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!
Communications in Computer and Information Science
150
Tai-hoon Kim Hojjat Adeli Rosslin John Robles Maricel Balitanas (Eds.)
Ubiquitous Computing and Multimedia Applications Second International Conference, UCMA 2011 Daejeon, Korea, April 13-15, 2011 Proceedings, Part I
13
Volume Editors Tai-hoon Kim Hannam University, Daejeon, Korea E-mail: [email protected] Hojjat Adeli The Ohio State University Columbus, OH, USA E-mail: [email protected] Rosslin John Robles Hannam University, Daejeon, Korea E-mail: [email protected] Maricel Balitanas Hannam University, Daejeon, Korea E-mail: [email protected]
ISSN 1865-0929 e-ISSN 1865-0937 ISBN 978-3-642-20974-1 e-ISBN 978-3-642-20975-8 DOI 10.1007/978-3-642-20975-8 Springer Heidelberg Dordrecht London New York Library of Congress Control Number: 2011926692 CR Subject Classification (1998): H.5.1, I.5, I.2, I.4, F.1, H.3, H.4
Ubiquitous computing and multimedia applications are areas that attract many academic and industry professionals. The goal of the International Conference on Ubiquitous Computing and Multimedia Applications is to bring together researchers from academia and industry as well as practitioners to share ideas, problems and solutions relating to the multifaceted aspects of ubiquitous computing and multimedia applications. We would like to express our gratitude to all of the authors of submitted papers and to all attendees, for their contributions to and participation in UCMA 2011. We believe in the need for continuing this undertaking in the future. We acknowledge the great effort of all the Chairs and the members of advisory boards and Program Committees of the above-listed event, who selected 15% of over 570 submissions, following a rigorous peer-review process. Special thanks go to SERSC (Science and Engineering Research Support Society) for supporting this conference. We are grateful in particular to the following speakers who kindly accepted our invitation and, in this way, helped to meet the objectives of the conference: Sabah Mohammed of Lakehead University and Peter Baranyi of Budapest University of Technology and Economics (BME). March 2011
Chairs of UCMA 2011
Preface
We would like to welcome you to the proceedings of the 2011 International Conference on Ubiquitous Computing and Multimedia Applications (UCMA 2011) which was held during April 13-15, 2011, at Hannam University, Daejeon, Korea. UCMA 2011 is focused on various aspects of advances in multimedia applications and ubiquitous computing with computational sciences, mathematics and information technology. It provided a chance for academic and industry professionals to discuss recent progress in the related areas. We expect that the conference and its publications will be a trigger for further related research and technology improvements in this important subject. We would like to acknowledge the great effort of all the Chairs and members of the Program Committee. Out of around 570 submissions to UCMA 2011, we accepted 86 papers to be included in the proceedings and presented during the conference. This gives an acceptance ratio firmly below 20%. Regular conference papers can be found in this volume while Special Session papers can be found in CCIS 151. We would like to express our gratitude to all of the authors of submitted papers and to all the attendees for their contributions and participation. We believe in the need for continuing this undertaking in the future. Once more, we would like to thank all the organizations and individuals who supported this event as a whole and, in particular, helped in the success of UCMA 2011. March 2011
Tai-hoon Kim Hojjat Adeli Rosslin John Robles Maricel Balitanas
Organization
Organizing Committee Honorary Co-chairs:
General Co-chairs:
Program Co-chairs:
Workshop Co-chairs:
Hyung-tae Kim (Hannam University, Korea) Hojjat Adeli (The Ohio State University, USA) Wai-chi Fang (National Chiao Tung University, Taiwan) Carlos Ramos (GECAD/ISEP, Portugal) Haeng-kon Kim (Catholic University of Daegu, Korea) Tai-hoon Kim (Hannam University, Korea) Sabah Mohammed (Lakehead University, Canada) Muhammad Khurram Khan (King Saud University, Saudi Arabia) Seok-soo Kim (Hannam University, Korea) Timothy K. Shih (Asia University, Taiwan)
International Advisory Board: Cao Jiannong (The Hong Kong Polytechnic University, Hong Kong) Frode Eika Sandnes (Oslo University College, Norway) Schahram Dustdar (Vienna University of Technology, Austria) Andrea Omicini (Universit` a di Bologna, Italy) Lionel Ni (The Hong Kong University of Science and Technology, Hong Kong) Rajkumar Buyya (University of Melbourne, Australia) Hai Jin (Huazhong University of Science and Technology, China) N. Jaisankar (VIT University, India) Gil-cheol Park (Hannam University, Korea) Ha Jin Hwang (Kazakhstan Institute of Management, Economics, and Strategic Research, Republic of Kazakhstan)
X
Organization
Publicity Co-chairs:
Publication Co-chairs:
Paolo Bellavista (Universit`a di Bologna, Italy) Ing-Ray Chen (Virginia Polytechnic Institute and State University, USA) Yang Xiao (University of Alabama, USA) J.H. Abawajy (Deakin University, Australia) Ching-Hsien Hsu (Chung Hua University, Taiwan) Deepak Laxmi Narasimha (University of Malaya, Malaysia) Prabhat K. Mahanti (University of New Brunswick, Canada) Soumya Banerjee (Birla Institute of Technology, India) Byungjoo Park (Hannam University, Korea) Debnath Bhattacharyya (MPCT, India)
Program Committee Alexander Loui Biplab K. Sarker Brian King Chantana Chantrapornchai Claudia Linnhoff-Popien D. Manivannan Dan Liu Eung Nam Ko Georgios Kambourakis Gerard Damm Han-Chieh Chao Hongli Luo Igor Kotenko J.H. Abawajy Jalal Al-Muhtadi
Javier Garcia-Villalba Khaled El-Maleh Khalil Drira Larbi Esmahi Liang Fan Mahmut Kandemir Malrey Lee Marco Roccetti Mei-Ling Shyu Ming Li Pao-Ann Hsiung Paolo Bellavista Rami Yared Rainer Malaka Robert C. Hsu Robert G. Reynolds
Rodrigo Mello Schahram Dustdar Seung-Hyun Seo Seunglim Yong Stefano Ferretti Stuart J. Barnes Su Myeon Kim Swapna S. Gokhale Taenam Cho Tony Shan Toshihiro Yamauchi Wanquan Liu Wenjing Jia Yao-Chung Chang
Table of Contents – Part I
A SOA-Based Service Composition for Interactive Ubiquitous Entertainment Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Giovanni Cagalaban and Seoksoo Kim
1
A Study on Safe Reproduction of Reference Points for Recognition on Screen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sungmo Jung and Seoksoo Kim
7
Maximized Energy Saving Sleep Mode in IEEE 802.16e/m . . . . . . . . . . . . Van Cuong Nguyen, Van Thuan Pham, and Bong-Kyo Moon
11
Propose New Structure for the Buildings Model . . . . . . . . . . . . . . . . . . . . . . Tuan Anh Nguyen gia
23
Solving Incomplete Datasets in Soft Set Using Parity Bits of Supported Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ahmad Nazari Mohd. Rose, Hasni Hassan, Mohd Isa Awang, Tutut Herawan, and Mustafa Mat Deris An Algorithm for Mining Decision Rules Based on Decision Network and Rough Set Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hossam Abd Elmaksoud Mohamed A Probabilistic Rough Set Approach to Rule Discovery . . . . . . . . . . . . . . . Hossam Abd Elmaksoud Mohamed
Oversampled Perfect Reconstruction FIR Filter Bank Implementation by Removal of Noise and Reducing Redundancy . . . . . . . . . . . . . . . . . . . . . Sangeeta Chougule and Rekha P. Patil
76
Designing a Video Control System for a Traffic Monitoring and Controlling System of Intelligent Traffic Systems . . . . . . . . . . . . . . . . . . . . . Il-Kwon Lim, Young-Hyuk Kim, Jae-Kwang Lee, and Woo-Jun Park
Edge Detection in Grayscale Images Using Grid Smoothing . . . . . . . . . . . Guillaume Noel, Karim Djouani, and Yskandar Hamam
110
Energy-Based Re-transmission Algorithm of the Leader Node’s Neighbor Node for Reliable Transmission in the PEGASIS . . . . . . . . . . . . Se-Jung Lim, A.K. Bashir, So-Yeon Rhee, and Myong-Soon Park
120
Grid-Based and Outlier Detection-Based Data Clustering and Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Kyu Cheol Cho and Jong Sik Lee
A Study on Utilization of Export Assistance Programs for SMEs and Their Exportation Performance in Korea . . . . . . . . . . . . . . . . . . . . . . . . . . . . Woong Eun, Sangchun Lee, Yong-Seok Seo, and Eun-Young Kim
290
The Entrance Authentication and Tracking Systems Using Object Extraction and the RFID Tag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dae-Gi Min, Jae-Woo Kim, and Moon-Seog Jun
313
A Privacy Technique for Providing Anonymity to Sensor Nodes in a Sensor Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jeong-Hyo Park, Yong-Hoon Jung, Hoon Ko, Jeong-Jai Kim, and Moon-Seog Jun Study on Group Key Agreement Using Two-Dimensional Array in Sensor Network Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Seung-Jae Jang, Young-Gu Lee, Hoon Ko, and Moon-Seog Jun
327
336
XIV
Table of Contents – Part I
Design and Materialization of Location Based Motion Detection System in USN Environments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Joo-Kwan Lee, Jeong-Jai Kim, and Moon-Seog Jun Realization of Integrated Service for Zoos Using RFID/USN Based Location Tracking Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jae-Yong Kim, Young-Gu Lee, Kwang-Hyong Lee, and Moon-Seok Jun
350
361
Real-Time Monitoring System Using Location Based Service . . . . . . . . . . Jae-Hwe You, Young-Gu Lee, and Moon-Seog Jun
A Smart Error Protection Scheme Based on Estimation of Perceived Speech Quality for Portable Digital Speech Streaming Systems . . . . . . . . . Jin Ah Kang and Hong Kook Kim
1
MDCT-Domain Packet Loss Concealment for Scalable Wideband Speech Coding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Nam In Park and Hong Kook Kim
11
High-Quality and Low-Complexity Real-Time Voice Changing with Seamless Switching for Digital Imaging Devices . . . . . . . . . . . . . . . . . . . . . . Sung Dong Jo, Young Han Lee, Ji Hun Park, Hong Kook Kim, Ji Woon Kim, and Myeong Bo Kim Complexity Reduction of Virtual Reverberation Filtering Based on Index-Based Convolution for Resource-Constrained Devices . . . . . . . . . . . Kwang Myung Jeon, Nam In Park, Hong Kook Kim, Ji Woon Kim, and Myeong Bo Kim Audio Effect for Highlighting Speaker’s Voice Corrupted by Background Noise on Portable Digital Imaging Devices . . . . . . . . . . . . . . . . . . . . . . . . . . Jin Ah Kang, Chan Jun Chun, Hong Kook Kim, Ji Woon Kim, and Myeong Bo Kim
19
28
39
Detection of Howling Frequency Using Temporal Variations in Power Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jae-Won Lee and Seung Ho Choi
The Eye Movement and Data Processing Due to Obtained BOLD Signal in V1 : A Study of Simultaneous Measurement of EOG and fMRI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hyo Woon Yoon, Dong-Hwa Kim, Young Jae Lee, Hyun-Chang Lee, and Ji-Hyang Lim
66
XVI
Table of Contents – Part II
Face Tracking for Augmented Reality Game Interface and Brand Placement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Yong Jae Lee and Young Jae Lee
72
On-line and Mobile Delivery Data Management for Enhancing Customer Services . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hyun-Chang Lee, Seong Yoon Shin, and Yang Won Rhee
79
Design of LCL Filter Using Hybrid Intelligent Optimization for Photovoltaic System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jae Hoon Cho, Dong-Hwa Kim, M´ aria Virˇc´ıkov´ a, and Peter Sinˇc´ ak
90
Distributed Energy Management for Stand-Alone Photovoltaic System with Storages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jae Hoon Cho and Dong-Hwa Kim
98
Framework for Performance Metrics and Service Class for Providing End-to-End Services across Multiple Provider Domains . . . . . . . . . . . . . . . Chin-Chol Kim, Jaesung Park, and Yujin Lim
107
Design of a Transmission Simulator Based on Hierarchical Model . . . . . . . Sang Hyuck Han and Young Kuk Kim Recommendation System of IPTV TV Program Using Ontology and K-means Clustering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jongwoo Kim, Eungju Kwon, Yongsuk Cho, and Sanggil Kang A Novel Interactive Virtual Training System . . . . . . . . . . . . . . . . . . . . . . . . Yoon Sang Kim and Hak-Man Kim
114
123 129
Recommendation Algorithm of the App Store by Using Semantic Relations between Apps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Yujin Lim, Hak-Man Kim, Sanggil Kang, and Tai-hoon Kim
139
A Comparative Study of Bankruptcy Rules for Load-shedding Scheme in Agent-Based Microgrid Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hak-Man Kim and Tetsuo Kinoshita
145
The Visualization Tool of the Open-Source Based for Flight Waypoint Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Myeong-Chul Park and Seok-Wun Ha
153
Space-Efficient On-the-fly Race Detection Using Loop Splitting . . . . . . . . Yong-Cheol Kim, Sang-Soo Jun, and Yong-Kee Jun A Comparative Study on Responding Methods for TCP’s Fast Recovery in Wireless Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mi-Young Park, Sang-Hwa Chung, Kyeong-Ae Shin, and Guangjie Han
162
170
Table of Contents – Part II
XVII
The Feasibility Study of Attacker Localization in Wireless Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Young-Joo Kim and Sejun Song
180
Efficiency of e-NR Labeling for On-the-fly Race Detection of Programs with Nested Parallelism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sun-Sook Kim, Ok-Kyoon Ha, and Yong-Kee Jun
191
Lightweight Labeling Scheme for On-the-fly Race Detection of Signal Handlers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Guy Martin Tchamgoue, Ok-Kyoon Ha, Kyong-Hoon Kim, and Yong-Kee Jun
201
Automatic Building of Real-Time Multicore Systems Based on Simulink Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Minji Cha and Kyong Hoon Kim
209
A Study on the RFID/USN Integrated Middleware for Effective Event Processing in Ubiquitous Livestock Barn . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jeonghwan Hwang and Hyun Yoe
Image Data Hiding Technique Using Discrete Fourier Transformation . . . Debnath Bhattacharyya and Tai-hoon Kim
315
Design of Guaranteeing System of Service Quality through the Verifying Users of Hash Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Yoon-Su Jeong, Yong-Tae Kim, and Gil-Cheol Park
324
Design of RSSI Signal Based Transmit-Receiving Device for Preventing from Wasting Electric Power of Transmit in Sensor Network. . . . . . . . . . . Yong-Tae Kim, Yoon-Su Jeong, and Gil-Cheol Park
An Approach to Roust Control Management in Mobile IPv6 for Ubiquitous Integrate Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Giovanni Cagalaban and Byungjoo Park
354
A Software Framework to Associate Multiple FPMN MSISDNs with a HPMN IMSI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dongcheul Lee and Byung Ho Rhe
360
Uplink Interference Adjustment for Mobile Satellite Service in Multi-beam Environments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ill-Keun Rhee, Sang-Am Kim, Keewan Jung, Erchin Serpedin, Jong-Min Park, and Young-Hun Lee Protecting Computer Network with Encryption Technique: A Study . . . . Kamaljit I. Lakhtaria
371
381
Table of Contents – Part II
A Study on the Strategies of Foreign Market Expansion for Korean IT Venture Company . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Woong Eun and Yong-Seok Seo Binding Update Schemes for Inter-domain Mobility Management in Hierarchical Mobile IPv6: A Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Afshan Ahmed, Jawad Hassan, Mata-Ur-Rehman, and Farrukh Aslam Khan
XIX
391
411
A Pilot Study to Analyze the Effects of User Experience and Device Characteristics on the Customer Satisfaction of Smartphone Users . . . . . Bong-Won Park and Kun Chang Lee
421
Exploring the Optimal Path to Online Game Loyalty: Bayesian Networks versus Theory-Based Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . Nam Yong Jo, Kun Chang Lee, and Bong-Won Park
428
The Effect of Users’ Characteristics and Experiential Factors on the Compulsive Usage of the Smartphone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bong-Won Park and Kun Chang Lee
438
Leadership Styles, Web-Based Commitment and Their Subsequent Impacts on e-Learning Performance in Virtual Community . . . . . . . . . . . . Dae Sung Lee, Nam Young Jo, and Kun Chang Lee
447
Test Case Generation for Formal Concept Analysis . . . . . . . . . . . . . . . . . . . Ha Jin Hwang and Joo Ik Tak
457
Mobile Specification Using Semantic Networks . . . . . . . . . . . . . . . . . . . . . . . Haeng-Kon Kim
Pairwise Protein Substring Alignment with Latent Semantic Analysis and Support Vector Machines to Detect Remote Protein Homology . . . . . Surayati Ismail, Razib M. Othman, and Shahreen Kasim Jordan Pi-Sigma Neural Network for Temperature Prediction . . . . . . . . . . Noor Aida Husaini, Rozaida Ghazali, Nazri Mohd Nawi, and Lokman Hakim Ismail Accelerating Learning Performance of Back Propagation Algorithm by Using Adaptive Gain Together with Adaptive Momentum and Adaptive Learning Rate on Classification Problems . . . . . . . . . . . . . . . . . . . Norhamreeza Abdul Hamid, Nazri Mohd Nawi, Rozaida Ghazali, and Mohd Najib Mohd Salleh Developing an HCI Model: An Exploratory Study of Featuring Collaborative System Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Saidatina Fatimah Ismail, Rathiah Hashim, and Siti Zaleha Zainal Abidin Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
526 547
559
571
581
A SOA-Based Service Composition for Interactive Ubiquitous Entertainment Applications Giovanni Cagalaban and Seoksoo Kim* Department of Multimedia, Hannam University, Daejeon, Korea [email protected], [email protected]
Abstract. As mobile and embedded computing devices become more ubiquitous, it is becoming obvious that the nature of interactions between users and computers have evolved. The capabilities of mobile and wireless devices to be seamlessly integrated into daily objects and tasks allow a plethora of services particularly in the field of entertainment and gaming. Particularly, this paper proposes a service composition for interactive ubiquitous entertainment applications based on service oriented architecture (SOA). It aims to establish collaborative relationships between heterogeneous devices to provide users an interactive and ubiquitous entertainment and fun. The robust approach is to create a seamless entertainment environment which can cope with highly dynamic environments in which resources such as network connectivity and services frequently vary over time. Keywords: service-oriented architecture, ubiquitous entertainment, service composition, ubiquitous computing.
1 Introduction The rapid advancement of miniaturized mobile or embedded information and communication technologies (ICT) with some degree of intelligence and context awareness, network connectivity, and advanced user interfaces allow continuous and uninterrupted access to information and communication technology services. The nature of devices will change to form augmented environments in which the physical world is sensed and controlled in such a way that it becomes merged with the virtual world. The ubiquitous, unobtrusive, supportive and informative functions can be readily available as these devices “weave themselves into the fabrics of everyday life until they are indistinguishable from it” [1]. In ubiquitous computing, information processing has been thoroughly integrated into everyday objects and activities. Mobile devices can play a major role in entertainment systems as one of the devices used by people to perform their work, play games, or entertain themselves while in motion. Context awareness may be applied more flexibly with mobile computing with any moving objects such as people bearing the mobile devices where they can sense their environment. Coupled with the recent trend in 3G technology, the portable nature of mobile and wireless devices is rapidly changing technologically *
from normal telephone conversations into interactive and fun multimedia games. The gaming devices, hardware, and platforms used have showed rapid growth in terms of complexity and richness of features. As a result, ubiquitous entertainment systems are now a potential major new application area in ubiquitous computing services. Establishing collaborative relationships between mobile computing devices and wireless computing devices provide users flexibility and shared services. Serviceoriented architecture (SOA) promises a robust approach that can easily handle these dynamic and heterogeneous environments. SOA has established itself as a prevailing software engineering practice in recent years [2]. Its characteristics of loose-coupling, stateless, and platform-independence make it an ideal candidate for integrating heterogeneous devices. In this research, we design a service composition based on SOA for robust ubiquitous entertainment systems to establish collaborative relationships between heterogeneous devices. This can be done in such a way that existing services are orchestrated into one or more new services that fit better the ubiquitous applications. The proposed service oriented architecture enables mobile users to find themselves in environments rich in ubiquitous services that their mobile devices can take advantage of entertainment, gaming, and fun. Particularly, this paper presents an agenda for future investigations in ubiquitous computing for interactive entertainment systems.
2 Related Work Despite the availability of infrastructural elements for entertainment systems, researches have shown that there are limitations in building ubiquitous entertainment systems due to the integration between all technologies used [3]. Numerous software infrastructures and frameworks have been proposed [4] to eliminate these barriers. Integration of different technologies such as WSN, RFID and wireless sensor networks and using standardized frameworks are studied by [5][6][7]. Also, an entertainment system built on ubiquitous computing and networking technologies [8] is capable of providing tangible interaction over a wide outdoor area. Additionally, with its interoperability of services with operating systems and other technologies that support applications, service-oriented architecture has received attention and has also been proposed for these purposes, such as Web services due to the near-ubiquity of the Internet such as in [9]. However, these researches do not provide the necessary requirements of an interactive design of a ubiquitous entertainment system.
3 Service Composition The design of the ubiquitous entertainment system is based on a high-level conceptual model consisting of computing devices, user, application services and user interfaces. The nature of the hardware design and software development is significant for providing ubiquitous entertainment services. With ubiquitous entertainment services, users are provided with unique experience and interactivity that can gain importance and relevance. It can be used to evaluate if such a service can be made as a benchmark
A SOA-Based Service Composition for Interactive Ubiquitous Entertainment Applications
3
for the development of ubiquitous entertainment systems. Figure 1 shows a ubiquitous entertainment where a mobile user is playing a game in a mobile device surrounded by sensor network and is equipped with context awareness to seamlessly perform other activities.
Fig. 1. Ubiquitous environment
Service composition of a ubiquitous entertainment systems are significant in the development of customized services often by done by discovering, integrating, and executing existing services of ubiquitous entertainment systems. The criteria of service composition involve interactive design, application interoperability and heterogeneous development. The interactive design of a ubiquitous entertainment system depends on the following features: the dimension of the user internal world model; the presentation effect of the interface; the perception mechanism; and the conceptualization of the world dimensions in the user’s model [10]. A major problem in designing user interfaces is the fact that interaction designers do not have any established metrics or benchmarks for applying the optimal software and hardware system within the context of ubiquitous entertainment systems. Table 1 presents the features of a ubiquitous entertainment system. Heterogeneous devices will be required to interact seamlessly, despite wide differences in hardware and software capabilities. Due to heterogeneous computing power, interface, and network bandwidth, ubiquitous services could not be implemented to provide the same capability, quality, and performance in each platform. It is necessary to design the system to provide seamless interaction with the user despite differences in hardware devices and software platforms. The concurrent distributed setup of ubiquitous entertainment systems are developed using heterogeneous hardware and platforms. It is important to determine the distributed settings for creating a collaborative relationship between the elements for development.
4
G. Cagalaban and S. Kim Table 1. Ubiquitous entertainment systems features Features Scalability User actions Multitasking Context awareness Social context User needs Interruption Timing and duration
Functionalities Plug and play modules Seamless interaction, multiple granularity levels Multi-modality, association and clear separation Connection and relation between the immediate surroundings and the entertainment services Multi-user integration and experiences Relevant applications Gradual fading, transition from foreground to background Conform user focus and interests, provide continuous entertainment and events
4 Service Oriented Architecture In the service oriented architecture for the ubiquitous entertainment system, the main concept is the existence of a ubiquitous service center where several ubiquitous environment and services are available for use by mobile devices. The ubiquitous service center has more computing power compared to the mobile devices that people use. Examples of such ubiquitous environment may be in a conference room, coffee shop, restaurant, train station, airport, or shopping mall, etc. The ubiquitous service centers can be configured in controlled and relatively stable environments, such as homes, offices, or public areas that experience a high density of mobile users to provide ubiquitous services. Discovering and using the services should be intuitive to users. In the ubiquitous entertainment system, a mobile user should be able to access and consume the services provided by the ubiquitous entertainment server through the ubiquitous environment in a seamless manner. The proposed service oriented architecture for the ubiquitous entertainment system is shown in Figure 2. The ubiquitous entertainment system is based on the interactions between the major service components: ubiquitous service center, ubiquitous environment, entertainment server, and mobile devices. The ubiquitous service center serves as the intermediary system that facilitates the management and operation of services to the mobile devices that queries the service center. In the service oriented architecture of the ubiquitous entertainment system, the main focus of the design is the service interface. Here, we consider that all the service components are deployed in ubiquitous environment and are visible for mobile devices to invoke or consume over the network. Also, the nature of SOA is focused on creating services using components with well defined interfaces, which allows devices to be loosely coupled with each other. Due to this, we propose an XML-based description for describing device capabilities and other elements of the ubiquitous entertainment systems.
A SOA-Based Service Composition for Interactive Ubiquitous Entertainment Applications
5
Fig. 2. Service oriented architecture for ubiquitous entertainment system
The design of the ubiquitous service center is capable of performing registration and query of services, and session management. The registration and discovery of services in a ubiquitous service center is done by the service manager. The service manager is responsible for facilitating the registration service where mobile devices can discover and select from a set of services. In session management, if a user of a mobile device discovers that there are available services in the ubiquitous service center that are of his interest, he can query and request the service manager for services to be used by his mobile device. When requested by a mobile device, the service manager sets up the session where the requested service is reserved for the requesting mobile device. In general, the service oriented architecture of the ubiquitous entertainment system needs to be carefully designed by considering service component allocation and distribution across cross-platforms. The architecture design affects on performance, reliability, availability, security and other software qualities.
5 Conclusion As mobile and embedded computing devices become more ubiquitous, it is becoming obvious that the nature of interactions between users and computers have evolved. We have presented the significance of introducing a service-oriented architecture for ubiquitous entertainment system. The main focus of the service composition is the creation of flexible and extensible interfaces for the interactive delivery of services for a ubiquitous entertainment. The design of a service-oriented architecture is an integral part of ubiquitous technology. We concluded with an agenda for future research that highlights the need for further assessment in adaptation, heterogeneity
6
G. Cagalaban and S. Kim
and user interface and also to integrate security mechanisms for the ubiquitous entertainment system by implementing security algorithms to enhance reliability and improved performance of ubiquitous services.
Acknowledgements This paper has been supported by the 2010 Hannam University Research Fund.
References 1. Weiser, M.: The computer for the twenty-first century. Scientific American 265(3), 94– 104 (1991) 2. McCoy, D.W., Natis, Y.V.: Service-Oriented Architecture: Mainstream Straight Ahead, Gartner Research Report (2003), http://www.gartner.com/pages/story.php.id.3586.s.8.jsp 3. Davies, N., Gellersen, H.W.: Beyond prototypes: Challenges in deploying ubiquitous systems. IEEE Pervasive Computing 1, 26–35 (2002) 4. Endres, C., Butz, A., MacWilliams, A.: A Survey of Software Infrastructures and Frameworks for Ubiquitous Computing. Mobile Information Systems Journal 1(1), 41–80 (2005) 5. Salehi, A., Aberer, K.: GSN, quick and simple sensor network deployment. In: European Conference on Wireless Sensor Networks (EWSN), Delft, Netherlands (January 2007) 6. Sánchez-López, T., Daeyoung, K.: Wireless Sensor Networks and RFID integration for Context Aware Services. Technical Report, Auto-ID Labs White Paper (2008) 7. Sung, J., Lopez, T.S., Kim, D.: The EPC sensor network for RFID and WSN integration infrastructure. In: IEEE International Conference on Pervasive Computing and Communications Workshops, pp. 618–621. IEEE Computer Society, Los Alamitos (2007) 8. Cheok, A.D., Fong, S.W., Goh, K.H., Yang, X., Liu, W., Farzbiz, F.: Human Pacman: A Sensing-based Mobile Entertainment System with Ubiquitous Computing and Tangible Interaction. In: Proceedings of the Second Workshop on Network and System Support for Games (2003) 9. Ubiquitous Web Applications Working Group, website at http://www.w3.org/2007/uwa/ 10. Rauterberg, M., Szabo, K.: A design concept for N-dimensional user interfaces. In: Proc. International Conference INTERFACE to Real & Virtual Worlds, pp. 467–477 (1995)
A Study on Safe Reproduction of Reference Points for Recognition on Screen Sungmo Jung and Seoksoo Kim* Dept. of Multimedia, Hannam Univ., 133 Ojeong-dong, Daedeok-gu, Daejeon-city, Korea [email protected], [email protected]
Abstract. In order to show multi-objects in AR images, a marker, a reference point for recognition, should be reproduced on screen. However, if reproduced markers are randomly output, they may overlap with each other, so we need to employ area protection and collision evasion technologies. Also, the number of markers needs to be limited or the size reduced so as to safely reproduce reference points for recognition. In this study, therefore, we suggested a method of safely reproducing markers on screen, limiting the number of markers in order to improve efficiency as a reference point for recognition, and automatically adjusting the sizes. Keywords: Augmented Reality, Recognition Standard Points, Limited Output, Objects Autosize.
1 Introduction More recently, research on AR (Augmented Realty) [1] is of growing interest, and more school-industry efforts are being put into applying the technology to real life. Previous marker-based AR technology usually allows one object to be augmented for one marker that appears on screen. Considering use of contents in the future, however, we need to solve a problem of adding one marker. To that end, we may use multi-object loading and reproducing technology [2] and, in this research, we suggest a method of safely reproducing a marker on screen, limiting the number of markers in order to increase its efficiency as a reference point, and automatically adjusting the size of a marker.
2 Method of Safely Reproducing the Recognition Reference Point on Screen 2.1 Limited Number of Markers When a screen in divided into 9 blocks and the actual marker in the center is deemed one block, 8 markers, which are not overlapped with the actual marker and used as a reference point for recognition, can be reproduced as depicted in the following figure. *
Fig. 1. Position of 8 Markers that can be Reproduced on 9-block Screen
In this way, when the sccreen is divided into a minimum number of blocks so tthat the actual marker can be used u as a reference point for recognition and when coppied markers are created the same s size as the actual marker, only 8 objects can be augmented on the screen. Thus, T if we leave a minimum area for a reference point and divide each of the 8 areas into 4 blocks again, more objects can be augmented. T The following figure shows th he maximum marker reproducing on one screen whhile allowing the actual marker to be used as a reference point.
Fig. 2. Position of 32 3 Markers that can be Reproduced on 36-block Screen
Here, the number of co opied markers should be limited to 32 per a screen. IIf a screen is divided into block ks exceeding the number, the copied markers would be too small to be recognized by th he reference point. 2.2 Automatic Adjustmen nt of Marker Sizes When a 3D AR image is acctually applied, it is often necessary to adjust marker siizes automatically in order to co ontrol the number of markers. An extracted marker shoould
A Study on Safe Reprroduction of Reference Points for Recognition on Screen
9
be deemed an image and then a function should be defined taking into accoount strength and directions of feeatures. In this study, we first established a vector field tthat expresses directions of majjor elements in an image and stabilized direction fields in order to reduce noises an nd to maintain continuity/connection around the flow w of features. Some of the tecchnologies of calculating stabilized direction fields are Scattered Orientation Interpolation[3][4], Orientation Diffusion[5][6][7], Structture Tensor [8], and non-linear direction filtering[9][10][11]. In this study we emplo yed Edge Tangent Flow[12] in order o to make a direction field. Usually, a size is adjusteed in the directions of axis x and y. However, in view off the nature of AR images show wing 3D images in real-time, axis z (vertical) shouldd be taken into account. Weightt for a vertical direction, ωv(x), can be expressed with the following formula. ·
(1)
Here, ey = (0, 1) is a unit vector v for axis y, which is weight to maintain a segmennt of axis y when an image is ad djusted in the direction of axis x. Similarly, weight for aaxis x, ωh(x), can be obtained wiithout difficulty. Based on the above, a fo ormula for a seam in the horizontal direction can be defiined as follows. |
|
(2)
Similarly, a formula for a seeam in the vertical direction can be defined as follows. |
|
(3)
ev is used for adjustment in n the horizontal direction and eh for the vertical directiion. The following figure show ws the final outcome of 2 markers (one marker from a 9block screen and one markeer from a 36-block screen) and adjustment of marker sizzes.
Fig. 3. 3 Automatic Adjustment of Marker Sizes
3 Conclusion In this study we suggested a method of safely reproducing markers on screen, limitting the number of markers in n order to improve efficiency as a reference point for recognition, and automaticaally adjusting the sizes.
10
S. Jung and S. Kim
The strength of this method is that the markers reproduced in a limited number can play a role of a reference point for recognition and that the maximum number of markers can be produced on one screen. However, in order to apply this technology to contents, not only marker-based but also non marker-based production technology should be developed in the near future.
Acknowledgement This paper has been supported by the 2011 Hannam University Research Fund.
References 1. Azuma Ronald, T.: A Survey of Augmented Reality. Teleoperators and Virtual Environments 6(4), 355–385 (1997) 2. Sungmo, J., Seoksoo, K.: A Study on M2M-based AR Multiple Objects Loading Technology using PPHT. In: The 12th WSEAS International Conference on Automatic Control, Modelling & Simulation, Cantania, Italy, pp. 420–424 (2010) 3. Litwinowicz, P.: Processing Images and Video for an Impressionist Effect. In: Proc. ACM SIGGRAPH, pp. 407–414 (1997) 4. Hays, J., Essa, I.: Image and Video-based Painterly Animation. In: Proc. NonPhotorealistic Animation and Rendering, pp. 113–120 (2004) 5. Xu, C., Prince, J.L.: Snakes: Shapes and Gradient Vector Flow. IEEE Transactions on Image Processing 7(3), 359–369 (1998) 6. Perona, P.: Orientation Diffusions. IEEE Transactions on Image Processing 7(3), 457–467 (1998) 7. Tschumperlé, D., Deriche, R.: Orthonormal Vector Sets Regularization with PDE’s and Applications. International Journal of Computer Vision 50(3), 237–252 (2002) 8. Weickert, J.: Anisotropic Diffusion in Image Processing. PhD thesis, Dept. of Mathematics, University of Kaiserslautern, Germany (1996) 9. Pham, T.Q.: Spatiotonal Adaptively in Super-Resolution of Under sampled Image Sequences, PhD thesis, Delft University of Technology (2006) 10. Paris, S., Brincen, O.H., Sillion, F.: Capture of Hair Geometry from Multiple Images. ACM Transactions on Graphics 23(3), 712–719 (2004) 11. Kang, H., Lee, S., Chui, C.: Coherent Line Drawing. In: Proceedings of ACM Symposium on Nonphotorealistic Animation and Rendering, pp. 43–50 (2007) 12. Kang, H., Lee, S., Chui, C.: Flow-based Image Abstraction. IEEE Transactions on Visualization and Computer Graphics 15(1), 62–76 (2009)
Maximized Energy Saving Sleep Mode in IEEE 802.16e/m Van Cuong Nguyen, Van Thuan Pham, and Bong-Kyo Moon Computer Science and Engineering, Dongguk University, Seoul, South Korea {cuongnv84,thuanhut}@gmail.com, [email protected]
Abstract. In this paper, we propose a new energy saving mechanism in IEEE 802.16e/m. By choosing dynamically sleep interval T for various traffic types and sending by Base Station (BS) via mobile-traffic-indication (MOB-TRFIND) message, a MSS only use constant sleep interval T in whole sleep mode duration until next transaction from awake mode to sleep mode. The new mechanism can be applied for PSC I and PSC II by changing T corresponding delay requirement of service. This mechanism also does not use MOB-SLPREQ/MOB-SLP-RSP that causes more energy consumption because of staying in waiting stage. The analytical results and simulation show that the proposed mechanism gains more than 36% of energy saving in maximized saving mode or 23% in optimized saving mode if it keeps response time change a little in compared with the standard sleep mode mechanism. Keywords: Maximized energy saving, dynamical sleep mode, optimized sleep interval, IEEE 802.16e/m.
mechanism can be applied for both PSC I and PSC II. The new mechanism has higher energy saving gain with the allowable transmission response time delay. The rest of this paper is organized as follows. In Section II, the standard sleep mode operation for IEEE 802.15e is introduced and what is our motivation. Section III describes our proposed MES mechanism and section IV shows the results of comparison between standard sleep mode and our proposal. The final section V is the conclusions about new mechanism.
2 Power-Saving Mechanism in IEEE 802.16e/m This section introduces the operation of the standard PSM for PSC I in the IEEE 802.16e and its analytical model [2] [3]. In general, an MSS has two modes: awake mode and sleep mode, shown in Fig. 1. After sending or receiving packets, an MSS needs to go to sleep stage by sending a sleep request message (MOB-SLP-REQ) which includes information such Tmin, Tmax, listening interval L and so on to its serving BS to get approval. If the serving BS accepts the sleep request from MSS, it sends MSS a response message (MOB-SLPRSP) which includes the beginning time of the sleep mode (TS), Tmin, Tmax, L and so on. After receiving the MOB-SLP-RSP, MSS starts sleep mode operation. During sleep mode operation, an MSS needs listening for a while to be sure if there is any packet transmitting to it. Sleep mode duration consists of sleep intervals and listening intervals. In 802.16e [1], sleep intervals are defined as follows: At the first sleep interval, a minimum sleep interval Tmin is used. After first sleep interval, an MS transits into a listening interval to wait for a MOB-TRF-IND message. This message indicates whether or not any traffic addressed to the MSS during previous sleep interval. If MOB-TRF-IND is negative, the MSS goes to the next sleep interval whose duration is doubles from the preceding sleep interval and the sleep mode operation continues. This process is repeated until the sleep interval reaches Tmax and then next sleep interval keeps unchanged. In j-th listening interval, if serving BS sends MSS a positive MOB-TRF-IND message, MSS leaves sleep mode and wakes up to process packets transmitted to. If we call the duration of first sleep interval T1=Tmin, then the duration of j-th sleep interval is
7M
PLQ^ M 7PLQ 7PD[ `
(1)
In the studied work of Xiao [2], [3], the performance of the sleep mode protocol in WIMAX/802.16e under variations of minimum and maximum sleep interval in sleep mode operation. The results are summarized as below: In this paper, we use the notation E[.] to stand the mean/average function. Let n denote the number of sleep intervals, D denote the duration of sleep mode before the MSS goes to awake mode. ES, EL denote the energy consumptions units per unit time in the sleep interval and the listening interval, respectively. R denotes the frame response
Maximized Energy Saving Sleep Mode in IEEE 802.16e/m
13
Fig. 1. Standard sleep mode in IEEE 802/16e
time which is defined as the delay a frame destined to an MSS has to wait before it is delivered. Let ej denote the event that there is at least one packet arrival during the monitor period j. Note that in our definition, listening intervals are also belong to the sleep mode.
WUXH H
3U H M
O 7 M /
(2)
The term Pr[n=j] represent the probability of success in exactly the j-th iteration, which is also the probability of failure in iteration 1 to j-1 and success in the j-th. The number of sleep cycles is an independent random variable.
3U Q
3U H
WUXH H
3U Q M 3U H
O 7 M /
IDOVH H M
IDOVH H M
WUXH
(3)
M
3U HL
WUXH 3U H M
WUXH
L
H
(>Q@
O
v
M
¦ 7L / L
H
¦
M 3 U Q
v
O
O 7 M /
(4)
M
M
¦
M
MH
M
¦ 7L / L
v
¦
M
MH
O
M
¦ 7L / L
(5)
14
V.C. Nguyen, V.T. Pham and B.-K. Moon v
M
¦ 3U Q
(> ' @
M ¦ 7N /
M v
¦H
O
N
M
¦ 7L / L
N
M v
¦ H
O
M
¦ 7L / L
M
¦ 7
N
/
N
v
v
¦H
O
M ¦ 7N ( 6 /( / N
M
¦ 7L / L
¦ H
M
¦ 7 ( N
M v
(6) M
¦ 3U Q M
6
/( /
6
/( /
N
O
M
¦ 7L / L
M
¦ 7 ( N
M
(> 5 @
/
N
M
( > (QHUJ\ @
M
¦ 7
N
v
¦ 3U Q
(7)
M 7 M /
M
v
¦H
O
M
¦ 7L / L
7 M /
M v
¦H
O
M
¦ 7L / L
7 M /
M
(8)
3 Maximized Energy Saving in IEEE 802.16E/M 3.1 Maximized Energy Saving for IEEE 802.16e/m Based on the above analysis, we can see that these parameters Tmin, Tmax are main factors affecting on the energy consumption and response time (delay). In general, each pair of (Tmin, Tmax) gives different results of energy consumption. So, if Tmin,Tmax are limited, we certainly choose a pair (Tmin, Tmax) which cause energy consumption is minimum correspond to a certain λ. The algorithm to choose Tmin-λ, Tmax-λ is below: Start For tmin=Tmin to Tmax For tmax=tmin to Tmax If(getEnergy(tmin,tmax,λ)<min_energy) Tmin_λ=tmin; Tmax_λ=tmax; min_energy=getEnergy(tmin,tmax,λ); end end end End
(9)
Maximized Energy Saving Sleep Mode in IEEE 802.16e/m
15
By scanning all possible cases of Tmin, Tmax, we choose the best case which has minimum energy consumption for given λ. So, let denote ψ(λ) is a function whose output is a pair (Tmin, Tmax) so that energy consumption calculated by (7) is minimum. ψ(λ) looks like below
To compare new mechanism in this paper with standard mode mechanism studied in Xiao work [2] [3], I chosen Tmin=1, Tmax=1024, L=1. Sleep energy consumption and listening energy consumption are chose based on [10], ES=30, EL=1. We also assume that packet arrival rate to a MSS follows Poisson distribution with mean λ. We consider λ in range of [0.001; 0.2]. Based the algorithm (9) we have following functions ψ(λ)
⎧(256,256) if 0.0010 ≤ λ ≤ 0.0016 ⎪ (128,128) if 0.0016 < λ ≤ 0.0058 ⎪ ⎪ (64, 64) if 0.0058 < λ ≤ 0.0194 Ψ (λ ) = ⎨ ⎪ (32,32) if 0.0194 < λ ≤ 0.0601 ⎪ (16,16) if 0.0601 < λ ≤ 0.1672 ⎪ ⎩ (8,8) if 0.1672 < λ ≤ 0.2
(11)
There is an interesting thing in results. We can recognize that Tmin=Tmax for given λ. It means that we always get a pair Tmin=Tmax corresponding to given λ so that the energy consumption by (7) is minimum.
3.2 Proposed Mechanism for Maximized Energy Saving (MES) A problem of standard sleep mode PSC I is to always use the fixed Tmin and fixed Tmax for various traffic types. It is not always good for all traffic rates. Another problem is switching time. An MSS has to send MOB-SLP-REQ message and wait to receive MOB-SLP-RSP if it wants transiting from awake mode to sleep mode. During switching time, the MS consumes the energy by staying in awake mode. Furthermore, the MSS also need energy for sending and receiving MOB-SLP-REQ/RSP. These problems motivate us to introduce a new sleep mode mechanism for IEEE 802.16e/m. Our energy saving energy mechanism does not use MOB-SLP-REQ/RSP to get sleep permission from BS. The mechanism lets BS choose the best sleep interval T for served MSS. An MSS uses constant sleep interval T send from its serving BS via MOB-TRF-IND message. Our new proposed mechanism operates as follow: after
16
V.C. Nguyen, V.T. Pham and B.-K. Moon
transmitting all packets in buffer, BS calculates the mean λ base on the packets which the BS has received in a given duration Δt. By using function ψ(λ) BS chooses the most suitable T (T=Tsuitable or T=Tsuitable/2 will be explained in section IV) for an MSS and send to MSS via MOB-TRF-IND message. MSS receives the MOB-TRF-IND message and goes to sleep mode operation. The MSS sleeps in the first sleep interval and listens in short time (L) to check whether there is any packet sent to by seeing MOB-TRF-IND is positive or negative. The next sleep interval is still T instead of 2T as in standard sleep mode. The value T is constant in the same sequence of sleep intervals until next transaction from awake mode to sleep mode. The new proposed energy saving mechanism is depicted in Fig. 2. In case of Tmin=Tmax=T, the formula (5)-(9) are rewrote as follows. Let p equal to
e− λ (T + L ) . Then n is a geometric random variable (>Q@
v
¦
M 3U Q
v
¦ M S
M
M
M
S
M
S S
(12)
Fig. 2. Maximized sleep mode mechanism for IEEE 802.16e/m
v
(> ' @
M
¦ 3U Q
M ¦ 7N /
M v
¦S
M
N
S M 7 /
M
S
v
¦ MS M
M
S 7 /
7 / S
(13)
Maximized Energy Saving Sleep Mode in IEEE 802.16e/m v
¦ 3U Q
( > (QHUJ\ @
M
v
¦S
M
17
M
M ¦ 7N ( 6 /( / N
S M 7( 6 /( /
M
S
v
¦ MS
M
S 7( 6 /( /
M
v
(>5@
¦3U Q
7( 6 /( / S
(14)
M 7M /
M
v M 7 / S S 7 / ¦ M
(15)
4 Performance Evaluation We made a simulation using MATLAB to evaluate our proposed mechanism. I also analytical evaluate and compare new mechanism with standard sleep mode in IEEE 802.16e. In both simulation and analytical evaluation, we use follow parameters: Tmin=1, Tmax=1024, L=1, ES=1, EL=30. We choose T=Tsuitable or T=Tsuitable/2 to get better response time. The total simulation time is 109 unit times. Fig. 3 shows simulation and analytical results of E[Energy]. In case of T=Tsuitale, Fig. 4, 5, 6, 7 show the comparison results between standard sleep mode and our scheme for IEEE 802/16e/m. Fig. 8 shows energy saving of our scheme in comparison with standard sleep mode. The formula of energy saving is
Energy saving(100%)=
Energy PSM − Energy MES x100% Energy PSM
Fig. 3. Analytical result simulation result (T=Tsuitable)
18
V.C. Nguyen, V.T. Pham and B.-K. Moon
Fig. 4. PSM vs MES on Energy consumption (T=Tsuitable)
Fig. 5. PSM vs MES on E[R] (T=Tsuitable)
Fig. 6. PSM vs MES on E[n] (T=Tsuitable)
Maximized Energy Saving Sleep Mode in IEEE 802.16e/m
Fig. 7. PSM vs MES on E[D] (T=Tsuitable)
Fig. 8. Saving Energy (T=Tsuitable)
Fig. 9. PSM vs MES on Energy consumption (T=Tsuitable/2)
19
20
V.C. Nguyen, V.T. Pham and B.-K. Moon
Fig. 10. PSM vs MES on E[R] (T=Tsuitable/2)
Fig. 11. PSM vs MES on E[n] (T=Tsuitable/2)
Fig. 12. PSM vs MES on E[D] (T=Tsuitable/2)
Maximized Energy Saving Sleep Mode in IEEE 802.16e/m
21
Fig. 13. Energy Saving (T=Tsuitable/2)
We can see in these Figs above, we have saved energy more than 36%. The response time, however, is also increased. From the formula (15), the response time is (T+L)/2. So, in case of T=Tsuitable/2, we get better both energy consumption result and response time results.
5 Conclusions In this paper, we have proposed a new energy saving mechanism without MOB-TRFREQ/RSP by using dynamical sleep interval corresponding to variation of traffic rates. The proposal lets BS decide which sleep interval time suitable for MSS by observing average packet arrival time. We provide analytical model for choosing sleep interval time to get maximized energy saving or optimized energy saving. Our mechanism can be applied for both PSC I and PSC II in IEEE 802.16e/m.
References 1. IEEE 802.16e-2006, IEEE Standard for Local and Metropolitan area networks, Part 16: Air Interface for Broadband Wireless Access Systems C Amendment for Physical and Medium Access Control Layers for Combined Fixed and Mobile Operation in Licensed Bands (February 2006) 2. Xiao, Y.: Energy saving mechanism in the IEEE 802.16e wireless man. IEEE Commun. Lett. 9(7), 595–597 (2005) 3. Xiao, Y.: Performance analysis of an energy saving mechanism in the IEEE 802.16e wireless MAN. In: Proc. IEEE CCNC, Las Vegas, NV, vol. 1, pp. 406–410 (January 2006) 4. Jung, W.J., Ki, H.J., Lee, T.J., Chung, M.Y.: Adaptive sleep mode algorithm in IEEE 802.16e. In: Proc. APCC, Bangkok, Thailand, pp. 483–486 (October 2007) 5. Xu, F., Zhong, W., Zhou, Z.: A novel adaptive energy saving mode in IEEE 802.16e system. In: Proc. MILCOM, Washington, DC, pp. 1–6 (October 2006)
22
V.C. Nguyen, V.T. Pham and B.-K. Moon
6. Jang, J., Han, K., Choi, S.: Adaptive power saving strategies for IEEE 802.16e mobile broadband wireless access. In: Proc. APCC, Busan, Korea, pp. 1–5 (August 2006) 7. Vasta, O.J., Raj, M., Kumar, R., Panigrahy, D., Das, D.: Adaptive power saving algorithm for mobile subscriber station in 802.16e. In: Proc. IEEE COMSWARE, Bangalore, India, pp. 1–7 (January 2007) 8. Xiao, J., Zou, S., Ren, B., Cheng, S.: An enhanced energy saving mechanism in IEEE 802.16e. In: Proc. IEEE GLOBECOM, San Francisco, CA, pp. 1–5 (November 2006) 9. Kim, M.G., Kang, M., Choi, J.Y.: Remaining energy-aware power management mechanism in the 802.16e MAC. In: Proc. IEEE CCNC, Las Vegas, NV, pp. 222–226 (January 2008) 10. Kim, M.G., Choi, J.Y., Kang, M.: Enhanced Power-Saving Mechanism to Maximize Operational Efficiency in IEEE 802.16e Systems. IEEE Transactions on Wireless Communications 8(9), 4710 (2009)
Propose New Structure for the Buildings Model Tuan Anh Nguyen gia [email protected]
Abstract. Urban data model (UDM) is a three-dimensional (3D) geography information systems (GIS) data model. The paper proposes the time class into UDM to manage historical of changes on the spatial properties of Point (0D), Line (1D), Surface (2D) and Body (3D) object. The changes would store explicit in database. The LOD (levels of detail) class of 3D objects
adds also to show buildings at four levels from simple to complex. Event class in the model should record the reasons to create changes on the 3D objects in their evolution. The last model named LT-UDM. Keywords: 3D, GIS data model, spatio-temporal model.
relationships the topology of space objects as well as lack of information about the semantics. While semantic information is needed, for example, we need to know the build time of a building A or the owner of the house B [4]. In addition, the studies of temporal-spatial data models also remarked. These issues also attracted many researchers for years [2, 8]; many models have been proposed with different authors: Event-Based model of Peuquet 2001, Chen 2000, Workboy 2005, State-Based model of Armstrong 1988, Langran 1992, and Liu 2006. 3D object-oriented data models Worboy 2005, Raza 1999. Most of these models performed for 2D objects [2]. These remarks show that these available 3D GIS models have the disadvantages:
The time dimension absences. 3D objects are only represented by one level.
The objectives of paper are to limit these disadvantages. Contents of the paper include presentation summarize for UDM and disadvantages of it (2). Propose adding the levels of detail on 3D objects by requirements the user. Levels of detail are four levels: 0, 1, 2, and 3 (3.A). Propose integration dimensions of time and events to keep track of changes on the properties of space-time. The changes will be stored fully and explicitly in the database (3.B). (4) Represent the experiments in Oracle 11g and C# language.
2 UDM and Disadvantages UDM (Urban data model) which is a model of 3D GIS for urban management proposed by Coors in 2003 [1, 11]. UDM built on four main subjects 0D, 1D, 2D, 3D by Point, Line, Surface and Body. Model (fig.1) used two geometric objects Node and Face. Node is described by three coordinates X, Y, Z. A Face is a triangle; it is represented by three Nodes. UDM is a 3D model that only describes the spatial properties of objects in a real world. UDM model has two main characteristics: size of data storage is small because the model omitted objects Arc (1D) and the model-represented surface of 3D block is good by triangulation. In 2010, Tuan Anh N.G proposed some innovations for 2D and 3D objects of UDM [10]. 3D objects were specified by cylinder, prism, cone and pyramid. These bodies were represented by new methods instead of represented by triangles. The innovations reduced query time and data size, name of this model is EUDM.
BODY
SURFA CE
LINE
1 *
*
*
POINT 1
* *
FA CE
1 NODE
*
3
Fig. 1. Urban Data Model
Propose New Structure for the Buildings Model
25
However, the model UDM or EUDM is still limited and this is the basis for recommending the proposes of this paper . These disadvantages of UDM are:
UDM has not mentioned the time dimension to manage the evolution of the objects. The representation of the geometric objects in UDM only describe in a fixed level.
3 Proposals 3.1 Additional Display of 3D Building According to the Different Levels of Detail Displaying 3D objects depend on the following: viewpoint, distance between the observer and the position of objects, the size of real 3D objects, a range of importance of objects in a specific application and different requirements of users. Table 1 presents the criteria levels of detail in the building management applications. Levels of detail divided into four (table 1). Table 1. Levels of detail (Lod) to represent a building Model Level of overview Structure of the roof Size (can change) Sub construction
Lod0 2D
Lod1 3D Buildings as blocks
Lod2 3D Buildings have floors
None
Flat
Flat
Lod3 3D Buildings have roof, windows, balcony Close to reality
>5x5x5(m)
>3x3x3(m)
>2mx2mx2m
Important construction
As blocks
Close to reality
Figure 3 describes a building is presented in levels 0, 1, 2, 3. Line is specified (fig.2) by Real-Line and Edge. Edge class is used to describe the segments in showing the building at Lod2, Lod3. Real-Line class describes 1D Lines that has in real world as streets, rivers. Surface is specified (fig.2) by Polygon and Window. Window class is used to describe windows of apartments in showing the building at Lod3 (fig.2). LINE
R EA L-LINE
SUR FA CE
EDGE
POLYGON
Fig. 2. Specifing Line and Surface class
W INDOW
26
T.A. Nguyen gia
LOD
Show
0
1
2
3
Fig. 3. Lod0, Lod1, Lod2, Lod3 of a building
Analyzing figure 3, three rules are recognized. A Building has many Bodies at many LODs. A Body exists in many Buildings at many LODs. A Building has many Surfaces at many LODs. A Surface exists in many Buildings at many LODs. A Building has many Lines at many LODs and a Line exists in many Buildings at many LODs. So there are links between objects: Building, LOD, Body; Building, LOD, Surface and Building, LOD, Line. The data model after proposal 3.A has as figure 4. BUILDING (#IDBD, NAME, DESC) LOD (#IDLOD, NAME, DESC1, DESC2, DESC3, DESC4) LODLINE (#IBBD, #IDL, #IDLOD) SURFACELOD (#IDBD, #IDS, #IDLOD) BODYLOD (#IDBD, #IDB, #IDLOD) BODY (#IDB, DESC) SURFACE (#IDS, DESC) LINE (#IDL, DESC)
3.2 Additional Time Dimension to UDM Non-temporal GIS only represents states of spatial properties. Temporal GIS represents only states but also events and evidences. The historical management of the spatial changes over time is an objective of GIS applications. To sole this problem, time class will be added to the model. Classes and links added to dimensional time in the model as figure 5. Time class is divided into two types of time, instant or an interval of time. For example, a building B built on 01/01/2010; 01/01/2011 is instant time. If B built on January 2, 2009 to January 2, 2011 then [2/01/2009, 2/01/2011] is an interval of time. DMY class is described by the day, month, year, hour, minute and second attributes. Event-Type class: is described by the name attributes of event. Event class: Event provides the information about the changes of objects. Event has the relation with the evolution of objects. Each event has a begin time and end time in real world and database. Begin time and end time can be instant or interval time. Table 2 is an example for state, event and evidence. Table 2. Example for state, event and evidence State House Bridge Road
Event Planning Earquake Flood
Evidence Survey Aerial Photography Remote sensing
28
T.A. Nguyen gia
Link between Time and Body: describes time of begin and the end of each Body. A Body has both times, begin time and end time in real world and database. Begin time and end time can be instant or interval time. Similar, link between Surface and Time: describes for the time begins and ends of Surface in real world and database. Link between the Line and Time: describes time that Line begins and ends in real world and database. Link between the Point and Time: describes time that Point begins and ends in real world and database. Link between Body and Event describe a Body is created by what event. Thus, the objects 0D, 1D, 2D, 3D always has 4 times, the begin time, the end time in real world and database. They are IDT1, IDT2, IDT3, IDT4. These times may be instant time or interval time. If it is instant time then INT-INST=0. UDM after two improvements used to integrate time, LOD classes has as figure 5 and the model named LT-UDM.
Fig. 5. Model LT-UDM ater adds recursive link
The model as figure 5 is analyzed into 16 relations BUILDING (#IDBD, NAME, DESC) LOD (#IDLOD, NAME, DESC1, DESC2, DESC3, DESC4) NODE (#IDN, X, Y, Z) LINENODE (#IDL, #IDN, SEQ) TRIANGLE (#IDT, IDN1, IDN2, IDN3, IDS) DMY (#IDMY, D/M/Y, H/M/S) EVENTYPE (#IDET, NAME) TIME (#IDT, IDDMY1, IDDMY2, INT-INST) EVENT (#IDE, IDT1, IDT2, IDT3, IDT4, IDET) LODLINE (#IBBD, #IDL, #IDLOD)
Propose New Structure for the Buildings Model
29
SURFACELOD (#IDBD, #IDS, #IDLOD) BODYLOD (#IDBD, #IDB, #IDLOD) BODY (#IDB, DESC, IDT1, IDT2, IDT3, IDT4). SURFACE (#IDS, DESC, IDT1, IDT2, IDT3, IDT4) LINE (#IDL, DESC, IDT1, IDT2, IDT3, IDT4) POINT (#IDP, DESC, IDT1, IDT2, IDT3, IDT4, IDN). 3.3 Parent and Children Relationship of Buildings Body class describes the Bodies. The recursive link of Body describes their parent and children relationship. For example, at 2006 as figure 6, with four Bodies B1, B2, B3, B4. At 2008: B1, B2, B3, B4 are merged to create B5. At 2010, B5 is divided into B6, B7. B5 is said the child of B1, B2, B3 and B4. B1, B2, B3 and B4 are said the parent of B5.
Fig. 6. B1, B2, B3, B4, B5, B6 at years 2006, 2008, and 2010
The model LT-UDM adds a recursive link of Body class as figure 7
Fig. 7. Model LT-UDM ater adds recursive link
30
T.A. Nguyen gia
A Body can has many parent Bodies and a Body can has many child Bodies. This model creates a new relation about parent and child of Bodies: BODY-CHIL-PAR BODY-CHIL-PAR (#IDBPAR, #IDBCHIL) #IDBPAR: Parent Bodies. #IDBCHIL: Child Bodies. Data in BODY-CHIL-PAR describes figure 6 as table 3. Table 3. Data in BODY-CHIL-PAR #IDBPAR B1 B2 B3 B4 B5 B5
#IDBCHIL B5 B5 B5 B5 B6 B7
B1, B2, B3, B4 have child B5 or parents of B5 are B1, B2, B3, B4. B5 is parent of B6, B7. The model LT-UDM is designed to integrate easy with EUDM.
4 Experiments The following experiments are installed in Oracle 11g and C #. Relational database management system does not support spatial data type and the operations in these data, so relational database management system need to extent. The result of extent creates the new database management system: relational-object [9]. This new system supports for spatial data type, and Oracle 11g have chosen. Two relations BODY (#IDB, DESC) and BODYFACE (#IDB, #IDF) in relational database would merge one relational-object BODY. Data structure of table BODY in Oracle 11g is created by: Create table BODY. (Idb number primary key, Desc varchar2 (50), Shape mdsys.sdo_geometry); In Oracle 11g, data type mdsys.sdo_geometry uses the storage set data. Code to insert one data row in BODY: Insert into BODY values (150,'1 cylinder block', 26, 35, 15, mdsys.sdo_geometry (2003, null, null, mdsys.sdo_elem_info_array (1, 1003, 1), mdsys.sdo_ordinate_array (25,0, 33,6, 23,6, 21,4, 16,4, 18,6, 8,6, 0,0, 25,0))); The sample data in database have inserted 30 buildings B1 and 20 buildings B2 (fig.8) and they are built in different years. The content of sample query is showing the buildings that are built from 2008 to 2009. Results of spatial query for time and LOD are presented through the forms. Form1 illustrates two the buildings were built from 2004 to 2005 and their data are shown in the level 0 as planes (fig.9). Form2 illustrates the buildings was built above time and their data are shown in the level 1 as the blocks (fig.9).
Propose New Structure for the Buildings Model
31
Fig. 8. (Left) buildings B1 and (right) B2
Fig. 9. Form1 (left) and form2 (right)
Form3 illustrates the buildings was built as query and their data are shown as the building has 10 floors (fig.10). Form4 illustrates the buildings was built from 2008 to 2009 and their data are shown as the building has 10 floors and windows of apartments (fig.10).
Fig. 10. Form3 (left) and form4 (right)
32
T.A. Nguyen gia
5 Conclusion The paper offers time dimension on the concept data model to record the evolutionary history of 0D, 1D, 2D, 3D objects over time. Time topology relationship between the buildings is described explicitly in the database. Finally, the paper also suggested adding LOD class into the model to server visualization of buildings at different levels. Last model reflects full GIS information for managing the building in the urban, including a full range of attributes: spatial, temporal and LOD. Particularly changes in the properties of spatial-temporal objects are stored completely in the database for the exploitation of information needs over time.
References 1. Abdul-Radman, A., Pilouk, M.: Spatial Data Modeling For 3D GIS, pp. 24–43. Springer, Heidelberg (2007) 2. Sabau, A.: The 3SST Relational Model, Studia Universitatis. Informatica LII (1), 77–88 (2007) 3. Koussa, C., Koehl, M.: A Simplified Geometric And Topological Modeling of 3D Building Enriched By Semantic Data: Combination of Surface-Based And Solid-Based Representations. In: ASPRS 2009 Annual Conference Baltimore, Maryland (2009) 4. Döllner, J., Kolbe, T.H., Liecke, F., Sgouros, T., Teichmann, K.: The Virtual 3D City Model Of Berlin - Managing, Integrating, And Communicating Complex Urban Information. In: Proceedings of the 25th Urban Data Management Symposium UDMS 2006, Aalborg, DK (2006) 5. Döllner, J., Buchholz, H.: Continuous Level-Of-Detail Modeling Of Buildings In 3D City Models. In: GIS 2005: Proceedings of the 13th Annual ACM International Workshop on Geographic Information Systems, pp. 173–181. ACM, New York (2005) 6. Hu, M.Y.: Semantic Based LOD Models Of 3D House Property. In: Proceedings of Commission II, ISPRS Congress, Beijing (2008) 7. Pelekis, N., Theodoulidis, B., Kopanakis, I., Theodoridis, Y.: Literature Review Of SpatioTemporal Database Models. Knowledge Engineering Review (2005) 8. Schmittwilken, J., Saatkamp, J., Förstner, W., Kolbe, T.H., Plümer, L.: A Semantic Model of Stair in Building Collars. Photogram-metric, Fernerkundung, Geoinformation (2007) 9. Schön, B., Laefer, D.F., Morrish, S.W., Bertolotto, M.: Three-Dimensional Spatial Information Systems: State of The Art Review. Recent Patents on Computer Science, 21– 31 (2009) 10. Tuan Anh, N.G.: Propose New Primitives For Urban Data Model. In: Proceedings of 3rd International Conference on Computer and Electrical Engineering, China, pp. 582–586 (2010), IEEE catalog number CFP1049I –PRT, ISBN: 978-4244-7224-6 11. Coors, V.: 3D GI in Networking Environments. In: International Workshop on 3D Cadastres, Delft (2003) 12. Zlatanova, S.: 3D GIS For Urban Development. PhD Thesis, ITC The Netherlands (2000)
Solving Incomplete Datasets in Soft Set Using Parity Bits of Supported Sets Ahmad Nazari Mohd.Rose1, Hasni Hassan1, Mohd Isa Awang1, Tutut Herawan2, and Mustafa Mat Deris3 1
FIT, Universiti Sultan Zainal Abidin, Terengganu, Malaysia 2 FSKKP, Univerisiti Malaysia Pahang, Pahang, Malaysia 3 FTMM, Universiti Tun Hussein Onn Malaysia, Johor, Malaysia (anm,hasni,khalid,isa)@unisza.edu.my, [email protected], [email protected]
Abstract. The theory of soft set proposed by Molodtsov [2]in 1999 is a new method for handling uncertain data and can be redefined as a Boolean-valued information system. The soft set theory has been applied to data analysis and decision support systems based on large datasets. Using retrieved datasets, we have calculated the supported values and then determine the even parity bits. Using the parity bit, the problem of missing values from the retrieved datasets can be solved. Keywords: Information system; Incomplete Datasets; Parity Bit, Supported Value; Soft set theory.
with incomplete datasets using soft set since most incomplete datasets were dealt using the approach of rough sets [8 – 10]. Recent work by Zou and Xiao in 2008 used the method of calculating the weighted-average of all possible choice values of the object and deciding the weight of each possible choice value by the distribution of other objects [11]. Our intention is to introduce another method that exploits the use of parity bit in supported sets in order to overcome the problem of missing attributes from the retrieved datasets. The concept of supported value of the Boolean-valued information based on soft set theory in reducing the data has been shown by M. Rose et al.[7]. Using the same technique of supported set as in [7], we then would like to propose a remedy for missing attribute values in datasets by utilizing the parity bits of the processed data. The main purpose of the proposed technique is to ensure that any attribute values that are missing after the data has been processed can still be recovered, hence maintaining the integrity of the data. Common method of applying parity bit is when an extra bit is added to the data in order to make the total quantity of 1’s on the data to be odd or even, depending upon whether one is employing odd parity checking or even parity checking Stalling[13]. It has been claimed that the use of parity bit checking is not foolproof due to its inability to detect multiple bit errors Held[14]. In contrast, we argued that owing to its simplicity in utilizing just one bit, we could solve the problems of multiple missing attributes by recursively applying the parity bit checking in the information systems of soft set theory. The rest of this paper is organized as follows. Section 2 describes the fundamentals of an information system and soft set theory. Section 3 presents the concepts of the supported sets. The definition of the used parity bits are clarified in Section 4. Section 5 will introduce the discussion regarding to the problem of missing attributes, solved using supported sets. Finally, we conclude our work in section 6.
2 Soft Set In this section, the concepts and definition of soft sets are been defined. 2.1 Information System An information system can be represented as a 4-tuple S = (U , A,V , f ) (also called
{
information (knowledge) function) where U = u1 , u2 ,L, u u
{
}
} is a non-empty finite set
of objects, A = a1 , a2 ,L, a A is a non-empty finite set of attributes, V =
U
ei ∈A
Ve , Ve i
i
is the domain (value set) of attribute a, f : U × A → V is an information function
such that f : U × A → V , f (u, a) ∈ Va , for every (u, a ) ∈ U × A . 2.2 Definitions
Throughout this section U refers to an initial universe, E is a set of parameters, P(U ) is the power set of U and A ⊆ E .
Solving Incomplete Datasets in Soft Set Using Parity Bits of Supported Sets
35
Definition 1. (See [2].)A pair (F, E ) is called a soft set over U, where F is a
mapping given by F : E → P(U ) hence
F : A → P(U ) .
In other words, a soft set over U is a parameterized family of subsets of the universe U. For ε ∈ A , F (ε ) may be considered as the set of ε -elements of the soft set
(F, E ) or as the set of ε -approximate elements of the soft set. Clearly, a soft set is not a (crisp) set. To illustrate this idea, let us consider the following example. Example 1
Lets illustrate the concept of a soft set by considering a soft set (F, E ) which describes the “attractiveness of apartment promotions” that Mr. X is considering to rent. Let assume that there are twenty apartment promotions in the universe U that are under consideration, U = {a1 , a 2 , L , a 20 } , and A is a set of decision parameters, E = {p1 , p 2 , p3 , p 4 , p 5 , p 6 , p 7 } , where p1 stands for the parameter “Expensive”, p2
stands for the parameter “Near to workplace”, p3 stands for the parameter “Condominium”, p4 stands for the parameter “Gated Security”, p5 stands for the parameter “Internal Alarm System”, p6 stands for the parameter “Close Proximity to Public Amenities” and p7 stands for the parameter “Swimming Pool”. Consider the
mapping F : E → P(U ) given by “credit card promotions (⋅) ”, where (⋅) is to be filled e∈ E . Suppose that the in by one of parameters a , a , a , a , a , a , a , a , a , a , a , a } , F ( p1 ) = {a1 , a2 , a3 , a4 , a5 , a6 , a7 , a8 , 9 10 11 12 13 14 15 16 17 18 19 20 a15 , a16 } , F ( p3 ) = {a2 , a3 , a4 , a7 , a11 , a12 , a15 , a16 , a18 , a19 } , F ( p 2 ) = {a2 , a3 , a11 , F (e4 ) = {a2 , a3 , a15 , a16 , a18 } , F (e5 ) = {a1 , a 4 , a 7 , a12 , a13 , a14 , a19 } , F (e6 ) = {a 4 , a7 , a11 , a12 , a19 } ,
}
F (e7 ) = {a1 , a 2 , a 3 , a 5 , a 6 , a8 , a 9 , a10 , a13 , a14 , a15 ,a16 , a17 , a18 . Thus, we can view the soft set
(F , E ) as a collection of approximations as illustrated below:
A.N. Mohd.Rose et al. Table 1. Tabular representation of a soft set in the above example Promotion
p1
p2
p3
p4
p5
p6
p7
a1 a2 a3
1
0
0
0
1
0
1
1
1
1
1
0
0
1
1
1
1
1
0
0
1
a4
1
0
1
0
1
1
0
a5
1
0
0
0
0
0
1
a6 a7 a8
1
0
0
0
0
0
1
1
0
1
0
1
1
0
1
0
0
0
0
0
1
a9 a10
1
0
0
0
0
0
1
1
0
0
0
0
0
1
0
1
1
0
0
1
0
1
0
1
0
1
1
0
a11 a12 a13
1
0
0
0
1
0
1
a14 a15 a16
1
0
0
0
1
0
1
1
1
1
1
0
0
1
1
1
1
1
0
0
1
a17 a18
1
0
0
0
0
0
1
1
1
1
1
0
0
1
a19 a20
1
0
1
0
1
1
0
1
0
0
0
0
0
1
In Table 1, a “1”denotes the presence of the described attributes while a “0” means that the attribute is not part of the description of the apartment. As observed from Table 1, apart from the use of “1” and “0” to denote whether the attribute can be used for the description of the apartment, the table can also be viewed as a Boolean-valued information system. Next, the relation between a soft set and a Boolean-valued information system is given as follows: Definition 2. (See [7].)If (F, E ) is a soft set over the universe U, then (F, E ) is a
binary-valued information system S = (U , A, V{0,1} , f ) .
Proof. Let (F, E ) be a soft set over the universe U, the mapping is defined as follows:
F = { f 1 , f 2 , L, f n } ,
Solving Incomplete Datasets in Soft Set Using Parity Bits of Supported Sets
37
where
⎧1, x ∈ F (ei ) f i : U → Vi and f i (x ) = ⎨ , for 1 ≤ i ≤ A . ⎩0, x ∉ F (ei ) Hence, if A = E , V =
U
ei ∈ A
Ve , where Ve = {0,1} , then a soft set (F, E ) can be coni
i
sidered as a binary-valued information system S = (U , A, V{0,1} , f ) .
□
From Definition 2, it is therefore permissible to represent a binary-valued information system as a soft set. Thus, we can construct a one-to-one correspondence between (F, E ) over U and S = (U , A, V{0,1} , f ) .
3 Supported Sets Throughout this sub-section the pair (F, E ) refers to the soft set over the universe U
representing a Boolean-valued information system S = (U , A,V{0,1} , f ) .
Definition 3. Let (F, E ) be a soft set over the universe U and u ∈ U . Support of an
object u is defined by supp(u ) = card({e ∈ E : f (u, e) = 1}) .
Definition 4. (See [2].) S = (U , A,V{0,1,*} , f ) .If (F, E ) is a soft set over the universe U,
then (F, E ) is an incomplete binary-valued information system
If any of the attribute value which should either be of a binary value, but contains null value, then it will be filled with * as an indicator for missing value. Hence, if A = E , V = e ∈A Ve , where Ve = {1,0,*} , then a soft set (F, E ) can be
U
i
i
considered as an incomplete binary-valued information system S = (U , A,V{0 ,1} , f ) . i
4 Parity Bits In this section, we will introduce a a concept based on the parity checking. A parity bit is added to a data packet for the purpose of error detection. In the even-parity convention, the value of the parity bit is chosen so that the total number of '1' digits in the combined data plus parity packet is an even number. In the following definitions, are definitions for the even parity checking used to check row and column wise. Definition 4.1. Let (F, E ) be a soft set over the universe U and u ∈ U . The even
parity bit for an object u is been defined as p bit = supp(u )mod2 .
Definition 4.2. Let (F, E ) be a soft set over the universe U and e ∈ A . The even
parity bit for an an attribute e is been defined as c bit = (∑in=1 f (u, ei )) mod 2 .
38
A.N. Mohd.Rose et al.
5 Discussions As for our discussion in solving the incomplete dataset, lets assume as in Example 4.1. Example 4.1. Let a soft set ( F , E ) representing studies the communication prowess among selected university student. Let assume that there are ten students that has been surveyed in the universe U with U = {u 1 , u 2 , L , u 10 } , and E is a set of parameters representing communication facilities that is been used by the student surveyed, E = {p1 , p 2 , p 3 , p 4 , p 5 , p 6 } , where p1 stands for the parameter for using communica-
tion facilities such as “email”, p2 stands for the parameter “facebook”, p3 stands for the parameter “blog”, p4 stands for the parameter “friendsters”, p5 stands for the parameter “yahoo messenger” and lastly p6 stands for the parameter “sms” . Consid-
er the mapping F : E → P(U ) given by “student communication prowess (⋅) ”, where (⋅) is to be filled in by one of parameters p ∈ E . As for example, F ( p1 ) means communication by email that is been used by and being represented functional value of {u1 , u3 , u 4 , u8 , u 9 , u10 } . While F ( p2 ) means communication by facebook is been
used by and being represented functional value of {u 2 , u3 , u 4 , u5 , u8 , u 9 } . And also F ( p3 ) means communication by sms is been used by and being represented functional value of {u 2 , u 3 , u9 , u10 } . Thus, the overall approximation can be represented as the follows: ⎧ email = {u1 , u3 , u 4 , u8 , u9 , u10 }, ⎫ ⎪ facebook = {u , u , u , u , u }, ⎪ 2 4 5 8 9 ⎪ ⎪ ⎪ blog = {u1 , u 2 , u 4 , u6 , u8 , u9 }, ⎪ (F , E ) = ⎨ ⎬ ⎪friendsters = {u 2 , u3 , u 4 , u8 , u10 },⎪ ⎪ ⎪ ym = {u 2 , u3 , u8 , u9 }, ⎪ ⎪ sms = {u 2 , u3 , u9 , u10 } ⎭ ⎩ Fig. 3. The soft set
The previous example can be represented in the following Boolean-valued information system. Using the calculated even parity bit as the guideline, we then are able determine the missing value in the case of missing values. Our strategy was first to solve the single missing values. The approach of solving single missing values was done horizontally then vertically. But in the case of more than one missing value, the support and aggregate value of each column will also be used in determining the missing values.
Solving Incomplete Datasets in Soft Set Using Parity Bits of Supported Sets Table 2. Tabular representation of a soft set from Example 4.1
U /P
p1
p2
p3
p4
p5
p6
u1
1
0
1
0
0
0
u2
0
1
1
1
1
1
u3
1
0
0
1
1
1
u4
1
1
1
1
0
0
u5
0
1
0
0
0
0
u6
0
0
1
0
0
0
u7
0
0
0
0
1
0
u8
1
1
1
1
1
0
u9
1
1
1
0
1
1
u10
1
0
0
1
0
0
As an example from Table 2, we have obtained the following values : Supp( pi ) = 5 , i = 2,8,9
( )
Supp p j = 4 , j = 3,4
Supp( pk ) = 2 , k = 1,10
Supp( pl ) = 1 , l = 5,6,7
Fig. 4. The support of each parameter Table 3. Integrating Even Parity bit into Tabular representation of a soft set of Table 2
U /P
p1
p2
p3
p4
p5
p6
p bit
u1
1
0
1
0
0
0
0
u2
0
1
1
1
1
1
1
u3
1
0
0
1
1
1
0
u4
1
1
1
1
0
0
0
u5
0
1
0
0
0
0
1
u6
0
0
1
0
0
0
1
u7
0
0
0
0
1
0
1
u8
1
1
1
1
1
0
1
u9
1
1
1
0
1
1
1 0
u10
1
0
0
1
0
0
c bit
0
1
0
1
1
1
39
40
A.N. Mohd.Rose et al.
In the following Table 4, incomplete attribute value will be denoted by symbol “*”. Table 4. An Example of missing attribute values
U /P
p1
p2
p3
p4
p5
p6
p bit
u1
1
0
*
0
0
0
0
u2
0
*
1
*
1
1
1
u3
1
0
*
*
*
1
0
u4
*
1
1
*
*
0
0
u5
0
1
0
*
0
0
1
u6
0
0
1
0
*
0
1
u7
0
0
0
0
1
0
1
u8
1
1
1
1
1
0
1
u9
1
1
1
0
1
1
1
u10
1
0
0
1
0
0
0
c bit
0
1
0
1
1
1
From the Table 4, it can be observed that there cases where there is only one missing value for object u1 and u 6 . Therefore, based on the obtained p bit = 0 , the missing value for p2 of u1 is determined to be 1. This is because only when the value
f (u1 , p3 ) = 1 , it will be able to produce p bit = 0 . As for u 6 , since p bit = 1 , therefore the missing value for f (u6 , p5 ) = 0 . We will then obtained the following table.
From the Table 6, it can be observed that there are cases of single value missing if viewed from the parameter perspective. As for parameter p1 , p 2 and p 3 , the derived c bit will be used to determine the missing values. For example, as can be observed, that
f (u3 , p3 ) = 0 is the most suitable option, since its c bit = 0. If
f (u3 , p3 ) = 1 , then we will obtain c bit = 1 , which is invalid. The same approach is also used to rectify the value of f (u2 , p2 ) = * , which should be corrected to f (u2 , p2 ) = 1 as to maintain c bit = 1 . As can been seen from the previous highlighted
single missing value examples, the parity bit can be used as the guidance in determining the missing values. It does not matter, whether to correct the missing values horizontally or vertically. By solving the single missing value horizontally and vertically based on p bit or c bit , there is the probability that the single missing values will recur in the outcome . So the step of solving single missing value by horizontal or vertical must be done recursively until there is no more single missing values. The following are steps that was taken to solve missing values from Table 6.
Solving Incomplete Datasets in Soft Set Using Parity Bits of Supported Sets
41
Table 5. Solved single missing value for objects
U /P
p1
p2
p3
p4
p5
p6
p bit
u1
1
0
1
0
0
0
0
u2
0
*
1
*
1
1
1
u3
1
0
*
*
*
1
0
u4
*
1
1
*
*
0
0
u5
0
1
0
*
0
0
1
u6
0
0
1
0
0
0
1
u7
0
0
0
0
1
0
1
u8
1
1
1
1
1
0
1
u9
1
1
1
0
1
1
1
u10
1
0
0
1
0
0
0
c bit
0
1
0
1
1
1
Table 6. Solved single missing value for parameters
U /P
p1
p2
p3
p4
p5
p6
p bit
u1
1
0
1
0
0
0
0
u2
0
1
1
*
1
1
1
u3
1
0
0
*
*
1
0
u4
1
1
1
*
*
0
0
u5
0
1
0
*
0
0
1
u6
0
0
1
0
0
0
1
u7
0
0
0
0
1
0
1
u8
1
1
1
1
1
0
1
u9
1
1
1
0
1
1
1
u10
1
0
0
1
0
0
0
c bit
0
1
0
1
1
1
Since we have already solved all single missing values. Then the next step that will be solving the next minimal number of missing values either by objects or parameter. Using Table 6, as the example for determining the missing attribute values, we hereby present the following steps:
42
A.N. Mohd.Rose et al.
5.1 Solving the Missing Two Values of u3
As for example for object u3 , there are two missing values and pbit = 0 . Since we have already known that supp( u3 ) is 4, therefore it is very obvious the two missing values are f (u3 , p4 ) = 1 and f (u3 , p5 ) = 1
5.2 Solving the Missing Two Values of u4
To solve the missing attribute values for u4 , is to first to try to solve the missing val-
ues as single missing values. This is achievable, for f (u3 , p4 ) = * and f (u3 , p5 ) = * it
can be solved vertically by using the c bit as the guidance. As for f (u3 , p4 ) = * , it can
be deduced that f (u3 , p4 ) = 1 , in order to maintain c bit = 1.
6 Conclusion In this paper, we have defined supported sets in accordance with the boolean information systems. While in [7], supported sets has proven feasible in reducing the data set, but in our paper we have used even parity bits and supported set to solve incomplete information systems of soft set theory. We started off by solving recursively the single missing values for objects and parameters. We have also shown that in cases where there more than one single value missing, it also feasible to determine the the missing value by using supported value and parity bit. The importance of this solution is that it helps to maintain the integrity of the file by successfully recovering the correct missing values. We intend to elaborate more on other solutions for solving incomplete information systems of soft set theory in different kind situation.
Acknowledgement This work was supported by the FRGS of Ministry of Higher Education, Malaysia, with reference to FRGS/2/2010/SG/UDM/02/2,
References [1] Zhao, Y., Luo, F., Wong, S.K.M., Yao, Y.: A general definition of an attribute reduct. In: Yao, J., Lingras, P., Wu, W.-Z., Szczuka, M.S., Cercone, N.J., Ślȩzak, D. (eds.) RSKT 2007. LNCS (LNAI), vol. 4481, pp. 101–108. Springer, Heidelberg (2007) [2] Molodtsov, D.: Soft set theory-first results. Computers and Mathematics with Applications 37, 19–31 (1999) [3] Maji, P.K., Roy, A.R., Biswas, R.: An application of soft sets in a decision making problem. Computer and Mathematics with Application 44, 1077–1083 (2002)
Solving Incomplete Datasets in Soft Set Using Parity Bits of Supported Sets
43
[4] Chen, D., Tsang, E.C.C., Yeung, D.S., Wang, X.: The Parameterization Reduction of Soft Sets and its Applications. Computers and Mathematics with Applications 49, 757–763 (2005) [5] Kong, Z., Gao, L., Wang, L., Li, S.: The normal parameter reduction of soft sets and its algorithm. Computers and Mathematics with Applications 56, 3029–3037 (2008) [6] Pawlak, Z., Skowron, A.: Rudiments of rough sets. Information Sciences 177(1), 3–27 (2007) [7] Rose, A.N.M., Herawan, T., Mat Deris, M.: A Framework of Decision Making based on Maximal Supported Sets. In: Zhang, L., Lu, B.-L., Kwok, J. (eds.) ISNN 2010. LNCS, vol. 6063, pp. 473–482. Springer, Heidelberg (2010) [8] Thiesson, B.: Accelerated quantification of Bayesian networks within complete data. In: The First International Conference on Knowledge Discovery and Data Mining, Montreal, Canada, pp. 306–311 (1995) [9] Zhang, D.X., Li, X.Y.: An absolute information quantity-based datamaking-up algorithms of incomplete information system. Comput. Eng. Appl. 22, 155–197 (2006) [10] Quinlan, J.R.: Unknown attribute values in induction. In: Proceedings of the 6th International Machine Learning Workshop, SanMateo, Canada, pp. 164–168 (1989) [11] Zou, Y., Xiao, Z.: Data Analysis approaches of sets under incomplete information. Knowledge Based Systems 21, 941–945 (2010) [12] Herawan, T., Rose, A.N.M., Mat Deris, M.: Soft set theoretic approach for dimensionality reduction. Database Theory and Application Communications in Computer and Information Science 64, 171–178 (2009) [13] Stalling, W.: Data and Computer Communications, 7th edn. Pearson Education, Inc., Pearson Prentice Hall, New Jersey (2004) [14] Held, G.: Understanding Data Communications – From Fundamentals to Networking, 2nd edn. John Wiley and Sons, West Sussex (1997)
An Algorithm for Mining Decision Rules Based on Decision Network and Rough Set Theory Hossam Abd Elmaksoud Mohamed Department of mathematics, College of Science and Humanitarians studies, Al-Kharj University, Saudi Arabia eng_hossam21@ yahoo.com
Abstract. Knowledge acquisition continues to be a challenging and time consuming task in building decision support systems. Rule induction is a data mining process for acquiring knowledge in terms of decision rules from a number of specific 'examples' to explain the inherent causal relationship between conditional factors and a given decision. This paper introduces a method of rules extraction for fault diagnosis based on rough set theory and decision network. The fault diagnosis decision system attributes are reduced firstly, and then a decision network with different reduced levels is constructed. Initialize the network’s node with the attribute reduction sets and extract the decision rule sets according to the node of the decision network. In addition, the coverage degree and the certainty factor were applied to filter noise and evaluate the extraction rules. The proposed methodology cannot only set up rational and succinct diagnosis model for large complicated power system but also it can dispose uncertainty information of substation and get correct diagnosis result under incomplete information. At last an example is given; the result indicates that the method can diagnosis single fault and multi-fault efficiently and could be used to assist operators in their decision-making processes. Keywords: Knowledge discovery, Rough set, Decision network, Decision Rules, fault diagnosis, Rule extraction, Attributes Reduction.
An Algorithm for Mining Decision Rules Based on Decision Network
45
Networks[12], fuzzy sets [17], inductive learning [2], wavelet transformation [3], etc…,In addition, the Chaos and Fractal Theories already begun to attract people’s attention. Through above ways have their advantages, but also exist some demerits. For example: expert diagnosis system needs complete and correct information of field, but the practical status is not so, the fault information acquired may be incomplete and uncertainty. Though ANN has ability of fault-tolerance, but its slow convergence speed effects its application. Genetic Algorithm (GA) has ability of global optimization, but it can’t get correct result under condition of uncertainty and incomplete information. Recently, as a mathematic tool which is newly used to deal with fuzzy and uncertain knowledge, rough set theory has been successfully applied in fault diagnosis of distribution networks [1, 7]. Rough Set Theory (RST), was first introduced by Pawlak [5] in 1982 and it has been applied to machine learning, knowledge acquisition, decision analysis, pattern recognition, and decision support systems etc. Rough set theory provides rigorous mathematical techniques for creating approximate descriptions of objects for fault diagnosis of power transformer, and it does not need the domain or prior knowledge. In this paper, a new fault diagnosis method for Distribution System based on rough set theory and Decision network is presented. By reducing the fault diagnosis attributes, then a decision network with different reduced levels is constructed. Initialize the network’s node with the attribute reduction sets and extract the decision rule sets according to the node of the decision network. In addition, the coverage degree and the certainty factor were applied to filter noise and evaluate the extraction rules. The proposed method can find more objective and effective diagnostic rules and has yielded promising results
2 Review of Rough Set Theory The Rough Set Theory is a new mathematical tool presented to dispose incomplete and uncertainty problem by Pawlak in 1982. He defined the knowledge according to new point of view, and regarded it as partition of universe. The concept of a rough set can be defined quite generally by means of topological operation, interior and closure, called approximations. Rough set theory can discover implicit knowledge and open out potentially useful rule by efficiently analyzing and dealing with all kinds of imprecise, incomplete and disaccord information. 2.1 Decision System Definition 1): In rough set theory, an information system can be considered as system S =(U, A,V, f) , where U is the universe; A = C ∪ D is the sets of fault attribute, the subset C and D are disjoint sets of fault symptoms attribute and fault decision V = ∪ Va attributes respectively; ,where Va is the value set of fault symptoms r∈R
46
Hossam A. Nabwey
attribute a ,is named the domain of attribute a .Each attribute a ∈ A ; f is an information function f : UxA → V , and f ( x,a ) ∈ Va , in which x ∈ U .
2.2 Equivalent Relations
Definition 2): In decision system S = (U, A, V, f), every attributes subset, an indiscernible relation (or equivalence relation) IND(B) defined in the following way:
IND(B) =
{( x,y ) ∈ Ux U ∀a ∈ B, f ( x, a ) = f ( y, a )}
(1)
The family of all equivalence relation of IND( B), a partition determined by B, denoted by U/IND(B),[x]B can be considered as equivalence classes, and defined as follows:
[ x ]B = {y ∈ U ∀a ∈ B, f ( x, a ) = f ( y , a )}
(2)
[ x ] IND ( B ) = ∩ [ x ] B
(3)
And
2.3 Reduction and Core
Definition 3): In decision system S=(U, A,V ,f),Let b ∈ B and B ⊆ A , if posB(D) = posB −{b}(D), attribute b is redundant to B, which relatives to D, otherwise the attribute b is indispensable. If IND(B) = IND(A) and POSB (D) ≠ POSB −{b} (D) , then B is called a reduction for information system S , are denoted as RED( A) ; the intersection of these reduction sets is called core, denoted as CORE = ∩ RED( A) .
3 Decision Networks and Decision Rules The main problem in data mining consists in discovering patterns in data. The patterns are usually expressed in form of decision rules, which are logical expressions in the form “if X then Y” where X and Y are referred to as predecessor (conditions) and successor (decisions) of the rule, respectively. Any set of decision rules is called a decision algorithm. Thus knowledge discovery from data consists in representing hidden relationships between data in a form of decision algorithms. However, for some applications, it is not enough to give only set of decision rules describing relationships in the database. Sometimes also knowledge of relationship between decision rules is necessary in order to understand better data structures. To this end we propose to employ a decision algorithm in which also relationship between decision rules is pointed out, called a decision network.
An Algorithm for Mining Decision Rules Based on Decision Network
47
The decision network is a finite, directed acyclic graph, nodes of which represent logical formulas, whereas branches are interpreted as decision rules. Thus every path in the graph represents a chain of decisions rules, which will be used to describe compound decisions. Let U be a non empty finite set, called the universe and let X, Y be logical formulas. The meaning of X in U, denoted by X is the set of all elements of U, that satisfies X in U. The truth value of X denoted val ( X ) is defined as card X card (U ) where card(X) denotes cardinality of X.
The number Supp ( X ,Y ) = Card ( X ∧Y ) will be called a support of the rule [6].
With every decision rule X → Y we associate its strength defined as: str ( X , Y ) = supp( X ,Y ) card (U )
(4)
Moreover, the certainty factor defined as: cer ( X , Y ) = str ( X ,Y ) val ( X )
(5)
And the coverage factor defined as : cov ( X , Y ) = str ( X ,Y ) val (Y )
(6)
4 Fault Diagnosis Method Based on Rough Set Theory and Decision Network The fault diagnosis of distribution system can be considered as a problem of mode classifying. So the decision table of rough set is fit for it. For distribution system fault diagnosis the information of circuit breakers and protection relay is adopted to judge fault components, such as busbar, feed line, and main transformer. As shown in Figure 1 a part of main wiring diagram of distribution system. It contains five components (busbar A, busbar B, transformer T, line L1 and line L2), Four breakers (CB1, CB2, CB3, CB4) and ten protection relays (Am, Bm, Tam, TBm, L1m, L2m, L1p, L2p, Tap and TBp). Here m and p represent primary protection relay, backup relay respectively, Am, Bm is primary protection of busbar A and B respectively, TAm and TBm is primary protection of transformer, Tap and TBp is backup protection of transformer. The fault conditional attributes of classification consist of the information of relays and breakers. The decision attributes is made up of those dubitable fault elements, such as busbar, transmission line, transformer and zones. In the diagnosis decision table “1” represents the condition that the breakers or relays have been disconnected. On the other side, “0” represents the contrary condition.
48
Hossam A. Nabwey
Fig. 1. A part of main wiring diagram of distribution system
The diagnosis method based on rough set theory and decision network as follows
Step1. Set up two fault sample sets of distribution system. One is about power cut area, and form decision table 1; another is about fault component and form decision table 2. The partition method of power cut area is as follows: consider every busbar as reference point and the making partition begin from the power supply entering busbar; end in the exporting busbar where the lines are connected. There are some busbars then there are some areas. In this example there are two partitions. Figure 2 is a partition illustration of Figure1. Step2. Make reduction for decision table 1. Decision table 1 is reduced by using software called Rosetta (a Rough Set tool kit for analysis of data). The GA is adopted in Reduction process. Eleven reductions are produced; they are shown in table 3. Step3.The decision attribute of table 1 is used as condition attribute in decision table 2, also make reduction for decision table 2 Similarly as done in step 2, 30 reductions are produced; they are shown in table 4.
Fig. 2. A partition illustration
An Algorithm for Mining Decision Rules Based on Decision Network
49
Table 1. Decision Table 1
U X1 X2 X3 X4 X5 X6 X7 X8 X9
CB1 CB2 CB3 CB4 Am Bm TAm TBm L1m L2m TAp TBp L1p L2p 1 1 1 0 0 0 0 0 0
0 1 0 1 1 0 0 0 0
0 0 0 0 1 1 0 1 0
0 0 0 0 1 0 1 0 1
1 0 0 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0
0 0 0 0 0 1 0 0 0
0 0 0 0 0 0 1 0 0
0 1 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0
0 0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 1
0 0 1 0 0 0 0 0 0
0 0 0 1 0 0 0 0 0
Power cut zone Z1 Z1 Z1 Z1 Z2 Z2 Z2 Z2 Z2
Table 2. Decision Table 2
U CB1 CB2 CB3 CB4 Am Bm TAm TBm L1m L2m TAp TBp L1p L2p X1 X2 X3 X4 X5 X6 X7 X8 X9
Redu Lengt h 3 CB4, Power cut zone} CB3, CB4} 3 CB3, Am} 3 CB3, Power cut zone} 3 CB4, Am} 3 CB4, Am} 3 CB2, CB4} 3 CB2, CB3} 3 CB2, TBm, TBp} 4 TAm, TAp, Power cut zone} 4 CB3, TBm, TBp} 4 CB2, TAm, TAp} 4 TBm, TBp, Power cut zone} 4 CB4, TAm, TAp} 4 CB4, L1m, L2m} 4 Am, Bm, Power cut zone} 4 CB4, L1m, L1p} 4 Am, Bm, Power cut zone} 4 CB3, Am, Bm} 4 CB4, L2m, L2p} 4 Am, Bm, L1m, L2m} 5 CB4, Am, Bm, TAp} 5 Am, Bm, L1m, L1p} 5 CB4, Bm, L1m, L1p} 5 CB4, Am, TBm, TBp} 5 CB4, Am, Bm, TBm} 5 Am, Bm, TBm, TBp} 5 Am, Bm, TAm, TAp} 5 Bm, L1m, L1p, Power cut 5 Am, Bm, TAm, TAp} 5
Step4. Construct Fault decision network for decision table 1 and decision table 2. As shown from table 3 and Table 4, that there is 11 reducts for decision table 1 and 30 reducts for decision table 2. So if we construct the decision network it will be complicated. In order to get a simple decision network the frequency of appearance of each conditional attribute in Table 3 is shown in Figure 3, and By ranking conditional attributes in the descending order of appearance frequency, obtain for each reduct set the sum of frequency of attributes corresponding to it and length equal three, one can see that the most significant reduced sets are: ({CB1, CB2, CB3}, {CB1, CB2, CB4}, {CB1, CB2, Bm}). Hence the fault decision network for table 1 is constructed as shown in Figure 4.
An Algorithm for Mining Decision Rules Based on Decision Network
51
Fig. 3. The frequency of appearance of conditional attributes in Table 3
Step5. Extracting the fault diagnosis rules From the decision network shown in figure 4, it can be seen that all the nodes with decision rules except the bottom empty node, which ensures extracting effective diagnosis rules from the incomplete diagnostic information. Table 5. Power Cut Zone Decision Rules Rule IF (CB3 = 0) And (CB4 = 0) THEN ( ZONE = Z1) IF (CB3 = 1) And (CB4 = *) THEN ( ZONE = Z2) IF (CB3 = *) And (CB4 = 1) THEN ( ZONE = Z2)
Table 6. Fault Component Decision Rules Rule IF (CB2 = 0) And (CB4 = 0) And (Z = Z1) THEN ( fault device = A) IF (CB2 = 1) And (CB4 = 0) And (Z = Z1) THEN ( fault device = T) IF (CB2 = 1) And (CB4 = 1) And (Z = Z2) THEN ( fault device = B) IF (CB2 = 0) And (CB4 = 0) And (Z = Z2) THEN ( fault device = IF (CB2 = 0) And (CB4 = 1) And (Z = Z2) THEN ( fault device =
Fig. 4. The decision network for decision table 1
52
Hossam A. Nabwey
For each node of the decision network, calculate the rule’s accuracy and coverage degree. Extract decision rules whose accuracy is greater than the threshold value μ0 . Set μ0 = 0.8 , if all the confidence degrees are less than 0.8 put the biggest two confidence degree items in rule set. The corresponding decision rules of table 1 are shown in table 5. Similarly, the corresponding decision rules of table 2 are shown in table 6.
5 Example of Fault Diagnosis In order to validate the correction of the mined diagnosis rules, the judgment for local distribution network shown in figure 1 is validated on the condition of two kinds of fault information. 5.1 On the Condition of Complete Information
Suppose that fault information is; CB1=1, TAp=1, TBm=1, CB2=1. According to the above method, firstly the power cut zone Z1 can be diagnosed by decision rules shown in table 5, then the fault component T is diagnosed by decision rules shown in table 6. The conclusion is true. 5.2 On the Condition of Incomplete Information
Suppose that fault information which should be gained is; CB1=1, TAp=1, TBm=1, CB2=1, but because of the fault of communication and device the information of CB1 and TBm aren’t transmitted to control center. On this condition the fault component T still can be diagnosed. In this example the lost information isn’t necessary core attribute of decision. If the lost information is core attribute of decision, then Hamming error will be adopted to judge, and the result is educed after synthetic estimation.
6 Conclusion This paper has presented a method of rules extraction for fault diagnosis based on rough set theory and decision network. The fault diagnosis problem is denoted by a decision network with different simplify level. According to the node in different level of the decision network, the uniform length rules of fault decision can be acquired. In addition, the concepts of coverage degree and certainty factor were applied to filter noise and improve diagnosis rules extraction efficiency. The proposed method can find more objective and effective diagnostic rules and has yielded promising results. It cannot only set up rational and succinct diagnosis model for large complicated power system but also it can dispose uncertainty information of substation and get correct diagnosis result under incomplete information.
An Algorithm for Mining Decision Rules Based on Decision Network
53
Acknowledgment I thank the President of Al-Kharj University, Deanship of Scientific Research in AlKharj University, Dean of College of Science and Humanitarians studies and all the Staff of Department of mathematics for their continuous support, encouraging and for their useful and fruitful discussion.
References [1] Zhao, D., Han, Y., Gao, S.: A New Method of Decision Table Reduct about Power Networks Fault Diagnosis. Automation of Electric Power Systems 28(4), 63–66 (2004) [2] YueShun, H., QiuLin, D.: Fault Mode Analyze of Power System Based on Data Mining. In: Proceedings of the 2009 International Symposium on Web Information Systems and Applications (WISA 2009) (2009) [3] Huang, Y.: A new data mining approach to dissolved gas analysis of oil-insulated power apparatus. IEEE Transactions on Power Delivery 18, 1257–1261 (2003) [4] Kaminaris, S.D., Moronis, A.X., Kolliopoulos, N.I., Sakarellos, A.P.: PTME – a new expert system application for Power Transformer Troubleshooting and Maintenance. In: Proceedings of the 6th WSEAS Int. Conf. on Artificial Intelligence, Knowledge Engineering and Data Bases, Corfu Island, Greece, February 16-19, pp. 52–57 (2007) [5] Pawlak, Z., Grzymala-Busse, J., Slowinski, R., et al.: Rough sets. Communication of the ACM 38(11), 89–95 (1995) [6] Qhrn A.: Discernibility and Rough Sets in Medicine: Tools and Applications, Ph.D dissertation, Dep. Compu. Sci. Inform. Sci., Norwegian Univ. Sci. Technol., Trondheim, Norway (2000) [7] Zhang, Q., Han, Z., Wen, F.: Fault Diagnosis Method of Power System Based on Rough Set Theory and a New Method on Alarm Manipulation. China Power 31(4), 32–35 (1998) [8] Hongxia, Q., Zhangzhuo, D., Qihong, S., et al.: An expert system for substation Fault Diagnosis and Restorative Operation Based on Object-oriented Technique Part1: Design of Expert System and Modeling. Automation of Electric Power System 20(9), 17–25 (1996) [9] Yang, Q.: Distribution Networks. China Power Publishing Company, Beijing (November 1998) [10] Salat, R., Osowski, S.: Accurate fault location in the power transmission line using support vector machine approach. IEEE Trans. Power Systems 19(2), 979–986 (2004) [11] Thukaram, D., Khincha, H.P., Vijaynarasimha, H.P.: Artificial neural network and support vector machine approach for locating faults in radial distribution systems. IEEE Trans. Power Delivery 20(2), 710–721 (2005) [12] Vasilic, S., Kezunovic, M.: Fuzzy ART neural network algorithm for classifying the power system faults. IEEE Trans. Power Delivery 20(2), 1306–1314 (2005) [13] Chen, W.-H., Liu, C.-W., Tsai, M.-S.: Fault diagnosis in distribution substations using CE-nets via Boolean rule matrix transformation. IEEE Power Engineering Society Summer Meeting 1, 416–420 (2000)
54
Hossam A. Nabwey
[14] Huang, Y.-C., Yang, H.-T., Huang, C.-l.: A new intelligent hierarchical fault diagnosis system for power networks. IEEE Transaction on Power System 12(1), 349–356 (1997) [15] Park, Y.M., Kim, G.-W., Sohn, J.-M.: A Logic Based Expert System (LBES) for Fault Diagnosis of Power System. IEEE Trans. on PWRS 12(1), 363–369 (1997) [16] Zhang, D., Liu, Z., Yang, Y.: Exolanation Fault Diagnosis Expert system for substation. Journal of North China Electric Power University 25(1), 1–7 (1998) [17] Zhang, N., Kezunovic, M.: Coordinating fuzzy ART neural networks to improve transmission line fault detection and classification. In: Proc. IEEE PES General Meeting, San Francisco (June 2005)
A Probabilistic Rough Set Approach to Rule Discovery Hossam Abd Elmaksoud Mohamed Department of mathematics, College of Science and Humanitarians studies, Al-Kharj University, Saudi Arabia eng_hossam21@ yahoo.com
Abstract. Rough set theory is a relative new tool that deals with vagueness and uncertainty inherent in decision making. In this paper we suggest a new rough algorithm for reducing dimensions and extracting rules of information systems using expert systems. This paper introduces a probabilistic rough set approach to discover grade rules of transformer evaluation when there is a missing failure symptom of transformer. The core of the approach is a soft hybrid induction system called the Generalized Distribution Table and Rough Set System (GDTRS) for discovering classification rules. The system is based on a combination of Generalized Distribution Table (GDT) and the Rough Set methodologies. With every decision rule two conditional probabilities associated, namely the certainty factor and the coverage factor. The probabilistic properties of the Decision rules are discussed. Keywords: Knowledge discovery, Rough set, Rule extraction, Attributes Reduction Missing attribute values, Generalized Distribution Table (GDT), Rule discovery.
The core of the approach is a soft hybrid induction system called the Generalized Distribution Table and Rough Set System (GDT-RS) for discovering classification rules. The system is based on a combination of Generalized Distribution Table (GDT) and the Rough Set methodologies. Table 1. An incompletely specified decision table
transformer 1
Valid utilization degree a1
Attributes Maintenance cost b1
Decision Reliability
Grade
c1
II
c0
I
2
-
b0
3
a1
b0
c1
II
4
a1
b0
I
5
a0
b0
* c0
II
6
a0
?
-
II
7
a1
c1
II
8
a1
? b1
c0
I
2 Rough Set and Missing Attribute Values Missing attribute values are commonly existing in real world data set. They may come from the data collecting process or redundant diagnose tests, unknown data and so on. Since the main concern is learning from examples, and an example with a missing decision value, (i.e., not classified) is useless [5], we will assume that only attribute values may be missing. Discarding all data containing the missing attribute values cannot fully preserve the characteristics of the original data. So In data analysis two main strategies are used to deal with missing attribute values in data tables. The former strategy is based on conversion of incomplete data sets (i.e., data sets with missing attribute values) into complete data sets and then acquiring knowledge. The process to change the incomplete data set into complete data set, say to transform the missing data into specified data via some technique, is called completeness of data set. Multiple approaches on filling in the missing attribute values were introduced [6],[7], such as selecting the “most common attribute value”, the “concept most common attribute value”, “assigning all possible values of the attribute restricted to the given concept”, “ignoring examples with unknown attribute values”, “treating missing attribute values as special values”, “event covering method” and so on. In this strategy conversion of incomplete data sets to complete data sets is a preprocessing to the main process of data mining. In the later strategy, knowledge is acquired from incomplete data sets taking into account that some attribute values are missing. The original data sets are not converted into complete data sets.
A Probabilistic Rough Set Approach to Rule Discovery
57
The later strategy is exemplified by the C4.5 approach to missing attribute values [8] or by a modified LEM2 algorithm [9, 10]. In both algorithms original data sets with missing attribute values are not preprocessed. This paper will concentrate on the later strategy used for rule induction, i.e., it will be assumed that the rule sets are induced from the original data sets, with missing attribute values, not preprocessed as in the former strategy. The next basic assumption is that there are three approaches to missing attribute values [11]: The first approach is that an attribute value, for a specific case, is lost. For example, originally the attribute value was known, however, due to a variety of reasons, currently the value is not available. Maybe it was recorded but later it was erased. The second approach is that an attribute value was not relevant , the case was decided to be a member of some concept, i.e., was classified, or diagnosed, in spite of the fact that some attribute values were not known. For example, it was feasible to diagnose a patient in spite of the fact that some test results were not taken (here attributes correspond to tests, so attribute values are test results). Since such missing attribute values do not matter for the final outcome, we will call them "do not care" conditions. The third approach is a partial "do not care" condition, we assume that the missing attribute value belongs to the set of typical attribute values for all cases from the same concept. Such a missing attribute value will be called an attribute-concept value. Calling it concept "do not care" condition would be perhaps better, but this name is too long. In the sequel it is assumed that all decision values are specified. Also, all missing attribute values are denoted either by "?" or by "*", or by "-", lost values will be denoted by "?", "do not care" conditions will be denoted by "*", and attribute-concept value will be denoted by "-". Additionally, it is assume that for each case at least one attribute value is specified. An example of an incompletely specified table is presented in Table 1 Obviously, in rough set theory any decision table defines a function ρ that maps the set of ordered pairs (case, attribute) into the set of all values [12]. For example, in table1, Ρ (1, Valid utilization degree) = a1 . Rough set theory [13] is based on the idea of an indiscernibility relation. The indiscernibility relation IND(B) is an equivalence relation. Equivalence classes of IND(B) are called elementary sets of B and are denoted by [x]B. The indiscernibility relation IND(B) may be computed using the idea of blocks of attribute-value pairs. Let a be an attribute and let v be a value of a for some case. For complete decision tables if t = (a, v) is an attribute-value pair then a block of t, denoted [t], is a set of all cases from U that for attribute a have value v. For incomplete decision tables the definition of a block of an attribute-value pair must be modified as follow : •
If for an attribute a there exists a case x such that ρ(x, a) = ?, i.e., the corresponding value is lost, then the case x should not be included in any block [( a, v )] for all values v of attribute a.
58
Hossam A. Nabwey
• •
If for an attribute a there exists a case x such that the corresponding value is a "do not care" condition, i.e., ρ (x, a) = *, then the corresponding case x should be included in blocks [(a, v)] for all specified values v of attribute a. If for an attribute a there exists a case x such that the corresponding value is a attribute-concept value, i.e., ρ (x, a) = –, then the corresponding case x should be included in blocks [(a, v)] for all specified values v of attribute a that are members of the set V(x, a), where V(x, a) = { ρ (y, a) | y ∈ U , ρ (y, d) = ρ (x, d)},
[(Maintenance cost, b0 )] = {2, 3, 4, 5}, [(Reliability, c1 )] = {1, 3, 4, 6, 7}, [(Reliability, c 0 )] = {2, 4, 5, 6}. These modifications of the definition of the block of attribute-value pair are consistent with the interpretation of missing attribute values [11] lost, "do not care" conditions, and attribute-concept values. Also, note that the attribute-concept value is the most universal, since if V(x, a) = Ø, the definition of the attribute-concept value is reduced to the lost value, and if V(x, a) is the set of all values of an attribute a, the attribute-concept value becomes a "do not care" condition.
3 Generalized Distribution Table Generalized Distribution Table (GDT) is a table in which the probabilistic relationships between concepts and instances over discrete domains are represented [14], [15]. Any GDT consists of three components: possible instances, possible generalizations of instances, and probabilistic relationships between possible instances and possible generalizations. The possible instances, which are represented at the top row of GDT, are defined by all possible combinations of attribute values from a database, and the number of the possible instances is m
∏n i =1
i
(2)
Where m is the number of attributes, n is the number of different data values in each attribute.
A Probabilistic Rough Set Approach to Rule Discovery
59
The possible generalizations for instances, which are represented by the left column of a GDT, are all possible cases of generalization for all possible instances, and the number of the possible generalizations is
⎛ m ⎞ ⎛ m ⎞ n 1 + − ( ) ∏ i ⎜ ⎟ ⎜ ∏ ni ⎟ − 1 ⎝ i =1 ⎠ ⎝ i =1 ⎠
(3)
A wild card ` * ' denotes the generalization for instances, For simplicity, the wild card will sometimes be omitted in the paper. For example, the generalization a0 * c0 means that the attribute b is superfluous (irrelevant) for the concept description. In other words, if an attribute b takes values from {b0 , b1 } and both a0b0c0 and a0b1c0 describe the same concept, the attribute b is superfluous, i.e. the concept can be described by a0c0 . Therefore, the generalization a0*c0 used to describe the set { a0 b0 c0 , a0 b1 c0 } The probabilistic relationships between possible instances and possible generalizations, represented by entries Gij of a given GDT, are defined by means of a probabilistic distribution describing the strength of the relationship between every possible instance and every possible generalization. The prior distribution is assumed to be uniform if background knowledge is not available. Thus, it is defined by
Gij = p(PI j \ PGi )
⎧ 1 if PGi is a generalization of PI j ⎪ = ⎨N PG i ⎪ otherwise ⎩ 0
⎫ ⎪ ⎬ ⎪ ⎭
(4)
where
PI j
is the j th possible instance,
PG i
is the ith possible generalization, N PG and i is the number of the possible instances satisfying the ith possible generalization . Rule Strength
In this approach, the rules are expressed in the following form: X → Y with That is, “ if X then Y with strength S ”. Where
S
X : denotes the conjunction of the conditions that a concept must satisfy, Y :denotes a concept that the rule describes, and S : is a “measure of strength” of which the rule holds. From the GDT, we can see that a generalization is 100% true if and only if all of instances belonging to this generalization appear. Let us use the example shown in
60
Hossam A. Nabwey
Table 1. Considering the generalization {b0 , c1}, if instances both {a0 b0 c1} and { a1 b0 c1} appear, the strength s({b0 , c1}) is 1; if only one of { a0 b0 c1} and { a1 b0 c1} appears, the strength s( {b0 c1}) is 0.5, as shown in Figure 1. 0.5
a1b0c1
a1b0c1
0.5
a1b0c0
0.5
a1
b0c1 a0b0c1
0.5
Fig. 1. Probability of a generalization rule
It is obvious that one instance can be expressed by several possible generalizations, and several instances can be also expressed by one possible generalization. For the example shown in table 1, the instance { a1 b0 c1} can be expressed by { a1 b0}, { b0 c1}……., or { c1}. Every generalization in upper levels contains all generalizations related to it in lower levels. That is, {a1} ⊃ {a1 b0} , { a1 c1} , {a1 b0} ⊃ { a1 b0 c1} In other words, if the rule {a1} → y is true, the rule {a1 b0} → y and { a1 c1} → y are also true. Otherwise, if {a1 b0} → y or { a1 c1} → y is false, the rule {a1} → y is also false. Figure 2 gives the relationship among generalizations. a1
b0
c1
a1b0
a1c1
b0c1
a1b0c1 Fig. 2. The relationship among generalizations
A generalization that contains the instances with different classes is contradictory , and it cannot be used as a rule. In contrast, a generalization that contains the instances with the same class is consistent, so From Table 1, we can see that the generalizations can be divided into three groups: contradictory, belonging to class y, and belonging to class n.
4 Searching Algorithm for an Optimal Set of Rules We now outline the idea of a searching algorithm for a set of rules based on the GDTRS methodology. a sample decision table shown in Table 1 is used to illustrate the idea.
A Probabilistic Rough Set Approach to Rule Discovery
61
Algorithm Step 1. Create the GDT Since : ∈ { a 0 , a1 }
over-current protection action
Exceeding of winding insulation resistance
⇒ n1 = 2
∈ { b 0 , b1 }
Unbalance of three-phase winding direct current resistance
⇒
n2 = 2
∈ { c0 , c1 }
⇒ n3 = 2
Hence : the number of attributes ( m ) = 3 , from Eq.(2) number of the possible instances is 8 , from Eq.(3) number of the possible generalizations is 18 , By deleting all of the instances and generalizations un-appeared in the example database shown in Table 1. Step 2 . simplify the GDT
From table 1 The instances appeared with respect to cases 1, 3, 5, 8 are
From Eq. ( 1 ) and table 1 the instance appeared with respect to case 2 is
{a1b0c1} ;
From Eq. ( 1 ) and table 1 the instance appeared with respect to case 4 may be one of
{{a b c } , {a b c }} ; 1 0 0
1 0 1
From table 1 , the instance appeared with respect to case 6 may be one of
{{a b c } , {a b c } , {a b c } ,{a b c }} 0 0 0
0 0 1
Similarly ,
0 1 1
the instance appeared with respect to case 7
{{a1b0c1} , {a1b1c1}} instance is
0 1 0
{a1b1c1} .
but
{a1b0c1}
may be one of
is not consistent with table. 2 . so the appeared
Step 3 . group the generalizations Generalizations can be divided into three groups contradictory, belonging to class yes, and belonging to class no . The contradictory generalizations, containing the instances belonging to different decision classes, cannot be used as the rules. Hence they are ignored. In other words, we are just interested in the generalizations belonging to class yes or no, which will be selected as the rules. Step 4. Rule Selection There are several possible ways for rule selection. For example:
• Selecting the rules that contain as many instances as possible .
62
Hossam A. Nabwey
• Selecting the rules in the levels of generalization as high as possible according to the number of “ * “ in a generalization . Selecting the rules with larger strengths Table 2. The GDT for the decision table shown in table 1
(Note the elements that are not displayed are all zero) a0 b0 c0 * b0 c0
a 0 b 0 c1
a0 b1 c0
a0 b1 c1
1/2
* b0 c1
1/2 1/2
1/2
1/2
1/2 1/2
1/2
a1 * c 0
1/2
a1 * c 1
1/2 1/2
1/2
1/2
1/2
a1 b 0 *
1/2
1/2
a1 b 1 *
a0 * *
1/2 1/4
1/4 1/4
1/4
1/4
1/4 1/4
1/4
1/4
1/4
* b1 *
1/4
1/4 1/4
1/4
1/4
1/4
1/4
1/4
Table 3. The generalizations belonging to class yes
a1 * c 0 a1 b 1 *
a1 b0 c0 1/2
1/2
1/4
1/4
a1 * * * b0 *
1/2
1/2
a0 b 1 *
* * c1
a1 b1 c1
1/2 1/2
a0 * c 1
* * c0
a1 b1 c0
1/2
* b1 c1
a0 b 0 *
a1 b0 c1
1/2
* b1 c0
a0 * c 0
a1 b0 c0
a 0 b1 c 0 1/2 1/2
a 1 b1 c 1
1/2
1/4
1/4
1/4
1/4
A Probabilistic Rough Set Approach to Rule Discovery
63
Table 4. The generalizations belonging to class no
a 0 * c0 a 0 * c1 a0 b0 * a0 b1 * a**
a0 b0 c0 1/2
a0 b0 c 1
1/2
1/2 1/2
1/4
1/4
a0 b1 c0 1/2
a0 b1 c 1
1/2 1/2 1/4
1/2 1/4
Since the purpose is to simplify the decision table and simpler results of generalization (i.e., more general rules) are preferred, the first priority will be to the rules that contains more instances, then to the rules corresponding to an upper level of generalization. and the third priority to The rules with larger strengths . Thus, from table 3 and table 4 the final rule set is
{a 1c o } → {a 1b 1 } → {a 0 } →
y e s , w it h S = 1 y e s , w ith S = 1 no
, w ith S = 1
Results The induced Rules can be written as:
• If (Valid utilization degree, a1 ) and (Maintenance cost, , not appearing) then (Grade, II) • If (Valid utilization degree, a1 ) and (Maintenance cost, b1 ) then (Grade, II) • If (Valid utilization degree, not appearing) then (Grade, I)
5 Conclusions ♦ Rough set theory and statistics are related to analyze the data from the rough set perspective. ♦ Three approaches to missing attribute values are presented in a unified way. It is shown that all three approaches to missing attribute values may be described using the same idea of attribute-value blocks. ♦ An approach of rule discovery based on Rough Sets and Generalization Distribution Table was presented. The basic concepts and an implementation of the methodology was described. Main features of that methodology can be summarized as follows: ♦ It can discover If-Then rules from very large, complex databases . ♦ It represent explicitly the uncertainty of a rule including the prediction of possible instances in the strength of the rule.
64
Hossam A. Nabwey
♦ Lost values are considered during the process of rule induction . ♦ It can flexibly select biases for search control. ♦ It can effectively handle noisy data, missing data.
Acknowledgment I thank the President of Al-Kharj University, Deanship of Scientific Research in AlKharj University, Dean of College of Science and Humanitarians studies and all the Staff of Department of mathematics for their continuous support, encouraging and for their useful and fruitful discussion.
References [1] Wang, G.: Extension of Rough Set under Incomplete Information Systems, National Science Foundation of china (No. 69803014 ), PD program of P.R. China [2] Grzymala-Busse J. W.: Three Approaches to Missing Attribute Values - A Rough Set Perspective. Accepted for the Workshop on Foundations of Data Mining, Associated with the Fourth IEEE International Conference on Data Mining, Brighton, UK, November 1-4 (2004) [3] Grzymala-Busse, J.W., Wang, A.Y.: Modified algorithms LEM1 and LEM2 for rule induction from data with missing attribute values. In: Proc. of the Fifth International Workshop on Rough Sets and Soft Computing (RSSC 1997) at the Third Joint Conference on Information Sciences (JCIS 1997), Research Triangle Park, NC, March 25, pp. 69–72 (1997) [4] Grzymała-Busse, J.W., Hu, M.: A comparison of several approaches to missing attribute values in data mining. In: Ziarko, W.P., Yao, Y. (eds.) RSCTC 2000. LNCS (LNAI), vol. 2005, pp. 378–385. Springer, Heidelberg (2001) [5] Grzymała-Busse, J.W.: Data with missing attribute values: Generalization of indiscernibility relation and rule induction. In: Peters, J.F., Skowron, A., GrzymałaBusse, J.W., Kostek, B.z., Świniarski, R.W., Szczuka, M.S. (eds.) Transactions on Rough Sets I. LNCS, vol. 3100, pp. 78–95. Springer, Heidelberg (2004) [6] Grzymala-Busse, J.W.: On the unknown attribute values in learning from examples. In: Raś, Z.W., Zemankova, M. (eds.) ISMIS 1991. LNCS, vol. 542, pp. 368–377. Springer, Heidelberg (1991) [7] Grzymala-Busse J. W.: In: Rough Set Strategies to Data with Missing Attribute Values. In: Proceedings of the Workshop on Foundations and New Directions in Data Mining, Associated with the Third IEEE International Conference on Data Mining, Melbourne, FL, USA, November 19-22, pp. 56–63 (2003) [8] Bjanger, M.S.: Vibration Analysis in Rotating Machinery using Rough Set theory and ROSETTA. Tech. report, Univ. of Norwegian (1999) [9] Zhong, N., Obsuga, S.: Using Generalization Distribution Tables as a Hypotheses Search Space for Generalization. In: Proc. 4th International Workshop on Rough Sets, Fuzzy Sets, and Machine Discovery (RSFD 1996), pp. 396–403 (1996) [10] Zhong, N., Dong, J.Z., Ohsuga, S.: Discovering Rules in the Environment with Noise and Incompleteness. In: Proc. 10th International Florida AI Research Symposium (FLAIRS 1997) edited in the Special Track on Uncertainty in AI, pp. 186–191 (1997)
A Probabilistic Rough Set Approach to Rule Discovery
65
[11] Zhong N., Dong J. Z., Ohsuga S.: Soft Techniques to Rule Discovery in Data. In: Proc. the Fifth Congress on Intelligent Techniques and Soft Computing (EUFUT 1997) edited in the Invited Session on Soft Techniques in Knowledge Discovery, pp. 212–217 (1997) [12] Zhong, N., Fujitsu, S., Ohsuga, S.: Generalization Based on the Connectionist Networks Representation of a Generalization Distribution Table. In: Proc. First Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD 1997), pp. 183–197. World Scientific, Singapore (1997) [13] Zhong, N., Skowron, A.: A rough set-based knowledge discovery process. Int. J. Appl. Math. Comput. Sci. 11(3), 603–619 (2001) [14] Zhu, W., Zhang, W., Fu, Y.: An Incomplete Data Analysis Approach using Rough Set Theory. In: Proceedings of the 2004 International Conference on Intelligent Mechatronics and Automation, Chengdu, China (August 2004) [15] Pawlak, Z., Busse, J.G., Slowinski, R., Ziarko, W.: Rough Sets. Communication of the ACM 38(11), 89–95 (1995) [16] Quinlan, J.R.: C4.5: Programs for Machine Learning. Morgan Kaufmann Publishers, San Francisco (1993)
A Robust Video Streaming Based on Primary-Shadow Fault-Tolerance Mechanism Bok-Hee Ryu, Dongwoon Jeon, and Doo-Hyun Kim* Dept. of Internet & Multimedia Engineering, Konkuk University Seoul, Republic of Korea {fbqhrgml,dongwoon,doohyun}@konkuk.ac.kr
Abstract. One-to-any multimedia streaming has been a challenging issue for many researchers due to the lack of stability and reliability. Furthermore, it is highly demanded that high-level software reliability and accurate timeliness for critical tasks such as medical or disaster monitoring system. In order to assure the robustness of video streaming, fault tolerance mechanism, this paper proposes a mechanism and experimental system for detecting fault and recovery. This system includes a fault-free master-node to provide one-to-any video streaming applications with enhanced sustainability. The experiments showed that the fault detection and recovery time of the proposed system were fairly small. The fault detections were done within 137ms in average, and the fault recoveries were completed within 33ms in average. Keywords: fault-tolerance, video streaming, primary-shadow replica.
1 Introduction With increase of communication bandwidth and computing capability of end hosts, one-to-any video applications used in such as remote education became attractive. In the structure of one-to-any video streaming service, heavy data transmission is carried out by a video streaming server where crash or malfunction may be occurred at streaming phase. On the other hand, if there are faults caused by software or hardware at video server, it will affect the reliability of whole system. Because these reasons, one-to-any video streaming service requires fault-tolerance mechanism for stable streaming service. Generally, multimedia streaming applications usually use UDP protocol in order to overcome additional delay by retransmission as in TCP for reliable data transmission. Furthermore, naive usage of UDP protocol cannot satisfy the quality of service nor reliable transmission. But, error-sensitive applications such as medical or disaster monitoring requires high reliability and stability. For example, disaster prevention service using UAV (Unmanned Aerial Vehicle) with real-time video streaming facilities needs more reliable delivery from the air to the head quarter. The remote surgery definitely requires fault-free real-time video delivery mechanism, too. In order to achieve reliable streaming, this paper applied existing schemes; TMO and PSTR. *
A Robust Video Streaming Based on Primary-Shadow Fault-Tolerance Mechanism
67
TMO (Time-triggered Message-triggered Object) [1] is a object-based scheme for distributed computing. The PSTR (Primary-Shadow TMO Replication) [2][3][4] is a mechanism for networked fault-tolerance which consists of two TMO replicas; one takes a role as the primary and the other is the shadow. When a fault is occurred in the primary TMO, the shadow TMO will detect the fault and undertake role of primary. There will be new primary and delivering the output it has prepared. Once the old primary detects its own fault, it will enter the local recovery mode with the goal of becoming a healthy shadow without disturbing the new primary which provides application services. This paper proposes resilient video streaming with handling of fault by measuring acceptable amount of packet loss and timeout value. The aim of this paper is to detect failure of primary-node then let shadow-node take over functions of primary-node with continuous streaming service. Furthermore, recognition of primary-node’s failure by any participated nodes is one of purpose. This paper suggests TCM (TMO configuration Manager) node that maintains information of all participated nodes, and performs these functions; detection, report, and recover of failure.
2 Related Works There are many researchers who focus on resilient streaming issue [6][7][8][9]. One of them proposed redundancy both network paths and data [6]. They introduced multiple distributions coding to provide redundancy. In rStream [7] proposed rateless coding technique for efficient transmission at link level. In DagStream [8] proposed streaming framework for resilient streaming using locality in distributed environment. In [9] focused on resilient streaming system at link and repository failure. Major of researches implemented on entire fault-free system (hardware). However, faults can be generated from hardware itself. This study approaches different point of view against previous studies at resilient streaming system. The TMO (Time-triggered Message-triggered Object)[1] is an object-oriented programming scheme devised with purposes to reduce design efforts and improve productivity in developing real-time systems especially such as networked and time coordinated systems. The TMO contains two types of methods; 1) time triggered methods (or spontaneous methods, SpMs) and 2) service methods (SvMs). The SpM executions are triggered whenever the RT clock reaches the time values determined at the design time. On the other hand, SvM executions are triggered by calls from client objects in the form of service request messages. The PSTR scheme[2][3][4] is a result of incorporating the basic underlying principles of the DRB/PSP(Distributed Recovery Block/ Pair of Self-checking Processing) scheme[5], called the primary-shadow active replication principle, into the TMO structuring scheme. Two partner TMOs, i.e., Primary TMO and Shadow TMO, are hosted at two different nodes connected mostly through Local Area Network (LAN). In the PSTR mechanism, each partner object has the same duplicated external inputs and own independent data members. The methods of both partner objects independently perform functionally equivalent computations and updates data members in parallel. The Shadow TMO takes the role of primary when it decides the primary has a fault. The behavior of the shadow TMO is described in the next section.
68
B.-H. Ryu, D. Jeon, and D.-H. Kim
3 The Fault-Tolerance Mechanism for Video Streaming 3.1 Design of Fault Tolerance System TMO (Time-Triggered & Message-Triggered Object) TMO (Time-triggered Message-triggered Object) is one of real-time architecture. It supports various APIs and components to establish middleware and software which need timing requirements. TMO provides concurrency of programmed elements and guaranteed execution time bounds and maximum invocation rates. Streaming data and fault-message will be delivered RMMC (Real-time Multicast and Memory-replication Channel) that supported by TMO scheme. PSTR (Primary-Shadow TMO Replication) The PSTR (Primary-Shadow TMO Replication) station consists of two TMO replicas, one is the primary and the other is the shadow. When the system is fault-free, the primary and the shadow TMOs process the identical incoming requests, but only the primary TMO is allowed to perform output actions. When a fault occurs in the primary TMO, the shadow TMO is able to detect the fault and cover up the fault by switching its role to that of the new primary and delivering the output it has prepared. Once the old primary detects its own fault, it will enter the local recovery mode with the goal of becoming a healthy shadow without disturbing the new primary which provides application services. TNCM (TMO Network Configure Manager) The TNCM [6] subsystem takes an important role of management for system reconfigurations. TNCM is a component of TMO. It has functions as following: (1) Accepting TMO registrations and performing maintenance of TMO participants’ information. (2) Supporting TMO migration service. (3) Handling fault detection reports from the subsystems and components including OS level. (4) Redistribution and reconfiguration for included TMO nodes after fault occurred. In addition, TNCM maintains information on relationships among TMO replicas. There are many assumptions with significant restraints which restrict session establishment rigorously between TMO-based applications. Thus, during the failure takes place in the primary node of core-role, the service is not sustained as far as all other participating worker node(client node) are executed again even if the shadow node switch the role immediately. With the restraint of TMO mentioned above, in the one-to-any application, it is required to supplement the node (TCM node in this paper) which is not affected by the failure and switch the role of formal master node to the worker node. At last, all participant node including primary node and shadow node act the role as worker node.
A Robust Video Streaming Based on Primary-Shadow Fault-Tolerance Mechanism
69
Fig. 1. Proposed PSTR architecture with TCM node
Figure 1 illustrates new PSTR structure with TCM node. There are 3 nodes at this system primary, shadow and TCM node. Primary node performs streaming server unless there is malfunction on system. Shadow node is spare for primary’s breakdown. TCM node is supervisor of whole system. It implemented as the master node among TMO objects to apply one-to-any application on TMO. System is structured for detection and recovery to prevent system failure. In this paper, the fault will be detected either sequential ID or time out value which reported through RMMC channel. TCM node checks whether FTO, FTB, FTBfreq variables conform the criterion or not. Entire participants including primary and shadow nodes share fault status variable, it enables this system suitable for fault-tolerance system. 3.2 Detection of Faults in Streaming There are two cases of faults in video streaming. One is the interruption of service with an error which is caused by the device itself, the other is determined of deficient quality of service which is caused of network status or operating system. Thus, in these two occasions, TCM node for non-stoppable video streaming should be able to detect and recover these faults. In this paper, three indicators are suggested to determine of two cases of fault. Table 1 contains explanations of them.FTO (Fault-tolerance Timeout) is the indicator which determines fault by observing mute period. It is timeout value for the fault determination to detect fault from device itself. The initial FTO value is set sufficiently for TMO’s constraints. TMO demands registration period for all participated nodes. FTO value will be updated in every specific period to maintain quality of service properly. In another point of view, the quality of video stream application depends on assurance of streaming continuousness. The fault can be determined by observing continuousness of streaming or decline of the quality of service. There are two indicators for fault determination caused by poor network condition as depicted figure 2; FTB, FTBfreq. FTB (Fault Tolerance Bound) is the indicator which can be defined the criterion count of the packet loss. FTBfreq is defined as criterion frequency of message loss for fault determination. At this point, the fault is determined when
70
B.-H. Ryu, D. Jeon, and D.-H. Kim Table 1. Indicators for fault determination
Indicator
Explain
FTO
Fault tolerance Timeout
FTB
Fault Tolerance Bound(Number of continuous packet loss)
FTBfreq
FTB Frequency
Fig. 2. Measuring FTB and FTO
message loss is occurred by observation of the amount of FTB with FTBfreq frequency. Fault which is determined by FTB means this system does not allow low quality of service by considering former and current message-loss while primary node works properly. 3.3 Fault Determination and Recovery In this paper, two kinds of ATs (Acceptance Test) are performed in the fault-tolerant system for streaming. FTB <= IDr – IDp
(1)
In the first acceptance test, fault will be reported by examining sequential message IDs which are sent from primary node. This fault comes from either low service quality or OS level problem. In this test, as it is shown in formula (1), the amount of message-loss will be compared FTB value continuously by subtracting previously received message ID (IDp) from recent message ID (IDr). If FTB exceeds FTBfreq times, the first acceptance test will be ended up with fault generation. FTO <= Tspm – Trec
(2)
A Robust Video Streaming Based on Primary-Shadow Fault-Tolerance Mechanism
71
In the second acceptance test, it will watch device’s malfunction. At this indicator, fault will be determined by measuring mute period which is defined as FTO value above and it is shown at formula (2). TCM node performs acceptance test to determine fault. If these acceptance tests are passed in each phase, two ATs will be performed consciously from at the first step with updating recent status.
Fig. 3. Illustration of mean time to repair
Figure 3 is explanation of MTTR (Mean Time to Repair) period. MTTR stands for period from fault detection to recovery it. Let suppose fault occurs at ‘a’ point however, the system cannot recognize it at that moment. ‘A’ is damages section which the system cannot aware fault. At ‘B’ section this system can detect fault by examining change of FTO and FTB value. Section ‘C’ is for recovery after detecting, and at this phase shadow node will take over primary’s tasks. After spot ‘d’, the system will work own jobs normally. We measured MTTR value and it referred at next chapter. Minimizing MTTR period is critical issue for FT system.
4. Experiments and Evaluation In this section, we will discuss evaluation of proposed PSTR-based video streaming with TCM node. Experimental analyses indicate this fault-tolerant system contributes resilient streaming service. We applied this system to video conference application. There can be many kind of faults occurred at video conference among multiple participants. Particularly, we assumed one of node’s stream come from UAV. Evaluation system is composed of 5 nodes as illustrated at figure 4. Other participants will share stream from UAV and participants can exchange opinions through this system also. We verified performance of fault-tolerance scheme which improves reliability of video streaming. In this evaluation, local network is set to minimize packet-loss between nodes and facilitate at the simulation. However, in order to detect the fault of FTB, irregular fault is inserted for pretending temporary interruption of data transmission. Also, for
72
B.-H. Ryu, D. Jeon, and D.-H. Kim
detection of FTO, intentional and immediate intervention is generated at discretionary time. There is gap between actual fault occurred time at the primary node and awareness time at TCM node or client node. In this section, we will show measured actual fault detection time and fault recovery time.
Fig. 4. Evaluation environment
4.1 Time Difference in Fault Occurrence and Awareness The fault determination process which is suggested in this paper can have time difference between fault occurrence at primary node and awareness at TCM or client node. The time differences in the fault occurrence and awareness were experimentally measured with 50 experiments. It is also shown in the experiments that these differences are no longer increased. The average difference of fault occurrence and awareness was measured as 12,520μs in FT-TCM mode. Minimum was 2500μs, maximum was 36,500μs. The distribution of the degree for the differences is depicted at figure 5. As a result, most of faults can be detected within 2,500 μs.
Fig. 5. Time difference in fault occurrence and awareness
A Robust Video Streaming Based on Primary-Shadow Fault-Tolerance Mechanism
73
Fig. 6. Fault detection time
4.2 Fault Detection Time We took a measurement of fault detection time within video streaming a hundred times. Fault detection time is gotten when the accumulated missing message number is higher than FTB value. As a result, the fault detection time was 136,771μs on the average. Minimum was 13,000μs and maximum was 199,000μs. Figure 6 shows distribution of fault detection time.
Fig. 7. Recovery time corresponding FTB
74
B.-H. Ryu, D. Jeon, and D.-H. Kim
4.3 Fault Recovery Time The last experiment is evaluation of the fault recovery time with the test of a hundred times. It is evaluated the fault recovery time through the difference of the time at which data was received from primary node and the time at which the data was first received from the shadow node. The fault recovery time was 33,241μs in FT-TCM mode on the average. Minimum was 21,000μs and maximum was 59,000μs. Figure 7 shows distribution of fault recovery time according to FTB value.
5 Conclusions and Future Works While video streaming is applied in many field of industry, especially in the area of medical and safety, it has been significant issue for the video streaming to be required not only real-time processing, but also the reliability and sustainability has become significant issue. This paper suggested new architecture of PSTR-based fault-tolerance utilizing TMO to improve the sustainability of the video streaming which is for the specific purposes mentioned above. In the traditional PSTR that does not contain the additional node for the specific task, the shadow node recovers the detected fault with duplicate deployed primary node that performs core function. However, in the one-to-any application, the restraint of TMO requires the master node that does not allow the fault. This paper supplements TCM node to let the primary node detect the fault and make the shadow node recover it. It suggests FTB of packet-loss limit and FTBfreq of packet-loss frequency to detect the fault that is caused by low quality of service from packet-loss which is the result from deficient network status or operating system trouble. And the FTO timeout value is also proposed to detect physical fault of the device itself. The general method of video streaming is maintained in the fault-tolerant system that this paper suggested. Especially, in FT-TCM mode, it does not require substantial revision of the legacy system. However, in the application which is highly sensitive in real-time, the difficulty of this scheme is that MSDT (Maximum Switchover Delay Time) is not disqualified very often. Therefore, in the future research, it requires improvement of the fault detect time and the fault recovery time with the algorithm detecting the fault and the enhancement through the integration with TMO middleware. And, with the occasion that the fault occurs in the early service, the scheme to solve the problem of increasing fault detect time is also needed.
Acknowledgment “This research was supported by the MKE(Ministry of Knowledge Economy), Korea, under the ITRC(Information Technology Research Center) support program supervised by the NIPA(National IT Industry Promotion Agency)" (NIPA-2010C1090-1031-0003).
A Robust Video Streaming Based on Primary-Shadow Fault-Tolerance Mechanism
75
References 1. Kim(Kane), K.H.: APIs for Real-time Distributed Object Programming. IEEE Computer 33(6), 72–80 (2000) 2. Kim(Kane), K.H., Moon, K.-D., Park, J.H., Zheng, L., Zhou, Q.: Fault-Tolerance in Universal Middleware Bridge. In: 11th IEEE Symposium on Object Oriented real-Time Distributed Computing (ISORC) (2008) 3. Kim(Kane), K.H., Liu, J.J.Q.: Techniques for Implementing Support Middleware for the PSTR Scheme for Real-Time Object Replication. In: Proceedings of the 7th IEEE International Symposium on Object-Oriented Real-Time Distributed Computing (ISORC) (2004) 4. Zhou, Q.: Robust Integration of Multi-Level Fault Detection Mechanisms and Recovery Mechanisms in a Component-based Support Middleware Model for Fault Tolerant RealTime Distributed Computing. Doctor of Philosophy in Electrical and Computing Engineering of University of California (2009) 5. Pullum, L.: Software Fault Tolerance Techniques and implementation. Artech House computing library, Boston (2001) 6. Padmanabhan, V.N., Wang, H.J., Chou, P.A.: Resilient Peer-to-Peer Streaming. In: 11th IEEE International Conference on Network Protocols (2003) 7. Wu, C., Li, B.: rStream: Resilient and Optimal Peer-to-Peer Streaming with Rateless Codes. IEEE Transactions on Parallel and Distributed Systems 19 (2008) 8. Liang, J., Nahrstedt, K.: DagStream: Locality Aware and Failure Resilient Peer-to-Peer Streaming. Multimedia Computing and Networking (2006) 9. Shah, S., Dharmarajan, S., Ramamritham, K.: An Efficient and Resilient Approach to Filtering and Disseminating Streaming Data. In: Proceedings of the 29th International Conference on Very Large Data Bases, VLDB 2003 (2003)
Oversampled Perfect Reconstruction FIR Filter Bank Implementation by Removal of Noise and Reducing Redundancy Sangeeta Chougule and Rekha P. Patil
Abstract. This paper proposes design algorithm for real valued, even order oversampled five channel FIR filter banks with selection of frequency bands and filter order, which reduces inband aliasing and amplitude distortion and removal of noise. This filter bank is designed using direct form-II transposed structure. Analysis of two dimensional (2D) image is carried out using five channel oversampled filter bank. This system is suitable for any images format. DCT Image compression technique is used to compare output of filter banks with respect to filter orders by using mean square error and peak signal to noise ratio. This algorithum of oversampled filter bank shows output near perfect replica of input image. Here improvement in noise and reducing residual redundancy by selecting oversampled factor 2. Keywords: oversampled, FIR, filter bank, subsampled, Multirate.
Oversampled Perfect Reconstruction FIR Filter Bank Implementation
77
bank. Application in data communications or high quality audio can be very sensitive to such errors. Another possibility to achieve the decomposition of the input signals is to use oversampled filter banks (OSFB), i.e. which consists of real valued filter banks and use of band pass filters. The term oversampled refers to filter banks where the sampling factor is less than the number of channels; that is, 1≤ K< M; the term critically sampled refers to the case K = M. The interest in oversampling filter banks is due to some improvements over critically decimated filter banks, such an addition design flexibility, improved frequency selectivity. Experimentation is carried out on filter banks having non integer oversampling ratio M/K, and non uniform bandwidth filters. Review of previous work: Taking a step towards design of oversampled filter bank, initally uniform and dyadic filter banks are designed using real valued casual FIR filters [5] . Uniform filter bank was developed for four channel and dyadic filter bank developed for two level symmetric filter bank. Here we found that we are not able to develop perfect reconstruction output because of spectral gap between transistion bands. Our main goal is to develop oversample filter bank which gives the output near input image[3,4] . Harteneck, Stephan Weiss, and Robert W. Stewart[3,4] have developed one oversampled filter bank for one dimensional signal. There is open problem to design oversample filter bank without inband aliasing and without amplitude distortion for two dimensional images. Number of researcher have developed filter bank using lattice structure filters. But we have developed filter banks using direct form –II transposed structure real valued filters for even order. We have tried above filter banks with different filter orders for two dimensional images. Number of researchers designs synthesis filter bank like wise which gives output near input image. In this proposed work all analysis and synthesis filters design parameters are same and selection of subsampled factor, filter orders of filters and frequency selectivity are most important factor which gives output near input image. Outline of the Paper: Section-2 shows oversampled five channel filter bank, which is designed for subsampling factor 2 and 4. Section 3 gives Experimental result and discussion. Section 4 highlights Experimentation for Five Channel Filter Bank for Reducing Redundancy and Improvement in Noise and 5 shows conclusion and section 6 highlights references.
2 Oversampled Five Channel Filter Bank A basic multirate filter bank is shown in Fig. 2a. Multirate filter banks are so named because they effectively alter the sampling rate of a digital system, as indicated by the decimators (downsamplers) following the analysis filters, A0 and A1, and the expanders (upsamplers) preceding the synthesis filters, S0 and S1. Properly designed analysis and synthesis filters combined with the properties of decimation and expansion allow filter banks to partition a wideband input signal into multiple frequency bands (often called subbands or channels) and to recombine these subband signals back into the original signal. In the case of Fig. 2, the analysis filters, A0 and A1, are typically complementary lowpass and highpass filters that mirror each other about the digital frequency, π/2, as shown in Fig. 2b. Such filters are often called quadrature mirror filters (QMF), since π/2 corresponds to one fourth the sampling frequency.
78
S. Chougule and R.P. Patil
X[n]
Y[n]
H
F
H
F
H
F
H
F ↓
H 4[ ]
2
F ↑ 2
4[ ]
Fig..1. Five channel oversampled FIR filter bank
Fig. 2. [a] Block diagram of a simple two-channel multirate system, and [b] approximate magnitude responses of analysis filters, A0 and A1
The concept of multirate filtering relies on the two processes that effectively alter the sampling rate, decimation and expansion. Decimation or downsampling by a factor of M essentially means retaining every Mth sample of a given sequence. Decimation by a factor of M can be mathematically defined as (1) 1. Expansion or upsampling by a factor of M essentially means inserting M-1 zeros between each sample of a given sequence. Expansion by a factor of M can be mathematically defined as, (2) or equivalently, (3)
Oversampled Perfect Reconstruction FIR Filter Bank Implementation
79
The QMF bank is prone to three types of distortion: aliasing distortion, amplitude distortion, and phase distortion. Practical filters have a non-zero transition band. In order for the QMF bank to remain lossless, then, the analysis filters must overlap as shown in Fig. 2b. Hence, the analysis filters are not strictly band-limiting and the subsequent decimation causes aliasing. Fortunately, aliasing can be completely cancelled by properly designing the synthesis filters, S0 and S1 [6]. (4) The aliasing component in y(n) becomes zero when (5) or, more specifically, when (6) Thus, aliasing in one branch is completely cancelled by the synthesis filter in the opposite branch. The entire alias-free filter bank can now be expressed as the single transfer function, (7) Then, letting z = ejω (8) If |H(ejω)| isn’t constant and non-zero for all ω (allpass), the filter bank suffers from amplitude distortion. Likewise, if H(ejω) does not have linear phase (if φ (ω) is not of the form a⋅ ω + b) the filter suffers from phase distortion. It is important, at this point, to emphasize that these two conditions apply to the filter bank as a whole (Equation 8), not the individual analysis and synthesis filters. When the filter bank is free from aliasing, amplitude distortion, and phase distortion, it is called a perfect reconstruction (PR) filter bank. Polyphase representation is, in simple terms, a method of re-organizing the coefficients of a given transfer function, h(n). The transfer function can be represented in terms of its even and odd coefficients.
(9) Therefore, (10)
80
S. Chougule and R.P. Patil
Where (11) This idea can be extended to further decomposition by a factor of M [6]. Then H(z) takes on the "Type 1 polyphase" form (12) Where (13) With (14) The "Type 2 polyphase" form is a permutated version of the Type 1 form and is given by
(15) Where (16) By combining the polyphase representation and the noble identities, it is possible to implement more efficient analysis filter/decimator and expander/synthesis filter blocks Harteneck, Stephan Weiss, and Robert W. Stewart[3,4] says that to design oversampled filter bank, add one more channel in fig.2 as bandpass filter to pass frequency components near to pi*1/2. For practical implementation selection of third channel frequency band and pass band ranges of first two channels are most important. Normalized frequency response of three channel filter bank with different filter orders are shown in fig.3. Implementation of this filters of filter banks using type-I direct form II transposed structure. These filters are transformed into two dimensional forms for analysis of 2D image. In fig.1 three channel FIR filter bank is shown, which shows that decimation and interpolation factor is less than no. of channels as per the oversampled condition. In the oversampled case more design freedom and improved numerical properties as compared with critically sampled FB’s, and they have noise reducing properties [1]. The design freedom increased since for a given oversampled analysis FB, there exist a whole class of synthesis FB’s providing near perfect reconstruction [NPR].
Oversampled Perfect Reconstruction FIR Filter Bank Implementation
81
After design of filters of filter bank in one dimension it is transformed into two dimension using frequency transformation technique as below. H(w1,w2)=B(W)│cosw=T(w1,w2) B(W) is the Fourier transform of the one dimensional filter. N -jwn B(W) = ∑ b(n) e n = -N -jw1n1 -jw2n2 T(w1,w2)= ∑ ∑ t(n1,n2) e e n1 n2
(17) (18)
The returned filter h is the inverse Fourier transform of H(w1,w2). Filters of analysis and synthesis filter banks design by using above design method. This proposed filter bank is simplest possible filter bank that uses equal subsampling factor. This filter bank preserves the property of alias free output of two dimensional Lena image. In this design all five filters. Filers cover all frequency components of input signal except frequency. All filters used in this FB are band pass filters of passband normalized frequency ranges are Ch1=0 to 0.3, Ch2=0.3 to 0.35, Ch3=0.35 to 0.65, Ch4=0.65 to 0.7, Ch5=0.7 to 0.9999. All filters are subsampled by M . All frequencies must be covered by at least one filter. This filer bank is designed using 1D FIR filter design technique which is then transformed into two dimensional FIR filter using frequency transformation technique. Requirement for filter bank is that it yields the perfect or near-perfect reconstruction property, i.e.y(k) = x(k-∆), where ∆ is a fixed and delay chosen a priori, and therefore common zeros in all analysis filters Hi(z) are ruled out as information is lost at these frequencies.
3 Experimental Results and Discussion Initially design of proposed oversampled FIR filter bank using type-I direct form II transposed structure FIR filters are developed in one dimensional (1D) form. Initially input signal is divided into 2 channels but at transition gap some frequency components are lost means not passed for further processing. Therefore to overcome this problem extra bandpass filter is selected at the spectral gaps and all the frequency components are passed at output stage. Design of downsampler and upsampler of analysis and synthesis filter bank is carried out by using bilinear interpolation technique. In downsampler block of each channel size of image with rows and columns are reduced by downsample factor. Similarly at the upsampler block, image size is restored by selecting same upsampler factor. Initially we have selected subsample factor 2 which is at oversample rate. Frequency and phase spectrum of analysis and synthesis filters are shown in Fig.3. With different filter orders investigation of oversampled filter bank near perfect reconstruction of input is carried out. One can apply this filter bank for different images. Here for filter bank investigation of oversample factor with proper filter orders are carried out, which shows resultant near perfect input. Resultant of Lena (gif image format).
S. Chougule and R.P. Patil Magnitude Response (dB) 0 -10
Magnitude (dB)
-20 -30 -40 -50 -60 -70
0
0.1
0.2
0.3 0.4 0.5 0.6 0.7 Normalized Frequency (×π rad/sample)
0.8
0.9
Fig. 3a. Normalised frequency response of 5-ch FB for filter orders 20 Magnitude Response (dB) 0
-10
Magnitude (dB)
-20
-30 -40
-50
-60
0
0.1
0.2
0.3 0.4 0.5 0.6 0.7 Normalized Frequency (×π rad/sample)
0.8
0.9
Fig. 3b. Normalised frequency response of 5-ch FB for filter orders 30
1.2 1 Magnitude
82
0.8 0.6 0.4 0.2 0 1 0.5
1 0.5
0
0
-0.5 F
-0.5 -1
y
-1
F
x
Fig. 3c. 2D Frequency responses of analysis filter 3
Oversampled Perfect Reconstruction FIR Filter Bank Implementation
83
Phase Response 0 -50
Phase (radians)
-100 -150 -200 -250 -300 -350 -400
0
0.1
0.2
0.3 0.4 0.5 0.6 0.7 Normalized Frequency (×π rad/sample)
0.8
0.9
Fig. 3d. 2D Phase response analysis filter 2
The normalized frequency response for 5-channel FB with filter orders 20 and 30 with above mentioned frequency bands are shown in fig 3a and 3b. Fig,3c shows two dimensional filter responses of analysis filter 3 of filter bank. By using real valued direct form-II transposed type-I FIR bandpass filters, design of 5-channels FB is carried out. Selection of frequency bands are important factor for FB, because all the frequency components should pass at the output side without loss of frequency and also without overlapping of frequency components. After design of 1-D FIR filter, there is necessity to convert into 2D form for image application. Therefore by using frequency transformation technique 1D signal is transformed into 2D form. These filters are suitable for even order filters only. For odd order filters this FB will not work. Fig,3 d shows two dimensional phase responses of analysis filter 2 of filter bank. This phase response maintains linearity in 2D form also. Therefore this FB called as linear phase oversampled FIR filter bank. After Analysis bank design signal passed through downsampler and upsampler blocks. By using linear interpolation technique downsampler and upsampler are performed. By using downsampling algorithim image size is reduced by downsample factor of each channel. Similarly by using upsampling algorithim image size is restored as original image size. Then signal is passed through synthesis filter bank. The designs of all synthesis filters are same as analysis filters. Then all the outputs of each channel are combined together and the resultant of filter bank achieved. But to analyze output of FB directly not gives proper difference between input and output image as well as to save memory image compression technique is used. Therefore direct cosine transform (DCT) image compression technique is used. Image compression technique is useful for differentiating outputs of FBs. This filter bank is implemented for different filter orders and analysis for perfect reconstruction is carried out. By using MSE (mean square error) and PSNR (peak signal to noise ratio) in dB, all FBs resultants checked. This filter bank is applied for Lena (Gif) and Barba (Tif) images. Fig. 4a shows Lena input image (gif format) of size 512*512. Resultant of FB shown in Fig. 4 b, which shows that input image restored at the output. From histograms of input and output images, shows that at filter order 20 and subsampled factor 2 we achieve exact replica of input image.
84
S. Chougule and R.P. Patil
Barba (tif image format) image is also applied for filter bank for analysis. Out of the image compression techniques available, transform coding is the preferred method. Since energy distribution varies with each image, compression in the spatial domain is not an easy task Images do however tend to compact their energy in the frequency domain making compression in the frequency domain much more effective. Transform coding is simply the compression of the images in the frequency domain. Transform coefficients are used to maximize compression. For lossless compression, the coefficients must not allow for the loss of any information. The DCT is fast. It can be quickly calculated and is best for images with smooth edges like photos with human subjects. The DCT coefficients are all real numbers unlike the Fourier Transform. The Inverse Discrete Cosine Transform (IDCT) can be used to retrieve the image from its transform representation. Here results of compressed image of each oversampled filter bank compared with input image. It is most easily defined via the mean squared error (MSE) which for two m×n monochrome images I and K where one of the images is considered a noisy approximation of the other is defined as:
Comparison of all the results of compressed output of filter banks with input image in terms of MSE is as shown in table1. After image compression all the results are applied for peak signal to noise ratio [PSNR] estimation. The PSNR is most commonly used as a measure of quality of reconstruction in image compression. It is most easily defined via the mean squared error (MSE). The PSNR is defined as:
input image
50 100 150 200 250 300 350 400 450 500 50
100
150
200
250
300
350
400
450
Fig. 4a. Input Lena image for Filter bank
500
Oversampled Perfect Reconstruction FIR Filter Bank Implementation
85
output image
50 100 150 200 250 300 350 400 450 500 50
100
150
200
250
300
350
400
450
500
Fig. 4b. Output Lena image of five channel Filter bank
Here, MAXI is the maximum pixel value of the image. Comparison of all the results of filter banks in terms of PSNR , MSE for different filter orders with oversample factor 2 is as shown in table1. PSNR and MSE calculated with respect to input compressed image. Histogram of input image and Oversampled filter bank output with filter orders 20 and 40 are shown in fig.5a and 5b respectively . Histogram which gives information in terms of intensity values of input and output images. . For lower i.e. below 20 and higher i.e. above 20 filter orders we can not achieve perfect reconstruction. Fig. 5a and 5b shows histogram error [ error = histogram of i/p – histogram of o/p] between input and output Lena image of FB for filter orders 20, 30, respectively. Here for filter order 20 only shows less histogram errors and also output image near equal to input image 4
x 10
Histogram of difference between I/P & O/P of FB
0
50
2.5
2
1.5
1
0.5
0 100
150
200
Fig. 5a. Histogram error for filter order 20
250
86
S. Chougule and R.P. Patil
Histogram of difference between I/P & O/P of FB
10000
8000
6000
4000
2000
0
0
50
100
150
200
250
Fig. 5b. Histogram error for filter order 30 Table 1. Sr . No.
Filter Order
MSE Lena(Gif Image)
PSNR[ dB] Lena
MSE Barba(Tif Image)
PSNR[dB] Barba
1. 2.
10 20
0.58 11.426
50.4 37.5
1.67 33.12
45.91 32.93
3. 4. 5. 6.
30 40 50 60
49.08 71.89 78.18 93.99
31.2 29.5 29.2 28.4
56.06 64.70 66.94 72.57
30.64 30.02 29.87 29.52
with histogram matches. For lower i.e. below 20 and higher i.e. above 20 filter orders we can not achieve perfect histogram. Fig. 5 shows histogram error [error = histogram of i/p – histogram of o/p] between input and output Lena image and barba images are carried out. Here for filter order 20 only shows less histogram errors and also output image near equal to input image with histogram matches. Table 1 shows filter orders for 5 channels FB with subsample factor 2 and PSNR of compressed input and output image and MSE of Lena and Barba images. The graph of PSNR verses filter orders are shown in Fig.6. Therefore at PSNR approximately at 37dB for Gif image and PSNR approximately at 33dB for Tif image, output of this proposed FB shows near input image. As per as filter order increases above 20, frequency spectrum becomes narrower, means loss of frequency components at transistion gap and less filter orders signal overlapping. Therefore at specific filter order means FB covers all frequency components and for specific subsample factor perfect reconstruction is achieved. DCT Image Compression Out of the image compression techniques available, transform coding is the preferred method. Since energy distribution varies with each image, compression in the spatial domain is not an easy task. Images do however tend to compact their energy in the frequency domain making compression in the frequency domain much more effective. Transform coding is simply the compression of the
Oversampled Perfect Reconstruction FIR Filter Bank Implementation
87
images in the frequency domain. Transform coefficients are used to maximize compression. For lossless compression, the coefficients must not allow for the loss of any information. The DCT is fast. It can be quickly calculated and is best for images with smooth edges like photos with human subjects. The DCT coefficients are all real numbers unlike the Fourier Transform. The Inverse Discrete Cosine Transform (IDCT) can be used to retrieve the image from its transform representation.
DCT: (6.15) IDCT: (6.16) After image compression all the results are applied for peak signal to noise ratio [PSNR] estimation. The PSNR is most commonly used as a measure of quality of reconstruction in image compression. It is most easily defined via the mean squared error (MSE) which for two m×n monochrome images I and K where one of the images is considered a noisy approximation of the other is defined as:
(6.17) The PSNR is defined as: (6.18) Here, MAXI is the maximum pixel value of the image. 5-ch oversampled FB with oversample factor 2 55 Lena Barba 50
PSNR in dB
45 Perfect reconstruction 40
35
30
25 10
15
20
25
30 filter order
35
40
45
50
Fig. 6. From table1 graph for oversampled 5 channel filter bank with oversample factor 2 for images Barba and Leena
88
S. Chougule and R.P. Patil
Comparison of all the results of filter banks in terms of PSNR , mse filter orders with different subsample factors is as shown in table 1. Finaly percentage error is estimated by using formula as below. %Error = [(Input Image – Output Image)/Input Image ] *100. And % Error for five channel oversampled FB for filter order 20 and subsampled factor 2 for all channels is 4 %.
4 Experimentation for Five Channel Filter Bank for Reducing Redundancy and Improvement in Noise Improvement in noise and reducing residual redundancy by selecting oversampled factor 2. For oversampled factor 4, channel 1 subband signal output is shown in fig. 7a Fig. 7b shows histogram difference between input and output image. Similarly For oversampled factor 2, channel 1 subband signal output is shown in fig. 8a Fig. 8b shows histogram difference between input and output image. Here improvement of noise and less redundancy for subsampled factor 2. From table 2 there is improvement in energy, SNR and MSE for subsampled factor 2. sub1 image
Fig. 8b. Histogram difference between input and output image for 5-ch and oversampled factor 2 Table 2.
Subsampled factor
SNR
MSE
Energy difference
4
16
19.9523
0.0782
2
4
11.4256
0.0448
90
S. Chougule and R.P. Patil
5 Conclusion This filter bank is designed with real valued direct form II structure, FIR filters with even filter orders. The investigation of five channel oversampled filter bank is carried out with subsample factor 2 with different filter orders for 2D images and also with subsampled factor 4. This filter bank gives result near perfect reconstruction of input image with filter order 20 only and proper frequency bands per channel for subsampled factor 2. Below filter order 20 we found distortion in FB output image because of more side lobes in transition gap of filters and some frequency components are lost at transition gap. For more filter order i.e. above 20, narrow frequency band and loss of frequency occurs and histogram becomes narrow means output histogram not matches with input image histogram. The analysis is carried out with the help of image compression, PSNR MSE and histogram of images. Inband aliasing that is reducing redundancy and noise removal is carried out with this proposed filter bank.
References 1. Xu, Z., Makur, A.: Theory and lattice Structure for oversampled linear phase paraunitary filter Banks with arbitary filter length, [email protected], [email protected] 2. Tran, T.D., Nguyen, T.Q.: M-channel linear phase FIR filter banks and application in image Compression. IEEE Tran. on Signal Processing 45(9) (September 1997) 3. Harteneck, M., Stewart, R.W., Paez-Borrallo, J.M.: A filterbank design for oversampled filter banks without aliasing in the subbands. In: Proc. TFTS 1997, Warwick, UK (1997) 4. Harteneck, M., Weiss, S., Stewart, R.W.: An oversampled filter bank with different analysis and synthesis filters for the use with adaptive filters. In: Proc. Asilomar Conf. Signals, Systems and Computers, vol. 2, pp. 1274–1278 (November 1997) 5. Chougule, S.R., Patil, R.S.: Uniform and Dyadic FIR filter bank applied for 2D image. In: NCSPA National Conference at DYPCOE Pimpri Pune (September 2007) 6. Vaidyanathan, P.P.: Multirate systems and filter banks. Pearson Education, London (1993) 7. Harteneck, M., Weiss, S., Stewart, R.W.: Design of near perfect reconstruction oversampled filter banks for subband adaptive filters
Designing a Video Control System for a Traffic Monitoring and Controlling System of Intelligent Traffic Systems Il-Kwon Lim, Young-Hyuk Kim, Jae-Kwang Lee*, and Woo-Jun Park Department of Computer Engineering, Hannam Univ., Daejeon, Korea {iklim,yhkim,jklee,wjpark}@netwk.hannam.ac.kr
Abstract. In order to resolve various traffic problems caused by increased traffics, a number of studies have been conducted to improve the existing traffic system by applying state of the art electronic, information and communication technologies and to develop an intelligent traffic system. The traffic monitoring and controlling system equipped with a remote controlling system for such an intelligent traffic system is the system which collects state information of transportation facilities such as a traffic camera and a road signal. On this study, the existing traffic monitoring and controlling system utilizes TCP/IP network, and then it expresses transmitted image with H.264 and implements it with DirectShow. Therefore, the system has been designed to transmit images to 1 to N, unlike the existing transmitting system which only allows 1 to 1 image transmission. Keywords: ITS, Intelligent Traffic System, Traffic Image Transmission, Traffic Monitoring and Controlling System.
1 Introduction The modern society has been stepped up along with development of technologies of networks and vehicles. Despite their positive influences, they have some noted downsides as well. Due to increased number of vehicles, the mobility of a vehicle has experienced fast declination. Moreover, the number of relevant accidents has been inclined. The table 1 shows that the total number of 2.34 million car accidents was incurred from 2000 to 20009, which took 70,000 peoples’ lives and caused 3.64 million injured people. A car accident takes 99.6%, 9.9%, and 95.5% of the total traffic accidents, the dead caused by a traffic accident, and the injured by a traffic accident respectively. Railroad accidents and vessel accidents are reasons of the dead by 3.2% and 2.0%. [1]. Amid these states, a number of studies have been carried out to resolve traffic conditions by more efficiently utilizing the existing traffic system with the state of art network technologies or providing new traffic service. An intelligent traffic system is defined as the information project which allows safe and convenient traffics and an *
*: Including an accidents caused by ultra-light flying vehicles efficient traffic system by providing appropriate information to passenger along with efficient management of traffic facilities, and this can be achieved by adding innovative electronic, information and communication technologies on the existing traffic facilities such as road and cars. In other words, it is an integrated system of traffic networks and telecommunication networks. By adding software technologies such as communication, electronics, control and computing technologies on the infrastructures including roads and vehicles, this integrate system allows that vehicles and traffic infrastructures can complement each other to provide more safe, convenient and efficient traffic [2]. The traffic monitoring and controlling system, which utilizes the remote monitoring network, collects state information of traffic facilities in real time, and it controls the traffic signal based on the received information to cope with road conditions. Unlike this, the existing traffic monitoring and controlling system is composed with PSTN (Public switched telephone network) with 2,400Kbps and a modem. Also, the traffic video controlling system atmosphere of the existing system is designed with the 1:1 video controlling system, which prevent concurrent video controlling on plural clients. Thus, in this study, the video controlling environment of a traffic controlling and monitoring system for an intelligent traffic system is designed to support 1:N video controlling with TCP/IP networks.
Designing a Video Control System for a Traffic Monitoring and Controlling System
93
2 Related Study 2.1 Traffic Signal Control System An intelligent traffic system can be classified into 5 major fields; Advanced traffic management system (ATMS), advanced traffic information system (ATIS), advanced public transportation system (APTS), CVO and advanced vehicles and highway system (AVHS). [2][3] A traffic signal control system belongs to a real-time traffic control service (ATC) of ATMS, and the [figure 1] shows ITS structure constituted with aforementioned 5 systems.
Fig. 1. ITS configuration
A traffic signal control system is the way of controlling traffics on trunk roads and road networks by dividing traffics upon time differences, and it maximizes the efficiency rate of the existing road by efficiently managing traffic signals as well as devices and equipments to enhance the capacity of road and to minimize congestion. The most appropriate traffic signal term is calculated by a computer in real time to control a traffic signal. To support the calculation, data is collected at a certain interval (usually 5 min) by a vehicle monitor installed underground. A traffic signal control system is composed of a central computer system, local controllers, a vehicle monitor and a communication device. 2.2 H.264 After the development of H.263+ and MPEG-4 standards, wireless communication has been widely spread. Consequently, the necessity of video compression technology standards which can be applied to various communication environments was heightened, and then new video compression technologies were approved by ITU-T and ISO/IEC as the final standard under the name of H.264 and MPEG-4 Part 10.
94
I.-K. Lim et al.
H.264, which are the next generation video compression standard, enacted by both ITU-T and ISO, have shown many advantages comparing to the existing technology standard such as MPEG-2 and MPEG-4(Part 2) on the aspect of flexibility and encoding efficiency. Although H.263 uses the similar method of hybrid encoding as the existing standards, it still shows technological advances such as enhanced entropy encoding, block transformation as the size of variable block, enhanced movement estimation and compensation, and in-loop deblocking filter. Comparing to the existing video compression standard including MPEG-2 and MPEG-4, H.264 has a higher compression capacity and flexibility. Nevertheless, it has some negative points as well. The complexity of an encoder and a decoder gets dramatically increase. On the aspect of an encoder, it requires to determine much increased number of parameters and encoding modes comparing to the existing standard. As for an encoder, the amount of calculation gets increased due to deblocking filter and compensation for the movement in a unit of 1/4 pixel. [4] [Figure 2] is the basic structure of an encoder of H.264 [5].
Fig. 2. Block diagram of H.264 encoder
There is H.264/SVC(Scalable Video Coding), which can carry out more efficient operations on the aspect of network channel capacity than H.264/AVC. However, it requires higher Bit-rate than of AVC if it guaranteed the same quality, and it even has 10 to 30% overhead. Therefore, it is more insufficient than AVC on the current network environment [6][7]. 2.3 DirectShow DirectShow is a Multimedia Framework based on Filter, developed by Microsoft. As different platforms on the same base, Quick-Time Framework by Apple and multimedia frameworks such as GStreamer and Xine are competing with DirectShow. DirectShow was developed to replace media process frameworks which cannot control codec, distributes frames in the process of video compression. DirectShow is on the first spot since it can easily manage to deliver high-quality capture easily and fast by using multimedia stream on various video filming devices. With DirectShow, multimedia applications can be developed faster and more easily with hardware elevation support. The exiting traffic monitoring camera can carry out only 1:1 method which can transmit video with exclusive line and fixed IP. However, with
Designing a Video Control System for a Traffic Monitoring and Controlling System
95
DirectShow, the traffic monitoring camera shall be able to carry out multicasting. Not limited to that, DirectShow based on Windows supports more wide range of formats than Renux which is based on multimedia framework, including ASF(Advanced Systems Format), WMV(Windows Media Video), MPEG(Motion Picture Experts Group), AVI(Audio-Video Interleaved), MP3(MPEG Audio Layer-3), H.264 [8]. 2.4 Real-Time Video Communication Protocol In general, a video communication system uses RTP and RTPC based on UDP (User Datagram Protocol) for real-time data transmission. RTP includes media and basic information of media, and Packet includes markerbit, sequence number, and time stamp. The structure of RTP Packet is shown in [figure 3] [9].
Fig. 3. Structure of RTP Packet
RTCP Packet has a structure of the figure 4 and it is classified into 5 types; RR(Receiver Report), SR(Sender Report), SDES(Source DEScription Message), APP(Application Specific RTCP), and BYE(Bye Message). SR and RR packets are the packet relevant to transmit and receive. The difference is that SR Packet has a sender information section of 20 byte.
Fig. 4. Structure of RTCP RR Packet
96
I.-K. Lim et al.
3 System Design A traffic monitoring and controlling system receives basic information form a loop sensor and a video sensor. The information is then processed by a traffic monitoring and controlling system on the site as the first stage. The processed information goes through a series of algorithm which provides the processed information to the central server at the set interval. The central server transmits a signal plan back to the traffic monitoring and controlling system, and the system calculates appropriate signal terms and then sends it to a traffic light. The existing real-time signal system is composed with PSTN and a modem equipped with 2400Kbps network. On the existing system, Sensor Gateway is equipped to design it to be able to employ TCP/IP networks. In the study, it is designed to view it from several system monitors after setting multicasting IP address on the received monitor of the specially designed system. As can be seen from the [figure 5], the existing system only 1:1. In that case, one ID is granted to one client. Therefore, if received video was transmitted to each system, each ID has to implement it, and if the screen is updated, update has to be processed by each ID which means more works are to be done to implement. Thus, it requires larger system resource.
Fig. 5. The Method of Video Transmission of the Existing System
Fig. 6. Transforming to Grid for Transmitting Multicasting Video
Designing a Video Control System for a Traffic Monitoring and Controlling System
97
Fig. 7. Enhanced Video Transmission Method
By using DirectShow, the picture which requires to be transmitted to every client is be turn to grid. So the picture to be transmitted shall be processed as one and granted on ID. Through such process, only 1 ID is required to be processed in order to update screens. This process is so much easier to implement and causes less load on the system. Also, it allows transmit updated videos to other client and the central server. Therefore, the method of video transmission is enhanced like the [figure 7].
4 Displaying Monitoring Video for a Traffic Monitoring and Controlling System For transmitting traffic condition video, a traffic monitoring and controlling system is designed to be able to carry out multicasting to transmit video to 1:N, rather than 1:1, by using DirectShow media replay environment, utilizing H.264/AVC video codec.
Fig. 8. Displayed Video Monitoring System
98
I.-K. Lim et al.
Basically current 1:1 environment can maintain a sole session between a client monitoring CCTV and a monitoring center. However, with the 1:N of a multi-cast group, suggested by this study, not only video transmission between clients monitoring CCTV for each area is allowed, but also video transmission between a client and a monitoring center. It is reasonable to assume that this system provides more efficient and abundant information requires for monitoring traffics and analyzing traffic flows. On the study, multicasting IP address is set on the received screen of the system so it can be displayed on several system monitors.
5 Conclusion An intelligent traffic system is expected to provide the most efficient traffic solution which provides intelligent links between people, vehicles and national infrastructures. An intelligent traffic system has made a quite developments and enhancements. For the study, the monitoring video is expressed with H.264 and displayed with DirectShow along with ALL-IP of networks of a traffic monitoring and controlling system which monitors and controls traffic control. H.264 is a video transmission codec for assessing traffic for a traffic controlling and monitoring system. This is a video encoding standard with the most superior compression rate. With multiple block coding modes and multi reference picture, it has twice higher compression rate of MPEG0-2 and 40% more than of MPEG-4 ASP with the same quality. Also, comparing to H.263, it is designed to have better quality at 50% less of bit rate, which enables it have superior efficiency. Also, the existing monitoring camera is designed to transmit to 1:1 unlike this system which provides 1:N multicasting transmission with DirectShow and have higher efficiency. Despite such advantages, a traffic monitoring and controlling system, one of public facility, is usually exposed to the public. Therefore, proper security measures shall be prepared for data transmission in order to apply it to the actual system. Acknowledgments. This research was supported by "Industry-University Partnership Laboratory Supporting Business".
References 1. 2010 Traffic Safety Annual Report, Ministry of Land, Transport and Maritime Affairs (August 2010) 2. Don, M.H.: Domestic and International ITS Market conditions and Implications, 15 products market conditions report (June 2003) 3. ITS standardization roadmap investigation, Korea national computerization agency (December 1998) 4. Kim, S., Cho, H.: H.264/AVC Baseline decoder design technologies, IT SoC Magazine (November 2005) 5. Suk, J., Kim, B., Lee, J., Cho, C.: development of HD level H.264 technologies (February 2006)
Designing a Video Control System for a Traffic Monitoring and Controlling System
99
6. Lee, S.-H.: An Efficient SVC Transmission Method in an IP Network. KICS Journal 34(4) (2009) 7. Chen, P., et al.: A network – adaptive SVC Streaming Architecture. In: ICACT 2007 (2007) 8. MSDN DirectShow Webpage, http://msdn.microsoft.com/en-us/library/dd375454(v=VS.85).aspx 9. Schulzrinne, H.: RTP: A transport protocol for real-time applications, Internet Engineering Task Force, RFCC 3551 (July 2003) 10. Kim, B.-Y., Lee, S.-O., Jung, K.-S., Sim, D.-G., Lee, S.-Y.: Real-Time Video Quality Assessment of Video Communication Systems. IEEK Journal 46(3) (2009)
Approximate Reasoning and Conceptual Structures Sylvia Encheva Stord/Haugesund University College, Bjørnsonsg. 45, 5528 Haugesund, Norway [email protected]
Abstract. In this paper we are applying approximate reasoning methods for extracting conceptual structures from collected data. Stabilities of previously obtained concepts are investigated by removing attributes from the data set. Another search for interesting patterns by build nested lattices and compare the obtained concepts with the ones resulting from applying the first two approaches is also enclosed. Keywords: Clustering, formal concept analysis, fuzzy logic, predictions.
1 Introduction This work is aimed at finding interesting patterns in knowledge discovery applications. We first apply formal concept analysis methods for deriving concepts from particular data sets. We also investigate concepts stabilities by removing attributes from the same data set. Another way to search for interesting patterns is to build nested lattices and compare the obtained concepts with the ones resulting from applying the first two approaches. As a practical application we discuss master students enrolled in research methods course. They are taking tests in each chapter and a midterm test, all web based. The goal is to see which student might not pass the final exam and which student might not have obtained sufficient knowledge in order to continue a PhD study. Data from previous years is used to build up clusters indicating possible developments. The clusters represent concepts resulting from application of formal concept analysis. New students are placed in existing clusters applying a fuzzy function. The rest of the paper is organized as follows. Related work and supporting theory may be found in Section 2. The decision process is presented in Section 3. Conclusions and future work can be found in Section 4.
2 Preliminaries Fuzzy reasoning methods are often applied in intelligent systems, decision making and fuzzy control. Some of them present a reasoning result as a real number, while others use fuzzy sets. Fuzzy reasoning methods involving various fuzzy implications and compositions are discussed by many authors, f. ex. [1], [5], [11], [13], [14], [15], and [18]. Fuzzy linear programming problems are considered as multiple fuzzy reasoning schemes in [6]. A novel statement of fuzzy mathematical programming problems is introduced in [2] as well as a method for finding a solution to such problems. T.-h. Kim et al. (Eds.): UCMA 2011, Part I, CCIS 150, pp. 100–109, 2011. c Springer-Verlag Berlin Heidelberg 2011
Approximate Reasoning and Conceptual Structures
101
An interesting prediction method presented in [7] applies formal concept analysis and fuzzy inference. In particular it shows how to calculate the value of a membership function of an object if the object belongs to a known concept. The authors however do not state whether classical set theory or fuzzy set rules are applied in establishing an object’s belonging to a concept. It seems that the method can be applied only to objects with predetermined concept membership. Subsequently the method does not show how to handle objects that partly belong to more than one concept. In this paper we propose solutions to these questions. Another interesting question that is not addressed in [7] is how the number of objects belonging to different concepts effects the outcome a particular prediction. Our approach involves a fuzzy function that allow changes in the number of objects in any concept to influence the final conclusion. Distances in probabilistic metric spaces are modelled by triangular norms [12]. The logical connective and in fuzzy sets theory is modelled by triangular norms. Definition 1. [12] A mapping T : [0, 1] × [0, 1] → [0, 1] is a triangular norm (t-norm for short) if and only if it is symmetric, associative, nondecreasing in each argument and T (a, 1) = a, for all a ∈ [0, 1].
3 Concepts Let P be a non-empty ordered set. If sup{x, y} and inf {x, y} exist for all x, y ∈ P , then P is called a lattice [3]. In a lattice illustrating partial ordering of knowledge values, the logical conjunction is identified with the meet operation and the logical disjunction with the join operation. A context is a triple (G, M, I) where G and M are sets and I ⊂ G×M . The elements of G and M are called objects and attributes respectively [3] and [17]. For A ⊆ G and B ⊆ M , define A = {m ∈ M | (∀g ∈ A) gIm}, B = {g ∈ G | (∀m ∈ B) gIm} where A is the set of attributes common to all the objects in A and B is the set of objects possessing the attributes in B. A concept of the context (G, M, I) is defined to be a pair (A, B) where A ⊆ G, B ⊆ M , A = B and B = A. The extent of the concept (A, B) is A while its intent is B. A subset A of G is the extent of some concept if and only if A = A in which case the unique concept of the which A is an extent is (A, A ). The corresponding statement applies to those subsets B ∈ M which is the intent of some concepts. The set of all concepts of the context (G, M, I) is denoted by B(G, M, I). B(G, M, I); ≤ is a complete lattice and it is known as the concept lattice of the context (G, M, I).
102
S. Encheva
The sum-of-1-criterion [8] states that Σi∈Mi mi (x) = 1, ∀x ∈ χ where Mi , i = 1, ..., k denotes all possible membership terms {mi , i = 1, ..., k} of a fuzzy variable in some universe of discourse χ. Suppose a formal context C has objects Pψ , ψ = 1, ..., r, attributes Tφ , φ = 1, ..., r, and sub attributes Tφς , φ = 1, ..., r, ς = 1, ..., q. By θ we denote the threshold for membership values above which an entry is regarded as significant, as in [7]. This is achieved by computing the arithmetic mean of all entries within a column and take it as a threshold. Other means can be considered. Our goal is to predict whether an object Pr+s , Pr+s ∈ / C possesses an attribute witha Tj
i value regarded as significant. For this purpose we construct a fuzzy function Υ Pr+s , where j μt Pt , Tij · v(Pt , Tij ) Ti Υ Pr+s = |T |
and μt Pt , Tij =
i=i
v Pt , Tij
η The meaning of the used notations is as follows: Υ - fuzzy function indicating the grade to which object Pr+s possesses attribute Tij , Tij - jth sub attribute of the ith attribute, v Pψ , Tφς - value of attribute Tφς of object Pψ ,
Tij - specifically chosen sub attributes, possessed by a single object Pt , t ∈ T , where i = i , j = j since the value v Pr+s , Tij is unknown, η - number of attributes’ values v Pt , Tij , T - the set of concepts to which object Pr+s belongs, |T | - the cardinality of T .
We first select all concepts to which object Pr+s belongs to with respect to sub attribute Tij . The selection criterion is based on comparison of all sub attribute values of object Pr+s with the corresponding values in the same sub attribute column of a table representing the formal context. The concept with the closest sub attribute value is taken. If there are several concepts with the same sub attribute value we include all of them. Ones the concepts are identified, we calculate the value of each j v P , T t i i=i μt Pt , Tij = , η where j refers to sub attributes of similar meaning. Suppose sub attributes have meaning ’high’ or ’low’ and we are interested in a sub attribute denoted as ’high’. For calcu-
Approximate Reasoning and Conceptual Structures
103
lating μt Pt , Tij we then take only sub attributes denoted as ’high’. The influence of the rest of the attributesis included due to the sum-of-1-criterion. Tj
i Finally if a value Υ Pr+s ≥ θ Tij we say that the degree to which object Pr+s
possesses attribute Tij is significant. A small example of a formal context is presented. Suppose object Pr+s belongs to concepts with intents P1 , P2 , P3 , P4 and P5 . We know the values of all the sub attributes’ values for any of the first r objects and sub attributes’ values for object Pr+s , denoted by a • in Table 1, Table 2 and Table 3. The value v Pr+s , T43 is calculated 3 T4 below based on the fuzzy function Υ Pr+s . v P1 , T13 + v P1 , T23 + v P1 , T33 μ1 P1 , Ti3 = |Ti3 |
v P2 , T13 + v P2 , T23 + v P2 , T33 = |Ti3 |
v P3 , T13 + v P3 , T23 + v P3 , T33 = |Ti3 |
v P4 , T13 + v P4 , T23 + v P4 , T33 = |Ti3 |
v P5 , T13 + v P5 , T23 + v P5 , T33 = |Ti3 |
μ2 P2 , Ti3
μ3 P3 , Ti3
μ2 P4 , Ti3
μ3 P5 , Ti3 v Pr+s , T43 =
μ1 (P1 ,T43 )·v(P1 ,T43 ) 5
+
+
μ4 (P4 ,T43 )·v(P4 ,T43 ) 5
μ2 (P2 ,T43 )·v(P2 ,T43 ) 5
+
+
μ2 (P5 ,T43 )·v(P5 ,T43 ) 5
μ3 (P3 ,T43 )·v(P3 ,T43 ) 5
If v Pr+s , T43 ≥ θ43 than we say that object Pr+s possesses sub attribute T43 . Students are divided in two groups according to their exam grades. The first group contains students who have obtained at least 70% correct answers while the second group of students have obtained between 50% and 70% correct answers. The first group is encouraged to continue with PhD studies and the second group is advised to put additional efforts in learning if further education is planned. Outcome of tests related to Chapter 1, Chapter 2, a midterm test, Chapter 3, Chapter 4 and Chapter 5 is presented in Fig. 1. The initial data can be seen in Table 1 and Table 2. The first halve of the first group has slightly better results than the second halve. It seems that increased learning activities in the second halve of the semester still cannot quite compensate for the lesser attention at the beginning of the semester. Example 1.
Fig. 1. Outcome of tests related to Chapter 1, Chapter 2, a midterm test, Chapter 3, Chapter 4 and Chapter 5
The approach was tested with data from another group of students. The outcome conformed the data with 80% accuracy. A closer look at the lattice in Fig. 1 shows that results from tests C12, C23, C33 point to the best students. These students also dominate while counting number of tests with results above the average. When it comes to students with 50% correct answers to the final exam the picture is not that clear, f.ex. one of them obtained 10% correctness on test C53 while the other one obtained 90%. For working with fuzzy inference systems we suggest application of Tsukamotos fuzzy reasoning method [16]. Let Aij be a value of the linguistic variable xj defined in the universe of discourse, y = y1 , ..., yn is a crisp vector and αi = T (Ai1 (y1 ), ..., Ain (yn )), i = 1, ..., m where T often is the minimum or the product t-norm. In the procedure for obtaining the crisp output, z0 , the overall system output is defined as the weighted average of the individual outputs, where associated weights are the firing levels, i=1,...,m αi zi z0 = i=1,...,m αi i.e. z0 is computed by the discrete Center-of-Gravity method. 3.1 Formal Concepts Stabilities Stability of a formal concept is discussed in [9] and [10].
Fig. 2. Reduced context related to Chapter 1, Chapter 2, a summary test, Chapter 3, Chapter 4 and Chapter 5
Approximate Reasoning and Conceptual Structures
107
Fig. 3. Nested lattice relating outcomes from Chapter 1 and Chapter 2
Definition 2. Let (A, B) a formal concept of B(G, M, I). Stability of (A, B) is |{C ⊆ A|C = A = B}| 2|A| The stability index of a concept indicates how much the concept intent depends on particular objects of the extent. Given a concept (A, B), the stability index measures the number of elements of G that are in the same equivalence class of A, where an equivalence class is defined as follows. γ(A, B) =
Definition 3. Let X ⊆ G, we denote by X the equivalence class of X where: X = {Y ⊆ G|Y = X } Note that when X is closed, any Y in X is a subset of X. Thus, considering a formal concept (A, B), definition 3 can be rewritten as: γ(A, B) =
|A | 2|A|
108
S. Encheva
Then, the larger the equivalence class of an extent is (with respect to extent size), the more stable the concept is. The idea behind stability is that a stable concept is likely to have a real world interpretation even if the description of some its objects (i.e. elements in the extent) is ”noisy”. The lattice in Fig. 2 corresponds to the lattice in Fig. 1 3.2 Nested Lattices A nested lattice is the product of two concept lattices, graphically represented by a nested line diagram, sometimes referred to as inner and outer lattice. A nested line diagram consists of an outer diagram that contains inner diagrams in each node. Inner diagrams are not necessarily congruent but only substructures of congruent diagrams. Congruent diagrams are shown as structures possessing some unrealized concepts. Nested line diagrams, similar to the tree structure, make the focus concepts the roots and associate each sequence of concepts below the focus with a path like the trunk of a tree, [4]. The formal concept analysis nested line diagrams are based on combining multiple partial views of the data represented in the context, [4]. The overall effect of building nested line diagrams is mapping concepts into a clear identifiable picture of reality. That means having several complete lattices of a partial context nested into one another rather than a partial lattice of a context, [4]. The nested lattice in Fig. 3illustrates dependencies between different tests results.
4 Conclusion The goal of this work is to find correlations between results from continuous evaluation and exam results. Considering the relatively small sample of initial data we would say that early feedback presented to students has positive effect on their final grades.
References 1. Bellman, R.E., Zadeh, L.A.: Decision making in a fuzzy environment. Management Sciences, Series B 17, 141–164 (1970) 2. Carlsson, C., Fuller, R.: Optimization under the fuzzy if-then rules. Fuzzy Sets and Systems 119(1) (2001) 3. Davey, B.A., Priestley, H.A.: Introduction to lattices and order. Cambridge UniversityPress, Cambridge (2005) 4. Fang, K., Chang, C., Chi, Y.: Using Formal Concept Analysis to Leverage Ontology-Based Acu-Point Knowledge System. In: Zhang, D. (ed.) ICMB 2008. LNCS, vol. 4901, pp. 115– 121. Springer, Heidelberg (2007) 5. Felix, R.: Relationships between goals in multiple attribute decision making. Fuzzy Sets and Systems 67, 47–52 (1994) 6. Fuller, R., Zimmermann, H.-J.: Fuzzy reasoning for solving fuzzy mathematical programming problems. Fuzzy Sets and Systems 60, 121–133 (1993) 7. Herrmann, C.S.: Fuzzy logic as inferencing techniques in hybrid AI-Systems. In: Martin, T., L. Ralescu, A. (eds.) IJCAI-WS 1995. LNCS, vol. 1188, pp. 69–80. Springer, Heidelberg (1997)
Approximate Reasoning and Conceptual Structures
109
8. Herrmann, C.S., Holldobler, S., Strohmaier, A.: Fuzzy conceptual nowledge processing. In: ACM Symposium on Applied Computing, 1996, pp. 628–632 (1996) 9. Kuznetsov, S.O.: On stability of a formal concept. Annals of Mathematics and Artificial Intelligence 49, 101115 (2007) 10. Kuznetsov, S.O., Obiedkov, S., Roth, C.: Reducing the representation complexity of latticebased taxonomies. In: Priss, U., Polovina, S., Hill, R. (eds.) ICCS 2007. LNCS (LNAI), vol. 4604, pp. 241–254. Springer, Heidelberg (2007) 11. Mamdani, E.H.: Applications of fuzzy logic to approximate reasoning using linguistic synthesis. IEEE Transactions on Computers 26(12), 1182–1191 (1977) 12. Schweizer, B., Sklar, A.: Associative functions and abstract semigroups. Publ. Math. Debrecen. 10, 69–81 (1963) 13. Sugeno, M.: Fuzzy measures and fuzzy integrals: a survey. In: Gupta, M.M., Saridis, G.N., Gaines, B.R. (eds.) Fuzzy Automata and Decision Processes, pp. 89–102. North-Holland, NY (1977) 14. Sugeno, M.: Industrial applications of fuzzy control. Elsevier Science Pub. Co., Amsterdam (1985) 15. Takagi, T., Sugeno, M.: Fuzzy identification of systems and its applications to modeling and control. IEEE Transactions on Systems, Man and Cybernetics, 116–132 (1985) 16. Tsukamoto, Y.: An approach to fuzzy reasoning method. In: Gupta, M.M., Ragade, R.K., Yager, R.R. (eds.) Advances in Fuzzy Set Theory and Applications (1979) 17. Wille, R.: Concept lattices and conceptual knowledge systems. Computers and Mathematics with Applications 23(6-9), 493–515 (1992) 18. Zadeh, L.A.: The concept of linguistic variable and its applications to approximate reasoning, Parts I, II, III. Information Sciences 8, 199–251 (1975); 8, 301–357 (1975); 9, 43–80 (1975)
Edge Detection in Grayscale Images Using Grid Smoothing Guillaume Noel, Karim Djouani, and Yskandar Hamam French South African Institute of Technology, Tshwane University of Technology, Pretoria, South Africa
Abstract. The present paper focuses on edge detection in grayscale images. The image is represented by a graph in which the nodes represents the pixels and the edges reflect the connectivity. A cost function is defined using the spatial coordinates of the nodes and the grey levels present in the image. The minimisation of the cost function leads to new spatial coordinates for each node. Using an adequate cost function, the density of points in the regions with large gradient values is increased. The new grid is then fed into an edge detector, which uses the geometric characteristics of the graph. The result is a sub-graph representing the edges present in the original image. The algorithm is tested on real images and the results are compared to existing edge detection techniques. Keywords: Grid smoothing, Edge detection, Non-linear optimisation.
1
Introduction
In image processing and computer vision, edge detection focuses on the localisation of significant variations in the grayscale image. The edge detector process is very important for a large class of applications like motion estimation, image enhancement, compression and so on [1]. Edge detection in grayscale images is a complex task and it is difficult to design a general edge detection technique which performs well in many contexts. Most of the techniques rely on a filtering (smoothing) of the image and image differentiation. The filtering leads to a modification of the information contained in the image while the differentiation is an ill-defined problem [1]. The main challenges of edge detection include the detection of weak edges, a low number of false edges detection and a proper spatial positioning of the detected edges. On top of the filtering and the image differentiation, all the edge detection techniques developed rely on the conventional definition of an image, that is to say a matrix of uniformly distributed pixels. As a result, even in the ideal situations, a clear straight edge whose orientation is 45 degrees is represented by a staircase-like line. As a consequence, the results of conventional edge detectors often lead to an inaccurate shape definition of the objects present in the image. Various solutions have been proposed to overcome this issue. Super-resolution algorithms were developed to enhance the resolution of the image while conserving its attributes ([3], [4]and [5]). Conventional edge detection algorithms are then applied to the enhanced images. T.-h. Kim et al. (Eds.): UCMA 2011, Part I, CCIS 150, pp. 110–119, 2011. c Springer-Verlag Berlin Heidelberg 2011
Edge Detection in Grayscale Images Using Grid Smoothing
111
Other approaches focus on the application and consequently a priori knowledge, to recover the information in the images. For example, in remote sensing, researchers pre-process the image using low-pass filter [7], contextual filter [9], adaptive filtering [10] and others [8]. All the techniques mentioned above rely on the definition of the pixel (even it is a sub-pixel in super-resolution) or are application-oriented. The gridsmoothing technique presented in the present paper modifies the uniform grid according to the grayscale levels of the images. The result of the gridsmoothing is no longer an image in the conventional sense. By fitting the grid to the objects present in the image, the grid smoothing is facilitating the edge detection. Previous work on the grid smoothing or interpolation may be found in [11] where the image is modelled as a non resistive power grid, in [12] where strong constraints on the shape of the object are assumed, and in [13], where hierarchical grid construction is introduced. None of the work cited above used a graph-based approach combined with optimisation techniques. Section 2 of this paper presents the graph-based representation of an image while section 3 discusses the general framework of the grid smoothing. Section 4 introduces the new edge detection algorithm associated to gridsmoothing. Section 5 presents and discusses the results of the simulations. Conclusion and recommendations may be found in section 6.
2
Graph-Based Image Representation
Our input data is a graph G = (V, E), embedded in the 3D Euclidean space. Each edge e in E is an ordered pair (s, r) of vertices, where s (resp. r) is the sending (resp. receiving) end vertex of e [6]. To each vertex v is associated a triplet of real coordinates xv , yv , zv . Let Cve be the node-edge incidence matrix of the graph G, defined as: ⎧ ⎨ 1 if v is the sending end of edge e Cve = −1 if v is the receiving end of edge e (1) ⎩ 0 otherwise In the rest of the paper, the node-edge matrix Cve will also be denoted C. Considering an image with M pixels, Z, Y and Z respectively represent t t t [x1 , ..., xM ] , [y1 , ..., yM ] and [z1 , ..., zM ] . X and Y are at first uniformly distributed (coordinates of the pixels in the plane), while Z represents the grey level of the pixels. Each pixel in the image is numbered according to its column and then its rows. . We denote L the number of vertices in the graph. C is consequently a matrix with L rows and M columns.
3 3.1
Optimisation-Based Approach to Grid Smoothing General Framework
A cost function is introduced to fit the objects in the image with the grid. The main idea is that the regions where the variance is small (low gradient)
112
G. Noel, K. Djouani, and Y. Hamam
require less points than the region with a large variance (large gradient). The grid smoothing technique moves the points of the grid from small variance regions to large variance regions. To achieve this goal, a cost function J is defined as follows: J = JX + JY (2) where JX = and JY =
t 1 ˆ Q X −X ˆ + θ1 X t AX + θ2 X t A2 X X −X 2
(3)
t 1 Y − Yˆ Q Y − Yˆ + θ1 Y t AY + θ2 Y t A2 Y 2
(4)
where Q and A = C t ΩC are diagonal matrices. The matrix Ω is diagonal and defined as follows: 2 Ωk,k = (zi − zj ) (5) where node i is the sending end of the vertex k and node j the receiving end. The other elements of Ω are equal to zero. Ω is a square matrix L × L. The first term in the expression of the cost function is called the attachment as it penalises the value of the cost function if the coordinates of the points are too far from the original values. It is introduced to avoid large movement in the grid [6]. The second term is called first order term, while the third one is called second order term. θ1 and θ2 are real numbers and are acting as weighting factors between the terms of the cost function. As a result of the definition of Ω, the minimisation of J is leading to the reduction of the areas of the triangle formed by two connected points and the projection of one of the point on the Z-axis. The edges in the image acts as attractors for the points in the grid. As a consequence, the edges are better defined in terms of location and steepness in the smoothed grid. 3.2
Convergence of the First Order Cost Function with Attachment
This section focuses on showing the existence of a unique solution for the minimisation problem presented above. The solution is presented for JX only. The proof for JY can be derived in a similar manner. The cost function of the first order with attachment may be expressed as:
t t 1 ˆ ˆ JX = X − X Q X − X + θ X AX (6) 2 The gradient of the first order cost function JX with attachment is: ˆ + θAX ∇x J X = Q X − X
(7)
At the optimum, the gradient is equal to zero. Let Xopt be the optimal solution for X. Xopt may be expressed as: Xopt = (Q + θA)
−1
ˆ QX
(8)
Edge Detection in Grayscale Images Using Grid Smoothing
113
The equation above shows that a unique optimal solution exists (the existence of the inverse of Q + θA may be shown) for the minimisation problem and that for small scale problem, the solution can be derived easily. For large scale problem, a gradient descent method may be used. Let Xn+1 and Xn be respectively the values of X at iteration n + 1 and n. Xn+1 is equal to ˆ + θAX Xn+1 = Xn − αn ∇x JX = Xn − αn Q X − X (9) αn is the step and may be chosen optimal or not. An optimal step leads to a smaller number of iterations while increasing the number of operations required for the optimisation. The optimal step αn may be expressed by: αn =
∇x
∇x J t ∇x J (Q + θA) ∇x J
Jt
(10)
t
If [λ1 , . . . , λM ] are the eigenvalues of A,the fixed step α which satisfies the monotonic convergence, may be expressed by: 1 α= (11) 1 + max ([λ1 , . . . , λM ]) 3.3
Convergence of the Second Order Cost Function with Attachment
This section focuses on showing the existence of a unique solution for the minimisation problem presented above. The cost function of the second order with attachment may be expressed as:
t t 2 1 ˆ ˆ JX = X−X Q X −X +θ X A X (12) 2 The gradient of the first order cost function JX with attachment is: ˆ + θA2 X ∇x JX = Q X − X
(13)
At the optimum, the gradient is equal to zero. Let Xopt be the optimal solution for X. Xopt may be expressed as: −1 ˆ Xopt = Q + θA2 QX (14) The equation above shows that a unique optimal solution exists (the existence of the inverse of Q + θA2 may be shown) for the minimisation problem and that for small scale problem, the solution can be derived easily. For large scale problem, a gradient descent method may be used. Let Xn+1 and Xn be respectively the values of X at iteration n + 1 and n. Xn+1 is equal to ˆ + θA2 X Xn+1 = Xn − αn ∇x JX = Xn − αn Q X − X (15) αn is the step and may be chosen optimal or not. The optimal step αn may be expressed by: ∇x J t ∇x J αn = (16) ∇x J t (Q + θA2 ) ∇x J
114
3.4
G. Noel, K. Djouani, and Y. Hamam
Convergence Speed and Processing Power
As mentioned in the previous sections, when the gradient descend method is applied, a fixed step or an optimal step may be used. A fixed step method requires less operations than the optimal step gradient descent while converging towards the solution in a monotonic manner. On the other hand, the optimal step gradient descent converges faster towards the optimum but oscillations in the values of the cost function may appear. In the grid smoothing approach, the convergence towards the optimal solution depends on the value of θ. The convergence is quicker with large values of theta as shown in the Fig. 1. 5
Fig. 1. Convergence of the first order cost function according to θ (fixed step)
The experiments show that the choice of the gradient descent method to use (fixed or optimal step) depends mainly on the value of θ. If θ is small, the number of iterations required is small. The fixed step gradient descent is chosen. If θ is large, the number of iterations required is large for a fixed step method.In this case, the steepest gradient descent is selected. The optimisation of the first order cost function takes half of the time required for the second order cost function optimisation. It may be explained by the characteristics of the matrix A. A is a sparse matrix. A2 is still a sparse matrix. However the number of non-zero elements in A2 is greater than in A (the ratio is about two).
4
Edge Detection
Edge detectors are based on the measurement of the gradient in the image. Their performances depend on the approximations of the gradient implemented as well as the pre-processing stage used (filtering). For example, the Canny edge detector implements a Gaussian filtering of the image and its gradient [2]. The pixels with high gradient are then flagged (using two thresholds) and represent the edges. The detected edges are then improved in terms of continuity and localisation using various techniques (estimation of the direction of the gradient, streaking...)[2].The implementation of the classical edge detectors, while very efficient in terms of speed and results, ends up with edge maps defined at a pixel precision of smoothed images.
Edge Detection in Grayscale Images Using Grid Smoothing
115
The edge detection technique using the grid smoothing approach is a simpler process using only the geometric properties of the grid. As presented in the sections above, the smoothing of the image grid leads to a compression of the grid in the areas of the image with high gradient and a relaxation of the grid otherwise. Moreover, the connections in the grid are aligned, by the grid smoothing process, with the direction of the gradient. In other terms, the connections fit the objects in the image. If the connectivity of the pixels is chosen equal to four, the initial grid is composed by squares which are compressed along one or several directions during the smoothing. The results are either diamond-shape quadrilaterals or rectangles. For a quadrilateral Q composed by X1 , X2 , X3 and X4 , let us denote d1 and d2 , the lengths of the two diagonals. d1 represents the distance between X1 and X4 . d2 represents the distance between X2 and X3 . For each quadrilateral, a feature fQ is extracted and is equal to: 2 2 (z1 − z4 ) (z2 − z3 ) fQ = max , (17) d1 d2 A large value of fQ may be found in compressed quadrilaterals while a low value of fQ denotes very little change in the initial square. By definition of the cost function, the quadrilaterals belonging to an edge may be determined by using a threshold on fQ .The selected quadrilaterals are then represented by the diagonal which is not selected for fQ . The choice of the threshold σ depends on the application. A large value of σ leads to the loss of weak edges in the image while a small value increases significantly the number of false edges detected.
5
Simulation Results
The simulations were performed using a standard laptop(1.87 GHz processor, 2GB RAM and Windows Vista SP1 as operating system) and Matlab R14 Service Pack 2. The algorithms were tested on an image obtained from the Matlab library. The process including grid smoothing and edge detection takes about 5 seconds for a 100 × 100 pixels image. The first set of simulations focuses on the comparison in performance between the first order and the second order grid smoothing. Two details of the original image are extracted and processed (see Fig. 2). The first detail represents an eye while a mouth and the tip of a nose may be found in the second. The first order and second order grid smoothing are applied to the two details. A value of θ = 1 is used for the eye and θ = 0.01 for the mouth. The results show that the objects in the image may be recognised in both cases using the two orders of smoothing. It may be observed that the level of noise in the case of a first order smoothing is significant while the second order leads to a clean grid. However, the definition of the edges is far better with the first order smoothing. The second order smoothing gives a powerful fitting of the objects which are well defined with a low level of noise. On the other hand, the first order smoothing includes
15
20
25
30
35
40 10
15
20
25
30
35
40
45
50
45
40
35
5
15
20
25
30
35
40
−50
−45
−40
−35
15
20
25
30
35
40
45
(e) First order grid smoothing
10
50
10
15
20
25
30
35
40
45
5
10
15
20
25
30
35
40
45
(f) Second order grid smoothing
−50
−45
−40
−35
−30
−25
−20
Fig. 2. Conparison between first order and second order grid smoothing
(d) Original image
10
−25
25 −30
−20
−15
−15
20
30
5
(c) Second order grid smoothing
−50
−45
15
5
50
−5
(b) First order grid smoothing
5
−10
−50
−5
50
50
−10
45
45
−40
−35
−30
5
(a) Original image
10
−45
−40
−35
−25
10
50
45
40
35
5
−25
25 −30
−20
20
30
−15
−15
15 −20
−5 −10
−5 −10
5
10
50
50
116 G. Noel, K. Djouani, and Y. Hamam
10
20
30
40
50
60
70
80
90
100
−100
−90
−80
−70
−60
−50
−40
−30
−20
−10
10
20
30
40
50
60
70
80
90
100
20
40
60
80
100
−100
−90
−80
−70
−60
−50
−40
−30
−20
−10
10
20
30
40
50
60
70
80
90
100
Fig. 3. Results of the edge detection
(d) Edge detected using θ = 1 and σ = 500 (e) Edge detected using θ = 0.005 and (f) Edge detected using θ = 0.005 and σ = 300 σ = 500
−100 0
100
−100
90
−90
−90
80
−80
−80
70
−70
−70
60
−60
−60
50
−50
−50
40
−40
−40
30
−30
−30
20
−20
−20
10
−10
−10
(a) Result of the grid smoothing with (b) Result of the grid smoothing with (c) Edge detected using θ = 1 and σ = 300 θ=1 θ = 0.005
−100
100
−100
90
−90
−90
80
−80
−80
70
−70
−70
60
−60
−60
50
−50
−50
40
−40
−40
30
−30
−30
20
−20
−20
10
−10
−10
Edge Detection in Grayscale Images Using Grid Smoothing 117
118
G. Noel, K. Djouani, and Y. Hamam
more details in the fitting at the expense of a greater impact of the noise. For the rest of the simulation, the first order smoothing was chosen for its capabilities to detect details in the image. The second set of simulations focuses on the influence of θ on the grid smoothing. The results are presented for two values of θ. The size of the original image is 100 × 100 pixels. It represents a part of a face. The eyes, the nose, one nostril, the cheek, the eyebrows and the upper lip may be clearly identified. The size of the image is chosen relatively small as it is difficult to display the grid for an higher number of pixels. Fig. 3(a) and Fig. 3(b) show the results of the first order grid smoothing for θ equals, respectively, 1 and 0.005. The main characteristics of the face may be easily seen in both Fig. 3(a) and Fig. 3(b). The density of points around the edges are greater with a large value of θ (less penality from the initial coordinates). However, a large value of θ leads to a greater level of noise in the grid. On both Fig. 3(a) and Fig. 3(b), it may be easily seen that the curve of all the elements of the face are smooth and comparable to the original image. The third set of simulations presents the results of the edge detector. The values for the threshold σ are chosen equals to 300 and 500 and θ equals to 1 and 0.002. It may be seen that a small value for the threshold leads to an accurate detection of both strong and weak edges. However, the edges are large in dimensions and the level of noise seems high. On the other hand, a large value for the threshold leads to an accurate detection of the edges while removing a large part of the noise. The level of noise in the edges is decreasing with θ. As a conclusion, the edge detection process is more sensitive to the choice of the threshold than the choice of θ. Most of the edges are detected accurately. The proposed method outperforms the Canny edge detection on particular aspects. For example, the proposed edge detector is able to detect the eyelashes, which is not the case of the Canny method. In this method, the smoothing of the image prevents the detector from being able to detect edges which are spatially close. Another improvement may be found on the spatial properties of the edges. The smoothness is preserved and their locations are accurate.
6
Conclusions
A new framework to detect edges in complex grayscale images in presented and implemented is the present paper using a grid smoothing approach. The results are encouraging in terms of edge properties (smoothness), localisation and low number of false edges detected. The main challenge of the method presented is the choice of the parameter θ for the grid smoothing and the threshold for the edge detection. These two parameters should be chosen according to the desired application. Further improvements of the method in terms of computing speed will be researched, focusing mainly on the properties of the various sparse matrices used. Finally, the grid smoothing approach will be combined with the graph-based mesh smoothing method presented in [6] to detect edges in noisy complex grayscale images.
Edge Detection in Grayscale Images Using Grid Smoothing
119
References 1. Ziou, D., Tabbone, S.: Edge detection techniques: An overview. International Journal of Pattern Recognition and Image Analysis 8(4), 537–559 (1998) 2. Canny, J.: A computational approach to edge detections. IEEE Trans. Pattern Anal. Mach. Intell. 8(6), 679–698 (1986) 3. Liyakathunisa, Kumar, C.N.R., Ananthashayana, V.K.: Super Resolution Reconstruction of Compressed Low Resolution Images Using Wavelet Lifting Schemes. In: Second International Conference on Computer and Electrical Engineering, ICCEE 2009, vol. 2, pp. 629–633 (2009) 4. Caramelo, F.J., Almeida, G., Mendes, L., Ferreira, N.C.: Study of an iterative super-resolution algorithm and its feasibility in high-resolution animal imaging with low-resolution SPECT cameras. In: Nuclear Science Symposium Conference Record, 2007, NSS 2007, October 26-November 3, vol. 6, pp. 4452–4456. IEEE, Los Alamitos (2007) 5. Toyran, M., Kayran, A.H.: Super resolution image reconstruction from low resolution aliased images. In: IEEE 16th Signal Processing, Communication and Applications Conference, SIU 2008, April 20-22, pp.1–5 (2008) 6. Hamam, Y., Couprie, M.: An Optimisation-Based Approach to Mesh Smoothing: Reformulation and Extensions. In: Torsello, A., Escolano, F., Brun, L. (eds.) GbRPR 2009. LNCS, vol. 5534, pp. 31–41. Springer, Heidelberg (2009) 7. Hai, J., Xiaomei, Y., Jianming, G., Zhenyu, G.: Automatic eddy extraction from SST imagery using artificial neural network. In: Proceedings of the International Archives of the Photogrammetry, Remote Sensing and Spatial Information Science, Beijing (2008) 8. Guindos-Rojas, F., Canton-Garbin, M., Torres-Arriaza, J.A., Peralta-Lopez, M., Piedra Fernandez, J.A., Molina-Martinez, A.: Automatic Recognition of Ocean Structures from Satellite Images by Means of Neural Nets and Expert Systems. In: Proceedings of ESA-EUSC 2004 - Theory and Applications of KnowledgeDriven Image Information Mining with Focus on Earth Observation (ESA SP-553), Madrid, Spain, March 17-18 (2004) 9. Belkin, I.M., O’reilly, J.E.: An algorithm for oceanic front detection in chlorophyll and SST satellite imagery. Journal of Marine Systems 78(3), 319–326 (2009) 10. Lim Jae S., Two-Dimensional Signal and Image Processing, Englewood Cliffs, NJ, Prentice Hall, 1990, p. 548, equations 9.44 – 9.46. 11. Feijun, J., Shi, B.E.: The memristive grid outperforms the resistive grid for edge preserving smoothing. In: Circuit Theory and Design, ECCTD 2009, pp. 181–184 (2009) 12. Shuhui, B., Shiina, T., Yamakawa, M., Takizawa, H.: Adaptive dynamic grid interpolation: A robust. In: Ultrasonics Symposium on High-Performance Displacement Smoothing Filter for Myocardial Strain Imaging, IUS 2008, November 2-5, pp. 753– 756. IEEE, Los Alamitos (2008) 13. Huang, C.L., Hsu, C.Y.: A new motion compensation method for image sequence coding using hierarchical grid interpolation. IEEE Transactions on Circuits and Systems for Video Technology 4(1), 42–52 (1994)
Energy-Based Re-transmission Algorithm of the Leader Node’s Neighbor Node for Reliable Transmission in the PEGASIS Se-Jung Lim, A.K. Bashir, So-Yeon Rhee, and Myong-Soon Park* Department of Computer and Radio Communications Engineering Korea University, Seoul, Korea {limsejung,ali,syrhee,myongsp}@korea.ac.kr
Abstract. The one among well-known protocols for the data gathering in wireless sensor network is the Power Efficient-Gathering in Sensor Information System (PEGASIS) [1]. In the PEGASIS, the base station (BS) is far away from sensor network. All the sensor nodes form a single chain using greedy algorithm. The leader node is randomly selected among the sensor nodes and transmits aggregated data to the BS. In order to gather data, the leader node sends a small token to the end nodes along the chain, and then non-leader nodes on the chain send the sensing data along the chain toward the leader node from the end nodes. In the PEGASIS, nonleader node's transmission distance decreased. However, still the leader node's transmission distance has not decreased since the BS is far away from the network field. Therefore, the leader node consumes more energy for data transmission than non-leader nodes. If the leader node's energy exhaustion has arisen, the leader node's data transmission is unreliable. As well as, the sensor nodes have to form a rechaining for the next round. In order to resolve this problem in this paper, we proposed energy based re-transmission of the leader node’s neighbor nodes in the PEGASIS. Through our approach, we have proved more high reliable transmission on the same simulation condition. Also, our approach decreases the number of rechaining during the whole round. Keywords: WSN, PEGASIS, Reliable Transmission.
1 Introduction In wireless sensor networks, the sensor nodes are deployed to gather the necessary information from the network field. The sensor nodes have a limited battery power, limited bandwidth, transmission range, computation capability and memory capacity and the BS commonly is far away from network field [2]. The sensor nodes have to transmit sensing data to the BS. To transmit sensing data, if all the sensor nodes have to directly communicate to the BS, this is impracticable in terms of sensor node’s limited battery and bandwidth because the sensor nodes battery exhaustion will occur rapidly and channel interference as the long distance transmission [3]. Therefore, it is necessary for protocols that node’s lifetime increase [4] and reduce bandwidth consumption [5]. *
Energy-Based Re-transmission Algorithm of the Leader Node’s Neighbor Node
121
In the PEGASIS, the BS is fixed and is far away from the network field. The sensor nodes with same capability randomly deploy in the network field and form a single chain for data transmission using the greedy algorithm from the node farthest to the BS. The BS randomly selects the leader node among the sensor nodes for every round and the selected leader node transmits aggregated data to the BS. To gather data, the leader node sends a small token to the end nodes along the chain and then non-leader nodes send own sensing data by multi-hop along the chain toward the leader node from the end nodes. The sensor nodes fuse data at every node on the chain with the exception of the end nodes. Each node fuses own sensing data with the received data from neighbor node on the chain, and then transmits to the other neighbor node on the chain. In the above process, if node’s fault has arisen, the sensor nodes form re-chaining with the exception of fault node. In the PEGASIS, non-leader node's transmission distance decreased. However, still the leader node's transmission distance has not decreased since the BS is far away from the network field. Therefore, the leader node consumes more energy for data transmission than non-leader nodes. Besides, the leader node’s sufficient energy can’t guarantee as the leader node selection without consideration for energy. Hence, the leader node's fault ratio as energy exhaustion is high. If the leader node's fault as energy exhaustion has arisen, the leader node's data transmission is unreliable. In other words the leader node's fault in terms of reliable transmission is critical issue because only the leader node can transmit aggregated data to the BS. As well as, the sensor nodes have to form a rechaining for the next round. And thus, the leader node’s great importance was researched [6-8]. In order to resolve this problem, we propose energy based re-transmission of the leader node’s neighbor nodes algorithm which is an improvement in the PEGASIS. The leader node’s neighbor nodes are relatively more short distance than the BS. So, in case the BS can’t receive aggregated data from the leader node, the leader node’s neighbor nodes can receive aggregated data from the leader node, and then the leader node’s neighbor node can transmit aggregated data to the BS. Moreover, our approach can transmit aggregated data to the BS for the next round without forming a rechaining of the sensor nodes. Through our simulation, we have proved more high reliable transmission on the same simulation condition. Also, we make decrease the number of re-chaining during the whole round. The rest of this paper is organized as follow. In the second section, we review the PEGASIS and we explain our approach in the third section. The fourth section shows the simulation results, and then final section is conclusion.
2 Related Work 2.1 PEGASIS Protocol In the PEGASIS, the BS is far away from network field. The sensor nodes have the uniform energy and the same capability, but the sensor nodes are not mobile. Also, the sensor nodes with global ID for identification randomly deploy in the network field. The sensor nodes form a single chain using the greedy algorithm starting from the node farthest to the BS like Figure 1.
122
S.-J. Lim et al.
Fig. 1. Example of PEGASIS’s Chain Form
After a single chain is formed, the BS broadcast the chain’s information and randomly selected leader node among the sensor nodes. The leader node transmits aggregated data in the network field to the BS. To aggregate, the leader node sends a small token to the end nodes along the chain, and then non-leader nodes send own sensing data by multi-hop along the chain toward the leader node from the end nodes like Figure 2. The sensor nodes fuse data at every node on the chain with the exception of the end nodes. Each node fuses own sensing data with the received data from neighbor node on the chain, and then transmits to the other neighbor node on the chain. If node’s fault has arisen, the sensor nodes form re-chaining with the exception of fault node.
Fig. 2. PEGASIS’s Data Gathering Approach
The BS is far away from network field and the leader node transmits aggregated data to the BS like Figure 2, therefore, the leader node consumes more energy than the non-leader nodes. Besides, the leader node’s sufficient energy doesn't guarantee because the leader node is randomly selected. Hence, the leader node's fault ratio as energy exhaustion is high. If the leader node's fault has arisen, the leader node's data
Energy-Based Re-transmission Algorithm of the Leader Node’s Neighbor Node
123
transmission is unreliable. We will explain in the next section how using energy based re-transmission of the leader node’s neighbor node for reliable transmission.
3 Energy Based Re-transmission of the Leader Node’s Neighbor Node for Reliable Transmission In the PEGASIS, the leader node's fault in terms of reliable transmission is critical issue because only the leader node can transmit aggregated data to the BS. We propose energy based re-transmission algorithm of the leader node’s neighbor nodes for reliable transmission in case the leader node's fault has arisen. In our approach, the BS calculates receiving time of aggregated data per round from the leader node. If the BS does not receive aggregated data from the leader node within the calculated receiving time, the BS broadcast an error message in the whole network. When the leader node’s neighbor nodes receive the error message from the BS, the leader node’s neighbor nodes make a direct connection for energy comparison between the two. A node with more energy among the two nodes allocates leader node’s role, and then the node can transmit aggregated data to the BS. Our approach is as follows. 3.1 The Leader Node’s Aggregated Data Transmission and the Fault Occurrence Non-leader nodes send own sensing data by hop-to-hop along the chain toward the leader node from the end nodes. The leader node transmits aggregated data by this way to the BS on the far from the network field. In here, the leader node's data transmission has some problems. The leader node’s sufficient energy can’t guarantee as the leader node selection without consideration for energy. Also, the leader node needs more energy than non-leader nodes for aggregated data transmission to the BS. If the leader node doesn’t have sufficient energy for data transmission to the BS, the BS can’t receive aggregated data from the leader node. In this case, the located leader node’s neighbor nodes within relatively more short distance than the BS can receive aggregated data from the leader node like Figure 3. After the leader node’s neighbor nodes receive aggregated data from the leader node, the leader node’s neighbor nodes wait until receive error message from the BS like Figure 4. 3.2 Energy Based Re-transmission of the Leader Node’s Neighbor Node The sensor nodes receive the chain information from the BS after a single chain formed. And thus, the sensor nodes know the other sensor node’s information on the chain. Therefore, it is possible to make a direct connection for energy comparison between the two nodes with the exception of leader node like Figure 3 after the leader node’s neighbor nodes receive error message from the BS. If node C2 and node C4 complete directly connection, one node that is located more near to the BS among the two nodes sends own energy to the other the leader node's close neighbor node. This meaning is that since node C2 is located more near to the BS like in Figure 1, node C2 sends own energy to node C4.
124
S.-J. Lim et al.
Fig. 3. Re-Transmission of the Leader Node’s Neighbor Node
As shown in Figure 4, if node C4 received energy of node C2 from node C2, node C4 compares the own energy and the received energy from node C2. As a comparison, if node C4 has more energy than node C2, node C4 transmit aggregated data to the BS, otherwise node C4 re-sends own energy to node C2. When node C2 receive energy of node C4, node C2 has to transmits aggregated data to the BS. In this case, node C2 does not need to compare energy between node C2 and node C4. If all above process is successful, the BS randomly select new leader node for next round without forming a re-chaining of the sensor nodes.
Fig. 4. Re-Transmission Sequence for Reliable Transmission
In the next section, we will discuss measurement result of energy based retransmission of the leader node’s neighbor node.
Energy-Based Re-transmission Algorithm of the Leader Node’s Neighbor Node
125
4 Simulation 4.1 Radio Model We use radio model [9] for performance comparison of the PEGASIS and our approach. Before this model use, we need to define about some parameters, so defined in the table1. Table 1. Parameters Parameter
Value
Parameter
Value
Eelec
50nJ
Energy
0.25J
100pJ
k
2000bit
∈
amp
In the table 1, Eelec is the transmitter or receiver circuitry and amp is the transmitter amplifier. Energy is allocated initial energy and radios have power control. k is bit message and d is distance. In order words, source and destination distance is d and two-way packet size is k-bit. In the table 2, we explain the transmitting radio model and the receiving radio model. Table 2. Radio Model Radio Model Transmitting Receiving
4.2 Simulation Environment Our simulation environments are as follows. The network range is a 100m * 100m field and the BS is located at (50, 300) also fixed. The number of sensor node in the network is 100 the nodes randomly deployed and the sensor nodes are fixed and initial energy of each node is 0.25J. The size of the data packet is 2000bits and we assume that the size of data packets is the same size during hop-to-hop data aggregation. We didn’t consider delay time and token passing energy in this simulation and our simulation is continued until all the sensor nodes have fault. 4.3 Simulation Results In this section, we use abbreviations our approach called "EBRT" for readability in the graphs. First, we measured the distribution of fault node during the whole round. The leader node's fault is shown high on the graph both of the PEGASIS and our approach as shown in Figure 5.
126
S.-J. Lim et al.
(a) Fault Node Distribution in PEGASIS
(b) Fault Node Distribution in EBRT Fig. 5. Fault Node Distribution
Fig. 6. Point of Re-Chaining
Energy-Based Re-transmission Algorithm of the Leader Node’s Neighbor Node
127
Second, we compared the point of re-chaining in both the PEGASIS and our approach. As shown in the Figure 6, re-chaining is evenly distributed during the whole round in the PEGASIS. But, our approach is distributed from the middle point of the graph. In terms of Numerical analysis, our approach decreased the number of re-chaining by over 50 percent during the whole round. Third, we measured retransmission ratio and re-transmission distribution of the leader node's close neighbor node during the whole round like in Figure 7, 8. Our approach obtained reliable retransmission around 80 percent of number of the leader node fault on the same simulation condition like in Figure 7.
Fig. 6. Number of Fault Node
Fig. 7. Successful Re-Transmission Distribution of Leader Node Fault in EBRT
5 Conclusions In the PEGASIS, the leader node's fault in terms of reliable transmission is critical issue because only the leader node can transmit aggregated data to the BS. As well as, the sensor nodes have to form a re-chaining for the next round in the leader node's fault case. To resolve this problem, we proposed energy based re-transmission algorithm of the leader node’s neighbor nodes which is an improvement in the
128
S.-J. Lim et al.
PEGASIS. Through simulation, we proved reliable re-transmission around 80 percent of number of the leader node fault during the whole round on the same simulation condition. Also, our approach decreased the number of re-chaining by over 50 percent.
Acknowledgment Professor Myong-Soon Park is the corresponding authors. We are thankful for his supervision to this research.
References 1. Lindsey, S., Raghavendra, C.S.: Power-Efficient Gathering in Sensor Information Systems, vol. 3, pp. 3-1125–3-1130. IEEE, Los Alamitos (2002) 2. Akyildiz, I.F., Su, W., Sankarasubramaniam, Y., Cayirci, E.: A Survey on Sensor Networks. IEEE Communicatoins Magazine 40(8), 102–114 (2002) 3. Shukla, I., Meghanathan, N.: Impact of Leader Selection Strategies on the PEGASIS Data Gathering Protocol for Wireless Sensor Networks. Accepted for publication in Ubiquitous Computing and Communication Journal 4. Singh, S., Woo, M., Raghavendra, C.S.: Power-Aware Routing in Mobile Ad Hoc Networks. In: Proceedings ACM/IEEE Mobicom 1998 (1998) 5. Zorzi, M., Rao, R.R.: Energy Management in Wireless Communications. In: Proceedings 6th WINLAB Workshop on Third Generation Wireless Information Networks (March 1997) 6. Shah, M.J., Flikkema, P.G.: Power-based leader selection in ad-hoc wireless networks. In: IEEE International on Performance, Computing and Communications Conference, IPCCC 1999, pp. 134–139 (February 1999) 7. Lee, J.-E., Kim, K.: Diamond-Shaped Routing Method for Reliable Data Transmission in Wireless Sensor Networks. In: Proceedings of the IEEE International Symposium on Parallel and Distributed Processing with Applications, pp. 799–801 (December 2008) 8. Rahman, M.M., Abdullah-Al-Wadud, M., Chae, O.: Performance analysis of Leader Election Algorithms in Mobile Ad hoc Networks. International Journal of Computer Science and Network Security 8(2), 257–263 (2008) 9. Heinzelman, W., Chandrakasan, A., Balakrishnan, H.: Energy-Efficient Communication Protocol for Wireless Microsensor Networks. In: Proceedings of the Hawaii Conference on System Sciences (January 2000)
Grid-Based and Outlier Detection-Based Data Clustering and Classification Kyu Cheol Cho and Jong Sik Lee School of Computer Science and Engineering Inha University #253, YongHyun-Dong, Nam-Ku, Incheon 402-751, South Korea [email protected], [email protected]
Abstract. Grid computing has been noticed as an issue to solve complex problems of large-scale bioinformatics applications and helps to improve data accuracy and processing speed on multiple computation platforms. Outlier detection helps classification success rate high and makes processing time reduce. This paper focuses on a data clustering and classification method with outlier detection which is an important bioinformatics application in grid environment. This paper proposes a grid-based and outlier detection-based clustering and classification(GODDCC) using grid computational resources with geographically distributed bioinformatics data sets. This GODDCC is able to operate large-scale bioinformatics applications in guaranteeing high bio-data accuracy with reasonable grid resources. This paper evaluates performance of GODDCC in comparing to the data clustering and classification(DCC) without outlier detection. The average of processing time of the GODDCC model records the lowest processing time and provides the highest resources utilization than the other DCC models. The outlier detection method reduces processing time for DCC models with maintaining high classification success rate and grid computing gives a great promise of high performance processing with geographically distributed and large-scale bio-data sets in bioinformatics applications.
transduction is a basic biological phenomenon in physiological functions. Especially, the Adaptive Resonance Theory 1(ART1)[3] as one of neural networks, solves nerve network problems, makes learned clusters with learning dataset, and selects a cluster for a new classification category. The ART1 ascertains functions and structures of protein in constituting new data families. Bioinformatics is growing to an innovative research field of science, which is applied to high performance systems using grid resources. Bioinformatics demands and encompasses high performance computing for performance evaluation. A grid computing for bioinformatics solves processing problems of large-scale bioinformatics data and improves the performance of data accuracy and processing time. This paper is organized as follows: Section 2 describes the grid computing and bioinformatics and HLA middleware in grid computing. Section 3 illustrates outlier detection, effects and processing flow of the outlier detection and illustrates a grid-based and outlier detection-based data clustering and classification(GODDCC). In section 4, the design of the GODDCC from this paper is demonstrated, by the analysis of experiment results. Finally, conclusions are in section 5.
2 Related Work 2.1 Grid Computing and Bioinformatics Currently, bioinformatics demands and encompasses a high performance computing for performance evaluation and data clustering and classification with increasing of bio-data processing. These demands focused on expanding a high performance computing, that is grid computing. A union of bioinformatics and grid computing is used to bioinformatics application such as bio-data analysis. The open bioinformatics grid(OBIGrid)[4] is mapped out initiative bioinformatics for parallel bioinformatics processing of security policy issues and network transparency. And in grid environment, the BioGrid[5] counts bio-data through computer resources. The BioGrid model is based on interface about statistical information and processing result. A tree structure of bio-data uses meta-data and manages raw bio-data and processing results for management. The grid computing can be used by bioinformatics implementation. For instance, UK BioGrid is an e-science grid project that is called MyGrid[6] for automatic job flow. North carolina BioGrid supplies data store and network for understanding genome variation, that is called NC Grid[7]. And, ClustalG[8] is applied to singapore BioGrid for implementing a sequence alignment and the other bioGrid projects are Asia-Pacific BioGrid[9], EuroGrid[10] and so forth. 2.2 HLA Middleware in Grid Computing High Level Architecture (HLA)[11] system implements grid simulation for merging geographically distributed federates with various services and uses to handling communication of simulation and visualization components. And the HLA run effectively large-scaled distributed simulation that studied dynamic discovery of HLA federates with complementary solutions[12]. HLA is come in handy for time management, useful for time-driven and event-driven interactive simulations and is of use to facilitate migration to dynamic load balancing of grid resources[13]. HLA helps data
Grid-Based and Outlier Detection-Based Data Clustering and Classification
131
distribution management and supports to build collaborative environments consisting of simulations in geographically distributed systems. The HLA simulation runs dynamic joining and resigning of simulation with flexible subscription and notification mechanisms between ownership of resources[14]. The G-HLAM system is used in the HLA-Speaking Service as the local services interfacing federates[15] and supports the Open Grid Services Infrastructure(OGSA) to allows the collaborative environment[16]. HLA–based FederationGrid(Fed-Grid)[17] supports real-time collaborative simulation for a scalable grid computing.
3 Outlier Detection-Based Data Clustering and Classification (ODDCC) in Grid Computing This section explains that an outlier detection method which is applied to the data clustering and classification. And this section looks about effects and the processing flow of ODDCC. And this section explains principal DCC models for the GODDCC’s performance comparisons with progress course of DCC models. 3.1 Outlier Detection in the DCC In this paper, we apply an outlier detection method[18] in the data clustering and classification (DCC). The outlier detection is an interesting data mining research filed. And many DCC algorithms try to minimize the influence of outliers. Most of data group, data sets may contain data objects that do not comply with the general behavior of the data sets likes noises, exceptions or external domain data. These data objects call outliers. The outlier is usually not defined rigorously. Most data mining methods discard these data and readjust data domain. Outlier methods may be detected using statistical tests that assume a distribution or classification model for the data. The DCC procedures need a data clustering process with binding each similar data and a classification process which divides data with comparing to clusters. In DCC processing, when classification machines are generated represent clusters, every data is placed to the most similar cluster among the represent clusters. The outlier data can be filtered to a cluster. This process improves to characters induction of the clusters. For example, G-Protein Coupled Receptor (GPCR)[19] data managed by clusters and the management system has two cluster classes. These data have been revealed in current genome research and have been noticed in demand of classifier for new protein classification prediction. If classA have outliers, after all classB data positioned in classA clusters, the character of classA largely change by outliers (classB data) and move the center of clusters weight. So status of the cluster get bad and the cluster become invalidity. For preventing cluster characters from changing, the cluster needs to manage and monitor with outlier detection. Figure 1 shows the outlier detection in the DCC and data set is divided two classes. And two classes are to say classA and the others class. In the figure, a spot is mapped a GPCR data and data are grouped by clusters. A plus symbol(+) means centers of clusters and lies to the distance of a classification threshold value. Some clusters are merged by two standard classes but few clusters are failed to grouping. These failed clusters and the belonged data of the clusters are detected to outliers.
132
K.C. Cho and J.S. Lee
Fig. 1. Outlier detection in DCC
The outliers are filtered and extracted from learning data. Because the wrong data filtered, this method makes the classification success rate high. And because the number of compared clusters is decreased, the method makes the process time reduce. Because cluster methods induct various recognition results and outliers, this classification shows various results with outlier detection in the comparison between classes or sub classes. 3.2 Processing Flow of Outlier Detection-Based DCC(ODDCC) The ODDCC method guarantees the maximum classification success rate and searches the minimum DCC processing time. The ODDCC method adjusts cluster information, DCC processing times, and classification success rates through controlling a threshold value. Figure 2 shows a state transition diagram of the ODDCC which procedure guarantees the maximum classification success rate and uses binary pattern that is a half value of a previous value. The algorithm within the ODDCC chooses the first input as a representative pattern of the first cluster. After, if an input comes into a layer, the input is compared with the first representative patterns. If distance of the first representative cluster is shorter than a set threshold value, the input is classified into the first cluster. If not so, the input generates a new cluster. This process applies to all inputs. Therefore, the number of clusters is gradually increased as time elapses. The ODDCC generates different results according to the distance measurement methods between inputs and a representative pattern of a cluster. In the clustering, the method generates different result according to distance measurement method between input and representative pattern of clusters. A vigilance threshold decides discord tolerance between input pattern and stored pattern. If a vigilance threshold is large, the clustering is a little difference between input pattern and expectation pattern and classifies new category. But, if a threshold value is small, the clustering is a great difference between input pattern and expectation pattern and roughly divides input patterns. The makes processing time and classification success rate either increase or reduce through adjustment of a threshold value.
Grid-Based and Outlier Detection-Based Data Clustering and Classification
133
Fig. 2. State transition diagram of the ODDDC
3.3 Performance Comparison of DCC Models In this paper, we compare classification time of the outlier detection-based data clustering and classification(ODDCC) to backpropagation[20] and SVM[21] for analysis of classification recognition capacities. The backpropagation is a supervised learning algorithm uses for a training artificial neural networks(ANN). It is most useful for feed-forward networks. And the SVM solving pattern classification prediction problems with statistical study and the SVM is known to an optimal solution of binary classification problems. The figure 3 presents our research works for the DCC models in bioinformatics. A classification success rate of the DCC is high, but the model needs to search an optimal threshold value for an optimal result induction. To solve this problem, the optimal-based data clustering and classification(ODCC)[22] is developed. Thus the
Fig. 3. Progress course of DCC models
134
K.C. Cho and J.S. Lee Table 1. Strengths and weaknesses of DCC methods
classification model
Single platformbased Model
Gridbased Model
Strengths
Weaknesses ▪ Requires for much time and costs to find the optimal threshold ▪ Hard to process a large-scale data
Data Clustering and Classification (DCC)
▪ Simple and useful when the optimal decision range is known
Optimal-based Data Clustering and Classification (ODCC)
▪ Readjusts a optimal threshold value ▪ Guarantees a reliable recognition success rate
▪ Hard to process a large-scale data
Grid-based Data Clustering and Classification (GDCC)
▪ Enables to process a large-scale data ▪ Guarantees an extension ▪ Static job load allocation ▪ Useful when computing capacities of resources are almost equal
▪ Waiting time of resources is high ▪ Waiting time to get all result is high ▪ processing time and resource utilization is low
Grid-based and Outlier Detection-based Data Clustering and Classification (GODDCC)
▪ Enables to process a large-scale data ▪ Static load dispersion with multi-depth job allocations ▪ Reduce processing times with outlier detection
▪ Waiting time of resource is high ▪ processing time and resource utilization is low ▪ Requires additional time for filtering outlier clusters ▪ Agents for communication are need.
ODCC helps to induct an optimal threshold value and assures the optimal classification recognition rate. But these two models operate in a single platform environment, the model does not assure expand for large-scaled data. Thus we develops the gridbased data clustering and classification(GDCC)[23] in a grid environment. The outlier detection model reduces the number of comparing clusters and processing times. The strengths and weaknesses of all DCC model presents table 1.
4 Experiment and Result This section designs the GODDCC model and evaluates that the GODDCC model is useful for the bio-data classification with providing high resource utilization and rapidly computing abilities. And we compare classification processing times of the GODDCC model to that of the gird-based data clustering and classification(GDCC) model.
Grid-Based and Outlier Detection-Based Data Clustering and Classification
135
4.1 Implementation of GODDCC We implement a simulation model of the GODDCC. To simulate the GODDCC, we develop a grid test-bed using the HLA(High-Level Architecture)[11] middleware specification and RTI(Run-Time Infrastructure)[24] implementation. As figure 4 illustrates, the inter-federate communication works on the GODDCC federation. The federation includes one rate readjust federate, three agents and twelve training federates. The RTI message passing for data management among federates depends on the inter-federate communication inside the federation. In the platform setting, we develop the grid system using RTI implementation which operates on windows operating systems. The total of sixteen federates are allocated to four machines, respectively, and they are connected via a 10 Base T Ethernet network. And we use classification data-set which is used GPCR[19] data which is one of the receptor-ligand interactions in signal transduction and which takes an important part at recent biology. Data controller agents and data training machines depend on the inter-component communication inside components.
Fig. 4. GODDCC federation
4.2 Experiment 1: Average of DCC’s Processing Time (SVM vs. Backpropagation Neural Network vs. DCC Using ART1 vs. ODDCC Using ART1) The experiment 1 demonstrates the average of DCC time of the backpropagation neural network, the SVM, the DCC models using ART1 and the ODDCC model using ART1. This experiment is available to present effects the outlier detection model with comparing to the average of DCC time per data unit. And we expect improved processing abilities to reduce DCC times. The figure 5 shows variations of the average of DCC times by adjusting iteration numbers and clustering rates. The average of DCC time is summation times of learning times and test times per the iteration numbers of data. The average of DCC times are decreased when clustering rates and iteration numbers are increased. And, the average of processing times of ODDCC model using ART1 are lower than that of the DCC model using ART1, the SVM and the backpropagation neural network in the whole iteration numbers and clustering rates. This result demonstrates that the ODDCC model using ART1 is speedier and has powerful computing abilities than the other three classifications.
K.C. Cho and J.S. Lee
Average of Processing Time (sec(s))
136
0.0075
SVM BackPropagation Neural Network DCC using ART1 ODDCC using ART1
0.006 0.0045 0.003 0.0015 0
300 3%
500 5%
1000 10%
1500 15%
3000 30%
5000 10000 Iteration Number 50% 100% Clustering Rate
Fig. 5. Average of classification processing times (SVM vs. Backpropagation Neural Network vs. DCC using ART1 vs. ODDCC using ART1)
4.3 Experiment 2: Average Processing Time of Data Clustering and Classification (DCC) Models The experiment 2 compares the average of processing time of the data clustering and classification (DCC) models with changing of iteration and clustering rate. We expect improved calculation abilities and the outlier detection method in the grid environment. The figure 6 shows average of processing time of the DCC models. The average of processing time of the DCC model is regular. The average of processing time of the DCC model recorded 0.002 seconds or over in all case and the highest time among the models. And single platform models (the DCC and the optimal-based data clustering and classification(ODCC) model) are higher than the gridbased data clustering and classification(GDCC) model and the grid-based and outlier detection-based data clustering and classification(GODDCC) model. Especially the average of processing time of the GODDCC records about 0.001 seconds in all case and the times are the lowest time among the models. These results demonstrate that the GODDCC model improves resource calculation ability.
Average Processing Time(sec(s))
0.004 DCC
ODCC
GDCC
GODDCC
0.003
0.002
0.001
0 300
500
1000
1500
3000
5000
3%
5%
10%
15%
30%
50%
10000 # of It erat ion 100% Clustering Rate
Fig. 6. Average of processing time of DCC models(DCC vs. ODCC vs. GDCC vs. GODDCC)
Grid-Based and Outlier Detection-Based Data Clustering and Classification
137
5 Conclusion Grid technology has been noticed as an issue to solve large-scale bioinformaticsrelated problems and improves data accuracy and processing speed on multiple computation platforms with distributed bio-data sets. In this paper, we suggest the grid-based and outlier detection-based data clustering and classification(GODDCC) method to operate large-scale bioinformatics applications in guaranteeing high biodata accuracy with reasonable grid resources. This paper demonstrates the GODDCC which enables to operate in maintaining high performance on a grid environment and improve accuracy and reduce processing time of DCC with outlier detection. The GODDCC provides high performance DCC processing in saving overall execution time with large-scale bio-data sets separated on geographically distributed computing resources. A grid computing improves data accuracy and processing speed for high performance systems on multiple computation platforms. We implement the GODDCC model and measure system performance of the model. And the result demonstrates the GODDCC model can process large-scaled data in grid computing environment and the model reduce data classification time with the outlier detection method. The outlier detection method reduces processing time for the DCC in maintaining high classification success rate. And, the average of classification time of the GODDCC model shows the lowest processing time than the others DCC models. The GODDCC model provides high performance classification processing in saving overall execution time with large-scale bioinformatics data sets separated on grid computing resources.
References 1. Foster, I., Kesselman, C.: The Grid: Blueprint for a New Computing Infrastructure. Morgan Kaufmann, San Francisco (1998) 2. Rajapakse, J.C., Wong, L., Acharya, R.: Pattern Recognition in Bioinformatics: An Introduction. In: Rajapakse, J.C., Wong, L., Acharya, R. (eds.) PRIB 2006. LNCS (LNBI), vol. 4146, pp. 1–3. Springer, Heidelberg (2006) 3. Carpenter, G.A., Grossberg, S.: Adaptive resonance theory: Stable self-organization of neural recognition codes in response to arbitrary lists of input patterns. In: Proceedings of the 8th Conference of the Cognitive Science Society, Hillsdale, NJ, pp. 45–62 (1988) 4. Fumikazu, K., Hiroyuki, U., Kenji, S., Akihiko, K.: A network design for Open Bioinformatics Grid(GBIGrid). In: Proc. The 3rd Annual Meeting, Chem-Bio Informatics Society, pp. 192–193 (2002) 5. Stevens, R.D., Robinson, A.J., Goble, C.A.: MyGrid: personalised bioinformatics on the information grid. Bioinformatics, 302–304 (2003) 6. http://www.myGrid.org.uk/ 7. http://www.ncbioGrid.org/ 8. Li, K.B.: Clustal W-MPI:ClustalW Analysis Using Distributed and Parallel Computing. Bioinformatics 19, 1585–1586 (2003) 9. http://www.apbionet.org/grid/ 10. http://www.eurogrid.org/ 11. DMSO, HLA RTI-1.3 NG Programmer’s Guide Version 3.2
138
K.C. Cho and J.S. Lee
12. Zong, W., Wang, Y., Cai, W., Turner, S.J.: Grid Services and Service Discovery for HLABased Distributed Simulation. In: 8th IEEE International Workshop on Distributed Simulation and Real-Time Applications, pp. 116–124. IEEE Computer Society, Los Alamitos (2004) 13. Cai, W., Yuan, Z., Low, M.Y.H., Turner, S.J.: Federate migration in HLA-based simulation. Future Generation Computer Systems, 87–95 (2005) 14. Rycerz, K., Bubak, M., Malawski, M., Sloot, P.M.A.: HLA Grid Based Support for Simulation of Vascular Reconstruction. In: Proceedings of the CoreGRID Workshop: Integrated Research in Grid Computing, pp. 165–174 (2005) 15. Rycerz, K., Bubak, M., Malawski, M., Sloot, P.M.A.: A Framework for HLA-Based Interactive Simulation on the Grid. Simulation, 67–76 (2005) 16. Rycerz, K., Bubak, M., Malawski, M., Sloot, P.M.A.: A Grid Service for Management of Multiple HLA Federate Processes. In: Wyrzykowski, R., Dongarra, J., Meyer, N., Waśniewski, J. (eds.) PPAM 2005. LNCS, vol. 3911, pp. 699–706. Springer, Heidelberg (2006) 17. Vuong, S., Cai, X., Li, J., Pramanik, S., Suttles, D., Chen, R.: FedGrid: An HLA approach to federating grids. In: Bubak, M., van Albada, G.D., Sloot, P.M.A., Dongarra, J. (eds.) ICCS 2004. LNCS, vol. 3038, pp. 889–896. Springer, Heidelberg (2004) 18. Bolton, R., Hand, D.J.: Statistical Fraud Detection: A Review. Statistical Science 17(3), 235–255 (2002) 19. Watson, S., Arkinstall, S.: The G-protein Linked Receptor Facts Book. Academic Press, Burlington (1994) 20. Jefferson, M.F., Narayanan, M.N., Lucas, S.B.: A neural network computer method to model the INR response of individual patients anticoagulated with warfarin. Br. J. Haematol. 89(1), 29 (1995) 21. Weston, J., Watkins, C.: Multi-class support vector machines, Technical Report CSD-TR98-04, Royal Holloway, University of London (1998) 22. Cho, K.C., Park, D.H., Ma, Y.B., Lee, J.S.: Optimal Clustering-based ART1 Classification in Bioinformatics: G-Protein Coupled Receptors Classification. In: Jiao, L., Wang, L., Gao, X.-b., Liu, J., Wu, F. (eds.) ICNC 2006. LNCS, vol. 4221, pp. 588–597. Springer, Heidelberg (2006) 23. Cho, K.C., Park, D.H., Lee, J.S.: Computational Grid-based ART1 Classification for Bioinformatics Applications. In: ICCSA 2006, Glasgow, UK, pp. 131–133 (2006) 24. Kapolka, A.: The Extensible Run-Time Infrastructure (XRTI): An Experimental Implemen-tation of Proposed Improvements to the High Level Architecture. Master’s Thesis, Naval Postgraduate School (2003)
Performance Comparison of PSO-Based CLEAN and EP-Based CLEAN for Scattering Center Extraction In-Sik Choi Department of Electronic Engineering, Hannam University, 133 Ojung-dong, Daeduk-Gu, Daejeon 306-791, Republic of Korea [email protected]
Abstract. Particle swarm optimization (PSO) is a new high-performance optimizer that can be easily implemented. We investigated the performance of PSO-based CLEAN and EP-based CLEAN for extracting target scattering centers. Simulation results using artificial and measured data show that PSObased CLEAN is faster than EP-based CLEAN without degradation of accuracy. Keywords: particle swarm optimization, evolutionary programming, scattering center extraction.
Recently, a particle swarm optimization (PSO) algorithm has been applied to electromagnetic optimization problems [7−9], because the basic algorithm of PSO is very simple and easy to implement. This paper proposes a PSO-based CLEAN that uses PSO instead of EP for the optimization of the defined cost function. Because of its fast convergence behavior, PSO-based CLEAN can save a considerable amount of computation time compared with EP-based CLEAN without any degradation of accuracy. To compare the performance, this paper uses the artificially created data and measured target data in a compact range.
2 EP-Based CLEAN vs. PSO-Based CLEAN EP is motivated by Darwin’s theory of evolution and “survival of the fittest”. Therefore, EP uses the mutation and selection features for evolution. The mutation is performed by adding a Gaussian noise vector. Therefore, EP performance is largely a function of random number generators. EP may require more evaluations than classical methods in solving simple, differentiable, uni-modal functions because no gradient and problem knowledge are utilized. PSO is another optimization technique that offers attractive benefits to the user, including the fact that the algorithm can be easily understood and implemented.
(a) Evolutionary Programming
(b) Particle Swarm Optimization
Fig. 1. Comparison of flowcharts of EP and PSO
Performance Comparison of PSO-Based CLEAN and EP-Based CLEAN
141
Compared to EP, PSO is based on the simulation of the social behavior of bird flocks and fish schools [10]. In PSO, individual particles in a swarm represent potential solutions, which move through the problem search space finding an optimal or good solution. In Fig. 1, we compare the flowcharts of typical EP and PSO routines in detail. In our previous research, we proposed the EP-based CLEAN which estimates the parameters of scattering centers via a step-by-step method [4−6]. EP-based CLEAN uses the concept of the conventional CLEAN algorithm and extracts the parameters using EP instead of FFT. After finding one scattering center on the target, EP-based CLEAN subtracts the components of the estimated scattering center from the phase history (which is called the frequency-domain CLEAN). The data model which is used in EP-based CLEAN is an undamped exponential model similar to the one used in the FFT-based CLEAN. The undamped exponential model in the high-frequency region is as follows [1]: K
E ( f m ) = ∑ ak exp(− j 4πf m Rk / c ), k =1
m = 1, L , M
(1)
where K is the number of scattering centers, M is the number of frequency samplings, Rk is the location of the kth scattering center, ak is the associated amplitude and c is the speed of light. During each EP-based CLEAN iteration, we define the new cost function and find the location and amplitude of the scattering center minimizing the defined cost function. The cost function J k of the kth iteration is defined as follows [4]: M
J k = ∑ E k ( f m ) − ak exp( − j 4πf m Rk / c
2
(2)
m =1
This paper compares the performance of EP and PSO for optimization of this cost function. In the EP and PSO subroutine, each individual or particle vector is composed of the location of the scattering center ( Rk ) and the real and imaginary part of the associated amplitude ( ak ). The EP and PSO subroutines are terminated when the available execution time has passed because we do not know the minimum value of Eq. (2).
3 Simulation Results 3.1 Simple Tests Using Ideal Point Scatterers To compare the performance of EP-based CLEAN and PSO-based CLEAN, this paper first uses artificially created data composed of five ideal point scatterers. Table 1 lists the ranges and amplitudes of these five ideal point scatterers. The fourth scattering center is located within 1 Fourier bin (Fbin) from the third scattering center to demonstrate the high-resolution characteristics of the proposed algorithm.
142
I.-S. Choi
Table 1. Ranges and amplitudes of 5 ideal point scatterers
number 1 2 3 4 5
range [m] 1.0 2.0 2.3 2.335 4.0
amplitude 0.2+j0.3 0.5+j0.5 1 0.5+j0.65 0.3+j0.4
The backscattered fields are generated in the frequency band 8.3−12.11 GHz with 128 point samplings, so that the bandwidth (BW) is 3.81 GHz. The initial number of populations in EP and number of particles in PSO are selected as 80 and the EP and PSO subroutines are terminated upon reaching 200 generations.
(a) EP-based CLEAN
(b) PSO-based CLEAN
Fig. 2. Cost evaluation curve of each method for extraction of 5 scattering centers
Fig. 2 shows the cost evaluation curves of EP-based CLEAN and PSO-based CLEAN for five scattering center extractions. The minimum cost of each method decreases as the number of generations increases. After 200 generations, the minimum cost values of EP-based CLEAN and PSO-based CLEAN are almost the same for each scattering center extraction. However, the computation time of the PSO-based CLEAN is 4.03 s, whereas EP-based CLEAN takes 7.39 s. To quantify the performance of EP-based CLEAN and PSO-based CLEAN, the relative error is defined as follows:
Rerr ≡
rp − rrp rp 2
2
(3)
Performance Comparison of PSO-Based CLEAN and EP-Based CLEAN
143
where US DQG U US represent the original range profile vector and reconstructed range profile vector, respectively and
2
denotes the 2-norm. To compare the relative
error and computation time of both methods, each algorithm was carried out 10 times and the average values were obtained. Table 2. Average relative error ( Rerr ) and computation time, 10 Monte Carlo simulations, K=5
methods quantities Average relative error Average computation time [s]
EP-based CLEAN
PSO-based CLEAN
0.089 7.417
0.064 4.025
Table 2 lists the average relative error and average computation time for 5 scattering center extractions. In this paper, MATLAB was selected as the programming language and a PC with an Intel® Core2 CPU-2.4 GHz was used. The computation time for PSO-based CLEAN was 54% of that of EP-based CLEAN. However, the accuracy of the extracted parameters by EP-based CLEAN and PSO-based CLEAN was almost same (PSO-based CLEAN is slightly better). The computation time is not a serious problem when the analysis is performed as an off-line process. However, in an on-line process, such as in immediate radar target recognition, this is very important. Therefore, PSO-based CLEAN is more applicable for on-line processing where immediate results are required. The reconstructed range profile, obtained from the IFFT (inverse fast Fourier transform) of the reconstructed field via Eq. (1), is shown in Fig. 3. Fig. 4 shows five extracted scattering centers using EP-based CLEAN and PSO-based CLEAN. The extracted scattering centers (o) agree very well with the actual scattering centers (*) for both EP-based CLEAN and PSO-based CLEAN. Furthermore, the third and fourth scattering centers, which are located within 1 Fbin and were not resolved in Figure 3, are also resolved for our methods. These results show that PSO-based CLEAN can be efficiently applied to radar target recognition, because the computation time is faster and the accuracy is higher than EP-based CLEAN. 3.2 Simulation Results Using Measured Data In this section, the performance is compared using data measured at POSTECH (Pohang University of Science and Technology) compact range. The bandwidth and number of sampling points are the same as the artificially created data of the previous section. The target used for the measurement was 1:15 scale model of an F-14, and the aspect angle was 45° with respect to the head.
144
I.-S. Choi
(a) EP-based CLEAN
(b) PSO-based CLEAN
Fig. 3. Reconstructed range profile using extracted parameters for each method, K=5
(a) EP-based CLEAN
(b) PSO-based CLEAN
Fig. 4. Extracted scattering centers using each method, K=5 Table 3. Average relative error ( Rerr ) and computation time, 10 Monte Carlo simulations, K=16
methods quantities Average relative error Average computation time [s]
EP-based CLEAN
PSO-based CLEAN
0.2478 13.880
0.2429 7.127
Table 3 lists the average relative error and computation time of 10 Monte Carlo simulations. PSO-based CLEAN has a better relative error and is faster than EP-based
Performance Comparison of PSO-Based CLEAN and EP-Based CLEAN
(a) EP-based CLEAN
145
(b) PSO-based CLEAN
Fig. 5. Extracted scattering centers using each method for F-14 at 45° , K=16
CLEAN (51% of EP-based CLEAN). If we increase the number of scattering centers (K), the relative error will decrease and the computation time will increase. The 16 extracted scattering centers (o) are shown in Figure 5. The first scattering center is from tip diffraction of the nose. The highest peak (7th along the range) corresponds to the inlet and the cockpit of the plane, because at an angle of 45°, the inlet and the cockpit are located at the same range point. Figure 5 shows that PSO-based CLEAN achieves robust detection of the dominant scattering centers with high resolution for the measured data.
4 Conclusion In this paper, we proposed a PSO-based CLEAN algorithm that can be generally applied for extracting scattering centers from the response of a radar target. The proposed algorithm uses PSO instead of EP for optimization of the defined cost function. In the simulation results, we showed that the computation time can be remarkably reduced while the accuracy is enhanced using PSO-based CLEAN for target scattering center extraction. Therefore, PSO-based CLEAN is a good candidate algorithm for application to on-line processing (where immediate decision is required).
Acknowledgment This work was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology (No. 2010-0016513).
146
I.-S. Choi
References 1. Tsao, J., Steinberg, R.D.: Reduction of sidelobe and speckle artifacts in microwave imaging: the CLEAN technique. IEEE Trans. on Antennas and Propagation 36, 543–556 (1988) 2. Hurst, M.P., Mittra, R.: Scattering center analysis via Prony’s method. IEEE Trans. on Antennas and Propagation 35, 986–988 (1987) 3. Carriere, R., Moses, R.L.: High resolution radar target modeling using a modified Prony estimator. IEEE Trans. on Antennas and Propagation 40, 13–18 (1992) 4. Choi, I.-S., Kim, H.-T.: One-dimensional evolutionary programming-based CLEAN. IEE Electronics Letters 37, 400–401 (2001) 5. Choi, I.-S., Kim, H.-T.: Two-dimensional evolutionary programming-based CLEAN. IEEE Trans. on Aerospace and Electronic Systems 39, 369–373 (2003) 6. Choi, I.-S., Seo, D.-K., Bang, J.-K., Kim, H.-T., Rothwell, E.J.: Radar target recognition using one-dimensional evolutionary programming-based CLEAN. Journal of Electromagnetic Waves and Applications 17, 763–784 (2003) 7. Boeringer, D.W., Werner, D.H.: Particle Swarm Optimization Versus Genetic Algorithms for Phased Array Synthesis. IEEE Trans. on Antennas and Propagation 52, 771–779 (2004) 8. Robinson, J., Sinton, S., Rahmat-Samii, Y.: Particle swarm, genetic algorithm, and their hybrids: Optimization of a profiled corrugated horm antenna. IEEE Antennas Propagation Soc. Int. Symp. Dig. 1, 314–317 (2002) 9. Malik, N.N.N.A., Esa, M., Yusof, S.K.S., Marimuthu, J.: Suppression of antenna’s radiation sidelobes using particle swarm optimization. In: PIERS Proceedings, pp. 18–21 (2009) 10. Kennedy, J., Eberhart, R.: Particle swarm optimization. In: Proc. IEEE Int. Conf. Neural Networks, vol. 4, pp. 1942–1948 (1995)
A Framework of Federated 3rd Party and Personalized IPTV Services Using Network Virtualization Md. Motaharul Islam, Mohammad Mehedi Hassan, and Eui-Nam Huh Department of Computer Engineering, College of Electronics and Information Kyung Hee University, 1 Seocheon-dong, Giheung-gu, Yongin-si, Gyeonggi-do, 446-701, Republic of Korea {motahar,hassan,johnhuh}@khu.ac.kr
Abstract. Increasing demand of 3rd party and personalized Internet Protocol Television (IPTV) services require cost effective strategies for content distribution network and quality of service provisioning. However, current IPTV architecture does not explicitly define any standard components, interfaces and cost effective policies for emerging 3rd party and personalized IPTV services. So in this paper we introduce a framework of federated 3rd party and personalized IPTV services using virtual network (VN). The federated 3rd party services can provide competitive IPTV services which will be cost effective and ensured QoS service provision. Also VN can provide flexible IPTV services with lower cost as compared to existing IP overlay network by sharing same physical network among multiple service providers. Through this framework, present IPTV services can be enhanced in terms of quality of service provisioning of the content delivery and resource utilization. Keywords: 3rd party IPTV service, Network Virtualization, Hosting service, Personalized IPTV content.
infrastructure requires significant investments for the distribution network, in terms of guaranteed bandwidth as well as available storage capacity for hosting service. However, the present internet architecture cannot provide desired services to the end user for its inherent limitations. So current IPTV service providers are forced to use IP overlay network but it also adds complexity and increased cost to service providers [3][11]. Recent works in the field of virtual networks offer a viable alternative that promises to cut costs by sharing the infrastructure among different service providers [4]. The key on network virtualization is of dividing the physical network infrastructure into several slices and associating them to different virtual providers. However, there are no explicit components and interfaces for new 3rd party and personalized IPTV services with network virtualization in the current IPTV Architecture [5][6][10]. So in this paper we propose a framework of federated 3rd party and personalized IPTV services through VN. A number of 3rd parties can join together and may form a consortium (federation) to provide quality oriented cost effective services. VN can offer the possibility to test, debug, and roll out new network architectures for 3rd party and personalized IPTV services. The paper is organized as follows: in section 2, we present the related works. In section 3 we give the details of the proposed framework with service delivery architecture and federated 3rd party and personalized IPTV services architecture and hosting services using NV. Section 4 describes the future research issues. Section 5 finally concludes the paper.
2 Related Works In comparison to traditional analog TV, IPTV provides the digital television services over Internet Protocol (IP) from the content providers to the end users at a cost effective approach. To enrich quality oriented service delivery, the 3rd party and personalized IPTV content providers can contribute a lot. Addition of VN with current IPTV distribution network can greatly enhance the network architecture and service quality. However, no work has been found in the literature regarding the delivery of content from federated 3rd party using VN architecture. There are few approaches proposed in the literature regarding virtualization in IPTV distribution network. In [2], the authors discuss some runtime aspects by examining control interfaces and signaling protocols necessary for the management of VN architecture. They also discuss how an IPTV application service provider may benefit from a network virtualization concept and describe the different roles of player and stakeholders in the VN architecture. However, they do not consider the 3rd party content provider issues in the architecture. In [1] authors discuss about P2P IPTV, in which user serve as peers and participate in video data sharing. This concept can be enriched and be fruitful if the peers form a group and provide the content as a 3rd party. Although IPTV is considered as a killer application of the next generation Internet, its success depends on quality oriented and cost effective contents development. So we propose a framework of federated 3rd party and personalized IPTV services. In this framework we also focus the problem of overlay based IPTV distribution network and give the solution of existing problem by using virtual network approach.
A Framework of Federated 3rd Party and Personalized IPTV Services
149
3 Proposed Framework 3.1 IPTV and Network Virtualization Here we briefly describe the proposed architecture of the federated 3rd party and personalized IPTV services through network virtualization. In our scheme we proposed that 3rd party and personalized content provider will be connected to the virtual network. VN can provide services through its virtual nodes and virtual links. In this framework IPTV content will be provided by the 3rd party. We consider two types scenario. Either the 3rd party and personalized content provider may contribute individually or can have a consortium to provide quality oriented and cost effective services. IPTV service provider can use these content provided by the 3rd party and personalized content provider thorough virtual network. It helps to dynamically share and host the content as required by the service provider and the end user. Now a day in most of the cases, IPTV access network can be of two types: i. IMS (IP Multimedia Subsystem) ii. Non-IMS. In our proposed scheme we considered both IMS and Non-IMS which is included in the virtual network attachment protocol module. Through the proposed scheme it is possible to render the IPTV services to the end user in cost effective approach. In figure 1 federated 3rd party and personalized IPTV scenario has been depicted.
Fig. 1. Network Virtualization for IPTV
150
M.M. Islam, M.M. Hassan, and E.-N. Huh
3.2 Service Delivery Architecture Figure 2 presents the service delivery architecture of 3rd party and personalized IPTV services based on VN. The major components of this service delivery architecture are described as follows: Network virtualization: It is an effective technique for service providers to use substrate network. To provide compelling value-added services, a connection must be made between ASs and two networks: the call control network that provides presence, instance messaging, and other services and a non-(IP Multimedia Subsystem) IMS network such as the Internet that provides web services using hypertext transfer protocol (HTTP). The network virtualization layer provides functions for making seamless connections between ASs and these two networks. AS virtualization: It is effective for achieving provision of value added services. The AS virtualization layer provides intelligence log-in, routing, and access control functions and provides a function for adding ASs independently of the network. Intelligence function lies at the point where information is received from the network [9]. It consists of flow identification and real-time data storage. Access control function performs various control functions such as regulating the number of simultaneous connections based on service level agreement with the subscriber, regulating access when the usage limit has been exceeded in case of prepaid charging, and controlling competition between services. The routing function receives requests from the network and decides which AS to send each request.
Fig. 2. Service delivery architecture of IPTV base on VN
A Framework of Federated 3rd Party and Personalized IPTV Services
151
Service orchestration: This function is needed to provide new value added services quickly and economically. The service orchestration layer consists of service components that can be used in common by multiple applications, basic scenarios that define combinations of components, and execution environments for executing those scenarios. Policy management: Policy management is divided into four parts: system policies, AS policies, hosting policies and subscriber policies. System policy management provides a function for managing policies defined for the system in terms of a service delivery platform. AS policy management provides a function for managing policies defined for each AS. Hosting policy provides a function to host 3rd party content in a virtualized storage environment for the easy access by the end users and prosumers. Subscriber policy management provides a function for managing policies defined for each subscriber. Access Network: There are two types of access networks. i. IMS network. It may consist of home subscriber server (HSS) which is an open source and standard for use in IMS context and serving call session control (S-CSC). ii. Non-IMS Network. It provides basic messaging framework upon which web services can be built up [8]. Hosting Service: It facilitates the federated 3rd party and personalized IPTV content provider to host the services in the particular AS in the VN environment. The AS represents individual substrate physical node that may host multiple virtual node of the different content provider. The detail descriptions of hosting services are explained in article 3.3 and in the figure 4. 3.3 Federated 3rd Party IPTV Using VN We assume that there will be multiple infrastructure providers required to enable communication between different locations and which provide end users with access to their networks. In this case IPTV service provider uses VN to provide application specific services hosted by 3rd party. Usually, the IPTV service provider would specify the desired network topology for the VN operator. The 3rd party IPTV content provider interacts within the VN. Individual or federated 3rd party content providers may access the VN in order to provide the necessary content, e.g., by hosting TV program and video onto streaming servers. If end-users (IPTV customer, prosumers) want to get access to the IPTV VN to which it is subscribed, connectivity should be established by the VN attachment protocol interfaces. It means the end-users node will automatically discover virtual access points of the VN it needs to connect. Figure 3 demonstrates a few 3rd party content providers forming a federation and belongs to a virtual network. In the same way many consortium may co-exist in terms of the member of different virtual networks. Dynamic collaboration can be done among the content providers of the same virtual network.
152
M.M. Islam, M.M. Hassan, and E.-N. Huh
Fig. 3. 3rd party IPTV services using VN
3.4 Federated 3rd Party IPTV Hosting Services Using VN Figure 4 describes basic substrate node architecture to implement federated 3rd party and personalized IPTV hosting services using network virtualization. It shows a substrate physical node with two real network interfaces. The physical substrate node hosts virtual nodes of two different virtual networks: i. Virtual Network No-1(VN #1) and ii. Virtual Network No-2 (VN #2). For addressing and identification purposes from control planes and data planes we need a unique VN-Identifier (VN-ID). It is required for the following purposes: 3rd party and personalized IPTV content provider & End User Attachment: In order to attach to the desired virtual networks from any place, a global unique identifier for virtual networks is most important. It facilitates virtual access of 3rd party content hosting nodes of the corresponding VN by the 3rd party and personalized IPTV content providers & end users. 2. Accounting, Authorization and Authentication: A globally unique identifier eases assignment of resource usage if multiple infrastructure providers are providing resources to a virtual network. 3. Uniqueness across multiple infrastructure providers: Since VN may span different infrastructure providers; the VN-ID also should be globally unique for accounting purposes. One option may be considered here is that a VN Provider generates the VN-ID as a cryptographic ID, e.g., as hash value of a generated public key. This could be used for improving the security: in case a VN Provider wants to modify a VN configuration, infrastructure providers can verify that the VN Provider possesses the corresponding private key that belongs to the aforementioned public key. Furthermore, the VN Provider can supply credentials to the VN Operator and the involved infrastructure providers, so that only authorized access to control functions is possible.
1.
A Framework of Federated 3rd Party and Personalized IPTV Services
153
Fig. 4. Node hosting federated 3rd party IPTV services using VN
There are different important components in the figure 4. These components are discussed below: i.
The physical substrate node control consisting of VNode Control and VLink Control allows for setup and modification of virtual node slices and virtual links and must therefore only be accessible by the infrastructure provider. ii. The (De-) Multiplexing and QoS Mechanisms component is responsible for de-multiplexing of multiple incoming / outgoing virtual links via one substrate link. iii. The Hypervisor / Resource Control are responsible for actual creation of virtual nodes and manage the resources assigned to them. iv. VN Management Access allows VN Operators to access each of their virtual nodes in case of initial setup, misconfigurations, or failures inside the virtual network and permits reboot, serial console access, and further management functionalities. VN Operators are allowed to access their virtual nodes after they have been properly authenticated and authorized. The access to this interface may be proxied by a management node of the infrastructure provider. From a security perspective, this interface is highly critical and requires extremely careful engineering.
4 Future Research Issues Success of IPTV depends on quality oriented content delivery. For this reason researches for 3rd party and personalized content delivery are going on. In this regards important research issues may be dynamic bandwidth allocation to the 3rd party content provider, dynamic collaboration among the 3rd party content provider,
154
M.M. Islam, M.M. Hassan, and E.-N. Huh
security in 3rd parties and virtualization of 3rd party content itself. The other issues mentioned as follows: i. Internetworking interface of 3rd party and personalized IPTV contents with IPTV service provider ii. Interoperable middleware to support innovative content classification and hosting of different 3rd party or personalized IPTV contents source. iii. Contents protection scheme such as digital right management (DRM) including conditional access system (CAS) and downloadable conditional access system (DCAS) to provide appropriate adaptation of 3rd party and personalized IPTV content service. iv. Appropriate policy management for application server flow control and intelligent logging with proper authorization and authentication.
5 Conclusion In this paper we present a framework of integrating federated 3rd party and personalized IPTV services with network virtualization architecture. As the IPTV subscribers are increasing rapidly, the quality oriented contents must be delivered. In this regard, federated 3rd party and personalized IPTV content providers can meet up this future demands. VN can ensure the content hosting services in a cost effective approach that can be easily reachable to the worldwide IPTV subscribers. Here we only provide the NV architecture and hosting service. As the next step to our research in this field, we are planning to design a control interface, hosting policy, feasible middleware prototype over virtual networking platform and schedulability of the resources to the 3rd party and personalized IPTV content providers.
Acknowledgement This research was supported by the MKE (The Ministry of Knowledge Economy), Korea, under the ITRC (Information Technology Research Center) support program supervised by the NIPA (National IT Industry Promotion Agency)" (C1090-10110001)). The corresponding author is Eui-Nam Huh.
References 1. Xiao, Y., Du, X., et al.: Internet Protocol Television (IPTV): The killer Application for the Next-Generation Internet. IEEE Commutations Magazine, 126–134 (2007) 2. Bless, R., Werle, C.: Control Plane Issues in the 4WARD Network Virtualization Architecture. Electronic Communications of the EASST 17 (2009) 3. Yu, M., Yi, Y., et al.: Rethinking virtual network embedding: substrate support for path splitting and migration. SIGCOMM Comput. Communication Rev. 38(2), 17–29 (2008) 4. Jain, R.: I Want My IPTV. IEEE Multimedia 12(3), 95–96 (2005) 5. Taplin, J.: The IPTV revolution, http://www-bcf.usc.edu/~jtaplin/IPTV.pdf 6. Park, S.: Integrated Session Control for Peer-to-Peer IPTV Services. In: International Conference on Convergence and Hybrid Information Technology (2008) 7. Global IPTV: Market Analysis and Forecast to 2011, http://www.researchandmarkets.com/reports/573665
A Framework of Federated 3rd Party and Personalized IPTV Services
155
8. Mikocyz, E. et al.: IMS based IPTV services-Architecture and Implementation. In: Proceedings of the 3rd International Conference on Mobile Multimedia Communications, Greece, August 27-29 (2007) 9. Froloshki, H., Pencheva, E.: Enabling Architecture for Third Party Applications in Intelligent Network. Cybernetics and Information Technologies 7(1) ( 2007) 10. The Evolving IPTV Service Architecture, White paper, Cisco Systems Inc. 11. Virtualization in the Core of the Network, White Paper, http://www.juniper.net/us/en/local/pdf/whitepapers/ 2000299-en.pdf
Faults and Adaptation Policy Modeling Method for Self-adaptive Robots Ingeol Chun1, Jinmyoung Kim2, Haeyoung Lee2, Wontae Kim2, Seungmin Park2 and Eunseok Lee1 1
School of Information & Communication Engineering, SungKyunKwan University, 300 Cheoncheon-dong, Jangan-gu, Suwon, 440-760, Korea [email protected], [email protected] 2 Electronics and Telecommunications Research Institute (ETRI), 138 Gajeongno, Yuseong-gu, Deajon, 305-700, Korea {jm.kim,haelee,wtkim,minpark}@etri.re.kr
Abstract. Owing to the proliferation of robots in the ubiquitous world, it is imperative to develop methods and theories to integrate robots with their operational environment for creating highly reliable systems. However the ubiquitous world where robots are operated has much uncertainty and uncontrollable conditions, so that it is impossible to make robots suitable to all situations. To achieve user satisfaction and overcome abnormal situation of the ubiquitous world, robots must be dependable, safe, scalable and adaptive. Especially robots could be more intelligent in the adaptation to deal with uncertainty and uncontrollable condition. Keywords: Intelligent system, Adaptable software, Self-adaptation, Fault model, Autonomic computing.
Faults and Adaptation Policy Modeling Method for Self-adaptive Robots
157
The first consideration to start our research is to understand the characteristics of the physical world. The physical world has much uncertainty and uncontrollable conditions. But in general development methodology, uncertainty and uncontrollable conditions are not allowed. Moreover, all of the user requirements and the system states are welldefined and described in a requirement specification document. Consequently we do not make systems suitable to all situations in the physical world because it is impossible to define all system status. To solve that problem, the home service robot must be smart and reliable to unexpected conditions and adaptable to abnormal situation. We have been finding a solution from a self-adaption technology. The self-adaptive software reasons about its state and environment, and adapts itself at runtime automatically and dynamically in response to changes. In this paper, we propose faults and adaptation policy modeling method to develop a self-adaptive robot[3].
2 Related Works In 1997, the DARPA Board Agency Announcement (BAA) defined self-adaptive software as follows [4]: “Self-adaptive software evaluates its own behavior and changes behavior when the evaluation indicates that it is not accomplishing what the software is intended to do, or when better functionality or performance is possible.” The primary reason why we need self-adaptive device is the increasing cost of handling the complexity of software systems to achieve their goals. Traditionally, a significant part of the research on handling complexity and achieving quality goals has been focused on software verification and validation (V&V) based on its internal quality attributes. However, in recent years, there has been an increasing demand to handle these problems at runtime. The self-adaptive device aims to adjust various artifacts or attributes in response to changes in the self and in the context of a software system[5]. Researchers in this area have proposed several solutions to incorporate adaptation mechanisms into software systems. In this section, we look into the related researches. Garlan propose an architecture-based adaptation framework called Rainbow[6]. The rainbow framework consists of an adaptation infrastructure and system-specific adaptation knowledge. The adaptation infrastructure implements self-adaptive capabilities to monitor, detect, decide, and act, based on the adaptation knowledge. Tivoli Risk Manager provides an integrated security management structure by filtering and correlating the data from different sources and then applying dynamic policies, such as server reconfiguration, security patch deployment and account revocation[7]. This system manages security incidents from a single security console that centrally manages security incidents and vulnerabilities. Recovery-Oriented Computing (ROC) provides a recovery mechanism that can restore most of the same failures as does a full reboot[8]. The mechanism addresses online verification of the recovery, isolation and redundancy problem. This can be achieved by micro reboots of just the offending modules within the operating system or application instead of re-launching the entire OS. Mukhija and Glinz propose a Contract-based Adaptive Software Architecture (CASA) framework that supports both application-level and low-level (for example, middleware) adaptation actions through an external adaptation engine[9]. CASA provides an integrated approach to include all kinds of service parameters across different application domains within the same framework. Dynamic adaptation on the CASA framework is
158
I. Chun et al.
achieved via runtime reconfiguration of the components of an application, according to the adaptation policy specified in the application contract. J3 presents a modeldriven framework for application-level adaptation based on three modules J2EEML, JAdapt, and JFense for modeling, interpreting and runtime management of selfadaptive J2EE application[10]. J3 addresses the need for component-level autonomic computing. Additionally J3 provides the J3 Toolsuite, an open-source1 MDE environment that supports the design and implementation of EJB autonomic applications. DEAS propose a framework for identifying the objectives, and analyzing alternative ways of how these objectives can be met[11]. The DEAS framework defines the design method for the system that supports all or some of these alternative behaviors using goal models. MADAM proposes modeling support tools and middleware that facilitate adaptive application development for mobile computing by representing architecture models at runtime[12]. The main contributions of this project are a middleware that provides the dynamic adaptation of component-based applications, and a model-driven development methodology that is based on abstract adaptation models and corresponding model-to-code transformations. The Middleware (M-Ware) project in Georgia Tech's College of Computing addresses distributed applications subject to performance constraints when moving and operating on large data volumes[13]. The M-ware provides agility (adapting application components and then, dynamically deploying new components and change component structures), resource-awareness, runtime management and openness in distributed applications. Multi-Level Intrusion Detection System (ML-IDS) detects network attacks by inspecting and analyzing the traffic using an autonomic computing concept to automate control and management. This automation allows ML-IDS to detect network attacks and proactively protect the operating system against them.
3 Faults and Adaptation Policy Modeling Method To specify a system model is the starting point to make self-adaptive software. We have selected ECML (ETRI CPS Modeling Language) as a system modeling language. ECML is developed by ETRI (Electronics and Telecommunications Research Institute) to model a hybrid system that includes both continuous and discrete properties. Because ECML is derived from DEV&DESS, it has same formalism like DEV&DESS[14]. Hybrid systems are described as a set of structure models communicating over a set of shared variables in an asynchronous way in ECML. The structure models may be grouped together in a hierarchical way into composite structure models starting from the most primitive ones called atomic structure models. Information flow inside a composite structure model may be hidden to the outside world. The grouping of structure models into composite structure models gives the architecture of the hybrid system. Atomic structure models may be endowed with a set of parameters that can be instantiated in different ways. Thus an atomic or composite structure model may also be understood as an architectural pattern that may be instantiated that is reused in different contexts that match the pattern. For example, at a lower level, a robot may be understood as the composition of a sensing structure model, a controller structure model, and an actuator structure model. At a higher level, one may consider a team of cooperating robots, communicating with each other in order to achieve a common goal.
Faults and Adaptation Policy Modeling Method for Self-adaptive Robots
159
The behavior of an atomic structure model is given by a set of behavior model that are linked together by a set of transitions. Each behavior model represents a particular behavior of the structure model and an associated dynamics given by a set of algebraic and differential constraints. The dynamics may be further constrained by a set of invariants. Also behavior models are grouped together in a hierarchical way to form composed behavior models starting from the most primitive ones called leaf behavior models. Moreover, each behavior model may declare its own set of local variables that is hidden outside the behavior mode, but is accessible to its sub behavior models. In other words, a behavior is a sequential, communicating, hierarchical state machine with well-defined dynamics and interface to other behaviors. Variables of a hierarchical behavior model have scoping rules similar to scoping in structured programming languages. A behavior model may also have a number of parameters and be regarded as a behavioral pattern that can be instantiated. 3.1 Faults Modeling Method In general, developers describe specific system faults in the system model whereas we define the normal status of the system to detect anomaly. Anomalies are patterns in data that do not conform to a well-defined notion of normal behavior. Anomalies might be induced in the data for a variety of reasons, such as malicious activity, for example credit card fraud, cyber-intrusion, terrorist activity or breakdown of a system, but all of the reasons have a common characteristic that they are interesting to the analyst. The "interestingness" or real life relevance of anomalies is a key feature of anomaly detection. Most of system developers are only interested in the normal phase of system because system can always provide its own service regardless of the system's errors or failures. As mentioned earlier, system developers are not interested in the abnormal phase but the normal phase of system so that the normal status constraints(NSC) is defined to specifying the normal condition in a certain phase. In order to express the normal condition, the extension of ECML is defined. First, the NSC is inserted in a CPS behavior model(CBM) to specify normal condition as shown in Fig. 1. Each state of a behavior model may have the NSC that describe as NSCname[NSCCondition]. The NSC name is a unique name of the NSC and the NSC condition is the condition that describes normal status. CBM VehicleController startMoving [true == running] /
Ready d (engine_torque) = 0
Always_on[engine_power > 0]
Move stopMoving [false == running] /
d (engine_torque) = acceleration Normal_movement [remains > d(engine_torque)]
Fig. 1. ECML example of Behavior Model with Normal Status Constraints
160
I. Chun et al.
Next, the fault monitor is inserted in the structure model to specify system components where faults occur as shown in Fig. 2. Each structure of a CPS structure model(CSM) may have FM that describes a potential fault as FM[FaultyComponent Name]. CSM GroundVehicle [A] Velocity : Vector
Fig. 2. ECML example of Structure Model with Fault Monitor
3.2 Adaptation Policy Modeling Method To apply the adaptation policy modeling method to a self-adaptive robot, 4 steps must be achieved as follows. First of all, customer requirements and functions are gathered because the objective of this robot is to meet customer satisfaction through providing a seamless service. Next we design a system model including the normal status. As customer requirements are the very important factor to make the system model, a developer should divide system components in accordance with requirements, and then refine the system model referring to functions of the robot. Finally the normal status constraint and the normal status condition are specified at the system model. Third, the Normal Status Table(NST), the Fault Monitor Table(FMT) and the Adaptation Policy Table(APT) derived from the system model is made. However the Adaptation Strategy Table(AST) is predefined according to the adaptation capability Table 1. Normal Status Table &DWHJRU\
Faults and Adaptation Policy Modeling Method for Self-adaptive Robots
161
of a target system. The NST consists of category, NSC name, CBM name, CPM name, state name and condition as illustrated in table 1. The CBM name, the CPM name and the state name are used to specify the monitoring target state. That is, if a certain normal condition is confirmed, the SAManager monitors the NSC condition of the target state. For example, in the case that the system want to know that it moves forward correctly, the SAManager monitors the condition “Rotated_angle = 0 OR L-motor_rotation_rate = R-motor rotation rate” that is defined in the state "Forward" of the CPM "Tracking" of the CBM "VehicleController". The FMT consists of category, FM name, CSM name, substructure name, fault description as illustrated in table 2. The CSM name and substructure name are used to specify the target structure where a fault may occur. That is, the FaultManager only monitors a designated structure so as to check the occurrence of a certain fault. For example, in the case that the system want to know the occurrence of the fault "PowerShortage", FaultManager monitors the fault status “SpecifiedBatteryVoltage > CurrentVoltage” that is defined in the substructure "Batterypower Gauge" of the CBM "GroundVehicle". Table 2. Fault Monitor Table )0QDPH
&60QDPH
%URNHQ0RWRU
*URXQG9HKLFOH 5LJKWVLGH0RWRU
6XEVWUXFWXUHQDPH
5HYROXWLRQ3HU0LQXWH WKUHVKROG
%LDVHG5RWDWLRQ
*URXQG9HKLFOH &RPSDVV6HQVRU
JHW&RPSDVV6HQVRU6WDWXV IDLO
3RZHU6KRUWDJH
*URXQG9HKLFOH %DWWHU\SRZHU*DXJH
6SHFLILHG%DWWHU\9ROWDJH!&XUUHQW9ROWDJH
:KHHO%URNHQ
*URXQG9HKLFOH /RFDWLRQ6HQVRU
JHW*36GDWD IDLO
8QNQRZQ2EMHFW *URXQG9HKLFOH 9LVLRQ6HQVRU
)DXOW6WDWXV
LPDJH$QDO\VLV IDLO
The APT is made by system developers using the FMT and the AST because they know system structures and potential faults of the system. The APT consists of policy name, FM name, strategy name and parameters as illustrated in table 3. The APT defines the relationship between a fault and a adaptation strategy. For example, if the potential fault “wheelbroken” occurs, the system must achieve the strategy "SWUpdate" with parameters "SW ID". Table 3. Adaptation Policy Table 3ROLF\1DPH
)01DPH
6WUDWHJ\1DPH
3DUDPHWHUV
6WRS0RYLQJ
%URNHQ0RWRU
.LOO
SURFHVV,'
&KDQJH7RUTXH
%URNHQ0RWRU
&KDQJH9DOXH
^YDULDEOHYDOXH`
&DOLEUDWLRQ&RPSDVV6HQVRU
%LDVHG5RWDWLRQ
&KDQJH9DOXH
^YDULDEOHYDOXH` ^YDULDEOHYDOXH`
0DLQWDLQ%DWWHU\
3RZHU6KRUWDJH
3RZHU'RZQ
$GG$YLRGDQFH2EMHFW/LVW
8QNQRZQ2EMHFW
&KDQJH.QRZOHGJHEDVH ^64/TXHU\`
&KDQJH0LVVLRQ
:KHHO%URNHQ
6:8SGDWH
6:,'
8SJUDGH6:
8QNQRZ2EMHFW
6:8SGDWH
6:,'
162
I. Chun et al.
Finally, the AST is defined during the systems modeling step as illustrated in table 4. The AST consists of strategy name, action and type. In the system modeling step, the AST has been defined according to the adaptation capability of a target system. The AST has two types of the adaptation strategy. Internal type is that the adaptation is executed in the adaptable software. On the other hand, external type is that the adaptation agent executes adaptation processes outside of the adaptable software. Table 4. Adaptation Strategy Table 6WUDWHJ\1DPH
$FWLRQ
7\SH
&KDQJH9DOXH
&KDQJHLQWHUQDOYDULDEOHVRIDGDSWDEOHVRIWZDUH
,QWHUQDO
.LOO
.LOO3URFHVV
([WHUQDO
5HVWDUW
5HVWDUW3URFHVV
([WHUQDO
3RZHU'RZQ
6ORZGRZQRUWXUQRIIXQXVHGHTXLSPHQW
([WHUQDO
6:8SGDWH
8SGDWHFXUUHQW6:WRQHZ6:
([WHUQDO
8SJUDGH6:
8QNQRZ2EMHFW
6:8SGDWH
The adaptable software and the knowledge for adaptation that is generated for above 4 steps deploys to a target system. All tables are translated to XML documents or DB files according to the system type. At a runtime, the adaptation agent generates the fault tree dynamically to analyze the current system status as indicated in Fig. 3. The Fault Tree analysis (FTA) is a failure analysis method in which a faulty state of a system is analyzed using Boolean logic to combine a series of lower-level events. The fault tree composed of the normal status and the faulty condition. The normal status is the user desirable status of the system and the faulty condition is the criteria to judge the status of the system. For example, “Normal_forward” status is judged by the combination of the faulty condition “RevolutionPerMinute < threshold” and ”getCompassSensor Status() = fail”. In the same manner, the faulty condition “RevolutionPerMinute < threshold” is composed of the normal status “Wheel_driving” and “MainBattery_charged”. This fault tree analysis method is generally used in the field of safety engineering to determine the probability of a safety hazard quantitatively.
Fig. 3. Fault Tree for Self-adaptive Robot
Faults and Adaptation Policy Modeling Method for Self-adaptive Robots
163
4 Case Study In this case study, two examples of self-adaptive functions is described. First function is to avoid a sudden-appeared obstacle. This robot is developed to operate in a predefined flat space that has no sudden obstacles and only has a few adaptive functions because of the lack of a memory. It patrols and monitors a house for performing its objective (NST {NSC_name : Object_recognization, State_name : Vehicle Controller/ Tarcking/SensingObject, Conditioin : Dectected _Ojbect = normal}). If it encounters any object, the adaptation agent analyzes the detected object and decides how to act against it. In the case of the pre-defined object such as people, wall, door, furniture and so on, the robot operates its own jobs. On the other hand, the detected object is not identified - that is, the condition (NST {Condition : Detected_Object = normal}) is violated - the robot starts a analysis process to decide the type of the occurred fault. The situation-aware monitor finds the type of the occurred fault to use a heuristic algorithm and a neuro fuzzy algorithm based on granular-neuro-evolutionary computing (FMT {FM_name : UnknownObject, Strucutre _name : GroundVehicle/ Vision_Sonsor, Fault_status : image Analysis() = fail}). Then referring to the APT, the robot builds the adaptation plan according to the shape, size, and movement speed of the object (APT{Policy_name : Upgrade SW, FM_name : UnknownObject, Strategy_name : SW Update}) and executes the adaptation strategy (AST {Strategy _name : SWUpdate, Action : Update current SW to new SW, Type : External}). Finally the robot applying new SW can, therefore, avoid or remove the obstacle. The second function is to adapt to some hardware failures (NST {NSC_name : Move_forward, State_name : Vehicle Controller/Tarcking/Forward, Conditioin : Rotated_angle = 0 OR L-motor_rpm = R-motor_rpm}). If the left step motor of the robot has a problem, it cannot go straight. Even though the robot is ordered to go straight, it veers to the right because the revolutions per minute of the left step motor and that of the right step motors differs. That is, the condition (NST {Condition : Rotated_angle = 0 OR L-motor_rpm = R-motor _rpm}) is violated. In this case, the situation-aware monitor decides the type of fault through the analysis process (FMT {FM_name : BrokenMotor, Strucutre_name : Ground Vehicle/Right-side_Motor, Fault_status : RevolutionPer Minute < threshold}). Then referring to the APT, the robot builds the adaptation plan that decrease the torque value of the normal motor (APT {Policy_name : ChangeTorque, FM_ name : BrokenMotor, Strategy_name : ChagneValue}) and executes the adaptation strategy (AST {Strategy_name : ChangeValue, Action : Change internal variables of adaptable software, Type : Internal}). As a result, the robot may move slowly, but can go straight.
5 Conclusion The self-adaptation technology that handles uncertainty and unpredictable situation of the physical world is a new paradigm to increase the robustness and the reliability of software. In this paper, we proposed the faults and the adaptation policy modeling method to develop a self-adaptive robot. The faults modeling is to find out anomalies of the robots whereas the adaptation policy modeling is to recover the robots from the detected faulty status. To apply proposed method to the self-adaptive robot, we have
164
I. Chun et al.
verified that the robot satisfies user requirements, overcomes the system failure, and then provides seamless service to users without intervention. At this time, the prototype of the self-adaptive robot has been developed and we are achieving additional tests to demonstrate the feasibility of the proposed method. Also we are planning to study about improving the efficiency and the reliability of our research. Acknowledgments. This work was supported by the IT R&D Program of MKE/KEIT [10035708, “The Development of CPS(Cyber-Physical Systems) Core Technologies for High Confidential Autonomic Control Software”].
References 1. Gates, B.: A Robot in Every Home. Scientific American Magazine (2007) 2. Chetan, S., Ranfanthan, A., Campbell, R.: Towards Fault Tolerant Pervasive Computing. IEEE Technology and Society 24(1), 38–44 (2005) 3. Cheng, B.H.C., de Lemos, R., Giese, H., Inverardi, P., Magee, J.: Software Engineering for Self-adaptive devices: A Research Roadmap 4. Laddaga R.: Self-adaptive software, Technical Report. 98-12. DARPA BBA (1997) 5. Salehie, M., Tahvildari, L.: Self-adaptive software: Landscape and research challenges. ACM Transaction on Autonomous and Adaptation Systems 4(2) (2009) 6. Garlan, D., Cheng, S.-W., Huang, A.-C., Schmerl, B., Steenkiste, P.: Rainbow: Architecture-based self-adaptation with reusable infrastructure. IEEE Computing 37(10), 46–54 (2004) 7. Tuttle, S., Batchellor, V., Hansen, M. B., Sethuraman, M.: Centralized risk management using tivoli risk manager 4.2, Technical report, IBM Tivoli Software (2003) 8. Candea, G., Kiciman, E., Kawamoto, S., Fox, A.: Autonomous recovery in componentized internet applications. Cluster Computing 9(1), 175–190 (2006) 9. Mukhija, A., Glinz, M.: Runtime adaptation of applications through dynamic recomposition of components. In: Proceedings of the International Conference on Architecture of Computing Systems, pp. 124–138 (2005) 10. White, J., Schmidt, D.C., Gokhale, A.S.: Simplifying autonomic enterprise java bean applications via model-driven development: A case study. In: Briand, L.C., Williams, C. (eds.) MoDELS 2005. LNCS, vol. 3713, pp. 601–615. Springer, Heidelberg (2005) 11. Lapouchnian, A., Liaskos, S., Mylopoulos, J., Yu, Y.: Towards requirements-driven autonomic systems design. In: Proceedings of the Workshop on Design and Evolution of Autonomic Application Software, pp. 1–7 (2005) 12. Floch, J., Hallsteinsen, S., Stav, E., Eliassen, F., Lund, K., Gjorven, E.: Using architecture models for runtime adaptability. IEEE Software, 62–70 (2006) 13. Kumar, V., Cooper, B., Cai, Z., Eisenhauer, G., Schwan, K.: Middleware for enterprise scale data stream management using utility-driven self-adaptive information flows. Cluster Computing 10(4), 443–455 (2007) 14. Zeigler, B.P., Kim, T.G., Praehofer, H.: Theory of Modeling and Simulation: Integrating Discrete Event and Continuous Complex Dynamic Systems, 2nd edn., p. 510. Academic Press, Boston (2000)
Reducing Error Propagation on Anchor Node-Based Distributed Localization in Wireless Sensor Networks* Taeyoung Kim1, Minhan Shon1, Mihui Kim2, Dongsoo S. Kim3, and Hyunseung Choo1,** 1 School of Information and Communication Engineering, Sungkyunkwan University, Korea [email protected], [email protected], [email protected] 2 Department of Computer Engineering, Hankyong National University, Korea [email protected] 3 Department of Electrical and Computer Engineering, Indiana University-Purdue University Indianapolis (IUPUI), IN46202, USA [email protected]
Abstract. This paper proposes a scheme for reducing error propagation (REP) based on low-cost anchor node-based distributed localization (LADL) that are advanced localization schemes using connectivity information between nodes. Even though the localization accuracy of LADL is higher than that of previous schemes (e.g., DRLS), estimation error can be propagated when anchor node ratio is low. Thus after the initial position of each normal node is estimated by LADL, estimated position is corrected using anchor nodes located several hops away so that error propagation is reduced. The simulation results show that LADL with REP has higher localization accuracy and lower message transmission cost than DRLS. In addition, even though message transmission cost is increased, the localization accuracy of LADL with REP is higher than that of DRLS. Keywords: Localization, Anchor node, Error propagation, Error correction, Two-hop flooding, Grid scan algorithm.
1 Introduction Many Applications in wireless sensor networks, such as object tracking, environment and habitat monitoring, intrusion detection, and geographic routing, are based on the location information of sensor nodes [1]. Thus, more accurate the position information is, more reliable and efficient the services are. The simplest method to give accurate position information to sensor nodes is to equip GPS [2] on each sensor *
node. However, it is not suitable that relatively expensive devices such as GPS are equipped on all of low-cost sensor nodes. Moreover, due to the fact that GPS is not usable at indoor environment, the new methods to estimate position for all of sensor nodes are required. Therefore, the localization research without additional devices is necessary [3]. Localization schemes are classified into range-based and range-free scheme. Range-based schemes [4-5] measure the distance or the angle between sensor nodes and estimate position using the information. However, range-based schemes are not suitable for the real wireless sensor networks due to additional devices to measure the distance or the angle between sensor nodes, and interferences such as noise and fading. Thus range-free schemes that estimate position using connectivity information between sensor nodes are proposed [6-8], In range-free schemes, however, message transmission cost to gather the connectivity information among sensor nodes and computational cost to estimate positions of sensor nodes increase, and position estimation error is relatively high. Recently, distributed range-free localization scheme (DRLS) [9] is proposed to increase the localization accuracy. However, the vector-based refinement to increases the estimation accuracy could calculate inaccurate position of normal nodes according to the position balance of anchor nodes, and needs square-root calculation that requires high computational cost. In addition, the normal nodes without neighbor anchor nodes estimate their position using the location information of neighbor normal nodes that include localization error, so that localization error might be propagated. Therefore, low-cost anchor node-based distributed localization (LADL) [10], our previous work, was proposed to overcome such the weaknesses of DRLS. LADL substitutes grid scanning using anchor nodes within two-hop distance from normal nodes for the vector-based refinement of DRLS. Thus, computational cost occurred by refinement of DRLS is diminished and localization accuracy was improved. However, even in DRLS, the initial estimated positions are affected by the error propagation in the environment with the small number of anchor nodes. In this paper, we advance the LADL with a method for reducing error propagation (REP); it corrects initial estimated position by LADL for the normal nodes without neighbor anchor nodes. Our contributions are followings:
We develop a REP scheme based on LADL to decrease the propagation error in the case of normal nodes without neighbor anchor nodes. The simulation results show that our scheme outputs higher localization accuracy and lower message transmission cost than DRLS. In addition, even though message transmission cost is increased, the localization accuracy of LADL with REP is higher than that of DRLS.
The remainder of the paper is organized as follow. Section 2 presents related work and assumptions. Section 3 describes LADL with REP in detail. Section 4 discusses simulation results. Finally, section 5 concludes the paper.
Reducing Error Propagation on Anchor Node-Based Distributed Localization
167
2 Preliminaries & Related Work 2.1 Assumptions and Definition The purpose of the proposed scheme is more accurate localization by improving the range-free localization schemes. The followings are the basic assumptions of the range-free localization schemes. Sensor network consists of normal nodes that do not know their position and a few anchor nodes that know their position by GPS. All sensor nodes are randomly deployed in the sensor field, and do not move after deployment. In addition, all sensor nodes have their unique IDs and the same data transmission radius [9]. In the proposed scheme, each normal node estimates its position by one- and twohop anchor nodes. When the data transmission radius is r, one-hop anchor nodes are defined as the anchor nodes located within the data transmission radius r of a normal node. Moreover, two-hop anchor nodes are defined as the anchor nodes that are located in a circle, whose center is the position of a normal node and radius is 2r, that are not one-hop anchor nodes, and that a two-hop path exists from the normal node to. Fig. 1 shows an example of one- and two-hop anchor nodes. Anchor nodes A1 and A2 are one-hop anchor nodes since A1 and A2 are located within the transmission radius of a normal node N1. Anchor node A3 is two-hop anchor node of N1 since N1 obtains the position information of A3 through a normal node N2.
Fig. 1. An example of one- and two-hop anchor nodes
2.2 Related Work DRLS [9] uses location information of anchor nodes within not only one-hop distance but also two-hop distance for localization. A normal node N assumes an estimative rectangle (ER) that is overlapped region of rectangles tangent to transmission radius of anchor nodes within one-hop distance from the normal node. Subsequently, ER is divided to small grid-shaped cells, and how many times each cell overlaps with transmission range of anchor nodes is calculated (i.e., grid scan). The average position of cells with the highest value is the initial estimated position of N, (i.e., N’). As the refinement phase, the initial estimated position N’ is amended based on the virtual force (VF). Refer the DRLS [9] for more detail.
168
T. Kim et al.
However, DRLS has some problems as followings. First, two-hop flooding of DRLS is based on anchor nodes. Each anchor node exchanges its position information with anchor nodes that are located within two-hop distance, and provides all of information to its neighbor normal nodes. Even though normal nodes obtain some position information during the information exchange between anchor nodes, the normal nodes ignore the information. Second, the computational cost of the squareroot operation in vector-based refinement phase is so high that the operation is not suitable for the low-cost sensor networks [11]. Thirds, in refinement phase, localization accuracy can be reduced in accordance with the distribution of sensor nodes. Fourth, normal nodes that do not have any neighbor anchor nodes estimate their position with the position information of neighbor normal nodes that have localization error, and thus error propagation is occurred seriously. Therefore, a scheme that substitutes the vector-based refinement and reduces error propagation should be proposed to solve the four problems above.
3 Anchor-Node Based Distributed Localization with Error Correction This section presents the basic LADL and advances the LADL with error correction to reduce the error propagation. 3.1 Basic LADL Basic LADL consists of three steps: anchor node selection, ER construction, and grid scan. We introduce the steps and limitation of LADL. Anchor Node Selection In the anchor node selection step, two-hop flooding is used for normal nodes to obtain the information of anchor nodes within two-hop distance. Two-hop flooding is started from each anchor node, and each anchor node broadcasts its ID and position information to its neighbor sensor nodes. All normal nodes that receive the information broadcasted decide the anchor nodes of the information obtained as their one-hop anchor nodes. Then, the normal nodes that have the information of one-hop anchor nodes broadcast the information of their one-hop anchor nodes to its neighbor normal nodes. The normal nodes that receive the information of the one-hop anchor nodes from their neighbor sensor nodes consider the anchor nodes of the information obtained as two-hop anchor nodes. ER Construction Every normal node can construct an estimative rectangle (ER) with collected position information of both its one- and two-hop anchor nodes; ER was designed for normal nodes estimate their position [6]. To reduce the computation overhead for the localization, the normal node uses the rectangle that circumscribes the transmission radio range of each anchor node. Thereafter, an ER is constructed by calculating the overlapped region of these rectangles. Fig. 2(a) shows an example of the ER construction. Normal node N keeps the position information of one-hop anchor node A3 and two-hop anchor nodes A1 and A2,
Reducing Error Propagation on Anchor Node-Based Distributed Localization
169
and constructs an ER (i.e., dark shaded area in the figure) applying different rectangle drawing mechanisms to one- and two-hop anchor nodes, respectively. Then, the normal node should be located inside the ER. Distributed Grid Scan The ER calculated by a normal node includes a region that is larger than the region in which the normal node can actually exist. Accordingly, the localization accuracy can be enhanced by excluding the region where a normal node cannot actually exist from the ER using the grid scan algorithm. After dividing the ER into a set of grid cells (i.e., the length of one side of each cell is set at 0.1r.), the normal node decides if which grid cells should be excluded by scanning each cell as shown in Fig. 2(b). Additionally, the value of the cell is set at 1, and the coordinate of the center of gravity in each cell is selected as the coordinate value of the position of the cell. Here, 1, the value of each cell, means that a normal node could exist in the cell. Finally, normal node N obtains final estimated position N´ by computing the average of the representative coordinate values in each shaded cell.
(a)
(b)
Fig. 2. An example of ER construction and grid scan ((a) ER construction and (b) Grid Scan)
In LADL, normal nodes that do not have any information of neighbor anchor nodes, wait until neighbor normal nodes complete localization and estimate their position using the information of neighbor normal nodes. However, this information used for estimating positions is not accurate but estimated information, and thus error propagation occurs frequently. Especially, when two anchor nodes of a normal node are located together, the normal node estimates its position on the extended line that connects these two anchor nodes. The frequency of the case is not high, but the influence of the case is high and error propagation could not be guessed when the case occurs. 3.2 Reducing Error Propagation REP handles only the error propagation that is occurred by two anchor nodes located together since the error propagation should be reduced due to its influence. Through collecting information of other anchor nodes and repositioning with anchor nodes, REP corrects the position information of normal nodes by rotating the estimated positions centering on anchor nodes.
170
T. Kim et al.
Collecting Information of Other Anchor Nodes To reduce error propagation, other information is necessary for adjusting position information. REP uses the information of anchor nodes that is multi-hop away from normal nodes. Normal nodes can obtain the information of anchor nodes multi-hop away when other normal nodes finishing to estimate the initial position and broadcasting their position to their neighbor normal nodes. When sending information of themselves to their neighbor sensor nodes after completing to estimate their initial position, normal nodes that have neighbor anchor nodes send not only their estimated position information but also the position information of an one-hop anchor node and the distance between the normal node and the one-hop anchor node. The anchor node information obtained first is used as the center of the rotation. Moreover, the center of the rotation of each normal node that does not have any neighbor anchor nodes is fixed for the center of the rotation of the neighbor normal node. The distance between the normal node and the center of the rotation is determined by hop progress (i.e., an estimated hop distance using probability based on the density of sensor nodes) [12]. The higher the density of sensor nodes is, the longer the estimated one-hop distance is. The density of normal nodes means the average number of sensor nodes in the transmission radius of a sensor node. Hop progress can be derived from (1) [12, 13]. In (1), E(R) is the estimated hop progress, Ec is the sensor node density, and r0 is the transmission radius. Hop progress is calculated by only the sensor node density and the data transmission radius.
2( EC + 1) sin θ E ( R) = πr0 2
∫
r0
0
2
l e
−
EC +1
π
θ (1−
l2 r0 2
)
dl
(1)
Each normal node estimates the distance between itself and the center of the rotation using hop progress. Then, the normal node broadcasts the position information of the center of the rotation and hop progress to the position for neighbor normal nodes that do not have any information of anchor nodes. This broadcast can be efficient by adding the information of the center of the rotation to the broadcast packet after the grid-scan phase. Through this procedure, all normal nodes have position information of 1~3 anchor nodes and hop progress to the anchor nodes as well as their initial position. Repositioning with Anchor Nodes Each normal node that has the position information of three anchor nodes and hop progress to the anchor nodes corrects its initial position using its information obtained. Each normal node calculates two cross points of two circles whose center point is the center of the rotation and radius is the hop progresses of two anchor nodes. These two cross points are regarded as the position in which the normal node can be located. However, since the hop progresses are based on the probability, the accuracy of these two cross points are low. Thus, these two points are used not to correct the initial position directly, but to decide the direction of the rotation. The direction of the rotation is decided as a cross point of the two circles that is nearer by the third anchor node that is not used to make two circles. The reason of the selection between two cross points is that the cross point far from the third anchor node cannot receive the information of the third anchor node. Fig. 3 shows an example of this procedure.
Reducing Error Propagation on Anchor Node-Based Distributed Localization
171
Fig. 3. Deciding the direction of the rotation with the information of three anchor nodes
To correct the initial position of a normal node by rotation, the angle of the rotation is necessary. As knowing the initial position and the center and the direction of the rotation, the normal node can obtain the angle of the rotation. Fig. 4 shows an example of the center and the angle of the rotation, and the angle of the rotation can be obtained by (2).
cos θ =
d ( A1 , D) 2 + d ( A1 , N1 ' ) 2 − d ( N1 ' , D) 2 2 ⋅ d ( A1 , D) ⋅ d ( A1 , N1 ' )
(2)
Ɵ
In (2), d(a,b) is the Euclidean distance between the point a and b. The value of sin can be a positive or negative value, so the normal node should decide the presence of the negative sign. When the coordinate of A1 is ( x A1 , y A1 ), the coordinate of N1’ is ( x N1 ' ,
y N1 ' ), and the coordinate of the rotation direction D is ( xD , y D ), sinƟ is cal-
culated in (3) by the relation between the line A1N1’ and the position D. ⎧ 1 − (cos θ ) 2 , if ( x − x ≥ 0 and ( x − x )( y − y ) ≥ ( y − y )( x − x )) N1 ' A1 N1 ' A1 D A1 N1 ' A1 D A1 (3) ⎪⎪ sin θ = ⎨ or ( x N1 ' − x A1 < 0 and ( x N1 ' − x A1 )( y D − y A1 ) < ( y N1 ' − y A1 )( x D − x A1 )) ⎪ 2 ⎪⎩− 1 − (cos θ ) , otherwise
Then, the coordinate of N1”, ( x N1 " , the angle of rotation
y N1 " ) that is the rotatory translation of N1’ with
Ɵ is obtained in (4).
⎛ x N1" ⎞ ⎛ cosθ ⎜ ⎟ ⎜ y N " ⎟ = ⎜⎜ sin θ ⎝ 1 ⎠ ⎝
− sin θ ⎞⎛ x N1 ' − x A1 ⎞ ⎛ x A1 ⎞ ⎟+⎜ ⎟ ⎟⎜ cosθ ⎟⎠⎜⎝ y N1 ' − y A1 ⎟⎠ ⎜⎝ y A1 ⎟⎠
(4)
172
T. Kim et al.
In Fig. 4, N1” that is the corrected position by rotatory translation is closer to N1, that is the real position of the normal node, than the initial position N1’. The normal nodes that complete REP broadcast their estimated position information to apply REP to the normal nodes that have one or two anchor node information.
Fig. 4. Amending the position with the initial estimated position and the direction
4 Performance Evaluation 4.1 Simulation Environment In the simulation of LADL and LADL with REP, data transmission radiuses of all sensor nodes are same as r, and unit dist model [14] is used to transmit data. The same sensor nodes are deployed randomly on sensor field whose size is 10r x 10r. A few anchor nodes that equip GPS know their real position and all sensor nodes are same except for the existence of GPS. The collision occurred during the message transmission for the position estimation is not considered. The cell size of grid scan phase in LADL is fixed to 0.1r x 0.1r. All simulation results are based on 100 repeats. The metrics for the performance evaluation are the localization accuracy (i.e., the average of the Euclidean distance between the real positions and the estimated positions of all normal nodes), the mean squared error (MSE) (i.e., the degree of variance about the estimated position of normal nodes), and the message transmission cost (i.e., the number of messages exchanged). The parameter for the simulation is the ratio of anchor nodes; it is changed from 5 to 30%. 4.2 Simulation Results Fig. 5 shows the localization accuracy and the MSE of LADL and LADL with REP. In Fig. 5(a), the localization accuracy of LADL with REP is highest, and that of DRLS is smallest, since LADL increases the localization accuracy by estimating position of normal nodes without the vector-based refinement of DRLS and LADL with REP amends the initial estimated position normal nodes in LADL by applying REP. The lower the ratio of anchor nodes is, the fewer the number of the position
Reducing Error Propagation on Anchor Node-Based Distributed Localization
173
information of anchor nodes is; therefore, localization accuracy is decreased. As shown in Fig. 5(a), LADL with REP has the lowest MSE, and DRLS has the highest MSE since LADL with REP has the small variance of the estimated position by reducing error propagation. Fig. 6 shows the message transmission cost for the position estimation (i.e., defined as the number of messages exchanged). Due to the fact that two-hop flooding is performed based on anchor nodes in DRLS and based on normal nodes in LADL, the message transmission cost of LADL is lowest. The reason of the highest data transmission cost of LADL with REP is that each normal nodes broadcasts the information of their center of the rotation additionally after applying REP. Generally, DRLS transmits messages more as many as the number of anchor nodes than LADL, and LADL with REP transmits messages more as many as the number of normal nodes than LADL. The simulation results show that the message transmission cost of LADL with REP is higher than LADL when the ratio of anchor nodes is low, but the gap of costs between LADL with REP and LADL declines as the ratio of anchors increases.
(a)
(b)
Fig. 5. Mean error and mean square error
Fig. 6. Number of messages
174
T. Kim et al.
5 Conclusions The paper proposed a scheme to reduce the error propagation on LADL, a LADL with REP. It makes normal nodes to estimate their position more accurately by rotating the initial estimated positions whose centers are anchor nodes. The simulation results showed that localization accuracy is higher, but the data transmission cost is higher than DRLS. Due to the fact that the data transmission cost of LADL with REP is high, we should decide a scheme among LADL and LADL with REP according to the application. We will do in-depth simulation on various environments to evaluate our approach from the diverse viewpoints.
References 1. Akyildiz, I.F., Weilian, S., Sankarasubramaniam, Y., Cayirci, E.: A Survey on Sensor Networks. IEEE Communications 40, 102–114 (2002) 2. Djuknic, G.M., Richton, R.E.: Geolocation and Assisted GPS. IEEE Computer 34, 123– 125 (2001) 3. Wang, C., Xiao, L.: Sensor Localization under Limited Measurement Capabilities. IEEE Network 21, 16–23 (2007) 4. Savvides, A., Han, C.-C., Strivastava, M.B.: Dynamic Fine-Grained Localization in AdHoc Networks of Sensors. In: ACM MobiCom, pp. 166–179 (2001) 5. Hightower, J., Borriello, G.: Location Systems for Ubiquitous Computing. IEEE Computer 34, 57–66 (2001) 6. Doherty, L., Pister, K.S.J., Ghaoui, L.E.: Convex Position Estimation in Wireless Sensor Networks. In: IEEE INFOCOM, vol. 3, pp. 1655–1663 (2001) 7. Hu, L., Evans, D.: Localization for Mobile Sensor Networks. In: ACM MobiCom, pp. 45– 47 (2004) 8. Rudafshani, M., Datta, S.: Localization in Wireless Sensor Networks. In: IPSN, pp. 51–60 (2007) 9. Sheu, J.P., Chen, P.C., Hsu, C.S.: A Distributed Localization Scheme for Wireless Sensor Networks with Improved Grid-Scan and Vector-Based Refinement. IEEE Transactions on Mobile Computing 7, 1110–1123 (2008) 10. Kim, T., Shon, M., Choi, W., Song, M., Choo, H.: Low-cost two-hop anchor node-based distributed range-free localization in wireless sensor networks. In: Taniar, D., Gervasi, O., Murgante, B., Pardede, E., Apduhan, B.O. (eds.) ICCSA 2010. LNCS, vol. 6018, pp. 129– 141. Springer, Heidelberg (2010) 11. Acharya, M., Girao, J., Westhoff, D.: Secure Comparison of Encrypted Data in Wireless Sensor Networks. In: IEEE Symposium on Modeling and Optimization in Mobile, Ad Hoc, and Wireless Networks, pp. 47–53 (2005) 12. Wang, Y., Wang, X., Wang, D., Agrawal, D.P.: Range-Free Localization Using Expected Hop Progress in Wireless Sensor Networks. IEEE Transactions on Parallel and Distributed Systems 20, 1540–1552 (2009) 13. Wang, X.: QoS Issues and QoS Constrained Design of Wireless Sensor Network. PhD dissertation, Univ. of Cincinnati (2006) 14. Clark, B.N., Colboum, C., Johnson, D.S.: Unit disk graphs. Discrete Mathematics 86, 165– 177 (1990)
Attention Modeling of Game Excitement for Sports Videos Huang-Chia Shih Department of Electrical Engineering, Yuan Ze University, 135 Yuandong Rd., Chungli, Taiwan, R.O.C. [email protected]
Abstract. This paper presents a novel attention modeling method that combines the visual attention features with the contextual game status information for sports videos. Two critical issues for attention-based video content analysis are addressed in this paper. (1) It illustrates the approach of extracting the visual attention map and illustrates the algorithm for determining the contextual excitement value. Semantic contextual inference is used to simulate how the video content attracts the subscribers. (2) It presents the fusion methodology of visual and contextual attention analysis based on the characteristics of human excitement. The experimental results demonstrate the efficiency and the robustness of our system by means of some baseball game videos. Keywords: attention modeling, key-frame detection, content analysis, contextual modeling, excitement curve, content-based video retrieval.
In this paper, we proposed a novel attention analysis of game excitement by integrating the object-oriented visual attention maps and the contextual on-going game outcomes. Using an object-based visual attention model combined with contextual attention model not only precisely determines the human perceptual characteristics but also effectively determines the type of video content that attracts the attention of the user.
2 Game Excitement In this paper, we adopted the contextual attention score to determine the game excitement, while the key-frames being determined based on the visual attention score. 2.1 Key Shot Selection Prior to determining the key-frame, the key shot will be quickly selected using the information from the SCB template. There are two different kinds of key shot: SCBappeared key shot and content-changed key shot. The former is defined as the shot taken whenever the SCB appears, whereas the latter is taken when the SCB content changes. In this paper we assume that the SCB template does not change during the entire video program so that we may use the SCB color model to identify its presence, even though the SCB may be transparent. The similarity measure between model hR and the potential SCB is formulated by the Mahalanobis distance as d(hiMscb, hR)=(hiMscbhR)TCm-1(hiMscb-hR), where hiMscb denotes the color distribution of the potential SCB region in the ith frame. If a shot is being detected that may change the game status, or if the SCB appears, then that shot will be selected as the key shot. 2.2 Key-Frame Selection The frame-level attention score can be quantitatively measured by means of the object-based visual attention score which is defined as the combination of all the visual attention scores with bias. Basically, a key-frame selection is based on two rules: (1) key-frames must be visually significant, and (2) key-frames must be temporally representative. It is obvious that combining all attention feature maps can meet the first rule. In the present study we adopted the camera motion characteristics and treated them as the balancing coefficient to support the second rule. A numerical value was then derived from the visual characteristics of all segmented objects within the frame. The denominator of each part is the normalization term, which is the sum of the attention maps for all frames belonging to the same video shot. Based on the predefined number of key-frames, the Ri key-frames {Fk*} with the largest visual attention score ψvt, is calculated as follows Ri
Fk* =
∪ arg max [ ψ ( f )× M i=1
fi
t v
i
camera
]
( fi ) ,
(1)
Attention Modeling of Game Excitement for Sports Videos
177
FM spatial ( f i )
(2)
Where ψ tv ( f i ) = ε 1 ×
Ns
∑ FM
spatial
j= 1
(fj)
+ ε2 ×
FM temporal ( f i ) Ns
∑ FM j= 1
temporal
(fj)
+ ε3 ×
FM facial ( f i ) Ns
∑ FM
facial
,
(fj)
j= 1
where ε1, ε2, and ε3 denote the weighting coefficients among the visual feature maps, Mcamera(fi) denotes the model of global camera motion in the moment of frame i .
3 Visual Attention Based on the types of features, the object-based visual attention model can be determined by four types of feature maps, i.e., spatial, temporal (local), facial, and (global) camera motion. 3.1 Spatial Feature Maps We selected three spatial features to be extracted and to be used as the spatial feature map: intensity, color contrast (red-green, blue-yellow), and orientation. The local orientation feature was acquired from I using the oriented Gabor pyramids G(σ, θ), where σ represents the scale and θ indicates the orientation. The spatial feature map FMspatial is then defined by the weighted sum of these three conspicuity maps. 3.2 Temporal Feature Maps The motion activity (MA) was computed for each frame with W×H macroblocks. The temporal feature map FMtemporal was integrated with the neighboring MAs within the object boundary which implies the moving energy of the object and reflects the information regarding the texture of the object. Consequently, the MAs in the background are ignored because they attract little attention. 3.3 Facial Feature Maps In this study we adopted the skin color instead of a face detection scheme because it requires less time. Empirically, we set the range of the skin tone in the (r, g, b) color space based on the following criteria to obtain the most satisfactory results, (1) r > 90, (2) (r-g) ∈ [10, 70], (3) (r-b) ∈ [24, 112], (4) (g-b) ∈ [0, 70]. The facial feature maps FMfacial can be described as having the probability that each pixel belongs to the skin-color tone. 3.4 Global Camera Motion When a highlight occurs, they track the key object for a while. Usually they move the camera or change the focal length. Thus, the global camera motion is very important information to infer excitement. In this paper we take the camera motion into consideration for computing the visual attention and replace the time consuming 2-D calculation with two 1-D calculations by projecting the luminance values in the vertical and horizontal directions [7]. Based on this algorithm, we can acquire the model of global camera motion Mcamera.
178
H.-C. Shih
4 Contextual Attention The contextual information shown in the superimposed caption box (SCB) allows us to determine the contextual excitement. The outcome statistics in a sports video are usually shown on the screen in a SCB. 4.1 Caption Template Construction Due to the limited image resolution, it is a nontrivial problem to achieve a good character recognition rate. To recognize the characters, we use a simple multi-class SVM classifier method [8] in which the feature vector of the character block consists of Area, Centroid, Convex Area, Eccentricity, Equivalent Diameter, Euler Number, Orientation, and Solidity, and the size of the character block is normalized to 30×40. Once a new character is recognized, we can generate a specific caption template defined as
CTi = [Pos, text/digit, SV,CB],
(3)
where Pos indicates the relative location in the SCB. The text/digit bit indicates that the caption is either text or digit, SV indicates a set of support vectors for this character, and CB is the corresponding character block. The caption template can be used to identify the characters in the succeeding videos. Table 1. The reachable contextual information of the SCB
Annotation
Context
Description
λ1
INNS
The current innings
λ2
RUN-
The base that are occupied by runners
λ3
RUNS
λ4
OUTS
The number of outs
λ5
BALLS
The number of balls
λ6
STRIKES
The number of strikes
The score difference
4.2 Modeling the Contextual Attention Different from visual attention which varies frame-by-frame, contextual attention varies depending if it is shot-based or event-based. Unfortunately, we were unable to obtain all of the statistical information from video the frames. Therefore, in this paper, we not only adopted the available information from the SCB, but also employed the historical statistics data. Normally, the contextual information embedded in the SCB consists of six classes (Λ={λi|i=1,2,…,6}) for the baseball game, as listed in Table 1.
Attention Modeling of Game Excitement for Sports Videos
179
4.2.1 Implication Factors In this paper, the contextual information is divided into three classes. These three classes are based on the relationship between the value of the context and the degree it excites the viewer, proportionally, specifically, or inversely. After classifying the contextual description, a group of implication factors {Fl |l=1,2,…, l’,…,L-1, L} are used to model the human excitement characteristics. Then, each of these implication factors can be classified as one of the three classes as Ω(Fl’) ∈ {ωp, ωs, ωi}, where ωp, ωs, and ωi represent the corresponding factor as a proportional type, specific type, or an inverse type respectively. It not only uses the implication factor in the current moment (i.e., Ft), but it also takes into account the implication factor from the historic statistics (i.e., F0:t-1). Let ψc(fi) indicate the viewer’s contextual attention score of frame f, which can be contributed to the implication factors Ft={Fk|k=1,2,…,K} which consist of all the probable annotations of the SCB at that moment. Also, the historical statistics are taken into consideration, in which the implication factors from the historical statistics are represented by F0:t-1={Fq|q=1,2,…,Q}. Thus ψct(fi)= F t+ F 0:t-1
(4)
Four implication factors from the current observation are considered in determining the contextual attention score, which include
F1t : The score difference The scoring gap between two teams greatly attracts the attention of the viewer. When the runs scored are the same or are very close, it indicates that the game is very intense. F2t : The number of BALLS and STRIKES pairs The ratio between BALLS and STRIKES can be applied to model user attention. In a baseball game that is being broadcasted, the number of balls is repeatedly updated from 0 to 3 and the number of strikes and outs are updated from 0 to 2. When λ5 or λ6 reaches 3 or 2, it indicates that the game is getting a high level of attention, because the current player will soon be struck out or get to walk soon. F3t : The number of the inning being played Generally speaking, the less the number of remaining innings, the more attention the game will attract. Therefore, we can design an implication factor from the number of the inning λ1. In general, the maximum number of innings is 9. F4t : The number of outs Obviously, the higher the number of OUTS, the higher the amount of excitement, and thus the higher the attention that is given. Here, we are also concerned with the past statistical data. For baseball games, a lot of fans and researchers like to analyze the relationship between the game’s statistics and the probability of scoring points. Jim Albert [9] collected case studies and applied statistical and probabilistic thinking to the game of baseball. He found that there is an
180
H.-C. Shih
active effort by people in the baseball community to learn more about baseball performance and strategy through the use of statistics. The on-base situation usually attracts much user interest. In addition, coaches tend to change their strategy depending on the situation so as to score as many runs as possible in the remainder of the inning. In [9], authors conducted a thorough statistical analysis of the data from the National League for the 1987 season, the play-by-play data. They archived the statistics of the expected runs scored in the remaining inning from each of 24 possible pairs (λ2, λ4). They then estimated the possible scoring under different base-occupied and number of out scenarios using the historic game statistics. Hence, based on the statistics, the implication factor was then adjusted by using the weighting sum of the past attention score given λ4 .
5 Experimental Results Our proposed object-based attention model is used as an effective scheme to measure the attention score. Table 2 shows the attention scores in different modules for six representative frames shown in Fig. 1. Table 2. The examples with different Attention Scores
#Frame
Integral Visual Attention
Spatial
Temporal
Facial
Camera Motion DVv DVH
573
0.243
0.053
0.688
0.107
144
241
650
0.093
0.051
0.110
0.039
0
11
931
0.691
0.506
0.672
0.562
82
242
1716
0.051
0.069
0.041
0.091
0
11
3450
0.186
0.038
0.253
0.038
24
129
4001
0.286
0.056
0.394
0.047
26
41
The values of each column represent the attention score, and accurately reflect the attention to the content via spatial, temporal, facial and global motion. Frame #931 zooms in for a close-up, but the object itself is stable, resulting in a high visual attention. Frame #4001 is a mid-distance view with local motion and the camera zooming. However, the face it zooms into is clear and near the center of the frame, resulting in a high attention score. Frames #650 and #1716 are globally static, so they have low temporal attention scores and low camera motions resulting in decreased visual attention. Frame #5763 has high attention due to the rapid panning to capture the pitcher throwing the ball. Frame #3450 is a mid-distance view with middle face attention, and with the camera panning, which also increases attention.
Attention Modeling of Game Excitement for Sports Videos #573
#650
#931
#1716
#3450
#4001
181
Fig. 1. The representative frames for testing
Fig. 2 shows the result of key-frame detection of the sample. It is evident that the proposed scheme is capable of extracting a suitable number of key-frames. Based on the visual attention score, the extracted key-frames are highly correlated with human visual attention. We suppose that from each of these shots at least one key-frame was selected based on the frames’ attention scores. That is why a few of the key-frames look like normal average play. Nevertheless, Fig. 2 shows that an exciting moment can be properly preserved by means of our proposed method.
Fig. 2. Sample results of the proposed key-frame detection method
6 Conclusions In this paper, we proposed a novel attention modeling method by integrating the visual attention model and the contextual game-status information and successfully employing for key-frame detection. We have illustrated an approximate distribution of a viewer’s level of excitement through contextual annotations using the semantic
182
H.-C. Shih
knowledge and visual characteristics. The key-frame selection depends on the combination of all the visual attention scores with bias. Employing the object-based visual attention model integrates with the contextual attention model not only produces precise human perceptual characteristics, but it will also effectively determine the type of video content that will attract the attention of viewers. The proposed algorithm was evaluated using commercial baseball game sequences and showed promising results.
References 1. Naphade, M.R., Kozintsev, I., Huang, T.S.: A Factor Graph Framework for Semantic Video Indexing. IEEE Trans. on CAS for VT 12(1), 40–52 (2002) 2. Doulamis, A., Doulamis, D.: Optimal Multi-Content Video Decomposition for Efficient Video Transmission over Low- Bandwidth Networks. In: IEEE ICIP (2002) 3. Shih, H.C., Huang, C.L., Hwang, J.-N.: An Interactive Attention-Ranking System for Video Search. IEEE MultiMedia 16(4), 70–80 (2009) 4. Tsotsos, J.K., Culhane, S.M., Wai, W.Y.K., Lai, Y.H., Davis, N., Nuflo, F.: Modeling visual-attention via selective tuning. Artifical Intelligence 78(1-2), 507–545 (1995) 5. Itti, L., Koch, C., Niebur, E.: A Model of Saliency-based Visual Attention for Rapid Scene Analysis. IEEE Trans. Pattern Anal. Mach. Intell. 20(11), 1254–1259 (1998) 6. Shih, H.C., Huang, C.L.: Content Extraction and Interpretation of Superimposed Captions for Broadcasted Sports Videos. IEEE Trans. on Broadcasting 54(3), 333–346 (2008) 7. Kim, M.K., Kim, E., Shim, D., Jang, S.L., Kim, G.: An Efficient Global Motion Characterization Methods For Image Processing Application. IEEE Trans. Consum. Electron. 43(4), 1010–1018 (1997) 8. Crammer, K., Singer, Y.: On the Algorithmic Implementation of Multi-class Kernel-based Machines. J. of Machine Learning Research 2, 265–292 (2001) 9. Albert, J.: Teaching Statistics Using Baseball, The Mathematical Association of America (2003)
Discovering Art in Robotic Motion: From Imitation to Innovation via Interactive Evolution Mária Virčíková and Peter Sinčák Center for Intelligent Technology, Dpt. of Cybernetics and Artificial Intelligence, Faculty of Informatics and Electrical Engineering, Technical University of Kosice, Letna 9, 04200 Kosice, Slovakia {maria.vircikova,peter.sincak}@tuke.sk
Abstract. The proposed research deals with a usage of artificial intelligent techniques in humanoid robotics. The objective is to design an interactive robotic behavior. The focus is on social robotics and how to improve the interaction between human and robot. In the first part user interactively cooperates with the System and this is the way how the robot acquires knowledge. Our approach is based on real manipulation where the robot freely interacts with humans and objects, and learns during continuous interaction with the environment. The experimental setup is the humanoid robot NAO. In the second part of the system we use the Interactive Evolutionary Computation – a technique which optimizes systems based on subjective human evaluation. So far, the algorithm is applied to a system of evolution of pleasant motions or dances. A user can build a robot behavior without any programming knowledge and then the evolutionary algorithm helps user to discover new possibilities of motions. Keywords: human-robot interaction, interactive evolutionary computation, learning from demonstration, robot Nao, sensor Kinect.
1 Introduction “Art begins in imitation and ends in innovation.” Mason Cooley (1927-2002)
Since AI’s earliest days there have been thoughts of building truly intelligent autonomous robots. In academic research circles, work in robotics has influenced work in Al and vice versa [2]. Recent generations of humanoid robots (HRs) increasingly resemble human in shape and capacities. HRs are robots that are at least loosely based on the appearance of the human body. In the Section 2. A. we employ this feature of our robots for mapping a human body to his points of the body. We work with humanoid robot Nao developed by Aldebaran Robotics. The robot has a human-like appearance and various sensors for interacting with humans. Humanoid robot Nao comes with 25 degrees of freedom for great mobility. The inertial sensor and closed loop control provide great stability while moving and enables positioning within space. Its state-of-the-art onboard actuators give Nao extreme precision in its movements. It supports multiple programming environments and programming languages like URBI Script, C++, Python, .Net Available on Linux. It can be programmed and controlled using Linux, Windows or Mac OS. The most used simulators for Nao are Microsoft Robotics Studio, Gostai Urbi Studio, Pyro, Choreographe and Webots. The position in angles of the degrees of freedom – servos or motors of the robot will represent the phenotype of the genetic algorithm The robot is running on a Linux platform and can be programmed using a proprietary SDK called NaoQi. It is based on a client-server architecture, where NaoQi itself acts as a server. The modules plug into NaoQi either as a library or as a broker, with the latter communicating over IP with NaoQi. Running a module as a broker thus has some IP-related overhead, but also allows the module to be run on a remote machine, which is very useful for debugging. NaoQi itself comes with modules which we use. As said in [3] new trends in research consider the robots ability to interact with people safely and in natural way. The current leading topics in HRs are so-called service and social robotics – where the target is getting robots closer to people and their social needs. There are many attempts to create robots capable of expressing emotions or communicating in a natural language. Our work has two parts: learning motion skills from observation (Section 2) and evolution of the motions by evaluating them (Section 3).
2 Learning Motion Skills from Observation 2.1 Human-Robot Interaction and Social Robotics An enduring challenge in social robotics has been to improve the interaction between human users and robots – it is still unnatural, slow and unsatisfactory for the human interlocutor. Robots have different abilities, knowledge, and expectations than humans. They need to react appropriately to human expectations and behavior. With this respect, scientific advances have been made to date for applications in entertainment and service robotics that largely depend on intuitive interaction.
Discovering Art in Robotic Motion: From Imitation to Innovation
185
Fig. 1. According to [7] these are the fields of major impact. Note that “collective robots” and “social robots” overlap where individuality plays a lesser role.
Dautenhahn and Billard [8] called the type of systems which we are trying to develop social robots: Embodied agents that are part of a heterogeneous group: a society of robots or humans. They are able to recognize each other and engage in social interactions, they possess histories – perceive and interpret the world in terms of their own experience – and they explicitly communicate with and learn from each other. It is a huge research field including gesture and natural language communication, perceiving emotions and expressing their own artificial emotions, establishing social relationships, exhibition of different personalities and characters, recognition of interaction partner and many others. In this paper we present our work in learning motion tasks from observation. 2.2 Imitation Learning Authors [9] formulated a statement: Imitation takes place when an agent learns a behavior from observing the execution of that behavior by a teacher. They also enumerated the advantages a learning mechanism should give: adaptation, efficient communication, compatibility with other learning mechanisms, efficient learning and thus, learning by imitation seems to be one of the major components of general intelligent behavior.
186
M. Virčíková and P. Sinčák
Fig. 2. Recognition of user (left) and Nao robot (right) using Kinect sensor
Imitation encompasses a set of different competences such as recognizing other people's actions, recognizing the goal of a particular action and the objects and/or subjects involved. This project attempts to implement a similar set of abilities in a humanoid robot following a biologically motivated perspective. Approach "Learning these tasks from demonstrations according to [10] is a convenient mechanism to customize and train a robot by transferring task related knowledge from a user to a robot. This avoids the time-consuming and complex process of manual programming. The way in which the user interacts with a robot during a demonstration plays a vital role in terms of how effectively and accurately the user is able to provide a demonstration. Teaching through demonstrations is a social activity, one that requires bidirectional communication between a teacher and a student. The key question was how to map a teacher action to an agent action. These agent actions in our experiments are the motor capabilities of the robots. We mimic some user movements using Microsoft Kinect sensor device, a horizontal bar connected to a small base with a motorized pivot and is designed to be positioned lengthwise above or below the video display. The device features an "RGB camera, depth sensor and multi-array microphone running proprietary software" [11], which provide fullbody 3D motion capture, facial and voice recognition capabilities. Figure 2 illustrates how Kinect sensor tracks a user and a robot. 2.3 Robots Learning from Each Other We are trying to model different robot´s personalities via its interaction with several human users and this way we personalize robot´s behaviors according to the interaction of the specific human user. This way various models for multiple robot
Discovering Art in Robotic Motion: From Imitation to Innovation
187
approach are developed. Robots are interacting between each other and are learning new motion tasks in the community of robots.
3 Discovering New Behaviors According to User’s Preference via Interactive Evolutionary Computation Jorge Luis Borges in his The Library of Babel writes about a library which consists of indeterminate and infinite number of books. Most of them contain nonsense and rational relation is almost miraculous exception, but somewhere there are books that contain all the knowledge of the universe. A traveler in such a Library wanting to find something rational has a very difficult task, as the most of the books are just a cluster of incomprehensible letters. How we could help this traveler to find a “masterpiece” in the endless space of options”. We need a tool that would fly in this space effectively and thanks to it we would find something inspiring. One of such a tool could be an evolutionary algorithm. An evolutionary algorithm is an optimization algorithm which uses some mechanisms inspired by biological evolution coming out from Darwinian Theory. Evolutionary computation is a biologically inspired general computational concept and includes Genetic Algorithms (GA), Genetic Programming and Evolutionary Strategies. The EC is a population-based searching algorithm and outputs multiple candidates, each called an individual, as system outputs. The space of all feasible solutions (it means objects among those the desired solution is) is called search space (also state space). Each point in the search space represents one feasible solution. Each feasible solution can be "marked" by its value or fitness for the problem. The chromosome should in some way contain information about solution which it represents. The most used way of encoding is a binary string. Crossover selects genes from parent chromosomes and creates a new offspring. After a crossover is performed, mutation takes place. This is to prevent falling all solutions in population into a local optimum of solved problem. Mutation changes randomly the new offspring. General GA process is as follows: 1) Initialize the population of chromosomes. 2) Calculate the fitness for each individual in the population using fitness function. 3) Reproduce individuals to form a new population according to each individual’s fitness. 4) Perform crossover and mutation on the population. 5) Go to step (2) until some condition is satisfied. GA provides a very efficient search method working on population, and has been applied to many problems of optimization and classification. There are two types of target systems for system optimization: systems whose optimization performances are numerically - or at least quantitatively - defined as evaluation functions and systems whose optimization indexes are difficult to specify. IEC is an optimization method that adopts EC among system optimization based on subjective human evaluation. It is simply an EC technique whose fitness function is replaced by a human user. We can say that the IEC is a technology that embeds
188
M. Virčíková and P. Sinčák
Fig. 3. The process of Genetic Algorithm and Interactive Genetic Algorithm where the fitness function is replaced by the subjective user evaluation [5]
human preference, intuition, emotion, psychological aspects, or a more general term, KANSEI, in the target system. There are many applications when the fitness function cannot be explicitly defined. Interactive evolutionary algorithm uses human’s response as fitness value. This enables algorithm to be applied to artistic domains, and we propose a dance choreography design aid system for humanoid robots using it. The article published by Takagi [5] gives a survey of all achievable knowledge about the Interactive Evolutionary Computation (IEC). The goal of our design of pleasant movements is to create some ’nice motion performance’. However, there is no general standard of ’beauty of motions’, and it is almost impossible to organize fitness function. IGA might be a solution for this. IGA can reflect personal preference because it percepts fitness directly from user instead of computing using some function. Users evaluate robotic motions and give them marks according to their preferences. The process starts from motion performances learned before. After user evaluates the first generation of motions of the robots, the genetic algorithm produces the next generation. We use the tournament selection method. The winner of each tournament is the one with the highest fitness value and is selected for crossover. The crossover is an important source of variation in genetic algorithm. We used a uniform crossover where the bits that are exchanged depend on the masks. The mutation is another source of variation in genetic algorithms. The probability of mutation of every individual is 80 percent but in a very close neighborhood of each bit (less than 10 percent), so the changes are ’small’. Our goal was to discover new frontiers of humanoid behaviors for aesthetic pleasure, entertainment or communication. In the field of developing robotic dances,
Discovering Art in Robotic Motion: From Imitation to Innovation
189
numerous works are presented in [1]. Researchers at the University of Tokyo have developed the learning-from-observation training method, which enables a robot to acquire knowledge of what to do and how to do it from observing human demonstrations. At the Kyoto University apply a method called intermodality mapping to generate robot motion from various sounds and also to generate sounds from motions using the back-propagation through-time algorithm. Other approach from Tokyo University is using Chaos to trade synchronization and autonomy in a dancing robot. Dancing Robot Partner built at the Tohoku University is well-known in the domain of robotics. Their robot acts as a female partner and realizes ballroom dances in coordination with a male human dancer by estimating his intention. Although some of the mentioned systems are examples where interaction between human and robot is the key factor of their success, they are exceptions in the world of robotic dance. The majority of dance systems of robots are only pre-programmed motions and after a while user find them boring. The proposed system in this work interacts with human and the motion is evolving in accordance with his evaluation of the seen behaviors learnt in the first part of the project. We performed an experiment to evaluate the developed system and analyzed if it is a good aid for inexperienced observers to create their own “robotic dance”. The result is that every person can adapt, in our case, the dance choreography, in accordance to his own expectations and preferences.
4 Conclusion In the first part of the project we engage people in face-to-face interaction with a humanoid robot Nao. An inexperienced human teacher can develop new motion tasks via interaction process. In the second part we presented a multi-robot system capable of evolution of motions using an interactive evolutionary computation. We believe that the interactive evolutionary computation will make interactive robots more personal in interacting with humans. The next step is to improve the system of how will the robot store his knowledge. Our global dream is for Nao to acquire its own mental model of people. Currently, he does not reason about the emotional state of others. It means that we want to extract the information about the own preferences of human during his evaluation of the behavior of the robot in the interactive evolutionary algorithm and make this process autonomous. According to Brooks and Breazeal [2], the ability to recognize, understand, and reason about another’s emotional state is an important ability for having a theory of mind about other people, which is considered by many to be a requisite of adult-level social intelligence. Acknowledgement. This work is the result of the project implementation: Development of the Center of Information and Communication Technologies for Knowledge Systems (ITMS project code: 26220120030) supported by the Research & Development Operational Program funded by the ERDF.
190
M. Virčíková and P. Sinčák
References 1. Aucouturier, J.J., Ikeuchi, K., et al.: Cheek To Chip: Dancing Robots and Ai’s Future. IEEE Intelligent Systems 23(2), 74–84 (2009) 2. Brooks, R.A., Breazeal, C., et al.: The Cog Project: Building A Humanoid Robot. In: Nehaniv, C. (ed.) CMAA 1998. LNCS (LNAI), vol. 1562, pp. 52–87. Springer, Heidelberg (1999) 3. Garcia, E., Jimenez, M., A., et al.: The Evolution Of Robotics Research, From Industrial Robotics To Field And Service Robotics. IEEE Robotics And Automation Magazine (2007) 4. Gouaillier, D., Hugel, V., et al.: The Nao Humanoid: A Combination of Performance And Affordability. IEEE Transactions on Robotics (2008) 5. Takagi, H.: Interactive Evolutionary Computation. Cooperation Of Computational Intelligence and Human Kansei. Proceedings of the IEEE 89(9), 1275–1296 (2001) ISSN: 00189219 6. Vircikova, M.: Artificial Intelligence In Humanoid Systems, Diploma Thesis (2010) 7. Fong, T., Nourbakhsh, I., Dautenhahn, K.: A Survey of Socially Interactive Robots: Concepts, Design and Applications, Technical report no. Cmu-Ri-Tr-02-29, Carnegie Mellon University (2002) 8. Dautenhahn, K., Billard, A.: Bringing up Robots or the Psychology of Socially Intelligent Robots: from Theory to Implementation. In: Proceedings of the Third Annual Conference on Autonomous Agents, Agents 1999. ACM, New York (1999) ISBN 1-58113-066-X 9. Berthouze, L., Bakker, P., Kuniyoshi, Y.: Learning of Oculo-Motor Control: A prelude to Robotic Imitation. In: Proceedings of the 1996 Lee/RSJ International Conference on Intelligent Robots and Systems (1996) 10. Koenig, N., Takayama, L., Mataric, M.: 2010 Special Issue: Communication and Knowledge Sharing in Human-Robot Interaction and Learning from Demonstration. Journal Neural Networks 23(9) (2010) 11. Totilo: Natal Recognizes 31 Body Parts, Uses Tenth Of Xbox 360, Computing Resources, Kotaku, Gawker Media (2010) (retrieved: November 25, 2010) 12. Ishiguro, H., Ono, T., Imai, M., Kanda, T.: Development of an interactive humanoid robot “Robovie”—An interdisciplinary approach. In: Jarvis, R.A., Zelinsky, A. (eds.) Robotics Research. Springer, New York (2003) 13. Calinon, S., Billard, A.: Incremental learning of gestures by imitation in a humanoid robot. In: Proceedings of the 2nd ACM/IEEE International Conference on Human_Robot Interactions, HRI 2007 (2007) 14. Ogino, M., Toichi, H., Yoshikawa, Y., Asada, M.: Interaction rule learning with a human partner based on an imitation faculty with a simple visuomotor mapping. The Social Mechanisms of Robot Programming by Demonstration, Robotics and Autonomous Systems 54 (5) (2006)
I/O Performance and Power Consumption Analysis of HDD and DRAM-SSD Hyun-Ju Song1 and Young-Hun Lee2,* 1
Maseer course of Electronic Eng., Hannam University, Ojeong -dong, Daedeok-gu, Daejon 306-791, Korea Tel.: + 82-63-629-8001 [email protected] 2 Dept. of Electronic Eng., Hannam University, 133 Ojeong-dong, Daedeok-gu, Daejon 306-791, Korea Tel.: +82-42-629-7565 [email protected]
Abstract. This paper is to acompared and analyzed power consumption and performance of HDD and DRAM-SSD Storage according to the data I/O accurance. As the analyzed results, the power consumption and performance of DRAM-SSD is small 2.9 times than one of the HDD Storaage at the given performance evaluation condition.And we conform, unit time power consumption was smaller along the data I/O increase. Keywords: SSD, DRAM-SSD, HDD, Storage.
1 Introduction The speed of the CPU increased 570 times during the past 20 years, while the speed of the HDD increased only 20 times. Thus SSD created for to solve the serious data I/O bottlenecks due to the speed difference of the HDD and CPU. Much research has been progressing for increasing the storage device function of the SSD. Data I/O performance difference was found through the DRAM-SSD and HDD storage performance measurement and analysis from the existing research results [1]~[3]. At this point, we will research that how much difference exist between HDD and SSD at the point of view performance and power consumption .
2 Test Environment Configuration The environment for this experiment was constructed as Figure 1, the test target Server is a Linux CentOS 5.3 environment. And the tool for performance measurement used the Postmark. In addition, when the data I/O occurs built in electrodynamometer to measure that power consumption of storage device in power input [4]~[6]. *
Fig. 1. The test environment block diagram for power consumption measurement
HDD storage was used as a EVA4400, and DRAM-SSD storage was used as a TJ128ST. Table 1 is a experimental condition. at Table 1 we measured performance and power consumption such as a three kinds(Low, Medium, High) of Postmark standard measurement conditions. Measuring the Block Size = 4KB, File Size = 327KB measured at a fixed total of 10 times, the average was calculated. Table 1. Experimental condition
ΠΤ ΤΥΞΒΣΜ͑
΅ΖΤΥ͑ΔΠΟ ΟΕΚΥΚΠΟ ΄Φ ΦΓΕΚΣΖΔΥΠΣΪ Ϊ͑
ͷΚΝΖ͑
ΟΦΞΓΖΣ
ΟΦΞΓΖΣ
ͽΠΨ
͢͡
͢͡͝͡͡͡
ͦ͡ ͡͝͡͡͡
;Ζ ΖΕΚΦΞ
͢͡
ͤ͡͝͡͡͡
ͦ͡ ͡͝͡͡͡
ΚΘΙ
͢͡͡
ͨ͡͝͡͡͡
͢͡͡͝͡͡͡
΅ΖΤΥ͑ͽΖΧΖΝ͑
΅ΣΒΟΤΒΔΥ ΥΚΠΟ͑ΟΦΞΓΖΣ Σ
Shell scripts of postmark for Data I/O occurring are same lower part. Set size 327,680 Set number 10,000 Set seed 42 Set transaction number 50,000 Set location /hdd Set read 4096 Set write 4096 Set buffering false Set bias read 5 Set bias create 5 Set report terse Quit Every test in dignity shell script number, transaction number, location part modify accord with each condition.
I/O Performance and Power Consumption Analysis of HDD and DRAM-SSD
193
3 Analysis of Experimental Results Table 2 is a Power consumption measurement results of the storage according to the load occur, the no-load(the data I/O state in storage does not occur) power consumption of HDD storage less approximately 55 percent than power consumption of D-SSD storage, the data I/O increase due to the transfer of storage power consumption increases slightly, but we found that the two storage(HDD and D-SSD) variations in the rate of consumption does not change, suggesting that the data I/O power consumption due to changes in the proportional increase in that can be evaluated. Table 2. Power consumption measurement results ΠΤΥΞΒΣΜ͑ ΅ΖΤΥ͑ΣΖΤΦΝΥ͙ΠΨΖΣ͑ΔΠΟΤΦΞΡΥΚΠΟͫ͑ͼΈ͚ ΅ΖΤΥ͑ͽΖΧΖΝ͑
͵͵
͵͞΄΄͵
ΒΥΖ
ͿΠ͞ΝΠΒΕ
ͣͧͤͩ͟
ͥͨͦͨ͟
ͦͦ͟͡
ͽΠΨ
ͣͧͥͩ͟͢
ͥͨͧͦͦ͟
ͦͦ͟͡
;ΖΕΚΦΞ
ͣͧͥͧͤ͟
ͥͩͥͨͥ͟
ͦͦ͟͡
ΚΘΙ
ͣͩͣͧ͢͟
ͦͩ͟͢͡͡
ͦͦ͟͡
Fig. 2. Power consumption measurement results
Table 3 is a measured result the total running time of the load generation (performance) according to the test condition. At the Low load conditions (Low Level) there was no difference in the performance. But at the High load conditions (High Level), the performance difference was about 5.2 times.
194
H.-J. Song and Y.-H. Lee Table 3. Performance(time) measurement result ΠΤΥΞΒΣΜ͑
΅ΖΤΥ͑ΣΖΤΦΝΥ͙ΡΖΣΗΠΣΞ͑ΥΚΞΖ͑ͫ͑΄ΖΔ͚
΅ΖΤΥ͑ͽΖΧΖΝ͑
͵͵
͵͞΄΄͵
ΒΥΖ
ͽΠΨ
ͦ͢
ͦ͢
͡
;ΖΕΚΦΞ
ͦͤ͢
ͤͥ
ͥͦ͟
ΚΘΙ
ͤͩͩ
ͨͥͤͤ͟
ͦͣ͟
Fig. 3. Performance(time) measurement result
Table 4 is a Total power consumption amount compared result from the load generation total performance time, at this table, when low level that data I/O is small amount, HDD evaluated less power consumption because of no-load power consumption difference. However,along the data I/O amounts is increased ,we found that D-SSD has less Power consumption amount . Table 4. Total power consumption amount compare ΠΤΥΞΒΣΜ͑
΅ΖΤΥ͑ΣΖΤΦΝΥ͙ΡΖΣΗΠΣΞΒΟΔΖ͑ΥΚΞΖ͑Ή͑ΡΠΨΖΣͭΈΙ͚ͯ
΅ΖΤΥ͑ͽΖΧΖΝ͑
͵͵
͵͞΄΄͵
ΒΥΖ
ͽΠΨ
ͤͪͧͣͨ͟͝
ͨͥͩͣͦ͢͟͝
ͦͦ͟͡
;ΖΕΚΦΞ
ͥͥͩͩͤͪ͟͡͝
ͧͥͩͧ͢͢͟͢͝
ͣͥͧ͟
ΚΘΙ
ͪͣͩͩͩ͢͢͟͡͝
ͤͨͨͧͤͩ͟͝͡
ͣͩͪ͟
I/O Performance and Power Consumption Analysis of HDD and DRAM-SSD
195
6QVCNEQPUWORVKQPRQYGT
+'' '66'
/RZ
0HGLXP
+LJK
6GUV.GXGN
Fig. 4. Total consumption power amount compare
4 Conclusion In this paper, power consumption performance is compared and analyzed according to the data I/O of a HDD and SSD storage device, As the analyzed results, the power consumption and performance of DRAM-SSD is small 2.9 times than one of the HDD Storaage at the given performance evaluation condition. And we conform, unit time power consumption was smaller along the data I/O increaseHowever, because of D-SSD are many power consumption of the initial at the no-load, low data I/O and long waiting time service environment HDD is advantageous and D-SSD is a profitable choice at the many data I/O required applications in a short time. Therefore, D-SSD storage is needed a plan that can reduce the total power consumption as the development of low power semiconductor device.
Acknowledgments This paper has been supported by 2011 Hannam University Research Fund.
References [1] Park, K.-H., Choe, J.-k., Lee, Y.-H., Cheong, S.-K., Kim, Y.-S.: Performance Analysis for DRAM-based SSD system. Korean Information Technical Academic Society Dissertation 7, unit 4, 41–47 (2009) [2] Kang, Y.-H., Yoo, J.-H., Cheong, S.-K.: Performance Evaluation of the SSD based on DRAM Storage System IOPS. Korean Information Technical Academic Society Dissertation 7, unit 1, 265–272 (2009)
196
H.-J. Song and Y.-H. Lee
[3] Cheong, S.-K., Ko, D.-s.: Technology Prospect of Next Generation Storage. Korean Information Technical Academic Society Dissertation Summer Synthetic Scientific Announcement Thesis, p. 137 (2008) [4] Solid Data systems, Comparison of Drives Technologys for high-Transaction Database, Solid Data systems, White paper (2007) [5] Bowen, J., Ault, M.: A new architecture that maximizes oracle performance, reliability, and grobal resiliency, TMS white paper (November 2009) [6] Cheong, S.-K., Ko, D.-S.: Load performance evaluation of the SSD according to the number of concurrent users. In: Kim, T.-h., Chang, A.C.-C., Li, M., Rong, C., Patrikakis, C.Z., Ślęzak, D. (eds.) FGCN 2010. Communications in Computer and Information Science, vol. 119, pp. 132–136. Springer, Heidelberg (2010)
A Measurement Study of the Linux Kernel for Android Mobile Computer* Raheel Ahmed Memon and Yeonseung Ryu Department of Computer Engineering, Myongji University Yongin, Gyeonggi-do, Korea {raheel,ysryu}@mju.ac.kr
Abstract. Performance evaluation of the computer system helps in determining how well it is performing and whether any improvements need to be made. Recently, Android has been the fastest growing mobile platform ever in the history of the world. However, the performance of Android platform has not known well. In this paper, we present our project on measuring the performance and the behaviors of Linux kernel on Android mobile computer, and introduce some preliminary results of experiment. The measurement data from this study can be used to improve the performance of Android mobile platform. Keywords: Android, Linux, Mobile Computer, Performance Measurement.
1 Introduction Google with the alliance of about 30 companies introduced Android a new operating system for mobile devices and used Linux kernel as its core. There are many reasons of using Linux kernel for core system services such as security, memory management, process management, network stack, and driver model. The kernel also acts as an abstraction layer between the hardware and the rest of the software stack [1]. Evaluating performance and understanding problems of any system is the best way to improve the efficiency of that system. As Android is using Linux kernel in its core so the kernel must aware of all activities of any application, because they involve system calls or traps. So by evaluating underlying kernel of android one can diagnose the latency issues in the kernel. As previously done work for evaluating performance of different operating systems such as Linux, FreeBSD and Solaris [2], Windows for workgroup, Windows NT, NetBSD [3] on different computer architectures. Also World Wide Web is observed for understanding server performance [4]. In this paper we are contributing the study of behaviors of Linux kernel in Android platform. This study enables the programmers to fetch information about program’s execution, where execution time is important. Using this approach one can diagnose the problems and can improve the system behaviors accordingly. As it is necessary to *
This research was supported by Basic Science Research Program through the National Research Foundation of Korea(NRF) funded by the Ministry of Education, Science and Technology(2010-0021897).
understand how system behavior differs in response to different types of requests. Such as playing audio or video files, browsing and capturing pictures, and other heterogeneous requests and the response of Linux kernel for each request. For recording the flow of instruction, and to observe the overall performance of applications, we are using LTTng; this is a tracing tool for both user and kernel space. We have chosen this tracing because of its capability to generate traces at a high rate with very low overhead, and its visual analyzer, the LTTV viewer. LTTV can handle traces more than 10GB. And it is easy to analyze the problems graphically furthermore it is also capable to analyze text based trace data [5]. The rest of this paper is organized as follows; Section 2 introduces the Android and LTTng architectures. In section 3, we present experiment setup and some experimental results. Finally, we present future work and conclude the paper in section 4.
2 Background 2.1 Previous Works As Linux kernel is traceable with various tracing tools. The popular open source tracers which are currently available are LTTng, Dtrac, Ftrace and SystemTap. LTTng is a best performance analyzing tool, which can show the behavior of Linux kernel or user space in textual and graphical mode [5]. Ftrace derived from two tools latency tracer and logdev used on rt-tree and debugging the Linux kernel. Ftrace is a small utility that uses the frysk engine to trace system calls in a similar manner to strace [6]. Dtrace is dynamic tracing framework created by Sun Microsystems. Originally this tracing tool is developed for Solaris. It can troubleshoot the real time problem in kernel and applications [7]. SystemTap is a project from Redhat, provides a simple command line interface and scripting language for writing instrumentation [8]. 2.2 Android Architecture Android is an open source software platform for high-end mobile phones, It includes an operating system, middleware, user-interface and applications. It intends to be a complete stack of software applications in mobile. Fig.1 is showing the five distinct layers of Android system [1], in Fig 1 the items in green are written in C/C++, items in blue are written in Java and run in the Dalvik Virtual Machine. Linux kernel: Linux kernel is used because it provides a proven driver model for memory management, process management, networking and security models, and in a lot of cases it also provides drivers. Android Runtime: This is consists of two components, Core Libraries and Dalvik Virtual Machine. Core libraries contain collection of all core libraries of java. And Dalvik VM Work likes a translator between the application and operating system. It runs .dex files, .dex files work well on small processors, and these are efficient bitecodes that are the result of converted .jar and .class files. Every application which runs on Android is written in java language.
A Measurement Study of the Linux Kernel for Android Mobile Computer∗
199
Fig. 1. Android Architecture
Libraries: Native libraries written in C and C++, this level is core power of android operating system. These includes the Surface Manager (for compositing windows), 2D(SGL graphics) and 3D graphics(penGL ES) 3D if the driver having 3D chip on it, one can also combine 2D and 3D graphics in the same application. Media Framework like MPEG-4, H264, AE, MP3, and all audio and video codecs need to build a rich media experience, Freetype to render the fonts, the SQL database SQLite, lib and the web browser engine WebKit (opensource browser engine). Application Framework: This is written in java, it is a toolkit that all applications to use, its components are: Activity managers, package managers, windows managers, telephony managers, resource managers, location managers, Notification managers, content providers and view system. Applications: Final layer of software stack on top, where all the applications get written includes the home, contacts, phone browsers, also applications from users, developers resides here. 2.3 LTTng and Its Architecture Linux Trace Toolkit next generation (LTTng) tracer helps tracking the performance and latency issues for debugging problems involving multiple concurrent process and threads, also it is possible to perform tracing of multiple systems with LTTng. LTTng also allows to obtain more than one traces, and enables to view all these traces simultaneously by using LTTV (the viewer of LTTng Traces) [5].
200
R.A. Memon and Y. Ryu
There are three main parts in LTTng: a user space command-line application, lttctl; a user space daemon, lttd that waits for trace data and writes it to disk; and a kernel part that controls kernel tracing. Fig. 2 shows the architecture of LTTng for remote target. lttctl is the command line application used to control tracing. It starts a lttd and controls kernel tracing behavior through a library module bridge which uses a netlink socket [9].
Fig. 2. LTTng Architecture LTTng is a set of patches for modifying linux kernel, integrated to the kernel. LTTng is used along with LTT-control (a toolchain) to control the tracing; controlling and retrieval libraries are available inside LTT-control. The LTTV is a viewer for analyzing and showing the traces, this viewer is capable to handle 10 GB traces, LTTV provides both the textual and graphical mode for displaying the trace information [5, 14].
3 Experiment 3.1 Experiment Setup In this section, we describe the porting of LTTng to ARM architecture. As LTTng is a flexible enough to work with 64, 32 or lower bit hardware [5]. We are using Android Development Platform based on Samsung S5PC100 Application Processor this is ARMv7 architecture using Android éclair 2.1 and Linux 2.6.29 [10]. So for porting LTTng to android, we need compatible version of the LTTng patchset for kernel, and controllers of LTTng which work as a toolchain in filesystem. Furthermore LTTV is needed at host machine for analyzing the traces [5, 12].
A Measurement Study of the Linux Kernel for Android Mobile Computer∗
201
Fig. 3 shows the porting setup, serial Cable required for Hyperterminal and Ethernet connection required for network boot between target and host system. For experimenting with LTTng host machine must have Linux installed on it. Further Table 1 is about all the hardware and software specifications used for this experiment. Kernel instrumentation must include architecture specific events and kernel also must configure with events of LTTng. When the program reaches to the instrumented options of kernel than a new event will be generated. As far as portability of data structures used in LTTng, the tracer is portable between various platforms and usable between different endianness [11, 13].
Fig. 3. Porting Setup
Table 1. Hardware and Software specifications
Hardware/Software Operating system Embedded Device Android Extra Development Board Kernel for Android Compiler Bootloader LTTng Patch set Ltt-control Ltt-viewer Cables
Versions/Specifications Ubuntu 9.04 AchroHD Eclair 2.1 Achro PC 100 Linux-2.6.29 GCC version 4.2.2-eabi u-boot-1.1.6 u-boot-1.3.2 patch-2.6.29-lttng-0.124 ltt-control-0.75-15112009 Lttv-0.12.14-15052009 Ethernet cable and 9 pin parallel cable
202
R.A. Memon and Y. Ryu
3.2 Some Results As the traces collected from Android are needed to analyze by using LTTV, this is viewer of traces, LTTV can be use in two different modes (i.e; Text mode and GUI mode). Text mode use to display information in human understandable statements by using textDump in command shell.
Fig. 4. Color Code of Control flow view
While LTTV uses the graphical view of the traces. LTTV uses different colors for defining each trace as Fig.4 is showing the colors and their representation. And the time stamps of the LTTng are in nanoseconds for providing accurate kernel spends time for the events. Fig. 5 shows the event IRQ (traces are rotated for defining purpose). The IRQ are hardware lines over which device can send interrupt signals to the microprocessor, in figure 5 (a) the entry of IRQ is shown and it is represented by orange color. Fig. 5 (b) shows the exit of IRQ, and also there are other events visible in Fig 5 (b) so those are defined as follows: pink is representing softirq handler event, Pale blue is representing system call event and dark yellow representing waiting for CPU event. Fig. 6 shows another view of traces of IRQ events entry and exit of the events are encircled in red. Here analyzing another traced data from android, due to space restrictions giving just the surface information about IRQ traces. The workload we have selected is a common scenario which is frequently used in everyday application that is playing a video file on Android. We used an mp4 file with length of 194 seconds and size 14.9MB. Measurement performed on same hardware. Table 2 describes that how many events occurred and how much those events consumed the CPU time. This analysis allows tracer to analyze the interrupts and to detect the latency issues during any process.
A Measurement Study of the Linux Kernel for Android Mobile Computer∗
Fig. 5. Entry in IRQ, exit from IRQ and other events
Fig. 6. Entry and exit from IRQ
203
204
R.A. Memon and Y. Ryu Table 2. Analysis of IRQ events
As in first experiment we defined that IRQ are hardware lines over which device can send interrupt signals to the microprocessor. Hence, if those signals are causing the latency issues in Android device than analyzer can analyze them easily using our approach.
4 Conclusion and Future Work In this paper, we presented on-going our project which studies the performance and the behavior of the Linux kernel on Android mobile platform. We also proposed the approach to trace the underlying processes of Android platform. This approach is useful for the developers of android, what is wrong, if developed application is not responding you for some moments or it is taking too much time than expectation, by using this approach one can easily analyze the underlying problems with developed application on android. From collected traces data we selected only IRQ traces to described, while this approach analyze all the traces (syscal, softirq, timer, wait, swap and all other traces) simultaneously. For the future work, we are going to delve into subsystems of the linux kernel such as file system and virtual memory system by using the various performance measuring tools. Also, we are going to study on improving the performance of Android mobile platform by using the measurement data.
References 1. Developers guide of Android, http://developer.android.com/guide/basics/ what-is-android.html 2. Lai, K., Baker, M.: A Performance Comparison of UNIX Operating Systems on the Pentium. In: 1996 USENIX Technical Conference (January 1996) 3. Bradley Chen, J., Endo, Y., Chan, K., Mazières, D., Dias, A., Seltzer, M., Smith, M.D.: The Measured Performance of Personal Computer Operating Systems. ACM Transactions on Computer Systems 14(1) (February 1996)
A Measurement Study of the Linux Kernel for Android Mobile Computer∗
205
4. Almeida, J.M., Almeid, V., Yates, D.J.: Measuring the Behavior of a World-Wide Web Server. In: IFIP HPN 1997: High Performance Networking Conference. White Plains, NY (April 1997) 5. LTTng Project Website, http://www.lttng.org 6. Bird, T.: Measuring Function Duration with Ftrace. In: Japan Linux Symposium (2009) 7. Cantrill, B.M., Shapiro, M.W., Leventhal, A.H.: Dynamic Instrumentation of Production Systems. In: USENIX 2004 (2004) 8. SystemTap, http://sourceware.org/systemtap/wiki 9. Fournier, P.-M., Desnoyers, M., Dagenais, M.R.: Combined Tracing of the Kernel and Applications with LTTng, Symposium (2009) 10. Huins AchroHD PC100, http://www.huins.com/m14.php?m=rd&no=65 11. Desnoyers, M., Dagenais, M.R.: Deploying LTTng on Exotic Embedded Architectures. In: Embedded Linux Conference (2009) 12. Desnoyers, M., Dagenais, M.: LTTng: Tracing across execution layers, from the Hypervisor to user space. In: Ottawa Linux Symposium (2008) 13. Desnoyers, M., Dagenais, M.R.: The LTTng tracer: A low impact performance and behavior monitor for GNU/Linux Instrumentation. In: Ottawa Linux Symposium (2006) 14. Prasad, V., Cohed, W., Eigler, F.C., Hunt, M., Keniston, J., Chen, B.: Location System Problem Using Dynamic Instrumentation. In: Ottawa Linux Symposium (2005)
RF Touch for Wireless Control of Intelligent Houses David Kubat1, Martin Drahansky1, and Jiri Konecny2 1
Department of Intelligent Systems, Faculty of Information Technology, Brno University of Technology, Bozetechova 2, 612 66 Brno, Czech Republic {ikubat,drahan}@fit.vutbr.cz 2 ELKO EP, s.r.o., Palackeho 493, 769 01 Holesov, Czech Republic [email protected]
Abstract. This article describes the principles and capabilities of the new intelligent electronic household control system called RF Control. The main part is focused on the RF Touch control unit, its functions and programming options for the connected devices. Keywords: Control, device, electronic, house, intelligent, remote, RF, touch, unit, wireless.
1 Introduction RF Touch is a new product of the company ELKO EP. This company provides intelligent electronic systems and solutions for a comfortable household control. RF Touch is the main control unit of the new wireless system generation called RF Control. How this unit and the entire system work and what are its capabilities will be described in this article.
2 RF Control System This RF control system [1] allows the user to control and maintain the entire building - from lights and sun-blinds, through heating system, to garage door and garden swimming pool. Every RF unit connected to some specific device like light switch or heating thermoregulator is communicating with the RF Touch control unit by a wireless protocol, so there is no need to damage the walls and strain new wires when installing the system. The RF units are mounted between the original switch and the device, so the function of the primary switch is preserved.
RF Touch for Wireless Control of Intelligent Houses
207
Fig. 1. Visualization of a RF Control system utilization
in various categories based on the theme of their role in the system (e.g. lights or heating). This main unit [3] communicates with the peripheries (sometimes called actuators) and is responsible for all the actions taken in the system. These actions may be invoked either by the user himself (sending real-time commands from any control device connected) or triggered automatically depending on any behavior scheme programmed in the device. 3.1 Unit Specifications The unit itself is manufactured in two versions [2]. The first is a stand-alone type designed to be hung on a wall or laid on a table. It is powered by a 12V DC adapter (2,1mm jack) or by 85-230V AC supply voltage (push-in terminal on the back side). The second type should be installed to the wall electric box with a 230V AC supply voltage. The color of the unit can also be customized as well as the color and material of the outside frame. Communication between the main control unit and the other units (as was mentioned before) is realized via a wireless protocol. This is running on the frequency of 868,5 MHz and has a maximum active range of 200 meters – depending on the construction materials used in the building (the worst in this case is steel/metal, followed by reinforced concrete). The graphical user interface is represented by a 3,5“ touch screen with a resolution of 320×240 pixels and 262.144 colors, with an active white LED backlight. It is possible to check the state of all the connected units (sorted by their function or location) from this screen and send direct commands and control them by only a few taps. The interface also allows the user to update the connected device list, to rename the items and the main thing – to program their function and behavior (more in the next chapter).
208
D. Kubat, M. Drahansky, and J. Konecny
In the standby mode, the screen shows a big clock along with actual date and some more predefined information fields like inside/outside temperature, running heating program etc. In addition, some more common tools like calendar, calculator or message-board can be displayed and used. Last interface included is a micro SD card reader, so the user can backup his programs, update the unit firmware or load some more GUI skins.
Fig. 2. RF Touch control unit
3.2 Device Programming As it is mentioned above, there are various types of connected units in the RF Control system. These RF units can be thematically divided into five groups [3]: heating, switching, lights, sun-blinds and detectors. Each of these categories has a little bit different control interface and also can be programmed in a different way. The heating group contains various thermal sensor and heat regulator units logically divided into rooms, so the system can keep the desired temperature in every room included. In the “initial“ mode there are 3 schemes predefined – economic, common and party – each of them adjusted to a different temperature, which can be changed. Second mode is called “heating program“, and allows the user to set different temperatures to every room, hour and day of the week. The third - “holiday” mode allows exceptions from the previous one for the case of uncommon situations. Switching and lights groups are used to control lights and any other devices, which can be switched remotely (e.g. garden watering system). From the menu, the user can check the status of every switch and manually turn it on or off. There is also an option to set a time offset to specify the moment of the command execution. In addition, the programmable week calendar is also available. When using light units with a darkfall effect, it is also possible to specify the desired brightness level and even the duration of the brightening / darkening effect. This can imitate the feeling of a sunrise or nightfall. Sun-blinds group contains all devices with the end position sensor like marquise or garage door. When initializing such devices, the correct time needed to move from
RF Touch for Wireless Control of Intelligent Houses
209
one end position to the opposite one is measured. After that, the user is able to fully open / close the device or stop in any position between. The programmable week calendar is included as well. The last group of units are the detectors. These simple devices can be installed to the door or window frames, so the user can simply check whether it is closed or not. Another type of this sensor is a motion detector. All these units are supposed to be used together with other devices. For example such detector can turn on the alarm in case of an unauthorized intrusion.
a)
b)
c)
d)
Fig. 3. RF unit examples [2]: a) 6-channel multifunctional switch; b) sun-blinds actuator; c) thermoregulator head; d) various detectors
There is one last programmable function in the RF Control system, which is called “the fast control” [3]. This feature unites various units from all the categories listed above in order to execute several commands by only one touch. For example in the “movie” scenario the sun-blinds are closed, the lights are darkened to some low level brightness level and the room temperature is slightly increased.
4 Remote Controllers Last pieces of hardware in the RF Control system are additional remote controllers – there are two types of them [2]. The first one is small and often paired with a single RF unit. This is called RF Key and for instance can be carried with the car key to open and close the garage door. The second one is called RF Pilot. It is much bigger, has its own OLED display and a 4-way control button. This device partly doubles the main RF Touch control unit, so it can be placed in any room in the house to ensure even more comfortable control from any place the user needs.
5 Conclusion In this paper we have presented a new intelligent system for easy building maintenance. This solution is based on wireless communication between its components. The key role in this system has the main touch screen control unit, which manages all the connected peripheries and allows their programming. The RF Control system increases the living comfort and security in small houses or flats as good as in bigger residences.
210
D. Kubat, M. Drahansky, and J. Konecny
Acknowledgments. This research and development has been done under the support of the project MPO FR-TI1/392 “Multimedial and wireless extension intelligent electroinstallation INELS” and CEZ MSMT MSM0021630528 “Security-Oriented Research in Information Technology”.
References 1. Wang, S.: Intelligent Buildings and Building Automation, p. 264. Spon Press, London (2009) ISBN 978-0415475716 2. RF Control and RF Touch product pages, http://www.elkoep.com/rf-touch-5/ (cited on January 24, 2011) 3. ELKO EP s.r.o.: RF Touch control unit user manual, Holesov, p. 36 (2010)
Facing Reality: Using ICT to Go Green in Education Robert C. Meurant Director, Institute of Traditional Studies • Seojeong College University, Yongam-Ri 681-1, Eunhyun-Myeon, Yangju-Si, Gyeonggi-Do, Seoul, Korea 482-863 Tel.: +82-10-7474-6226 [email protected] http://web.me.com/rmeurant/INSTITUTE/HOME.html
Abstract. The potential of ICT in institutional education is being realized in the implementation of distance and blended learning. The Internet, Smartphone, iPad, Apps, App stores and WiFi networking are providing ubiquitous computing, as e- and m-learning rapidly develop. Meanwhile, Earth suffers from worsening environmental pollution, climate change, and its sixth mass extinction. As environmental consciousness develops, educational institutions seek to adopt responsible environmental practices. One means is to encourage blended learning environments, within which online distance components are then implemented. Integrated scheduling allows students to participate in distance online learning for a significant portion of their study time, and effect environmental savings associated with reduced commuting costs and usage of institutional facilities, while providing increased capacity to service students. Learning Management Systems will likely merge with educational MUVEs, with Open Learning Networks incorporating Personal Learning Environments to provide effective autonomous learning environments that support traditional, blended, and distance learning. Keywords: ICT, EFL/ESL, MUVE, LMS, MALL, PLE, OLN, SLA, education, climate change, environment, sustainability, mass extinction, blended learning, distance learning, e-learning, autonomous learning, convergence, Korea.
communication. EFL/ESL courses offer special potential to help raise awareness and act as a conduit to access a multiplicity of relevant English-language online resources, notably global research and educational material.
2 Environmental Pollution + Climate Change = Mass Extinction 2.1 Environmental Pollution and Climate Change The 2007 Intergovernmental Panel on Climate Change (IPCC) reached consensus that climate change is happening and that it is largely related to human activities [1]. Estimates of global warming during the next century vary, but generally fall in the range of 2°C to 4°C. Rises as high as 7°C are projected for much of the United States and Europe, with even higher temperatures expected in northern Eurasia, Canada, and Alaska. In the UK, the environment ministry Defra, faced with the latest projections that suggest the potential for major change which include southern England being on average 2.2-6.8°C warmer by the 2080s, has recently unveiled climate-proofing plans. These include building roads to the same standards as the scorching south of France; moving fish from the overheated Lake District to cooler waters in Scotland; taking action to protect lighthouses threatened by rising seas; and raising concerns about keeping railway passengers cool in heatwaves, ensuring that rail lines do not buckle in high temperatures, and preventing embankments collapsing as a result of flooding [2]. 2.2 Mass Extinction Events Wake and Vredenberg detail how, in each of five great mass extinctions on this planet, a profound loss of biodiversity occurred during a relatively short period [3]. The oldest mass extinction occurred at the Ordovician-Silurian boundary (≈439 Mya), when approximately 25% of the families and nearly 60% of the genera of marine organisms were lost. The next great extinction was in the Late Devonian (≈364 Mya), when 22% of marine families and 57% of marine genera disappeared. The PermianTriassic extinction (≈ 251 Mya) was by far the worst of the five mass extinctions, when 95% of all species (marine as well as terrestrial) were lost, including 53% of marine families, 84% of marine genera, and 70% of land plants, insects, and vertebrates. The End Triassic extinction (≈199-214 Mya), associated with the opening of the Atlantic Ocean by sea floor spreading related to massive lava floods that caused significant global warming, most strongly affected marine organisms, where 22% of marine families and 53% of marine genera were lost, but terrestrial organisms also experienced much extinction. The most recent mass extinction was at the Cretaceous-Tertiary boundary (≈65 Mya), when 16% of families, 47% of genera of marine organisms, and 18% of vertebrate families were lost. Merely 74,000 years ago, humanity nearly became extinct, as a consequence of the Toba supereruption plunging the planet into a volcanic winter. The world’s population plummeted to perhaps 2,000, a genetic bottleneck from which all modern humans are thought to be descended [4]. And currently, the BBC is drawing attention to the urgent need to alleviate projected global suffering in the next 20 years by increasing worldwide food supplies by 40%, fresh water by 30%, and energy by about 50% [5].
Facing Reality: Using ICT to Go Green in Education
213
2.3 The Sixth Mass Extinction According to Wake and Vredenberg [3], many scientists argue that we are now either entering or in the midst of a sixth great mass extinction, and that this is happening as a consequence of intense human pressure - both direct and indirect - that is having profound effects on natural environments. Habitat destruction and climate change particularly impacts upon narrowly adapted and distributed species, and little time may be left to stave off a potential mass extinction. Scientists from many fields warn of burgeoning threats to species and habitats. The evidence of such threats as human population growth, habitat conversion, global warming and its consequences, impacts of exotic species, and new pathogens, suggests that a wave of extinction is either upon us or is poised to have a profound impact. Unlike previous mass extinction events, the principal cause of the present extinction spasm appears to be human activities, which are associated directly or indirectly with nearly every aspect. The sheer magnitude of the human population has profound implications because of the demands placed on the environment. Population growth, which has dramatically increased since industrialization, is connected to nearly every aspect of the current extinction event. Other (non-human) species have been severely impacted by habitat modification and destruction, which frequently has been accompanied by the use of fertilizers and pesticides. In addition, many other pollutants that are byproducts of human activities have negative effects on other species. Humans have been direct or indirect agents for the introduction of exotic organisms. Furthermore, with the expansion of human populations into new habitats, new infectious diseases have emerged that have real or potential consequences, both for humans, and for many other taxa. Perhaps the most profound impact is the human role in climate change, the effects of which so far may have been relatively small, but which may shortly be dramatic [6]. Extrinsic forces, such as global warming and increased climatic variability, increase the susceptibility of high-risk species that have small geographic ranges, low fecundity, and specialized habitats. Multiple factors acting synergistically contribute to the disease, declines and global extinctions of certain species, and the critical loss of biodiversity that ensues.
3 The Imperative to Go Green If we are to act responsibly as a species, it therefore follows that there is a clear imperative to rapidly adopt informed environmental attitudes, and to work together towards avoiding a global catastrophe. This catastrophe is, I argue, a direct consequence of our global civilization’s widespread denial of reality, immature world-view, and unhealthy lifestyle. Educational institutions need to recognize and must inevitably bear a major responsibility to meaningfully and effectively address these issues. 3.1 Green Content The most obvious way in which EFL/ESL educators can promote green-aware and responsible lifestyles is by including green issues into educational content. Study topics can include green issues, and tasks can be set that address matters relating to environmental consciousness. Oral examination topics I have set pairs of students include:
214
R.C. Meurant
• You are a committed environmentalist, who is very worried about global pollution. Convince your partner to buy a bicycle, then go for an imaginary bike ride together in Seoul or in the country, and talk about what you both experience. • One of you is strongly in favor of the Grand Canal scheme linking Busan to Seoul proposed by President Lee. The other is firmly opposed to it. Debate the issue, and try to convince one another of whether it is a good idea or not. • You have recently learned of the Sixth Mass Extinction. Describe it to your partner. Partner, ask appropriate questions, and suggest what might be the consequences for Korea and for Koreans. • Discuss with your partner the causes and problems of Global Warming, how they are likely to affect you in Seoul and in Korea, and what you might do to deal with the problem. • You are a committed environmentalist, who is very concerned about pollution in Korea. Your partner is a litterbug, who regularly leaves his/her rubbish behind after having a picnic at the beach. Convince your partner to be responsible for the mess he/she makes. Then switch roles. Recent tasks that required students to post online responses have included topics that allow for environmental issues to be introduced and addressed: • Where would you like Korea to be in the next 10 years? How should that happen? • What is the most urgent problem confronting Korea: what should be done about it? 3.2 The Green L2 College A more pervasive and effective means of promoting an environmentally responsible lifestyle might lie in restructuring the educational environment at different levels, so that students, teachers and administrators consciously adjust their lifestyles to live more in accord with green principles. At the institutional level, initially this may not even need to involve ICT, and might simply involve a rescheduling of classes and contact hours to reduce commuting time of staff and students, which with some planning need not imply a reduction in contact hours. Case Study 1: Teacher Commutes to and from a Suburban College. A suburban college employs 12 native English-speaking teachers. Teachers are required to teach 4 classes of 3 hours per week, plus keep office hours of a further 3 hours per week. Most teachers have just one class per day, and so are required to be present 4 days per week, with office hours being worked on a chosen class day. The college is situated on the edge of Seoul, and accessible by public transport by a regular bus service from a nearby subway station. Most teachers live in Seoul, and commute times of teachers range from 40 minutes to 2 hours or more each way. The typical commute time is of perhaps 90 minutes each way, and involves walking or busing from one’s apartment to a nearby subway station, a subway ride with or without transfers, and a bus trip from the destination station to the college. Students also tend to commute for substantial periods, as few live nearby. Analysis. The significant environmental costs associated with requiring staff and students to be physically present in a classroom at particular times could be reduced.
Facing Reality: Using ICT to Go Green in Education
215
Proposed Solution. A restructuring of class times could reduce course hours to one class of two hours per week per student. Most students are of relatively low academic level, and for many, the motivation for learning English is not high. This should therefore entail little loss in educational quality, as teachers agree that the third hour is quite unproductive. Teachers would then be required to teach 6 classes of 2 hours per week (plus office hours as before). The 150% increase in classes would be offset by reducing class sizes to 2/3rds of the present level, which would substantially improve the quality of education. Teachers would face some increase in administrative duties because of the extra number of classes. But the total number of students each teacher managed would remain constant, and as the classes taught by a teacher tend to have the same level, textbook and syllabus, little extra work is involved. A significant improvement for teachers would then lie in rescheduling classes, so that teachers taught 2x2 hour classes per day, and were thus only required to be present for 3 days of the week rather than 4. This provides an incentive for attracting and retaining wellqualified teachers, while reducing the overall commuting time and associated energy costs for teachers by 1/4. Similar savings are possible for student commuting times. 3.3 The Green L2 Classroom As I elsewhere discuss [7, 8], ICT and Convergence are having a radical impact on society, and the effects of educational pedagogy and institutions are profound. Telecommunications and multimedia are providing for highly effective distance learning, communication and collaboration. Most recently, mobile computing among students is becoming ubiquitous, following a progression of cell-phone, SMS and dedicated electronic bilingual/multilingual dictionary, smart-phones, email and instant messaging, and associated apps, laptops, and now tablet computing as exemplified in the Apple iPad [9]. Teachers are gradually adopting e-learning through a progression of home or office use of a computer for preparation and administration, through a teacher’s console in classrooms with associated OHP and Internet access, through intermittent use of fixed computer labs for online tasks, quizzes and exams, to experimental use of student cell phones and smart phones for educational tasks, culminating in laptop or most recently tablet computing, where students use (and may have been issued on enrolment with) personal devices. Blended learning environments are developing to integrate the best qualities of traditional face-to-face learning with online learning, multimedia, access to online resources, and distance communication, and are gradually being adopted. Learning Management Systems are being implemented that integrate administrative duties; provide for social networking between students and potentially between classes whether local or distant; provide access to learning tasks, quizzes and exams; provide access to task uploading and teacher assessment; provide for automatic scoring of quizzes and exams and a medium for efficient teacher grading of set tasks and essay style answers; provide a medium for teacher assessment and grading including overall scores; and provide a means of students accessing their grades while preventing them from viewing other student grades [10]. In that process, the status of the traditional classroom is increasingly being called into question. Give the rapidly developing sophistication of telecommunications, is it
216
R.C. Meurant
still necessary - or even advisable - for students and their teacher to always be present in the same physical space each classtime for effective learning to occur? The learning space is increasingly becoming virtual, as students use course homepages, social networking sites, blogs and online forums to learn from and teach one another and interact with their teacher. While web 2.0 technologies that enable students to create, load and edit content are exposing the limitations of commercial Learning Management Systems (LMS), Bates claims LMS still have major advantages: they provide an institutionally secure environment, enable the management of learning, and integrate with administrative systems [11]. Once blended learning is implemented allowing ready teacher-student and student-student online communication and collaboration, physical presence in the same classroom need not be full-time. Blended learning can incorporate components of online distance learning, which together with appropriate scheduling can reduce the need for commuting, and for physical meeting spaces and services. Naturally, virtualization requires a measure of autonomy on the part of the student, who increasingly needs to take responsibility for her own learning. Students thus partake in a global shift that is taking place from consumer- to participatoryculture, which Jenkins, Purushotma, Clinton, Weigel and Robison describe as having: • • • • •
relatively low barriers to artistic expression and civic engagement; strong support for creating and sharing one’s creations with others; informal mentorship as what is known by the experienced is transmitted to novices; members who believe that their contributions matter; and, members who feel some degree of social connection with one another and who care what other people think about what they have created [12].
Case Study 2: Student Commutes to and from a Suburban College. Using the same example of a suburban college, most students also face lengthy commutes five days a week during semester to and from college, as few students live within close proximity. Substantial brick and mortar facilities are needed to cater for peak usage, but are often unused for substantial periods of the day. Analysis. Offering blended learning across all courses (not just in EFL/ESL) permits all courses to include distance learning components. This helps reduce the number of commutes students are required to undertake, while making for effective use of college facilities that constitute a major investment, such as buildings and classrooms. Proposed Solution. Two solutions are suggested that implement blended courses with integrated schedules and online distance learning components. Courses would therefore employ mixed usage of traditional classroom use and virtual online classes. The virtual online classes would mainly be distance learning, with students not physically present, though some time would be allocated to using virtual facilities in computer labs on campus (with students physically present), for orientation, familiarization with online learning and collaboration, introduction of work segments, presentation and evaluation of course work, and quizzes and exams. This would require students to exercise discipline in participating online in the distance component of courses each week. Given Korea’s very high penetration of high-speed Internet access and ubiquitous supply of PC rooms, this should prove feasible, as those without personal computers have easy access to cheap commercial computer facilities (typically 1,000 won per hour). Virtual online classes would include collaborative as well as
Facing Reality: Using ICT to Go Green in Education
217
self-directed engagement and learning. Students who lived near one another might elect to work together in local groups, taking advantage of the rapid uptake of WiFienabled mobile computing such as hotspots in cafés and home WiFi networking in apartments. With some imagination, a teacher could structure all students into diverse localized groups, and take it in turns to visit and be physically present with one such group each class, while maintaining virtual contact with all groups and students. This could be combined with the topics and themes that are addressed in class, so localities might include (group) visits to museums, art galleries, zoos, exhibitions etc. Such mobility is a key affordance provided by the iPad. The first alternative is to implement blended courses with integrated schedules that require students to physically attend college on just three days a week, but that require online distance participation in courses on the other two days of the week. Classes that involved online distance components would be staggered by course (presuming 5 day-a-week student use of facilities is desired). In this first case, student commutes and requirements for physical facilities such as classrooms could be cut by 40%. The second alternative is again to implement blended courses with integrated schedules, but this time pair classes so that partner classes alternate weekly between physical attendance and distance participation. An effective way to implement this would be to also pair their teachers - so that one would be tasked with traditional teaching and the other with online teaching. The teachers would alternate classes each week, with students alternating weekly between physical attendance and online distance participation. Further, teachers conducting online distance components could in principle do so from home, without themselves needing to commute to college. In this second case, student (and teacher) commutes and requirements for physical college facilities such as classrooms (and offices) could be cut by 50%. In both cases, while significant investment is required to recruit skilled and motivated teachers, and provide them with adequate technological and administrative support, environmental savings would be substantial, and extend to the educational institution as well as to the students. In addition to the considerable time saved in avoiding unnecessary commuting, typical yearly commute fees for a student would be halved: 0.5 x 2 semesters x 2 ways x 16 weeks x 5 days x 1,000 won = 160,000 won per year. Case Study 3: Student Handouts for Courses. Over seven years I have moved my classes towards blended learning environments, but with distance online activities limited to asynchronous tasks. Substantial compromises have been needed, arising from lack of access to adequate institutional computing facilities. Last year I taught four classes of freshman students, with average class sizes of 28 students, for a total of 2 x 4 x 28 = 224 students. Presuming no computing facilities are available and no LMS or online exam management system is used, each class on average requires 28 x 25 = 700 A4 pages for handouts. But in practice significantly more are required, as more handouts are needed for forgetful students, so conservatively estimate 800 A4 pages per class (though wastage is higher). This is 8 x 800 = 6,400 A4 pages per year. Analysis. A significant environmental cost in terms of paper supply, copying, and room cleaning is associated with providing students with physical hardcopy handouts for syllabus and course information, and particular quiz, task and exam instructions and content. Handouts are often left abandoned in classrooms, and students quite often request further copies, which is also rather time-consuming. This is hardly an isolated issue; in
218
R.C. Meurant
December 2010, WWF, the world’s leading environmental organization, announced its .wwf file format in a bid to stop unnecessary printing and wastage of paper. The .wwf format, based on PDF, is unprintable to encourage the avoidance of printed products and printing documents onto paper. Amongst other environmental benefits, this should save trees, notwithstanding union opposition [13]. Proposed Solution. Implement a Learning Management System (such as Moodle) together with an online Assessment System (such as Cognero) to manage content, create and assign tests, and deliver tests through a secure online test center, while providing complete reporting and data dissemination. Early in each semester, create a course homepage in the Moodle LMS, with needed instructions and hotlinks to online tasks, quizzes and exams to be provided when required. (While the instructor is afforded some capability to customize the course homepage, it would be beneficial to use an LMS that affords considerable latitude to students to also customize and decorate the class site, particularly to encourage participation and a degree of ownership of the site pages). Train students to expect to log in to that homepage to locate desired course information and to access tasks, quizzes and exams, and encourage them to bookmark the login page on their PC. To manage classes effectively, use local institutional Student ID numbers for their IDs and for their passwords, but encourage them to edit their profiles, personalize their shared virtual environment, and utilize social networking site capabilities. Design tasks that are to be performed and submitted online e.g. online questionnaires using Google Forms or Moodle Survey activity, or Moodle Q and A forum for blog posts and responses. Where scoring is required (rather than a simple binary determination of whether the task been performed or not), this should preferably be conducted automatically, so make quizzes online, e.g. either use Moodle Quiz activities, or use Cognero to provide online tests using for example Multiple Choice, Multiple Response, Completion, Matching, True/False, Yes/No type questions. Note that Opinion Scale/Likert, Subjective short answer and Essay type questions can also be used, that require the instructor to assess the response rather than use automatic grading; ensure that this assessment can be effectively conducted online (for example, blog posts in Moodle Q and A forums can be graded online as well as responded to, but at least in Moodle 1.9, the process is not particularly userfriendly, requiring a number of quite unnecessary dialogue boxes to be navigated for each grading or teacher response/post).
4 Strategies for Integrating Online Distance Learning With blended learning implemented, online distance learning components can be to: • Simply provide asynchronous online tasks for students to work on individually, in pairs or groups, including online quizzes and exercises such as online crosswords. • Provide links to online content that students are expected to view and respond to. • Provide lecture material online, for students to view; e.g. podcasts that could initially be audio only, but later introduce podcasts in video format. • Require students to use synchronous communication tools, e.g. chat platforms. As d’Eça observes [14], good ones are available on the World Wide Web, and some are free. These can be employed in text, audio or video modes.
Facing Reality: Using ICT to Go Green in Education
219
• Provide MALL (Mobile Assisted Language Learning) programs that employ student-focused, media-rich, flexible and collaborative learning strategies. • Be available online for videochat counseling of individuals or small groups. • Host videoconferencing sessions where students are expected to participate (e.g. SKYPE now allows the hosting of videoconference sessions for a moderate $5 per day or $9 per month fee. Free alternatives for group videoconferencing include iChat (Mac OS X-only), QNext, Tokbox, Tinychat, and Sifonr [15]. • Use, modify or develop educational MUVEs (multi-user virtual environments). • Use OLNs (open learning networks) to integrate PLE (personal learning environments), as suggested by Mott to overcome the limitations of LMSs and PLEs [16]. 4.1 MUVEs for Teaching and Learning MUVEs (multi-user virtual environments), while commonplace to gamers (especially in Korea), also offer considerable potential for substantive teaching and learning, where as Dieterle and Clarke show, they can be used to support the situated and distributed nature of cognition within an immersive, psychosocial context [17]. In general, MUVEs enable multiple simultaneous participants to: • • • • •
access virtual contexts; interact with digital artifacts; represent themselves through avatars; communicate with other participants and with computer-based agents; and take part in experiences incorporating modeling and mentoring about problems similar to those in real world contexts.
Dieterle and Clarke discuss River City as a case study of how MUVEs can be designed to support the situated and distributed nature of learning, thinking and activity; cognition is situated within both a physical and a psychosocial context, and distributed between a person and his/her tools. Distributed cognition and situated learning are complementary and reciprocal. Cognitive processes need no longer be confined within the head of an individual, but include cognitive activities that are distributed across internal human minds, external cognitive artifacts, groups of people, and space and time. The mental burdens of activity can therefore be understood as dispersed physically, socially, and symbolically between individuals and their tools. Through collaborative experiences of teaching and learning from other students and virtual agents in the world, students distribute cognition socially. Cognition can be symbolically distributed through symbolic systems such as mathematical equations, specialized vocabularies and representational diagrams. Concept maps transform thoughts and notions into tangible systems, and may be used as graphical organizers, tools for collaborative knowledge construction, and assessment instruments. Students learn to use the specialized language, customs, and culture of the scientific community to overcome barriers to symbolic distribution of cognition in the classroom resulting from a dearth of language for thinking and need to cultivate a common vocabulary about inquiry, explanation, argument and problem solving. The user’s avatar is also an example of symbolic distribution of cognition, as the virtual, symbolic embodiment of the user within the virtual space. Customizing expression,
220
R.C. Meurant
gestures, facial expressions, clothing and other symbols or symbolisms used to define identity in face-to-face settings are virtually created and projected by participants in MUVEs, and define who or what participants want to be, providing the user with the ability to create singular or multiple online identities, thus allowing her to explore how an individual is recognized or known. In the situated perspective of cognition, learning is a phenomenon that occurs in the course of participation in social contexts. Concepts are not independent; rather, activity, concept and culture are entwined among the physical and social contexts for knowing and understanding. Knowing is an activity, not a thing; always contextualized, not an abstraction; reciprocally constructed between an individual and his/her environment, not defined objectively or created subjectively; and a functional stance based on interaction and situation, not a truth. Cognitive apprenticeships offer a three-part sequence of modeling, coaching and fading. MUVEs change both what and how students learn and teachers teach, and lend themselves to capturing student learning. One reason for developing MUVEs is their ability to leverage aspects of authentic learning conditions that are hard to cultivate in traditional classroom settings. They also allow for the design of situations that are not possible or practical in the real world. MUVEs can offer scenarios with real-world verisimilitude that are safe, cost-effective, and directly target learning goals. Cultures are coupled with the technological advances that evolve with them. MUVEs have become a major force, shaping how we communicate, participate, learn, and identify ourselves. But teaching practices have not changed to embrace such technologies [17]; schools focus on individual performance, unaided thinking, symbolic thinking, and general skills. Cognition outside schools is usually socially distributed and tool use is prominent, involving particularization and contextualization of abstraction, and learning that focuses on situation-specific ideas. The best learning environments for students are authentic, situated, and distributed across internal and external sources; but these conditions are difficult to create in classroom settings. MUVEs can create learning experiences that are not only authentic, situated and distributed, but also provide a context to change standards by which student achievements are judged and the methods by which their accomplishments are assessed. 4.2 Integrating MUVEs from the Present Pedagogical Situation Clearly a full integration of MUVEs into readily available educational technology is yet to occur. Perhaps the uptake of MUVEs in education will parallel the adoption of tablet computing. Tablet computers have been available for quite some years now, but it was not until Apple provided an adequate solution when it introduced the iPad together with a supportive environment of iOS 4, multi-touch technology, apps and app stores that tablet computing has finally taken off. Educational MUVEs have yet to see adequate solutions that offer the required mix of virtual environmental sophistication for the student with ease of design and customization for the teacher. Meanwhile, Dieterle and Clarke’s summation of the affordances MUVEs offer [17] suggests that existent LMS such as Moodle can be regarded as rudimentary MUVEs: they allow multiple participants access to virtual contexts which can be simultaneous; a limited ability to interact with digital artifacts; a very constrained ability to represent themselves as avatars; an ability to communicate with other participants in chat
Facing Reality: Using ICT to Go Green in Education
221
rooms; and the potential to take part in experiences incorporating modeling and mentoring. An interim solution for developing widespread distance learning componentry to blended environments is then to offer a virtual replacement to the traditional classroom that teachers can readily adopt, structure and implement, and students can readily access and operate. So the traditional sense of community in a classroom and ability to collaborate in learning could in transition be provided in a virtual environment, where students express and develop their own identity, and use web 2.0 technologies to structure their classroom in a virtual environment that they come to own. But this is just an interim solution, where new technology is used to create a virtual similitude of the traditional physical reality of the classroom. As we move towards a post-convergence age [18], the virtual environment will inevitably develop beyond simply replicating existent physical forms. One problem to be dealt with in using MUVEs is what we presently encounter as the disparate nature of virtual and physical reality. But as technology evolves, we look towards more integrated relationships, so that the boundaries do not remain hard and distinct, but allow for soft interpenetrations and hazy definitions. In that realization, we move effortlessly back and forth between physical and virtual worlds, as the ubiquitous computing provided by the iPad and WiFi networking already allow. And we envisage an augmented age in which reality is greatly enriched, as we learn to more properly inhabit the infinite possibilities of the virtual realm, within which lie secure islands and reserves of physical reality.
5 Conclusion Faced with critical issues of environmental pollution, global warming, and the sixth mass extinction of species, society urgently needs to adopt a sustainable lifestyle. Educational institutions, courses, teachers, administrators and students can contribute to that needed reorientation, particularly in EFL/ESL. Educational content should address green issues, explore potential solutions, and move aggressively towards integrating ICT for research, collaboration and presentation. Blended learning environments should be utilized, where, together with the provision of effectively ubiquitous computing provided by WiFi networked campuses together with the iPad, environmental advantages of distance e- and m-learning can then be integrated and exploited. While LMS may be regarded as rudimentary MUVEs, the two will likely merge, as sophisticated educational MUVEs are developed to accommodate LMS functions. OLNs may integrate LMSs with PLEs, that learners modify to facilitate their learning [11], [16], to provide effective autonomous learning environments that fully support traditional, blended, and distance learning in an environmentally responsible manner.
References 1. Parry, M.L., Canziani, O.F., Palutikof, J.P., van der Lindern, P.J., Hansen, C.E. (eds.): Contribution of Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge (2007) 2. Shukman, D.: Defra’s UK climate-proofing plans unveiled (2011), http://www.bbc.co.uk/news/science-environment-12302555
222
R.C. Meurant
3. Wake, D.B., Vredenburg, V.T.: Are we in the midst of the sixth mass extinction? A view from the world of amphibians. Proceedings of the National Academy of Sciences of the United States of America 105 (suppl-1), 11466–11473 (2008) 4. Whitehouse, D.: BBC News: When humans faced extinction (2003), http://newsvote.bbc.co.uk/mpapps/pagetools/print/ news.bbc.co.uk/2/hi/science/nature/2975862.stm 5. Ghosh, P.: BBC News: Science & Environment: Report: Urgent action needed to avert global hunger (2011), http://www.bbc.co.uk/news/science-environment12249909 6. Jackson, J.B.C.: Ecological extinction and evolution in the brave new ocean. Proceedings of the National Academy of Sciences of the United States of America 105 (suppl_1), 11458–11465 (2008) 7. Meurant, R.C.: Applied Linguistics and the Convergence of Information Communication Technologies: Collected refereed papers 2006-2009. The Opoutere Press, Auckland (2010) 8. Meurant, R.C.: An Overview of Research on the Effects of ICT and Convergence on SLA. In: Yoon, J.-Y., et al. (eds.) WICLIS-2011: Linguistic Theories and Their Application, 10. Hankookmunhwasa, Seoul. Language Education: #2, pp. 1–11 (2011) 9. Meurant, R.C.: The iPad and EFL Digital Literacy. In: Kim, T.-H., Pal, S.K., Grosky, W.I., Pissinou, N., Shih, T.K., Slezak, D. (eds.) SIP/MulGraB 2010. CCIS, vol. 123, pp. 224– 234. Springer, Heidelberg (2010) 10. Meurant, R.C.: How Computer-Based Internet-Hosted Learning Management Systems such as Moodle Can Help Develop L2 Digital Literacy. International Journal of Multimedia and Ubiquitous Engineering 5(2), 19–26 (2010) 11. Bates, T.: e-learning outlook for 2011 (2011), http://www.tonybates.ca/2011/01/16/e-learning-outlookfor-2011 12. Jenkins, H., Puroshotma, R., Clinton, K., Weigel, M., and Robison, A.J.: Confronting the Challenges of Participatory Culture: Media Education for the 21st Century (2005), http://www.newmedialiteracies.org/files/working/NMLWhitePape r.pdf 13. De Abrew, K.: Planet PDF: Enterprise & Government: Unite demands meeting with WWF over unprintable PDF (2011), http://www.planetpdf.com/enterprise/article.asp?ContentID=Un ite_demands_meeting_with_WWF_over_unprintable_PDF&gid=8145&n l=pp 14. d’Eça, T.A.: The Use of Chat in EFL/ESL. In: Sokolik, M. (ed.) TESL-EJ, vol. 7 (2003) 15. Tabini, M.: MacWorld: MacUser: Five free videoconferencing solutions for Mac users (2011), http://www.macworld.com/article/157478/2011/02/videoconferen cing_alternatives.html?lsrc=top_3 16. Mott, J.: Envisaging the Post-LMS Era: The Open Learning Network. Educause Quarterly 33 (2010), http://www.educause.edu/EDUCAUSE+Quarterly/EDUCAUSEQuarterly MagazineVolum/EnvisioningthePostLMSEraTheOpe/199389 17. Dieterle, E., Clarke, J.: Multi-User Virtual Environments for Teaching and Learning. In: Pagani, M. (ed.) Encyclopedia of Multimedia Technology and Networking, 2nd edn. Idea Group, Inc., Hershey (2005) 18. Clemens, J., Nash, A.: ACVA Australian Centre of Virtual Art: Seven Theses on the Concept of Post-Convergence (2011), http://www.acva.net.au/blog/detail/seven_theses_on_the_ concept_of_post-convergence
A Network Coding Based Geocasting Mechanism in Vehicle Ad Hoc Networks Tz-Heng Hsu1, Ying-Chen Lo1, and Meng-Shu Chiang2 1
Department of Computer Science and Information Engineering, Southern Taiwan University, Tainan, Taiwan, R.O.C. 2 Department of Computer Science and Information Engineering, Far East University, Tainan, Taiwan, R.O.C. [email protected]
Abstract. Geocasting can be used to perform regional broadcast to deliver geographic-related safety, commercials, and advertisements messages. The challenging problem in geocasting is how to deliver packets to all the nodes within the geocast region with high efficiency but low overhead. Network coding is a special in-network data processing technique that can potentially increase the network capacity and packet throughput in wireless networking environments. In this paper, a network coding based transmission architecture for delivering geocast packets over VANET is proposed. The proposed algorithm can increase packet delivered ratio at each mobile node. Keywords: Geocasting, Network Coding, VANET.
may send out packets that are linear combinations of previously received information. The main benefits of network coding are: (1) network throughput improvements and (2) robustness enhancement on data transmission. In this paper, a network coding based transmission architecture for delivering geocast packets over VANET is proposed to increase packet delivered quality with network coding requirement at each vehicle. By equipping network coding technique for geocast packets according to network status, the proposed architecture can enhance the utilization of packet transmission, and can satisfy the requirement of adaptive telematic services over VANETs. The rest of this paper is organized as follows: Section 2 introduces related works about geocasting. Section 3 depicts the proposed network coding based transmission architecture for delivering geocast packets over VANETs. Section 4 shows the experimental results. Section 5 concludes the proposed transmission scheme.
2 Related Works In this section, works related to geocasting are presented. 2.1 Geocasting Geocast protocols can be mainly categorized based on flooding the network or on forwarding a geocast packet on a particular routing path [3]. In [4], Ko and Vaidya present two different location-based multicast schemes to decrease delivery overhead of geocasting packets. The proposed algorithms limit the forwarding space for transmitting packets to the forwarding zone. Simulation results show that the proposed schemes can reduce the message overhead significantly. Meanwhile, it is possible to achieve accuracy of multicast delivery comparable with multicast flooding. In [5], two location-aided routing (LAR) protocols are proposed to decrease overhead of route discovery by utilizing location information for mobile hosts. The LAR protocols use location information to reduce the search space for a desired route, which results in fewer route discovery messages. In [6], Stojmenovic et al. propose geocasting algorithms using Voronoi diagram to forward messages to neighbors who may be the best choices for a possible position of destination. The V-GEDIR routing method is based on determining those neighbors of current that may be closest to a possible location of the destination. The CH-MFR is based on the convex hull on neighboring node. The proposed V-GEDIR and CH-MFR algorithms are loop-free, and have smaller flooding rate compared to other directional geocasting methods. In [7], Boleng et al. propose three mesh approaches for delivering packets to the geocast group: FLOOD, BOX, and CONE. The mesh provides redundant paths between the source and the destination region against host mobility and link failures. In the proposed mesh approaches, the mesh is created in initial step for discovering redundant paths. Nodes inside the destination region will join the mesh when receiving the initial packet. After a unicast reply is sent back to the sender according to the reverse path, the flooding is stopped. The FLOOD approach has the highest control overhead and network-wide data load, but provides the highest level of reliability. The CONE approach has the smallest control overhead and network-wide data load,
A Network Coding Based Geocasting Mechanism in Vehicle Ad Hoc Networks
225
but provides the smallest level of reliability. The BOX approach fall between the FLOOD and CONE approaches. In [8, 9], Ko and Vaidya present a novel protocol called GeoTORA for geocasting packets in ad hoc networks. The GeoTORA protocol is obtained using a small variation on the TORA anycasting protocol. In GeoTORA, flooding is limited to nodes within a small region. GeoTORA can significantly reduce the overhead of geocast delivery. Our transmission scheme differs from above mentioned works in that we try to (1) increase the packet delivery ratio with multi-forwarding zone and (2) recover packet loss caused by broadcast storm within the geocast region at high node density using network coding.
3 The Network Coding Based Geocast Transmission Mechanism In this section, we introduce the traditional geocast transmission mechanism and the proposed network coding based geocast transmission mechanism. 3.1 Traditional Geocast Transmission Mechanism Geocasting is a kind of regional broadcasting, which sends messages to a number of nodes within a specific region. A source node may reside inside or outside the geocast region. When the source node is resided inside the geocast region, the source node floods the packets to all nodes within the geocast region for disseminating locationbased information. When the source node is resided outside the geocast region, a forwarding zone (FZ) is defined as a partitioned network between the source node and the geocast region. The size of the forwarding zone is determined by (i) the size of the geocast region and (ii) the location of the source node. The current location is determined using GPS receivers at each node. When a node outside the geocast region receives a geocast packet and the packet was not received previously, it will forward the packet to its neighbors if it is belongs to the forwarding zone; otherwise, it will discard the packet. Once a geocast packet is reached at the geocast region and a node is received, the node will accept the packet. The node will also broadcast the packet to its neighbors, if it has not received the packet previously (repeated reception of a packet is checked using sequence numbers in the packet header). There are several forms of forwarding zone, e.g., BOX and CONE. In a BOX forwarding zone, the forwarding zone is defined as the smallest rectangle that covers both the source node and the geocast region. In a CONE forwarding zone, the forwarding zone is defined as a cone rooted at source node, such that angle made by the cone is large enough to include the forwarding zone [4]. Consider a source node Server that needs to send geocast packets to all nodes that are currently located within a certain geographical region, i.e. geocast region. Reducing the area of the forwarding zone can reduce control overhead and network-wide data load. However, a geocast packet may be lost due to no node exists in the reduced area of forwarding zone. In a CONE forwarding zone, such case occurs when the angle made by the cone is not large enough. Furthermore, a geocast packet may be lost due to there do not have enough nodes in forwarding zone (FZ) to forward packets to the geocast region. Figure 1 shows an example of packet loss caused by no enough nodes in the
226
T.-H. Hsu, Y.-C. Lo, and M.-S. Chiang
Fig. 1. An example of packet loss caused by no enough nodes in the forwarding zone (FZ)
forwarding zone (FZ). In Figure 1, node A, B, and C cannot receive the geocast packets from Server due to no enough nodes can forward packet to them. Although node D and E can receive packets from Server, however, node D and E drop the received packets and will not forward the packets to other nodes, e.g., B and C, because of node D and E are not belonging to the forwarding zone. Forwarding zone adaptation can solve the above mentioned issues. In a CONE forwarding zone, the forwarding zone can be expanded into a BOX forwarding zone to avoid packet loss caused by the small angle made by the cone. In a BOX forwarding zone, a parameter can be used to extend the forwarding zone [4]. When is positive, the rectangular forwarding zone is extended. Figure 1 illustrates an example of transmitting geocast packets using extended forwarding zone. In Figure 2, node A cannot receive the geocast packets from Server due to no nodes can forward packet to it. However, node B and C can receive geocast packets from node D and E because of node D and E are belonging to extended forwarding zone in this case. Node B and C then broadcast the geocast packets to all nodes within the geocast region. In case of still no node exists in the extended forwarding zone, the source node can use the simplest flooding mechanism to flood geocast packets to the destined geographical region. Source node broadcasts the geocast packet to all its neighbors. When a node receives the packet, compares the geocast region's coordinates with its own location. If the node’s location is within the geocast region, the node will accept the packet. The node will also broadcast the packet to its neighbors, if it has not received the packet before. If the node is located outside the geocast region and the packet was not received previously, it just broadcasts the packet to its neighbors. The flooding mechanism can increase the possibility that packets can be relayed to the destined geocast region. Meanwhile, the flooding mechanism is simple and easy to implement. However, it will increase the transmission overhead and network-wide data load as a trade-off between transmission efficiency and packet reachability.
δ
δ
A Network Coding Based Geocasting Mechanism in Vehicle Ad Hoc Networks
227
3
D
Fig. 2. The proposed multi-forwarding zone (MFZ) geocasting mechanism in a CONE forwarding area
3.2 The Proposed Multi-Forwarding Zone Geocasting Mechanism In urban areas, VANET may suffer from greatly high node density. To increase the throughput of geocast packets that reached at the geocast region with high node density, we propose a geocasting mechanism using the multi-forwarding zone concept. The proposed multi-forwarding zone (MFZ) geocasting mechanism splits the original forwarding zone in to two or more small forwarding zones. The simplest way is to draw a virtual cutting line from the center of the geocast region to the source node. The virtual cutting line splits the original forwarding zone into two small forwarding zones: upper forwarding zone (UFZ) and lower forwarding zone (LFZ). The size of upper forwarding zone (UFZ) and lower forwarding zone (LFZ) are the same. For nodes that are equipped antennas with multi-input multi-output (MIMO), geocast packets can be forwarded simultaneously using the upper forwarding zone (UFZ) and lower forwarding zone (LFZ). For nodes with single antenna, geocast packets can be forwarded interleavedly using the upper forwarding zone (UFZ) and lower forwarding zone (LFZ). Figure 2 illustrates the proposed multi-forwarding zone (MFZ) geocasting mechanism in a CONE forwarding area. In Figure 3, the upper forwarding zone (UFZ) is responsible for delivering geocast packet Pa; the lower forwarding zone (LFZ) is responsible for geocast packet Pb. The throughput of geocast packets that reached at the geocast region will be increased by using the proposed multi-forwarding zone (MFZ) geocasting mechanism. In case of no enough nodes exist in the extended multi-forwarding zone (MFZ), forwarding zone adaptation policy is adopted to increase the possibility that packets can be relayed to the destined geocast region. In a CONE multi-forwarding zone, the multi-forwarding zone can be expanded into a BOX multi-forwarding zone to avoid packet loss caused by the small angle made by the cone. In a BOX multi-forwarding zone, we adopt the parameter to extend the multi-forwarding zone as listed in [4]. Figure 3 shows an example of delivering geocast packets using a BOX multi-forwarding zone (MFZ). If there still no node exists in the extended multiforwarding zone (MFZ), the source node will use the simplest flooding mechanism to
δ
228
T.-H. Hsu, Y.-C. Lo, and M.-S. Chiang Network Space G
H
F Forwarding Zone
Pa
C
Pa
I
Pa
A Pa
Geocast Region
Server Pb
Pb
D
Pb
B
Pb
E
Fig. 3. An example of delivering geocast packets using a BOX multi-forwarding zone (MFZ)
flood geocast packets to the destined geographical region. The flooding mechanism can increase the possibility of packets that can be relayed to the destined geocast region. 3.3 Network Coding with Multi-Forwarding Zone (MFZ) Geocasting Mechanism Using the multi-forwarding zone (MFZ) geocasting mechanism can transmit packets to the geocast region quickly and efficiently. However, it will cause the broadcast storm when too many geocast packets are arrived at the destined geocast region in a certain period. In the geocast region, when multiple nodes receive a geocast packet, these nodes will all re-broadcast the packet to its neighbors if the packet has not been received before. When multiple nodes re-broadcast the packet to its neighbors simultaneously, it will make severe contention on the channel. Blindly flooding induces the broadcast storm, which wastes precious bandwidth by sending redundant packets that will probably collide. Table 1. The estimated number of recoverable packets using network coding in node A 1HWZRUN&RGHG3DFNHW У У У УУ
0D[LPXP5HFRYHU1XPEHU
A Network Coding Based Geocasting Mechanism in Vehicle Ad Hoc Networks
D
E
F
G Fig. 4. An example of packet recover using network coding
229
230
T.-H. Hsu, Y.-C. Lo, and M.-S. Chiang
Fig. 5. The topology of the simulated geocast environment
In order to recover packet loss caused by broadcast storm within the geocast region, we adopted network coding technique to reduce the number of retransmissions using the minimum number of network coded packets. Figure 4 illustrates an example of packet recover using network coding for nodes within the geocast region. In Figure 4 (a), node A and B broadcast packet 1 to its neighbors simultaneously, which makes severe contention on the radio channel. Only nodes C and D can receive the packet 1 due to collisions. In Figure 4 (b), Node A then broadcasts packet 2 to its neighbors, however, the hidden node problem occurs when node E is receiving packets from other nodes outside the geocast region. Node E does not receive the packet 2 from node A. In Figure 4 (c), node A continues to broadcasts packet 3 to its neighbors. A packet collision occurred due to node C is sending a geocast packet at the same time. Only node B and E receive the packet 3. In Figure 4 (d), node A snoops on the medium, obtains the status of its neighbors, and detects coding opportunities. In such a case, node A have 7 combinations to recover the lost packets: single packet 1, 2, and 3; network coded packet 1У2, 2⊕3, 1У3, and 1У2У3. If node A broadcasts a single packet to recover lost packets, the estimated number of recoverable packets using network coding is listed in Table 1. According to Table 1, node A will code as many packets as the recipients can recover the lost packets, i.e., network coded packet 1У3. Node A then broadcasts network coded packet 1У3 to its neighbors, node B and E can recover packet 1, node C and node D can recover packet 3 after receiving the network coded packet 1У3. If node A snoops on the medium and find that other nodes broadcast the network coded packet 1У3 before its scheduled sending time, node A will select the network coded packet with 2nd maximum recover number as the next coded recover packet. If the network coded packets in queue all have the same maximum recover number, the node will select a network coded packet in a random manner and then broadcasts the network coded packet to recover lost geocast packets.
4 Performance Evaluations To evaluate the performance of the proposed algorithms, we use the NS2 network simulator to simulate the geocast environment. There are 19 mobile nodes in a 5000 meter by 5000 meter grid in the simulation environment; 9 mobile nodes are within the forwarding zone; 10 mobile nodes are within the geocast region. We generated a scenario file with 1 source node, which sends out a 512 bytes packet per 10ms using
A Network Coding Based Geocasting Mechanism in Vehicle Ad Hoc Networks
231
Fig. 6. The results of delivering geocast packets using multi-forwarding zone (MFZ) mechanisms
constants bit rate (CBR) generator. The parameters in the simulated IEEE802.11b WLAN environment are based on Orinoco 802.11b Card [10]. The simulation runs with durations of 30 seconds. To simulate the mobility of mobile nodes, two nodes will move inside the geocasting region at time 10 and 15. A node will move out the geocast region during time 13 to 20. Nodes within geocast region will perform network coding if the nodes receives the packet recover request from its neighbors; the packet recover request will be put into a recover request queue. A node will perform the network coding after 0.3 seconds, which can gather more recover requests. The node then codes as many request packets as the recipients can recover the lost packets. Comparisons of multi-forwarding zone (MFZ) mechanisms with/without network coding are performed. Figure 5 shows the topology of the simulated geocast environment. Figure 6 shows the results of delivering geocast packets using multi-forwarding zone (MFZ) mechanisms. The x-axis denotes the simulation time and the y-axis denotes the average packet throughput in kbytes. The red-line denotes the simulation result of delivering geocast packets using multi-forwarding zone (MFZ) mechanism without network coding, the green-line denotes the simulation result of delivering geocast packets using multi-forwarding zone (MFZ) with network coding. In Figure 6, the source node sends geocast packets to the geocast region. When nodes in the geocast region receive the geocast packets, these nodes then re-broadcast the geocast packets to its neighbors. The packet throughput increases rapidly at beginning and then drops at time 5 due to packet loss that caused by severe contention on the radio channel. In Figure 6, the green-line stays in top of red-line, which means that delivering geocast packets using multi-forwarding zone (MFZ) with network coding can have higher average packet throughput.
232
T.-H. Hsu, Y.-C. Lo, and M.-S. Chiang
5 Conclusion By equipping network coding technique for geocast packets according to network status, the proposed architecture can enhance the utilization of packet transmissions. Simulation results show that the proposed multi-forwarding zone (MFZ) mechanism with network coding can have higher average packet throughput. Acknowledgments. This research is supported by the National Science Council of the Republic of China under the grant NSC 99-2221-E-218 -018.
References 1. Ahlswede, R., Cai, N., Li, S.-Y.R., Yeung, R.W.: Network information flow. IEEE Transactions on Information Theory 46(4), 1204–1216 (2000) 2. Li, S.-Y.R., Yeung, R.W., Cai, N.: Linear network coding. IEEE Transactions on Information Theory 49(2), 371–381 (2003) 3. Maihofer, C.: A Survey of Geocast Routing Protocols. IEEE Communications 6(2), 32–42 (2004) 4. Ko, Y.-B., Vaidya, N.H.: Geocasting in Mobile Ad Hoc Networks: Location-Based Multicast Algorithms. In: 2nd IEEE Workshop on Mobile Computing Systems and Applications, New Orleans, pp. 101 – 110 (1999) 5. Ko, Y.-B., Vaidya, N.H.: Location-Aided Routing (LAR) in Mobile Ad Hoc Networks. Wireless Networks 6(4), 307–321 (2000) 6. Stojmenovic, I., Ruhil, A.P., Lobiyal, D.K.: Voronoi diagram and convex hull based geocasting and routing in wireless networks. In: 8th IEEE International Symposium o Computers and Communication, Canada, pp. 51–56 (2003) 7. Boleng, J., Camp, T., Tolety, V.: Mesh-based geocast routing protocols in an ad hoc network. In: 15th International Parallel and Distributed Processing Symposium, San Francisco, pp. 1924–1933 (2001) 8. Ko, Y.-B., Vaidya, N.H.: GeoTORA A Protocol for Geocasting in Mobile Ad Hoc Networks. In: 8th International Conference on Network Protocols (ICNP), Osaka, Japan, pp. 240–250 (2000) 9. Ko, Y.-B., Vaidya, N.H.: Anycasting-Based Protocol for Geocast Service in Mobile Ad Hoc Networks. Computer Networks Journal 41(6), 743–760 (2003) 10. Xiuchao, W.: Simulate 802.11b Channel within NS2 (2004), http://www.comp.nus.edu.sg/~wuxiucha/research/reactive/ report/80211ChannelinNS2_new.pdf
An In-network Forwarding Index for Processing Historical Location Query in Object-Tracking Sensor Networks Chao-Chun Chen1 and Chung-Bin Lo2 1
2
Dept. of Comp. Sci. & Info. Engr., Southern Taiwan University, Taiwan [email protected] Dept. of Comp. Sci. & Info. Engr., National Cheng-Kung University, Taiwan [email protected]
Abstract. Object tracking in wireless sensor networks is critical for many applications, e.g., military and biology, and has been studied in past years. Based on the object tracking technology, many services become feasible in wireless sensor networks, such as location management and nearest query. However, most services in the related work are for retrieving the current information of a moving object or sensor network. In this paper, we investigate the processing of historical location query (HLQ) in the wireless sensor network. We propose the two-tier HLQ processing architecture to management the object locations: the first tier, the database server, is a query gateway and maintains volumes of object location history; the second tier, in-network forwarding index, maintains the recent location history of a moving object to save communication cost by using aggregation techniques. Then, we invent the forwarding indexbased query processing scheme for history location queries. We conduct a primary version of experiments to observe the performance characteristic of our proposed method. The results reveal that the proposed method is indeed effective for processing HLQ. Keywords: wireless sensor network, query processing, spatio-temporal data management, forwarding, index design.
1
Introduction
Object tracking in wireless sensor networks is critical for many applications, e.g., military and biology, and has been studied in past years [1]. Based on the object tracking technology, location-based services become feasible in wireless sensor networks, such as location management and nearest query. For example, a biologist equips a sensor node on a wild animal which lives in a monitored sensing field, and thus, he/she can look up the movement trajectory of the wild animal for past hours. The historical location query (HLQ) has been studied in the traditional spatial database [12,13]. When a user issues a HLQ to a traditional spatial database, the database server would find out the result of the HLQ from the spatial database. T.-h. Kim et al. (Eds.): UCMA 2011, Part I, CCIS 150, pp. 233–242, 2011. c Springer-Verlag Berlin Heidelberg 2011
234
C.-C. Chen and C.-B. Lo
The previous research on processing HLQ focuses on minimizing the number of disk accesses or increase the scalability of the number of data tuples. However, in wireless sensor networks, due to the poor resource of sensor nodes (e.g., limited power capacity, communication, storage, and processing ability), the data obtained from sensor nodes are preferred to be maintained inside the sensor network [11,9] in order to reduce the frequent data communication. Particularly, a good communication control scheme can efficiently prolong the network life, because the communication is the most critical component for energy consumption. Hence, processing HLQ in wireless sensor networks needs to minimize the transmission cost. Moreover, since sensor storage is limited, the great amount of obtained locations is not possible to be permanently stored in wireless sensor network. Thus, a mechanism for managing location data is needed as well. In this paper, we investigate the processing of historical location query (HLQ) in the wireless sensor network. We consider two questions which are incurred by wireless sensor networks: the first is to deal with the great amount of obtained locations; the second is to offer HLQ service with minimal communication cost. In order to solve the first question, we proposed the two-tier HLQ processing architecture to integrate the sensor storage and the central storage. We treat the sensor storage as a buffer of the central database. Hence, the obtained locations can be maintained in the sensor network, and there is no need to transmit data back to the server each time obtaining a data item. Hence, the communication cost is reduced. In other words, the first tier, the database server, maintains volumes of object location history; the second tier, in-network forwarding index, maintains the recent location history of an object to save communication by using aggregation techniques. In order to solve the second question, we proposed the forwarding index-based query processing method (FIQP) to maintain the obtained location in wireless sensor network, and answer HLQ with minimal communication cost. We conduct a primary version of experiments to observe the performance characteristic of our proposed method. The results reveal that the proposed method is indeed effective for processing HLQ.
2
Preliminaries
A wireless sensor network consists of hundreds or thousands of sensor nodes. These nodes can communicate each other by using a multi-hop protocols [11]. A sensor node basically has five components: computing component, sensing component, transmission components, storage components, and power component [1]. In other words, a sensor node can use the sensing component and communication component to obtain the locations of an object based on the localization methods[4,5]. Each location can be saved to the storage component. After collecting certain locations, a sensor node can aggregate locations by using its computing component, and transmit data to a remote node by using the transmission component. Since the sensor node can localize an object and transmit data to a remote node, the object tracking issue emerges and is studied in recent years. In the
An In-network Forwarding Index for Processing Historical Location Query
gateway
Internet
sensing unit
235
processing unit processor storage
S2 transceiver
S1
power unit
Multi-hop communication
Fig. 1. A reference architecture of wireless sensor networks
object-tracking sensor network, sensor nodes cooperatively track the object so that users can continuously extract the locations of the object, even through Internet. Many object tracking schemes have been proposed for wireless sensor networks, such as PES [4]. All these schemes aim at minimizing the sensor energy consumption on tracking the object. Although the sensor storage is limited, the storage of sensor nodes can be united as a large-size distributed storage by the network connection. The management of the distributed sensor storage is studied in [10]. Basically, the design of the distributed storage retrieval is based on the application characters. Because the sensing data is similar in the same region, the cluster head aggregates the sensing data to minimize the storage.
3
Two-Tier HLQ Processing Architecture
Figure 2 shows the two-tier HLQ processing architecture. In the architecture, two components are designed to maintain the location history of moving objects: one is the query gateway, the other is the in-network storage. The query gateway equips with a central database that is used to permanently maintain the location history of moving objects. The query gateway is assume to be a fixed station, thus, it has powerful resources (e.g., computation and storage) and need not consider the energy consumption problem. The in-network storage use the storage of sensor nodes to maintains the location history of moving objects. Since a sensor network is naturally a distributed environment, the in-network storage across sensor nodes is with different network topology (we use tree topology in this paper [6,7]), and needs communication cost to the retrieve history data. The location history of moving objects are maintained by the two-tier architecture to optimize the energy consumption of the sensor network. The earlier location history is maintained in the query gateway to avoid the memory overflow in the sensor network, and this can reduce the energy consumption for querying location history as well. On the other hand, we maintain the more recent location history in the in-network storage to reduce the energy consumption on transmitting location data to the query gateway after movements. By the cooperation between the query gateway and the in-network storage, the architecture
236
C.-C. Chen and C.-B. Lo Q[ts,tm]
gateway
Internet database
Q[tm,te]
Fig. 2. Two-tier HLQ processing architecture
can minimize the communication cost for both answering location query and managing object movement in the sensor network. When acquiring location history of a moving object, a request Q[ts , te ] is issued and sent to the query gateway, where ts is start time and te is end time. Let tm be the last timestamp of location history in the central database of the query gateway. Thus, if te ≤ tm , such query can be process by using traditional query processing methods in the central database. For general cases, we assume ts < tm < te 1 . The query gateway divides the query period into two sub-periods [ts , tm ] and [tm , te], and processes the two sub-queries Q[ts , tm ] and Q[tm , te ] in the query gateway and the in-network storage, respectively. The sub-query Q[ts , tm ] is processed in the central database, and thus, it can be answered by using existing methods in the spatial database research area, such as [12,13]. The sub-query Q[tm , te ] is processed in the wireless sensor network. The in-network storage will retrieve the requested data from the distributed sensor nodes. The details of the query processing will be presented in Section 4. When an object moves across territory of the current node, the locations the movement are designed to be maintained the in-network storage so that these recent history data can be aggregated while they are transmitted to the central database in the future. A movement is processed as follows, and is illustrated in Figure 3. When the object is first detected in the sensor network, the detecting node reports the location and the object id to the query gateway. The first node of the forwarding index is called the anchor, because the query gateway searches for the location history through the anchor. When the object moves to the territory of the next node, the new node would inform the previous node to leave a forwarding link so that the query can follow forwarding links to retrieve the location history. Notice that the aggregation techniques can be applied to the history data not only over distributed nodes, but also in a single node. If an object resides in the territory of a node for long time, the node could obtain certain locations for sampling periods. Under such situations, the node can aggregate the locations to save the sensor storage [3]. For example, a nodes maintain five locations of 1
If tm ≤ ts , the query is completely processed in the sensor network. The queries belonging such cases can be processed under our proposed scheme as well.
An In-network Forwarding Index for Processing Historical Location Query
237
query gateway query-delivery path
E D A
anchor
t0
B
C t1
t2
t3
building forwarding links for movements
Fig. 3. Building forwarding links for object movements
object o1 as (t1 , (x1 , y1 )), (t2 , (x2 , y2 )), (t3 , (x3 , y3 )), (t4 , (x4 , y4 )), (t5 , (x5 , y5 )). The five location can be aggregated as a linear regression model to represent the movement pattern, and can be represented as the following form: (o1 , [t1 , t5 ], y = ax + b) In this manner, the data size can be greatly reduced. Also note that too long forwarding links would increase the communication cost in querying locations of a moving object. Thus, a mechanism of reducing the length of forwarding links is necessary needed to improve the performance of the sensor system. The reduction of forwarding links is an optimization issue, and needs to consider cost analysis of the sensor system. We tackle the issue in the following sections.
FIQP includes two functions: (1) transmitting the location history to the query gateway and (2) retrieving the object locations of a given period. The first function is used to deal with the object movements, and aims at reducing the the energy consumption when the object incurs too long length of the forwarding index. The second function, retrieving the object locations, is used to answer a location query. In order to reduce the energy consumption, we transmit the location history while a location query is injected to the sensor network, that is, the location history transmission and location query are combined together. Such design principle is like the piggyback technique in the networking research topic. Notice that the condition of executing the location history transmission is triggered by the movements of the object. Thus, we need the history-dumping flag fdump and a history-dumping evaluator to determine whether it is worth to transmit the location history to the query gateway. The history-dumping evaluator is to
238
C.-C. Chen and C.-B. Lo
evaluate whether the cost saving in the history collection is greater than the in-network query cost. If the above condition is hold, the history-dumping flag fdump is set to true; otherwise, fdump is set to false. Then, when the next location data arrives and fdump ==true, the location history in the sensor network will be sent back to the central database. The details of history collection cost and query cost will be derived in Section 5. Retrieving data from the forwarding index in the wireless sensor network is illustrated as follows. Assume the query gateway divides the query Q[ts , te ] into two sub-queries Q[ts , tm ] and Q[tm , te ], as shown in Figure 2. Q[ts , tm ] is processed in the central database, and Q[tm , te ] and the history-dumping flag fdump are delivered to the anchor in the wireless sensor network. Remind that the history collection and the requested location retrieval are combined together to save the communication cost as mentioned above. Thus, if the history-dumping flag fdump is true, the location history from the anchor to the last node of the forwarding index are collected by using aggregation techniques and transmitted to the central database through query gateway. After executing this step, all history data have been maintained in the central database. Hence, the forwarding index is reset and the locations in previous nodes are erased. On the other hand, if the history-dumping flag fdump is false, only the locations of the requested period is collected and transmitted to the query gateway/central database. The HLQ Q[tm , te ] collects location data from the node containing locations of tm to the node containing locations of te by following the forwarding links from the anchor. After the location data in [tm ,te ] are collected and transmitted, the query gateway merges the location data from the sensor network (i.e., [ts ,tm ]) and those from the central database (i.e., [tm ,te ]) as the result of the HLQ Q[ts , te ].
5
Cost Analysis
Recall that we compare the history collection cost and the query cost to decide whether the location data in the sensor network should be collected and transmitted to the central database when next query arrives. In this section, we derive the above two costs so that a sensor node can apply FIQP scheme at run time. 5.1
History Collection Cost
The history collection cost is the communication cost used to collect the location history in the sensor network to the central database. The history collection cost consists of three parts: (i) the history collection operation is delivered from the query gateway to the anchor, (ii) the location data are aggregated and forwarded from the anchor to the current node, (iii) the aggregated data are transmitted from the current node to the query gateway. Thus, the history collection cost can be represented as follows. history collection cost = (cost from query gateway to anchor) + (cost from anchor to current node)+(cost from current node to query gateway)
An In-network Forwarding Index for Processing Historical Location Query
239
The cost estimations of the above terms are derived as follows. For the first part in the above equation, since the history collection command is small and can be encapsulated into a transmission packet, the cost from query gateway to anchor can be calculated as the one-hop communication cost times the number of nodes between the query gateway and the anchor (that is, the length of the query-delivery path shown in Figure 3), i.e., dist(query gateway, anchor) × α, where α is communication cost for transmitting a packet from a node to its one-hop neighboring node, and dist(a, b) means the number of hop count from sensor node a to sensor node b. For the second part, the location data of a node, say ni , are aggregated with the previous aggregated data and the aggregated data (whose size is represented by using aggregate(tm , tni )) would be transmitted to the next node, ni+1 . Since the process of the location data collection would scan all the nodes in the forwarding chain, the total cost of the second part summarizes the communication cost in each node of the forwarding k−1 chain, that is, i=0 dist(ni , ni+1 ) × aggregate(tm , tni ) × α. For the third part, all location data (whose size is represented as aggregate(tm , te )) in the current node are transmitted to the query gateway, thus, the communication cost is equal to dist(current node, query gateway) × aggregate(tm , te ) × α. Therefore, history collection cost can be further represented as follows. history collection cost = (dist(query gateway, anchor) × α) + (
The above form will be used to estimated the history collection cost in our simulation, shown in Section 6. 5.2
Query Cost
The query cost is the communication cost used to access the latest location of the requested object in the sensor network. The query cost consists of three parts: (i) the communication cost from query gateway to anchor, (ii) the communication cost from anchor to the last node, (iii) the communication cost from the last node to the query gateway. Thus, the query cost can be represented as follows. query cost = (dist(query gateway, anchor) × α) + k−1
In our simulation, we divide the sensor field into 512 × 512 grids, and the edge length of a grid is 17 meters. We deploy at least one sensor node to each grid for tracking moving objects. For each sensor node, the sensing range and the
240
C.-C. Chen and C.-B. Lo
radio range are set to 40 meters. Additionally, the multi-hop routing algorithm, GPSR [2], is applied to transmit message between sensors. To model the moving patterns of the target, the random walk model is used to generate target trajectories. The velocity is set to 1 meter/second. The query arrival rate is 0.5, that is, the query arrives every two movement on average. In the experiments, we compare our proposed scheme to the central storage (CS) scheme and DIST scheme [7]. 6.1
Effect of Simulation Time
total cost ( logscale (messages) )
Figure 4 depicts the performance of various schemes over different simulation times which varies between 5000 and 30000 seconds in the horizontal axis. The vertical axis is the total cost which is measured as the number of messages. From the result, FIQP performs more outstanding than other three schemes. This shows that FIQP efficiently maintain certain trajectory data in the innetworking storage to reduce the number of messages for object’s movements, while, at the same time, successfully alleviate the increasing number of messages for querying object’s historical locations. In addition, the difference of FIQP and related schemes does not reduce as simulation time. This means FIQP performs quite stable, and is suitable for long-running applications. 5 × 108
5 × 107
5 × 10
6
5000
CS DIST FIQP 10000 15000 20000 25000 simulation time (sec)
30000
Fig. 4. Effect of Simulation Time
6.2
Effect of Velocity
Our second experiment studies the effect of velocity for various schemes, and the results are shown in Figure 5. We vary the velocity from 0.5 to 2.5 meters/second. As our expect, the total costs of three schemes, except CS, increase as increasing velocity. This is because these three schemes need additional communication cost to track the object if the query period includes the current time, and the amount of needed communication messages is proportional to the velocity. Under high velocity (i.e., velocity is greater than 2 m/s), CS even becomes the best performance, while other three schemes spends much communication cost for
total cost ( logscale (messages) )
An In-network Forwarding Index for Processing Historical Location Query
241
5 × 109
5 × 108
5 × 10
CS DIST FIQP
7
0.5
1
1.5 velocity (m/s)
2
2.5
Fig. 5. Effect of Velocity
dealing with high velocity. Note that since CS transmits the object’s locations to the central server immediately after each movement of the object, thus, the velocity incurs the least impact for CS. In the figure, we can see the curve of CS is very close to a horizontal curve.
7
Conclusions and Future Work
In this paper, we study the history location query in the wireless sensor network. In order to avoid a great amount of communication on data transmission in wireless sensor network, we first propose a two-tier architecture which can efficiently maintain the location history in the query gateway or the sensor network. Then, we invent the forwarding index-based query processing scheme to retrieve the historical location history and maintain the length of the forwarding index. We conducte a set of experiments to measure the performance of our proposed scheme. The results reveal that the proposed method indeed performs outstanding than related schemes.
Acknowledgements This work is supported by National Science Council of Taiwan (R.O.C.) under Grants NSC 99-2221-E-218-035 and NSC 98-2221-E-218-036.
References 1. Juang, P., Oki, H., Wang, Y., Martonosi, M., Peh, L.-S., Rubenstein, D.: Energy efficient computing for wildlife tracking: Design tradeoffs and early experiences with zebranet. In: Proceedings of the Tenth International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS 2002), pp. 96–107 (2002)
242
C.-C. Chen and C.-B. Lo
2. Karp, B., Kung, H.T.: GPSR: Greedy Perimeter Stateless Routing for Wireless Networks. In: Proceedings of the Sixth Annual International Conference on Mobile Computing and Networking, Massachusetts, Boston (August 2000) 3. Xu, Y., Lee, W.-C.: Compressing Moving Object Trajectory in Wireless Sensor Networks. IJDSN 3(2), 151–174 (2007) 4. Xu, Y., Winter, J., Lee, W.-C.: Prediction-based Strategies for Energy Saving in Object Tracking Sensor Networks. In: Proceedings of 2004 IEEE International Conference on Mobile Data Management (MDM 2004), Berkeley, California, USA, p. 346 (January 2004) 5. Xu, Y., Winter, J., Lee, W.-C.: Dual Prediction-based Reporting for Object Tracking Sensor Networks. In: Proceedings of First Annual International Conference on Mobile and Ubiquitous Systems: Networking and Services (MobiQuitous 2004), Cambridge, Massachusetts USA, pp. 154–163 (August 2004) 6. Madden, S., Franklin, M.J., Hellerstein, J.M., Hong, W.: TinyDB: an acquisitional query processing system for sensor networks. Proceedings of ACM Trans. Database Syst. 30(1), 122–173 (2005) 7. Meka, A., Singh, A.K.: DIST: a distributed spatio-temporal index structure for sensor networks. In: Proceedings of the 2005 ACM CIKM 2005, Bremen, Germany, pp. 139–146 (November 2005) 8. Lin, C.-Y., Peng, W.-C., Tseng, Y.-C.: Efficient In-Network Moving Object Tracking in Wireless Sensor Networks. Proceedings of IEEE Trans. Mob. Comput. 5(8), 1044–1056 (2006) 9. Xu, J., Tang, X., Lee, W.-C.: A New Storage Scheme for Approximate Location Queries in Object-Tracking Sensor Networks. Proceedings of IEEE Trans. Parallel Distrib. Syst. 19(2), 262–275 (2008) 10. Meka, A., Singh, A.K.: Distributed Spatial Clustering in Sensor Networks. In: Ioannidis, Y., Scholl, M.H., Schmidt, J.W., Matthes, F., Hatzopoulos, M., B¨ ohm, K., Kemper, A., Grust, T., B¨ ohm, C. (eds.) EDBT 2006. LNCS, vol. 3896, pp. 980–1000. Springer, Heidelberg (2006) 11. Ratnasamy, S., Karp, B., Yin, L., Yu, F., Estrin, D., Govindan, R., Shenker, S.: GHT: a geographic hash table for data-centric storage. In: Proceedings of the First ACM International Workshop on Wireless Sensor Networks and Applications(WSNA), Atlanta, Georgia, pp. 78–87 (September 2002) 12. Pfoser, D., Jensen, C.S., Theodoridis, Y.: Novel Approaches in Query Processing for Moving Object Trajectories. In: Proceedings of 26th International Conference on Very Large Data Bases (VLDB), Cairo, Egypt (September 2000) 13. Tao, Y., Papadias, D.: MV3R-Tree: A Spatio-Temporal Access Method for Timestamp and Interval Queries. In: Proceedings of 27th International Conference on Very Large Data Bases (VLDB), Roma, Italy (September 2001)
Embedding Methods for Bubble-Sort, Pancake, and Matrix-Star Graphs Jong-Seok Kim1, Mihye Kim2, Hyun Sim3, and Hyeong-Ok Lee3,∗ 1
Department of Information and Communication Engineering, Yeungnam University, 214-1 Dae-dong, Gyeongsan-si, Gyeongbuk, South Korea [email protected] 2 Department of Computer Science Education, Catholic University of Daegu, 330 Hayangeup Gyeonsansi Gyeongbuk, South Korea [email protected] 3 Department of Computer Education, Sunchon National University, 413 Jungangno Suncheon Chonnam, South Korea {simhyun,oklee}@scnu.ac.kr
Abstract. Bubble-sort, pancake, and matrix-star graphs are interconnection networks with all the advantages of star graphs but with lower network costs. This study proposes and analyzes embedding methods for these graphs based on the edge definition of graphs. Results show that bubble-sort graph Bn can be embedded in pancake graph Tn with dilation 3 and expansion 1, while bubblesort graph B2n can be embedded in matrix-star MS2,n with dilation 5 and expansion 1. Keywords: Interconnection network, Embedding methods, Bubble-sort graph, Pancake graph, Matrix-star graph.
1 Introduction Parallel-processing computers can be classified, based on their memory structure, into either multiprocessor systems with shared memory or multicomputer systems with distributed memory. Each processor in a multicomputer system has its own memory and is linked with the others through a static interconnection network. Inter-processor communication takes place in a data-driven computation model using message passing via an interconnection network [1], [2]. The most well-known topologies of interconnection networks are mesh, hypercube, bubble-sort, star, macro-star, matrixstar, and pancake graphs. The architecture of an interconnection network for the connection of the processors in a multicomputer system greatly influences its overall performance and scalability. Therefore, research on interconnection networks is the foundation of parallelprocessing computer development and the need for such research is on the rise. The most common parameters used for evaluating the performance of interconnection networks are degree, diameter, symmetry, fault tolerance, scalability, broadcasting, and embedding [1], [3]. ∗
The embedding parameter is related to mapping the communication links of the processors of an arbitrary interconnection network G onto the communication links of the processors of another interconnection network H. In other words, embedding research concerns examining the efficiency with which an algorithm developed for one certain interconnection network G can perform with a second interconnection network H. If the algorithm developed for G can be efficiently embedded in another network H with lower cost, then the algorithm developed for G can be used with H at lower cost [4], [5]. The common evaluation measures for the cost of embedding are dilation, congestion, and expansion [6], [7]. This paper examines embedding methods for bubble-sort, pancake, and matrix-star graphs (networks), which are well-known Cayley graphs. In previous work [8], [9], we analyzed embedding algorithms for star, matrix-star, Rotator-Faber-Moore, and pancake graphs. Note that the theoretical background for this paper has been partially reported elsewhere [8], [9]. This paper is organized as follows. Section 2 reviews the existing literature on the graphs used in this paper for embedding. Section 3 analyzes and presents the embedding methods among bubble-sort, pancake, and matrix-star graphs. Section 4 summarizes and concludes the paper.
2 Related Work An interconnection network can be defined as an undirected graph G = (V, E) with each processor presented as a node v of G, and the communication channel between those processors presented as an edge (v, w). V(G) and E(G) denote the set of nodes and edges of graph G, respectively; that is, V(G) = V(G) = {0, 1, 2, …, n–1} and E(G) consists of pairs of distinct nodes from V(G). There exists an edge (v, w) between two nodes v and w of G if and only if a communication channel between v and w exists. The network parameters for measuring the performance of interconnection networks are degree, diameter, symmetry, and scalability, fault tolerance, broadcasting, and embedding [4]. If we classify the interconnection networks based on the number of nodes, these can be categorized into three variations: the mesh with n×k nodes, the hypercube with 2n nodes, and the star graph with n! nodes. An n-dimensional bubble-sort graph Bn is composed of n! nodes and n(n–1)!/2 edges [10]. The address of each node is represented as a permutation of n distinct symbols and there exist an edge between two arbitrary nodes v and w if and only if the corresponding permutation to the node w can be obtained from that of v by interchanging two adjacent symbols v and w in the permutation. Bubble-sort graph Bn can be defined as the following formulas, where n distinct symbol sets = {1, 2, .., n}, and a permutation of , B = b1b2...bn, bi [11]:
∈
∈
V(Bn) = {(b1b2...bn)│bi , i≠j, bi≠bj} and E(Bn) = {(b1b2...bibi+1...bn)(b1b2...bi+1bi...bn)│(b1b2...bi...bn)
∈V(B ), 1≤i≤n–1}. n
Since there is an edge between the permutation consisting of n distinct symbols and the node where two symbols in continuous positions on the permutation is exchanged, the bubble-sort graph Bn is a regular graph of degree n–1 and a hierarchical interconnection network because it can partition the graph with the edge as the center. Fig. 1 shows a four-dimensional bubble-sort graph. The bubble-sort
Embedding Methods for Bubble-Sort, Pancake, and Matrix-Star Graphs
245
graph Bn is node- and edge-symmetric as well as bipartite. It has a diameter of n(n–1)/2 and includes Hamiltonian cycles. 2134
1243
2314 3214 3124
3241
4123
4213 1342
2431 2341
1423
2143 2413
1234 1324
4231 4321
3142 3412
3421
1432 4132
4312
Fig. 1. Example of a four-dimensional bubble-sort graph
An n-dimensional pancake graph Pn consists of n! nodes and n(n–1)!/2 edges with node symmetry. The address of each node can be represented as a permutation of n distinct symbols {1, 2, 3, ..., n} and the node set V(Pn) = {(p1p2...pn)│pi , i≠j, pi≠pj}. The adjacency (edge) between two arbitrary nodes v and w is defined as follows: v1v2...vi...vn is adjacent to w1w2...wi...wn via an edge of dimension i with 2≤i≤n if wj = vi-j+1 for all 1≤ j ≤i and wj = vj for all i<j≤n [12]. Pancake graph Pn can be defined by the following formulas, where n distinct symbol sets = {1, 2, ..., n}, and a permutation of , P = p1p2...pn, pi [6]:
∈
∈
∈
V(Pn) = {(p1p2...pn)│pi , i ≠ j, pi ≠ pj} and E(Pn) = {(p1p2...pi...pn)(pipi-1pi-2...p2p1pi+1pi+2...pn)│(p1p2...pi...pn)
∈V(P ), 2≤i≤n}. n
Even though a pancake graph includes Hamiltonian cycles, it is not a bipartite graph because cycles can exist with an odd length in a pancake graph Pi of dimension four or greater, i.e., 4≤i≤n. Fig.2 shows a four-dimensional pancake graph. 1234
4321
a 3214 2314
2134
3421
3124
3241 4231 2413
1324 3142 4132 1432 b
b
2341
2431
3412
1342
4213 4312
1243
4123
1423 a
2143
Fig. 2. Example of a four-dimensional pancake graph
246
J.-S. Kim et al.
In a matrix-star graph MS2,n [13], a node is represented by a 2 × n matrix … … , which consists of 2n distinct symbols {1, 2, 3, ..., … … 2n}. An edge exists between two arbitrary nodes v and w if and only if the node w is obtained by applying one of the following matrix operations (1), (2), and (3) on the … … . node v. Let node v be … … (1) Matrix w, which exchanges the (1, 1)th symbol with the (1, i) th symbol of v: … … . … … (2) Matrix w, which exchanges the first row vector with second row vector of v: … … . … … (3) Matrix w, which exchanges the (1, 1)th symbol with the (2,1) th symbol of v: … … . … … In matrix-star graph MS2,n, three different types of edges are defined: edge that connects node v and the node obtained by the matrix operation (1), edge obtained using matrix operation (2), and edge R obtained using matrix operation (3). MS2,n consists of (2n)! nodes because it can generate a matrix as large as the number of permutations represented by 2n distinct symbols using the above edge definition, and it is a regular graph of degree n–1 (n ≥ 2). Figure 3 shows an example of MS2,2 in which the nodes of the graph are represented by 2×2 matrices [13]. In this paper, we use the terms node and matrix interchangeably.
21 34 31 24
12 34 32 14 23 14 13 24
42 24 31 31 34 21 43 21
32 41 23 41
34 14 12 32
43 12 13 41 32 31 42 42
14 23 41 23
24 13 42 13 12 43 21 43
Fig. 3. Example of a four-dimensional matrix-star graph
Embedding Methods for Bubble-Sort, Pancake, and Matrix-Star Graphs
247
3 Embedding Methods The embedding of a specific graph G1 into another graph G2 is a mapping to observe whether graph G1 is included in the structure of graph G2, and/or how they are interrelated. The embedding of graph G1 into a graph G2 can be defined as a function f = (ø, ρ) where ø is a function that maps each of vertices in V(G1) onto each of vertices in V(G2), and ρ is a function that maps an edge e = (v, w) in G to a path in H that connects nodes ø(v) and ø(w). Dilation, congestion, and expansion are commonly used evaluation parameters for measuring the cost of an embedding method [14], [15]. Let us observe the evaluation parameters when graph G1 embeds graph G2. The dilation is the length of the path ρ(e) in G2 when edge e in G1 maps an edge in G2, and the dilation of embedding f is the maximum value of dilations which can be obtained in G1 when edge e in G1 maps an edge in G2. The congestion of edge e' in G2 is the number of ρ(e) included in e', and the congestion of embedding f is the maximum number of all edge congestions in G2. The expansion of embedding f is the ratio of the number of vertices in G2 to the number in G1 [8], [9]. Theorem 1 A bubble-sort graph Bn can be embedded in a pancake graph Pn with dilation 3 and expansion 1. Proof. Bubble-sort graph Bn and pancake graph Pn with n! nodes can be mapped oneto-one using identical node numbers as follows: node B (=b1b2b3...bi...bn) in Bn is mapped onto node P (=p1p2p3...pi...pn) in Pn, and node B' connected to B by an idimensional edge in Bn, is mapped to node P' in Pn. The dilation of this embedding can be analyzed through the number of edges required to generate the address (permutation) of node P' from the address of P in Pn. We analyze this embedding by dividing the edges of Bn into three cases based on their dimension. Case 1. One-dimensional edge (i=1) The permutation of node B' adjacent to node B (=b1b2b3...bi...bn) via a onedimensional edge in Bn is b2b1b3...bi...bn. Node P (=p1p2p3...pi...pn) in Pn is connected to node P' (=p2p1p3...pi...pn) by a two-dimensional edge. Therefore, the two nodes B and B' adjacent via a one-dimensional edge in bubble-sort graph Bn can be embedded in nodes P and P' in pancake graph Pn with dilation 1. Case 2. Two-dimensional edge (i=2) The node adjacent to node B (=b1b2b3b4...bi...bn) via a two-dimensional edge in bubble-sort graph Bn is B' (=b1b3b2b4...bi...bn), but node P (=p1p2p3p4...pi...pn) is not adjacent to node P' (=p1p3p2p4...pi...pn) in Pn. Therefore, we analyze this embedding based on the number of edges required for the shortest path routing from node P to P' in Pn. We assume that the dimensional edge sequence required for optimal routing from P (=p1p2p3p4...pi...pn) to P' is . Following this edge sequence, we can first obtain node P2(P) (=p2p1p3p4...pi...pn) adjacent to node P (=p1p2p3p4...pi...pn) via dimensional edge P2, then node P3(P2(P)) (=p3p1p2p4...pi...pn) adjacent to node P2(P) through edge P3, and finally node P2(P3(P2(P))) (= p1p3p2p4...pi...pn) adjacent to node P3(P2(P)) via edge P2. Here, we can see that the address of node P2(P3(P2(P)))
248
J.-S. Kim et al.
(=p1p3p2p4...pi...pn), which is obtained by sequentially applying the edge sequence to node P, is the same as the address of node P' (=p1p3p2p4...pi...pn). Hence, two nodes B and B' adjacent via a two-dimensional edge in bubble-sort graph Bn can be embedded in pancake graph Pn with dilation 3. Case 3. Greater than three-dimensional edge (i ≥ 3) The permutation of node B' connected to node B (=b1b2b3...bi–1bibi+1bi+2...bn) by an idimensional edge in Bn is b1b2b3...bi–1bi+1bibi+2...bn. However, nodes P (=p1p2p3...pi– 1pipi+1pi+2...pn) and P' (=p1p2p3...pi–1pi+1pipi+2...pn) are not connected to each other in Pn. Therefore, we analyze the dilation of this case using the number of dimensional edges required for the shortest path routing from node P to P' in Pn. We assume that the dimensional edge sequence for optimal routing from P to P' in Pn to be . Following this edge sequence, we first obtain node Pi+1(P) (=pi+1pipi– 1...p3p2p1pi+2...pn) adjacent to node P (=p1p2p3...pi–1pipi+1pi+2...pn) via dimensional edge Pi+1, and node P2(Pi+1(P)) (=pipi+1pi–1...p3p2p1pi+2...pn) adjacent to node Pi+1(P) through edge P2, then node Pi+1(P2(Pi+1(P))) (=p1p2p3...pi–1pi+1pipi+2...pn) adjacent to node P2(Pi+1(P)) via edge Pi+1. Accordingly, we can see that the address of node Pi+1(P2(Pi+1(P))) (=p1p2p3...pi–1pi+1pipi+2...pn), which is obtained by sequentially applying the edge sequence to node P, is identical to the address of node P' (=p1p2p3...pi–1pi+1pipi+2...pn). Therefore, two nodes B and B' adjacent via an edge of dimension three or greater (i ≥ 3) in bubble-sort graph Bn can be embedded in pancake graph Pn with dilation 3. Theorem 2. The dilation cost of embedding a pancake graph Pn into a bubble-sort graph Bn is O(n2). Proof. The permutation of node P' is pipi–1pi–2...p3p2p1pi+1...pn, where node P (=p1p2p3...pi–1pipi+1pi+2...pn) is adjacent to node P' via an i-dimensional edge in Pn. But nodes B (=b1b2b3...bi–1bibi+1bi+2...bn) and B' (=bibi–1bi–2...b3b2b1bi+1...bn) in Bn are not adjacent to each other. Therefore, we analyze the dilation of this embedding using the number of dimensional edges used for the shortest path routing from B to B' in Bn by assuming that this embedding has maximum dilation n2 when node P' (=pipi–1pi– 2...p3p2p1pi+1...pn) is adjacent to node P (=p1p2p3...pi–1pipi+1pi+2...pn) via an idimensional edge. The permutation of node B' (=bibi–1bi–2...b3b2b1bi+1...bn) can be generated from the permutation of node B (=b1b2b3...bi–1bibi+1bi+2...bn) in Bn by reversing in descending order the first symbol b1 to the ith (bi) symbol of B. Because the edge sequence required for positioning the first symbol b1 to the ith (bi) symbol of a given node in Bn is <1,2,3,...,i–1>, the number of edges needed for this reversal is i–1. In addition, i–2 edges are required for positioning the second symbol b2 through to the (i–1)th symbol of a node in Bn, because the edge sequence needed for this transformation is <1,2,3,...,i–2>. If we generalize these transformation processes, we obtain the sum of 1 , because the total number of edges this dilation with ∑ required for this conversion from the permutation of B to that of B' is (i–1) + (i–2) + (i–3) + ... + 1. Therefore, we can say that the dilation cost of this embedding is O(n2).
Embedding Methods for Bubble-Sort, Pancake, and Matrix-Star Graphs
249
Theorem 3. A bubble-sort graph B2n can be embedded into a matrix-star graph MS2,n with dilation 5 and expansion 1. Proof. Analyzing Theorem 3 requires reformulating the address of a node represented by a permutation to a 2×n matrix form for the nodes in B2n. This is done by placing symbols from the first (b1) to the nth in the permutation of node B (=b1b2... bn– 1bnbn+1...b2n) into the first row and the remainder of the symbols (bn+1...b2n) into the … . Here, if we regard the symbols second row of the 2×n matrix, i.e., … … denoted by bi in B2n as xi (1 ≤ i ≤ 2n), node B in B2n can be mapped … … one-to-one onto node X in MS2,n. … In bubble-sort graph B2n, the edge that connects node B (=b1b2... bi–1bibi+1...bn) and the node whose permutation is b1b2...bi–1bi+1bi...b2n, in which the two symbols bi and bi+1 in continuous positions on the permutation of B are exchanged, is called an idimensional edge; the node connected to node B by an i-dimensional edge is Bi. Then, we prove Theorem 3 by dividing it into five cases according to the dimensional edges of B2n. Case 1. {(i, i+1) | i =1}
… , and the matrix We assume that the 2×n matrix of node B in B2n is … … of node X in MS2,n is . Then, the matrix of node B1 connected to … … , and the matrix of node node B by a one-dimensional edge in B2n is … … . Here, we can see X' connected to node X by edge in MS2,n is … that two nodes B and B1 adjacent via a one-dimensional edge in B2n can be mapped in MS2,n with dilation 1. onto nodes X and X' adjacent via edge Case 2. {(i, i+1) | 2 ≤ i ≤ n } In bubble-sort graph B2n, the matrix of node Bi connected to B by an i-dimensional … … (2≤ i ≤n). In a matrix-star graph MS2,n, we edge is … … … …… … … obtain the matrix by exchanging the (1, i)th symbol … … … …… … … , and call it with the (1, i+1)th symbol of node X … … … … … node X'. When we try to map nodes B and Bi in B2n onto nodes X and X' in MS2,n, we can see that the nodes X and X' in MS2,n are not adjacent to each other. Therefore, we route from node X to node X' by sequentially applying the edge sequence < , , >. This routing process is as follows: … … … … … … … … … … … …… …
… … … … … … … … … . … … … ……
250
J.-S. Kim et al.
Consequently, two nodes B and Bi connected by an i-dimensional edge in bubble-sort graph B2n can be mapped one-to-one onto nodes X and X' connected by a path of length 3 in matrix-star graph MS2,n.:
Case 3. {(i, i+1) | i = n } The matrix of node Bn adjacent to B via an n-dimensional edge in B2n is … … , and we obtain the matrix of node X' by … … … … interchanging the (1, n)th symbol with the (2, 1)th symbol of node X … in MS2,n. Let us map nodes B and Bn in B2n onto X and X' in MS2,n. … However, nodes X and X' in MS2,n are not adjacent to each other. Therefore, we route from node X to node X' using the edge sequence < , , > as follows: … … …
… … …
… … …
… . Here, we can see that the two nodes B and Bn adjacent via an … … n-dimensional edge in B2n can be mapped onto nodes X and X' adjacent via a path of length 3 in MS2,n. Case 4. {(i, i+1) | i = n+1} In bubble-sort graph B2n, the matrix of node Bn+1 connected to B by an (n+1)… . In a matrix-star graph MS2,n, we obtain dimensional edge is … … the matrix of node X' by exchanging the (2, 1)th symbol and … … . Here, we can see that the nodes X the (2, 2)th symbol in node X … and X' in MS2,n are not connected to each other when mapping nodes B and Bn+1 in B2n onto nodes X and X' in MS2,n. Hence, we route from X to X' in MS2,n by sequentially applying the edge sequence < , , > as follows: … … … … … … … . Consequently, the two nodes B and Bn+1 connected by an … (n+1)-dimensional edge in a bubble-sort graph B2n can be mapped one-to-one onto nodes X and X' connected by a path of length 3 in a matrix-star graph MS2,n. Case 5. {(i, i+1) | n+2 ≤ i ≤ 2n–1} The matrix of node Bi adjacent to B via an i-dimensional edge in B2n is … … … …… (n+2 ≤ i ≤ 2n–1). We obtain the matrix of node X' … … … … … …… by interchanging the (2, i)th symbol with the (2, … … … … … … … in MS2,n. Because the nodes i+1)th symbol of node X … … X and X' in MS2,n are not connected to each other, let us route from node X to node X'
Embedding Methods for Bubble-Sort, Pancake, and Matrix-Star Graphs
251
using the edge sequence < , , , , >. The following shows this routing process: … … … … … … … … … … … … … … … … … … … … … … … … … … …… … … … … … …… . Here, we can see … … … …… … … that two nodes B and Bi adjacent via an i-dimensional edge in bubble-sort graph B2n can be mapped one-to-one onto nodes X and X' adjacent via a path of length 5 in matrix-star graph MS2,n. Therefore, a bubble-sort graph B2n can be mapped to a matrix-star graph MS2,n with dilation 5 according to the proof for the five cases above.
4 Conclusion In this paper, we propose and analyze methods for embedding bubble-sort, pancake, and matrix-star graphs, which are interconnection networks with node symmetry, recursive structure, and maximum fault-tolerance that improve upon the network cost of a star graph. The proposed embedding methods are based on mapping two adjacent nodes in a source graph to nodes in a target graph based on the edge definition of the graphs, assuming that the graphs have the same number of nodes. The dilation of embedding is then analyzed by computing the minimum number of edges used in a target graph when a source graph is mapped to the target graph. Results of this study show that bubble-sort graph Bn can be embedded in pancake graph Pn with dilation 3 and expansion 1, and bubble-sort graph B2n can be embedded in matrix-star graph MS2,n with dilation 5 and expansion 1. The results suggest that the embedding method developed for a bubble-sort graph can be simulated in both pancake and matrix-star graphs with additional constant cost. Acknowledgement. This research was supported by Mid-career Researcher Program through National Research Foundation of KOREA (NRF) grant funded by the Ministry of Education, Science and Technology (MEST).
References 1. Feng, T.: A Survey of Interconnection Networks. IEEE Computer, 12–27 (December 1981) 2. Ranka, S., Wang, J., Yeh, N.: Embedding Meshes on the Star Graph. Parallel and Distributed Computing 19, 131–135 (1993) 3. Wu, A.Y.: Embedding of Tree Networks into Hypercubes. Parallel and Distributed Computing 2, 238–249 (1985) 4. Azevedo, M.M., Bagherzaeh, N., Latifi, S.: Low Expansion Packing and Embeddings of Hypercubes into Star Graphs: A Performance-Oriented Approach. IEEE Parallel and Distributed Systems 9(3), 261–274 (1998)
252
J.-S. Kim et al.
5. Ghafoor, A., Bashkow, T.R.: A Study of Odd Graphs as Fault-Tolerant Interconnection Networks. IEEE Trans. Computers 40(2), 225–232 (1991) 6. Berthome, P., Ferreira, A., Perennes, S.: Optimal Information Dissemination in Star and Pancake Networks. IEEE Trans. on Parallel and Distributed Syst. 7(12), 1292–1300 (1996) 7. Akers, S.B., Harel, D., Krishnamurthy, B.: The Star Graph: An Attractive Alternative to the n-Cube. In: Proc. International Conference on Parallel Processing, 393–400 (August 1987) 8. Kim, M., Kim, D.W., Lee, H.O.: Embedding Algorithms for Star, Bubble-Sort, RotatorFaber-Moore, and Pancake Graphs. In: Hsu, C.-H., Yang, L.T., Park, J.H., Yeo, S.-S. (eds.) ICA3PP 2010. LNCS, vol. 6082, pp. 348–357. Springer, Heidelberg (2010) 9. Lee, H.O., Sim, H., Seo, J.H., Kim, M.: Embedding algorithms for bubble-sort, macro-star, and transposition graphs. In: Ding, C., Shao, Z., Zheng, R. (eds.) NPC 2010. LNCS, vol. 6289, pp. 134–143. Springer, Heidelberg (2010) 10. Yeh, C.H., Varvarigos, E.A.: Macro-Star Networks: Efficient Low-Degree Alternatives to Star Graphs. IEEE Trans. Parallel and Distributed Systems 9(10), 987–1003 (1998) 11. Chou, Z., Hsu, C., Sheu, J.: Bubblesort Star Graphs: A New Interconnection Network. In: 9th International Parallel Procession Symposium, pp. 41–48 (1996) 12. Lin, C.K., Huang, H.M., Hsu, L.H.: The super connectivity of the pancake graphs and the super laceability of the star graphs. Theoretical Computer Science 339, 257–271 (2005) 13. Lee, H.O., Kim, J.S., Park, K.W., Seo, J., Oh, E.: Matrix-Star Graphs: A New Interconnection Network Based on Matrix Operations. In: Srikanthan, T., Xue, J., Chang, C.-H. (eds.) ACSAC 2005. LNCS, vol. 3740, pp. 478–487. Springer, Heidelberg (2005) 14. Qiu, K., Akl, S.G., Meuer, H.: On Some Properties and Algorithms for the Star and Pancake Interconnection Networks. Journal of Parallel and Distributed Computing 22, 16– 25 (1994) 15. Corbett, P.F.: Rotator Graphs: An Efficient Topology for Point-to-Point Multiprocessor Networks. IEEE Transaction Parallel Distributed System 3(5), 622–626 (1992)
Development of a 3D Virtual Laboratory with Motion Sensor for Physics Education∗ Ji-Seong Jeong1, Chan Park1, Mihye Kim2, Won-Keun Oh1, and Kwan-Hee Yoo1, ** 1
Department of Information Industrial Engineering, Department of Information Communication Engineering, Department of Physics Education, Department of Computer Education and IIE, Chungbuk National University, 410 Seongbongro Heungdukgu Cheongju Chungbuk, South Korea {farland83,szell,wkoh,khyoo}@chungbuk.ac.kr 2 Department of Computer Science Education, Catholic University of Daegu, 330 Hayangeup Gyeonsansi Gyeongbuk, South Korea [email protected]
Abstract. This paper proposes a three-dimensional (3D) virtual laboratory simulation system for physics education. The system uses readily available motion sensors by taking into account the advantages of both 2D and 3D virtual environments. Students can simulate the motion of objects in 3D virtual spaces using a controller with a motion sensor as if in a real laboratory experiment. The system displays 3D virtual spaces as 3D images, and allows students to simulate the motion of objects visually with real-time graphs to make the experience more realistic. The proposed system will help students easily understand the physical concepts of dynamics while reducing misconceptions that can arise during physics lessons on force and motion. Keywords: 3D virtual laboratory, physics education, motion sensor.
1 Introduction Learning about force and motion are fundamental to mastering physics. Since concepts of force and motion are also formed intuitively, based on everyday experiences, they not only interfere with learning [1], but can also lead to incorrect conclusions about scientific concepts [2]. Thus, experimental activities during physics education play a significant role in students learning concepts correctly and accurately [3]. In practice, however, conducting an experiment to illustrate the concepts in each physics class is difficult due to the time required to set up the experiment, as well as the difficulty in manipulating laboratory equipment and problems with experimental accuracy and precision. Real-time measurement and analysis of experimental data are ∗
also difficult because experiments are usually conducted on moving objects, especially during lessons related to force and motion. To address these issues, computer simulations using personal computers (PCs) have been used early on. Computer simulations can solve structured problems and can control experimental situations by simplifying experimental variables. Moreover, many students today prefer computer-based learning situations because they are familiar with computer communications, information searching, and games. However, most existing simulation systems for teaching physics are based on twodimensional (2D) virtual environments with simple forms [1–4]. In 2D environments, circumstances can arise where observations change depending on the reference point of the coordinates because 2D simulations systems display the results of object movement based on a fixed reference point. Moreover, realistic simulation of falling or colliding objects is difficult. Due to these shortcomings, a 2D virtual simulation environment does not provide students with the desired learning effect. Thus, it is important to introduce three-dimensional (3D) simulation to enable students to experience the realistic motion of objects from diverse viewpoints. This paper proposes a 3D virtual simulation system that can simulate the magnitude and direction of object movement in 3D virtual space using a motion sensor controller. The system can help students in reducing misconceptions by viewing the simulation results in real-time graphs. Most simulation parameters can also be modified freely over a wide range in a 3D environment, so the motion of objects can be understood as vectors. Note that this work has been partially reported in earlier work [5-7]. This paper is organized as follows. Section 2 reviews the existing literature on 3D virtual simulation applications. Section 3 presents the proposed 3D virtual laboratory simulation system for physics education. Section 4 introduces a 3D virtual lab developed on a PC to demonstrate and evaluate the value of the proposed system. Section 5 concludes the paper and describes future research ideas.
2 Related Work Existing 2D virtual simulation systems for physics education simulate the motion of objects from only a static observation point. This makes perception of the physical concepts of dynamics difficult. Additionally, because such systems control the main parameters of object motion using a keyboard and/or a mouse, simulations often differ from those in an actual experiment [8-10]. To address these problems with 2D systems, three-dimensional simulation systems are being developed. A previous study [11] developed a virtual reality application for physics education that enabled students to create their own experiments in a 3D virtual environment and simulate physical experiments correctly and accurately. That system was based on the PhysX physics engine [12]. However, the system can run only on special equipment, such as a head-mounted display or a personal interaction panel. The International Institute of Information Technology in India also developed a 3D virtual physics laboratory based on open-source components to provide a high-quality physics education environment [13]. Students can create their own experiments in 3D virtual environments, manipulate the simulation parameters, and observe the results. Another
Development of a 3D Virtual Laboratory with Motion Sensor for Physics Education
255
virtual simulation learning system [14] was proposed that permits students and educators, represented as avatars, to cooperate with each other in conducting physics experiments in 3D virtual environments. That system is a work-in-progress and plans exist to integrate it with other related projects. In another approach to the development of a 3D virtual simulation environment, application programming interfaces (APIs) for physics engines have been developed to enable easier development of simulated 3D experiments by end users. Some physics engines are freely available and open source, including PhysX from AGEIA [12], Havok [15], and Open Dynamic Engine (ODE) [16]. The virtual simulation system proposed in [8] was based on PhysX and ODE was used in [13]. This study proposes a 3D virtual lab simulation system for physics education based on a readily available motion sensor controller. It capitalizes on the strengths of 2D and 3D virtual environments. The system simulates the motion of objects using simulation input values from a motion sensor rather than from a keyword and/or mouse. As a result, students can understand physics concepts more readily and avoid misconceptions, especially in lessons concerning force and motion.
3 Design of a 3D Virtual Lab Simulation System with Motion Sensor for Physics Education 3.1 System Architecture Fig. 1 shows the architecture of the proposed system. The 3D Virtual Physics Education Simulation Engine (3D VESE) manages and controls the overall simulation. It also collects motion information from a controller equipped with a motion sensor, and specific input data from a keyword and/or mouse. It creates a 3D virtual space using 3D objects, such as a laboratory, desk, building, and time recorder, which are generated in advance using Autodesk 3DS Max modeling, animation, and rendering software [17]. Furthermore, the system provides a function that enables rendering of the simulation results produced by 3D VESE to demonstrate the motion of objects more realistically. The system also displays simulation information in real time, numerically and graphically, and represents textures by replacing the surface of images depending on the distance between objects and by adjusting the transparency of the background color. 3.2 3D Object Configuration and Rendering First, the 3D objects that are required in the experiments are created using Autodesk 3DS Max. Then, a 3D virtual simulation space is created by rendering a 3D virtual space and arranging the created objects accordingly. Fig. 2 shows examples of 3D objects created using 3DS Max and Fig. 3 shows examples of 3D virtual spaces using these 3D objects. For example, Fig. 3 (c) shows an example of a 3D virtual space designed to simulate the uniform acceleration caused by gravity when a ball is dropped from the top of a building.
256
J.-S. Jeong et al.
Fig. 1. Architecture of the proposed 3D virtual lab simulation system for physics education
Fig. 2. Examples of 3D objects created using 3DS Max
Fig. 3. Examples of visualized 3D virtual spaces created using 3D objects
3.3 Communication Using a Motion Sensor Controller The proposed system uses a motion sensor controller to control the dynamic interactions between objects to provide a more realistic experimental environment; that is, the interface of the proposed system was developed to be accomplished with the interactions between the system and the motion controller trough Bluetooth communication. A Wii-remote motion sensor controller, which includes Bluetooth communication, an infrared sensor, and a triaxial acceleration sensor, was used.
Development of a 3D Virtual Laboratory with Motion Sensor for Physics Education
257
Fig. 4 shows the data communication process between the system and the Wiiremote controller. The communication process is divided into two stages: the synchronization and execution stages. In the synchronization stage, the 3D VESE connects the PC and the motion controller via Bluetooth synchronization. In the execution stage, the various data inputs from the user such as the movement of the motion controller are first detected and analyzed. Then, the input information and related objects are displayed in a 3D virtual space depending on the type of movement. Analyzing the movement of the controller requires knowledge of the coordinates used. The proposed system used roll, pitch, and yaw rotations for this.
Fig. 4. Data communication process between the system and the motion controller
3.4 Object Movement Simulation Using a Motion Sensor Controller To allow students to perceive force and motion concepts correctly, simulation should be performed in a 3D virtual space using a motion sensor that can generate object movements that are similar to those in real life. For this reason, the proposed system used the Wii-remote motion controller and defined several function keys to manipulate simulation action as the controller moves. Table 1 shows the function keys defined in this study. The motion controller’s ‘arrow key’ and ‘A button (click on) + movement’ were used to control the data from the camera installed in the 3D virtual space, ‘A button (click on) + movement + A button (click off)’ was used to control the object simulations, and ‘A button (click on) + movement’ was used to control objects in the 3D virtual space. Also, ‘A button’ was used to display the simulation results in graphs and other forms.
258
J.-S. Jeong et al. Table 1. Function keys defiend in the system Type
Function Keys
Simulation Action
Arrow key Change camera position in a 3D virtual space to XZ Plane Camera control mode A button (click on) + movement Change the direction of camera’s lens Simulation A button (click on) + movement Simulate direction of motion controller’s movement control mode + A button (click off) B button Change a 3D object to other object 3D object control mode A button (click on) + movement Change 3D object’s initial position Graph display A button Turn display of real-time graph on and off mode
Determining the specific movements of the motion controller requires calculation of the direction vector using the motion sensor’s rotation data. The free Virtual Trackball software [14] with roll, pitch, and yaw rotations was used. The direction vector D = (Dx, Dy, Dz) is calculated using Eq. (1) where ΔY is the magnitude of the change in the pitch and ΔX is the magnitude of the change in the yaw. Eq. (2) (p(t)) computes the 3D position coordinates of a moving object at time t with the direction vector calculated in Eq. (1) where p0 is the initial 3D coordinates of the object, v0 is the initial velocity of the object, and a is the acceleration applied to the object. The system can present the movement of objects using these equations. The system also shows information about the objects’ movement in real time by calculating the distance they move using Eq. (3). cos ∆
sin ∆ 1 1
;
sin ∆
2
2
;
cos ∆
cos ∆
(1) (2) (3)
4 Prototypes and Experiments Several prototypes were developed to demonstrate and evaluate the proposed system. These prototypes were implemented using a number of tools, such as MS Visual Studio 2008, Microsoft Foundation Class (MFC), WiiremoteLib, and Autodesk 3DS Max in a Windows Vista, Dual Core, Geforce 900GT environment. The first system was intended to produce a 3D virtual simulation system that would allow students to simulate an object’s movement in a specific direction with uniform velocity or uniform acceleration. The second prototype was a 3D virtual highway simulation system that allowed students to experience speed. A third prototype provided a 3D virtual building free-fall simulation space [5, 6]. Here, the first prototype is used to describe the proposed system. Fig. 5 shows the configuration of the 3D virtual lab simulation system. A student can move each object by calculating the distance it moves in the virtual space in a certain time using Eq. (3), and by computing the 3D position coordinates of the object with the direction vector used in the object movement. The system then displays dots on a paper tape based on the movement distance of the object over time, and the movement information is displayed on a real-time graph and a time recorder.
Development of a 3D Virtual Laboratory with Motion Sensor for Physics Education
259
Fig. 5. Configuration of the 3D virtual laboratory simulation system
Fig. 6 shows the initial screen of the 3D virtual laboratory simulation system. The real-time numerical information about the object’s movement (velocity, movement distance, and acceleration) is displayed near the top-left corner of the screen. The object’s movement is displayed in a real-time graph near the top right-hand corner of the screen with time on the x-axis and distance on the y-axis.
Fig. 6. Initial screen of the 3D virtual laboratory simulation system
The system was designed to be able to freely modify the user’s viewpoint and camera direction. Fig. 7 show examples of the modification of the camera’s direction using the output date of the motion controller or the right button of the mouse. This show also examples of the modification of the user’s viewpoint using the arrow keys or the special key input of the motion sensor.
260
J.-S. Jeong et al.
Fig. 7. Changing the camera direction and user viewpoint using the motion sensor controlller
The left side of Fig. 8 sh hows the simulation results when an object is moved witth a uniform velocity of 1 m/s in one direction; the right side of Fig. 8 shows unifoorm he same direction. Time progresses from top to bottom and acceleration of 1 m/s2 in th the distance moved clearly varies with elapsed time.
Fig. 8. Simulation results wheen the object is moved with uniform velocity (left) and unifform acceleration (right) in the samee direction
Development of a 3D Virtual Laboratory with Motion Sensor for Physics Education
261
The system was also used to demonstrate acceleration in the 3D virtual environment by confirming the distance an object moves in uniform time units using a time recorder. Fig. 9 shows examples of the dots recorded by the time recorder on a paper tape on a desk when an object moves with a specific velocity or acceleration in the 3D virtual laboratory.
Fig. 9. Simulation results vary depending on the velocity and acceleration of an object ((a): 0.5 m⁄s, (b): 1 m⁄s, (c): 0.5 m⁄s , (d): 1 m⁄s )
5 Conclusion This study described a 3D virtual laboratory simulation system equipped with a controller with a motion sensor for effective physics education. One-on-one experiments were conducted with second-year students (11 male and two female students) from 12 middle schools (Guahm, Kuksabong, Nanwoo, Namkang, Dongjak, MoonChang, Seongnam, Sungbo, Shinlim, and Inhun middle schools, and Chungang University’s middle school) to assess the system’s educational effectiveness. The results of the experiment showed that the system generally helped students understand the concepts of dynamics more easily. Most of the students reported that in particular, the simulated speed and acceleration experiments presented in real-time graphs helped them understand the relevant concepts. These results could support the hypothesis that the proposed system not only allows users to easily understand the concepts of dynamics, but also reduces possible misunderstanding. Several issues still remain related to further development and evaluation of the system through practical use in actual physics classes. The performance of the system was much slower on Windows XP than on Windows Vista systems because it was developed using MFC. This disparity could be eliminated by re-implementing the system in .NET. The system should be able to obtain more accurate movement information from the users. One possible solution to this would be to attach an
262
J.-S. Jeong et al.
additional motion sensor to the existing motion controller. This would permit obtaining more precise movement information using the differences between the two motion sensors. As well as addressing these problems, the system should support mobile devices, such as smart phones, tablets, and other portable devices to accommodate a variety of users.
References 1. Gunstone, R.F., Champagne, A.B., Klopfer, L.E.: Instruction for understanding: A case study. Australian Science Teachers Journal 27(3), 27–32 (1981) 2. Viennot, R.: Spontaneous reasoning in elementary dynamics. European Journal of Science Education 1(2), 205–221 (1979) 3. Strike, K.A., Posner, G.J.: Conceptual change and science teaching. European Journal of Science Education 4(3), 231–240 (1982) 4. McDermott, L.C.: Research on conceptual understanding in mechanics. Physics Today 37(7), 24–32 (1984) 5. Jeong, J.S.: Design and Implementation of Virtual Simulation Systems for 3D Dynamics Education using Motion Sensors. Master Thesis, Chungbuk National University (2011) 6. Jeong, J.S., Park, C., Jang, Y.H., Kim, A.R., Oh, W.K., Yoo, K.H.: Design and Implementation of Basic Dynamics Education System using Motion sensor in 3D virtual space. Journal of Korean Society for Computer Game 2(20), 141–147 (2010) 7. Jeong, J.S., Park, C., Oh, W.K., Yoo, K.H.: 3D Virtual Simulation System for Physics Education and its Application. In: International Conference on Convergence Content, ICCS 2010, vol. 8(2), pp. 239–240 (2010) 8. Science Love of KimJungSik and HuMyeongSung, http://sciencelove.com/ 9. Science All, http://www.scienceall.com/ 10. Information Center for Physics Research, http://icpr.snu.ac.kr/ 11. Kaufmann, H., Meyer, B.: Physics Education in VR: An Example. TR-188-2-2010-03, Technical Reports, Vienna University of Technology, Austria (2009) 12. AGEIA Technologies: PhysX Documentation, http://www.ageia.com/ 13. Singh, J., Sampath, H., Sivaswamy, J.: An Open Source Virtual Lab for School Physics Education. TR-2009-239, Technical Reports, International Institute of Information Technology, India (2009) 14. dos Santos, F.R., Guetl, C., Hailey, P.H., Harward, V.J.: Dynamic Virtual Environment for Multiple Physics Experiments in Higher Education. In: IEEE EDUCON 2010, pp. 731– 736 (2010) 15. Havok Physics Engine, http://www.havok.com/index.php?page=havokphysics/ 16. ODE (Open Dynamics Engine), http://www.ode.org/ 17. Autodesk 3DS Max, http://usa.autodesk.com/adsk/servlet/pc/index? id=13567410&siteID=123112 18. Virtual Trackball free software, http://www.umnet.com/free-software/ 13786-Virtual_Trackball_v12 19. Shoemake, K.: ARCBALL: A User Interface for Specifying Three-Dimensional Orientation Using a Mouse. In: Proceedings of Graphics Interface 1992, pp. 151–156 (1992)
An Adaptive Embedded Multi-core Real-Time System Scheduling Liang-Teh Lee1, Hung-Yuan Chang1,2, and Wai-Min Luk1 1
Dept. of Computer Science and Engineering, Tatung University, Taiwan 2 Dept. of Electronic Engineering, Technology and Science Institute of Northern Taiwan, Taiwan [email protected], [email protected], [email protected]
Abstract. Due to the popularity of multi-core systems in recent years, it is important to design a practical multi-core scheduling for multi-core system, to improve the performance of the embedded multi-core real-time system. The majority of well-known real-time scheduling algorithm is applying prioritydriven, to ensure hard real-time tasks to be completed before their deadlines, and to service soft real-time tasks using remaining time. The proposed adaptive scheduler, an adaptive scheduling algorithm for embedded multi-core real-time systems, can dynamically assign task and cross over global and local scheduler well, and take into account both of soft and hard real-time tasks. Moreover, by established a special non-preemptive region, the hard-heavy tasks can execute independently, so as to decrease the context switch overhead and improve the system performance. The experiment results show that, compared with other multi-core schedulers, the proposed method can get better results. Keywords: embedded, multi-core, real-time system, scheduling.
system mixes a large amount of hard real-time tasks and soft real-time tasks. The majority of well-known real-time scheduling algorithm is to use the setting of priority-driven, to ensure that hard real-time tasks can complete before miss deadline, and the rest of the service time, the system provides services of soft real-time tasks. But by using this kind of scheduling method, we often ignore the requirements of the soft real-time tasks [7][8]. In this paper, an improved real-time scheduling method has been proposed. The purposed method is not only to guarantee service of hard real-time tasks, but also to meet the needs of soft real-time tasks. The proposed AFS (Adaptive Feedback Scheduling) can dynamically assign tasks and cross over global and local scheduler well, and take into account two different kinds of soft and hard real-time tasks. Moreover, by established a special non-preemptive region, the hard-heavy tasks can execute independently, so as to decrease the context switch overhead and improve the system performance. The rest of the paper is organized as follows. Section 2 will describe some background knowledge of techniques about the real-time system, overview of traditional schedule algorithm method. We present our approach of adopting the architecture of AFS and proposing a scheduling strategy in section 3. In section 4, we will introduce the environment of the experiment to set up and show the experimental results. Finally, section 5 is the conclusion.
2 Related Work All the tasks have some degree of urgency of the need to be completed before a certain time deadline in real-time systems. These tasks will give an appropriate response according to some events that occurred outside. Thus, it is very important for a real-time operating system to have an efficient task scheduler for ensuring tasks to meet their deadlines. Each task in a real-time system, the correctness of its operation is not only considering its execution result but also has to consider its deadline. If there is a task which execution result is correct, but it is not able to meet the requirements of its deadline, then this result can not be accepted. Each task has its own deadline in real-time systems, and it also must meet its deadline [3]. 2.1 Classification of Real-Time Tasks Tasks can be divided into soft and hard real-time tasks in the real-time system. Soft real-time tasks are allowed to miss their deadlines. In an embedded system, for example, when performing the phone call, the user must press the button immediately to establish links, but can tolerate slightly delay. Conversely, hard real-time tasks are not allowed to miss their deadlines, such as automatic Engine Control Unit, it must perform computations and be processed before its deadline, if miss its deadline, the engine will not work correctly. Based on previously characteristics, we know that a real-time system with mixed soft and hard real-time tasks, it does absolutely not allow the soft real-time tasks to have higher priorities than that of the real-time tasks.
An Adaptive Embedded Multi-core Real-Time System Scheduling
265
2.2 Traditional Schedule Algorithm Scheduler manages the execution order of tasks in the system. Considering a real-time system, assume that there are p cores, and a periodic task set N = {n1, n2, …, nm}. A task ni has its period Ti, computation time Ci and deadline Di, will be placed in a ready queue until the scheduler allocates it to the core for execution. The well-known scheduling algorithms for real-time systems are EDF (Earliest Deadline First Scheduling) and H-EDF (Hierarchical-EDF) algorithm, respectively. We will also discuss the advantages and shortcomings of these scheduling algorithms and provide some guidelines to construct the proposed scheduling algorithm. EDF scheduling algorithm is the most famous and efficient in real-time operating systems. In the single-core system, EDF is an optimal scheduling algorithm [4]. It is a dynamic priority scheduling algorithm; the task priorities are not fixed but can be changed depending on the closeness of their absolute deadlines. In such a framework, we assume that there is a set of m periodic tasks as mentioned previously, the relative deadline of each task is equal to its period. If the total utilization of the task set is no greater than 1, the task set can be feasible scheduled on a single-core system by the EDF algorithm, as shown in the equation (1):
∑ CT ≤ 1 m
U=
i =1
i
(1)
i
Where task ni is with computation time Ci and the period Ti. The above equation is also implying that maximum utilization of the system is 1. In other words, if EDF cannot schedule a task set on a single-core system, there are no other scheduling algorithms can [1][7]. However, the algorithm with the best performance in the single-core platform, cannot assure to play a good role in the multi-core platforms. It is also well known that optimal scheduling for multiprocessor or multi-core system is an NP-Hard problem [2]. AS (Adaptive Scheduling) algorithm belongs to one of H-EDF algorithms that a simple hierarchical scheduling architecture similar to the EDF algorithm. For the HEDF scheduling, in the first layer, tasks are classified according to task’s category. In the second layer, tasks are scheduled according to their priorities, to ensure hard realtime tasks are not affected by soft real-time tasks to prevent from missing their deadlines. But this kind of algorithm is often ignore the requirements of the soft real time tasks, thus degrades the system performance. The proposed AFS method can be viewed as an improved version of the H-EDF algorithm. It is not only to keep the advantage of H-EDF but also to improve the schedulability of soft real-time tasks. Full migration can achieve better performance [5]. The proposed algorithm is also an improvement of the restricted migration scheduling policy, which separates the migration and no migration tasks to reduce the regular overhead of migration caused by full migration.
3 Multi-core Scheduling Scheme In this section, the proposed AFS scheduling algorithm will be introduced. We propose a mechanism for multi-core real-time scheduling algorithm. The multi-core
266
L.-T. Lee, H.-Y. Chang, and W.-M. Luk
scheduling method often needs to be modified for adapting to different situations or conditions, because the embedded platforms have different hardware, software environments with various applications. If an appropriate scheduling type in a scheduling algorithm can be chosen by developer according to different scenarios, the system performance will be improved effectively. 3.1 System Hypothesis
In a real-time system with p multi-cores, we assume that there is a task set N = {n1, n2, …, nm}. According to the equations (2) and (3) proved by Joel Goossens et al. [1], any instance of tasks can be EDF scheduled on p processors when its utilization less than p / (2p - 1). EDF algorithm in a multi-core system can be successfully scheduled if these two equations can be satisfied. As long as the task utilization in a multi-core system meets these two equations, all the tasks can be completed before their deadlines. ui ≤ p / (2p - 1),
1≤i≤m
U ≤ p2 / (2p - 1) .
(2) (3)
ui is utilization of task ni as shown in the equation (4), that is: ui = Ci / Ti .
(4)
U is the total utilization of all the tasks in the system as shown in the equation (5): Ci . i =1 Ti
U( n ) = ∑ n
(5)
3.2 Task Priority
According to the equation derived by Joel Goossens et al., a priority-driven algorithm, called EDF-US[p/(2p-1)] [2][10], for periodic task on multi-core systems has been proposed. Equations (6) and (7) are used for determining the weights of tasks. By this rule, it will determine whether the task is heavy task or a light task in the system. In this paper, we will use this rule to calculate the weights of all tasks. Heavy Task:
if ( Ci / Ti ) > ( p / ( 2p - 1 )), and 1 ≤ i ≤ m
(6)
Light Task:
if ( Ci / Ti ) ≤ ( p / ( 2p - 1 )), and 1 ≤ i ≤ m
(7)
3.3 Adaptive Feedback Scheduling Algorithm
AFS scheduling algorithm presents cross global and local scheduling algorithm. At the beginning, the tasks will be scheduled by the global scheduler. In the global scheduler, weights of tasks will be calculated by using EDF-US [p/(2p-1)] rule. Each task will also be classified a soft real-time task or a hard real-time task. Then the
An Adaptive Embedded Multi-core Real-Time System Scheduling
267
priority of the task will be decided by its weight and classification. As a result, tasks will be classified into four types: 1. 2. 3. 4.
According to the task type, a task will be assigned to a core and enter the field of local scheduling which is further divided into two categories: preemptive region and nonpreemptive region. Preemptive region follows the general single-core scheduling rules, such as RM, EDF or LLF, and so on. For the non-preemptive region, when a task enters this zone, it will be executed according to the first come first serve policy and will not be preempted by other tasks. Thus, tasks in this region can be guaranteed that it will be executed and complete quickly, Hard-heavy tasks can be executed independently to reduce the context switching overhead. In addition, to ensure the process in the non-preemptive region, once the task assigned to this category of scheduler, it will not be re-scheduled by outside global scheduler until the task is terminated [7][9]. For the strategy of distributing tasks to the local scheduler, Hard-heavy task will always be assigned to the non-preemptive region without conditions. The remainder tasks will be assigned to the preemptive region according to their weights, and then assigned to the core with the lightest loading. After entering the local scheduler, a low priority task may miss its deadline caused by waiting for execution. In local scheduler, though the deadline is approaching, the priority of the task can not be dynamically changed between cores. In order to improve this kind of situation, an adaptive feedback mechanism is adopted in the proposed scheme. Once there is a task terminated in its period, global scheduler will start re-scheduling to re-calculate weights of all tasks in the system and reallocate tasks between cores to improve the system schedulability. By selecting the appropriate computation of feedback mechanism intervals, will avoid shortcomings of over-frequent context switch which produced in the LLF algorithm [6]. When any task is terminated in its period, the system will automatically start the adaptive feedback mechanism as illustrated by straight arrow lines in the Fig. 1. Fig. 1 presents the AFS algorithm of the entire process. At first, global scheduler computes weight and classifies the type of each task, and then tasks are assigned to the corresponding local scheduler. THH will be decided whether there is enough space to enter the non-preemptive region. If there is no more space for scheduling the THH, the task will be assigned to other preemptive region with the higher priority. Adaptive feedback mechanism will be performed when any task is terminated in its period, the global scheduler will be activated again, and reallocate tasks between cores. However, tasks in non-preemptive region will not be rescheduled to guarantee accuracy of the region and decrease the context switch overhead. The AFS algorithm is described as follows: 1.
First of all, calculating the weight of each task in the global scheduler according to the equations described in subsection 3.2.
268
L.-T. Lee, H.-Y. Chang, and W.-M. Luk
Fig. 1. Adaptive Feedback Scheduling Mechanism
2.
Classifying tasks into four types: (1) (2) (3) (4)
Assign THH to the non-preemptive region. If the Hard-heavy task’s remaining time is insufficient to enter any nonpreemptive region, scheduler will assign it into the lightest loading preemptive region area. Assign remaining tasks into preemptive region according to their weights. Scheduling tasks in the local scheduler. If there is a task terminated in its period, the global dynamic scheduler will start re-scheduling to re-calculate weights of all tasks in the system, then reallocate tasks between cores excluding the tasks in the non-preemptive region. Repeat steps 2 to 7.
For applying the proposed mechanism to embedded systems with different environments, the number of preemptive regions and non-preemptive regions can be adjusted by the developer to obtain a better performance. To decide the appropriate number of these regions is very important. If system has too many non-preemptive regions, it may result in other tasks can not be scheduled successfully. In contrast, if the number of non-preemptive regions is not enough, Hard-heavy task will compete
An Adaptive Embedded Multi-core Real-Time System Scheduling
269
with other tasks with lower priorities. By selecting the appropriate number of preemptive regions and non-preemptive regions for the scheduler, system will not only maintain the accuracy of hard real-time tasks and reduce the context switch overhead, but also improve the schedulability of the soft real-time tasks.
4 Experimental Results In this section, we will compare the proposed AFS algorithm with the H-EDF algorithm. In the experiment, simulations act on the premise that all hard real-time tasks can be scheduled to be compared with the miss rate of soft real-time tasks. Through simulations, the effectiveness of the proposed scheduling mechanism will be demonstrated. 4.1 Experimental Environment
We establish the experiment environment is as follows. The multi-core number in hardware environment consists of 4 and 8 cores. The ratio of the number of nonpreemptive regions and preemptive regions is 1:3 and miss rate is measured in each type with different loadings. Here assume that all tasks are periodic and independent, and their periods are equal to their deadlines. In addition, the context switch overhead of the two algorithms in this simulation is not practical to be considered. The system loading is also classified into two types: heavy loading and over heavy loading. In the simulation, light loading task set is not considered because the total loading of the light loading task set in the system will less then 100% and all real-time tasks will meet their deadline, the simulation results will be no significant difference. The heavy loading task set is more modest degree of system loading, the total loading is about 105%, while the over heavy loading task set is higher degree of system loading, the total loading is about 110%. Every single task’s loading does not exceed 60%. With heavy and over heavy system loading testing, more soft real-time tasks will miss their deadlines, thus, the performance difference between two algorithms can be presented clearly. 4.2 Simulation
We will compare the proposed AFS algorithm with the H-EDF algorithm. In the simulation, simulations act on the premise that all hard real-time tasks can be scheduled to compared with the miss rate of soft real-time tasks. In the first simulation, light loading task set is not considered because the total loading of the light loading task set in the system will not exceed 100%, the simulation results will be no significant difference. With heavy and over heavy loading testing, more soft real-time tasks will miss their deadlines, thus, the performance difference between two algorithms can be presented clearly. Fig. 2 shows the results of the comparison of two different algorithms under the 4 cores environment. The proposed method gets lower miss rate than H-EDF, the difference becomes clearer especially under the over heavy loading testing. It means that the more the loading grows, the better effect will be illustrated by applying the proposed AFS algorithm. AFS algorithm can dynamically re-schedule the task to
270
L.-T. Lee, H.-Y. Chang, and W.-M. Luk
prevent tasks from missing their deadlines, so the schedulability of the task set can be improved significantly. In experiments, hard real-time tasks will not miss their deadlines, since the number of hard real-time tasks does not exceed the number of the total cores in the proposed system. In the experiments with 8 cores, similar results are obtained as shown in Fig. 3 the experimental results still show that difference of miss rate between two algorithms becomes larger when system loading changed from heavy loading to over heavy loading. Comparing with the previous simulation results in the 4-core environment, we can still find that applying the proposed AFS algorithm can achieve better performance than H-EDF. From the data we can find that whatever the system in the heavy or over heavy loading situation, comparing with the H-EDF algorithm, the proposed AFS algorithm can effectively reduce the occurrence of miss rate and greatly improve the system performance. The following experiments are made by gradually changing the loading of the system for comparing the proposed AFS with the H-EDF. The loading of the task set is gradually increasing from 60% to 120%. From Fig. 4 and Fig. 5, we can find that when the loading reaches 90% or more, the gap between the two algorithms would
Fig. 2. Miss Rate Diagram under 4 Cores
Fig. 3. Miss Rate Diagram under 8 Cores
An Adaptive Embedded Multi-core Real-Time System Scheduling
271
Fig. 4. Miss Rate in Different Loadings under 4 Cores
Fig. 5. Miss Rate in Different Loadings under 8 Cores
obviously be greater. Miss rate occurs rarely when the loading does not exceed 80%, and the results can not be distinguished clearly. As the loading gradually increases, the difference of miss rate between two algorithms became apparent, especially when the loading reaches 110%. The miss rate increases abruptly in both algorithms when the loading becomes larger than 110%. Because in the heavy loading environment, it has seriously exceeded the loading that system can support. But comparing with the H-EDF algorithm, the miss rate of AFS algorithm is relatively low.
5 Conclusion In the multi-core real-time platform, we propose a modified scheduling algorithm with adaptive scheduling and the feedback mechanism which can take care of both hard real-time and soft real-time tasks. It is a multi-level dynamic priority-driven algorithm that is based on the weight-driven to assign the task to the appropriate local scheduler, and by using feedback mechanism to monitor the system for reducing the miss rate of the soft real-time tasks. This scheduling algorithm can be easily adapted and changed to fit a variety of embedded environments. The simulation results show
272
L.-T. Lee, H.-Y. Chang, and W.-M. Luk
that the proposed scheduling strategy can effectively improve the schedulability of the system. In addition to ensuring the validity of a hard real-time task, the meet radio of soft real-time task can also be increased.
References 1. Goossens, J., Funk, S., Sanjoy, B.: Priority-Driven Scheduling of Periodic Task Systems on Multiprocessors. Real-time Systems 25(2-3), 187–205 (2003) 2. Banús, J.M., Arenas, A., Labarta, J.: Dual Priority Algorithm to Schedule Real-Time Tasks in a Shared Memory Multiprocessor, pp. 2–12. IEEE Computer Society, Washington, DC (2003) 3. Jean, J.: Labrosse, MicroC OS II The Real Time Kernel, 2nd edn. CMP Books (2002) 4. Chen, K.-Y., Liu, A., Lee, C.-H.L.: A Multiprocessor Real-Time Process Scheduling Method. In: Proceeding of Fifth International Symposium on Multimedia Software Engineering, pp. 29–36 (2003) 5. Huerta, P., Castillo, J., Martinez, J.I., Pedraza, C.: Exploring FPGA Capabilities for Building Symmetric Multiprocessor Systems. In: Proceedings of 3rd Southern Conference on Programmable Logic, pp. 113–118 (February 2007) 6. Oh, S.-H., Yang, S.-M.: A Modified Least-Laxity-First Scheduling Algorithm for RealTime Tasks. In: Proceedings of Fifth International Conference on Real-Time Computing Systems and Applications, pp. 31–36 (1998) 7. Pathan, R.M., Jonsson, J.: Load Regulating Algorithm for Static-Priority Task Scheduling on Multiprocessors. In: 2010 IEEE International Symposium on Parallel & Distributed Processing (IPDPS), April 19-23, pp. 1–12 (2010) 8. Tan, P.: Task Scheduling of Real-time Systems on Multi-Core Architectures. In: International Symposium on Electronic Commerce and Security, ISECS 2009, May 22-24, pp. 190–193 (2009) 9. Bertogna, M., Cirinei, M., Lipari, G., Member, IEEE: Schedulability Analysis of Global Scheduling Algorithms on Multiprocessor Platforms. IEEE Transactions on Parallel and Distributed Systems, 553–566 (April 2009) 10. Easwaran, A., Lee, I., Shin, I.: Hierarchical Scheduling Framework for Virtual Clustering of Multiprocessors. In: Euromicro Conference on Real-Time Systems, ECRTS 2008, July 2-4, pp. 181–190 (2008)
Query Processing Systems for Wireless Sensor Networks Humaira Ehsan and Farrukh Aslam Khan Department of Computer Science, National University of Computer and Emerging Sciences, A.K. Barohi Road H-11/4, Islamabad, Pakistan {humaira.ehsan,farrukh.aslam}@nu.edu.pk
Abstract. Wireless Sensor Networks (WSNs) have been widely used during the past few years and have wide areas of applications. In today’s era, the users require a sophisticated state of the art system to control and monitor the sensor network. The old centralized systems to collect sensor data have become obsolete because of the lack of flexibility in data extraction and scalability problems. Motivated by the success of distributed database systems, the concept of viewing the sensor network as a sensor database system was proposed which gained a lot of popularity. Based on that concept, many systems have been developed for query processing in WSNs. In this paper, we discuss all those existing systems and compare them based on their working and performance. Keywords: Wireless Sensor Networks (WSNs), Query Processing Systems, Query Optimization.
limited computation and communication capabilities. The first and far most important goal of WSN applications is minimizing energy consumption in order to have a long living network. WSN users are typically interested in continuous streams of sensed data from the physical world. Query processing systems provide a high-level user interface to collect, process, and display continuous data streams from sensor networks. These systems are high-level tools that facilitate the application developers and ad hoc users to rapidly develop and use wireless sensor network applications. In contrast, writing WSN applications in a systems language such as C or Java is tedious and error-prone. A query processing system abstracts the users from tasks such as sensing, forming an ad-hoc network, multi-hop data transmission, and data merging and aggregation [3]. The responsibility to deploy and manage these networks is usually allocated to an owner that acts as a single controlling entity. Research has shown that it is so far not possible to deal with a secure multi-purpose federated sensor network, involving tens of thousands of sensor nodes running different applications in parallel and able to reconfigure dynamically to run others. Due to this reason, a sensor network is often dedicated to a single application. Various factors are taken into consideration while allocating sensing bandwidth and computation resources including the query load and the priority and urgency of each application. Special care is taken to provide desirable quality-of-service, whilst preserving fairness, secure operation and privacy across applications. During the past few years, many systems have been developed by researchers for query processing in WSNs. In this paper, we discuss all those existing systems and provide a comprehensive comparison of these systems based on their working and performance. The rest of the paper is organized as follows: In sections 2, various existing query processing systems are discussed in detail. Section 3 discusses various specific features of these systems and presents their comparison. Finally, section 4 concludes the paper.
2 Query Processing Systems The data centric approach of tasking WSNs was first formally introduced in [4]. After that there have been many other approaches proposed. In general, there are two broad categories of query processing; the centralized approach and the distributed approach. In centralized approach a predefined set of data is regularly delivered from the sensors to a central location where it is stored in a database. User queries that database through some interface provided by the system. This approach is similar to the warehouse approach of traditional database systems but it is not very suitable for WSNs. The problem is that most of the WSN applications require real time data and offline data is of no use in such scenarios. Secondly, communicating bulk of data from sensors to sink periodically wastes lots of resources. The second approach is distributed approach where data is kept on the sensors and part of the processing is done there and only the required data is sent to the sink. Distributed query processing in WSNs has been an active research area over the last few years. In the distributed
Query Processing Systems for Wireless Sensor Networks
275
approach, the data that ought to be extracted from sensors is determined by the query workload. The distributed approach is, therefore, not only flexible such that different queries extract different data from the sensor network, but it is also efficient ensuring extraction of only relevant data from the sensor network. TinyDB [5] [6] and Cougar [7] represent the first generation of distributed query processing systems in WSNs. Other systems that will be discussed are Corona [8] [9], SINA [10] and SenQ [11]. 2.1
Sensor Information Network Architecture (SINA)
Sensor Information Network Architecture (SINA) [10] is a middleware designed for querying, monitoring and tasking of sensor network. The functional components of the SINA architecture are: Hierarchical clustering, attribute-based naming, and location awareness. Due to the large number of sensors nodes, nodes are divided into clusters and each cluster is assigned a cluster head. In this way a hierarchy of clusters is formed and all the information filtering, fusion and aggregation is performed through cluster heads. Most of the sensor applications heavily depend on the physical environment and location of the sensors, therefore, location information is very important component of SINA. Location information can be obtained through GPS, but because of economical reasons all sensor nodes cannot be equipped with GPS. A number of techniques are available to solve this issue and any of them can be used in this component. Sensor Query and Tasking Language (SQTL) [10] is a procedural scripting language which is used as a programming interface between applications and SINA middleware. A sensor execution environment (SEE) runs on every node, which is responsible for receiving, interpreting and dispatching SQTL messages. SQTL has many arguments which are used to generate various types of actions. SINA provides various information gathering methods and according to the application requirements, combinations of those methods are used appropriately. In dense sensor network, generation of response from each node and passing it to the sink cause response implosion. Some applications may not need response from every node; response of some of the nodes from certain area may be enough. Through experiments, authors have shown that large amount of collisions can be caused if none of the information gathering techniques is used. The diffused computation technique performs better than all others. 2.2
TinyDB
TinyDB [6] is an acquisitional query processing system which is designed to work on UC Berkeley motes. They focus on the fact that sensors have control over where, when, and how often data is physically acquired. It is the most widely used system. Its prominent features are intelligent query processing, query optimization, and power efficient execution. It does fault mitigation by automatically introducing redundancy and avoiding problem areas. Fig. 1 illustrates the basic architecture of the system.
276
H. Ehsan and F.A. Khan
Fig. 1. Basic Architecture of TinyDB
In TinyDB, the sensor tuples belong to a table sensor which, logically, has one row per node per instant in time, with one column per attribute (e.g. light, temperature, etc.). It uses TinyOS platform and TinySQL for writing declarative queries. TinySQL queries have the form: SELECT , [FROM {sensors | }] [WHERE <predicates>] [GROUP BY <exprs>] [SAMPLE PERIOD | ONCE] [INTO ] [TRIGGER ACTION ] Given a query specifying user’s data interests, TinyDB collects that data from motes in the environment, filters it, aggregates it together, and routes it out to a PC. To use TinyDB, TinyOS components need to be installed onto each mote in the sensor network. TinyDB provides a simple Java API for writing PC applications that query and extract data from the network; it also comes with a simple graphical querybuilder and result display that uses the API. TinyDB uses a flooding approach to disseminate the queries throughout the network. The system maintains a routing tree rooted at the user. Every sensor node has its own query processor that processes and aggregates the sensor data and maintains the routing information. The important features that TinyDB includes are: metadata management, network topology and multiple queries handling. 2.3
COUGAR
COUGAR: The Network Is The Database [12] was a project of Cornell University database systems group. They believe that declarative queries are very well suited for WSN applications. They have proposed a query layer for declarative query processing. As in WSNs, computation is cheaper than communication in terms of energy efficiency, they have also proposed in network aggregation which suggests
Query Processing Systems for Wireless Sensor Networks
277
that instead of communicating all of the raw data to the base station, results should be aggregated at intermediate nodes and then communicated towards the base station. Because of the diversity of WSN applications, the requirements in terms of energy consumption, delay and accuracy vary from application to application. This system can generate different query execution plans according to the requirements of different applications. The query plans are normally divided into two components: communication component and computation component. A query plan decides how much computation will be pushed into the network and specifies the responsibility of each sensor node i.e., how to execute the query, and how to coordinate the relevant sensors. The network is viewed as a distributed database which has multiple tables where each table corresponds to a sensor type. Their proposed software component which should be deployed on each sensor is called a query proxy. The proposed query template is given below: SELECT FROM WHERE GROUP BY HAVING DURATION EVERY
{attributes, aggregates} {Sensordata S} {predicate} {attributes} {predicate} time interval time span e
The long running periodic queries are supported by “DURATION” and “EVERY” clause. Authors have proposed three approaches for in-network aggregation. First is Direct Delivery, in which the leader nodes do the aggregation and each sensor sends its data towards the leader. Second is Packet Merging, in which several records are merged into a single packet and that packet is sent; in this way packet overhead is incurred only once. Third is Partial Aggregation, in which each node computes the partial results and those results are sent to the leader. The last two techniques need modification in routing protocol as the packets need to be intercepted and modified packets need to be generated. To perform packet merging or partial aggregation, synchronization between the sensors is required. They have not yet developed a complete working system but their ideas have been partially tested using NS-2. 2.4
Corona
This project Corona [9] was previously named as Sun SPOT Distributed Query Processing (SSDQP) [3]. Corona is a distributed query processor, developed at the School of IT, University of Sydney. The system is implemented on Sun SPOTs which is new state of the art sensor network hardware with full java support. The platform provides much more memory and computational power than previous generation of sensor nodes i.e., Berkley Motes. The system is fully written in Java on top of the Sun SPOT’s Squawk VM, a lightweight J2ME virtual machine, which makes is easy to maintain and extend. The system consists of three components as shown in Fig. 2, i.e., 1. The query engine that is executed on the Sun SPOTs 2. The host system on the user’s PC that is connected to the base station 3. A GUI client which connects via TCP/IP to the host system.
278
H. Ehsan and F.A. Khan
Fig. 2. Basic Architecture of Corona
Corona uses a variant of an acquisitional SQL which provides all the features of a querying language. A unique feature of corona query processor is that it can execute multiple queries simultaneously, due to which the same network can be used for different applications. As energy efficiency is the primary goal of every system designed for WSNs, Corona also has components to ensure efficient energy utilization such as in-network clustering operator which is resource-aware and dynamically adapts its processing granulites to keep the number of transmitted messages small. 2.5
SenQ
SenQ [11] is an embedded query system for interactive wireless sensor networks (IWSNs). IWSNs are human centric and interactive. Applications of this area require very different category of features. The key challenges that SenQ addresses are heterogeneity, deployment dynamics, in-network monitoring, localized aggregation, and resource constraints. General architecture of SenQ is illustrated in Fig. 3.
Fig. 3. Architecture of SenQ
It has a layered system design. The lowest two layers require lesser computation and storage and these layers reside on the embedded sensor devices. Layer 3 of query management and data storage resides on the micro server. Layer 4 is a declarative language like SQL and it is called SenQL. These layers are loosely coupled to deal
Query Processing Systems for Wireless Sensor Networks
279
with diverse application requirements. The design of SenQ is very flexible and can be easily adapted according to the application requirements. SenQ supports two types of queries: snapshots and streaming. Streaming queries collect data and report back results continuously until a stop command is executed, while snapshots provide efficient point in time samples of data. Authors have evaluated SenQ's efficiency and performance in a test bed for assisted-living.
3 Features Comparison The above discussed sensor network query processing systems are compared in this section. First, the comparison parameters are discussed and then the comparison is provided in tabular form in Table 1. 3.1 Event-Based Queries These queries are executed only when interesting events happen, for example, button pushed, some threshold sensor reading sensed, bird enters nest etc. Events in TinyDB are generated explicitly, either by another query or by a lower-level part of the operating system. SenQ uses EventSensor drivers in sensor sampling and processing layer to generate data sporadically. SINA has arguments in SQTL which can be used to trigger events periodically or when a message is received by the node. 3.2 Life-Time Based Queries The sensor network should have a way to specify long running periodic queries parameter. In TinyDB, the SQL clause “LIFETIME <>” is used to create life time based queries. In Cougar, the same is accomplished by using DURATION and EVERY clauses of SQL. SenQ provides the same feature through streaming queries. 3.3 In-network Aggregation In this technique instead of passing raw values in the network, the sensor nodes pass on aggregated data along the routing path. This technique is very efficient in saving constrained resources of the network. TinyDB provides various techniques for innetwork aggregation. Cougar was the first project in which in-network aggregation was introduced and then implemented. Cougar have done this by modifying network layer in NS-2. SenQ provides the features of temporal aggregation and spatial aggregation in its sensor sampling and processing layer. SINA has implemented this feature in information gathering component through diffused computation operation. 3.4 Multi-query Optimization In practice, sensor network query systems supports many users accessing the sensor network simultaneously, with multiple queries running concurrently. The simple approach of processing every query independent of the others incurs redundant communication cost and draining the energy reserves of the nodes [13]. In multi query optimization the concept of optimal result sharing for efficient processing of
280
H. Ehsan and F.A. Khan
multiple aggregate queries is suggested. Corona claims to do the multi-tasking optimally. SenQ does support multi-query execution but it is not very clear whether they do optimization or not. 3.5 Heterogeneous Network The network nodes can be heterogeneous in terms of energy, bandwidth, computing and storage capabilities, availability of pre-computed results etc. SenQ is designed to deal with all kinds of heterogeneity in the network. 3.6 Time Synchronization As WSNs can be dynamic, nodes leave and join very often. To keep the nodes synchronized efficient time synchronization technique is required which consumes minimum energy and gives accurate results. 3.7 Scalability in Network Size The system should be scalable for larger networks. Generally, WSNs consist of large number of sensor nodes so the performance of the system should not degrade with the increase in network size. The centralized query processing systems had the scalability problems but distributed approach normally scale well with the network size. Table 1. Comparison of various query processing techniques Criteria Platform
SQL type query interface GUI In-network aggregation Multi query optimization Event Based Queries Life Time based Queries Heterogeneous Networks Support Time Synchronization Scalability in network size
TinyDB Berkeley Motes + TinyOS Yes
Cougar Simulation
Corona SUN SPOT
SenQ TinyOS
SINA Simulation
Yes
Yes
Yes
Yes
Yes
No
Yes
Yes
No
Yes
Yes
Yes
Yes
Yes
No
No
Yes
Yes
No
Yes
No
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
No
No
No
Yes
No
Yes
Partial
Yes
Yes
No
Not Clear
Not Clear
Yes
Yes
Yes
Query Processing Systems for Wireless Sensor Networks
281
3.8 Interfaces Interfaces are very important for the users of the system. Cougar provides no graphical user interface (GUI). TinyDB provides the GUI, SQL and the programming abilities. Corona also has both the GUI and the SQL interfaces. SenQ has the most sophisticated interfaces for all categories of users i.e., database experts, application programmers, and ad-hoc users. SINA provide interface through procedural scripting language but no GUI is present.
4 Conclusion A review of existing query processing systems for Wireless Sensor Networks (WSNs) is presented in this paper. Generally, TinyDB is the most widely used system because of its availability. It is a ready to use system in standard mica-mote networks in which the user enters simple SQL-like queries into base station PC. High degree of optimization is possible. But it requires modification of underlying network layer or development of a “wrapper” around the layer to provide the required functionality. Cougar is yet to be implemented on a real test bed. Corona is latest of all the systems and is good as the platform used has more capabilities than the Mica motes. It can be a system for more powerful sensor networks of next generation. SenQ is mainly targeted for a sub-domain of WSNs i.e., interactive WSNs, but because of its loosely coupled layered architecture it can be adapted for any kind of WSN. Its support for heterogeneous networks makes it suitable for all kinds of applications. A feature comparison of the most famous query processing systems has been presented in this paper. However, a true performance comparison of the existing systems is still required in which the energy-efficiency, accuracy, delay and results of these systems need to be compared. Standard benchmarks should also be designed to test the performance of such systems.
References 1. Trigoni, N., Guitton, A., Skordylis, A.: Chapter 6: Querying of Sensor Data. In: Learning from Data Streams: Processing Techniques in Sensor Networks, pp. 73–84. Springer, Heidelberg (2007) 2. Amato, G., Baronti, P., Chessa, S.: Query Optimization for Wireless Sensor Network Databases in the MadWise System. In: Proc. of SEBD 2007, Torre Canne, Fasano, BR, Italy, pp. 242–249 (2007) 3. Scholz, B., Gaber, M.M., Dawborn, T., Khoury, R., Tse, E.: Efficient time triggered query processing in wireless sensor networks. In: Lee, Y.-H., Kim, H.-N., Kim, J., Park, Y.W., Yang, L.T., Kim, S.W. (eds.) ICESS 2007. LNCS, vol. 4523, pp. 391–402. Springer, Heidelberg (2007) 4. Intanagonwiwat, C., Govindan, R., Estrin, D.: Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks. In: Sixth Annual ACM/IEEE International Conference on Mobile Computing and Networking, Boston, USA (2000)
282
H. Ehsan and F.A. Khan
5. Madden, S.R., Franklin, M.J., Hellerstein, J.M., Hong, W.: TinyDB: An Acquisitional Query Processing System for Sensor Networks. ACM Trans. Database Syst. 30(1), 122– 173 (2005) 6. TinyDB: http://telegraph.cs.berkeley.edu/tinydb/overview.html 7. Demers, A., Gehrke, J., Rajaraman, R., Trigoni, N., Yao, Y.: The Cougar Project: A Work in Progress Report (2003) 8. The Corona Website (2009), http://www.it.usyd.edu.au/~wsn/corona/ 9. Khoury, R., et al.: Corona: Energy-Efficient Multi-query Processing in Wireless Sensor Networks. In: Kitagawa, H., Ishikawa, Y., Li, Q., Watanabe, C. (eds.) DASFAA 2010. LNCS, vol. 5982, pp. 416–419. Springer, Heidelberg (2010) 10. Jaikaeo, C., Srisathapornphat, C., Shen, C.: Querying and Tasking in Sensor Networks. In: SPIE’s 14th Annual Int’l. Symp. Aerospace/Defense Sensing, Simulation, and Control, Orlando, FL, (2000) 11. Wood, A.D., Selavo, L., Stankovic, J.A.: SenQ: An Embedded Query System for Streaming Data in Heterogeneous Interactive Wireless Sensor Networks. In: Nikoletseas, S.E., Chlebus, B.S., Johnson, D.B., Krishnamachari, B. (eds.) DCOSS 2008. LNCS, vol. 5067, pp. 531–543. Springer, Heidelberg (2008) 12. Cougar, http://www.cs.cornell.edu/bigreddata/cougar/index.php 13. Demers, A., Gehrke, J., Rajaraman, R., Trigoni, N., Yao, Y.: Directions in Multi-Query Optimization for Sensor Networks. In: Advances in Pervasive Computing and Networking, pp. 179–197. Springer, Heidelberg (2004)
Educational Principles in Constructivism for Ubiquitous Based Learning Sung-Hyun Cha, Kum-Taek Seo, and Gi-Wang Shin
Abstract. This study aims to explore educational principles applicable to ubiquitous based learning from constructivism. For this, characteristics of ubiquitous and principles of constructivism are investigated, and then compared and extracted common characteristics for ubiquitous based learning such as interactivity, social contextuality, and individuality. Keywords: constructivism, ubiquitous, ubiquitous based learning.
In order to develop more effective u-learning programs, however, it is necessary to make a solid investigation of characteristics of ubiquitous and educational principles that can be applied to ubiquitous environments. In this study, we considered constructivism as a concept of educational principle for ubiquitous based learning and explored common characteristics between ubiquitous and constructivism. Furthermore this study would present a theoretical basis that determines which characteristics of the constructivism can be applied to ubiquitous based learning.
2 Background Studies 2.1 Characteristics of Ubiquitous Paradigm Ubiquitous paradigm could be summarized into the following eight characteristics applicable to education[5]. First, it is the connectivity that makes works with ease whenever and wherever they are required. Second, it is the reality that represents an actual space that is combined by electronic and physical spaces instead of a virtual space. Third, it is the mobility that performs easy movements using portable devices rather than desktop PCs. Fourth, it is the ubiquity that makes an easy work through not only mobile devices but also computers existed in around objects and environments. Fifth, it is the convergence that plays a role in carrying various functions in a single device. Sixth, it is the intelligence that performs a work through intelligence devices, which are able to implement autonomous sensing, environment adaptation, context awareness, and so on. Seventh, it is the personalization that provides a proper learning for each person, considering learners' personal characteristics. Eighth, it is the variety that creates a new learning type, which did not exist before then. Among eight characteristics of ubiquitous paradigm, variety, convergence, reality, and intelligence are performed through personalization. While these are closely related to interactivity and connected to social context, other characteristics such as connectivity, mobility, and ubiquity are the unique characteristics of ubiquitous. 2.2 Characteristics of the Constructivism The constructivism stresses the composition of knowledge by learners themselves. In this viewpoint, the knowledge composed by individuals are through social tuning and then become meaningful[6]. Vygotsky represented social constructivism and proposed that knowledge exists in the social context and is not presented as an isolated figure in a person's mind but shared other people in the origin[7]. From this social constructivism, adolescence is a growth period that can be influenced by the relationship with other people, and the interaction between people in daily life is very important in this period. Education activities in this period are naturally performed through usual interactions and
Educational Principles in Constructivism for Ubiquitous Based Learning
285
communications with other people in which they reestablish their experience and knowledge and learn sociality. According to the social constructivism, people refines their own ideas and helps forming other people's ideas through interaction. That is, social knowledge, or social adaptability is build up through such process as accommodation, assimilation, and control in social contexts. Piaget mentioned that the concept of equilibration is not to be considered as a static state but a dynamic state for fully understanding it. It means that the equilibrium is not a continuous process that shows a sequential process of assimilation, complication, and control but a dynamic process that attempts to make an equilibrium in adaptation and construction, growth and change like dancing[8]. In addition, according to Piaget, structure represents three properties, such as whole, transformation, and self-regulation. The whole means a comprehensive systematic idea more than the sum of parts. However, the parts are related to each other and interact between them. Also, the parts are not separated from the whole and represent no meaning in the parts themselves. That is, the parts have only its value as they are related to the whole or represent the relation to the parts themselves. Finally, transformation explains the relation to the parts. That is, it represents how a part becomes to an another part. Based on this process, it is possible to recognize the process that makes a change in the parts. In addition, each structure basically represents self-maintenance, self-organization, and self-regulation for closure. Based on such principles and methods of constructivism above, the characteristics of the constructivism could be presented in [Table 1]. Table 1. Characteristics and methods in the constructivism Characteristics
3 Relationship between Ubiquitous and the Learning Principle of Constructivism Learning activities based on constructivism linking with ubiquitous paradigm and its technologies could be characterized as follows[9]. First, it is possible to perform customized learning for each person. It can provide most proper learning contents for learners by identifying learners' awareness level,
286
S.-H. Cha, K.-T. Seo, and G.-W. Shin
sensibility, hobby, personality, and so on at all times, using intelligence terminals and inherent sensors. Second, it emphasizes self-directed learning. According to the process of customized learning for each person, learners determine their own learning objectives by themselves instead of accepting the provided learning objectives and select the resources required in their learning. Third, it stresses learning activities based on cooperation. As the free access to learning spaces is allowed without time, interactions between learners are made naturally and instantaneously. Fourth, periodic evaluations for learning activities are to be performed, because learning activities of learners are always checked and analyzed for providing most proper learning information to learners. Fifth, it provides learning resources flexible to learners. It freely provides ways of transformation and delivery of information and resources that are satisfied for each learner and supports the hardware, software, and interface in order to increase the learning immersion under the environment that represents lots of movements. Sixth, it provides multi-dimensional learning spaces. It freely connects physical and cyber spaces as well as online and offline learning organizations that provides customized learning resources based on various representation formats timely. Also, it provides community learning environments. Seventh, it provides integrated learning of cognition, affection, and psychomotor. It provides learning resources for various learning objectives and moral and emotional development education. Eighth, it serves as a facilitator. A teacher guides learners to the proper information and helps to have constructive interactions. As educational applications of constructivism using ubiquitous technologies are presented above, teachers should play a role in a facilitator and learners should have a self-directed learning attitude. In addition, customized and integrated instructionlearning methods could be applied. Also, periodic and formative evaluation methods could easily be used. Moreover, the ubiquitous based learning is effectively employed to establish learning environments of constructivism due to the following characteristics [10]. First, the ubiquitous based learning is useful to determine instructional objectives. In the viewpoint of constructivism, learning objectives can be presented as an authorized project in practical environments according to the interest of learners. Second, the ubiquitous based learning provides varieties in learning material presentation. Learners could make their meanings through the presented materials. Here, the materials could be modified and reorganized according to the instructionlearning objectives. Third, because the ubiquitous based learning presents virtual learning environments quite similar to the actual environments, it may provide experiences similar to the real world. Fourth, the ubiquitous based learning provides various interactions. Fifth, the ubiquitous based learning enables learners to listen other people's various ideas. Also, it forms the knowledge structure in learners' mind by identifying the results of their learning and provides feedbacks to review the required sections.
Educational Principles in Constructivism for Ubiquitous Based Learning
287
Table 2. Educational applicability of constructivism with the characteristics of ubiquitous Advantages
Application Possibility of Constructivism Lee Young-Mi(2006)
Kim Ran(2007)
Lim SunYoung(2006)
Immediacy
Providing immediate feedbacks -Providing learning information by experts -Possible to learn within a short term based on the objectives of education
-Supporting to immediately solve problems because required information can be obtained whenever and wherever it is required. -Possible to write or record the questions occurred in a site
―
Individuality/Variety
-Accessing the learning management information of learners -Simple evaluation of learners -Providing real-time learning information of SRM
―
-Remote learning provides varieties in material presentation -Learners can compose the meaning of learning using the presented materials.
―
-Learners can interact with experts, teachers, and fellows simultaneously and unsimultaneously.
-Providing various interactions -Learners can establish their own knowledge structure and provide feedbacks to review the required sections.
Easy access to learning information -Supporting selfdirected learning -Performing resource based learning
-Easy access to the document, data, video, and etc. made by learners -Supporting selfdirected learning
―
Applicable to field experience learning -Field trip activities -Contextual learning
-Learning exists in daily life, and problems and related knowledge can naturally be presented as a closed form to daily life.
-Providing virtual learning environments quite similar to actual environments -Providing learners' practical experiences similar to the real world
Interactivity
Accessibility in information
Contextuality
288
S.-H. Cha, K.-T. Seo, and G.-W. Shin
The educational applicability of constructivism can be extracted from the advantages of ubiquitous, as presented above. [Table 2] shows a summary of previous studies that presented the educational applicability of constructivism. From the literature review above, educational applicability of constructivism with ubiquitous can be summarized as immediacy, individuality/variety, interactivity, easy accessibility in information, and contextuality. In the comparison of these elements with the characteristics of constructivism presented in this study, the interactivity and social contextuality are common elements. other characteristics - dynamics, openness, mobility, transformation, and whole are presented as the elements of individuality/ variety. In addition, the immediacy and the easy access in information could be considered as the internal characteristics of ubiquitous. Therefore, the common characteristics between ubiquitous and constructivism are convergent to the social interactivity, social contextuality, and individuality. It means that the principle of constructivism exists in the ubiquitous based learning. The interactivity is a core principle of the constructivism representing mutual understanding and communication, and also are presented by the communication between tools and organisms in ubiquitous paradigm. The social contextuality largely means the time and space that learning happens in a social context. In this regard, social constructivism stresses schools as a social context that learning happens, as the interactivity between learners occurs in schools. Ubiquitous paradigm also emphasizes such social contextuality in terms that knowledge and information can be obtained through social life, which interact between a person and other people. The individuality is another core characteristic of constructivism and ubiquitous paradigm. The principles of constructivism such dynamics, openness, transformation and whole are closely related to individuality, in terms that these principles take into account of learns' individual characteristics including learning of level. Also, ubiquitous paradigm deals with such individuality as a key element representing personalization for a proper learning for each person.
4 Conclusion In recent years, developments in ubiquitous technologies largely affect not only politics, economics, society, and cultures but also education - especially instructionlearning methods. In order to extend usefulness of ubiquitous in education beyond a simple and convenient tool, educational principle is to be combined and applied. In this rationale, this study compared inherent characteristics of ubiquitous paradigm and principles of constructivism. The three common characteristics are extracted from these two sides; interactivity, social contextuality, and individuality. In order to directly apply principles of constructivism to ubiquitous based learning environments, however, there is a condition; creation of knowledge and information are to be encouraged in self-directed ways within a whole education system. Ubiquitous based learning can be more effective, ensuring that the extracted principle of constructivism is to be fully reflected.
Educational Principles in Constructivism for Ubiquitous Based Learning
289
References 1. KERIS: Development of teaching and learning model in ubiquitous learning environment, CR 2005-12 (2005) 2. Zheng, Y.L., Li, L.Y., Ogata, H., Yano, Y.: Using context-awareness to support peer recommendation in e-learning settings. In: Cantoni, L., McLoughlin, C. (eds.) Proceedings of ED-MEDIA 2005, Montreal, QB, Canada, pp. 2367–2373. Association for the Advancement of Computing in Education, Chesapeake (2005) 3. Lee, Y.-M.: The Effect of Question-Generation Strategy on Reflection in Web-based Discussion. The Korean Journal of Educational Methodology Studies 18(1), 95–117 (2006) 4. Kim, R.: A study on the effect of u-learning integrated project studying program for improving self-directed learning ability. Unpublished master’s dissertation, Chonnam National University (2007) 5. Ministry of Information and Communication: Cross-border circulation of personal onformation in the age of ubiquitous computing : A study of relevant policies in foreign countries and korea’s response measures (2005) 6. Brown, J.S., Collins, A., Duguid, P.: Situated cognition and the culture of learning. Educational Researcher 18, 32–42 (1989) 7. Vygotsky, L.S.: Mind in society: The development of higher psychological processes. Harvard University Press, Cambridge (1978) 8. Catherine, T.F. (ed.): Constructivism: Theory, perspectives and practice, 2nd edn. Teachers college press, Columbia University, New York (1989) 9. KERIS: Study on the development and application of a ubiquitous-based classroom environment model, RR 2007-1 (2007) 10. Lim, S.-Y.: A study on the implementation and application of the courseware for learning computers in elementary school. Unpublished master’s dissertation, Yeungnam University (2006)
A Study on Utilization of Export Assistance Programs for SMEs and Their Exportation Performance in Korea* Woong Eun, Sangchun Lee**, Yong-Seok Seo**, and Eun-Young Kim*** Hannam University, 133 Ojeong-dong, Daeduk-gu, Daejeon 306-791, Korea [email protected]
Abstract. This study aims to explore how export assistance programs influence exportation performances for export-oriented SMEs in Korea. Recently, SMEs are focusing on overseas marketing as a means to expand their export. Many other advanced countries are also presenting various export assistance policies to develop their export-oriented SMEs. This study will verify the causal relationship between the utilization of export assistance programs and exportation performances based on comparative analyses of export assistance programs of Korea and other countries. In addition, this study will identify the barriers of export marketing and operational characteristics of export assistance programs to assess the effectiveness and validity of the current various export assistance programs. Keywords: Export assistance programs, Exportation performances, Export assistance policies, Utilization, SMEs.
1 Introduction According to an OECD statement in 20091, as for a country's competitiveness, global Small and Medium Enterprises (SMEs) will perform critical roles in the global environment in the 21st century. In OECD member states, SMEs form more than 95% of all businesses and provide 60-70% of employment. Hence, every country is strengthening their policies to cultivate SMEs which have become the center of economic growth and job creation. Furthermore, all countries consider the competitiveness of SMEs to be a major element in determining the competitiveness of an industry or a country, which increases interest in SME development as a major national policy. While the recent free trade agreements with United States and European Union are likely to expand the export market of SMEs, it is not easy for SMEs to explore overseas markets on their own. In reality, SMEs in Korea have their own limit in terms of capacity. Under a management environment like this, one of the tasks to be fulfilled by SMEs in Korea is to expand export through entry into overseas markets. *
This research was supported by the 2011 Hannam University Grant. Worked at Kookmin University. *** Corresponding author. 1 European Commission enterprise and industry directorate-general (2007), Final Report of the expert group, Supporting the Internationalization of SMEs, p.5. **
A Study on Utilization of Export Assistance Programs for SMEs
291
Active governmental support will be necessary to make this come true in addition to the efforts of the companies themselves. As a solution from red letters in the balance of international payments to unemployment, export expansion of SMEs has already become the choice to central and local governments of various countries including the USA, and have come up with various export assistance programs to support SMEs 2 . Also as countries like USA, Japan, and Europe recreate barriers to export through non-tariff barriers like the environmental regulations, the export environment for SMEs is deteriorating due to world financial crisis. As a result, many countries provide various support policies to increase the competitiveness of SMEs and the Korean government will need to increase various support policies and explore ways of improvement to increase export by SMEs. Currently, in Korea, central and local governments such as the Small and Medium Business Administration (SMBA), the Ministry of Knowledge and Economy as well as export-related organizations such as the Small and Medium Business Corporation (SMBC), the Korea International Trade Association (KITA), and the Korea Trade and Investment Agency (KOTRA), are carrying out various supportive policies to promote the export by SMEs. Nevertheless, the effectiveness use of such policies tend to decrease for most SMEs or they do not have information at all3. Also, in some export assistance programs, are overlapped and lead to criticism of their ineffectiveness4. Hence, this study will verify a causal relationship between the level of export assistance programs usage and exportation performances based on the previous studies on export assistance programs. It will also provide characteristics of SMEs, barriers to export marketing, and operational characteristics of export assistance programs to show the effectiveness and validity of the current export support programs. Especially, this study aims to verify the level of awareness and use of export assistance programs by institution to present concrete plans to increase export performances. Furthermore, it will find problems with export assistance programs and present points of improvement with a specific purpose to secure export competitiveness of Korean SMEs in global markets.
2 The Support Policies for Export-Oriented SMEs in Korea and Other Countries Governmental institutions, export-related organizations, financial institutions, private organizations and local governments are providing various export assistance programs for Korean SMEs. This study will explore the relationship between export assistance program and its performance but limits its subject to export assistance programs of governmental institutions and export-related institutions for analyses. Financial support programs, among other various export assistance programs will be excluded. This is because SMEs may find it difficult to receive financial benefits due to the 2
3
4
Yeoh, Poh-Lin(1994), "Entrepreneurship and Performance: A Proposed Conceptual Model."In S.Tamer Cavusgil, & Catherine N. Axinn, Advances in International Marketing, Vol.6, pp.43-68. Jie, Yong-hee (1994), “Plans to Improve International Competitiveness of SMEs”, ⌜Seogang Harvard Business⌟, pp.86-92. Kim, Soo-yong (2004), “Plans to Support Marketing Facilitation of SMEs”, Economy Research Center, Industrial Bank of Korea.
292
W. Eun and E.-Y. Kim
WTO's prohibition of export subsidies and most of the support programs closely related to export performances of SMEs are of export marketing assistance. Therefore, this study will focus on export marketing to explore the types and characteristics of assistance programs related to export marketing. Export assistance institutions for SMEs are established based on the Regulation on Establishment and Operation of Export Assistance Centers for SMEs. Based on exploration of export assistance programs provided by the Korean government and export-related organizations, there were several overlapping assistance programs despite the unique characteristics and functions of each organization. The following, Table 1 summarizes the most significant assistance programs by institution among various assistance programs in Korea. Table 1. Types of Assistance Programs by Institution Type
Institution
Assistance Area Transformation of companies for domestic consumption to export-oriented companies Designation prospects
SMBA
of
SMEs
with
good
export
Assistance with security of overseas standards certification 10-year Export-oriented SMEs 500 Program Dispatching SMEs trade promotion teams Development of global brands of export-oriented SMEs Dispatching a trade mission
KOTRA
Assistance with developing overseas branches Supporting exhibitions
Overseas marketing assistance
Provide export consultancy conferences Assistance with operation of export incubators SMBC
Development of global brands of export companies Facilitation of overseas private networks
KITA
Supporting participation in overseas exhibitions
Korea Chamber of Commerce & industry
Dispatching and receiving economic and trade missions Participation in overseas exhibitions
Korea Federation of Small and Medium Business
Establishment of export consortiums Supporting overseas investments Dispatching overseas market exploration teams
A Study on Utilization of Export Assistance Programs for SMEs
293
Table 1. (continued) SMBA
Developing overseas market developers
KOTRA
Operation of the KOTRA Academy
Human resources/education and training SMBC
SMBA
Invitation of advanced overseas human resources Cultivation developers
of
overseas
market
Assistance with marketing overseas portal sites
through
Online export assistance E-business assistance
KOTRA
Assistance with overseas marketing of electronic commerce
SMBC
Online export assistance programs
KITA
Online promotion of export products
Ministry of Knowledge and Economy
SMEs assistance services International trends and FTA trends Working as an exploration agent
Information assistance
KOTRA
overseas
market
Operation of a library for commercerelated materials Production and distribution of export product catalogs
KITA
Collection of commerce information and provision of materials
Source: summarized by the author Table 2. Differences of Export Assistance Programs provided by Korea and Other Countries Difference Major contents Other countries
Korea
Export assistance by province
- Operation of export assistance programs by area (USA) - Although the size of exhibition support is small, there is systemic management of participation performances based on follow-up evaluation
- Overseas exhibitions and market exploration are assisted by province - Expansion of assistance to other areas is necessary
Assistance by item
- Export support by item through global networks (Hong Kong)
- There are no export assistance programs differentiated by item
Establishment of export product and business DB
- A database of export products and exporters, establishing a cooperative system among trade institutions (Japan's JETRO, TDC in Hong Kong)
- In terms of the internet marketing assistance, Korea has more business databases for overseas promotion
294
W. Eun and E.-Y. Kim Table 2. (continued)
Export subsidies
- Subsidies for participation in overseas exhibitions (Netherlands)
- Direct costs for overseas exhibitions and overseas market development are provided as subsidies
Overall export assistance
- Government-centered export assistance programs (USA) - Integrated services among governments and trade institutions (UK)
- Integration of governmental and trade organizations is yet to be developed. It is not systemic, yet.
Source: summarized by the author The following is a brief summarization of the implications and differences of the export assistance programs for SMEs in Korea and advanced countries. In US, Japan, and UK, assistance is given at an earlier stage of the business according to the advanced industrial structure. Other countries stress the principles of market economy rather than the cultivation of SMEs, establish demand-oriented support programs through interactions among central and local governments and private organizations, and focus on the management globalization of SMEs. The following is a summarization of comparative analyses of export assistance programs by province in Korea and other countries.
3 Literature Reviews Several articles have addressed the relationship between the use of export assistance programs and export performances. Kim Soo-yong 5 showed active governmental support for export assistance programs and cooperation between large companies and SMEs to facilitate export marketing of SMEs. The study suggested overseas market development via co-branding of products of SMEs with large companies brands which have established their own brand images in overseas markets. Lee Gang-bin 6 found that prioritization of export assistance programs would be necessary. Lee stated that while export assistance programs such as overseas exhibitions, overseas market exploration teams, and electronic commerce were supported based on original goals and budgets of various organizations such as SMBA, SMBC, KITA, KOTRA, and local governments as well as the Ministry of Knowledge and Economy, there was no legal ground and support was provided over a short period of time, resulting in problems such as overlapped or excess investments and lack of mid and long-term visions. The study suggested promotion of efficiency through effective networking tools among institutions based on legislation.
5
6
Kim, Soo-yong (2004), “Plans to Support Marketing Facilitation of SMEs”, Economy Research Center, Industrial Bank of Korea. Lee, Gang-bin (1999), “Plans to Expand Local SMEs' Export in the IMF Era”, ⌜Korea Trade Research Association⌟, Vol. 24. Issue 3, Korea Trade Research Association, pp.36-37.
A Study on Utilization of Export Assistance Programs for SMEs
295
Wiedersheim-Paul et al. 7 explained that separate export assistance programs reflecting the differences of factors determining export would be necessary to maximize the export size and number of export-oriented SMEs. They showed factors dealing with international attitudes and abilities such as information on overseas markets rather than financial incentives were effective export assistance programs. Moini8 clarified that the efficiency of various export assistance programs run by central and local governments should have an adequacy of export assistance programs and that the awareness and perceptions of SMEs' CEOs, demanders, were important determinant factors. He proved the effect of export assistance programs to be higher with active export-oriented companies. In addition, he said that there were differences in efficiency, substantial effects of export assistance programs according to globalization stages of SMEs, in other words, the degree of export activities. Wiedersheim-Paul, Olson, Welch9 said that effects of export assistance programs should vary according to the level of human resources placed by export-oriented companies and that the lack of awareness of export assistance programs should reduce the efficiency of such programs. As can be seen from studies of Kjell & Lorentzen10 and Leoidous et al.11, export assistance policies increased their knowledge and understanding of overseas markets, reduce costs for export activities, and improve effiency of export activities while they may not influence internal characteristics of companies. Gencturk & Kotabe12 have analyzed the impacts of export assistance programs on export performances of 8,761 companies. In order to do this, they divided export assistance programs into ‘direct export assistance programs’ for the competitive edges of companies and ‘indirect export assistance programs’ for the profitability of companies, and analyzed which was more useful for export performances. As a result, it was suggested that in terms of the impact of export assistance programs on export performances, direct export assistance programs were more useful than indirect export assistance programs to increase corporate competitiveness. Jeong Jae-seung13 positively analyzed 222 Korean export-oriented SMEs in terms of performances of export assistance programs. It was found that among other SMEs 7
F. Wiedersheim-Paul, Olson H. C, & Welch L. S(1978), "Pre-Export Activity: The Frist Step in Internationalization", Journal of International Business Studies, Vol. 9, No.1, pp.47-58. 8 A. H. Moini (1998), "Small Firms Exporting: How effective Are Government Export Assistance Program?", Journal of Small Business Management, Vol.33. No.3, pp.3-25. 9 F. Wiedeersheim-Paul , Olson H. C, & Welch L. S , op. cit., pp.47-58. 10 G.Kjell and T.Lorentzen(1988), "Exporting the Impact of Government Export Subsidies", European Journal of Marketing, Vol.17, No. 2. 11 L. C. Leoidou, C. S. Katsikeas and N. F. Piercy(1998), "Indentifying Managerial Influences on Exporting: Past Research and Future Direction", Journal of International Marketing, Vol. 6, No. 2. 12 E. F. Gencturk & M. Kotabe(2001), "The Effect of Export Assistance Program Usage on Export Performance: A Contingency Explanation", Journal of International Marketing, Vol. 9. No. 2, pp.51~72. 13 Jeong Jae-seung (2006), “A Positive Study on Performances of Korea's Export Assistance Programs for SMEs”, A Ph.D Thesis, Chungang University.
296
W. Eun and E.-Y. Kim
characteristics, in terms of export commitment, the number of employees in relation to export and export portions as well as the number of countries for export in terms of export experiences significantly influenced awareness, use, and export performances of export assistance programs. Park Cheol and Lee Jae-jeong 14 found that barriers to export marketing most significantly influenced the deterioration of export competitiveness among all other barriers to export of SMEs. Also, export competitiveness had direct negative impacts on export performances. They maintained that governmental facilitation of export assistance programs and companies' efforts would be necessary to solve marketingrelated problems with export. Gabbitas, O. and Gretton, D15, in their study to identify relations between sizes of companies and export performances, measured export performance based on export intensity and sales revenues. Stephen Redding and Anthony J. Venables16, in their study to identify relations between geographical locations and export performances, measured export performances based on export growth rates. As for concrete previous studies on operational characteristics and export performances arising from application of export assistance programs, Woong Eun17 showed that the reasons for decreasing performances of export assistance programs were complex participation procedures (66.6%), questions related to export promotion (33%), and excessive cost arising from participation (23%). Also Park Gwang-seo and Ahn Jong-seok18 found that the level of SMEs' use of export assistance programs was extremely low. They showed that the reasons included complicated procedures of export assistance programs, convoluted applications, and poor effects in reality. Based on the theoretical exploration of the use of export assistance programs and export performances, this study will analyze how the use of export assistance programs provided by various institutions influence export performances, identify assistance programs that are unique to each institution, and study various problems arising from operation of these programs to provide plans to improve export assistance programs for SMEs. Park Cheol ‧ Lee Jae-jeong (1998), “A Study on Relations between Barriers to SMEs' Export and Performances-With an Emphasis on Development and Verification of a Relational Model for Barriers to Export, Export Competitiveness, and Export Performances”,⌜SME Studies⌟, Volume 20. Issue 1, Korea SMEs Academy, pp.113-136. 15 Gabbitas, O and Gretton, D.(2003), "Firm Size and Export Performance : Some Empirical Evidence, " Media and Publications Productivity Commission. 16 Stephen Redding, Anthony J. Venables(2003), "Geography and Export Performance: External Market Access and Internal Supply Capacity", National Bureau of Economic Research. 17 Eun Woong (2009), “Research on expanding export for SMEs through government export subsidy policies”, The Journal of Korea Research Society for Customs, Volume 11 Issue 1, Korea Research Society for Customs, p.193. 18 Park Gwang-seo ‧ Ahn Jong-seok (2001), “A Study on Use of Export Assistance Programs for Export-oriented SMEs”, Korea Trade Research Association, Volume 26, Issue 1, Korea Trade Research Association, pp.137~160. 14
A Study on Utilization of Export Assistance Programs for SMEs
297
4 A Research Model and Hypotheses In this chapter, a research model and hypotheses will be established based on the previous studies. There are many factors for SMEs to make a successful entry into the world market. Among others, this study aims to analyze export performances through export assistance programs of governments and export-related institutions. However, it is necessary to review variables that influence the use of export assistance programs before analyses of export performances through export assistance programs. Hence, this study will analyze impacts of corporate characteristics, barriers to export marketing, operational characteristics of export assistance programs, and awareness of export assistance programs by institution as well as export performances through the use of such programs. The following is a summarization of a model for this study.
Fig. 1. A Research Model
The following hypotheses can be constructed based on the research model. (1) A hypothesis on characteristics of companies and the utilization. This study arranged the following hypotheses to verify relations between characteristics of companies and the utilization of export assistance programs. As companies' characteristics can be measured based on their sizes, export experiences, and portions of export, the size of a company was defined based on the number of employees. The operational definition of export experiences was based on the company's exports to other countries or it's ability for international transactions. Export shares were defined based on the export portion in the sales of an SME. H1: Characteristics of a company will significantly influence the utilization of export assistance programs. H1-1: Export experiences, among other corporate characteristics, will significantly influence the utilization. H1-2: The number of employees, among other corporate characteristics, will significantly influence the utilization. H1-3: Export shares, among other corporate characteristics, will significantly influence the utilization.
298
W. Eun and E.-Y. Kim
(2) A hypothesis on barriers to export marketing and the utilization Based on the previous studies, this study established a hypothesis to verify the impacts of barriers to export marketing on the utilization of export assistance programs. Barriers to export marketing arise in the process of an SME's exporting. Among other various factors, the operational definition consists of the lack of information on overseas markets, the lack of overseas marketing abilities, and the external environmental factors. As for external environment, environmental factors such as increased prices of raw materials from increased exchange rates or prices were provided as variables. H2: Barriers to export marketing will significantly influence the utilization. H2-1: Lack of information on overseas markets, among other barriers to export marketing, will significantly influence the utilization. H2-2: Lack of abilities at overseas marketing, among other barriers to export marketing, will significantly influence the utilization. H2-3: External environment, among other barriers to export marketing, will significantly influence the utilization.
(3) A hypothesis on the operational characteristics and the utilization of export assistance programs This study constructed the following hypothesis to verify impacts of operational characteristics of export assistance programs' utilization. Operational characteristics of export assistance programs are barriers to the Korean SMEs use export assistance programs. They become barriers due to the perception that the procedures of export assistance programs are complicated, takes a long time with poor results. These operational characteristics were defined as variables to find how they affect the utilization. H3: Operational characteristics of export assistance programs will significantly influence the utilization. H3-1: Among other operational characteristics of export assistance programs, complicated procedures will significantly influence the utilization. H3-2: Among other operational characteristics of export assistance programs, that it takes long to get support will significantly influence the utilization. H3-3: Among other operational characteristics of export assistance programs, the perception of poor substantial effects will significantly influence the utilization.
(4) A hypothesis on the awareness and the utilization of export assistance programs by institution This study established the following hypothesis to select the most well-known assistance program among export assistance programs by institution and to verify its impact on the utilization. For awareness of export assistance programs by institution, this research used export assistance program as variables by type to explore how well SMEs were aware of assistance programs of each type by institution and to identify relations between the awareness and the utilization. Among others, operational definition of a variable was given to a specific assistance program with the highest
A Study on Utilization of Export Assistance Programs for SMEs
299
average awareness among institutions. As for the KOTRA, its assistance of overseas branches is to provide support the overseas sales routes from discovery of transaction routes to contracting while the overseas works as an overseas branch of an exporting company. Its assistance for exhibitions is to provide a Korean Pavilion for SMEs in a famous overseas exhibition and to give export-related support and finance. The SMBA's program to designate SMEs with healthy export prospects will select those with growth potentials and support them through networks with export-related institutions covering export finances and overseas marketing. Its program to convert companies from producing goods for domestic use to export-oriented companies is to provide professional services to start-up companies so that they can explore overseas markets. The SMBC has export incubating programs, which establishes export incubators in major overseas trade areas so that SMEs can settle down at an earlier stage. Its overseas networks in the private sector support Korean SMEs wishing to enter into overseas markets with the help of consulting or marketing companies based in other countries. The KITA's support for participation in overseas exhibitions is to help SMEs taking part in overseas exhibitions while its online services support promotion of export products based on registration of electronic catalogs of KITA member companies. H4: Awareness of export assistance programs by institution will significantly influence the utilization. H4-1: Awareness of the KOTRA's support in generating overseas branches and exhibitions will significantly influence the utilization. H4-2: Awareness of the SMBA's designation of SMEs with good export prospects and conversion of domestic goods manufacturers to export-oriented companies will significantly influence the utilization. H4-3: Awareness of the SMBC's support for operation of export incubators and use of private overseas networks will significantly influence the utilization. H4-4: Awareness of the KITA's support for participation in overseas exhibitions and online promotion of export goods will significantly influence the utilization.
(5) A hypothesis on the utilization of export assistance programs and export performances This study arranged the following hypothesis to verify impacts of the utilization of export assistance programs on export performances. The previous studies defined export performances as the sales and export growth rates as for export performances as a dependent variable.
H5: The utilization of export assistance programs by institution will significantly influence export performances. H5-1: The utilization of export assistance programs by institution will significantly influence the sales. H5-2: The utilization of export assistance programs by institution will significantly influence export growth rates.
300
W. Eun and E.-Y. Kim
5 Empirical Result This study used the SPSS 12.0 to positively analyze materials collected in relation to various variables and the following positive analyses took place. Regression analysis was conducted for each variable to confirm relations. (1) Verification of Hypothesis 1 H 1-1: Export experiences, among other corporate characteristics, will significantly influence the utilization.
The results for the regression analysis of export experiences as a corporate characteristic and utilization are provided in Table 3. Table 3. Results for Regression Analysis of Export Experiences and Utilization Parameter
Estimate Standard error
df
t
P - value
Intercept
3.243
0.089
202
36.402
0.000
[Export experiences=1]
0.132
0.201
202
0.656
0.513
[Export experiences=2]
-0.012
0.146
202
-0.081
0.936
[Export experiences=3]
0.091
0.123
202
0.739
0.461
[Export experiences=4]
0.003
0.141
202
0.023
0.981
F
0.282
P - value
0.889
Note: When the dependent variable: When the utilization / export experience is around 5, all the contrast coefficients will be -1.
Based on the results from the F test, significance probability was 0.889, meaning poor explanatory power of the regression equation. Significance probability of each parameter estimate was 0.05 or above, meaning statistical insignificance. However, based on the parameter estimates, when export experiences were low (around 1), the increase means a decreasing level of use. This means that when a company is just starting and has little export experiences, it uses export assistance programs but as it gets more export experience, it uses the programs less. H 1-2: The number of employees, among other corporate characteristics, will significantly influence the utilization.
Table 4 shows the results for the regression analysis of the number of employees as a corporate characteristic and the utilization.
A Study on Utilization of Export Assistance Programs for SMEs
301
Table 4. Results from Regression Analysis of Number of Employees and Utilization Parameter
Estimate Standard error
df
t
P - value
Intercept
3.151
0.075
202
42.028
0.000
[Number of employees=1]
0.072
0.233
202
0.307
0.759
[Number of employees=2]
0.372
0.146
202
2.548
0.012
[Number of employees=3]
0.109
0.129
202
0.844
0.399
[Number of employees=4]
0.215
0.119
202
1.812
0.072
F
P - value
1.925
0.108
Note: When the dependent variable: When the utilization / export experience is around 5, all the contrast coefficients will be -1
Based on the result from the F Test, significance probability was 0.108, meaning that the explanatory power of the regression equation is poor. Except for the level of [Number of employees=2], based on each parameter estimate, the significance probability was 0.05 or above, which is statistically insignificant. However, based on parameter estimates, when the number of employees was around 1 or 2, the utilization increased while the level decreased. This means that when the number of employees is small, a company will use export assistance programs but as the number of employees increases, it is likely that the company will use them less. H 1-3: Export shares, among other corporate characteristics, will significantly influence the utilization.
Table 5 shows the results from the regression analysis of export shares as a corporate characteristic and the utilization. Table 5. Results for Regression Analysis of Export Shares and Utilization Parameter
Estimate
Standard error df
t
P - value F
Intercept
2.339
0.243
202
9.623
0.000
[Export shares=1]
0.827
0.276
202
2.995
0.003
[Export shares=2]
0.922
0.258
202
3.573
0.000
[Export shares=3]
1.092
0.254
202
4.303
0.000
[Export shares=4]
0.902
0.262
202
3.441
0.001
5.057
P - value
0.001
Note: When the dependent variable: When the utilization / export experience is around 5, all the contrast coefficients will be -1
302
W. Eun and E.-Y. Kim
Based on the F Test, the significance probability was 0.001, indicating good explanatory power of the regression equation. Based on each parameter estimate, the significance probability is below 0.05, indicating statistical significance. Based on parameter estimates, when the sales are around 1, 2, or 3, the utilization increase while when the sales are around 4 or 5, the level of use falls. This means that when export shares are low, a company actively uses export assistance programs compared to when export shares exceed a certain point, the company will use less of them. (2) Verification of Hypothesis 2 H 2-1 : Lack of information on overseas markets, among other barriers to export marketing, will significantly influence the utilization.
Table 6 shows the regression analysis results between lack of information on overseas markets and utilization. Table 6. Regression results between Lack of Information on Overseas Markets and the Utilization Unstandardized coefficient Parameter B Intercept
Standard error
1.978
.228
.365
.063
Standardized coefficient beta
t
P F value
P value
8.659 .000 33.667 0.000
Lack of information overseas markets
on
.376
5.802 .000
Dependent variable: Utilization
Based on the F Test, the significance probability was <0.001, indicating good explanatory power of the regression equation. As for the lack of information on overseas markets, the significance probability was <0.0001, which shows statistical significance. This means that when there is less information on overseas markets, a company will try to use export assistance programs more actively. H 2-2: Lack of abilities at overseas marketing, among other barriers to export marketing, will significantly influence the utilization.
Table 7 shows the results for regression analysis of lack of marketing abilities and utilization.
A Study on Utilization of Export Assistance Programs for SMEs
303
Table 7. Results for Regression Analysis of Lack of Overseas Marketing Abilities and Utilization Unstandardized coefficient
Parameter
B Intercept
2.018
Lack of overseas marketing .341 abilities
Standard error
Standardized coefficient t beta
.220
P value
P F value
9.183 .000 34.281 0.000
.058
.379
5.855 .000
Dependent variable: Utilization
Based on the F Test, the significance probability was <0.001, indicating good explanatory power of the regression equation. As for the lack of overseas marketing abilities, the significance probability was <0.001, which is statistically significant. This means that when a company has insufficient overseas marketing abilities, it will try to use export assistance programs more actively. H 2-3: External environment, among other barriers to export marketing, will significantly influence the utilization.
Table 8 shows the results for regression analysis of external environment and utilization. Table 8. Results for Regression Analysis of External Environment and Utilization Unstandardized coefficient Parameter
Standardized coefficient beta t B
Intercept
Standard error
2.388 .263
9.084
P F value
P value
.000 11.826 0.001
External .238 environment
.069
.234
3.439
.001
Dependent variable: Utilization
Based on the F Test, the significance probability was 0.001, indicating good explanatory power of the regression equation. As for the external environment, the significance probability was <0.001, indicating statistical significance. This means that a company will use export assistance programs actively as the external environment becomes poor.
304
W. Eun and E.-Y. Kim
(3) Verification of Hypothesis 3 H 3-1: Among other operational characteristics of export assistance programs, complicated procedures will significantly influence the utilization.
Table 9 shows the results from the regression analysis of complicated procedures and utilization. Table 9. Results for Regression Analysis of Complicated Procedures and Utilization
Parameter
Unstandardized coefficient B
Standard error
Intercept
3.947
.220
Complicated procedures
-.187
.060
Standardized coefficient beta
P F value
t
P value
17.940 .000 9.632 0.002 -.212
-3.104 .002
Dependent variable: Utilization
Based on the F Test, the significance probability was 0.002, indicating good explanatory power of the regression equation. The significance probability of complicated procedures was 0.002, which is significantly significant. This means that the level of use of export assistance programs falls because of the complicated procedures. H 3-2: Among other operational characteristics of export assistance programs that it takes long to get support will significantly influence the utilization.
Table 10 shows the results from regression analysis of a long period of time to get support and the utilization. Table 10. Results for Regression Analysis of a Long Period of Time to Get Support and Utilization Unstandardized coefficient
Parameter
B
Standard error
3.835
.167
A long period of time to get -.167 support
.048
Intercept
Standardized coefficient beta
t
Pvalue
F
Pvalue
22.901 .000 11.964 0.001
Dependent variable: Utilization
-.235
-3.459 .001
A Study on Utilization of Export Assistance Programs for SMEs
305
Based on the F Test, the significance probability was 0.001, indicating good explanatory power of the regression equation. The significance probability for the factor of a long period of time to get support was 0.001, which is statistically significant. This means that as it takes longer to get support of export assistance programs, the utilization falls. H 3-3: Among other operational characteristics of export assistance programs, the perception of poor substantial effects will significantly influence the utilization.
Table 11 shows the results from the regression analysis of the perception of poor substantial effects and utilization. Table 11. Results for Regression Analysis of the Perception of Poor Substantial Effects and Utilization Unstandardized coefficient Parameter
Intercept
B
Standard error
4.281
.215
Standardized coefficient beta
t
Pvalue
F
Pvalue
19.870 .000 22.573 0.000
Poor substantial effects
-.286
.060
-.315
-4.751
.000
Dependent variable: Utilization
Based on the F Test, the significance probability was <0.001, indicating good explanatory power of the regression equation. The significance probability when the substantial effects were low was <0.001, which is statistically significant. This confirms that when substantial effects of export assistance programs are considered to be low, the utilization decreases. (4) Verification of Hypothesis 4
H 4-1: Awareness of the KOTRA's support in generating overseas branches and exhibitions will significantly influence the utilization.
Table 12 shows the results from regression analysis of the KOTRA's support and utilization.
306
W. Eun and E.-Y. Kim Table 12. Results for Regression Analysis of the KOTRA's Support and Utilization Unstandardized coefficient
Parameter
Standardized coefficient beta
Pvalue
t
Pvalue
8.173
.000 .000 102.70 0.000
B
Standard error
Intercept
1.229
.150
Support in establishing overseas branches
.210
.041
.278
5.158
Exhibitions
.418
.041
.550
10.210 .000
F
Dependent variable: Utilization
Based on the F Test, the significance probability was <0.001, indicating good explanatory power of the regression equation. The significant probabilities for establishment of overseas branches and exhibitions were <0.001, which is significant statistically. This means that a higher awareness of the KOTRA's support in establishing overseas branches and exhibitions are related to increased utilizations. H 4-2: Awareness of the SMBA's designation of SMEs with good export prospects and conversion of domestic goods manufacturers to export-oriented companies will significantly influence the utilization.
Table 13 shows the results from regression analysis of the SMBA's support and utilization. Table 13. Results for Regression Analysis of the SMBA's Support and Utilization
Parameter
Unstandardized coefficient Standard B error
Standardized coefficient beta
t
Pvalue
Intercept
1.105
.165
Designation of SMEs with good export prospects
.397
.044
.503
9.123 .000
Helping companies producing domestic goods to become .266 export-oriented
.046
.317
5.758 .000
Dependent variable:Utilization
F
Pvalue
6.699 .000
95.049 0.000
A Study on Utilization of Export Assistance Programs for SMEs
307
Based on the F Test, the significance probability was <0.001, indicating good explanatory power of the regression equation. The significant probabilities for designation of SMEs with good export prospects and conversion of domestic goods manufacturers into export-oriented companies were <0.001, which is statistically significant. This confirms that higher awareness of the SMBA's designation of SMEs with excellent export prospects and conversion of domestic manufacturers into exportoriented companies relates to increased utilizations. H 4-3: Awareness of the SMBC's support for operation of export incubators and use of private overseas networks will significantly influence the utilization. Table 14 shows the results from regression analysis of the SMBC' support programs and utilization. Table 14. Results for Regression Analysis of the SMBC' Assistance Programs and Utilization
Parameter
Unstandardized coefficient
Standardized coefficient beta
t
Pvalue
F
Pvalue
B
Standard error
1.299
151
Support for operation of export .280 incubators
.045
.365
6.190 .000 91.220 0.000
.045
.426
7.230 .000
Intercept
Private networks
.326
8.620 .000
Dependent variable: Utilization
Based on the F Test, the significance probability was <0.001, indicating good explanatory power of the regression equation. The significance probabilities for support in operation of export incubators and use of networks were <0.001, which is statistically significant. This means that a higher awareness of SMBC' support for operation of export incubators and private networks increases the utilization. H 4-4: Awareness of the KITA's support for participation in overseas exhibitions and online promotion of export goods will significantly influence the utilization.
Table 15 shows the results from regression analysis of the KITA's support programs and utilization.
308
W. Eun and E.-Y. Kim
Table 15. Results for Regression Analysis of the KITA's Assistance Programs and Utilization
Parameter
Unstandardized coefficient
Standardized coefficient beta
t
Pvalue
F
Pvalue
B
Standard error
Intercept
1.080
.146
Support for participation in overseas exhibitions
.260
.040
.351
6.430 .000 119.31 0.000
Online promotion of export goods
.391
.043
.496
9.084 .000
7.381 .000
Dependent variable: Utilization
Based on the F Test, the significance probability was <0.001, indicating good explanatory power of the regression equation. The significant probabilities for assistance with participation in overseas exhibitions and online promotion of export goods were <0.001, which is statistically significant. This confirms that higher awareness of the KITA's support for participation in overseas exhibitions and online promotion of export goods relates to increased utilizations. (5) Verification of Hypothesis 5 H 5-1: The utilization of export assistance programs by institution will significantly influence the sales.
Table 16. Results for Regression Analysis of Utilization of Export Assistance Programs by Institution and Sales
Parameter
Unstandardized coefficient
Standardized coefficient beta
t
Pvalue
F
Pvalue
B
Standard error
Intercept
-10.533
6.370
The level of use of the SMBC' assistance programs
1.516
2.666
.058
.569
The level of use of the SMBA's assistance programs
3.713
2.987
.137
1.243 .215 11.138 0.000
The level of use of the KOTRA's 5.399 assistance programs
2.658
.213
2.032 .044
2.503
.066
.680
The level of use of the KITA's assistance programs Dependent variable: Utilization
1.703
.100 1.654 .570
.497
A Study on Utilization of Export Assistance Programs for SMEs
309
Table 16 shows the results from regression analysis of the utilization of export assistance programs by institution and sales. Based on the F Test, the significance probability was <0.001, indicating good explanatory power of the regression equation. As for significance probabilities of independent variables, only the significance probability of the level of use of the KOTRA's assistance programs was below 0.05, which is statistically significant. The KOTRA's standardized coefficient beta is the highest, among others, which tells that it has the greatest impact on export performances. This means that an increase of the level of use of the KOTRA's assistance programs relates to increased sales. H 5-2: The utilization of export assistance programs by institution will significantly influence export growth rates.
Table 17 shows the results from regression analysis between utilization of export assistance programs by institution and export growth rates. Table 17. Results for Regression Analysis between Utilization of Export Assistance Programs by Institution and Export Growth Rates
Parameter
Intercept
Unstandardized coefficient Standard B error
Standardized coefficient beta
t
Pvalue
F
Pvalue
1.053
0.211
4.989 0.000
The utilization of the KOTRA's 0.277 assistance programs
0.088
0.279
3.141 0.002
The utilization of the SMBA's assistance programs
0.090
0.099
0.087
0.912 0.363 30.896 0.000
The utilization of the SMBC' assistance programs
0.330
0.088
0.342
3.745 0.000
The utilization of the KITA's assistance programs
-0.034
0.083
-0.035
0.680 0.413
Dependent variable: Utilization
Based on the F Test, the significance probability was <0.001, indicating good explanatory power of the regression equation. As for significance probabilities of independent variables, only the significant variables of the utilization of assistant programs provided by the SMBC and the KOTRA were below 0.05, which is statistically significant. Standardized coefficient beta values of the SMBC and the KOTRA were high, meaning that they have the greatest impact on export growth rates. This means that an increase of the utilization of assistant programs provided by the SMBC and the KOTRA relates to increased export growth rates.
310
W. Eun and E.-Y. Kim
6 Conclusion Korea has set export as the axis of economic growth since the 1960's and implemented various export assistance programs to promote the export of companies. These export assistance programs greatly contributed to the rapid export increase in Korea. Based on the country's economic structure, large companies have led the economic growth and benefits, and export assistance have been enjoyed by these companies. However, since the SMEs' roles for the national economy began to be emphasized since the late 1980's, the support for SMEs was provided in various ways. Hence, various export support programs are provided to SMEs by central and local governments as well as trade-related organizations. As mentioned, institutions providing export assistance programs to promote export of SMEs include the SMBC, the KITA, the KOTRA, and to the SMBA. Results from previous studies show that the utilization of export assistance programs significantly influence export performances. However, there are several reasons for poor use of various assistance programs. First, awareness of export assistance programs by institutions tend to be low. Based on research on the actual conditions, there are results of high awareness of export assistance programs by institution but awareness of recent export assistance programs tend to be low. Also, there are many similar assistance programs among institutions, making it difficult for the institutions to accurately identify them all. Thus, promotion strategies to actively inform export assistance programs are necessary. Second, problems arising from operation of assistance programs should be solved to increase the utilization of export assistance programs. A well-organized integrated system is also necessary among governments and trade-related associations. The assistance procedures also need to be simplified, these programs shall be supplemented to promote substantial export. Each institution should implement more effective programs actively while removing effective ones. There are many overlapping programs and institutions should reorganize them according to their own characteristics and purposes. If these institutions become specialized, their assistance programs will become more effective. For example, it is desirable for the KOTRA to take responsibility for overseas market exploration as the organization is based in various countries. The KITA might focus on education, while the SMBA could excavate companies with good export prospects, and the SMBC could take responsibility for financing along with financial institutions. To achieve this, each institution needs to evaluate themselves where they can professionally support, so that an effective assistance system can be established. Third, as SMEs taking part in overseas marketing need to be competitive in terms of export to achieve substantial performances, an institutional tool should be arranged to increase competitiveness when selecting companies to take part in the export assistance programs. Professional associations by item should be induced for participation with the KOTRA, and there must be an cooperative body of trade associations to excavate excellent companies that can actually export in an overseas market. Fourth, plans to manage SMEs taking part in overseas marketing assistance programs should be arranged. They should be systemically managed so that they can
A Study on Utilization of Export Assistance Programs for SMEs
311
maintain transactions with foreign buyers based on a follow-up evaluation of their performances after participation. This study aimed to present ways to improve export assistance programs for the consistent development of Korean export-oriented SMEs. Although this study maintains that the most effective assistance programs should be expanded and less effective ones removed based on the needs of various export assistance programs by institution, this is an institutional issue and will take time to materialize. While this study conducted a nationwide questionnaire survey, the sample size is not enough to represent the whole country. Hence, this presents limitations to generalize the results from this study and the application of more elaborate statistical analysis techniques were difficult. So in the future, positive analyses based on larger sample sizes of SMEs in the country should take place. Also, while this study analyzed the variables influencing the utilization of export assistance programs with an emphasis on characteristics of companies, barriers to export marketing, awareness of export assistance institutions, and operational characteristics of export assistance programs, there are many more variables that influence the utilization of export assistance programs including resources, capacity, and characteristics of CEOs. If a more quantitative analysis of impacts of the utilization of export assistance programs on export performances take place in the future based on more variables, there will be more detailed positive analyses.
References 1. Moini, H.: Small Firms Exporting: How effective Are Government Export Assistance Program? Journal of Small Business Management 33(3), 3–25 (1998) 2. European Commission enterprise and industry directorate-general, Final Report of the expert group, Supporting the Internationalization of SMEs, p. 5 (2007) 3. Gencturk, E.F., Kotabe, M.: The Effect of Export Assistance Program Usage on Export Performance: A Contingency Explanation. Journal of International Marketing 9(2), 51–72 (2001) 4. Woong, E.: Research on expanding export for SMEs through government export subsidy policies. The Journal of Korea Research Society for Customs 11(1), 193 (2009) 5. Wiedersheim-Paul, F., Olson, H.C., Welch, L.S.: Pre-Export Activity: The Frist Step in Internationalization. Journal of International Business Studies 9(1), 47–58 (1978) 6. Kjell, G., Lorentzen, T.: Exporting the Impact of Government Export Subsidies. European Journal of Marketing 17(2) (1988) 7. Gabbitas, O., Gretton, D.: Firm Size and Export Performance: Some Empirical Evidence. Media and Publications Productivity Commission. SMEs Academy, 113–136 (2003) 8. Jae-seung, J.: A Positive Study on Performances of Korea’s Export Assistance Programs for SMEs, Ph.D Thesis, Chungang University (2006) 9. Jie, Y.-h.: Plans to Improve International Competitiveness of SMEs. Seogang Harvard Business, 86–92 (1994) 10. Kim, S.-y.: Plans to Support Marketing Facilitation of SMEs, Economy Research Center, Industrial Bank of Korea (2004) 11. Lee, G.-b.: Plans to Expand Local SMEs’ Export in the IMF Era. Korea Trade Research Association 24(3), 36–37 (1999)
312
W. Eun and E.-Y. Kim
12. Leoidou, L.C., Katsikeas, C.S., Piercy, N.F.: Indentifying Managerial Influences on Exporting: Past Research and Future Direction. Journal of International Marketing 6(2) (1998) 13. Cheol, P., Jae-jeong, L.: A Study on Relations between Barriers to SMEs’ Export and Performances-With an Emphasis on Development and Verification of a Relational Model for Barriers to Export, Export Competitiveness, and Export Performances. SME Studies 20(1) (1998) 14. Gwang-seo, P., Jong-seok, A.: A Study on Use of Export Assistance Programs for Exportoriented SMEs. Korea Trade Research Association 26(1), 137–160 (2001) 15. Redding, S., Venables, A.J.: Geography and Export Performance: External Market Access and Internal Supply Capacity. National Bureau of Economic Research (2003) 16. Yeoh, P.-L.: Entrepreneurship and Performance: A Proposed Conceptual Model. In: Tamer Cavusgil, S., Axinn, C.N. (eds.) Advances in International Marketing, vol. 6, pp. 43–68 (1994)
The Entrance Authentication and Tracking Systems Using Object Extraction and the RFID Tag Dae-Gi Min, Jae-Woo Kim, and Moon-Seog Jun Department of Computer Science, Soongsil University, Sangdo-Dong, Dongjak-Gu, Seoul 156-743, Korea {daegi min,saypeace,mjung}@ssu.ac.kr
Abstract. In this paper, the proposal system can achieve the more safety of RFID System with the 2-step authentication procedures for the enhancement about the security of general RFID systems. After authentication RFID Tag, additionally, the proposal system extract the characteristic information in the user image for acquisition of the additional authentication information of the user with the camera. In this paper, the system which was proposed more enforce the security of the automatic entrance and exit authentication system with the cognitive characters of RFID Tag and the extracted characteristic information of the user image through the camera. The RFID system which use the active tag and reader with 2.4GHz bandwidth can recognize the tag of RFID in the various output manner. Additionally, when the RFID system have errors, the characteristic information of the user image is designed to replace the RFID system as it compare with the similarity of the color, outline and input image information which was recorded to the database previously. In the result of experiment, the system can acquire more exact results as compared with the single authentication system when it using RFID Tag and the information of color characteristics. Keywords: RFID, 2-Step Authentication, Image Processing.
1
Introduction
Today’s rapid development of computer network technique and information communication technique has been dramatically progressed. The technology that can automatically identify a specific thing and organism or the digital contents are supplied familiarly to us. Numerous service has been offered from networks as well. As a result, manifold informations has been produced. To control this informations, maintaining database is the prerequisite. Until now many database security model that complement its security problem has been researched. The security model which is researched during that time can protect just from the inner part of the system. Of course the security technique has become diffusion by adding DRM technique or the electronic library, field T.-h. Kim et al. (Eds.): UCMA 2011, Part I, CCIS 150, pp. 313–326, 2011. c Springer-Verlag Berlin Heidelberg 2011
314
D.-G.Min, J.-W. Kim, and M.-S. Jun
and grafting of fingerprint recognition etc. But it is not able to detect a direction and illegal use of documented data. In order to misuse and outflow of the data, UBIQUITOUS RFID(Radio Frequency Identification) SYSTEM which can exchange information by using radiowave remote control and recognize is possible to provide application service in broad industry and personal life. Progress of information network and electronic communication technique has been advanced. It is able to identify specific thing and organism as an automatic movement. Especially RIFD CARD SYSTEM has been expected being diffused in entrance and exit control, fee levy, the electronic money and medical insurance field. To solve these problem against documented data and an entrance and exit control system we suggest a systems using realtime object extraction and the RFID Tag. The Entrance Authentication Systems that we suggested abstracts information of RFID Tag and face feature information of user. Like this process, we suggest the system that can track the object in the camera range. As a face detection method we will test a color base face area detection method, an edge base eye area detection and a 3 steps face area normalizer. The composition of the paper is like this. Section 2 will detail the related research, Section 3 will discuss about system that we suggest, Section 4 will describe a efficiency evaluation of proposal system. Finally Section 5 contains a conclusion and future work.
2 2.1
Related Work RFID Systems
The RFID is a automatic recognition technologies by reader. It uses a radio frequency and automatically recognize a saved data from a tag that has builtin micro chip (Tag), label and card etc. The RFID the system is a important technologies of next generation. It made up for the weak point of the existing bar code or self magnetic equipment. Convenience for user has been improved. And capacity also expanded so it can hold more information [3]. Right now the RFID system is using in variety field such as national defense, medical treatment, circulation, traffic, security, manufacture, construction, service and administration field. This system is composed of Tag (or transponder), Reader (or interrogator), Back-end-Server [9]. Tag. Tag is a device that attaches on person, thing and the animal gadfly and transmits it’s direct or indirect identification and recognition. Generally, tag is composed of a IC chip and an antenna. Each tag has it’s own specific information Id(TID). And it could be classified with Active Tag and Passive Tag [2]. Active tag has it’s own battery so it is possible to perform self-operation. And transmitting range of data reach until hundreds of thousands of meter. But the price of tag is expensive because of built-in battery. Badly tag’s life depends
The Entrance Authentication and Tracking Systems
315
Fig. 1. RFID System
on the battery’s life as well. Active tag is often applied to trade container and location recognition, Healthcare field. Passive tag does not have it’s own battery. To get power, it use a electric wave sent from a reader and induced the electromagnetic. Passive tag’s transmission power is weak so it’s transmission range is shorter than Active tag. Usually, it is mainly using for a short-range network. Passive tag’s cost is reasonable and it is possible to reduce it’s size as smaller. It has long life span so it can apply to diversity field. We expected that passive tag will mainly used in the administration of physical distribution. Reader. RFID Reader is a transmitting device in order to collect information of the tag. And it also forward the collected information from the tag to middleware [11]. RFID reader includes a RF analog department and a digital signal control department. RF analog department transmits electricity and deals with analog signal. And digital signal control department composed of decoder and encoder etc. The RFID the leader can classified into two basic types: portable and fixed. Fixed RFID reader receive a tag information by wireless. And then, it delivers received information to server. Portable RFID reader has built-in wireless network interface for transmitting tag information that is received in the leader to server Reader’s main role is transmitting tag information to subsystem and backend server so it can analyze the data [10]. 2.2
Face Area Detection and Recognition
Face area detection is searching face area which has specific feature that include eyes, nose and mouth. face recognition is a finding same person’s image from the database through input image [6]. To face detection and recognition, face outline, skin color, head shape, eyes, nose and mouth is important feature. and each person has their own style. so
316
D.-G.Min, J.-W. Kim, and M.-S. Jun
Fig. 2. shows 2.45GHz RF Active Tag. Existing RFID Tag use only 1 output (dbm), but the system we want to develop’s tag is able to transmit electric wave that’s identification distance is variable.
for the feature detection must be sampled with great exactitude. If face image damaged, we need to restore that to the original. So that improve the rate of face area recognition. Face restoration is a progress to restore damaged face image to original image [4]. face detections. First process for a face area detection is separate face existence area and background area from input image. As a face detection method, we will use Fleck’s skin filter and detected face area from input image. Firstly, applying a Theorem (1) will convert the value of R, G and B to I, Rg, By as a value of log-opponent color representation. In this Theorem (1), the value of n is a random noise that existence between [1], [5](L(x)=1052 log10(x+1+n)). I = L(G), Rg = L(R) = L(C), By = L(B) −
L(R) + L(G) 2
(1)
By Theorem (2), We can get a value of hue as H and value of saturation as S. R = tan− 1(Rg/By), s = Rg 3 + By (2) When Theorem (2)’s operation is ends, we can detect only face area from the input image. The value of Human skin’s Hue and chroma is a) hue = [ 110, 150 ], chroma = [ 20, 60 ] or b) hue = [ 130, 170 ], chroma = [ 30,130 ]. If we mark every pixel that satisfied those condition, we can get only face area [7].
The Entrance Authentication and Tracking Systems
317
Pre-operation step. In order to find out damaged part of face area, We have to get a value of center line and point of face. At this moment, we have to operating Sobel edge operation, Binary-coded decimal, and morphology operation as a preoperation step. A. Face outline detection The way to detects face outline is changing value of light against differential operator and using mask. Sobel mask method is effective because speed of calculation is fast. Where size of input image I is M*N, this input image I need to separate like a Fij from Theorem (3) in order to apply the Sobel mask. Assuming Fij ’s separated image and sobel mask S1, S2 is equal to (Theorem 3), (Theorem 4), edge of face area could be defined as (Theorem 5) [8]. Pi−1j−1 Pi−1j Pi−1j+1 Fi j = Pij−1 Pij Pij+1 i+1j−1 Pi+1j Pi+1j
B. Binary-coded decimal and morphology operation Binary-coded decimal is a image operation method to simplify whole information of image. Normally it use as a Pre-operation step for main image control. The value of light intensity is putted as 0 and 1 based on critical value. After Binary-coded operation, dilation operation is needed to erase noises on face area. Dilation operation expanded a outer line pixel of face area. so empty space of face can be filled and broken line can be connected. Conversion of Logarithmic - Polar Coordinates. Converting absolute coordinate to Logarithmic - Polar Coordinates. There are some important points about this conversion method. Firstly, based on the center point, important information is densely distributed over a center area. Secondly, a axis of Z is radius of a circle and a axis of H is rotative angle in this Logarithmic - Polar Coordinates [12]. ’s a, b, c, d in absolute coordinate area can be converted as following in Logarithmic - Polar Coordinates. left and right area of face image is divided into section b, section c, section a and section d.
318
D.-G.Min, J.-W. Kim, and M.-S. Jun
Fig. 3. absolute coordinate and log - polar coordinate
Extracting of a automatic interrelation feature. Automatic interrelation p(s) can be found by using (Theorem 6) and (Theorem 7) per each row of log - polar coordinate. Assuming X(t) is a one row who has ingredients as much as N, solve a (Theorem 6) and find out a value of R(s). When s=o, R(s) can have biggest value. P(s) is defined by (Theorem 7). R(s) =
N −1 1 (x(t) − X)(x(x + t) − x) N i−1
− 1 ≤ p(s) −
R(s) ≤1 R(0)
(6)
(7)
Face recognitions. The specific face is a face that is expressed through specific vectors that comes from Principal ingredient analysis technique. The principal ingredient analysis technique need reorganizing high-dimensional data to lowergrade one. and Training-set is necessary. Assuming is a random face image from training group, M*N is a size of image and number of image is n=1.....M, (Theorem 8) the value of average face(ψ) of training group can be calculated. ψ=
M 1 Γ (n − 1...........M ) M
(8)
M 1 φk φk M
(9)
wk = uΓk (T − ψ)
(10)
Ω Γ = [w1 w2 ........wk]
(11)
k−1
C=
k−1
The Entrance Authentication and Tracking Systems
319
Calculate a vector from original image in database minus a value of average face. After that, calculate a covariance matrix through (Theorem 9). Covariance matrix become a size of N2*N2, and this will be used to calculate a eigenvalues uk and a distribution of rank high data which has well expressed data range as much as M. The specific value is arranged by biggest value. If we rearrange those specific vector as N*N, it becomes face image like real one so-called Specific face. By (Theorem 10), a specific vector is used for weight wk that is k=1.....M. Each weight from each specific vector can be expressed as vector like (Theorem 11). When image inputted, we can get a value of input image’s Ω. It recognize a closest face image with input image that has smallest Euclidian distance Ek = ||(Ω − Ωk )||). But, if Ek is bigger than critical value, it is recognized as a new face recognition.
3
Proposal System
The proposal system in this paper is automatic approval system by two step confirmational operation. This system is automatically approve a permitted user. For the first step RFID reader read a data of RIFD tag. After first approval progress Tag users can be permitted. And for the second step user’s approval, tag owned-user’s image feature information is putted on the comparison list. At the second step, camera extracts user’s face image to get approval information. Extracted face images which is composition elements of the face has information of face outline, hue and shape. It goes through character based progress. Then admits Tag user. At the same time, user’s face feature information is compare it’s previous information that saved in database. And it also check the user’s admittance information. (Figure 4) shows our suggesting system.
Fig. 4. Structure of entire system
3.1
RFID System
The RFID the System is composed of Tag, Reader and Backend server. This tag is a 2.4GHz band active tag. Various RF output is available as well. So it is possible to adjust a communication distance between tag and reader through Air Interface.
320
D.-G.Min, J.-W. Kim, and M.-S. Jun
The Reader receives the data of the Tag and it separates additional information and Tag ID. After that, reader transmits it to server and approve users by using registration information data and input image feature information from camera. 3.2
System Management Server
The system management Server send an image information that comes from Tag and Camera to backend DB. These collected information is checking the similarity of image feature information and saved Tag ID in backend DB. If it is not match together, each tag’s input information, image data, image feature information and time stamp supposed to save immediately. And this system will send a alert E-mail or SMS message about outsider’s approach and security breakdown. 3.3
Face Recognition Methods
Face recognition is familiar to people and non-compulsive. Right now, some studies about face confirmation software for general low-cost camera or poor background and illumination is going on. There are 2 way of face recognition method : Holistic approach method and Analytic approach method. Holistic approach method considers the entire area’s quality of pattern. Eigenface, Template matching-based method are Holistic approach method. Analytic approach method is based on geometric feature extraction. Recognition speed of this method is fast and required only small memory. But selecting and extracting a feature point is difficult. So we suggested a Mesh type face area searching method that is most effective for real-time input image. From extracted face area, we will get the shape information due to feature point of face contour, head shape, eyes, nose and mouth. Comparing this information with data in database and RFID authentication will increase the security level of user confirmation.
Fig. 5. Mesh type face area search
The Entrance Authentication and Tracking Systems
321
Mesh type Face Area Searching. Through pixel testing, real-time image is available to extract a contour point. If we search this whole image per pixel, it required over-operation. So we will not count on whole pixel of image and set up the Pixel Space like (Figure 5). First test and second test will proceeding after this. Select the object’s square blocks by using contour point that extracted from Pixel Space. After First test, save coordinates points of x1,x2..,xn as extracted contour point. Next, measure the distance of the coordinate values which are located in same scanned line. And calculate the maximum width value of the square block of the object. When second test is done, calculate coordinates points of y1,y2..yn as extracted contour point. Calculate maximum height value of the square block (Theorem 12).
F aceRect
F aceXL arg e = M ax(x1 , x2 , ...xn )F aceXSmall = M in(xx , x2 , ...xn ) F aceYL arg e = M ax(y1 , y2 , ...yn )F aceYSmall (12) = M in(y1 , y2 , ...yk ) = RECT (F aceXSmall , F aceYSmall , F aceXL arg e , F aceYL arg e )
For the area division of the each square block, equally divide into 3 pieces. then save the value of divided area. Extracting of Hue feature information. RGB hue in extracted face area which extracted from Mesh type Face Area search method shows sensitive reaction about change of color tone. So each pixel’s color has to convert to HSI model consists of hue, chroma and luminosity. Save those converted HSI ingredients in database as a value of hue per each area. Following (Theorem 13) and (Theorem 14) shows the conversion process of RGB model to HSI model. I = 0.3R + 0.59G + 0.11B V1 = R − I = 0.7R − 0.59G − 0.11B V2 = B − I = −0.3R − 0.59G + 0.89B −1
H = tan
V1 V2
,S =
2 V+V 21 2
(13)
(14)
According to (Theorem 13), (Theorem 14), get values of hue, chroma and luminosity per each channel. Value of H is 0 and 180. If B>G, H = 360 - H. Value of S is between 0 and 1. Then, H and S have to normalize until 0 and 255. With this value of HSI, calculate each channel’s average and standard deviation. (Theorem 15), (Theorem 16), (Theorem 17) is followed example. Ei =
Si =
c 1 Pij M · N j=1
C 1 2 (Pij − Ej ) M · N j=1
(15)
(16)
322
D.-G.Min, J.-W. Kim, and M.-S. Jun
3 σi =
C 1 3 (Pij − Ei ) M ·N
(17)
j=1
Pij is a value of ”i”th pixel in input image MN’s ”j”th channel. Ei is the average against each channel. Si is a standard deviation. σ is a skewness. C represents each channel. Finally, hue feature information is expressed as a Sk (Ej ,Sj , σ). Sk is a whole image or section K’s color feature information. Extracting face feature area. Through these described process, background and object was separated while inputting. Face area and body area was separate from separated-object by hue feature information and extracted contour point. Mesh type search method can extract face outline, head shapes, contour of eyebrows, eyes, nose and mouth. Due to each point of outline has a coordinate point, acquiring shape information is possible.
Fig. 6. Specific area extraction of face
As we see (Figure 6), Extracting face feature information through Step 1 and step 6. And save these information to database. Here are our 6 steps. 1. Extract the entire outline of face and obtain the feature information from Mesh search method. 2. Use extracted face shape information and hue feature information. Obtain contour of head and shape information. (Assuming shape of head, eyebrows, eyes, nose and mouth is located in certain rates) 3. Acquisition of eyebrow’s outline and shape information. 4. Acquisition of eye’s outline and shape information acquisition. 5. Acquisition of nose’s outline and shape information acquisition. 6. Acquisition of mouth’s outline and shape information acquisition. Compare similarity of shape information. Extracted feature area and shape information that comes from face feature extraction method is comparing it’s similarity with saved information in database. Shape information is a key value to calculate it’s moment. This method can use as a measure of the similarity. In other words, the moment value like (Theorem 18) is nonchangeable even
The Entrance Authentication and Tracking Systems
323
though it moves or rotates. Measuring a polyhedron and non-convex surface’s similarity is possible. mM(p+q) =
i
ip j q F image(i, j)
(18)
j
But, calculating moment needs extra calculation formula. So it is not suitable for real-time confirmation system. Our suggested system is effective way to compare face feature area. Improved moment formula such as (Theorem 19) is good to use. mM (p+q) = (i − j)p (i − j)q (19) (i,j)∈C
Extracted face feature area is not always perfectly match with one saved in database. So it is quiet difficult to get proper results of similarity. Therefore, set limits and critical value of moment. And calculates it’s similarity like (Theorem 20). SShape (Qquery , Idbimage ) = n wShape i=1 (Qquery (mm(p+q) )i − Idbimage (mm(p+q) )i)
(20)
In this (Theorem 20), is a extracted shape information and similarity of database’s shape information. is a weight of new target shape information’s value of moments and database’s value of moments.
4 4.1
Results Test Environments
For the best results, we choose about 100 tag datas and some place where is difficult to changed it’s pixel. We also use a web camera for the test and background image and input image was controling in real-time. This system was operated by Intel Pentium 4 PC (CPU 2.0GHz, 1G RAM Visual C++ 6.0 (Service Pack 6)). We also considered about transmission of input image, so we used 320X240 size of RGB 24bit color image. 4.2
Test Results
Distance efficiency test of Tag. As a Active tag’s recognition efficiency test, we saved 100 tags to database. And we also put a reader in front of exit. We did various street test with 4 different wave bands such as -15db, -10db, -5db and 0db. We measured tags that recognized in each street. In view of the results, distance of Active Tag is vary with Frequency intensity. So we can use this system as variety purpose. Proposal system used -15db.
324
D.-G.Min, J.-W. Kim, and M.-S. Jun Table 1. Measuring distance by frequency intensity of the Tag Tag
1
Distance (m) Frequency intensity of the Tag2 -15db -10db -5db -0db 1 O O O O 2 O O O O 3 O O O O 4 O O O O 5 X O O O 10 X O O O 15 X X O O 20 X X X O 25 X X X O Table 2. Tags recognition efficiency test Tag 1 2 3 4 5 6 7 8 9 10
Recognition efficiency test of Tag. We decided distance of the Tag from those test. In this time, we did recognition efficiency test of Tag within the limit of recognition distance. In this recognition efficiency test of tag, first 10 and 40 tag’s recognition efficiency was great. But when it comes over 50, frecuency of error were increased. The reason was lose phenomenon. Which means, too many tag data was sented from reader and it need time to transmit and compare. Therefore, we limits a mount of simultaneous recognition within maximum 50. face recognition efficiency tests. For the face recognition which is a second confirmation step, we used a camera that is located in the entrance. We got a input image of entrance user’s upper body. And then, we tested a rate of recognition. Saved feature information was compared to input image. 1 2 3 4
Amount of tested Tags are 100. There was not any covers in the test area. Total 100 Tag was used. 10 at the first time and 100 for the last. The frequency of recognition efficiency test is average value of total 10 tests.
The Entrance Authentication and Tracking Systems
325
Table 3. Face Recognition Efficiency Test Feature Users U1 6 U2 U3 U4 U5 UX6 7 UX7 UX8 UX9 UX10
Similarity of face feature information (%) A B C D E 98 95 96 100 97 99 96 95 95 96 97 95 96 99 97 98 95 95 94 99 99 99 99 97 98 75 79 65 88 51 79 88 87 84 85 84 81 75 76 88 75 71 69 88 69 71 82 84 75 81
5
Confirm O O O O O X X X X X
Fig. 7. Face recognition efficiency graph
(Table 3) and (Figure 7) is a face recognition ability test. From U1 to U5 is the users who is register1)ed in database. We compared their input image which created from the camera. From UX6 to UX10 is the users who is not registered in database. Only if this shape information’s similarity is over 95%, system will recognize this user as a same person. We are 95% accuracy of this test except some special situation such as environmental problem.
5
Results and Future Work
We suggested 2 steps entrance recognition system in order to strengthen previous system that only use a RFIDTag. proposal system is based on RFID technique and image control. Existing entrance system is using only RFID Tag. If stranger 5 6 7
U1 and U5 : Users who is on the database UX6 and UX10 : Users who is not on the database A: face outline feature, B: head shape feature, C: eyebrows shape feature, D: eyes shape feature, E: mouth shape feature
326
D.-G.Min, J.-W. Kim, and M.-S. Jun
found this he can actually enter. But the case of proposal system, for the first step, RFID system extracts user’s informaton through Activc tag. Next, the camera located at enterance extracts user’s upper body input image and compare with the information in database. Thus no matter you have a identification card or not, if first confirm information and second confirm information are different, entering is restricted. In consequence, we realized that using RFID System is efficient for the entrance recognition system. Advanced study for real-time object tracking will be need in the future. We need complete confirmation by complete face recognition .
References 1. Yang, K.-M., Sun, M.-T., Wu, L.: A family of vlsi designs for the motion compensation block-matching algorithm. IEEE Transactions on Circuits and Systems 36, 1317–1325 (1989) 2. Wang, J., Wang, D., Zhao, Y.: A novel anti-collision algorithm with dynamic tag number estimation for rfid systems. In: International Conference on Communication Technology, ICCT 2006, pp. 1–4 (November 2006) 3. Joo-Sang, P., Young-Il, K., Yong-Joon, L.: Security considerations for rfid technology adoption. In: The 7th International Conference on Advanced Communication Technology, ICACT 2005, vol. 2, pp. 797–803 (2005) 4. Hu, W., Farooq, O., Datta, S.: Wavelet based sub-space features for face recognition. In: Congress on Image and Signal Processing, CISP 2008, vol. 3, pp. 426–430 (May 2008) 5. Wan, S.-Y., Higgins, W.: Symmetric region growing. In: Proceedings of 2000 International Conference on Image Processing, 2000, vol. 2, pp. 96–99 (September 2000) 6. Romer, K., Schoch, T., Mattern, F., Dubendorfer, T.: Smart identification frameworks for ubiquitous computing applications. In: Proceedings of the First IEEE International Conference on Pervasive Computing and Communication (PerCom 2003), pp. 253–262 (March 2003) 7. Ikonomatakis, N., Plataniotis, K., Zervakis, M., Venetsanopoulos, A.: Region growing and region merging image segmentation. In: Proceedings of 13th International Conference on Digital Signal Processing, DSP 1997, vol. 1, pp. 299–302 (July 1997) 8. Yang, J., Waibel, A.: A real-time face tracker. In: Proceedings 3rd IEEE Workshop on Applications of Computer Vision, WACV 1996, pp. 142 –147 (December 1996) 9. Li, S., Tao Weng, X.: Electromagnetic environment of uhf-rfid. In: 10th Electronics Packaging Technology Conference, EPTC 2008, pp. 1034–1039 (December 2008) 10. Tsang, P., Yung, K., So, K., Cheung, K.: Integration of video surveillance and rfid for remote scene monitoring. In: Asia-Pacific Microwave Conference, APMC 2008, pp. 1–4 (December 2008) 11. Meng, X.-L., Song, Z.-W., Li, X.-Y.: Rfid-based security authentication system based on a novel face-recognition structure. In: 2010 WASE International Conference on Information Engineering (ICIE), vol. 1, pp. 97–100 (August 2010) 12. Revankar, S., Sher, D.: Constrained contouring in polar coordinates. In: Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition, CVPR 1993, pp. 688–689 (June 1993)
A Privacy Technique for Providing Anonymity to Sensor Nodes in a Sensor Network Jeong-Hyo Park1, Yong-Hoon Jung1 , Hoon Ko2 , Jeong-Jai Kim3 , and Moon-Seog Jun1 1
3
Department of Computer Science, Soongsil University, Sangdo-Dong, Dongjak-Gu, Seoul 156-743, Korea {helios914,s0178,mjung}@ssu.ac.kr 2 Knowledge Engineering & Decision Support Research Group(GECAD), Institute of Engineering-Polytechnic of Porto (ISEP/IPP) [email protected] #68 Yatap-Dong, Bundang-Gu, Seongnam-Si, Gyeonggi-Do 463-816, Korea [email protected]
Abstract. Research on sensor network had been focused on security services that provide authentication, confidentiality, integrity and availability until recently, but now there is growing interest in tackling the problem of actual sensor IDs being exposed. Many techniques for providing anonymity to the source in an ad-hoc network have been proposed, but they are not suitable. Thus, a technique that is well-suited to the characteristics of sensor networks is needed. This paper, limiting the type of attack against a sensor network to that of eavesdropping, proposes a new technique for providing anonymity using Phantom ID and SMAC. The degree of anonymity provided by the proposed technique was analyzed using an entropy-based modeling technique. The results showed that the anonymity is high when the proposed technique is used. The key factor responsible for the improved anonymity had to do with disguising the sensor ID so that it cant be found easily. Keywords: Sensor Network, Anonymity, Phantom ID, SMAC.
1
Introduction
For sensor networks, wireless communication infrastructure is essential, but it is limited in terms of resources [4]. For example, it only has low computing capability and limited power supply capability [3]. It also has to be implemented at a low cost [17]. Thus, security vulnerability of sensor networks is vastly greater than for other types of networks. Sensor networks suffer from many potential problems: as sensors are exposed in an outside environment they can be physically attacked or a person with ill-intent can plant malicious code to cause malfunction [15]; the wireless transmission signals are exposed [9]; sensor-related equipment has been made mainstream [10]; environmental characteristics sometimes have to be taken into account when installing outdoors [5]; only limited amount of energy T.-h. Kim et al. (Eds.): UCMA 2011, Part I, CCIS 150, pp. 327–335, 2011. c Springer-Verlag Berlin Heidelberg 2011
328
J.-H. Park et al.
can be used and there is only limited computing capability available [16]; and the network organization is atypical and sensors can easily join or leave the network, reorganizing the network [6]. For these reasons, there has been research into security techniques specifically designed for sensor networks [14]. But research into security for sensor networks had focused on security services that provide authentication, confidentiality, integrity and availability until recently. Now, however, there is growing interest in the problem of guaranteeing anonymity of nodes in a sensor network [7], [11]. If an actual ID rather than a fake ID is used to transmit a message, attackers can easily sniff network traffic to analyze it or easily find out not only the ID of the transmitting sensor that is communicating with the BS (Base Station) but track the location of the sensor as well [2]. Thus it is extremely important to make the IDs of the transmitter and the receiver in communication exchanging data difficult to identify by a third party in an environment where a sensor node is performing routine work of detection or object tracking. In addition, in the case of monitoring an important asset, data obtained by the sensor should not be exposed to a third party, and so a security device such as encryption will be used in order to protect the data. But notice that although it may be that data on the assets or the monitored environment, or contents, are important, it also may be that the source that transmitted the data is even more important. Thus an additional device to prevent exposure of the source is needed, that which is appropriate for the characteristics of the sensor network, besides protection by way of encryption. Put in another way, it is necessary to use anonymous receiver/transmitter IDs for messages sent over the network, so that attackers are unable to find out the actual IDs [1], [8], [12]. This paper is organized as follows: Chapter 2 describes problems with existing techniques for providing anonymity; Chapter 3 proposes a privacy technique for ensuring anonymity for sensor nodes in a sensor network; Chapter 4 analyzes the degree of source anonymity provided by the proposed technique; finally, Chapter 5 gives the conclusions.
2
Motivation for Our Research
A privacy problem in WSNs is the naming of nodes and base stations. In particular, we have in mind the problem of identity hiding in the local single-hop communication. Previously described schemes do not address this problem in its full scope. Although some of them have the need to somehow identify the sender or the receiver, no satisfactory solution providing anonymity is proposed. Using the node real ID in this context may be considered as a vulnerable, because an adversary is able to identify individual nodes [17]. A naive solution is to use a key shared with the recipient and hide the IDs by encryption. Nevertheless, in the worst case, the recipient has to perform decryption with the keys shared with each other neighbor, in order to find out whether the message is addressed to her or not. This approach may be computationally expensive and may introduce significant delays. So a better solution is to use some kind of pseudonyms that are
A Privacy Technique for Providing Anonymity to Sensor Nodes
329
transmitted in a plaintext header of the message and help to identify the sender and/or the receiver. However, fixed pseudonyms provide as little anonymity as the real IDs. Therefore dynamically changing pseudonyms should be adopted to provide the node anonymity. The problem of hiding the identity of nodes and the base station in WSNs was first addressed by Misra and Xue [13]. They have proposed the Simple Anonymity Scheme (SAS) that utilizes a contiguous range of pseudonyms to conceal the real node identity. Each node is pre-assigned with a certain number of non-overlapping pseudonym ranges. After the deployment, in the set-up phase, the node associates each pseudonym range with one of its neighbors. Each two neighboring nodes then exchange information on mutually assigned pseudonym ranges. Afterwards, if the node wants to send a message to its neighbor, it uses random pseudonyms from the appropriate pseudonym ranges in place of the sender and the receiver ID. Misra and Xue have proposed also another anonymity scheme called the Cryptographic Anonymity Scheme (CAS). In this scheme, the pseudonyms are generated on per-message basis using keyed hash function HK and are computed as Pseudonym = HKuv (xuv ⊕sequv ) where Kuv is the key shared between communicating nodes u and v, xuv is a random seed shared between u and v, sequv is a message sequence number and ⊕ denotes exclusiveor operation. When compared to the SAS, the CAS is more computationally expensive in exchange for the memory eciency. Ouyang et al. have proposed the Hashing based ID Randomization (HIR) scheme to replace the real node ID [18]. Similarly to the CAS, it utilizes the keyed hash function. However, unlike the CAS where the sequence number is used to change the resulting pseudonym message by message, in the HIR, a new pseudonym is obtained by hashing the previous one, creating a keyed hash chain. The advantage over the 12 CAS arises when the node is compromised. If the HIR is used, it is more dicult for the adversary to reveal the old pseudonyms used. The HIR can be also modified to limit the possibility of an attacker to impersonate the node when the keys are compromised. In this case, the hash chain is pregenerated and used in the reverse order. Nonetheless, when the reverse hash chain is used, each node has only limited number of pseudonyms available.
3
Privacy Scheme That Takes into Account Sensor Node Anonymity
This chapter proposes a technique for providing privacy for sensor node data. Phantom ID is a technique that makes it difficult for attackers to obtain sensor node data simply by analyzing the traffic. It also confuses attackers by creating redundant IDs. As for SMAC, the existing MAC technique was modified to be suitable for the sensor network so that only destination sensors are able to identify transmitter sensors accurately and quickly.
330
3.1
J.-H. Park et al.
Modification to the Data Frame
This section proposes a modification to the data frame as a way to hide the sensor ID.
Fig. 1. Original Data Frame
Fig. 2. Proposed Data Frame
As shown in the beyond figures, when transmitting data, Source ID is replaced with Phantom ID, HV (Hidden Vector) and TS (Timestamp). In addition, CRC is replaced with SMAC (Simple Message Authentication Code). In the discovery stage before communication takes place, sensors communicate using Phantom IDs of neighboring sensors. In particular, HV (Hidden Vector) makes it so that only the destination sensor can identify the source. Using HV, the destination sensor is able to identify the source quickly and at once (see Expression (1)). HV = SID ⊕ KD ⊕ RN D
(1)
Even if at some point during the process HV is exposed, analysis or manipulation is not possible even if RND is known, provided that SID (which only the source knows) and private key KD (which only the destination knows) are not known. Furthermore, as HV is replaced regularly through RND, replay attacks can be prevented. 3.2
Redundant Phantom IDs
In this section, for complete guarantee of source anonymity, PID is used. The above figure shows the cluster composition and the associated table. By using PID, eavesdroppers cant get at sensor data from simply knowing PID. PID is created by each sensor. The process by which PID is created and exchanged among neighboring sensors is as follows. 1. After the initial distribution of sensors, each sensor creates PID according to Expression 2. P ID = q SID mod p
(2)
2. In the neighbor discovery process, PIDs are exchanged among neighboring sensors. The PIDs get encrypted using pair-wise keys that these neighboring sensors have.
A Privacy Technique for Providing Anonymity to Sensor Nodes
331
Fig. 3. Cluster composition and associated table
3. Each sensor stores and maintains PIDs of neighboring sensors in a mapping table. 4. When mapping tables are exchanged, each sensor checks the mapping table that is being maintained. 5. When exchanging IDs, the mapping table is checked. If the same PID exists, the link status of the mapping table is set to UP/DOWN and saved. 6. PIDs are renewed regularly. 3.3
Phantim ID Indentification
This section proposes a technique that prevents exposure of sensor IDs that may occur when local sensors transmit data to the BS or the destination sensor, and using which only the destination sensor can identify the sensor IDs. The above figure shows the path of data as they travel from Sensor A to the destination sensor BS, under the configuration from Section 3.2. Suppose that there are eavesdroppers D and G somewhere along the compromised path from A to BS. Among the values exchanged between A and BS, the values that the eavesdroppers can obtain are PID, HV, RND, SMAC and TS. Security analysis of each of these values is as follows. 1. PID Although PID is created based on source ID, going the opposite way or trying to get at SID based on PID is made difficult by prime factorization. PID is created according to the following expression. P ID = q SID mod p
(3)
2. HV, RND As for Hidden Vector, as even if values of HV and RND are obtained source ID cant be obtained if the private key of BS is not known, it is not possible to get at the source ID. HV is created according to the following expression. HV = SID ⊕ KD ⊕ RN D
(4)
332
J.-H. Park et al.
Fig. 4. Data transmission path from A to BS
According to the above expression, even if values PID, HV, RND, SMAC and TS are exposed to local eavesdroppers D and G, it wont be possible to find out the actual ID of the sensor. 3.4
SMAC(Simple Message Authentication Code) Creation
In this section, CRC (Cyclic Redundancy Check) was substituted by SMAC, taking into account the characteristics of sensor networks where only limited amount of energy is available. As well, factors of security were also taken into account. SMAC is used to ensure integrity of K and TS. It was designed for the energy constraints of sensor networks and is superior to MAC used for wired/wireless networks in this respect. SMAC is created and checked according to Expression 5 and 6 below. 1. Creation (sensor) K = AV SID mod p SM AC = h(k ⊕ T S)
(5)
2. Identification (BS) K = P ID KD mod p SM AC = h(k ⊕ T S)
(6)
In order to check the integrity of data and PID, the transmitting sensor creates K and TS and sends SMAC to the destination sensor, while the destination sensor creates K and checks SMAC against SM AC to ensure integrity of the message.
4
Privacy Evaluation Model
In this chapter, performance of the proposed technique is analyzed and quantified as degree of anonymity. It was shown that the proposed technique provides superior degree of anonymity compared to not using any technique for sensor node privacy.
A Privacy Technique for Providing Anonymity to Sensor Nodes
4.1
333
Degree of Anonymity
To quantify the degree of anonymity, a relational expression of entropy was used (see Expression (7)). H(X) = −
N
[Pi × log2 Pi ]
(7)
i=1
Pi is the likelihood that, when an event of source transmitting data occurs, ith sensor is the source, as judged by the eavesdropper. If the total number of nodes is N, each node can be defined as N1 . When this is applied to the all sources, N i=1 Pi = 1 results. HM is the max. value of entropy (see Expression (8)). HM
N 1 1 =− [ × log2 ] = log2 N N N i=1
(8)
The degree of anonymity is defined as follows, from the definition of entropy. d = 1−
HM − H(X) H(X) = HM HM
(9)
It can be seen that the degree of anonymity is 0 ≤ d ≤ 1 according to the above definition. If a local eavesdropper is completely unable to know which among the sensors is the source, entropy gets the max. value, which denotes the highest degree of anonymity. 4.2
Measuring Degree of Anonymity
This section analyzes the degree of anonymity of the proposed technique using the relational expression of entropy defined in Section 4.1. As can be seen, the degree of anonymity is in the range of 0 ≤ d ≤ 1 according to the following expression. d=
H(X) HM
(10)
Max. value of entropy denotes the highest degree of anonymity while min. value of entropy denotes the lowest degree of anonymity. That is, as d approaches the max. value, HM ≈ H(X) is satisfied. HM = −
N 1 1 [ × log2 ] = log2 N N N i=1
(11)
HM is the entropy value when the likelihood of getting identified by an attacker is the same across all sensors. H(X) is the entropy value after the event and can be measured after the signal has occurred by a sensor node. H(X) = −
N i=1
[Pi × log2 Pi ]
(12)
334
J.-H. Park et al.
In determining d, it can be seen that, when H(X) and HM are compared, the most important factor is Pi , which denotes the likelihood of an eavesdropper being able to find out the source of the sensor when a data transmission event occurs at the source. If Pi = N1 , the values of H(X) and HM are the same, so the degree of anonymity d can said to be the greatest. Pi = Pr (X = i)
(13)
In a network with N number of sensors, the likelihood of attackers being able to find out the source of the event depends on them having knowledge of the times during which sensors are active and their ability to track the location of the sensors. That is, for a network in which anonymity technique is not used, as IDs and data values for all sensors are known accurately, Pi =1 as given by the following expression. But using the proposed technique, IDs of all sensors are created according to the following expression and are used redundantly. Thus Pi of each sensor node is Z1∗ ≈ N1 ( Z1∗ ⊂ n: make n as close as possible). n
n
Zn∗ = {a ∈ Zn | (a, n) = 1}
(14)
Applying Pi of the proposed technique to H(X), the following equation results. H(X) = −
N 1 1 [ × log2 ] N N i=1
(15)
As a result, H(X) gives HM , so it can be proven that the value of d is the max. value by the following equation. d=
5
H(X) =1 HM
(16)
Conclusions
This paper proposed a new privacy scheme for protecting data by providing anonymity to sensor nodes, the design of which took into account characteristics of sensor networks. To this end, the type of passive attacks that can invade privacy of sensor nodes in a sensor network was defined as that of eavesdropper, and in order to provide anonymity to sensor nodes a technique for creating and identifying Phantom IDs was proposed, as well as a SMAC technique. In a sensor network, redundant Phantom IDs are created, which have values unrelated to the original ID of the source. Thus high degree of anonymity can be maintained and replay attacks can be prevented by the regular replacement of values. Furthermore, the degree of anonymity in a sensor network was mathematically analyzed and quantified using an entropy-based modeling technique. The results confirmed the high degree of anonymity provided by the proposed technique, and it was shown that the crucial factor for providing anonymity is to cause confusion for attackers trying to find out the sensor ID.
A Privacy Technique for Providing Anonymity to Sensor Nodes
335
References 1. Li, N., Zhang, N., Das, S.K., Thuraisingham, B.: Privacy preservation in wireless sensor networks: A state-of-the-art survey. Ad Hoc Networks 7(8), 1501–1514 (2009) 2. Liu, D., Ning, P., Liu, A., Wang, C., Du, W.K.: Attack-resistant location estimation in wireless sensor networks. ACM Trans. Inf. Syst. Secur. 11(4), 1–39 (2008) 3. Pietro, R.D., Mancini, L.V., Mei, A.: Energy efficient node-to-node authentication and communication confidentiality in wireless sensor networks. Wirel. Netw. 12(6), 709–721 (2006) 4. Akkaya, K., Younis, M.: A survey on routing protocols for wireless sensor networks. Ad Hoc Networks 3, 325–349 (2005) 5. Gruteser, M., Grunwald, D.: Anonymous usage of location-based services through spatial and temporal cloaking. In: Procs. of the 1st Int. Conf. on Mobile systems, Applications and Services, pp. 31–42. ACM, New York (May 2003) 6. Nezhad, A.A., Makrakis, D., Miri, A.: Anonymous topology discovery for multihop wireless sensor networks. In: Procs. of the 3rd Workshop on QoS and Security for Wireless and Mobile Networks, pp. 78–85. ACM, New York (October 2007) 7. Sweeney, L.: Achieving k-anonymity privacy protection using generalization and suppression. International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems, 571–588 (2002) 8. Wadaa, A., Olariu, S., Wilson, L., Eltoweissy, M., Jones, K.: On providing anonymity in wireless sensor networks. In: ICPADS 2004, pp. 411–418 (2004) 9. Karlof, C., Wagner, D.: Secure routing in wireless sensor networks: attacks and countermeasures. Ad Hoc Networks, 293–315 (2003) 10. Mehta, K., Liu, D., Wright, M.: Location privacy in sensor networks against a global eavesdropper. In: ICNP 2007, pp. 314–323 (2007) 11. Samarati, P.: Protecting respondents’ identities in microdata release. IEEE Trans. Knowl. Data Eng., 1010–1027 (2001) 12. Agrawal, R., Srikant, R.: Privacy-preserving data mining. In: SIGMOD Conference 2000, pp. 439–450 (2000) 13. Misra, S., Xue, G.: Efficient anonymity schemes for clustered wireless sensor networks. IJSNet, 50–63 (2006) 14. Goldschlag, D.M., Reed, M.G., Syverson, P.F.: Hiding routing information. In: Anderson, R. (ed.) IH 1996. LNCS, vol. 1174, pp. 137–150. Springer, Heidelberg (1996) 15. Xi, Y., Schwiebert, L., Shi, W.: Preserving source location privacy in monitoringbased wireless sensor networks. In: Procs. of the 20th Int. Parallel and Distributed Processing Symposium. IEEE Computer Society, Los Alamitos (April 2006) 16. Govindan, R., Estrin, D.: Directed diffusion: A scalable and robust communication paradigm for sensor networks. In: Procs. of the 6th ACM Int. Conf. on Mobile computing and Networking. ACM, New York (August 2000) 17. Ozturk, C., Zhang, Y., Trappe, W.: Source-location privacy in energy-constrained sensor network routing. In: Procs. of the 2nd ACM Workshop on Security of Ad Hoc and Sensor Networks, pp. 88–93. ACM, New York (2004) 18. Ouyang, Y., Le, Z., Xu, Y., Triandopoulos, N., Zhang, S., Ford, J., Makedon, F.: Providing anonymity in wireless sensor networks. In: International Conference on Pervasive Services, vol. 0, pp. 145–148 (2007)
Study on Group Key Agreement Using Two-Dimensional Array in Sensor Network Environment Seung-Jae Jang1, Young-Gu Lee1, Hoon Ko2, and Moon-Seog Jun1 1
Department of Computer Science, Soongsil University, Sangdo-Dong, Dongjak-Gu, Seoul 156-743, Korea {hezc81,ad3927,mjun}@ssu.ac.kr 2 Institute of Engineering-Polytechnic of Porto [email protected]
Abstract. These days, with the implementation of ubiquitous computing concept, sensor networks that collect analyze and process information of things and environment through sensors have become a huge interest as the technology that would realize this concept. However, sensor network technology has wireless communication infrastructure as its foundation and fundamentally has security weakness with limitations on low computing capacity, power supply and low budget. The core of group telecommunications between sensor network sensor nodes is the group key management. Managing group key requires appreciating group key security and management mechanism efficiency. This paper takes the characteristics of sensor network into consideration and not only increases the group key agreement efficiency through Two-dimension array key set distribution and not the existing group key method to satisfy both the network function and security but also protects information between each nodes thereby enhancing the sense of security. Keywords: Sensor Network, Group Key Agreement, Two-Dimensional Array.
1 Introduction With the implementation of ubiquitous computing concept the importance of sensor network is being noticed more than ever as the technology needed to realize this concept. Sensor network is bound in wireless infrastructure and recognizes the situation through installed sensors and connects the recognized data to the utilization service [1]. Long distance communication shall become possible outside the existing short range communications and the sensor nodes are available for use in more industries as low cost, low electricity and high capacity sensor nodes are being mass produced [5]. There have been various studies on sensor networks so far but in order to guarantee a basic security for wireless channels there needs to be a different approach than the existing wireless networks. Not only the common weaknesses of the existing wireless communication but through unique characteristics of the sensor networks makes it
Study on Group Key Agreement Using Two-Dimensional Array
337
hard to implement diverse security schemes. The existing password technology should be made lightweight and realized [7]. In consideration of the sensor network environment characteristics, this paper suggests a mechanism that considers security and real time traits with low calculation amount that uses a preliminary random number distribution method that creates group keys and identity keys and conducts OTP(One time password) functions with simple calculation through KeySet Pre distribution in Two-Dimension array that would fulfill confidentiality, certification and integrity among sensors to enhance network function and security. The formation of this paper is composed of Chapter 2 on group key based key management method and Quorum system needed for group key and Identity key creation and Chapter 3 suggesting group key creation and group key exchange algorithm using Two-Dimensional array. Chapter 4 is consists of the suggested system realization and system analysis and is followed by the final Chapter 5.
2 Related Work 2.1 SPINS (Security Protocols for Sensor Networks) SPINS protocol consists of a SNEP(Secure Network Encryption Protocol) and uTESLA(Timed Efficient Stream Loss-tolerant Authentication). Designing and developing SNEP (Secure Network Encryption Protocol) providing data confidentiality, two party data authentication, and data freshness, with low overhead. Designing and developing uTESLA (the “micro” version of TESLA), providing authenticated streaming broadcast [8]. SNEP provides a number of unique advantages. First, it has low communication overhead; it only adds 8 bytes per message. Second, like many cryptographic protocols it uses a counter, but we avoid transmitting the counter value by keeping state at both end points. Third, SNEP achieves semantic security, a strong security property which prevents eavesdroppers from inferring the message content from the encrypted message (see discussion below). Finally, the same simple and efficient protocol also gives us data authentication, replay protection, and weak message freshness. Authenticated broadcast requires an asymmetric mechanism, otherwise any compromised receiver could forge messages from the sender. Unfortunately, asymmetric cryptographic mechanisms have high computation, communication, and storage overhead, making their usage on resource constrained devices impractical. uTESLA overcomes this problem by introducing asymmetry through a delayed disclosure of symmetric keys, which results in an efficient broadcast authentication scheme [9], [10]. 2.2 Group Key Based Key Management Method Group key based key management method is a method of using group key to maintain security within the group while geographically nearby node form and operate groups together [11]. The most representative method is a group key management study with a basic structure of center aggregator with base station and cluster structure at the
338
S.-J. Jang et al.
core. Among such group key related studies there is a particular research conducted by Jing Deng, Richard Han, and Shivakant Mishra [12], [13]. In this study the assumption is that each sensor node has a 1:1 secret key with the base station beforehand and a three type mechanism that would deliver safe group key with a single echo hash function and uTESLA. Figures 1 shows each stage of this method. First of all, (a) refers to the group announcing stage where the groups are formed beforehand, aggregator is selected and the base station broadcasts aggregator information for each group to all sensor nodes. In the second stage (b) the base station delivers nodes and group key to the aggregator of each group. Stage (c) involves aggregator of each group delivering group recognition ID and group key to nodes in the respective group. The last stage (d) is where each node recognizes the group they belong to individually and safely sends the group key.
Fig. 1. Group key management method
2.3 Quorum System Quorum can be defined with more than one element of the two intersections [14]. In other words, if two intersections have been arbitrarily calculated the meeting of gathering that does not have an empty set can be called a Quorum System. For instance, {1,2,3,4,}, {2,5,6,8}, and {3,8,9,0} is a Quorum system. Intersection calculation with {1,2,3,4}∩{2,5,6,8}={2}, {2,5,6,8}∩{3,8,9,0}={8}, or {1,2,3,4} ∩{3,8,9,0}={3} does not create an empty set and thus is a Quorum System. If the groups are {1,2,3,4,}, {2,5,6,7}, and {3,8,9,0} , intersection of {2,5,6,7} and {3,8,9,0} would be an empty set and thus would not create a Quorum System.
Study on Group Key Agreement Using Two-Dimensional Array
339
Fig. 2. Grid Quorum System
2.3.1 Grid Quorum System In the Grid Quorum System the elements are aligned on the Two-Dimensional level field in line. Each gathering picks a column and a row from the Figures 2 to form a Quorum System. The elements on the selected column and the row shall be collected and be designated as the elements of one’s own. Such gathering shall follow a certain rule and shall be arbitrarily designated. Arbitrarily chosen two gatherings within the system will show at least 2 intersections. Such Quorum System will select 2 -1 element among the total n number of elements and will guarantee more than 2 of common elements [15].
3 Group Key Agreement Using a Two-Dimensional Array The system suggested in this paper is consisted of a BS BS (Base Station) and SN (Sensor Node), as seen on the Total System Configuration in Figures 3. BS takes the core role of creating and distributing group keys when forming group networks and each SN and BS is assumed to have a random number r that has OTP (One Time Password) role and a KeySet that can create keys. With this assumption in minds, the group key creation and distribution process are as follows.
340
S.-J. Jang et al.
Fig. 3. Group Key Agreement System
1. BS sends Group Key Generation Message to the SN_A(Sensor Node A). 2. BS and SN_A randomly selects one column and one row from KeySet. 3. SN_A creates MaskRow, MaskColumn value by taking hash value of column and row value with rA shared with IDA and XOR calculation. 4. SN_A sends IDA, MaskRow, MaskColumn value to the BS. 5. BS sends necessary column and row to SN_A that is necessary for SN_A to create Identity Key A. 6. BS creates a random number rA that is to be newly shared with the Group Key. 7. The created Group Key and new_rA will be encoded as the Identity Key A and is delivered to SN_A.
Fig. 4. KeySet for Group Key Creation
Study on Group Key Agreement Using Two-Dimensional Array
341
Here BS and SN (Sensor Node) receives a Keyset in Two-Dimensional array shown in Figures 4. Separately delivered KeySet can input a 10x10 value and each cell an input Capital A-Z, small a-z, and numbers 0-9. The cells will be filled with random functions. BS and SN makes calculations to create group key and recognition key with the Keyset. Two-dimensional field is the numbers that represent columns and rows and the line that divides cells which has been divided by numbers. Using such double grid will create a grid space that allows drawing shapes while being a Two-Dimensional array field. 3.1 Group Key Creation Algorithm Using a Two-Dimensional Array Assumption of 3 SN Groups with A, B, C is made before each SN goes through the following procedure to create group key. 1. BS sends group key creation message to each SN. 2. BS and each SN randomly selects line and row from the Keyset at hand. A's Select : KeySet(XA,YA) B's Select : KeySet(XB,YB) C's Select : KeySet(XC,YC)
(1)
M's Select : KeySet(XM,YM)
3. Each SN creates MaskRow, MaskColumn value by taking hash value of line and row value with r value commonly shared with ID and calculating XOR and sends it to the BS with personal ID.
4. BS confirms random number r through the ID value sent by the SN from the database and deducts line and row values from the MaskRow and MaskColumn value sent by SN. 5. BS goes through process to deduct value from own Keyset as shown on Figures 5 with line and row value sent by each SN. The extracted value is shown on Table 1. Table 1. Extracted Value to KeySet Group’s Value A’s Value B’s Value C’s Value BS’s Value
Column 1 9 3 6
Row 1 7 8 4
SEED VALUE 6 5 1 J
6. Group key is created using a hash function by attaining the center of gravity value of the shape drawn with the selected values and coordinates.
∥ ∥1∥J∥{Centroid(x,y)}]
Group Key = Hash[6 5
(3)
BS will attain group key after the aforementioned process. 3.2 Group Key Exchange Algorithm In order to safely send the group key to each SN, confidentiality, integrity and nonrepudiation, etc needs to be guaranteed. Safe group key distribution requires encoding the created group key where the group member uses the same key used in encoding to conduct decoding the encoded group key. In order to achieve that there needs to be a method to safely deliver the key used in encoding and this is called Key exchange algorithm. Figures 6 displays the order of process required for recognition exchange. BS will send group key creation message to each SN and BS and each SN will randomly select line and row from one’s Keyset. Next, the procedure of each SN sending its ID, MaskRow, MaskColumn value to BS is the same as that of the group key creation procedure. BS with the created group key will go through the following process and exchange the group key to safely deliver the group key safely to each SN. 1. BS that received column and row value delivered by each SN, and SN A, SN B, SN C that received column and row value selected by BS shall deduce intersection elements from the Keyset at hand as shown in Figures 7.
Study on Group Key Agreement Using Two-Dimensional Array
Fig. 6. Group Key Creation and Identity Key Exchange Algotithm
Fig. 7. Identity Key Creation Method
343
344
S.-J. Jang et al.
The extracted value is displayed on Table 2. Table 2. Intersection Element Group A B C BS
Set {D, 6, T, u, o, B, 3, 8, f, L, A, k, p, I, 4, g, 5, c, R} {z, R, J, Q, G, e, X, 5, d, L, I, 8, i, B, h, 3, 7} {o, p, 8, a, t, Z, 1, 5, M, f, K, w, H, z, d} {1, g, S, q, J, x, 2, h, t, m, U, o, j, H, v, C, e, G}
∩M {o, g} {J, G, e} {o, t, 1, H}
2. Identity key is created using a hash function with the calculated intersection element.
∥g) Identity Key B = Hash(J∥G∥e) Identity Key C = Hash(o∥t∥1∥H) Identity Key A = Hash(o
(4)
3. BS encodes the random number r that is to be newly shared with the group key and delivers it to each SN. EncryptionIdentity_Key_A(Group Key, new_rA) EncryptionIdentity_Key_B(Group Key, new_rB)
(5)
EncryptionIdentity_Key_C(Group Key, new_rC) 4. Each SN that receives the encoded group key will conduct decoding with own Identity key and will attain random number r that is to be newly shared with the group key which will be used for actual data delivery. DecryptionIdentity_Key_A(Group Key, new_rA) DecryptionIdentity_Key_B(Group Key, new_rB)
(6)
DecryptionIdentity_Key_C(Group Key, new_rC)
4 System Realization and Analysis 4.1 System Realization System for function evaluation on the suggested system is realized using C# of Intel(R) Core(TM)2 Quad CPU Q9400 @ 2.66GHz 2.67Ghz, RAM 3.00GB’s Window 7’s operation system Ultimate K 32bit.
Study on Group Key Agreement Using Two-Dimensional Array
345
Hash function needed for group key and Identity key creation used SHA-1 function where as AES parallel key encoding algorithm have been applied to encode the created key and delivery message. Experiment is composed of BS for group key creation and SN A, SN B and SN C that would previously have a KeySet and a random number r. Each SN will connect to BS and deliver the selected value (column and row) to the BS. BS will determine SN A and B as the geographically close nodes and create group key by forming a SN A, B group. Delivering the message to after encoding it with the group key to each SN A, B, and C via created group key will allow normal message display through decoding the encoded message sent to SN A and B that are connected to the group but SN C not connected to the same group can only look up the encoded message. Figures 8 displays SN A and B creating the recognition key with BS determining only SN A and B as groups even though all SN A, B, C is connected and sends the selected X and Y value.
Fig. 8. Identity Key Creation of SN
Figures 9 shows group key delivery encoded in each SN’s recognition key upon group key creation button on the BS. SN C shall not receive group key since it is not included in the group.
346
S.-J. Jang et al.
Fig. 9. Group Key Creation and Key Exchange
Figures 10 is the screen shown when “Test Message” text is sent to SN A, B, and C from the BS.
Fig. 10. Send Message
Study on Group Key Agreement Using Two-Dimensional Array
347
Here, SN A and SN B that has the group key will receive the original text through decoding the encoded message but SN C that does not have the group key will receive a non-recognizable message since it is incapable of decoding the encoded text. 4.2 System Analysis Key creation method using personal information such as ID or MAC and IP is the most simple one but it allow allows easy exposure or change of such information in case of a malicious user. In other words, since the user certification direction is a oneway it may be exposed to risks such as personal privacy invasion, etc. On the other hand, Diffie-Hellman method or the suggested system conducts a two-way certification for user certification and can be said safe from fraud identification attacks and is also safer than other methods regarding personal privacy issues since it does not involve using personal information for key creation. In general hash function takes the role of elongating or reducing a certain value to a certain extent. Also, hash function is often used for key creation method due to its quick calculation speed. This paper also has used hash function for key creation in which a value necessary for key creation is deducted within a Two-Dimensional array closed frame. For instance, creating a key using hash function after selecting 4 values will create n! number of possible keys and will thus create 4! = 4 3 2 1 = 24 different keys. However, the suggested method in this paper will create not a 4! but a 5! Number of keys since multi-angle shape’s center of gravity value is calculated and added despite selecting 4 values. In other words, selecting n number of values will create (n+1)! number of possibilities and will make it difficult for the malicious user to recognize the key. Table 3 is a comparison of number of vertexes for the necessary time to calculate multi-angle shape’s center of gravity value needed for group key creation.
ⅹⅹⅹ
Table 3. The necessary time to calculate muti-angle shape’s center of gravity
1 2 3 4 5 6 7 8 9 10 Average
3-polygon 8.1ms 8.4 ms 8.2 ms 8.2 ms 8.1 ms 8.3 ms 8.1 ms 8.2 ms 8.2 ms 8.3 ms 8.21 ms
4-polygon 8.8 ms 8.7 ms 8.4 ms 8.7 ms 8.9 ms 8.8 ms 8.6 ms 8.6 ms 8.7 ms 8.8 ms 8.7 ms
5-polygon 8.5 ms 8.8 ms 8.9 ms 9.1 ms 8.5 ms 8.8 ms 9.2 ms 9.1 ms 8.9 ms 8.9 ms 8.87 ms
6-polygon 8.6 ms 9.6 ms 8.4 ms 9.5 ms 8.4 ms 8.3 ms 8.4 ms 9.3 ms 8.5 ms 9.3 ms 8.81 ms
7-polygon 8.8 ms 8.4 ms 9.8 ms 8.3 ms 8.5 ms 8.6 ms 8.8 ms 8.2 ms 9.1 ms 8.8 ms 8.73 ms
348
S.-J. Jang et al.
As shown in the table, triangle shows average value of 8.21ms, square 8.7ms, pentagon 8.87ms, hexagon 8.81ms, and heptangular with 7 points- average of 8.73ms. What we can see here is that the amount of calculation will not increase to a great extent despite the increasing number of the vertex and it consumes approximately under 0.1 second to find the center of gravity for a multi-angle shape. Figures 11 is a graph version of Table 4 and makes it easy to recognize that the average speed of calculating center of gravity of a multi-angle shape is approximately 8.5ms.
Fig. 11. The necessary time to calculate muti-angle shape’s center of gravity
5 Conclusion Sensor network environment emphasizes limits on calculation capability, storing mechanism and electronic mechanism with wireless network formation between nodes and makes it its purpose to create a lighter version of the existing mechanism. Also, the broadcasting characteristics of wireless make it easy to have message fraud/change issues and confidentiality impeding attacks such as sensor information taps, message reuse, etc. In consideration of sensor network environment, this paper creates group key through a simple calculation via a Two-Dimensional array KeySet Pre-Distribution which will satisfy confidentiality, certification, and integrity between BS and SN to enhance network function and security while suggesting a mechanism that will increase efficiency and security using a preliminary random number distribution method that conducts a OTP (One Time Password) role. The suggested method takes a Two-Dimensional array that has a double coordinate to calculate necessary group key creation to minimize key and personal information exposure and resolves key overlapping problems by evading key colliding risks in a closed frame by utilizing multi-angle shape’s center of gravity. A method that uses double coordinate array and shape information regarding center of gravity used in this paper shows that calculating multi-angle shape’s center of gravity requires a very subtle amount of calculation for the users. It shows
Study on Group Key Agreement Using Two-Dimensional Array
349
possibilities for the future adaptability to any types of environment that requires various values that do not overlap within the confined resources and not the domain limited to group key creation.
References 1. Culler, D., Estrin, D., Srivastava, M.: Overview of Sensor Network. IEEE Computer Society, UC Berkeley (2004) 2. Goldreich, O., Goldwasser, S., Micali, S.: How to construct random function. Journal of the ACM 33(4), 792–807 (1986) 3. Haekins, D., Carrel, D.: The Internet Key exchange (IKE): RFC 2409, Information Sciences Institute, University of Southern California (1988) 4. Deng, J., Han, R., Mishra, S.: Enhancing Base Station Security in Wireless Sensor Networks. Technical Report CU-CS-951-03, Department of Computer Science, University of Colorado (2003) 5. Jolly, G., Kuscu, M.C., Kokate, P.: A Low - energy KeyManagement Protocol for Wireless Sensor Networks. In: Proc. of the 8th IEEE International Symposium on Computers and Communications (2003) 6. Chan, H., Perrig, A., Song, D.: Random key pre-distribution schemes for sensor network. In: IEEE Symposium on Research in Security and Privacy (2003) 7. Chan, H., Perrig, A., Song, D.: Random Key-Assignment for Secure Wireless Sensor Networks. In: Proc. of the 1st ACM Workshop on the Security of Ad Hoc and Sensor Networks (SASN) (2003) 8. Perrig, A., Szewczyk, R., Wen, V., Culler, D., Tygar, J.: SPINS: Security Protocols for Sensor Networks. In: Proceeding of Seventh Annual ACM International Conference on Mobile Computing and Networks (Mobicom) (2001) 9. Perrig, A., Canetti, R., Song, D., Tygar, J.D.: Efficient and secure source authentication for multicast. In: Network and Distributed System Security Symposium, NDSS 2001 (2001) 10. Perrig, A., Canetti, R., Tygar, J., Song, D.: Efficient authentication and signing of multicast streams over lossy channels. In: IEEE Symposium on Security and Privacy (2000) 11. Eschenauer, L., Gligor, V.: A Key - Management Scheme for Distributed Sensor Network. In: Proc. of the 9th ACM Conference on Computer and Communications Security (2002) 12. Deng, J., Han, R., Mishra, S.: Security Support for In-Network Processing in Wireless Sensor Networks. In: Proc. of the 1st ACM Workshop on the Security of Ad Hoc and Sensor Networks (SASN) (2003) 13. Zhu, S., Setia, S., Jajodia, S.: LEAP: Efficient Security Mechanism for Large-Scale Distributed Sensor Network. In: Proceedings of the 10th ACM CCS 2003 (2003) 14. Colbourn, C.J., Dinitz, J.H., Stinson, D.R.: Quorum systems constructed from combinatorial designs. Information and Computation, 160–173 (2001) 15. Lang, S.D., Mao, L.J.: A torus quorum protocol for distributed mutual exclusion. In: Proc. of the 10th Internat. Conference on Parallel and Distributed Computing and Systems, pp. 635–638 (1998)
Design and Materialization of Location Based Motion Detection System in USN Environments Joo-Kwan Lee1 , Jeong-Jai Kim2 , and Moon-Seog Jun1 1
2
Department of Internet Information, Seoil University, Myeonmok-Dong, Jungnang-Gu, Seoul 131-702, Korea {jklee0131,mjung}@ssu.ac.kr #68 Yatap-Dong, Bundang-Gu, Seongnam-Si, Gyeonggi-Do 463-816, Korea [email protected]
Abstract. Regardless of the technical significance held by USN, the unmanned surveillance using wireless network cameras, the sensing technologies using the trespassing surveillance sensors as well as the technologies of encoding the data occurring within such systems, the technologies of transferring and processing them through the sensor network and the efficient person positioning and placement method using RFID have become a single integrated system in the industrial field for the security of facilities and have not yet been provided as services. The present paper proposes the user tracking and management systems using RFID, sensor and microwave intrusion detecting to replace the composition of network cameras and the existing infrared intrusion detecting sensors. Keywords: USN, RFID.
1
Introduction
Changes of various paradigms including people to people, people to objects and objects to objects are predicted in the ubiquitous environment. Such ubiquitous environment is made possible through the network that acquires and manages the information through the sensor attached to all objects. In other words, it is enabled through the Ubiquitous Network(USN). Due to its characteristic of being able to connect the physical world to the digital world, the developments of USN can be applied to many different fields. For example, it is possible to use trespassing detectors or fire sensors in particular networks to take measures against burglary and fires ahead of time and as it is used for distribution management of cognitive service or facilities in industrial sites, the active studies are currently being progressed in numerous fields including various medical systems or scientific fields [1]. Regardless of the technical significances held by USN, the methods of unmanned surveillance using wireless network cameras, surveillance technologies using trespassing detection sensors and encoding the data that occurs in such systems to transfer and process the data through sensor network as well as the efficient person positioning and placement method using RFID, have all yet to be serviced as one integrated system within the industrial field of facility security. T.-h. Kim et al. (Eds.): UCMA 2011, Part I, CCIS 150, pp. 350–360, 2011. c Springer-Verlag Berlin Heidelberg 2011
Design and Materialization of Location Based Motion Detection System
351
The suggested system will not only develop and verify the key infrastructure technology of the ubiquitous era but the development and utilization of the application system using this foundation will hold a great significance within the technical aspects. For this, the suggested system is related to user tracking and management using RFID and microwave trespassing detection sensor for replacing the existing infrared trespassing detection sensors and the composition of network cameras [2]. The goals of the developed system are managing facilities using RFID, USN and wireless cameras, on-site detection of personnel as well as improving and suggesting the efficiency of the system.
2 2.1
Related Studies Sensor
Infrared sensors are largely classified as thermal types and photon types according to the motion principles [3]. The thermal types are sensors that alter the infrared rays into heat and output in the forms of resistance variation or electromotive force. On the contrary, the photon type uses the photovoltaic effects caused by photoconductivity effects using the moving energy absorption of the semiconductor. In general, the photon type detects the abandoned photon exposed by the infrared rays, the semiconductor materials of InSb, MCT(HgCdTe), INAs and PbS are mostly used. Furthermore, the photon types hold the advantages of great reception, fast response and enabled to apply the semiconductor manufacturing techniques, the cooling of extremely low temperature is essential for the element to hold sufficient amounts of sensitivity. The microwave using sensor is applied with the X-Band(10.525GHz) using the Doppler principle to hold the semi-permanent lifespan with the method of specially controlling the power through latching relay and semiconductor element and also holds the characteristics of measuring the movements of liquids for the power of underground parking lots in public buildings by going through obstacles. Through controlling the power during motion surveillance, it is enable to actualize energy reduction [4]. 2.2
Camera
The system called the closed circuit television delivers the video information to specific users for it to be the industry accomplishing the boom in domestic regions as well as exportations. This CCTV industry began with the foundation of small scaled equipments of manufacturers in the beginning but upon entering the 1980s, the CCTV system equipment investments began based on foreign parts to vitalize production and at the present time, the majority of products are being localized by large corporations as well as smaller enterprises. As a type of the security industry, the future-oriented added value is being created based on the electronic digital technologies while appertaining to information communication
352
J.-K. Lee, J.-J. Kim, and M.-S. Jun
and since the acceleration of information society based on the competitive pursuit multimedia by social safety and productivity is forming various application technologies and integrated markets, it holds great potential for development. In particular, it is being developed as the wireless CCTV with no requirement for circuit work based on the computer network multimedia technologies and the storage media has also expanded from DVR to wireless DVR [5],[6]. 2.3
RFID Tag
The system for real-time positioning of objects using RFID can be materialized and by using the RFID Active Tag, the cognitive range may be wider than the RFID Passive Tag but there is the tendency that the it is used only for particular purposes rather than the general application of RFID [7]. For example, it can be applied to high-priced assets requiring accurate positioning, the positioning management of assets that requires accurate status due to it being shared by various people, the migration routes of outsiders and confirmation of inside personnel and outsiders in restricted areas. There are various systems using infrared rays, ultrasonic waves, RFID, UWB and signal strength in the location recognition system. The summarization of the characteristics for each of the location recognition systems are as follows [8]. Table 1. Location Recognition System Feature Location Location Recognition Accuracy Scale Recognition System Technology Distance GPS measurements using Within 10m 24 satellites around wireless mobility time 24 satellites around Active Badge
Infrared cellular proximity method
About size of room
RFID
proximity method
1m
RADAR Easy Living
802.11 RF scene analysis and triangulation Vision and triangulation
3
4.3m
Variable
Limit
Outdoor
Sunlight A base station / interference of Badge of a room infrared Should know A sensor of each location of location sensor 3 Base station of each floor 3 small camera
Need wireless NIC Ubiquitous public camera
However, different limitations are involved for each system since they operate in different environments. For example, the limitation held by the GPS system is that it is unable to detect satellite signals from the interiors. To solve this, there is the method of installing the GPS repeater on the roof of a building to broadcast the signal once more from the ground. Meanwhile, the reader of the RFID tagging system is unable to accurately distinguish the tag when numerous tags exist, the effective anti-collision algorithm must be developed to prevent any collisions [9].
Design and Materialization of Location Based Motion Detection System
3
353
Suggested System
The subjects for application of USN in the suggested system can be divided into the areas to detect trespassing, areas for surveillance and areas for management of dangerous articles. When there exist abnormalities in the managing subjects after blending the sensor and camera as well as the tag and sensor(hybrid reader) technologies for the characteristics of each area, the USN must be composed so that quick measures can be taken or the efficient personnel positioning and management is possible using the tag and sensor(hybrid reader) technology.
Fig. 1. USN structure of each kind of sensors
The area for trespassing detection is to be composed for constant monitoring by using cameras and microwave sensors for surveillance. When trespassing is detected in the monitored area by the sensors, the sensor will transfer the sensor ID, sensed data and location information to the sever as the sever uses the sensor data to move the camera viewpoint to the location of the sensor where the trespassing has been detected to ultimately save image or video as well as inform the manager with the warning. To efficiently manage the personnel placed in areas for facility management, efficient allotment of working hours, management of employee schedules and surrounding environment information are gathered by having the personnel possess Active Tags enabled for wireless ID recognition and the Hybrid Readers attached on the structures. If an abnormality were to occur in a different area, it is enabled to send the closest personnel to solve the problem or warn the occurrence to the manager through the central server to take appropriate measures. 3.1
Motion Flowchart of RF and USN Hybrid Reader
The process of receiving data from the Active Tag to deliver the location information of the Tag to the server is as follows.
354
J.-K. Lee, J.-J. Kim, and M.-S. Jun
Fig. 2. Motion flowchart of RF and USN hybrid reader
3.2
Motion Flowchart of Sensor and Camera
In the case of a trespasser in the subject of facility management, the detection is made through the microwave sensor that transfers the sensor information to the server to move the cameras viewpoint to the location of the sensor in storing as image and video or uses the motion detection technique to store the time of occurrence along with the location information of the camera in protecting the facility by warning the manager.
Fig. 3. Motion Flowchart of Sensor and Camera
3.3
Microwave Detector
Since the existing infrared and thermal trespassing detection sensors do not penetrate obstacles, there are many restrictions for installations in locations with numerous obstacles. Generally, the nonmetallic objects hold the characteristics of detecting through the thickness of approximately 5cm it holds advantages of not having blind spots for indoor and outdoor areas for the methods of sending and receiving frequencies. Moreover, it is possible to voluntary control the range of detection and is enabled to detect up to 15M. The following displays the interior block diagram for microwave detection.
Design and Materialization of Location Based Motion Detection System
355
Fig. 4. interio block diagram of Microwave Detector
As displayed in Diagram 4, when the PWM signal of Duty Ratio 4% and 2Khz output from the control part is supplied to the X-band sensor power control, the X-band sensor part delivers the frequency to form the detection domain. When motion of objects are detected in the domain, the IF signal is outputted as the phase deviation is created for the transmission and receive frequency. Since this IF signal is low in level, it is inputted into the A/D converter of the control part by going through the filter and the amplification process and the A/D converter makes the distinction of whether it is the static signal or the detection signal of moving objects. When the signal is determined as the detection of a moving object, the timer sends output signals for 2 seconds to operate the lead relay for 2 seconds. 3.4
DVR Motion Detection Event Using Network Camera
When motion signal is detected in DVR, the motion is detected with the blockmatching method. It was designed and developed so that a single picture is split into 64 parts for detection and the scanning would be accomplished once every second. Furthermore, it was also designed and developed so that the data will be stored into the event record database to review the past records through searches in the future. 3.5
Location Analysis System Using RFID Tag and USN Reader
As the system applied with the method of Received Signal Strength Indicator(RSSI) to the RTLS technology, the location tracking engine was composed by using the Reference Tag and the virtual Reference Tag. The 2.45GHz RFID Active Tag Data is received from the police employees and visitors through the all of the sensor network-based USN Readers inside the store to analyze the positioning analysis by using the functions of gathering, filtering and the location tracking engine.
356
J.-K. Lee, J.-J. Kim, and M.-S. Jun
Fig. 5. Motion search using block matching method
4 4.1
Method of Experiment Detection Experiment of 2.45GHz RFID Active Tag and USN Reader
To test and evaluate the functions of the possessed/developed RFID and USN hardware equipments and their applied solutions, the actual test environment was constructed within the company to conduct the test so that it will be ready for direct application when the building like the public office was actually constructed. The cognitive rates and distances were tested by comparing the cases with and without screens for the RFID equipments and the sensor data of the USN Reader determine the standard value for each usage to analyze the data collected from the sensor. Moreover, the applied solutions enabled for the employees and outsiders to test the actually used results according to each scenario for verifying the safety and efficiency of the solutions. The transferred data sends the data to the control system through the USN Reader and sends the data required to each of the location analysis engine. Here, the agent server converts the data for transfer so that it can be stored in the database of the intelligent server. In the same method as mentioned above, the location detection was accomplished to gain the results as shown in Table 2. The motion speed of Table 2 was tested with 3 items of slow, medium and fast. Each item viewed the average speed of a persons walk as being medium to consider walking slower to be slow and walking faster to be fast for the test. The motion distance displays the total distance from the starting point to the end point of the detection and the number of motion determines the movements made from the starting point to the end point for the test to be conducted 100
Design and Materialization of Location Based Motion Detection System
357
Table 2. Location Detection Test Motion Motion Motion Location Detection Type speed Distance Counting Recognition Counting Recognition Rate(%) Slow 15M 100 100 100.00 F1 1 Medium 15M 100 100 100.00 Fast 15M 100 99 99.50
times for each item. As displayed in the table, 1 cognitive error occurred during the 100 movements when moving fast and such cases were caused because there was a difference in the periodic reception speed of the USN Reader(0.3 sec ∼ 1 sec) and the movement speed of the person for the detection to be made in the next location rather than being in the present location. This occurs when location of the present USN Reader and the next one both recognize the Tag data while collecting the USN Reader(strengths of the frequencies are sorted to recognize the side that reads the smaller frequency first). 4.2
Experiment of Trespassing Detection Sensor
As the product that unified the X-Band sensor into the PCB, it was designed and developed to detect motion from the minimum of 1m to maximum of 15m in distance. In particular, it was installed with the variable resistance to enable the control of detection domains and sensitivity. Table 3. Motion Detection Test Motion Motion Location Detection speed Distance Recognition Counting Recognition Rate(%) Slow 15M 100 100% Medium 15M 100 100% Fast 15M 100 100% Slow 1M 100 100% Medium 1M 100 100% Fast 1M 100 100%
4.3
Remote Control Function of Network Camera
The corresponding program and DVR equipment is not the method of processing in the existing analogue signal but rather connected through the IP Address of the Network and the corresponding equipments were developed to process 100 users at once. Next displays the screen of remote controlling with each device by using corresponding programs within other Network Addresses. It was designed and developed to enable the functions of Pan/Tilt/Zoon through the DVR program from exteriors and by clicking the mouse point it 1
F1 : Location detection test on USN Reader with only 2.45GHz RFID Active Tag.
358
J.-K. Lee, J.-J. Kim, and M.-S. Jun
Fig. 6. Tested in an external network DVR program
is possible to freely rotate the screen. Furthermore, automatic expansion function operates when dragging the desired area and also supports auto-focus.
Fig. 7. Remote Contorol Feature
This DVR equipment uses the Wavelet compression technique to store highresolution images and holds the functions of controlling the number of frames that can be received per second as well as the functions of image extraction. 4.4
Experiments on the Conveyance of Alarm to Trespassing Alert System from the Control Monitoring System
The cognitive rates to the control monitoring systems were measured by the RFID/USN experiments mentioned above and this clause displays the experimentation for the transference to the trespassing alarm system after concluding that the received signal is the signal of trespassing.
Design and Materialization of Location Based Motion Detection System
359
Fig. 8. Recived bandwidth per second settings
Based on the judgment of trespassing, the message is conveyed as the TCP/IP socket transfer when sending the message to each serve system and through the Multi-Casting as the IP of each serve system, the number of message conveyance and error lists were prepared for the experiment and confirmed that all were safely conveyed. The port number used at the time was 7000 and in the cases without firewalls, they were safely delivered. Table 4. Information Transfer Test Sending Location Detection Type Alarm Sending Counting Recognition Rate(%) Sending to U-Terminal Trespassing Alarm 100 100% Trespassing Alarm 100 100% Sending to user PC Sending Notice 100 100%
There were no cases of incorrect indications of corresponding areas of trespassing for warnings and confirmed good operations without any errors of the emergency lights.
5
Conclusion
Regardless of the fact that the detection using cameras, the detection technologies using sensors and the technologies of encoding the data that occurs in such systems to transfer and process through sensor networks, were enabled for utilization in same industrial fields, the technologies and solutions developed at the present time could not integrate and use these functions as a single system. However, the technology mentioned above is definitely the most essential element technology, in other words, infrastructure, demanded in the era of ubiquitous and when used, it will enable the surveillance against trespassers of various facilities
360
J.-K. Lee, J.-J. Kim, and M.-S. Jun
and application for efficient personnel placement using RFID and furthermore, will allow to take quick measures by perceiving fires and terrors in real-time when using temperature, illumination and smoke sensors. Through the system suggested in this paper, the development and verification of the key infrastructure of the ubiquitous era as well as the development and utilization of the applied system using this foundation will hold a great significance. As a future study, there are goals to introduce the system used in the present study to real life and make arrangements through linking with other applications.
References 1. Kim, E.E.: Usn standard and technology. In: 10th International Conference on Advanced Communication Technology, ICACT 2008, vol. 2, pp. 1415–1418 (February 2008) 2. Sheng, Q., Li, X., Zeadally, S.: Enabling next-generation rfid applications: Solutions and challenges. Computer 41, 21–28 (2008) 3. Kolev, I., Stoeva, I.: Development of testers for indication of infrared radiation. In: 27th International Spring Seminar on Electronics Technology: Meeting the Challenges of Electronics Technology Progress, 2004, vol. 3, pp. 448–450 (May 2004) 4. Abegaonkar, M.P., Karekar, R.N., Aiyer, R.C.: Miniaturized nondestructive microwave sensor for chickpea moisture measurement. Review of Scientific Instruments 70, 3145–3149 (1999) 5. Ta’eed, O.: Getting past big brother-transforming cctv into mis. In: 30th Annual 1996 International Carnahan Conference on Security Technology, 1996, pp. 176–183 (October 1996) 6. Broadley, E.T.: M11 cctv. In: Road Transport Information and Control - RTIC 2008 and ITS United Kingdom Members’ Conference, IET, pp. 1–6 (May 2008) 7. Moen, H., Jelle, T.: The potential for location-based services with wi-fi rfid tags in citywide wireless networks. In: 4th International Symposium on Wireless Communication Systems, ISWCS 2007, pp. 148–152 (October 2007) 8. Huang, H.-P., Chang, Y.-T.: Robust design for rfid system testing and applications. In: IEEE International Conference on Robotics and Automation, 2007, pp. 4594– 4599 (April 2007) 9. Hightower, J., Borriello, G.: Location systems for ubiquitous computing. Computer 34, 57–66 (2001)
Realization of Integrated Service for Zoos Using RFID/USN Based Location Tracking Technology Jae-Yong Kim1, Young-Gu Lee1, Kwang-Hyong Lee2, and Moon-Seok Jun1 1
Department of Computer Science, Soogsil University, Sangdo-Dong, Dongjak-Gu, Seoul 156-743, Korea {raient,ad3927,mjun}@ssu.ac.kr 2 Department of Computer Science, Seoil University, Myeonmok-Dong, Jungnagn-Gu, Seoul 131-702, Korea {dreamace}@seoil.ac.kr
Abstract. In addition to the diffusion of internet, the advancements in information communication technology have continued with the appearances of numerous new technologies and supply of new services. Ubiquitous computing has also been developed based on such technologies and is the most representative information communication technology that has been developed. The wireless communication technology of ubiquitous computing does not only provide useful services in industries but also in various environments including households, schools and hospitals. This paper suggests the system that accurately confirms the location of the moving body by fusing the communications network and USN based RFID and based on that provides various services. Based on the location confirmation of the moving body, the various service systems are being developed and provided in hospitals, ports and distribution centers and this paper proposes the integrated service system for zoos and theme parks. The sensor with its distinct ID is attached to children, who are the major subjects for protection in zoos and theme parks, to search the proximity contents with the location information gained through the RFID receiver located in the elements throughout the zoo and output the searched contents through the output device. Moreover, the current location information and course of the moving body is used to provide services to protect children from getting lost. Along with the supply of education contents suited to the situation of the moving body, the specialized integrated service is materialized to take measures when children get lost in zoos. Keywords: Sensor Network, Group Key Agreement, Two-Dimensional Array.
interface. As the key technology of ubiquitous computing, the RFID/USN technology is evolving to distinguish objects, tracking personal records, monitoring state information, real-time surveillance and control as well as autonomous services and is gaining the spotlight as the promising business for future growth as the key to IT technology fusion in various fields including distribution, construction, traffic, agriculture, national defense and medicine. The ubiquitous computing RFID/USN technology can provide services by applying on the overall field of human activity rather being restricted to the industrial field. Starting with the most basic field of households, it is possible to provide the services specialized to the location and characteristics including recreation sites and theme parks. The present paper realizes the integrated service for zoos using the RFID/USN based location tracking technology. This paper proposes the system that provides various services based on accurately confirming the location of the moving object by fusing the technologies of USN based RFID and communications network. Based on the location confirmation of the moving body, various service systems within hospitals, ports and distributions centers are being developed and provided. Moreover, this paper proposes the integrated service system for zoos and theme parks. The sensors with distinct IDs are attached to the children in zoos in materializing the specialized integrated service that allow to take direct measures when children get lost or can provide education contents suited to the situations of the moving bodies with the location information gained through the RFID receiver located in the elements throughout the zoo.
2 Related Work 2.1 RFID System RFID is the technology of automatically recognizing the data stored in the tag, label and cards with micro-chips from the reader using the wireless frequencies. The RFID system secures the weaknesses of existing barcodes or self-recognition devices and is the next generation technology that has improved to hold the convenience of usage and much information[1]. The RFID system is currently being used in numerous fields of applications including, national defense, medicine, distribution, traffic, security, manufacture, construction, service and administration. Such RFID system is accomplished with 3 components of tag(transponder), reader(interrogator) and back-end server[2]. 2.1.1 Tag Tag is the device that is attached to humans, objects and animals to transmit the direct or indirect distinguishable and cognitive information for the object. Generally, a tag is composed of 1 IC chip and 1 antenna. Tags hold their own distinct information ID(TID) and largely can be divided into the Active Tags and Passive Tags.
Realization of Integrated Service for Zoos Using RFID/USN
363
Fig. 1. RFID system
The Active Tags are installed with their own batteries inside the tag and are enabled for self arithmetic operations and its transference range of data also reaches up to more than tens of meters. However, it is high in price due to the internal battery and holds the weakness that the lifespan of the tag is subordinate to the lifespan of the battery. The Active Tag is mainly used on containers for trade, location recognition and healthcare[3]. Passive Tags do not carry its own batteries within itself and secures power through the electronic induction by using the frequencies sent from the reader. Since the transfer power of the tag is weak, it has shorter range of transfer in comparison to the Active Tags and therefore, is mostly used in short distance information communication. However, it is low in cost and small in size since it has no internal battery and also has a long lifespan to be used in many fields of application. It will be the tag mostly used from now on since it is appropriate for usage on distinguishing objects in short distances. 2.1.2 Reader The RFID reader is the device that transmits with the tag to collect the information of the tag and performs the function of transferring the collected data to the middleware. Furthermore, the RFID reader is composed of the RF analogue part and the digital signal process control part. The RF analogue part takes charge of the functions such as power transference and analogue signal processing while the digital signal process control part is accomplished of the decoder and encoder. RFID reader can be classified as the fixed type and the mobile type. 2.1.3 Back-end-Server Back-end Server is the server system that processes the tag information transferred from numerous readers. The back-end server manages the tag and related information by storing as database and for the purpose of efficiency it can be operated in various servers. Moreover, the back-end server is considered as the system that is reliable within the aspects of security[1].
364
J.-Y. Kim et al.
2.2 Trend in Location Tracking Technologies There are various forms of real-time location tracking technologies for moving objects including the Cell tracking method, GPS tracking method and the Network method and each of these methods hold different location error ranges and technical features[3]. The RFID method used for the location tracking in this paper have been used mainly for interiors in comparison to those previously mentioned, it has also verified the efficiencies for real-time location tracking for moving objects of exterior environments with the proposals of various systems[4,5] that can reduce the errors when grasping the location of moving objects along with the strength that it is possible to distinguish the objects as well as track locations[1,5].
3 Integrated Service for Zoos Using RFID/USN Based Location Tracking Technology 3.1 Overview of Integrated Service for Zoos Using RFID/USN Based Location Tracking Technology The goal of this study is to provide the system to prevent lost children and provide the educational contents suitable to each location by grasping the location information of children who have been selected as the subjects within locations of zoos and theme parks. For this, the accurate location tracking of moving objects using the RFID tag and the receiver must be performed in advance.
Fig. 2. Overall organization of the system
Realization of Integrated Service for Zoos Using RFID/USN
365
After acquiring such information, it must be managed through the information management server and through the integrated control system, the services for providing contents and preventing lost children will be provided. The following figure is the overall organization of the system. Unlike how the past location tracking using the RFID had been limited to the restricted areas of mainly interiors and distribution storages, the accurate location tracking device is required for wide external spaces like zoos. The RFID receiver that receives the location information of Active Tags should be attached on the fixed structures such as light fixtures and outer walls of animal cages. After being attached on the fixed structures, the RFID receiver will be able to receive stable power and use the location coordinates of corresponding structures as the standard points to confirm the real-time location of the moving objects more accurately. The location information of the moving object carrying the RFID Tag, confirmed through the RFID receiver, is collected and managed from the location information management server. As the integrated control system, the location information management server shares the location information to go through the collective module of location information of integrated control system, location information knowledge base and service adjustment module to provide the education contents and services for prevention against lost children. 3.2 Location Tracking The provision of integrated service suggested in this paper begins with the tracking for the accurate location information of the moving object. Amongst the many location determination methods, this paper applies the TDOA[6] location determination method of Range-Based Approach. The location information is calculated by the location relationship with close neighboring nodes in the RangeBased Approach and the absolute distance between the sender and receiver can be measured by the properties of over one communication signal from the sender to the receiver. The TDOA determines the location of the moving object by using the geometric relationship based on the differences of the signal conveyance time observed based on 3 or more location reference points. The Active Tag attached to the moving objects are granted with the distinct ID value that distinguishes each Tag and the location information of the Active Tag acquired through the RFID receiver is managed in real-time through the location information management server. Furthermore, the location information management server transfers and shares the information of the Active Tag’s distinct ID, time information and location information to the integrated control system. The following figure is the output screen for the information of the Active Tag.
366
J.-Y. Kim et al.
Fig. 3. Tag Location Information
3.3 Supply of Integrated Service The present paper provides the services for providing contents and prevention against lost children through specialized integrated service system for locations of zoos and theme parks. The location information of the moving objects gained through the location information management server is transferred and shared through the integrated control system and the organization of the integrated control system is as shown in the following figure.
Realization of Integrated Service for Zoos Using RFID/USN
367
The integrated control system is composed of the location information collecting module, the location information knowledge base and the service control module. Moreover, the location information collecting module compares and analyzes the collected data directly with the knowledge base and determines the occurrence of lose children and differentiates the contents near each moving object. Based on the data compared and analyzed between the knowledge base and collected module, the service determination module, the service determination module provides the contents to devices such as smartphones and PDAs.
4 Realization Environment The integrated service for zoos realized in this paper is divided into the mobile client module for providing integrated service contents and the RFID location tracking technology sensor module for the location tracking of moving objects. The realization environments of each are as follows. Table 1. Realization Environment Module Mobile Client Module
5 Conclusion The RFID/USN technology of the past had numerous restrictions due to problems of great margins of error, transfer capabilities of simple information within limited short distances and inefficient power management. However, the necessities of wireless
368
J.-Y. Kim et al.
communication technologies within various environments and national technical developments have continued to radiant developments of RFID/USN technologies. The RFID/USN technology provided autonomous service, real-time surveillance and control, monitoring of status information, personal record tracking and distinction of objects and soon began to provide various application services such as status information process and location tracking. Moreover, it can be stated as the key technology of ubiquitous computer that can naturally use the desired service at any place and time. This paper proposes the integrated service for zoos using the location information of RFID/USN technology based. The specially integrated service system is suggested for the zoo composed with the applied service technologies and USN/middleware platform that provides the services through creating knowledge contents and status recognition information by storing, processing and integrating the collected data along with the RFID/USN technology that efficiently detects and conveys the various status information of all objects and physical environments. The integrated system that can provide contents suitable to each situation and take immediate measures for lost children by accurately confirming the real-time location of the moving object through using the location information of the RFID receiver attached to the fixed structure is materialized and suggested. It is anticipated to find the solution through future studies.
References 1. Cha, M., Kyo, J.: A USN Based Mobile Object Tracking System for the Prevention of Missing Child. Journal of Computing Science and Engineering (2008) 2. Choi, J.W.: Indoor positioning using received signal strength and reference points in wireless LAN. Seoul University (2006) 3. Want, R.: Enabling ubiquitous sensing with RFID. IEEE Computer (2004) 4. Ni, L.M., Liu, Y., Lau, Y.C., Patil, A.P.: Landmark: Indoor Location Sensing Using Active RFID. In: IEEE International Conference on Pervasive Computing and Communications (2004) 5. Cha Maeng, Q., Jin, C.A.: A Study on the Real Time Location Sensing of Moving Object using Active RFID Reference Points of Road Lamp. Korean Society of Civil Engineers (2007) 6. Ho, K., Chan, Y.: Solution of performance analysis of geolocation of TDOA. IEEE Transactions on Aerospace and Electronic Systems (2004) 7. Lim, J.H.: study on enhanced position estimation of sensor node under the location attack in wireless sensor networks. Korea Universty (2010) 8. Garfinkel, S., Rosenberg, B.: RFID Applications, Security, and Privacy (2005) 9. Draft ITU-T Recommendation X.1171 (X.nidsec-1), Threats and Requirements for Protection of Personally Identifiable Information in Applications using Tag-based Identification (2008) 10. Walsh, W.P.: Research and application of RFID Technology to enhance aviation security. IEEE, Los Alamitos (2000) 11. Ward, A., Jones, A., Hopper, A.: A New Loacation Technique for the Active Office. IEEE Personal Comminications (1997)
Real-Time Monitoring System Using Location Based Service Jae-Hwe You, Young-Gu Lee, and Moon-Seog Jun Department of Computer Science, Soongsil University, Sangdo-Dong, Dongjak-Gu, Seoul 156-743, Korea {hwe100,ad3927,mjung}@ssu.ac.kr
Abstract. In this thesis, the future-oriented system using RFID technology was developed for figure out the number of people following by the guide promptly and effectively every time location changes. By having the tourists carry RFID tag, the group leader can check the number of the tourists by collecting the information on surrounding tags realtime through PDA with attached RFID reader. To make this feasible, the RFID based location check for the group members were made possible. To examine the safety of the proposed system, the directionality of antenna was tested. The result showed that the presence of the followers can be easily checked using the reader, processing the signals from the active tag, it could overcome the limitations of the location which the existing the prevention of lost children system in theme parks. Keywords: RFID, Active Tag, LBS, RTLS.
1
Introduction
The terms, RFID and ubiquitous are no longer the words that are not unfamiliar to lay people [1]. Each city and district governing authorities are implementing the pilot or actual project related to RFID and making progress with the pilot project and service through the investments and support from each department of state agencies. In addition, most of the large companies have formed the exclusive team for RFID which is already been used in our daily lives and accelerated the preparation in response to the age of RFID and ubiquitous. Despite such trends, the method of orally checking or directly counting the group and number of tourists in culture and tourism areas have persisted have persisted, which caused much inconvenience[6]. In this thesis, in response to such demands, the future-oriented system was developed to promptly and effective check the people in charge by the tourist guide using RFID and ubiquitous technology whenever the domestic and foreign tourist groups relocate to different places. By having the tourists carry around RFID tag, the group leader can collection information from the surrounding tags real-time using PDA with attached RFID reader. To check the presence of the tourists(audience) to tourist spots such as cultural heritage, museums and galleries by the leader(guide) viable, RFID based location identification of the T.-h. Kim et al. (Eds.): UCMA 2011, Part I, CCIS 150, pp. 369–379, 2011. c Springer-Verlag Berlin Heidelberg 2011
370
J.-H. You, Y.-G. Lee, and M.-S. Jun
group members was made possible. To explain this in depth, for instance, four electric waves such as 0 dbm (recognition distance: 30m), -10 dbm (recognition distance: 15m), -20 dbm (recognition distance: 5m) can be transmitted at once and these can be used for each guide check system. It was developed to be used for future tourist location recognition and the given reader obtains only necessary information for the system by distinguishing and using them for the basic data for the system operation[1].
2 2.1
Related Works Types of Localization Technology
Location detection technology (LDT), or localization technology, refers to the technology used in the service or the application system which utilizes this by accurately positioning the location of the person or the object in question. Generally, Location-based service (LBS), refers to the location based service in the mobile networks and it can be classified diversely depending on the method and applicable range of the location detection technology. It is classified according to the scope and technology where LDT is applied. Furthermore, the characteristics, pros and cons and potential application location for each technology can be summarized as follows. GPS (Global Positioning System). GPS is the most familiar technology for us that has infiltrated deep into our lives. The technology used for GPS, identifies the location in the worldwide or extensive range as its name indicates. Based on the transmitted signals from 27 GPS satellites which rotate in separate orbits, the position of the given object can be determined[7]. To determine the location using the GPS receiver, the signals from at least 4 satellites has to be received. If the signal can be transmitted from more satellites, more accurate location would be obtained. However, the drawback to GPS technology is that the transmission of signals is not possible inside the building. GPS positioning has about 50 200m error but using error correction method such as DGPS(Differential GPS), it can be reduced to within 5m[8]. LBS (Location-Based Service). LBS refers to the location based service which determines the location of a person or object and offers additional service based on the location information using mobile phones or equipments with secured mobility in mobile networks. Depending on the service format, LBS can be divided into cell format using the mobile communication base station and GPS format which uses the satellite. Cell format uses the base stations distributed throughout the country, making the location determination with error of about 500 1500m possible. However in case of densely populated areas of the metropolitan cities, the base stations are installed in the interval of tens of meters, reducing the range of error within 100m. Cell format has a wide range of error, which is disadvantageous in that only rough location determination is possible. However, its advantage is that it can find the location within the building or in the basement using the gap filler[9].
Real-Time Monitoring System Using Location Based Service
371
RTLS (Real -Time Location System). RTLS is a real-time location system applicable to the narrowest area in terms of position tracking range. In other words, RTLS refers to the service which figures out the location of a person or a given object indoor such as inside the factory or office and outdoor within the limited range such as yards and parks, and offer additional features based on such location information[10].
Fig. 1. Real-Time Location System(RTLS)
Generally, it uses Wi-Fi technology which is used in wireless LAN and the adoption of next generations technology such as UWB and ZigBee is being thoroughly reviewed. In RTLS system also, as in the case with GPS and LBS, the proximity which are known for triangulation and presence features uses cell format which tracks the current location by checking the location of the cell in the space is divided into small cells. Among them, the location tracking method which is the most universal is the location tracking using triangulation and the location determination with triangulation would be based on RSSI or TDOA technology. RFID system. Tag can be attached to people, object or animals, and it is a device which transmits direct or indirect discrimination and recognition information about the object. Generally, tag consists of one IC chip and one antenna[2]. Each tag has unique information identification (TID) and can be largely classified as an active tag and a passive tag. Active tag is equipped with a battery on the tag, with independent algorithm execution and the range of data transmission can be up to tens of meters. However, since the battery is built in, the price of the tag is high. Its drawback is that the life span of the tag depends on the lifespan of the battery. Active tag is mainly used in trade containers, location recognition and health care. Passive tag does not have a battery on the tag itself, using the transmission from the reader and induce the electromagnet to secure the power. Tags transmission power is weak and thus compared to active tag, the transmission range is short which makes it viable for use the information communication in the short distance. However, it does not have a built in battery and it is very cost effective in the production price. Since it can also be produced in a small size with generally long user life span, it has a wide areas of application[3][4].
372
J.-H. You, Y.-G. Lee, and M.-S. Jun
RFID reader is a device which transmits to collect information from the tag. It functions in transmitting the collected information from the tag to the middle wear. RFID reader consists of RF analogue part and digital signal processing control part. RF analogue part functions in the power transmission and analogue signal processing while digital signal processing control part consists of decoder and encoder. RFID reader can be classified into fixed-type and portable-type. Fixed-type reader transmits the tag information wirelessly and performs the role of sending the received information to the server. Portable-type reader has the built-in wireless network interface to submit the received tag information on the reader by wireless to the server. The main function of the reader is to transmit the tag information to analyze the data from the tag in sub system or backend server[5].
3 3.1
Proposed System Overview of the Proposed System
As shown in , the user (generally, tourists, audience and kindergarteners) uses the active tag and the group leader uses round-type reader which is attached on PDA. For the group leader to confirm the users in the vicinity, for the basic concept of ubiquitous, anytime, anywhere, the ID value of active tag and the information of tourists are matched. Then active tags are designed to transmit different intensities of frequency. As a result, depending on the intensity of the given frequencies, the location of the tourist can be determined to see how far the person is located from the bus or the group leader. To develop such system, the basic material for the business in culture and tourism areas was examined and the plan is in progress to have the development system could be approached from the users position. With the cooperation of the tourism businesses, the tourist patterns of tourists upon the domestic and overseas travels would be obtained and how the development system was applied to actual travel product was modeled. below is the actual simulation results on the application of the developed technology by selecting one from the travel products. 3.2
2.45GHz Active Tag and Reader Development
As can be confirmed from , RF Modem with 2.45GHz requires diverse specification. What is first required for M:N format communication method is for the simultaneous use for one group leader and several tourists(audience). Moreover it can be used for various application systems when it can simultaneously identify several active tags at once. In order to support such functions, the selection of RF Modem proved to be crucial in the design. In addition, to secure more than 6 months of Battery Life Time and the process of tuning RF equipments would be required. Lastly, using the output value, the distance adjustment of 3m 20m is secured, transmitting 3 frequencies such as -5 dbm (recognition distance: 8m),
Real-Time Monitoring System Using Location Based Service
373
Fig. 2. Overview of the proposed location checking system
Fig. 3. Road map of travel pattern analysis
-10 dbm (recognition distance: 5m) and -20 dbm (recognition distance: 3m). Connecting such information to various services such as the guide person check system, traffic analysis system for the tourists (audience) and interaction with them, the system was designed. In case of recognition distance, the system was designed and developed to identify maximum 30 m using tests and tuning. The issues which can occur here is adjusting the recognition distance of the reader which is attached to the transportation device (bus) or the reader used in connection with the handheld PDA. Reader for PDA will be tuned to maintain 8m recognition distance and reader for the transportation device, 3m. After the transmission of the signal from 3m distance for use and through the reception adjustment tuning of antenna, about 1.5m of recognition distance should be maintained. With such development, the reader which is attached on the transportation device at the entrance of the transportation, the entry of the tourists (audience) can be accurately determined.
374
J.-H. You, Y.-G. Lee, and M.-S. Jun
Fig. 4. Developing hardware spec and operation principle
The communication regulations are as follows. Table 1. Communication Protocol Type Command Response Reader RESET 0x10 R : Reset, O : OK Network status 0x20 N : Network Status, O : OK Data transmission 0x30 Q : None, D : Data Tag registration 0x40 I : Information
A. Reader/Controller Reset - Reset Command should be the same for Reader/Controller Table 2. Command Type Length Command CheckSum Contents TotalSize Type response Length ∼ Command (XOR value) Size(Byte) 1 1 1 1 Value 4 0X10 ’R’
B. receiver/Controller network status information - receiver/Controller network status information, Command is the same for receiver/Controller
Real-Time Monitoring System Using Location Based Service
375
Table 4. Command Type Length Command CheckSum Contents TotalSize Type response Length ∼ Command (XOR value) Size(Byte) 1 1 1 1 Value 4 0X20 ’N’ Table 5. Response Type Length Command CheckSum Contents TotalSize Type response Length ∼ Command (XOR value) Size(Byte) 1 1 1 1 Value 4 0X20 ’O’
C. Tag registration information - Pack sent from the exclusive tag registration reader Table 6. Response Type Length Command CheckSum Contents TotalSize Type response Length ∼ Command (XOR value) Size(Byte) 1 1 1 1 Value 4 0X40 ’O’
D. Data transmission - Length * The overall length is variable. After an interval of a set time (500ms or 1Sec), the received ID has to be collected and sent at once, the basic length 4 + data number * 6 - Tag ID * RID : Receiver ID. 0x00 0x0F 1byte(A maximum of 16 installation in a controller). * TID : Tag ID. 4Byte. Length can be modified Table 7. Send Data Type Length Command
Contents Total Size Type Response Tag ID Receiver Status
Size(Byte) Value 1 4+data*6 1 0X30 1 ’D’ Data Reserved 1 bit Action type 1bit Length ∼ Data (XOR Value) Receiver type 3bit none Battery type 1bit Output type 2bit Check-Sum Length ∼ Data (XOR Value) 1
376
E. -
J.-H. You, Y.-G. Lee, and M.-S. Jun
Data Format changes CR, LF added CR(0x0A), LF(0x0D) added to the end of all data The length of all data is the same as 12 Bytes Because TCP/IP communication module transmits by unit of 12 Bytes, in case of short data, command or response, the remaining parts of 12 byes should be added with ”0x00” and later CR and LF is to be added.
Table 8. Example - Command(Hex Code : 04 10 R XX 00 00 00 00 00 00 CR LF) Type Length Command CheckSum Contents TotalSize Type response Length ∼ Command (XOR value) Size(Byte) 1 1 1 1 Value 4 0X10 ’R’
Table 9. Example - Response(Hex Code : 04 10 O XX 00 00 00 00 00 00 CR LF) Type Length Command CheckSum Contents TotalSize Type response Length ∼ Command (XOR value) Size(Byte) 1 1 1 1 Value 4 0X10 ’R’
- Command of ’Command’ and ’Response’ except received ID packet can be applied.
4 4.1
Analysis Measurement Results from the Electromagnetic Measurement Center
To determine the intensity of frequency and directionality of this RFID equipment, the measured results from the electromagnetic measurement center were requested. Readers measurement results. The results of the measurement on the readers antenna radiation characteristics, as shown in , the radiation characteristics 45 behind on the left was the best from the top. Better radiation characteristics were measured from the right than from the left because the IC and other parts are built in on the left according to the board design of RFID Reader and GND copper interferes with the electromagnetic radiation. In case of a top part or bottom part with antennas, GND copper and other displays are not present, which makes the electromagnetic radiation active. On the other hand, it was not active in the PCB section. For propagation loss, the loss ranges from the minimum of -5dBm to maximum of -10dBm.
Real-Time Monitoring System Using Location Based Service
377
Fig. 5. Electromagnetic waves radiation pattern of reader on horizontality location
Fig. 6. Electromagnetic waves radiation pattern of reader on verticality location
Fig. 7. Electromagnetic waves radiation pattern of necklace tag on horizontality location
378
J.-H. You, Y.-G. Lee, and M.-S. Jun
Fig. 8. Electromagnetic waves radiation pattern of necklace tag on verticality location
Tag measurement results. As shown in and , in case of necklace-type tag, the overall propagation loss is significant. Furthermore, the necklace-type tag shows similar radiation characteristics to the reader. The reason for it seems to be the same reason with the reader and since the size of PCB is small, in turn GND copper is quite small as well. Thus very irregular radiation pattern is observed.
5
Conclusion
In this thesis, the guide or the leader of the group would carry the portable tourist guide terminal (PDA or PDA phone) with 2.45 GHz RF Reader attached which would allow him or her to check the presence of the group members through the signals from the active tag in possession of the member with ease. As a result, the limitations in terms location which the existing lost children prevention system in theme parks overseas can be overcome. In case of the reader and tag used in this system, they all had directionality with more marked difference in the side than in front/back side. For future design of PCB, the directionality of antennas must be considered along with the layout of the parts and location of antenna. The antenna and matching parts must be well designed to minimize the electrical loss of the electric waves radiated into the space. Due to the expansion of theft prevention system of the exhibits and relics in the tourism spots, the tags must be attached to them, making the real-time monitoring of the asset management possible. To make the monitoring of the manager through notice process possible once they overstep the designated area, the theft prevention system and portable tag would have to be expanded. Furthermore, using portable tag in possession of the tourists, the ways to make applications for making payments and using point (coupon) information in various facilities such as the restaurants and shops in the tourism spot will be further researched.
Real-Time Monitoring System Using Location Based Service
379
References 1. Hightown, J., Borriello, G.: Location System for Ubiquitous Computing. IEEE Computer Society 34(8), 57–66 (2001) 2. Yeoh, W.G., et al.: A 2.45-GHz RFID Tag with On-Chip Antenna. In: IEEE RFIC Symp. Dig., pp. 365–368 (June 2005) 3. Barras, D., Ellinger, F., Jackel, H., Hirt, W.: Low-Power Ultra-Wideband Wavelets Generator with Fast Start-Up Circuit. IEEE Transactions on Microwave Theory and Techniques 54, 2138–2145 (2006) 4. Chee, Y.H., Niknejad, A.M., Rabaey, J.M.: An Ultra-Low-Power Injection Locked Transmitter for Wireless Sensor Networks. IEEE Journal of Solid-State Circuits 41, 1740–1748 (2006) 5. Kim, J.G.: A Divide-and-Conquer Technique for Throughput Enhancement of RFID Anti-collision protocol. IEEE Communications Letters 12(06), 0474–0476 (2008) 6. Bagci, F., Petzold, J., Trumler, W., Ungerer, T.: Ubiquitous mobile agent system in a p2p-network. In: UbiSys-Workshop at the Fifth Annual Conference on Ubiquitous Computing, Seattle, USA, October 12-15 (2003) 7. Kaplan, E.D.: Understanding GPS: Principles and Applications. Artech House Publishers, Boston (December 2005) 8. Rempel, R.S., Rodgers, A.R.: Effects of differential correction on accuracy of a GPS animal location system. J. Wildl. Manage. 61(2), 525–530 (1997) 9. Chen, F., Yang, C., Yu, W., et al.: Research on Mobile GIS Based on LBS. In: Geoscience and Remote Sensing Symposium (IGARSS 2005) (2005) 10. Cho, H., Jung, Y., Choi, H., Jang, H., Son, S., Baek, Y.: Precise Location Tracking System based on Time Difference of Arrival over LR-WPAN. In: The First ACM International Workshop on Mobile Entity Localization and Tracking in GPS-less Environments, pp. 67–72 (September 2008)
