARQ), so the observed benefits are due to the joint channel coding across nodes. This method of combining NC and channel coding is by far the most studied in this area because of its natural connections to these two techniques and relative ease of implementation. However, at least two important exceptions arise. Yang et al. [7] study a P = 2, N = 3 uplink (two sources, one shared cooperator), whose relay will forward the log likelihood ratios (LLRs) of the network coded packets received from the sources. Note that these quantities can be computed also if one or both of the information units are corrupted; hence, the relay can also cooperate with

16

unreliable information about the original data, while in the previous approach [5, 6] redundancy is generated only on correctly decoded frames. However, this comes at the price of additional bandwidth consumption, since the LLRs rather than the bits themselves are sent. In the third approach [8], there are no additional relays (i.e., N = P = 2); nonetheless, a diversity order of 2 is achievable. Such a result is possible because each source knows its own information, and this leads to an encoding scheme in which each partner transmits the algebraic superposition of its local and relayed information. Decoding at the destination is then carried out by iterating between the codewords from the two partners. Reference [8] is important also because it highlighted the so-called Cooperative Dilemma: if a terminal has to transmit two packets at the same time (e.g., because the first packet is a frame of its own and the second is a relayed one), it can either sum their Galois symbols and then modulate them, or apply superposition coding by summing their modulated signals. In the latter case, the node must strike a trade-off between the power to allocate between the first and the second packet. This can be very detrimental, because subtracting power from the relayed packet will reduce the effectiveness of cooperation, while subtracting power from the endogenous frame will make it harder to decode that packet both by the intended destination and by any possible relay. The scheme in [8] uses Galois arithmetic rather than superposition coding, because multiplexing two bit streams by means of Boolean addition does not require to divide the transmitted power between the two frames. The importance of this detail will be apparent also in Physical Layer Network Coding, where it will be shown that superposition coding can imply a loss of 3 dB with respect to Galois arithmetic. Table 1 summarizes the performance of these three strategies. The shown parameters are P, N, the coding rate, the achieved diversity order (i.e., the slope of the bit error rate [BER] curve vs. signal-to-noise ratio [SNR] in a log-log scale under Rayleigh fading), and the scenario for which these schemes are designed. If N > P, there are more transmitters than information units. It can be noticed that diversity is achieved in all schemes by means of an additional relay. The only exception is [8], for the aforementioned reasons. As an additional remark, [9] has shown that separate channel and network coding cannot achieve full diversity order N. In order to attain this goal, lower channel code rates are needed. This is indirectly confirmed by all the previous network-channel coding systems, since their code rate is relatively low (at most 1/2). To summarize, the studies on hybrid channel/network coding have delineated a few important points: • The performance gap between separate and joint channel/network coding is about 2–3 dB [5, 6, 8]. • The usage of incorrectly decoded frames to generate additional redundancy leads to bandwidth expansion [7]. • The achievable diversity order increases as the code rate is decreased [8, 9].

IEEE Wireless Communications • February 2011

ROSSETTO LAYOUT

2/7/11

10:35 AM

Page 17

In this research area, significant results have been achieved for idealized settings. Nonetheless, some issues are still open. First of all, all efforts focus on metrics taken from digital communication theory (e.g., diversity order, BER) or information theory (e.g., capacity regions), while networking issues have not received much attention so far. Another problem is the lack of evaluation of these techniques in realistic environments, either because the systems are impractical (and mainly show performance bounds, rather than evaluating real world protocols or network layer issues spurred by this techniques) or the communication protocols are too idealized. In conclusion, the design and performance of these systems in more realistic scenarios including even a minimal protocol stack is yet to be investigated in detail, and appears to be a promising area of research.

x1

A

IEEE Wireless Communications • February 2011

R

y3

B

x4 Traditional routing

x1

A

y2

R

B

x⊕y 3

PHYSICAL LAYER NETWORK CODING Another approach that has aroused much excitement has been so-called analog NC or physical layer NC [3, 14]. Let us refer to Fig. 2, which illustrates the well-known two-way relay channel. 1 Nodes A and B must exchange two packets (X from A and Y from B) through relay R. If the channels are assumed to be error-free, this frame exchange would entail four transmissions in traditional routing: first X is sent from A to R, then Y from B to R, X from R to B, and finally Y from R to A. Two phases can be distinguished: first, A and B must send their packets to R (multiple access phase), and then R has to send these frames to the destinations (broadcast phase). Digital NC (DNC) [1] works only on the broadcast phase, by sending an invertible function of X and Y. Nodes A and B can recover all information since they know X and Y, respectively. Such strategy reduces the number of required transmissions from 4 to 3. Physical layer Network Coding takes off from the observation that R need not know X and Y separately, but it is enough to decode a function of them (e.g., their sum).2 Hence, in PNC multiple access phase, A and B are allowed to transmit simultaneously. The channel additivity (i.e., the collision, in networking jargon) naturally computes a linear combination of the two packets and this function will then be broadcast to A and B. The clear advantage is that just 2 slots are required to deliver the frames, rather than 3 in DNC or 4 in traditional routing. Of course, several details need to be filled in this picture, and the vast majority of the research has focused on the two-way relay channel as a simple and neat starting point. We shall assume throughout this section that this is the case of interest. Three main approaches have emerged [3, 10–17]: •Amplify and Forward PNC (AF-PNC): Also known as Analog Network Coding (ANC). According to this idea [10, 11, 14, 16], the relay decodes neither X, Y nor their sum. Hence, it deals only with the analog signals that hºave collided during the packet reception. The resulting signal is amplified and broadcast, and the two extreme nodes decode the intended packet after

y2

Digital network coding x+y1

A

R

B

f(x+y)/g(x,y) 2 Analog network coding

Figure 2. Different protocols for the two way relay channel. The subscript to a packet is the time slot in which this frame is sent.

subtracting the frame that they sent (which is known, of course). The advantage of this scheme, compared to the other flavors of PNC, is its simplicity at the relay. On the other hand, the relay also amplifies the noise with which the final destinations (A and B) must cope. And last but not least, A and B need to carry out non-trivial interference cancellation that requires channel information. This sophisticated signal processing has actually been implemented, but it requires considerable effort and works only in some specific proof-of-concept scenarios [10]. It is yet unknown how to generalize it to less particular settings at affordable complexity. •Decode and forward PNC type 1 (DF-PNC 1): Also known as compute and forward, this strategy [3, 12, 15] suggests decoding the sum of X and Y at the relay, but at neither X nor Y individually. This strategy is inherently more complicated than ANC, because the sum of the signals from A and B is less structured than, for instance, the modulation to which X and Y belong. All these authors try to solve this issue with lattices, which are mathematical subsets of R N such that the sum of any two elements of these subsets still belongs to the lattice. Under simplifying assumptions on the channel coefficients between A, R and B, this scheme works well, because it is not subject to error amplification, and can effectively support higher rates

1 Note that the two-way relay channel is a special case of Fig. 1, with two sources, one relay, and two destinations. We also remark that each source is the destination for the other source. 2 This statement is certainly true for the two-way relay channel. However, R can no longer compute any other function of X and Y except for the one it has received over the air. This can be a problem in networks where nodes should keep on producing innovative packets, because in this case R cannot compute, for instance, any linear combination of X and Y independent of what has been received. Whether such a problem offsets the potential gains of PNC in large networks is yet to be explored.

17

ROSSETTO LAYOUT

4

2/7/11

10:35 AM

Page 18

C DF − PNC 1 =

DNC ANC DF-PNC 1 DF-PNC 2

3.5

4. DF-PNC 2 [13] C DF − PNC

Max throughput (bits/s/Hz)

3

2

1 ⎛1 ⎞ = min ⎜ log(1 + 2 Λ ), log(1 + Λ )⎟ ⎝4 ⎠ 2

2.5

2 1.5

1

0.5 0 -10

-5

0

5

10 SNR (dB)

15

20

25

Figure 3. Comparison of DNC and the three main approaches to PNC. than ANC. However, lattice-based schemes are extremely sensitive to timing and carrier synchronization; that is, A and B must be symbol and phase synchronous at R, which may be hard to achieve. •Decode and forward PNC type 2 (DF-PNC 2): In this approach [13] the relay decodes X and Y out of their superimposed signal by means of multiuser detection. While ANC and DF-PNC 1 base their transmission on a complex linear combination of the transmitted signals X and Y, DFPNC 2 separately decodes the two superimposed frames. After that, R transmits a linear combination of the digital packets. This approach is not affected by noise propagation (which besets ANC), but requires to decode more information than DF-PNC 1. As we will show, this matter can heavily affect the system performance. In order to compare these three approaches, we have computed the achievable capacities for ANC, DF-PNC 1 and DF-PNC 2 for the two way relay network with AWGN channels (see also [17]). Let us assume, for simplicity, that both A-R and B-R links are subject to the same SNR?. The maximum rate for each flow is the minimum between the capacity of the multiple access phase and the capacity of the broadcast phase. The four protocols to be compared are DNC, ANC, DF-PNC 1 and DF-PNC 2, and their capacities can be computed as follows: 1. DNC 1 C DNC = log(1 + Λ ) 3 2. ANC [10, 11, 13, 14, 16] C ANC =

⎛ 1 Λ2 ⎞ log ⎜ 1 + 2 3Λ + 1 ⎟⎠ ⎝

3. DF-PNC 1 [12, 15]

18

1 ⎛1 ⎞ log ⎜ + Λ ⎟ ⎝2 ⎠ 2

The results are plotted in Fig. 3. As can be noticed, DF-PNC 1 is the overall winner, because it can successfully suppress the noise at the relay (hence it is not affected by error propagation, unlike ANC). On the other hand, the necessity for DF-PNC 2 to decode both X and Y induces a big performance loss also with respect to DNC. As a final remark, despite these works, some important matters have not been dealt with yet. For instance, all these ideas work for frequency flat channels, and it is unknown how to modify them in a frequency selective channel. In addition, tight symbol synchronization is often assumed (i.e., the packets must be symbol- and often also phase-synchronized at the relay). An exception is [10], which was able to overcome some of these problems, although in some very simple, standard topologies where these issues are not as troublesome as in a generic, random network. Finally, almost all the aforementioned papers study information-theoretic based metrics, such as the achievable rate regions, while an actual MAC protocol that could work in a variety of settings has not yet been designed beyond the first steps of [10, 16]. In spite of these open issues, physical layer network coding has the inherent and paramount merit of truly merging Network Coding and the multiple access channel: the benefits of Network Coding for the multicast/broadcast channel have received immense attention by the NC community and this situation is well understood. However, it is less clear how NC impacts multiple access schemes, whereas physical layer network coding provides an original and new point of view on this issue, although many challenges are still ahead.

MIMO NETWORK CODING Network coding is inherently a MIMO scheme. It codes together multiple information units (the inputs) to yield multiple coded packets (the outputs), and all techniques to recover the information units (e.g., the Gaussian elimination procedure) are effectively vector detection algorithms. Recovering a vector of received information units from a vector of received samples is one of the key issues in MIMO, and some papers have investigated how ideas drawn from MIMO can be used to improve some aspects of NC [9, 18, 19]. For instance, MIMO is well known to provide diversity and power gains and hence it is robust to errors and noise. Given this property, MIMO may also be able to retrieve information even if some of the antennas are subject to strong fading. Such features are especially desirable in NC, as the loss of a coded packet may delay the whole decoding process. Since this area is still rather unexplored, it has been left as the last in this article.

IEEE Wireless Communications • February 2011

ROSSETTO LAYOUT

2/7/11

10:35 AM

Page 19

The assumption of having channel state information at the receiver (coherent detection) or not (noncoherent detection) constitutes one of the basic distinctions in MIMO signal processing. Network coding approaches based on coherent MIMO processing have found more direct application to cooperation so far; thus, we focus on them. In [9], usage of coherent MIMO signal processing for NC detection is investigated. The underlying principle is to use the channel state estimates and NC coefficients to perform joint demodulation and NC decoding based on the received analog signals. The scheme is made practical by using efficient MIMO algorithms such as Sphere Decoding, which leads to nearoptimal performance at affordable complexity. Such a method offers some relevant advantages over conventional NC, of which the most important is the ability to use corrupted or redundant frames (i.e., linearly dependent on the already received coded packets). This ability is definitely interesting in wireless NC, since the loss of even a single coded packet might lead to a rank deficient network coding matrix and thus preclude decoding. Instead, by leveraging the possibility to use also corrupted frames, such an event is indeed far less likely. In addition, the joint demodulation-decoding process always yields a BER no larger than conventional NC, hence improving the error correction capabilities. On the other hand, [9] remarked that conventional NC encoding cannot perform well in a wireless environment when network and channel coding are separated. For instance, if N nodes transmit one coded packet each out of the same pool of P information units, the diversity order at any receiver will be at most N – P +1, even with maximum likelihood detection. This is in stark contrast to MIMO, where the diversity order with N receive antennas, M transmit antennas, and M independent, spatially multiplexed transmitted streams with maximum likelihood detection is N. Such a problem has been implicitly recognized by [8], where it is shown that even with N = P = 2, a diversity order of 2 can be achieved, but only if channel and network coding are jointly designed. A straightforward application of this principle is Phoenix [19], so far one of the two hybrid cooperative/network coding protocols that have been tested in a full-fledged discrete event simulator or in a testbed, the other being the Analog Network Coding protocol of [10]. Phoenix is a Carrier Sense Multiple Access (CSMA) based protocol that reduces the bandwidth inefficiency of cooperative retransmissions. Let us consider Fig. 4, where node A has unsuccessfully transmitted a packet X to node D. In a cooperative protocol, one of A and D’s neighbors is elected as relay (called R), and has to perform a retransmission on behalf of A. In a standard cooperative protocol, R retransmits a copy of X; hence R helps A, but receives no direct reward. If R sent a coded packet, which was a linear combination X ⊕ Y of X and one of R’s packets, called Y, R would also deliver its own traffic, albeit at the price of a slightly higher packet error rate. However, standard NC would not be able to recover X and Y, because A’s coded packet (the

IEEE Wireless Communications • February 2011

y

R x/x⊕y x

S

x

D

Figure 4. Reference topology for NC/cooperative protocols [19].

first version of X) is corrupted and hence unusable by ordinary NC algorithms. Instead, if D employs MIMO_NC, D can jointly decode X and X ⊕ Y, and could potentially recover both X and Y. Phoenix encourages nodes to help each other, since they can perform a retransmission and pursue their own interest. Cooperative behavior is especially useful in multihop networking, because a route composed by multiple hops needs to pursue higher link reliability than single-hop communication in order to achieve satisfactory performance. An example is reported in Fig. 5 [19]. This picture reports the aggregate throughput for two-hop and four-hop routes delivered by an ad hoc network with 25 nodes and 1 Mb/s link data rate, where all terminals transmit data to a gateway, possibly through a multihop path; three protocols are compared: IEEE 802.11 (carrier sense multiple access, CSMA), a conventional decode-and-forward cooperative protocol (cooperative CSMA, CCSMA), and the hybrid NC/cooperative version (Phoenix). Phoenix gains as much as 18 percent over CCSMA for two-hop traffic and even more for four-hop traffic. In addition, the chance to send a new packet along with a retransmission is especially useful to relieve congestion in bottlenecks. For instance, if many flows converge to the same node (e.g., a gateway), any packet loss would delay the traffic of all the flows converging there. Total aggregate throughput can gain as much as 25 percent over CCSMA [19]. Phoenix is able to relieve this network-level problem by its cooperative/NC retransmissions. For completeness, we also briefly summarize the main results on Network Coding techniques inspired by non-coherent MIMO [18]. Reference [18] points out that if the NC coefficient matrix is unknown to the receiver, information is brought not by the information units themselves, but by the vector space they span. Such a property is strongly related to noncoherent MIMO detection, since in that case as well information is not brought by the transmitted symbols but by the subspace they span. This parallel enables to deeply understand the limits and possibilities of error correction techniques for NC and to design better codes. Many concepts of classic channel coding theory (such as the sphere packing, sphere covering, and singleton bounds) are extended to non-coherent NC, also thanks to this parallel with MIMO.

19

ROSSETTO LAYOUT

2/7/11

10:35 AM

Page 20

12 CSMA, 2 hops CSMA, 4 hops CCSMA, 2 hops CCSMA, 4 hops PHOENIX, 2 hops PHOENIX, 4 hops

Aggregate throughput [kb/s]

10

8

6

4

2

0

2

3

4

5

6

7 8 Load [pk/s]

9

10

11

12

Figure 5. Aggregate throughput for two- and four-hop traffic [19].

THE CHALLENGES AHEAD The quest for highly efficient and practical hybrid NC/cooperative protocols started about four years ago and has already energized a whole sector of the research community. In this relatively short period of time, three main approaches to this cross-layer area have emerged that have looked at this field with different perspectives, yet interactions among them have shed light on some deep and important issues on the topic. These are some of the lessons that have been learned together with the challenges they open: What will the role of PHY be in this area? All these roads suggest that interaction between NC and cooperation must have the physical layer (PHY) as a pivotal element. Moreover, the three PHYs adopted by the described approaches are radically different from each other, since each of them must be designed according to the different needs and philosophy behind every technique. Hence, the adoption of one of these systems implies a clear choice on the physical layer. No new PHY can be effective without a proper medium access control (MAC) or network layer, and MIMO networking is a recent and telling example. For the time being, MAC layer issues or opportunities brought by hybrid NC/ cooperative techniques have gone largely unaddressed. The only notable exceptions are Phoenix [19], where an 802.11-like protocol uses MIMO_NC [9] to “hide” packet losses, and ANC [10]. In addition, can this PHY be supported by conventional off-the-shelf radios, or should a brand new chip be designed? Can the NC architecture remain the same for a wireless cooperative protocol? •The encoding phase of NC must be changed in order to exploit the full gain of NC in the wireless environment [9]. Work in the channel/network coding area and MIMO_NC have shown that separation of network coding and channel coding implies non-negligible performance degradations.

20

•Combining these two techniques entails more complexity in the system, and the question of what is the trade-off between performance, redundancy, and complexity has only been partly investigated and is still largely uncharted territory. What have we learnt from physical layer NC? •Physical layer NC is a very interesting and potentially beneficial idea. One of the basic findings is that in spite of exploiting the “analog” nature of the waveforms, especially for the possibility of “combining” signals in the air, it is more beneficial to retain the use of digital algebra, because of its ability to suppress noise much better than amplify-and-forward protocols. Better performance can be achieved by decoding a function of the transmitted information units (DF-PNC 1) rather than decoding each single packet (DF-PNC 2) or not decoding anything and amplifying the aggregate received signal (AF-PNC). •Several practical issues must be answered before actual deployment in a generic random network is viable. So far, PNC requires tight symbol synchronization, the same modulation for colliding signals, and flat fading channels. In conclusion, cooperation and network coding have shown opposite features and problems, and their cutting-edge union may enhance their virtues while minimizing their problems. The potential for very high performance is unquestionable, as the preliminary but clear indications in information and digital communication theory have shown.

ACKNOWLEDGMENTS Francesco Rossetto is supported in part by the Space Agency of the German Aerospace Center and teh Federal Ministry of Economics and Technology based on the agreement of the German Bundestag with the support code 50YB0905 and by the European Commission within the Seventh Framework Programme (FP7) under theSAPHYRE Project (C.A. #2480011). The authors would like to thank Elena Fasolo, Andrea Munari, and Bobak Nazer for insightful discussions on these topics. They would also like to thank the Associate Editor and anonymous reviewers for their insightful comments.

REFERENCES [1] R. Ahlswede et al., “Network Information Flow,” IEEE Trans. Info. Theory, vol. 46, no. 4, July 2000, pp. 1204–16. [2] J. Laneman, Cooperation in Wireless Networks: Principles and Applications, Springer, 2006, “Cooperative Diversity: Models, Algorithms, and Architectures,” pp. 163–88. [3] S. Zhang, S. C. Liew, and P. P. Lam, “Physical-Layer Network Coding,” ACM MOBICOM, Los Angeles, CA, Sept. 2006. [4] P. Larsson and N. Johansson, “Multi User ARQ,” VTC Spring, Melbourne, Australia, May 2006. [5] C. Hausl and P. Dupraz, “Joint Network-Channel Coding for the Multiple Access Relay Channel,” IEEE IWWAN, New York, NY, June 2006. [6] X. Bao and J. Li, “Adaptive Network Coded Cooperation (ANCC) for Wireless Relay Networks: Matching Codeon-Graph with Network-on-Graph,” IEEE Trans. Wireless Commun., vol. 7, no. 2, Feb. 2008, pp. 574–83. [7] S. Yang and R. Koetter, “Network Coding over a Noisy Relay: a Belief Propagation Approach,” IEEE ISIT, Nice, France, 9 Jan. 2007. [8] L. Xiao et al., “A Network Coding Approach to Cooperative Diversity,” IEEE Trans. Info. Theory, vol. 53, no. 10, Oct. 2007, pp. 3714–22. [9] E. Fasolo, F. Rossetto, and M. Zorzi, “Network Coding meets MIMO,” NetCod 2008, Hong Kong, China, Jan. 2008.

IEEE Wireless Communications • February 2011

ROSSETTO LAYOUT

2/7/11

10:35 AM

Page 21

[10] S. Katti, S. Gollakota, and D. Katabi, “Embracing Wireless Interference: Analog Network Coding,” ACM SIGCOMM 2007, Kyoto, Japan, 27–31 Aug. 2007. [11] S. Katti et al., “Joint Relaying and Network Coding in Wireless Networks,” IEEE ISIT 2007, Nice, France, 24–28 June 2007. [12] B. Nazer and M. Gastpar, “Computation over Multiple Access Channels,” IEEE Trans. Info Theory, vol. 53, no. 10, Oct. 2007, pp. 3498–516. [13] B. Rankov and A. Wittneben, “Spectral Efficient Protocols for Half-Duplex Fading Relay Channels,” IEEE JSAC, vol. 25, no. 2, Feb. 2007, pp. 379–89. [14] P. Popovski and H. Yomo, “Bi-directional Amplification of Throughput in a Wireless Multi-Hop Network,” IEEE VTC Spring, Melbourne (Australia), May 2006. [15] –, “The Anti-Packets Can Increase the Achievable Throughput of a Wireless Multi-Hop Network,” IEEE ICC, Istanbul, Turkey, June 2006. [16] Z. Ding et al., “On the Study of Network Coding with Diversity,” IEEE Trans. Wireless Commun., vol. 8, no. 3, Mar. 2009, pp. 1247–59. [17] Y. Hao et al., “Achievable Rates for Network Coding on the Exchange Channel,” IEEE MILCOM, Orlando, FL, Oct. 2007. [18] R. Koetter and F. R. Kschischang, “Coding for Errors and Erasures in Random Network Coding,” IEEE Trans. Infor. Theory, vol. 54, no. 8, Aug. 2008, pp. 3579–91. [19] E. Fasolo et al., “Phoenix: A Hybrid Cooperative-Network Coding Protocol for Fast Failure Recovery in Ad Hoc Networks,” IEEE SECON 2008, San Francisco, CA, 16–20 June 2008.

BIOGRAPHIES FRANCESCO ROSSETTO [S’06, M’09] ([email protected]) received his Laurea (equivalent to M.S.) and Ph.D. degrees in telecommunications engineering in 2005 and 2009, respectively, from the University of Padova, Italy. In 2004–2005 he studied electrical engineering at the University of California, San Diego (UCSD) under a student

IEEE Wireless Communications • February 2011

exchange program. In 2008 he was on leave at UCSD, working for the MURI project, a multiuniversity initiative for the development of multihop MIMO networks. Since 2009 he has been with the DLR (German Aerospace Center), Munich. His research interests include satellite communication, network coding, and cross-layer design. His corporate experience includes a summer internship in 2006 at Ericsson Eurolabs, Aachen, Germany, working on HARQ for 3G/LTE cellular networks. MICHELE ZORZI [F’07] ([email protected]) received his Laurea and Ph.D. degrees in electrical engineering from the University of Padova in 1990 and 1994, respectively. During academic year 1992–1993 he was on leave at UCSD, attending graduate courses and doing research on multiple access in mobile radio networks. In 1993 he joined the faculty of the Dipartimento di Elettronica e Informazione, Politecnico di Milano, Italy. After spending three years with the Center for Wireless Communications at UCSD, in 1998 he joined the School of Engineering of the University of Ferrara, Italy, where he became a professor in 2000. Since November 2003 he has been on the faculty of the Information Engineering Department at the University of Padova. His present research interests include performance evaluation in mobile communications systems, random access in mobile radio networks, ad hoc and sensor networks, energy constrained communications protocols, and underwater communications and networking. He was Editor-In-Chief of IEEE Wireless Communications from 2003 to 2005, is currently Editor-In-Chief of IEEE Transactions on Communications, and serves on the Editorial Board of the Wiley Journal of Wireless Communications and Mobile Computing. He was also guest editor for special issues in IEEE Personal Communications (“Energy Management in Personal Communications Systems”) and IEEE Journal on Selected Areas in Communications (“Multimedia Network Radios” and “Underwater Wireless Communications and Networking”). He is a Member-at-Large of the Board of Governors of the IEEE Communications Society.

21

DOMINGO LAYOUT

2/7/11

10:26 AM

Page 22

ACCEPTED FROM OPEN CALL

SECURING UNDERWATER WIRELESS COMMUNICATION NETWORKS MARI CARMEN DOMINGO, BARCELONA TECH UNIVERSITY

ABSTRACT Underwater wireless communication networks are particularly vulnerable to malicious attacks due to the high bit error rates, large and variable propagation delays, and low bandwidth of acoustic channels. The unique characteristics of the underwater acoustic communication channel, and the differences between underwater sensor networks and their ground-based counterparts require the development of efficient and reliable security mechanisms. In this article, a complete survey of security for UWCNs is presented, and the research challenges for secure communication in this environment are outlined.

INTRODUCTION The authors present a complete survey of security for Underwater Wireless Communication Networks, and they outline the research challenges for secure communication in this environment.

22

Underwater wireless communication networks (UWCNs) are constituted by sensors and autonomous underwater vehicles (AUVs) that interact to perform specific applications such as underwater monitoring (Fig. 1) [1]. Coordination and sharing of information between sensors and AUVs make the provision of security challenging. The aquatic environment is particularly vulnerable to malicious attacks due to the high bit error rates, large and variable propagation delays, and low bandwidth of acoustic channels. Achieving reliable intervehicle and sensor-AUV communication is especially difficult due to the mobility of AUVs and the movement of sensors with water currents. The unique characteristics of the underwater acoustic channel, and the differences between underwater sensor networks and their groundbased counterparts require the development of efficient and reliable security mechanisms. This article discusses security in UWCNs. It is structured as follows. The following section explains the specific characteristics of UWCNs in comparison with their ground-based counterparts. Next, the possible attacks and countermeasures are introduced. Subsequently, security requirements for UWCNs are described. Later, the research challenges related to secure time synchronization, localization, and routing are summarized. Finally, the article is concluded.

1536-1284/11/$25.00 © 2011 IEEE

CHARACTERISTICS AND VULNERABILITIES OF UWCNS Underwater sensor networks have some similarities with their ground-based counterparts such as their structure, function, computation and energy limitations. However, they also have differences, which can be summarized as follows. Radio waves do not propagate well underwater due to the high energy absorption of water. Therefore, underwater communications are based on acoustic links characterized by large propagation delays. The propagation speed of acoustic signals in water (typically 1500 m/s) is five orders of magnitude lower than the radio wave propagation speed in free space. Acoustic channels have low bandwidth. The link quality in underwater communication is severely affected by multipath, fading, and the refractive properties of the sound channel. As a result, the bit error rates of acoustic links are often high, and losses of connectivity arise [1]. Underwater sensors move with water currents, and AUVs are mobile. Although certain nodes in underwater applications are anchored to the bottom of the ocean, other applications require sensors to be suspended at certain depths or to move freely in the underwater medium. The future development of geographical routing is very promising in UWCNs due to its scalability and limited signaling properties. However, it cannot rely on the Global Positioning System (GPS) because it uses radar waves in the 1.5 GHz band that do not propagate in water. Since underwater hardware is more expensive, underwater sensors are sparsely deployed. Underwater communication systems have more stringent power requirements than terrestrial systems because acoustic communications are more power-hungry, and typical transmission distances in UWCNs are greater; hence, higher transmit power is required to ensure coverage [1]. The above mentioned characteristics of UWCNs have several security implications. UWCNs suffer from the following vulnerabilities. High bit error rates cause packet errors. Consequently, critical security packets can be lost. Wireless underwater channels can be eavesdropped on. Attackers may intercept the infor-

IEEE Wireless Communications • February 2011

DOMINGO LAYOUT

2/7/11

10:26 AM

Page 23

mation transmitted and attempt to modify or drop packets. Malicious nodes can create out-ofband connections via fast radio (above the water surface) and wired links, which are referred to as wormholes. Since sensors are mobile, their relative distances vary with time. The dynamic topology of the underwater sensor network not only facilitates the creation of wormholes but it also complicates their detection [2]. Since power consumption in underwater communications is higher than in terrestrial radio communications, and underwater sensors are sparsely deployed, energy exhaustion attacks to drain the batteries of nodes pose a serious threat for the network lifetime.

ATTACKS ON UWCNS AND COUNTERMEASURES Both intervehicle and sensor-AUV communications can be affected by denial-of-service (DoS) attacks. Next, we summarize typical DoS attacks, evaluate their dangers, and indicate possible defenses to muffle their effects.

Sink AUV4 Event 1

AUV

3

Event 2

Event 3 AUV1 AUV2

Figure 1. Underwater sensor network with AUVs.

Sink

JAMMING A jamming attack consists of interfering with the physical channel by putting up carriers on the frequencies neighbor nodes use to communicate. Since underwater acoustic frequency bands are narrow (from a few to hundreds of kilohertz), UWCNs are vulnerable to narrowband jamming. Localization is affected by the replay attack when the attacker jams the communication between a sender and a receiver, and later replays the same message with stale information (an incorrect reference) posing as the sender (Fig. 2). Since jamming is a common attack in wireless networks, some of the solutions proposed for traditional wireless networks can be applied. Spread spectrum is the most common defense against jamming [3]. Frequency hopping spread spectrum (FHSS) and direct sequence spread spectrum (DSSS) in underwater communications are drawing attention for their good performance under noise and multipath interference. These schemes are resistant to interference from attackers, although not infallible. An attacker can jam a wide band of the spectrum or follow the precise hopping sequence when an FHSS scheme is used. A high-power wideband jamming signal can be used to attack a DSSS scheme. Underwater sensors under a jamming attack should try to preserve their power. When jamming is continuous, sensors can switch to sleep mode and wake up periodically to check if the attack is over. When jamming is intermittent, sensors can buffer data packets and only send high-power high-priority messages to report the attack when a gap in jamming occurs. In ground-based sensor networks, other sensors located along the edge of the area under attack can detect the jamming signal as higherthan-normal background noise and report intrusion to outside nodes. That will cause any further traffic to be rerouted around the jammed region [3]. However, this solution cannot be applied to

IEEE Wireless Communications • February 2011

Malicious node Underwater sensor node

Underwater sensor node

Message with stale information

Figure 2. Replay attack. UWCNs, since nodes underwater are usually sparsely deployed, which means there would not be enough sensors to delimit the jammed region accurately and reroute traffic around it. Another solution proposed for ground-based sensor networks against jamming is to use alternative technologies for communication such as infrared or optical [3]. However, this solution cannot be applied either, since optical and infrared waves are severely attenuated under water.

WORMHOLE ATTACK A wormhole is an out-of-band connection created by the adversary between two physical locations in a network with lower delay and higher bandwidth than ordinary connections. This connection uses fast radio (above the sea surface) or wired links (Fig. 3) to significantly decrease the propagation delay. In a wormhole attack the malicious node transfers some selected packets received at one end of the wormhole to the other end using the out-of-band connection, and re-injects them into the network [4]. The effect

23

DOMINGO LAYOUT

2/7/11

10:26 AM

Page 24

Sink

Wormhole link

local network topology (virtual layout) within two hops using multidimensional scaling (MDS). Since a wormhole contracts the virtual layout at certain regions, some nodes far away appear to be neighbors, and these contradictions can be detected visualizing the virtual layout. A wormhole indicator variable is defined to compute the distortion in angles; the distortion in edge lengths is computed as the difference between the measured distances among neighboring sensors and the lengths of the reconstructed connections. In [5] a suite of protocols is proposed to enable wormhole-resilient secure neighbor discovery with high probability in underwater sensor networks. This solution is based on the direction of arrival (DoA) estimation of acoustic signals, which depends on the relative locations of signal transmitters and receivers, and cannot be manipulated.

SINKHOLE ATTACK In a sinkhole attack, a malicious node attempts to attract traffic from a particular area toward it; for example, the malicious node can announce a high-quality route. Geographic routing and authentication of nodes exchanging routing information are possible defenses against this attack, but geographic routing is still an open research topic in UWCNs.

Distributed underwater sensor nodes

HELLO FLOOD ATTACK Figure 3. Underwater network with a wormhole link. is that false neighbor relationships are created, because two nodes out of each other’s range can erroneously conclude that they are in proximity of one another due to the wormhole’s presence. This attack is devastating. Routing protocols choose routes that contain wormhole links because they appear to be shorter; thus, the adversary can monitor network traffic and delay or drop packets sent through the wormhole. Localization protocols can also be affected by these attacks when malicious nodes claim wrong locations and mislead other nodes. One proposed method for wormhole detection in ground-based sensor networks consists of estimating the real physical distance between two nodes to check their neighbor relationship [4]. If the measured distance is longer than the nodes’ communication range, it is assumed that the nodes are connected through a wormhole. However, accurate distance estimation depends on precise localization (geographical packet leashes, wormhole detection using position information of anchors), tight clock synchronization (temporal packet leashes), or use of specific hardware (directional antennas) [4]. In underwater communications accurate localization and time synchronization are still challenging. The authors in [2] propose a distributed mechanism named Distributed Visualization of Wormhole (Dis-VoW) to detect wormhole attacks in three-dimensional underwater sensor networks. In Dis-VoW, every sensor collects the distance estimations to its neighbors using the round-trip time of acoustic signals; after these distances are broadcast by each sensor to its neighbors, every node is able to construct the

24

A node receiving a HELLO packet from a malicious node may interpret that the adversary is a neighbor; this assumption is false if the adversary uses high power for transmission. Bidirectional link verification can help protect against this attack, although it is not accurate due to node mobility and the high propagation delays of UWCNs. Authentication is also a possible defense.

ACKNOWLEDGMENT SPOOFING A malicious node overhearing packets sent to neighbor nodes can use this information to spoof link layer acknowledgments with the objective of reinforcing a weak link or a link located in a shadow zone. Shadow zones are formed when the acoustic rays are bent and sound waves cannot penetrate. They cause high bit error rates and loss of connectivity [1]. This way, the routing scheme is manipulated. A solution to this attack would be encryption of all packets sent through the network.

SELECTIVE FORWARDING Malicious nodes drop certain messages instead of forwarding them to hinder routing. In UWCNs it should be verified that a receiver is not getting the information due to this attack and not because it is located in a shadow zone. Multipath routing and authentication can be used to counter this attack, but multipath routing increases communication overhead.

SYBIL ATTACK An attacker with multiple identities can pretend to be in many places at once. Geographic routing protocols are also misled because an adversary with multiple identities can claim to be in multiple places at once (Fig. 4).

IEEE Wireless Communications • February 2011

DOMINGO LAYOUT

2/7/11

10:26 AM

Page 25

Authentication and position verification are methods against this attack, although position verification in UWCNs is problematic due to mobility.

Sink Malicious node C

SECURITY REQUIREMENTS

Nodes A and B want to send their data towards the sink.

In UWCNs the following security requirements should be considered.

AUTHENTICATION Authentication is the proof that the data was sent by a legitimate sender. It is essential in military and safety-critical applications of UWCNs. Authentication and key establishment are strongly related because once two or more entities verify each other’s authenticity, they can establish one or more secret keys over the open acoustic channel to exchange information securely; conversely, an already established key can be used to perform authentication. Traditional solutions for key generation and update (renewal) algorithms should be adapted to better address the characteristics of the underwater channel. In [6], a key generation system is proposed that requires only a threshold detector, lightweight computation, and communication costs. It exploits reciprocity, deep fades (strong destructive interference), randomness extractor, and robust secure fuzzy information reconciliators. This way, the key is generated using the characteristics of the underwater channel and is secure against adversaries who know the number of deep fades but not their locations.

CONFIDENTIALITY

B A a) Sink Malicious node C

B A

b) Sink

Confidentiality means that information is not accessible to unauthorized third parties. Therefore, confidentiality in critical applications such as maritime surveillance (Fig. 5) should be guaranteed.

Malicious node C

INTEGRITY B

It ensures that information has not been altered by any adversary. Many underwater sensor applications for environmental preservation, such as water quality monitoring [7], rely on the integrity of information.

A

RESEARCH CHALLENGES The security issues and open challenges for secure time synchronization, localization, and routing in UWCNs are summarized in the following sections.

SECURE TIME SYNCHRONIZATION Time synchronization is essential in many underwater applications such as coordinated sensing tasks. Also, scheduling algorithms such as timedivision multiple access (TDMA) require precise timing between nodes to adjust their sleep-wakeup schedules for power saving. For example, in

IEEE Wireless Communications • February 2011

A and B select the invented non-existent positions of malicious node C to forward their messages. Node C overhears them.

c)

AVAILABILITY The data should be available when needed by an authorized user. Lack of availability due to denial-of-service attacks would especially affect time-critical aquatic exploration applications such as prediction of seaquakes.

Malicious node C broadcasts advertisement messages of invented nonexistent positions of nodes (yellow nodes)

Figure 4. Sybil attack.

water quality monitoring [7], sensors are deployed at different depths because the chemical characteristics of water vary at each level. The design of a delay-tolerant time synchronization mechanism is very important to accurately locate the water contaminant source, set up the sleep-wakeup schedules among neighboring nodes appropriately, and log the water quality data correctly into the annual database with accurate timing information. Achieving precise time synchronization is especially difficult in underwater environments due to the characteristics of UWCNs. For this reason, the time synchronization mechanisms proposed for ground-based sensor networks cannot be applied, and new mechanisms have been

25

DOMINGO LAYOUT

2/7/11

Distributed underwater sensor nodes

10:26 AM

Page 26

Intruder submarine

Surface sink

Data path

Data transmitted to the on-shore command center

Figure 5. Intruder submarine detection.

proposed. Tri-Message [8] is a time synchronization protocol designed for high-latency networks with a synchronization precision that increases with distance. A multilateration algorithm is proposed in [9] for localization and synchronization in 3D underwater acoustic sensor networks. It is assumed that a set of anchors, several buoys on the ocean surface, already know their locations and time without error. A group of nearby sensors receives synchronization packets containing the coordinates and packet transmit times from at least five anchor nodes and performs multilateration to obtain their own locations. The sensors learn the time difference between themselves and each anchor node by comparing their local times at which they received the time synchronization packet with the transmit time plus propagation delays; these nodes subsequently become new anchor nodes and there-

26

after broadcast new synchronization packets to a larger range, and so on. MU-Sync [10] is a cluster-based synchronization protocol that estimates the clock skew by performing the linear regression twice over a set of local time information gathered through message exchanges. The first linear regression enables the cluster head to offset the effect of long and varying propagation delay; the second regression enables the clusterhead to obtain the final estimated skew and offset. None of the aforementioned time synchronization schemes [8–10] consider security, although it is critical in the underwater environment. Time synchronization disruption due to masquerade, replay and message manipulation attacks, can be addressed using cryptographic techniques [11]. However, countering other possible attacks such as delays (deliberate delaying the transmission of time synchronization messages) [11] and DoS attacks requires the use of other strategies. The countermeasures against delay attacks proposed in [11] for ground-based sensor networks are not applicable to UWCNs. They are based on the detection of outliers (malicious time offsets), but they do not distinguish between deliberate alterations and abnormal values resulting from long and variable propagation delays or node mobility. Moreover, they do not support global synchronization in multi-hop sensor networks. A correlation-based security model for water quality monitoring systems has been proposed in [7] to detect outlier timestamps due to insider attacks. The authors prove that the acoustic propagation delays between two sensors in neighboring depth levels fit an approximately normal distribution, which means that the timestamps between them should correlate. However, this correlation is lost if a captured inside node is sending falsified timestamps. With proper design of a timestamp sliding window scheme, insider attacks are detected. Each sensor should obtain timestamp readings from multiple sensors and calculate the correlation coefficient for each neighbor’s timestamp, obtaining a window of coefficients. If a coefficient of the window of data is below a threshold, it is an outlier value. If the abnormal percentage of data in one window (outlier percentage) is consistently (10 consecutive windows) higher than a predetermined threshold, the corresponding neighbor is flagged as a malicious node generating insider attacks. However, identifying a neighbor node as malicious is difficult, because sometimes timestamps can be corrupted due to propagation delay variations caused by the channel rather than deliberately. Because of wave motion, the signal multipath components undergo time-varying propagation delays. Node mobility due to water currents also modifies the propagation delays. In order to better distinguish between unintended and malicious timestamp alterations, the authors in [12] improve the proposed scheme by using as a second step a statistical reputation and trust model to detect outlier timestamps, and identify nodes generating insider attacks. It is based on quantitative measurements and on the assumption that identifying an insider attacker requires long-term behavior observations.

IEEE Wireless Communications • February 2011

DOMINGO LAYOUT

2/7/11

10:26 AM

Page 27

The following open research issues for secure time synchronization need to be addressed: • Because of the high and variable propagation delays of UWCNs, the time required to synchronize nodes should be investigated. • Efficient and secure time synchronization schemes with small computation and communications costs need to be designed to defend against delay and wormhole attacks.

SECURE LOCALIZATION Localization is a very important issue for data tagging. Sensor tasks such as reporting the occurrence of an event or monitoring require localization information. Localization can also help in making routing decisions. For example, the underwater sensors in [13] learn the location and speed of mobile beacons and neighbors during the localization phase; the position and motion of mobile beacons are used by the routing protocol to choose the best relay for a node to forward its data. Localization approaches proposed for ground-based sensor networks do not work well underwater because long propagation delays, Doppler effect, multipath, and fading cause variations in the acoustic channel. Bandwidth limitations, node mobility, and sparse deployment of underwater nodes also affect localization estimation. Proposed terrestrial localization schemes based on received signal strength (RSS) are not recommended in UWCNs, since non-uniform acoustic signal propagation causes significant variations in the RSS. Time of arrival (ToA) and time difference of arrival (TDoA) measurements require very accurate time synchronization (which is a challenging issue), and angle of arrival (AoA) algorithms are affected by the Doppler shift. Localization schemes can be classified into: Range-based schemes (using range and/or bearing information): The location of nodes in the network is estimated through precise distance or angle measurements. •Anchor-based schemes: Anchor nodes are deployed at the seabed or sea surface at locations determined by GPS. The propagation delay of sound signals between the sensor [9] or AUV and the anchors is used to compute the distance to multiple anchor nodes. •Distributed positioning schemes: Positioning infrastructure is not available, and nodes communicate only with one-hop neighbors and compute their locations using multilateration. Underwater sensor positioning (USP) has been proposed in [14] as a distributed localization scheme for sparse 3D networks, transforming the 3D underwater positioning problem into a 2D problem using a distributed non-degenerative projection technique. Using sensor depth information, the neighboring reference nodes are mapped to the horizontal plane containing the sensor to be localized. After projecting the reference nodes, localization methods for 2D networks such as bilateration or trilateration can be used to locate the sensor. •Schemes that use mobile beacons/anchors: They use mobile beacons whose locations are always known. Scalable localization with mobility prediction (SLMP) has been proposed in [15]

IEEE Wireless Communications • February 2011

as a hierarchical localization scheme. At the beginning, only surface nodes know their locations, and anchor nodes can be localized by these surface buoys. Anchor nodes are selected as reference nodes because of their known locations; with the advance of the location process more ordinary nodes are localized and become reference nodes. During this process, every node predicts its future mobility pattern according to its past known location information. The future location is estimated based on this prediction. Range-free schemes (not using range or bearing information): They have been designed as simple schemes to compute only coarse position estimates. A range-free scheme proposed in [16] estimates the location of a sensor within a certain area. None of the aforementioned localization schemes [9, 13–16] was designed with security in mind. Some localization-specific attacks (replay attack, Sybil attack, wormhole attack) have previously been described. Open research issues for secure localization are: • Effective cryptographic primitives against injecting false localization information in UWCNs need to be developed. • It is necessary to design resilient algorithms able to determine the location of sensors even in the presence of Sybil and wormhole attacks. • Techniques to identify malicious or compromised anchor nodes and to avoid false detection of these nodes are required. • Secure localization mechanisms able to handle node mobility in UWCNs need to be devised.

Routing is specially challenging in UWCNs due to the large propagation delays, low bandwidth, difficulty of battery refills of underwater sensors, and dynamic topologies. Therefore, routing protocols should be designed to be energy-aware, robust, scalable and adaptive.

SECURE ROUTING Routing is essential for packet delivery in UWCNs. For example, the Distributed Underwater Clustering Scheme (DUCS) [17] does not use flooding and minimizes the proactive routing message exchange. Routing is specially challenging in UWCNs due to the large propagation delays, the low bandwidth, the difficulty of battery refills of underwater sensors, and the dynamic topologies. Therefore, routing protocols should be designed to be energy-aware, robust, scalable and adaptive. Many routing protocols have been proposed for underwater wireless sensor networks. However, none of them has been designed with security as a goal. Routing attacks can disable the entire network’s operation. Spoofing, altering, or replaying routing information affects routing. Important routing attacks (selective forwarding, sinkhole attack, Sybil attack, wormhole attack, HELLO flood attack, acknowledgment spoofing) have been previously described. Although the attacks against routing in UWCNs are the same as in ground-based sensor networks, the same countermeasures are not directly applicable to UWCNs due to their difference in characteristics. Proposed broadcast authentication methods would cause high communication overhead and latency in UWCNs. Multipath routing would cause high communication overhead as well.

27

DOMINGO LAYOUT

2/7/11

10:26 AM

The proper functioning of these schemes is challenging because they do not work well in mobile environments, the time required to detect compromised nodes increases substantially in UWCNs due to the long propagation delays, and they must be adapted to tolerate short-term disruptions.

Page 28

Open research issues for secure routing are: •There is a need to develop reputation-based schemes that analyze the behavior of neighbors and reject routing paths containing selfish nodes that do not cooperate in routing. The proper functioning of these schemes is challenging because they do not work well in mobile environments, the time required to detect compromised nodes increases substantially in UWCNs due to the long propagation delays, and they must be adapted to tolerate short-term disruptions. •Quick and powerful encryption and authentication mechanisms against outside intruders should be devised for UWCNs because the time required for intruder detection is high due to the long and variable propagation delays, and routing paths containing undetected malicious nodes can be selected in the meantime for packet forwarding. •Sophisticated mechanisms should be developed against insider attacks such as selective forwarding, Sybil attacks, HELLO flood attacks, and acknowledgment spoofing. •There is a need to develop new techniques against sinkholes and wormholes, and improve existing ones. With Dis-VoW [2] a wormhole attack can still be concealed by manipulating the buffering times of distance estimation packets. The wormhole-resilient neighbor discovery in [5] is affected by the orientation error between sensors.

CONCLUSIONS In this article we have discussed security in UWCNs, underlining the specific characteristics of these networks, possible attacks, and countermeasures. The main research challenges related to secure time synchronization, localization, and routing have also been surveyed. These research issues remain wide open for future investigation.

ACKNOWLEDGMENT This work was supported by the Spanish Ministry of Education and Science under project TSI2007-66637-C02-01.

REFERENCES [1] I. F. Akyildiz, D. Pompili, and T. Melodia, “Underwater Acoustic Sensor Networks: Research Challenges,” Ad Hoc Net., vol. 3, no. 3, Mar. 2005.

28

[2] W. Wang et al., “Visualization of Wormholes in Underwater Sensor Networks: A Distributed Approach,” Int’l. J. Security Net., vol. 3, no. 1, 2008, pp. 10–23. [3] A. D. Wood and J. A. Stankovic, “A Taxonomy for Denial-of-Service Attacks in Wireless Sensor Networks,” chapter in Handbook of Sensor Networks: Compact Wireless and Wired Sensing Systems, M. Ilyas and I. Mahgoub, Eds., CRC Press, 2004. [4] L. Buttyán and J.-P. Hubaux, Security and Cooperation in Wireless Networks: Thwarting Malicious and Selfish Behaviour in the Age of Ubiquitous Computing, Cambridge Univ. Press, 2008. [5] R. Zhang and Y. Zhang, “Wormhole-Resilient Secure Neighbor Discovery in Underwater Acoustic Networks,” Proc. IEEE INFOCOM, 2010. [6] Y. Liu, J. Jing, and J. Yang, “Secure Underwater Acoustic Communication Based on a Robust Key Generation Scheme,” Proc. ICSP, 2008. [7] F. Hu, S. Wilson, and Y. Xiao, “Correlation-Based Security in Time Synchronization of Sensor Networks,” Proc. IEEE WCNC, 2008. [8] C. Tian et al., “Tri-Message: A Lightweight Time Synchronization Protocol for High Latency and ResourceConstrained Networks,” Proc. IEEE ICC, 2009. [9] C. Tian et al., “Localization and Synchronization for 3D Underwater Acoustic Sensor Networks,” in Ubiquitous Intelligence and Computing, LNCS, Springer, 2007, pp. 622–31. [10] N. Chirdchoo, W.-S. Soh, and K. Chua, “MU-Sync: A Time Synchronization Protocol for Underwater Mobile Networks,” Proc. WUWNet, 2008. [11] H. Song, S. Zhu, and G. Cao, “Attack-Resilient Time Synchronization for Wireless Sensor Networks,” Ad Hoc Net., vol. 5, no. 1, 2007, pp. 112–25. [12] F. Hu et al., “Vertical and Horizontal Synchronization Services with Outlier Detection in Underwater Acoustic Networks,” Wireless Commun. Mobile Comp., vol. 8, no. 9, 2008, pp. 1165–81. [13] M. Erol and S. Oktug, “A Localization and Routing Framework for Mobile Underwater Sensor Networks,” Proc. IEEE INFOCOM, Apr. 2008. [14] W. Cheng et al., “Underwater Localization in Sparse 3D Acoustic Sensor Networks,” Proc. IEEE INFOCOM, 2008. [15] Z. Zhou, J.-H. Cui, and A. Bagtzoglou, “Scalable Localization with Mobility Prediction for Underwater Sensor Networks,” Proc. IEEE INFOCOM, 2008. [16] Y. Zhou et al., “A Range-free Localization Scheme for Large Scale Underwater Wireless Sensor Networks,” J Shanghai Jiaotong Univ. (Science), vol. 14, no. 5, 2009, pp. 562–68. [17] M. C. Domingo and R. Prior, “Design and Analysis of a GPS-Free Routing Protocol for Underwater Wireless Sensor Networks in Deep Water,” Proc. UNWAT, 2007.

BIOGRAPHY M ARI C ARMEN D OMINGO ([email protected]) received her Lic. and Ph.D. degrees in electrical engineering from Barcelona Tech University, Spain, in 1999 and 2005, respectively. She currently works as an assistant professor in the Electrical Engineering Department of the same university. Her current research interests are in the area of network security, sensor, and wireless networks.

IEEE Wireless Communications • February 2011

MOESSNER LAYOUT

2/7/11

10:42 AM

Page 30

ACCEPTED FROM OPEN CALL

SPECTRUM SENSING FOR COGNITIVE RADIO SYSTEMS: TECHNICAL ASPECTS AND STANDARDIZATION ACTIVITIES OF THE IEEE P1900.6 WORKING GROUP KLAUS MOESSNER, UNIVERSITY OF SURREY HIROSHI HARADA, CHEN SUN, YOHANNES D. ALEMSEGED, AND HA NGUYEN TRAN, NATIONAL INSTITUTE OF INFORMATION AND COMMUNICATIONS TECHNOLOGY DOMINIQUE NOGUET, CEA-LETI RYO SAWAI AND NAOTAKA SATO, SONY CORPORATION

Cognitive engine/ data archive

Sensor A

The authors present the technical issues and IEEE standardization activities related to sensing information exchange. In particular, IEEE P1900.6 working group activities are discussed.

30

ABSTRACT The evolution of future wireless communication systems imposes a strong requirement on the efficiency of spectrum usage, which is expected to be leveraged by interacting and cooperating cognitive radios forming wider cognitive radio systems. Dynamic spectrum access is a potential means to improve spectrum usage. A key step in realizing DSA is to obtain spectral occupancy information provided by spectrum sensors. Subsequently, nodes in a CRS analyze spectrum usage to find unused spectrum (often referred to as white spaces). Then the system makes a decision on the best opportunities considering regulatory policy, transceiver capacity, and so on. In such an operation, sensing information exchange plays a fundamental and key role in enabling efficient DSA. This article presents the technical issues and IEEE standardization activities related to sensing information exchange. In particular, the IEEE P1900.6 working group activities aimed at standardizing logical interfaces and data structures required for exchange of sensing related information between sensors and their clients are discussed. By explaining the objective, use cases, reference model, data structure, data representation, and generic procedures developed so far, the article presents the main technical aspects of the IEEE P1900.6 sensing interface and its usefulness in CRSs.

INTRODUCTION The need for advanced radio systems that can accommodate more users, provide higher throughput, and support higher mobility is growing. This has led to an increased use of spectrum resources. However, radio spectrum is a limited

1536-1284/11/$25.00 © 2011 IEEE

resource, and the spectrum for each service has been allocated mostly through fixed spectrum assignment. Hence, spectrum scarcity is imminent. On the other hand, recent studies reveal that some frequencies allocated for some radio access technologies (RATs) are underutilized [1]. Dynamic spectrum access (DSA) promises to provide an efficient way to utilize spectrum resources through enabling technologies such as cognitive radio (CR). To achieve DSA, a cognitive radio system (CRS) must acquire radio environment knowledge. This will be done in many cases through spectrum sensing. Based on the sensing information, the system analyzes the spectrum usage and makes decisions on spectrum access [2, 3]. Apparently, the sensing function realized by spectrum sensors plays a fundamental role in the DSA process of a CRS. It is most important that the sensing information provided by spectrum sensors can be conveyed to other entities or units in the CRS, which in IEEE P1900.6 is defined as a client of the sensors. Recently proposed advanced radio systems based on sensing technology combine sensing and cognitive engines (CEs) that use the sensing results in proprietary transceiver and CR architectures (e.g., those being worked on within IEEE P802.22 [4]). This model of proprietary development usually reduces innovation and, in the past in cases like closed operating systems, has limited the opportunities for integrating new component technologies for better system performance. In the case of CR, proprietary interfaces also limit the degree and capability at which different types of sensors and clients can interoperate. In order to make the development of spectrum sensors independent of the evolution and development of advanced wireless communication systems, and to ensure the

IEEE Wireless Communications • February 2011

MOESSNER LAYOUT

2/7/11

10:42 AM

Page 31

compatibly and coexistence of different sensors from various manufacturers that implement a variety of sensing techniques, a standard is needed which defines the interface between spectrum sensors and their clients (i.e., the functional entities that exploit sensing information obtained from the sensors). As the IEEE DYSPAN Standards Committee (formerly Standards Coordinating Committee 41 — SCC41), one of the active working groups (WGs) within the IEEE P1900.6 WG was established in 2008 to address the above need to standardize the information exchange between spectrum sensors and their clients in radio communication systems [5]. The logical interface and supporting data structures used for information exchange are defined in an abstract manner without constraining the sensing technology, client design, or data link between sensor and client. By defining the spectrum sensing interfaces and data structures for DSA and other advanced radio communications systems, the standard will facilitate interoperability between independently developed devices and thus allow for separate evolution of spectrum sensors and other system functions. This article introduces the current status of IEEE P1900.6 standardization activities based on published contributions and the P1900.6 draft standard, which is subject to changes along the course of development. The next section presents the technical aspects of the IEEE P1900.6 standard in terms of scope and purpose, use cases, reference model, data structure and representation, and generic procedures, respectively. The following section gives the future direction of the WG. The final section concludes the article.

THE P1900.6 APPROACH This section provides the technical scope, use cases, and requirements considered in the WG.

SCOPE OF IEEE P1900.6 STANDARD The scope of the standard, as stated in the project authorization request (PAR), is to define the information exchange between spectrum sensors and their clients in radio communication systems [6]. The logical interface and supporting data structures used for information exchange are defined abstractly without constraining the sensing technology, client design, or data link between sensors and their clients. According to the scope of the standard, the WG defines the information exchange between spectrum sensors and their clients in radio communication systems. Figure 1 shows a scenario where sensing information is exchanged among sensors and their clients. The clients include the CE, data archive (DA), and sensors. The CE is defined as the portion of the CRS containing the policybased control mechanism and the cognitive control mechanism, which must have knowledge about the current state and a set of attainable states, and may have knowledge about the cost associated with (state) transitions of the reconfigurable radio platform. The DA is defined as a logical entity where sensing information obtained from spectrum sensors or other information sources, and regulatory and policy information

IEEE Wireless Communications • February 2011

Primary user base station

IEEE P1900.6 logical interface

Cognitive engine

Primary user

Standalone spectrum sensor

Secondary user

Embedded spectrum sensor

Data archive Secondary system base station

Figure 1. CRS scenario where the IEEE P1900.6 interface can be employed for sensing information exchange among different entities.

are stored systematically. The sensor can play a client role of another sensor; that is, a sensor can receive sensing information from one or more sensors and forward it to CE or DA. Figure 2 gives an abstract view of the above described scenario and illustrates different interfaces between spectrum sensors and their clients that are within the scope of the P1900.6 standard. The CE/DA-S interface is used for exchanging sensing information between a CE or DA and a sensor. As an example, the CE/DA-S interface is used in scenarios where a given CE or DA obtains sensing information from one or several sensors, or a given sensor provides sensing information to one or several CEs or a DA. The S-S interface is used for exchanging sensing information between sensors; this is needed in cases where one sensor may not be able to obtain all required information, or in scenarios where sensors A and B exchange sensing information for collaborative or cooperative sensing. The CE-CE/DA interface is used for exchanging sensing information between CEs or between a CE and a DA. As an example, the CE-CE/DA interface is used in scenarios where CEs A and B exchange sensing information for collaborative or cooperative sensing. The CE-CE/DA interface is also used in scenarios where a CE obtains sensing information as well as policies (e.g., sensing and access rules) and regulatory information from a DA.

USE CASES At the current stage the WG has defined 31 use cases within the scope and purpose of the standard. These use cases represent the situations

31

MOESSNER LAYOUT

2/7/11

10:42 AM

Page 32

Cognitive engine/ data archive

Cognitive engine

P1900.6 interfaces*

Sensor A

Sensor B

*The client role can be taken by a cognitive engine, sensor, or data archive. CE/DA-S interface between cognitive engine or data archive and sensor to exchange sensing information and sensing control information. S-S interface between sensor and sensor to exchange sensing information and sensing control information. CE-CE/DA interface between cognitive engine and cognitive engine or data archive to exchange sensing information and sensing control information.

Figure 2. Scope of the P1900.6 standard.

where the upcoming P1900.6 interface will be employed in a CRS. In this article we give one use case as an example. A complete description of the use cases is given in [7]. CRS with distributed sensors: The CRS analyzes spectrum usage and makes a decision on spectrum access based on the sensing information provided by sensors located at different physical positions. For example, these sensors can be deployed in certain service areas such as a city hot spot area or a university campus area to form a spectrum sensor network. A CR terminal (e.g., a cognitive mobile phone) can initiate the sensing function of the sensing network once it enters the service area of the sensor network.

System Model — Different applications that implement CR may have different ways of exchanging sensing information and of using sensing information interfaces. This subsection provides an abstract system model that is not dependent on a particular topology and spectrum usage. Based on the system model, the use cases are classified in the following subsection. The system model consists of three scenarios. Figure 3a shows a single CE/DA and single sensor scenario, which is denoted the 1:1 scenario. In this scenario, a single sensor provides sensing information to one CE/DA. This is the simplest scenario and is considered as a reference. Figure 3b shows a single CE/DA and multiple sensors scenario, which is denoted the 1:N scenario. N number of sensors provide sensing information to a CE/DA. In other words, one CE/DA can access sensing information from N sensors. In

32

cooperative sensing and collaborative sensing, the CE/DA can access multiple sensors and make DSA based on sensing information from distributed sensors. Also, multiple sensors can exchange information and provide the CE/DA with an improved sensing result. Figure 3c shows an M number of CEs/DAs and single sensor scenario, which is denoted the M:1 scenario, where multiple CEs/DAs access sensing information of one sensor. In other words, one sensor provides sensing information to multiple CEs/DAs. In a situation where the sensor can be accessed by multiple CEs/DAs, these CEs/DAs can share usage of the sensor. Or if one of the sensors is capable of accessing a sensor, other CEs/DAs can relay sensing information from a sensor to the CE/DA. For the case of multiple CEs/DAs and multiple sensors, the system can be decomposed into a combination of the first three scenarios. Note that this is understood within the context of model description and does not imply decomposition on the implementation level.

Use Case Classification and Analysis — Different applications that implement CR may have different ways of exchanging sensing information and of using sensing information interfaces. To analyze the use cases and extract sensing requirements, the WG classified various use cases based on how the sensing information and interface are utilized. For this classification, the system model described earlier is used. From the spectrum usage perspective there are two usage models: long-term and short-term. The long-term spectrum usage model refers to the situation where spectrum is used over a relatively long period. In most cases the spectrum is used as a substitute primary user service. In the short-term spectrum usage model spectrum is used over short periods (e.g., autonomous spectrum access). Within each usage model, use cases are further classified into three scenarios (i.e., 1:1, 1:N, and M:1) based on the system model given earlier. Such an approach clarifies which interfaces are involved during the information exchange. Furthermore, it assists in generating the sensing parameters for each use case. Table 1 presents the classification of selected use cases.

SENSING REQUIREMENTS In many scenarios involving CR, a communication device needs to capture the current usage of the spectrum before establishing its own communication. This behavior is referred to as detecting free bands, which means identifying frequency bands that are free of already established communications. Band BL can be considered free if the signal received in this band BL is only made of noise. On the other hand, if signals are detected in the presence of noise, the band is declared occupied. Thus, the function the detector has to perform is that of detecting signals in the presence of noise, which can be stated as the following hypothesis: H0: r(t) = n(t)

H1: r(t) = hs(t) + n(t),

where H 0 is the hypothesis that B L is a free band, and H1 assumes that BL is occupied. n(t) is

IEEE Wireless Communications • February 2011

MOESSNER LAYOUT

2/7/11

Cognitive engine/ Data archive

10:42 AM

Page 33

Cognitive engine/ Data archive

Cognitive engine/ Data archive

Cognitive engine

P1900.6 interfaces P1900.6 interfaces

Sensor

P1900.6 interfaces

Sensor A

(a)

Sensor B

(b)

Sensor

(c)

Figure 3. IEEE P1900.6 system model: a) 1:1 scenario; b) 1:N scenario; c) M:1 scenario.

noise, and s(t) is a telecommunication signal sent over communication channel h. The major issue in asserting a band is free is that the sensor must be able to detect signals at very low SNRs. Many techniques have been studied for signal detection and recently adapted to the CR case. These techniques can be classified according to the a priori knowledge of the sensor on the signal to detect. Simple energy detectors (EDs) have limited performance but exhibit low complexity and no a priori knowledge is needed, whereas feature detectors (e.g., cyclostationarity detectors) outperform EDs but at the cost of higher complexity and the additional requirement of some knowledge about the waveform to detect. A survey of spectrum sensing techniques has been carried out for the P1900.6 WG and is captured in [8]. This survey highlights the fact that the choice of a sensing technique very much depends on scenarios and hardware resources available. For instance, in TV white space scenarios, the primary users (PUs) are known, and the sensor can utilize a priori knowledge about the PU to improve its detection capability, whereas in more ambitious scenarios such as technology agnostic spectrum usage, a priori knowledge is almost impossible to consider. Bearing in mind this variety of cases and solutions, it is not possible for a group like P1900.6 to focus on specific sensing techniques in order to define the sensor interface requirements. The aim of the WG is rather to consider more abstract features that can be generalized to a large number of systems, while using the specific sensing scheme analysis to cross-check how these requirements stand against practical implementation, and to see if some specific cases lead to additional requirements.

REFERENCE MODEL To give the industry a guideline on how to implement the logical interface, the WG defines a reference model, shown in Fig. 4. In general, the P1900.6 logical interfaces in P1900.6 defined entities (i.e., sensor, CE, and DA) have the reference model shown in the figure. Sensors and their clients can have all service access points (SAPs) or a subset of the three SAPs (i.e., application SAP [A-SAP], communication SAP [C-

IEEE Wireless Communications • February 2011

SAP], and measurement SAP [M-SAP]); their logical positions and boundaries are depicted in Figs. 4 and 5. They utilize distinct SAPs to realize the logical interface. The P1900.6 services in the reference model refer to functional blocks inside sensors and clients required to realize logical interfaces between sensors and clients. In particular, these functional blocks are responsible for generating data structures defined by the P1900.6 standard for exchanging sensing and sensing control information. The A-SAP defines a set of generic primitives and data structures to control the P1900.6 entity and/or obtain the sensing results for application purposes. For example, this SAP is used for a P1900.6 entity to utilize sensing information for its purpose (e.g., policy investigation and analysis of spectrum usage). The A-SAP may provide functions to set up a configuration of P1900.6 entities (e.g., sensors and cognitive engine), to configure these for collaborative sensing, to start the data acquisition and processing (e.g., policy processing), and to obtain the results of P1900.6 processing in order to configure the radio frequency (RF) interface accordingly. The M-SAP is used by P1900.6 entities to access P1900.6 compliant services provided by the station’s hardware and/or firmware to control the spectrum measurement module, such as collocated physical spectrum measurement module (i.e., analog-to-digital/digital-to-analog converter [ADC/DAC], filtering, signal condition, etc.), and acquire measurement data from these. For example, a station (terminal) utilizes its RF interface during idle times for spectrum measurement and provides RF spectrum data to collocated P1900.6 sensor entities that registered at the local M-SAP. The M-SAP shall be instantiated by the station’s P1900.6 compliant measurement function. The C-SAP is used for sensing information (sensing message, sensor message, control message, and regulatory information) exchange between sensors and their clients. The client role can be taken by sensor, CE, and DA. It abstracts communication mechanisms from P1900.6 entities by providing a set of generic primitives and mapping these primitives on transport protocols. A P1900.6 compliant message transport service has to be provided by the station in order to

Many techniques have been studied for signal detection and recently adapted to the CR case. These techniques can be classified according to the a priori knowledge of the sensor on the signal to detect.

33

MOESSNER LAYOUT

2/7/11

10:42 AM

Page 34

locate a remote P1900.6 peer entity and establish a communication link with this entity. Message exchange then takes place as defined by this standard, and it is the responsibility of the transport service to map these message transfers to a suitable transport, network or link layer communication. For example, a P1900.6 entity CE can take the role of a client to a remote P1900.6 entity sensor by sending a request to configure the sensor for delivering measurement data to the requestor utilizing its C-SAP functionality. The C-SAP shall be instantiated by the P1900.6 compliant message transport service. The services available at these SAPs realize the logical interfaces (i.e., CE/DA-S interface, S-S interface, and CE/DA-CE interface) between different P1900.6 entities. Spectrum usage models

Scenarios of system model

Figure 5 provides alternative views of the reference model. The figure shows how the above mentioned P1900.6 services and SAPs are involved in the sensing information exchange between sensors and their clients. Figure 5a shows a situation where the CE/DA and sensor exchange sensing information. The controls/applications at the CE/DA generate sensing information requests (desired frequency band to sense, desired sensing performance, etc.). The P1900.6 services receive these requests at the A-SAP and convert these requests into a message format defined by the commonly understood interface. The message is sent to the C-SAP. The communication subsystem provides communication services and sends the requests as messages to a sensor that

Use cases and application examples

Load sharing to reduce blocking at peak traffic times: two networks that exhibit peak loads at different times and at different locations can relieve network congestion by offloading their peak traffic to one another using CR.

Long term spectrum usage

1:1

Introduction of new users or services: The use of CR to find and utilize spectrum that is underutilized at a specific time and location would allow the introduction of new users or services without significant delay. Worldwide mobility: A CR radio could do automatic RAT switching to avoid missed calls in the event that a user travels to a different country or location. Emergency services: CR allows all of the emergency personnel to find common usage channels throughout the disaster area which can be operated without interference from any of the surviving legacy communications

1:N Dynamic spectrum assignment: In the dynamic spectrum assignment use case, frequency bands are dynamically assigned to the RANs in order to optimize radio resource usage and improve quality-of-service.

M:1

Self management of uncoordinated spectrum: CR provides an effective means of self coordination so as to avoid interference with other networks while providing useful throughput, for instance in unlicensed bands where central coordination might not be practical. Dynamic spectrum assignment: In the dynamic spectrum assignment use case, frequency bands are dynamically assigned to the RANs in order to optimize radio resource usage and improve quality-of-service.

1:1

Add capacity for emergency: Dynamic spectrum access, or the ability for CRs to identify unused or underutilized spectrum, could be used in emergency scenario and provide a means for expanding capacity when needed. Network extension for coverage: This can be achieved by automatically reconfiguring the CRs to include a repeater capability to extend network coverage to areas where radios are otherwise cut off from their infrastructure.

Short term spectrum usage

1:N

Dynamic spectrum sharing: In the dynamic spectrum sharing use case, frequency bands assigned to RANs are fixed. However, a particular frequency band can be shared by several RANs to optimize radio resource usage and improve quality-of-service. Policy violation: This use case is focuses on the usage of spectrum sensing in CR for the investigation of policy violations.

M:1

Cognitive relay: In cognitive wireless network, multiple cognitive terminals form a wireless ad hoc network or mesh network. Such network can be considered as having multiple peer to peer links. Each radio in the peer to peer link detects the behavior of the radio frequency environment in multiple channels and provides efficient communication links and paths. Priority based sensing information provision: when multiple clients contend to access the same sensing information source at a time, clients that have a high priority level can access the sensing information first. Distributed radio resource usage optimization: In this use case, frequency bands assigned to RANs are fixed. Also, reconfiguration of RANs is not considered in this use case. Instead, reconfigurable terminals with or without multi-homing capability are considered.

Table 1. Classification of use cases.

34

IEEE Wireless Communications • February 2011

10:42 AM

Page 35

also has a communication subsystem through a radio channel (wireless LAN, ultrai wideband [UWB], etc.). The P1900.6 services at the sensor unpack the messages based on the commonly understood data structure and obtain the requests sent from the CE/DA. Then these sensing requests are sent to the spectrum measurement through the M-SAP. The hardware/firmware performs the spectrum measurement and produces sensing results. The P1900.6 services at the sensor obtain the sensing results from the M-SAP according to the commonly understood data structure. Through the C-SAP, the sensor sends the sensing results to the client CE/DA. The P1900.6 services at the CE/DA unpack the sensing results based on the commonly understood data structure and provide the sensing information to the controls/applications through the A-SAP. Similarly, Fig. 5b provides an alternative view of the reference model for the situation where two sensors exchange sensing information. Sensing information received from one sensor can be processed at another sensor (the client). For example, the sensing information from two sensors can be combined to provide improved sensing information. Finally, Fig. 5c provides an alternative view of the reference model for the situation where one CE can access one sensor and share sensing information with another CE/DA.

DATA STRUCTURE Based on the use case analysis, reference model, and sensing techniques the WG has listed four types of information (i.e., sensing information, sensor information, control information, and policy information). A sensing message relates to measurement data (e.g., sensing results) from sensors and clients. A control message relates to the control of sensing activity of sensors or clients, such as target performance and sensing duration. A sensor message relates to sensor specification, sensor capability, or sensor identity. Regulatory information is defined as sensing and access rules derived from radio spectrum regulation, such as required sensing duration and sensitivity levels. Of course, all these messages depend significantly on the regulatory requirements. For instance, if TV white space detection assumes that digital TV channels of 6 MHz must be detected with a sensitivity of –114 dBm, the client must ensure that the sensor can provide this performance level. Then, when all messages are identified, an important task is to define how these messages are represented in order to be well adapted to real case implementation. For instance, the message format is to be kept reasonably complex while generic enough to support all cases investigated. Each parameter of these four types of information is defined by giving a short textual description, parameter ID, name, unit, parameter type, and size, as well as range and resolution. The textual description explains the purpose of the parameter. The parameter’s name provides a unique identification of the parameter in human readable form such as start frequency, detection threshold, and sensor location,

IEEE Wireless Communications • February 2011

Control/ application

Spectrum measurement

IEEE P1900.6 service

Measurement SAP (M-SAP)

2/7/11

Application SAP (A-SAP)

MOESSNER LAYOUT

Communication SAP (C-SAP) Communication subsystem

Figure 4. Reference model of P1900.6 standard.

whereas the numerical ID is given to unambiguously identify the parameter in the process of information exchange between P1900.6 logical entities. The size of a parameter is defined as the number of elements contained in the parameter. Data types include primitive types such as integer and Boolean, as well as complex and derived types such as enumeration and structured. Data unit is used to describe the physical units of the data. For example, bandwidth is defined as an unbounded parameter with the unit frequency. It is defined by the lower and upper frequencies, thus having the array (frequency) type and size 2. Here, frequency is also a defined parameter and used as a data type of the elements in this parameter.

DATA REPRESENTATION To represent various sensing related information defined in the IEEE P1900.6 WG while allowing for extendibility of uses to add more sensing information, the object-oriented Unified Modeling Language (UML) is employed. Sensing related information is categorized into four classes: control information, sensor information, sensing information, and regulatory requirement. The control information class contains the information to control the transport of sensing information. It also contains the information to set the spectrum measurement, such as measurement objective, measurement profile, and measurement performance. The sensor information class contains information to describe spectrum sensors. For example, this class contains the sensor’s physical profile, such as antenna profile and location profile, and manufacturer’s information. Information that describes the spectrum measurement capability is also included in this class. The sensing information class contains mainly the measurement data such as measured values of signals, channels, and RATs. It also contains the information related to the measurement that has been carried out, such as where, when, and how the measurement has been done. Finally, the regulatory requirement class contains different regulatory information related to different primary signals in different regions and countries.

35

MOESSNER LAYOUT

2/7/11

10:42 AM

Page 36

USE OF THE INTERFACES: GENERIC PROCEDURE This subsection gives a generic procedure and its usage examples. The generic procedure can be seen as a basic building block for a more complex procedure required for the exchange of sensing related information between one or multiple sensors and their clients. To express the procedure, the notions of service user and service provider model as specified by ITU-T X.210 [9] are used as a reference. The service provider refers to an abstraction of the totality of those entities that provide an P1900.6 service to the service user. The P1900.6 service provider includes the P1900.6 service (as indicated in the reference model in the previous section), and the associated SAPs that instantiate the logical interface for exchange of sensing related information between the P1900.6 servers and clients. The service user refers to the IEEE P1900.6 logical entities that shall implement or use the P1900.6 logical interface to exchange

Control/ application A-SAP

Spectrum measurement M-SAP

P1900.6 service

P1900.6 service

C-SAP

C-SAP

Communication subsystem (data bus, wired or wireless channel) Client (CE/DA)

Sensor

Control/ application

A-SAP

(a)

Spectrum measurement M-SAP

Spectrum measurement M-SAP

P1900.6 service

P1900.6 service

C-SAP

C-SAP

Communication subsystem (data bus, wired or wireless channel) Client (Sensor)

FUTURE DIRECTION Sensor

(b)

Control/ application A-SAP

Control/ application A-SAP

P1900.6 service

P1900.6 service

C-SAP

C-SAP

CE/DA (c)

Figure 5. Alternative views of the reference model.

36

The P1900.6 draft standard passed the sponsor ballot in late 2010 and is currently with the Standards Board for approval. DYSPAN working group 6 continues its work, and is currently developing a new project aiming to complement the definitions of the current draft standard.

CONCLUSION

Communication subsystem (data bus, wired or wireless channel) Client (CE)

sensing information. For example, the P1900.6 logical entity that shall use the logical interface to obtain sensing related information plays the client role. Here we give a simple use case where the CE obtains sensing information from a logical entity sensor to exemplify the procedure. Both the CE and sensor are P1900.6 service users. They are using the services provided by the P1900.6 service provider to realize information exchange, where the CE is playing the client role and the sensor is playing the server role of a client/server model: 1. When a CE needs to obtain sensing information, it issues a request to the P1900.6 service provider. 2. Then the P1900.6 service provider forwards the request along with its parameters toward a sensor. The parameters from the P1900.6 logical interface form the service data unit (SDU), which shall be communicated between CE and sensor as the payload of a protocol data unit (PDU) used by the protocol layers beneath. Upon reception of an SDU, the P1900.6 service provider generates an indication toward the sensor. The actual procedure and protocol used in the communication subsystem is out of the scope of this standard. Upon receiving the indication, the sensor performs further processing to consider or reject the request. 3. In response to the indication, the sensor issues a response to the P1900.6 service provider to send the results (e.g., the sensing information) as requested by the CE. 4. The P1900.6 service provider generates a confirmation toward the CE while providing the SDU as originated by the sensor. For this type of synchronous communication flow, the confirmation indicates that the P1900.6 request has been completed, and the CE may now check if it has obtained valid results. Note that in the above communication message flow, only the service request from the client (i.e., CE) is depicted. In this case the sensor is stated as acceptor while the CE is stated as requestor. The requestor and acceptor roles change depending on who is initiating the service request.

This article presents the activities of the IEEE P1900.6 WG, which is currently defining an interface of sensing information exchange between sensors and their clients. By explaining the scope, use cases, and system model reference models we have shown how the interface can be used in various use cases of CRSs. The interface defined in this standard will ensure the interoperability of sensors and other system components of the CRS. Thus, the standard will bring

IEEE Wireless Communications • February 2011

MOESSNER LAYOUT

2/7/11

10:42 AM

Page 37

freedom to manufacturers to develop their own advanced sensing techniques.

ACKNOWLEDGMENT The authors would like to express their sincere gratitude to all the participants of the IEEE P1900.6 WG.

REFERENCES [1] G. Staple and K. Werbach, “The End of Spectrum Scarcity,” IEEE Spectrum, vol. 41, no. 3, Mar. 2004, pp. 48–52. [2] J. Mitola III and G. Q. Maguire Jr., “Cognitive Radio: Making Software Radio More Personal,” IEEE Pers. Commun., vol. 6, no. 4, 1999. [3] S. Haykin, “Cognitive Dynamic Systems,” Proc. IEEE, vol. 94, no. 11, Nov. 2006, pp. 1910–11. [4] IEEE P802.22/ DRAFTv2.0, “Draft Standard for Wireless Regional Area Networks Part 22: Cognitive Wireless RAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications: Policies and Procedures for Operation in the TV Bands.” [5] IEEE SCC41; http://www.scc41.org/. [6] IEEE P1900.6 WG; http://grouper.ieee.org/groups/scc41/ 6/index.htm. [7] IEEE Std. 1900.6/D0.6, “IEEE Draft Standard for Spectrum Sensing Interfaces and Data Structures for Dynamic Spectrum Access and Other Advanced Radio Communication Systems,” Feb. 2010. [8] D. Noguet et al., “Sensing Techniques for Cognitive Radio — State of the Art and Trends,” IEEE SCC41 — P1900.6 Working Group,” White Paper, Apr. 2009; http://grouper.ieee.org/groups/scc41/6/documents/white _papers/P1900.6_WhitePaper_Sensing_final.pdf. [9] ITU Rec. X.210, “Information Technology (Open System Interconnection) Basic Reference Model: Conventions for the Definition of OSI Services.”

BIOGRAPHIES KLAUS MOESSNER ([email protected]) is a professorial research fellow in the Centre for Communication Systems Research at the University of Surrey, United Kingdom. He earned his Dipl-Ing (FH) at the University of Applied Sciences, Offenburg, Germany, an M.Sc. from Brunel University, and his Ph.D. from the University of Surrey. His research interests include dynamic spectrum allocation, cognitive radio networks reconfiguration management, service platforms, and adaptability of multimodal user interfaces. He is chair of the IEEE SCC41 WG6 defining interfaces and data structures for DSA. HIROSHI HARADA [M] ([email protected]) is director of the Ubiquitous Mobile Communication Group, National Institute of Information and Communications Technology (NICT), Yokosuka, Japan. He is also director of NICT Singapore Wireless Communication Laboratory. He joined the Communications Research Laboratory, Ministry of Posts and Communications (currently NICT), in 1995. His research interests include software-defined radio, cognitive radio, dynamic spectrum access networks, and broadband wireless access systems on the microwave and millimeter-wave bands. He currently serves on the Board of Directors of the SDR Forum, and has been Chair of the IEEE Standards Coordinating Committee 41 (IEEE SCC41; IEEE P1900) since 2009 and Vice Chair of IEEE P1900.4 since 2008. Moreover, he was Chair of the Institute of Electronics, Information, and Communication Engineers (IEICE) Technical Committee on Software Radio from 2005 to 2007. He was the recipient of the Achievement Award and made a Fellow from IEICE in 2006 and 2009, respectively, and received the Achievement Award from the Association of Radio Industries and Businesses (ARIB) in 2009 on the topic of SDR and cognitive radio research and development. C HEN S UN [S’02, M‘05] ([email protected]) received a B.E. degree in electrical engineering from Northwestern Polytechnical University, Xi’an, China, in 2000 and a Ph.D. degree in electrical engineering from Nanyang Technological University, Singapore, in 2005. From August 2004 to May 2008 he was a researcher with ATR Wave Engineering Laboratories, Kyoto, Japan. In June 2008 he joined the Ubiquitous Mobile Communications Group, NICT, Yokosuka, Japan, as an Expert Researcher working on cognitive radio. His research interests

IEEE Wireless Communications • February 2011

include spectrum sensing, dynamic spectrum access, smart antennas, and cooperative communications. He is a voting member of IEEE Standards Coordinating Committee 41. He is also a voting member of the IEEE 1900.6 Working Group and serves as the Technical Editor. YOHANNES D. ALEMSEGED [S’06, M‘08] ([email protected]) received a B.Sc. degree in electrical and electronic technology from Nazareth Technical College (currently Adama University), Nazareth, Ethiopia, in 1997, an M.Sc. degree in electrical engineering from Addis Ababa University, Ethiopia, in 2002, and a Ph.D. degree from Graz University of Technology, Austria, in 2008. He is currently an expert researcher with NICT, Yokosuka, Japan. His research interests include digital signal processing for communications, spectrum-sensing algorithms for cognitive radios, and lowcomplexity ultra-wideband transceivers. He is a voting member of IEEE Standards Coordinating Committee 41 and serves as the Committee Secretary. He is also a voting member of the IEEE 1900.6 Working Group. H A N GUYEN T RAN [M‘08] ([email protected]) received his B.E. and M.E. degrees in electronics and information engineering and a Ph.D. degree in information science and technology from the University of Tokyo, Japan, in 2000, 2002, and 2005, respectively. He is currently an expert researcher with the New Generation Wireless Communications Research Center, NICT, Yokosuka, Japan. His research interests include wireless networking and cognitive radio. He is a voting member of IEEE Standards Coordinating Committee 41. He is also a voting member of the IEEE 1900.6 Working Group and serves as the Working Group Secretary.

The interfaces defined in this standard will ensure the interoperability of sensors and other system components of the CRS. Thus, the standard will bring freedom to manufacturers to develop their own advanced sensing techniques.

D OMINIQUE N OGUET joined CEA-LETI in 1998 where he has carried out digital communication hardware architecture research. His research activities cover reconfigurable and flexible radio, and more recently cognitive radio. Since 2001 he has had several national and international project management positions in the field of wireless communication. He currently leads flexible radio activities (WPRC) within the European Network of Excellence NEWCOM++ and is the technical manager of QoSMOS, a major EU project on cognitive radio. He received a best paper award and the best Ph.D. award from INPG. He has authored or co-authored about 40 papers in peer reviewed journals and conferences. He is currently head of the Digital Architectures and Prototypes group at LETI, where he also leads cognitive radio activities. RYO SAWAI ([email protected]) received his B.E., M.E. and Ph.D. degrees in electrical and electronic engineering from Chuo University, Tokyo, Japan, in 1998, 2000, and 2002, respectively. From July 1999 to September 1999 he joined the Nokia student exchange program in Helsinki, Finland. From October 1999 to May 2002 he was a student trainee with NICT, Yokosuka, Japan. In April 2002 he joined the Ubiquitous Technology Laboratories, Sony Corporation, Japan. He is currently a researcher at System Technology Laboratories, Sony Corporation, Japan. His research interests include advanced coding design (e.g., turbo code/ decoder and low density parity check code/decoder), reconfigurable hardware architecture design, dynamic spectrum access, and multi-input and multi-output signal processing technologies. He received the IEEE VTS Japan Researcher’s Encouragement Award in 2001. He is a voting member of IEEE Standards Coordinating Committee 41. He is also a voting member of the IEEE 1900.6 Working Group. NAOTAKA SATO ([email protected]) received B.E. and M.E. degrees in electrical engineering from Tokyo University of Science, Japan, in 1991 and 1993. From April 1993 to April 1995 he was an engineer in Sony Corporation, Tokyo, Japan. From April 1995 to May 1999 he was an RF engineer in Sony Electronics, Inc., San Diego, California. From May 1999 to September 2001 he was a senior RF engineer in Sony Corporation, Tokyo, Japan. From October 2001 to March 2005 he was senior RF engineer in Sony Ericsson Mobile Communications Japan, Inc., Tokyo, Japan. In April 2005 he joined the Communication Technology Laboratory, Sony Corporation, Tokyo, Japan, as a senior researcher working on cognitive radio and 4G cellular technology. His research interests include dynamic spectrum access and RF system architecture. He is a voting member of IEEE Standards Coordinating Committee 41. He is also a voting member of the IEEE 1900.6 Working Group.

37

JIANG LAYOUT

2/7/11

10:43 AM

Page 38

ACCEPTED FROM OPEN CALL

MULTICARRIER MODULATION AND COOPERATIVE COMMUNICATION IN MULTIHOP COGNITIVE RADIO NETWORKS TAO LUO, BEIJING UNIVERSITY OF POSTS AND TELECOMMUNICATIONS FEI LIN, SHANDONG INSTITUTE OF LIGHT INDUSTRY TAO JIANG, HUAZHONG UNIVERSITY OF SCIENCE AND TECHNOLOGY MOHSEN GUIZANI, KUWAIT UNIVERSITY WEN CHEN, SHANGHAI JIAOTONG UNIVERSITY

ABSTRACT

PUt

SUt1

SUt2 Cooperative spec communication be

The authors combine CR with the cooperative diversity technique, and construct three cooperative diversity cognitive models: the collaborative spectrum sensing model, the cooperative communication cognitive model, and the hybrid model.

For high-data-rate wireless communication systems, two major issues are the underutilization of limited available radio spectrum and the effect of channel fading. Using dynamic spectrum access, cognitive radio can improve spectrum utilization. Almost all proposed CR systems are based on multicarrier modulation since multiple users can access the MCM systems by allocating subcarriers. Generally, MCM mainly includes two different schemes, orthogonal frequency-division multiplexing and filtered multitone modulation. Considering mutual interference elimination, synchronization, and transmission efficiency, we conclude that FMT is better than OFDM in MCM-based CR systems. Additionally, cooperative diversity can reduce the fading effect since the space diversity gain can be obtained through the distributed antennas of each user. Hence, in this article, we combine CR with the cooperative diversity technique, and then construct three cooperative diversity cognitive models: the collaborative spectrum sensing model, the cooperative communication cognitive model, and the hybrid model. Additionally, radio resources can be extended from time-frequency dimensions to space-time-frequency dimensions in the proposed models, which effectively improves both spectrum utilization and MCM-CR system performance. Finally, extensive simulations are conducted to show the validity and effectiveness of the proposed models.

INTRODUCTION As we know, almost all existing wireless communication networks are allocated a fixed spectrum, resulting in a large portion of the assigned spectrum being used sporadically [1–3]. According to a report by the Federal Communications Com-

38

1536-1284/11/$25.00 © 2011 IEEE

mission (FCC), the percentage of the assigned spectrum that is occupied ranges only from 15 to 85 percent, varying widely in time and geographical position [2]. Hence, the limitation and underutilization of available spectrum resource accelerates new research. Ultra-wideband (UWB) and cognitive radio (CR) are candidate techniques to improve utilization of the assigned spectrum. Under the power spectral density emission limit of Part 15, which is –41.3 dBm/MHz or significantly lower (as low as –75 dBm/MHz), UWB can share wideband spectrum with other existing wireless systems. However, the application of UWB is limited because of its ultra-wideband frequency range, poor agility, and high complexity. The term cognitive radio was first introduced by Joseph Mitola. As a promising candidate, CR has the ability to share or reuse spectrum in an opportunistic manner by employing spectrum overlay and/or spectrum underlay approaches, which results in an increase of spectrum utilization [1–3]. In [1], the authors defined CR as an intelligent wireless communication technology that is aware of its surrounding environment, uses the methodology of understanding-by-building to learn from the environment, and then adapts its internal states to statistical variations in the incoming radio frequency stimuli by making corresponding changes in certain operating parameters (e.g., transmit power, carrier frequency, and modulation strategy) in real time. Moreover, high-data-rate wireless communication systems are limited not only by the limited spectrum, but often more significantly by the fading effects due to multipath propagation, the Doppler effect, and the angle spread of the wireless channel. Diversity is one of the most effectual methods to resist the fading effect. As we know, multiple-input multiple-output (MIMO), multicarrier modulation (MCM) and code-division multiple access (CDMA) are

IEEE Wireless Communications • February 2011

JIANG LAYOUT

2/7/11

10:43 AM

Page 39

commonly considered as candidates. The MCM schemes, such as orthogonal frequency-division multiplexing (OFDM) and filtered multitone (FMT) modulation [4], are approaches to overcome the intersymbol interference (ISI) caused by multipath propagation. CDMA can suppress narrowband noise and interference by spreading the signal bandwidth to a wideband spectrum. Hence, MCM such as OFDM and multicarrier CDMA (MC-CDMA) are hailed as promising candidates for realizing spectrum overlay and spectrum underlay CR applications [2, 5], respectively. In MC-CDMA-based CR (MC-CDMA-CR) systems, it is possible to abandon distributed sensing in a way that the transmitting secondary user (SU) can spread its signal across the entire band, including that occupied by the primary user (PU), which results in a base station and/or signaling channel not being needed. Furthermore, MCCDMA allows narrowband PU interferers to be excluded locally at the SU receiver, hence improving its performance. In other words, MC-CDMA technology is more suitable for spectrum underlay CR applications. However, we mainly focus on spectrum overlay CR systems in this article. Therefore, MC-CDMA-CR is not discussed below. Additionally, MIMO can improve the channel capacity and performance of wireless communication systems by using space and time resources. However, it is not sufficient that only one antenna be built in the mobile station due to the limitations of its cost, size, and complexity. Moreover, the MIMO technique does not work well when the fading is large-scale. Consequently, cooperative diversity [6], also called virtual MIMO, has been proposed, in which users can transmit data by sharing antennas of other users surrounding them. Similarly, a cooperative communication system can obtain the space diversity gain and improve reliability. The basic ideas behind cooperative communication can be traced back to the groundbreaking work of Cover and El Gamal on the information theoretic properties of the relay channel with additive white Gaussian noise (AWGN). However, the aim of relays is only to help the source transmit information, whereas the users in cooperative communication systems can act as both information sources and relays. Nosratinia et al. have proven that even though the interuser channel is noisy, cooperation can still lead not only to an increase in capacity for both users, but also to a more robust system [6]. In this case, the achievable rates of users are less susceptible to channel variations. There are three main cooperative signaling protocols: amplify-and-forward (AF), detect-and-forward (DF), and coded cooperation (CC) methods [6]. In this field, the key techniques mainly include power allocation, cooperative partner selection, performance evaluation, and so forth. In fact, CR and cooperative communication have developed rapidly in their own fields. However, there is little advantage in improving spectrum utilization when only cooperative communication is used, whereas CR is not good for improving the symbol error rate performance of each user. Consequently, combining CR with

IEEE Wireless Communications • February 2011

cooperative communication may be a good solution to both problems. Ghasemi [7] and Ganesan [8] first proposed the collaborative (exchanging spectrum hole vectors with each other) and cooperative communication scheme to improve the detection probability of spectrum sensing in CR systems. Nevertheless, both of them are only based on AF and do not take the locations of SUs into consideration. After that, Devroye introduced cooperative communication to data transmission, and obtained fundamental limits of achievable rates in CR systems. Obviously, this model is ideal and needs cooperative communications among SUs. Finally, Simeone et al. studied the cognitive relaying scheme between PUs and SUs [9], which is of course a simple and elementary discussion. Consequently, in this article we expand the models of Ghasemi and Devroye, and give an overview of the cooperative diversity cognitive models, which combine MCM, cooperative diversity, and CR techniques together. This article is organized as follows. In the next section we discuss MCM techniques in the multihop CR system. Then three cooperative diversity cognitive models are proposed. Simulation results show the effectiveness of the proposed models in the following section. Conclusions are drawn in the final section.

The longer the delay spread of the channel is, the longer the length of the CP is, and thus the less the efficiency is. Obviously, these methods conflict with the concept of improving the spectrum utilization of the CR technique.

MULTICARRIER MODULATION TECHNIQUES IN MULTIHOP CR NETWORKS Recently, some work has been reported on multihop CR networks [10], in which each node has a list of available frequency bands and must work adaptively among these frequency bands because of dynamic spectrum access. It is well known that two nodes cannot communicate if they work on different frequency bands. Hence, routing in multihop CR networks becomes a critical and challenging issue. In general, solutions of this problem mainly focus on the methods in the network layer, whose processing delay is on the order of milliseconds. However, the high-speed wireless channel in multihop CR networks varies on the order of microseconds due to multipath fading, the Doppler effect, and dynamic occupancy of the subchannel by PUs. Therefore, the solutions proposed in the network layer may cause heavy interference to PUs. To this end, adopting MCM to intersection nodes (i.e., RUs) of multihop CR networks may be a good solution in the physical layer. Because of the usage of MCM, the intersection node can allocate some unused subcarriers to different information flows; thus, all flows can be transmitted simultaneously. In fact, considering that the access of multiple users can be implemented by the allocation of subcarriers in an MCM system, almost all of the proposed spectrum overlay CR systems are based on MCM technology, specifically the MCM-CR system. Moreover, almost all proposed MCM-CR systems are based on OFDM [2], such as IEEE 802.22, the spectrum pooling system proposed by Timo A. Weiss, and the Next Generation (xG) communication networks proposed by the Defense Advanced Research

39

JIANG LAYOUT

2/7/11

10:43 AM

Page 40

0 -10

Amplitude /dB

-20 -30 -40 -50 -60

-70 -80

0

0.02

0.04

0.06

0.08

0.1

fT/Nc

Figure 1. Frequency response of the first five subcarriers in MCM system with Nc = 64 (solid: FMT, dashed: OFDM). Projects Agency (DARPA) [2]. Recently, cosine modulated multitone (CMT)-based CR (CMTCR) and FMT-based CR (FMT-CR) systems have also been proposed [3]. In an MCM-CR system, a maximum likelihood detection (MLD) model is deduced under the constraint of the interference temperature [3]. Moreover, the optimal detection region, and the probability of detection and false alarm are obtained in [3]. Spectrum allocation and access algorithms are well studied in [2, 3]. Based on the MLD scheme and Markovian chain prediction (MCP) model, an efficient SU access algorithm based on spectrum hole vector (SHV) is proposed to meet the quality of service (QoS) requirements of the SUs in a centralized MCMCR system in [3]. However, one issue of the MCM-CR system is that it requires a large size inverse discrete Fourier transform (IDFT)/DFT operation due to the varying locations of each subcarrier in a wide spectrum, resulting in a heavy implementation complexity and serious delay. To this end, a scheme combining CooleyTukey’s recursive algorithm with pruning algorithm can be adopted. Now, the main issue becomes: which technique is better in an MCMCR system, OFDM or FMT? We discuss more detail in the following. Recently, some challenges in the physical layer of OFDM-CR systems have been analyzed (e.g., mutual interference and synchronization) [2]. In OFDM-CR systems, spectrum partitioning is realized in the form of overlapping subbands, in which adjacent subcarriers are at the nulls of the sinc(f) function. Therefore, spectrum efficiency is high. However, the requirement of synchronization is also very strict, especially for frequency synchronization. If the subcarriers are not orthogonal, the sidelobes of the sinc-shaped spectrum on each subcarrier may fully interfere with PUs even if the parameter of the OFDM system used by PUs and SUs is the same. Moreover, the worse case is that OFDM is not adopt-

40

ed by PUs, or the parameters of OFDM for both PUs and SUs are different even though OFDM is used by PUs. To decrease the interference, the received OFDM signal is windowed in the time domain before it is fed into the operation of Fourier. Another method is to leave some virtual subcarriers (VCs) free. Furthermore, in an OFDM-CR system, the cyclic prefix (CP) or socalled guard interval is added to each transmitted symbol to avoid ISI, which occurs in multipath channels and destroys orthogonality. Unfortunately, like VCs, CP leads to a loss of transmission efficiency. The longer the delay spread of the channel, the longer the length of the CP, and thus the lesser the efficiency . Obviously, these methods conflict with the concept of improving the spectrum utilization of the CR technique. Compared with OFDM-CR systems, an FMTCR system does not require CP between several continuous transmitted symbols [4]. Instead, the bandwidth of each subcarrier is chosen to be quasi-orthogonal in the frequency domain, which is also called subcarrier spectral containment, and it can be achieved by the use of steep rolloff bandpass filters (i.e., filter bank). As a result, the time domain response of these filters may overlap in several successive transmitted symbol periods. Therefore, it is necessary for equalization per subchannel to reduce any remaining ISI, even if the channel is in an ideal state. A high level of subcarrier spectral containment is good for CR systems because the leakage of signal energy between adjacent subchannels may be neglected since it is as low as –70 dB where the subcarriers are closely spaced, as shown in Fig. 1. Figure 1 illustrates the frequency response of the first five subcarriers in an MCM system, where the solid and dashed lines denote FMT and OFDM, respectively. Due to the tight spectral containment achieved by the prototype filter in FMT, negligible power leaks into adjacent banks. Hence, fewer VCs are needed to comply with the regulatory power spectral mask than in an OFDM system. Therefore, it is obvious that only a few VCs are needed since the CP is not necessary for FMT systems; thus, the transmission efficiency of FMT systems is better than that of OFDM systems. Without loss of generality, we define the transmission efficiency as η=

Nc N c − N vc N c − N vc = × 100%, N c + LCP Nc N c + LCP

(1)

where N c , N vc , and L CP denote the number of subcarriers, the number of VCs, and the length of the CP, respectively. For example, when the bandwidth of occupied spectrum equals 20 MHz, N c = 64, LCP = 16, and N vc = 12 in an OFDM system based on the standard of HIPERLAN/2 or IEEE 802.11a, while LCP = 0 and Nvc = 2 ~ 4 in an FMT system. Accordingly, the efficiency of OFDM and FMT is ηOFDM = 65 percent and ηFMT ≈ 94 percent, respectively. Moreover, synchronization among different users is not serious in an FMT system because of its tight spectral containment. The disadvantage of the FMT is its complexity due to the filter bank and equalization per subchannel. However,

IEEE Wireless Communications • February 2011

JIANG LAYOUT

2/7/11

10:43 AM

Page 41

PUt

PUr

PUt

PUr

SUt1

SUr

SUt1

SUr

SUt2

SUt2

Between PUs and SUs

Among SUs

Figure 2. Proposed collaborative spectrum sensing model.

it may not be a serious issue in the future due to the fast development of digital signal processing techniques. In summary, taking mutual interference suppression, synchronization, and transmission efficiency into consideration, it is much better to adopt FMT than OFDM in MCM-CR systems.

PROPOSED COOPERATIVE DIVERSITY COGNITIVE MODELS IN AN MCM-CR SYSTEM When the idea of CR is taken into account in the cooperative communication system, the cooperative communication system with cognitive relay can be proposed to further improve the spectrum efficiency. On the other hand, cooperative diversity CR systems can also be proposed when the idea of cooperative diversity is considered in a CR system. Thus, based on Chasemi and Devroye’s model, combining MCM, cooperative diversity, and CR techniques, we propose three cooperative diversity cognitive models, including the collaborative spectrum sensing model, the cooperative communication cognitive model, and the hybrid model, as shown in Figs. 2, 3, and 4, respectively. Solid and dashed ellipses denote PUs in the primary network and SUs in the secondary network, respectively. Furthermore, dash-dotted lines denote the link to transmit the control data, including signaling and spectrum hole vector information. The solid and dashed lines denote the transmissions of the users’ self data and the partners’ data for cooperative retransmission, respectively.

COLLABORATIVE SPECTRUM SENSING MODEL In the process of spectrum sensing, the fading effect, noise, and interference may cause some errors (result in false alarm detection and false dismissal detection), which are very harmful because they may introduce serious interference to PUs, resulting in a decrease of the spectrum utilization ratio. The left model in Fig. 2, which is one of the proposed collaborative spectrum sensing models between PUs and SUs, means that a PU leaves or broadcasts the spectrum hole information to SUs over the air or through a wired channel. Obviously, the PU is the

IEEE Wireless Communications • February 2011

licensed user, who would not do this. However, if a PU would like to allow SUs access to his/her licensed spectrum when he/she was not using them, he/she should offer the spectrum hole information to help SUs. Certainly, this proposed collaborative spectrum sensing model is simple and introduces no any interferences to PU because spectrum hole information is very accurate in this case. Nevertheless, the disadvantage of this model is that it needs an additional specialized channel to transmit the spectrum hole information from PU to SUs, as well as the permission and authorization of the PU. The proposed collaborative spectrum sensing model among SUs is shown in the right part of Fig. 2. Collaborative means that SUs can exchange or cooperatively retransmit their SHV information with each other. For some reasons, such as deep fading or distance from PUs, some SUs (e.g., SUt2) may not detect the used spectrum successfully, which will cause a false dismissal detection. However, their neighbors (e.g., SU t1 ) can do them a favor. Therefore, the detection probability of SU t2 can be improved by cooperatively communicating with its adjacent SUs (e.g., SUt1) or exchanging SHV information with each other. SHV exchange and cooperative communication were first proposed by Chasemi [7] and Ganesan [8], respectively. However, only AF was considered in [8]. Obviously, there are still many key techniques of the proposed collaborative spectrum sensing model that need to be further studied, such as the spectrum sensing algorithm, the influence of the detection probability and locations of SUs, the choice of cooperative partner, and so forth.

The disadvantage of FMT is its complexity due to the filter bank and equalization per subchannel. However, it may not be a serious issue in the future due to the fast development of digital signal processing techniques.

COOPERATIVE COMMUNICATION COGNITIVE MODEL In this proposed model, we assume that the SHV has been detected successfully by SUs. The top part of Fig. 3 shows the cooperative communication between PUs and SUs. PUs (e.g., PU t and PU r ) are source/receiver, and SUs (e.g., SU t1 or SU t2 ) act as relays to retransmit PUs’ data. The rationale of this choice is that helping the PU to increase its throughput entails (for a fixed demand of rate by the PU) a diminished transmission time for the PU, which leads to more transmission opportunities for the SUs. That is to say, the PUs’ data can be transmitted

41

JIANG LAYOUT

2/7/11

10:43 AM

In the proposed hybrid model, which may be close to the systems easily to carry out, there are two steps needed: collaborative spectrum sensing and cooperative communication. Certainly, they can also be implemented simultaneously.

Page 42

PUt

PUr

SUt1

SUr

SUt2 Between PUs and SUs

PUr

PUt

SUt1

PUt

PUr

SUt1

SUr

SUt2

SUr2

SUr

SUt2 Half-duplex

Full-duplex Among SUs

Figure 3. Proposed cooperative communication cognitive model.

quickly, and some seconds are then saved for SUs to access the idle spectrum. The bottom part of Fig. 3 shows the cooperative cognitive model among SUs, where the SUs act as cooperative partners with each other. In such a case, SUs can work in either half-duplex (left bottom) or full-duplex (right bottom) mode. When it is in half-duplex mode, time-division multiple access (TDMA) is often adopted by SUs, and SUs as relays (e.g., SUt2 in the left bottom of Fig. 3) do not transmit data themselves. For full-duplex mode, CDMA is always selected by SUs, and SUs act as both source and relay (e.g., SUt1 and SUt2 in the right bottom of Fig. 3). Similar to the proposed collaborative spectrum sensing, there are also many key techniques that need to be studied, such as the transmitter scheme of cooperative communication, capacity and diversity gain analysis, power allocation algorithm, receiver designing, partner selection algorithm, and so on.

HYBRID MODEL As aforementioned, the proposed model of collaborative spectrum sensing is helpful for spectrum detection, while the proposed model of cooperative cognitive communication is mainly for data communication. However, spectrum detection and data communication will be as a whole in future practical CR systems. Therefore, the combination of the aforementioned two models (i.e., a hybrid model) is proposed in this subsection, which is shown in Fig. 4. Obviously, in the proposed hybrid model, there are two steps needed:

42

collaborative spectrum sensing and cooperative communication. Certainly, they can also be implemented simultaneously. Both collaborative spectrum sensing and cooperative communication are between PUs and SUs on the left of Fig. 4, while they are both among SUs on the right . Apparently, the same collaborative partners are not needed during the different steps. For example, collaborative spectrum sensing occurs between PUs and SUs during the first step, while cooperative communication is among SUs during the second step. Therefore, the probability of spectrum detection is very important. In the future, further research is needed on the hybrid model, including spectrum sensing, the influence of probability of spectrum detection, the selection of cooperative users, the synchronization and channel estimation algorithm, and so forth. Note that we need to point out that the MCM technology can be used in any of the three proposed models. Moreover, the space diversity gain can be obtained by adopting cooperative communication technology. Therefore, the introduction of cooperative diversity expands CR research from two dimensions (time-frequency) to three dimensions (space-time-frequency). In summary, all resources can be fully and effectively utilized to improve the performance of MCM-CR systems. Finally, it is necessary to point out that the PU’s model to occupy the channel is no limitation in this article. This is because we mainly focus on the model description and performance

IEEE Wireless Communications • February 2011

JIANG LAYOUT

2/7/11

10:43 AM

Page 43

PUt

PUr

PUt

PUr

SUt1

SUr

SUt1

SUr

SUt2 Cooperative spectrum sensing and communication between PUs and SUs

SUt2 Cooperative spectrum sensing and communication among SUs in half-duplex

Figure 4. Proposed hybrid model.

analysis, which results in the detection probability of the spectrum being the chief factor considered in the three proposed models, regardless of the PU’s model to occupy the channel.

100

10-1

SIMULATION RESULTS

PERFORMANCE LOSS WITH PROBABILITY OF SPECTRUM DETECTION Here, we adopt the half-duplex model seen in the right part of Fig. 4, in which SHV information is exchanged for collaborative spectrum sensing and cooperative communication among SUs. We assume that there is only one communication link between PUt and PUr, and several links among SUs. Obviously, the collaborative probability of spectrum detection Pd = 1 – (1 – Pds)n, where Pds (we have assumed that it is the same for all SUs) is the probability of spectrum detection by each SU without any help, and n is the number of SUs selected to join in SHV exchange for collaborative spectrum sensing. Then we further assume that only one SU (e.g., SU t2 ) is randomly selected to be a relay partner for cooperative communication after collaborative spectrum sensing. Based on these assumptions, the unconditional symbol error rate (SER) performance in an AF cooperation communication system is illustrated in Fig. 5, in which Pds = 0.9, and equal power allocation (EPA) is used. In Fig. 5, for comparison, the curve with perfect spectrum sensing (Pd = 1) is also plotted. Obviously, it can be concluded from Fig. 5 that there is about 4.0 dB loss without cooperation (n = 1 and P d = P ds = 0.9)

IEEE Wireless Communications • February 2011

10-2

SER

To get insight into the effectiveness of the proposed models and validate some related analytical results, extensive computer simulations have been conducted in which M-phase shift keying (MPSK) modulation is used, and channel coefficients hi,j(i ∈ {s, r}, j ∈ {r, d}) are independent samples of zero mean complex Gaussian random 2 variables with variance σ i,j , where s, r, and d denote the source, relay, and destination, respectively. Moreover, we assume σ 2s,d = 1, σ 2s,r = 1, σ2r,d = 10, and the total transmitted power P = P s + P r , where P s and P r represent the power allocated to source and relay, respectively.

n=1 n=2 n=3 Perfect sensing

10-3

10-4

10-5

10-6

0

5

10

15

20

25

30

P/N 0 /dB

Figure 5. SER performance using the half-duplex model in Fig. 4 with Pds = 0.9.

when SER is 10–3, and the SER performance is very close to the results with perfect sensing information when the number of collaborative SUs equals 3 (n = 3 and Pd = 0.99). That is to say, there is about 4.0 dB gain when two more SUs join for collaborative spectrum sensing. Obviously, the complexity of the system increases because of the cooperative partners’ joining.

POWER CONTROL SCHEME OF SUS UNDER THE CONSTRAINT OF INTERFERENCE TEMPERATURE In this subsection, under the constraint of the interference temperature, we discuss a CR system based on the cooperative cognitive halfduplex model on the bottom left of Fig. 3 in which SUt1, SUt2, and SUr are source, relay, and destination, respectively, and PU t is the PU. Without loss of generality, Q denotes the interference temperature level. Hence, the problem becomes how to obtain the optimal performance

43

Ergodic capacity (b/s/Hz)

JIANG LAYOUT

2/7/11

10:43 AM

Page 44

ACKNOWLEDGMENTS

5

4

3

2

-2

-1

-1.5

-0.5

0 Q/dB

0.5

1

1.5

2

Ergodic capacity

4 EPA OPAE

3 2 1

REFERENCES

0

[1] S. Haykin, “Cognitive Radio: Brain-empowered Wireless Communications,” IEEE JSAC, vol. 23, no. 2. Feb. 2003, pp. 201–20. [2] I. F. Akyildiz et al. “Next Generation Dynamic Spectrum Access Cognitive Radio Wireless Networks: A Survey,” Comp. Net., Sept. 2006, pp. 2127–59. [3] T. Luo et al., “A Subcarriers Allocation Scheme for Multi-carrier Modulation Based Cognitive Radio Systems,” IEEE Trans. Wireless Commun., vol. 7, no. 9, Sept. 2008, pp. 3335–40. [4] I. Berenguer, Filtered Multitone (FMT) Modulation for Broadband Fixed Wireless Systems, Master’s thesis, Univ. of Cambridge, 2002. [5] V. Chakravarthy et al., “Novel Overlay/Underlay Cognitive Radio Waveforms Using SD-SMSE Framework to Enhance Spectrum Efficiency — Part I: Theoretical Framework and Analysis in AWGN Channel,” IEEE Trans. Commun., vol. 57, no. 12, Dec. 2009, pp. 3794–3804. [6] A. Nosratinia et al., “Cooperative Communication in Wireless Networks,” IEEE Commun. Mag., vol. 42, no. 10, Oct. 2004, pp. 74–80. [7] A. Ghasemi et al., “Collaborative Spectrum Sensing for Opportunistic Access in Fading Environments,” Proc. IEEE DySPAN ‘05, 2005. [8] G. Ganesan and Y. Le, “Cooperative Spectrum Sensing in Cognitive Radio, Part I: Two User and Part II: Multiuser Networks,” IEEE Trans. Wireless Commun., vol. 6, no. 6, June 2007, pp. 2204–22. [9] O. Simeone and U. Spagnolini, “Cooperation and Cognitive Radio,” IEEE ICC ‘07, 2007, pp. 6511–15. [10] Y. Shi and Y. T. Hou, “A Distributed Optimization Algorithm for Multi-Hop Cognitive Radio Networks,” IEEE INFOCOM, 2008, pp. 1292–1300.

0

0.2

0.4

0.6

0.8

1

σ 2 s,p

Figure 6. Ergodic capacity with different Q (top) and σ2s,p (bottom), respectively. of the SUs’ network under constraint Q. It is well known that power control is one simple solutions. In order to obtain the maximal ergodic capacity, the approximate optimal power allocation scheme (OPAE) is proposed in the conducted simulations, and an AF cooperative communication protocol is selected among SUs. Furthermore, cooperative communication is obviously not adopted when the source-to-destination channel is better than the relay-to-destination one. Figure 6 illustrates the ergodic capacity with the changing of Q and the channel variances of the source-to-PU (σ2s,p), respectively, where σ 2s,p = 0.1 in the upper part and Q = –1 dB in the lower part. Obviously, it can be seen from Fig. 6 that the capacity performance of the proposed OPAE scheme outperforms that of the existing EPA, and the capacity performance monotonically increases with increasing Q and monotonically decreases with increasing σ2s,p.

CONCLUSIONS In this article, we have first studied the MCM techniques in a multihop MCM-CR system. Considering mutual interference elimination, synchronization, and transmission efficiency, we conclude that FMT is better than OFDM in an MCM-CR system. Second, we propose three cooperative diversity cognitive models: the collaborative spectrum sensing model, the cooperative communication cognitive model, and the hybrid model. The introduction of cooperative diversity expands CR research from time-frequency dimensions to space-time-frequency dimensions. Therefore, all resources can be fully and effectively utilized to improve the performance of an MCM-CR system. Finally, some simulations have been conducted to verify that the collaborative spectrum sensing can improve the probability of spectrum detection, resulting in enhancement of the performance of the MCM-CR system.

44

The work presented in this article was supported in part by the National Science Foundation of China with Grants 60872049, 60971082, 60972073, 60872008, 60702039, and 60972031; National Key Basic Research Program of China with Grant 2009CB320407; the National High Technology Development 863 Program of China under Grant 2009AA011803; the Program for New Century Excellent Talents in the University of China under Grant NCET-08-0217; the Research Fund for the Doctoral Program of Higher Education of the Ministry of Education of China with Grant 200804871142; National Great Science Specific Project with Grants 2009ZX03003-001, 2009ZX03003-011; and the SEU SKL project with Grant W200907.

BIOGRAPHIES T AO L UO [M‘09] ([email protected]) is now working as a professor at Beijing University of Posts and Telecommunications, and Key Laboratory of Universal Wireless Communications, Ministry of Education, P.R. China. His recent research interests are in the areas of wireless communication theory and technologies including OFDM, MIMO, WAVE, cooperative communication, and cognitive radio technologies. FEI LIN ([email protected]) is currently is currently an associate professor at Shandong Institute of Light Industry, Jinan, China. Her current research interests are in the areas of wireless communications, especially cooperative communications, cognitive wireless access, and MIMO. TAO JIANG [M‘06, SM’10] ([email protected]) is currently a full professor at the Wuhan National Laboratory for Optoelectronics, Department of Electronics and Information Engineering, Huazhong University of Science and Technology, Wuhan, P.R. China. He has authored or co-authored over 70 technical papers in major journals/conferences and five books/chapters in the areas of communications. His current research interests include the areas of wireless communications and corresponding signal processing, especially for cognitive wireless access, vehicular technology, OFDM, UWB and MIMO, cooperative networks, nano

IEEE Wireless Communications • February 2011

JIANG LAYOUT

2/7/11

10:43 AM

Page 45

networks, and wireless sensor networks. He has served or is serving as a member of the symposium Technical Program Committees of many major IEEE conferences, including INFOCOM and GLOBECOM. He was invited to serve as TPC Symposium Chair for the International Wireless Communications and Mobile Computing Conference 2010. He is serving as Associate Editor of some technical journals in communications, including EEE Communications Surveys & Tutorials and Wiley’s Wireless Communications and Mobile Computing He was a recipient of Best Paper Awards at IEEE CHINACOM ’09 and WCSP ’09. M OHSEN G UIZANI [S’83, M’90, SM’98, F‘09] (mguizani@ ieee.org) is currently a professor and the associate dean of academic affairs at Kuwait University. He is also an adjunct professor at Western Michigan University (WMU). He was chair of the CS Department at WMU from 2003 to2008 and chair of the CS Department at the University of West Florida from 1999 to 2003. He received his B.S. and M.S. degrees in electrical engineering; M.S. and Ph.D. degrees in computer engineering in 1984, 1986, 1987, and 1990, respectively, from Syracuse University, New York. His research interests include wireless communications and

IEEE Wireless Communications • February 2011

mobile computing, and optical networking. He is Editor-inChief of the Wireless Communications and Mobile Computing Journal (Wiley) and the Journal of Computer Systems, Networks, and Communications (Hindawi, Inc.). He is the author of seven books and more than 250 publications in refereed journals and conferences. He has guest edited a number of special issues in IEEE journals and magazines. He has also served as member, Chair, and General Chair of a number of conferences. He received both the Best Teaching and Excellence in Research Awards from the University of Missouri-Columbia in 1999. He is the past Chair of TAOS and current Chair of WTC IEEE Communications Society Technical Committees. He is a member of ASEE and senior member of the ACM. WEN CHEN ([email protected]) received his Ph.D. from the University of Electro-Communications, Tokyo, Japan, in 1999. In 2001 he joined the University of Alberta, Canada. Since 2006 he has been a full professor in the Department of Electronic Engineering, Shanghai Jiaotong University, China. His interests cover network coding, cooperative communications, cognitive radio, and MIMO-OFDM systems.

45

TALEB LAYOUT

2/7/11

10:44 AM

Page 46

ACCEPTED FROM OPEN CALL

CHALLENGES, OPPORTUNITIES, AND SOLUTIONS FOR CONVERGED SATELLITE AND TERRESTRIAL NETWORKS TARIK TALEB, NEC EUROPE LTD. YASSINE HADJADJ-AOUL, UNIVERSITY OF RENNES 1 TOUFIK AHMED, UNIVERSITY OF BORDEAUX 1

ABSTRACT

Service/content provider

2

The authors define some issues related to the interworking operation between the satellite and terrestrial domains, and suggest some solutions and discuss their potential.

46

The current trend in telecommunications services provisioning is shifting toward global ubiquitous networking and unified service architecture. Given the diversity of access technologies, this global ubiquitous networking cannot be possible without efficient interworking between the different access players. This leads to the necessity of defining, implementing, and deploying common services control architecture, able to support a wide variety of services for users in a variety of roles (consumer, producer, or manager of communication and media). This article defines some issues related to the interworking operation between the satellite and terrestrial domains. It suggests some solutions and discusses their potential.

INTRODUCTION Wireless communications continues to pervade all aspects of our lives — wireless distribution of audio and video around the home, wireless solutions for logistics, wireless ticketing and access control, wireless sensors for agriculture, medical applications, and so on. While many people appreciate the profound impact that wireless communications are having and will have on our lives, it will be some time before the vision of wireless everywhere will be realized — mainly because introducing large-scale changes to the way many systems work is complex, and requires significant time, effort, and energy. While there have been many important advances in wireless technology in recent years, there are economic challenges in providing high-speed wireless access to less populated areas. This gap between those who benefit from digital technology and those who do not is known as digital divide. A key technology that can help to bridge the digital divide is satellite communications as it can be used in areas where there is no terrestrial alternative. In the developed world, satellite net-

1536-1284/11/$25.00 © 2011 IEEE

works can be interworked with existing terrestrial networks, be they wireless or fixed systems, core or access network, and function as a high-speed backbone network to support a wide variety of services for users in a variety of roles. In this regard, there are many lessons to learn from recent mobile satellite experience. Indeed, in urban/suburban areas, fixed and mobile technologies (e.g., asymmetrick digital subscriber line [ADSL], Global System for Mobile Communications [GSM]) are well advanced. Satellites, performing in isolation, cannot compete with terrestrial systems in these urban areas. They can only provide niche services to areas inaccessible to terrestrial technologies. While these markets are politically important, they are small and bring poor revenue for satellite operators. The future of nextgeneration satellite systems is clearly in an integrated architecture with terrestrial systems. Their success also hinges on their ability to provide, in full cooperation with terrestrial systems, broadband data rate applications similar in spirit to today’s Internet. This is also beneficial for terrestrial system operators as it will enable them to increase the capacity of their systems, support large-scale deployment of different emerging bandwidth-intensive services, and satisfy the ever growing community of Internet users. Two critical issues arise when considering satellite systems in this context. First, satellite systems are very costly in general; second, there are challenges in integrating satellite and terrestrial networks, particularly when terminal mobility is necessary. This article will give some insight toward solving both of these problems. In this article we focus on interworking between the satellite part of the network and its terrestrial counterpart. Interworking related operations are performed at newly defined entities called interworking gateways (IGWs). The scope of this article is to define the modules of the technological solutions that will be incorpo-

IEEE Wireless Communications • February 2011

TALEB LAYOUT

2/7/11

10:44 AM

Page 47

Connec

tion B

Connection A

IGW1 IGW2

Receiver A

Sender Receiver B

Figure 1. RTT-based connection setup + connection decoupling. rated in IGWs and evaluate their performances via computer simulations. The remainder of this article is organized as follows. The next section portrays the key components of the envisioned architecture. We then describe our proposed context-aware complete end-to-end QoS approach devised for interworking between the satellite and terrestrial domains. The article concludes in the final section.

ENVISIONED NETWORK TOPOLOGY Although geostationary Earth orbit (GEO) systems are widely in use, and low/medium Earth orbit (LEO/MEO) will come onto the scene in the longer term, this article does not target any particular satellite constellation type. The developed solutions will be designed to be applicable to all constellation types (GEO, MEO, and LEO). The satellites are only assumed to be bidirectional interactive, acquiring onboard processing (OBP) capabilities and intersatellite links (ISLs) [1]. Terminals are interactive. Terminals outside the reach of the terrestrial network have direct access to the satellites. Terminals within reach of the ground Internet infrastructure have the ability to connect either directly to satellites or via the IGWs. The overall objective of this article is to define the necessary intelligence that should be added to IGWs to guarantee context-aware complete end-toend quality of service (QoS) for users. Thus, different levels of convergence are considered. The first level of convergence concerns efficient data transmission based on IP. The second level refers to the control and signaling for providing resource allocation and management. The third level of convergence deals with the provisioning of a generic service delivery platform based in IP Multimedia Subsystem (IMS). Finally, mobility management and seamless connectivity are considered for both network-link handover and link-layer handover.

IEEE Wireless Communications • February 2011

In light of the rapid globalization of the Internet and the resultant universality of IP, the data traffic load to be generated from the interworked satellite/terrestrial networks is expected to be all-IP as well. Investigating the interactions of IP protocols with the network is of vital importance.

CONTEXT-AWARE COMPLETE END-TO-END QOS APPROACH EFFICIENT DATA TRANSMISSION First, in light of the rapid globalization of the Internet and the resultant universality of IP, the data traffic load to be generated from the interworked satellite/terrestrial networks is expected to be all-IP as well. Investigating the interactions of IP protocols with the network is of vital importance. Satellite systems are well known for their unique characteristics — long propagation delays, large delay-bandwidth product, errors due to propagation corruption and handovers, and variable round-trip time (RTT) and link handovers. These features put limitation on the working of most transmission. With this regard, the authors have recently developed the Recursive, Explicit, and Fair Window Adjustment (REFWA) method to enhance the efficiency and fairness of TCP in satellite systems [2]. The use of the REFWA scheme has been extended further to the case of hybrid wired/wireless networks as well. While the REFWA scheme exhibits good performance, its performance remains limited in large bandwidth environments (such as satellite systems) due to its windowbased nature. Development of a new rate-based congestion control protocol that is specifically tailored for satellites and can exploit well the large delay bandwidth product feature of the satellite systems is required. REFWA can be a good candidate for that by changing its window-based feature to a rate-based one. Indeed, this is possible by having IGWs send data at rates exactly equal to the feedback value computed by REFWA. This is similar in spirit to the concept of the Explicit Control Protocol (XCP). Using such rate-based congestion control mechanism, and similar in spirit to the connection splitting in [3], the full end-to-end path will be decoupled into

47

Average window size (100 TCP flows)

TALEB LAYOUT

2/7/11

20 15 10 5 0 20 15 10 5 0 20 15 10 5 0

10:44 AM

Page 48

REFWA

0

50

100

150

200

250

300 XCP

0

50

100

150

200

250

300 TCP/Reno

0

50

100

150 Time (s)

200

250

300

Figure 2. The average window size for REFWA, TCP, and XCP.

separate segments — end terminal to IGW, a segment traversing the satellite network, and the final segment between the IGW and the remote end terminal (Fig. 1). The use of the protocol pertains to the segment traversing the satellites. The other two segments will employ control mechanisms optimized for their characteristics (wireless or wired). Furthermore, the necessary intelligence required for coordination between the used data transmission mechanisms will be added to IGWs to ensure reliable delivery of data while meeting the end-to-end QoS requirements. To illustrate the idea with more clarity, we have conducted simulations using NS-2. The window-based nature of REFWA is replaced by a rate-based one as explained earlier. The performance of the modified REFWA is compared against that of XCP and TCP as shown in Fig. 2. We tested the system under homogeneous traffic conditions using 100 heavy FTP sources during 300 s, a duration long enough to capture and study the behavior of our proposed transport protocol. At the beginning of the simulation, the hosts behind each satellite terminal are activated randomly following a uniform distribution ranging from 10 to 100 ms. Figure 2 shows the average window size for the three protocols. From the figure, it is clear that the average window size, when REFWA is in use, converges immediately to the optimal window value. In contrast, in case of XCP, it takes 15 s before the system reaches its optimal window. With TCP/Reno, the system oscillates around the optimal value without reaching a steady state. The simulation results also demonstrate the good performance of the modified REFWA as it achieves the highest goodput. Indeed, REFWA outperforms XCP and TCP/Reno by 6.93 and 20.90 percent, respectively, in terms of goodput. For the sake of further transmission efficiency and better QoS, short RTT connections should be established via the terrestrial wireless network. Indeed, terminals communicating with nearby users do not have to drain their energy to connect directly to satellites. Long RTT connections can be set via satellites. Different tech-

48

niques can be used for periodic monitoring of the network conditions. IGWs will be constantly updated with feedback on network dynamics. Based on this feedback and the RTT of connections, IGWs decide the path for communication — either via the satellites or via only the terrestrial network. For this purpose and in order to blur the separation between the satellite and terrestrial domains, there is need for exchange of state information (e.g., instant link loads) between the two domains. A hierarchical architecture of gateways can be considered. With this regard, the number of levels of this hierarchy, the size of each level, the amount of control traffic that should be exchanged and the length of the monitoring interval time should be decided in a way that enhances the accuracy in the assessment of network dynamics while minimizing the overhead in terms of signaling messages. Such a context-aware routing scheme will yield a better load balancing over the entire network and will enhance the end-to-end (E2E) QoS [4]. In the considered interworked satellite/terrestrial network, the IGW also provides the interface for any service/content provider who desires to provide service/content over both the terrestrial and satellite networks. If the provided data is bursty in nature (e.g., video data) and the targeted population of users is potential, it will be highly useful to send data from the provider to users using satellite channels. In such a communication scenario (Fig. 3), a service subscriber issues a request for a particular data/video title to the IGW via the terrestrial network (wireless or wired). The IGW informs the service provider of the request, and the latter allocates the necessary resources to satisfy the user’s request. To ensure reliable transmission of data (depending on the underlying transport protocol) the client keeps acknowledging successful receptions of data to the IGW via acknowledgment (ACK) packets sent over the terrestrial network. IF the ACK packets are delayed or lost, the overall network performance may be impacted. Adding intelligence in terms of an adequate delayed acknowledgment mechanism along with a robust error recovery mechanism to IGWs can help cope with these issues.

RESOURCE ALLOCATION AND MANAGEMENT In light of the limited resources of any powerful network, QoS can be maintained only via efficient resource management/allocation mechanisms. In the case of converged satellite/ terrestrial networks, devising an efficient resource allocation method is a highly challenging task due to the fluctuating nature of the wireless links. Indeed, in DVB-S2 networks, for instance, a novel satellite-tailored adaptive coding and modulation (ACM) technique is introduced to cope with the wireless channel fluctuations. ACM renders the resource reservation process even more difficult since the channel capacity changes frequently as the channel experiences noisy periods. While there are many approaches to solving resource management issues in networks, one that is suitable when there are limited and costly resources is connection admission control (CAC).

IEEE Wireless Communications • February 2011

2/7/11

10:44 AM

Page 49

IEEE Wireless Communications • February 2011

3

Service/content provider

2

A large library of CAC schemes has been proposed in the literature. These techniques can be classified as either resource-reservation-based or statistical-multiplexing-based schemes. Resource reservation CAC systems have some known scalability issues and may often lead to self-induced congestion due to the heavy resource reservation process. Besides, static reservation falls short of satisfying the flexibility requirements of typical network operators. Furthermore, statistical multiplexing CAC approaches cannot completely eliminate congestion during some peak noisy periods. However, they enable resource sharing between users and yield reduced waste of resources. This feature renders statistical multiplexing schemes more suitable for converged satellite/ terrestrial networks. Several research works have devised different CAC schemes to guarantee a reasonable QoS level under different network conditions [5]. A common shortcoming of these schemes resides in their inefficiency in dealing with the varying nature of the physical layer capacity of the satellite network. The authors’ research work presented in [6] takes into account the satellite channel fluctuations and presents an interesting CAC mechanism that also ensures fairness among terminals competing for the capacity of the same satellite channel. A shortcoming of the proposed approach lies in its lack of a bandwidth allocation mechanism and multiservice support. Indeed, CAC should be exerted in conjunction with a bandwidth allocation mechanism, especially in converged satellite/terrestrial networks where the link capacity may vary as a result of the ACM mechanism. In this case, combined action among various layers, a cross-layer approach, of the networks is likely to improve the performance of the overall system by protecting, for example, prioritized flows from packet drops during congestion events. In this area of research, there is a particular interest in the development of a cross-layer bandwidth allocation mechanism that can assist CAC and further enhance its functionality. For this purpose, we suggest using a channel prediction mechanism based on the least mean square (LMS) algorithm to tackle the excessive delay incurred by the feedback in satellite networks. Based on the proposed model, a self-configuring mechanism to cope with variable network conditions is derived. From this crosslayer approach, both optimized bandwidth allocation and guaranteed per-class QoS are expected. To illustrate the idea, we have conducted simulations using Opnet. We test the system under homogeneous traffic conditions using FTP sources during 1500 s. In the first phase of the simulation, the system is maintained free from noise. At the beginning of the second phase of the simulation (i.e., t = 245 s), a source of noise is introduced, which directly impacts the signalto-noise ratio (SNR), which decreases by approximately 2 dB. At the third phase of the simulation (i.e., t = 790 s), another source of noise is introduced. Finally, at the last phase, which starts at t = 1000 s, the sources of noise are eliminated successively at t = 1000 s and t =

3

TALEB LAYOUT

1 IGW

Client

Figure 3. Service delivery pattern.

1210 s. This performance evaluation scenario allows us to see how quickly and accurately our scheme adapts to both degrading and improving conditions. Figure 4 shows the instantaneous SNR experienced by a satellite terminal vs. the predicted SNR. In the first stage of the simulation, we observe that the estimation error is relatively important. This is principally due to the random initialization used in LMS. After 105 s, we clearly see that the estimation becomes more precise, which demonstrates the effectiveness of our proposed prediction mechanism. At this point, the accuracy of the proposed algorithm is approximately 98.5 percent. Figure 5 indicates the performance of the proposed mechanism in terms of throughput. The figure clearly shows that the bandwidth manager using predicted SNR values, which allow selecting the appropriate modulation and code rate, outperforms the conventional approach. As a consequence, the experienced throughput is increased by approximately 4.6 percent. The proposed cross-layer CAC mechanism, which relies on predicted SNR values, protects the network from congestion while maintaining a good trade-off between bandwidth utilization and end-to-end delay. These performances are particularly interesting for the provision of delay-sensitive and bandwidth-intensive applica-

49

TALEB LAYOUT

2/7/11

10:44 AM

Page 50

22

Predicted SINR Instantaneous SINR

20 18

SINR (dB)

16 14

32 APSK (3/4, 4/5, 5/6, 8/9, 9/10)

12 10

16 APSK (2/3, 3/4, 4/5, 5/6)

8

8 PSK (2/3, 3/4)

6 4 2 QPSK (1/4, 1/3, 2/5, 1/2, 3/5, 2/3, 3/4, 4/5, 5/6, 8/9, 9/10)

0 0

250

500

750

1000

1250

1500

Time (s)

Figure 4. The instantaneous SNR vs. the predicted SNR.

tions over converged satellite/terrestrial networks.

MOBILITY MANAGEMENT The success of any communication system hinges on its ability to provide acceptable QoS. In the context of mobile environments, QoS provisioning depends in turn on an efficient management strategy for mobility. In satellite networks, mobility management is a challenging task. Indeed, supporting continuous communication over satellite systems may require changing spot beams (and links in LEO/MEO systems) as well as the IP address of the communication endpoints. Thus, both link-layer and network-layer handovers are required for satellite networking. In case of non-geostationary (NGEO) systems, mobility management becomes more complex as both the satellite network and mobile users are on the move. In satellite networks, handovers can be broadly classified into two categories: network-link handover and link-layer handover. The former occurs when one of the communication endpoints changes its IP address due to motion of satellites or mobility of the user terminal. The latter occurs when one or more links between the end terminals change. It consists of satellite handover, ISL handover, and spot beam handover. Spot beam handover is the most common type of handover. They occur frequently due to the small area covered by spot beams and the mobility of users (or high speed of NGEO satellites). In this article, we initially focus on spot beam handovers and then extend the study to other handover types.

Spot Beam Handover Management — For better frequency utilization, the footprint of an individual satellite is divided into smaller cells, called spot beams. To ensure uninterrupted ongoing communications, a current communication link should be handed off to the next spot beam when needed. A spot beam handover involves the release of the communication link between the user and the current spot beam and acquiring a new link from the next spot beam to continue the communication. Due to the small area covered by spot beams, users’ mobility, and high

50

satellite speed in the case of NGEO systems, spot beam handovers are the most common type of handovers experienced in satellite systems. Efficient management of handovers is particularly linked to the resource allocation problem discussed above. Indeed, the selection of a suitable policy for channel allocation can ensure channel availability during handover. Thus, channel allocation strategies and handover guarantee are the prime issues in managing handover requests. It should be noted that it is more desirable to ensure and guarantee smooth ongoing calls rather than block a newly arriving call. To solve the spot beam handover problem, several handover schemes have been proposed in the recent literature. A thorough survey on these techniques is available in [7]. Sophisticated network planning is required to assign more capacity to spot beams when a high traffic rate is expected. Statistical methods, coupled with a user behavior model and precise predictions of satellite tracks relative to the Earth surface, allow general characterization of the traffic load for a particular satellite or spot beam. Via a cross-layer design, this can help in anticipating imminent handover events and locating the new point of attachment to the satellite network [8]. While a cross-layer optimization can be implemented at either end devices or intermediate nodes in the network, such as IGWs, it is relatively easier and more feasible to implement changes at mobile nodes. Indeed, at the communication endpoint, the physical layer of a mobile host instantly measures the radio strength or link quality. When the mobile node moves into the overlapping area of two or more spot beams, and different signals are consequently detected by the physical and data link layers, a warning message notifying of an imminent handoff event along with a list of the new possible spot beams are sent to the application layer. In case of multiple spot beams, the application layer refers to a set of tools to sort out the spot beam to which the mobile node is most likely going to be connected. Indeed, the application layer may use history on the user’s mobility pattern to predict the new spot beam. Referring to a spatial conceptual map, along with the user’s personal information, its current position, and its velocity heading, the application layer can make an accurate prediction of the most probable future spot beam. Prior contextual knowledge of the coverage area of the satellite network and the type of application can further increase the accuracy of the prediction. Once the next spot beam is determined, the mobile host informs the IGW of the next spot beam. Based on the current conditions (e.g., maximum number of free channels) of each spot beam, the IGW decides whether the call should be accepted or denied. If the handover cannot be made without degrading QoS of already existing users or causing network congestion, the IGW denies the handover request and sends an immediate negative acknowledgment to inform the mobile host that the request has been turned down. Simultaneously, a list of available spot beams can be sent along with the negative acknowledgment to induce the mobile host to hand over with another spot beam. The mecha-

IEEE Wireless Communications • February 2011

2/7/11

10:44 AM

Page 51

nisms by which IGWs admit or turn down handover requests (from a user) should be consistent with the underlying resource allocation strategy. An actual design and implementation of this context-aware cross-layer architecture at the mobile terminals and the intelligence required by IGWs to manage handovers define an interesting topic of research in this particular field of research.

Satellite and ISL Handover Management in NGEO Systems — The solution suggested above may also be highly interesting when it is put in the context of mobile satellite communication systems (e.g., LEO, MEO). In these systems, satellite handovers are more important as users need to first choose among different satellites and will then be served by the spot beam covering the user. In addition to the above solution, there is a need to develop complementary solutions that select the most suitable satellite for communications that can reduce bandwidth waste and call blocking probability, and also fulfill the QoS requirements. In regard to QoS, the application type should be taken into account in the satellite handover management strategy. Another important issue in the context of converged terrestrial/satellite networks is how to manage network layer handovers. While there are IP-based mobility solutions which have already reached the marketplace, they are unsuitable for converged satellite/terrestrial networks. This is mainly because they result in a very large amount of signaling traffic when employed in a satellite context, due to the constant and rapid motion of the satellites. Consequently, alternative network-layer handover mechanisms are required.

SEAMLESS CONNECTIVITY While in the above subsections the focus was on defining issues pertaining to congestion control, resource allocation, and mobility management in the converged satellite/terrestrial network, and devising possible solutions, in the remainder of this article we discuss possible scenarios for seamless use of the interworked satellite/terrestrial network and provide guidelines for their realization.

IMS-Based Service Delivery Architecture — A possible solution for service integration between satellite and terrestrial networks can be based on IMS, which represents a key element in the satellite architecture, supporting seamless and universal access to personalized services. Indeed, the adoption of IMS will favor the rapid emergence of new secure services and will enable seamless provisioning of multimedia services. IMS provides a service delivery platform (SDP) on top of convergence network technologies. This will help in generating new revenues, reducing the complexity of the IWG while significantly decreasing the cost of the satellite network management, which impacts directly the satellite services cost. Additionally, IMS allows more efficient handling of the multicast and broadcast traffic initiated from the satellite terminals. The rest of this section describes how to handle multicast services using IMS-based architecture.

IEEE Wireless Communications • February 2011

4.0E+06 3.5E+06 3.0E+06 Throughput (b/s)

TALEB LAYOUT

2.5E+06 2.0E+06 1.5E+06 1.0E+06 5.0E+05

Bandwidth manager using instant SNR Bandwidth manager using predicted SNR

0.0E+00 0

250

500

750 Time (s)

1000

1250

1500

Figure 5. Throughput using instantaneous vs. predicted SNR. Current mechanisms for delivering multicast services over satellite links use snooping (layer 2) or proxying (layer 3) to allow the delivery of IGMP/MLD membership messages to a satellite gateway over the air interface. A proxy and snooper do not change anything in the IGMP/ MLD messages but only forward the request further toward theIGW. In fact, over the satellite network, the broadcast property exists only on the forwarding link (i.e., the satellite return link provides only directional links). The host cannot listen to the signaling reports transmitted by other hosts on the return link. This leads them to individually send out their report, which generates excessive traffic. This would result in flooding and high latency problems. Flooding occurs when many hosts (i.e., IGMP/MLD clients) reply to a broadcast request from the IGMP/MLD Querier sent out by the router to sense the presence of clients in each multicast group. As highlighted earlier, unlike LAN, the satellite return link does not provide a broadcast property but only a unidirectional connection. Typically, hosts cannot listen directly to replies from other hosts. Thus, all the hosts have to respond to the IGMP/MLD Querier after the expiration of their timer. Moreover, satellite multicast groups can be very large and very dynamic. This leads to a waste of bandwidth and CPU over the satellite link and the Gateway respectively. Another important issue of IP multicast behavior over satellite networks is the latency in stopping transmission after the last host leaves a multicast group. The latency is the delay needed for the Querier to become aware that the multicast group is empty in order to stop multicast forwarding on it. The authors explain in [9] how to tackle the flooding and latency issues in providing multicast services. In this context, the IMS service delivery platform is used, which allows the IGMP messages to be aggregated and transmitted as a Session Initiation Protocol (SIP)based message for managing multicast groups. The argument for using SIP/IMS protocols is to allow hosts to join and leave a multicast group as IGMP does, verify the authentication and authorization of the user, signal any cryptographic context (e.g., using MIKEY), and easily support any future extension/augmentation that can be implemented in SIP.

51

TALEB LAYOUT

2/7/11

10:44 AM

A possible solution for service integration between satellite and terrestrial network can be based on the IP Multimedia Subsystem that represents a key element in the satellite architecture, supporting seamless and universal access to personalised services.

52

Page 52

CONCLUDING REMARKS This article highlights some of the opportunities behind integrated satellite and terrestrial networks. For the realization of such converged networks, it addresses a number of issues pertaining to transmission efficiency, resource allocation and management, mobility management, and seamless connectivity. While the main objective of this article is to highlight the related issues and define new directions for the community of satellite researchers, it also suggests a number of solutions, as seen from the networking perspectives, that, once they are put together, form a complete context-aware end-to-end QoS approach that solves the aforementioned issues.

REFERENCES [1] J. Farserotu and R. Prasad, “A Survey of Future Broadband Multimedia Satellite Systems, Issues and Trends,” IEEE Commun. Mag., vol. 38, no. 6, 2000. [2] T. Taleb, N. Kato, and Y. Nemoto, “REFWA: An Efficient and Fair Congestion Control Scheme for LEO Satellite Networks,” IEEE/ACM Trans. Net. J., vol. 14, no. 5, Oct. 2006. pp. 1031–44. [3] M. Marchese, M. Rossi, and G. Morabito, “PETRA: Performance Enhancing Transport Architecture for Satellite Communications,” IEEE JSAC, vol. 22, no. 2, Feb. 2004, pp. 320–22. [4] T. Taleb et al., “Explicit Load Balancing Technique for NGEO Satellite IP Networks with On-Board Processing Capability,” to appear, IEEE/ACM Trans. Net. J. [5] R. Abi Fadel and S. Tomhe, “Connection Admission Control and Comparison of Two Differentiated Resources Allocations Schemes in a Low Earth Orbit Satellite Constellation” ACM Wireless Net. J., vol. 10, no. 10, May 2004. [6] Y. H. Aoul and T. Taleb, “An Adaptive Fuzzy-Based CAC Scheme for Uplink and Downlink Congestion Control in Converged IP and DVB-S2 Networks,” IEEE Trans. Wireless Commun., to appear. [7] P. K. Chowdhury, M. Atiquzzaman, and W. Ivancic, “Handover Schemes in Satellite Networks: State-of-theArt and Future Research Directions,” IEEE Commun. Surveys & Tutorials, vol. 8, no. 4, 4th qtr. 2006. [8] T. Taleb et al., “A Cross-Layer Approach for an Efficient Delivery of TCP/RTP-Based Multimedia Applications in Heterogeneous Wireless Networks,” IEEE Trans. Vehic. Tech., vol. 57, no. 6, Nov. 2008, pp. 3801–14. [9] T. Ahmed et al. “IMS-based IP Multicast Service Delivery over Satellite Network” 8th IEEE PIMRC ‘07, Sept. 2007.

BIOGRAPHY TARIK TALEB [S‘04, M‘05, SM‘10] ([email protected]) is currently working as a senior researcher at NEC Europe Ltd, Heidelberg, Germany. Prior to his current position until March 2009, he worked as am assistant professor at the Graduate School of Information Sciences, Tohoku University, Japan. From October 2005 to March 2006 he worked as a research fellow with the Intelligent Cosmos Research Institute, Sendai, Japan. He received his B.E. degree in information engineering with distinction, and M.Sc. and Ph.D. degrees in information sciences from GSIS, Tohoku University, in 2001, 2003, and 2005, respectively. YASSINE HADJADJ ([email protected]) is an associate professor at the University of Rennes 1, France, where he is also a member of the IRISA Laboratory. He received a B.Sc. in computer engineering with high honors from SENIA University, Oran, Algeria, in 1999. He received his Master’s and Ph.D. degrees in computer science from the University of Versailles, France, in 2002 and 2007, respectively. He was an assistant professor at the University of Versailles from 2005 to 2007, where he was involved in several national and European projects such as NMS, IST-ATHENA, and IST-IMOSAN. He was also a post-doctoral fellow at the University of Lille 1 and a research fellow, under the EUFP6 EIF Marie Curie Action, at the National University of Dublin, where he was involved in the DOM’COM and IST-CARMEN projects, which aim at developing mixed Wi-Fi/WiMAX wireless mesh networks to support carrier grade services. His main research interests concern the fields of wireless networking, multimedia streaming, congestion control and QoS provisioning, and satellite communications. His work on multimedia and wireless communications has led to more than 25 technical papers in journals and international conference proceedings. TOUFIK AHMED ([email protected]) is a professor at Institut Polytechnique de Bordeaux (IPB) in the ENSEIRB-MATMECA School of Engineering. He is doing his research activities in CNRS LaBRI Lab, UMR 5800 at the University of Bordeaux 1. He received a B.Sc. in computer engineering with high honors from the National Institute of Computer Science, Algiers, Algeria, in 1999, and M.Sc. and Ph.D degrees in computer science from the University of Versailles, France, in 2000 and 2003, respectively. In November 2008 he obtained his Habilitation à Diriger des Recherches degree from the University of Bordeaux 1 on adaptive streaming and control of video QoS over wired/wireless IP networks and P2P architectures. He was a visiting scientist at the School of Computer Science of the University of Waterloo in 2002 and a research fellow at PRiSM laboratory of the University of Versailles until 2004. His main research activities concern QoS for multimedia wired and wireless networks, end-to-end signaling protocols, P2P networks, and wireless sensor networks. His work on QoS and video delivery has led to many publications in major journals and conferences.

IEEE Wireless Communications • February 2011

ZHENG LAYOUT

2/7/11

10:46 AM

Page 54

ACCEPTED FROM OPEN CALL

INTERFERENCE COORDINATION FOR OFDM-BASED MULTIHOP LTE-ADVANCED NETWORKS KAN ZHENG, BIN FAN, JIANHUA LIU, YICHENG LIN, AND WENBO WANG, BEIJING UNIVERSITY OF POSTS & TELECOMMUNICATIONS

ABSTRACT

Power value level -75dBm -90dBm -90dBm -60dBm -97dBm -90dBm -93dBm -87dBm RS5

4

RS4

R

RS9

After RS groupin

BS1

RS2

cto

Se

RS3

RS1

r3

ector 1 BS2 RS6

r2

cto

Se

RS7

The authors present an overview of the interference coordination strategies in OFDM-based multihop cellular networks. They propose several typical static or semi-static interference coordination schemes to improve coverage and increase the cell edge data rate.

54

Recently there has been an upsurge of interests in the multihop infrastructures for orthogonal frequency division multiplexing-based cellular networks in both academia and industry. In this article, we first present an overview of the interference coordination strategies in OFDM-based multihop cellular networks. Then, based on the framework of third-generation LTE-Advanced networks with multihop relaying, several typical static or semi-static interference coordination schemes are proposed to improve the coverage and increase the cell edge data rate. By applying these schemes, the radio resources can be reused with certain limitations on either the frequency or time domain, or even both of them. Dynamic system-level simulations are also carried out to demonstrate the effectiveness of the proposed interference coordination schemes.

INTRODUCTION The specification of the third-generation (3G) Long-Term Evolution (LTE) radio interface was recently finished by the Third Generation Partnership Project (3GPP) [1]. It aims to provide downlink/uplink peak rates of at least 100 Mb/s/50 Mb/s and round-trip times of less than 10 ms. The first commercial LTE deployment took place in Stockholm, Sweden, and Oslo, Norway, in December 2009. However, the LTE system cannot meet the requirements of future broadband wireless networks, which is officially called International Mobile Telecommunications (IMT)-Advanced by the International Telecommunications Union — Radiocommunication Standardization Sector (ITU-R). The IMTAdvanced system is expected to support enhanced peak data rates on the order of 100 Mb/s for high-mobility and 1 Gb/s for low-mobility environments, respectively [2]. Also, it is able to provide a high degree of commonality of functionality worldwide while retaining the flexibility to support a wide range of services and applications in a cost-efficient manner. In order

1536-1284/11/$25.00 © 2011 IEEE

to meet these new challenges, 3GPP has started to develop further advancements for 3G LTE systems, referred to as LTE-Advanced, as a candidate for IMT-Advanced. In order to meet the requirements of IMTAdvanced, more spectrum bands are needed. Besides the existing spectrum for 3G mobile communication systems, spectrum bands located at 450–470 MHz, 698–790 MHz, 2.3–2.4 GHz, and 3.4–3.6 GHz have also been identified for 3G and IMT-Advanced systems by the ITU during World Radio Conference 2007 (WRC ’07) [3]. Most of them are above the 2 GHz band, where the radio propagation is more vulnerable to non-favorable channel conditions. With traditional cellular architectures, the density of base stations (BSs) has to be significantly increased to meet service coverage requirements, offering high data rates at these high-frequency bands. Obviously, this is not a favorable method since it would greatly increase deployment costs. Instead, a cost-effective solution would be the multihop cellular architecture with relaying, which shortens the transmission distance and increases the amount of users under good channel conditions, thus allowing for higher throughput. Recently, standardization efforts of integrating cooperative relaying technologies into LTE-Advanced networks have commenced [4]. In LTE and LTE-Advanced networks, orthogonal frequency-division multiplexing (OFDM) has been chosen as the multiple access method since it can provide high data rates and spectrum efficiency. In OFDM-based systems, users are multiplexed in time and frequency by means of a scheduler that dynamically assigns subcarriers to different users at different time instances according to predefined scheduling metrics. Therefore, the OFDM-based multihop transmission by means of relay stations (RSs) has been recognized as an efficient technique to meet the requirements of future broadband wireless networks. The RSs have the capability of forwarding the traffic between the base station (BS) and the mobile stations (MSs). The main objective of introducing multihop relaying technology into

IEEE Wireless Communications • February 2011

ZHENG LAYOUT

2/7/11

10:46 AM

Page 55

OFDM-based cellular networks is to achieve both higher throughput and better service coverage with the assistance of cost-effective relay architecture. Preliminary study on cooperative relaying technology indicates that multihop relaying offers certain performance advantages such as coverage extension and capacity improvement [5]. However, numerous challenges at the physical (PHY) and medium access control (MAC) layers in OFDM-based multihop cellular networks still remain [6]. For example, time and frequency resources are typically reused in multiple cells, thus leading to co-channel interference impairments among the coverage of neighboring BSs and/or RSs. Such co-channel interference between cells plays an important role in affecting the performance of OFDM-based multihop cellular networks. The impact of interference is more obvious for cell edge users, who are more susceptible due to poor channel gains with their serving BS or RS. Limited reception caused by interference at the cell edge is an issue of great importance for wireless operators who want to provide full coverage within their service areas and guarantee a prior agreed quality of service (QoS) to their subscribers. To the best of the authors’ knowledge, there have been few works in the literature on specific interference coordination schemes in OFDM-based multihop cellular networks. The scope of this article is hence to examine how users can share the available radio resources efficiently, in terms of bandwidth and time allocation, in order to mitigate intercell co-channel interference and thus enhance user throughput, especially for cell edge users. We first briefly introduce the state-of-the-art interference mitigation schemes in OFDM-based cellular networks. Then, with the introduction of the system framework based on 3G LTE specifications, several static or semi-static interference coordination schemes are proposed. We also analyze and discuss their performance extensively through simulations.

STATE-OF-THE-ART In OFDM-based cellular networks, interference coordination strategies have been studied in order to increase achievable reuse of the scarce spectrum with reasonable complexity and overhead. An exhaustive exposure of the state of the art is outside the scope of this article. Interference coordination aims at applying restrictions to radio resource management in a coordinated way among cells. These restrictions can be either on the available radio resources or in the form of restrictions on the transmit power that can be applied to certain radio resources. Such restrictions provide the possibility for improvement in signal-to-interference-plus-noise ratio (SINR), and consequently to the cell edge throughput and coverage. Interference coordination also requires communication between different nodes in the network to (re)configure resource restrictions. Based on the requirement of the inter-BS communication interval, most of the existing and currently in development interference coordination strategies can be categorized into three types [5]:

IEEE Wireless Communications • February 2011

Static coordination: Internode communication is very limited since it corresponds to the setup of restrictions only. Reconfiguration of the resource allocation restrictions among nodes is done on a timescale of days. Semi-static coordination: The corresponding signaling rate of internode communication is generally on the order of tens of seconds to minutes. Reconfiguration of the restrictions is done on a timescale of seconds or longer. Dynamic interference coordination: It requires much internode communication to exploit multiuser diversity among neighboring cells with high computational complexity. In this case, internode signaling or data transferring may be needed at each scheduling instant. In the LTE standardization process, many appealing and feasible interference coordination algorithms were extensively studied for OFDMbased networks. Typical interference coordination strategies, such as soft frequency reuse (SFR) and fractional frequency reuse (FFR), utilize the resources of frequency and radiated power to coordinate BS transmissions with predefined resource constraints for different types of users as follows: Soft frequency reuse: The whole available bandwidth is divided into multiple non-overlapping subbands. Each cell selects one subband as its major band and the others as its minor bands. Major bands can be used in the whole cell area with full transmit power while minor bands are only in the inner zone of the cell with reduced transmit power. The performance of the cell edge user (CEU) can be improved by using the major bands to mitigate intercell interference. Also, the high data rate of the cell center user (CCU) can be achieved since both major and minor bands are available for its transmission [7]. Fractional frequency reuse: It splits the given bandwidth into inner and outer subbands. The inner subbands are completely reused by all cells, while the outer subbands are divided among neighboring cells with a frequency reuse factor greater than one. Intercell interference is reduced at the cell edge by assigning resources on outer subbands to CEUs [8]. With the development of LTE-Advanced, multihop relaying techniques have been introduced into the cellular network. In a multihop cellular network, an RS usually transmits the same or a different format of the information as that received from the BS or MS, which is likely to be regarded as a certain kind of repetition. So the capacity of this relaying network is decreased from the system point of view. Therefore, it is quite necessary to design an efficient resource allocation scheme in a multihop cellular network, which turns to high radio resource reuse among RSs and BSs. However, compared with traditional cellular networks, more complicated and serious interference exists in an OFDMbased multihop cellular network. The downlink co-channel interference in such a network can be classified as: Intercell interference: The co-channel interference introduced by the frequency reuse between multiple cells; that is, interference from BS → RS links, BS → MS links, and RS → MS links in the neighboring cells, respectively.

Interference coordination aims at applying restrictions to the radio resource management in a coordinated way among cells. These restrictions can be either on the available radio resources or in the form of restrictions on the transmit power that can be applied to certain radio resources.

55

ZHENG LAYOUT

2/7/11

10:46 AM

Most interference coordination schemes for OFDM-based multihop cellular networks are only the extensions of existing coordination strategies used in traditional cellular networks. The opportunities of multihop relaying transmissions in dealing with co-channel interference problems have not been fully exploited.

Page 56

Intracell interference: The co-channel interference induced by resource reuse within the same cell; that is, interference from BS → MS and the RS → MS links in the same cell, respectively. Similar to a traditional OFDM cellular network, there are also three types of interference coordination strategies to deal with co-channel interference in OFDM-based multihop cellular networks: static, semi-static, and dynamic schemes. With static/semi-static interference coordination schemes, both inter- and intracell interference can be diminished by allocating resources under the constraint of the predefined resource coordination pattern. Also, it is critical to exploit the characteristics of multihop transmission to obtain a well-designed coordination pattern that can achieve throughput gain for all users in the cell. For dynamic coordination schemes, the power and resource allocation is dynamically coordinated among neighboring cells at each transmission time [5]. Optimal intercell resource coordination and large multiuser diversity gain can be obtained by centralized control or non-cooperative competition gaming among the cells. However, due to its huge signaling overhead and high complexity, such a dynamic scheme is not practical in current OFDM-based multihop cellular networks. As of today, most available interference coordination schemes for OFDM-based multihop cellular networks are only extensions of existing coordination strategies used in traditional cellular networks. The particularities and opportunities of multihop relaying transmissions in dealing with co-channel interference problems, however, have not been fully exploited yet.

FRAMEWORK OF A MULTIHOP RELAYING NETWORK In this section we briefly introduce the underlying multihop relaying framework for downlink transmission. As depicted in Fig. 1a, there are usually three types of links involved in end-toend communication in multihop relaying cellular networks: the link from BS to RS (BS → RS), the link from RS to MS (RS → MS), and the link from BS to MS (BS → MS). Furthermore, for the sake of clarity, we refer to the BS → RS link as the relay link, while both the RS → MS and BS → MS links are called access links. Relaying should happen only when it can improve the end-to-end throughput. For the sake of description, the time-divisionduplex (TDD) frame structure defined in LTE is used as an example to enable relaying technology in cellular networks. As shown in Fig. 1b, each radio frame of length Tf = 10 ms consists of two half-frames with length Thalf = 5 ms each. Usually, each half-frame includes four common subframes of length Tsub = 1 ms and three special fields: downlink pilot time slot (DwPTS), uplink pilot time slot (UpPTS), and guard period (GP). Each subframe comprises two slots with length T slot = 0.5 ms. For more detailed information on this frame structure, please refer to the 3GPP specifications [1]. In practical implementation, the half-duplex

56

RS is usually assumed to be deployed in LTEAdvanced networks, where the transmission and reception take place in different subframes. When multihop relaying happens, the complete two-way transmission over the air has four communication phases (i.e., BS → RS, RS → MS on the downlink [DL], and MS → RS, RS → BS on the uplink [UL]). In each phase one subframe is used for transmission; the basic transmission granularity in the time domain is one subframe, consisting of two successive time slots. Based on this framework, we present several advanced interference coordination schemes for multihop cellular networks in the next section, which can be also applied in frequency-division duplex (FDD) systems.

ADVANCED INTERFERENCE COORDINATION SCHEMES Different from traditional cellular networks, cochannel interference occurs not only between neighboring BSs or RSs but also between nearby BSs and RSs involved in multihop transmission. In order to mitigate these interferences, suitable radio resource allocation and scheduling mechanisms in such a multihop cellular network become vital. In this section we propose several static or semi-static interference coordination schemes for downlink transmission. In general, the interference between neighboring RSs or BSs is measured when the network is initialized, or the user distribution and services slowly vary. Then, according to measurement results, the RSs that do not cause severe interference to each other are grouped together and reuse the same radio resources. The restrictions on radio resource usage for BSs and RSs can be carried on along the frequency or time domain (i.e., one-dimensional), or both of them (i.e., two-dimensional) in an OFDMbased multihop cellular network. Then, with these restrictions, each BS or RS can schedule its serving users without explicit internode communication.

ONE-DIMENSIONAL INTERFERENCE COORDINATION SCHEMES In order to illustrate 1D interference coordination schemes clearly, we show a typical deployment scenario in Fig. 2a, where each cell is partitioned into three sectors and a fixed RS is put on the bore sight line of the directional antenna in each sector. First, every neighboring BS or RS takes turns transmitting the reference signal while others keep silent and measure the received signals during the measurement period. The received signal power from RSi at RSj (i ≠ j) is denoted P ji. Then each RS reports the measured Pji to its anchor BS. Next, an interference measurement table is formed at the anchor BS, which can be used as the guide for RS grouping. Define the interference power threshold Pth as the maximum value of interference power an RS can tolerate with acceptable communication. The simple grouping criterion is that RS i and RSj can be selected into the same resource reuse set only when Pji < Pth. For example, the inter-

IEEE Wireless Communications • February 2011

ZHENG LAYOUT

2/7/11

10:46 AM

Page 57

BS

MS

RS

MS

Access link

Relay link

Access link

Single hop

First hop

Second hop

End-to-end connection

End-to-end connection (a)

One radio frame One half-frame

One slot Subframe #0 One subframe DwPTS

Subframe #2 Subframe #3 Subframe #4 Subframe #5

DwPTS

GP UpPTS

Subframe #7 Subframe #8 Subframe #9

GP UpPTS

(b)

Figure 1. Illustration of the framework in a multihop cellular network: a) illustration of the different links in a multihop cellular network; b) frame structure.

ference power threshold Pth = –80 dBm is predefined in this section. According to measurement results in Fig. 2a, the nine RSs concerned are grouped into three RS reuse sets (i.e., G1 = {RS1, RS7, RS8}, G2 = {RS2, RS5, RS6}, and G3 = {RS 3 , RS 4 , RS 9 }). Note that every two sets have a null intersection (i.e., Gm ∩ Gn = ∅, if m ≠ n). In the resource partitioning stage, all RSs in the same reuse set can be allocated to the radio resources with the same pattern. Therefore, in the remaining parts of this article, we only concentrate on resource coordination among the RSs in the different groups (e.g., RS1 ∈ G1, RS2 ∈ G2, and RS3 ∈ G3).

Frequency Domain Interference Coordination Scheme — In this scheme different priorities of the radio resources for different users are defined in the frequency domain before resource allocation. At each BS or RS, the radio resources with high priority are first assigned to its remote users that experience strong interference with high probability. On these high-priority radio resources, the signals can be transmitted with full radiated power. If the data amount of the service demands exceeds the throughput the BS or RS can provide only with high-priority radio resources, other radio resources with low priority have to be allocated under the constraint of lower radiated power. In Fig. 2b we give an example of high-priority radio resource allocation for three RS groups, which is based on the measurement results in

IEEE Wireless Communications • February 2011

Fig. 2a. In this frequency domain (FD) interference coordination scheme, every two successive subframes are allocated for BS and RS transmission, respectively. In the first subframe a part of radio resources along the frequency domain is orthogonally allocated to the relay links (BS → RS) of the neighboring sectors. Meanwhile, since most single-hop users are usually close to the BS, they are insensitive to intercell interference. The other radio resources are shared by the single-hop transmissions between BS and MS in the neighboring sectors. The radio resources are not evenly divided among the BS → RS link and the BS → MS link, which is dependent on the ratio of the single-hop user number and two-hop user number. In the second subframe the RSs transmit to their serving MSs, the remote MS are first allocated high-priority radio resources, which are orthogonal to each other in the frequency domain in order to eliminate strong co-channel interference. Then, for those users near their serving RSs, low-priority radio resources can be used for transmission with lower radiated power without much performance loss. Usually, a lineof-sight (LOS) channel is assumed in the BS → RS link, and a non-line-of-sight (NLOS) in the RS → MS link. The quality of radio channels on the BS → RS link is much better than that on the RS → MS link. Hence, the higher modulation and coding scheme (MCS) for data transmission can be applied in BS → RS links. Consequently, the throughput balance between the BS → RS link and the RS → MS link can be

57

ZHENG LAYOUT

2/7/11

Received signal from RS RS1 RS2 RS3 RS5 RS6 RS7 RS8 RS9

10:46 AM

Power value level -75dBm -90dBm -90dBm -60dBm -97dBm -90dBm -93dBm -87dBm

RS4

RS5

Page 58

RS4

RS5

RS1

RS3 RS7

RS9

RS8

After RS grouping

BS1

RS2 r2 cto

Se

RS9

Se

RS3

cto r3

Sector 1 RS1

RS6

RS2

BS2 RS6

RS7

BS3 RS9

Received signal from RS RS1 RS2 RS3 RS4 RS5 RS6 RS7 RS9

Radio frame i Downlink Subframe #9

Power value level -80dBm -60dBm -90dBm -70dBm -82dBm -90dBm -90dBm -76dBm

Radio frame i+1 Uplink

Downlink Subframe #0

Subframe #2

Subframe #3

Frequency Frequency

RS3

BS → RS1 RS1 → MS

RS1

BS → MS

Time

1 subframe

1 subframe

Frequency

RS2

BS → RS2 BS → MS 1 subframe

(a)

BS → RS3 BS → MS RS3 → MS 1 subframe

1 subframe

Time

RS2 → MS 1 subframe

Time

(b) Radio frame i Downlink

Radio frame i+1 Uplink

Downlink

Subframe #8 Subframe #9 Subframe #0

RS1 RS1-MS 1 subframe

BS-MS

BS-RS1

Subframe #2 Subframe #3

RS3

1 subframe

RS2

1 1 subframe subframe

BS-MS

BS-RS2 1 subframe

RS2-MS

BS-RS1

RS1-MS

1 1 subframe subframe

BS-MS

1 1 subframe subframe

(c) Start

Start Coordination in the first frame

Resource allocation • Allocate orthogonal frequency resources to BSs in neighboring sectors for BS → RS link. • Allocate other resources to BSs for BS-MS link.

Resource allocation • Group each three successive subframes together as a basic coordination unit.

Power allocation • Full radiated power for the BS → RS link • Resources are used by its neighboring BSs for BS → RS link?

Yes Half radiated power for BS → MS link.

• In each sector, assign each subframe to one type of link, (i.e. BS → RS, BS → MS, or RS → MS link. • Let same type of links in the neighboring sectors be in different subframes

No Full radiated power for BS → MS link.

Coordination in the second frame Resource allocation • RSs in different groups allocate orthogonal frequency resources to their CEUs as high-priority resources • Each RS uses the remained frequency resources as its low-priority resources

Power coordination • Reduce the radiated power of the BS in the subframes for BS → MS link. • Use beamforming in the BS → RS link (optional)

Power allocation • Full radiated power on high-priority resources • Half radiated power on low-priority resources

End (d)

End (e)

Figure 2. Illustration of 1D interference coordination schemes in OFDM-based multihop cellular networks: a) deployment of one RS per sector; b) resource allocation by the frequency domain IC scheme; c) resource allocation by the time domain IC scheme; d) flowchart of the frequency domain IC scheme; e) flowchart of the time domain IC scheme.

58

IEEE Wireless Communications • February 2011

ZHENG LAYOUT

2/7/11

10:46 AM

Page 59

achieved with the suitable design. Figure 2d gives the flowchart of the FD interference coordination scheme.

Time-Domain Interference Coordination Scheme — Multihop relaying transmissions in the downlink usually consist of two phases in sequence (i.e., first from BS to RS and then from RS to MS). Considering this characteristic, the subframes allocated to different transmission phases can be coordinated among cells in the time domain. According to the channel path loss model defined in [9], the strongest interference to the remote users served by an RS mostly comes from their neighboring RSs. Thus, it is advisable to ensure that the resources for RS → MS links of the neighboring cells are kept orthogonal in the time domain to mitigate the interference. The time domain interference coordination scheme is proposed and described as follows. In the time domain, different subframes can be used by one of three types of links (i.e., BS → MS, BS → RS, and RS → MS). Figure 2c shows an example of resource allocation pattern for three neighboring sectors, where BS → MS represents the single-hop link from BS to MS; BS → RSi, i ∈ {1, 2, 3}, is the relay link of the twohop transmission; and the access link of the twohop transmission is denoted RSi → MS, i ∈ {1, 2, 3}. Different subframes are allocated to different types of links within one sector, while those of the same type of links among the neighboring sector are not overlapped in the time domain. Thus, every three successive subframes are used together as a unit for the coordination among these three links by the time domain interference coordination scheme. The flowchart of the time domain interference coordination scheme is also shown in Fig. 2e. By this allocation method, the interference from neighboring BSs becomes the dominant factor, affecting the performance of RS → MS links in the anchor cell. To overcome this problem, beamforming can be applied at the BSs for transmission of the relay link, which concentrates the transmit power in the direction of a particular user and minimizes the interference from BS → RS links in the neighboring cell to RS → MS links in the anchor cell. Furthermore, a neighboring BS can reduce its transmit power in BS → MS links for single-hop users located in the cell center, thus also mitigating the interference to RS → MS links in the anchor cell. Meanwhile, the interference on other links such as BS → MS and BS → RS can also be decreased. In addition, the ratio of radio resources between single- and two-hop links can be varied with the distribution of different types of users. Performance Comparison — The simulations are carried out to demonstrate the performances of the priority-based interference coordination (IC) schemes. Most simulation assumptions follow the evaluation methodology as in [9, 10] and are summarized in Table 1. The scenario with intersite distance (ISD) of 1500 m is considered, where the operating bandwidth of 10 MHz is located at the central frequency of 2 GHz for downlink transmission. A penetration loss of 10 dB is assumed for the access links (i.e., BS →

IEEE Wireless Communications • February 2011

MS and RS → MS). For simplicity, only a singleantenna configuration is assumed for each node. Statistics are collected from a total of 20 drops in the simulations. In each drop, users are uniformly distributed around each BS with a density of 90 users/cell. The simulation time span is 50 s (1000 radio frames) in each drop. The ideal hexagonal cell is assumed for each BS, and two tiers of cells are considered with respect to one reference cell in the center (i.e., a total of 19 hexagonal cells). Moreover, each cell is partitioned into three 120˚ sectors, where one 120˚ directional antenna for each sector is applied at the BS. For each sector, only one RS is deployed on the bore sight line of the directional antenna of the BS. The distance between the RS and its anchor BS is two thirds of the cell radius. According to the received SINR, all users can be classified into two types: CEU and CCU. The ratio between the number of CEUs and CCUs in each cell is set to be 1:2. If the frequency domain (FD) IC scheme is applied, the signal is transmitted on the high-priority resources with full radiated power, while only half of the full radiated power is allowed on the low-priority ones. With the time domain (TD) IC scheme, the radiated power ratio between each BS and RS in the networks is set to be 2:1 for fair comparison. We also provide the performance of the network without the IC scheme for comparison. The full-buffer traffic model is first assumed for simulations, in which there are infinite data packets in the queues. Figure 3a shows the cumulative distribution function (CDF) of the received SINR per cell in the network with/without 1D IC schemes. Since the interference power to the high-priority resources used by the CEUs is decreased by using the FD IC scheme, the SINR performance of CEUs can be improved. However, for the CCUs in the network with the FD IC scheme, the corresponding SINR values are decreased because the radiated power on the low-priority resources is reduced. Moreover, it can be seen that the SINR values of all users in the network with the FD IC scheme are limited to be no more than 16 dB. This is because the performance of the two-hop users is restricted by the quality of the relay link (BS → RS), whose SINR is no larger than 16 dB due to the intersector interference. Different from the FD IC scheme, the SINR performance is apparently improved for all users including CEUs and CCUs by using the TD IC scheme. Meanwhile, parts of interference to the relay link (BS → RS) come from the neighboring RSs with lower radiated power instead of the BSs by using the TD IC scheme. Therefore, the highest SINR value of the two-hop users, restricted by the relay link, is increased to around 18 dB. The comparison of per user throughput performance in the networks with/without 1D IC schemes is also given in Fig. 3b. Similar to the SINR performance, the FD IC scheme can achieve significant throughput improvement for the CEUs with little performance degradation of the CCUs. In the network with the TD IC scheme, the throughput is increased for both CEUs and CCUs. Furthermore, the aggregated cell throughput of the networks without or with 1D interference coordination schemes is given in Fig. 3c. Com-

Multihop relaying transmissions in the downlink usually consist of two phases in sequence (first from BS to RS and then from RS to MS). Considering this characteristic, the subframes allocated to different transmission phases can be coordinated among cells in the time domain.

59

ZHENG LAYOUT

2/7/11

10:46 AM

Page 60

Parameters

Values

Intersite distance (ISD)

1500 m

Bandwidth/carrier frequency

10 MHz/2 GHz

Antenna configuration (BS/RS/MS)

1/1/1

BS-MS

128.1 + 37.6 log10(R)

BS-RS LOS/NLOS

Type D LOS in [9]/128.1 + 37.6 log10(R), R in km

RS-MS LOS/NLOS

Type C/Type A in [9]

BS

46 dBm/32 m

RS

37 dBm/12.5 m

Pathloss

Transmit power/height

IC scheme, with 7.5 and 15.1 percent relative gain with the FD IC and TD IC scheme, respectively.

TWO-DIMENSIONAL INTERFERENCE COORDINATION SCHEMES For network deployment with more than one RS per sector, the coordination scheme operating solely in the FD degrades the potential frequency diversity gain, while either too small time granularity or a group with too many subframes is needed for the scheme operating only in the TD. It is therefore necessary to develop a 2D time-frequency interference coordination scheme with more flexibility for an OFDM-based multihop cellular network. Assume that there are two RSs deployed per sector as shown in Fig. 4a. According to the interference measurement results, all RSs are grouped into the following six sets (i.e., G1 = {RS 1, RS 15, RS 17}, G 2 = {RS 2, RS 16, RS 18}, G 3 = {RS 3, RS 12, RS 14}, G4 = {RS 4, RS 11, RS 13}, G5 = {RS5, RS8, RS9}, G6 = {RS6, RS7, RS10}). Recall that the RSs in the same set can reuse the same radio resources. The intersector cochannel interference can be mitigated by means of the TD coordination scheme, while the intrasector interference is dealt with by the FD coordination scheme. In Fig. 4b, for TD resource allocation, all RSs in six sets can also be combined into three pairs: P 1 = {RS 1 , RS 2 }, P 2 = {RS3, RS4}, and P3 = {RS5, RS6}. Then different kinds of links including RS → MS, BS → RS, and BS → MS in each paired RS set are distinguished in the TD within one sector. Moreover, the same kind of links between the neighboring sectors are allocated to different subframes without overlap in the TD. Then further radio resource allocation within one sector among different RS sets is performed in the FD. The total frequency bandwidth for each RS set is partitioned into two orthogonal bands,high-priority and low-priority. The highpriority bands of two different RS sets in the same sector are orthogonal. For each RS, it is necessary to first allocate the radio resources in the high-priority band to the remote MSs with full radiated power, while those in the low-priority band with lower radiated power can be used by the MSs that are close to the RS. In this way the intrasector co-channel interference, which comes from the different reuse sets, can be decreased efficiently. Also, we show the flowchart of the proposed 2D interference coordination scheme in Fig. 4c. Note that this scheme can be extended to cases with larger numbers of RSs. ~

~

Penetration loss

10 dB

~

Thermal noise spectral density

–174 dBm/Hz

Noise figure

5 dB

Traffic model

Full buffer

Table 1. Parameter assumptions in a multihop cellular network. pared to the network without the IC scheme, a large throughput gain for the CEUs (i.e., 15.9 percent) can be obtained with the FD IC scheme. However, it decreases the aggregated cell throughput of the CCUs. On the other hand, the TD IC scheme can improve the throughput performance of both CEUs and CCUs, with more than 18.3 and 15.6 percent throughput gain compared to the case without the IC scheme, respectively. Thus, we can conclude that the FD IC scheme can only improve the performance of the CEUs when imposing little restriction on the radio allocation. On the other hand, the performance of both CEUs and CCUs can be enhanced by the TD IC scheme. However, the TD IC scheme has less flexibility because three subframes have to be grouped together for resource allocation. Next, we study the effects of 1D IC schemes on real-time voice over IP (VoIP) service in Fig. 3c. Since the SINR performance of CEUs can be improved by applying the IC schemes, the VoIP capacity of the CEUs is increased, with 14.3 and 28.6 percent gain corresponding to the FD IC and TD IC schemes compared to the case without the IC scheme, respectively. Usually, at least one or two resource blocks (RBs) are needed to support the transmission of each VoIP packet for the CCUs. The small SINR degradation due to the FD IC scheme does not affect the number of allocated RBs, so the VoIP capacity of the CCUs does not decrease. With the TD IC scheme, the SINR performance of the CCUs is also enhanced and leads to VoIP capacity improvement. Thus, the VoIP capacity of all users in the network with the IC schemes outperforms that without the

60

~

~

~

Performance Comparison — The performances of the multihop cellular networks with the 2D IC scheme are also evaluated under the scenario with the parameters given in Table 1. In these simulations two RSs are deployed per sector while one 60˚ directional antenna is deployed for each BS → RS link. The number ratio of CEU to CCU is set as 1:2, and the transmit power ratio between each BS and RS is 3:1. Figure 5a gives the cumulative distribution function (CDF) of the received SINR in the net-

IEEE Wireless Communications • February 2011

ZHENG LAYOUT

2/7/11

10:46 AM

Page 61

1.0

1.0

0.9

0.9

0.8

0.8

0.7

CEU

CDF

CDF

All users

0.6

0.6 0.5

0.5

0.4

0.4

0.3

0.3 0.2

0.2 w/o IC with TD IC with FD IC

0.1 0

CEU

0.7

All users

-4

-2

0

2

4

6

8 10 SINR (dB)

12

14

16

18

w/o IC with TD IC with FD IC

0.1 0

20

0

100

200 300 400 500 Per user throughput (kb/s)

(a)

(b)

10

7.25 6.27 6.21

5

1.49 1.26 1.46

61

w/o IC with FD IC with TD IC

57

60 53

55 VoIP Capacity (users/MHz/sector)

Aggregated cell throughput (Mb/s)

65

w/o IC with FD IC with TD IC

8.73 7.53 7.67

600

50

7.5% 15.1%

45 40 35 30 25 20

14

15

16

18

14.3% 28.6%

10 5

0

0

All users

CCU

CEU

0

0

CEU

All users (d)

(c)

Figure 3. Performance comparison between OFDM-based multihop cellular networks with/without 1D IC scheme: a) SINR distribution; b) per user throughput distribution; c) aggregated cell throughput; d) VoIP capacity. work with or without the 2D IC scheme. With the resource coordination in both the TD and FD, the SINR performance of all users is improved by the 2D IC scheme, especially for the CEUs. Since the 60˚ directional antenna is applied to the BS → RS link in the case with two RSs per sector, less power leaking occurs compared to the case of only one RS per sector. Hence, the intersector interference on the BS → MS link is greatly decreased. In addition, as the main interference of the BS → RS links is not caused by leaking power of the directional antenna from neighboring sectors, the performance improvement of two-hop users with high SINR is not obvious when the 2D IC scheme is applied. In Fig. 5b the per user throughput performance of the network with/without the 2D IC scheme is also compared. Similarly, it can be seen that all users in the network can achieve significant throughput improvements, especially the CEUs, by using the 2D IC scheme. Moreover, we give the cell throughput of the network with or without the 2D IC scheme in Fig. 5c. Compared to the case without the IC scheme, aggregated cell throughput is increased by more than 25.2 per-

IEEE Wireless Communications • February 2011

cent by using the 2D IC scheme, while about 16.4 and 61.9 percent throughput gains are obtained for the CEUs and CCUs in the networks, respectively. In Fig. 5d we compare the VoIP capacity of the network with or without the 2D IC scheme. Both CEUs and CCUs can obtain SINR performance gain when the 2D IC scheme is applied in the networks, resulting in improvement of the VoIP capacity of all users. For example, about 46.6 percent VoIP capacity gain is achieved by the 2D IC scheme when all users are considered.

CONCLUSIONS Interference coordination is essential for OFDM-based cellular multihop networks in order to solve the problem of co-channel interference and achieve high spectral efficiency. In this article we present several semi-static time/frequency interference coordination schemes, demonstrating their applicability and efficiency in 3G LTE networks toward further advancements. In our analysis we have shown that the frequency domain interference coordi-

61

ZHENG LAYOUT

2/7/11

RS8

RS7

10:46 AM

Page 62

RS13

BS1

Radio frame i Downlink Subframe Subframe #8 #9

RS14 RS5

Radio frame i+1 Uplink

Downlink Subframe #0

Subframe #2

Subframe #3

RS11 Frequency

RS1

RS2

RS9

BS2

Frequency RS-MS BS-MS

RS4

RS10 BS 3

RS17

RS6

RS12

RS2-MS RS3

RS15

RS16

BS-RS1

RS5 RS6

RS1 RS2

BS-RS2

1 1 1 subframe subframe subframe

BS-MS

BS-RS5

RS5-MS

BS-RS6

RS6-MS

1 1 1 subframe subframe subframe

RS Time 4 RS3

Time

Frequency

RS18

BS-RS3

RS3-MS

BS-RS4

RS4-MS

BS-MS

1 1 1 subframe subframe subframe

(a)

Time

(b)

Start Time-domain coordination (mitigate the inter-sector interference) Resource allocation • Conjuct each three successive subframes together, and group the same kinds of links in each sector. • Assign the subframes to different kinds of links, each for certain kind of link. • Keep the same kind of links in the neighboring sectors in different subframes.

Power coordination • Half radiated power in the BS → MS subframes. • Use beamforming in the BS → RS subframes (optional). Frequency-domain coordination (mitigate the intra-sector interference) Resource allocation • In the subframes for BS → RS and RS → MS links, orthogonally allocate the frequency resource to BSs and RSs in different groups as high-priority resources. • Assign the remained frequency resources to BSs and RSs as low-priority resources. Power allocation • Full radiated power on high-priority resources. • Half radiated power on low-priority resources.

End (c)

Figure 4. Illustration of 2D interference coordination schemes in OFDM-based multihop cellular networks: a) deployment of two RSs per sector; b) resource allocation by the 2D IC scheme; c) flowchart of the 2D IC scheme. nation scheme can only improve the performance of CEUs, while the performance of both CEUs and CCUs can be enhanced by the time domain interference coordination scheme with a slight strict limitation. Furthermore, the frequency-time domain interference coordination scheme is proposed to achieve throughput gain with high flexibility.

62

ACKNOWLEDGMENT This work was supported in part by the China NSFC under Grant 60802082, National Key Technology R&D Program of China under Grant 2009ZX03003-008-01, and Research Fund for the Doctoral Program of Higher Education under Grant 200800131023.

IEEE Wireless Communications • February 2011

ZHENG LAYOUT

2/7/11

10:46 AM

Page 63

1.0

1.0

0.9

0.9

0.8

0.8

0.7

0.7

0.5 0.4

0.5 0.4

All users

0.3

0.3

0.2

0.2

0.1 0

All users

0.6

CEU

CDF

CDF

0.6

CEU

0.1

w/o IC with 2D IC -2

0

2

4

6

8

10 12 SINR (dB)

14

16

18

20

0 0

22

w/o IC with 2D IC 100

200

300 400 500 600 Per user throughput (kb/s)

90

11.13

w/o IC with 2D IC

85

7.18

5 2.77 1.71

VoIP capacity (users/MHz/sector)

8.36

w/o IC with 2D IC

80

10 Aggregated cell throughput (Mb/s)

800

(b)

(a)

8.89

700

46.6%

70

58

60 50 40

27

30 58.8% 17

20 10

0

0

All users

CCU

CEU

0

0

All users

(c)

CEU (d)

Figure 5. Performance comparison between OFDM-based multihop cellular networks with/without 2D IC schemes: a) SINR distribution; b) per user throughput distribution; c) aggregated cell throughput; d) VoIP capacity.

REFERENCES

BIOGRAPHIES

[1] 3GPP TR 36.211 v. 8.6.0, “Evolved Universal Terrestrial Radio Access (E-UTRA), Physical Channels and Modulation (Release 8),” Mar. 2009. [2] ITU-R M.1645 Rec., “Framework and Overall Objectives of the Future Development of IMT-2000 and Systems Beyond IMT-2000,” June 2003. [3] ITU, “Final Acts — WRC-07,” WRC ‘07, Dec. 2007; http://www.itu.int/publ/R-ACT-WRC.8-2007/en. [4] Y. Yang et al., “Relay Technologies for WiMAX and LTEAdvanced Mobile Systems,” IEEE Commun. Mag., vol. 47, no. 10, Oct. 2009, pp. 100–5. [5] IST WINNER D3.5.1, “Relaying Concepts and Supporting Actions in the Context of CGs”; http://www.istwinner.org/WINNER2Deliverables/D3.5.1v1.0.pdf. [6] B. Can et al., “Implementation Issues for OFDM-Based Multihop Cellular Networks,” IEEE Commun. Mag., vol. 45, no. 9, Sept. 2007, pp. 74–81. [7] 3GPP R1-050507, “Soft Frequency Reuse Scheme for UTRAN LTE,” Huawei, 3GPP RAN WG1 Meeting #41, May 2005. [8] 3GPP R1-050738, “Interference mitigation-Considerations and Results on Frequency Reuse,” Siemens, 3GPP RAN WG1 Meeting #42, Aug. 2005. [9] IEEE 802.16j-06/013r3, “Multihop Relay System Evaluation Methodology (Channel Model and Performance Metric),” Feb. 2007. [10] 3GPP TR 36.814 v. 1.4.1, “Further Advancements for E-UTRA Physical Layer Aspects (Release 9),” Sept. 2009.

KAN ZHENG [M‘03, SM‘09] ([email protected]) received his B.S., M.S., and Ph.D. degree from Beijing University of Posts & Telecommunications, China, in 1996, 2000, and 2005, respectively, where he is currently an associate professor. He worked as a researcher in companies including Siemens and Orange Labs R&D, Beijing, China. His current research interests lie in the field of signal processing for digital communications, with an emphasis on cooperative communication and heterogeneous networks.

IEEE Wireless Communications • February 2011

BIN FAN received his B.S. and Ph.D. from Beijing University of Posts & Telecommunications, China, in 2005 and 2010, respectively. Now he works in Orange Labs R&D, Paris, France. His current research interests lie in the field of cooperative communication and radio resource management. JINHUA LIU received a B.S. from Beijing University of Posts & Telecommunications in 2008, where he is currently working toward an M.S. degree. His current research interest lies in the field of signal processing for cooperative communication. YICHENG LIN received his B.S. and M.S. from Beijing University of Posts & Telecommunications, China, in 2007 and 2010, respectively. His current research interest lies in the field of signal processing for radio resource management.

63

LIU LAYOUT

2/7/11

10:47 AM

Page 64

ACCEPTED FROM OPEN CALL

DISTRIBUTED AUTOMATED INCIDENT DETECTION WITH VGRID BEHROOZ KHORASHADI, FRED LIU, DIPAK GHOSAL, MICHAEL ZHANG, AND CHEN-NEE CHUAH, UNIVERSITY OF CALIFORNIA, DAVIS

ec. 5

Sec. 6

3

ec. 5

1

ec. 5

Sec. 6

Sec. 7

2

Sec. 6

ABSTRACT

Sec. 7

Sec. 7

The authors study an ad hoc distributed automated incident detection algorithm for highway traffic using vehicles that are equipped with wireless communications, processing, and storage capabilities (referred to as VGrid vehicles).

In this article, we study an ad hoc distributed automated incident detection algorithm for highway traffic using vehicles that are equipped with wireless communications, processing, and storage capabilities (referred to as VGrid vehicles). Each VGrid vehicle periodically broadcasts beacon messages with its speed, location, and lane information. Using these beacons, each VGrid vehicle builds and maintains information about different sections of the road. Using such information, each VGrid vehicle independently performs an anomaly detection algorithm based on the traffic density, speed, and the number of lane changes to identify incidents. The robustness of the detection is improved by a voting scheme in which a consensus, among participating VGrid vehicles, is achieved when a fixed number of votes are accumulated. We use a simulation tool called VGSim to study the performance of our detection algorithm in a highway scenario. The results show that our distributed incident detection algorithm has low false positive rate, zero false negative rate, and can still achieve incident detection with as little as 10 percent penetration of VGrid vehicles.

INTRODUCTION Every year freeway accidents and obstructions result in traffic congestion [1], environmental pollution, and damages in property, personal injury, and/or fatalities. The overall cost to society is in the billions of dollars and increasing [1]. With the availability of advanced sensor technologies, preventing and/or minimizing the impact of these occurrences has been a major focus in highway incident detection research in the past decade. An important goal in these efforts has been to reduce the response times in detecting the incident. Timely and accurate inci1

Nodes refer to VGrid vehicles that can store and process information and, using wireless communication, communicate with other VGrid vehicles. Throughout this article we use the terms VGrid vehicle and node interchangeably.

64

1536-1284/11/$25.00 © 2011 IEEE

dent detection is critical to notify approaching drivers, dispatch emergency response teams to deal with the injured, clear the road of the impeding obstruction, to return traffic patterns to the normal flow, and prevent subsequent secondary incidences. Automated incident detection (AID) methods have been developed for detecting potential incidents since the 1970s. However, early deployments of these systems were relatively ineffective since they had either high false alarm rates (which rendered them ineffective) or low detection rates (which rendered them unreliable). The recent advances in sensor technologies and wireless devices have created a new paradigm for design and development of AID systems. Recent systems have incorporated complex algorithms to predict and detect incidents and their locations [2]. Unfortunately, even these elaborate methods have unacceptable levels of false positive and false negative rates. In this article, we propose an AID system that leverages vehicular ad hoc networks (VANETs) and is implemented as an application on the VGrid framework [3]. Within a geographic area, an ad hoc network of vehicles equipped with sensors and in-vehicle processing capability can form an ad hoc cluster of sensors and grid computers, which we refer to as VGrid. VGrid is a control and management infrastructure framework, where VGrid collects and processes realtime traffic data using in-vehicles sensors and computers to make distributed control decisions. VGrid can complement and extend the existing fixed infrastructure by distributing previously centralized control functions, such as traffic statistics collection and dissemination of traffic advisory messages, to local intelligent agents, including in-vehicle sensors and computers. As part of its design, VGrid vehicles broadcast beacon messages that include information about the vehicle’s speed, position, and lane. These messages are then collected by nodes 1 that are within the broadcast range of transmitting vehicles. Using the aggregated information, each node independently estimates and maintains information on the occupancy, lane change, and speed for different sections of the

IEEE Wireless Communications • February 2011

LIU LAYOUT

2/7/11

10:47 AM

Page 65

road. The information is processed to detect anomalous traffic patterns to identify static obstructions in the road. The initial individual detection phase is used to detect all potential incident points, and then collaboration and verification among the nodes is used to filter out erroneous predictions. The proposed method decreases both the false positive and false negative rates to acceptable levels. In addition, because each vehicle acts as a low-cost sensor that collects and processes the information received from other sensors, the detection time is, in most cases, very low (on average about 20 s). The low cost of each device also allows more efficient deployment compared to systems that require devices that must be installed at roadsides or in the road itself. The rest of this article is organized as follows. In the next section we discuss related work on AID. We then give a detailed description of the proposed incident detection approach. We briefly describe the simulation tool and the scenario for which we evaluate the performance of our proposed incident detection algorithm. We then discuss the results and finally, we conclude with a summary of our results and outline future work.

RELATED WORK In general there are four main types of AID detection algorithms. They are based on pattern recognition, catastrophe theory, statistical analysis, and artificial intelligence (AI) techniques [4]. Of these four types, statistical and pattern recognition based methods were the first to be developed in the 1970s. Later, with the emergence of more high-powered computers, AI techniques were applied. The earlier work in AID used techniques based on decision trees for traffic pattern recognition [5], time series analysis [6], and Kalman filters [7]. Based on these techniques, three major AID algorithms were developed: the California Algorithm, the McMaster Algorithm, and the Minnesota Algorithm. The California Algorithm utilizes decision trees to identify anomalous lane densities between detection points. The California Algorithm, also referred to as Traffic Services Corporation (TSC) Algorithm-2, uses a five-minute roll-wave suppression logic that helps reduce false alarms due to shock waves from downstream. The key advantages of the California Algorithm is that it is comprehensive in nature and has low false positive rates. Unfortunately, the California Algorithm makes decisions based solely on the occupancy rate of the road; other parameters such as vehicle speed and volume are not used. The McMaster Algorithm is based on catastrophe theory [8]. It classifies traffic conditions into subcategories and determines if traffic patterns fall within these categories. They are robust in the sense that they are not affected by upstream input failures. However, the McMaster Algorithm only tracks data from the fast lane. Finally, the Minnesota Algorithm, which is similar to the California Algorithm, uses the occupancy rates. This, however, leads to the same disadvantage as the California Algorithm but

IEEE Wireless Communications • February 2011

also has the added limitation of creating a large number of false alarms in low traffic density scenarios. Examples of AID systems based on statistical methods include the DES algorithm summarized in [4], and the Standard Normal Deviation (SND) [4] and Single-Station Incident Detection (SSID) algorithms [9]. SSID uses the standard statistical T-test to analyze the differences in occupancy. The introduction of AI techniques has resulted in new AID approaches. The study reported in [2] uses a combination of fuzzy logic and genetic algorithms (GAs). While this approach is highly adaptive, both fuzzy logic and GAs are highly complex algorithmic approaches that have yet to be proven to be efficient. The AI-based approaches for AID systems have also focused on artificial neural networks which contain multiple layers, multiple inputs, and a complex structure. This approach has been shown to perform better than fuzzy logic and GA-based approaches; however, complexity and accuracy are still major drawbacks. The use of AI in conjunction with modern camera technology was investigated in [10]. Unfortunately, this approach is very expensive since cameras must be installed and maintained. An example of this is the video or 3D image processing adopted in [11]. In recent years, VANET has provided a unique framework to develop new techniques for AID. In the research reported in [12, 13], a centralized detection approach is proposed for AID. These methods rely on information collected and reported periodically at roadside nodes by VANET vehicles. These roadside nodes then aggregate the information collected from individual vehicles to detect obstructions upstream from the roadside node. Although this work addresses a similar problem (AID using VANET), there are fundamental differences that distinguish it from the work presented in this article. A major difference is that the method proposed in this article is a fully distributed approach that leverages the fully distributed VGrid computing framework in which vehicles form an ad hoc network with no infrastructure coordination to detect anomalies in vehicular traffic flows. This distributed design of the AID scheme has significant advantages, as mentioned later in this article. In addition, the proposed detection method is evaluated using VGSim [14], a simulation platform designed specifically to simulate a realistic mobility model for vehicles and an integrated wireless networking protocol stack. To date, the most effective and rapid detection system seems to be based on cell phone callers and/or roadside call boxes [4]. It has been argued that using cell phones to call in roadside incidents can be dangerous, and could result in secondary accidents and/or other traffic flow problems. Furthermore, studies have shown that as cell phone conversations increase, the likelihood that some highway traffic situation will go unreported and/or unnoticed also increases [15]. The use of call boxes can be very effective but requires the overhead of deployment; the more frequent the call box stations, the greater the cost.

The key advantages of the California Algorithm is that it is comprehensive in nature and has a low false positive rates. Unfortunately, the California Algorithm makes decisions based solely on the occupancy rate of the road; other parameters such as vehicle speed and volume are not used.

65

LIU LAYOUT

2/7/11

10:47 AM

Over time each VGrid vehicle builds an independent and local view of the road and its traffic pattern. Each vehicle periodically searches this aggregated information for specific markers that indicate road obstructions.

Page 66

Field

Type

Size

Source IP address

Integer

4 bytes

Position

Integer

4 bytes

Lane number

Byte

1 bytes

Time-to-live (TTL)

Byte

1 bytes

Message timestamp

Long

8 bytes

Vehicle speed

Short

2 bytes

Road ID

Integer

4 bytes

Total bytes

24 bytes

Table 1. Beacon message fields. All the previously mentioned AID techniques sort through data collected from roadside sensors or other similar data collection devices. As such, these algorithms are all constrained by the same limitation; they can only achieve a certain level of accuracy and efficiency. VANET technologies and the proposed VGrid framework [14] enable new approaches to AID. They provide both spatially and temporally fine-grained data regarding the speed, position, and lane changes of vehicles. Furthermore, the distributed grid computing (VGrid) framework implemented over VANET allows complex coordinated computation over the data to accurately and robustly detect various patterns in the traffic flow and thereby detect incidents.

OBSTRUCTION DETECTION ALGORITHM In order for vehicles to detect potential road obstruction, vehicles must exchange traffic information. The VGrid framework [14] provides this in the form of traffic beacon messages. These message are broadcasted periodically by each VGrid vehicle and, in-turn, collected by all VGrid vehicles that are within the transmission range of the broadcast. Over time each VGrid vehicle builds an independent and local view of the road and its traffic pattern. Each vehicle periodically searches this aggregated information for specific markers that indicate road obstructions. Once a vehicle detects an obstruction, it initiates a voting phase, in which nodes within the transmission range vote to reach a distributed consensus on the detection. In the following subsections we describe the details of the algorithm.

BEACON MESSAGES AND ROAD PROFILING The core of this algorithm relies on the dissemination of vehicle traffic information by all VGrid vehicles. The beacon messages are broadcast at 4 Hz (4 times/s) and contain the fields shown in Table 1. The Manual on Uniform Traffic Control Devices (MUTCD) standardizes the partitioning of roads into logical sections by demarcating highway sections into 100 ft increments. Although this is primarily used for the design of highway systems, it additionally pro-

66

vides coordination between vehicles to acquire road ID and position information which can be used to organize obstruction location information. To determine lane information multiple methods exist, and we believe the simplest method is to use slightly higher accuracy GPS systems than are available in today’s vehicles. Any GPS unit that can obtain ±1 ft accuracy is sufficient to determine the lane information needed for this detection method. More accurate GPS units are also available that have subcentimeter accuracy; however, that level of detail far exceeds the needs of lane determination. Through detailed experiments we found that a beaconing rate between 0.5 Hz and 20 Hz was required to ensure the required performance of our proposed AID scheme. For beaconing rate less than 0.5 Hz there was noticeable degradation of the detection rate of the algorithm. For beaconing rate greater than 20 Hz, there were packet drops due to collisions, which in turn resulted in higher detection time. Indeed, the impact was higher for higher traffic densities, which were caused by local congestion near the obstruction point. The results shown in this article have been obtained for a beaconing rate of 4 Hz. Note that the size of the beacon messages is small (24 bytes), resulting in minimal impact on network congestion when beacon frequency is within a given threshold. A more detailed analysis of the impact of beacon frequency on performance is reported in [14]. The beacon messages are collected by each VGrid vehicle, and then used to form a dynamic view of the road and the traffic patterns. Each VGrid vehicle gathers traffic information and uses it to track the movement and position of other VGrid vehicles. Tracking is done by simply storing the beacon messages of each distinct vehicle. Since beacon messages are periodic when a new beacon is received, the receiving vehicle can infer the trajectory of the source vehicle. For example, assume that a vehicle x receives a message from vehicle y at time 2 s with a beacon message that indicates a position of 150 in lane 3. Later, vehicle x receives another beacon from vehicle y at time 5 s with position 200 and lane 3. Vehicle x infers that vehicle y traveled from position 150 to 200 in 3 s. In the event of a lane change, vehicle x makes the assumption that y changed lanes halfway between points 150 and 200. Each VGrid vehicle maintains three information arrays for each lane, where each cell in the array corresponds to a particular section of road. We choose road sections to be the length of a vehicle in the simulation (7.5 m or 75 cells in the simulation). In order to compensate for the reduction in penetration rate (the percentage of VGrid enabled vehicles) we then partition the road inversely proportional to the penetration rate: road_section_size = vehicle_length/penetration_rate (1) These road section divisions are not relative to the vehicle itself but to the location. The coordination between vehicles can be achieved relatively simply as highways have already been naturally partitioned using mile markers. Sec-

IEEE Wireless Communications • February 2011

LIU LAYOUT

2/7/11

Sec. 1

10:47 AM

Sec. 2

T=0

Page 67

Sec. 3

Sec. 4 2

Sec. 5

Sec. 6

1

Sec. 1 5

Sec. 2

Sec. 8

Sec. 9 Sec. 10 4

3

Sec. 3

Sec. 4

Sec. 5

Sec. 6

Lane 1 Lane 2

Sec. 7

Sec. 8

Sec. 9

Sec. 10 Lane 1

T=1

1

Sec. 1

Sec. 7

Sec. 2

Sec. 3

Sec. 4 5

Sec. 5

2

Sec. 6

T=2

3

Sec. 7

Sec. 8

Sec. 9 Sec. 10 2

1

Lane 2

Lane 1 Lane 2

X VGrid Vehicle X

The coordination between vehicles can be achieved relatively simply as highways already have been naturally partitioned using mile markers. Sections zero can start at the beginning of each unique mile marker on the highway.

Road obstruction

Figure 1. Illustration of how vehicles move through in time. Table 2 represents the corresponding array values as seen by vehicle 1 or 2 as both vehicles are present in all three time instances.

tions zero can start at the beginning of each unique mile marker on the highway. We use the following three arrays to collect information used for the detection algorithm. The road section count (RSC) array indicates the number of vehicles that have been tracked over a particular section. In order to create the RSC array, vehicle x will increment each cell in the RSC when a vehicle is either tracked over a section or is located in a particular section. Consider the previous example. If vehicle x receives a beacon from vehicle y located in section 150 of the road, x will increment its count or RSC array by one at index 150 for lane 3. When the second beacon is received x will increment indexes 151–200 in lane 3. If y were to change lanes the increment would be for 151–175 in lane 3 and 175–200 in lane 2 (assuming lane 2 was the lane vehicle y moved into). The road section speed (RSS) array contains the weighted average of speeds tracked over a particular section of the road. The indices (which correspond to road sections) in the RSS array are initialized to the maximum, which for simulation study reported in this article is set to 30 m/s (≈60 mph). Vehicles are tracked over road sections in the same fashion as in the RSC array. Speed is either calculated by tracking a vehicle over time or extracted directly from the beacon message itself. The RSS for an index i in the array is updated or calculated using an exponentially weighted moving average given by RSS[lane][i] = (1 – α)RSS[lane][i] + α * vtracked_speed,

(2)

where 0 < α ≤ 1, lane is the lane which the vehicle is being tracked, and v tracked_speed is the observed speed of the vehicle being tracked. Again, as a vehicle y is tracked over multiple sec-

IEEE Wireless Communications • February 2011

tions, Eq. 2 is applied to each section vehicle y passes over. A weighted moving average is used to quickly reflect the speed changes in traffic over time. We chose an α = .25 to quickly exhibit changes in traffic speeds in a particular section but not overly emphasize the information disseminated by any one particular vehicle. The end goal was to have newly disseminated traffic information about a recently created obstruction quickly dampen stale data in a road section where an obstruction did not previously exist. The vehicle lane change (VLC) array contains a counter for each section of the road. If a vehicle is observed changing out of a lane in a particular section the index is decremented. If a vehicle is observed entering a lane at a particular section, the index in VLC is incremented. The lane change point is an estimation of a vehicle’s actual location of change. However, because beacon messages are so frequent (4 Hz, which is 4 beacon messages/s), and the sections of the road are large, it was found that this estimation was adequate for the detection process. Figure 1 illustrates the obstruction detection scenario at three different time instances, and Table 2 shows the RSC, RSS, and VLC array values for each time frame. In this example, we assume that all vehicles are within each other’s transmission range and there are no packet losses (due to interference and collision). It is important to note that the proposed method relies on accurate positioning information. We assume that the VGrid vehicles have a GPS device that is accurate to within a few feet to pinpoint lane location. While currently midrange GPS devices are accurate to within 10–15 ft, the higher end units are accurate within 1–2 ft and would satisfy the requirements to accurately determine the vehicles, lane, and position. We also found that there are production quality

67

LIU LAYOUT

2/7/11

Time

10:47 AM

Array

Page 68

Lane #

Sec: 1

Sec: 2

Sec: 3

Sec: 4

Sec: 5

Sec: 6

Sec: 7

Sec: 8

Sec: 9

Sec: 10

Lane 1

0

0

0

1

0

0

0

0

1

0

Lane 2

0

1

0

0

0

1

0

0

0

0

Lane 1

30

30

30

29

30

30

30

30

28.5

30

Lane 2

30

27.5

30

30

30

29

30

30

30

30

Lane 1

0

0

0

0

0

0

0

0

0

0

Lane 2

0

0

0

0

0

0

0

0

0

0

Lane 1

1

0

0

1

1

0

0

0

1

1

Lane 2

0

1

1

1

2

2

1

1

1

0

Lane 1

29

30

30

29

7

30

30

30

28.5

29

Lane 2

30

27.5

20

20

14

20

29

29

29

30

Lane 1

0

0

0

0

–1

0

0

0

0

0

Lane 2

0

0

0

0

1

0

0

0

0

0

Lane 1

1

1

1

2

1

0

1

1

2

1

Lane 2

0

1

1

1

2

3

3

1

1

0

Lane 1

29

28.5

28.5

28

7

30

27

27

26.5

29

Lane 2

30

27.5

20

20

14

21.5

25.2

25

29

30

Lane 1

0

0

0

0

–1

0

+1

0

0

0

Lane 2

0

0

0

0

1

0

–1

0

0

0

RSC

T=0

RSS

VLC

RSC

T=1

RSS

VLC

RSC

T=2

RSS

VLC

Table 2. Representation of RSC, RSS, and VLC arrays for vehicle 2 in Fig. 1.

units that can pinpoint position accuracy to within 2 cm.2 These devices are designed specifically to measure vehicle positions and are currently used to collect acceleration and braking distance measurements. Accuracy within 2 cm is, of course, not a requirement for functional performance of our proposed AID system. However, it is important to note that such technology is available and can be applied in the future. Additional technologies that are able to track lane change include devices that track a vehicles tire movement, accelerometers coupled with GPS systems, or other types of sensors such as those embedded within the road itself designating each lane. Although lane determination is a required component of this proposed AID system, we have described several technologies that are available for lane detection. Consequently, we believe that this requirement would not be a cause for barrier to entry for our proposed AID system.

DETECTION METHOD 2

An example of such a device can be found at http://www.racelogic.co.u k/?show=vbox

68

Detection is carried out by each vehicle by periodically scanning the three arrays in search of anomalous traffic patterns. In order to quantify the search object, we first formulated the look of

a road obstruction from a purely human perspective. Figure 2 depicts what we would expect from tracking vehicles around an obstruction point. The most prominent characteristic of an obstruction point will be the lack of vehicles in the lane downstream from the obstruction. In addition, vehicles upstream from the obstruction, in the same lane, would attempt to merge out of that lane while no vehicles should enter that lane. Conversely, vehicles after the obstruction point should attempt to merge back into the lane due to the available space. Finally, vehicles stuck behind the obstruction point will travel at much slower rates than the normal traffic, and vehicles past the obstruction point should be able to travel at average traffic speeds (where average speed is calculated by averaging all speeds collected from beacon messages within the area). The detection method consists of two phases. In the first phase individual VGrid vehicles independently search through the accumulated and aggregated traffic information collected from beacon messages to identify potential obstruction points. Once a vehicle identifies a potential obstruction point, it moves to the second phase of the algorithm, in which it broadcasts an

IEEE Wireless Communications • February 2011

2/7/11

10:47 AM

Page 69

obstruction detection vote. The vote contains the vehicles unique id and also the details concerning the obstruction location. The first vehicle to collect enough votes exceeding some pre-defined vote_threshold, broadcasts an alert message verifying the obstruction point. In this article, we consider positive detection only once an alert message is generated. Once a consensus is reached, alert messages can be generated to all VGrid vehicles by the method described in [14]. We define the time to detection as the time at which the obstruction occurs until the first alarm message is generated. Vehicles that have not detected the obstruction point themselves are restricted from broadcasting an alert message. Only vehicles that have both locally detected the obstruction point and acquired enough votes are permitted to generate an alert message.

Phase 1 (Independent Search) — Phase one of the detection algorithm begins by periodically (every 1 second) scanning only the RSC array. As previously mentioned, the point of, and following, an obstruction can be classified as a low density area, where few or no vehicles entered. In the case of the RSC, a search is conducted for large drops in observed count between adjacent sections in the road. For example, if road section i contains an obstruction, each vehicle scans for a significant contrast between RSC[lane][i – 1] and RSC[lane][i]. This is done by building an orthogonal array RSCdif, which is defined as follows:

VGrid enabled vehicle Non-VGrid enabled vehicle

VGrid enabled vehicle

Road obstruction

Non-VGrid enabled vehicle

Expected vehicle trajectory

Area of low traffic density

Figure 2. Figure depicts what is expected from a road obstruction. Directly after the obstruction point it is expected that average density would be lower than the rest of the lanes. Before the obstruction vehicles should be attempting to exit the lane and return after the obstruction point.

Road section count (RSC) Lane 1 Lane 2 Lane 3 Lane 4

80 70 60 Count

LIU LAYOUT

50 40 30 20

RSCdif[lane][i] = abs(RSC[lane][i] – RSC[lane][i – 1]),

10 0

where i is the section of road, lane is the lane in the road, and abs(x) is the absolute value of x. Since RSCdif contains the absolute value change between road sections, vehicles then search for a large fluctuation in RSCdif. The scope of the investigation is made more efficient by implementing a sliding search window, which reduces the range of the search to 200 m before the current vehicle’s position and 100 m after. This is done to increase the efficiency of searches by avoiding an entire scan every cycle. We can assume that few messages will be received from outside of the search window range and that the search should not be conducted in road sections where the vehicle has not passed or reached (in terms of transmission range). The asymmetry of the search range is used to limit the scope of the search upstream from the vehicle (200 m before) based on the relative range of the vehicles transmission. The downstream (100 m ahead) limit is due to the fact that sufficient information may not have been collected for an accurate search since the vehicle has not yet reached the area in question. In order to classify a significant change in RSCdif, we track the mean and standard deviation of the values in RSCdif. Changes that are three standard deviations from the mean are flagged as significant drops in road count. When such a point is detected, vehicles examine the section in question and use the VLC and RSS to verify those anomalous points. This is done through analysis of the VLC to verify that lane changes 100 m before the flagged section are outbound, and all lane changes 100 m after are

IEEE Wireless Communications • February 2011

0

2,500

5,000

7,500 10,000 Position (cells)

12,500

15,000

Figure 3. Road Section Count (RSC) array for a particular vehicle collected during actual simulation. Penetration rate = 100 percent, traffic density = 15 percent, and transmission range = 500 m (19.2 dBm, which is a transmission power that yields a transmission range of approximately 500 m).

inbound. If this is verified the RSS is then used to check that the average speed of section i – 1 is less than 20 percent of the observed average speed of surrounding vehicles and that average speed of section i + 1 is at least 80 percent of the observed average speed. In essence, vehicle speeds behind the obstruction should be slow and speeds in front of the obstruction faster. Phase 1 can be summarized as follows: RSCdif[lane][i] > mean + 3 * stdev

(3)

VLC[lane][i – k] < 0 ∀ k < 100 m

(4)

VLC[lane][i + k] > 0: ∀ k < 100 m

(5)

RSS[lane][i – 1] < avgSpeed * .2

(6)

RSS[lane][i + 1] > avgSpeed * .8

(7)

We use Eq. 3 to locate a potential obstruction point and then use Eqs. 4, 5, 6, and 7 as a filtering mechanism to filter out false positives. Figure 3 represents a snapshot of the RSC array collected from a particular vehicle, in a simula-

69

LIU LAYOUT

2/7/11

10:47 AM

Page 70

SIMULATION STUDY

RSCdif Array 45 40 RSC(i) – RSC(i – 1)

SIMULATION TOOL

Lane 1 Lane 2 Lane 3 Lane 4

35 30 25 20 15 10 5 0 0

2500

5000

7500 10,000 Position (cells)

12,500

15,000

Figure 4. The RSCdif generated from RSC array illustrated in Fig. 3. Penetration rate = 100 percent, traffic density = 15 percent and transmission range = 500 m (19.2 dBm).

Sleep for 1 s

Phase 1

Phase 2

Search for anomalies in RSCdif array (Equation 3)

Voting threshold is reached and alert is generated

No anomaly detected

Anomaly detected Verify anomaly with VLC and RSS arrays (equations 4,5,6 & 7)

No detection

Insufficient votes

Anomaly verifed by vehicle

Vehicle totals all votes received thus far for this location

Vehicle broadcasts a “vote”

Figure 5. A flow diagram of the detection method. tion in which an obstruction was present. The RSCdif generated from Fig. 3 is represented in Fig. 4.

3

Details of VGSim can be found at http://wwwcsif.cs.ucdavis.edu/VGrid 4

This tool can be found at http://jist.ece.cornell.edu/ 5

Documentation for SWANS can be found at http://www.aqualab.cs.nor thwestern.edu/projects/sw ans++/

70

Phase 2 (Coordinated Voting) — In the second phase, vehicles cast votes to verify an obstruction point. We found that in phase 1, approximately 98 percent of the votes cast were accurate. However, we argue that not only does the second phase guarantee that votes do not generate inaccurate alerts, but it avoids even a 1 percent false positive rate, which can render an AID system unreliable [4]. It can also protect against misbehaving nodes, which could broadcast bogus votes. In this article we do not explore the effect of misbehaving nodes. Further study of the second phase of this algorithm showed that for particular traffic situations the voting mechanism was critical in reducing the false alarm rate while still ensuring reasonable detection times. In particular, we found that a voting threshold of three was sufficient to weed out the false positives generated in the first phase of the algorithm. Figure 5 gives a highlevel overview of the detection method as executed by an individual vehicle.

In order to carry out simulations we use VGSim,3 which is a simulator based on Java in Simulation Time (JiST). 4 In VGSim, the network simulation module is based on SWANS, 5 a Java based network simulator. In VGSim, vehicular movements and applications are transformed into events that are processed by the JiST event driven platform. The network simulator and the vehicular traffic model run on a feedback loop that enables the interaction between the networking simulation and the vehicular mobility. Information obtained from the SWANS network simulator is fed into the mobility model and then based on the mobility model, updated antenna positions are determined for the SWANS network simulator. The work done in [3] proposes and validates a mobility model specifically for the VGSim platform. The vehicular mobility module of VGSim is a Cellular Automata (CA) model, which implements a modified version of Nagel and Schreckenberg (N-S) model. A detailed description of this integration of the modified N-S model and the VGSim simulation platform is described in [14]. The simulation tool also has a graphical user interface (GUI) which provide a visual interface to the simulations. The SWANS network simulator and vehicle mobility simulator both update a graphical interface that allows network and vehicle mobility parameters to be changed dynamically.

SIMULATION PARAMETERS VGrid vehicles are equipped with low cost processor and memory to store data and carry out computation on the data. In addition each VGrid vehicle carries an onboard GPS unit and wireless network enabled following the IEEE 802.11b standard. As part of our study, we vary the percentage of VGrid vehicles and refer to this percentage as the penetration rate. In order to quantify our algorithm, we use the following three common AID performance measurements: false alarm rate (FAR), detection rate (DR), and mean time to detection (MTTD), FAR = DR =

# of false alerm cases × 100% 20 seconds

# of detected incidents total # of incidents

MTTD =

1 n ∑ (tdetection − t start ) n i =1

(8) (9)

(10)

where tdetection is the time of the first alert message and t start is the start time of the incident. We chose an interval of 20 s to calculate FAR because in the majority of the studies presented in [4], the AID algorithms used the same FAR metric. We consider a 1.5 km stretch of highway. Incidents are injected at cell position 9000 (900 m) in lane 2 at time 0. The simulation

IEEE Wireless Communications • February 2011

2/7/11

10:47 AM

Page 71

ends when the first alarm is generated. Transmission range is based on an estimation of the corresponding decibel level designated in the simulation. A transmission range of 500 m refers to the maximum transmission distance achievable given the transmission decibel level (19.2 dBm). As a result, the average transmission range is far less than the designated transmission range. When the transmission range is set to 500 m, the average transmission is closer to 250 m. The traffic injection is based on the traffic density parameter. The road is initially populated in probabilistic fashion. If the traffic density parameter is 20 percent, enough vehicles are placed on the road to fill 20 percent of the cells on the road. Since each vehicle occupies 75 cells (7.5 m), calculating the number of vehicles to place on the road for a given density is a simple linear equation. When placing the vehicles the initial positions of each vehicle is chosen at random. The injection pattern is based on exit the frequency of each vehicle. When a vehicle exits the road, a new vehicle is injected at the beginning of the road. If the new vehicle cannot be injected, the vehicle is queued and injected when sufficient room is available. Vehicle are randomly marked as VGrid vehicles upon creation (both in the initial population of vehicles or upon injection) with a probability based on the chosen penetration rate.

Average detection time, all TX ranges

MTTD (s)

LIU LAYOUT

70 65 60 55 50 45 40 35 30 25 20 15 10 5 0

Penetration 1.0 Tx tange (meters) 250 Penetration 1.0 Tx tange (meters) 250 Penetration 1.0 Tx tange (meters) 750 Penetration 1.0 Tx tange (meters) 1000

0.05

0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 Traffic density

Figure 6. The MTTD for transmissions ranges of 250, 500, 750, and 1000 m and penetration rate of 100 percent.

Penetration rate

Traffic density

DR

5%

All

0%

10%

5%

7%

10%

10%

20%

10%

15%

77%

RESULTS

10%

20%

90%

EFFECTS OF TRANSMISSION RANGE

20%

5%

96%

Figure 6 shows the impact of transmission range on the mean detection time (MTTD) for a penetration rate of 100 percent. Figures representing the other penetrations are omitted due to space constraints. Our study found that a transmission range of 250 meters was not sufficient for a 100 percent DR in all cases. In fact, for penetration rates less than 25 percent, DR was almost zero except when traffic density was very high. Conversely, for higher transmission ranges, such as 1000 meters, it was possible to increase the detection rate even with low penetration rates and low traffic densities. We also observed some anomalous behavior in the 20 to 30 percent traffic density ranges seen by the erratic values for MTTD. This behavior is a result of traffic mobility transitioning between free flow (vehicles can travel at maximum velocity) and congested traffic (vehicle velocity is constrained by downstream traffic). In these transition ranges the variance in velocity of individual vehicles is much greater across the entire road, resulting in a higher variance in MTTD. The result in Fig. 6 shows that in order to enable detection, VGrid vehicles must form a connected grid or mesh to coordinate detection. If the transmission range is too small and the VGrid vehicles are sparse, inter-vehicle communication becomes sporadic if not entirely disabled. As a result, detection of obstructions (through coordination) becomes unlikely or impossible. Results clearly show that there is a minimum required transmission range to enable detection.

IEEE Wireless Communications • February 2011

Table 3. DRs less than 100 percent. Penetration rate

Traffic density

DR

Obstruction present

50%

50% and 60%

1.1%

No

50%

40%

0.56%

Yes

50%

50% and 60%

1.1%

Yes

25%

60%

0.56%

Yes

75%

60%

0.13%

Yes

Table 4. FAR greater than 0 percent.

EFFECTS OF PENETRATION RATE AND TRAFFIC DENSITY To test the limits of our detection algorithm, we focused on penetration rates and traffic density parameters in ranges that were less than ideal for detection. The detection rate is a function of penetration rate and traffic density. In general, if enough VGrid vehicles were present, detection was possible. With a penetration rate of 5 percent, detection was not possible for any traffic densities. Detection became possible with a 10 percent penetration rate. However, at the lower

71

LIU LAYOUT

2/7/11

10:47 AM

Page 72

3D Comparison of penetration rate and traffic density vs MTTD

500 450 400 MTTD (s)

350 300 250 200 150 100 50 0 0 0.2 0.4

0.4 Traffic density

0.6 0.8

1

0.8

0.2

0

0.6 Penetration rate

Figure 7. The MTTD as a function of both penetration rate and traffic density (transmission range is set at 500 m).

traffic densities (5, 10, 15, 20 percent), DR was still less than 100 percent. Table 3 represents all points were DR was less than 100 percent. For other penetration rate and traffic density values the DR was 100 percent. An important metric of any AID system is its false alarm rate (FAR). If FAR is too large, the system become unreliable. Even a FAR of 1 percent, can make the system unreliable [4]. For the proposed approach, we categorize the FAR for two different scenarios. The first is false alarms generated in a traffic scenario where an accident exists (an alarm generated for an obstruction other than the actual obstruction). The second case is false alarms generated in the situation where no obstructions exist (control simulations). Table 4 shows the data for which a FAR greater than 0 percent was observed. When considering the FAR for the first scenario (an obstruction is present) it is important to point out that all false alarms were generated for locations upstream from the actual obstruction. This was due to the traffic congestion generated by the downstream road blockage. In highly congested traffic vehicles move very slowly, which is falsely detected as obstructions. Figure 7 shows a 3D plot of the effects of penetration rate and traffic density on MTTD. Excluding the low penetration rate and traffic density scenarios, we found that MTTD was, in general, less than 1 min. It is interesting to note the effect of traffic density on MTTD. We found that the more vehicles present, the faster the detection time. This is clearly shown in the figure as detection time is much slower for the 5 and 10 percent traffic densities. We also found, however, that for the higher traffic densities there was also a slight increase in MTTD. The increase in MTTD for the congested traffic scenarios is due to the fact that the detection time is correlated with the movement of the vehicles. Our algorithm is based on each vehi-

72

cle tracking traffic pattern over time and space. With very high traffic congestion, vehicles move slowly, which has the same effect as slowing down time. When the vehicles move slowly, it is more difficult to quickly identify areas of low traffic density. Table 5 is a comparison of the proposed method (referred to as VGrAID) and other AID algorithms. When presenting our results in Table 5, we choose to aggregate our findings by penetration rate. The average FAR for a given penetration rate across all traffic densities is less than the previously stated FAR considering both the traffic density and the penetration rate. The results and values presented for other AID algorithms are referenced from the “Summary of Algorithms” table in [4] with results of VGrAID appended at the bottom. Note that the values for DR, TTD, and FAR may not have been measured in the same way. The table only gives a qualitative comparison of the different algorithms that exist.

CONCLUSION AND FUTURE WORK In this article we have presented a distributed AID method that leverages ad hoc networking and computing in vehicles with storage, computing, and wireless networking capabilities. We show that information collected by the vehicles, analyzed, and shared in a distributed manner can improve AID detection times compared to those of traditional systems. In addition, since nodes are mobile and the information they generate is shared, the data is more accurate and timely than other AID systems. With the low cost of each device, we believe that implementation of such a system could potentially be more cost effective than other traditional infrastructure-based systems. Finally, the collaborative nature of the system also greatly reduces the false alarm rate and can, in the future, protect against malicious behavior such as a malfunctioning sensor or intentional disruptive behavior. As part of the future work we will explore the concept of categorizing incidents based on additional information collected by the vehicular grid network. Furthermore, in this work we categorize an obstruction as a fixed blockage in the roadway. We can expand this definition to include slow moving or partially disabled vehicles. As an extension to this work, we also plan to modify the proposed algorithm to adapt to a centralized infrastructure where roadside sensors can be leveraged in order to accumulate traffic data collected and dumped by moving vehicles. This could greatly reduce the number of VGrid vehicles needed to facilitate detection as the roadside nodes could collect and aggregate information over a period of time. In addition to studying variations in which VGrid AID can be applied, we will also study the communication impact of this AID system and how it could impact other concurrent safety and infotainment applications.

ACKNOWLEDGMENT The authors would like to thank the reviewers for their comments. This research was funded by NSF grant CMMI-0700383. Dr. Behrooz Khorashadi is now employed with Qualcomm.

IEEE Wireless Communications • February 2011

LIU LAYOUT

2/7/11

10:47 AM

Page 73

REFERENCES [1] D. Schrank and T. Lomax, “The 2007 Urban Mobility Report,” Texas Transportation Institute, Texas A&M Univ., Sept. 2007. [2] D. Srinivasan, R. L. Cheu, and Y. P. Poh, “Hybrid Fuzzy Logic-Genetic Algorithm Technique for Automated Detection of Traffic Incidents On Freeways,” Proc. IEEE Intelligent Transportation Sys. Conf., Aug. 2001. [3] C.-N. Chuah et al., “Distributed Vehicular Traffic Control and Safety Applications with VGrid,” IEEE Wireless Hive Net. Conf., Aug. 2008. [4] T. Martin et al., “Incident Detection Algorithm Evaluation,” prepared for Utah Dept. Transportation, Mar. 2001. [5] H. J. Payne, E. D. Helfenbein, and H. C. Knobel, “Development and Testing of Incident Detection,” Federal Highway Administration, Research Methodology and Results, vol. 2, 1976, report no: FHWA-RD-76-20. [6] S. R. Ahmed and A. R. Cook, “Application of Time-Series Analysis Techniques to Freeway Incident Detection,” Transportation Research Record, 1982, pp. 19–21. [7] S. Willsky et al., “Dynamic Model-Based Techniques for the Detection of Incidents on Freeways,” IEEE Trans. Automatic Control, vol. 25, no. 3, 1980, pp. 347–60. [8] N. Persaud, F. L. Hall, and L. M. Hall, “Congestion Identification Aspects of the McMaster Incident Detection Algorithm,” Transportation Research Record, tech. rep. 1287, 1990, pp. 167–75. [9] C. Antoniades and Y. Stephanedes, “Single-Station Incident Detection Algorithm (SSID) for Sparsely Instrumented Freeway Sites,” Transportation Eng., 1996. [10] S. Kamijo et al., “Traffic Monitoring and Accident Detection at Intersections,” Proc. IEEE ITSC, Oct. 1999. [11] T. Martin et al., “Video-Based Automatic Incident Detection for Smart Roads: The Outdoor Environmental Challenges Regarding False Alarms,” IEEE Trans. Intelligent Transportation Sys., vol. 9, no. 2, June 2008. [12] M. Abuelela, S. Olariu, and G. Yan, “Enhancing automatic incident detection techniques through vehicle to infrastructure communication,” 11th IEEE ITSC ‘08, Oct. 2008, pp. 447–52. [13] M. Abuelela and S. Olariu, “Automatic Incident Detection in VANETs: A Bayesian Approach,” Proc. IEEE VTCSpring, Barcelona, Spain, Apr. 2009. [14] B. Khorashadi, Enabling Traffic Control and Data Dissemination Applications with VGrid — A Vehicular Ad Hoc Distributed Computing Framework, Ph.D. thesis, UC Davis, 2009; http://wwwcsif.cs.ucdavis.edu/~VGrid/ VGrid/Publications.html. [15] J. McKnight and A. McKnight, “The Effect of Cellular Phone Use upon Driver Attention,” Nat’.l Public Services Research Inst., 1991.

BIOGRAPHIES DIPAK GHOSAL received his B.Tech. degree in electrical engineering from the Indian Institute of Technology, Kanpur, in 1983, his M.S. degree in computer science and automation from the Indian Institute of Science, Bangalore, in 1985, and his Ph.D. in computer science from the University of Louisiana in 1988. He is currently a professor in the Department of Computer Science at the University of California, Davis. His primary research interests are in the areas of high-speed networks, wireless networks, vehicular ad hoc networks, next-generation transport protocols, and parallel and distributed computing. CHEN-NEE CHUAH is currently a professor in the Electrical and Computer Engineering Department at the University of California, Davis. She received her B.S. in electrical engineering from Rutgers University, and her M. S. and Ph.D. in electrical engineering and computer sciences from the University of California, Berkeley. Her research interests lie in the area of computer networks and wireless/mobile computing, with emphasis on Internet measurements, network anomaly detection, network management, multimedia, online social networks, and vehicular ad hoc networks. She received the NSF CAREER Award in 2003 and the Outstanding Junior Faculty Award from the UC Davis College of Engineering in 2004. In 2008 she was selected as a Chancellor’s Fellow of UC Davis. She has served on the executive/technical program committee of several ACM and IEEE conferences, and is currently an Associate Editor for IEEE/ACM Transactions on Networking. MICHAEL ZHANG is currently a professor in the Civil and Environmental Engineering Department at the University of California, Davis. His research is in traffic operations and control, transportation network analysis, and intelligent

IEEE Wireless Communications • February 2011

Name

DR (%)

TTD (min)

FAR (%)

APID

86

2.50

0.05%

DES

92

0.70

1.87%

ARIMA

100

0.40

1.50%

Bayesian

100

3.90

0%

California

82

0.85

1.73%

Low-pass filter

80

4.00

0.30%

McMaster

68

2.20

0.0018%

Neural networks

89

0.96

0.012%

SND

92

1.10

1.30%

SSID

100

not reported

0.20%

TSC 7

67

2.91

0.134%

TSC 8

68

3.04

0.177%

Video image processing

90

0.37

3.00%

Cell phones

100

—

5.00%

VGrAID 10% penetration

74%

3.58

0.0%

VGrAID 20% penetration

99.6%

1.29

0.0%

VGrAID 25% penetration

100%

1.06

0.046%

VGrAID 50% penetration

100%

.45

0.16%

VGrAID 75% penetration

100%

0.37

0.01%

VGrAID 100% penetration

100%

0.48

0.0%

Note: This table provides a high level qualitative comparison of the DR, TTD, and FAR metrics. These variations are the result of the various measurement methodologies used in each study.

Table 5. Comparison of VGrAID with other AID algorithms. transportation systems. He received his B.S. degree in civil engineering from Tongji University, and M.S. and Ph.D. degrees in engineering from the University of California, Irvine. He is an Area Editor of the journal Network and Spatial Economics and an Associate Editor of Transportation Research — Part B: Methodological. B OJIN L IU ([email protected]) is a Ph.D. student in the Computer Science Department, University of California, Davis. He received his Bachelor’s degree in computing from Hong Kong Polytechnic University. His research interests include vehicular ad hoc networks, wireless networks, and parallel and distributed systems. BEHROOZ KHORASHADI received his Bachelor’s degree from the University of California, Berkeley in 2004. He is a Ph.D. graduate from the Department of Computer Science at the University of California, Davis in 2009 and is currently working at Qualcomm’s Bay Area Research and Development (BARD) facility, Santa Clara, California. His research interests include vehicular ad hoc networks, wireless networks, parallel and distributed systems, and network protocol optimization. His current work at Qualcomm includes projects dealing with indoor location services on mobile devices and multicore systems on mobile devices.

73

ESMAILPOUR LAYOUT

2/7/11

10:34 AM

Page 74

ACCEPTED FROM OPEN CALL

TOPOLOGICAL-BASED ARCHITECTURES FOR WIRELESS MESH NETWORKS AMIR ESMAILPOUR AND NIDAL NASSER, UNIVERSITY OF GUELPH TARIK TALEB, NEC EUROPE LTD.

A

ISP3

The authors provide an overview of architectural design approaches for WMN, then summarize the state-of-theart research findings and suggest further topics that need to be addressed. Additionally, we identify three different types of architectures for WMN: campus mesh, downtown mesh, and long haul mesh.

74

ABSTRACT The wireless mesh network and the associated IEEE 802.11s standard have attracted an enormous amount of research in the wireless research community in the past few years. Nevertheless, WMN architecture has not received much needed attention compared to other topics in this area of research. Based on topological differences, various network architectures are possible for WMNs, and we believe such architectures could affect wireless characteristics differently. In this article we provide an overview of architectural design approaches for WMNs, then summarize the state-of-the-art research findings and suggest further topics that need to be addressed. Additionally, we identify three different types of architectures for WMNs: campus mesh, downtown mesh, and long-haul mesh. Furthermore, we discuss and investigate different WMN characteristics that could be affected by commonly deployed architectures. Among the considered characteristics we select routing, network management, and network performance for further analysis, and look at the challenges these architectures face with respect to those characteristics. To illustrate these challenges, we perform a simple experiment to show that LHM and DTM under identical network environments show significant differences in performance parameters such as throughput and delay.

INTRODUCTION Wireless mesh networks (WMNs) have been the subject of much discussion in recent years for the promising improvements they bring to many problems of ad hoc networks, as well as their practicality and effectiveness in areas of difficult terrain. A vast amount of research has been performed in different areas of WMNs, such as network performance, routing, and applications. However, most research work does not identify the type of architecture design in which they work and fail to address whether their solutions could generate the same results when implemented for different types of WMN architectures. WMN architecture could take different topologies based on structural design and orientation of

1536-1284/11/$25.00 © 2011 IEEE

its components with respect to each other and to the network environment. Such variations in the architecture of WMNs could pose fundamental differences in the physical characteristics and performance of the network, which in turn could yield different results in the performed studies. Such differences necessitate architecture-dependent solutions for each problem. Thus, an in-depth study of different types of architectures is important and critical for WMNs, and could potentially elucidate some of the challenges facing WMNs in research areas such as applications, routing, network management, and network performance. In this article we identify three mainstream architectures for WMNs: campus mesh (CM), downtown mesh (DTM), and long-haul mesh (LHM), and explain their fundamental differences as well as some important factors that could be affected by the type of architecture. We then discuss how the dynamics of routing, network management, and performance of WMNs could be affected by the characteristics of the architecture, without getting into detailed analysis or evaluation of those factors. The remainder of this article is organized as follows. In the next section we give a brief literature review. We then identify three distinct types of topological architectures. We then outline major network characteristics that could be affected by the type of the deployed architecture, and show performance variations using system throughput for different architectures. We conclude our findings and make recommendations to be considered in WMN research studies in the final section.

RELATED WORKS The IEEE 802.11s is associated with WMNs, and defines the interoperability between WMNs and wireless ad hoc networks. The IEEE 802.11s medium access control (MAC) layer defines architecture for WMNs including protocols to support broadcast, multicast, and unicast connections for a self-configuring multihop network combining a backbone of fixed wireless mesh routers (WMRs) and clusters of mobile nodes (MNs) [1]. To the best of our knowledge, few research papers exist in the literature clearly addressing

IEEE Wireless Communications • February 2011

ESMAILPOUR LAYOUT

2/7/11

10:34 AM

Page 75

the importance of the architecture of WMNs. Akyldiz et al. [2] identify, for the first time, three distinct architectures for WMN. They classify them as infrastructure/backbone, client, and hybrid WMNs. Their classification is based on the functionality of WMRs vs. MNs. Nodes in the backbone architecture have different functionality than those in the client architecture. In infrastructure WMNs, the authors introduce WMRs in the backbone, connecting to MNs and collecting traffic from them. Then WMRs forward the traffic to the Internet backbone via access points. In client WMNs the authors introduce peer-to-peer connectivity of MNs and totally remove the need for WMRs. The hybrid WMN is basically a combination of the other two architectures. Waharte and Boutaba proposed a tree-based architecture for WMNs [3]. They identify two types of WMRs, access points (APs) and network gateways (NGs). In their tree-based architecture, APs collect traffic from MNs and pass it to NGs in a two-layer hierarchical fashion, where MNs connect to the Internet through NGs rather than exchanging peer-to-peer traffic. Therefore, traffic streams will mainly be directed toward/from the NGs, causing a bottleneck in the access network. Based on these findings, the authors conclude that a tree-based or hierarchical architecture would suit WMNs most perfectly. Seamless mesh (sMesh) architecture is introduced in [4]. The authors propose an architecture entirely based on the backbone and WMRs. It is also totally transparent to MNs. This means the entire WMN is seen as a single AP from the point of view of MNs. They add a connectivity monitoring system to the functionality of WMRs that constantly monitors connectivity power for each MN, and finds the strongest connected AP to which the MN will communicate. In this architecture a higher density of WMRs in the backbone ensures that all MNs always have good connections to the WMN. Recently, BelAir Networks has taken a new approach to categorizing different architectures of WMNs [5]. The authors distinguish different structures based on the number of radios used on WMN nodes. They identify three types of architectures: single-radio, dual-radio, and multiradio mesh networks. Each structure includes a string of APs connecting several clusters of MNs. In this type of WMN categorization, a single-radio wireless mesh has low capacity and does not effectively scale to implement a complete network solution, whereas a dual-radio mesh architecture could scale to a metro dimension, since using different radios, it separates traffic and reduces interference to improve capacity. On the other hand, a multiradio mesh system separates wireless access and backbone by using a dedicated point-to-point link to form a wireless backbone. This provides for a high-capacity system that can support large networks with wireless broadband service for the end user. WMNs can be implemented using different types of wireless technologies such as wireless local area network (WLAN) or (WiMAX, or cellular technologies such as Universal Mobile Telecommunications System (UMTS) or Long Term Evolution (LTE). In recent years, WiMAX networks and a new generation of cellular net-

IEEE Wireless Communications • February 2011

works such as LTE have been proposed as the backhaul for the WLAN to build a new generation of WMNs that integrate various wireless technologies to fulfill one of the promises of fourth-generation (4G) wireless technologies [6, 7]. Wireless technology vendors have recently addressed a new architecture for WMN called MetroMesh, which covers areas on a metropolitan scale [8]. However, they have not addressed other possible architectures to include smaller environments such as campuses or larger environments such as long-haul, and stopped short of characterizing WMNs based on different architectures.

DESCRIPTION OF THE WMN ARCHITECTURES WMN is a self-configured, self-organizing, multipath, and multi-hop network, which consists of, fixed WMRs in the backbone and MNs in the access network. In this section, we first introduce the general or baseline architecture for WMN including all equipment, interfaces, links, and protocols. Then we describe how this architecture is implemented in three completely different operating environments, namely campus, downtown, and long haul mesh. All architectures follow the general structure outlined in the baseline. However, they differ on how their backbone and access networks are topologically designed and connected to each other. These architectures could be combined in order to build more complex hybrid structures, and could be customized to fulfill specific requirements set by clients.

All architectures follow the general structure outlined in the baseline. However, they differ on how their backbone and access networks are topologically designed and connected to each other. These architectures could be combined in order to build more complex hybrid structures.

WIRELESS MESH NETWORK BASELINE ARCHITECTURE The WMN baseline architecture is based on the connectivity and physical orientation of different types of WMN nodes. There are essentially two types of nodes, WMRs and MNs, as illustrated in Fig. 1. WMRs can be classified into three types: backbone mesh router (BMR), access mesh router (AMR), and Internet access point (IAP) depending on in which part of the mesh network they are located. Each MN is connected through an access link to an AMR, which serves as a gateway to the backbone network. The BMR is in the core of the mesh network and does not have access functionality, nor does it have any MNs connecting to it. IAPs serve as gateways to the Internet for the entire WMN. All WMRs have gateway/bridging functionality, which is not required for MNs. Although other technologies such as WiMAX and UMTS have been proposed for the backbone of WMNs, in this study all assumptions are primarily of WMNs based on the IEEE 802.11 WLAN with a/b/g/s amendments. The backbone interfaces are equipped with 802.11a (in the infrastructure mode of operation), access interfaces with 802.11b/g (in the ad hoc mode of operation), and the mesh definitions are based on 802.11s. A general overview of WMNs is presented in Fig. 1. Generally, a WMN has two distinct parts: a backbone network and an access network. The backbone network is a collection of various types

75

ESMAILPOUR LAYOUT

2/7/11

Generally, a WMN has two distinct parts: a backbone network and an access network. The backbone network is a collection of various types of WMRs connected to each other. The access network, on the other hand, is a collection of several clusters of MNs.

10:34 AM

Page 76

next section we identify the fundamental differences in each type of architecture.

of WMRs connected to each other. The access network, on the other hand, is a collection of several clusters of MNs. Each AMR acts as a clusterhead for its corresponding cluster of MNs, collects traffic from MNs and forwards it to other AMRs or BMRs for uplink delivery to the Internet via IAPs. The backbone is a group of different types of WMRs organized in circular, ad hoc, or longitudinal fashion depending on whether the architecture is CM, DTM, or LHM, respectively. WMRs are often fixed and stable, and have access to unlimited power supply. They use proactive routing protocols such as Open Shortest Path First (OSPF). The access network is a group of clusters, each containing several MNs. These clusters are highly mobile and unstable, use temporary sources of power, and use ondemand and ad hoc routing protocols such as Ad Hoc On-Demand Distance Vector (AODV). All the links in the backbone and access networks are wireless, established using 802.11a and 802.11b/g respectively. AMRs are equipped with two physical interfaces; one to connect other WMRs in the backbone, and the other to connect to MNs in the access network. All WMRs in the backbone use multiple virtual interfaces to connect to multiple peer WMRs to build a partial mesh as depicted in Fig. 1. MNs use 802.11b/g contention MAC to access shared channels and connect to their clusterhead AMR. The baseline details mentioned here are shared by the three identified architectures in the next section. The difference is in the topology; the geographical location and physical orientation of the equipment with respect to one another and to the network environment. In the

CAMPUS MESH ARCHITECTURE In the CM architecture (Fig. 2), a limited number of buildings are located in a campus environment, with generally good line of site (LOS), and a central management and administration unit. WMRs could simply be installed on existing infrastructure in campus. The number of MNs in such environments is usually fixed, and MNs have little or no mobility, since the wireless equipment in a campus environment has little or no movement once they are stationed in a location. The entire network is usually under a single administration and is controlled by a single Internet service provider (ISP). Traffic can easily be monitored, and the amount of exchanged traffic can easily be predicted during different time periods, resulting in a more static and predictable network requirement. Thus, the network in a CM architecture is generally easy to deploy, monitor, manage, and upgrade. These features provide a highly flexible environment for deployment of a WMN. Due to single administration provisioning, it is easy to monitor and control different aspects of network management, such as routing, congestion, and interference control. CM is the most flexible environment of the three architectures. It is also the simplest architecture to deploy. Figure 2 shows a typical CM architecture, where there are two rings of WMRs in the backbone: inner and outer rings. The outer ring essentially represents AMRs connecting MNs to the backbone. The BMRs are in the inner ring, with

Public internet L3: Internet gateway network infrastructure

IAP

IAP

Wired links Wireless links 802.11a Routing: OSPF

BMR

L2: Backbone network partial mesh

BMR

BMR

AMR AMR

AMR MN

L1: Access network ad hoc MN

Wireless links 802.11b Routing: AODV

MN

MN MN

MN D_MN S_MN

WMR=Wireless mesh routher (BMR, AMR, IAP) with 2 physical and several virtual interfaces MN=Wireless mobile node with 1 interface

: Wired link : Wireless link S_MN: a Source MN D_MN: a Destination MN

Figure 1. Wireless mesh network baseline architecture.

76

IEEE Wireless Communications • February 2011

ESMAILPOUR LAYOUT

2/7/11

10:34 AM

Page 77

no direct connection to the access network. Due to high concentration of MNs in a CM, some of the inner ring BMRs could also act as AMRs. The inner ring has several major functions; to act as a redundant array of routers or a backup path in case of congestion or disconnection, to provide multipath routing options to the AMRs, and to collect traffic and pass it to Internet access points (IAPs) for Internet connectivity.

Public internet

IAP

AMR

DOWNTOWN MESH ARCHITECTURE In the DTM architecture, many buildings ranging from small to large are scattered over several blocks in a downtown environment, as shown in Fig. 3. This type of architecture introduces many challenges in terms of deployment, management, and control. Generally, LOS is not adequate, and towers are not available or accessible in many locations. The number of MNs varies with time, and MNs tend to change their locations frequently around the downtown area. The network could be under different administration and management, or even different ISPs, which introduces more technical and billing difficulties, such as roaming and network sharing among ISPs. In terms of traffic load and prediction of traffic behavior, this type of architecture is quite different from the CM. In a CM, the majority of traffic is generated by the users on the campus, such as employees in an enterprise or students in a university campus. The number of employees, and the type of operation, application, and usage are well known to the administration over time. DTM, on the other hand is usually the harshest environment, where one does not know what to expect in terms of real-time traffic. The number of users passing by, the type of traffic they are using, the time and day they are passing by, and other factors could very well change the fluctuations of the amount of traffic. Exchanged traffic is highly bursty depending on different client operations, and different times of day, week, month, or even year. Different ISPs provide different types of services to their clients, which makes it extremely difficult for them to coordinate with one other. The complication in management and billing coordination could increase significantly. DTM thus requires a more advanced highcapacity network and costly equipment for deployment. Coordination between ISPs is required, and constant city involvement and licensing issues should be taken into consideration. Such technical and management difficulties could make the solution not as viable as originally thought.

LONG HAUL MESH ARCHITECTURE In the LHM there are no buildings around, but rather a long set of WMRs along a stretch of highway inside a city or in suburban areas, where there is no infrastructure in place, or it is difficult and costly to deploy one. The WMRs could be as far apart as their transmission range allows. Lack of abstraction allows for long LOS using single powerful unidirectional antennas between each pair of adjacent routers. Deployment could prove simple, where antennas are positioned at great heights, kilometers away from each other, depending on their transmission power. A second set of routers (BMRs) could be deployed on the other side of the roadway for redundancy, as depicted in Fig. 4.

IEEE Wireless Communications • February 2011

AMR

BMR

BMR

BMR AMR AMR

Inner ring of BMRs in the backbone used for redundancy

Cluster of MNs

Outer ring of AMRs in the access network

Figure 2. Campus mesh architecture. LHM is a phenomenal solution for networking and communications, because it eliminates the need for extensive and costly infrastructure, as required by traditional wired and wireless technologies. In [9] the authors propose LHM architecture for the first time, along with a routing scheme that involves OSPF and Border Gateway Protocol (BGP) routing protocols in the backbone and AODV with an alternative routing path through the access network. It is generally agreed that multipath routing is a viable solution for WMN routing. On the contrary, with multipath routing, there is not much gain for LHM architecture, since there are not many possible paths between a pair of source and destination nodes. The only backup solution is an array of redundant BMRs that runs along with the main backbone of AMRs.

DIFFERENCES BETWEEN CM, DTM, AND LHM ARCHITECTURES The three identified architectures are fundamentally different in terms of technical details as well as management. Each type of architecture possesses advantages and pitfalls. There are numerous studies that evaluate the performance of various WMNs in their respective environment and unique architecture [2]. Performance of each type of architecture is adjusted and optimized to its specific characteristics and environment. For instance, physical and MAC layer characteristics could be adjusted or links with different capacities could be used to improve performance of one type vs. another. In this article, however, the main objective is to highlight the differences and their consequences without getting into details of performance analysis. In the next section we create a simple case to highlight one of the areas that shows clear differences among architec-

77

ESMAILPOUR LAYOUT

2/7/11

With multipath routing, there is not much gain for LHM architecture, since there are not many possible paths between a pair of source and destination nodes. The only backup solution is an array of redundant BMRs that runs along with the main backbone of AMRs.

10:34 AM

Page 78

Public internet

BMR

IAP

BMR

ISP5

AMR

Public internet

IAP

ISP4 BMR

ISP3

BMR BMR

BMR

AMR

AMR BMR

AMR BMR

BMR AMR

AMR

ISP1

IAP Public internet

ISP2

Figure 3. Downtown mesh architecture. tures. Differences between CM and DTM fall into two major categories: • Differences between the two in ISP management problems • Differences in network issues such as routing, congestion, and interference due to cluster density of MNs The major difference between LHM and the other two architectures consists of the structure itself, the type of equipment, and the lack of multipath routing in LHM. Table 1 summarizes the basic characteristics and differences of the three types of architecture.

PRETENTIOUS NETWORK CHARACTERISTICS In this section we present some factors or network characteristics that could be seriously affected and severely constrained by the implemented architecture in WMNs. We explain the most important factors that have direct effects on the WMN when a particular architecture is used. However, a long list of such factors can be pointed out based on applications and solutions proposed for different areas of WMNs.

ROUTING IN WMNS Several routing protocols are proposed for WMNs such as AODV, Dynamic Source Routing (DSR), and OSPF. OSPF is proposed and

78

deployed in the backbone by many research groups and vendors, such as Nortel Networks. However, a difficulty with implementing OSPF in the backbone lies in the fact that OSPF works in a hierarchical fashion. This structure includes a backbone area (i.e., area0) in the root and several other areas all connected through area0. Hierarchical OSPF areas would nicely fit into the DTM architecture where the main buildings are surrounded by smaller offices, and to a lesser extent for the CM architecture. However, in the LHM architecture it would be impractical to deploy OSPF in a hierarchical fashion. This is due to the fact that traffic from all other areas needs to go through area0 before reaching its destination. This causes an enormous amount of traffic to pass through area0, which results in an area0 bottleneck. Multipath routing in the backbone is introduced to improve the quality and performance of WMN routing. Several multipath routing protocols have been proposed for WMNs along with their extensions as well as new metrics for performance measurement [10]. Multipath routing could be applied in DTM or CM architectures. However, in an LHM architecture where the WMRs are stretched longitudinally along hundreds of kilometers of highways, LHM could not gain much by using multipath routing.

IEEE Wireless Communications • February 2011

ESMAILPOUR LAYOUT

2/7/11

10:34 AM

Page 79

Public internet

IAP Long haul array of access routers (AMRs)

IAP

Redundant array of mesh routers (BMRs)

Long range point-to-point directional antenna

BMR AMR

BMR AMR

Clusters of MNs

Figure 4. Long haul mesh architecture.

Architecture

Area (km2)

Management

ISP

Line of site

Users

Campus

1

Enterprise

Single

Good

100s

Downtown

10

City

Single/multiple

Inadequate

1000s

Long-haul

1000

State

Multiple ISPs

Excellent

Unlimited

Hierarchical OSPF areas would nicely fit into the DTM architecture where the main buildings are surrounded by smaller offices, and to a lesser extent for the CM architecture. However, in the LHM architecture, it would be impractical to deploy OSPF in a hierarchical fashion.

Table 1. Basic characteristics and differences of the identified three architectures.

NETWORK MANAGEMENT One of the issues facing WMNs is network management and network ownership by multiple ISPs. As the size of a network increases, more than a single ISP could get involved in managing the entire network [2, 11]. Traditionally, each WLAN is managed by a single ISP. However, for larger networks that could range over hundreds of kilometers or hundreds of buildings in a downtown area, different parts of the network most likely fall into territories of different ISPs, and could pose complications in management such as roaming, billing and handoff between different networks. Although ISP management could introduce serious issues in LHM and DTM architectures, it does not pose any problem in a CM architecture. In a CM architecture where most of the network management is handled by a single ISP or even occasionally handled locally by in-house network administration, there will be no need for ISP or management coordination and consideration. Therefore, individual solutions could be developed for CM problems that do not concern management issues similar to those of DTM or LHM. Using multiple ISPs in large networks has also been proposed for load balancing as well as other network management issues. In wired networks, BGP is the protocol of choice for network management and employment of multiple ISP solutions. WMN architectures such as LHM and DTM could also use BGP in the backbone, and provide a viable solution and replacement for the last-mile networks.

NETWORK PERFORMANCE Network performance for WMNs has been the subject of much debate in the wireless research community in the past few years. Many studies

IEEE Wireless Communications • February 2011

have introduced new and improved solutions throughout various TCP/IP layers to optimize network performance for WMNs [2]. Others studied performance issues that come from interference. Usually, MNs in a cluster engage in communication with AMRs and with other MNs, causing multiple levels of interference. Interference could be a major obstacle in a DTM network where backbone routers are closer to each other and ad hoc MNs are moving. On the other hand, when LHM architecture is deployed in a suburban area, there are only few WMRs, and they are deployed far apart. In this kind of structure, interference is minimal and does not affect the functionality of other nodes and routers. In [9] the authors show that the performance degradation is caused by contention among MNs. In this section we evaluate and compare the performance of various architectures. We perform throughput and delay measurements to highlight differences among the different architectures. One can find more detailed performance analysis for WMNs in the literature [2, 9]. Our throughput hypothesis states that performance degradation is due to reduction in throughput caused by contention among MNs for medium access on the link to the AMR, and such a performance measure is highly affected by the orientation of the AMRs and MNs, and the type of architecture in the WMNs. Therefore, it is expected that throughput values for different architectures show significant differences. A simulation model and experiments were implemented and carried out in the OPNET modeler 14.5 PL1. We implemented CM, LHM, and DTM architectures with the same network environment such as amount of equipment, applications, and traffic. In each model there are 12 routers and six clusters with two MNs in each

79

ESMAILPOUR LAYOUT

2/7/11

10:34 AM

Page 80

b/s (x106) 6 MNs

4 MNs

3.0

2.5

4 MNs 6 MNs

2.0

2 MNs 1.5

2 MNs

1.0

0.5

0.0 0.0

1.0

2.0

3.0

4.0

0.0

1.0

2.0

3.0

Min

Min

(a)

(b)

4.0

Figure 5. Overall system throughput for architectures: a) LHM; b) DTM. cluster. OPNET has different options for generating traffic at various layers with unique specifications, such as mobile ad hoc network (MANET) traffic. MANET traffic is generated between a pair of source and destination MNs. In each simulation experiment we gradually increase the number of MNs in the source cluster from two to six and monitor the changes in the system throughput values. Figure 5a shows the changes in throughput for LHM architecture. As the number of MNs increases from two to four, there is a twofold increase in the system throughput. However, by further increasing MNs to six, the contention problem causes the throughput to decrease to just above that of two MNs. Figure 5b shows corresponding results for DTM architecture. These results are significantly different from those of LHM. As we increase the number of MNs from two to four, the throughput increases by 25 percent. As we increase the number of MNs to six, the throughput increases at a steady rate by another 25 percent. This shows that the contention is not affecting the network performance as severely as in the LHM architecture. This linear performance change can be attributed to several reasons. Intuitively, in a dense DTM environment when an MN has many contending neighbors, it seeks other paths to reach the destination, thereby increasing the system throughput at a more linear way than the LHM Architecture. Similar results were achieved when we tried the link throughput between the D_AMR and D_MN. In terms of delay analysis, as illustrated in Fig. 6, in both cases the delay is the least for the scenario with two MNs. As we move to four MNs, the delay increases for DTM at a slow rate by less than 5 percent for four MNs and less than 8 percent for six MNs. However, in case of

80

LHM, the delay follows a slightly higher increase to over 10 percent. However, as we move to 6 MNs, the delay increases dramatically to over 70 percent compared to the case of two MNs. The results show that there is no proportional relation between the number of MNs and the system throughput or delay in LHM compared to DTM. Clearly, contention exists in both cases. However, its effect becomes insignificant when other architectural features as well as other constraints are included in the equation. In this article we are not trying to generalize the results or declare which type of architecture is best. We merely highlight the impact of these architectures on network performance. Regardless of the reasons, the results support the hypothesis that different types of architectures generate significantly different results in performance analysis and measurements. Furthermore, they show that differences are so significant that proof of a point in one type of architecture could not support the same point in another type of architecture.

CONCLUSION AND RECOMMENDATIONS Recent research on WMNs has not adequately addressed various WMN architectures. A WMN can assume different types of network architectures, and the type of architecture can affect wireless characteristics differently. In this article we identify three types of architectures and various network characteristics that can be differently affected by each type. We point out three major areas in which the differences are highlighted among the three architectures, and show by simulations that performance measures such as throughput and delay could vary significantly depending on the underlying architecture. We recommend that WMN architecture should be considered as an integral part of

IEEE Wireless Communications • February 2011

ESMAILPOUR LAYOUT

2/7/11

10:34 AM

Page 81

Seconds 1.0

0.8 6 MNs 2 MNs 0.6

6 MNs

4 MNs

4 MNs

2 MNs

0.4

0.2

0.0 0.0

1.0

2.0

3.0

4.0

5.0

0.0

1.0

2.0

3.0

Min

Min

(a)

(b)

4.0

5.0

Figure 6. End-to-end delay results for a)LHM; b) DTM architectures.

research activity, and that experiments in this area clearly distinguish and identify the scope of the network as well as its type of architecture. Various solutions in different areas of WMN have been proposed in recent years, and experiments have been developed to prove or disprove proposals based on single architecture types. We further recommend investigating the validity of such proposals under various types of architectures. In the future, we plan to propose standard structural definitions for the identified architectures and other hybrid architectures that could be built based on the three mainstream architectures mentioned in this article. We will also identify a complete list of factors at different layers (e.g., application, network, and MAC layers) that could be affected by the architecture type. We would further like to investigate the accuracy of several proposed solutions based on different architectures, and investigate their reliability to see if they apply to all kinds of architectures or only one.

REFERENCES [1] IEEE 802.11s, “Amendment for Mesh Networking to the IEEE802.11”; http://ieeexplore.ieee.org/xpl/standards.jsp. [2] I. F. Akyildiz, X. Wang, and W. Wang, “Wireless Mesh Networks: A Survey,” Comp. Net., vol. 47, no. 4, 2005, pp. 445–87. [3] S. Waharte and R. Boutaba, “Tree-Based Wireless Mesh Network Architecture: Topology Analysis,” Proc. 1st Int’l. Wksp. Wireless Mesh Net., Budapest, Hungary, July 2005. [4] Y. Amir et al., “Fast Handoff for Seamless Wireless Mesh Networks,” Proc. 4th MobiSys, Uppsala, Sweden, 2006, pp. 83–95. [5] BelAir Networks, “Capacity of Wireless Mesh Networks,” White Papers, 2008; http://belairnetworks.com/ resources/. [6] M. Cesana, “The Evolution toward Heterogeneous High Speed Mesh Networking,” ICT Mobile SUMMIT-CARMEN Project Wksp. Carrier Grade Mesh Net., Santander, Spain, June 2009. [7] N. Bayer et al., “Towards Carrier Grade Wireless Mesh Networks for Broadband Access.” German Ministry of

IEEE Wireless Communications • February 2011

Education and Research, 2009; http://www.deutschetelekom-laboratories.de/~karrer/papers/opcomm06.pdf. [8] StrixSystems, “WiFi Mesh Business Case Technical Research Paper,” 2009; http://www.strixsystems.com/ case-studies/WiFi-Mesh-business-case.asp. [9] A. Esmailpour et al., “Ad-Hoc Path: An Alternative to Backbone for Wireless Mesh Networks,” Proc. IEEE ICC, Glasgow, U.K., June 2007, pp. 3752–57. [10] N. S. Nandiraju, D. S. Nandiraju, and D. P. Agrawal, “Multipath Routing in Wireless Mesh Networks,” Proc. IEEE MASS, Oct. 2006, pp. 741–46. [11] B. V. Daggett, “Localizing the Internet: Five Ways Public Ownership Solves the U.S. Broadband Problem,” Inst. Local Self-Reliance, tech. rep., Jan. 2007.

BIOGRAPHIES AMIR ESMAILPOUR ([email protected]) is currently a Ph.D. candidate at the University of Guelph, Canada. He received his Bachelor of Science at the University of Ottawa and Master’s of Applied Science from Ryerson University, Toronto, Canada. He worked at Nortel Networks as a software engineer and Daimler Chrysler as a network engineer for seven years, and returned to academic studies and research to pursue his Ph.D. degree. His area of research is in wireless mesh networks and quality of service for the IEEE 802.16 standard. He is presently working on his Ph.D. thesis in radio resource management and QoS for mobile WiMAX. NIDAL NASSER ([email protected]) completed his Ph.D. in the School of Computing at Queen’s University, Kingston, Ontario, Canada, in 2004. He is currently an associate professor in the Department of Computing and Information Science at the University of Guelph. He is an associate editor of the Journal of Computer Systems, Networks, and Communications, Wiley’s International Journal of Wireless Communications and Mobile Computing, and Wiley’s Security and Communication Networks Journal. TARIK TALEB [S‘04, M‘05, SM‘10] ([email protected]) is currently working as a senior researcher at NEC Europe Ltd, Heidelberg, Germany. Prior to his current position until March 2009, he worked as am assistant professor at the Graduate School of Information Sciences, Tohoku University, Japan. From October 2005 to March 2006 he worked as a research fellow with the Intelligent Cosmos Research Institute, Sendai, Japan. He received his B.E. degree in information engineering with distinction, and M.Sc. and Ph.D. degrees in information sciences from GSIS, Tohoku University, in 2001, 2003, and 2005, respectively.

81

MARCO LAYOUT

2/7/11

10:48 AM

Page 82

ACCEPTED FROM OPEN CALL

SYNCHRONIZATION OF MULTIHOP WIRELESS SENSOR NETWORKS AT THE APPLICATION LAYER ÁLVARO MARCO AND ROBERTO CASAS, UNIVERSITY OF ZARAGOZA JOSÉ LUIS SEVILLANO RAMOS, UNIVERSITY OF SEVILLE ’

VICTORIA’N COARASA AND ANGEL ASENSIO, UNIVERSITY OF ZARAGOZA MOHAMMAD S. OBAIDAT, MONMOUTH UNIVERSITY

N4

2

N3 N8 N7

6

0

N11

N15

N12 N16

The authors present a method that allows accurate synchronization of large multi-hop networks, working at the application layer while keeping the message exchange to the minimum.

82

ABSTRACT Time synchronization is a key issue in wireless sensor networks; timestamping collected data, tasks scheduling, and efficient communications are just some applications. From all the existing techniques to achieve synchronization, those based on precisely time-stamping sync messages are the most accurate. However, working with standard protocols such as Bluetooth or ZigBee usually prevents the user from accessing lower layers and consequently reduces accuracy. A receiver-to-receiver schema improves timestamping performance because it eliminates the largest non-deterministic error at the sender’s side: the medium access time. Nevertheless, utilization of existing methods in multihop networks is not feasible since the amount of extra traffic required is excessive. In this article, we present a method that allows accurate synchronization of large multihop networks, working at the application layer while keeping the message exchange to a minimum. Through an extensive experimental study, we evaluate the protocol’s performance and discuss the factors that influence synchronization accuracy the most.

tasks) time synchronization [1]. From a practical point of view, WSNs are networks composed of a large number of small devices that take measurements, process them, and communicate with other devices coordinating their operations. This collaboration enables a complex sensing task, named data fusion [2]. Data fusion requires synchronization for two tasks: time scheduling and timestamping. The first is needed when the nodes coordinate to perform cooperative communications. The second is commonly used when data is fused taking into account the collecting instant; for example, to perform event detection, tracking, reconstruction of a system’s state for control algorithms, and offline analysis. In this article we describe in detail and evaluate the use in WSNs of the Multihop Broadcast Synchronization (MBS) protocol. An early version of this scheme was first introduced by us in [3]. The proposed protocol is very well suited for WSNs and fills an existing gap when synchronizing WSNs with a global network time. As seen in the following sections, MBS helps to achieve high accuracy and energy efficiency when timestamping at lower layers is not possible in multihop networks.

INTRODUCTION

RELATED WORK

Time synchronization entails an important function in wireless sensor networks (WSNs), and this function can be performed at different layers depending on the objective of synchronization. For instance, sharp timing is fundamental at low layers as it helps increase data rates (short bit times), and enhance noise immunity (frequency hopping) and time-division multiple access (TDMA)-based scheduling. Furthermore, since radio is usually the most energy-consuming part of a node, keeping the nodes awake the minimum required time to exchange data is a common practice that requires synchronization. Synchronization is also of great interest at higher layers. Akyildiz et al. identify in the application layer the sensor management protocol, which includes (among other administrative

Creating a common temporal reference using the nodes communication capabilities has been widely studied [2]. According to the strategy, we could distinguish between a posteriori and a priori synchronization [4]. A posteriori methods keep devices’ clocks running free, gathering information between relative clocks and rearranging timestamps once the measurement processes are finished. These methods are usually the most energy-efficient because they optimize the number of messages exchanged, but they do not offer real-time capabilities. On the other hand, a priori methods overcome this by synchronizing all the nodes with a common time reference (global network time [GNT]) using regular clock corrections. A common drawback of these techniques is overload of the network

1536-1284/11/$25.00 © 2011 IEEE

IEEE Wireless Communications • February 2011

MARCO LAYOUT

2/7/11

10:48 AM

Page 83

due to the messages required to estimate the communication delays. Another key issue is whether there is a sender transmitting the current clock values as timestamps (sender-to-receiver) or not (receiver-toreceiver). According to the time analysis performed by Maróti et al., the most significant delays when transmitting messages over a wireless link are those from the send, receive, and access processes [5]. The problem with senderto-receiver methods is the uncertainty time introduced by the send and access processes. As a receiver-to-receiver method, the Reference Broadcast Synchronization (RBS) protocol proposes the use of reference broadcast messages to establish a common time reference and get rid of the transmitter-side non-deterministic error sources (it is assumed that all devices listening to the broadcast get the message at the same time). This eliminates the time uncertainty introduced by the send and access processes, and sets a temporal reference shared by all the nodes [4]. The biggest drawback of receiver-to-receiver synchronization methods is how to propagate the local timestamps of the broadcast-receivers to set a GNT. Elson and Estrin propose a post facto synchronization method, that is, a method that performs synchronization only when it is needed [6]. The scattering method they propose does not give a common time reference to the broadcast sender; it only synchronizes receivers. However, various authors point out the need of setting a network time to propagate the synchronism over a multihop network using broadcasts [4, 7]. Thus, nodes in the broadcast domain need to share timing information among them to determine the GNT. On the other hand, Timing-sync Protocol for Sensor Networks (TPSN) is a sender-to-receiver protocol that achieves real time with high accuracy optimizing message exchange. It avoids the indeterminism working at the medium access control (MAC) layer to precisely timestamp messages at the exact moment they are sent [8]. The Flooding Time Synchronization Protocol (FTSP) also uses MAC layer timestamping at both the sender and receiver sides. The protocol proposes a multihop propagation scheme that does not need any initial configuration to propagate synchronization info. This ad hoc structure also enables dynamically overcoming node and link failures in the network [5]. However, these sender-to-receiver synchronization schemes require accessing the lower layers, which is not always possible when using standard or complex protocols. Current WSN applications are developed using a wide variety of hardware, software, and communication protocols. Some of them are proprietary; others use common development platforms that have become de facto standards (Motes, i-beans); and others implement the two current standards suitable to be used in WSNs: ZigBee and Bluetooth. In all these cases accessing the lower layers is not possible, and this prevent us from using sender-to-receiver synchronization. The Multihop Broadcast Synchronization protocol is a receiver-to-receiver synchronization scheme that nonetheless obtains a GNT working at the application layer. MBS is suitable for

IEEE Wireless Communications • February 2011

large multihop networks, keeping the number of messages in the same order of magnitude when compared to the sender-to-receiver methods.

MULTIHOP BROADCAST SYNCHRONIZATION PROTOCOL Many of the functions performed by sensor nodes require a precise time measurement (e.g., bit time calculation in communications). The devices commonly used to provide an accurate time base are oscillators. Although these clocks ideally provide a heartbeat at a constant rate, all of them present a frequency tolerance (whose limits are provided by its manufacturers) that is also slightly affected by the operation temperature and aging. Thus, two clocks, even manufactured in the same process, will not have a priori the same frequency. This way, nodes in a WSN measure time with oscillators at slightly different rates. This divergence, called clock skew, can go up to 150 parts per million (PPM) (i.e., each second the nodes will commit an error than can go to 150 μs). Apart from the clock skew, there is an offset among clocks because each node starts at a different instant. Given a node i, with clock skew si and offset ki, its clock reading ti can be written as ti = sit + ki, i = 1…n.

Although these clocks ideally provide a heartbeat at a constant rate, all of them present a frequency tolerance that is also slightly affected by the operation temperature and aging. Thus, two clocks, even manufactured in the same process, will not have a priori the same frequency.

(1)

Thus, synchronizing the clock of node i implies estimating and compensating clock skew and offset (s i, k i). The most used procedure to perform this adjustments is broadly described in the literature [2, 4, 5, 9]. There is a reference clock tr to which all the nodes are synchronized. A sync-point k is defined as a pair of timestamps collected at the same time t k in the reference node and in the nodes that want to be synchronized: {t ik , t rk }. Once each node stores several sync-points at different instants, the offset and skew (ki*, si*) differences with the reference can be calculated using linear regression: si* =

∑ k (trk − tr ) (tik − ti ) ∑ k (trk − tr ) ki* = ti − si* t r ,

2

(2)

where the bar indicates average. This way, every node can estimate the global time (tr*) from its local clock: t r* =

ti − ki* si*

.

(3)

Once a node is synchronized, it can propagate the estimated t r* to others, creating new sync-points that spread the GNT in multihop networks [5]. Strictly speaking, it is not possible to obtain a sync-point at the same instant in two nodes not physically linked; there will be unknown delays in the synchronization message exchange. While deterministic uncertainties in wireless links slightly affect the synchronization accuracy, nondeterministic ones drastically reduce it [1, 4, 5,

83

MARCO LAYOUT

2/7/11

10:48 AM

Page 84

N4

N2

N1

N3 N8

N5 N9 N13

N7

N6

N10 N14

N11

N15

Synchronization link Global time provider

N12 N16

Timestamper Propagator Common node

Figure 1. A grid network where propagator nodes send broadcast messages, and time stamper nodes reply to them with the arrival time of the broadcast. All the nodes also have direct communication links with their neighbors.

8]. TPSN and FTSP eliminate the biggest uncertainties (send, receive, and access time) by timestamping at the MAC layer the send and receive instant of a message. This way, to obtain the required sync-points accurately, only one message is necessary. When access to the low layers is not possible, reference broadcast messages eliminate the uncertainty at the sender side. Unfortunately, although the reception instant of a broadcast message is tight, this procedure does not provide the required sync-points directly because the sender is not synchronized [4, 10]. To the best of our knowledge, existing methods require additional message exchange between every receiver node to set the GNT. This makes multihop propagation very inefficient [4, 7]. Our MBS protocol obtains sync-points using reference broadcasts with considerably less message overhead than other methods that also use reference broadcasts, and manages to synchronize all the nodes (even the broadcast sender) as well, making multihop propagation easy.

PROTOCOL DESCRIPTION To illustrate how MBS works, we consider the example network in Fig. 1. Nodes in MBS perform two different tasks. Propagators are those that spread the GNT broadcasting synchronization messages. Timestampers are nodes that can notify the propagators about the timestamp when the previous broadcast message arrives. In Fig. 1, N3, N6, N9, and N11 are propagators. N1, N7, and N10 are timestampers; the provided timestamps are referred to the GNT, so they have to be synchronized when notifying the corresponding propagator node. To overcome this requirement when initiating the process, N1 will be the node whose local clock will be the GNT (i.e., N1 is the global time provider [GTP]). The synchronization will be made in two hops, the same number of hops FTSP would need. In the first hop, N6 will synchronize the nodes in the dashed ring to N1’s clock (using a technique similar to that of RBS). Then, N3 using N7 as timestamper, and N9 and N11 using N10, will synchronize the nodes in the dotted rings and propagate the GNT one hop away. N6 must also be synchronized, and, as it is the only propagator connected to the GNT, it will be synchronized in the second hop by any of the other propagators.

84

This process will be done using two different messages: SyncBC and TimeUC. The first type is broadcast by propagators and is equivalent to the reference broadcast in RBS: messages that trigger timestamping of the receiving instant at the sender side. It contains three fields: the propagatorID identifying the sender, the sequenceNumber of the message, and the timeStamp when the previous SyncBC arrived. These messages are used to propagate the GNT the same way as synchronization messages in FTSP [5]. TimeUC messages are unicast messages used by timestampers to notify the propagators about the time when the last SyncBC arrived. They have the following fields: the timeStamperID identifying the sender, and the corresponding propagatorID, sequenceNumber, and timeStamp about which they are informing. Now we explain the sequence to synchronize the network in Fig. 1. We indicate the message sent specifying the fields: SyncBC (propagatorID; sequenceNumber; timeStamp) and timeUC (timeStamperID; propagatorID; sequenceNumber; timeStamp). The first hop would be as follows: 1. N6 initiates the synchronization process by sending a SyncBC (N6; 0; void) message. The nodes that receive the message (N1, N2, …) timestamp the arrival of the SyncBC from node 6 with sequence number 0; that is, TSN1{N6, 0}, TSN2{N6, 0}, …. 2. N1 informs N6 about the timestamp when it received the last SyncBC message: TimeUC (N1; N6; 0; TSN1{N6, 0}). 3. N6 sends the timestamp of the previous SyncBC message and sets a new reference point for timestamping: SyncBC (N6; 1; TSN1{N6, 0}). At this instant, all Ni neighbors of N6 have their respective sync-points from the first SyncBC: [TSNi{N6, 0}, TSN1{N6, 0}]. N1 does not need it because it rules the GNT. From now on, steps 2 and 3 will be repeated, causing all nodes in range of N6 to have a collection of sync-points. Then, using linear regression, they get synchronized, calculating their offset and skew differences to the reference clock. In that first hop, all nodes within the dashed ring will be synchronized to N1’s clock. Note that N6 does not have the global time. To fix this and to propagate the clock one hop away, any other propagator, such as N3, initiates the above described sequence. The subsequent hops will be performed following the same procedure. Each propagator broadcasts SyncBC messages including the timestamp of the previous SyncBC message. Timestampers notify the propagators about the last timestamp. All nodes in a range get a collection of sync-points that allow them to synchronize to GNT. Accuracy of the sync-points will be degraded as nodes are farther away from the clock generator (N1), similar to sender-to-receiver methods. When synchronizing the nodes situated within the dotted rings, the timestampers used by the propagators (N7 and N10) are one hop away from the GNT (N1), which will increase the error committed. In case of near failure (i.e., low batteries), propagators and timestampers can transfer their responsibilities to neighboring nodes. In any case, similar to the scheme proposed by Maróti

IEEE Wireless Communications • February 2011

MARCO LAYOUT

2/7/11

10:48 AM

Page 85

et al., a network could be autonomous and selfhealing in terms of synchronization using node identifiers to automatically assign roles [5].

PRACTICAL ISSUES Earlier we described the MBS protocol without taking into account practical issues like how to initialize the algorithm, how to determine the role to be performed by every node, and so on. To understand the general case, let us first consider several examples. Let us assume that we have a node, N1, that is already synchronized, and a couple of nodes, N2 and N3, that are not yet synchronized. In order to synchronize N2 and N3 with respect to N1, an intermediate node, N4, that can send broadcast messages to N1, N2, and N3 is needed. Thus, N4 will act as a propagator node to its neighbors, and N1 will act as the timestamper for N4. Thus, the general rule is that all the neighbors of the neighbors of N1 can be synchronized with respect to N1 provided that these nodes have a one-hop estimation of the time of N1. In the same manner, N1 will have a one-hop estimation of the time in the node with respect to which it is synchronized, which in turn will have a one-hop estimation, and so on until the GTP is reached. The number of hops needed to reach the GTP can be used to define the quality of a node synchronization, which we call HopId. Now consider a node N1 that is not synchronized. In order to become synchronized, a nearby — synchronized — node must assume the role of timestamper. Analogous to the previous case, all the neighbors of the neighbors of N1, which are already synchronized, can behave as timestampers. Therefore, the one with the best GNT estimation (i.e., the lowest HopId) must be chosen as timestamper. If more than one node have the lowest HopId, the one that has more synchronizable nodes will be chosen, and the node that is neighbor of both the timestamper and N1 will act as the propagator. In order to improve performance, if a node listens to more than one propagator, it must only use sync messages from the propagator whose timestamper has lower HopId. Also, if a node is acting as timestamper for a propagator, it cannot use incoming SyncBC messages from this propagator to synchronize itself. Now, the mechanism should be clear. First, all nodes but the GTP (with HopId equal to 0) are not synchronized. Nodes close to the GTP will be synchronized with respect to the GTP, and they will be assigned a HopId equal to 1. Synchronization will spread across the entire network following the rules discussed above. Still, several criteria are possible to set the GTP, such as minimizing the total synchronization error expressed as the sum of the HopId of every node, maximizing the number of nodes with lower HopId, or setting an upper bound for the HopId.

MBS PROTOCOL EVALUATION AND COMPARISON We have evaluated the MBS protocol in two different architectures. Both are used to build multihop WSNs following two standard protocols: ZigBee and Bluetooth. In the case of ZigBee,

IEEE Wireless Communications • February 2011

we use a common platform: an Atmel microcontroller (ATMega128) with the Chipcon’s CC2420 transceiver [9]. Developing the entire stack to make the network ZigBee-compliant requires a lot of work, thus we used the EmberZNet embedded software. This way, we built a multihop, auto-routing and self-healing ZigBee network working at the application layer. For the Bluetooth architecture, a Microchip PIC16F876 microcontroller manages a Mitsumi Bluetooth module (WML-C20) through HCI, the standard Bluetooth Host Controller Interface. This design enables us to access low-power modes and implement tree topology networks using scatternets. To evaluate MBS’s performance, we connect every node to a wired bus, where one of them periodically generates a pulse. All the nodes, which are synchronized with MBS, timestamp the receiving instant with their GNT estimation and send back the data to the PC, where the timestamp differences with respect to the GTP are considered to obtain the synchronization error.

SINGLE-HOP SYNCHRONIZATION

All the nodes, which are synchronized with MBS, timestamp the receiving instant with their GNT estimation and send back the data to the PC, where the timestamp differences with respect to the GTP are considered to obtain the synchronization error.

When synchronizing wireless nodes, all methods use one of the following strategies: to timestamp at the sender and receiver side, or use reference broadcasts, timestamping only at the arriving side. In Table 1 we compare the alignment errors of some synchronization schemes presented in the references. The timing accuracy among nodes depends mainly on the hardware and firmware architecture: how the sending and arriving moments are detected, and which times (propagation, access, etc.) are affected and their uncertainty. Errors shown in Table 1 determine the accuracy of each sync-point that will be used to perform the linear regression in Eq. 2. Of course, the fewer the errors, the more accurate the synchronization. Other factors that also affect the estimation are the following: • The distribution of the errors (uniform, Gaussian, etc.) will determine how the linear regression eliminates them and the quality of the clock estimation. The time difference between reception instants of broadcast messages follows a Gaussian distribution with the architectures described before [3, 4]. • The local oscillator drift is influenced by the initial accuracy (difference between the oscillator output frequency and the specified frequency at 25°C at the time of shipment by the manufacturer), temperature stability, and aging. Its behavior can also condition the precision and the timing lifetime. • Finally, the frequency and number of syncpoints used in the estimation will determine the expected accuracy. As stated by van Greunen and Rabaey, not all sensor networks applications have the same sync needs in terms of accuracy [11]. In order to provide the reader with some guidelines that could help in deciding on the best suited hardware (transceivers, crystals, etc.) and firmware (sync-message rate, number of data points to perform regression) in each application, we have characterized synchronization behavior in several

85

MARCO LAYOUT

2/7/11

10:48 AM

Page 86

Average error (μs)

Worst case error (μs)

Sender — Receiver synchronization ZigBee (Motes 2.4 GHz) [9]

14.9

61.0

TPSN (Motes 916 MHz) [8]

16.9

44.0

FTSP (Motes 433 MHz) [5]

1.4

4.2

Receiver — Receiver synchronization RBS (Motes) [4]

21.9

93.0

MBS (Bluetooth)

4.5

18.0

MBS (ZigBee)

22.2

52.0

Table 1. Comparison of alignment error in synchronization methods. scenarios. We compared the two architectures described above, Bluetooth and ZigBee, each with different alignment errors, as shown in Table 1. Both alternatives have been tested with two local oscillators with different frequency stabilities of 40 PPM and 150 PPM. We have also changed the synchronization rates (30 s and 300 s) and number of sync-points used in the regression (3, 6, 8, 12, 20, and 50). In Table 2 we show the average and maximum synchronization error with 95 percent probability for both scenarios. The first interesting result is that synchronizaZigBee 40 PPM

N=3

N=6

N=9

N = 12

N = 20

N = 50

tion precision depends mainly on the number and accuracy of the sync-points used to estimate parameters by regression. Thus, if we have very low computation resources or need low-accuracy timing we can use a few sync-points (i.e., a lightweight method) [11]. In contrast, if the highest accuracy is needed, we need powerful hardware to use the maximum number of points in the regression. Contrary to what people might expect and agreeing with what Maróti et al. have found out, synchronization rate and oscillator accuracy barely affect precision [5]. This makes sense if we consider the local clock stable in the short to medium term, something that fortunately is so in most cases. Aging has a negligible effect on stability: one to three PPM each year. Temperature influences clock drift more severely: tens of PPMs in the operating temperature range. In worst cases, this translates into an error of 1 μs/°C. The reduced influence of synchronization rate can be used in many ways. By increasing it, we can reduce the initial settling time in multihop networks, quickly synchronize a new node, or maintain accuracy in sudden temperature variations. Lowering the rate will be useful to minimize extra traffic and reduce power consumption. Despite the results shown in Table 2, the synchronization interval cannot be as long as we want. Linear regression estimates the skew and offset of the local clock referred to the GNT, and there will always be errors. According to Eq. 1, the more time from the last resynchronization, the larger the error of

ZigBee 150 PPM

Bluetooth 40 PPM

Bluetooth 150 PPM

Avg.

Max.a

Avg.

Max.a

Avg.

Max.a

Avg.

Max.a

tSync = 30 s

39.26

105.20

39.45

106.42

7.67

20.92

8.10

21.70

tSync = 300 s

39.99

108.09

39.59

109.05

8.00

21.61

7.73

20.88

tSync = 30 s

21.12

52.65

22.06

55.23

4.17

10.30

4.34

10.77

tSync = 300 s

21.90

53.52

21.29

52.07

4.41

10.99

4.36

10.93

tSync = 30 s

17.27

40.95

16.58

40.66

3.30

8.19

3.30

8.25

tSync = 300 s

16.10

39.38

15.72

38.90

3.28

8.22

3.36

8.39

tSync = 30 s

13.33

32.83

14.39

35.13

2.82

6.84

2.79

6.89

tSync = 300 s

13.99

34.52

13.77

34.87

2.73

6.62

2.68

6.63

tSync = 30 s

11.29

27.20

9.83

24.23

1.96

4.71

2.13

5.14

tSync = 300 s

10.61

26.13

10.46

25.51

2.19

5.29

2.13

5.30

tSync = 30 s

7.03

17.46

6.77

15.58

1.23

2.93

1.27

3.10

tSync = 300 s

6.22

14.80

7.54

18.43

1.30

3.27

1.32

3.21

N is the number of sync-points used to perform regression, and tSync is the synchronization interval. a Maximum synchronization error with 95% probability.

Table 2. Synchronization error in the MBS method with one hop (in μs).

86

IEEE Wireless Communications • February 2011

2/7/11

10:48 AM

Page 87

this estimation. Figure 2 shows the synchronization error between two Bluetooth nodes and the reference node. Here we used sync messages every 30 s, and after 500 messages (about 4 h) the synchronization process was stopped. We can see how after the stopping instant, the synchronization error of each node grows depending on the last estimation of the skew and offset. Additional results may be found in [3].

80 60 Error (μs)

MARCO LAYOUT

20 0 0:00

MULTIHOP GNT PROPAGATION

CONCLUSIONS Local clocks of nodes in wireless sensor networks have different offset and accuracy. Synchronization is mainly achieved by collecting sync-points (pairs of timestamps collected at the same time in the reference node and in the node that wants to be synchronized) and performing

IEEE Wireless Communications • February 2011

1:00

2:00

3:00

4:00 5:00 Time (hours)

6:00

7:00

8:00

Figure 2. Illustration of synchronization error vs. time with sync messages every 30 s and 20 pairs used to perform regression. Synchronization process stops after 500 sync messages.

500

N=3 N=6 N=9 N = 12 N = 20 N = 50

450 400 Synchronization error (μs)

Cumulative errors when propagating the network time only depend on the estimation’s accuracy of the corresponding timestamper’s clock. When the timestamper is the one setting the reference, the precision will be as described earlier. Differences arise when it owns an n-hop estimation of the network clock. We have evaluated the behavior of MBS in two different scenarios. With the same philosophy of the single-hop case, we have tested its performance with different numbers of syncpoints. The first scenario is a four-hop Bluetooth scatternet. Bluetooth networks are made up of piconets, i.e., star networks with one master and up to seven slaves, where only the master can send broadcast messages. Piconets form scatternets using nodes playing both master and slave roles. Time-stampers are nodes that must be slaves in two different piconets. In the case of ZigBee, we implemented a mesh network similar to that of Fig. 1 but having four-hop depth. Synchronization errors for the Bluetooth and ZigBee networks are shown in Figs. 3 and 4, respectively. As in the case of one hop, synchronization error depends chiefly on the number of points used to perform regression and the alignment error in the establishment of sync-points. We can see how as the number of hops increases, synchronization error becomes sensitive to the number of sync-points used. Again, if the application needs higher synchronization accuracy than the alignment error between nodes, it is possible to reduce error by increasing the number of sync-points to perform regression. We find no point in comparing precision among methods because it mainly depends on the accuracy estimating sync-points and the number of pairs used. That is to say, architecture (communication transceiver, protocol, memory available, etc.) will be much more relevant than the synchronization protocol used. On the other hand, the amount of messages needed will have a big influence on the applicability of the method. This is just one of the strongest points of MBS; it drastically reduces the number of messages compared to other receiver-to-receiver protocols [4, 7]. Indeed, it is on the same order of magnitude as FTSP (the most efficient senderto-receiver method) [5].

40

350 300 250 200 150 100 50 0

0

1

3

2

4

Sync level

Figure 3. Average synchronization error for Bluetooth network with 4 hops and sync messages every 30 s; here N is the number of points used to perform regression, and the dashed bars represent maximum error with 95 percent probability.

linear regression to compensate for differences among nodes’ clocks. Medium access time is a non-deterministic error that hinders accurate timestamping of transmission instants when working at the highest layers of protocol stacks. In these cases, several techniques based on a receiver-to-receiver scheme have to be used. Nevertheless, these methods set a network time shared by all the nodes (including the broadcast senders) at the expense of a high network load. This drawback may be inadmissible for large networks. In this article we have presented MBS, a multihop broadcast synchronization protocol that is able to efficiently set a common global time. The key issue of the technique is that each reference broadcast informs about the timestamp of the previous one. This way, message exchanging is minimized, similar to other sender-to-receiver methods. We have implemented the MBS protocol in two different architectures (Bluetooth and Zig-

87

MARCO LAYOUT

2/7/11

1000

Page 88

[7] S. PalChaudhuri, A. Saha, and D. B. Johnson, “Adaptive Clock Synchronization in Sensor Networks,” Proc. 3rd IEEE IPSN, Berkeley, CA, 2004, pp. 340–48. [8] S. Ganeriwal, R. Kumar, and M. B. Srivastava, “TimingSync Protocol for Sensor Networks,” ACM SenSys, Los Angeles, CA, 2003, pp. 138–49. [9] D. Cox, A. Milenkovic, and E. Jovanov, “Time Synchronization for ZigBee Networks,” Proc. 37th South-Eastern Symp. Sys. Theory, Tuskegee, AL, 2005, pp. 135–38. [10] R. Casas et al., “Synchronization in Wireless Sensor Networks using Bluetooth,” 3rd Int’l. Wksp. Intelligent Solutions in Embedded Sys., 2005, pp. 79–88. [11] J. van Greunen and J. Rabaey, “Lightweight Time Synchronization for Sensor Networks,” Proc. 2nd ACM Int’l. Conf. Wireless Sensor Net. Apps., San Diego, CA, 2003, pp. 11–19.

N=3 N=6 N=9 N = 12 N = 20 N = 50

900 800 Synchronization error (μs)

10:48 AM

700 600 500 400

BIOGRAPHIES

300

ÁLVARO MARCO ([email protected]) received his degree in electrical engineering in 2000 and his Ph.D. in electronic engineering in 2007, both from the University of Zaragoza, Spain. Currently, he is a senior researcher in the Department of Electrical Engineering and Communications at the same university, and his research interests include sensor networks, ambient intelligence, and assistive technology.

200 100 0

0

1

3

2

4

Sync level

Figure 4. Average synchronization error for Zigbee network with 4 hops and sync messages every 30 s; here N is the number of points used to perform regression, and the dashed bars represent maximum error with 95 percent probability.

Bee), evaluating its performance. Through exhaustive experimentation we have identified the factors that influence synchronization accuracy the most. We have analyzed hardware (transceivers, crystals) and firmware (sync-message rate, number of data points to perform regression) issues to provide the application designer with guidelines that could help in deciding on the best suited technology for each application, and we have characterized synchronization behavior in several scenarios. These results allow the application designer to decide on the hardware architecture and protocol scheme best suited to achieve the required synchronization accuracy.

ACKNOWLEDGMENTS This work was supported in part by the Spanish MCYT under AmbienNet project (TIN200615617-C03) and by the European Commission under the MonAMI project.

REFERENCES [1] I. F. Akyildiz et al., “A Survey on Sensor Networks,” IEEE Commun. Mag., vol. 40, no. 8, 2002, pp. 102–14. [2] F. Sivrikaya and B. Yener, “Time Synchronization in Sensor Networks: A Survey,” IEEE Network, vol. 18, no. 4, 2004, pp. 45–50. [3] A. Marco et al., “Multi-Hop Synchronization at the Application Layer of Wireless and Satellite Networks,” Proc. IEEE GLOBECOM ’08, New Orleans, LA, 2008, pp. 1–5. [4] J. Elson, L. Girod, and D. Estrin, “Fine-Grained Time Synchronization using Reference Broadcasts,” Proc. 5th Symp. Op. Sys. Design Implementation, Boston, MA, 2002, pp. 147–63. [5] M. Maróti et al., “The Flooding Time Synchronization Protocol,” Proc. 2nd ACM SenSys ’04, 2004, pp. 39–49. [6] J. Elson and D. Estrin, “Time Synchronization for Wireless Sensor Networks,” Proc. 15th Int’l. Parallel & Distrib. Process. Symp., San Francisco, CA, 2001, pp. 1965–70.

88

ROBERTO CASAS received his degree in electrical engineering in 2000 and his Ph.D. in electronic engineering in 2004, both from the University of Zaragoza. Currently, he is an assistant professor in the Department of Electrical Engineering and Communications at the same university, and his research interests include sensor networks, digital electronics, and assistive technology. J OSÉ L UIS S EVILLANO R AMOS ([email protected]) received his Ph.D. from the University of Seville, Spain, in 1993. Since 1996 he has been an associate professor of computer architecture at the University of Seville. Currently, he is coordinator of the Telefónica Chair on Intelligence in Networks at the same university. He also serves as VP Membership of the Society for Modeling & Simulation International, SCS. VICTORIA’N COARASA is a technical researcher in the Department of Electrical Engineering and Communications at the University of Zaragoza, Spain. His research interests include sensor networks, digital electronics, and assistive technology. He received his M.S. degree in wireless engineering in 2007 from the University of Zaragoza, and is currently applying to obtain a Ph.D. degree in electronic engineering. ’

ANGEL ASENSIO is a senior researcher in the Department of Electrical Engineering and Communications at the University of Zaragoza. His research interests include sensor networks, digital electronics, and assistive technology. He received his M.S. degree in electrical engineering in 2005 from the University of Zaragoza and is currently applying to obtain a Ph.D. degree in electronic engineering. MOHAMMAD S. OBAIDAT [F‘05] ([email protected]) is an internationally known academic/researcher/scientist. He received his Ph.D. in computer engineering from Ohio State University. He is currently a full professor of computer science and software engineering at Monmouth University. Among his previous positions are chair of the Computer Science Department and director of the Graduate Program at MU. He has received extensive research funding, and authored/co-authored 10 books and over 475 refereed scholarly journal and conference articles. He has served as a consultant for several corporations worldwide and is editor of many scholarly journals including being Editor-inChief of Wiley’s International Journal of Communication Systems. He is President of the Society for Modeling & Simulation International, SCS. He was awarded the distinguished Nokia Research Fellowship and the Distinguished Fulbright Award. He has been invited to lecture and give keynote speeches worldwide. He was the recipient of the Best Paper Award for one of his papers accepted for IEEE AICCSA 2009 and IEEE GLOBCOM 2009. He also received the SCS prestigious McLeod Founder’s Award in recognition of his outstanding technical and professional contributions to modeling and simulation. He is a Fellow of SCS.

IEEE Wireless Communications • February 2011

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

Recommend Documents