Communication and Networking: International Conference, FGCN 2011, Held as Part of the Future Generation Information Technology Conference, FGIT 2011, ... in Computer and Information Science, 266) 3642272002, 9783642272004

Citation preview

Communications in Computer and Information Science

266

Tai-hoon Kim Hojjat Adeli Wai-chi Fang Thanos Vasilakos Adrian Stoica Charalampos Z. Patrikakis Gansen Zhao Javier García Villalba Yang Xiao (Eds.)

Communication and Networking International Conference, FGCN 2011 Held as Part of the Future Generation Information Technology Conference, FGIT 2011 in Conjunction with GDC 2011 Jeju Island, Korea, December 8-10, 2011 Proceedings, Part II

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] Wai-chi Fang National Chiao Tung University, Hsinchu, Taiwan, R.O.C. E-mail: [email protected] Thanos Vasilakos University of Western Macedonia, Kozani, Greece E-mail: [email protected] Adrian Stoica Jet Propulsion Laboratory, Pasadena, CA, USA E-mail: [email protected] Charalampos Z. Patrikakis National Technical University of Athens, Greece E-mail: [email protected] Gansen Zhao Sun Yat-sen University, Guangzhou, China E-mail: [email protected] Javier García Villalba Universidad Complutense de Madrid, Spain E-mail: [email protected] Yang Xiao University of Alabama, Tuscaloosa, AL, USA E-mail: [email protected] ISSN 1865-0929 e-ISSN 1865-0937 ISBN 978-3-642-27200-4 e-ISBN 978-3-642-27201-1 DOI 10.1007/978-3-642-27201-1 Springer Heidelberg Dordrecht London New York Library of Congress Control Number: Applied for CR Subject Classification (1998): C.2, H.4, I.2, H.3, D.2, F.1 © Springer-Verlag Berlin Heidelberg 2011 This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, re-use of illustrations, recitation, broadcasting, reproduction on microfilms or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer. Violations are liable to prosecution under the German Copyright Law. The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. Typesetting: Camera-ready by author, data conversion by Scientific Publishing Services, Chennai, India Printed on acid-free paper Springer is part of Springer Science+Business Media (www.springer.com)

Foreword

Future generation communication and networking is an area that attract many professionals from academia and industry for research and development. The goal of the FGCN conference 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 future-generation communication and networking. We would like to express our gratitude to all of the authors of submitted papers and to all attendees for their contributions and participation. We acknowledge the great effort of all the Chairs and the members of Advisory Boards and Program Committees of the above-listed event. Special thanks go to SERSC (Science and Engineering Research Support Society) for supporting this conference. We are grateful in particular to the speakers who kindly accepted our invitation and, in this way, helped to meet the objectives of the conference. December 2011

Chairs of FGCN 2011

Preface

We would like to welcome you to the proceedings of the 2011 International Conference on Future Generation Communication and Networking (FGCN 2011) — one of the partnering events of the Third International Mega-Conference on Future-Generation Information Technology (FGIT 2011) held during December 8–10, 2011, at Jeju Grand Hotel, Jeju Island, Korea. FGCN 2011 focused on various aspects of advances in future-generation communication and networking. 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 the FGCN 2011 Chairs, International Advisory Board, Committees, Special Session Organizers, as well as all the organizations and individuals who supported the idea of publishing this volume of proceedings, including the SERSC and Springer. We are grateful to the following keynote, plenary and tutorial speakers who kindly accepted our invitation: Hsiao-Hwa Chen (National Cheng Kung University, Taiwan), Hamid R. Arabnia (University of Georgia, USA), Sabah Mohammed (Lakehead University, Canada), Ruay-Shiung Chang (National Dong Hwa University, Taiwan), Lei Li (Hosei University, Japan), Tadashi Dohi (Hiroshima University, Japan), Carlos Ramos (Polytechnic of Porto, Portugal), Marcin Szczuka (The University of Warsaw, Poland), Gerald Schaefer (Loughborough University, UK), Jinan Fiaidhi (Lakehead University, Canada) and Peter L. Stanchev (Kettering University, USA), Shusaku Tsumoto (Shimane University, Japan), Jemal H. Abawajy (Deakin University, Australia). Last but not the least, we give special thanks to Ronnie D. Caytiles and Yvette E. Gelogo of the graduate school of Hannam University, who contributed to the editing process of this volume with great passion. We would like to express our gratitude to all of the authors and reviewers of submitted papers and to all attendees, for their contributions and participation, and for believing in the need to continue this undertaking in the future. December 2011

Tai-hoon Kim Hojjat Adeli Wai-chi Fang Thanos Vasilakos Adrian Stoica Charalampos Z. Patrikakis Gansen Zhao Javier Garc´ıa Villalba Yang Xiao

Organization

Honorary Chair Dae-sik Ko

Mokwon University, Korea

General Co-chairs Wai-chi Fang Thanos Vasilakos Adrian Stoica

National Chiao Tung University, Taiwan University of Western Macedonia, Greece NASA JPL, USA

Program Co-chairs Charalampos Z. Patrikakis Gansen Zhao Javier Garc´ıa Villalba Tai-hoon Kim Yang Xiao

National Technical University of Athens, Greece Sun Yat-sen University, China Universidad Complutense of Madrid, Spain GVSA and University of Tasmania, Australia University of Alabama, USA

Workshop Chair Byungjoo Park

Hannam University, Korea

Publicity Co-chairs Houcine Hassan Damien Sauveron Qun Jin Irfan Awan Muhammad Khurram Khan Yang Xiao J.H. Abawajy

Polytechnic University of Valencia, Spain University of Limoges, France Waseda University, Japan University of Bradford, UK King Saud University, Saudi Arabia The University of Alabama, USA Deakin University, Australia

Publication Chair Maria Lee

Shih Chien University, Taiwan

X

Organization

International Advisory Board Hsiao-Hwa Chen Gansen Zhao Han-Chieh Chao Hamid R. Arabnia Gongzhu Hu Byeong-Ho Kang Aboul Ella Hassanien Tughrul Arslan Jianhua Ma Sankar K. Pal Xiaofeng Song

National Sun Yat-Sen University, Taiwan Sun Yat-sen University, China National Ilan University, Taiwan The University of Georgia, USA Central Michigan University, USA University of Tasmania, Australia Cairo University, Egypt The University of Edinburgh, UK Hosei University, Japan Indian Statistical Institute, India Nanjing University of Aeronautics and Astronautics, China Oslo University College, Norway

Frode Eika Sandnes

Program Committee Aboul Ella Hassanien Ai-Chun Pang Aggeliki Sgora Albert Banchs Andres Iglesias Prieto Andrzej Jajszczyk Antonio Lagana Benahmed Khelifa Bogdan Ghita Chao-Tung Yang Chia-Chen Lin Christophe Fouquer´e Chu-Hsing Lin Clement Leung Damien Sauveron Dimitrios D. Vergados Don-Lin Yang Driss Mammass Farrukh A. Khan Gianluigi Ferrari Hong Sun Hsiang-Cheh Huang

Hsin-Hung Chou Hui Chen Huirong Fu J. Vigo-Aguiar Janusz Szczepanski Jiann-Liang Jieh-Shan George Yeh Jiming Chen Juha Ro”ning Kin Keung Lai Kwok-Yan Lam Li Shijian Luis Javier Garc´ıa Villalba Marc Lacoste Matthias Reuter Michel-Marie Deza Ming-Yen Lin Feng Mohammad Moghal Nashwa El-Bendary Neveen I. Ghalii Nikolaos Pantazis

Special Session Organizers Hong Kook Kim Tae-Young Byun Y. Byun

N. Jaisankar Ning Gui Omar Soluiman P.R. Parthasarathy Ricky Yu-Kwong Kwok Robert Goutte Rui L. Aguiar Shun-Ren Yang Soon Ae Chun Stephen Huang Sun-Yuan Hsieh Tae (Tom) Oh Terence D. Todd Victor C.M. Leung Viktor Yarmolenko Vincent Oria Vincenzo De Florio Weili Han Witold Pedrycz Yung-Hui Li Feng Yvette E. Gelogo Ronnie D. Caytiles

Table of Contents – Part II

Studies on the Key Technologies of Multi-Platform Mobile Thin Client System: Cross-Layer Isolation and Session Allocation . . . . . . . . . . . . . . . . . Biao Song, Wei Tang, Tien-Dung Nguyen, Sang-Ho Na, Jun-Hyung Lee, and Eui-Nam Huh

1

LDPC Equalizer for Compensating the CFO and Phase Noise in OFDM System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Do-Hoon Kim and Heung-Gyoon Ryu

11

TC-HMIPv6: A Study of HMIPV6 Handover Management for Packet Transmission Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sung-Gyu Kim, Farkhod Alisherov, and Byungjoo Park

20

A Multi-hop Communication Scheme for IEEE 802.11p Based V2V Communication Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Woong Cho and Hyun Seo Oh

26

A Political Communication Scheme of Citizen Network System on Disembedding and Embedding Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jang-Mook Kang and Bong-Hwa Hong

34

Web Contents Mining System for Real-Time Monitoring of Opinion Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ho-Bin Song, Moon-Taek Cho, Young-Choon Kim, and Suck-Joo Hong An Energy-Efficient Cluster-Based Routing in Wireless Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Seongsoo Cho, Bhanu Shrestha, Keuk-Hwan La, Bong-Hwa Hong, and Jongsup Lee A Management of Resource Ontology for Cloud Computing . . . . . . . . . . . Hwa-Young Jeong and Bong-Hwa Hong

43

57

65

Development of an Algorithm for Video Quality Measurement for Broadcasting Communications Services . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sang-Soo Kim, Hae-Jong Joo, and Euy-Soo Lee

73

An Effective Resource Managements Method Using Cluster-Computing for Cloud Services . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Seong-Sik Hong and Jin-Mook Kim

83

XII

Table of Contents – Part II

Study on Micro-processing Implementation of USN Environment Data by a Graphic-Based Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Young-Wook Lee

90

Web Based Remote Robot Control for Adjusting Position on Manufacturing System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hwa-Young Jeong and Bong-Hwa Hong

96

Discrimination of Speech Activity and Impact Noise Using an Accelerometer and a Microphone in a Car Environment . . . . . . . . . . . . . . . Seon Man Kim, Hong Kook Kim, Sung Joo Lee, and Yun Keun Lee

104

Crosstalk Cancellation for Spatial Sound Reproduction in Portable Devices with Stereo Loudspeakers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sung Dong Jo, Chan Jun Chun, Hong Kook Kim, Sei-Jin Jang, and Seok-Pil Lee Perceptual Enhancement of Sound Field Reproduction in a Nearly Monaural Sensing System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chan Jun Chun, Hong Kook Kim, Seung Ho Choi, Sei-Jin Jang, and Seok-Pil Lee

114

124

Quality-Aware Loss-Robust Scalable Speech Streaming Based on Speech Quality Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jin Ah Kang, Seung Ho Choi, and Hong Kook Kim

132

Artificial Bandwidth Extension of Narrowband Speech Signals for the Improvement of Perceptual Speech Communication Quality . . . . . . . . . . . Nam In Park, Young Han Lee, and Hong Kook Kim

143

Improvements in Howling Margin Using Phase Dispersion . . . . . . . . . . . . . Jae-Won Lee and Seung Ho Choi Secure Client-Side Digital Watermarking Using Optimal Key Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jing-Jing Jiang and Chi-Man Pun Effective Electronic Advertisement Auction System . . . . . . . . . . . . . . . . . . . Tokuro Matsuo and Satoshi Takahashi

154

162

169

Energy-Efficient Fire Monitoring Protocol for Ubiquitous Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Heemin Kim, Ae-cheoun Eun, Sunyoung Han, and Young-guk Ha

179

Design of Optimal Combination for New and Renewable Hybrid Generation System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Kun Hyun Park, Chul Uoong Kang, Gi Min Lee, and Jong Hwan Lim

189

Table of Contents – Part II

Parameter Optimization of UWB Short Range Radar Detector for Velocity Measurement in Automobile Applications . . . . . . . . . . . . . . . . . . . Purushothaman Surendran, Chul-Ung Kang, and Seok-Jun Ko Data Signature-Based Time Series Traffic Analysis on Coarse-Grained NLEX Density Data Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Reynaldo G. Maravilla Jr., Elise A. Tabanda, Jasmine A. Malinao, and Henry N. Adorna Automated Video Surveillance for Monitoring Intrusions Using Intelligent Middleware Based on Neural Network . . . . . . . . . . . . . . . . . . . . . Ana Rhea Pangapalan, Bobby D. Gerardo, Yung-Cheol Byun, Joel T. De Castro, and Francisca D. Osorio

XIII

199

208

220

SMS-Based Automatic Billing System of Household Power Consumption Based on Active Experts Messaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mark Dominic Cabioc, Bobby D. Gerardo, Yung-Cheol Byun

229

Hierarchical Clustering and Association Rule Discovery Process for Efficient Decision Support System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bobby D. Gerardo, Yung-Cheol Byun, and Bartolome Tanguilig III

239

Implementation of Energy Efficient LDPC Code for Wireless Sensor Node . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sang-Min Choi and Byung-Hyun Moon

248

A Multi-layered Routing Protocol for UWSNs Using Super Nodes . . . . . . Abdul Wahid, Dongkyun Kim, and Kyungshik Lim Experimental Measurement for EVM Performance Enhancement of Wireless Repeater System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Daesik Ko and Hwase Park

258

268

Power Model and Analysis of Wireless Transceiver System . . . . . . . . . . . . Jae-Hoon Choi and Heung-Gyoon Ryu

274

Feedback Scheduling for Realtime Task on Xen Virtual Machine . . . . . . . Byung Ki Kim, Kyung Woo Hur, Jae Hyuck Jang, and Young Woong Ko

283

DTAR: Deduplication TAR Scheme for Data Backup System . . . . . . . . . . Sung Woon Kang, Ho Min Jung, Jung Geun Lee, Jin Haeng Cho, and Young Woong Ko

292

Effect of Maximum Node Velocity on GA-Based QOS Routing Protocol (QOSRGA) for Mobile Ad Hoc Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jiwa Abdullah

301

XIV

Table of Contents – Part II

Application of Wireless Accelerometer System for Evaluating Osteoarthritis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dong Rak Kwon and Ho-Cheol Lee

312

A Performance Evaluation of a Novel Clustering Scheme Considering Local Node Density over WSN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jeong-Sam Kim and Tae-Young Byun

320

Performance Analysis of DRAM-SSD and HDD According to the Each Environment on MYSQL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hyun-Ju Song, Young-Hun Lee, and Seung-Kook Cheong

330

Dynamic Channel Adjustable Asynchronous Cognitive Radio MAC Protocol for Wireless Medical Body Area Sensor Networks . . . . . . . . . . . . Byunghwa Lee, Jangkyu Yun, and Kijun Han

338

A Multiple-Metric Routing Scheme for QoS in WMNs Using a System of Active Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jangkyu Yun, Byunghwa Lee, Junhyung Kim, and Kijun Han

346

Implementation of Log Analysis System for Desktop Grids and Its Application to Resource Group-Based Task Scheduling . . . . . . . . . . . . . . . Joon-Min Gil, Mihye Kim, and Ui-Sung Song

354

FM Subcarrier Multiplexing Using Multitone Modulation for Optical Coherent Communications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hae Geun Kim and Ihn-Han Bae

364

An Ontology-Based ADL Recognition Method for Smart Homes . . . . . . . Ihn-Han Bae and Hae Geun Kim

371

Analysis of User Preferences for Menu Composition and Functional Icons of E-Book Readers in a Smartphone Environment . . . . . . . . . . . . . . . Mihye Kim, Joon-Min Gil, and Kwan-Hee Yoo

381

Dynamic Transmission Target Selection Scheme for Load-Balancing in WSN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Seok-Yeol Heo, Wan-Jik Lee, and Won-Yeoul Lee

393

Organizing Virtual Research Groups with Light Path Technology . . . . . . Min-Ki Noh, Won-Hyek Lee, Seung-Hae Kim, and Joon-Min Gil Remote Monitoring Information Management System for Preventing Performance Degradation of Database . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Myung-Ju Kim, Un-Bai Lee, and Kwang Sik Chung Noise Reduction Scheme for Precise Indoor Localization . . . . . . . . . . . . . . Inseok Moon and Won-Kee Hong

403

412 419

Table of Contents – Part II

Development of a Korean Language-Based Augmentative and Alternative Communication Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chang-Geol Kim, Soo-Won Kwak, Ryu Juang Tak, and Byung-Seop Song

XV

429

Adaptive Power Management for Nanoscale SoC Design . . . . . . . . . . . . . . Jeong-Tak Ryu and Kyung Ki Kim

437

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

447

Table of Contents – Part I

Wireless Multimedia Sensor Networks Testbeds and State-of-the-Art Hardware: A Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Muhammad Omer Farooq and Thomas Kunz An Energy-Efficient Cluster-Based Routing in Wireless Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Seongsoo Cho, Bhanu Shrestha, Keuk-Hwan La, Bonghwa Hong, and Jongsup Lee

1

15

Event-Driven Test Script Methodology for SOA System . . . . . . . . . . . . . . . Youngkon Lee

23

Business-Context Based SLA Parameters for SOA Management . . . . . . . . Youngkon Lee

31

Prescription-Level Based Test Assertion Model for SOA . . . . . . . . . . . . . . . Youngkon Lee

39

Multithreaded Power Consumption Scheduler Based on a Genetic Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Junghoon Lee, Gyung-Leen Park, and Hye-Jin Kim Design of a Composite Sensor Node in Agricultural Ubiquitous Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Junghoon Lee, Gyung-Leen Park, Hye-Jin Kim, Ho-Young Kwak, Seongjun Lee, Jong-Heon Lee, Bong-Soo Kang, and Yun-Hyuk Kim Millimetric Waves Technologies: Opportunities and Challenges . . . . . . . . . Jahangir Dadkhah Chimeh and Saeed Bashirzadeh Parapari

47

53

59

A Reduced Complexity Subcarrier Switching Scheme for PAPR Reduction in OFDM System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sabbir Ahmed and Makoto Kawai

67

An Indoor Location-Aware System Based on Rotation Sampling in Positioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chu-Hsing Lin, Jung-Chun Liu, Chien-Hsing Lee, and Tang-Wei Wu

77

Research on the ZigBee-Based Indoor Location Estimation Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chu-Hsing Lin, Jung-Chun Liu, Sheng-Hsing Tsai, and Hung-Yan Lin

82

XVIII

Table of Contents – Part I

Visible Watermarking Based on Multi-parameters Adjustable Gamma Correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chu-Hsing Lin, Chen-Yu Lee, Tzu-Chien Yang, and Shin-Pin Lai

87

A Tree Overlapping-Based Mesh Routing Protocol for Wireless Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Inwhee Joe, Yeonyi Choi, and Dongik Kim

93

Performance Comparison among MIMO Techniques at Different Interference Levels for LTE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mohammad T. Kawser, Md.K. Syfullah, Nawshad U.A. Chowdhury, and Md.T. Hoq Handover Method Considering Power Consumption and Video Quality Satisfaction at the Mobile Node . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hyun Jong Kim and Seong Gon Choi

103

111

Shape Retrieval Combining Interior and Contour Descriptors . . . . . . . . . . Solima Khanam, Seok-Woo Jang, and Woojin Paik

120

Hardware Architecture of Bilateral Filter to Remove Haze . . . . . . . . . . . . . Eun-Kyoung Kim, Jae-Dong Lee, Byungin Moon, and Yong-Hwan Lee

129

An Efficient Interworking Architecture of a Network Processor for Layer 7 Packet Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Kyeong-ryeol Bae, Seung-Ho Ok, Hyeon-Sik Son, Sang Yoon Oh, Yong-Hwan Lee, and Byungin Moon A Rectification Hardware Architecture for an Adaptive Multiple-Baseline Stereo Vision System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hyeon-Sik Son, Kyeong-ryeol Bae, Seung-Ho Ok, Yong-Hwan Lee, and Byungin Moon

136

147

Building Self-organizing Autonomic Agent Based on a Mobile Cell . . . . . Kiwon Yeom

156

Modeling u-Healthcare Frameworks Using Mobile Devices . . . . . . . . . . . . . Haeng-Kon Kim

166

Design and Implementation of Smart Meter Concentration Protocol for AMI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Byung-Seok Park, Cheoul-Shin Kang, and Young-Hun Lee Biologically-Inspired Optimal Video Streaming over Wireless LAN . . . . . Yakubu S. Baguda, Norsheila Fisal, Rozeha A. Rashid, Sharifah K. Yusof, Sharifah H. Syed, and Dahiru S. Shuaibu

179

188

Table of Contents – Part I

Emerging of Mobile Ad-Hoc Networks and New Generation Technology for Best QOS and 5G Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jahangir Khan, Zoran S. Bojkovic, and Muhammad Imran Khan Marwat Intelligent Hybrid Anomaly Network Intrusion Detection System . . . . . . . Heba F. Eid, Ashraf Darwish, Aboul Ella Hassanien, and Tai-hoon Kim Remote Data Acquisition and Touch-Based Control of a Mobile Robot Using a Smart Phone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Yong-Ho Seo, Hyo-Young Jung, Chung-Sub Lee, and Tae-Kyu Yang The Performance Analysis of LT Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ling Yang, ShiLi Song, Wei Wei Su, Yi Fan Wang, and Hong Wen Unseen Visible Watermarking for Gray Level Images Based on Gamma Correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chu-Hsing Lin, Chen-Yu Lee, Shu-Yuan Lu, and Shih-Pei Chien People Counting Using Object Detection and Grid Size Estimation . . . . . Oliver C. Agustin and Byung-Joo Oh The Research of Serially Concatenated FQPSK Demodulation Based on LDPC Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Gao Yuan Zhang, Hong Wen, Liang Zhou, Ling Yang, and Yi Fan Wang

XIX

198

209

219 227

236 244

254

Design and Implementation of Efficient Reed-Solomon Decoder for Intelligent Home Networking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ik Soo Jin

261

A Cost-Effective Multicasting Using an FP-LD Modulator in a WDM-PON . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hyuek Jae Lee

269

A Wireless CCTV Converter Based on Binary CDMA Technology . . . . . . Yeong-Jin Baek and Sang-Hoon Lee A Soft QoS Supporting Multi-path Routing Scheme for Mobile Nodes in MANETs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Tz-Heng Hsu, Yi-Da Li, and Meng-Shu Chiang An Adaptive Query Optimization in a Hierarchical Mediator System . . . Nam Hun Park and Kil Hong Joo

277

283 293

XX

Table of Contents – Part I

On Implementation of a KVM IaaS with Monitoring System on Cloud Environments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chao-Tung Yang, Bo-Han Chen, and Wei-Sheng Chen

300

RFID and Supply Chain Management: Generic and Military Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Tae Hwan Oh, Young B. Choi, and Rajath Chouta

310

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

325

Studies on the Key Technologies of Multi-Platform Mobile Thin Client System: Cross-Layer Isolation and Session Allocation Biao Song, Wei Tang, Tien-Dung Nguyen, Sang-Ho Na, Jun-Hyung Lee, and Eui-Nam Huh Department of Computer Engineering Internet Computing and Network Security Lab KyungHee University Global Campus, South Korea {bsong,wtang,junhyung,johnhuh}@khu.ac.kr, {ntiendung,shna}@icns.khu.ac.kr

Abstract. Virtualization was considered as the best way to isolate independent thin client sessions on a physical machine. However, the hypervisor, guest OS and guest remote server not only consumes a considerable amount of memory but also degrades the processing power of CPU. In this paper, we propose a novel cross-layer isolation technology to support independent user sessions with only one OS and one remote server. Furthermore, a session allocation/migration algorithm is introduced in this paper. The algorithm solves the multi-user to multi-machine allocation/migration problem within thin client environment. Keywords: Thin client, Multi-user, Isolation, Allocation, VM Migration.

1

Introduction

Nowadays, the rapid development of network promotes the investigation of thin client (remote display) technology. Using thin client system, users are able to remotely control other computers (servers) and delegate actual information processing to them. Thus, thin client technology provides a powerful way to break the barrier between diverse applications and insufficient local hardware/software environment. For example, a mobile device with thin client system permits its user to use the applications running on different mobile platforms (Android, iOS, and so on) or PC platforms (Windows, Linux, and so on). In our previous paper [1], we introduced Multi-platform Mobile Thin Client architecture and invented the concept of crossplatform application market running on Cloud server where ubiquitous access to individual applications and PaaS/SaaS are enabled through thin client system. The features of the Cloud server, such as virtualization, flexibility, security, and dynamic management can be fully utilized to support a large number of thin client users. However, the previous architecture cannot achieve mass acceptance until we find practical solutions to the following technical challenges. First of all, one physical machine should be able to support multiple thin client users. VM, as a completely isolated operating system installation within your normal T.-h. Kim et al. (Eds.): FGCN 2011, Part II, CCIS 266, pp. 1–10, 2011. © Springer-Verlag Berlin Heidelberg 2011

2

B. Song et al.

operating system, can be viewed as a feasible way to achieve isolation for multi-user sessions. However, attempting to create VM for every single user session is not always efficient as VM implementation itself consumes considerable amount of memory, and degrades the capacity of CPU. Most of existing thin client systems focused on supporting collaborative multi-user sessions. This focus has resulted in the majority of the work done in this area to be centered on the cursor management and remote display protocol towards heterogeneous client devices. While these approaches have resulted in improvements for collaborative scenarios, they could not provide an effective solution for independent multi-user sessions running on one physical machine. Secondly, the server side should be capable to manage and allocate a multitude of user sessions using its local resources. Although the task placement problem in Cloud environment has been studied in a number of works [4-6], the problem we are facing is distinct from their problem. For one thing, VM is not necessary but alternative carrier for single user session; for another, the metrics of thin client QoS differs from the tradition QoS in Cloud environment. In this context, we first introduce a novel thin client technology to support independent multi-user sessions running on one physical machine. This work is inspired by both THINC system [2] and VNC system [3] where the former one hooks data from server’s device driver layer and the latter one hooks data from server’s hardware frame buffer layer. Unfortunately, both of them are not directly applicable to our scenario. On the one hand, the device driver hooking requires complex translations among different device drivers as well as the hardware support from client device. Thus, using this model in mobile thin client environment may lose the generality since the hardware (especially the graphic component) of mobile client device are highly heterogeneous and usually insufficient to handle various applications. On the other hand, the hardware frame buffer hooking also has a nontrivial drawback in multi-user session environment, which is the failure of audio isolation. The audio information hooked from audio card frame buffer is a mixed signal including all the audio information from multiple user sessions. To address these shortcomings, we adopt a cross-layer approach, which is an effort to provide light weight isolation of remote display, audio playback and input functionalities for multiple user sessions. Comparing with VM isolation, the cross-layer isolation approach consumes less CPU and memory resources. However, the cross-layer isolation approach also has drawbacks. The input interference is one of the problems that may degrade the QoS. For example, when one user is dragging the cursor, other users’ operations have to be paused until the cursor is released by that user. Besides, VM isolation also benefits workload balancing as VM can be migrated without losing running state. Thus, we consider both isolation approaches are alternative, and apply a QoS based selection approach considering the differences in application requirements. Meanwhile, a user session allocation algorithm is proposed. The two goals of session allocation are achieving better resource utilization and reducing service interruption time caused by input interference or VM migration.

Studies on the Key Technologies of Multi-Platform Mobile Thin Client System

3

The rest of this paper is organized as follows. Section 2 discusses some related work in thin client domain. Section 3 gives a brief overview of multi-platform mobile thin client architecture. A description of cross-layer isolation approach is given in section 4. Session allocation algorithm can be found in section 5. We conclude our work in section 6.

2

Related Work

Many collaborative supports have been done in the field of thin client. We found that the main theme among all of these researches was the need of “floor-control management”, which could be used to handle control of a synchronous task. Boyd introduced “fair dragging” where control of a user gains control of the floor once the mouse is dragged [7]. Another well-known study was collaborative VNC [8]. Collaborative VNC was a patch applied to the TightVNC server and client that provides managed collaborative sessions over the RFB protocol. With collaborative VNC, one user has the “floor” (i.e. controls the desktop) at any given time. Other users have the power to take control from or give control to other users at any time. Every user’s cursor is displayed, with each cursor assigned one of several colors. In [9], a multi-user collaborative support named “THINCing Together” was introduced to extend THINC system. The proposal contains a protocol that allows for asynchronous and synchronous multi-user session. It implemented centralized cursor management to optimize bandwidth usage for multiple users. These existing studies provide good solutions for the user input management in an independent multi-user scenario. However, they are not sufficient in terms of remote display and audio support for independent user sessions since they consider same screen and audio output for all users. The remote display protocols can be categorized into three distinctive groups. At application/OS layer, Remote Desktop Protocol (RDP) is a typical protocol developed by Microsoft, which concerns providing a user with a graphical interface to another computer. RDP clients exist for most versions of Microsoft Windows (including Windows Mobile), Linux, Unix, Mac OS X, Android, and other modern operating systems. At device driver layer, THINC uses its virtual device driver to intercept display output in an application and OS agnostic manner [2]. It efficiently translates high-level application display requests to low-level protocol primitives and achieves efficient network usage. At hardware frame buffer layer, VNC uses the RFB protocol to remotely control another computer [3]. Server only sends the graphical screen updates back to the client. In [10], a hybrid protocol was proposed to handle multimedia streaming and interactive gaming applications. Both VNC RFB-protocol and THINC protocol are alternative in this system. Among these protocols, RDP and THINC require graphical hardware support from client device which is hardly provided by mobile thin client device. VNC RFB-protocol, on the other hand, is more general and flexible since it can fully utilize the GPU of server to support mobile clients. Thus, we choose VNC RFB-protocol with non-overlapping window placement to support remote display for independent multi-user sessions.

4

3

B. Song et al.

Multi-platform Mobile Thin Client Architecture

This section introduces the Multi-Platform Mobile Thin Client Architecture. In Fig.1, we present an overview of the architecture. The architecture is composed by several mobile thin clients and a multi-platform thin client server. To receive services, every mobile device must install a thin client viewer which provides all functionalities of remote access. Traditionally, a thin client ran a full operating system for the purposes of connecting to server. A newer trend is called a zero client, which no longer runs a full operating system: the kernel instead merely initializes the network, begins the networking protocol, and handles display of the server's output. Thus, it is not necessary to have an OS on mobile thin client device.

Fig. 1. Multi-Platform Mobile Thin Client Architecture

As Fig.1 shows, the multi-platform thin client server contains a number of modules. At first, the mobile terminal sends a request to authentication module. The request includes the user’s ID, password and terminal information. Then the user selects applications from application store, and provides necessary information to the payment system. The application request is delivered to the task manager. The task manager receives current resource condition and application profiles from the QoS monitor and the internal data repository, respectively. By analyzing the information, the task manager performs task allocation and real-time migration on local resources (only for VM). The QoS monitor continuously monitors the QoS information of running applications as well as the resource condition of local resources. QoS monitor also detects QoS violation, and send reports to task manager for task migration (only for VM). The remote display module, audio support module and remote input module are the three components of thin client protocol. They provide all functionalities of remote access on the server side. Local resources are composed by physical machines, OSs, hypervisors, VM images, and applications. The actual processing power, platform, and services are provided by local resources. The local resources exchange VM image

Studies on the Key Technologies of Multi-Platform Mobile Thin Client System

5

and application data with the internal data repository in order to save or load user sessions. Meanwhile, the application running on local resources may interact with the external data source at any time. The following section will focus on the isolation technology that allows one physical machine to support multiply independent user sessions without using VM.

4

Cross-Layer Isolation Technology

The two issues to be discussed when designing a multi-user thin client system are how to get graphical/audio updates and how to isolate input/output for multiple user sessions. Fig.2 contains a three-layer interception model illustrating the approaches that allow the graphical/audio updates to be retrieved and redirect to the client. This three layer model was first proposed by R.Baratto in his Ph.D. thesis [2]. The original version demonstrates only the display pipeline while we extend the same concept to the audio/input pipeline as well. There exists three interceptions points in the pipeline: (1) between the applications and the device independent layer, (2) between the device independent layer and the device dependent layer, and (3) between the device dependent layer and the hardware layer. To utilize them, the server side must be able handle the application/OS interfaces, the device drivers and the hardware frame buffer, respectively. The requirement for client side also differs from one interception point to another. The interception between the applications and the device independent layer requires client’s OS support and hardware processing capability. Intercepting between the device independent layer and the device dependent layer needs client’s hardware processing capability. Hardware frame buffer interception merely requires the A/V playback functions on client device.

Fig. 2. Cross-layer isolation vs. VM based isolation

Since the gap between the capacity of mobile graphic card and the requirement of PC applications always exists, using the first or the second interception point to support remote display cannot be the proper choice in our scenario. However, the hardware frame buffer interception also has two disadvantages. First, display updates consisting of raw pixels along are typically too bandwidth-intensive. Second, the intercepted signals including audio and video are mixed already, which means we

6

B. Song et al.

need extra isolation technology to support multiple user sessions. The first problem can be solved by optimizing the remote display protocol and the network condition. A successful example is the latest version of VNC, which is able to provide fluent video display under current network circumstance. To solve the second problem, we invent a new isolation approach named cross-layer isolation. Fig.2 also shows the detail information of cross-layer isolation and the well-known VM based isolation. Unlike the VM based isolation in which the hypervisor takes the responsibility to divide and manage the hardware resources, the cross-layer technique has three isolation components deployed between the user sessions and the remote server. The motivation of our design is from the fact that the VM based isolation consumes considerable amount of resources to maintain VM, guest OS and remote server for every single user. Since the modern OSs have the capability to manage several processes running concurrently, the QoS of user sessions can be guaranteed by OS as long as the server has sufficient resources to handle the tasks. The reason we choose the name “cross-layer” is that the audio isolation, the video isolation and the input isolation are implemented in different layers. To isolate the graphical output of one user session from others, the server assigns a non-overlapping rectangle area for each user session. While hooking the whole screen information from the hardware frame buffer, the server can easily extract any user session from the picture by using the coordinates of the corresponding rectangle. Then using RFB protocol, the updates are distributed to the users continuously. The audio isolation utilizes device driver interception. Before the audio signals from multiple user sessions are mixed, the server hooks them and sends to the corresponding clients. For one thing, it is very hard to extract one session’s audio information from a mixed audio signal retrieved from hardware frame buffer; for the other, the mobile clients are able to provide the hardware processing capability for audio. The RTP (Real-time Transport Protocol) is adopted as the protocol for audio transmission. Unlike video and audio isolation, the input isolation intercepts on the client side and performs the actual input on the server side. The input isolation on server side needs the APIs of server OS. Taking our implementation on Microsoft Window XP as an example, the input isolation module requires the handle of each user session. The user inputs are managed by a multi-queue system. When the server wants to perform a user input, it first activates the corresponding session using the handle, and then simulates the mouse or keyboard event. We adopt Shortest-Remaining-Size-First (SRSF) preemptive scheduler known to be optimal for minimizing mean response time and improving the interactivity of a system [11]. In terms of resource consumption, the cross-layer isolation approach is more efficient than the VM-based isolation approach; nevertheless it can be only an alternative solution since it has two disadvantages. First, input interference may occur. Second, session migration is not possible. In next section, we will discuss about these problems and give a primal solution to choose the isolation technology and to allocate user sessions.

Studies on the Key Technologies of Multi-Platform Mobile Thin Client System

5

User Session Management

5.1

Isolation Selection Approach

7

As we introduced in section four, all the user sessions supported by cross layer isolation are running on same OS. While one session gains the control of mouse/keyboard, the inputs of other sessions cannot be performed until the control is released. Since the control shift among these sessions should be made in a quick and implicit manner, we need to figure out which operation may hinder others from getting the control. Any mouse/keyboard operation can be viewed as one of the following two events: instant event and continuous event. Instant events usually do not interfere with each other whereas the total amount of instant events cannot be larger than that can be handled by the input isolation component. Continuous events, such as mouse dragging, greatly interfere with instant events and themselves. Therefore, we need to profile the averaging number of instant event and the cumulative time of continuous event for each application. Let NI j be the number of instant events generated by the user of application j within one minute,

TC j be the cumulative time of continuous events generated by

the user of application

j within one minute. If the value of NI j or TC j highly

depends on user’s behavior, we need to create profile for each user. Let MAX ( NI ) be the maximum number of instant events that can be handled by the input isolation component within one minute, MAX (TC ) be the maximum time that can be occupied by continuous events within one minute. The value of MAX ( NI ) depends on the hardware and software environment. The value of

MAX (TC ) should guarantee that no user feels the existence of other user. We define the first part of selection approach as follow: given a user session is application j , if NI j > MAX ( NI ) × 50% or TC j > MAX (TC ) × 50% , the cross layer isolation cannot be applied to this session. Using

MAX ( NI ) × 50% and

MAX (TC ) × 50% rather than higher values leaves enough space for allocating other sessions supported by cross-layer isolation. The CPU consumption is also considered as an important factor. Since the crosslayer isolation technology does not support session migration, it is not suitable for the sessions whose CPU consumption fluctuates all the time. Let t be the duration between the time when the CPU is overloaded and the time when QoS monitor triggers a migration. Given an application j , the profiling of its CPU consumption takes time n × t where n should be large enough to present the averaging CPU consumption of application j . By that, we can get n averaging CPU consumption information

{ AC j1 , AC j 2 ,..., AC jn } and an overall averaging CPU consumption

8

B. Song et al.

information

AC j . Let FC be the free CPU capacity to buffer the unexpected

workload burst. We define the second part of selection approach as follow: given a user session is application j , if max{ AC jk | (1 < k < n)} − AC j > FC × 50% , the cross layer isolation cannot be applied to this session. The reason of using FC × 50% is also to facilitate session allocation. 5.2

Session Allocation Approach

Given an application

i and a physical machine j , let cij , mij , gij and bij be the

percentages of resource usage regarding CPU, memory, GPU and network bandwidth, respectively. If the application is supported by VM, the extra resource consumption of VM, OS and remote server should be also included. Let fc j , fm j , fg j and fb j be the percentages of idle CPU, memory, GPU and network bandwidth resources, respectively, on machine j . For any resource, at least 25% free capacity should be kept to buffer the unexpected workload burst, and should not be counted in the available idle resources. The application can be allocated to that physical machine only if the following condition can be fulfilled:

cij Pload then 11.C(t) Eq.(6) 12. PLP 0 13. end 14. if Ptot < Pload then 11. C(t) Eq. (7) 12. PLP Eq.(10) 13. end Fig. 1. Optimal design algorithm

10 9 8 7 6 5 4 3 2 1 0 1

6

11

16

21

26

31

36

Fig. 2. Wind data

41

46

51

56

61

66

196

K.H. Park et al.

Fig. 3. Current data

1200 (j/M2) 1000 800 600 400 200 0 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 (t) Fig. 4. Irradiance data Table 1. Design Conditions

Cost Type

Unit size

Total size (Won/kWh)

Wind

5W

To be designed

105

PV

30 W

//

640

Tide

5W

//

201

Battery

5 Wh

//

320

Wind

5W

20 Wh

-

Design of Optimal Combination for New and Renewable Hybrid Generation System

197

Table 2. Results of optimal combinations Size

cost

remarks

combination wind (W) PV(W)

battery(Wh)

(won)

-

145

95.3

-

tide(W)

Wind-PV

100

60

Wind-Tide

5

-

270

135

100.43

-

PV-Tide

-

30

105

95

71.65

-

Wind-PV-Tide

40

30

45

80

58.45

optimal

5

Conclusion

In this paper, method for design of optimal combination for hybrid renewable generation system has been studied. A simple mathematical modeling for typical new and renewable energy resources such as wind, PV, tidal energy resources were developed. Using the models, the method of designing the optimal combination of the hybrid system was developed. The method is based on RLP (Ratio of Lack of Power) and economical model. This aims at finding the configuration, among a set of system components that meets the desired system requirements, with the lowest value of the energy cost. The method was applied to hybrid wind-PV-tide generation system. The availability of the methodology was successfully demonstrated with the field data acquired from sets of experiments. Acknowledgement. Following are results of a study on the "Human Resource Development Center for Economic Region Leading Industry" Project, supported by the Ministry of Education, Science & Tehnology(MEST) and the National Research Foundation of Korea(NRF).

References 1. Kellogg, W., Nehrir, M.H., Venkataramanan, G., Gerez, V.: Optimal unit sizing for a hybrid PV/wind generating system. Electric Power System Research 39, 35–38 (1996) 2. Chedid, R., Saliba, Y.: Optimization and control of autonomous renewable energy systems. Int. J. Energy Res. 20, 609–624 (1996) 3. Karaki, S.H., Chedid, R.B., Ramadan, R.: Probabilistic performance assessment of autonomous solar-wind energy conversion systems. IEEE Trans. Energy Conv. 14, 766–772 (1999) 4. Bagul, A.D., Salameh, Z.M., Borowy, B.: Sizing of stand-alone hybrid PV/wind system using a three-event probabilistic density approximation. Solar Energy 56, 323–335 (1996) 5. Musgrove, A.R.D.: The optimization of hybrid energy conversion system using the dynamic programming model – RAPSODY. Int. J. Energy Res. 12, 447–457 (1988)

198

K.H. Park et al.

6. Yokoyama, R., Ito, K., Yuasa, Y.: Multi-objective optimal unit sizing of hybrid power generation systems utilizing PV and wind energy. J. Solar Energy Eng. 116, 167–173 (1994) 7. Yang, H.X., Burnett, J., Lu, L.: Weather data and probability analysis of hybrid photovoltaic wind power generation systems in Hong Kong. Renewable Energy 28, 1813–1824 (2003) 8. Beyer, H.G., Langer, C.: A method for the identification of configurations of PV/wind hybrid systems for the reliable supply of small loads. Solar Energy 57, 381–391 (1996) 9. Protogeropoulos, C., Brinkworth, B.J., Marshall, R.H.: Sizing and techno-economical optimization for hybrid solar PV/wind power systems with battery storage. Int. J. Energy Res. 21, 465–479 (1997) 10. Diaf, S., Diaf, D., Belhamel, M., Haddadi, M., Louche, A.: A methodlogy for optimal sizing of autonomous hybrid PV/wind system. Energy Policy 35, 5708–5718 (2007) 11. Nelson, D.B., Nehrir, M.H., Wang, C.: Unit Sizing of Stand Alone Hybrid Wind /PV/Fuel Cell Power Generation Systems. IEEE Power Engineering Society General Meeting 3, 2116–2122 (2005)

Parameter Optimization of UWB Short Range Radar Detector for Velocity Measurement in Automobile Applications Purushothaman Surendran1, Chul-Ung Kang1, and Seok Jun Ko2,* 1

The authors are with Jeju National University 2 Jeju National University Jeju, Korea [email protected]

Abstract. For designing UWB-SRR detectors we must understand the required design parameters and specific algorithm for target detection. In this paper, we optimize the parameters such as the number of coherent integration, pulse repetition interval, Doppler frequency resolution and FFT measurement time of UWB Short Range Radars for measuring the relative velocity of the target in automobile applications. The proposed detector with optimized parameters can measure the target minimum relative velocity of about 6.99 km/hr in very short measurement time. The minimum number of FFT points required to process Doppler Frequency by Fast Fourier Transform (FFT) is 32 points. The detection is based on one transmitted pulse against a background of white Gaussian noise (WGN). The performance of the proposed detector with optimized parameters is analyzed and simulation has been done in order to verify. Keywords: FFT, Velocity resolution, Coherent integration, UWB-SRR Detector, Doppler Frequency, Pulse Repetition Interval.

1

Introduction

In automobile industries, the vehicle safety has improved in the last decades with the increase in new safety technologies. The 24 GHz Short Range Radars are mainly used to give information such as range and velocity of the target to the driver and in some cases they are connected to a computer that performs some guiding actions to reduce collision and minimize the injuries. The source for target detection is the radar signals reflected by the target that is a mixture of noise and varied signals. The designed system must provide the optimal technique to obtain the desired target detections and to measure the relative velocity between the radar and target. The preferred detection can be determined by using specific algorithm for measuring the energy of the received signals. For this reason radar systems in 24 GHz have good performance in range and velocity measurement therefore they can be applied in different automobile applications like parking-aid, pre-crash detection [1]. *

Corresponding author.

T.-h. Kim et al. (Eds.): FGCN 2011, Part II, CCIS 266 , pp. 199–207, 2011. © Springer-Verlag Berlin Heidelberg 2011

200

P. Surendran, C.-U. Kang, and S.J. Ko

Previous works [4]-[5] has been focused on influence of increasing range resolution on detection ability of targets and also designed the algorithm with Fast Fourier Transform (FFT) method mainly. In this paper, we present the parameters optimization such as the number of coherent integration, the number of FFT points and the velocity resolution by using very narrow pulse width such as UWB signal. Our paper is made on the assumption of single moving target and shows the results by using the Monte Carlo simulation. The organization of this paper is as follows. In Section 2, the system model is described. In Section 3, we propose the detector for velocity measure. In Section 4, the parameter optimization method is expressed. In Section 5, simulation result is shown. Finally, conclusion is presented in Section 6.

2

System Description

The block diagram of a UWB radar system as shown in figure 1 is split in to two parts, the transmitter part and the receiver part.

Fig. 1. Block diagram of UWB radar system

In the transmitter part, the pulses are initiated by the Pulse Repetition Frequency (PRF) generator which triggers the pulse generator which in turn generates Gaussian pulses with sub-nano second duration as shown in figure 2. The Pulse Repetition Interval (PRI) is controlled by the maximum range of the radar. The maximum range for unambiguous range depends on the pulse repetition frequency and can be written as follows Rmax =

c 2 ⋅ f PRF

(1)

Parameter Optimization of UWB Short Range Radar Detector

201

where fPRF is the pulse repetition frequency and c is the velocity of light. And the range resolution can be written as ΔR =

c ⋅ TP 2

(2)

where Tp is the pulse width and c is the velocity of light. Then the transmitted signal can be written as follows s (t ) = AT ⋅ cos( 2πf c t + ϕ 0 ) ⋅

+∞

 p (t − n ⋅ T

PRI

)

(3)

n = −∞

where p(t) is the Gaussian pulse as follows   t  p (t ) = exp − 2π   τ    p 

2

   

(4)

And the parameters employed in this UWB radar system are described as follows; AT is the amplitude of single transmit pulse, φ0 is the phase of the transmit signal, fc is the carrier frequency, and TPRI is the pulse repetition time.

Fig. 2. Transmitted signal and received baseband signal

Since the range resolution of the UWB radar system is much less than the extent of the target, the echo signal is the summation of the time-spaced echoes from the individual scattering centers that constitute the target [3]. Therefore, in this paper, we can assume that the target has L independent reflecting cells. The target model is written as L −1

h(t ) =  α l ⋅ δ (t − τ l )

(5)

l =0

where the number of scatters L, the amplitude of the scatters αl, and the time delays of the scatters τl are all unknown. If the target is moving with relative velocity ν [km/hr]

202

P. Surendran, C.-U. Kang, and S.J. Ko

between the radar and the target, then the baseband complex received signal r(t) is written as r (t ) = AT

+∞ L −1

 α

l

⋅ e jθl p (t − nTPRI − τ l ) + n(t )

(6)

n = −∞ l =0

where n(t) is the complex additive white Gaussian noise (AWGN) with two-sided power spectral density N0/2 and θl is the arbitrary phase of l-th scatter that can be written as θl = -2πfcτl+ϕ0. The sampling rate is same to the pulse width and the wavelength λ is c/fc and c is the velocity of light, the Doppler shift is denoted as ωd = ±4πν /λ = ±4πνfc /c. In the Doppler shift, the positive sign (+) indicates the closing target and the negative sign (-) means the receding target.

3

Proposed Detector

In this section, we propose the algorithm for measuring the relative velocity of the target. First, as shown in Figure 3, we can detect the relative velocity of the target by performing DFT. Typically DFTs are performed with Fast Fourier Transform (FFT). The proposed detector consists of coherent integration, Discrete Fourier Transform (DFT) algorithm and square law detector. The sampling frequency is set depending on the pulse width, the baseband received signal is sampled in an in phase (I) and quadrature (Q) channel at every Tp, and the sampling rate Tp is same to the pulse width of 2ns and the range resolution can be 30cm from (2) and also it is assumed that the maximum target range can be 20m by using (1). From the above mentioned range resolution and maximum target range, the range gates of atleast 67 are required to detect the target range and velocity. Therefore the range gates are equal to the number of memory in the coherent integration. The sampled valued at every TP is applied to switch I and Q of the coherent integration. The switch I is shifted at every Tp sample, i.e., the range gate. It takes N·TPRI to coherently integrate and dump for all range gates. The coherent integration for the i-th range gate in I branch can be expressed as follows X I (i ) =

1 Nc

Nc

 Re {r n =1

n

( iT P )}

(7)

where L−1

rn (iTP ) = AT αl ⋅ e jθl p((nTPRI + iTP ) − nTPRI −τ l ) + n' (nTPRI + iTP )

(8)

l =0

The DFT formed for a range cell provides a direct measure of the Doppler frequency. The result for one complete measurement cycle is a matrix of range gates and Doppler frequencies. Detection statistics (such as square law detection statistic) are performed for DFT outputs according to the prescribed detection strategy.

Parameter Optimization of UWB Short Range Radar Detector

203

Fig. 3. Block Diagram of the proposed detector

The DFT algorithm performed by FFT operates at every N·TPRI and the 32 point output of the DFT algorithm is a complex output which is squared and added can be represented as Y (i ) =

4

1 N

 {X N

n =1

I

(i ) e − jΩ n

} + {X 2

Q

(i ) e − jΩ n

}

2

(9)

Parameter Optimization Method

The detector of UWB radar must determine that a signal of interest is present or absent, and then the UWB radar processes it for some useful purpose such as range determination and velocity measure. In this paper, we propose the importance of optimizing various design parameters such as the number of coherent integration, sampling time, velocity resolution, Doppler frequency resolution, FFT measurement time and the required FFT length. The most important part of radar receiver is to understand how the ultra-short pulses are sampled in resolution (range-Doppler) cell. The required sampling frequency depends on the Doppler frequency range which corresponds to the relative velocity between the detector and the target. The minimum Doppler frequency resolution that can be achieved is given by

204

P. Surendran, C.-U. Kang, and S.J. Ko

Δf D = (1 / TFFT ) / N

(10)

where TFFT is the sampling time of the FFT input which is the product of number of coherent integration (Nc) and pulse repetition interval (TPRI). And N is the number of FFT points TFFT = N c × TPRI

(11)

The relation between Doppler frequency and relative velocity is fD = −

2vf c c

(12)

where v is the relative velocity and fc is the transmitter frequency. Thus the velocity resolution that can be achieved from the minimum Doppler frequency is Δv = −

Δf D ⋅ c 2 ⋅ fc

(13)

The required FFT measurement time also depends on the velocity resolution as shown in the figure below, For a FFT length of 32 points the FFT measurement time is given by TFFT measure = N × TFFT

(14)

If the number of FFT points increases then the measurement time also increases correspondingly.

5

Computer Simulation Result

The purpose of the simulation is to find the relative velocity between the UWB detector and the target. In the simulation we assume that the total energy reflected from the target is 1 and the system parameters are used as mentioned in table I. The signal-to-noise ratio (SNR) is defined as Ē/N0, where Ē is the total average energy reflected from the target. Fig. 6 shows the simulation result of a single moving target in frequency domain against a background of additive white Gaussian noise (AWGN). Fig. 7 shows the variance (dB) versus the Signal-to-noise ratio (Ē/N0) in time domain and frequency domain. A large enough number of trials are performed to obtain the variance at various Ē/N0. The number of trials is about 1000000 times. We can predict that the variance (which is proportional to noise power of AWGN) in frequency domain is less than the variance in the time domain, therefore the performance of the detector will be superior in frequency domain than compared to time domain.

Parameter Optimization of UWB Short Range Radar Detector 650

FREQUENCY RESOLUTION (Hz)

600

COHERENT INTEGRATION = 200

550 500 450 400 350 300 250 200 150 100 50 0 0

16

32

48

64 FFT POINTS

80

96

112

128

14

16

Fig. 4. FFT points Vs Doppler frequency resolution

MEASUREMENT TIME (ms)

250

200

150

100

50

0 0

2

4

6 8 10 12 VELOCITY RESOLUTION (Km/Hr)

Fig. 5. Velocity resolution Vs measurement time Table 1. System Parameters Parameters

Notation

Value

Pulse Repetition Interval

TPRI

500ns

Pulse Width

Tp

2ns

Maximum Target Range

Rmax

20m

Range Resolution

ΔR

30cm

Number of Coherent Integration

Nc

200

Velocity Resolution

Δv

6.9948 km/hr

Number of FFT points

N

32

205

206

P. Surendran, C.-U. Kang, and S.J. Ko

Fig. 6. 3D Graph of Detector output

1 VARIANCE_TIME_DOMAIN

0.1

VARIANCE_FREQUENCY_DOMAIN

VARIANCE (dB)

0.01

0.001

0.0001

0.00001

0.000001 -15

-12

-9

-6

-3

0

3

Ē/N0

Fig. 7. Ē/N0 Vs Variance

6

9

12

15

Parameter Optimization of UWB Short Range Radar Detector

6

207

Conclusion

In this paper, we have optimized the design parameters such as coherent integration, pulse repetition interval, target velocity resolution and the FFT measurement time of Ultra Wide Band Short Range Radar (UWB-SRR) detectors in Automobile applications. Thus the optimized parameters enhance the performance of the proposed detector, so that the target relative velocity of up to 6.9948 km/hr can be detected in very short FFT measurement time of about 3.2ms by using minimum FFT points of 32 efficiently. Finally, we show that the noise power is reduced in frequency domain when compared with the time domain thus increases the detection probability of the detector. Therefore the proposed detector is memory and time efficient detector for automobile applications which is the critical problem in conventional FFT method. Acknowledgments. This work was supported by the Korean Ministry of education & Science Technology, 331-2007-1-D00265.

References [1] Strohm, K.M., et al.: Development of Future Short Range Radar Technology. In: Radar Conference (2005) [2] Taylor, J.D., et al.: Ultra-Wideband Radars Technology: Main Features Ultra-Wideband (UWB) Radars and differences from common Narrowband Radars. CRC Press (2001) [3] Taylor, J.D., et al.: Introduction to Ultra-Wideband (UWB) Radars systems, Boca Raton, FL (1995) [4] Klitz, M.: An Automotive Short Range High Resolution Pulse Radar Network. Ph. D. Dissertation (January 2002) [5] Minkler, G., Minkler, J.: The Principles of Automatic Radar Detection In Clutter: CFAR. Magellan Book Company (1990) [6] Proakis, J.G.: Digital Communications. McGraw-Hill (2001) [7] Surendran, P., Ko, S.J., Kim, S.-D., Lee, J.-H.: A Novel Detection Algorithm for Ultra Wide Band Short Range Radar in Automobile Application. In: IEEE VTC 2010. Springer, Heidelberg (2010) [8] Skolnik, M.I.: Introduction to radar systems, 3rd edn. McGraw-Hill (2001) [9] Oppermann, I., Hamalainen, M., Iinatti, J.: UWB Theory and Applications. John Wiley & Sons Ltd (2004)

Data Signature-Based Time Series Traffic Analysis on Coarse-Grained NLEX Density Data Set Reynaldo G. Maravilla Jr., Elise A. Tabanda, Jasmine A. Malinao, and Henry N. Adorna Department of Computer Science (Algorithms and Complexity Lab) University of the Philippines, Diliman, Quezon City 1101, Philippines {jamalinao,hnadorna}@dcs.upd.edu.ph

Abstract. In this study, we characterize traffic density modeled from coarse data by using data signatures to effectively and efficiently represent traffic flow behavior. Using the 2006 North Luzon Expressway Balintawak-North Bound (NLEX Blk-NB) hourly traffic volume and time mean speed data sets provided by the National Center for Transportation Studies (NCTS), we generate hourly traffic density data set. Each point in the data was represented by a 4D data signature where cluster models and 2D visualizations were formulated and varying traffic density behaviors were identified, i.e. high and low traffic congestions, outliers, etc. Best-fit curves, confidence bands and ellipses were generated in the visualizations for additional cluster information. We ascertain probable causes of the behaviors to provide insights for better traffic management in the expressway. Finally, from a finer-grained 6-minute interval NLEX Blk-NB density data set, the coarser-grained hourly density data set were validated for consistency and correctness of results. Keywords: Data Signatures, Traffic Density Analysis, North Luzon Expressway, Non-Metric Multidimensional Scaling.

1

Introduction

Previous traffic behavior studies dealt only with volume analysis[2]. If we are to consider congestion, density is an accurate indicator. Density considers the occupied space of the road and the speed of the vehicles and it can give a better estimate of the real behavior of the traffic flow. Expressways, most of the time, should exhibit very low densities and high speeds, but spikes in the density graphs of 2006 North Luzon Expressway Balintawak - North Bound (NLEX BlkNB) segment data show otherwise. Domain experts identified inconsistencies and pointed out that the outliers determined are unrealistic for expressways. The study aims to show that the proposed density model is effective in estimating the traffic behavior of NLEX, with emphasis on u being the space mean speed and not time mean speed. The data set recorded and provided by National Center for Transportation Studies (NCTS) is in time mean speed. The data set, T.-h. Kim et al. (Eds.): FGCN 2011, Part II, CCIS 266, pp. 208–219, 2011. c Springer-Verlag Berlin Heidelberg 2011 

Data Signature-Based Time Series Traffic Analysis

209

therefore, will be preprocessed to produce and represent realistic characterizations of traffic flow in NLEX. With the densities produced by the model from a study recently conducted[1], we will analyze the traffic flow by building a model for hourly traffic space mean speed and volume in NLEX. Data signatures will then be produced to represent the hourly traffic density data points. These data signatures and the time-domain data set cluster model will be visualized using the Non-Metric Multidimensional Scaling[11] and Data Images, respectively. Then, the intercluster and intracluster relationships of these data points will be examined. Data set outliers and potential outliers will be identified and analyzed using the methods in [2,8]. We validate our results in the hourly density data with the 6-minute data set. Section 1.1 discusses the definitions and basic notations used in this paper. Section 2 shows the the step by step building of the density-based model for the NLEX Blk-NB segment. This section also shows how the density data set is represented as a data signature and further visualized using nMDS, and Data Image. The results of these steps are discussed in Section 3. Finally, the conclusions and recommendations for further studies are in Section 4. 1.1

Definitions

The Data Sets. The data set provided by NCTS in this study on the NLEX Blk-NB segment in the year 2006 is periodic. The first data set contains hourly time mean speed and mean volume of the said segment. The second data set contains 6-minute time mean speed and mean volume of the said segment. The data set is preprocessed in a previous study in which average time mean speeds must meet the minimum speed requirement of 40 kph. Eleven weeks are eliminated for the first data set, leaving us with 41 weeks at 168 hours each. To maintain consistency, eleven weeks are also eliminated for the second data set, leaving us with 41 weeks at 1680 hours each. Traffic Flow 1. Volume q. Volume is the hourly mean of the number of vehicles per lane. 2. Time Mean Speed ut . Time mean speed is the mean of the speeds ui of the n vehicles  npassing through a specific point within a given interval of ui time, i.e. ut = i=1 . n 3. Space Mean Speed us . Space mean speed is the speed based on the average travel time of n vehicles in the stream within a given section of road, i.e. us =  nn ui . i=1 4. Density k. Density k is the number of vehicles over a certain length of a road. It is estimated as in k = uqs . Space mean speed is used in estimating the density because it considers the space between the vehicles.

210

R.G. Maravilla Jr. et al.

5. Estimation of Space Mean Speed from Time Mean Speed. Since the data set provided contains only the time mean speed and space mean speed is required in determining density, we estimate the space mean speed from the time mean speed using Rakha-Wang equation[4] to get us , where σ2 us ≈ ut − utt There will be a 0 to 1 percent margin of error in the estimation. Data Signature. A data signature, as defined in [6] is a mathematical data vector that captures the essence of a large data set in a small fraction of its original size. These signatures allow us to conduct analysis in a higher level of abstraction and yet still reflect the intended results as if we are using the original data. Various Power Spectrum-based data signatures[7,8] had been employed to generate cluster and visualization models to represent periodic time series data. Fourier descriptors such as Power Spectrums rely on the fact that any signal can be decomposed into a series of frequency components via Fourier Transforms. By treating each nD weekly partitions in the NLEX BLK-NB time-series traffic volume data set[8] as discrete signals, we can obtain their Power Spectrums through the Discrete Fourier Transform(DFT) decomposition. Power Spectrum is the distribution of power values as a function of frequency. For every frequency component, power can be measured by summing the squares of the coefficients ak and bk of the corresponding sine-cosine pair of the decomposition and then getting its square root, where k = 0, 1, . . . , n − 1. The Power Spectrum Ak of the signal, k = 0, 1, . . . , n − 1 is given by, Ak = ak 2 + bk 2 . Studies have shown that the set {A0 , A7 , A14 , A21 } is an optimal data signature for both visualization[7] and clustering [9]. Methods in [9] validate the optimality of the 4D data signature by showing an improved Dunn-like index. The 4D data signature used for clustering achieved statistical competence among all other data signatures. The study achieved ≈ 97.6% original data reduction for production of an optimal cluster model for Dunn-like variables. Data Visualization. In this study, we incorporate two methods, namely NonMetric Multidimensional Scaling and Data Images to analyze the data set. The first method projects the 4D signatures into a simpler 2D visualizations for traffic analysis. The second one is used to present the time-domain traffic density data. 1. Non-Metric Multidimensional Scaling. Non-Metric Multidimensional Scaling (nMDS)[11] is another visualization technique that maps multidimensional data set onto a 2D space. It computes the dissimilarity of the data points using Euclidean distance, Correlation, Mahalanobis, and other distance measures discussed in the literature[11]. nMDS includes a minimization of the stress or loss function to determine an optimal projection of the points in the Euclidean space given the known relationships in the higher dimension.

Data Signature-Based Time Series Traffic Analysis

211

2. Data Image. Data image is a graphical representation that transforms the given multidimensional data set into a color range image. Observations are made through the colors’ given characteristics and respective magnitudes. In our given data set, weeks are represented by the y-axis arranged by their cluster membership and days by the x-axis (with 1 as Sunday, 2 as Monday, and so on). The weeks are arranged according to their clusters. Clusters are determined by using the X-Means Clustering algorithm[5] that takes the 4D data signatures of the weeks in the data set as its input. 3. Confidence Intervals. Confidence Interval is a statistical analysis on given data sets to determine the reliability of the estimated population parameter. – Best Fit Curve. Linear, quadratic, cubic, quartic, and quintic curves are generated and fitted to their respective clusters. Root Mean Squared Deviation (RMSD) formula is applied to the curves to determine the best fit curve. RMSD gets the difference of the actual nMDS y values observed and the predicted y values (ˆ y ) of the curve model. The one with the lowest value will then be the best fit curve. – Confidence Bands. From the constructed best fit curve, the confidence √  SS band is extended above and below the curve by c DF tα (DF ), where c = G|x × Σ × G |x, G|x is the gradient vector of the parameters at a particular value of x, G |x is the transposed gradient vector, Σ is the variance-covariance matrix, SS is the sum of squares for the fit, DF is the degrees of freedom, and tα (DF ) is the student’s t critical value based on the confidence level α and the degrees of freedom DF . – Confidence Ellipse. A confidence ellipse, as defined in [12], uses intervals for both X and Y values of the scatterplot. The interval is projected horizontally and vertically respectively. The confidence ellipse is formed using the equation Z ± R × I, where Z is the mean of either X or Y , R is the range of either X or Y , and I is the confidence level 1 − α. 4. Potential Outliers. Potential outliers, as previously defined in [8,12] are points projected “near”or at the periphery of a region occupied by its cluster in the 2D visualization. – Absolute Potential Outliers. An absolute potential outlier is a data point that lies outside the confidence band and ellipse of its respective cluster. This point is not represented by its cluster’s best fit curve. – Valid Potential Outliers. A valid potential outlier is a data point that lies outside the confidence ellipse, but lies within the confidence band of its cluster and is still represented by the best fit curve. – Ambiguous Potential Outliers. An ambiguous potential outlier is a data point that is bounded by either two confidence bands or two confidence ellipses of different clusters or is inside a confidence ellipse, but outside of its cluster’s confidence band.

212

2 2.1

R.G. Maravilla Jr. et al.

Methodology Building an Effective Model from Sparse Data Points

1. From the preprocessed data set, we extract the mean volume and time mean speed per hour. 2. We estimate the space mean speed from the time mean speed by first getting the variances among the time mean speeds of the four lanes. We apply the Rakha-Wang equation to get the space mean speed per hour of the segment. To maintain consistency, the computed space mean speeds undergo preprocessing to eliminate values that are below 40 kph. 3. We estimate the density k using the given mean volume and space mean speed per hour. 4. To validate the produced hourly density data, steps 1-3 for modeling hourly traffic density are also conducted for modeling 6-minute traffic density. 2.2

Data Signature-Based Cluster and Visualization Models of the NLEX Traffic Density Model

1. Given the hourly density data generated from the previous section, the values of each week is transformed from its time domain to its frequency domain representation through the Discrete Fourier Transform and generate its 168D (and 1680D) Power Spectrum values for the hourly (and 6-minute interval) data set. Then, a 4D data signature is constructed from the Power Spectrum values of each week consisting of the components A0 , A7 , A14 , A21 . 2. Using all the data signatures of the weeks in the density data set as input to the X-Means clustering algorithm[5], we build the data set’s cluster model to identify groups of weeks that may have high, regular, and low traffic density (i.e. congestion) and pinpoint outliers and potential outliers in the model. We also pinpoint time frames where these various traffic behavior are identified. The traffic density analysis will be presented to the domain experts for their assessment. With the resulting assessment, we will provide suggestions to traffic control management for business-related decisions. 3. Produce the Non-Metric Multidimensional Scaling 2D visualization using the 4D data signature representations of each week in the data set, incorporating the results of X-Means clustering algorithms by coloring the 2D projections of the data signatures with respect to the assigned color information of their cluster. 4. Generate confidence bands, confidence ellipse, and best fit curve at 90% confidence interval per cluster to determine its set of potential outliers. 5. Visualize the traffic density values of the time domain data set where rows represent the values of each week, structured contiguously based on the clustering result, and each pixel is colored based on the actual values of the density in a time slot. Darkened lines separate the clusters and outliers from one another.

Data Signature-Based Time Series Traffic Analysis

213

6. We also perform the data signature visualization of the 6-minute density data set, producing this data set’s own set of nMDS scatterplot and Data Image. We then compare its clusters’ data points and Data Image with the hourly density data set’s. Using this data set’s nMDS scatterplot, we also construct confidence bands, confidence ellipse, and best fit curve per cluster to determine its own set of potential outliers. We compare the produced outliers of the 6-minute density data set with those of the hourly density data set. After validating the hourly density data set with the 6-minute data set, we perform analysis on this data set’s potential outliers to find their enclosing cluster’s and its elements’ behavior.

3 3.1

Results and Discussions Density Graphs of the Data Sets

From the preprocessed data, we computed the variances of the hourly time mean speed. The variances are consistent, but there are relatively high values of variance are found. It is because other lanes are congested during a specific hour, therefore, time mean speed variation is evident. From the computed variances, the hourly space mean speeds of the segment are produced. Spikes from processed space mean speeds are still observed, but they are relatively shorter than the spikes produced in the raw space mean speed graph. With the new set of space mean speeds (from 6888 hours, we now have 6880 hours), consistency in the density graph is expected. The calculated hourly and 6-minute densities from the mean volume and processed space mean speeds are shown in Figure 1(a) and 1(b), respectively. The graphs shows consistent values of density except on spikes where traffic incidents could have happened. The graphs show similar behavior of traffic density. Consistency in this matter shows us that the produced hourly density model is precise. To further validate the hourly density model, we then perform data visualization techniques in comparing the two data sets.

Fig. 1. Hourly Densities of the Segment

214

3.2

R.G. Maravilla Jr. et al.

Data Signature-Based Cluster Models

Hourly Data Set Visualization. The nMDS visualizations of the data signatures from the hourly traffic density data set is shown in Figure 2, respectively. The points which are reflected with the same symbols and coloring belong to the same cluster. The shown cluster model was generated using the data signatures of the hourly traffic density data set. Using the nMDS visualization, curves are fitted to each cluster’s data points. The best fit curves determined by RMSD are found to be linear. No curve was fitted for Cluster 3 due to the lack of data necessary in constructing the curve. The best fit curves are used to determine the confidence bands and confidence ellipse of each cluster. With the resulting best fit curves, confidence ellipses, and confidence bands, outliers and potential outliers are examined. The resulting visualizations with each cluster’s best fit curves, confidence bands and ellipse is shown in Figure 2.

Fig. 2. Data Signature-based nMDS Visualization of the Hourly Traffic Density Data

Cluster 4’s ellipse is not considered because it only has a few points and it covers all points of Cluster 2, making all these points ambiguous. Cluster 4 of the hourly density data set contains no potential outliers because it is relatively far from the points of other cluster, preventing them from being covered by other clusters’ ellipses. Cluster 3 is an outlier of the hourly density data set because it is the only has one data point. All potential outliers are found to be ambiguous since all points are covered by their own confidence ellipse. The details and evidences of the points being potential outliers in their clusters is found in the appendix.

Data Signature-Based Time Series Traffic Analysis

215

Table 1. Potential Outliers of Hourly Density Data Set’s Clusters Cluster Ambiguous Potential Outliers 0 Wk1, Wk8, Wk28, Wk29, Wk31, Wk35 1 Wk2, Wk3, Wk5, Wk7, Wk11, Wk33, Wk37 2 Wk13, Wk16, Wk18, Wk20, Wk23, Wk38, Wk45, Wk46, Wk47

The Data Image of the hourly density data set is also analyzed to determine the time frames of regular and irregular densities. The weeks of the hourly data set’s Data Image are arranged according to their clusters. Figure 3 shows the Data Image of the hourly traffic density data set.

Fig. 3. Data Image of the Segment’s Hourly Traffic Density Data

The densities that are consistent in their value with respect to the same day throughout the year are the regular densities. Irregular density values are inconsistent with respect to the regular values of their day. As seen in Figure3, week 18’s Day 2 (Monday) has irregular density because it has relatively lower density than the other Monday’s of the year. Week 15’s Day 4 (Wednesday) has higher density than the other Wednesday’s of the year, making it irregular. The weeks of Cluster 2 should be the time frames of regular density since Cluster 2 has the highest number of weeks among the clusters. But since incidents are inevitable, irregular densities can be observed in Cluster 2. The details describing these discovered irregularities are discussed in the appendix. 6-minute Interval Data Set Visualization. Using the DSIV tool, the data signature for the 6-minute density data set is produced and the clusters of the weeks are generated. The weeks of the hourly and 6-minute density data sets

216

R.G. Maravilla Jr. et al.

have the same cluster model except for two variations. Cluster 2 in the hourly model is split into two different clusters (Clusters 2 and 4). The weeks of Cluster 4 of the first model were separated as outliers in the second one. Thus, Clusters 3, 5, 6, 7, and 8 in the 6-minute density data set are all outliers. These divisions occurred because of the weeks’ high intracluster distance. 6-minute density data set has more points to consider than the hourly density data set. The Data Image of the 6-minute density data set is also generated. The weeks of the 6-minute data set’s Data Image are arranged in such a way that they follow the order of the weeks of the hourly data set’s Data Image. This is done for a more convenient comparison between the two Data Images.

Fig. 4. Data Image of the Segment’s 6-minute Traffic Density Data

Figures 3 and 4 exhibit similar behaviors. This further validates the accuracy of the hourly density model. With the validation, the hourly data set is sufficient in representing the whole traffic data set. From the data signatures of the 6-minute density data set, we produce its cluster model, nMDS visualizations, and the best fit curves, confidence bands and ellipses. The 6-minute density data set has more potential outliers than the hourly density data set. It can be ascertained that most of the potential outliers in the hourly model are also discovered as such in the 6-minute data, validating our initial result.

4

Conclusions and Recommendations

We have shown in this paper that data signature-based density analysis can provide an efficient and effective representation of traffic behavior. Using the space mean speed instead of time mean speed produces realistic results because it considers the rate of movement of vehicles within a given section. Density analysis, together with thorough preprocessing of the data set, produces an effective congestion indicator.

Data Signature-Based Time Series Traffic Analysis

217

With the data signature representation of the hourly density output data, analysis on traffic outliers can be conducted efficiently. With the same preprocessing and procedures done on the 6-minute density data set, comparison with the hourly density data set yielded similar results. Thus, the hourly density data set is validated and is accurate enough to be used in traffic congestion analysis. With the validation of a larger scaled data set, there are less data points to process, providing efficiency without compromising its accuracy. With the outliers and potential outliers determined by our study, expressway management can have an efficient analysis of traffic behavior that can be used in anticipating traffic flow patterns. Traffic incidents can be addressed more efficiently to reduce accidents and other traffic obstructions. Additionally, to come up with a more generalized behavior of the whole expressway, it is recommended that other NLEX segments be also analyzed. Acknowledgements. The researchers like to thank Dr. Ma. Sheilah GaabucayanNapalang and Dr. Jose Regin Regidor for validating our results and providing the data sets. This work is partially supported by a grant from DOST-PCIEERD through an ERDT project entitled Information Visualization via Data Signatures.

References 1. Maravilla, R., Tabanda, E., Malinao, J., Adorna, H.: Traffic Density Modeling on NLEX Time Series Data Segment. In: Proceedings of the National Conference for Information Technology Education (2011) 2. Malinao, J., Juayong, R.A., Corpuz, F.J., Yap, J.M., Adorna, H.: Data Signatures for Traffic Data Analysis. In: 7th National Conference on IT Education (2009) 3. Sigua, R.G.: Fundamentals of Traffic Engineering, 42–66 (2008) 4. Rakha, H., Wang, Z.: Estimating Traffic Stream Space-Mean Speed and Reliability from Dual and Single Loop Detectors (2005) 5. Pelleg, D., Moore, A.: X-means: Extending K-means with efficient Estimation of the Number of Clusters. In: Proceedings of the 17th International Conf. on Machine Learning (2000) 6. Wong, P., Foote, H., Leung, R., Adams, D., Thomas, J.: Data Signatures and Visualization of Scientific Data Sets. In: Pacific Northwest National Laboratory. IEEE, USA (2000) 7. Malinao, J., Juayong, R.A., Oquendo, E., Tadlas, R., Lee, J., Clemente, J., Gabucayan-Napalang, M.S., Regidor, J.R., Adorna, J.: Gabucayan-Napalang, Ma.S., Regidor, J.R., Adorna, J.: A Quantitative Analysis-based Algorithm for Optimal Data Signature Construction of Traffic Data Sets. In: Proceedings of the 1st AICS/GNU International Conference on Computers, Networks, Systems, and Industrial Engineering, CNSI 2011 (2011) 8. Malinao, J., Juayong, R.A., Becerral, J., Cabreros, K.R., Remaneses, K.M., Khaw, J., Wuysang, D., Corpuz, F.J., Hernandez, N.H., Yap, J.M., Adorna, A.: Patterns and Outlier Analysis of Traffic Flow using Data Signatures via BC Method and Vector Fusion Visualization. In: Proc. of the 3rd International Conference on Human-centric Computing, HumanCom-2010 (2010)

218

R.G. Maravilla Jr. et al.

9. Malinao, J., Tadlas, R.M., Juayong, R.A., Oquendo, E.R., Adorna, H.: An Index for Optimal Data Signature-based Cluster Models of Coarse- and Fine-grained Time Series Traffic Data Sets. In: Proceedings of the National Conference for Information Technology Education (2011) 10. Johnson, R.: Visualization of Multidimensional Data with Vector-fusion. IEEE Trans., 298–302 (2000) 11. Cox, T., Cox, M.: Multidimensional Scaling, 42–69 (1994) 12. Oquendo, E.R., Clemente, J., Malinao, J., Adorna, H.: Characterizing Classes of Potential Outliers through Traffic Data Set Data Signature 2D nMDS Projection. Philippine Information Technology Journal 4(1) (2011)

Appendix For the hourly traffic density data set, the we were able to find evidences of some of the points in the cluster model that we have produced as being potential outliers in Table 1, as shown below. We also state here the identified in traffic flow using Figure 3. We look for the events that triggered the irregular densities in the Data Image by focusing on the time frames of relatively high, relatively low, and extremely high density values. – Certain days of Cluster 0’s weeks 1, 8, 29, and 35 have relatively lower density than the common density for that particular day in the cluster. For example, week 1’s Sunday has lower density than other Sundays of the same cluster. On the other hand, certain days of this cluster’s weeks 28, and 31 have relatively higher density than the common density for that particular day in the cluster. – Some of Cluster 1’s weeks have similar behavior with Cluster 0’s weeks. This is because the confidence ellipses of Clusters 0 and 2 cover these weeks, making them ambiguous. However, these weeks belong to Cluster 1 because of distinct behaviors exhibited only by the weeks in Cluster 1. We can see this in week 2, where its density is relatively lower on Friday, but relatively higher on Saturday which is a characteristic of Cluster 1. – Cluster 2’s potential outliers have days that have a different density value than the common one. Week 13, for instance, has relatively lower densities on Friday and Saturday, as compared to its co-weeks in the cluster. – Extremely low densities are not classified specifically because this is a regular occurrence in expressways during midnight until early morning. – Extremely high density values are observed during important holidays in the country. This is reflected by the time frames with a dark red range. The weeks are weeks 15 and 44. The start of Holy Week happens on week 15 (Day 4) and the All Saints’ Day holiday happens on week 44 (Day 3). These are the days when people visit their provinces located at the north of Luzon. – Sudden increase of density values during days with consistent density values usually observed are irregular densities that are classified as relatively high density values. Reasons behind this sudden increase include accidents on weeks 12 (Day 6), 20 (Day 7), and 28 (Day 3) and departure for holidays on weeks 50 and 52.

Data Signature-Based Time Series Traffic Analysis

219

– Relatively low densities are observed on sudden decrease in densities the same way relatively high densities are observed on sudden increase in densities. We observe that the days after the highest density value occurred have relatively low density values. This is due to the large number of people departing at the same time at the official start of the vacation for the Holy Week. Majority of the people planning on a vacation have already left, leaving a few to depart on the following days (week 15’s days 5, 6, and 7 and week 16’s day 1). – During weeks 39 and 44, typhoons Milenyo and Paeng, respectively, struck the country. Travel advisory from weather domain experts prevented the people from traveling which is why there is a low density turnout on the said time frames. There is also a low density turnout during some holidays. Christmas Eve (week 32’s day 1) and New Year (week 1’s day 1) are observed in the country with people staying inside their houses to celebrate. Day 2 of week 18 also has relatively low density. This is due to the Labor Day holiday. Most of the professional drivers who pass by NLEX are on holiday. – The Data Image also reflects relatively low densities during Day 1 of weeks 4, 27, and 47. This is due to the many people watching the fights of the boxer Manny Pacquiao. Pacquiao-Morales 2 happened during week 4 (January 22, Philippine time). Pacquiao-Larios happened during week 27 (July 2, Philippine time). Pacquiao-Morales 3 happened during week 47 (November 19, Philippine time). For some of the potential outliers in the 6-minute density data set’s model which are not found in the hourly density data set’s, the following observations were derived. Cluster 0’s week 27 has relatively lower density on Sunday than the usual density value of Sundays in the same cluster. Week 43 of Cluster 2 and week 42 of Cluster 4 are both similar to each other’s cluster. The same goes for the majority of the potential outliers in both clusters. This is due to their clusters’ confidence ellipses enclosing each others points. This is further supported by the hourly data set’s clustering wherein both clusters’ points belong to one cluster (Cluster 2 of hourly data set).

Automated Video Surveillance for Monitoring Intrusions Using Intelligent Middleware Based on Neural Network Ana Rhea Pangapalan1, Bobby D. Gerardo1, Yung-Cheol Byun2,*, Joel T. De Castro1, and Francisca D. Osorio1 1

Institute of ICT, West Visayas State University Luna St., Lapaz, Iloilo City, Philippines [email protected] 2 Dept. of Computer Engineering, Jeju National University Jeju City, Korea [email protected]

Abstract. Automated Video Surveillance Using Intelligent Middleware presented a Java based system that detects human activities in a security sensitive area and provides alarm for illegal activities identified. The system was developed using Netbeans IDE 6.8 [7] as the working environment, while Java as the programming language [8]. This study enhanced and strengthened existing security, therefore minimizing possibility of missed events which might be a threat to an area. The system composes three major processes: Motion Detection, Subject Identification and Behavior Classification. Motion Detection captures image of any movement detected. Subject Identification screens every captured image by classifying whether the motion is made by human and eliminating those which are caused by wind, animals and other non-human entity. Behavior Classification categorizes the image passed as to what action and outputs alarm if it is considered as illegal. In order to carry out these complex functionalities, a middleware was utilized to maintain continuous data flow from capturing to image processing and to reduce bulk of inputs that are being processed. Neural network [9, 10] was employed as the information processing paradigm for human or non-human and behavior classification. The result shows that the system processed video continuously as it classified behavior automatically. Keywords: Intrusion detection, intelligent middleware, neural network, security.

1

Introduction

Security of human lives and property has always been a major concern for civilization for several centuries. Video surveillance systems have been widely used to solve issues arising from illicit human activities. As more and more surveillance cameras are deployed in a facility or area, the demand for automatic methods for video processing is increasing [6], since in a conventional system one important event could be missed by a simple distraction from the one in charge of monitoring and watching the video streams. *

Corresponding author.

T.-h. Kim et al. (Eds.): FGCN 2011, Part II, CCIS 266, pp. 220–228, 2011. © Springer-Verlag Berlin Heidelberg 2011

Automated Video Surveillance for Monitoring Intrusions

221

Rapid developments from the past years to fully automate these systems have occurred since motion analysis in video has attracted many researchers. Smart Surveillance in itself is one of the most challenging problems in computer vision [1] due to its promising applications in many areas. With these developments, issues regarding the functional aspect of the automated systems arise. Common processes of these systems are motion detection, extraction of human body structure from images, tracking across frames, and behavior classification. In motion detection itself, different changes in input like fluctuating illumination and shadows can create movement therefore providing useless data for processing. Different algorithms have been developed to support complex processes and inconsistency from inputs due to changes in real world scenarios. From time to time, new algorithms are created and new methods are applied to overcome limitations of previous developed systems. These issues motivated the researchers in developing the proposed system which covered motion detection, subject identification and behavior classification. This paper documents all the information and procedures necessary to achieve the goals of the system.

2

Related Studies

In the study of Oh et al. [10] about View-Point Insensitive Human Pose Recognition Using Neural Network, they proposed a view-point insensitive human pose recognition using neural network. Their system consists of silhouette image capturing module, data driven database, and neural network. First, they capture 3D human model from different camera angle to generate a 2D silhouette image. Different poses from different camera are captured. Captures are stored in a database and will be used for neural network training. They use 2/3 features of all and 1/3 features are utilized as test data. The researchers used trained and non-trained data for testing. It was observed that the average precision is 75.3% at non-training data, and 81.24% at training data. The paper on ACE (Annotated Critical Evidence) Surveillance [5] presents an automated video surveillance technology that is developed by National Research Council of Canada (NRC) in 2008 which states that, in real-time monitoring mode, the problem is that an event may easily pass unnoticed due to false or simultaneous alarms and lack of time needed to rewind and analyze all potentially useful video streams. The system was developed for the purpose of enabling more efficient use of surveillance systems and one of its concerns is data storage and manageability for post-incident investigation. Developed by the Video Recognition Systems team of the National Research Council of Canada’s Institute of Information Technology (NRC), it was implemented as a software that runs on an ordinary desktop computer, performs real-time analysis of captured video streams for the purpose of automatically extracting and annotating Critical Evidence Snapshots, which are used to automatically alarm the system and which allow efficient summarization and browsing of captured video data.

222

A.R. Pangapalan et al.

On the other hand, a paper on Video-Based Human Motion Estimation System [11] came-up with the system designed to estimate body silhouette representation from sequences of images. Milanova and Bocchi in the same paper seeks to explore the hypothesis that the two building blocks of the accuracy of human motion estimation, the measured data and the prior model are critical components, using extremely high accuracy measured data and shape of body motion priors, so that the objective function is more precise and less noisy, resulting in an easier solution. Their main goal is to develop a new module for extracting accuracy measured data from video imagery. In a paper on Middleware for Distributed Video Surveillance [4], the researchers used middleware for the system and suggested the use of intelligent middleware for addressing the real-life challenges that include managing a large scale of cameras which will be difficult only when utilizing computer vision algorithms required in detecting and interpreting activity in video. Also, video surveillance systems manned by security personnel will be ineffective because even trained operators lose concentration and miss a large percentage of significant events. They utilized a middleware for support of network of cameras where the input is coming from. In the paper about Robust Techniques for Background Subtraction in Urban Traffic Video[2], explored the different background subtraction techniques in urban traffic video sequences have been compared and tested. For experiment, Cheung and Kamath [4] have selected four publicly-available urban traffic video sequences. Two algorithms that produce good performance are Mixture of Gaussian and Median Filtering. The MoG (Mixture of Gaussian) appears to have the best precision and recall and MF (Median Filtering) is a very close second. MoG still, has its own drawbacks. It is computationally intensive and its parameters require careful tuning and are very sensitive to global illumination. The researchers concluded that MoG produces best results while adaptive median filtering offers a simple alternative but still having a competitive performance. In [3, 12], the researchers build a motion decomposition approach to analyze the periodicity in human actions and also study about motion detection. In this paper [12], the researcher proposed a novel video compression idea to compress these extracted periodic activities from the videos. The method exploits the correlation between the frames over longer length of time, of the order of the period of the activity, as compared to the traditional video compression algorithms which use correlation between a few neighboring frames for motion prediction and compensation. He also considers the problem of silhouette normalization for activity analysis.

3

System Architecture

One of the goals of this study was to develop middleware that supported the simultaneous data flow from capturing to video processing. During the simulation and testing, the researchers utilized only one camera to minimize the size of input data that would be handled by the system for real time processing. The system was limited only to single, stationary cameras and video input; thus, pan, tilt and zoom features were not included. Another concern was that, from different area, varying behaviors were perceived to be illegal. For the scope to be visible, a list of different behaviors

Automated Video Surveillance for Monitoring Intrusions

223

was given and activities which were illegal from the area covered were selected, if not, the default settings would be followed where illegal activities were pre-defined. Due to limited knowledge in the language and the paradigm used, more time had been spent on researching causing time allocation to be affected therefore, creating some constraints on coding, debugging and polishing. 3.1

Framework

Figure 1 shows the foundation of the proposed system. It starts from a video stream that serves as the main input. From there motion detection initializes which is followed by the activation of neural networks. Take notice that middleware at that stage then subject detection and behavior classification until the alarm is generated takes place.

Alarm

Behavior Classification Subject Detection Middleware

Activation of Neural Network Motion Detection

Video Stream

Fig. 1. Framework of the proposed system

3.2

Architectural Design

Figure 2 shows the Architectural Design. The hardware included in the proposed system is a camera and a workstation/computer unit. The data being passed by a middleware includes the raw and the classified images. Those were being processed until the system outputs an alarm.

224

A.R. Pangapalan et al.

Fig. 2. Architectural Design of the Proposed System

3.3

Procedural Design

When the system is initiated, it will ask for a user name and a password for user authentication and it will give an option to start the surveillance. The user will then be prompted to choose a human behavior to be identified as illegal by the system. Through this, networks which correspond to the chosen behaviors will be activated. However, if the user fails to choose anything the network for all default illegal behaviors will be used by the system. The procedural design of the proposed system is presented in figure 3. Start

Input user

Yes

Start

Activate Networks

Surveillance

No Start

Fig. 3. Procedural Design of the Proposed System

Automated Video Surveillance for Monitoring Intrusions

4

225

Simulation and Results

As the system starts-up the user is required to input the password for authentication to start the surveillance. After that, user can define behaviors which are considered to be illegal in the area chosen for the system to deploy. A list is given and if the user fails to select, pre-defined/default human behaviors will be perceived as illegal and all neural networks will be activated. Output comes in the form of notification and if an instance of intrusion occurs. The log-in window and the action selector is shown in figure 4 and 5, respectively. Figure 5 also shows action that could be selected for activation of Neural Network. While launching of motion detection software is shown in figure 6.

Fig. 4. Log in window of the system

Fig. 5. Motion Selector of the system

226

A.R. Pangapalan et al.

Fig. 6. Launching of Motion Detection Software

Fig. 7. Sample Output for Identified Illegal Action

Whenever a human is detected, the middleware saved the image and passed it for behavior classification. Lastly, when an illegal behavior is detected, an alarm is generated and prompted as illustrated in figure 7.

5

Conclusions and Recommendations

The purpose of the proposed system was to offer security over an area without much need of human judgment and monitoring. It was planned, developed, tested and evaluated and was functional and working according to the processes. In one way or another, though limited, the system still achieved its goals and objectives since the researchers were able to identify the key processes which supported the complexity of an automated video surveillance system, determined the right algorithm that ensured efficiency, carried out real-time processing through the middleware, found ways to reduce the bulk of data needed and utilized only the inputs necessary to run the system.

Automated Video Surveillance for Monitoring Intrusions

227

Furthermore, it can also be concluded that the system served a good start and a foundation for a more sophisticated and polished security application. It shows that the limitations in terms of the system’s capacity can be solved since the idea and the basic concepts, processes and functions were achieved and made. For the improvement of the proposed system the researchers recommend that the future system should be capable of deleting previous captured and stored images to maximize space, place activation of networks, capturing and system alert in the same window for easier navigation and show detected image together with a sound alert. Acknowledgments. This research was financially supported by the Ministry of Education, Science Technology (MEST) and Korea Industrial Technology Foundation (KOTEF) through the Human Resource Training Project for Regional Innovation. This research is not possible without the help of the research team composed of Yssel Tarra Modina, Alvin Jorge Tambalo, Christy Villano, Joy Vista and some Faculty of the Institute of ICT at West Visayas State University.

References 1. Kosmopoulos, D.I., Antonakaki, P., Valasoulis, K., Katsoulas, D.: Monitoring human behavior in an assistive environment using multiple views. In: Proceedings of PETRA (2008) 2. Cheung, S.-C.S., Kamath, C.: Robust techniques for background subtraction in urban traffic video. In: Proceeding of SPIE, vol. 5308, p. 881 (2004) 3. Danjou, N.: Motion Detection (2006), http://noeld.com/programs.asp?cat=video#MDetect (retrieved March 1, 2010) 4. Detmold, H., van den Hengel, A., Dick, A., Falkner, K., Munro, D., Morrison, R.: Middleware for Distributed Video Surveillance. IEEE Distributed Systems Online 9(2) (2008) 5. Gorodnichy, D.O.: ACE Surveillance: The Next Generation Surveillance for Long-Term Monitoring and Activity Summarization. First International Workshop on Video Processing for Security (VP4S-2006), June 7-9, Quebec City, Canada 6. Kotikalapudi, U.K.: Abnormal event detection in video, Master’s Thesis, Supercomputer Education Research Center, Indian Institute of Science (2010), http://www.serc.iisc.ernet.in/graduation-theses/Uday.htm (retrieved March 1, 2010) 7. Java (2010), http://www.java.com/en/download/whatis_java.jsp (retrieved March 1, 2010) 8. NetBeans, http://netbeans.org/community/releases/68/relnotes.html (retrieved March 1, 2010) 9. Neural Network, http://www.merriam-webster.com/dictionary/neural%20network (retrieved March 1, 2010)

228

A.R. Pangapalan et al.

10. Oh, S., Lee, Y., Hong, K., Kim, K., Jung, K.:: View-point insensitive human pose recognition using neural network (2009), http://www.waset.org/journals/waset/v44/v44-50.pdf (retrieved 1 March, 2009) 11. Milanova, M.G., Bocchi, L.: Video-Based Human Motion Estimation System. In: HCI, vol. (11), pp. 132–139 (2009) 12. Varsha Chandrashekhar, H.: Human Activity Representation, Analysis and Recognition. (2006), http://www.security.iitk.ac.in/contents/publications/more/ human/activity/representation.pdf

SMS-Based Automatic Billing System of Household Power Consumption Based on Active Experts Messaging Mark Dominic Cabioc1, Bobby D. Gerardo1, Yung-Cheol Byun2, 1

2

*

Institute of ICT, West Visayas State University Luna St., Lapaz, Iloilo City, Philippines [email protected]

Dept. of Computer Engineering, Jeju National University Jeju City, Korea [email protected]

Abstract. The study about SMS-based Automatic Billing System of Power Consumption aimed to change the conventional way the power utility provider gathers and handles billing data. The system is composed of two basic parts such as the remote site and the base station. The former calculates and sends power consumption while the latter retrieves meter readings, calculates billing charges and processes payment of the customers. Microsoft Visual Studio 2008 was used to develop the Main Server Software with Visual Basic 2008 as the Integrated Development Environment (IDE) or programming tool and Visual Basic as the programming language. The database was created in Microsoft SQL Server 2005 and ActiveXperts Messaging Server 4.1 was use as an SMS framework that allows the system to send, receive and process SMS. The use of this system will give greater benefit to the electric company and its customers because of the ease and less impediment in gathering meter readings in remote locations and an instant delivery of billing statements to the customer’s cellular phones through SMS technology. Keywords: SMS messaging, automatic billing, Active Expert, power management.

1

Introduction

Conventional electricity billing system in the country has been lagging in terms of technology in gathering and processing power consumption data for billing purpose. It is done by an assigned person who visits each meter location periodically and reads the meter manually. Data collected are then processed either manually or via specialized software into the customers‟ bills which in turn are distributed manually to each respective customer. As such, conventional meter reading poses several problems wherein adverse effects extend to the customer himself. Misreading of the power meter for instance can *

Corresponding author.

T.-h. Kim et al. (Eds.): FGCN 2011, Part II, CCIS 266, pp. 229–238, 2011. © Springer-Verlag Berlin Heidelberg 2011

230

M.D. Cabioc et al.

inadvertently increase or decrease the customer’s bill. The recent technology of automatic meter reading (AMR) [1] has solved the problems of conventional meter reading. AMR has promised fast and accurate collection of meter readings, however, incorporating this with manual computation of electric bills does not fully utilized its functionality. With the development of the text messaging in the country and the dependence of the people on it, the researchers took advantage of the GSM network and SMS technologies available to solve these inefficiencies. The convergence of SMS and GSM network allow more mobile and wireless applications to be implemented such as automatic meter reading and billing. Although many related projects have been presented in some other countries, the development and implementation of this study in the Philippines is not yet totally accomplished. The proposed SMS-based Automatic Billing System of Power Consumption will elaborate further on some existing technologies and studies that focus on Automatic Meter Reading and Remote Meter Reading since it will give more effort and attention in developing the billing system interface of power consumption. The development of the system will not only focus on the communication network but on the application side wherein a secure and reliable Automatic Billing System will be developed. The use of this system will give greater benefit to the electric company and its customers because of the ease and less impediment in gathering meter readings in remote locations and an instant delivery of billing statements to the customer's cellular phones.

2

Related Studies

The study on Fixed-Network Automatic Meter Reading (AMR) System [2] was a graduation project of Awad and Gosh presented to the Department of Electrical Engineering, University of Jordan. The project discusses about utilizing already available fixed communication networks (e.g., the cellular network) for exchanging data to minimize cost and human effort. The purpose of their project was to introduce a Fixed-Network AMR design that manages the reading of the electricity meters at the consumers' side. This design was intended to replace the existing manual methods of gathering data. The approach to the solution for this problem was made with the use of the GSM network and a custom RF solution. Different hardware modules were introduced to help exchange data between a central office and any node in the system (i.e., customer's side). The GSM and RF communication media were fully utilized by introducing their own GSM and RF protocols. A paper on Networked Remote Meter-Reading System Based on Wireless Communication Technology [4] studied meter-reading system based on Bluetooth wireless communication technology and GSM. The remote meter-reading system employs distributed structure, which consists of measure meters, sensors, intelligent terminals, management centre and wireless communication network. The intelligent terminal which designed based on embedded system and Bluetooth technology was used to realize acquisition information submitted from meters and sensors control the energy-consuming devices moreover in residence.

SMS-Based Automatic Billing System of Household Power Consumption

231

While in a paper on Real-time Energy Management over Power-lines and Internet [17], explore the creation of an infrastructure for energy management that should enable the enhancement of existing applications like automatic meter reading, distribution grid management, and remote control. The system allows for direct communication with equipment at the customers’ premises via a two stage hierarchical power line communication system and an IP (Internet Protocol)-based private network. A paper on Novel Approach for Remote Energy Meter Reading Using Mobile Agents [15] incorporated power or energy meter systems with embedded controllers such as micro web-servers with Ethernet port to transmit the reader data over the Internet. Such data can be then fed and integrated into existing energy management systems located at power companies and organizations. Mobile agents are executing programs that migrate during execution and present methods for maintaining and using distributed systems. However, the problem of efficiently collecting data from a large number of distributed embedded Web-servers in the energy meters is still a challenging problem. On the other hand, in [1] developed hardware structure consisting of a digital energy meter module, another digital meter for water and a telephone module, all lined with a single chip microcontroller were equipped with credit card reading capability to automatically read and charge the consumption on site. Also, all service metering modules were facilitated with an automatic service connection and disconnection based on the available credit. The software structure commands the whole process via the microcontroller input/output ports. Furthermore, the study of Jabundo et al. [9] about prepaid kilowatt-hour meter aimed at reducing certain problems such as pilferage and excessive use of electric energy in establishments. This system was composed of a customized digital power meter that was prepared for prepaid use. Occupants of the establishment will purchase energy load–equal to its corresponding energy based on the current energy rate–from the establishments’ owner. When the occupant totally consumes the energy load, the current will be cut off by the magnetic switch. Based on the reviewed literatures, the researchers are motivated to study an automatic billing system of power consumption that automatically acquires meter readings from a remote station, calculates billings and sends SMS bills to the customers. The proposed system used a digital power meter designed to send meter readings directly to the system through GSM network, an SMS Messaging Server as its middleware and develop a multi-level system that handles data collection, billing computation and payment processing.

3

System Architecture

Theoretically, the proposed system will help ease the collection of billing data, thus saving the utility provider and customer time, money, effort and inconvenience. It also eliminated the cost incurred by the traditional meter data collection process, thus lessening the burden cast upon the customers. The proposed automatic billing system is composed of a remote site and base station as shown in Figure 1. The digital meter that is located in the remote site (homes and buildings) will send the meter readings in the base station’s billing system via SMS using the stable GSM network for cellular phones.

232

M.D. Cabioc et al.

Fig. 1. The proposed System Architecture

3.1

Remote Site

The remote site is the digital power meter customized to receive queries and send its own meter reading to the base station via SMS. The device consists of a microcontroller unit, GSM modem, LCD display, current and voltage sensor attached to the main power supply.

Fig. 2. Architectural design of the remote site

As shown in Figure 2, the remote site’s current and voltage sensors receive energy that passes through the meter reader from the power line to the customer’s workload — appliances or other AC devices at a maximum of 2 amperes. The two sensors measure current and voltage respectively, which can be viewed from the meter reader’s LCD display.

SMS-Based Automatic Billing System of Household Power Consumption

233

Microcontroller unit calculates the input from the current and voltage sensors, converting the results into kilowatt per second, representing the customer’s power consumption. The resulting data are then sent to the GSM modem in the event the remote site receives an SMS request from the base station with this format:’check data’. The data are then processed into an SMS message in this format: “Meter ID, Reading”. The message is then sent to the server. 3.2

Base Station

The base station acts as the central server where all the commands and processing of data are executed. It is composed of another GSM modem, an SMS Messaging Server and the main server software.

Fig. 3. Architectural Design of the Base Station

As shown in Figure 3, GSM modem is used by the base station to communicate with the remote site by sending an SMS request to the customized digital meter, which sends back the current kWh reading in return. The GSM modem is also used to send billing information to the customer via SMS.

4

Simulation and Results

The SMS Messaging Server manages SMS messages sent and received by the GSM modem. Running in the background, it acts as the system’s middleware. Meanwhile, the Main Server Software is multi-level software that handles processes such as retrieval of meter readings directly from the remote station, computation of bills and processing of payments for the said bills. The software is composed of a database and data collection, computation and analysis, SMS billing statements generation and payment processing modules. The Main Server Software’s Database is connected to all the modules of the software. This is where data in the system is stored and retrieved. The Data Collection Module communicates data with the SMS Messaging Server. From there it

234

M.D. Cabioc et al.

retrieves relevant SMS messages such as meter readings to be saved in the database. The computation and analysis, module retrieves data from the database, calculates and analyzes the overall consumption and relays the information to the billing statements generation module.

Fig. 4. Input Window for Customer’s Details for Registration

Fig. 5. Assigning meter details for each customer

Fig. 6. Updating charge rates

SMS-Based Automatic Billing System of Household Power Consumption

235

The SMS billing statements generation module generates detailed reports of customers monthly bill consisting of the current and previous kWh readings, total kWh used, charge rates, amount due and due date. Billing statements are sent to the consumer’s cellular phone in SMS format. This is presented in Figure 7.

Fig. 7. Processing Customers’ Payment

On the other hand, the payment processing module is responsible for processing the payments of the customer. It also includes the handling of unpaid due amounts and other functions related to payment system.

Fig. 8. Billing Information (SMS) Received by the Customer

236

M.D. Cabioc et al.

The GSM network, when interfaced with an SMS gateway like ActiveXperts can be utilized as an effective data communications system. The ActiveXperts SMS gateway on the other hand, is good third-party software for this matter. It efficiently manages received data into databases and is capable of sending data via SMS. Figure 8 shows the SMS sent to the customer for its billing information.

Fig. 9. Customer’s Monthly Consumptions Report

On the other hand, Figure 9 shows the monthly power consumption of the customer. The system could also provide master list of the customers and as well as daily report of consumption and payment reports.

5

Conclusions and Recommendations

In this study, a user-friendly software application for the base station was created. We demonstrated the automation of the retrieval, storage and sending of data. In addition, the system’s software was able to compute the data acquired from the remote site and generate billing statements for the customer. A prototype digital meter that measures power consumption and sends data via SMS was successfully adopted and integrated into the system. Moreover, effective communication between the remote site and the base station was successfully established using the GSM network gateway. To further improve the efficiency of the software system, the researchers recommend that functions such as bulk sending of SMS billing information to the customers, one click computation of all customers’ monthly bills and receiving of inquiries from customers via SMS shall be included in the future study. Also the researchers recommend that energy consumers should monitor their power consumption by inquiring their own meter readings directly from the customized digital meter at any given time.

SMS-Based Automatic Billing System of Household Power Consumption

237

Acknowledgments. This research was financially supported by the Ministry of Education, Science Technology (MEST) and Korea Industrial Technology Foundation (KOTEF) through the Human Resource Training Project for Regional Innovation. We would like to express our gratitude to the efforts made by the research team composed of Elera Marie O. Joaquin, Denmark S. Padua, Jey Mark P. Palma, John Niño C. Requina, , Christian Roy S. Somcio, and some Faculty members of the Institute of ICT of West Visayas State University.

References 1. Al-Qatari, S.A., Al-Ali, A.R.: Microcontroller-based Automated Billing System (1995) 2. Awad, A.J., Abu Ghos, R.T.: Fixed-Network Automatic Meter Reading (AMR) System. Department of Electrical Engineering, University of Jordan. FixedNetwork-AutomaticMeter-Reading-AMR-System, http://www.scribd.com/doc/950863/ (retrieved January 2010) 3. Base station. In: TechTerms.com, The Tech Terms Computer Dictionary. from http://www.techterms.com/definition/basestation (Retrieved 2005) 4. Cao, L.: Networked Remote Meter-Reading System Based on Wireless Communication Technology. In: IEEE International Conference on Information Acquisition (2006) 5. Database: Dictionary of Computer and Internet Terms (10th ed.). Barron’s Educational Series, Inc., New York (2009) 6. Downing, D., Covington, M.: Dictionary of Computer and Internet Terms (10th ed.) Barron‟s Educational Series, Inc.,New York (2009) 7. GSM. In: VoIP Dictionary (2003), http://www.voipdictionary.com (retrieved January 2010) 8. GSM Modem: In Now SMS (2002). from http://www.nowsms.com/gsm%20modems.htm (retrieved January 2010) 9. Jabundo, D.J.R., et al.: Prepaid Kilowatt-hour Meter. Prepaid Kilowatt-hour Meter. College of Engineering, Central Philippine University, Iloilo City, Philippines (2010) 10. Microsoft, SQL. Tutorials, http://www.sqlcourse.com (retrieved from 2011) 11. Microsoft Visual Basic (2008), http://www.microsoft.com (retrieved 2008) 12. Ofrane, A., Harte, L.: Introduction to Wireless Billing. Usage Recording, Charge Processing, System Setup, and Real Time Billing 2006, http://www.billingdictionary.com/billing_dictionary_billing_ system_definition.html (retrieved January 2010) 13. Power Consumption. In: Dictionary Babylon (1997), http://www.dictionary.babylon.com/ (retrieved January 2010) 14. SMS: In Tech Terms, The Tech Terms Computer Dictionary (2005). SMS Messaging Server, http://www.techterms.com/definition/sms (retrieved January 2010) 15. Tahboub, R., et al.: Novel Approach for Remote Energy Meter Reading Using Mobile Agents. In: Third International Conference on Information Technology: New Generations, ITNG (2006) 16. Thiele, T.: Power Meter - How an Electrical System Works (2008), http://electrical.about.com/od/panelsdistribution/ss/elecsys works_3.htm (retrieved January 2010)

238

M.D. Cabioc et al.

17. Treyl, A., et al.: Real-Time Energy Management Over Power Lines and Internet. In: Proceeding of the 2003 Joint Conference of the 4th International Conference on Information and Communications and Signal Processing, 2003 Pacific Rim Conference on Multimedia, pp. 144–147 (2003) 18. Visual Basic Tutorials, http://www.msdn.microsoft.com (retrieved from 2011)

Hierarchical Clustering and Association Rule Discovery Process for Efficient Decision Support System Bobby D. Gerardo1, Yung-Cheol Byun2,*, and Bartolome Tanguilig III3 1

Institute of ICT, West Visayas State University Luna St., Lapaz, Iloilo City, Philippines [email protected] 2 Dept. of Computer Engineering, Jeju National University Jeju City, Korea [email protected] 3 Technological Institute of the Philippines, Cubao, Quezon City [email protected]

Abstract. This paper proposed a model based on hierarchical Clustering and Association Rule, which is intended for decision support system. The proposed system is intended to address the shortcomings of other data mining tools on the processing time and efficiency when generating association rules. This study will determine the data structures by implementing the cluster analysis which is integrated in the proposed architecture for data mining process and calculate for associations based on clustered data. The results were obtained using the proposed system as integrated approach and were rendered on the synthetic data. Although, our implementation uses heuristic approach, the experiment shows that the proposed system generated good and understandable association rules, which could be practically explained and use for the decision support purposes. Keywords: Data mining, decision support system, clustering, association rules.

1

Introduction

Often, there are many attributes or dimensions that are contained in the database, and it is possible that subsets of such dimensions are highly associated with each other. The dimensionality of a model is determined according to the number of input variables used. Clustering can be used to group data into clusters so that the degree of association is strong between members of the same cluster and weak between members of different clusters [1], [9]. Thus, each cluster describes the class to which its members belong. For that reason, cluster analysis can reveal similarities in data which may have been otherwise impossible to find. Data cubes allow information to be modeled and viewed in multiple dimensions and such cubes are then defined by the dimensions and facts [1]. They defined *

Corresponding author.

T.-h. Kim et al. (Eds.): FGCN 2011, Part II, CCIS 266, pp. 239–247, 2011. © Springer-Verlag Berlin Heidelberg 2011

240

B.D. Gerardo, Y.-C. Byun, and B. Tanguilig III

dimensions as entities with respect to which an organization wants to keep records of. Data cubes may be used in theory to answer query quickly, however, in practice they have proven exceedingly difficult to compute and store because of their inherently exponential nature [7]. Moreover, issues that other researchers observed in the data mining tasks were computing speed, reliability of the approach for computation, heterogeneity of database, and vast amount of data to compute [1], [2], [7]. This paper explore the formulation of the cluster analysis technique as integrated component of the proposed model to partition the original data prior to implementation of other data mining tools. The model that we proposed uses the hierarchical nearest neighbor clustering method and apriori algorithm for association mining implemented on transactional databases.

2

Related Studies

Association rule mining tasks includes finding frequent patterns, associations, or causal structures among sets of items or objects in transactional databases and relational databases. Data mining uses various data analysis tools such as from simple to complex and advanced mathematical algorithms in order to discover patterns and relationships in dataset that can be used to establish association rules and make effective predictions. 2.1

Cluster Analysis

The goal of cluster analysis is categorization of attributes like consumer products, objects or events into clusters or groups, so that the degree of correlation is strong between members of the same cluster and weak between members of different clusters. Each group describes the class in terms of the data collected to which its members belong. It may show structure and associations in data, although not previously evident, but are sensible and useful once discovered. The results of cluster analysis [9] may contribute to the definition of a formal classification scheme, such as in taxonomy for related animals, insects or plants; suggest statistical models with which to describe populations; indicate rules for assigning new cases to classes for identification and diagnostic purposes; provide measures of definition, size and change in what previously were only broad concepts. 2.2

Apriori Algorithm

There are varieties of data mining algorithms that have been recently developed to facilitate the processing and interpretation of large databases. One example is the association rule algorithm, which discovers correlations between items in transactional databases. The Apriori algorithm is used to find candidate patterns and those candidates that receive sufficient support from the database are considered for transformation into a rule. This type of algorithm works well for complete data with

Hierarchical Clustering and Association Rule Discovery Process

241

discrete values. Some limitations of association rule algorithms, such as the Apriori is that only database entries that exactly match the candidate patterns may contribute to the support of that candidate pattern. In the past years, there were lots of studies on faster, scalable, efficient and cost-effective way of mining a huge database in a heterogeneous environment. Most studies have shown modified approaches in data mining tasks which eventually made significant contributions in this field. However, there are limitations on generated rules, like producing enormous, unclear and sometimes irrelevant rules.

3

System Architecture

The proposed architecture for the data mining system is shown in Figure 1. Its refinement is presented in the subsequent sections. Phase 1

Phase 2

Phase 3

Hierarchical Clustering

Association Rule Mining

Distributed DB Discovered Knowledge Aggregated Data

Clustered data

Fig. 1. The general view of the proposed model

Figure 1 shows the proposed three phase architecture, where the first phase is the data preprocessing stage that performs data extraction, transformation, loading and refreshing. This will result to an aggregated data cubes as shown in the same figure. Phase 2 shows the implementation of the hierarchical nearest neighbor clustering, while Phase 3 is the implementation of Apriori algorithm to generate rules. Phase 2

Hierarchical clustering using nearest neighbor method

Randomly Select k objects

Calculate Mean op for each cluster value

Iterate until criterion function converges

Aggregated Data

Clustered Data

Fig. 2. Refinements of the model at Phase 2

242

B.D. Gerardo, Y.-C. Byun, and B. Tanguilig III

Figure 2 shows the implementation of the cluster analysis using the hierarchical nearest neighbor clustering algorithm while Figure 3 is the implementation of association rule discovery method. Figure 3 shows the refined view of Phase 3. In this illustration, association rule algorithm is used as part of the data mining process. The successions of transforms for association rule algorithm which are represented by bubbles are shown in the shaded rectangle. Phase 3

Association Rules Generation

Calculate frequent sets

Clustered Dataset

Compute association rules

Generate rules based on constraints

Discovered Association Rules

Fig. 3. Refinements of the model at Phase 3

Phase 3 is the final stage in which the association rule algorithm will be implemented to generate the association rules. This calculates for the frequent itemsets and then compute for the association rules using the threshold for support and confidence. The output is given by the last rectangle showing the discovered rules. In this study, the discovered rules are provided in the table showing the support count and the strength of its confidence which are presented in section 5. The process allows the data to be modeled and viewed in multiple dimensions. Cluster analysis will generate partitions of the dataset, and then the association rule discovery process will be employed. The data cubes will reveal the frequent dimensions, thus, could generate rules from it. The final stage is utilization of the result for decision support. The proposed architecture will implement the association rule generation on a clustered database and would expect better data mining results.

4

Cluster Analyses and the Proposed Model

Among the most popular hierarchical clustering methods are Nearest-Neighbor, Farthest-Neighbor, and Minimal Spanning Tree while for non- hierarchical methods are K-Means, Fuzzy K-Means, and Sequential K-Means. This study put more emphasis on the use of hierarchical method as shown in the experimental results. 4.1

Types of Cluster Analysis

Cluster analysis is a method used for partitioning a sample into homogeneous classes to create an operational classification. Such classification may help formulate hypotheses concerning the origin of the sample, describe a sample in terms of a typology, predict the future behavior of population types, optimize functional processes for business site locations or product design, assist in identification as used

Hierarchical Clustering and Association Rule Discovery Process

243

in the medical sciences, and measure the different effects of treatments on classes within the population [9]. Nearest-Neighbor clustering is one of the simplest agglomerative hierarchical clustering methods, which is also known as the nearest neighbor technique. The defining feature of the method is that distance between groups is defined as the distance between the closest pair of objects, where only pairs consisting of one object from each group are considered [10]. An agglomerative hierarchical clustering procedure produces a series of partitions of the data, Pn, Pn-1 until P1. The first Pn consists of n single object clusters, while the last P1 consists of single group containing all n cases. At each particular stage the method joins together the two clusters which are closest together or are most similar. Figure 4 shows the algorithm for the nearest neighbor clustering. Given: A set X of objects {x1,...,xn}, A distance function dis(c1,c2) 1. for i = 1 to n ci = {xi} end for 2. C = {c1,...,cb} 3. l = n+1 4. while C.size > 1 do a) (cmin1,cmin2) = minimum dis(ci,cj) for all ci,cj in C b) remove cmin1 and cmin2 from C c) add {cmin1,cmin2} to C d) l = l + 1 end while

Fig. 4. The nearest neighbor clustering algorithm

Although there are several other hierarchical clustering methods, in this study, the nearest neighbor had been utilized as part of our proposed model for clustering the data. 4.2

Implementation of the Proposed Model

The models in section 3 as reflected in Figures 1, 2 and 3, respectively, will be implemented in a heuristic process. In our experiment, we will calculate for the clustered data based on the proposed model. And then the outputs will be processed by implementing the Apriori for association mining. Tables showing the comparison of the results on original dataset, the proposed model and the discovered rules are presented in section 5.

5

Simulation and Results

The simulation was done on the database containing 30 attributes comprising of six (6) major dimensions and a total of 1,000 tuples of e-commerce and transactional

244

B.D. Gerardo, Y.-C. Byun, and B. Tanguilig III

types of data. The evaluation platforms utilized in the study were IBM compatible computer, Windows OS, C++, and Python. For the purposes of illustrating the database used in the experiment, we present the Dataset showing partially the data as revealed in Table 1. The abbreviated notations for the attributes stand as follows: An= books and its corresponding subcategories, Bn = Electronics, Cn = Entertainment, Dn= Gifts, En = Foods, and Fn = Health. Furthermore, An Book attribute is consist of subcategories like A1= Science, A2=social, A3=math, A4=computer, A5=technology, A6=religion, and A7=children books. Other dimensions are written with notations similar to that of An. The discrete values indicated by each record are corresponding to the presence or absence of the attribute in the given tuples. Supposed that we consider the problem of determining how often consumers buy products and the probability of purchasing some items online. The results which will be presented in the subsequent sections will answer this problem. Most literatures assumed that the hierarchical clustering procedure is suitable for binary or counts data types [1], [7], [8], [10]. The method that we considered for cluster analysis, which is integrated in proposed model is just suitable for the dataset that we assumed. Consumers respond to questions by giving their agreement or disagreement on buying some products online. 5.1

Hierarchical Clustering Results

The simulation will identify relatively homogeneous groups of cases based on selected characteristics. It is observed that a total of 4 clusters had been created and the group membership of each case is shown in Table 1. In the clustering result, the minimum distance of each case indicates its membership to the cluster. In summary, cluster 1 has a total of 433 cases (43.3%), cluster 2 has 235 cases (23.5%), cluster 3 has 165 (16.5%) and cluster 4 has 167 cases (16.7%). 5.2

Comparison of Data Mining Result after the Implementation of the Model

The data mining results using the two approaches are shown on Table 1 which also shows their corresponding values. The same table presents the number of cases that belong to the respective clusters. After implementing the clustering, we then employed the association rule algorithm (Apriori property). The results is shown in Table 2. The use of such algorithm is for discovering association rules that can be divided into two steps: (1) find all itemsets (sets of items appearing together in a transaction) whose support is greater than the specified threshold. Itemsets that meet the minimum support threshold are called frequent itemsets, and (2) generate association rules from the frequent itemsets. All rules that meet the confidence threshold are reported as discoveries of the algorithm.

Hierarchical Clustering and Association Rule Discovery Process

245

Table 1. Clusters and the Discovered Rules, Support >0.90

Original Dataset

Clustered (Cases, 433, 235, 165, 176)

Clusters 

Number of Rules

All

1

2

3

4

(1,000)

(433)

(235)

(165)

(176)

1,758

1,154

708

650

548

Table 2. Comparison of the Discovered Rules Models

Discovered Rules (showing first 5 rules generated)

Support

Confidence

Original (1,758 rules)

A6=Buy -> A2=Buy F4=Buy A6=Buy -> A2=Buy A3=Buy A6=Buy -> A2=Buy C4=Buy F4=Buy A6=Buy -> A2=Buy D2=Buy F4=Buy A6=Buy -> A2=Buy A3=Buy C4=Buy

0.935 0.927 0.916 0.915 0.910

0.942 0.934 0.922 0.921 0.916

Cluster Analysis 1(1154rules)

A6=Buy -> A2=Buy F4=Buy A6=Buy -> A2=Buy A3=Buy A6=Buy -> A2=Buy D2=Buy F4=Buy A6=Buy -> A2=Buy C4=Buy F4=Buy A6=Buy -> A3=Buy F4=Buy

0.924 0.919 0.905 0.903 0.901

0.939 0.934 0.920 0.918 0.915

2 (708 rules)

A6=Buy -> A2=Buy F4=Buy A6=Buy -> A2=Buy A3=Buy A6=Buy -> A2=Buy A6=Buy -> A2=Buy C4=Buy A6=Buy -> A2=Buy D2=Buy

0.912 0.910 0.963 0.942 0.940

0.923 0.921 0.974 0.953 0.951

3 (650 rules)

D2=Buy -> F3=Buy D2=Buy -> F2=Buy D2=Buy -> C4=Buy D2=Buy -> A2=Buy D2=Buy -> A3=Buy

0.921 0.909 0.958 0.958 0.945

0.938 0.926 0.975 0.975 0.963

4 (548 rules)

D2=Buy -> F3=Buy D2=Buy -> F2=Buy D2=Buy -> A2=Buy D2=Buy -> C4=Buy D2=Buy -> A6=Buy

0.922 0.910 0.958 0.952 0.946

0.939 0.927 0.976 0.970 0.963

246

B.D. Gerardo, Y.-C. Byun, and B. Tanguilig III

The result only shows the first five rules generated for each of the cluster. The support threshold that we set prior to the experiment was 0.90. In the original dataset, those who buy A6 (books on religion) will most likely buy A2 (books on social science) and F4 (Health supplement) with support of 0.935 and confidence of 0.942 (94.2% probability of buying). The same fashion of explanation and analysis could be done to other rules. In cluster 1, those who buy A6 (books on religion) will most likely buy A2 (books on social science) and F4 (Health supplement) with support of 0.924 and confidence of 0.939 (93.9%). Similar approach of analysis could be made for other rules in this cluster. And a similar fashion of explanation could also be done for other rules discovered such as in clusters 2, 3 and 4, respectively. In principle, there would be an improvement in processing time since the computation of rules is based on chunks of data, i.e. clustered data. Shorter processing time had been observed to compute for smaller clusters attributes implying faster and ideal processing period than processing the entire dataset. 5.3

Further Analysis and Implications

The blending of cluster analysis and association rule generation in the proposed model specifically isolate groups of correlated cases using the hierarchical nearest neighbor clustering and then using of the extended data mining steps like the algorithm for association rule generation. The model identify relatively homogeneous groups of cases based on selected characteristics and then employed the Apriori algorithm to calculate for association rules. This resulted to some partitions where we could conveniently analyze specific associations among clusters of attributes. This further explains that the generated rules were discovered on clusters indicating highly correlated cases which will eventually implies simplification of analysis of the result, thus beneficial to be used for decision support purposes.

6

Conclusions and Recommendations

The model reveals clusters that have high correlation according to predetermined characteristics and generated isolated but imperative association rules based on clustered data which in return could be practically explained for decision support purposes. The rules generated based on clustered attributes indicates simple rules, thus it could be efficiently used for decision support system such as in policy making or top level decision making. For future works, upgrade of the model based on extended clustering methods like divisive and non-hierarchical clustering may be needed to check if it performs well with other mechanisms. Acknowledgments. This research was financially supported by the Ministry of Education, Science Technology (MEST) and Korea Industrial Technology Foundation (KOTEF) through the Human Resource Training Project for Regional Innovation.

Hierarchical Clustering and Association Rule Discovery Process

247

References [1] Han, J., Kamber, M.: Data Mining Concepts & Techniques. Morgan Kaufmann, USA (2001) [2] Pressman, R.: Software Engineering: a practitioner’s approach, 6th edn. McGraw-Hill, USA (2005) [3] Hellerstein, J.L., Ma, S., Perng, C.S.: Discovering actionable patterns in event data. IBM Systems Journal 41(3) (2002) [4] Multi-Dimensional Constrained Gradient Mining, ftp://fas.sfu.ca/pub/cs/theses/2001/JoyceManWingLamMSc.pdf [5] Chen, B., Haas, P., Scheuermann, P.: A new two-phase sampling based algorithm for discovering association rules. In: Proceedings of ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (2002) [6] Margaritis, D., Faloutsos, C., Thrun, S.: NetCube: A Scalable Tool for Fast Data Mining and Compression. In: 27th Conference on Very Large Databases (VLDB), Roma, Italy (September 2001) [7] Han, E.H., Karypis, G., Kumar, V., Mobasher, B.: Clustering in a high-dimensional space using hypergraph models (1998), http://www.informatik.uni-siegen.de/~galeas/papers/general/ Clustering_in_a_High-Dimensional_Space_Using_Hypergraphs_ Models_%28Han1997b%29.pdf [8] Cluster Analysis defined, http://www.clustan.com/what_is_cluster_analysis.html [9] Determining the Number of Clusters, http://cgm.cs.mcgill.ca/soss/cs644/projects/siourbas/ cluster.html#kmeans [10] Using Hierarchical Clustering in XLMiner, http://www.resample.com/xlminer/help/HClst/HClst_intro.htm [11] Ertz, L., Steinbach, M., Kumar, V.: Finding Topics in Collections of Documents: A Shared Nearest Neighbor Approach. In: Text Mine 2001, Workshop on Text Mining, First SIAM International Conference on Data Mining, Chicago, IL (2001) [12] Hruschka, E.R., Hruschka Jr., E.R., Ebecken, N.F.F.: A Nearest-Neighbor Method as a Data Preparation Tool for a Clustering Genetic Algorithm. In: SBBD, pp. 319–327 (2003)

Implementation of Energy Efficient LDPC Code for Wireless Sensor Node Sang-Min Choi1 and Byung-Hyun Moon2 1

D&D Division Taihan Electric Wire Co., Ltd. 785, Gwangyang-2dong, Dong-gu, Anyang, Gyeonggi, 431-810, Korea [email protected] 2 Dept. of Computer and Communication Engineering, Daegu Univ. 15, Neari, Jinryang, Gyeongsan, Gyeongbuk, 712-714, Korea [email protected]

Abstract. The energy efficiency of error control scheme is very important because of the strict energy constraints of wireless sensor networks. Wireless sensor node requires simple error control schemes because of the low complexity request of sensor nodes. Automatic repeat request(ARQ) and forward error correction(FEC) are the key error control strategies in wireless sensor networks. In this paper, we implemented the efficient QC-LDPC encoder which does not require matrix inversion to improve the complexity of the encoder. It is shown that the efficient QC-LDPC code obtained 17.9% and 36% gain respectively in the mean number of transmission for the transmission power of -19.2dBm and -25dBm. Keywords: QC-LDPC code, Wireless sensor networks.

1

Introduction

In recent years, the idea of wireless sensor networks has produced lots of research, because of wireless sensor networks can be applied widely in many fields. In wireless sensor networks, erroneous transmission can happen by wireless channel noise. Automatic repeat request (ARQ) detects errors using cyclic redundancy check (CRC) and retransmission of data is used as error control scheme for wireless sensor networks. Sensor node requires long lifetime with limited battery, but retransmission of data is the primary source of energy consumption and reduces the lifetime of sensor node. Therefore, error control scheme of forward error correction (FEC) for wireless sensor networks is necessary [1]. Low-density parity-check (LDPC) codes were first introduced by Gallager in 1962 and rediscovered by Mackay and Neal in 1996 and come into the spotlight for next generation communication system[2][3]. The LDPC codes construction can be categorized into Mackay's random construction and sub-block based structured construction. The Mackay's random construction LDPC codes show very good performance. However, this construction is computationally intensive implementation because of large memory requirement. The sub-block based structured construction T.-h. Kim et al. (Eds.): FGCN 2011, Part II, CCIS 266, pp. 248–257, 2011. © Springer-Verlag Berlin Heidelberg 2011

Implementation of Energy Efficient LDPC Code for Wireless Sensor Node

249

scheme can be implemented with less complexity than the Mackay’s random construction [4]. In this paper, the efficient quasi-cyclic(QC) LDPC code on ATmega128 based sensor node using sensor network OS platform(SenWeaver OS) is implemented. The efficient QC-LDPC code for wireless sensor networks has small size parity check matrix and can be implemented easily. The encoding scheme of efficient QC-LDPC code is simplified version of Richardson’s encoder and does not require matrix inversion so that this encoding scheme is suitable for sensor node which has limited computation ability. This paper is organized as follows. In section 2, the efficient QCLDPC encode is given. In section 3, the specifications of the sensor node and the OS platform (SenWeaver OS) are introduced. In section 4, the experiment results for the efficient QC-LDPC code are given. Finally, the conclusion is made in section 5.

2

The Efficient QC-LDPC Code

2.1

QC-LDPC Code

For wireless sensor node applications, we consider a subclass of LDPC code, QCLDPC, whose parity-check matrix consists of circulant permutation matrices or the zero matrix. Ii is the Ns × Ns permutation matrix which shifts the identity matrix I to the right by i-times for any integer i, 0 ≤ i ≤ Ns. Let I1 be the Ns × Ns permutation matrix given by 0 0  I1 =  #  0 1 

1 0 " 0 0 1 " 0 # # % #  0 0 " 1 0 0 " 0

(1)

Using this notation parity check matrix H can be defined by

 I s0 , 0  I  s1, 0 H =  I s2 , 0   # I s  ( m −1), 0

I s0 ,1 I s1,1

I s0 , 2 I s1, 2

I s2 ,1

I s2 , 2

# I s( m −1),1

# I s( m −1), 2

I s0 ,( n −1)  I s1,( n −1)  " I s2 ,( n −1)   % #  " I s( m −1), ( n −1)   " "

(2)

where si , j is the shift value corresponding position (i, j ) sub-block. This value is one of the {0, 1, 2, …, Ns-1} for nonzero sub-block and si , j = − 1 for zero matrix. The size of H is mNs × nNs. [4].

250

S.-M. Choi and B.-H H. Moon

2.2

Encoding of LDPC Codes

Block type LDPC encodin ng process has Richardson’s encoding algorithm if M × N Parity check matrix is divid ded into the form

A B T  H =  C D E 

(3)

where A is (M-l) × (N-M), B is (M-l) × l, T is (M-l) × (M-l), C is l × (N-M), D is l × l, E is l × (M-l). All these matrices are sparse and T is a lower triangular w with one along the diagonal [5]. Fig. 1 shows the parity check matrix for the Richardsoon’s encoder.

Fig. 1. Parity check matrix for Richardson’s encoder

Let the codeword c={u, p1, p2} where u denotes the systematic part, p1 and p2 dennote the parity parts, p1 has leng gth l, and p2 has length (M-l). The codeword c satisfies the following equations.

 uT  A B T   T  T HcT =    p1  = 0 C D E   T   p2  Au T + Bp B 1T + Tp2T = 0

(− ET

−1

)

(

)

A + C u T + − ET −1 B + D p1T = 0

(4)

(5)

Implementation of Energy Efficient LDPC Code for Wireless Sensor Node

251

Let Φ = − ET −1 B + D and d assume that Φ is nonsingular. Then, we can obttain parity bits as the following equations

( (Au

)

p1T = − Φ −1 − ET −1 A + C u T p =−T T 2

2.3

−1

T

+ Bp

T 1

)

(6)

The Efficient QC-L LDPC Code

The encoding scheme of efficient e QC-LDPC code is to simplify the Richardsoon’s encoder. The goal is to elim minate the matrix inversion of Φ −1 and T −1 in equationn (6). If Φ and T are identity y matrixes, the encoder can be simplified as shownn in equation (7). Also, the sim mplified encoder does not require matrix inversion and encoding can be done without sub-matrix D and T. This will reduce amountt of computations required to en ncode at wireless sensor nodes. Fig. 2 shows the simpliffied Richardson’s encoder.

p1T = − (− EA + C )u T

(

p2T = − Au T + Bp1T

)

(7)

Fig. 2. The simplified Richardson’s encoder

The parity check matrix forr the simplified Richardson’s encoder is constructed byy 3 × 6 sub-matrices as shown n in equation (8).

 A1,1  H =  A2,1  A3,1 

A1, 2

A1, 3

A1, 4

A1, 5

A2, 2

A2,3

A2, 4

A2, 5

A3, 2

A3, 3

A3, 4

A3,5

A1, 6   A2, 6  A3, 6 

(8)

In equation (8), to mak ke T matrix as identity matrix, let A1,5 = A2,6 = I and

A1, 6 = A2,5 = I −1 = 0 . To makke Φ as an identity matrix from equation (9), equattion (10) has to be satisfied.

252

S.-M. Choi and B.-H H. Moon

Φ = − ET −1 B − D  A1, 4  A3, 6   + A3, 4  A2, 4  = A3,5 A1, 4 + A3, 6 A2, 4 + A3, 4 = I

[

]

= A3,5

(9)

Also, if we let A3, 4 = I wh hich corresponds to matrix D as identity matrix, equattion (10) is obtained.

A3,5 A1, 4 + A3,6 A2, 4 = 0

((10)

Assume A3,5 A1, 4 = A3,6 A2, 4 , then Φ becomes identity matrix. In order to satiisfy equation (10), we let A1, 4 in B matrix and A3, 6 in E matrix as an identity mattrix. a equivalent matrixes. The parity check matrix for the And, let A3,5 and A2, 4 are efficient QC-LDPC code is shown in Fig. 3.

Fig. 3. Parity y check matrix for the efficient QC-LDPC code

In this paper, we propose 96 × 192 parity check matrix that is made by shifting thee 32 × 32 identity matrix as shown in the following equation (11). I H =  I1  I13

I

I

I

I

I3 I11

I5 I9

I7 I

I −1 I7

I −1  I  I 

((11)

Implementation of Energy Efficient LDPC Code for Wireless Sensor Node

253

Fig. 4. Parity check k matrix of 96 × 192 for the efficient QC-LDPC code

Code rate 1/2 Mackay's ran ndom constructed LDPC code with column weight 3 and code length 192 has six cycle-4. c However the efficient QC-LDPC code has zzero cycle-4 and girth of 6. The parity check matrix for the efficient 96 × 192 QC-LDPC C is shown in Fig. 4

3

Specifications off Sensor Node and OS Flatform

We have implemented an energy e efficient QC-LDPC code on an ATmega 128 baased sensor node as shown in Fig. F 5. The detailed specifications of the sensor node are shown in Table 1. The exteended 32Kbyte SRAM by Chiplus is used to overcome the shortage of the memory during the encoding and decoding of QC-LDPC. T The CC2420 RF module is useed to transmit the LDPC coded message. The multi-laayer chip antenna for the freq quency range between 2400-2455MHz with 100 M MHz bandwidth is used. The sensor node uses SeenWeaver OS that is developed by the UTRC(Ubiquittous Technology Research Centter) in Daegu University [6]. The features of SenWeaaver OS are as following    

Priority based schedulling/multi-threading Vertical and horizontaal layered architecture Provide ANSI C based d API(Application Program Interface) Semantic modular architecture (energy efficiency by dynamic softw ware reprogramming)  Provide multiprogram mming like PC that operate a number of API withhout alteration  Provide hierarchical reconstruction of API  Provide abstraction and a automatic code generation of variable sensor nnode hardware. The software architecture for f sensor nodes designed such that a layer modular iis a separate block. This allowss easy replacement a block according to hardware withhout changing upper layers. The SenWeaver OS architecture overview is shown in Fig. 66.

254

S.-M. Choi and B.-H H. Moon

Fig. 5. Sensor nodee used for the efficient QC-LDPC code implementation

Fig. 6. SenWeaver OS architecture overview Table 1. Detailed specifications of a sensor node

MCU

Model ATmegaa128L

Ext. SRAM RF module

CS18LV V02563 CC2420 0

Antenna

SWBBL L1

Specification 8bit processor, 128Kbytes flash, 4Kbyte SRAM 32Kbytes SRAM 2.4GHz, IEEE 802.15.4/ZigBee-ready RF transceiver Multilayer chip antenna 2400~2485MHz, 100MHz BW

Implementation of Energy Efficient LDPC Code for Wireless Sensor Node

4

255

Experiment Results

As shown in Fig. 6, the transmitter and the receiver are located on the ceiling to ensure the line of sight communication. By monitoring the 2.43GHz frequency band, no other source of interference is observed. In order to measure performance of the efficient QC-LDPC code, 15,000 12byte long MAC frames as shown in Fig. 7 are transmitted.

Fig. 7. Experiment environment of performance measurement

Fig. 8. LDPC coded simple MAC frame

The experiment results for the uncoded and the efficient QC-LDPC coded cases are summarized in Table 2. The transmitting power is varied from -7 dBm to -25 dBm.. Table 2. Measured BER performance of efficient QC-LDPC code TX power -7 dBm -10 dBm -15 dBm -19.2 dBm -25 dBm

Uncoded case -6 3.4787 × 10 -5 8.5301 × 10 -4 8.3478 × 10 -3 7.0767 × 10 -2 1.6713 × 10

Efficient QC-LDPC coded case 0 -5 5.7418 × 10 -4 4.6648 × 10 -3 5.0378 × 10 -2 1.2139 × 10

256

S.-M. Choi and B.-H. Moon

Given the packet length and BER performance, we calculate the PER (Packet Error Rate) to be

Pep = 1 − (1 − BER ) L

(12)

where Pep is PER, L is packet length[7]. The calculated PER using equation (12) is shown in table 3. Table 3. PER performance analysis of efficient QC-LDPC code TX power -7 dBm -10 dBm -15 dBm -19.2 dBm -25 dBm

Uncoded case -4 3.3390 × 10 -3 8.1558 × 10 -2 7.7043 × 10 -1 4.9428 × 10 -1 8.0171 × 10

Efficient QC-LDPC coded case 0 -3 5.4971 × 10 -2 4.3804 × 10 -1 3.8421 × 10 -1 6.9040 × 10

The mean number of transmission, R, required for success can be calculated as ∞

R = (1−Pep )  ( k + 1) Pepk = k =0

1 1 − Pep

(13)

Also, the gain of mean number of transmission Rgain is given by

Rgain =

Rcoded Runcoded

(14)

Where Runcoded is mean number of transmission for uncoded case, Rcoded is that of the efficient QC-LDPC coded case. Table 4 shows the mean number of transmission using equations (13) and (14). Table 4. Mean number of transmission for success TX power

Uncoded case

-7 dBm -10 dBm -15 dBm -19.2 dBm -25 dBm

1.0003 1.0082 1.0835 1.9774 5.0430

Efficient QC-LDPC coded case 1.0000 1.0055 1.0458 1.6239 3.2300

Rgain 0.0003 0.0027 0.0348 0.1787 0.3595

It is shown that the implemented QC-LDPC code showed limited gain for the TX powers between -7dBm and -10dBm. When the TX powers are between -19.2dBm

Implementation of Energy Efficient LDPC Code for Wireless Sensor Node

257

and -25dBm, the implemented QC-LDPC code obtained about17.9% and 36% gain over uncoded case, respectively.

5

Conclusion

In this paper, we propose FEC using efficient QC-LDPC code for the error control in the wireless sensor networks. The efficient QC-LDPC code has small size parity check matrix and easy implementation scheme. The encoding of the efficient QCLDPC code that constructs sub-matrix T and Φ of Richardson’s encoder into identity matrix and requires less computation compare to the Mackay’s random constructed LDPC code. We implemented the efficient QC-LDPC code on ATmega128 based sensor node using sensor network OS platform (SenWeaver OS) and measured the BER performance at TX powers between -7dBm to -25dBm. Also, we calculated the packet error rate and the mean number of transmission required for successful transmission. It is shown that the implemented QC-LDPC code obtained performance gain in the mean number of transmission about 17.9% and, 36% gain over the uncoded case for the transmission powers of -19.dBm and -25dBm, respectively. Acknowledgments. This research was financially supported by Daegu University Research Grant, 2011.

References 1. Jeong, J., Tien Ee, C.: Forward error correction in sensor Networks, U.C.Berkeley Technical Report (May 2003) 2. Gallager, R.G.: Low-Density parity-Check Codes. IRF Trans. on Info. Theory 8, 21–28 (1962) 3. MacKay, D.J.C., Neal, R.M.: Near Shannon limit performance of low density parity check codes. Electron, Lett. 32(18), 1645–1646 (1996) 4. Fossorier, M.P.C.: Quasi-Cyclic Low Density Parity Check Codes from Circular Permutation Matrices. IEEE Trans. Information Theory 50, 1788–1794 (2004) 5. Richardson, T.J., Urbanke, R.L.: Efficient Encoding of Low-Density Parity-Check Codes. IEEE Trans. IT 47, 638–656 (2001) 6. Kim, T.-H., Kim, H.-C.: Software Architecture for Highly Reconfigurable Sensor Operating System. Journal of IEMEK 2(4), 242–250 (2007) 7. Lettieri, P., Fragouli, C., Srivastava, M.: Low power error control for wireless links. In: Proceedings of 3rd annual ACM/IEEE Intl. conference on Mobile computing and networking (MOBICOM) (1997)

A Multi-layered Routing Protocol for UWSNs Using Super Nodes Abdul Wahid, Dongkyun Kim*, and Kyungshik Lim Kyungpook National University, Daegu, Korea [email protected], {dongkyun,kslim}@knu.ac.kr

Abstract. Underwater Wireless Sensor Networks (UWSNs) have peculiar characteristics such as high propagation delay, limited bandwidth and high error rates. Therefore, communication protocols for UWSNs are highly required to cope with them. Specifically, the design of an efficient routing protocol for UWSNs is one of the most important challenges. The routing protocols can take advantage of the localization of sensor nodes. However, the localization itself is a crucial issue in UWSNs, which remains to be solved yet. Hence, the design of a non-localization based routing protocol is a preferable alternative. In this paper, we therefore propose a non-localization based routing protocol named MRP (Multi-layered Routing Protocol) for UWSNs. In MRP, the concept of super nodes is employed. The super nodes are the nodes having high capacity such as high energy and transmission power. Our proposed protocol, MRP, consists of two phases: layering and data forwarding phases. During the layering phase, multiple layers are formed by the super nodes, whereas, in the data forwarding phase, data packets are forwarded based on these layers. Based on the simulations using NS2, we observe that our proposed routing protocol, MRP, contributes to the performance improvements. Keywords: Underwater wireless sensor networks, routing, super node.

1

Introduction

Underwater Wireless Sensor Networks (UWSNs) have attracted much research attention both from academia and research community due to the need of a number of applications e.g. harbor inspection, ecological monitoring, oil/gas spills monitoring, homeland security etc. Typically, in UWSNs, acoustic signals are employed as a physical media because of the poor performance of radio and optical signals in water. The radio signals propagate long distance at extra low frequencies (30 – 300 Hz), which requires large antennas and high transmission power. The optical signals do not suffer from such high attenuation, but are affected by scattering. Due to the employment of the acoustic signals, UWSNs have some unique challenges which are different from the terrestrial wireless sensor networks. The acoustic signals have large propagation delay (i.e.1500 m/sec), limited bandwidth (i.e. l, l’ is aggregated data size for transmitting from a cluster head j to sink node. Total energy consumption in a round in D-LEACH may be written as preclustering steady _ cluster E(total + E(setup round ,i , j ) ≈ E( round ) round ,i , j ) + E( i , j )

4

(16)

Performance Evaluation

We enumerate the details of simulation parameters in Table 1. Most parameters except node distribution, several types of packet sizes and the location of sink node are identical to those of LEACH. Table 1. Simulation parameters Parameters

Unit

Values

Area for WSN

Meter2

100m × 100m

The number of initial sensor nodes

Integer

100

The location of sink node

(x, y)

(50, 350)

Eelec, Elocal-density, ESchedule, Edata-aggr., EDecision-CH, EDecison-nonCH, Eselect-CH,

nJ/bit

50

eamp

pJ/bit/m2

100

l (normal packet size)

Byte

500

ldisco., lCH-ADV, lJoin, lSchedule

Byte

25

Initial residual energy

Joule

1

The ratio of CHs to all of the nodes

Percentage (%)

5

Sensor node distribution model

Percentage (%)

uniform distribution, changes from 30% to 90% by 10%

326

J.-S. Kim and T.-Y. Byun

It is important to adjust the density of node for measuring the energy consumption and the lifetime of network in D-LEACH and LEACH. In addition to uniform distribution of the nodes, we add a variety of node distribution scenarios, where some nodes among all of the nodes are uniformly distributed placed across area for WSN. To measure performances according to varying local node density, some nodes are intentionally placed in specific area on WSN, and a variety of the node distribution scenarios are used in our simulation. We evaluate some performance metrics in both D-LEACH and LEACH as follows • The number of executions as cluster head per node • The ratio of round of pre-clustering to total rounds over WSN lifetime • The progress of live nodes according to changes in local node density • The progress of active nodes considering local node density Fig. 2 shows the progress of the number of executions as cluster head per node in both D-LEACH and LEACH. As the figure implies, the gap between D-LEACH and LEACH significantly increases when nodes over 70% among total live nodes are excessively placed on specific area. This is caused by extended lifetime in D-LEACH compared with that in LEACH, so each node generally have more chance as cluster head in D-LEACH.

Fig. 2. The number of executions as cluster head per node over network lifetime according to varying nodes distribution

The ratio of round executing pre-clustering to total rounds is shown in Fig. 3. Preclustering stage is executed in 10% rounds to total rounds on an average regardless of various nodes distribution. In LEACH, as the local density increases, network energy is abruptly exhausted due to increase of energy consumption of cluster heads after a node firstly dies, thereupon, network lifetime may be shorten. However, network lifetime in D-LEACH can remarkably be extended in comparing with that in LEACH[9]. So, the period from first-node-die to last-nod-die is greatly extended in comparing in that in LEACH as shown in Fig. 4. In D-LEACH, as the number of participating nodes in area with higher node density decreases, each live node averagely consumes less energy, finally results in prolongation of network lifetime.

A Performance Evaluation of a Novel Clustering Scheme

327

Fig. 3. The ratio of round of pre-clustering to total rounds according to varying the node distribution

(a) LEACH

(b) D-LEACH Fig. 4. The progress of live nodes over network lifetime

328

J.-S. Kim and T.-Y. Byun

The progress of active or participating nodes according to a variety of nodes distribution scenarios is shown in Fig. 5, here, the number of active nodes is measured by 1 round and 50 rounds respectivly. As mentioned above, increase of local node density leads to extension of network lifetime. Especially, as nodes over 60% to total initial nodes are deployed on specific area, which means relatively high local density, prolongation of network lifetime increases in comparing with that in LEACH.

(a) The progress of active nodes in case of sampling by 1 round

(b) The progress of active nodes in case of sampling by 50 rounds Fig. 5. The number of active nodes in clusters according to changes in the node distribution in D-LEACH

A Performance Evaluation of a Novel Clustering Scheme

5

329

Conclusion

In this paper, we mathematically analyzed details of energy consumption in DLEACH, a novel clustering protocol for WSN that minimizes global energy usage by adjusting the number of active nodes considering local node density per each node in a cluster. Also, we showed some performance metrics related with energy consumption of both D-LEACH and existing LEACH. Due to nature of D-LEACH operation, which selects some active nodes among all live nodes by considering local node density, D-LEACH has advantages of prolonging network lifetime.

References 1. Basagni, S.: Distributed Clustering Algorithm for Ad-hoc Networks. In: International Symposium on Parallel Architectures, Algorithms, and Networks, I-SPAN (1999) 2. Kwon, T.J., Gerla, M.: Clustering with Power Control. In: Proceeding of MilCOM 1999 (1999) 3. Amis, A.D., Prakash, R., Vuong, T.H.P., Huynh, D.T.: Max-Min D-Cluster Formation in Wireless Ad Hoc Networks. In: Proceedings of IEEE INFOCOM. IEEE Press (2000) 4. Heinzelman, W.R., et al.: Energy-Efficient Communication Protocol for Wireless Microsensor Networks. In: Proceedings of the 33rd HICSS 2000. IEEE Press (2000) 5. Banerjee, S., Khuller, S.: A Clustering Scheme for Hierarchical Control in Multi-hop Wireless Networks. In: Proceedings of IEEE INFOCOM. IEEE Press (2001) 6. Chatterjee, M., Das, S.K., Turgut, D.: WCA: A Weighted Clustering Algorithm for Mobile Ad Hoc Networks. Cluster Computing, 193–204 (2002) 7. Bandyopadhyay, S., Coyle, E.: An Energy-Efficient Hierarchical Clustering Algorithm for Wireless Sensor Networks. In: Proceedings of IEEE INFOCOM (2003) 8. Younis, O., Fahmy, S.: Distributed Clustering in Ad-hoc Sensor Networks: A Hybrid, Energy-Efficient Approach. In: Proceedings of IEEE INFOCOM. IEEE Press (2004) 9. Kim, J.-S., Byun, T.-Y.: A Density-based Clustering Scheme for Wireless Sensor Networks. CCIS, vol. 195, pp. 267–276. Springer, Heidelberg (2011)

Performance Analysis of DRAM-SSD and HDD According to the Each Environment on MYSQL Hyun-Ju Song1, Young-Hun Lee1,*, and Seung-Kook Cheong2 1

Dept. of Electronic Eng., Hannam University, Ojeong -dong, Daedeok-gu, Daejon 306-791, Korea [email protected], [email protected] 2 Dept. Principal Member of Engineering Staff.. ETRI, 161, Gajeong-dong, Yuseong-gu, Daejeon, 305-700, Korea [email protected] Abstract. Recently, Users needed storage for processing high-capacity data. Meanwhile, the HDD was used primarily as storage device, but SSD was developed as a fast access device. So, the use of SSD increased a lot of capacity-process. Using a SAN Switch, we tested in order to resolve data process faster. So in this paper, DRAM-SSD and HDD’s performance ability in the data process will be confirmed using TPC-H Benchmark on MYSQL at SAN environment. From the performance analysis results, performance difference of HDD and DRAM-SSD is little when database size is low at san switch connecting with storage. But ad-hoc query process ability difference between DRAM-SSD and HDD increased. Based on these results in a SANbased, DRAM-SSD has a better performance than the HDD. Therefore San is judged to be more effective when dealing with the large amounts of data using a SAN to manage data. Keywords: SSD, HDD, Mysql, TPC-H, SAN.

1

Introduction

Recently, Users needed storage for processing high-capacity data. Meanwhile, the HDD was used primarily as storage device, but the development of SSD is faster access as storage devices and a lot of research is being accelerated. From the research of HDD and SSD, the difference of data I/O processing performance was progressed by comparing performance of storage device of each.[1-3] Basic storage device’s manage data and use of DBMS which provided efficient and convenient way increased. So in this paper, we will compare DRAM-SSD to HDD and confirm showing good performance in data processing, using TPC-H Benchmark through SAN Switch on MYSQL, The process of this paper is as follow. In chapter 2 we will introduce used technology that evaluates performance of HDD and SSD, in chapter 3 we will introduce analysis environment and condition by thinking each storage device and using tool. And in chapter 4 analyzes test results will be shown with conditions described in chapter 3, and chapter 5 is concluded. *

Corresponding author.

T.-h. Kim et al. (Eds.): FGCN 2011, Part II, CCIS 266, pp. 330–337, 2011. © Springer-Verlag Berlin Heidelberg 2011

Performance Analysis of DRAM-SSD and HDD According to the Each Environment

2

Related Study

2.1

MYSQL

331

MYSQL is the relative database management system of open source that uses SQL which is standard database quality language, which is very fast, flexible, and easy to use. MYSQL offers client/server environment, and a server installed MYSQL has MYSQL daemon called mysqld, so client program connects via network by this daemon so that it can control data. [5]

mysqld Client Network

mysqld Database

Client mysqld

MySQL Server

Fig. 1. Performance structure of MYSQL

2.2

SAN

SAN stands for Storage Area network and say move available high-speed network for large-capacity between storage equipment which unrelated distributed sort of Host. i.e., say all together configured network to communication storage component and system component at fiber channel networks. SAN integrate and share storage, and supply high-performance link to data device, and add overlapping-link to storage, and speeding up data backup, and support available-high-clustered systems. SAN have advantage by building implementing a highly available data access, integration resources and management of storage, required window to backup and traffic reduce, occupation solution of host CPU cycle, data retention function through disaster Accepted techniques. [6]

Fig. 2. General

storage server VS SAN

structure

332

2.3

H.-J. Song, Y.-H. Lee, and S.-K. Cheong

TPC-H

TPC stands for Transaction processing performance and is non-profit organization that has multiple hardware and software companies and small number of user organizations. Usually TPC called transaction processing performance evaluation committee, but TPC-alphabet notation tells the benchmark model. TPC becomes standard evaluating processing performance of on-line transaction processing systems. TPC defines transaction processing and benchmarks of database and is used to performance measure of total system including disk I/O and s/w. As a benchmarking tool to measure that how fast it can handle complex SQL, and defines 22 SQL statements and DB schema, and set of data about 1GB. TPC-H benchmark is public performance test that is used SQL that Business-oriented ad-hoc Query and concurrent data modifications made by the combination about large data. Fig 3 showing business environment of TPC-H, impromptu Query and modification transactions was performed at table from multiple users, model situation entered data from database of decision support system from OLTP system. [7]

Fig. 3. TPC-H's business environment

3

Performance Analysis Environment and Conditions

3.1

Performance Test Environment

This paper comparative analyzed the performance of each, which the HDD as a storage device in modern mainly used and the SSD as a storage device in modern society used increasingly. HDD Storage and DRAM-SSD Storage were constructed in conjunction with a SAN switch for test Environment. Test Environment constituted as fig 4, and test subject server formed as Linux CentOS 5.3 version, performance measure tool used TPC-H Benchmark, to use a tool as the database was installed MYSQL 5.0.90 version.

AM-SSD and HDD According to the Each Environment Performance Analysis of DRA

333

Fiig. 4. Test environment configuration

3.2

d Condition Test Procedure and

orage device comparative analyzed to configure SAN Performance of each sto environment using TPC-H Benchmark. Test using the TPC-H Benchmark compaared to performance analysis thrrough total three step procedure with the results. Load T Test is the first step, which maake up database and store to generate data. Next stepp is power test, which was anallyzed by measuring the ability when Single active useer is run the Query. Final step is Throughput Test, which was analyzed by measuring the user is run Query at same time. The performance of eeach ability when Multi active u storage device analyzed thrrough Power TEST and Throughput Test combined results in SAN environment. Data generation and total of eight tables produced to sttore generated data in data storaage of Mysql. Read data in the created file store data whhich get by running stored-codee in the DBMS, confirm insert play time of stored dataa. A n the TPC-H benchmark, which the play-time checkk by total of 22 Query given in running through the Query Browser, total of 22 Query stored for multi active user test through vim command in TPC-H installed the directory. After run total of 22 Querry a T script run to delete set by checking the result. Result of insert execution time and 22Query execution time aand delete execution time calculate value of Power T Test 1. apply in below expression 1

(1) n time of I second query, RI (j, 0) as the running timee of QI (i, 0) is the execution nd the data generated by the size of SF, i.e., the value off the insert and delete unit secon accordingly practice ability of multiple users at the database. When each of Query Q same time make amount of data as table 1 number of users to differ.

334

H.-J. Song, Y.-H. Lee L , and S.-K. Cheong Ta able 1. Conditions for throughput test Datab base capacity

User number

1GB

2

5GB

2

10GB

3

The fixed users perform m as shown in above table1 parallel 22 Query of each. A And time check was expressed in units of sec from first user starts the first Query to last user finish last Query. By substituting s in Equation 2 is to evaluate performance tiime to check on the recipe. (2) In the above expression 2, is performance time, S is number of user, SF shhow amount of data. Through th he result of expression 1 and expression 2, expression 3 has produced result to get final result to analysis performance of each storage device. (3) QphH@Size which is a calculated value from Equation 3 is reflected value of the many features of Query processing p system, these features are performed Queery, Query processing ability when w performed Query, Query performed by multiple ussers simultaneously including alll the processing ability. Thus, QphH@Size shows ad-hoc query capabilities that to haandle hourly capacities of the database.

creation n

Tool (DBGen, QGen)

Database

application

n creation

Queries

Fig. 5. TPC-H H applies to get the results the block Diagram

Performance Analysis of DRAM-SSD and HDD According to the Each Environment

4

335

Test Results

Test results using TPC-H Benchmark comparative analyzed performance at SAN environment in according to described result calculation method in Section 3.2. Some of total 22 Query for test when Database capacity 1GB, 5GB, 10GB, by modifying comparative analysis performance for each size Shorten the time the test case. Table 2. SAN environment HDD performance measurements

HDD

Database capacity

Test result

1GB

5GB

10GB

Power@size

1.0891E-09

3.8770E-30

3.6293E-31

Throughput@size

15.7283

7.372

7.0681

QphH@size

1.3088E-04

5.3461E-15

1.6016E-15

Table 3. Local environment DRAM-SSD performance measurements

DRAM-SSD

Database capacity

Test result

1GB

5GB

10GB

Power@size

9.7911E-09

4.6299E-20

8.8710E-22

Throughput@size

19.3928

34.7933

80.1214

QphH@size

4.3575E-04

1.2692E-09

2.6660E-10

As can be seen through above Table 2 and Table 3, the practical value for comparing the performance of DRAM-SSD Storage and HDD Storage in SAN environment is QphH@size, compared result to only database-capacity of a storage device of each same below picture. QphH@size value that can handle ad-hoc queries, the little difference find out between ad-hoc query handle-capabilities of HDD Storage and DRAM-SSD Storage in low carrying a load 1G, but DRAM-SSD storagere can be seen that much higher which per hour to handle ad-hoc query capabilities more than HDD storage when amounts of data is increasing.

336

H.-J. Song, Y.-H. Lee, and S.-K. Cheong

Fig. 6. SAN environment database capacities of each storage device QphH@size value

5

Conclusion

In this paper, we analyzed the data processing performance of a DRAM-SSD and HDD as a data processing storage using the TPC-H Benchmark on mysql in a san environment. From the performance analysis results, when database Size increased performance compared analysis value QphH@size was increased with TPC-H, but difference of the ad-hoc query processing ability of DRAM-SSD and HDD are increased when compared of each database capacity. Based on these results, using by connecting SAN Switch and DRAM-SSD storage is judged to be effective at large amounts of data I/O required field or applications. Also mentioned in the introduction, SSD has benefits if SSD price is stabled. Using SSD is thought to be effective in case of industry side which required large amounts of data I/O or Media Server, using storage device of computer and notebook computer. In the future, we will be analyzed the power consumption from the TPC-H Benchmark at SAN Environment. Also, SAN Environment and Local Environmental TEST results will be analyzed to process the data.

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(4), 41–47 (2009)

Performance Analysis of DRAM-SSD and HDD According to the Each Environment

337

[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(1), 265–272 (2009) [3] Cheong, S.-K., Jeong, Y.-W., Jeong, Y.-J., Jeong, J.-J.: Input-Output Performance Analysis of HDD and DDR-SSD Storage under the Streaming Workload. Korean Information Technical Academic Society Dissertation, 322–325 (2010) [4] Cheong, S.-K., Ko, D.-S.: Technology Prospect of Next Generation Storage. Korean Information Technical Academic Society Dissertation Summer Synthetic Scientific Announcement Thesis, 137 (2008) [5] Kim, H.: Learn to MYSQL Database Programming, http://www.Young-jin.com [6] Judd, J., Beaucbamp, C.: Building SANs Witch Brocade. Syngess, 9–16 (November 1996) [7] http://www.tpc.org/tpch/spec/tpch2.13.0.pdf

Dynamic Channel Adjustable Asynchronous Cognitive Radio MAC Protocol for Wireless Medical Body Area Sensor Networks Byunghwa Lee1, Jangkyu Yun1, and Kijun Han2,

*

1 The Graduate School of Electrical Engineering and Computer Science, Kyungpook National University, 1370, Sankyuk-dong, Buk-gu, Daegu, 702-701, Korea 2 The School of Computer Seience and Engineering, Kyungpook National University, 1370, Sankyuk-dong, Buk-gu, Daegu, 702-701, Korea {bhlee,kyu9901}@netopia.knu.ac.kr, [email protected]

Abstract. Medical body area networks (MBAN) impose several requirements to the medium access control layer which have various contexts: energy efficiency, QoS providing, reliability. And a cognitive radio (CR) network should be able to sense its environment and adapt communication to utilize the unused licensed spectrum without interfering with licensed users. As CR nodes need to hop from channel to channel to make the most use of the spectrum opportunities, we consider asynchronous medium access control (MAC) protocols to be solution for these networks. The DCAA-MAC protocol presented in this paper has been designed has been designed specifically for wireless body area network with cognitive radio capability. The DCAA-MAC protocol has energyefficiency, low latency, and no synchronization overhead by provide asynchronous and fast channel switching. Analytical models are shown to perform at low energy consumption, to scale well with network size. Keywords: Cognitive radio, Wireless body area network, MAC.

1

Introduction

The convergence of all media and data services is disseminated throughout innovation of wireless communication technologies. In recent years there has been also increasing interest in implementing ubiquitous monitoring system in hospital and house for patient. These monitoring systems are normally observed by on-body sensors that are connected by a controller of the medical body area network (MBAN). The MBAN is a promising solution for eliminating wires, thus allowing sensors to reliably and inexpensively collect multiple parameters simultaneously and relay the monitoring information wirelessly so that clinicians can respond rapidly [1]. MBANs for wireless patient monitoring is an essential component to improving patient outcomes and lowering healthcare costs. *

Corresponding author.

T.-h. Kim et al. (Eds.): FGCN 2011, Part II, CCIS 266, pp. 338–345, 2011. © Springer-Verlag Berlin Heidelberg 2011

Dynamic Channel Adjustable Asynchronous Cognitive Radio MAC Protocol

339

Quality of service is a key requirement for MBANs, and hence the importance of having a relatively clean and less crowed spectrum band. The 2.4Ghz industrial, scientific and medical (ISM) band is not suitable for life-critical medical applications due to the interference and congestion for wireless networks in hospital and house. Cognitive radio (CR) is a novel technology, which improves the spectrum utilization by seeking and opportunistically utilizing radio resources in time, frequency and space domains on a real time basis. Cognitive radio is an emerging technology to address critical challenges of spectrum efficiency, interference management, coexistence and interoperability associated with current and future wireless networks. Cognitive wireless communications and networks have high potentials to bring enormous economic benefits to both customer and operators. In recent years, cognitive wireless communications and networking attract wide interests from both academia and industry. These features and requirements raise the demand of CR for MBAN implementation. We explore novel technique to provide CR MAC protocol for WBAN.

2

The Dynamic Channel Adjustable Asynchronous MAC Protocol in Wireless Medical Body Area Sensor Networks as Enabler for CR (DCAA-MAC)

In this section, we will present a new CR MAC protocol for MBAN that enables efficient spectrum sharing by borrowing licensed spectrum. The protocol is designed to protect primary users (PUs) from CR devices’ interference since borrowing licensed spectrum has to be protected. Basically, the DCAA-MAC based on asynchronous MAC paradigm to provide energy efficiency, low latency, reconfigurability to MBAN. At initialization time, each node scans channel to join the network. If it fails to find the network, the device selects the channel which has best condition to make the network. (E.g. the least frequently used channel, SNR, etc.) To minimize listen cost, every node in the network wakes up only in check interval which sensing communication channel. Each node goes to sleep and wake up periodically and independently of the others. When a node wants to communicate with a neighbor, it first send a preamble which contains address of destination node that lasts an entire duty cycle, and receive the ACK message from the destination node which hears preamble, just after that, it sends the data. The receiver detects the preamble which has same address of its when it wakes up, it remains awake in order to receive the data. After data transmission is done, nodes which transmitted and received data go to the sleep mode. Advantage of this technique is that it works in a completely unsynchronized environment.

340

B. Lee, J. Yun, and K. Han

Start Check Interval Duration Time Over

CCA

Busy Target Address Checking Target Address == My Address

Clear Target Address != My Address Unidentified

Sleep Target Address != My Address Listen ≤ 1 Check Interval Unidentified

ACK Message Transmission

Control Channel Open

Data Reception

Listen ≤ 1 Check Interval

CSI Reception Failed CSI Reception Success

DACK Message Transmission

CSACK Message Transmission

CSACK Reception Failed CSPreamble Transmission

CSACK Reception Success Channel Switching

Fig. 1. Activity diagram of DCAA-MAC

If when a node detects interferences on its current channel, either due to PU or noise appearance, it switches to another one so that communications can be maintained. The operation over multiple communication channels allows to maintain communications even when facing PUs appearance or interference. When a node detects signal which has over the power of threshold during CCA with fast sensing period on its current channel, listen to signal to find the address of destination during maximum one check interval. If it finds the address of destination and the address is same with own address, it sends ACK message to source of preamble, else it sleep to next check interval. If it does not find any address type, it determines that PU or interference appears and try to switch to another channel for PU protection and QoS provision. That is provided CR capability which based on result of energy detection during CCA with fast sensing period in MAC layer. Spectrum sensing is an important requirement for the realization of CR networks. Feature detection and energy detection are the most commonly used for spectrum sensing in CR networks. Feature detection determines the presence of PU signals by extracting their specific features. Although feature detection is most effective for the nature CR networks, it is

Dynamic Channel Adjustable Asynchronous Cognitive Radio MAC Protocol

341

computationally complex and requires significantly long sensing time [2]. Energy detection is optimal to detect the unknown signal if the noise power is known. In the energy detection, CR users sense the presence/absence of the PUs based on the energy of the received signals. It is difficult to provide QoS with channel switching with result of feature detection because of feature detection’s these characteristic. Therefore, channel switching is performed with only result of energy detection in DCAA-MAC. After determine channel switching, node switch the current channel to control channel. And then listen during maximum one check interval because try to find another node which broadcast channel switching preamble (CSPreamble). If receive the preamble which have the channel switching information, node transmits the channel switching ACK (CSACK) message to the source node and switch the channel as follow the channel switching information. If there are not any signals on control channel during one check interval, nodes transmitCSPreamble that include channel switching information which has channel number to move and after that receive CSACK message from destination node, switch the channel. These fast channel switching mechanisms provideQoS for MBAN by low latency. Fig.2 shows the example of the proposed protocol.

Fig. 2. Example of the proposed MAC protocol for CR.

3

Performance Analysis

3.1

System Model

Performance analyses are based on some assumptions. We define that a radio can be in one of three steady states: Transmission, Reception and Sleep. Furthermore, four

342

B. Lee, J. Yun, and K. Han

transient states are defined: SetupTx, SetupRx, SwitchTxRx, SwitchRxTx. The time spent in a transient state is a TTrState, the power consumption in each state is PState and the energy cost of a transition from one steady state to another is ETrState. Table 1 shows the parameters we have used for the performance analysis in detail. Table 1. Parameters for performance analysis Parameter

Value

PTx, PSetupTx PRx, PSetupRx PSleep PSwTxRx, PSwRxTx TSwTxRx TSetpuTx TSetupRx TCCA TW TM TPreamble TACK TCS

56.4 mW 25.5 mW 0.06 mW 54.3 mW 160 μs 12μs 192μs 128μs 500 ms 6.9 ms 128 μs 1.02 ms 61.5 μs 250 kbps

Bit rate

We assume that a network has N devices. The network is fully connected and each node can directly communicate with all the others. Each node generates traffic according to an exponential distribution with parameter = 1/ : the mean interval between two packets is L seconds. Each node receives as much data as it is transmitting. 3.2

Power Consumption

In case of CSMA protocol, power consumption is evaluated as follows: =

1 (1)

On average, during a time L, a radio using the CSMA must send a packet, receive a packet and sleep the rest of the time.

Dynamic Channel Adjustable Asynchronous Cognitive Radio MAC Protocol

343

In DCAA-MAC makes use of wake-up preambles. A node using this protocol must perform one carrier sensing, transmit data, receive data and sleep in the rest time of one period. This shows to the following expression: ,

= where: = =

1

1 =

1

(2)

=

Traffic adaptation shows how power consumption evolves with an increase of data rate.

Fig. 3. Protocol power consumption depending on traffic rate

Fig.3 shows the power consumption and energy efficiencies of CSMA and our DCAA-MAC protocols as a function of the mean time L between packets. CSMA and DCAA-MAC exhibit the same qualitative behavior. DCAA-MAC scale remarkably well and never far from the CSMA power consumption.

344

B. Lee, J. Yun, and K. Han

Network density is more important when considering node mobility. This can be the case for body area networks. 3.3

Delay

A simply analysis of average delay of protocol can be made by using standard results of queueing theory. We assumed that packets are generated at each sensor device according to a Poisson process of parameter λs, and the sum of these N independent and identically distributed random point processes also follows a Poisson distribution, of parameter λ=λs, If packets are constant size, the service time is deterministic and the queueing model M/D/1 can be applied. The average delay is given by: = where

= ,

,

is the service time

(3)

For CSMA, -1=TM+2TSIFS is approximation of the service time. And for DCAAMAC, μ‐1=TW/2because the packet can arrive at the MAC layer at any time.

Fig. 4. Average packet latency for DDCA-MAC and CSMA

Fig.4 shows bilogarithmic scale the average packet delay as a function of the average number of packets per node, for various network densities (from 5 to 100).

Dynamic Channel Adjustable Asynchronous Cognitive Radio MAC Protocol

4

345

Conclusion

In this Paper, we have proposed a dynamic channel adjustable asynchronous MAC protocol as enabler for CR in wireless medical body area sensor network. In the proposed approach, each node goes to sleep and wake up periodically and independently of the others. Channel switching is performed not with result of feature detection but result of energy detection because QoSproviding. Analytical models are shown to perform at low energy consumption, to scale well with network size and to allow coexistence with simultaneously operating independent networks. Acknowledgments * This work was supported by the second phase of the Brain Korea 21 Program in 2011. * This work was supported by National Research Foundation of Korea Grant funded by the Korean Government.

References 1. Petel, M., Wang, J.: Applications, Challenges, and Prospective in Emerging Body Area Networking Technologies. IEEE Wireless Commun. 17(1), 80–88 (2010) 2. Hur, Y., Park, J., Woo, W., Lee, J.S., Lim, K., Lee, C.-H., Kim, H.S., Laskar, J.: A Cognitive Radio (CR) System Employing A Dual-Stage Spectrum Sensing Technique: A MultiResolution Spectrum Sensing (MRSS) and A Temporal Signature Detection (TSD) Technique. In: Proceedings of the IEEE Globecom 2006, pp. 1–5 (2006)

A Multiple-Metric Routing Scheme for QoS in WMNs Using a System of Active Networks Jangkyu Yun1, Byunghwa Lee1, Junhyung Kim1, and Kijun Han2, 1

*

The Graduate School of Electrical Engineering and Computer Science, Kyungpook National University, 1370, Sankyuk-dong, Buk-gu, Daegu, 702-701, Korea 2 The School of Computer Seience and Engineering, Kyungpook National University, 1370, Sankyuk-dong, Buk-gu, Daegu, 702-701, Korea {kyu9901,bhlee,jhkim}@netopia.knu.ac.kr, [email protected]

Abstract. Wireless mesh networking is emerging as an important architecture for the future generation of wireless communications systems. The challenging issue in WMNs is providing Quality of Service (QoS). So, this paper proposes a multiple-metric routing scheme for QoS in WMNs using a system of active networks. The Active Network paradigm offers the attractive capability of being able to carry executable payloads that can change the characteristics of a given platform. In other words, network nodes not only forward packets, but also perform customized computation on the packets flowing through them. It provides a programmable interface to the user. Keywords: wireless mesh networks, active networks, AODV.

1

Introduction

Wireless mesh networks (WMNs) will play an increasingly important role in futuregeneration wireless mobile networks. A WMN normally consists of mesh routers and clients, and can be independently implemented or integrated with other communications systems such as conventional cellular networks [1]. WMNs are characterized by their dynamic self-organization, self-configuration, and self-healing to enable quick deployment, easy maintenance, low cost, great scalability, and reliable services, as well as enhanced network capacity, connectivity, and resilience[2]. A method of providing QoS is the key technology for traffic management in WMNs. There have been various researches conducted on how to provide QoS in WMNs. A previous research for providing QoS has been studied with routing metrics such as hop count, Expected Transmission Rate (ETX) and Expected Transmission Time (ETT). The hop count is the base metric and is a simple measure of the number of hops between the source and destination of a path [3]. The ETX is a measure of link and path quality and the ETT is the Expected Transmission Time of a data transmission in a direct link [3][4][5]. Although there have been many studies on how to provide QoS, it is not suitable to offer QoS in a dynamic environment. *

Corresponding author.

T.-h. Kim et al. (Eds.): FGCN 2011, Part II, CCIS 266, pp. 346–353, 2011. © Springer-Verlag Berlin Heidelberg 2011

A Multiple-Metric Routing Scheme for QoS in WMNs

347

In this paper, an active network architecture is used to provide QoS in WMNs. Active Networking is primarily a DARPA funded project focusing on mechanisms, applications, and operating systems research to develop a reconfigurable network infrastructure. The active network paradigm offers the attractive capability of carrying executable payloads that can change the characteristics of a given platform. The key point of an active network architecture is the active node construction because the Active packets are operating on the active nodes. The active nodes architecture, which was put forward by DARPA, depicted the data processing flow. The logical constructions of active nodes involve three parts: Node OS, Execution Environment and Active Application. The Node OS is similar to the general operating kernel which through a fixed interface to provide resources and render services for the execution environment. The Execution Environment is a transparent, programmable space and is unrelated to the platform. It operates in each active node and user terminal node; a multi-execution environment could operate on the same active node at the same time. The execution environment provides various network application interfaces for higher level applications. The Active Application is a series of user-defined procedures. By executing the network API provided by the executive environment, the necessary resources can be obtained when running a program. Finally, a customized function can be realized. The active node (routers, etc.) in Active Networks are not only data forward packets, but also perform customized computation on the packet flowing through them. The active node operating codes are initially contained in the active node. This code can also be dynamically inserted into the forwarding packet to configure them according to the needs of the applications in execution. This way, packets have the capacity of carrying not only data but also the code to be executed in remote nodes. Therefore, the user has the possibility of “programming” the network, providing that the programs to be used by the routers and switches to execute their computations are available. In an active network, the difference between network internal nodes (routers, switches, etc.) and user nodes is tenuous, since both are able to perform the same computations. Hence, the user can view the network as a part of his/her application and can adapt the network to obtain the best performance of his/her application [6].

2

Proposed Scheme

It has already been mentioned that, many performance metrics, such as ETX and ETT so on have been considered for QoS in WMNs. Since each individual routing metric considers some features, it is difficult to satisfy all the requirements of WMNs by using a single metric. Therefore, it is proposed that a multiple-metric routing scheme be used. An active networking technique for QoS was grafted onto an existing routing protocol. The basic routing protocol of the proposed scheme is the Ad hoc Ondemand Distance Vector (AODV). In this paper, this protocol is referred to as the

348

J. Yun et al.

Active AODV. A routing table and a route discovery process make the difference between these two protocols. The modification performed is the addition of a QoS Type field to the routing table. It determines what type of QoS to use such as reliability, delay, interference and so on. And RREP, RREQ and RERR packets are also modified. A QoS field is added to reserve the fields of these three packets, since they have reserved fields. The QoS field also determines what type of QoS is required for the application. Nodes fill the QoS field when sending the route discovery message. Nodes compare just the destination ID field when the process route discovery is in pure AODV. However, nodes compare not only the destination ID but also the QoS field of the discovery message. Therefore, each routing metric is used individually to select the demanded path in the Active AODV. In other words, the Active AODV determines where the suitable path is. For example, it determines the best path for the highest success rate for a packet, which will result in providing successful service. Further, it determines the fastest path for the delay demanded packet. There are many QoS types for applications. Nevertheless, this paper considers only five metrics: HOP, ETX, ETT, AB and EI. AB and EI mean Available Bandwidth and Expected Interference, respectively. The AB of the link is defined as following equation 1. ABl = Bl − UBl

(1)

Bl and UBl denote link bandwidth and using bandwidth of link, respectively. Also, The EI of the link is defined as following equation 2. EI l = ETTl × N l

(2)

Nl denotes the neighbor number of the link. On the active protocol level, several solutions for active networks have been proposed so far. The base protocol of this scheme is the Active Network Encapsulation Protocol (ANEP). Further, payloads are encapsulated within ANEP packets and ANEP packets are encapsulated within an IP packet as shown in Fig. 1. 



+2*GCFGT 8GTUKQP #0'2*GCFGT





+2 *GCFGT 4QWVGT#NVGT1RVKQP (NCI

6[RG+&

#0'2*GCFGT .GPIVJ

#0'22CEMGV.GPIVJ 1RVKQP

#1&82CEMGV 44'3 44'24'44

Fig. 1. Structure of encapsulated packet



A Multiple-Metric Routing Scheme for QoS in WMNs

349

All active packets are demultiplexed by ANEP. The Type ID field of the ANEP header indicates the evaluation environment of the packet. The active node should evaluate the packet in the proper environment. Yet, traditional nodes cannot demultiplex active packets, so it just forwards the packet. This means this scheme can operate on both networks Fig. 2 shows the active node structure. When the node receives a packet, node OS it dispatched to a suitable EE. An active packet is dispatched to ANEP, but a normal packet is dispatched to IPv4 of IPv6. After that, an active packet is demultiplexed by the ANEP and the AODV packet is processed on the Active AODV

...

Active AODV AA

ANEP

IPv4

IPv6

...

Management EE

EE

Dispatch

Security Enforcement Engine

Transmit

Store ...

.. .

Node OS

Input

Output

Policy

channels

channels

DB

Fig. 2. Structure of active node

Fig. 3 shows an example of the proposed multiple-metric routing. B C

A

D Routing Table of node C Destination

Next hop

Metric

QoS Type

A

D

4

2

A

B

5

3

…

. . .

Fig. 3. An example of a routing table

This is the routing table of node C. There are two entries to node A. Hence, node C forwards a packet to node D for the packet success rate. On the other hand, node C forwards a packet to node B for a delay required packet. It means every node has many entries to the same destination. The QoS Type field makes it possible.

350

3

J. Yun et al.

Evaluation

A simulation was conducted based on ns-2 to evaluate the Active AODV, which uses multiple metrics (HOP, ETT, ETX, AB, EI). In addition, a pure AODV and an Active AODV was used as a routing protocol. The simulations tried two ways of measuring the performance of the Active AODV according to the bandwidth in the networks. The first simulation used set nodes with varied bandwidths (1~4 Mbps) and the second simulation had a fixed bandwidth (1.5 Mbps). The topology is a 4 x 4 Grid, the number of nodes with regular deployment is 16 and an IEEE 802.11 which is a MAC protocol, was used in the simulation. Table 1 shows the detailed parameters in the simulations.

Table 1. Parameters used for simulation Parameters

Values

Topology

4 x 4 Grid

The number of node

16

The distance between nodes

90m

Radio rage

100m

MAC Protocol

IEEE 802.11

Packet size

512 Byte

Bandwidth

1~4 Mbps

Time of total simulation

200s

In the first simulation, the routings used in each HOP, ETX, ETT, AB and EI were compared according to nodes with increasing bandwidths from 1 to 4 Mbps.

Fig. 4. Packet delivery ratio (1~4 Mbps)

A Multiple-Metric Routing Scheme for QoS in WMNs

351

Fig. 4 and Fig. 5 present the result ratio and the delay of packet delivery. It shows that the experimental result of routing with ETX, ETT, AB and EI is better than HOP when pairs of routing nodes were increased in Fig. 4. However, AB and EI obviously decrease when there are 11 pairs of routing nodes. It can be observed that AB and EI are worse than other metrics at high loads. Fig. 5 illustrates that the performance of routing with ETT is greatly enhanced and the packet delivery delay is significantly reduced. It can be seen that routing with ETT provides an efficient path because it considers the packet delivery delay of a packet.

Fig. 5. Packet delivery delay (1~4 Mbps)

In the second simulation, the performance of routing with nodes with fixed 1.5 Mbps was tested according to the pairs of routing nodes in Fig. 6 and Fig. 7.

Fig. 6. Packet delivery ratio (1.5 Mbps)

352

J. Yun et al.

Fig. 6 represents similar results in the performance when pairs of nodes are 10 but routing using AB and EI sharply declined. Fig. 7 shows that delay of all routing rose according to increasing loads. Also, routing using ETX and ETT showed improved performance when pairs of routing node are 8. It was observed, through this simulation, that routing is influenced by metrics. There is a noticeable difference in the routing performance when it is measured in metrics. It has been proven that efficient routing is possible when a property metric is used according to the QoS of data since there is a varied bandwidth in WMNs. Therefore, a path can be selected by considering the success rate when the data delivery ratio is important or a path can be selected by considering delay when data delivery delay is important.

Fig. 7. Packet delivery delay (1.5 Mbps)

4

Conclusion

In this paper, the multiple-metric routing scheme for QoS in WMNs has been proposed by using a system of active networks. An active networking technique was grafted for QoS onto the existing AODV routing protocol. Moreover, the routing table and the routing discovery process of the AODV were modified. This makes the Active AODV determine where a suitable path is for the required QoS application. Plus, the simulation result proved this. The proposed scheme is especially useful in multiple-metric required environments in WMNs. Acknowledgments * This work was supported by the second phase of the Brain Korea 21 Program in 2011. * This work was supported by National Research Foundation of Korea Grant funded by the Korean Government.

A Multiple-Metric Routing Scheme for QoS in WMNs

353

References 1. Akyildiz, F., Wang, X., Wang, W.: Wireless Mesh Networks: A Survey. Elsevier Comp. Networks 47(4), 445–487 (2005) 2. Ci, S., et al.: Self-Regulating Network Utilization in Mobile Ad-hoc Wireless Networks. IEEE Trans. Vehic. Tech. 55(4), 1302–1310 (2006) 3. Gowrishankar, S., Sarkar, S.K., Basavaraju, T.G.: Performance analysis of AODV, AODVUU, AOMDV and RAODV over IEEE 802.15.4 in wireless sensor networks. Computer Science and Information Technology, 59–63 (2009) 4. Yang, Y., Wang, J., Kravets, R.: Designing routing metrics for mesh networks. WiMesh (2005) 5. Jiang, W., Zhang, Z., Zhong, X.: High Throughput Routing in Large-Scale Multi-Radio Wireless Mesh Networks. In: Wireless Communications and Networking Conference 2007 (2007) 6. Lu, Q., Ma, Y., Zhang, J.: A study of the active router structure. Computing, Communication, Control and Management 1, 157–160 (2009)

Implementation of Log Analysis System for Desktop Grids and Its Application to Resource Group-Based Task Scheduling Joon-Min Gil1 , Mihye Kim2 , and Ui-Sung Song3, 1

School of Computer & Information Communications Eng., Catholic Univ. of Daegu 2 Dept. of Computer Science Education, Catholic Univ. of Daegu, 13-13 Hayang-ro, Hayang-eup, Gyeongsan-si, Gyeongbuk 712-702, S. Korea {jmgil,mihyekim}@cu.ac.kr 3 Dept. of Computer Education, Busan National University of Education, 24 Gyodae-ro, Yeonje-gu, Busan 611-736, S. Korea [email protected]

Abstract. It is important that desktop grids should be aggressively deal with the dynamic properties arisen from the volatility and heterogeneity of resources. Therefore, it is required that task scheduling should positively consider the execution behavior that is characterized by an individual resource. In this paper, we implement a log analysis system which can analyze the execution behavior by utilizing actual log data of desktop grid systems. To verify the log analysis system, we conducted simulations and showed that the resource group-based task scheduling, based on the analysis of the execution behavior, offers faster turnaround time than the existing one even if few resources are used. Keywords: Desktop grids, Execution behavior, Log analysis system, Resource group-based task scheduling.

1

Introduction

With the popular use of Internet and high-performance of PC resources, it is possible to build a desktop grid environment, in which enables to build a virtual computing environment by binding the unused PC resources that is connected to Internet [1]. An important aspect in desktop grids is that each resource has volatility property, due to free withdrawal from execution participation even in the middle of task execution. Moreover, each resource has heterogeneity property as it has totally different computing environment (e.g., CPU performance, memory capacity, network speed, etc) [1,2]. Due to these two properties, it is not possible to expect the completion time of entire tasks. Therefore, if tasks are allocated without any consideration for the dynamic execution features of resources, execution failures will occur frequently due to the volatility and heterogeneity, and 

Corresponding author.

T.-h. Kim et al. (Eds.): FGCN 2011, Part II, CCIS 266, pp. 354–363, 2011. c Springer-Verlag Berlin Heidelberg 2011 

Implementation of Log Analysis System for Desktop Grids

355

thus the turnaround time of entire tasks becomes longer. However, execution failures can be prevented by selectively allocating tasks to those resources that are suitable for the current execution environment of desktop grids. As a result, it is necessary to analyze the execution behaviors of resources as a way to provide a stable execution environment for desktop grids. In this paper, we implement the log analysis system that can extract the execution behavior of resources from actual log data in Korea@Home system. The analysis results of the execution behavior can be used as the task allocation information to minimize the waste of resources and the execution delay in task scheduling. In particular, this study can be very useful as background on deciding task scheduling policies for various desktop grid systems as well as Korea@Home. The rest of this paper is organized as follows. In Section 2, we provide a brief description of desktop grids used in this paper. In Section 3, we define the execution behavior of resources as availability and credibility. The log analysis system implemented in this paper is presented in Section 4. This section also presents in detail the modules of the log analysis system. In Section 5. we present log analysis results and their application to resource group-based task scheduling. Section 6 concludes the paper.

2

Desktop Grid Environment

The desktop grid environment assumed in this paper is physically composed of a client (or an application submitter), a central server, and resources. The client is an entity submitting its own application to the central server. The central server takes responsibility of mediating resources and tasks and performs the functions such as task management, resource management, task allocation, etc. Each resource acts as a resource provider and executes the tasks sent by the central server during its idle cycles. Once a task is finished, the task result is sent back to the central server. The detailed steps to execute the parallel tasks submitted by a client in desktop grids are described as follows: Step 1 (Resource registration): Each resource registers its own information, such as CPU performance, memory capacity, OS type, etc., to the central server. Step 2 (Application submission): A client submits its own application to the central server. Step 3 (Task allocation): The central server allocates tasks in task pool to resources. Step 4 (Task execution): The resource to which tasks are assigned executes the tasks during its idle cycles. As soon as finishing task execution, it sends back task results to the central server. Step 5 (Result collection): The central server collects the task results received from resources and records them to databases. Step 6 (Application completion): The central server checks if all tasks are completed, and sends final results to the client.

356

J.-M. Gil, M. Kim, and U.-S. Song

The above described steps are based on the client-server architecture which has been typically used in the most desktop grid systems (e.g., BOINC [3], XtremWeb [4], Korea@Home [5], etc). The application, that is submitted to the central server by clients, is divided into hundreds of thousands of tasks, each of which is small enough to be executed in one resource. In addition, each task is mutually independent without any dependency between each others.

3

Execution Behavior of Resources

The performance of desktop grids is largely influenced by the dynamic features such as the execution join and withdrawn of resources [1,6]. Due to the task stops that can occur at any time even in the middle of task execution, task failures will be unavoidably encountered. Thus, the availability, that represents how much time each resource can spend executing tasks for a given time period, and the credibility, that determines the trustworthiness of task results in the present of failures, can be considered as an important factor to enhance the overall performance of desktop grids. Due to the dynamic features of desktop grids, it is strongly recommended that tasks should be allocated to the resources with high availability and credibility; these resources can return credible task results as many as possible within a given time period. In this paper, therefore, we classify resources by two factors, availability and credibility, and analyze the execution behavior of resources based on these two factors. Definition 1. Availability (A): a probability that can execute tasks in the presence of task failures. A=

M T T CF M T T CF + M T T CR

(1)

Definition 2. Credibility (C): a factor of how many task results can be returned after tasks are allocated to resources.  C=

r n,

0,

if n > 0 if n = 0

(2)

In Equation (1), MTTCF (Mean Time To Critical Failure) means the average failure time that is caused by execution failures or resource defects, and MTTCR (Mean Time To Correct and Recover) means the average execution time including recovery time after a failure occurs [7]. Thus, the availability defined as Equation (1) is calculated as the rate divided execution time by the total time that includes failure time and execution time. In Equation (2), n means total number of the allocated tasks to resources, and r means the number of task results for the n number of tasks allocated to resources. In this paper, we will implement the log analysis system that can classify resources by the availability and credibility presented in Definitions 1 and 2.

Implementation of Log Analysis System for Desktop Grids

ok

ok

start

Version

Login

joblist completion end

Logout Logout

357

Joblist

failure

ok

task execution

ok

StartWork StartWork

EndWork task completion

Fig. 1. Execution flow of each resource

4 4.1

Log Analysis System Execution Flow of Resources

Before implementing the log analysis system, it is important to understand first the execution flow of each resource to capture the dynamic features of desktop grids. Fig. 1 shows the execution flow of each resource in Korea@Home desktop grid system [5]. In this figure, a resource requests a function to the central server by message communications with an XML document and receives all the data and information needed to execute tasks from the central server. As we can see in Fig. 1, request-response functions include Login, Version, Joblist, StartWork, EndWork, and Logout, and are essentially required for analyzing the execution behavior of each resource. – Login: it is first phase to participate in the task execution of desktop grids. In this phase, an agent program installed in each resource performs the authentication function to check the credentials of the resource. – Version: it queries to the central server to check if there is newer agent program than one already installed in a resource. If so, the update process to reinstall the newer program is performed. – Joblist: it is a procedure to request tasks to be executed by a resource, and the central server provides the job list for the resource as a response. It also has the download function that sends the task program and data files needed for task execution from the central server to the resource. – StartWork: it is a function to inform the central server of task execution once a resource is ready to execute a task, – EndWork: it is a function to inform the central server of task completion after a resource finishes task execution. It also uploads task results to the central server. – Logout: it logs out of desktop grids. The time period between StartWork and EndWork can be seen as actual execution time of a task. As long as there is no failure during the time period,

358

J.-M. Gil, M. Kim, and U.-S. Song











(b) EndWork

(a) StartWork

Fig. 2. Example of XML messages Log Analysis System for Desktop Grids LogManager

Raw Log File

Log Data

LogParser

Filtered Log Data

XMLParser

LogAnayser

Information for Each Resource

Availability & Credibility

DBManager

Log DB

Fig. 3. Log analysis system for desktop grids

pure execution time is same to actual execution time. However, if there are failures, actual execution time is obtained by adding the pure execution time to the failure time. Fig. 2 shows an example of XML request messages for StartWork and EndWork. The XML messages of these types are exchanged between a resource and the central server for each request-response. In this paper, we extract the execution behavior of resources by parsing log information in the XML messages and analyze the availability and credibility of resources with the execution behavior. 4.2

Detail Modules of Log Analysis System

The log analysis system consists of five modules as shown in Fig. 3. The detailed function of each module is described as follows: – LogManager: it receives primitive log data from a desktop grid system as a type of files and performs preprocessing for the log data. – LogParser: it takes the responsibility of filtering unnecessary log from primitive log data so as to obtain only information relevant to the execution behavior of resources.

Implementation of Log Analysis System for Desktop Grids

359

Table 1. RESTful-based open APIs Name getAgent(agentid) getAvailability(agentid) getCredibility(agentid) getAgentByTask(taskid) getTaskByAgent(agentid) getCompletedTaskList() getCompletedTaskTime()

Description Get the information of a specific agent Get the availability of a specific agent Get the credibility of a specific agent Get agents having executed a specific task Get all of the tasks executed by a specific agent Get a list of the completed tasks Get total execution time of the completed tasks

Fig. 4. Example of log data fetched by open APIs

– XMLParser: it extracts meaningful information for each log through an XML parsing procedure and captures detailed information relevant to the execution behavior of resources. – LogAnalyzer: it takes responsibility of analyzing actual availability and credibility, utilizing the detailed information extracted by XMLParser. Besides, it takes responsibility of cooperating with DBManager to store and update the analysis results of availability and credibility in databases. – DBManager: it stores, updates, and retrieves the availability and credibility information of resources to/from databases. The log analysis system in this paper has been implemented with JDK 1.6.0, and JDOM 1.1 [8] has been used as an XML parser to extract task information from request-response XML messages. Meanwhile, we have used MySQL 5.5 [9] as the database to store and manage all the information relevant to the execution behavior of resources, including availability and credibility. The log analysis system also supports RESTful-based open APIs that allows users to access log analysis results more easily via web browsers. The users can simply acquire log analysis results using the open APIs presented in Table 1. These open APIs have been implemented by the Restlet [10] that supports REST web services based on Java. Fig. 4 shows an example of the log data fetched by the implemented open APIs in a web browser.

360

J.-M. Gil, M. Kim, and U.-S. Song

300000

250000 200000 150000 100000

50000

Mon

Tue

Thu

Wen

Fri

Sun

Sat

18:00:00

6:00:00

12:00:00

0:00:00

18:00:00

6:00:00

12:00:00

0:00:00

18:00:00

6:00:00

12:00:00

0:00:00

18:00:00

6:00:00

12:00:00

0:00:00

18:00:00

6:00:00

12:00:00

0:00:00

18:00:00

6:00:00

12:00:00

0:00:00

18:00:00

6:00:00

12:00:00

0:00:00

0

Fig. 5. Average execution time in a week basis

5 5.1

Log Analysis Results and Their Application to Task Scheduling Log Analysis Results

In this paper, log analysis results have been produced based on the primitive log data of the Korea@Home system which is the only desktop grid system in Korea. The used log data have been collected for one month (Dec. 2007). Fig. 5 shows the log analysis results acquired by analyzing these log data. The log analysis results are depicted in this figure as an average execution time in a week basis. In this figure, x- and y-axis represents the time in a unit of hour and total execution time by all the resources for one hour, respectively. On analyzing the results of this figure, we can find that execution behavior in weekday is different from that in weekend. Although execution behavior per day in weekday is not exactly same to each other, it has almost similar patterns; i.e., task execution time gradually increases from 9:00 to 12:00, and high execution time is maintained without any sharply fluctuation between 12:00 and 15:00. The task execution time from 15:00 to 24:00 steadily decreases. The task execution time from 0:00 and 9:00 is almost flat, but comparatively lower than that of other times. In fact, this tendency is due to the work style of white-colored workers. We can also observe from this figure that execution behavior in weekend is different from that of weekday; i.e., task execution time is kept at certain value or decreases steadily regardless of anytime for each day. Thus, we can see from this figure that different task scheduling policies should be applied according to weekday and weekend. Meanwhile, the availability and credibility of resources can have an uplifting effect on the performance of task scheduling in desktop grids. For example, resources with relatively high availability and credibility can execute more tasks for a given time period and moreover, return much credible task results to the central server. On the contrary, it is highly probable that resources with low availability and credibility will fail to return credible task results within a given time period. Therefore, if tasks are properly allocated according to the availability and credibility, resources will be efficiently utilized due to the reduction of task failures, resulting in shortening turnaround time for entire tasks.

1.0

1.0

0.8

0.8

0.6

0.6

Credibility

Credibility

Implementation of Log Analysis System for Desktop Grids

0.4

0.4

0.2

0.2

0

0 0

361

0.2

0.4

0.6

Availability

0.8

1.0

(a) Dec. 2007

0

0.2

0.4

0.6

0.8

1.0

Availability

(b) Jan. 2008

Fig. 6. Distribution of availability and credibility Table 2. Four sets of Resource Groups by Availability and Credibility Group R1 R2 R3 R4

Availability High High Low Low

Credibility High Low High Low

Fig. 6(a) shows the distribution of availability and credibility for the total 3,390 resources participating in Korea@Home system during one month (Dec. 2007). During this period, average availability and credibility of resources are 50.62% and 62.65%, respectively. On analyzing the distribution of availability and credibility in this figure, we can see that all resources are widely scattered and resources with more than average availability and credibility have much contributed to task execution than the others. 5.2

Application to Resource Group-Based Task Scheduling

Here, we apply the execution behavior of weekday and weekend to task scheduling. First, a set of resource groups is constructed with the availability and credibility of resources. Table 2 shows four sets of resource groups, R1 , R2 , R3 , and ˜ and credibility R4 . We used two kinds of thresholds, availability threshold (A) ˜ threshold (C) as a criterion to make resource groups. We determined the values ˜ considering the execution behavior presented in Fig. 5. Table 3 of A˜ and C, shows the values of A˜ and C˜ for weekday and weekend, respectively. Let us consider task scheduling based on the resource groups that are classified by the execution behavior of resources. In this paper, we use the task scheduling policy, in which a task is first allocated to a resource in the group with the highest availability and credibility (i.e., R1 ). If there are no more resources in R1 , the task is allocated to a resource in R2 . By a way of this process, tasks are allocated to resources in each group in order of R1 , R2 , R3 , and R4 .

362

J.-M. Gil, M. Kim, and U.-S. Song ˜ and C ˜ Table 3. Values of A (a) weekday Time zone A˜ 00:00∼09:00 0.8 09:00∼12:00 0.65 12:00∼15:00 0.5 15:00∼24:00 0.6

˜ C 0.8 0.7 0.6 0.7

(b) weekend ˜ Time zone A 00:00∼09:00 0.8 09:00∼12:00 0.8 12:00∼15:00 0.8 15:00∼24:00 0.8

30000

60000

Resource Group-based Scheduling Existing Scheduling

Resource Group-based Scheduling Existing Scheduling 50000

Number of Used Resources

Turnaround Time (min.)

25000

20000

15000

10000

5000

0

C˜ 0.8 0.8 0.8 0.8

1000

2000

3000

4000

5000

6000

7000

8000

Number of Tasks

(a) Turnaround time

9000 10000

40000

30000

20000

10000

0

1000

2000

3000

4000

5000

6000

7000

8000

9000 10000

Number of Tasks

(b) Used resources

Fig. 7. Comparison of scheduling performance

We conducted the simulation to evaluate the performance of desktop grids when the execution behavior of resources is applied to task scheduling. The log data of Jan. 2008, that are different from ones that have been used to extract the availability and credibility of resources, were used in the simulation (see Fig. 6(b)). For comparison, the existing task scheduling is also simulated without any consideration for the execution behavior. We used turnaround time and the number of used resources as performance measure. Fig. 7 shows the average turnaround time and the average number of used resources when 100∼1,000 tasks are used. From the results presented in Fig. 7(a), we can observe that the resource group-based task scheduling has much lower turnaround time than the existing task scheduling, regardless of the number of tasks. These results are caused from the use of different criterion on availability and credibility for each time zone. Meanwhile, as the existing task scheduling does not consider the execution behavior of resources, it allocates tasks to resources regardless of the execution environment of desktop grids. Thus, if resources have high availability and credibility, the return of task results can be fast. Otherwise, the fast return of task results can be hardly expected. As a result, the resource group-based task scheduling will obtain faster turnaround time by allocating tasks to resources according to the execution behavior of resources, compared to the existing task scheduling.

Implementation of Log Analysis System for Desktop Grids

363

The results of Fig. 7(b) show that the resource group-based task scheduling can complete entire tasks in spite of using much less resources, compared to the existing task scheduling. This results from the fact that our task scheduling can allocate tasks to the resources with as high availability and credibility as possible by analyzing the execution behavior of resources and thus task failures can be handled at a low level. On the contrary, in the existing task scheduling, tasks can be allocated to the improper resources for the execution environment of each time zone and hence the chances of task failures will increase dramatically. Once a task fails, it should be allocated to other resources again. Task failures will induce the waste of resources, resulting to low resource utilization. In conclusion, it is seen that fast turnaround time can be achieved with less resources if task scheduling is operated with the execution behavior of resources extracted from the log analysis system of this paper.

6

Conclusion

In this paper, we implemented a log analysis system which can extract the execution behavior of resources in desktop grid systems. Actual log data in Korea@Home systems were used to analysis the execution behavior. We observed from analysis results that there is a large difference between execution behaviors in weekday and weekend for each time zone. In the simulation to verify the applicability of the implemented log analysis system, resource group-based task scheduling was applied to actual desktop grid systems after grouping resources according to availability and credibility. Simulation results indicated that this scheduling can achieve faster turnaround time in spite of using less resources than the existing one. Therefore, we do expect that the log analysis system can be usefully utilized in improving the performance of new task scheduling policies or existing ones by analyzing task execution environment in advance before these policies are directly applied to actual desktop grid systems.

References 1. Kacsuk, P., Lovas, R., N´emeth, Z.: Distributed and Parallel Systems in Focus: Desktop Grid Computing. Springer, Heidelberg (2008) 2. Al-Azzoni, I., Down, D.G.: Dynamic Scheduling for Heterogeneous Desktop Grids. Journal of Parallel and Distributed Computing 70(12), 1231–1240 (2010) 3. BOINC: Berkeley Open Infrastructure for Network Computing, http://boinc.berkeley.edu 4. XtremWeb, http://www.xtremweb.net 5. Korea@Home, http://www.koreaathome.org 6. Kando, D., Fedaka, G., Cappello, F., Chien, A.A., Casanova, H.: Characterizing Resource Availability in Enterprise Desktop Grids. Future Generation Computer Systems 23, 888–903 (2007) 7. Shooman, M.L.: Reliability of Computer Sysytems and Networks. John Wiley & Sons Inc. (2002) 8. JDOM 1.1, http://www.jdom.org 9. MySQL 5.5, http://www.mysql.com 10. Lightweight REST framework for Java, http://www.restlet.org

FM Subcarrier Multiplexing Using Multitone Modulation for Optical Coherent Communications Hae Geun Kim and Ihn-Han Bae School of Computer and Information Communication Catholic University of Daegu 330 Kumrak-ri, Hayang-up, Kyungsan-si, 712-702, Korea [email protected] Abstract. FM (Frequency Modulation) subcarrier multiplexing using multitone modulation for optical coherent communication system have been firstly introduced to maximize the bandwidth utilization, where the optimized number of bits allocated in each subcarrier is defined. Each subcarrier transmitting different number of data bits is modulated by the M-QAM. When we implement FDM to divide the given bandwidth within FM channel, the spectrum of the subcarrier signals in the given FM bandwidth is the quadratic noise. We perform the numerical analysis the multicarrier modulation for only 4 subcarriers to prove applicability to FM subcarrier modulation, so the bit loading algorithm is not used. Firstly, we calculate the SNRs to obtain the BER of 10-9 for the cases of 4, 16, 64, 128-QAM at the given 4 subcarrier frequencies, 0.1, 0.15, 0,2, and 0.25 MHz. Secondly, we make the optimum choice the group composed of 20.7 dB for fsc=0.1 MHz, 18.4 dB for fsc=0.15 MHz, 20.3 dB for fsc=0.20 MHz, and 19.1 dB for fsc=0.25 MHz, where the SNR difference is lass than 2.3 dB. Hence the optimum modulators of FM subcarrier modulation are QPSK for fsc=0.1 MHz, 16-QAM for fsc=0.15 MHz, and 64QAM for both fsc=0.20 MHz and fsc=0.25 MHz. Keywords: Multitone Modulation, Narrowband Frequency Modulation Subcarrier Modulation, Optical Coherent Transmission.

1

Introduction

Frequency division multiplexing (FDM) with multitone modulation have been achieved the maximized reliable data rates over bandlimited communication channels [1]. Mostly, multitone modulation (MC) has been used in improving the performance of high bit rate digital subscriber lines. The main advantage of Multitone modulation maximizes the bandwidth utilization to optimize the number of bits allocated in each subcarrier. Besides, the MC employs bandwidth efficient modulation schemes such as QAM (quadrature amplitude modulation) [3][4]. A coherent FM-SCM optical communication system has been one of the essential solutions to transmit high speed optical data, because of recent development of high T.-h. Kim et al. (Eds.): FGCN 2011, Part II, CCIS 266, pp. 364–370, 2011. © Springer-Verlag Berlin Heidelberg 2011

FM Subcarrier Multiplexing Using Multitone Modulation

365

speed electronics. Multichannel data can be also handled in the radio frequency (RF) domain with one optical carrier in a coherent SCM system [2]. In a FM SCM receiver, the noise characteristic is not uniform, so we can not transmit the same data bits for all subcarriers. Hence, we need to decide the optimum number of data bits for each modulator [5]. In this paper, we propose the FM-SCM combined with MC where FM subcarrier multiplexing using multitone modulation for an optical coherent communication system. In multicarrier modulation, the number of transferable bits by each modulator is proportion to SNR of the channel, so the number of bits depending on the power of FM quadratic noise. Consequently, the BERs of all modulators in the FM band become equal. In chapter 2, we describe the FM subcarrier multiplexing using multitone modulation system and the spectrum of the subcarrier signals in the given FM bandwidth with the FM quadratic noise. In chapter 3, numerical results of the noise variance within the FM bandwidth of the SCM system is performed, then

2

FM Subcarrier Multiplexing Using Multitone Modulation

Multicarrier modulation with N subcarrier frequencies is shown in Fig. 1 [1]. Serial input bits are converted to parallel, then divided into n groups for FDM with subcarrier frequencies, fsc1, fsc2,…… fscN. When we let the frequency difference between fscN and fscN-1 be Δf, the total bandwidth is nΔf.

m1 bits Serial-toInput bits

m2 bits

fsc1 fsc2 Σ

Parallel

To FM Modulator

Modulators Converter mn bits

f

Fig. 1. Multicarrier modulation with N subcarrier frequencies

The outputs of modulators are summed for FM modulation, where each modulator is transmitting different bits depend on the noise power within the narrowband FM band. Because the spectrum of narrowband FM is similar to AM, we can multiplex the signal in FDM form as shown in Fig. 2(a). The subcarriers are modulated by the M-QAM, so the spectrum of each data is sinc function. When we implement FDM to divide the given bandwidth within FM channel, the spectrum of the subcarrier signals in the given FM bandwidth with the FM quadratic noise is shown in Fig. 2(b).

366

H.G. Kim and I.-H. Bae

In multicarrier modulation, the number of transferable bits by each modulator is proportion to SNR of the channel. Hence, if we can choose appropriate number of bits depending on the power of FM quadratic noise for assigned band, the BERs of all modulators in the FM band become equal. In Fig. 2(b), we assigned the bits per signal, which is proportion to the noise power. Hence, for example, the number of bits, 8, 6, 4, and 2, per symbol are transmitted with 256-QAM, 64-QAM, 16-QAM, and QPSK, respectively.

fcarrier BPF

fsc1 fsc2 fsc3

fscN

FM Carrier

…. Bandwidth W (a) Amplitude

Bits per Signal

FM quadratic Noise spectrum 8

8

8

8 6

6

… (b)

Fig. 2. The spectrum of the subcarrier signals in the given FM bandwidth with the FM quadratic noise

FM Subcarrier multiplexing using multitone modulation for an optical coherent communication system is shown in Fig. 3 [5]. Here, a single mode fiber is used in transmitting the FM modulated signal. In the transmitter, the output signal of a multicarrier modulator in Fig. 1 with N subcarrier frequencies is fed to an FM modulator, and then the FM signal modulates transmitting laser signal. Fig. 3 depicts the transmitter and receiver of the SCM system using FM as a principal modulation scheme where N SCM with electrically modulated signals are summed and enter into an FM modulator. Then the laser diode signal at λi, where i = 1,…, k, is FM modulated for transmission through the single mode fiber.

FM Subcarrier Multiplexing Using Multitone Modulation

367

At the optical receiver, the input optical signal is combined with the local laser signal. A PD (photodiode) transfers to the microwave signal consisting N channel multicarrier signals. The BPF produces an FM signal which is multiplied by an integration and dumpG detector with subcarrier frequencies, fsc1, fsc2,…… fscN.

G G G G G XG G G G G G G G G

Receiver 1

Multicarrier Transmitter 1 skG G ɅXG

tœ“›Šˆ™™Œ™G

šŠXG s–Šˆ“G skG G ɅXG

v—›Šˆ“G j–œ—“Œ™G

”–‹œ“ˆ›–•G

šŠYG

ΣG UG UG

UG

XG

mtG t–‹œ“ˆ›–™G

•GmŽUGXG

Multicarrier Transmitter k skG G Ʌ’G

tœ“›Šˆ™™Œ™G ”–‹œ“ˆ›–•G

kG UG tG lG UG UG |G tG UG UG G z•Ž“ŒG |G UG t–‹ŒG UG m‰Œ™G G UG UG

šŠuG

mtG kŒ›ŒŠ›–™G

iwmG

wkG

Receiver k

šŠXG s–Šˆ“G skG G Ʌ’G

v—›Šˆ“G j–œ—“Œ™G

šŠYG ΣG

’G mtG t–‹œ“ˆ›–™G

G

wkG

iwmG

tœ“Šˆ™Œ™G kŒ›ŒŠ›–™GXG

mtG kŒ›ŒŠ›–™G

UG UG UG

tœ“Šˆ™Œ™G kŒ›ŒŠ›–™G’G

šŠuG

Fig. 3. Multicarrier transmission using a FM channel with a single mode fiber

An integration and dump detector extracts a single channel SCM signal with the noise, After the signal with the bandwidth W. the noise autocorrelation becomes [5]

Rn (τ ) =

An

π

W

2  ω cos ωτ dω = 0

2 An

π

cos Wτ −

4 An

πτ 3

sin Wτ +

2 AnW 2 sin Wτ 4πτ

(1)

where the input signal of baseband filter has the cut-off frequency, W, and the anvelop An.

3

Numerical Results and Analysis of the Noise Calculation

Calculation of the noise variance within the FM bandwidth of the SCM system is performed using (1) which can be described as [5]

T T  E{N c2 } = E   Rn (t − t ) cos(ωsct ) cos(ωsct )dt dt  0 0 

(2)

where T is the period of the integration-and dump filter. When we assume there are only 4 subcarriers within the FM bandwidth, the noise power with a bandwidth, W, for each subcarrier is clearly defined as shown in Fig.4.

368

H.G. Kim and I.-H. Bae

The calculation results of the noise power of (1) at the stable bandwidth of 0.6 GHz in nW/Hz for the subcarrier frequency at 0.1, 0.15, 0.2, and 0.25 GHz are shown in Fig. 4. The relative gains between 0.1 GHz and 0.15 GHz, 0.2 GHz, and 0.25 GHz are - 2.68, - 4.58, and – 6.57 dB, respectively. Higher subcarrier frequency generates more noise power. So, we can allocate more data bits at the lower subcarrier frequency, typically known as bit loading. S (f) n

-B

-W

Sn(f)

W

0

B

f

f

FM quadratic Noise Calculated within Bandwidth W

fcarrier fsc1=0.1GHz

fsc2=0.15GHz

fsc4=0.25GHz fsc3=0.2GHz

f Calculated Noise Power [nW/Hz]

10.70

Gain [dB] Margin

0

19.83 - 2.68

32.68 - 4.58

48.56 - 6.57

Fig. 4. The calculated noise variance in nW/Hz and the relative gain in dB of the narrowband FM bandwidth

Bit loading divided into two categories: rate-adaptive and margin adaptive is a critical issue in multitone modulation. But, in this paper, since we perform the numerical analysis the multicarrier modulation for only 4 subcarriers to prove applicability, the bit loading algorithm is not used. First, we assume BER is set to 10-9, then the number of data bits for each subcarrier are defined. The bit error probability of M-ary QAM transmitting k bit/symbol within the Gaussian channel is denoted as [6]

pb =

 1 1  1 −  erfc  k M  

(M

3kEb 2 − 1 N0

)

   

where the Eb/N0 is the SNR and the erfc(x) is the complementary error function.

(3)

FM Subcarrier Multiplexing Using Multitone Modulation

369

When we consider the calculation results of [5], the SNRs required to achieve the BER = 10-9 in Figs. 6-9 for the QPSK is 12.5, 14.5, 16.5, and 19.1 dB, respectively. The SNRs to obtain the BERs of 10-9 for the cases of 16, 64, 128-QAM are also calculated in Table 1. Table 1. The BER calculation results to obtain the BER of 10-9 for the cases of 4, 16, 64, 128QAM at the given subcarrier frequencies, 0.1, 0.15, 0,2, and 0.25 MHz

Modulation Scheme QPSK 16-QAM 64-QAM 128-QAM

Subcarrier Frequency 0.1 MHz 12.5 dB 16.3 dB 20.7 dB 25.4 dB

0.15 MHz 14.5 dB 18.4 dB 22.6 dB 27.3 dB

0.2 MHz 16.5 dB 20.3 dB 24.5 dB 29.2 dB

0.25 MHz 19.1 dB 22.9 dB 27.1 dB 31.8 dB

In Table 1, we can make the group which consists of the the SNRs with about the same values in dB for each subcarrier frequency to optimize the number of bits loaded for the 4 modulation schemes. For example, when we group 25.4 dB for fsc=0.1 MHz, 22.6 dB for fsc=0.15 MHz, 24.5 dB for fsc=0.20 MHz, and 22.9 dB for fsc=0.25 MHz, the SNR difference is 2.8 dB. So, this case is still needed to diminish the SNR differences. With our FM channel case, the best choice is the group composed of 20.7 dB for fsc=0.1 MHz, 18.4 dB for fsc=0.15 MHz, 20.3 dB for fsc=0.20 MHz, and 19.1 dB for fsc=0.25 MHz, the SNR difference is lass than 2.3 dB. Hence the optimum modulators of FM subcarrier modulation are QPSK for fsc=0.1 MHz, 16-QAM for fsc=0.15 MHz, and 64-QAM for both fsc=0.20 MHz and fsc=0.25 MHz.

4

Conclusions

We introduce FM subcarrier multiplexing using multitone modulation for optical coherent communication system, where the optimized number of bits allocated in each subcarrier is defined. Multicarrier modulation is usually used in digital subscriber loop and its main advantage is to allocate the number of bits per subscriber based on the corresponding SNR. We successfully prove applicability to FM subcarrier modulation. In this paper we did not use the bit loading algorithm because only 4 channels are considered. We calculate the BERs to obtain the BER of 10-9, then make the optimum choice the number of bits for each channel. In later time, we plan to use the bit loading algorithm for multiple channels more than 100. Acknowledgments. This research was supported by the Research Grants of Catholic University of Daegu in 2011.

370

H.G. Kim and I.-H. Bae

References 1. Bingham, J.: Multicarrier Modulation for Data Transmission: An Idea Whose Time Has Come. IEEE Commu. Mag. 5 (May 5-14,1990) 2. Gross, R., Olshansky, R., Schmidt, M.: Coherent FM-SCM System Using DFB Lasers and a Phase Noise Cancellation Circuit. IEEE Photo. Tech. Let. 2 (January 1990) 3. Vaidyanadan, P., Lin, Y., Phoong, S.: Discrete Multitone Modulation With Principal Component Filter Banks. IEEE Trans. Circuit and Systems 49(10), 1397–1412 (2002) 4. Lee, S., Breyer, F., Randal, S., Cardenas, D., Boom, H., Koonan, A.: Discrete Multitone Modulation for High-Speed Data Transmission over Multimode Fibers using 850-nm VCSEL. In: OFC/NFOEC Conference (February 2009) 5. Kim, H.: High Speed Optical Coherent Transmission System using Narrowband FM Subcarrier Multiplexing. To appear in CCIS Proceedings of AST (September 2011) 6. Sklar, B.: Digital Communications. Prentice-Hall, Inc. (2001)

An Ontology-Based ADL Recognition Method for Smart Homes* Ihn-Han Bae and Hae Geun Kim School of Computer and Information Communication Engineering, Catholic University of Daegu, Gyeongbuk, Rep. of Korea {ihbae,kimhg}@cu.ac.kr Abstract. This paper presents a method for recognition of Activities of Daily Living (ADLs) in smart homes. Recognition of activities of daily living and tracking them can provide unprecedented opportunities for health monitoring and assisted living applications, especially for elderly and people with memory deficits. We present ARoM (ADL Recognition Method) that discovers and monitors patterns of ADLs in sensor equipped smart homes. The ARoM is consists of two components: smart home management monitoring and ADL pattern monitoring. This paper studies on the ontology base and the reasoning that are main parts of ADL pattern monitoring. The ontology base supports the semantic discovery for location, device, environments domains in smart homes. The reasoning system discovers the activity for a person and the appropriate service for a present situation. On detection of significant changes of context, the reasoning is triggered. We design the ontology model for ARoM and implement the prototype system of ARoM by using Protege and Jess tools. Keywords: Activities of daily living, ontology, semantic reasoner, smart home, web ontology language.

1

Introduction

Smart Homes, also known as automated homes, intelligent buildings, integrated home systems or domestics, are a recent design development. Smart homes incorporate common devices that control features of the home. Originally, smart home technology was used to control environmental systems such as lighting and heating, but recently the use of smart technology has developed so that almost any electrical component within the house can be included in the system. Moreover, smart home technology does not simply turn devices on and off, it can monitor the internal environment and the activities that are being undertaken whilst the house is occupied. The result of these modifications to the technology is that a smart home can now monitor the activities of the occupant of a home, independently operate devices in set predefined patterns or independently, as the user requires. Recently, research of intelligent spaces that grasp actions of living people inside and help them has been activated. Based on this liveliness, more and more information has become able to be handled quickly by development of information *

This work was supported by research grants from the Catholic University of Daegu in 2011.

T.-h. Kim et al. (Eds.): FGCN 2011, Part II, CCIS 266, pp. 371–380, 2011. © Springer-Verlag Berlin Heidelberg 2011

372

I.-H. Bae and H.G. Kim

technology. We can accumulate and deal with data of life behaviors or change of surroundings as time series data based on various sensors. There can be many supports based on the time series data acquired by sensor. For example, there are direct helps such as carrying loads when getting home or preparing ingredients in meals, and indirect helps such as batching our daily life and informing or warning in emergencies [1]. In this paper, we present ARoM that discovers and monitors patterns of ADLs in sensor equipped smart homes. The ARoM is consists of two components: smart home management monitoring and ADL pattern monitoring. This paper studies on the ontology base and the reasoning that are main parts of ADL pattern monitoring. The ontology base supports the semantic discovery for location, device, environments domains in smart homes. The reasoning system discovers the activity for a person and the appropriate service for a present situation. On detection of significant changes of context, the reasoning is triggered. We design the ontology model for ARoM and implement the prototype system of ARoM by using Protege and Jess tools. The remainder of this paper is organized as follows. Section 2 reviews OWL ontology and smart homes on related works. Section 3 designs the ontology model for ARoM using by Protege tool. Section 4 presents the ADL reasoning process with Jess, OWL reasoning engine. Section 5 concludes the paper and discusses future work.

2

Related Works

2.1

OWL Ontology

The term "ontology" originates from philosophy and refers to the discipline that deals with existence and the things that exist. Studer et al. [2] merged these two definitions stating that: "An ontology is a formal, explicit specification of a shared conceptualization." In computer science, an ontology is standardized representation of knowledge as a set of concepts within a domain, and the relationships between those concepts. It can be used to reason about the entities within that domain, and may be used to describe the domain [3]. Typical elements of ontologies are: (a) concepts and its attributes; (b) taxonomies to categorize concepts by generalization and specification; (c) relations between concepts; (d) axioms to define statements which are always true; and (e) individuals (or facts) are instances of concepts and its relations [4]. Ontology languages allow users to write explicit, formal conceptualization of domain models. The main requirements are: (a) a well-defined syntax; (b) a welldefined semantics; (c) efficient reasoning support; (d) sufficient expressive power; and (e) convenience of expression. OWL has been designed to meet this need for a Web Ontology Language. OWL is part of the growing stack of W3C (World Wide Web Consortium) recommendations related to the Semantic Web [5]. OWL is divided into three increasingly expressive sub-languages OWL-Lite, OWL-DL and OWL-Full. OWL-DL is most often used because it provides maximum expressiveness.

An Ontology-Based ADL Recognition Method for Smart Homes

2.2

373

Smart Homes

The recent emergence of ubiquitous environments, such as smart homes, has enabled the housekeeping, assistance and monitoring of chronically ill patients, persons with special needs or elderly in their own home environments in order to foster their autonomy in the daily living life by providing the required service when and where needed [6, 7]. By using such technology, we can reduce considerably costs, and alleviate healthcare systems. However, many issues related to this technology were raised such as activity recognition, person identification, assistance and monitoring. Activity recognition in smart environments is gaining increasing interest among researchers in ubiquitous computing and healthcare. Automatic recognition of activities is an important and challenging task. One of the typical applications in healthcare systems is the assistance and monitoring of the ADLs for persons with special needs and elderly to provide them with the appropriate services [8]. Several research works have been done, and several models are proposed to recognize activities for smart environments. B. Chikhaoui et al. [8] propose a new approach based on frequent pattern mining principle to extract frequent patterns in the datasets collected from different sensors disseminated in a smart environments. This approach adopt a hierarchical representation of activities, and generate patterns for each activity model. In order to recognize activities, a mapping function is used between the frequent patterns and the activity models. X. Hong et al. [9] presented homeADL that addressed the issues associated with the heterogeneous nature of storage and distribution of the data within the smart environment community. J. Xu [10] proposed an ontology-based framework to facilitate the automatic composition of appropriate applications. The system composed appropriate services depending upon the available equipments in each individual household automatically. P. A. Valiente and A. Lozano-Tello [11] presented IntelliDomo, an ontologies-based AmI system for the control of domestic systems. It used domestic components and its state values represented as instances of an ontology, and taken advantage of the power of the production rules specified by user in order to change the state of the system components in real time.

3

Ontology Model for ARoM

The use of Ambient Intelligence (AmI) is one of the areas which are rapidly gaining importance in the application of intelligent systems in companies and homes. Ubiquitous computing was suggested that computer and electric systems should be integrated into a physical environment and form part of it. AmI systems have sensors able to collect information in the environment. These systems are usually knowledge-based systems containing the specification of domestic elements and they are based on production rules that represent the system's reasoning elements. A correct way of representing domestic systems and behaving rules is throughout the use of ontologies and production rules based on the concepts established on these ontologies. In this paper, ARoM, the AmI system for smart homes is based on the apartments for elders or persons who live alone in the South Korea. Each apartment provides a means of independent living. A typical layout of an apartment is shown in Fig. 1.

374

I.-H. Bae and H.G. Kim

Position sensors in each room have the ability to monitor the movement of a person through the home environments, contact sensors detect if the window or the door has been opened/closed, and device sensors within electronic devices detect if the electronic devices have been turned on/off. Given the vast amount of information which may be generated from these sensors, it is necessary to discriminate between normal and abnormal situations. S2

S3

S1

Back Terrace S13

S8

Bathroom

Kitchen

S9

Dinngroom

Studyroom

S12

Front Door

S10

Dressroom

S5

S6

S11

Bedroom 1

Livingroom

S7

Bedroom 2

Front Terrace S4 S1-S4: Contact sensors, S5-S10: Position sensors, S11, S12: Temperature sensors, S13: Gas sensor

Fig. 1. Layout of space with various sensors to support independent living

We present ARoM, an ontology-based smart home system for discovers and monitors patterns of ADLs. Fig. 2 shows the overall architecture of ARoM system. A typical smart home environment contains sensors to detect the context and services to be invoked by users. Users interact with the system via user devices such as smart phones. The abstraction component maps the sensor data from various sensors to context information with a mapping function such as fuzzy membership functions. The context information is processed by the context provision component, which is a complex event processing system, producing formatted context events. The service management stores service descriptions, monitors and invokes home services. Service registration and un-registration are also performed by this component. The change detection decides when to trigger activity or service discovery. The ontology base component is introduced to support the semantic discovery for location, device, environments domains in smart homes. The reasoning component discovers the activity for a person and the appropriate service for a present situation. Sometimes the

An Ontology-Based ADL Recognition Method for Smart Homes

375

reasoning component forwards a ADL message to children or close relatives who live in elsewhere through mobile networks. After user identity certification process, the resident or the person who received the message can invoke a home service by sending a command to the service management component.

Fig. 2. System architecture of ARoM.

In order to infer ADL of a smart home from temporal contexts, ARoM is defined in this paper. In ARoM, the semantic web is used to represent the temporal contexts. The context model for the smart home is defined by OWL ontology, and the model is implemented by Protege tool, a graphical editor. Protege is a free, open-source platform that provides a growing user community with a suite of tools to construct domain models and knowledge-based applications with ontologies. At its core, Protege implements a rich set of knowledge-modeling structures and actions that support the creation, visualization, and manipulation of ontologies in various representation formats [12]. The ontology model for ARoM is shown in Fig. 3. It is composed of seven domains: AmIApplication, Device, DeviceStatus, Location, Person, Sensor and SensorStatus to represent the knowledge base in smart home systems. AmIApplication ontology describes the concepts related to the ambient intelligent applications for the smart home. AmIApplication class is consisted of two sunclasses: Activity and Service. Also, Service subclass is consisted of two subclasses: DailyLifeService and SafetyService. Sensor ontology describes the concepts related to various sensors which are installed in the smart home. Sensor class is consisted by three subclasses: ContextSensor, DeviceSensor and Timer. ContextSensor subclass has four individuals that are instances of concepts: Humidity, Light, Noise and Temperature. Fig. 4 shows the structure of AmIApplication class and it's OWL coding.

376

I.-H. Bae and H.G. Kim

Fig. 3. OWL-based context ontology for ARoM

...





...



...





...

Fig. 4. Structure of AmIApplication class and it's OWL coding

An Ontology-Based ADL Recognition Method for Smart Homes

377







...

Fig. 4. (continued)

The OWL structure of ARoM is shown in Fig. 5. Fig. 5(a) and Fig. 5(b) represent the classes and the properties of ARoM, respectively. The OWL model for ARoM has 75 individuals, but Fig. 5(c) represents the individuals of the subclass DailyLifeService of AmIAppication class in ARoM.

Fig. 5. OWL structure for ARoM. Table 1. Object properties in the OWL-based context ontology for ARoM Object property calls has hasDevice hasDuration hasSensing hasSensor hasStatus hasTime locatedIn uses

Domain Person Person Sensor Timer Sensor Device Device Timer Person Person

Range Service Activity Device TimeDuration SensorStatus Sensor DeviceStatus TimeInstance Location Device

Property characteristics Functional Functional InverseFunctional Functional Functional InverseFunctional Functional Functional Functional Functional

378

I.-H. Bae and H.G. Kim

Properties are binary relations on individuals i.e. properties link individuals together. Table 1 shows the object properties in the OWL-based context ontology for ARoM. For example, the property locatedIn might link the domain class Person to the range class Location. Properties can be limited to having a single value i.e. to being functional. They can also be either transitive or symmetric. For example, the inverse of hasDevice is hasSensor.

4

ADL Reasoning

A semantic reasoner, reasoning engine, rules engine, or simply a reasoner, is a piece of software able to infer logical consequences from a set of asserted facts or axioms. The notion of a semantic reasoner generalizes that of an inference engine, by providing a richer set of mechanisms to work with. The inference rules are commonly specified by means of an ontology language, and often a description language. A large number of rule engines work well with Java, and many are available as open source software. Some of the most popular engines include Jess, Algernon and SweetRules. We chose Jess as the semantic reasoner for ADL reasoning.

Fig. 6. OWL/Jess integration structure for ADL reasoning.

Jess (Java Expert System Shell) system [13] consists of a rule base, and an execution engine. Fig. 6 shows the OWL/Jess integration structure for ADL reasoning. Once the ARoM OWL concepts have been represented in Jess, the Jess execution engine can perform inference. As ADL rules fire, new Jess facts are inserted into the fact base. Those facts can be used in further inference. When the inference process completes, these facts must be transformed into OWL knowledge. Fig. 7 shows an example of ADL rule reasoning. Jess system has currently five facts that are represented in Fig. 7(a), Jess engine performs the ADL rule that is represented in Fig. 7(b), and Jess system displays the inference result message for the ADL rule, Fig. 7(c). After the ADL rule performed, new fact is generated by Jess engine, the new fact 'has Bae Sleeping' (the part of rectangular box in Fig. 8(a)) is inserted in fact base, and the short message that is generated by the inference is sent to children or close relatives who live in elsewhere through wireless networks (Fig. 8(b)).

An Ontology-Based ADL Recognition Method for Smart Homes

379

Fig. 7. Example of an ADL rule reasoning.

Fig. 8. Results of an ADL rule reasoning.

5

Conclusions and Future Work

Smart home provide a platform that allow us to monitor and assist with activities of daily living. By monitoring and recording such activities, automatic detection of changes in patterns of behaviors becomes possible. This information can subsequently reveal a decline in health, potential risks in the surrounding environment, and emergency situations.

380

I.-H. Bae and H.G. Kim

In this paper, we present ARoM that discovers and monitors patterns of ADLs in sensor equipped smart homes. The ARoM is consists of two components: smart home management monitoring and ADL pattern monitoring. This paper studies on the ontology base and the reasoning that are main parts of ADL pattern monitoring. The ontology base supports the semantic discovery for location, device, environments domains in smart homes. The reasoning system discovers the activity for a person and the appropriate service for a present situation. On detection of significant changes of context, the reasoning is triggered. We design the ontology model for ARoM and implement the prototype system of ARoM by using Protege and Jess tools. Our future works include studying on an fault tolerant ADL recognition method on the basis of variable precision rough sets and realization on the full version of proposed ARoM.

References [1] Mori, T., Fujii, A., Shimosaka, M., Noguchi, H., Sano, T.: Typical Behavior Patterns Extraction and Anomaly Detection Algorithm Based on Accumulated Home Sensor Data. Future Generation Communication and Networking 2, 12–18 (2007) [2] Studer, R., Benjamins, R., Fensel, D.: Knowledge engineering: Principles and methods. Data & Knowledge Engineering 25, 161–198 (1998) [3] Viinikkala, M.: Ontology in Information Systems, http://www.cs.tut.fi/~kk/webstuff/Ontology.pdf [4] Ay, F.: Context Modeling and Reasoning using Ontologies. University of Technology, Berlin (2007) [5] Antoniou, G., van Harmelen, F.: Web Ontology Language: OWL. In: Handbook on Ontologies in Information Systems, pp. 67–92. Springer, Heidelberg (2003) [6] Qin, W., Shi, Y., Suo, Y.: Ontology-Based Context-Aware Middleware for Smart Space. Tsinghua Science and Technology 2, 707–713 (2007) [7] Rashidi, P., Cook, D., Holder, L., Schmitter-Edgecombe, M.: Discovering Activities to Recognize and Track in a Smart Environment. IEEE Transactions on Knowledge and Data Engineering 23, 527–539 (2010) [8] Chikhaoui, B., Wang, S., Pigot, H.: A Frequent Pattern Mining Approach for ADLs Recognition in Smart Environments. In: International Conference on Advanced Information Networking and Application, pp. 248–255 (2011) [9] Hong, X., Nugent, C.D.: HomeADL for Adaptive ADL Monitoring within Smart Homes. In: Annual International IEEE Engineering in Medicine and Biology Society Conference, pp. 3324–3327 (2008) [10] Xu, J.: Ontology-based Smart Home Solution and Service Composition. In: Int. Conf. of Embedded Software and Systems, pp. 297–304 (2009) [11] Valiente-Rocha, P.A., Lozano-Tello, A.: Ontology-based expert system for home automation controlling. In: 23th International Conference on Industrial, Engineering & Other Applications of Applied Intelligent Systems, pp. 661–670 (2010) [12] Horridge, M.: A Practical Guide to Building OWL Ontologies using Protégé 4 and COODE Tools (2007), http://owl.cs.manchester.ac.uk/tutorials/protegeowltutorial/ resources/ProtegeOWLTutorialP4_v1_1.pdf [13] Mei, J., Bontas, E.P.: Reasoning Paradigms for OWL Ontologies. Technical Report B-0412, Department of Information Science. Peking University, p. 24 (2004)

Analysis of User Preferences for Menu Composition and Functional Icons of E-Book Readers in a Smartphone Environment Mihye Kim1, Joon-Min Gil2, and Kwan-Hee Yoo3 1 Department of Computer Science Education, School of Computer & Information Communication Engineering, Catholic University of Daegu, 330 Hayangeup Gyeonsansi Gyeongbuk, South Korea 3 Department of Computer Education and Information Industrial Engineering, Chungbuk National University, 410 Seongbongro Heungdukgu Cheongju Chungbuk, South Korea {mihyekim,jmgil}@cu.ac.kr, [email protected] 2

Abstract. With the rapid growth of the electronic-book (e-book) market, various types of e-book readers, such as dedicated e-book reading devices and e-book reader applications, have been released. However, the user interfaces of these e-book readers are highly diverse, which is becoming a major problem regarding usability. In this paper, user preferences for the menu composition of an e-book reader are analyzed via a survey, and an ideal menu composition for e-book readers is proposed on the basis of the survey results. The functional icons used in the menus of e-book readers are also analyzed, to investigate the necessity to standardize these icons. Keywords: E-books, E-book readers, Menu composition of e-book readers, Functional icons of e-book readers.

1

Introduction

The electronic book (e-book) industry is growing rapidly, far beyond expectations. The major online booksellers and publishers have already rushed into e-book production, and are leading the growth in the e-book market. Price Waterhouse Coopers (PwC) has forecast that the world e-book industry will reach a total of US$82 billion in 2013 (up from $24 billion in 2009), with the average annual growth of 27.2% [1], [2], and that 90 percent of reading material will be published in digital form by 2020 [3]. According to the Association of American Publishers [4], monthly e-book sales exceeded paper book sales in the United States in February 2010. In May 2011, Amazon.com also announced that they sold more e-books than printed books [5]. The increasing use of tablet PCs, such as iPad, Galaxy Tab, and PlayBook, as well as smartphones, seems to be accelerating this rapid growth in the e-book market. An e-book is a traditional printed book published in digital form “produced on, published by, and readable on computers or other electronic devices” [6, p.164]. T.-h. Kim et al. (Eds.): FGCN 2011, Part II, CCIS 266, pp. 381–392, 2011. © Springer-Verlag Berlin Heidelberg 2011

382

M. Kim, J.-M. Gil, and K.-H. Yoo

E-books can provide integrated resources with many additional functions, such as images, voices, videos, hyperlinks, searching and navigation, and dictionaries, while performing the traditional role of printed books without time or space limitations, using wired and wireless networks [7]. Users can easily read not only books, but also newspapers, magazines, and blogs on a screen enhanced with rich, riveting digital media technology, in much the same way they read printed books. Despite the remarkable growth of the e-book market, several fundamental issues exist. The major issues relevant to e-books are copyright problems pertaining to digital rights, interoperability problems associated with the current variety of exclusive e-book formats, and some negative psychological perceptions of e-books [3]. In an attempt to resolve the interoperability issue, the International Digital Publishing Forum [1] released the Electronic Publication format (abbreviated EPUB, ePub, Epub, or epub) to establish a free and open e-book standard. Another major ebook issue is usability. This issue is caused by the present diversity of e-book formats, as well as the variety of e-book readers with different menu compositions and functional icons. As many new e-book publishers and sellers are emerging in the ebook market, they provide their own dedicated e-book reading devices or e-book reader applications. The user interfaces of these e-book readers, including the menu composition for the main features and the functional icons displayed on these menus, are configured in various ways. In this paper, user preferences are analyzed for the menu composition and functional icons of the five most commonly used e-book readers: iBook, Kindle, Kyobo eBook, Stanza, and Wattpad. The analysis is based on the main features and functional icons of these e-book readers, with the objective of determining an ideal menu composition for an e-book reader in terms of usability, on the basis of user preferences. The necessity to standardize functional icons is also analyzed but limited to smartphones, which are emerging as universal e-book reader terminals. However, the menu structures of dedicated e-book readers and universal terminals are nearly identical in most cases. The remainder of this paper is organized as follows. Section 2 describes the theoretical background of this study, including the definition of an e-book and overviews of the five e-book readers, together with their main features and functional icons. Section 3 presents and discusses the method and questions of the survey, as well as the survey results. Section 4 summarizes the paper and concludes with possible directions for future research.

2

Theoretical Background

2.1

Definition of an E-book

A variety of definitions of e-books have appeared in the literature. The online Oxford Dictionary of English [8] defines an e-book as “an electronic version of a printed book which can be read on a computer or a specifically designed handheld device.” Rao defines an e-book “as a text in digital form; as a book converted into digital form; as digital reading material; as a book in a computer file format or electronic file of words; or as images with unique identifiers” [9, p. 364]. Gardiner and Musto define

Analysis of User Preferences for Menu Composition and Functional Icons

383

an e-book as follows: “An electronic book (also e-book, ebook, digital book) is a book-length publication in digital form, consisting of text, images, or both, and produced on, published through, and readable on computers or other electronic devices” [6, p. 164], [10]. Nelson [11] defines an e-book as an electronic publication that can be read with the aid of a computer, a personal digital assistant (PDA), a special e-book reader, or even a mobile phone. Similarly, Chrystal defines an e-book as follows: “An e-book is an electronic or digital representation of a given text, which scanned, typed, or programmed (for example, using HTML), having virtual “pages”, that are read using e-book reading software, either on a personal computer (PC), a PDA, a smartphone, or on a dedicated e-book reading device” [3, p. 2]. Accordingly, an e-book can be defined as an electronic or digital book that can be read on various types of devices, including desktop computers, notebooks, netbooks, tablet computers, smartphones, or dedicated devices using e-book reader applications. The term “e-book” refers not only to digital books, but also to digital newspapers, magazines, other digital publications, and all content in digital form, including blogs. 2.2

E-Book Readers

There is a variety of e-book-related terminology, such as e-book content, format, reader software, reader applications, and reading devices. The terms e-book (reader) software, e-book reader application, and e-book reading device are often considered interchangeable with the term e-book reader or simply e-reader. E-book devices can be classified into two types: dedicated e-book terminals and universal terminals [2]. Dedicated e-book terminals include e-reader devices such as the Amazon Kindle [12], the Barnes & Noble Nook [13], and the Sony Reader Touch Edition [14], and they are usually constructed with e-reader software. For example, Kindle is an e-book reading device with an embedded e-book reader application. On the other hand, to read e-books on a universal terminal (such as a PC, a tablet PC, or a smartphone), only an e-book reader application (i.e., software) is required. Thus, e-book sellers and publishers who produce dedicated e-book devices also support several different versions of their e-book applications, to enable users to read e-books on various types of general-purpose terminals. For instance, the version of Kindle that runs on PCs or smartphones is only an e-book application. Here, the terms e-book reading device and e-book reader application (or software) are both e-book readers. There are also many different types of e-book applications for general-purpose terminals. These include iBooks [15], Stanza [16], Wattpad [17], and Kyobo eBook [18]. The terms e-book reader application and e-book reader software are also referred to as e-book readers or simply e-books. In this paper, we analyze user preferences for the menu composition and functional icons of the five most commonly used e-book readers: iBooks, Kindle, Kyobo eBook, Stanza, and Wattpad. Fig. 1 shows screenshots of these e-book readers. Barnes & Noble, one of the world largest publishers, has its own e-book reading device, known as the Nook, but also uses iBooks and Stanza for universal terminals. The Nook is not included in the present analysis, because this study is limited to e-book readers for smartphones. Kyobo is included because it is the largest e-book seller in Korea.

384

M. Kim, J.-M. Gil, and K.-H. Yoo

iBooks. iBooks is an application for downloading and reading e-books on the iPad, iPod Touch, and iPhone by Apple, Inc. [15], [19]. iBooks uses the ePub format, and also supports PDF files via data synchronization with iTunes [19]. Users can download e-books from the iBookstore, as well as through Safari or Apple Mail. The downloaded books will then be available on the bookshelf (library) of the device. Kindle. Kindle is an electronic device for downloading and reading e-books, developed by the Amazon.com subsidiary Lab126. In other words, Kindle is an e-book reader serviced by Amazon.com, which allows users to read e-books, newspapers, magazines, blogs, and other digital media on its website [20], [21]. However, Kindle can be used to read e-books from other sellers or websites, as long as they are compatible. Some file formats such as HTML, JEPG, BMP, and PDF can also be converted into the Kindle format [21]. Amazon has released a number of versions of Kindle, from the initial version of 2007, to Kindle 2 in 2007 and Kindle DX in 2009, to Kindle DX Graphite and Kindle 3 in 2010, with enhanced features and technologies, including a longer-lasting battery, more storage capacity, converting function from text to speech, 3G and Wi-Fi connectivity, and new E-Ink Pearl display [12], [22]. Amazon has also released a number of e-book reader applications for universal terminals, such as Kindle for PCs and many different types of smartphones [22]. To read Kindle e-books on their devices, users must first register for an Amazon account, and then download and install an appropriate Kindle reader application. Kyobo eBook. Kyobo eBook is an e-book reader application developed by Kyobo Book Center Co., LTD, Korea’s largest bookstore chain. It supports desktop computers, notebooks, tablet PCs (iPad and Galaxy Tab), PDAs, Apple iPhones, and Android smartphones, as well as a number of dedicated e-book devices, including Samsung’s SNE-60k and iRiver’s Story [18]. It uses the ePub format, but also supports PDF and TXT file formats. Stanza. Stanza is an e-book application for downloading and reading e-books, and is compatible with a number of devices, including Apple iPad, iPod Touch, and iPhone, as well as desktop computers for Windows and Mac [16], [23]. It supports various e-book formats, including ePub, eReader, MS LIT, Amazon Kindle, Mobipocket, and PalmDoc, as well as general document formats such as HTML, PDF, MS Word, and Rich Text Format [23]. With its rich features, Stanza was the most popular e-book reader for the iPhone prior to the announcement of iBooks. Wattpad. Wattpad is one of the most popular e-book reading applications, and can run on most mobile phones, as well as on desktop computers [24]. Wattpad.com is also an e-book community for writers wishing to publish and share their content for free distribution [24]. After creating a free account, users can download almost all of the e-books for free, and can also upload their content to Wattpad without converting their file formats, and without any compatibility issues [25]. Wattpad is the only e-book reader that supports all major mobile phones [24], including Alcatel, Apple, BenQ-Siemens, BlackBerry, HTC, LG, Motorola, Nokia, O2, Panasonic, Pantech, Sagem, Samsung, Sharp, Sony Ericsson, Toshiba, VK Mobile and others [25].

Analysis of User Preferences P for Menu Composition and Functional Icons

ks (a) iBook

(c) Kyobo eBo

385

(b) Kindle

(d) Stanza

(e) Wattpad

he function menus of five e-book readers displaying e-books Fig. 1. Screenshots of th

2.3

Main Features of E-Book E Readers

Most e-book readers havee various features that allow users to read books m more conveniently. Table 1 show ws the main features of the five e-book readers consideered here. These features are av vailable while reading an e-book, but not all of them are included in any one device. The symbol ‘O’ indicates that an e-book reader supporrts a given feature, while ‘X’ ind dicates that it does not. In general, users can org ganize a library (i.e., bookshelf) of e-books, and can brow wse and search the books in thee library from a list sorted by title, author, or genre. T They can enjoy reading books by y touching or tapping the book icons and buttons with thheir fingers, and by switching between b portrait and landscape view. Users can turn paages by sliding the left or right sides of the screen, or by pressing a button to go forwardd or backward, as well as by flip pping or scrolling up or down through the pages. They can also view and edit book infformation such as the title, authors, and abstract of a boook, and can share their e-books via Twitter or Facebook, or e-mail them to others.

386

M. Kim, J.-M. Gil, and K.-H. Yoo Table 1. Main features of the five e-book readers E-book Readers

Category Scroll

Up and down

Slide

Gestures

Right and left One Center of page short Left of page Touch touch Right of page (Tab) Two short touch One long touch

Book information Share

Page turning Page turning effect Move to a special page Information on current page Font setting Search

Etc.

View book information Edit book information Move to library (bookshelf) Twitter Facebook E-mail By touching or tapping Sliding right and left By flipping through the page Scrolling up and down Like slide transition Flip style like book By table of contents (TOC) By bookmarks Move to beginning By page numbers By page navigator Chapter Page number Percentage Change font size Change font face Change font color Word search within book In built-in dictionary Insert bookmark Insert memo/note/annotation Changing backgound color Lock auto-rotation Change day and light mode

iBooks

Kindle

Kyobo eBook

Stanza

Wattpad

Adjust Page Scrolling brightness Page turning X View function menu Backward page turning Forward page turning Word selection X X X Auto scrolling Select words with Same as Same as Word X a magnifying glass iBooks iBooks selection X X O O O X X X O X O O O O O X O X O X X O X O O X X X X O O O O O O O O O O X O O X X X X X X X O X O O O O O X X O X O O O O X O O O O X X O X X X X O X X X O O O O O X X O O X O O O O O X O X O X O O O O O O X X O O X O X O O O O X O X O O X O X O O O O X O O X O X O X O O O X X O O X X X O O O X

X

X

Most e-books automatically save the place where a user left off, so that the user can easily resume reading from that place later. E-book readers usually provide movement and navigation features that enable users to go to a particular page in a book using the table of contents, bookmarks, or a page navigator (i.e., page scroll bar) located at the bottom of the screen, or by entering a specific page number. Users can look up information on the current page, such as the chapter the current page belongs to, the page number, and the current page location expressed as a percentage. They can also change the text size and the size, face, and color of the font. Most e-book readers also provide search features that allow users to find a specific word or phrase in a single book or a whole library, or look up a selected word or phrase in a built-in dictionary or

Analysis of User Preferences for Menu Composition and Functional Icons

387

via Google or Wikipedia. Users can highlight and insert memos/notes/annotations for specific words, phrases, or sentences, and can also bookmark favorite pages. They can control the foreground and background color, and adjust the screen brightness and contrast, as well as change the text and background colors by switching from day mode to night mode, or vice versa. Other features not shown in Table 1 can be found in references [11], [15], [16], [17], and [18].

3

User Preferences for the Menu Composition and Functional Icons of E-Book Readers

3.1

Survey Method

The analysis of user preferences for the menu composition and functional icons of the five e-book readers was conducted via a survey of university students using Google Androidbased smartphones or Apple iPhones. The students who participated in the survey were all majoring in computer education. One month before conducting the survey, we provided the participants with detailed instructions on the five e-book readers, and guided them in the downloading and use of free e-books from each of the e-book reader websites. A total of 25 students participated in the survey. A number of iPhones and iPads were prepared for the students using Android-based smartphones, to give them the opportunity to use iBooks and Stanza (which only support the Apple devices). Among the participants, 52% used Android-based smartphones and 48% used Apple iPhones. 3.2

Design of the Survey Questions

Most e-book readers have a variety of features that combine the traditional advantages of printed books with the additional benefits of digital media. Some functions, such as page turning and word selection, are continuously available while reading a book, whereas other features must be manipulated using a function menu. Accordingly, most e-book readers provide many of their features via a function menu with icons. This menu is displayed at the top and bottom of the page, generally when the user taps or touches the text anywhere around the center of the page. The function menu is typically composed of three lines, one at the top and two at the bottom of the page. The screenshots of Fig. 1 show examples of the function menus for each of the five e-book readers. Because the screen of a smartphone is small, all the features supported by an e-book reader cannot be displayed on a menu. Thus e-book readers normally display only 8 to 10 features on their function menus, represented by icons, while the remaining features are accessed through a settings feature. As Fig. 1 indicates, the features displayed on the function menu vary significantly from one reader to the next. Question 1 therefore targeted which features are most desirable on a function menu, in terms of usability. Fifteen features were selected from among the main features of Table 1. We then listed these 15 features in alphabetical order, and asked each respondent to choose 10 that would be most useful to him/her as menu items. Some of the main features listed in Table 1, such as “slide,” “touch,” and “insert annotation after selecting a word or text,” were excluded from the 15 features because they are activated directly, rather than by functional icons. Table 2 lists the 15 features presented to the respondents.

388

M. Kim, J.-M. Gil, and K.-H. Yoo Table 2. Fifteen features presented to the respondents in Question 1

Features 1) Adjust brightness 2) Change background color 3) Change day and light mode 4) Change font color 5) Change font face 6) Change font size 7) Display page navigator (i.e., page scroll bar)

Selection(√) □ □ □ □ □ □ □

Features 8) Information on current page 9) Insert bookmark 10) Lock auto-rotation 11) Move to library 12) Search 13) Share (Facebook, Twitter, E-mail) 14) View book information 15) View table of contents

Selection(√) □ □ □ □ □ □ □ □

The second question concerned the icons used in the function menus. As Fig. 1 shows, the features in the menus are denoted by icons. Table 3 lists the icons utilized in the function menus of the five e-book readers. As can be seen, some of the icons are very similar, whereas others are very different from one another. When circumstances compel a user to employ more than one e-book reader, different functional icons for the same feature can be confusing and inconvenient unless the icons clearly indicate the feature they represent. However, the meanings of some icons are not readily apparent from their images or shapes. This is why we listed the e-book reader icons together with their corresponding features, and then asked the respondents to choose the icon that best represents each feature.

Table 3. Functional icons used in the function menus of the five e-book readers (Question 2) E-Book Readers Features Font setting

iBook

Kindle

Change font size Change font face Change font color

Word search in a book Word search in a builtSearch in dictionary

Share

Stanza

X

X X X

(dic.) Auto connects to dic. when select a word

X (TOC)

X X

X X

(dic.)

X

(TOC)

X

(book inf.) X

X

(my lib.)

Twitter

X

X

X

Facebook

X

O

X

X

E-mail

X

X

X

X

Insert bookmark Etc

Wattpad

(settings) X

By table of contents View book information Book Edit book information Inf. Move to the library

Kyobo eBook

X O X

Insert memo/note/annotation

(memo)

Set day and night mode

X

X X

(night mode)

(annot.)

X

Analysis of User Preferences for Menu Composition and Functional Icons

3.3

389

Survey Results

Table 4 summarizes the results of Question 1. The features chosen by the respondents are listed in decreasing order of frequency. Under the assumption that a response rate ~ of Table 4 are the greater than 80% is significnat, we can say that items features users preferred for inclusion in a function menu.

① ⑧

Table 4. Questionnaire results on the menu features the respondents selected from Question 1 No

Features

Information on current £ page ¤ Insert bookmark ¥ Search ¦ Adjust brightness § Change font size page navigator ¨ Display (i.e., page scroll bar) © Move to library ª View table of contents

Selection (n)

%

No

Features

Selection (n)

%

25

100% « Lock auto-rotation

17

68%

25 25 23 23

100% 100% 92% 92%

16 13 8 6

64% 52% 32% 24%

2

8%

1 25x10=250

4% -

¬ ­ ® ¯

View book information Share (Facebook, Twitter, …) Change day and light mode Change font face

23

92% ° Change font color

22 21

88% ± Change background color 84% Total

Table 5 shows the features provided in the function menu of each of the five ebook readers, in order of user preference according to Table 4. The symbol ‘∆’ represents a feature supported indirectly via a settings feature in the function menu. As Table 5 indicates, only iBooks provides all of the features that more than 80% of the respondents selected in Question 1 (i.e., features of ~ in Table 4), although it does not support the features “lock auto-rotation,” “view book information,” and “share,” which more than 50% of the respondents preferred (i.e., ~ in Table 4). All of the features included in the questionnaire are features of provided by Stanza in its function menu, either by direct icon or indirect icon (i.e., a settings icon). The menus provided by Kindle, Kyobo eBook, and Wattpad are not in close accord with user preferences, especially in the case of Wattpad.

① ⑧

⑨

Table 5. Features supported by the function menus of the five e-book readers No

Features

£ Information on current page ¤ Insert bookmark ¥ Search

% of selection iBooks 100% O 100% O 100% O

O X O

Kyobo eBook O O X

Kindle

Stanza

Wattpad O X X

¦ Adjust brightness

92%

O

X

X

§ Change font size ¨ Display page navigator (i.e., page scroll bar) © Move to library ª View table of contents « Lock auto-rotation ¬ View book information ­ Share (Facebook, Twitter, E-mail) ® Change day and light mode ¯ Change font face ° Change font color ± Change background color

92%

O

O

∆

O ∆ O By sliding up & down ∆

92%

O

O

O

O

O

88% 84% 68% 64% 52% 32% 24% 8% 4%

O O X X X X O X O

O ∆ X X ∆ X X O X

O O ∆ O X O X X ∆

O O ∆ O ∆ O ∆ ∆ ∆

O X X O O ∆ ∆ ∆ ∆

X ∆

390

M. Kim, J.-M. Gil, and K.-H. Yoo

From the survey results, we can determine an ideal function menu for an e-book reader, namely the iBooks menu with an added settings feature icon, as in Stanza’s function menu. In other words, the ideal menu composition for an e-book reader consists of a total of nine icons: eight for features ~ in Table 4, and one as a settings feature, through which features ~ or others are indirectly supported. Table 6 shows the user preferences for the functional icons used in the five e-book readers. For the settings icon, 80% of the respondents chose that of Stanza, and 20% selected that of Kyobo, which employs the same image as Stanza, with the word “settings” inserted below the icon. Nobody selected the Kindle or Wattpad icons. Among the “move to library” icons, 84% selected those of iBooks and Kyobo, and 12% chose that of Kindle. Only one respondent (4%) selected the icons of Stanza and Wattpad. Regarding the “view book information” icon, 80% chose that of Stanza and 20% chose those of Kyobo and Wattpad. Among the “view table of contents” icons, similar user preferences were noted for those of iBooks and Kyobo: 56% preferred that of iBooks, and 40% chose that of Kyobo. Only one respondent selected the icons of Kindle and Stanza. For the “change font size” icon, there were similar user preferences for three different shapes, among which the icons of iBooks and Kindle were most often selected (by 44%). For the “show amount of reading” icon, 76% of the respondents chose that of Kyobo, 16% chose that of Kindle, and 8% selected the icon of Stanza. Regarding the “insert bookmark” icon, 88% chose that of iBooks and only 12% chose those of Kyobo and Stanza. Nobody preferred the Kindle icon. For the “insert memo/note/annotation” icon, users exhibited similar preferences for those of Stanza and iBooks: 48% chose that of Stanza, and 40% preferred that of iBooks. Only 12% of the respondents selected the Kindle icon. Regarding the “set day and night mode” icon, 76% chose that of Kyobo, and 24% selected that of Stanza.

⑨ ⑮

① ⑧

Table 6. User preferences of functional icons No

¤ ¥ ¦ § ¨ © ª «

E-book n Reader

Feature

Icon

%

Settings



Stanza 20 80%

Move to library View book information



iBooks, Kyobo 21 84%

View TOC



Icon E-book n Reader

Stanza 20 80% iBooks 14 56%



Change font iBooks,  Kindle size Show amount  Kyobo of reading Insert  iBooks bookmark Insert  Stanza memo/note/an (annotation) notation Set day and  Kyobo night mode (night mode)



3

12%



Kyobo, 5 Wattpad

20%

-

Kindle

5

-

-

4% -

9

36%

Kindle

5 20%

4

16%



Stanza

2

8%

12%



Kindle

0

0%

iBooks 10 40%



Kindle

3 12%

Stanza

-

-

Kindle





0%

Stanza

Kyobo, Stanza 3

19 76%

Wattpad 0 Stanza, Wattpad 1





(memo)

0%



19 76%



0

Kyobo 10 40%

11 44%

12 48%

Kindle

%

Kindle, Stanza 1



22 88%

Icon E-book n Reader

20%

 Kyobo settings  

%

6

24%

-

4%

-

Analysis of User Preferences for Menu Composition and Functional Icons

391

As the results of Question 2 indicate, there were several icons that nobody or only a few respondents selected (e.g., Kindle and Wattpad icons of , Kindle, Stanza, and Wattpad icons of , Kindle and Stanza icons of , Stanza icon of , Kyobo, Stanza and Kindle icons of , and Kindle icon of ). This seems to imply that the images of these icons do not clearly represent the corresponding features. Some icons were preferred at similar levels (e.g., the “view TOC” icons of iBook and Kyobo, the “change font size” icons of iBooks, Kindle, and Stanza, and the “insert memo/note/annotation” icons of Stanza and iBooks). In such cases, it seems to be relatively unimportant which of the icons is used in a function menu. We also observed an interesting trend regarding the use of two icon types: (1) icons that display only an image (i.e., icons without words) and (2) icons that display both an image and a word (i.e., a word that indicates the meaning of the icon). For example, for the “view TOC” feature, the iBooks icon is of the first type, and the Kyobo icon is of the second type. For icons whose images clearly represent a feature, the respondents tended to prefer icons without words (e.g., Stanza icons of and ). On the other hand, for icons whose images do not clearly represent a feature, the and ) or respondents preferred icons with words more (e.g., Kyobo icons of similarly (e.g., iBooks and Kyobo icons of ). This seems to suggest that if an icon cannot clearly convey its meaning with only an image, it would be better to add a word to the image to help users understand the exact meaning of the icon. From the results of Question 2, we can also notice that all icons utilized by iBooks adapt to user preferences, but most of the icons used by Kindle and some icons by Stanza and Wattpad are incompatible with user preferences.

②

⑦

⑧

④

4

④

①

⑥

① ⑥

③ ⑨

Conclusion

This paper presented an analysis of user preferences for the menu composition and functional icons of the five most commonly used e-book readers, based on their main features and functional icons. The objective was to suggest an ideal menu composition for an e-book reader on the basis of user preferences. The necessity to standardize functional icons was also analyzed. The analysis of user preferences was conducted via a survey of university students using Google Android-based smartphones or Apple iPhones. A total of 25 students participated in the survey. The survey results indicated that the menu compositions of iBooks and Stanza accommodate user preferences to some extent, but those of Kindle, Kyobo eBooks, and Wattpad are comparatively incompatible with those preferences. From the user preference results on menu composition, we were able to determine an ideal function menu for an e-book reader, namely the iBooks menu with an added settings icon to support the features that are not presented in the function menu. Furthermore, we confirmed that the use of function icons needs to be standardized to some extent. The icons that nobody (or only a few respondents) selected should be replaced with the icons preferred by the majority of the respondents. In such cases, users were actually confused by the icons, and did not understand their meaning. The survey results also suggested that when it is difficult to determine the exact meaning of an icon with only an image, it would be better to add a word to the image. However, considering the small screens used in smartphones, the icons should be as simple as possible.

392

M. Kim, J.-M. Gil, and K.-H. Yoo

This paper explored user preferences for the menu composition and function icons of e-book readers in terms of their usability. Nevertheless, several issues related to usability remain. These are related to the use of actions such as scrolling, sliding, and touching or tapping the e-book screen or icons. These issues will be investigated in future research.

References 1. The International Digital Publishing Forum (IDPE), http://idpf.org/ 2. Kim, W.: The status and prospects of the e-book market. Telecommunication Association 52, 72–77 (2010) 3. Chrystal, R.: The Evolution of e-Books: Technology and Related Issues. Digital Libraries, INFO 653 (2010) 4. Association of American Publishers (AAP), http://www.publishers.org/ 5. Amazon.com: New Release, http://phx.corporate-ir.net/ phoenix.zhtml?ID=1565581&c=176060&p=irol-newsArticle 6. Gardiner, E., Musto, R.G.: The Electronic Book. In: Suarez, M.F., Woudhuysen, H.R. (eds.) The Oxford Companion to the Book. Oxford University Press (2010) 7. Kim, M., Yoo, K.-H., Park, C., Yoo, J.-S.: Development of a Digital Textbook Standard Format Based on XML. In: Kim, T., Adeli, H. (eds.) AST/UCMA/ISA/ACN 2010. LNCS, vol. 6059, pp. 363–377. Springer, Heidelberg (2010) 8. Oxford Dictionaries: “e-book”, http://oxforddictionaries.com/ (accessed on August 2011) 9. Rao, S.: Electronic book technologies: an overview of the present situation. Library Review 53(7), 363–371 (2004) 10. Wikipedia: “E-book”, http://en.wikipedia.org/wiki/E-book (accessed on August 2011) 11. Nelson, M.R.: E-Books in Higher Education: Nearing the End of the Era of Hype? Educause Review 43(2), 40–56 (2008) 12. The Kindle, http://www.amazon.com/ref=gno_logo 13. NOOK Tech Specs – Barnes & Noble, http://www.barnesandnoble.com/ nook/features/techspecs/?cds2Pid=30195 14. Sony Reader Touch Edition, http://ebookstore.sony.com 15. iBooks, http://www.apple.com/ipad/built-in-apps/ibooks.html 16. Stanza, http://www.lexcycle.com/ 17. Wattpad, http://www.wattpad.com/ 18. Kyobo eBook, http://digital.kyobobook.co.kr/digital/guide/ guideMain.ink?guidePage=guide01&guide_menuNo=1 19. Wikipedia: “iBooks”, http://en.wikipedia.org/wiki/IBook 20. Dudley, B.: Kindle hacking, iPod parallels and a chat with the Kindle director. Seattle Times (November 19, 2007), http://blog.seattletimes.nwsource.com/ brierdudley/2007/11/chatting_with_amazons_kindle_d.html 21. Brinlee, D.: What Is a Kindle? AskDeb, http://www.askdeb.com/technology/kindle/what/ 22. Wikipedia: “Amazon Kindle”, http://en.wikipedia.org/wiki/Amazon_Kindle 23. Wikipedia: “Lexcycle Stanza”, http://en.wikipedia.org/wiki/Lexcycle_Stanza 24. Wikipedia: “Wattpad”, http://en.wikipedia.org/wiki/Wattpad 25. Wattpad – Free Mobile E-Book Reader With Free E-Books, Wattpad – Free Mobile EBook Reader With Free E-Books

Dynamic Transmission Target Selection Scheme for Load-Balancing in WSN* Seok-Yeol Heo1, Wan-Jik Lee1, and Won-Yeoul Lee2,** 1

Department of Applied IT & Engineering, Pusan National University of Pusan, Korea 2 Department of Cyber Police & Science Youngsan University of Yangsan, Kyungnam, Korea {syheo,wjlee}@pusan.ac.kr, [email protected]

Abstract. We studied a dynamic transmission target selection scheme for loadbalancing in WSN. The goal is for the energy consumption of packet transmission for all nodes to be nearly same as possible. For a load-balancing, it should be needed to predict the scale of the transmission energy of all nodes. In multihop transmission mode, a fixed transmission path is necessary for the loadbalancing of nodes that are at the same hop distance from the sink node (we call as horizontal hopping transmission), and a variable range transmission mode is needed for the load-balancing of nodes that are at different hop distances from the sink node(we call as vertical hopping transmission). In this paper, we developed a dynamic transmission target selection scheme for horizontalhopping and vertical-hopping transmission. The performance evaluation results show that the energy consumption of all the nodes was nearly the same, and the performance achieved was superior to that of existing load-balancing schemes. Keywords: WSN, dynamic transmission target selection, load-balancing, horizontal-hopping transmission, vertical-hopping transmission.

1

Introduction

It is not possible to derive an accurate load-balancing in WSN. However, it is possible to derive a transmission technology that can almost realize load balancing. We studied a transmission target selection for load-balancing of WSN providing periodic monitoring. We define the load-balancing to be a method that has uniform transmission energy consumption if packets are transmitted without errors from all nodes on the predetermined path. Load-balancing transmission technology involves transmission target nodes selection such that the packet transmission energy of all sensor nodes is the same. We assume that the WSN providing periodic monitoring has a single fixed sink node, and all the sensor nodes periodically transmit sensing data to the sink node. Moreover, the nodes have an ability of switching the transmission mode between multihop and direct. We assume this application area because this is the most widely used WSN. * **

This paper is supported by YoungSan University Corresponding author.

T.-h. Kim et al. (Eds.): FGCN 2011, Part II, CCIS 266, pp. 393–402, 2011. © Springer-Verlag Berlin Heidelberg 2011

394

S.-Y. Heo, W.-J. Lee, and W.-Y. Lee

In a multihop transmission environment, the energy consumption increases if the distance between the nodes and the sink node is small. In a direct transmission environment, the greater the distance between the nodes and the sink node, the higher is the energy consumption. For load-balancing of the WSN, we need to predict the energy consumption. To do this, a sink node must select the transmission target nodes of all nodes in advance. The sink node collects the location and neighbor information of all nodes when the network is initially organized. If the sink node calculates the target nodes of all nodes and delivers the calculation result to all nodes, the transmission target selection for load-balancing is complete. This load-balancing setting has to be reset only when there is a change in the network phase. By doing so, it is possible to reduce energy consumption due to control packet for collecting information of neighbor nodes, and excessive energy waste can be prevented. For load-balancing in multihop-transmission mode, if necessary, more than one node can be selected as the next nodes, and some packets must be transmitted directly to the sink node. That is, multi-path and direct transmission should be appropriately combined for load balancing. This Function for load balancing differs slightly from the legacy routing function. The routing function uses routing information and metric. But in transmission target selection function, packets are forwarded by the predetermined path which consists of the next nodes. And the next nodes can vary dynamically. We call this function as a dynamic transmission target selection function. We propose a dynamic transmission target selection and transmission methods for load balancing. We assessed the performance by comparing the proposed scheme with existing load balancing routing schemes and evaluated whether or not the hot spot problem was solved. Related research for equalizing the energy consumption of sensor networks is discussed in Section 2, and Section 3 presents our proposed scheme. The performance evaluation and the results are given in Section 4, and the conclusion and future work are presented in Section 5.

2

Related Researches

Many routing schemes have been proposed to increase the lifetime of sensor networks [1-6] by load-balancing. However, these schemes improve only partially the weighted energy consumption pattern. The LEACH (Low-Energy Adaptive Clustering Hierarchy) protocol tried to distribute the load of a head node that has excessive energy consumption [7]. However, this scheme cannot solve the imbalance problem of the entire network. The UCS (Unequal Clustering Size) scheme [8] and the energy aware routing scheme [9] proposed to directly solve the energy imbalance problem of sensor networks. The UCS tried to correct the load imbalance using cluster size. Nodes with above-average energy and mobility are moved to predetermined positions and become cluster head nodes. The cluster size is determined by the distance from the sink node. Cluster head nodes that are close to the sink node consume energy faster than those more distant because they receive more multihop transmission requests from distant clusters.

Dynamic Transmission Target Selection Scheme for Load-Balancing in WSN

395

UCS was shown to be about 10%–30% better than the ECS (Equal Clustering Size) scheme. It assumes that the cluster head nodes can be moved to predetermined positions. However, in practice this assumption is unrealistic. In addition, the difference between internal communication costs and communication costs between clusters was not considered. Each protocol has different between-cluster and internal communication costs, so the effectiveness of UCS seems to be low. The Energy Aware Routing scheme selects efficient paths in a multihop transmission environment. The basic idea is that the path that consumes the smallest amount of energy is not always the best path. This scheme instead uses the probability of the candidate paths. An energy-metric-based probability value is allocated to each path. When a packet is received, a path is randomly selected based on its probability value. In this scheme, it is possible to continuously assess and change paths. In sensor networks, this may be either an advantage or a disadvantage. The path determination is carried out at each packet transmission, which is an additional burden. Thus, this routing system cannot avoid unbalanced energy consumption.

3

Dynamic Transmission Target Selection for Load-Balancing

This paper assumes the WSN that all nodes have the variable transmission ranges. In multihop transmission, all nodes have the same transmission range. Moreover, all nodes except 1-hop distance nodes have the variable transmission range because the nodes have to transmit in both multihop and direct transmission mode. The network is divided into areas on the basis of the distance from the sink node as shown in figure 2. The load balancing of nodes in the same area is achieved by equalizing the number of received multihop transmission requests of each node from the previous area. Hence, we define horizontal-hopping transmission, which ensures that all nodes in the same area receive the same number of multihop transmission requests. Horizontal-hopping transmission ensures load balancing for nodes in the same area, but load imbalance with nodes in other areas can occur. Load balancing between different area nodes can be accomplished by a proper combination of direct and multihop transmission. If a distant node carries out direct transmission rather than multihop transmission, its own energy consumption increases but the number of multihop transmissions for nodes close to the sink node decreases. Thus, the transmission energy can be maintained at almost the same level by performing multihop transmission and direct transmission in rotation for the nodes located more than 2-hops from the sink node. We define the Vertical-hopping transmission as the method to have the balancing of energy consumption using multihop and direct transmission in rotation. The sink node executes selection the next nodes for multihop transmission and calculation of the vertical-hopping transmission ratio when network initiation and whenever the network status changes.

396

3.1

S.-Y. Heo, W.-J. Lee, and W.-Y. Lee

Dynamic Transmission Target Selection for Horizontal-Hopping Transmission

Figure 1 shows the horizontal-hopping transmission method. In Fig. 1(a), node 1 in area 1 receives two multihop transmission requests from area 2, and node 2 receives three such requests. This leads to a difference in the energy consumption of nodes 1 and 2. For load balancing, node 3 in area 2 transmits the packets to node 1 and node 2 by turns. We call these nodes as hopping node which has more than 2 multihop transmission target node. A node that is not a hopping node is called a designated node, and a designated node performs multihop transmissions with only one. In Fig. 1(b), node 3 is the hopping node. Hopping node requests in rotation to selected target nodes. In Fig. 1(b), if the nodes in area 2 send 10 packets each, 50 packets are delivered to area 1, and each node in area 1 receives 25 packets respectively.

Sink Node Area 1

Area 2

1

2

3

Sink Node 1

2

Area 1

3

Area 2

Horizontal-Hopping Transmission (a)target selection of legacy scheme

(b)target selection of our scheme

Fig. 1. Horizontal-hopping transmission method

Figure 2 shows our performance evaluation network model. Solid marked nodes are hopping nodes decided by sink node. The nodes connected through lines from the hopping nodes are target nodes of each hopping node. We can perceive from figure 2 that the transmission range of hopping node is practically acceptable. 3.2

Dynamic Target Selection for Vertical-Hopping Transmission

There is a difference in energy consumption between nodes in different areas. To eliminate this difference, we must increase the energy consumption for distant nodes and decrease the energy consumption for nodes close to the sink node. That is, distant nodes should transmit some packets directly to the sink node. This reduces the number of multihop transmissions, decreasing the energy consumption of forwarding nodes. Therefore, a dynamic transmission target for vertical-hopping is the next nodes or the sink node.

Dynamic Transmission Target Selection Scheme for Load-Balancing in WSN

397

100 90 80 70 60 50 40 30 20 10 0 0

10

20

Sink

30

40

50

Sensor Node

60

70

80

90

100

Hopping Node

Fig. 2. An example of target selection of hopping nodes

By vertical-hopping transmission, all nodes except 1-hop distance nodes transmit some packets via multihop transmission and the rest via direct transmission. Figure 3 illustrates vertical-hopping transmission. Sink Node

Area 1

Area 2

1

2

3

Sink Node

1

2

3

Area 1

Area 2

Vertical-Hopping Transmission Fig. 3. Vertical-hopping transmission method

For load balancing through vertical-hopping transmission, we must calculate the direct-to-multihop transmission ratio as we call vertical-hopping ratio for each area. Figure 4 shows the vertical-hopping ratio analysis model. In figure 4, Nm_j and Nm_i are the number of multihop transmissions for the node in area j and area i, respectively. Nd_j and Nd_i are the number of direct transmissions for node in area j and area i, respectively. Nodes in the same area have the same number of transmissions and receptions via horizontal-hopping, i.e., Nm_i1 = Nm_i2 and Nd_i1 = Nd_i2. N is the number of packets for one round per node. Only Nd_j packets are transmitted via direct transmission; the remainder are transmitted via multihop transmission.

398

S.-Y. Heo, W.-J. Lee, and W.-Y. Lee

Node i1:N

multi-hop transmission = N + 2*Nm_i – Nd_j Node j:N

direct transmission = Nd_i

Node i2:N

direct transmission = Nd_j

direct transmission = Nd_i

Fig. 4. Vertical-hopping ratio analysis model

To calculate the vertical-hopping ratio, we need to know the transmission and reception energy of the sensor nodes. We applied the energy consumption pattern developed in [7]. For the calculation of the vertical-hopping ratio, we assumed that all transmission and reception packet sizes are the same. We did not consider reception energy because it cannot be compared with transmission energy. Table 1 shows the parameters for the calculation of the vertical-hopping ratio. Table 1. Parameters for calculation of vertical-hopping ratio parameter K H Np_i Nd_i r D Ei ET_d

meaning hop distance from sink node maximum hop distance number of nodes of area i number of direct transmission of area i diameter of area maximum multihop transmission range after decision of load balancing routing path transmission energy of area i transmission energy when the distance is d

Given the model of figure 4, equation 1, 2, and 3 give the energy consumption of a node in each area. E1 in equation 1 and E2 in equation 2 are the energy consumed by a single node in areas 1 and 2 respectively; Ek in equation 3 is the energy consumed by a single node in area k. H  H    N p _ i − (N d _ i ⋅ N p _ i )  i =1 i =1  E E1 =   T_D N p_i    

(1)

Dynamic Transmission Target Selection Scheme for Load-Balancing in WSN H  H    N p _ i − (N d _ i ⋅ N p _ i )  i =2 E E2 =  i = 2 + N d _ 2 ⋅ ET _ 2 r   T _D Np_i     H H     N p _ i − (N d _ i ⋅ N p _ i )  i =k i =k  E Ek = + N d _ k ⋅ ET _ k ⋅r   T_D Np_i    

399

(2)

(3)

We set E1 = E2= ⋯ = Ek and, using equation 1, 2, and 3, obtain Nd_1, Nd_2, ⋯ , Nd_k. The vertical-hopping ratio is then the number of direct transmissions that equalizes the energy consumption of all the nodes. Equation 4 gives Nd_H, the number of direct transmissions in area H, and using this value, we can find the numbers of the remaining direct transmissions. Nd _ H

4

H    H −1   N p _ 1 −  N p _ i  ⋅ ET _ D +   N d _ i ⋅ N p _ i  ⋅E T _ D i =1   i =2  = N p _ 1 ⋅ (ET _ D − E T _ H ⋅r ) − N p _ H ⋅ ET _ D

(4)

Performance Assessments

We used simulation to assess the performance of the load-balancing routing system. We measured the number of surviving nodes for each round (specific period) and the remaining energy of each area. In addition, we compared our approach with the multihop transmission scheme, the direct transmission scheme, and the clustering scheme. The conditions for the performance assessment are given in Table 2. Table 2. Network model for performance assessment parameter network size number of node diameter of area(r) multihop transmission distance(D) processing energy(Eelec) amplitude energy(ε) initial energy packet size(k)

Meaning/value a quarter of circle with 100 meter radius 100 25 m Max_Tr 2.5 μJ/bit 1.8 μJ/bit/m2 15 kJ 400 bit

Figure 5 shows the number of surviving nodes for each round of our scheme. It can be seen that the energy of the nodes is depleted at nearly the same time. The first energy depletion occurs at about round 2800. The node energies are then depleted rapidly. Not all the nodes are depleted at the same round because of differences in the packet reception energy or differences introduced by approximation in the verticalhopping ratio.

400

S.-Y. Heo, W.-J. Lee, and W.-Y. Lee 100 90

Number of sensors still alive

80 70 60 50 40 30 20 10 0 0

500

1000

1500

2000

2500

3000

3500

4000

Round

Fig. 5. Number of surviving node by each round

% of residual energy

Figure 6 shows the remaining node energy of each area based on the distance from the sink node. The changes in the remaining energy are approximately constant in all the areas, confirming that the load balancing of each area is successful.

Area 4 Area 3 Area 2 Area 1

Round

Fig. 6. Remaining node energy by each area

Figure 7 shows the packet reception rates of the sink node. The packet reception rate is calculated as the total number of packets sent by sensor nodes versus the total number of packets received by the sink node. The packet reception rate drops rapidly if the connection is cut off because of a hot spot. For the multihop scheme, the packet reception rate drops rapidly because hot spots occur: the energy of nodes close to the sink node is quickly depleted. For the direct transmission, the packet reception rate drops because the energy of distant nodes is quickly depleted. The clustering scheme performs better than the multihop and direct transmission schemes, but it is not possible to avoid the hot-spot problem.

Dynamic Transmission Target Selection Scheme for Load-Balancing in WSN

401

The packet reception rate of the hopping routing scheme remains high for a longer period. That is, the network service is provided continuously without hot spots until the energy of each node is depleted. 100

Packet reception rate of sink node

90 80 70

호핑전송 proposed sheme direct trans. 직접전송 multihop trans. 멀티홉전송 clustering 클러스터링

60 50 40 30 20 10 0 0

1000

2000

3000

4000

5000

Round

Fig. 7. Packet reception rate of sink node

Number of sensors still alive

The network lifetime is an important performance indicator for sensor networks. Different applications define the lifetime differently. For periodic monitoring applications, FND (First Node Dies) is the reference point for the lifetime. Figure 8 shows the network lifetime of our scheme, the multihop transmission scheme, the direct transmission scheme, and the clustering scheme. The lifetime of our scheme is about 1.3~2 times longer than those of the other schemes.

proposed sheme Direct transmission Multihop transmission clustering

Round

Fig. 8. Performance comparison by the number of survival nodes

402

S.-Y. Heo, W.-J. Lee, and W.-Y. Lee

The transmission energy of all the sensor nodes is consumed almost equally when our scheme is applied. Moreover, the network lifetime is better than those of other systems, and our scheme nearly eliminates the hot-spot problem.

5

Conclusions and Future Work

We have proposed a dynamic transmission target selection scheme for load-balancing in WSN. Proposed scheme is possible to predict the energy consumption pattern of all the nodes of the network. The energy consumption of all the nodes is nearly equalized via horizontal-hopping transmission and vertical-hopping transmission. An increase of transmission range because of horizontal-hopping transmission can be avoided by selecting the proper hopping node. The vertical-hopping ratio is needed to find the number of direct transmissions in one round. We can get this ratio via the energy consumption model. Simulations showed that our proposed scheme is effective in solving the load imbalance problem.

References 1. Gomez, J., Campbell, A.T.: A case for variable-range transmission power control in wireless multihop networks. In: Proc. of the 23rd International Annual Joint Conference of the IEEE Computer and Communications Societies (INFOCOM 2004), pp. 1425–1436 (2004) 2. Xu, Y., Heidemann, J., Estrin, D.: Adaptive energy-conserving routing for multihop ad hoc networks. Research Report 527, USC/ISI (2000), http://www.isi.edu/johnh/PAPERS/Xu00a.html 3. Heinzelman, W., Chandrakasan, A., Balakrishnan, H.: An application specific protocol architecture for wireless microsensor networks. IEEE Transactions on Wireless Communications 1(4), 660–670 (2002) 4. Perillo, M., Cheng, Z., Heinzelman, W.B.: On the problem of unbalanced load distribution in wireless sensor networks. In: IEEE GlobeCom Workshops 2004, November 29December 3, pp. 74–79 (2004) 5. Younis, M., Youssef, M., Arisha, K.: Energy aware routing in cluster based sensor networks. In: Proceedings of the 10th IEEE/ACM International Symposium on Modeling, Analysis and Simulation of Computer and Telecommunication Systems (MASCOTS 2002), pp. 129–136 (2002); Fort Worth 6. Gao, J., Zhang, L.: Load-Balanced Short-Path Routing in Wireless Networks. IEEE Transactions on Parallel and Distributed Systems 17(4), 377–388 (2006) 7. Heinzelman, W., Chandrakasan, A., Balakrishnan, H.: An application specific protocol architecture for wireless microsensor networks. IEEE Transactions on Wireless Communications 1(4), 660–670 (2002) 8. Soro, S., Heinzelman, W.B.: Prolonging the Lifetime of Wireless Sensor Networks via Unequal Clustering. In: Proceedings of the 5th International Workshop on Algorithms for Wireless, Mobile, Ad Hoc and Sensor Networks (IEEE WMAN 2005), pp. 236–243 (April 2005) 9. Akyildiz, I.F., et al.: Wireless sensor networks: a survey. Computer Networks 38, 393–422 (2002)

Organizing Virtual Research Groups with Light Path Technology Min-Ki Noh1, Won-Hyek Lee1, Seung-Hae Kim1, and Joon-Min Gil2 1

Korea Institute of Science and Technology Information, 245 Daehangno, Yuseong-gu, Daejeon 305-806, Korea {mknoh,livezone,shkim}@kisti.re.kr 2 Catholic University of Daegu, 13-13 Hayang-ro, Hayang-eup, Gyeongsan-si, Gyeongbuk 712-702, Korea [email protected]

Abstract. Advances in network technologies lead not only the advent of new applications but also research form change from the individual research. Research and Development communities for specific cooperative workgroups over LP (Light Path) are required to support dedicated network resources, intelligent network management, and sophisticated access control and monitoring. In this paper, we introduce LP technologies and describe tightly remote cooperative work community, namely VRG (Virtual Research Group). Lastly, we address the function and framework of VRG with LP technologies Keywords: Optical network, Network resource allocation, Light path, Research network.

1

Introduction

Currently, global-scale collaborative researches have a feature that many researchers, organizations, and laboratory equipments require a tight cooperative work. In addition, the technology of network progress has influenced on research methods, types, and applications in mutual relevance. The network connected to each resource is the representative cutting-edge network services that can provide researchers and research groups with international collaboration works. Most researches on network activities has focused on providing the high quality of services (QoS) over networks such as bulk data transmission, resource sharing, real-time remote conferencing, and the remote control of equipment. In particular, the form of cooperative research is changed to the virtual expansion, and thus this research form needs more enhanced security and sensitive applications, sharing the resources between researchers. To support virtual environment, network resources are reserved to the researchers between end sites. As shown in Fig. 1, each site offers the required bandwidth through a dedicated path and thus can provide highquality and high-performance services on congestion-free networks. T.-h. Kim et al. (Eds.): FGCN 2011, Part II, CCIS 266, pp. 403–411, 2011. © Springer-Verlag Berlin Heidelberg 2011

404

M.-K. Noh et al.

Fig. 1. Optical network diagram creating E2E LP

However, most E2E (End-to-End) LP reservations consume lots of resources by static allocation. This results in considerable waste of resources. Therefore, there is a need of a capacity limitation when available resources are allocated to the E2E LP all at once [1]. In order to solve this problem, network technologies and researches on the virtualization of network resources and dynamic resource allocation have been variously presented in this literature. Virtualization network resources and dynamic resource allocation have been mainly used in such technologies and researches. The E2E LP, which is created over optical network, is suitable for the performance and quality of network required by research demands, but may suffer from the continuous waste of network resources. Therefore, it is necessary to manage E2E LP by efficient resource allocation. Moreover, it would be beneficial to create as newer LPs as possible under limited network resources [2] [3]. In this paper, we suggest the network design for VRG that can make with integrating the set of E2E LP technologies. In addition management of VRG for researchers which separated from Original Domain and support tight cooperative work. The rest of this paper is organized as follows. Section 2 describes the characteristics of E2E LP. In Section 3, we suggest the design of organizing the VRG that can be configured with optimal number of LPs. In this section, we also describe structure and management for VRG, and show the improvement of LP by testing on media transmission and data transmission. Lastly, we conclude the paper along with our plans for future work.

2

Characteristics of E2E LP

Typically, E2E LP is composed of several parts. These parts are connected by network equipments, such as an amplifier, DWDM (Dense Wavelength Division Multiplexing), borders, and cross links. Each of optical equipments makes the path in transport network part and connects to systems (e.g., users, equipments, servers, etc.) via end site network systems. The path established between end systems is LP. As the LP, EOS (Ethernet Over SONET/SDH) can efficiently manage network resources by encapsulation. The path, that is composed of Ethernet switching and optical signal

Organizing Virtual Research Groups with Light Path Technology

405

integration in transport layer, is conversed to Ethernet frame in consumer layer. The LP determines the path through the network composed of several nodes and it has the function of dividing and allocating the bandwidth of STM-N/OC-N (Synchronous Transport Module-N, Optical Carrier-N) units on E2E links. The E2E link configured by the LP manages end-to-end network bandwidth and meets the quality requirement of services such as High Speed Lambda Service (HSLS), On-Demand Communications Circuit (OCC), Optically Extended LAN (OEL), and so on [4]. The LP on circuit level was able to avoid congestion with other traffics and It could minimize the network delay (Serialized Delay) with away from network electronic element. The path constructed with reserved resources makes very stable network environment and provides the high available bandwidth and throughput.

Fig. 2. TCP throughput test on Light Path and IP network

Fig. 2 shows the throughput for LP (red line) of about 870Mbps on 1Gbps links. In this figure, we can see that the resources of more than 87% are available and thus throughput line is extremely stable than IP network (green line). Thus, we can confirm the performance and stability of LP, and the suitability of establishing virtual cooperative environment. Consequently, LP can be configured on distributed and long distance environment. Organizations, researchers, and communities can tightly connect and comprise VRG which close to the local network environment that not only connected stable network but also can provide excellent transmission performance while it is possible.

3

Related Work

We expect that E2E LP technologies will be employed in large-scale optical networks in the near future. To improve the flexibility and efficiency of the transport network resource utilization, virtualization technologies have been widely extended from the computer area to the network area [5]. Based on testing results presented in Section 2, the E2E LP can offer users the fully allocated bandwidth usage of wavelength by isolating resources in optical layer. We can assume that the E2E LP is an ideal service to the users who have a strong requirement on large and stable network resources on virtual research environment.

406

M.-K. Noh et al.

The researchers belonged to a VRG have increasingly used the LP to connect their partners or shared resources. However, network resource allocation and LP creation will meet to network resource limitation. One of the most important issue is the efficient management and allocation of the dedicate network resources and the creation of VRG. In this section, we address the issues on network organization and efficient management. 3.1

Design of Virtual Research Group with E2E LP

First of all, we have to define optical network elements composed of VRG. As shown in Fig. 3(a), VRG is comprised of 3 sites. In order to join in cooperative research community on E2E LP, they need several network elements. Core nodes (N) are connected other core nodes by WDM links. An important factor in optical network is the product of the data rate of single wavelength, and the number of wavelengths of a fiber line [3]. These factors mean that total capacity of fiber increases with the number of Lines and the number of direct connections to neighbor core nodes. In addition, a core node can provide interface to access links and can be extended to transport network layer (see Fig 3(b)).

(a) Links of VRG

(b) Network Components of E2E LP

Fig. 3. Network components of VRG

Fiber lines (L) in Fig. 3(b) are connected to each core node, allocating bandwidth access on the Fiber lines to reserve each LPs. Access Links are connected end sites electronic interface and interface of Multi Service Provision Platform (MSPP) on core node. In addition, each link consists of several Lambdas (λ) and the bandwidth allocated by control plan. Therefore, we can define components of LP which connect to VRG as each lines and lambda with bandwidth. Also, the number of lambdas can be allocated on WDM line. The bandwidth allocated on each lambda means the available bandwidth of total links and total lambda’s capacity.

Organizing Virtual Research Groups with Light Path Technology

407

Fig. 4. Connect to network node with E2E LP

Let us assume that the bandwidth of the access links (B) can serve to end-user. Given the above network diagram and network components, we can formulate the increasing number of the LP based on required bandwidth (see Fig. 4). =

=

(1)

In order to maximize the effectiveness of E2E LP classify researcher and organizations involved with the VRG. In order to increase the values of LP, it needs to remove ports or hops on nodes. To solve this problem, we installed the L1/L2 switch to reduce hop counts. By doing this, the total number of LPs assigning to the VRG can be increased. As shown in Fig. 5, the inputs are the physical topology of the above optical network and the number of λs on each node. Each λ can be more reserved than before when it has a bypass. The sum of paths between source (S) and destination (D) should be bigger than the sum of λ or same. |

|

(2)

The value of path number between destination and source is closed to the value “d-1”. |

|=

1

(3)

Equation (3) indicates that the destination of the hop count information can be accessed in the same VRG.

408

M.-K. Noh et al.

Fig. 5. Components of optical network for creating LP

The value of latency should be considered and the information of node located in optimal region that establish for research purposes. Lastly, we have to consider capacity and fiber on each node. We can’t create anymore LP on this site if there is no more capacity on a fiber. Thus, we have to increase the number of network components related with the core node (N), it can make more bypaths between source and destination. Consequently, we establish more paths on above optical network. In the previous study, we can know that to maximize the effectiveness of LP classified researcher and organizations involved with VRG. It means that not only can control location of virtual node with minimum wastage WDM links but also provide environment guaranteed network qualities required as the same. In other words, logical configuration for each virtual environment beyond of the existing research in the local domain, considered submitted requirements, and optimize LPs regarded design the network suggested by this paper. • The optical links independently connect to researchers with each collaborative groups and these links connect to VRG than existing local network domain. - Estimate the resource transformed by optical network components - Set the bandwidth with links and wavelengths. - Calculate total of access link connected to VRG with network resource • Regard as same Domain node that is located in a single hop, and we organized the LP network that requires maximum hop is (N-1). • If there are additional maximum (N-1) hops between source and destination. - Choose the location of extend LP to destination. - Reduce hop count of maximum less than (N-1) - Terminate E2E Light Path and form VRG which connected to researchers and organizations with same purpose

Organizing Virtual Research Groups with Light Path Technology

409

Organize flow of Virtual Research Group are as follows.

Fig. 6. Flow chat of organization of Virtual Research Group

3.2

Experimental Install and Performance Test

We conducted performance tested for a network environment. As shown in Fig. 7, we constructed the experimental network services for medical research group over KREONET (Korea National Research and Science Network). In case of medical research group in the network environment, particular network resource is required according to applications types. For example, high-resolution image data transmission application needed available bandwidth, sensitive video conferencing including video stream of patients need an excellent quality network. First, Fig. 7 shows the network diagram designed for medical researchers. Each of organization alternately serves the multiple stream connection for other researcher and other member being a client receiving the stream and image data. The result of test is shown in Fig 7. Above all we have confidence that LP can provide more number of channels and higher resolution image transmission with guaranteed latency and jitter. Since Bytes In/Out amount of data is very similar with every other traffic, in spite of fifth LP, site (2) is created on other network region and connected with fiber channel node via L2 switch. Moreover, network of low value of jitter and RTT is to guarantee real-time multimedia stream transmission with a satisfactory delay and buffer requirements.

410

M.-K. Noh et al.

Fig. 7. Experimental Extension S/W install and result

4

Conclusion and Future Plan

We proposed a new optical network service for planning and allocation of E2E LP links and optical network resource under organizing node and using layer 2 and layer 1 extension switch. The approach efficient allocation and performance evaluation through a node and path increase network capacity for optical network and guarantee the network-level QoS such as latency, jitter, packet loss, etc. Test results indicate that our network design can efficiently allocate E2E LP for researchers, and show the network performance made up with network components. It is designed to be quite flexible in terms of create LP, to be suitable for research group have to be on multi-links [6]. In this paper, we composed the LPs operated on L1/L2 infrastructure through the Lambda networking technique. This technique can be used to compose the improved networks in terms of stability and performance. However, there are some disadvantages associated with restricting direct resource selection by users. To solve these problems, we have the plan to conduct the dynamic resource allocation that can automatically organize LP composition on a middleware through information exchange with resources [7].

References 1. Benjamin, D., Trudel, R., Shew, S.: Optical services over the intelligent optical network. IEEE Communication Magazine 39(9), 73–78 (2001) 2. Rajagopalan, et al.: IP Over Optical Networks: Architectural Aspects. IEEE Commun. Mag. 38(9), 94–102 (2001)

Organizing Virtual Research Groups with Light Path Technology

411

3. Ramaswami, R., Sivarijan, K.N.: Optical networks–a practical perspective. Morgan Kaufmann Publishers, San Francisco (1998) 4. Smit, M.K., Dam, C.: PHASER-based WDM-devices: principles, design, and application. IEEE Journal of Selected Topics in Quantum Electron 2(2), 236–250 (1996) 5. Haque, A., Ho, P.-H., Boutaba, R., Ho, J.: Group shared protection (GSP): a scalable solution for spare capacity reconfiguration in mesh WDM networks. In: Proceedings of the 47th IEEE Global Telecommunications Conference, Dallas, TX, vol. 3, pp. 2029–2035 (2004) 6. Ho, P.-H., Mouftah, H.T.: On optimal diverse routing for shared protection in mesh WDM networks. IEEE Transactions on Reliability 53(6), 216–225 (2004) 7. Ramamurthy, B., Ramakrishnan, A.: Design of virtual private networks (VPNs) over optical wavelength division multiplexed (WDM) networks. SPIE Optical Networks Magazine 3(1) (2002)

Remote Monitoring Information Management System for Preventing Performance Degradation of Database Myung-Ju Kim1, Un-Bai Lee2, and Kwang Sik Chung2,* 1

2

Service Development Team, TELCOIN Co., Ltd. Dept. of Computer Science, Korea National Open University

[email protected], {lub,kchung0825}@knou.ac.kr Abstract. In order to guarantee the stability and reliability of database performance, database monitoring system and program are necessary. The previous commercial database monitoring servers and programs make database overhead and degrade performance of database management system. And there were no alert message or warning message for database administrators in the previous commercial database monitoring system. We propose the remote database monitoring information management system without degrading performance of database. In the proposed system, remote monitoring information database is constructed by collecting monitoring information from each database server and the remote database monitoring information server deliver monitoring information and each database state to administrators. Thus additional overhead of database monitoring does not occur in the proposed system. The proposed system uses a smart phone as administrator terminal and send alert or warning message to administrators. It provides monitoring service without limitation of time and place of administrators. Keywords: database monitoring system, smart phone, monitoring server, database monitoring information.

1

Introduction

In order to manage Database stably, many database monitoring systems were developed and are now installed and used commercially in many fields. In the recent, previous commercial monitoring systems are directly connected with target database systems through sessions, and, in order to collect monitoring information about database, regularly and frequently send queries to the database. In previous database monitoring systems, input/output transactions of database, session of connection and query processing overhead increases and performance of database decreases, as the number of database monitoring session increases. Since terminal with previous database monitoring software only can monitor the target database, monitoring operation place and time is limited. While an administrator does not run database monitoring software, urgent alert or urgent changes can not be monitored by the administrator. *

Corresponding author.

T.-h. Kim et al. (Eds.): FGCN 2011, Part II, CCIS 266, pp. 412–418, 2011. © Springer-Verlag Berlin Heidelberg 2011

Remote Monitoring Information Management System

413

Thus we propose new database monitoring system that can avoid the database monitoring overhead and monitoring place and time limitation. In the proposed database monitoring system, we use a smart phone as monitoring terminal, support real-time SQL processing state information for database tuning, and avoid performance degradation of database that comes from monitoring queries process.

2

Remote Monitoring Information Management System

In this chapter, we propose remote monitoring information management system architecture and implementation. Figure 1 is the proposed remote monitoring information management system architecture that can guarantee the stability of database performance that can be affected by monitoring query process.

Fig. 1. Remote Monitoring Information Management System Service Flow

In figure 1, the remote monitoring information server is connected with database server 1 ~ N with Java RMI mechanism. The remote monitoring information server protects database servers from malicious and abnormal access with unpermitted IP address and port number to database servers. A smart phone as administrator’s monitoring terminal uses Wifi networks or 3G networks for connection with the remote monitoring information server and can receive a warning message from the remote monitoring information server. And anywhere and anytime a database administrator can monitors state of database through the remote monitoring information server. Figure 2 is software architecture of the remote monitoring information system.

414

M.-J. Kim, U.-B. Lee, and K.S. Chung

Fig. 2. Remote Monitoring Information Management System Architecture

In figure 2, the remote monitoring information server is implemented with JSP and consists of MIP(Monitoring Information Provider), MIS(Monitoring Information Server), and MIA(Monitoring Information Analyst). MIP(Monitoring Information Provider) delivers monitoring information to a smart phone. MIS(Monitoring Information Server) collects monitoring information from database servers and stores collected monitoring information to monitoring information database. MIA(Monitoring Information Analyst) analyzes monitoring information to monitoring information database and decides whether current state is alerted to an administrator or not. MIC(Monitoring Information Client), monitoring information agent, in database servers regularly collects and processes monitoring information, and sends it to MIS. Lastly, on smart phone, MICA(Monitoring Information Client Application) has monitoring information display function and warning message receiving function are implemented with Android platform. MIA regularly analyzes the latest accumulated monitoring information on monitoring information database. If MIA detects abnormal state of database or abnormal monitoring information of database, then MIA makes alert message for an administrator. MIA is implemented with Java programming language and SQL. Measured monitoring information value is ‘S’, base standard value is ‘L’, an number of base standard excess is ‘C’, and delivery message is ‘M’. Figure 3 is an algorithm in which MIA detects base standard excess and makes and saves an alert message table.

Remote Monitoring Information Management System

415

After processing the request of MICA, MIP converts requests result into XML format file and sends it to MICA. MIP that is implemented with JSP, SQL, XML is run on Tomcat. MIP provides database monitoring information search service and alert message search service. Figure 3 and figure 4 are MICA request monitoring information with XML format from MIP.

Fig. 3. Database Network Traffic Request

Fig. 4. Database Session Request

3

Evaluation of the Remote Monitoring Information Management System

We measure database input/output traffic changes on proposed remote monitoring information management system and the previous system according to monitoring requests increase. Table 1 shows that measurement of database input/output traffic with 5 minutes interval changes according to the number of connection. We use Lab128 program for experiments, and, for accurate measurement, repeatedly measure database input/output traffic of the previous system and the proposed database monitoring system after 2 minute 36 seconds from the time of increasement of monitoring connection. Since connection session of Lab128 program is established for database input/output traffic measure, although the number of monitoring connection is 0, there is initiative traffic.

416

M.-J. Kim, U.-B. Lee, and K.S. Chung Table 1. Input/output Traffic Measurement Results Connection #

0

1

2

3

4

5

input (KB/s)

Previous system Proposed system

0.181 (0) 0.159 (0)

0.358 (0.177) 0.168 (0.009)

0.506 (0.148) 0.169 (0.001)

0.644 (0.138) 0.166 (-0.003)

0.774 (0.13) 0.168 (0.002)

0.931 (0.157) 0.167 (-0.001)

output (KB/s)

Previous system Proposed system

2.08 (0) 2.064 (0)

4.254 (2.174) 2.112 (0.048)

6.587 (2.333) 2.170 (0.058)

8.602 (2.015) 2.081 (-0.089)

10.856 (2.254) 2.101 (0.02)

13.337 (2.481) 2.311 (0.21)

In Table 1, database input traffic in the previous system increases by 0.15 KB per second and database output traffic in the previous system increases by 2.25 KB per second. In the previous system, as the number of monitoring connection increases, each connection session to the target database is respectively established which makes additionally input/output traffic. But, in the proposed monitoring information management system, only one connection session to the target database for monitoring information collection is needed. And although the number of administrator and query request increase, request reply traffic does not increase. The reason is that the proposed monitoring information management system needs only connection with the remote monitoring information server. Figure 5 shows the database input traffic changes on proposed remote monitoring information management system and previous system.

Fig. 5. Database Input Traffic Comparison

Remote Monitoring Information Management System

417

Figure 6 shows the database output traffic changes on the proposed remote monitoring information management system and previous system.

Fig. 6. Database Output Traffic Comparison

As the number of database monitoring connection and query increases, database input/output traffic and session management overhead increases in the previous system. But in the proposed remote monitoring information management system, instead of the direct connection with each database server, remote monitoring information server is needed and the number of administrator and connections session managements do not make additional overhead of each database server.

4

Conclusion and Future Work

The previous commercial database monitoring and tuning systems degraded performance of database system. And they limited place and time of administrator’s operation. Lastly they did not send any alert message to administrators while administrators are out of operation rooms. In order to decrease the overhead of database server occurred by database monitoring and alert administrators while out of operation room, we propose and develop the remote monitoring information management system and evaluation it. The remote monitoring information management system has several advantages as follows, from comparison with the previous database monitoring systems. First, since the remote monitoring information management system requests and collects database

418

M.-J. Kim, U.-B. Lee, and K.S. Chung

monitoring information from the remote monitoring information server, it degrades the performance of each database server. Second, since the remote monitoring information management system can use both of a desktop PC and a smart phone as a monitoring terminal of an administrator, there are no limitation for time and place of monitoring operation. Lastly, since the remote monitoring information management system uses a smart phone as a monitoring terminal, instant alert message can be sent to an administrator and seamless database monitoring operation is possible for an administrator. But the remote monitoring information management system can provide only accumulated database monitoring information on the remote monitoring information server to an administrator. And additional cost for developing the remote monitoring information management system is needed. We have a plan to study to avoid time gap between real-time database monitoring information and accumulated database information on the remote monitoring information server.

References 1. 2. 3. 4. 5.

Toad for Oracle, http://www.toad.co.kr/ Lab128, http://www.lab128.com/ Oracle Database, http://www.oracle.com/kr/index.html Android, Android Developers, http://developer.android.com/index.html Jang, S.: Monitoring System using Mobile Device for Database Administrators, Master Thesis, Graduate School of Computer Science and Technology, Korea University (2001) 6. An, T.: A study on establishment and management plan for database integration monitoring system in corporatio, Master Thesis, The Graduate School of International Affairs and Information, Dongguk University (2005)

Noise Reduction Scheme for Precise Indoor Localization Inseok Moon and Won-Kee Hong Dep. Of Information and Communication Engineering, Daegu University Gyeongsan, Gyeongbuk, Rep. of Korea [email protected], [email protected]

Abstract. The indoor localization is very useful and essential technique in several application area of the wireless sensor network. Various noises such as clock offset, clock drift, and other environmental noises are a hindrance to the accurate localization measurement of a mobile node. In this paper, a new indoor localization scheme is introduced that takes the Kalman Filter to reduce the noises and meausre distances between nodes accurately.The experimental results show that the proposed method improves the accuracy of distance measurement by 18%. Keywords: Localization, Distance Measurement, Kalman Filter, SDS-TWR.

1

Introduction

The wireless sensor network, where lots of tiny sensor devices monitor their surrounding environment continuously and in real time, demands the localization information of the sensor device. While the outdoor localization is easily resolved by the GPS [1], the indoor localization gives several challenging issues such as accurate measurement, real time computation and system load [2],[3],[4]. The indoor localization techniques can be categorized into range-based localization and rangefree localization depending on the requirement of distance measurement. In general, range-based localization provides an exact localization because of being based on the distance between nodes. The exact measurement of distance between two nodes is the basis of figuring out the position of a node. The IEEE 802.15.4a CSS (Chirp Spread Spectrum) was produced that is a standard protocol to measure the distance between nodes via wireless RF communication [5]. It allows the information of time to be sampled by the chirp signal. It uses TWR (Two Way Ranging) or SDS-TWR (Symmetric Double Sided Two Way Ranging) algorithm to compute the distance by using the sampled time. The TWR and SDS-TWR are the schemes to reduce the time error due to the time asynchronization between nodes [6]. However, they still suffer from the clock offset, clock drift and other environmental noise [7]. This is the significant problem in the indoor localization that only tolerates error with centimeters. T.-h. Kim et al. (Eds.): FGCN 2011, Part II, CCIS 266, pp. 419–428, 2011. © Springer-Verlag Berlin Heidelberg 2011

420

I. Moon and W.-K. Hong

In this paper, a new approach exploiting the Kalman Filter is presented to obtain an accurate localization. It is based on the IEEE 802.15.4a protocol and SDS-TWR for distance calculation. The Kalman Filter assumes any measured data on an object contains a probabilistic error and the current state of the object has a linear relation with its previous state. The Kalman Filter has the characteristics that it does not have to hold all the previous data. That is, it needs only current and previous data and to get a distance. The system values of the Kalman Filter are defined in this paper to reflect the noise in the distance measurement. Experiments are performed by comparing measurement of distance and location depending on the application of the proposed Kalman Filter based on the SDS-TWR scheme. The experimental results shows that the proposed method using the Kalman Filter improves the distance measurement accuracy by 18%. The organization of this paper is as follows. Section 2 introduces the related works about localization in WSN. In section 3, the proposed distance measurement scheme using Kalman Filter will be explained. The experimental results and analysis are presented in section 4. Finally, section 5 contains concluding remarks.

2

Related Works

In this section, researches on the localization in the literature are introduced, categorized with the range-based scheme and range-free scheme. Moreover, recent studies on the exact distance measurement in the ToA (Time of Arrival) based localization are described. The range-based scheme is a technique to figure out a node’s position based on the distance from beacon nodes which already know their position. The RF’s inherent characteristics such as time of flight, reception angle of antenna, or signal strength are exploited to get the distance between two nodes. The RSSI (Received Signal Strength Indicator) exploits the characteristics that as the distance is longer, RF’s signal strength will be weaker [8]. However, it has a difficulty to measure a distance exactly because the signal strength is very susceptible to the environmental noise and its fluctuation is very severe. The ToA (Time of Arrival) takes advantage of the time of flight to measure the distance [9],[10]. Basically, it performs the distance calculation with the time stamp received from the other node so that the time synchronization among nodes is assured to get the exact distance. The TDoA (Time Difference of Arrival) obtains the distance using the time difference between the different received signals. It requires time synchronization among beacon nodes and extra hardware cost due to the additional communication device [9],[10]. The AoA (Angle of Arrival) uses the directional angle of received signal and the crossing point between nodes [9],[10] requires at least 2 directional angles to calculate the distance. It has a difficulty in getting exact directional angles due to the multi-path fading. While the range-based scheme is a mathematical approach based on the distance between nodes, the range-free scheme is a parametrical approach to exploit surrounding information around a node such as surrounding images and neighbor nodes to estimate its position. The Centroid [11], APIT [12] and DV-Hop[13] are the

Noise Reduction Scheme for Precise Indoor Localization

421

representative range-free scheme. In the Centroid, there are fixed nodes that are deployed at equal interval. A mobile node gets the information packets, which holds the position coordinates of a fixed node, periodically from surrounding nodes. It calculates its location by averaging all the coordinates collected from the surrounding nodes. It requires high density of fixed nodes deployed to obtain an exact location, which results in high cost. The APIT is a method to figure out rough location of an object using surrounding node. It scans out neighboring nodes within a one-hop radius of a node requesting its location and then picks out three nodes forming the smallest area of triangle which includes the requesting node. The DV-hop calculates a node’s location with the number of hops assuming all the hop distances are the same. 2.1

Ranging Protocol in ToA Based Localization

In general, distance is obtained by multiplying time by velocity. Because the speed of light is a constant parameter, distance is determined by the time taken a message to be transferred between two nodes. That is, the transfer time can be a difference between sending time (t ) and arrival time (t ) of message shown in Fig. 1. Therefore, the distance dAB can be denoted by the following equation: =

(1)

Fig. 1. The basic distance calculation between node A and node B

However, the equation (1) does not consider the time error due to asynchrony between two nodes. Given that time error is ε, the distance dAB can be denoted by the following equation: (2) = This is a one way ranging (OWR) scheme that requires time synchronization between nodes to improve distance accuracy. The two way ranging (TWR) scheme removes the requirement of time synchronization by transferring two types of messages back and forth as shown in Fig. 2. Using TWR, the time difference (t P ) can be obtained like following: =

1 2

(3)

422

I. Moon and W.-K. Hong

Fig. 2. The communication flow between node A and node B in the TWR

In equation (3), the time error due to time asynchrony is removed but clock offset error within a node still remains. If the clock offset errors are εA and εB at node A and node B respectively, the time difference (t̃ P ) can be denoted by the following equation: 1 (4) t t t̃ = t 1 ε t 1 ε P

A

2

B

Thus, time difference error can be defined by the following equation: 1 t t εA t t εB t̃ P t P = (5) 2 1 t P εA = t εB B εA 2 t is much larger than t P since it includes the time to prepare a reply message. In general, it takes a few milliseconds, while propagation time is only several nanoseconds. It implies that the error largely depends on the reply time at node B and the difference between clock offsets of node A and node B. In order to reduce the time difference error, Symmetric Double Sided-Two Way Ranging (SDS-TWR) sends a ranging message twice each other as shown in Fig.3. Given with no clock offset, the propagation time (t P ) in SDS-TWR can be denoted by the following equation: t

t

1 tP = 4

= t

A

t

A

t

t

4t P

B

t

t

(6)

B

When clock offsets are εA and εB , the propogation time (t̃ P ) can be defined by the following equation: t̃ =

1 4

t

t

A

1

εA

t

t

B

1

εB

(7)

Noise Reduction Scheme for Precise Indoor Localization

423

Fig. 3. The communication flow between node A and node B in SDS-TWR

The time difference error can be derived from the above two equations: t̃ P

tP

= =

1 t 4 1 2

tP tA

t tB

A

1 4

εA

t

trelpyA

t trelpyB

B

εB

εB

(8)

εA

where t = 2t t = 2t P t B, t B As shown in equation (8), SDS-TWR reduces the time difference error by more than a half of that of TWR. This equation also implies that the error gets reduced as the preparation times and clock offset of each node get converged at some point.

3

Ranging Compensation with Kalman Filter

The Kalman filter is a recursive filter to track along noisy status in a dynamic linear system. The ranging compensation scheme with the Kalman filter designed in this paper removes the error from the distance obtained by the SDS-TWR passing through the predict phase and the correct phase of the Kalman Filter as shown in Fig. 4. The proposed method takes a set of measured distance Z as an input and produces a set of estimated distance X as an output. = =

| = 0,1, … , | = 1, … ,

424

I. Moon and W.-K. Hong

Fig. 4. Ranging compensation with Kalman Filter

What should be done to advance through the Kalman Filter is to model the localization system which consists of the system model and the measurement model. The system model and measurement model can be defined as follows respectively: = = The A is a coefficient that relates the estimated value at the previous step with the one at the current step without considering noise in the system. It is assumed that A is one because the distance should always be consistent in case that the object does not move and there is no noise. In the system model, the is a noise in the system and it has a Gaussian distribution. Given the system noise covariance is , it is defined by the following: ~

0,

,

where N (a, b) is the Gaussian function with mean a and covariance b. In this paper, it is assumed that there is no noise in the system. Therefore, can be set to 0. The H is a coefficient that relates the estimated value at the current step with the measured value at the same step without considering noise in the system. The H is set to one because the measured value should be exactly the same as the estimated value if there is no noise.

Noise Reduction Scheme for Precise Indoor Localization

425

In the measurement model, the is a noise occurred in the real measurement and it has a Gaussian distribution. Given the measurement noise covariance is , it is defined by the following: ~ 0, The measurement noise is assumed to be the difference between the measured data at the current step and the estimated value at the previous step. |

=|

where is a measured data at step i and is an estimated data at step i-1. Then, the measurement noise covariance can be defined by the following: = =

|

| |

|

The initial estimated value x and covariance Matrix P is needed. In this paper, it is assumed that x is z and P is 1.

4

Experimental Results and Analysis

In this section, results of experiments are presented and analyzed comparing measurement of distance and location depending on the application of the proposed Kalman Filter based on the SDS-TWR scheme. KF_ST stands for the SDS-TWR with Kalman Filter and ST stands for the SDS-TWR without Kalman Filter. The three beacon nodes which are denoted by A, B, and C and already know their locations and a node which wants to figure out it location are placed in an indoor space as shown in Fig. 5. The real distance between the node and three beacon nodes are 3, 4 and 5 meters respectively.

Fig. 5. Deployment of beacon node (A, B, C) and mobile node used in the experiments for performance evaluation

426

I. Moon and W.-K. Hong

Fig. 6 shows the results of comparing the distance measurement of the KF_ST and then ST when the real distances are 3, 4, and 5 meters respectively. While the measured data by the ST shows lots of fluctuation, those measured by KF_ST converges into a specific value as the number of measurement goes by. It is because the KF_ST can reduce the noise effectively. Table 1 shows the average of 100 distance measurements of the ST and the KF_ST. While the average distance error of ST is 58cm that of KF_ST is shown to be 40cm.

Fig. 6. Distance measurement of ST and KF-ST Table 1. Average measured distance A node[m]

B node[m]

C node[m]

ST

3.55

4.77

5.44

KF_ST

3.27

4.67

5.27

Noise Reduction Scheme for Precise Indoor Localization

427

The node location (x, y) can be obtained by the trilateration using the measured distances from the three beacon nodes. The real node location is (290, 130). Fig. 7 shows the frequency of x-coordinate and y-coordinate of the node location calculated 100 times by distance values measured by the KF_ST and the ST. Fig. 7 (a) shows all the x-coordinates corresponds between 300 and 330, which are very close to the real x-coordinate in case of the KF_ST but the ST has a wide distribution of x-coordinates. Fig. 7 (b) shows all the y-coordinates correspond between 290 and 320, which are very close to the real y-coordinate. Therefore, these results demonstrate the KF_ST provides the position of a node more accurately than the ST. In the KF_ST, the error rate of x-coordinate is 21% and that of y-coordinate is 12%. This is x-coordinate accuracy by 18% and y-coordinate accuracy by 24% lower than that of ST.

Fig. 7. Distribution of measured location

5

Conclusions

The exact measurement of distance between nodes is very important in the localization. The distance measurement scheme through wireless message packet transmission should resolve the time synchronization problem to obtain an exact distance. The TWR and the SDS-TWR are the distance measurement methods to overcome the time error due to clock drift or clock offset of a node. However, the distance error still remains and it is fatal defect in indoor localization application that requires higher location accuracy. In this paper, a method is proposed to perform an exact distance measurement by removing noise using the Kalman Filter. The Kalman Filter is used to remove the noise produced in the SDS-TWR. The performance evaluation is conducted in terms of distance measurement and location measurement using Nanotron NanoLOC. The distance error in the KF_ST that adapts the Kalman Filter to the SDS_TWR is 18% lower than that in the ST that does not use the Kalman Filter. Consequently, the coordinates calculated by the KF_ST is shown to be deployed densely around the real coordinates.

428

I. Moon and W.-K. Hong

Acknowledgments. This research was supported by the Daegu University Research Grant, 2010.

References 1. Per, E., Todd, W., Sam, P., Changdon, K., Yi-Chung, C., Yeou-Jyh, T.: Wide AreaAugmentation of the Global Positioning System. Proceedings of the IEEE (1996) 2. Morellu, C., Nicoli, M., Rampa, V., Spagnolini, U., Alippi, C.: Particle Filters for rssbased localization in wireless sensor networks an experimental study. In: ICASSP, pp. 957–960 (2006) 3. Hyuntae, C., Yeonsu, J., Hyunsung, J., Ingu, P., Yunju, B.: Precision Time Synchronization System over Wireless Networks for TDOA-based Real Time Location Systems. Journal of Korean Information and Communications Society 34(1) (2009) 4. Ahmad, H.: Application of Channel Modeling for Indoor Localization Using TOA and RSS. Worcester polytechnic institute in partial fulfillment of the requirements for the Degree of Doctor of Philosophy (2006) 5. Hach, R.: Symmetric double sided two-way ranging. IEEE 802.15.4a standard, doc. IEEE P.802.15-05-0334-00-004a (2005) 6. Jiang, Y., Leung, V.: An Asymmetric Double Sided Two-Way Ranging for Crystal Offset. In: Int’l Symposium on Signals, Systesms and Electronics (ISSSE 2007), pp. 525–528 (2007) 7. Heidarian, F., Schmaltz, J., Vaandrager, F.: Analysis of a Clock Synchronization Protocol for Wireless Sensor Networks. In: Cavalcanti, A., Dams, D.R. (eds.) FM 2009. LNCS, vol. 5850, pp. 516–531. Springer, Heidelberg (2009) 8. Bahl, P., Padmanabahan, V.: RADAR: An In-Building RF-based User Location and Tracking System. In: Nineteenth Annual Joint Conference of the IEEE Computer and Communications Societies, vol. 2, pp. 775–784 (2000) 9. Boukerche, A., Oliceira, H.A.B., Nakamura, E.F., Loureiro, A.A.F.: Localization systems for wireless sensor networks. IEEE Wireless Communications 14, 6–12 (2007) 10. Sayed, A.H., Tarighat, Khajehnouri, N.: Network-based wireless location: challenges faced in developing techniques for accurate wireless location information. IEEE Signal Processing Magazine 22, 24–40 (2005) 11. He, T., Huang, C., Blum, B.M., Stankovic, J.A., Abdelzaher, T.F.: Range-Free Localization Schemes in Large Scale Sensor Networks. In: Mobicom 2003, pp. 81–95 (2003) 12. Roxin, A., Gaber, J., Wack, M., Nait-Sidi-Moh.: Survey of Wireless Geolocation Technoques. In: Globecom Workshops 2007 IEEE, pp. 1–9 (2007) 13. Niculescu, D., Nath, B.: Ad-Hoc Positioning Systems (APS). In: IEEE GLOBECOM 2001, pp. 2926–2931 (2001) 14. Kleinbauer, R.: Kalman Filtering Implementation with Matlab. Study Report in the Field of Study, Geodesy and Geoinformatics at Universitat Stuttgart

Development of a Korean Language-Based Augmentative and Alternative Communication Application Chang-Geol Kim1, Soo-Won Kwak2, Ryu Juang Tak1, and Byung-Seop Song1 1

Deptartment of Rehabilitation Science & Technology, 2 School of Electronics, Daegu University, Jillyang-eup, Gyeongsan-si, Gyeongbuk, Korea {chang014,jryu,bssong}@daegu.ac.kr, [email protected]

Abstract. Communication is an essential element of human interaction with a community. People with communication disorders that would otherwise impede this interaction can interface with their communities via augmentative and alternative communication. The Korean government supports these disabled people with augmentative and alternative communication devices. However, most devices require high cost and have only simple functions. Furthermore, they require space for storage. As an effort to relieve the difficulties of using such devices, this study develops an augmentative and alternative communication application that can be mounted on widely spreading smartphones and tablet PCs. Keywords: Augmentative and Alternative Communication, Tablet PC, Application

1

Introduction

Communication is one of the most fundamental elements by which a person contacts others and lives as a social creature. Lack of communication results in restricted social interaction, emotional withdrawal, anxiety, and frustration. Furthermore, it severely restricts the ability to learn new things [1-3]. For those who have communication disorders, a supplementary approach called Augmentative and Alternative Communication (AAC) has been designed to assist and improve independent communication in every situation. An AAC system consists of symbols, assistive aids, techniques, and strategies. Symbols include body languages, pictures, gestures, facial expressions, paintings, words, line drawings, and Blissymbols. Assistive aids include physical tools such as communication boards, communication books, and communication devices used for exchanging messages. Techniques such as direct selection, scanning, and encoding determine how to deliver messages. Strategies are plans to efficiently utilize the symbols, aids, and techniques to improve communication [1]. T.-h. Kim et al. (Eds.): FGCN 2011, Part II, CCIS 266, pp. 429–436, 2011. © Springer-Verlag Berlin Heidelberg 2011

430

C.-G. Kim et al.

Another essential element of prompt interaction with a community in modern society is access to information, i.e. information technology equipment. People utilize information technology to learn skills for daily life and economic activities and to enjoy society and culture [4]. However, not everyone has access to this benefit. Those who cannot use information technology or access information due to their poverty or physical disabilities are alienated and isolated, resulting in yet another handicap in terms of social activity. In order to relieve these social disabilities via supporting disabled people’s access to information, the Korean government has been supplying information aids to disabled people since 2003. To some extent, governmental efforts can liberate the disabled from their isolation from information. However, living as an integrated member of a community and interacting with that community places a number of demands on a disabled person. These demands include purchasing new devices and learning new skills. Patients with complicated problems, such as physical disabilities or brain lesions, may need one or more items of supplementary equipment such as an electric wheel chair, AAC device, alternative input device for access to information, or environmental control unit (ECU). Most of this equipment demands familiarity with device-specific operation instructions. Furthermore, it comes in a variety of appearances and requires space for storage [5]. To make matters worse, these devices are expensive, so the user’s financial burden is significant. In an effort to relieve financial and spatial difficulties and to reduce the inconvenience of carrying large amounts of equipment, this study develops a Korean AAC application that can be mounted on smartphones and tablet PCs.

2

Survey of AAC Devices

This chapter outlines the AAC devices selected as the Korean government’s official supplies in 2011. It will give an overview of common AAC devices used in Korea. (Korea National Information Society Agency, 2011) 2.1

SuperTalker

The SuperTalker is a voice output device that holds a voice recording of up to 16 minutes. Its dimensions are 32 20 45 cm , and it has exchangeable panels of 2, 4, and 6 grids that correspond to the user's ability for direct selection. Up to 8 additional switches can be installed to further support direct selection. Although it is the product of a U.S. brand, Koreans can use it easily because the user can record its output voices instead of relying on pre-recorded machine voices. However, it is limited by the fact that it takes a maximum of only 8 words, requiring the user to rerecord the voices and change the words depending on the situation.

Development of a Korean Language-Based Augmentative

431

Fig. 1. SuperTalker

2.2

OK Toc Talk

The OK Toc Talk is a product of Korea, which holds up to 24 minutes of voice recording. Its dimensions are 28.2 14.8 2.8 cm . Its settings are versatile enough to accommodate 2 users, and it supports 6 channels. Its voices and images can be edited on a PC. When the user changes the picture panel, the corresponding voices can be turned on by changing the channel to the one that matches the picture card. However, this device is useful only when the user is capable of direct selection, and the sizes of the direct selection buttons are not changeable. Furthermore, the available words are limited to a total of 120, and the user needs additional assistance to change the picture card.

Fig. 2. OK Toc Talk

2.3

KidsVoice

KidsVoice is a TTS-based device developed in Korea. Its and 1Kg system is equipped with Korean-familiarized image symbols. Its word sets

432

C.-G. Kim et al.

Fig. 3. KidsVoice

are selectable according to the situation, and sentence outputs can be generated from combinations of words. The size of the button is flexible, and it supports direct selection and scanning via the USB switch as well. Furthermore, the vocabulary and symbols are editable so that various AAC strategies can be applied. However, its price is as high as 2,970,000 Won, and configuring its numerous functions is not trivial.

3

System Design

The conceptual diagram of the application described in this paper is shown in Fig. 4. The symbol system used in the application is flexible enough in choosing texts or image symbols to accommodate various strategies. The assistive aid employs an Android-based tablet PC, which is widely used in Korea. The technique of word selection supports both direct and indirect techniques, so that various AAC strategic choices are available. The size of the button for direct selection is automatically adjusted according to the user-specified number of cells displayed on the screen. If more words are stored than the screen can display on one page, another page is automatically added so that all the stored words are accessible. The user can explore the pages by either selecting the corresponding page button on the bottom of the screen or by dragging the page. In order to help the user construct sentences, words that can follow the selected word are suggested in a dynamic display. Indirect selection is designed to make efficient use of limited space by scanning words one by one when the user touches the screen. Furthermore, various pre-set word panels can be prepared in advance, which the user can choose according to the situation.

Development of a Korean Language-Based Augmentative

433

Fig. 4. Conceptual diagram of the system

4

Implementation of the System

The AAC application is constructed based on the Android OS and mounted on a Samsung Galaxy Tab as the platform. The application works as follows: First, tapping the icon loads the application. The first screen to display is the front page of situation options, shown in Fig. 5. Selecting a situation leads to the primary word set associated with that situation. Now the system is ready to use. When the user selects a word from the displayed list, the AAC application displays the corresponding text in the text box at the top of the screen and outputs the voice using the onboard TTS engine. The default words used in the application are stored in specified text files, which the user can edit. The words consist of 2-level sets: primary words, which are mainly used in normal situations, and secondary words used in conjunction with the primary words to make sentences. The primary words are registered as the user creates a text file in a specified folder and names the file to reflect the situation. The application recognizes the secondary words of a primary word as it finds words in the text file whose name is the same as the primary word. As the user selects the situation from the front page, the application automatically distributes the primary words onto pages depending on the number of primary words and the number of cells on the screen. The pages are generated in the order of the words recorded in the text file. The total number of pages and current page number are indicated in red at the top-right corner of the screen. The pages are turned as the user touches the right and left buttons at the bottom of the

434

C.-G. Kim et al.

screen. When the user selects a primary word, the application searches for the same file name as the selected primary word and, if it is found, the screen turns to the list of associated secondary words, and the system waits for the user’s selection. If no secondary word is found, the system only displays the text and outputs the TTS voice for the selected primary word. The user can return to the situation page by tapping the text box at the top of the screen or the information pad.

Fig. 5. Implemented AAC application

4.1

Symbol System

The symbol system of the AAC application employs both texts and images. Clicking the menu button of the tablet PC activates the menu of the application, as shown in Fig. 6. The configuration button at the right corner leads to the set-up menu, where the user decides whether or not to use images. Furthermore, different images can be used for words if the jpg file corresponding to the word, i.e. the file in the image folder under the same folder as the text file, is edited or switched.

Fig. 6. Process to change the symbol system

4.2

Strategy

To allow more choices for direct selection, the application employs direct communication with a screen keyboard for users who are able to manipulate their fingers to select fine print. Furthermore, as shown in Fig. 7, the number of cells in direct selection can be controlled depending on the user's manual dexterity. A setting that allows users to turn the page by dragging is also available for users with hand skills fine enough to drag. These settings are stored in separate files so that they can be customized for different users.

Development of a Korean Language-Based Augmentative

435

Fig. 7. Process to change the number of cells

In the configuration for indirect selection, both activation of indirect selection and the scanning period for each word can be specified, as shown in Fig. 8.

Fig. 8. Configuration and usage of scanning

5

Conclusion

An AAC application was developed using an Android OS–based tablet PC. The application supports text and image symbols to help the user access the AAC via suitable strategies and techniques. The application can accommodate both direct and indirect inputs. The size and the number of cells can be adjusted according to the user’s motor skills. The scanning speed for indirect selection can be controlled.

436

C.-G. Kim et al.

The AAC application described here may relieve some of the financial burden of people with communication disorders, and it may also reduce the inconvenience of carrying many devices. As a result, it will give people with disabilities the opportunity to participate more fully in the community. Furthermore, because the application has versatile function settings, those who want to learn how to use AAC devices can experience various strategies without changing the device itself. Therefore, it can be available at minimal cost, making it useful in the field of education as an exploratory alternative to expensive professional AAC devices for children. This application has the limitation that, as it is based on the tablet PC, the TTS voice options are limited. Acknowledgement. This work was supported by the Korea Health Industry Development Institute grant funded by the Korea government (A101986).

References 1. Jeong, H.D., Kim, J.Y., Park, E.H., Park, S.J.: AAC Education for students who have disabilities. Korea National Institute for Special Education, Anshan, Gyeonggi-do (1999) 2. Lee, M.H., Park, E.H.: Effect of Peer Training for Peer-Mediated Intervention with AAC on the Social Interaction of Children with Severe Physical Disabilities. The Korean Journal of Early Childhood Special Education 6(1), 109–127 (2006) 3. Park, E.H., Bak, S.H.: A Survey of Special Education Teachers’ Perceptions on Education for Students with Severe Disabilities. Korean Journal of Special Education 36(1), 29–55 (2001) 4. Kim, T.I., Do, S.G.: The Analysis of Digital Divide between Disabled and Non-disabled People. Social Welfare Policy 21, 341–365 (2005) 5. Kim, C.G., Song, B.S.: Development of Integrated Computer Interface for Power Wheelchair User. Journal of biomedical engineering research 31, 251–257 (2010)

Adaptive Power Management for Nanoscale SoC Design Jeong-Tak Ryu and Kyung Ki Kim* School of Electronic Engineering, Daegu University, Gyeongsan, South Korea {jryu,kkkim}@daegu.ac.kr

Abstract. The demand for power sensitive designs in system-on-chip (SoC) has grown significantly as MOSFET transistors scale down. Since portable battery powered devices such as cell phones, PDA's, and portable computers are becoming more complex and prevalent, the demand for increased battery life will require designers to seek out new technologies and circuit techniques to maintain high performance and long operational lifetimes. As process dimensions shrink further toward nanometer technology, traditional methods of dynamic power reduction are becoming less effective due to the increased impact of standby power. Therefore, this paper proposes a novel adaptive power management system for nanoscale SoC design that reduces standby power dissipation. The proposed design method reduces the leakage power at least by 500 times for ISCAS’85 benchmark circuits designed using 32-nm CMOS technology comparing to the case where the method is not applied. Keywords: Power management, Stand-by power, Leakage power, SoC.

1

Introduction

The demand for power sensitive designs in system-on-chip (SoC) has grown significantly as MOSFET transistors scale down. Since portable battery powered devices such as cell phones, PDA's, and portable computers are becoming more complex and prevalent, the demand for increased battery life will require designers to seek out new technologies and circuit techniques to maintain high performance and long operational lifetimes. Lowering power-supply voltage in the system is one of the most effective schemes to reduce the power dissipation. A number of methods have been proposed to scale down the power-supply voltage dynamically [1]-[5]. Even though they are effective in decreasing dynamic power dissipation, they do not help reduce leakage power effectively. As transistor geometries are scaled down aggressively, threshold voltage decreases to achieve high performance, resulting in the exponential increase in leakage current. Due to the continued scaling of the technology and supply/threshold voltage, leakage power has become a dominant portion in the power dissipation of nanoscale VLSI *

Corresponding author.

T.-h. Kim et al. (Eds.): FGCN 2011, Part II, CCIS 266, pp. 437–446, 2011. © Springer-Verlag Berlin Heidelberg 2011

438

J.-T. Ryu and K.K. Kim

systems. An analysis of trends based on the International Technology Roadmap for Semiconductors shows that the power lost to leakage is beginning to exceed the power spent on useful computation. Therefore, the leakage power is a very serious problem in portable electronic systems that operate mostly in standby mode. The impact of standby power is increasing steadily as process dimensions shrink. In nanometer MOSFET circuits, the main components of standby power are subthreshold, gate tunneling, and reverse-biased junction band-to-band-tunneling (BTBT) leakage current. The reduction of transistor geometries necessitates the reduction of the supply voltage to avoid an electrical break down and to obtain the required performance. However, to retain or improve performance, it is necessary to reduce the threshold voltage (Vth) as well, resulting in the exponential increase of sub-threshold leakage [6][7]. To control the short-channel effects and increase the transistor driving strength in deep sub-micron (DSM) circuits, gate-oxide thickness also becomes thinner as technology scales down. The aggressive scaling in the gate-oxide results in currents tunneling through the oxide, which is a strong exponential function of the oxide thickness and the voltage magnitude across the oxide. In scaled devices, the higher substrate doping density and the application of the "halo" profiles cause significantly large reverse biased junction band-to-bandtunneling (BTBT) leakage current because of drain-substrate and source-substrate junctions [6][7]. In order to minimize the leakage power dissipation, several circuit techniques have been proposed, such as multi-threshold voltage CMOS (MTCMOS) [8] and variable threshold voltage CMOS (VTCMOS) [9] using variable substrate bias voltage. To reduce the leakage power by increasing the threshold voltage of MOSFET transistors during standby mode, adaptive reverse body-biasing (ABB) technique has been proposed [10]. The ABB decreases the sub-threshold leakage current of the scaled MOSFET. However, it increases the depletion width of the MOSFET parasitic junction diode and rapidly increases the BTBT current between the substrate and source/drain, especially in halo implants. Recently, methods using forward body biasing in active mode have been introduced [11]. Forward biasing increases the dynamic range of the device threshold and improves the circuit performance by decreasing threshold voltage. In addition, a power-performance trade off methodology for microprocessors has been proposed [12]. Another research has shown that simultaneous supply voltage scaling and bidirectional body biasing (forward + reverse biasing) is more effective in achieving high performance in active mode and low-power dissipation in standby mode. Therefore, the optimal voltage scaling and bidirectional body biasing determine the optimal tradeoff between supply voltage and body-bias voltage [13]. However, these techniques require significant circuit modification and performance overhead for leakage reduction, and they have not been complete or robust enough to apply to VLSI systems since the process, voltage, and temperature (PVT) variations are not considered for leakage power, especially in nanometer technology.

Adaptive Power Management for Nanoscale SoC Design

439

Fig. 1. Graphical representation showing reduction of total active and standby leakage power

This paper proposes a novel power management system to achieve low power in standby mode by exploiting the supply-voltage scaling and body-bias-voltage scaling taking PVT variations into consideration.

2

Proposed Power Management System

As described in the previous section, leakage currents will become a large component of total power dissipation as technology scales down. Although total power dissipation (dynamic + leakage) during the active mode is reduced with scaling, further power gains can be achieved if leakage currents are controlled wherever possible because these currents will make up a larger percentage of overall power dissipation in future technologies. Furthermore, the overall idle power dissipation tends to increase during the idle mode or standby mode where no computation is taking place due to the large leakage currents as shown in Fig. 1 [14]. Therefore, power reduction in standby mode has to be aggressively controlled so that the aggregate power consumption for the circuit in standby mode can be minimized. In order to reduce the leakage power component during the standby period, this paper proposes a new power management system in standby mode as shown in Fig. 2. The proposed power management system is composed of the main control unit and several sub-control units. The main control unit monitors each functional unit and sends sleep and active signals to each sub-control unit. Also, the main unit controls the entire workload and clock synchronization. Each sub-control unit is contained in a functional unit (FU) and monitors the idle period and the PVT variation of the FU. The sub-control unit consists of an input pattern generator for minimum leakage power, a data retention latch, a body voltage biasing circuit, PVT monitoring circuits, and a VDD scaling circuit.

440

J.-T. Ryu and K.K. Kim

(a)

(b) Fig. 2. Block diagram of the proposed power management system: (a) Main control unit, (b) Sub-control unit

3

Protocol Design

To explain the proposed scheme, we define the following messages.  

SLEEP_IN : broadcast message which contains SLEEP_TABLE. SLEEP_FIND : request message of SLEEP_TABLE.

Before one terminal of the active mode enters the sleep mode, it broadcasts SLEEP _IN message. SLEEP FIND message is used when the terminal of the sleep mode wants to transmit the packet. Here, SLEEP_TABLE contains the sleep mode ID, the time to sleep, and sleep duration as shown in Fig. 3 (a). All the terminals should manage the SLEEP_TABLE independently. Moreover, all terminals have three modes as follows:  Active_mode : transmits or receives data packet.  Listen_mode: only listens through receiver and does not any transmit packet through transceiver.  Sleep_mode : neither transmits nor receives packet.

Adaptive Power Management for Nanoscale SoC Design

441

The transition among modes is shown in Fig. 3 (b)(c). The idea behind the power management of each sub-control unit is described with five operating regions depicted in the timing diagram and the state machine shown in Fig. 3. In the time diagram, the sleep signal is generated for clock gating at the first breakeven (Tsleep1). Throughout sleep #1 to #3 region, the clock gating state is stayed to disables the clock signal and save power consumption. The five operation regions are as follows: (1)

Active: In this state, the FU is being used, and the sub-control unit counts the idle time of the FU by monitoring the primary inputs/outputs of the FU.

(2)

Sleep#1: This state is reached if the FU stays in the idle state longer than the first breakeven point (Tsleep1). In this state, the sub-control unit starts to scale down the supply voltage of the FU primary input changes of the FU. The sub-control unit goes back to the active state if the primary inputs of the FU have new transitions or if the main control unit sends an active signal.

(3)

Sleep#2: This state is reached if the FU stays in the idle state of the sleep #1 longer than the second breakeven point (Tsleep2). In this state, the optimal body biasing is applied to the FU with the scaled supply voltage to reduce more leakage power consumption. The other operations are the same as the sleep #1 state.

(4)

Sleep#3: This state is reached if the FU stays in the idle state of the sleep #2 longer than the third breakeven point (Tsleep3). In this state, the lowleakage input pattern is used in the FU with the scaled supply voltage and the optimal body voltage to reduce much more leakage power consumption than previous states. Then, all the internal node states of the FU are changed based on the input pattern, unlike the sleep #1 and #2 states. However, the previous node states are held through data retention circuits. The sub-control unit goes back to the active state if the primary inputs of the FU have new transitions or if the main control unit sends an active signal with an enable command. The FU restores the previous node states after one clock period.

(5)

Wait: This state is reached if the main control unit sends an active signal without the enable command. In this state, the sub-control unit prevents the FU from sending its primary output signals and waits during a period requested from the main control unit. If the main control unit sends a reset signal to clear all the internal node states restored from the retention circuits, the sub-control unit sends a reset signal to each latch contained in the FU.

442

J.-T. Ryu and K.K. Kim

(a)

(b)

(c) Fig. 3. Power management in standby mode: (a) Sleep Table format, (b) Timing diagram (c), PMS management state machine

If one terminal neither transmits nor receives the packet during a certain period, it broadcasts the SLEEP_IN message and enters the sleep mode. If the terminals in the active mode receive the SLEEP_IN, they keep or update SLEEP_TABLE. The procedure that enters the sleep mode is shown in Fig. 4 (a). As shown in Fig. 4 (b), when the terminals in the sleep mode want to transmit the packet, they first enter the active mode and broadcast SLEEP_FIND message. The terminal in the active mode which receives SLEEP_FIND transmits SLEEP_TABLE including ACK signal after a random delay to avoid conflict. Then, after the source

Adaptive Power Management for Nanoscale SoC Design

443

terminal sees the information of destination in the SLEEP_TABLE, it determines whether it transmits right now or not. In case the destination terminal is in the sleep mode, the source terminal waits until the destination terminal in the sleep mode wakes. However, if the destination terminal is in the active mode (destination ID is not included in SLEEP_TABLE), the source terminal does not need to wait any longer and transmits the packet immediately. When the terminals in the active mode want to communicate with the terminal in the sleep mode to transmit the packet, they monitor the time to start to sleep and the sleep duration in the SLEEP_TABLE and wait until the terminal in the sleep mode wakes as shown in Fig. 4 (a). If the destination is active mode (destination ID is not included in SLEEP_TABLE), the packet is transmitted immediately. The terminals in the sleep mode periodically wake and enter the listen mode to check whether they need to receive the packet or not. If they do not need to receive any packet, they enter the sleep mode again, but if the receiving packet exists, they go to the active mode and receive the packet. There is a special case when all terminals except one terminal are in the sleep mode. In this case, the last active terminal does not enter the sleep mode although it neither transmits nor receives the packet during a certain time because at least one terminal should manage the SLEEP_TABLE. If all terminals do not have SLEEP _TABLE, they do not know when other terminals in the sleep mode wake and which terminal is in the sleep mode. Therefore, the last active terminal which wants the sleep mode waits until another terminal enters the listen mode or the active mode. If the terminal in the sleep mode enters the listen mode, the last active terminal sends SLEEP_IN to it and enters the sleep mode itself. The SLEEP TABLE is distributed to reduce the power consumption of a specific terminal to manage SLEEP_TABLE. If only one terminal manages SLEEP_TABLE, it should continually wake. Then, this terminal cannot use LPM and its power is soon exhausted.

4

Experimental Results

The proposed optimal control system using 32-nm MOSFET technology has been implemented and evaluated using ISCAS’85 benchmark circuits designed in the same technology. The number of transistors of the sub-controller is 851, and its power dissipation is 141 W. Table 1 shows the summary of the results for the proposed approach at 50oC and a typical corner. The average leakage power has been measured using random input test vectors at 0.9V supply voltage. As shown in Table 1, the new technique for the minimal standby power provides average 1000 times reduction in leakage power compared to the simulation results of benchmark circuits without any optimization techniques. In order to show the effects of temperature and process variations, three temperature conditions (-25, 50, 125±oC) and three process-corner conditions (slow, typical, fast) are considered in the ISCAS benchmark circuit simulation. The optimal VDD/VBody control technique gives at least 500 times reduction in leakage power dissipation compared to the simulation results of benchmark circuits

444

J.-T. Ryu and K.K. Kim

without any optimization techniques being used. Moreover, our simulation results presents that when the proposed technique is applied, the leakage power dissipation is far less sensitive to the temperature and process variations because the optimal supply voltage and body bias voltage are changed according to the temperature and process. All the simulation results demonstrate that the proposed system is very effective in reducing the standby power in the big circuits.

(a)

(b) Fig. 4. Flow chart of an active and a sleep mode operation: (a) An active mode operation of proposed system, (b) A sleep mode operation of proposed scheme

Adaptive Power Management for Nanoscale SoC Design

5

445

Conclusion

As technology scales down below 90 nm, leakage currents have become a critical issue. In the past, circuit techniques and architectures ignored the effects of these currents because they were insignificant compared to the switching currents and threshold voltages were high enough. However, in modern technologies, the role of the leakage currents cannot be ignored and becomes increasingly significant issue with further scaling. Therefore, new circuit techniques and design considerations must be developed to control leakage currents in standby mode in order to provide lowpower solutions. In order to reduce the standby power, this paper proposed a novel asynchronous power management system to reduce the leakage power dissipation during standby mode. The proposed system consists of main control units, a bus interface, and subcontrol units. To reduce the power consumption of each terminal, each terminal is allowed to enter the low power mode using Sleep_Table and has five operation regions. Also, the power management system includes a novel control system that uses an adaptive method to find the optimal scaling algorithm during standby mode. Based on the temperature and process conditions, the optimal supply voltages is generated to reduce the leakage power, and body-bias voltage is automatically adjusted continuously by the control loop to adapt to the PVT variations. The results show that the proposed control system is a viable solution for high energy reduction in nanoscale CMOS circuits. Table 1. Experimental results for the standby power

Circuit

# of gates

With optimization (Typical, T=85ºC)

Average leakage power measured in Hspice ( VDD=0.9V, T=85ºC)

C432

160

8.56 nW

10.14 μW

C499

202

29.48 nW

48.90 μW

C880

383

18.53 nW

23.92 μW

C1355

546

23.11 nW

31.63 μW

C1908

880

46.36 nW

69.86 μW

C2670

1193

74.34 nW

124.45 μW

C5315

2307

145.10 nW

180.08 μW

C6288

2388

97.36 nW

110.64 μW

Acknowledgments. This work was supported by IC Design Education Center (IDEC) – CAD Tools (Softwares) only.

446

J.-T. Ryu and K.K. Kim

References 1. Meijer, M., Pineda de Gyvez, J., Otten, R.: On-chip digital power supply control for system-on-chip applications. In: IEEE ISLPED, pp. 311–314 (August 2005) 2. Nakai, M., Akui, S., Seno, K., Meguro, T., Seki, T., Kondo, T., Hashiguchi, A., Kawahara, H., Kumano, K., Shimura, M.: Dynamic voltage and frequency management for a lowpower embedded microprocessor. IEEE J. Solid-State Circuits 40(1), 28–35 (2005) 3. Wang, W., Mishra, P.: System-Wide Leakage-Aware Energy Minimization Using Dynamic Voltage Scaling and Cache Reconfiguration in Multitasking Systems. IEEE Transactions on Very Large Scale Integration (VLSI) Systems 99, 1–9 (2011) 4. Nourivand, A., Al-Khalili, A.J., Savaria, Y.: Postsilicon Tuning of Standby Supply Voltage in SRAMs to Reduce Yield Losses Due to Parametric Data-Retention Failures. IEEE Transactions on Very Large Scale Integration (VLSI) Systems 99, 1–13 (2011) 5. Pavan, T.K., Jagannadha Naidu, K., Shekar Babu, M.: Implementation of delay and power monitoring schemes to reduce the power consumption. In: IEEE International Conference on Signal Processing, Communication, Computing and Networking Technologies (ICSCCN), pp. 459–464 (July 2011) 6. Elgharbawy, W.M., Bayoumi, M.A.: Leakage sources and possible solutions in nanometer CMOS technologies. IEEE Circuits Syst. Magazine 5(4), 6–17 (2005) 7. Agarwal, A., Mukhopadhyay, S., Raychowdhury, A., Roy, K., Kim, C.H.: Leakage power analysis and reduction for nanoscale circuits. IEEE Micro. 26(2), 68–80 (2006) 8. Anis, M., Elmasry, M.: Multi-Threshold CMOS Digital Circuits: Managing Leakage Power. Kluwer, Norwell (2003) 9. Inukai, T., Hiramot, T., Sakurai, T.: Variable threshold voltage CMOS (VTCMOS) is series connected circuits. In: Int. Symp. Low Power Electron. Des. (ISLPED), pp. 201–206 (2001) 10. Roy, K., Mukhopadhyay, S., Mahmoodi-Meimand, H.: Leakage current mechanisms and leakage reduction techniques in deep-submicrometer CMOS circuits. Proc. IEEE 91(2), 305–327 (2003) 11. Hokazono, A., Balasubramanian, S., Ishimaru, K., Ishiuchi, H., Liu, T.K., Hu, C.: MOSFET design for forward body biasing scheme. IEEE Electron Device Lett. 27(5), 387–389 (2006) 12. Kulkarni, M., Sheth, K., Agrawal, V.D.: Architectural power management for high leakage technologies. In: IEEE Southeastern Symposium on System Theory (SSST), pp. 67–72 (March 2011) 13. Ishibashi, K., Fujimoto, T., Yamashita, T., Okada, H., Arima, Y., Hashimoto, Y., Sakata, K., Minematsu, I., Itoh, Y., Toda, H., Ichihashi, M., Komatsu, Y., Hagiwara, M., Tsukada, T.: Low-voltage and low-power logic, memory, and analog circuit techniques for SoCs using 90 nm technology and beyond. IEICE Trans. Electon E89-C(3), 250–262 (2006) 14. Anis, M., Elmasry, M.: Multi-threhsold CMOS digital circuits. Kluwer Academic Publishers (2003)

Author Index

Abdullah, Jiwa II-301 Adorna, Henry N. II-208 Agustin, Oliver C. I-244 Ahmed, Sabbir I-67 Alisherov, Farkhod II-20

Gerardo, Bobby D. II-220, II-229, II-239 Gil, Joon-Min II-354, II-381, II-403

Bae, Ihn-Han II-364, II-371 Bae, Kyeong-ryeol I-136, I-147 Baek, Yeong-Jin I-277 Baguda, Yakubu S. I-188 Bojkovic, Zoran S. I-198 Byun, Tae-Young II-320 Byun, Yung-Cheol II-220, II-229, II-239 Cabioc, Mark Dominic II-229 Chen, Bo-Han I-300 Chen, Wei-Sheng I-300 Cheong, Seung-Kook II-330 Chiang, Meng-Shu I-283 Chien, Shih-Pei I-236 Chimeh, Jahangir Dadkhah I-59 Cho, Jin Haeng II-292 Cho, Moon-Taek II-43 Cho, Seongsoo I-15, II-57 Cho, Woong II-26 Choi, Jae-Hoon II-274 Choi, Sang-Min II-248 Choi, Seong Gon I-111 Choi, Seung Ho II-124, II-132, II-154 Choi, Yeonyi I-93 Choi, Young B. I-310 Chouta, Rajath I-310 Chowdhury, Nawshad U.A. I-103 Chun, Chan Jun II-114, II-124 Chung, Kwang Sik II-412 Darwish, Ashraf I-209 De Castro, Joel T. II-220 Eid, Heba F. I-209 Eun, Ae-cheoun II-179 Farooq, Muhammad Omer Fisal, Norsheila I-188

I-1

Ha, Young-guk II-179 Han, Kijun II-338, II-346 Han, Sunyoung II-179 Hassanien, Aboul Ella I-209 Heo, Seok-Yeol II-393 Hong, Bong-Hwa II-34, II-57, II-65, II-96 Hong, Bonghwa I-15 Hong, Seong-Sik II-83 Hong, Suck-Joo II-43 Hong, Won-Kee II-419 Hoq, Md.T. I-103 Hsu, Tz-Heng I-283 Huh, Eui-Nam II-1 Hur, Kyung Woo II-283 Jang, Jae Hyuck II-283 Jang, Sei-Jin II-114, II-124 Jang, Seok-Woo I-120 Jeong, Hwa-Young II-65, II-96 Jiang, Jing-Jing II-162 Jin, Ik Soo I-261 Jo, Sung Dong II-114 Joe, Inwhee I-93 Joo, Hae-Jong II-73 Joo, Kil Hong I-293 Jung, Ho Min II-292 Jung, Hyo-Young I-219 Kang, Bong-Soo I-53 Kang, Cheoul-Shin I-179 Kang, Chul-Ung II-199 Kang, Chul Uoong II-189 Kang, Jang-Mook II-34 Kang, Jin Ah II-132 Kang, Sung Woon II-292 Kawai, Makoto I-67 Kawser, Mohammad T. I-103 Khan, Jahangir I-198 Khanam, Solima I-120 Kim, Byung Ki II-283 Kim, Chang-Geol II-429

448

Author Index

Kim, Do-Hoon II-11 Kim, Dongik I-93 Kim, Dongkyun II-258 Kim, Eun-Kyoung I-129 Kim, Hae Geun II-364, II-371 Kim, Haeng-Kon I-166 Kim, Heemin II-179 Kim, Hong Kook II-104, II-114, II-124, II-132, II-143 Kim, Hye-Jin I-47, I-53 Kim, Hyun Jong I-111 Kim, Jeong-Sam II-320 Kim, Jin-Mook II-83 Kim, Junhyung II-346 Kim, Kyung Ki II-437 Kim, Mihye II-354, II-381 Kim, Myung-Ju II-412 Kim, Sang-Soo II-73 Kim, Seon Man II-104 Kim, Seung-Hae II-403 Kim, Sung-Gyu II-20 Kim, Tai-hoon I-209 Kim, Young-Choon II-43 Kim, Yun-Hyuk I-53 Ko, Daesik II-268 Ko, Seok-Jun II-199 Ko, Young Woong II-283, II-292 Kunz, Thomas I-1 Kwak, Ho-Young I-53 Kwak, Soo-Won II-429 Kwon, Dong Rak II-312

Lee, Seok-Pil II-114, II-124 Lee, Seongjun I-53 Lee, Sung Joo II-104 Lee, Un-Bai II-412 Lee, Wan-Jik II-393 Lee, Won-Hyek II-403 Lee, Won-Yeoul II-393 Lee, Yong-Hwan I-129, I-136, I-147 Lee, Young Han II-143 Lee, Young-Hun I-179, II-330 Lee, Youngkon I-23, I-31, I-39 Lee, Young-Wook II-90 Lee, Yun Keun II-104 Li, Yi-Da I-283 Lim, Jong Hwan II-189 Lim, Kyungshik II-258 Lin, Chu-Hsing I-77, I-82, I-87, I-236 Lin, Hung-Yan I-82 Liu, Jung-Chun I-77, I-82 Lu, Shu-Yuan I-236

La, Keuk-Hwan I-15, II-57 Lai, Shin-Pin I-87 Lee, Byunghwa II-338, II-346 Lee, Chen-Yu I-87, I-236 Lee, Chien-Hsing I-77 Lee, Chung-Sub I-219 Lee, Euy-Soo II-73 Lee, Gi Min II-189 Lee, Ho-Cheol II-312 Lee, Hyuek Jae I-269 Lee, Jae-Dong I-129 Lee, Jae-Won II-154 Lee, Jong-Heon I-53 Lee, Jongsup I-15, II-57 Lee, Jung Geun II-292 Lee, Junghoon I-47, I-53 Lee, Jun-Hyung II-1 Lee, Sang-Hoon I-277

Oh, Byung-Joo I-244 Oh, Hyun Seo II-26 Oh, Sang Yoon I-136 Oh, Tae Hwan I-310 Ok, Seung-Ho I-136, I-147 Osorio, Francisca D. II-220

Malinao, Jasmine A. II-208 Maravilla Jr., Reynaldo G. II-208 Marwat, Muhammad Imran Khan I-198 Matsuo, Tokuro II-169 Moon, Byung-Hyun II-248 Moon, Byungin I-129, I-136, I-147 Moon, Inseok II-419 Na, Sang-Ho II-1 Nguyen, Tien-Dung II-1 Noh, Min-Ki II-403

Paik, Woojin I-120 Pangapalan, Ana Rhea II-220 Parapari, Saeed Bashirzadeh I-59 Park, Byungjoo II-20 Park, Byung-Seok I-179 Park, Gyung-Leen I-47, I-53 Park, Hwase II-268 Park, Kun Hyun II-189 Park, Nam Hun I-293 Park, Nam In II-143 Pun, Chi-Man II-162

Author Index Rashid, Rozeha A. I-188 Ryu, Heung-Gyoon II-11, II-274 Ryu, Jeong-Tak II-437 Seo, Yong-Ho I-219 Shrestha, Bhanu I-15, II-57 Shuaibu, Dahiru S. I-188 Son, Hyeon-Sik I-136, I-147 Song, Biao II-1 Song, Byung-Seop II-429 Song, Ho-Bin II-43 Song, Hyun-Ju II-330 Song, ShiLi I-227 Song, Ui-Sung II-354 Su, Wei Wei I-227 Surendran, Purushothaman II-199 Syed, Sharifah H. I-188 Syfullah, Md.K. I-103 Tabanda, Elise A. II-208 Tak, Ryu Juang II-429

Takahashi, Satoshi II-169 Tang, Wei II-1 Tanguilig III, Bartolome II-239 Tsai, Sheng-Hsing I-82 Wahid, Abdul II-258 Wang, Yi Fan I-227, I-254 Wen, Hong I-227, I-254 Wu, Tang-Wei I-77 Yang, Chao-Tung I-300 Yang, Ling I-227, I-254 Yang, Tae-Kyu I-219 Yang, Tzu-Chien I-87 Yeom, Kiwon I-156 Yoo, Kwan-Hee II-381 Yun, Jangkyu II-338, II-346 Yusof, Sharifah K. I-188 Zhang, Gao Yuan I-254 Zhou, Liang I-254

449