Lecture Notes in Electrical Engineering 315

Hamzah Asyrani Sulaiman
Mohd Azlishah Othman
Mohd Fairuz Iskandar Othman
Yahaya Abd Rahim
Naim Che Pee
Editors

Advanced Computer
and Communication
Engineering
Technology
Proceedings of the 1st International
Conference on Communication and
Computer Engineering
Lecture Notes in Electrical Engineering

Volume 315

Board of Series editors

Leopoldo Angrisani, Napoli, Italy
Marco Arteaga, Coyoacán, México
Samarjit Chakraborty, München, Germany
Jiming Chen, Hangzhou, P.R. China
Tan Kay Chen, Singapore, Singapore
Rüdiger Dillmann, Karlsruhe, Germany
Haibin Duan, Beijing, China
Gianluigi Ferrari, Parma, Italy
Manuel Ferre, Madrid, Spain
Sandra Hirche, München, Germany
Faryar Jabbari, Irvine, USA
Janusz Kacprzyk, Warsaw, Poland
Alaa Khamis, New Cairo City, Egypt
Torsten Kroeger, Stanford, USA
Tan Cher Ming, Singapore, Singapore
Wolfgang Minker, Ulm, Germany
Pradeep Misra, Dayton, USA
Sebastian Möller, Berlin, Germany
Subhas Mukhopadyay, Palmerston, New Zealand
Cun-Zheng Ning, Tempe, USA
Toyoaki Nishida, Sakyo-ku, Japan
Federica Pascucci, Roma, Italy
Tariq Samad, Minneapolis, USA
Gan Woon Seng, Nanyang Avenue, Singapore
Germano Veiga, Porto, Portugal
Haitao Wu, Beijing, China
Junjie James Zhang, Charlotte, USA
About this Series

‘‘Lecture Notes in Electrical Engineering (LNEE)’’ is a book series which reports

the latest research and developments in Electrical Engineering, namely:

• Communication, Networks, and Information Theory

• Computer Engineering
• Signal, Image, Speech and Information Processing
• Circuits and Systems
• Bioengineering
LNEE publishes authored monographs and contributed volumes which present
cutting edge research information as well as new perspectives on classical fields,
while maintaining Springer’s high standards of academic excellence. Also con-
sidered for publication are lecture materials, proceedings, and other related
materials of exceptionally high quality and interest. The subject matter should be
original and timely, reporting the latest research and developments in all areas of
electrical engineering.
The audience for the books in LNEE consists of advanced level students,
researchers, and industry professionals working at the forefront of their fields.
Much like Springer’s other Lecture Notes series, LNEE will be distributed through
Springer’s print and electronic publishing channels.

More information about this series at http://www.springer.com/series/7818

Hamzah Asyrani Sulaiman
Mohd Azlishah Othman •

Mohd Fairuz Iskandar Othman

Yahaya Abd Rahim Naim Che Pee
•

Editors

Advanced Computer
and Communication
Engineering Technology
Proceedings of the 1st International
Conference on Communication
and Computer Engineering

123
Editors
Hamzah Asyrani Sulaiman
Mohd Azlishah Othman
Mohd Fairuz Iskandar Othman
Yahaya Abd Rahim
Naim Che Pee
Universiti Teknikal Malaysia Melaka
Melaka
Malaysia

ISSN 1876-1100 ISSN 1876-1119 (electronic)

ISBN 978-3-319-07673-7 ISBN 978-3-319-07674-4 (eBook)
DOI 10.1007/978-3-319-07674-4

Library of Congress Control Number: 2014947654

Springer Cham Heidelberg New York Dordrecht London

 Springer International Publishing Switzerland 2015

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or
information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed. Exempted from this legal reservation are brief
excerpts in connection with reviews or scholarly analysis or material supplied specifically for the
purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the
work. Duplication of this publication or parts thereof is permitted only under the provisions of
the Copyright Law of the Publisher’s location, in its current version, and permission for use must
always be obtained from Springer. Permissions for use may be obtained through RightsLink at the
Copyright Clearance Center. Violations are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, 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.
While the advice and information in this book are believed to be true and accurate at the date of
publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for
any errors or omissions that may be made. The publisher makes no warranty, express or implied, with
respect to the material contained herein.

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

Preface

The first International Conference on Communication and Computer Engineering

(ICOCOE 2014) was held in the historical city of Malacca, Malaysia, between
May 20 and 21, 2014. The conference focuses on industrial and manufacturing
theory and applications of electronics, communications, computing, and infor-
mation technology. The objectives of the conference were to provide an avenue for
researchers to present high-quality research and be involved in professional
interactions for the advancement of science, technology, and fellowship. More
than 300 participants from 15 countries registered for the conference. ICOCOE
2014 provided a platform for them to present their latest invention and advanced
technology and research, share their expertise, establish new contacts, and discuss
a range of communication and computer engineering topics for the betterment and
future advancement of these fields.
ICOCOE 2014 is the first ever event handled by Prime Systems, an educational
consulting sector provider. It was successful in getting overwhelming response
from public and private sectors, local and foreign universities, research institu-
tions, stakeholders, and various industries from all over the region. Nearly 200
papers were received with only around 102 papers accepted. Associate Professor
Dr. Muhammad Ramlee from Universiti Teknologi Malaysia gave a keynote
address on RF and microwave communication system in advanced systems while
Prof. Eryk Dutkiewicz from Macquarie University presented his latest work on
wireless communications. Several technical sessions were arranged for the
conference.
This book contains a selection of revised manuscripts presented at ICOCOE
2014 within the two main conference themes, which are communications and
computer engineering research areas. This book covers various aspects of
advanced computer and communication engineering, specifically on industrial and
manufacturing theory and applications of electronics, communications, computing,
and information technology. The development of latest technologies has been

v
vi Preface

highlighted, aimed at various computer and communication-related professionals

such as telecommunication engineers, computer engineers and scientists,
researchers, academicians, and students. Furthermore, several applications
involving cutting edge communication and computer systems are also discussed.
The editors wish to acknowledge Prime Systems for successfully organizing
and sponsoring this conference.

Hamzah Asyrani Sulaiman

Mohd Azlishah Othman
Mohd Fairuz Iskandar Othman
Yahaya Abd Rahim
Naim Che Pee
General Committee and Advisors

General Committee and Advisor/Editorial Board

Mohd Azlishah Bin Othman Universiti Teknikal Malaysia Melaka,

Malaysia
Ahmad Naim Bin Che Pee Universiti Teknikal Malaysia Melaka,
Malaysia
Hamzah Asyrani Bin Sulaiman Universiti Teknikal Malaysia Melaka,
Malaysia
Yahaya Bin Abdul Rahim Universiti Teknikal Malaysia Melaka,
Malaysia
Mohd Fairuz Iskandar Bin Othman Universiti Teknikal Malaysia Melaka,
Malaysia

International Advisor/Technical Program Committee

Prof. Hj. Kamaruzaman Jusoff

Dr. Abdel Rahman Mohammad Said Al-Tawaha
Prof. Dr. R.J. Godwin
Prof. Dr. Erik Valdemar Cuevas Jimenez
Prof. Dr. Arun Kumar Gupta
Prof. Robert Morelos-Zaragoza

vii
viii General Committee and Advisors

Running Committee

Chairman
Alishamuddin Enjom

Co-Chairman
Shahrizan Jamaludin

Secretary
Nurulhuda Abu Talib

Treasurer
Zamani Marjom

Technical Program Chair

Siti Zawiyah Iskandar
Contents

Part I Communication

1 RWA: Comparison of Genetic Algorithms and Simulated

Annealing in Dynamic Traffic . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Arturo Rodriguez, Antonio Gutierrez, Luis Rivera
and Leonardo Ramirez

2 Route Optimization in Proxy Mobile IPv6 Test-Bed via

RSSI APPs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Nur Haliza Binti Abdul Wahab, L.A. Latif, S.H.S. Ariffin,
N. Fisal and N. Effiyana Ghazali

3 Polytetrafluoroethylene Glass Microfiber Reinforced

Slotted Patch Antenna for Satellite Band Applications . . . . . . . . 29
M. Samsuzzaman, T. Islam and M.R.I. Faruque

4 2.4 GHz Circularly Polarized Microstrip Antenna for

RFID Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Rudy Yuwono and Ronanobelta Syakura

5 On Understanding Centrality in Directed Citation Graph . . . . . 43

Ismael A. Jannoud and Mohammad Z. Masoud

6 Channel Capacity of Indoor MIMO Systems in the

Presence of Spatial Diversity . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
M. Senon, M.N. Husain, A.R. Othman, M.Z.A. Aziz,
K.A.A. Rashid, M.M. Saad, M.T. Ahmad and J.S. Hamidon

ix
x Contents

7 Design of Multi-band Antenna for Wireless MIMO

Communication Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
M.M. Saad, M.N. Husain, M.Z.A. Aziz, A.R. Othman,
K.A.A. Rashid and M. Senon

8 Design of Linear Polarization Antenna for Wireless

MIMO Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
K.A.A. Rashid, M.N. Husain, A.R. Othman, M.Z.A. Aziz,
M.M. Saad, M. Senon, M.T. Ahmad and J.S. Hamidon

9 The Effect of IV Characteristics on Optical Control of

SDR Si IMPATT Diode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
T.S.M. Arshad, M.A. Othman, M.N. Hussain and Y.A. Rahim

10 Variable Intrinsic Region in CMOS PIN Photodiode for

I–V Characteristic Analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
M.A. Othman, N.Y.M. Yasin, T.S.M. Arshad, Z.A.F.M. Napiah,
M.M. Ismail, H.A. Sulaiman, M.H. Misran, M.A. Meor Said
and R.A. Ramlee

11 Variable Depletion Region in CMOS PN Photodiode for

I–V Characteristic Analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . 103
M.A. Othman, T.S.M. Arshad, Z.A.F.M. Napiah, M.M. Ismail,
N.Y.M. Yasin, H.A. Sulaiman, M.H. Misran, M.A. Meor Said
and R.A. Ramlee

12 U-Slot Rectangular Patch Antenna for Dual Band

Application. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
Mohammad Shawkat Habib, I.M. Rafiqul, Khaizuran Abdullah
and M. Jamil Jakpar

13 Analysis of Synthetic Storm Technique Based on Ku-Band

Satellite Beacon Measurements in Malaysia . . . . . . . . . . . . . . . 121
Ali K. Lwas, I.M. Rafiqul, Mohamed Hadi Habaebi,
Ahmad F. Ismail, Mandeep Singh, Jalel Chebil,
Al-Hareth Zyoud and Hassan Dao

14 The Evolution of Double Weight Codes Family in Spectral

Amplitude Coding OCDMA . . . . . . . . . . . . . . . . . . . . . . . . . . 129
N. Din Keraf, S.A. Aljunid, A.R. Arief and P. Ehkan
Contents xi

15 Performance Evaluation of LTE Scheduling Techniques

for Heterogeneous Traffic and Different Mobility
Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
Lukmanhakim Sukeran, Mohamed Hadi Habaebi,
Al-Hareth Zyoud, Musse Mohamud Ahmad, Shihab Hameed,
Amelia Wong and I.M. Rafiqul

16 Design for Energy-Aware IP Over WDM Networks with

Hibernation Mode and Group-Node Techniques . . . . . . . . . . . 151
M.N.M. Warip, Ivan Andonovic, Ivan Glesk,
R. Badlishah Ahmad, P. Ehkan,
Mohamed Elshaikh Elobaid Said Ahmed,
Shamsul Jamel Elias and Fazrul Faiz Zakaria

17 Content Based Image Retrieval Using Color Layout

Descriptor and Generic Fourier Descriptor . . . . . . . . . . . . . . . 163
Muhammad Imran, Rathiah Hashim and Noor Elaiza

18 Pilot Based Pre FFT Signal to Noise Ratio Estimation for

OFDM Systems in Rayleigh-Fading Channel . . . . . . . . . . . . . . 171
A.M. Khan, Varun Jeoti and M. Azman Zakariya

19 The Use of Convolutional Code for Narrowband

Interference Suppression in OFDM-DVBT System. . . . . . . . . . 183
Aizura Abdullah, Muhammad Sobrun Jamil Jamal,
Khaizuran Abdullah, Ahmad Fadzil Ismail and Ani Liza Asnawi

20 Two-Elements Crescent Shaped Printed Antenna for

Wireless Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
Wan Noor Najwa Wan Marzudi, Zuhairiah Zainal Abidin,
Ma Yue and Raed A. Abd-Alhameed

21 Wideband Linearly Polarized Printed Monopole Antenna

for C-Band . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
Touhidul Alam, Mohammad Rashed Iqbal Faruque
and Mohammad Tariqul Islam

22 A Novel Anti-collision Protocol for Optimization

of Remote Sensing in Dense Reader Network. . . . . . . . . . . . . . 213
Faiza Nawaz and Varun Jeoti
xii Contents

23 Double Square Loop Frequency Selective Surface (FSS)

for GSM Shielding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
Nur Khalida Binti Abdul Khalid
and Fauziahanim Binti Che Seman

24 Analysis of the Active Region of Archimedean Spiral

Antenna . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
Abdirahman Mohamoud Shire and Fauziahanim Che Seman

25 Optimization of BER Performance in the MIMO-OFDMA

System for Mobile WiMAX System Using Different
Equalization Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
Azlina Idris, Norhayati Abdullah, Nor Azlizan Hussein
and D.M. Ali

26 Performance Analysis of Polling Delay in Transparent and

Non-transparent Multi-hop Relay WiMAX Network . . . . . . . . 251
Mohd Daud A. Hassan, Habibah Hashim
and D.M. Ali

27 Blind Source Computer Device Identification from

Recorded Calls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
Mehdi Jahanirad, Ainuddin Wahid Abdul Wahab
and Nor Badrul Anuar

28 Visibility for Network Security Enhancement in Internet

Protocol Over Ethernet Networks . . . . . . . . . . . . . . . . . . . . . . 277
W.K. Alzubaidi, Longzheng Cai,
Shaymaa A. Alyawer and Erika Siebert-Cole

29 Comparative Analysis of Different Single Cell

Metamaterial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
Pankaj Rameshchandra Katiyar
and Wan Nor Liza Binti Wan Mahadi

30 Distributed Video Coding with Frame Estimation at

Decoder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
Kin Honn Chiam and Mohd Fadzli Mohd Salleh

31 Performance Analysis of an OCDMA System Based on

SPD Detection Utilizing Different Type of Optical Filters
for Access Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
Sarah G. Adbulqader, Hilal A. Fadhil, S.A. Aljunid
and Anuar Mat Safar
Contents xiii

32 Deployment of Optimized Algorithm for MPEG-4 Data

Over Wireless Multimedia Sensor Network . . . . . . . . . . . . . . . 321
Norlezah Hashim, Sharifah Hafizah Syed Ariffin,
Farizah Yunus, Fakrulradzi Idris and Norsheila Fisal

33 Partially Compensated Power Control Technique

for LTE-A Macro-Femto Networks . . . . . . . . . . . . . . . . . . . . . 331
Sawsan Ali Saad, Mahamod Ismail and Rosdiadee Nordin

34 Design and Development of the Visible Light

Communication System. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
Anuar Musa, Mazlaini Yahya, Nazaruddin Omar,
Mohd Kamarulzamin Salleh and Noor Aisyah Mohd Akib

35 The Embroidered Antenna on Bending Performances for

UWB Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
M.S. Shakhirul, A. Sahadah, M. Jusoh, A.H. Ismail
and Hasliza A. Rahim

36 The Embroidered Wearable Antenna for UWB

Application. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
M.S. Shakhirul, A. Sahadah, M. Jusoh, A.H. Ismail,
C.M. Nor and F.S. Munirah

37 Bowtie Shaped Substrate Integrated Waveguide Bandpass

Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
Z. Baharudin, M.Z.U. Rehman, M.A. Zakariya,
M.H.M. Khir, M.T. Khan and J.J. Adz

38 Logical Topology Design with Low Power Consumption

and Reconfiguration Overhead in IP-over-WDM
Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375
Bingbing Li and Young-Chon Kim

Part II Computer

39 Systematic Analysis on Mobile Botnet Detection

Techniques Using Genetic Algorithm. . . . . . . . . . . . . . . . . . . . 389
M.Z.A. Rahman and Madihah Mohd Saudi
xiv Contents

40 An Empirical Study of the Evolution of PHP MVC

Framework. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
Rashidah F. Olanrewaju, Thouhedul Islam
and N. Ali

41 Evolutionary Approach of General System Theory

Applied on Web Applications Analysis . . . . . . . . . . . . . . . . . . 411
Aneta Bartuskova, Ondrej Krejcar and Kamil Kuca

42 A Novel Distributed Image Steganography Method Based

on Block-DCT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
Rosemary Koikara, Dip Jyoti Deka, Mitali Gogoi and Rig Das

43 An Improved History-Based Test Prioritization Technique

Using Code Coverage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437
Avinash Gupta, Nayneesh Mishra, Aprna Tripathi,
Manu Vardhan and Dharmender Singh Kushwaha

44 Local Pricewatch Information Solicitation and Sharing

Model Using Mobile Crowdsourcing . . . . . . . . . . . . . . . . . . . . 449
Hazleen Aris

45 Enhancement of Nurse Scheduling Steps Using Particle

Swarm Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 459
Norhayati Mohd Rasip, A.S.H. Basari,
Nuzulha Khilwani Ibrahim and Burairah Hussin

46 Hardware Implementation of MFCC-Based Feature

Extraction for Speaker Recognition . . . . . . . . . . . . . . . . . . . . . 471
P. Ehkan, F.F. Zakaria, M.N.M. Warip, Z. Sauli
and M. Elshaikh

47 Parallel ASIP Based Design of Turbo Decoder. . . . . . . . . . . . . 481

F.F. Zakaria, P. Ehkan, M.N.M. Warip
and M. Elshaikh

48 A Comparative Study of Web Application Testing and

Mobile Application Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . 491
Maryam Ahmed and Rosziati Ibrahim

49 Multi-objective Functions in Grid Scheduling . . . . . . . . . . . . . 501

Zafril Rizal M. Azmi, M.A. Ameedeen
and Imran Edzereiq Kamarudin
Contents xv

50 Experimental Analysis on Available Bandwidth

Estimation Tools for Wireless Mesh Network . . . . . . . . . . . . . 525
Imran Edzereiq Kamarudin, M.A. Ameedeen
and Zafril Rizal M. Azmi

51 A Survey of Petri Net Tools . . . . . . . . . . . . . . . . . . . . . . . . . . 537

Weng Jie Thong and M.A. Ameedeen

52 Towards a Exceptional Distributed Database Model

for Multi DBMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553
Mohammad Hasan Ali and Mohd Azlishah Othman

53 Semantic Search Engine Using Natural Language

Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 561
Sudhakar Pandiarajan, V.M. Yazhmozhi and P. Praveen kumar

54 Integration of Mobile Based Learning Model Through

Augmented Reality Book by Incorporating Students
Attention Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 573
Zarwina Yusoff, Halina Mohamed Dahlan
and Norris Syed Abdullah

55 Law Reckoner for Indian Judiciary: An Android

Application for Retrieving Law Information Using Data
Mining Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 585
S. Poonkuzhali, R. Kishore Kumar and Ciddarth Viswanathan

56 Enhancing the Efficiency of Software Reliability

by Detection and Elimination of Software Failures
Through Univariate Outlier Mining . . . . . . . . . . . . . . . . . . . . 595
S. Poonkuzhali, R. Kishore Kumar and R. Kumar

57 A Survey on the Application of Robotic Teacher

in Malaysia. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 605
Noraidah Blar and Fairul Azni Jafar

58 A Novel Method for Distributed Image Steganography . . . . . . 615

Bismita Choudhury, Rig Das and Themrichon Tuithung

59 An Efficient Beam Scanning Algorithm for Hidden Node

Collision Avoidance in Wireless Sensor Networks . . . . . . . . . . 627
Moorthy Sujatha and Raghuvel Subramaniam Bhuvaneswaran
xvi Contents

60 Evaluation of Stereo Matching Algorithms and Dynamic

Programming for 3D Triangulation . . . . . . . . . . . . . . . . . . . . . 641
Teo Chee Huat and N.A. Manap

61 Image Enhancement Filter Evaluation on Corrosion

Visual Inspection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 651
Syahril Anuar Idris and Fairul Azni Jafar

62 A Framework for Sharing Communication Media

in Supporting Creative Task in Collaborative Workspace . . . . 661
Norzilah Musa, Siti Z.Z. Abidin and Nasiroh Omar

63 Joint Torque Estimation Model of sEMG Signal for Arm

Rehabilitation Device Using Artificial Neural Network
Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 671
M.H. Jali, T.A. Izzuddin, Z.H. Bohari, H. Sarkawi,
M.F. Sulaima, M.F. Baharom and W.M. Bukhari

64 Enhancement of RSA Key Generation Using Identity. . . . . . . . 683

Norhidayah Muhammad, Jasni Mohamad Zain,
M.Y.M. Saman and Mohd Fadhil Ramle

65 Rules Mining Based on Clustering of Inbound Tourists

in Thailand. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 693
Wirot Yotsawat and Anongnart Srivihok

66 Designing a New Model for Worm Response Using

Security Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 707
Madihah Mohd Saudi and Bachok M. Taib

67 Neural Network Training Algorithm for Carbon Dioxide

Emissions Forecast: A Performance Comparison . . . . . . . . . . . 717
Herrini Mohd Pauzi and Lazim Abdullah

68 Theorem Prover Based Static Analyzer: Comparison

Analysis Between ESC/Java2 and KeY . . . . . . . . . . . . . . . . . . 727
Aneesa Saeed and S.H.A. Hamid

69 Designing a New Model for Trojan Horse Detection Using

Sequential Minimal Optimization . . . . . . . . . . . . . . . . . . . . . . 739
Madihah Mohd Saudi, Areej Mustafa Abuzaid,
Bachok M. Taib and Zul Hilmi Abdullah
Contents xvii

70 An Access Control Framework in an Ad Hoc Network

Infrastructure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 747
Tanya Koohpayeh Araghi, Mazdak Zamani,
A.A. Manaf and Sagheb Kohpayeh Araghi

71 Enhancement of Medical Image Compression by Using

Threshold Predicting Wavelet-Based Algorithm . . . . . . . . . . . . 755
N.S.A.M. Taujuddin and Rosziati Ibrahim

72 Detection and Revocation of Misbehaving Vehicles from

VANET . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 767
Atanu Mondal and Sulata Mitra

73 A Novel Steganalysis Method Based on Histogram

Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 779
Bismita Choudhury, Rig Das and Arup Baruah

74 Pattern Recognition Techniques: Studies on Appropriate

Classifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 791
Sasan Karamizadeh, Shahidan M. Abdullah, Mazdak Zamani
and Atabak Kherikhah

75 Environmental Noise Analysis for Robust Automatic

Speech Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 801
N. Sai Bala Kishore, M. Rao Venkata and M. Nagamani

76 Performance Comparison of Selected Classification

Algorithms Based on Fuzzy Soft Set for Medical Data . . . . . . . 813
Saima Anwar Lashari and Rosziati Ibrahim

77 A Hybrid Selection Method Based on HCELFS and SVM

for the Diagnosis of Oral Cancer Staging. . . . . . . . . . . . . . . . . 821
Fatihah Mohd, Zainab Abu Bakar, Noor Maizura Mohamad Noor,
Zainul Ahmad Rajion and Norkhafizah Saddki

78 A Linear Assignment Method of Simple Additive

Weighting System in Linear Programming Approach
Under Interval Type-2 Fuzzy Set Concepts for MCDM
Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 833
Nurnadiah Zamri and Lazim Abdullah
xviii Contents

79 Hybridization Denoising Method for Digital Image

in Low-Light Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 843
Suhaila Sari, Sharifah Zahidah Hasan Al Fakkri,
Hazli Roslan and Zarina Tukiran

80 The Improved Models of Internet Pricing Scheme of Multi

Service Multi Link Networks with Various Capacity Links . . . 851
Fitri Maya Puspita, Kamaruzzaman Seman and Bachok M. Taib

81 Improving the Models of Internet Charging in Single Link

Multiple Class QoS Networks . . . . . . . . . . . . . . . . . . . . . . . . . 863
Irmeilyana Saidi Ahmad, Indrawati, Fitri Maya Puspita
and Lisma Herdayana

82 A New Aggregating Phase for Interval Type-2 Fuzzy

TOPSIS Using the ELECTRE I Method . . . . . . . . . . . . . . . . . 873
Nurnadiah Zamri and Lazim Abdullah

83 The Role of Green IT and IT for Green Within Green

Supply Chain Management: A Preliminary Finding
from ISO14001 Companies in Malaysia . . . . . . . . . . . . . . . . . . 883
K.S. Savita, P.D.D. Dominic and Kalai Anand Ratnam

84 Integrating e-Learning with Radio Frequency

Identification (RFID) for Learning Disabilities:
A Preliminary Study. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 895
Wan Fatin Fatihah Yahya, Noor Maizura Mohamad Noor,
Mohd Pouzi Hamzah, Mohamad Nor Hassan,
Nur Fadila Akma Mamat and Mohd Arizal Shamsil Mat Rifin

85 Palmprint Identification Using Invariant Moments

Algorithm Based on Wavelet Transform . . . . . . . . . . . . . . . . . 905
Inass Shahadha Hussein and M.J. Nordin

86 Auto Mobile Ad Hoc Mechanism in Delay Tolerant

Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 915
Muhammad Affandy Azman, Sharifah Hafizah Syed Ariffin,
Norsheila Fisal, Mazlan Abbas, Mohd Husaini Mohd Fauzi
and Sharifah K. Syed-Yusof

87 An Exploratory Study on Blind Users’ Mental Model

in Computer Accessibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . 925
Manoranjitham Muniandy and Suziah Sulaiman
Contents xix

88 Case-Based Reasoning and Profiling System for Learning

Mathematics (CBR-PROMATH) . . . . . . . . . . . . . . . . . . . . . . . 939
Nur Azlina Mohamed Mokmin and Mona Masood

89 What Is the Influence of Users’ Characteristics

on Their Ability to Detect Phishing Emails? . . . . . . . . . . . . . . 949
Ibrahim Alseadoon, M.F.I. Othman and Taizan Chan

90 Adaptive and Dynamic Service Composition

for Cloud-Based Mobile Application . . . . . . . . . . . . . . . . . . . . 963
R. Kanesaraj Ramasamy, Fang-Fang Chua and Su-Cheng Haw

91 Web Service Composition Using Windows Workflow

for Cloud-Based Mobile Application . . . . . . . . . . . . . . . . . . . . 975
R. Kanesaraj Ramasamy, Fang-Fang Chua and Su-Cheng Haw

92 An Effective Image Retrieval Method Based on Fractal

Dimension Using Kernel Density Estimation . . . . . . . . . . . . . . 987
Zhang Qin, Huang Xiaoqing and Liu Wenbo

93 Bio Terapi Solat: 3D Integration in Solat Technique

for Therapeutic Means . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1001
Arifah Fasha Rosmani, Noor Azura Zainuddin,
Siti Zulaiha Ahmad and Siti Zubaida Ramli

94 Enhanced Interactive Mathematical Learning Courseware

Using Mental Arithmetic for Preschool Children . . . . . . . . . . . 1013
Siti Zulaiha Ahmad, Noor Asmaliyana Ahmad,
Arifah Fasha Rosmani, Umi Hanim Mazlan
and Mohammad Hafiz Ismail

95 Comparative Evaluation of Ensemble Learning

and Supervised Learning in Android Malwares
Using Network-Based Analysis . . . . . . . . . . . . . . . . . . . . . . . . 1025
Ali Feizollah, Nor Badrul Anuar, Rosli Salleh
and Fairuz Amalina

96 Tailored MFCCs for Sound Environment Classification

in Hearing Aids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1037
Roberto Gil-Pita, Beatriz López-Garrido and Manuel Rosa-Zurera

97 Metamodelling Architecture for Modelling Domains

with Different Mathematical Structure . . . . . . . . . . . . . . . . . . 1049
Vitaliy Mezhuyev
xx Contents

98 Use Case Based Approach to Analyze Software Change

Impact and Its Regression Test Effort Estimation . . . . . . . . . . 1057
Avinash Gupta, Aprna Tripathi and Dharmendra Singh Kuswaha

99 A Review of Image Segmentation Methodologies

in Medical Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1069
Lay Khoon Lee, Siau Chuin Liew and Weng Jie Thong

100 The Utilization of Template Matching Method for License

Plate Recognition: A Case Study in Malaysia. . . . . . . . . . . . . . 1081
Norazira A. Jalil, A.S.H. Basari, Sazilah Salam,
Nuzulha Khilwani Ibrahim and Mohd Adili Norasikin
 Part I
Communication
Chapter 1
RWA: Comparison of Genetic Algorithms
and Simulated Annealing in Dynamic
Traffic

Arturo Rodriguez, Antonio Gutierrez, Luis Rivera

and Leonardo Ramirez

Abstract Modern telecommunications are supporting every day a progressive

demand for services, which in turn generates greater requirements from the
attention capacity in photonic transport networks. This phenomenon forces us to
improve the routing systems, to minimize the blocking probability and minimize
the use of the network, among other indicators, in order to attend current demand
and to have the capacity to attend future demand. This paper compares four studies
on routing and wavelength assignment with the aim of supporting the improve-
ment of the already mentioned indicators. A comparison is made between opti-
mizing algorithms and heuristic simulated annealing and genetic algorithms, using
comparative indicators such as blocking probability and the use of the network.
The results show that the heuristic algorithms are potentially better for a high load
dynamic demand (greater than 120 erlangs) that would function much better under
stress. GINT proposes genetic algorithms as a solution to the coming future
demand of data transport.

Keywords Simulated annealing  Genetic algorithm  NSFNET  Wavelength

A. Rodriguez (&)  A. Gutierrez  L. Rivera

Department of Industrial Technology, Research Group New Technologies (GINT),
Universidad Santiago de Chile, Santiago, Chile
e-mail: [email protected]
A. Gutierrez
e-mail: [email protected]
L. Rivera
e-mail: [email protected]
L. Ramirez
Division of Technology Development and Innovation Research Group on Telemedicine
(TIGUM), Universidad Militar de Nueva Granada, Bogotá, Colombia
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 3

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_1
4 A. Rodriguez et al.

1.1 Introduction

Currently, in the field of fiber optics, the discussion is centered on commutation,

which is preferable because of its high response speed to routing. The closeness of
the upper layers of OSI to the linkage layer has triggered the reflection of the
permanence of the IP layer, but not of the IP address. In any case, the search to
establish routes that determine a fast and secure connection should be solved
through different algorithms found in the literature or innovative algorithms pro-
posed by researchers, such as those of Dijkstra and Floyd-Warshall among others,
however those optimizing algorithms are not efficient with respect to optical
demand, which requires good routes rather than optimum routes, due to the need to
decrease the probability of blocking the network and improving its use.
The traffic supported by a network can be classified as static and dynamic [1]. In
the optical transport networks the traffic was of the static type because the demand
was not sufficient to congest the system; but the static characteristics of traffic have
turned toward dynamic traffic, and the RWA (Routing Wavelength Assignment)
problem has become important, requiring the selection of a route and a wavelength
to establish the connection, under the restriction that there should be no change in
wavelength along the chosen route, which is known as CCW (Continuity Con-
straint Wavelength). This restriction helps to improve the delay on the route of
present demand [2], but the use of the network increases, increasing the probability
of blocking the future demand, because the more routes are established, the scarcer
the roads become.
In general, this problem has been solved by subdividing it into two parts: The
first one solves the route to be followed based on minimizing some pre-established
condition [3, 4], while the second part solves the problem of assigning the
wavelengths. There are other proposals that solve the problem from an integral
perspective, i.e., without subdividing it [5]. The present paper makes a comparison
of the two methods from the heuristic standpoint [6, 3].

1.2 Description of the RWA Problem

In All Optical Networks (AON) the transmitters are connected with each receiver
at all wavelengths, and therefore as many signals are required as there are
wavelengths. When many sessions use the same wavelength, these signals cannot
be sent simultaneously along the same fiber to avoid collisions within the network,
consequently blocking the session. It must be ensured that if a receiver is awaiting
data at a given wavelength in a given time interval, only one signal at that
wavelength must get to the receiver; if two or more signals arrive under those
conditions, we face a phenomenon called contention, which can be avoided by
isolating the signals in space and/or in time, i.e., by sending the signal along
another road and/or at another time. Temporary isolation is in general what we are
trying to avoid, because it causes latency.
1 RWA: Comparison of Genetic Algorithms … 5

Fig. 1.1 Example of

contention in optical
networks

An optical network is formed by OXC (Optical Cross Connect) or optical

commuters linked by optical fibers. Depending on the technology of these com-
muters, they can commute fibers, wavelengths, wavelength bands, time division
multiplexing (TDM). The optical routes are formed by optical links that direct the
light beam (data) through the same wavelength according to the wavelength con-
tinuity restriction, but this possibility is exhausted when the demand exceeds the
number of possible sessions; when this restriction cannot be satisfied, the request is
blocked. Routing systems reusing the wavelength, where it is allowed to be
changed, have been developed, however the studies do not show significant changes
in routing performance [7]. Figure 1.1 shows 5 OXC that are part of a larger system.
The established routes with their corresponding wavelengths (colors) are (x-1, x,
x + 2/Red); (x-1, x, x + 1/Green) and (x-2, x, x + 1/Light blue); when a service
request arrives whose solutions goes through the (x-2, x, x + 2) links as shown in
blue color, the specified route cannot be assigned the red wavelength because it is
being used in the (x, x + 2) link, so either another wavelength should be used or a
new route should be found. This problem is called contention.
We must define some variables to formulate the problem. We know that in the
network there is current traffic and arriving or requested traffic. Traffic can be static
or dynamic, depending on the existing relations between the current and arriving
traffic. It is dynamic when the average connection time of current traffic is much
less than the average time between services request arrivals. This scenario is not
optimizable, so heuristic algorithms are used, most of which offer greater solutions
that are not necessarily optimum, but this richness provides much support when
the contentions problem arises, avoiding a new execution of the algorithmic
process. The problem is to establish the routes and the wavelength assignments in
the network for various criteria, with minimum probability of blocking the
requests, low transport latency along the route, low level of jitter, etc. The problem
is known as RWA, Routing Wavelength Assignment (routing and wavelength
assignment). Different strategies have been tested to satisfy the demand of an
optical network, and optimization criteria and algorithms have been used [8]. The
optimization criteria have been based on:
6 A. Rodriguez et al.

Fig. 1.2 RWA–CCW solution strategies

• Minimizing the probability of blocking new requests [5].

• Minimizing the use of the network’s wavelengths [9, 10, 11].
• Minimizing the use of wavelengths per link [12, 13].
• Minimizing dispersion on the route [14].
• Minimizing the ASE (Amplified Spontaneous Emission) [15]
The optimizations algorithms can be classified into conventional, which are
algorithms used in electronic routing or only as an initial population value for
heuristic algorithms, such as minimum cost roads, dijkstra; incremental cost roads,
bellman fulkerson; minimum delay roads, bellman ford; and Linear Programming
[16], and heuristics, which are widely used algorithms in the search for good routes
and not necessarily optimum routes, such as Ant [17, 18], Genetic [19, 6], Tabu
Search [20], and Simulated Annealing [3, 4]. The diagram in Fig. 1.2 shows the
processes located in the literature for solving the RWA problem. It is seen that they
go from using problem division strategies until it is completely solved, making use
of different criteria and algorithms that allow getting a route and a wavelength that
allow the information to be transported. It should be stressed that in optical
1 RWA: Comparison of Genetic Algorithms … 7

networks the aim is not necessarily to optimize, but what is sought is rather a route
that will work without generating a substantial increase of the probability of
blocking and without enlarging the network.

1.3 Model Description

In this demand model the nodes N participate in the routing request from the
network with a probability pl (for the lth access node connected with the lth optical
node Xl). This is the routing request probability that will be used as a simulation
parameter. The demand Ml of the lth access node follows a Poisson distribution
with an average rate le.

lm
e le
PðMl ¼mÞ ¼ e ð1:1Þ
m!

With this model we can have a varied range of requests around the whole
network and we can vary the load intensity at all the nodes, and the demand can be
changed from static to dynamic according to the study that it is desired to make.
The comparative work was done under dynamic demand.

1.4 Description of the Algorithms

The selected strategies are to divide the problem into two parts or to use it wholly.
Simulated annealing was used in the first case, and genetic Algorithms in the
second case, both of them carried out previously by the Grupo de Investigacion de
Nuevas Tecnologias (GINT) Industrial Technology Department, Facultad Tecno-
logica of the Universidad Santiago de Chile.

1.4.1 Network Demand

Requests arrive at the boundary nodes (with Poisson behavior, but different
probabilistic scenarios could be studied). These requests bring along four
parameters that must be satisfied, otherwise the request will be blocked.

vSi ¼ ðrO ; rD ; nC ; tC Þ ð1:2Þ

Where:
vSi is the vector that represents the ith request that arrives at the sth boundary
node (Edge Router)
8 A. Rodriguez et al.

rO is the identification number of the node of origin

rD is the identification number of the destination node of the arrival request
nC is the number of connections requested for the (rO,rD) pair
tC is the time of connections requested for the (rO,rD) pair

Three matrices have been established: The first linking matrix, C, that will
always be setting the available capacity of the links, so it can be used to monitor
when a link is not available; the second wavelength matrix, hh, which determines
the wavelength in use; and the third, the time matrix T, is a matrix that will have
the function of keeping count of the connection time of each wavelength in each
link, and must be updated dynamically. Its structure is similar to that of matrix
hh, but its elements will have the value 1 when the wavelengths are available in
the given link, and the time will be negative when it is in use, and it will decrease
as time goes by, and in this way it can be detected in the aptitude functions FA so
as not to use it. The definitions of the matrices are given below [21, 19, 6].
8 9
< cij =ðcij ¼ 0 8 i ¼ j ^ cij ¼ G NEE =
c¼ ^ cij ¼ g 8 i 6¼ j SEE ^ j 2 ½0; N  1 ð1:3Þ
: ;
^ k 2 ½0; nw  1

G= Very large number

g= Instantaneous cost of the link
NEE = There is no link
SEE = There is a link

8 9
< kijk =ðkijk ¼ 0 LOU _ NEEÞ _ =
k ¼ ðkijk ¼ k þ 1 LODÞ 8 i 2 ½0; N  1 ð1:4Þ
: ;
^ j 2 ½0; N  1 ^ k 2 ½0; nw  1

nW = Number of wavelengths in the network

8 9
< tijk =ðtijk ¼ tc LOU tijk ¼ 0 NEEÞ _ =
T ¼ ðtijk ¼ 1 LODÞ 8 i 2 ½0; N  1 ð1:5Þ
: ;
^ j 2 ½0; N  1 ^ k 2 ½0; nw  1

LOU = Wavelength in the link is being used

LOD = Wavelength is not being used
1 RWA: Comparison of Genetic Algorithms … 9

Fig. 1.3 Initial population 0 1 2 3 4 5 6 7 8 9

matrix
0 ⎡1 3 9 3 2 2 5 6 9 4⎤
1 ⎢⎢1 4 9 2 6 3 4 4 2 4 ⎥⎥
2 ⎢1 1 1 3 4 5 3 7 1 4⎥
⎢ ⎥
3 1 4 2 7 3 2 4 5 8 4⎥
S0 = ⎢
4 ⎢1 5 9 7 5 1 2 1 9 4⎥
⎢ ⎥
5 ⎢1 6 8 5 7 4 8 6 1 4⎥
6 ⎢1 7 8 2 3 6 9 7 6 4⎥
⎢ ⎥
7 ⎣⎢1 3 2 8 5 3 2 4 7 4 ⎦⎥

1.4.2 Simulated Annealing Algorithm

Let S0 be the two-dimensional matrix of order m x N, where m is the number of

rows of the matrix, which indicates the population sample used, in addition to
being a simulations variable, while N indicates the number of nodes.
8 9
< sxy =sxy ¼ a ^ a 2 ½0; N  1 =
Si ¼ ^ x 2 ½0; m  1 ^ y 2 ½0; N  1 ð1:6Þ
: ;
8 N; m ¼ 2k ^ k 2 Z þ

Futhermore:

s0y ¼ rO ^ sxN1 ¼ rD 8 x; y ð1:7Þ

When a request arrives at node 1, matrix S0 is established; if it arrives at node 2,

matrix S1 is established, and so on. When m015 ¼ ð1; 4; 3; 128Þ arrives, we read: 15
service request entering node 1, which requests route from node 1 to node 4 with
three 128 ms. connections. Therefore, according to (1.8), the first column has the
origin node and the last column has the destination node. The rest of the elements of
the initial population are filled randomly. For example, (Shown Fig. 1.3), for
m = 8 and N = 10, and the request for connection from node 1 to node 4.
Then rotation of the internal layers at different speeds must take place to find the
routes. Also, the wavelength assignments were made with First Fit, to be able to
compare them under the same conditions in terms of stopping criteria, aptitude
functions, and other parameters of the algorithm. They can be seen in detail in [3, 4].

1.4.3 Genetic Algorithms

Let P0 be the initial population of the genetic algorithm to be used, and nP the
number of chromosomes that constitute these initial populations. There will also
be ‘‘m’’ genetic algorithms for the ‘‘m’’ nodes that are requesting transport service.
10 A. Rodriguez et al.

Fig. 1.4 3D matrix of the

initial population

Within the genetic algorithms there are various implementations that allow dif-
ferent behaviors of nP (regeneration policies); in this research this value remains
constant and it is a simulation parameter. In this way, the three-dimensional matrix
P0 of nP x (N + 3nW + 2) x N elements is defined.
8 9
< pabc =pabc ¼ y ^ a 2 ½0; nP  1
> >
=
P0 ¼ ^ b 2 ½0; N þ 3nW þ 2 ^ c 2 ½0; m  1 ð1:8Þ
>
: >
;
_ y 2 ½0; N  1

P0bc ¼ rO ^ Pa0c ¼ rD 8 a; b; c ð1:9Þ

This matrix is filled randomly, the additional columns are for saving the cal-
culations of the aptitude functions referring to each available wavelength. La
Fig. 1.4 shows the 3D matrix of the initial populations that will serve for the
genetic process. For example, Fig. 1.4 shows the initial population of a network
with 6 (m = 6) nodes and 8 (nP = 8) chromosomes without intergenerational
numerical change, and in Fig. 1.5 each commuter is a gene that constitutes the
chromosome, whose information is the route that is sought. For example, in row 4
the chromosome is 2-3-3-2-1-4, where the first and the last genes are the origin and
the destination. The rest of the genes are filled randomly in the population matrix.
Assuming that the network has two wavelengths (nW = 2), in Fig. 1.5, we would
have one of six genetic algorithms with ‘‘a’’ from 0 to 9, ‘‘b’’ from 0 to 5 and ‘‘c’’
from 0 to 5.
If the request that arrives at node 2 is m015 ¼ ð2; 4; 3; 128Þ, then the matrix would
be:
1 RWA: Comparison of Genetic Algorithms … 11

Fig. 1.5 2D matrix of the 0 1 2 3 4 5 6 7 8 9 10

initial population of node 2, 0 ⎡2 1 2 2 3 4 ⎤
belonging to P0
1 ⎢⎢ 2 3 1 1 2 4 ⎥
⎥
2 ⎢2 4 3 3 1 4 ⎥
⎢ ⎥
3 ⎢2 2 4 4 2 4 ⎥
COMPUTATIONAL ⎥
4 ⎢2 3 3 2 1 4
Pab 1 = ⎢ USE ZONE ⎥
5 ⎢2 2 4 4 3 4 ⎥
6 ⎢2 1 5 5 4 4 ⎥
⎢ ⎥
7 ⎢2 2 3 3 1 4 ⎥
8 ⎢2 5 2 2 2 4 ⎥
⎢ ⎥
9 ⎢⎣ 2 2 3 1 1 4 ⎥⎦

Fig. 1.6 NSFNET network

used with the four methods

Then the algorithm is executed to find the routes; as to the stopping criteria,
aptitude functions and other parameters of the algorithm, they can be seen in detail
in [21, 19, 6].

1.5 Comparison Scenario

Once the simulation had been made, it was compared with two similar reports
under the same simulation conditions.
The network used for the test was the NSFNET (National Science Foundation
NETwork), which has 16 nodes and 25 fiber optics links (Shown Fig. 1.6). The
parameters used were similar to those presented in [1]. Comparisons were made of
probability of blocking and use of the network, varying the load in the [0,180]
interval, with 10-erlang increments. The number of connections made during the
simulation in both scenarios, simulated annealing (SA) and genetic algorithms
(GA), was 108 connection requests.
12 A. Rodriguez et al.

Fig. 1.7 Comparison of blocking probability and network use in the NSFNET

1.6 Results

Comparisons were made of the simulated annealing [3], genetic algorithms [19], of
Dr. Xiao Wen [2], and of Dr. Hui Zang [1] simulations with the blocking prob-
ability indicator, which measures the probability of blocking the service requests
under a given load, and the network use indicator, which measures network
occupation under a given load. Figure 1.7 graphs the blocking probability,
showing that up to 120 Erlangs the GA has better performance than SA, but above
120 Erlangs the SA is better than GA, but in no case are they better than that of
Hui Zang; the excellent performance of the Xiao Wen algorithm is worth noting,
but it is because 40 wavelengths are used, compared to eight wavelengths that
were used in the other three studies. Figure 1.7, graphs the percentage use of the
network, comparing the network use with the previously mentioned reports. The
SA algorithm improves the network use for loads smaller than 50 Erlangs, but for
greater loads the GA algorithm uses fewer network resources. But, under 110
Erlangs the SA and GA algorithms are not better than that of [1], but the GA
algorithm is better for loads greater than 110 Erlangs. Note that work with SA and
GA reaches 180 Erlangs, while the other reports involve simulations with lower
load intervals. On the other hand, the performance of the algorithm reported in [2]
involves high network consumption, even with many wavelengths, generating a
problem to satisfy future load demand.

1.7 Conclusions

The work reported in [2, 1] is always aimed at optimum solutions, while the work
based on SA and GA looks for solutions without the need to go over the whole
possible universe, thereby saving a large amount of operating time. From the
results it can be concluded that heuristic algorithms have great potential for future
demand because they function much better at high loads and small dynamic traffic,
and the latter should become very important because it allows the use of the
1 RWA: Comparison of Genetic Algorithms … 13

network with an optimum sense. This research team agrees that it will be necessary
to develop the re-use of wavelengths to be able to remove the restrictions given in
CCW and improve the results while better proposals are developed.

References

1. Zang, H., Jue, J.: Dynamic lightpath establishment in wavelength-routed WDM networks.
IEEE Commun. Mag. 39, 100–108 (2001)
2. Chu, X., Li B., Zhang, Z.: A dynamic RWA algorithm in a wavelength-routed all-optical
network with wavelength converters, INFOCOM 2003. Twenty-second annual joint
conference of the IEEE computer and communications, vol. 3, pp. 1795–1804. IEEE
Societies (2003)
3. Rodríguez, A., Saavedra, F., Ramírez, L.: Simulated Annealing una Propuesta de Solución al
Problema RWA en Redes Fotónicas. IEEE Intercon UNI, Lima-Perú (2011)
4. Guenduez, H., Kadir, H.: A well-arranged simulated annealing approach for the location-
routing problem with time windows. In: 46th Hawaii International Conference on System
Sciences (HICSS), pp. 1144–1153 (2013)
5. Liu, L., Xue, Q., Zhang, J., Wang, F., Wu, J., Tong, Lin, J., Bin, L.: Investigation of uniform
models for analysing network blocking probability in ASON, communications, networking in
China, 2006. ChinaCom ‘06. First International Conference on, pp. 1–4 (2006)
6. Rodríguez, A., Saavedra, F., Ramírez, L.: Solución Simultánea Del Enrutamiento Y
Asignación De Longitud De Onda en redes WDM Con Algoritmos Genéticos. IEEE
COLCOM, Bogota, Colombia (2008)
7. Ramaswami, R.: Optical networking technologies: what worked, what didńt. IEEE Commun.
Mag. 44, 132–139 (2006)
8. Balasis, F., Wang, X., Xu, S., Tanaka, Y.: A dynamic physical impairment-aware routing,
wavelength assignment scheme for 10/40/100 Gbps mixed line rate wavelength switched
optical networks. In: 15th International Conference on Advanced Communication
Technology (ICACT), 2013, pp. 116–121 (2013)
9. Zang, H., Jue, J., Mukherjee, B.: A review of routing, wavelength assignment approaches for
wavelength-routed optical WDM networks. Opt Netw. Mag. 1, 47–60 (2000)
10. Ramaswami, R., Sivarajam, K.: Design of logical topologies for wavelength-routed optical
networks. IEEE J. Sel. Areas Commun. 14, 840–851 (1996)
11. Wang, J., Qi, X., Chen, B.: Wavelength assignement for multicast in all optical WDM
networks with splitting constraints. IEEE/ACM Trans. Networking 14:169–182 (2006)
12. Kuri, J., Puech, N., Gagnaire, M., Dotaro, E., Douville, R.: Routing, wavelength assignement
of scheduled lightpath demands. IEEE J. Sel. Areas Commun. 21, 1231–1240 (2003)
13. Saengudomlert, P., Modiano, E., Gallager, R.: Dynamic wavelength assignment for WDM
all-optic tree networks. IEEE/ACM Trans. 13, 895–905 (2005)
14. Gurzi, P., Steenhaut, K., Nowe, A.: Minimum cost flow based R&WA algorithm for
dispersion, OSNR limited all-optical networks. In: 15th International Conference on Optical
Network Design, Modeling (ONDM), 2011, pp. 1–6 (2011)
15. Feres, M., Trevelin, L.: RWA algorithm aware of PMD, ASE for all-optical networks. In:
11th International Conference on Transparent Optical Networks, 2009. ICTON ‘09, pp. 1–4
(2009)
16. Ozdaglar, E., Bertsekas, D.: Routing, wavelength assignment in optical networks. IEEE/
ACM Trans. Networking 11, 259–272 (2003)
17. Zeng, P., Yu, H.: An ant-based routing algorithm to achieve the lifetime bound for target
tracking sensor networks. In: IEEE International Conference on Communications, ICC ‘06,
vol. 8, pp. 3444–3449 (2006)
14 A. Rodriguez et al.

18. Tan, S.: Ant-based Physical Attack, Amplifier Spontaneous Emission-aware routing.
Communication Technology (ICCT), 2012 IEEE 14th International Conference on,
pp. 650–653 (2012)
19. Rodríguez, A., Saavedra, F.: Enrutamiento y asignación de longitudes de onda en redes
WDM: Solución simultánea basada en algoritmos genéticos. IEEE Intercon UNSA,
Arequipa-Perú (2009)
20. Assis, K., Ferreira, D., Giozza, W.: Hybrid algorithms for routing assignment wavelengths in
optical networks. Lat. Am. Trans. 8, 214–220 (2010)
21. Barpanda, R., Turuk, A., Sahoo, B., Majhi, B.: Genetic algorithm techniques to solve routing,
wavelength assignment problem in wavelength division multiplexing all-optical networks. In:
Third International Conference on Communication Systems, Networks (COMSNETS),
pp. 1–8 (2011)
22. Zang, H., Sahasrabuddhe, L., Jue, J., Ramamurthy, S., Mukherjee, B.: Connection management
for wavelength-routed WDM networks, global telecommunications conference, 1999.
GLOBECOM ‘99, vol. 2, pp. 1428–1432 (1999)
Chapter 2
Route Optimization in Proxy Mobile IPv6
Test-Bed via RSSI APPs

Nur Haliza Binti Abdul Wahab, L.A. Latif, S.H.S. Ariffin,

N. Fisal and N. Effiyana Ghazali

Abstract Proxy Mobile IPv6 is the new protocol, but there is a problem on
Localized Routing (LR) algorithm which it is not ready build in PMIPv6 protocol.
As for this in PMIPv6 in RFC5213 address the needed to enable Localized Routing
(LR) but it not specify a complete procedure to establish Route Optimization (RO).
RFC6279 state the problem statement on LR issue with several scenarios to tackle
down. LR is important especially to minimize data delay and decrease handover
latency due to the un-optimized data route. Data packets on PMIPv6 protocol
without LR always need to travel to Local Mobility Anchor (LMA) with result in
end-to-end delay. Therefore, this paper propose the new LR algorithm to optimized
data route for selected scenarios to reduce handover and data packet delay. This
paper also will illustrate the setting up of the PMIPv6 and test-bed performance.

2.1 Introduction

Users demand on mobility dedicate Internet Engineering Task Force (IETF) to

designed a standard communications protocol named as Mobile IP or IP mobility
to allow Mobile Node (MN) to maintain their permanent IP address while moving
around from one network to another network.
Mobile IP for Internet Protocol version 4 (IPv4) addresses first described in
IETF and documented in [1] while Mobile IPv6 (MIPv6) protocol was designed by
IETF to support mobility implementation for the next IP generation of IPv6 [2].

N.H.B.A. Wahab
L.A. Latif (&)
N. Fisal
N.E. Ghazali

UTM Razak School, Universiti Teknologi Malaysia, UTM, Kuala Lumpur, Malaysia
e-mail: [email protected]
N.H.B.A. Wahab
e-mail: [email protected]
S.H.S. Ariffin
UTM MIMOS Center of Excellence, Universiti Teknologi Malaysia, UTM, Skudai,
Malaysia

 Springer International Publishing Switzerland 2015 15

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_2
16 N.H.B.A. Wahab et al.

Table 2.1 Differences between Mipv6 and Pmipv6

Protocol criteria MIPv6 PMIPv6
Mobility scope Global Local
Location Yes Yes
management
Required Home Local mobility anchor (LMA), mobile access gateway
infrastructure agent (MAG)
MN modification Yes No
Handover latency Bad Good
Localized routing Yes No

Although MIPv6 enable a MN to move but this ability is still not sufficient for
true mobility as MIPv6 requires the MN to modify its client functionality in the
IPv6 stack. Plus, the need of MN to involve in IP related to signaling and enabling
efficient handover is an additional and critical requirement thus IETF again
introduces new protocol to tackle the issues occur in MIPv6. The new protocol
known as Proxy Mobile IPv6 (PMIPv6). Table 2.1 shows the differentiation
between MIPv6 and PMIPv6.
PMIPv6 provide new approach named as Network-Based Localized Mobility
Management (NetLMM) which has been developed to relocate mobility proce-
dures from mobile devices to network components to support in IP networks.
There are three main advantages [1, 2] of PMIPv6 and they are:-
• Handover performance optimization.
PMIPv6 can reduce latency in IP handovers by limiting the mobility man-
agement within the PMIPv6 domain. Therefore, it can largely avoid remote
service which not only cause long service delays but consume more network
resources.
• Reduction in handover-related signaling overhead.
The handover-related signaling overhead can be reduced in PMIPv6 since it
avoids tunneling overhead over the air as well as the remote Binding Updates
either to the HA or to the Correspondent Node (CN).
• Location privacy.
Keeping the mobile node’s Home Address (MN-HoA) unchanged over the
PMIPv6 domain dramatically reduces the chance that the attacker can deduce
the precise location of the mobile node.
Although there are many advantages on PMIPv6 rather than MIPv6, there is a
lack on PMIPv6 protocol. As state in Table 2.1, there are no Localized Routing
(LR) protocol build-in in PMIPv6. Therefore [3] address the need to enable this
LR features.
There are two main entities in PMIPv6. Local Mobility Anchor (LMA) and
Mobile Access Gateway (MAG). The function of LMA is to act as the persistent HA
for MN and as the topological anchor point for MN prefix assignments. LMA also
2 Route Optimization in Proxy Mobile IPv6 … 17

function as default multicast upstream for the corresponding MAG which manages
the links to maintain the state of MN.
LMA is responsible to detect MN movements and change of attachment.
Packets sent to and received from MN are routed via tunnels between LMA and
the corresponding MAG. MAG performs the mobility-related signaling on behalf
of MN which is attached to it. MAG acts as an Access Router for MN, which is the
first-hop router in the Localized Mobility Management (LMM).
Figure 2.1 shows the basic operation of PMIPv6 [3] as the signaling informa-
tion transverse when a Mobile Node connects to the PMIPv6 network. The sig-
naling flow starts when a MN enters a PMIPv6 domain and MN attaches to an
access link provided by a MAG.
MAG on that access link performs the access authentication process which it
identifies MN, the LMA’s address and other configuration parameters. To update
the LMA about the current location of MN, MAG sends Proxy Binding Update
(PBU) to the LMA which also contain MAG address with identity of the MN. Upon
receiving this request, the LMA allocates MN’s home network prefix and sends
back to the MAG as Proxy Binding Acknowledgement. It also creates the Binding
Cache entry and sets up its endpoint of the bi-directional tunnel to the MAG.
While MAG sets up its endpoint of bi-directional tunnel to the LMA, it also sets
up the forwarding channel for MN’s traffic. MAG then sends Router Advertise-
ment messages to MN on access link including the prefix allocates to MN. Upon
receiving the Router Advertisement messages on the access link, MN attempts to

Fig. 2.1 Signaling when Mobile node connects to PMIPv6 network [3]
18 N.H.B.A. Wahab et al.

Fig. 2.2 Overview of

scenario A21

configure an address using interface either stateful or stateless address configu-

ration modes.
After obtaining the initial address configuration in PMIPv6 domain, if MN
moves, the MAG updates the location of the MN and will signal the LMA and
advertises the same prefix to MN. MN keeps the address configured when it first
enter the LMA domain. This means that MAG configured the same link local
address for specific MN.
PMIPv6 cannot directly apply the LR protocol in MIPv6 as MN is not involved
in mobility signaling thus it cannot perform signaling to set up a RO [3, 4]. In
PMIPv6, data packets have to always go through the LMA even if the CN and MN
sit on the same network.
In [3], PMIPv6 protocol does not fully specify the procedure to establish such
localization routing state. The problem specified in [4] which have 5 scenarios to
be considered.
All scenarios refer to ‘‘A’’ cases as assumption that both the MN and the CN are
registered with an LMA according to the PMIPv6 protocol [3, 94]. The scenarios
represent as ‘A = (number of MAGs)(number of LMAs)’ which scenario A11,
A21, A12, and A22 [4]. Scenario that has similar topology as in the office or indoor
environment is scenario A21. Scenario A21 is scenario where there are two MAGs
and one LMA in the network. Figure 2.2 shows the overview of scenario A21.
LR for scenario A21, needs the routing information at both MAGs so that data
packet can directly forwarded between MN’s MAG and CN’s MAG. As LMA is
the common anchor for the MN and the CN, LMA should maintains location
information for both nodes.
The remainder of this paper is organized as follows: Section II will discuss
related research work done in this field; Section III will describe the setup of the
test-bed; Section IV will present the localized routing algorithm; Section V will
present the results and discussion; and lastly Section V will conclude the paper.
2 Route Optimization in Proxy Mobile IPv6 … 19

2.2 Related Research

Over the past few years, there has been a huge interest in PMIPv6 research area
after it was proposed to overcome the long registration delay problem. With
PMIPv6, it can avoid tunneling overhead over the air and support the hosts without
any involvement in the mobility management side.
Several related research has been proposed in this field such as in [9]. Authors
in [9] proposed some enhancements for PMIPv6 as their studies show that PMIPv6
may cause high handover latency is the LMA is located far from the current MAG
through simple mathematical model.
Paper [10] presented a new novel localized routing scheme called LRP. Jun
et al. [10] claim that their proposed design has shorten the response time of LR
compared with a classical protocol. An experimental evaluation on PMIPv6 under
different implementation configurations and they evaluated the impact on the
performances on PMIPv6 real test-bed.
Research journal in [11] also presented the real PMIPv6 test-bed implemen-
tation but it focus on Network Mobility (NEMO). Minoli [11] propoed a NEMO
supporting scheme, which supports MN’s mobility between PMIPv6 networks and
mobile networks as well as the basic NEMO.
Iapichino and Bonnet [12] focused only hardware or real test-bed. Their test-
bed OS used Ubuntu 9.04 and the kernel is 2.6.31. Software that used in [12] is
OpenWRT Kamikaze 7.09. Linksys WRT54GL v1.1 is used as the Access Point in
[12]. Iapichino and Bonnet [12] only compare their system performance result
with previous work. Another related research is [13] where looking at improve-
ment localized routing handover. Lee et al. [13] is a project which run the sim-
ulation without test-bed development.

2.3 Test-Bed Setup

The PMIPv6 test-bed architecture for this work is shown in Fig. 2.3. In this
architecture, we can see that the PMIPv6 test-bed consists of several entities which
three (3) computers, two (2) notebooks, two (2) access points (AP) and one (1) hub
or router.
The three computers are needed to be setup as LMA, MAG1 and MAG2 while
notebooks are represent and acted as MN and CN. The hub or router is needed to
work as interfaces that connect LMA, MAG1 and MAG2. AP will acted as AP for
MAG1 and MAG2 as well.
dAPs is the distance between AP1 and AP2. Communication or handover pro-
cess are different and vary when dAPs change. An experimental are done with
different dAPs to see the correlation between distance and handover latency.
20 N.H.B.A. Wahab et al.

Fig. 2.3 Proxy mobile IPv6 architecture

2.3.1 Software and Hardware Requirement

The hardware specification for this PMIPv6 test-bed project are listed in Table 2.2.
Specification shows all the IP address for every components in the PMIPv6 test-
bed and their hardware details.
Hardware needed for LMA, MAG, MN and CN did not require specific type of
processor, RAM, type of devices (computer or notebook) and more. But, for LMA
and MAGs, the hardware require should use computer as it need more than two
Ethernet card to be used to connect to AP, MN or CN.
The type of AP that used should be an AP that can be configure or modified
their setting and program because the AP should be work on masquering mode to
fulfill the PMIPv6 test-bed. In this project, Cisco Aironet 1200 Series AP are used
which can be configure and modified to be used in masquering mode.
2 Route Optimization in Proxy Mobile IPv6 … 21

Table 2.2 Hardware specification

Device Hardware specification IP address
LMA Processor :Intel CoreTM i5-2310 eth0 : 2001:100::1
Memory : 3.2 GiB eth1 : 2001:2::1
CPU :2.90 Ghz
MAG1 Processor :Intel Pentium
DualCoreTM i5-2310 eth0 : 2001:100::2
Memory : 3.2 GiB eth1 : 2001:1::1
CPU :2.90 Ghz
MAG2 Processor :Intel Pentium
DualCoreTM i5-2310 eth0 : 2001:100::3
Memory : 3.2 GiB eth1 : 2001:1::2
CPU :2.90 Ghz
MN Processor : Inter CoreTM i3-M350
Memory : 1.8 GiB wlan0 : 2001:100:6:5000:
CPU : 2.27 GHz 214:6CFF:FE53:180A
CN Processor :Intel CoreTM 2 Duo eth0 : 2001:2::2
Memory :1.9 GiB
CPU :[email protected] GHz
AP Cisco Aironet 1200 Series –
2.4 GHz@5 GHz 54 Mbps

Several software used to build up this PMIPv6 test-bed project. The main
software needed is Linux Open Source (OS) for the base system. The Linux type
used in this work is Ubuntu 10.04 LTS that chosen to be used in all components
(LMA, MAG1, MAG2, MN and CN) in this PMIPv6 test-bed.
As Linux was chosen to be as the platform system, the next important thing to
look down is the kernel version for the Linux. In this project, the kernel version
used is 2.6.32.59 + drm33.24 which a modification kernel from kernel version
2.6.32.

2.3.2 Kernel Setup and Installation

As mention previously, kernel version used in this PMIPv6 test-bed is

2.6.32.15 + drm.33.24. This kernel is modified version from kernel version
2.6.32. The kernel needed to enable several features to support PMIPv6 envi-
ronment. After modified the kernel, several packages needed to be installed to
fully run this PMIPv6 test-bed project.
22 N.H.B.A. Wahab et al.

2.3.3 Access Point Cisco Aironet 1200

Access Point (AP) that used in this project is Cisco Aironet 1200 Series, as
mention in previous section. This AP needed to be configured to enable their
SYSLOG client and acted in masquering mode. This is because, the AP should be
a separator for the MAG function that connected to it.

2.4 Localized Routing Algorithm

The Route Optimization (RO) is Using Multicasting Binding Cache approach to

MAGs method, add on RSSI APPs and Prediction method. RSSI APPs is the APPs
that been installed to the MN so that MN will keep sending Signal Strength of
every APs of MAGs detected and sent to the LMA.
LMA is a Dynamic LMA which it will calculate and compare all the Signal
Strength given by MN via RSSI APPs and keep update the MAGs with all the RSSI
using Multicasting Binding Cache approach. Since the LMA is in the local domain,
it can always multicast its binding cache to all the MAGs to keep update cache ahead
before any operation (Handover, data communication, etc.) take happen.
The Prediction method will change the data path route automatically when MAGs
detect AP-MAG1 = AP-MAG2 or AP-MAG1 [ AP-MAG2 for the first time follow
to the topology. It’s mean, before handover happen, the path route of data communi-
cation already change to the shorter path or known as Route Optimization (RO),
without need to wait for the handover process to end to change the data path route.
The data travel to maximum path if there is no RO implementation. Data from
CN will be sent to MAG1 to LMA, than back to MAG1 and lastly to MN
(CN ? MAG1 ? LMA ? MAG1 ? MN), rather than with LR via RSSI APPs
the data will be travel by shorter path which data from CN to MAG1 than direct to
MN because they attach to the same MAG (CN ? MAG1 ? MN).
While MN moving, it keep sending the signal strength of both MAGs (MAG1
and MAG2) to LMA so that LMA are up to date of the MN location. As LMA which
always sending BU to MAGs using multiple binding cache approach to MAGs will
keep update the caches to MAGs before any operation (handover) happen.
The prediction method will change the data path when MAGs detect the signal
strength of AP-MAG1 equal to signal strength of AP-MAG2 or when AP-MAG1
higher than signal strength AP-MAG2 (AP-MAG1 C AP-MAG2). This mean,
before handover happen, the data path already change to the shorter path, without
need to wait for the handover process to end to change the path flow. The data flow
with LR via RSSI APPs and without LR shows in Fig. 2.4. In Fig. 2.4, we can see
that data paths flow is the black arrows and the red arrow is the deleted or reduction
path. The data flow with LR via RSSI APPs and without LR shows in Fig. 2.5.
2 Route Optimization in Proxy Mobile IPv6 … 23

Fig. 2.4 Data flow with RO via RSSI APPs and without RO

Fig. 2.5 Data flow with and without route optimization in PMIPv6

2.5 Result and Discussion

The aims of this project is to develop PMIPv6 test-bed that can be used for
experimental work. By having this PMIPv6 test-bed, other work can be implement
in this test-bed for enhancement especially as next work for this project is to look
forward for handover and localized routing problem in PMIPv6.
24 N.H.B.A. Wahab et al.

Fig. 2.6 Successfully enable PMIPv6 features

Fig. 2.7 MN successfully ping CN

To allow IPv6 addresses to be used in the system, the kernel setting should
enable the IPv6. As mention in previous section, the kernel 2.6.32 should be setup
to meet IPv6 forwarding. Figure 2.6 shows that the LMA system has been suc-
cessfully enabled in the PMIPv6. To see this result, the LMA need to be run and
the script in Fig. 2.6 will be seen.
Final result to prove that this PMIPv6 test-bed project is successfully developed
is by making communication between MN and CN. Figure 2.7 shows that MN can
communicate with CN with ‘ping’ procedure.
Experiment has done for topology shown in Fig. 2.5 with two scenarios which
dAPs 5 and 10 m. The experiment has done for handover without LR and handover
with LR to check and compare the handover latency. As to mention again, to
2 Route Optimization in Proxy Mobile IPv6 … 25

Fig. 2.8 Handover latency versus packets send for scenario 1

Fig. 2.9 Handover latency versus packets send for scenario 2

enable multimedia communication (seamless multimedia handover) while hand-

over happen, the handover latency should be less than 50 ms.
Figure 2.8 shows the handover latency versus packets send for scenario 1. Sce-
nario 1 is the scenario where distance between AP-MAGs is 5 m (dAPs = 5 meter).
We can see that the handover latency without LR is more than 50 ms while data
packets send more than 30 packets (64 bytes every packets). As for this, it cannot be
used for seamless multimedia handover. When the LR algorithm (RSSI APPs LR
with Dynamic LMA for PMIPv6) used, the handover latency will be minimize until
less than 10 ms for 10–100 packets send. This conclude that the LR algorithm can be
used for seamless multimedia handover as it reduce the handover delay.
While Fig. 2.9 shows the handover latency versus packets send for scenario 2.
Scenario 2 is the scenario where distance between AP-MAGs is 10 m
(dAPs = 10 m).
We can see that the handover latency without LR is more than 50 ms while data
packets send more than 20 packets (64 bytes every packets). As for this, it cannot
26 N.H.B.A. Wahab et al.

be used for seamless multimedia handover. When the LR algorithm (RSSI APPs
LR with Dynamic LMA for PMIPv6) used, the handover latency will be minimize
until less than 20 ms for 10–100 packets send. This conclude that the LR algorithm
can be used for seamless multimedia handover as it reduce the handover delay.
The delay is higher than scenario 1 because when the dAPs is large, the delay time
will increase.

2.6 Conclusion

This PMIPv6 test-bed was successfully run without errors and handover experi-
ment had been done to see the handover latency for different distance between AP.
This proposed work has successfully develop real test-bed for PMIPv6. The
optimized routing problem in [8] has completely solved with reducing the hand-
over delay.

2.7 Future Work

For future work, the algorithm will be run on the simulation also by using NS3.

References

1. Perkins, C. (ed.): IP Mobility Support for IPv4, Revised, RFC5944

2. Johnson, D.: Rice University, Mobility Support In IPv6, RFC3775
3. Kato, T., Takechi, R., Ono, H.: A study on mobile IPv6 based mobility management
architecture, FUJITSU Science Technology Journal 37, 1, June 2001. M. Clerc, the swarm
and the queen: towards a deterministic and adaptive particle swarm optimization. In:
Proceedings of the IEEE Congress on Evolutionary Computation (CEC), pp. 1951–1957
(1999)
4. Gundavelli, S., Leung, K., Devarapalli, V., Chowdhury, K., Patil, B.: Proxy mobile IPv6,
RFC 5213, Aug 2008. RFC5213-3 H.H. Crokell, specialization and international
competitiveness. In: Etemad, H., Sulude, L.S. (eds.) Managing the Multinational
Subsidiary, Croom-Helm, London (1986)
5. Liebsch, M.: PMIPv6 localized routing problem statement, draft-ietf-netext-pmip6-lr-ps-02,
RFC6279, Jan 2010. In: Deb, K., Agrawal, S., Pratab, A., Meyarivan, T. (eds.) A Fast Elitist
Non-dominated Sorting Genetic Algorithms for Multiobjective Optimization: NSGA II,
KanGAL report 200001, Indian Institute of Technology, Kanpur, India (2000)
6. Jeong, S., Shin, M.-K., Kim, H.-J.: Performance comparison of route optimization schemes in
proxy mobile IPv6’’, ICACT 2010. In: 12th International Conference on Advanced
Communication Technology, Gangwon-Do, Korea (2010)
7. Lee, K.-W., Seo, W.-K., Choi, J.-I., Cho, Y.-Z.: A simple route optimization detection
scheme for multiple LMAs in PMIPv6 domain. In: Asia-Pacific Conference on Wearable
Computing Systems, APWCS 2010, Shenzhen, China, IEEE Computer Society, 17–18 Apr
2010
2 Route Optimization in Proxy Mobile IPv6 … 27

8. Rasem, A., St-Hilaire, M., Makaya, C.: A comparative analysis of predictive and reactive
mode of optimized PMIPv6. IWCMC, 722–727 (2012)
9. Lei, J., Fu, X.: Evaluating the benefits of introducing PMIPv6 for localized mobility
management. In: The International Wireless Communications and Mobile Computing
Conference 2008 (IWCMC 2008), Chinia, Greece, 6–8 Aug (2008)
10. Jun, T., Changxing, P., Bo, S.: Research on the localized routing mechanism for the PMIPv6.
J. Xidian University, 2012-01
11. Minoli, D.: Mobile Video with Mobile IPv6. Wiley, New York (2012)
12. Iapichino, G., Bonnet, C.: Experimental evaluation of proxy mobile IPv6: an implementation
perspective. In: IEEE Wireless Communications and Networking Conference (WCNC),
Sydney, Australia, 18–21 Apr 2010
13. Lee, H.-B., Han, Y.-H., Min, S.-G.: Network mobility support scheme on PMIPv6 networks.
In: International Journal of Computer Networks and Communication (IJCNC), vol. 2, Sept
2010
14. Melia, T., Bernardos, C.J, de la Oliva, A., Giust, F., Calderon, M.: IP flow mobility in
PMIPv6 based networks: solution design and experimental evaluation. In: Wireless Personal
Communications Journal, Springer, doi: 10.1007/s1277-011-0423-3, Oct 2011
15. Kempf, J., Leung, K., Roberts, P., Nishida, K., Giaretta, G., Liebsch, M.: Problem Statement
for Network-Based Localized Mobility Management, RFC 4830, IETF (2007)
16. Wahab, N.H.A.: Location-Assited Session Transfer for Real Time Applications, Master
Thesis. Universiti Teknologi Malaysia, Skudai (2010)
17. Nagamalai, D., Renault, E., Dhanuskodi, M.: Advances in parallel, distributed computing. In:
First International Conference on Parallel, Distributed Computing Technologies and
Applications, PDCTA 2011, Tirunelveli, Tamil Nadu, India, Proceedings Volume 203 of
Communications in Computer and Information Science, Published by Springer, 23–25 Sept
2011
18. Udugama, A., Iqbal, M.U., Toseef, U., Goerg, C., Fan, C., Schlaeger, M.: Evaluation of a
network based mobility management protocol: PMIPv6. In: IEEE 69th Vehicular Technology
Conference, VTC Spring (2009)
19. Lee, J.-H., Lim, H.-J., Chung, T.-M.: Preventing out-of-sequence packets on the route
optimization procedure in proxy mobile IPv6. In: 22nd International Conference on
Advanced Information Networking and Applications (aina 2008)
20. Chadchan, S.M., Akki, C.B.: 3GPP LTE/SAE: an overview. Int. J. Comput. Electr. Eng. 2(5)
(2010)
21. Nishida, K., Tanaka, I., Koshimiz, T.: Basic SAE management technology for realizing All-
IP network. NTT DOCOMO Tech. J. 11(3)
22. Ali-Yahiya, T.: Understanding Lte and Its Performance. Springer, Berlin (2011)
23. Abed, G.A., Ismail, M., Jumari, K.: Modeling and performance evaluation of LTE networks
with different TCP variants. Int. J. Inf. Commun. Eng. 6, 4 (2010)
24. Kim, J.H., Haw, R., Hong, C.S.: Development of PMIPv6 based 6LoWPAN sensor node
mobility scheme. In: The third AsiaFI Winter School, Seoul National University, Seoul,
Korea, 27 Feb 2010
25. Idserda, K.: Simultaneous binding proxy mobile IPv6. Master’s Thesis, University of Twente,
Enschede, The Netherlands (2008)
26. PMIPv6 Open Source, http://www.openairinterface.org
27. Soto, I.: Universidad Politécnica de Madrid; Carlos J. Bernardos, and María Calderón,
Universidad Carlos III de Madrid; and Telemaco Melia, Alcatel Lucent Bell Labs, PMIPv6:
A network-based localized mobility management solution. Internet Protoc. J. 13(3)
28. Wahab, N.H.A., Latiff, L.A., Ariffin, S.H.S., Fisal, N.: Development of proxy mobile IPv6
using flat domain model test-bed. In: 7th International Conference on Computing and
Convergence Technology ICCCT2012 (ICCIT, ICEI and ICACT), Seoul, Korea, 3–5 Dec 2012
29. Zhang, X., Zhou, X.: Lte-Advanced Air Interface Technology. CRC Press, Bosa Roca (2012)
30. Lucent, A.: Strategic White Paper
Chapter 3
Polytetrafluoroethylene Glass Microfiber
Reinforced Slotted Patch Antenna
for Satellite Band Applications

M. Samsuzzaman, T. Islam and M.R.I. Faruque

Abstract In this paper, a new configuration of triangular and diamond slotted

patch shape microstrip fed low-profile antenna is introduced on Polytetrafluoro-
ethylene glass microfiber reinforced material substrate for satellite applications.
The antenna is composed of a rectangular shape patch slotted with eight triangles
and two diamonds and elliptical slotted ground plane. The rectangular shape patch
is obtained by cutting two diamond slots in the middle of the rectangular patch and
six triangular slots on the left and right side of the patch, and two triangular slots in
the up and down side of the patch. The slotted radiating patch, the elliptical slot
ground and microstrip fed allow widening the matching bandwidth. Using a finite
element based simulator, a parametric investigation was performed for the opti-
mization. The proposed antenna offers a—10 dB impedance bandwidth of 1,020
and 1,350 MHz, respectively, and relatively stable and omnidirectional radiation
patterns in the matching band.

3.1 Introduction

In modern wireless communication system, wide and multiband antennas demand

is increasing due to support multi user and to provide more information with
higher data transmitting and receiving rates. Among different kinds of antennas,

M. Samsuzzaman (&)
T. Islam

Department of Electrical Electronic and Systems Engineering, Universiti Kebangsaan,
Kebangsaan, Malaysia
e-mail: [email protected]
T. Islam
e-mail: [email protected]
M.R.I. Faruque
Centre of Space Science, Universiti Kebangsaan, Kebangsaan, Malaysia
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 29

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_3
30 M. Samsuzzaman et al.

microstrip antennas are one of the most prominent structures due to their light
weight, compatibility, low-profile, ease of fabrication, multifrequency capability
and low cost. Compared to conventional three different types of antennas planar
microstrip antennas printed on small pieces of printed circuit board (PCB)
becomes familiar with recent wireless communication. Because it can be easily
embedded into wireless devices or integrated with other radio frequency (RF)
circuitry. Generally, a planar structure can be used to minimize the volumetric
dimension of a wide band antenna by replacing three dimensional radiation ele-
ments with their planar design [1].
Various types of antennas have already been designed for wide and multi-band
applications. A variety of dielectric materials have been used for designing and
prototyping of these antennas. Basically, a dielectric material chosen for the design
of wideband antennas is preferably needed to feature a higher permittivity and
lower dissipation factor [2]. Materials with higher dielectric constant have higher
capability to store charge and produce larger electromagnetic fields but limited
isolation between conductors. On the other hand, material with lower permittivity
are good insulators for lower frequency signals requiring high isolation in densely
packed circuits such as mobile and satellite communications [3, 4]. The search for
latest improved materials that could exchange the aluminum or other conductors is
an important task for many applications in the wireless industry. Size and weight
reduction, more tolerance to fatigue and easiness of manufacturing for complex
structures are some advantages of using such materials. Such kinds of character-
istics can lead to the operational cost reduction and performance intensification.
The woven glass fabric with epoxy resin (FR4) composite is a successful and
popular example of such materials in many applications. Several studies have been
made in recent years to explore this material performance for antenna develop-
ment. Moreover, by using a material with higher permittivity a compact antenna
can be designed that is capable of achieving very wide or multiple operating band
[4–8]. Ceramic-polytetrafluoroethylene composite material-based miniaturized
split-ring multiband patch antenna was designed [5]. The proposed antenna
obtained operating bandwidths (reflection coefficient \-10 dB) range from 5.0 to
6.5 GHz, 9.1 to 9.6 GHz, and 10.7 to 11 GHz. However, the antenna was designed
on high dialect and costly substrate. A miniaturized modified circular patch
antenna was designed on ceramic-polytetrafluoroethylene (PTFE) composite
material with dimension 0.22k 9 0.29k 9 0.23k; the proposed antenna achieved
multi-band characteristics. However, the antenna failed to fulfil the requirement of
Wi-Fi-WiMAX applications [6]. The dual-band operation of a microstrip patch
antenna (20 mm 9 20 mm) for Ku- and K-bands has been presented [9] on glass
microfiber reinforced PTFE. A compact square loop multiband patch antenna
design on high dielectric ceramic composite material was proposed [10]. Though
the reported antenna achieved multiband but impedance bandwidth was low and
substrate cost is high compare to epoxy resin fiber. A printed wide-slot antenna
designed and prototyping on available low-cost polymer resin composite material
fed by a microstrip line with a rotated square slot for bandwidth enhancement and
defected ground structure for gain enhancement [11]. A wideband pentagon shape
3 Polytetrafluoroethylene Glass Microfiber Reinforced … 31

Fig. 3.1 Proposed antenna W

geometry layout

Patch

Ground

Feed line

microstrip slot antenna was designed on epoxy resin composite material [12]. The
proposed design antenna obtained 124 % impedance bandwidth, but its use in
portable communication devices was limited due to large ground plane.
This article presents a simple modified rectangular shape printed antenna that
exhibits dual band characteristics. An elliptical slot is etched on the ground plane
for more coupling with slotted patch radiator. The simulated reflection coefficients
and far field radiation pattern are presented. The simulation and analysis of the
presented antenna are performed by using Finite Element Based high frequency
structural simulator HFSS.

3.2 Antenna Design, Architecture and Optimization

The proposed, modified, rectangular-shaped, dual-band antenna is illustrated in

Fig. 3.1. For the design studied here, the radiating element and feeding line are
printed on the same side of the microstrip patch substrate, which has a thickness of
1.575 mm, a dielectric constant of er = 2.33 and tand = 0.002, while the other side
is the ground plane of the antenna. The microstrip transmission line is used to feed
the antenna. The basic antenna structure begins as a rectangular-shaped patch. The
radiating patch is created by cutting two diamond-shaped slots and two equatorial
triangular slots in the middle as well as three triangular slots on both sides of the
32 M. Samsuzzaman et al.

rectangular patch. An elliptical slot is etched on the ground plane of the proposed
antenna. Different arm lengths of different slots and widths have been optimized to
obtain optimum outputs. The dual-frequency, wide-impedance matching capability
is enabled via the electromagnetic coupling effect of the ground plane to the feed
line and the radiating patch. The design parameters are L = 38 mm, W = 30 mm.

3.3 Properties and Performance Analysis of Different

Dielectrics Substrate

Four different substrate materials, the properties of which are summarized below
are used in the simulation studies presented herein.
The FR4 substrate material consists of an epoxy matrix reinforced by woven
glass. This composition of epoxy resin and fiber glass varies in thickness and is
direction dependent. One of the attractive properties of polymer resin composites
is that they can be shaped and reshaped without losing their material properties.
The composition ratio of the material is 60 % fiber glass and 40 % epoxy resin.
Taconic TLC laminates are engineered to provide a cost effective substrate
suitable for a wide range of microwave applications. TLC laminates offer superior
electrical performance compared to thermoset laminates (e.g. FR-4, PPO, BT, and
Polyimide and cyanate ester). TLC’s construction also provides exceptional
mechanical stability.
NeltecNX 9240 has superior mechanical and electrical performance, making it
the material of choice for low-loss, high-frequency applications, such as wireless
communications. This material is specially designed for very low-loss antenna
applications. The enhanced N9000 materials reduce passive intermodulation issues
in antenna and high-power designs.
RT/Duroid 5870-filled PTFE composites are designed for exacting strip-line
and microstrip circuit applications. The unique filler results in a low density,
lightweight material that is beneficial for high-performance, weight-sensitive
applications. The very low dielectric constants of such RT/Duroid 5870 laminates
are uniform from panel to panel and are constant over a wide frequency range.
This substrate material is a high-frequency laminate and PTFE (Polytetrafluoro-
ethylene) composite amplified using glass microfibers. Table 3.1 represents the
different substrate dielectric properties.
The effect of the different substrate material on the reflection coefficient of the
proposed antenna is shown in Fig. 3.2. It can be clearly observed that the proposed
antenna offers a wider bandwidth and an adequate return loss value compared with
other reported materials. Although the antenna with epoxy resin fiber material
substrate provides a lower return loss value due to its higher dielectric constant,
but bandwidth is narrower, and the desired resonance is shifted. Moreover, the
epoxy resin fiber loss tangent is higher compared to others. Others two material
3 Polytetrafluoroethylene Glass Microfiber Reinforced … 33

Table 3.1 Properties of different substratea

Parameters FR4_epoxy Rogers RT/duroid 5870 Taconic Neltec NY
Dielectric constant 4.6 2.33 2.4 2.4
Loss tangent 0.02 0.0012 0.0016 0.0016
Water absorption (%) \0.25 0.02 \0.02 \0.02
Tensile strength \310 MPa 450 MPa – –
Volume resistivity 8 9 107 2 9 107 1 9 107 1 9 107
(MX.cm)
Surface Resistivity (MX) 2 9 105 3 9 107 1 9 107 1 9 107
Breakdown voltage (kV) 55 [60 – [60
Peel Strength (N/mm) 9 5.5 12 12
a
All information collected from the different substrate data sheet from their website

Fig. 3.2 Effect of reflection

coefficient of different
substrate material

substrate material have achieved resonance, but low performance. The dielectric
properties of the substrate material are tabulated in Table 3.1.
The far-field radiation patterns for the proposed wideband dual-frequency
slotted antenna are also examined. Figure 3.3 shows the measured co—and
cross—polarized radiation patterns including the horizontal (E plane) and vertical
(H plane) polarization pattern for the antenna at lower band of 8.25 GHz and upper
band 9.95 GHz. These results represent that the patterns are stable across the
operating matching band. One can note that there is small asymmetry in the both E
and H plane patterns which is due to the radiation of the microstrip line and the
patch. Omnidirectional radiation pattern is observed in both E and H plane.
34 M. Samsuzzaman et al.

(a)

(b)

E plane H plane

Co polar
Cross polar

Fig. 3.3 E- and H-plane radiation patterns of the proposed antenna at a 8.25 GHz b 9.95 GHz

3.4 Conclusion

A novel dual band configuration of triangular and diamond slotted patch with
slotted ground plane antenna is proposed. The antenna is designed on Polytetra-
fluoroethylene glass microfibre reinforced substrate. Compared to the other
reported material, the proposed material antenna exhibits better performance in
terms of impedance bandwidth, and stable radiation pattern. In this design a large
bandwidth has been obtained by adding different slot in the patch and the ground
plane. In addition, a parametric investigation was carried out to optimize the
proposed design. Simulated results show that the proposed antenna can offer an
impedance bandwidth of 1,020 and 1,350 MHz, respectively. The proposed shape
dual band antenna has a very simple structure, which makes the design simpler and
fabrications easier, and is very suitable for applications in the access points of
satellite application.
3 Polytetrafluoroethylene Glass Microfiber Reinforced … 35

References

1. Balanis, C.A.: Antenna theory: analysis and design. Wiley-Interscience, Hoboken (2012)
2. OOHIRA, K.: Development of an antenna material based on rubber that has flexibility and
high impact resistance. NTN Tech. Rev. 76(Special Issue), 58–63 (2008)
3. Art Aguayo, S.: Analyzing advances in antenna Materials. Antenna Syst. Technol. 14–15
(2010)
4. Samsuzzaman, M., Islam, M.T., Mandeep, J.S.: Parametric analysis of a glass-micro fibre-
reinforced PTFE material, multiband, patch-structure antenna for satellite applications.
Optoelectron. Adv. Mater. Rapid Commun. 7(9), 760–769 (2013)
5. Ullah, M.H., Islam, M.T., Mandeep, J., Misran, N.: Ceramic-polytetrafluoroethylene
composite material-based miniaturized split-ring patch antenna. Sci. Eng. Compos. Mater.
21(3)‘ 405-410 (2014)
6. Ullah, M.H., Islam, M.T.: Miniaturized modified circular patch monopole antenna on
ceramic-polytetrafluroethylene composite material substrate. J. Comput. Electron. 1–6 (2013)
7. Assis, R.R.d., Bianchi, I.: Analysis of microstrip antennas on carbon fiber composite material.
J. Microw. Optoelectron. Electromagn. Appl. 11, 154–161 (2012)
8. Samsuzzaman, M., Islam, T., Abd Rahman, N.H., Faruque, M.R.I., Mandeep, J.S.: Coplanar
waveguide fed compact wide circular-slotted antenna for Wi-Fi/WiMAX applications. Int.
J. Antennas Propag. 2014, 10 (2014)
9. Islam, M., Islam, M., Faruque, M.: Dual-band operation of a microstrip patch antenna on a
duroid 5870 substrate for ku-and k-bands. Sci. World J. 2013, 378420 (2013)
10. Ullah, M.H., Islam, M.T.: A compact square loop patch antenna on high dielectric ceramic–
PTFE composite material. Appl. Phys. A 113(1), 185–193 (2013)
11. Samsuzzaman, M., Islam, M.T., Mandeep, J.S., Misran, N.: Printed wide-slot antenna design
with bandwidth and gain enhancement on low-cost substrate. Sci. World J. 2014, 10 (2014)
12. Rajgopal, S.K., Sharma, S.K.: Investigations on ultrawideband pentagon shape microstrip slot
antenna for wireless communications. IEEE Trans. Antennas Propag 57(5), 1353–1359
(2009)
Chapter 4
2.4 GHz Circularly Polarized Microstrip
Antenna for RFID Application

Rudy Yuwono and Ronanobelta Syakura

Abstract Due to deployment of the wireless technology, the microstrip antenna

can be used for RFID application especially for tag antenna. The existed RFID tag
antenna usually use dipole model to get the maximum performance. However, the
dipole antenna which has linear polarization performance has disadvantages on
alignment that makes data detection on RFID can not be accurate. To reduce error
during data detection caused of misalignment between reader and transponder, the
tag antenna should have circularly polarized performance. The antenna was made
with microstrip-line fed on FR-4 Epoxy substrate and works covers 2.4 GHz of its
frequency range of 2 up to 3.3 GHz which S-parameter level below -9.54 dB and
the direction of the transmitted power omnidirectionally and has circular polari-
zation at frequency range of 2.39 until 2.46 GHz which has axial ratio below 3 dB.

Keywords Circularly polarized
RFID tag
Microstrip antenna

4.1 Introduction

The deployment of the wireless technologies is going rapidly especially in the

antenna that supports the wireless system to receive and transmit the information.
The antenna should be easy to fabricate, has various models and has proportional
size with the wireless system. The antennas type which has these criteria is the
microstrip antenna [1, 2, 7].

R. Yuwono (&)
R. Syakura

Electrical Engineering Department, University of Brawijaya MT, Haryono 167, Malang,
Jawa Timur, Indonesia
e-mail: [email protected]
R. Syakura
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 37

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_4
38 R. Yuwono and R. Syakura

Microstrip antenna can be applied in many wireless technologies. One of the

wireless technologies which applied the microstrip antenna is in Radio Frequency
Identification System (RFID), as the tag antenna that transmits the saved infor-
mation in the transponder into reader to detect the information [8–10]. Existed tag
antenna mostly used printed dipole as the model to get the maximum performance
during data detection [3, 12].
Usage of the printed dipole has disadvantages especially in alignment of
propagation. Printed dipole which has linear or elliptical polarization can cause
problem if the tag and reader get the misalignment [4, 13, 14]. The reader can not
detect the information accurately because of that misalignment. To solve the
problem, the tag antenna should be circularly polarized.

4.2 Antenna Design

The tag antenna is designed with the dimension of substrate 30 9 40 mm with the
microstrip line-fed. Microstrip-line fed method is chosen because it is easy to
fabricate and match impedance. For the shape of the microstrip-line chosen the L
shape to get orthogonal phase magnitude which can occur circular polarization
[5, 6].
The applied substrate material is FR-4 Epoxy that has dielectric constant (er)
4.4 with thickness 0.8 mm. Antenna models is shown on Fig. 4.1.

4.3 Result

The fabricated antenna was shown on Fig. 4.2.

The Observed Parameter antenna covers result of S-parameter (S11), directivity
pattern and axial ratio.

4.3.1 S11 Result

S11 which obtained the frequency of the antenna that can work shown on Fig. 4.3
Figure 4.3 Shows that the antenna can work covers frequency of 2.4 GHz in the
work frequency range of 2.2 until 3.3 GHz. The frequency range of the antenna
obtained from the S11 level below 9.54 dB which is the maximum tolerance level
of the antenna that can work [11].
4 2.4 GHz Circularly Polarized Microstrip Antenna … 39

(a) (b)
Ws Wg1

Lg2

Ls Lf2 Lg1 Lp1

Wm
Rc
Rp

Lf1 Wi

Wm Wg2 Lp2

Fig. 4.1 The antenna design; X = the excitation port; a front view b rear view, with 0.8 mm
FR-4 substrate (er = 4.4) Ls = 40 mm; Ws = 30 mm; microstrip-line fed with the dimension of
Wm = 1 mm; Wi = 2 mm; Rc = 5 mm; Lf1 = 12.5 mm; Lf2 = 10.2 mm; ground plane with the
dimension of Lg1 = 40 mm; Wg1 = 30 mm; Lg2 = 20 mm; Wg2 = 15.5 mm; Lp1 = 5 mm;
Lp2 = 4.1 mm; Rp = 10 mm

Fig. 4.2 The fabricated antenna; a front view; b rear view

40 R. Yuwono and R. Syakura

Fig. 4.3 S11 parameter

versus frequency

Fig. 4.4 Polar diagram of

directivity pattern of the
antenna at frequency of
2.4 GHz

4.3.2 Directivity Result

The directivity result of the antenna at frequency of 2.4 GHz shown on polar
diagram as directivity pattern which is shown at Fig. 4.4
Figure 4.4 shows which the antenna has directivity pattern of omnidirectional.
It is obtained from the directivity pattern that can transmit the power in all
direction.

4.3.3 Axial Ratio Result

The result of the axial ratio shown on Fig. 4.5

Figure 4.5 show that the antenna has circular polarization at frequency range of
2.39 up to 2.46 GHz obtained from the axial ratio level that has value below 3 dB.
With that performance the antenna can be used for RFID tag with circular
polarization.
4 2.4 GHz Circularly Polarized Microstrip Antenna … 41

Fig. 4.5 Axial ratio with the

function of frequency

4.4 Conclusion

The RFID tag antenna which made with microstrip-line fed method on 0.8 mm
FR-4 epoxy with the dimension of the substrate 30 9 40 mm has performance can
work covers frequency of 2.4 GHz in the range of frequency 2 up to3.3 GHz with
the directivity pattern omnidirectional. The antenna also has circular polarization
which is obtained from the axial ratio at frequency of 2.39 up to 2.45 GHz which
is used for RFID application.

Acknowledgments The Authors would like to express sincere gratitude to University of

Brawijaya for their support under contract number SK DEKAN No 26/UN10.6/PG/2013, June 3,
2013

References

1. Balanis, C.A: Antenna Theory Analysis and Design. Wiley, New York (2005)
2. Bhartia, P., Bahl, Inder, Garg, R.: Ittipiboon: Microstrip Antenna Design Handbook. Artech
House, Boston (2001)
3. Deavours, D.: UHF RFID Antennas. In: Bolic, M., Simplot-Ryl, D., Stojmenovic, I. (eds.)
RFID Systems Research Trends and Challenges, p. 68. Wiley, New York (2010)
4. Finkenzeller, K.: Klaus: RFID Handbook, 2nd edn. Wiley, London (2003)
5. Joseph, R., Fukusako, T.: Circularly polarized broadband antenna with circular slot on
circular ground plane. Prog. Electromagnet. Res. C 26, 205–217 (2012)
6. Kumar, G., Ray, K.P.: Broadband Microstrip Antennas. Artech House, Boston (2003)
7. Preradrovic, S., Karmakar, N.C.: Design of fully printable planar chipless RFID transponder
with 35-bit data capacity. In: Proceeding of the 39th European Microwave Conference (2009)
8. Preradrovic, S., Karmakar, N.C.: Design of chipless RFID tag for operation on flexible
laminates. In: IEEE Antennas and Propagation Letters, vol. 9 (2010)
9. Preradrovic, S., Karmakar, N.C.: Multiresonator-Based Chipless RFID-Barcode of the
Future. Springer, New York (2012)
10. Nakar, P.S.: Design of a compact microstrip patch antenna for use in wireless/cellular
devices. Electronic Theses, Treatises and Dissertations Paper 2790, http://diginole.lib.fsu.
edu/etd/2790, (2004)
42 R. Yuwono and R. Syakura

11. Yuwono, R., Baskoro, A., Erfan, A.D.: Design of circular patch microstrip antenna for
2.4 GHz RFID applications. In: Lecture Notes in Electrical Engineering, vol 235, pp. 21–28.
Springer, Heidelberg (2013)
12. Yuwono, R., Wahyu, D.A., Muhammad, F.E.P.: Design of circular patch microstrip antenna
with egg slot for 2.4 GHz ultra-wideband radio frequency identification (UWB RFID) tag
applications. Applied Mechanics and Materials. vol 513–517 Trans Tech Publications
Zurich-Durnten Switzerland 3414–3418 (2014)
13. Ruengwaree, A.; Yuwono, R.; Kompa, G.: A noble rugby-ball antenna for pulse radiation.
In: The European Conference on Wireless Technology, pp. 455–458. IEEE Conference
Publications (2005)
14. Yuwono, R., Wendaria, P.: Design of circular patch microstrip antenna with rugby ball slot
for ultra wideband applications. In: The 2nd International Conference on Radar, Antenna,
Microwave, Electronic and Telecomunication (ICRAMET) pp. 130–132 (2013)
Chapter 5
On Understanding Centrality in Directed
Citation Graph

Ismael A. Jannoud and Mohammad Z. Masoud

Abstract Modeling complex networks as directed/undirected graphs is considered

one of the most common methods in network science. Citation graph is a directed
graph of scientific published papers. This graph has been studied massively in the past
decade. Citation graph can be utilized to study relationships between authors and
papers. It can be used to study the characteristics of citation network to demonstrate
the growth model, graph type and to predicted hot new topics. In this paper, we
attempt to study the relationship between popularity of a paper and the publication
date. The purpose of this study is to demonstrate the relation between paper quality
and hot topics. Betweenness metric has been used to measure the popularity of a
published paper. Moreover, a comparison between betweenness and citation count
(node degree) has been conducted to show that papers may have a small citation
count, however, they may have a great impact in research field. We have generated a
directed citation graph by crawling paper information from CiteSeerx. Our study
shows that date of publication is important to write a popular paper. However, high
quality papers get opportunity to be popular regardless the date of publication.

Keywords Node degree  Citation graph  Graph metrics  Cluster coefficient 

Betweenness

5.1 Introduction

Citation is the process of quotation information, results and conclusions from

book, paper, or websites in a research work. Scientific research papers (SRP)
contain from few references to few hundreds. Citation is one of the metrics that

I.A. Jannoud (&)  M.Z. Masoud

Al-Zaytoonah University of Jordan, Amman, Jordan
e-mail: [email protected]; [email protected]
M.Z. Masoud
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 43

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_5
44 I.A. Jannoud and M.Z. Masoud

used to measure the popularity and the importance of SRP. Moreover, citation may
be used to find the roots of science fields. Citation process generates a complex
directed graph, which is called citation graph or citation network. In this graph,
nodes (papers) are connected with directed edges that begin from newly published
papers and point to old papers.
Citation in these networks are the edges, papers are the nodes. This definition
converted these networks into complex directed graphs. The emerged of network
science, its applications and tools increased the possibilities in studying complex
networks. Citation network is one of the complex networks that have been studied
heavily in the past decade. Paper popularity, author ranking, author relations and
community studies are some examples of the vast range of studies that have been
conducted in this field.
Mapping graph parameters and properties into physical meanings that reflects
useful meaning is the niche of network science. This mapping helped in predicting
special phenomenon of citation networks, such as, power-law distribution, graph
type and the expanding models of these networks. However, many graph param-
eters have not been studied and mapped in citation networks, such as, betweenness
and closeness.
In this work, paper importance will be measured. Paper centrality in the citation
network will be extracted. Timing of papers and their centrality will be examined.
The purpose of this work is to demonstrate the important of hot topics time and the
position of papers in citation networks. Graph betweenness will be utilized to
measure the centrality of papers. To conduct our experiment massive amount of
papers in a certain field must be harvested. To facilitate our experiment, a web
crawler has been implemented to collect paper information from CiteSeerx web-
site. To this end, we have generated a direct graph from the harvested data. We
seek to answer the following questions using graph analysis:
• What is the meaning of centrality in citation network?
• What the relationship is between hot topics timing and paper centrality?
• Can researchers write a paper with high citation value without considering the
hot topic?
The rest of this paper is organized as follows; this section ends with the related
works that have been conducted in this area. Section 5.3 introduces the perfor-
mance metrics. Section 5.4 provides a description of the experiment that has been
implemented. This section ends with the results and discussion. Finally, Sect. 5.6
concludes this work.

5.2 Related Work

Studying real network as graphs inspired researchers over the years. They heavily
studied social, information, technological, and biological graphs [1]. These studies
gain researchers more insights of how these networks may evolve and how bugs,
errors and diseases may separate.
5 On Understanding Centrality in Directed Citation Graph 45

Many types and kinds of networks have been studied as graphs. For example, in
[2, 3] it has been reported that the World Wide Web (WWW) follows an expo-
nential degree distribution with more than 269 thousands nodes and around 1 and
half million edges. This study provided that the WWW is a small world graph. In
the measurement works [4], authors attempted to generate Internet graph to study
its properties. They have produced an undirected graph with 10 thousands nodes
and 31 thousands edges. A 0.035 global cluster coefficient value has been com-
puted. These values have been computed over the years again to show the
development of the Internet [5].
A third type of networks have been studied in [6], the author of this study
generated a software-classes directed-graph. With 1,377 nodes and 2,213 edges,
the author studied the properties of this small graph. The author reported a mean
node-node distance with 1.5 hops. In addition, global clustering coefficient and
degree correlation coefficient were computed. This study showed that software
classes don’t follow the small world phenomenon. Furthermore, in the works of
[7, 8], authors generated a small undirected graph based on harvested data to
simulate P2P network. A graph of 880 nodes and 1,296 edges has been imple-
mented and studied. The author found that the average shortest path in this graph is
4.2 hops and the global clustering coefficient is 0.012. Unfortunately, their
implemented graph was too small to mimic a swarm of P2P networks. In [9],
authors attempted to study human’s neural networks as a directed graph. They
have constructed a directed graph with 307 neural and 2,359 edges between them.
Subsequently, they studied the properties of this graph. They reported a global
clustering coefficient of 0.18 and an average shortest path value of 3.97. These
values demonstrated that human’s neural network is a small world graph.
Finally, Citation networks have not been forgotten by researchers. In [5], the
author studied the citation network as a directed-graph. The author computed node
number and vertexes. However, the cluster coefficient and degree correlation were
not computed. These results made it hard to predict node distribution in citation
network. In [10] a citation graph has been generated to study the impact of
co-authors on the popularity of papers. In addition, co-authors are connected if
they have shard work. Our work in the citation network differs from these work in
two points. First, we utilized a new graph parameter to study the centrality and
popularity of a paper in a certain field. Second, time has been utilized to dem-
onstrate the relation between time and popularity.

5.3 Performance Metric

This section introduces the metrics that have been utilized from graph theory.
Graph centrality through betweeenness and node degree will be utilized. Graph
centrality is a metric to evaluate the importance of any node in a graph. Many
methods are utilized to measure the centrality of a node. Node degree, eigenvector,
closeness and betweenness are the most common methods to calculate the
46 I.A. Jannoud and M.Z. Masoud

centrality of nodes. In citation graph, centrality metric may be used to extract the
core nodes of the graph. These core nodes may be introduced as the root of a
scientific field. Any research in a scientific area starts from an idea in the past.
However, there is a time when this field emerges as a hot topic. Core nodes are the
nodes that convert a topic into a hot topic.
In this work, we have utilized the betweenness and node degree of the nodes in
the citation graph to measure graph centrality. In the next section node degree and
betweenness will be defined.

5.3.1 Betweenness and Node Degree Centrality

Betweenness of a node is defined as total number of paths that flow through this
node [11]. Equation 5.1 shows the mathematical method of calculating betweeness
of a node. In a directed graph such as citation graph, betweenness may be used to
demonstrate the importance of a scientific paper.
X
GðvÞ ¼ Qsi ðvÞ=Qsi ð5:1Þ

Node degree is defined as the total number of incoming and outgoing connections
that connects a node with other nodes.
Node degree may be used as centrality metric. However, some nodes may be
important and they may have less degree than other nodes. This fact is emerged in
betweenness centrality more than node degree centrality. We believe that
betweenness centrality results are more accurate for two reasons. First, in scientific
research timing is an important factor. The important of a paper depends on the
time. Old popular paper will not get more citation after a long time. Rich will be
richer if they are not too old unlike other graphs, such as, the AS graph. Second, an
important paper may be cited by little number of papers that may become more
and more popular. These two facts can be found crystal clear in the betweenness
centrality. Figure 5.1 shows an example of the differences between node degree
and node betweenness. Nodes 1–7 have node degree more than node 8. However,
node 8 is more important since it connects between this node and nodes: 9, 10.

5.4 The Experiment

A web-crawler has been implemented to harvest scientific paper information from

CiteSeerx website. CiteSeerx is a scientific literature digital library. It focus primary
on scientific research papers in the fields of computer and information science. It has
many functionalities and methods that allow it to index postscript and PDF articles.
Our web-crawler starts by retrieving the search results of a specific topic. We have
used ad hoc as our main topic to generate our graph. The maximum allowed retrieved
results are 500 papers. Subsequently, the crawler uses the retrieved 500 paper names,
5 On Understanding Centrality in Directed Citation Graph 47

Fig. 5.1 Example of the

differences between node
degree and node betweenness
centrality

Fig. 5.2 Example of the

generated direct citation
graph

which are the seed list of the crawler, to retrieve their cited paper. Each cited retrieved
paper is added into the list to retrieve its citations and so on. The harvesting process
started in 15th of January and continued for 10 days. The crawling process stopped
when encountered papers that have no information in the CiteSeerx web-site.
The harvested data of the conducted experiment has been utilized to generate a
directed citation graph. Figure 5.2 shows an example of the generated direct
citation graph.
Table 5.1 lists the properties of this graph. We can notes from the graph that the
clustering coefficient of this graph is too small comparing with other graphs, such as,
the AS graph. In addition, the average path length is longer since it’s a directed graph.

5.5 Results

Figure 5.3 shows the CDF of node betweenness value. From this figure we can
observe that more than 88 % of the nodes have a betweenness value less than 1. In
addition we can observe that less than 4 % of the nodes have a betweeness value
48 I.A. Jannoud and M.Z. Masoud

Table 5.1 The constructed Property Citation graph

citation graphs properties
Number of nodes 91,211
Number of links 221,677
Average node degree 2.43
Cluster coefficient 0.067
Average path length 7.4
Graph diameter 24

Fig. 5.3 The CDF of node

betweenness value

Fig. 5.4 The relation

between publication year and
the betweenness value

less than 100. More over we can observe that a tiny number of nodes (0.02 %)
have a vast betweeness value. These nodes are the central of our citation graph.
Figure 5.4 shows a relation between publication year and the betweenness
value. To compute this figure, 10 K papers have been excluded from the result.
These papers have no publication year information in CiteSeerx web-site. The rest
5 On Understanding Centrality in Directed Citation Graph 49

Fig. 5.5 The relationship

between betweenness value
and node degree

Fig. 5.6 The CDF of node

citation count

of the papers (more than 8 K) have been used to generate this relation. We can
observe that the figure is a random figure. We notice that for each year of pub-
lication, the betweenness value varies. In addition, we observed that the highest
betweenness value was in 2005. However, the papers in ad hoc started before this
year. This figure proves that high quality papers we be in the core of any field
regardless of the year of publication.
Figure 5.5 shows the relation between betweenness value and node degree or
citation count. This figure shows that the relation between betweenness and cita-
tion count is fuzzy. With the increased number of citation count, betweenness
value increases.
However, some node with high citation value has small betweenness. In
addition, we can observe that the node with highest betweenness value is not the
50 I.A. Jannoud and M.Z. Masoud

node with the highest citation count. The figure has a fuzzy slop to the right. This
figure shows the differences between betweenness centrality and node degree
centrality. This figure demonstrates that some nodes with less citation count may
be more important than others. Betweenness may be used to show this fact.
Finally, Fig. 5.6 shows the CDF of node citation count (node degree). We can
observe the similarity between the CDF of betweenness value in figure and this
figure. However, unlike the betweenness CDF that has less than 10 % of the nodes
with significant betweenness value, we can observe that more than 33 % of the
nodes have significant node degree (more than 10). This fact demonstrate how
betweenness centrality can be used to filter a massive number of nodes to obtain
the most central and important once in a graph.

5.6 Conclusion

Researches utilized the citation graph to study and introduce the publication net-
works for various reasons. The last decade witnessed a heavy study of the prop-
erties of this graph. In this work, we have studied the impact of the publication
date on the popularity of published papers. Three main graph properties have been
utilized to study the relationship between popularity and hot topic (date of pub-
lication). Betweenness, global and local cluster coefficients and node degree dis-
tributions has been computed. To this end we have generated a citation graph by
crawling paper information of ad hoc (research field) from CiteSeerx web-site. Our
results demonstrated three main points. First, the probability of getting high
citation count increases with the time or the date of publication. Second, there is a
probability that a high quality paper will get a good citation count even if it is late
in the field. Finally, the data shows a fuzzy slope in the graphs. This uncertainty
requires more experiments.

References

1. Newman, M.E.J.: The structure and function of complex networks. SIAM Rev. 45, 167–256
(2003)
2. Barabasi, A.-L., Albert, R., Jeong, H.: Scale-free characteristics of random networks: the
topology of the world-wide web. Phys. A 281, 69–77 (2000)
3. Albert, R., Jeong, H., Barabasi, A.-L.: Diameter of the world-wide web. Nature 40, 130–131
(1999)
4. Dorogovtsev, S.N., Mendes, J.F.F.: Language as an evolving word web. Proc. R. Soc. Lond.
Ser. B 268, 2603–2606 (1999)
5. Masoud, M., Hei, X., Cheng, W.: A graph-theoretic study of the flattening internet as
topology. In: IEEE ICON, December 2013 (2013)
6. Valverde, S., Cancho, R.F., Sole, R.V.: Scale-free networks from optimal design. Europhys
Lett. 60, 512–517 (2002)
5 On Understanding Centrality in Directed Citation Graph 51

7. Ripeanu, M., Foster, I., Iamnitchi, A.: Mapping the gnutella network: properties of large-
scale peer to-peer systems and implications for system design. IEEE Internet Comput. 6(1)
(2002)
8. Adamic, L.A., Lukose, R.M., Puniyani, A.R., A. Huberman, B.: Search in power-law
networks. Phys. Rev. E 64, 046135 (2001)
9. White, J.G., Southgate, J.N.T.E., Brenner, S.: The structure of the nervous system of the
nematode in C. Elegans. Philos. Trans. R. Soc. Lond. 314, 1–340 (1986)
10. Healy, K.: A co-citation network for philosophy, June 2013 (2013)
11. Lewis, T.G.: Network science theory and practice. Wiley, Hoboken (2008)
Chapter 6
Channel Capacity of Indoor MIMO
Systems in the Presence of Spatial
Diversity

M. Senon, M.N. Husain, A.R. Othman, M.Z.A. Aziz,

K.A.A. Rashid, M.M. Saad, M.T. Ahmad and J.S. Hamidon

Abstract The capacity of Multiple Input Multiple Output (MIMO) systems has
received much attention in recent years. This paper analyze the capacities of MIMO
channel model for indoor propagation with increasing the distance between transmit
and receive antenna through simulation and measurement. A spatial diversity
method is employed during measurement and simulation process. The investigation
on the channel capacity for various distance and spacing of both transmitter and
receiver antenna have been done. The investigations of channel capacity are
included with difference distance between transmitter and receiver sides and dif-
ferent in element antenna spacing. For the simulation, the path loss for the free space
and physical effect are been considered. The 2 9 2 rectangular microstrip patch
array antenna is used in order to characterize channel parameter at 2.4 GHz
operating frequency. The system measurement was been conducted in UTeM
Microwave Laboratory, according to the real situation in indoor environment.

6.1 Introduction

Today, demand for high data rate and channel bandwidth is increasing due to the
modern and future application requirement in wireless communication systems.
The Multiple-input-multiple-output (MIMO) system was invented to make the
multipath propagation mechanism as an advantage in order to increase the channel
capacity. This system was characterized of multiple antennas that used at the
transmitter and receiver sides and can increase the channel capacity (b/s/Hz)
without increased the bandwidth and transmit power [1, 2]. The used of diversity
was to increase the probability at the receiver end where at least one of the signals

M. Senon (&)  M.N. Husain  A.R. Othman  M.Z.A. Aziz  K.A.A. Rashid
M.M. Saad  M.T. Ahmad  J.S. Hamidon
Faculty of Electronics and Computer Engineering, Universiti Teknikal Malaysia Melaka,
Melaka, Malaysia
e-mail: [email protected]; [email protected]

 Springer International Publishing Switzerland 2015 53

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_6
54 M. Senon et al.

were received correctly [3]. The spatial diversity was one of the diversity tech-
niques [4–7]. It can be done by space apart between the antennas but when used at
the limited volume or space.
In WLAN technology, the signal will propagate through numerous paths which
effected from the indoor environment. This phenomenon is call multipath propa-
gation. So the MIMO channel system is introduced to solve this problem by
exploiting the richness of multipath propagation. Additionally, the use of multiple
antennas at both the transmitter and receiver provides significant increase in
wireless channel capacity.
The MIMO channel models can be divided into the non-physical and
physical models [8–10], in order to characterize the channel performance, the
statistical model is used to find the channel capacity for every scenario. The
MIMO channel model can be characterized by modelling the channel with con-
sideration of physical parameter such as distance and scattering for every scenario.
This paper will discuss and analyzed the simulation of capacity effect to the
wireless MIMO communication channel model for different spatial diversity
is applied at the both sides of transmitter and receiver. Then the measured chan-
nel capacity is compared to channel capacity obtain from Kronecker and
Weischselberger model.

6.2 Channel Model and Channel Capacity

6.2.1 MIMO Channel Model for LOS Scenario

Wireless propagation is dominated by the daily changes by environment features

such as reflection, diffraction and transmission depending on the location of
transmitter and receiver.
In this project, reflection is avoided because LOS is considered as a major
signal. Since no reflection between n-th transmitter and m-th receiver, the channel
coefficient h, only consists a direct path between these antennas. Because of the
operation in free-space condition, the power receive of Friis free-space equation is
considered [11].

Pt Gt Gr k2
Pr ¼ ð6:1Þ
ð4pÞ2 rmn
2 L

Where Pt is the transmitted power, Gt and Gr are transmitter and receiver

antenna gain and k is the wavelength in meters. The system loss, L, such as cable
loss or antenna efficiency are not related to free-space propagation.
6 Channel Capacity of Indoor MIMO Systems … 55

6.2.2 System Configuration

In this paper, parameters of MIMO channel model identified with construct a

different configuration in indoor environment. Figure 6.1 shows the antenna
configuration and measurement setup. The development of algorithm is starting by
calculating pathloss of the system. Then the correlation coefficient and eigen
analysis are calculated sequentially to find the channel capacity of MIMO wireless
communication system.
The Matlab simulation tool is use to compare MIMO channel models based on
the theoretical data and measured data. The measurement campaign is done to
collect the data required to verify the MIMO channel model (Kronecker model and
Weichelberger model) and to estimate the model parameters that charaterize dif-
ferent configuration.
For this project, fixed wireless system is more to line-of-sight environment
while the distance (d) is assumed much higher than antenna spacing (l). From
geometrical arrangement, the different path length (rn,m) for LOS arrangement
between transmitter nth and receiver mth is calculate such as Eq. 6.2 [12, 13].
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rn;m ¼ d2 þ ðlðn  mÞÞ2 ð6:2Þ

For the second scenario, both transmitter and receiver antenna is placed with
different antenna spacing (l). Thus, the path length (rn,m) is depending on the
shifting of the antenna spacing. The paths length (rn,m) for different antenna
spacing at transmitter (lm) and receiver (ln) is given by:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rn;m ¼ d2 þ ðlm  ln Þ2 ð6:3Þ

The third configuration is a multiple altitude antenna configuration which is the

uniform antenna spacing is placed at both transmitters and receiver antenna front
end but with different altitude.

6.2.3 MIMO Channel Capacity

The data collected from the measurement are analyzed for every configuration.
Equation (6.4) show the MIMO channel matrix H where N and M representing the
number of transmitting and receiving antennas [14].
2 3
q11 q12 q13 q14
6 q21 q22 q23 q24 7
H¼6
4 q31
7 ð6:4Þ
q32 q33 q34 5
q41 q42 q43 q44 MxN
56 M. Senon et al.

Input 1 Output 1
lm ln
Output 2 Spectrum
Input 2 analyser
Output 3
Input 3
d Output 4
Input 4

Fig. 6.1 Antenna configuration

The channel capacity is calculate by using Eq. (6.5) [15].

X
i  
SNR
CMIMO ¼ log2 1 þ ki b=s=Hz ð6:5Þ
i¼1
N

6.3 Results and Discussion

Experimental work has been done by collecting data with configurations of typical
and spatial diversity technique. The different measurement tools are helping in
characterizing MIMO channel parameter.

6.3.1 Typical and Spatial Diversity

The measurement work was started by collecting the data from the typical setup.
Table 6.1 show thesimulation and measurement for typical and spatial diversity
setup with lm and ln is antenna spacing at transmitter (lm = ln = k).
From the table, the Kronecker and Weichselberger model shows huge different
compared to the measured data. However the Weichselberger model is closer to
the measured data because this model uses joint correlation at both ends. It also
prove the [16] work that Kronecker model is fail to predict the channel capacity for
the system used more than 2 9 2 antenna.

6.3.2 MIMO System with Spatial Diversity Technique

Measurement for spatial diversity setup has been done by varying the spacing
(l) between antennas at transmitter by fixing antenna spacing at receiver and vice
versa. From the observation, by changing the spacing between the antennas, it will
change the value of eigenvalue and capacity.
Table 6.2 shows the MIMO channel capacity for configuration by changing l at
receiver with fixed antenna spacing at transmitter.
6 Channel Capacity of Indoor MIMO Systems … 57

Table 6.1 Comparison of MIMO channel capacity for simulation and measured data of typical
configuration at l = k
Simulation Measurement
Kronecker Weichselberger
42.3868 b/s/Hz 30.3454 b/s/Hz 7.86 b/s/Hz

Table 6.2 Result for MIMO channel capacity by changing l at receiver with fixed antenna
spacing at transmitter
Antenna spacing at receiver, ln
Antenna spacing at transmitter lm = k k/4 k/2 k 2k 3k
Channel capacity (b/s/Hz) 6.92 6.134 7.7265 7.4282 7.8934

Fig. 6.2 Comparison 120

between measurement result 103.2822
and simulation result for 100 91.392
spatial diversity at receiver
80

60
42.3868

40 31.3755 41.0702
43.8203
24.5111 30.3454
26.0801
20 23.8862
6.92 6.134 7.7265 7.4282 7.8934

antenna spacing at receiver

Measure Kron Wiechs

The result from Table 6.2 is a comparison of measurement result for MIMO
Channel capacity by changing l at receiver with fixed antenna spacing at trans-
mitter. From the table, the largest MIMO channel capacity is 7.8934 b/s/Hz at 3k
antenna spacing configuration. Due to the previous literature on spatial diversity,
the capacity increase by expansion of spacing between the antennas.
From the Fig. 6.2, the result shows the channel capacity of Weichselberger
model is closer to the measured data compare to Kronecker model.
The observation result of MIMO channel of that graph also shown capacity is
increase by increasing the number of antennas. However, the capacity from
measured data does not increase linearly by increasing antennas spacing.
Table 6.3 shows the result for antenna spacing at transmitter is changing with
fixed l at receiver. Compared to typical configuration, the MIMO channel capacity
is lower when applying spatial diversity at the transmitter which is the inter
58 M. Senon et al.

Table 6.3 Result for antenna spacing at transmitter is changing with fixed l at receiver
Antenna spacing at transmitter (Im)
Antenna spacing at receiver ln = k k/4 k/2 k
Channel capacity (b/s/Hz) 7.3274 6.3301 7.7265

Fig. 6.3 MIMO channel 120

capacities with spatial
diversity at transmitter 103.2822
100
91.392
80

60

44.9238
42.3868 42.1257
40
31.3755 30.3454
24.5111 24.0597
20 19.3999
7.3274 6.3301 7.7265 6.7639
7.0673
0

Antenna spacing at transmitter

Measure Kron Wiechs

Fig. 6.4 Eigenvalue for 5

10 * Eigenvalue

measure data with spatial 0

diversity at transmitter -5
-10
-15
-20
-25
-30

antenna spacing
L1 L2 L3 L4

element antenna spacing is shifted from l = k/4 until 3k. From the observation,
the highest MIMO channel capacity is 7.7265 b/s/Hz while the lowest value of the
system is 6.3301 b/s/Hz when applying k/2 inter element antenna spacing. The
differentiation between highest and lowest value is 18.07 %.
Figure 6.3 show the comparison between simulation and measurement result of
MIMO channel capacity with spatial diversity at transmitter. The result has shown
similar to the system with spatial diversity at transmitter where the simulation
result is much differs from the measurement result. The MIMO channel capacity
increase by increasing the spacing between the antennas for all models.
Table 6.4 Correlation coefficient at lm = 0.5
Receiver (ln = k/4) Receiver (ln = k)
-0.08116 - 0.07135 0.278584 0.184097 -0.22019 -0.05203 -0.06486
0.25826
-0.36418 0.320501 -0.02726 0.14998 0.53665 -0.30272 0.12837
0.073011
0.013381 0.069882 -0.51134 -0.20811 -0.23324 0.401047 -0.2107
-0.1958
6 Channel Capacity of Indoor MIMO Systems …

0.168741 0.428591 0.447742 0.331321 0.037825 -0.36922 0.280644

0.105638
-0.08116 0.07135 0.278584 0.184097 -0.22019 -0.05203 -0.06486
-
0.25826
-0.36418 0.320501 -0.02726 0.14998 0.53665 -0.30272 0.12837
0.073011
59
60 M. Senon et al.

Table 6.5 Result average ln = k/4 ln = k

correlation coefficient
Average correlation 0.033713 0.024304

Fig. 6.5 Eigenvalue for 5

0

Eigenvalue
MIMO system with spatial
-5

10 log
diversity at receiver
-10
-15
-20
-25
-30

Antenna Spacing
L1 L2 L3 L4

6.3.3 Eigenvalue Analysis and MIMO Channel Correlation

Coefficient

Figure 6.4 shows the result of Eigenvalue analysis for measurement result of
spatial diversity at transmitter.
From the graph, the highest eigenvalue of the system is L1 especially when the
antennas are placed with k/2 inter element spacing but the L2 and L3 is lower than
the system with k antenna spacing.
Correlation coefficient is used in order to calculate MIMO channel capacity.
Table 6.4 summarize the correlation coefficient of MIMO system with spatial
diversity at the receiver. The Table 6.5 had shown the average of correlation
coefficient from the previous table. The average correlation is decrease by
increasing the spacing (Lrx). Due to the result of average correlation, the capacity
of the system is better when the correlation coefficient is higher.
Figure 6.5 show the eigen value of the MIMO system when applying the spatial
diversity at receiver. The channel capacity increase when the eigenvalue of the
system close within each other such as at ln = 3k. Other than the L1, the eigen-
value of L4 also give a significant contribution in enhancing the capacity at
ln = 3k because the value is highest at that configuration compared to others.

6.4 Conclusion

The paper presented a brief analysis of the MIMO channel capacity by using spatial
diversity technique at 2.4 GHz operating frequency. The analysis on the experiment
data shows the MIMO channel capacity increase by increasing antenna spacing.
6 Channel Capacity of Indoor MIMO Systems … 61

The capacity of Weichelberger model is more suitable to the measured channel

capacity compared to Kronecker model when physical effect were considered.

References

1. Jensen, M.A., Wallace, J.W.: A review of antennas and propagation for MIMO wireless
communications (invited paper). IEE Trans. Antennas Propagat. 52, 2810–2824 (2004)
2. Foschini, G.F., Gans, M.J.: On limits of wireless communication in a fading environment
when using multiple antennas. Wirel. Pers. Commun. 6(3), 311–335 (1998)
3. Duman, T.M., Ghrayeb, A.: Coding for MIMO Communication Systems. John Wiley & Sons
Ltd, England (2007)
4. Poon, A.S.Y.: A Spatial Channel Model For Multiple—Antenna Systems. 2004
5. Anreddy, V. R & Ingram, M. A. (2006). Capacity of measured Ricean and Rayleigh indoor
MIMO channels at 2.4 GHz with polarization and spatial diversity. In: Anreddy, V.R.,
Ingram, M.A. (eds.) IEEE Wireless Communication and Networking Conference,
pp. 946–951 (2006)
6. Kermoal, J.P., et al.: Experimental investigation of multipath richness for multi-element
transmit and receive antenna arrays. In: Kermoal, J.P., Mogensen, P.E., Jensen, S.H.,
Andersen, J.B., Frederiksen, F., Sorensen, T.B., Pedersen, K.I. (eds.) IEEE Vehicular
Technology Conference Proceedings, vol. 3, pp. 2004–2008 (2000a)
7. Molina-Garcia-Pardo, J.-M., et al.: Polarized indoor MIMO measurements at 2,45 GHz. In:
Molina-Garcia-Pardo, J.-M., Castillo O.I., Egea-Garcia, F., Juan-Llacer, L. (eds.) IEEE
Antennas and Propagation Society International Symposium, pp. 5335–5338 (2007)
8. Almers, P., Bonek, E., Burr, A., Czink, N., Debbah, M., Degli-Esposti, V., Hofstetter, H.,
Kosti, P., Laurenson, D., MAtz, G., Molisch, A.F., Oestges, C., Özcelik, H.: Survey of
Channel and Radio Propagation Models For Wireless MIMO Systems (Research Article).
EURASIP J. Wirel. Commun. Networking 2007 (2007)
9. Botonjiæ, A.: ‘‘MIMO channelmodels’’, Diploma Thesis,Examensarbete utfört i
Elektronikdesign vid Linköpings Tekniska Högskola,Campus Norrköping
10. Yu, K., Ottersten, B.: Models for MIMO propagation channels, a review. In: Wiely Journal
on Wireless Communications and Mobile Computing Special Issue on Adaptive Antennas
and MIMO Systems, 2002-07-08
11. Richard J.E., Oscar F., Torres, R.P.: Empirical analysis of 2 9 2 MIMO channel in outdoor–
indoor scenarios for BFWA applications. IEEE Antennas Propag. Mag. 48(6) (2006)
12. Rappaport, T.S.: Wireless Communications (Principle and Practice), 2nd edn. Prentice Hall
of India, India (2007)
13. Schott, J.R.: Matrix Analysis for Statistics, p. 24. Wiley, New York (1997)
14. Liu, L., Hong, W.: Characterization of Line-of-Sight MIMO Channel for Fixed Wireless
Communications. 2007
15. Pérez F. F., Mariño E. P.: Modeling The Wireless Propagation Channel. Wiley, New york
(2008)
16. Özcelik, H.: Indoor MIMO channel models. PhD Thesis, submitted to Institute für
Nachrichtentechnik and Hochfrequenztechnik of the Technische Universität Wien. Dec 2004
Chapter 7
Design of Multi-band Antenna
for Wireless MIMO Communication
Systems

M.M. Saad, M.N. Husain, M.Z.A. Aziz, A.R. Othman,

K.A.A. Rashid and M. Senon

Abstract This paper present the design of band antenna for wireless Multiple Input
Multiple Output (MIMO) communication system. The advantage of MIMO system
is able to enhanced the overall wireless channel capacity. Besides system perfor-
mances, the size and mobility of the communication devices are also contributed
the successfulness of proposed wireless communication system. The multi-band
antenna design has been discussed by many researchers for various application.
This paper present the design of multiband antenna by using slot techniques. The
design was simulated by using microwave CST studio. The results show that, the
proposed antenna can operate at 2.4, 3.5 and 5.2 GHz frequency band.

7.1 Introduction

High capacity is very crucial in wireless as it is required for future application. It is

because, the data transferred by users are becoming more larger nowadays. For
example, in the early years the data transferred consists of only text or voice only.
As time goes by, the user is demanding higher channel capacity as they wish to
send voice with picture, or text with video or text, picture, and video at the same
time. These data are required more capacity to be transferred. One of the main
challenges in wireless communication is to gain high capacity in order to fulfil
future application necessities. It is important in increasing the data transferred
simultaneously. Today’s channel suffers from attenuation due to multipath in the
channel [1]. The increasing demand for capacity in wireless systems has motivated
considerable research aimed at achieving higher throughput on a given bandwidth.
It is proven in [2–5] that wireless communication systems using multiple antennas

M.M. Saad (&)
M.N. Husain
M.Z.A. Aziz
A.R. Othman

K.A.A. Rashid
M. Senon
Faculty of Electronics and Computer Engineering, Universiti Teknikal Malaysia Melaka
(UTeM), Durian Tunggal, Melaka, Malaysia
e-mail: [email protected]

Ó Springer International Publishing Switzerland 2015 63

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_7
64 M.M. Saad et al.

at both transmitter and receiver sides, denoted as Multiple Input Multiple Output
(MIMO) antenna systems, enable great enhancements of channel capacity com-
pared to the single input single output (SISO) system.
From [1–8], it is mentioned that multiple antennas at both transmitter and
receiver can improve the wireless channel capacity within the same bandwidth and
power received. Besides improving performance in terms of higher channel
capacity which enables a wireless device to transfer and receive data with higher
data rate, MIMO also proven to reduce the multipath fading. Also, it is mentioned
that multiple antennas are proven to introduce robustness against channel fading
and interference.
Current communication devices are portable and multi-applications supported
such as Bluetooth, Wi-Fi, GPS, and so much more. These applications are used
different operating frequencies in order to avoid interferences between them. The
conventional way to support this demand is by placing two or more antennas with
different resonant frequency in that device. Clearly, this method is not efficient
because integrating two or more antennas proves difficult as they are likely going to
couple to each other causing degradation of the received or transmitted signals [9].
Placing two antennas in a device also bring in the space issue since it is usually
designed to be a portable device and by placing multiple antennas in the device
required more spaces. Another way to support multi-frequency demand without
facing any of those problems is by using multi-band antennas which consist only
one simple structure antenna that can support dual frequency ranges and it is
clearly being more efficient.
As mentioned before, in MIMO system, at least two antennas for each operating
frequency should be placed in the wireless device in order to support the
requirement of the multi-application device. This is quite challenging since gen-
erally, according to [10], placing two antennas at the distance of half wavelength is
necessary to achieve good isolation. Then, if the antennas are in large size, more
space required for the spacing and for the antennas itself which are not practical
and inefficient. Then, an efficient antenna especially for MIMO application is an
antenna that compact and can support multi-frequency operation.

7.2 Antenna Design

There are several design steps that has been done in designing the multiband
antenna in this project. Dual-band antenna which is Design A is designed first and
the method approached in designing a dual-band antenna in this project is by
embedding a pair of slots on the antenna’s patch. There are several slot shapes that
designed and studied and they are divided to three designs which are Design A1(i),
A1(ii), and A1(iii). All of the designs are founded by the shape of slot in Design
A1(i) which is C-shaped slots. Design A1(ii) is the Three-shaped slot which
are double-up the C-shaped slots and Design A1(iii) is continuous from the
7 Design of Multi-band Antenna for Wireless MIMO Communication Systems 65

Three-shaped slot, but instead of having the equal diameter at both upper and
lower slots, this design has a different diameter of the upper and lower slot and
produce a pumpkin-like shape, hence named as The Pumpkin-shaped slot.

7.2.1 C-Shaped Slot Antenna [Design A1(i)]

The antenna structure of C-Shaped Slot Antenna and all other designs are similar
to the basic rectangular patch antenna which consist patch, substrate (dielectric),
and the ground plane and the material of each plane are similar. The structure of
antenna’s geometry is as shown in Fig. 7.1. There is a pair of slots that in
C-shaped which gives the additional second frequency, then makes the antenna act
as a dual-band antenna.
The effects of the antenna parameters such as the slot width and the slot radius,
position of slots, and the separation between the slots are studied. But, in this
section only the studies that give significant impact on the design presented, which
are slot width, a, distance between slots, x, and the slot radius, r.
From Table 7.1, it shows that the parametric study of slot width and distance
between slot respectively. The second resonance frequency, f2 tends to shift larger
as the slot gets wider and as the slots get further from each other. The parameter of
the slot width, a is studied from 0.5 to 2 mm, and the distance between slots, x is
studied from 1 to 29.5 mm as it is the minimum and the maximum value, but the
result presented is the one that gives the significant impact on the result. In terms
of return loss, it gets smaller as the slots get wider but, conversely become worse
when the slots gets further from each other.
From the results, it shows that as the slot radius, r increases, the second res-
onant frequency, f2 shifting to left, which is smaller, and the return loss becomes
increases. The parameter studied included all values that possible, but radius of
3 to 7 mm gives the significance effects to the response. From all the studies, it
occurs that all three parameters only affect the second resonance frequency, f2
while first resonance frequency, f1 remain constant or not much difference.
Table 7.1. shows the summary of all three antenna parametric studies.

7.2.2 Three-Shaped Slot Antenna [Design A1(ii)]

Three-Shaped Slot Antenna is the design by combining two of the C-shaped slots in
the previous design and creates a slot that looks like the shape of the number three. It
also has the same structure and material used as the dielectric and conductor
material which are FR-4 and copper respectively. Figure 7.2 shows the antenna’s
structure and in this design, all the antenna parameters that studied are the same as
the design A1(ii). However, only two of them are considered giving the significance
effects to the response, which are slot width, a and the slot radius, r.
66 M.M. Saad et al.

Fig. 7.1 C-Shaped slot antenna design

Table 7.1 Summary of the parametric studies of Design A1(i)

Parameters f1 Return loss f2 Return loss
(GHz) (dB) (GHz) (dB)
Slot width, a (mm) 0.5 2.392 -26.33 5.158 -6.86
1.0 2.386 -23.81 5.480 -12.13
1.5 2.380 -23.38 5.590 -17.72
2.0 2.374 -24.03 5.752 -20.77
Distance between slots, 6 2.380 -23.25 5.368 -9.94
x (mm) 8 2.380 -26.69 5.500 -14.52
10 2.380 -23.38 5.590 -15.72
14 2.380 -23.78 5.704 -19.11
Slot radius, r (mm) 3 2.404 -28.59 6.190 -21.84
4 2.392 -28.77 6.022 -22.86
5 2.380 -23.38 5.590 -15.72
6 2.356 -20.56 3.196 -7.57
7 2.338 -19.34 2.896 -4.29

Table 7.2 represents the analysis study of slot width, a that embedded on the
patch and the studied are started with a equal to 0.5 until a equal to 2.0 mm. From
figure, it shows that as slot width increases, the second resonance frequency, f2
also increases, contrary to the return loss which decreases as the slots get wider.
Table 7.2 also shows the analysis study for the parameter of slot radius, r. The
table shows that the second resonant frequency, f2 will be increased as the radius
increases. Noticed that the same effects occurred in the frequency of Design A1(i).
As for the return loss parameter, it will decrease as the slot radius, r increases. All
parameters are affecting only the upper frequency and not much impact to the
lower one. Table 7.2 shows the summary of all parametric studies.
7 Design of Multi-band Antenna for Wireless MIMO Communication Systems 67

Fig. 7.2 Three-shaped slot antenna design

Table 7.2 Summary of the parametric studies of Design A1(ii)

Parameters f1 (GHz) Return loss (dB) f2 (GHz) Return loss (dB)
Slot width, a (mm) 0.5 2.434 -20.35 4.678 -12.42
1.0 2.428 -20.89 4.634 -11.38
1.5 24.160 -19.72 5.200 -40.26
2.0 24.160 -20.04 5.530 -30.35
Slot Radius, r (mm) 3 2.434 -23.60 5.926 -20.21
4 2.422 -20.48 5.278 -37.30
5 2.404 -23.08 5.056 -29.17

7.2.3 Pumpkin-Shaped Slot Antenna [Design A1(iii)]

The Pumpkin-Shaped Slot Antenna is design based on Three-Shaped antenna. The

only difference is the lower and upper part of the slot is not in the same radius
which makes it looks like a pumpkin-shaped slot. This design has the same
geometry of design structure and the same material used in the previous designs.
Figure 7.3 shows the geometry of the design.
In this design, the parameters that give the significance impact to the response
are the distance between the slots, x and the differences of slots radius, r. Table 7.3
shows that the same results are occurred as the slots get further from each other
and the differences of slots radius become larger because both of them makes the
second resonance frequency, f2 shifting and become larger. However, it is not the
same case for the return loss as it gets increased when the distance between slots,
x increases and decreases when the differences of slots radius, o increases. The
return loss of the first resonance frequency f1 is slightly increased as the differ-
ences of slots radius, r become larger, but still not affected by the distances
between slots, x. Table 7.3 shows the overall analysis of all both parameters.
68 M.M. Saad et al.

Fig. 7.3 Pumpkin-shaped slot antenna design

Table 7.3 Summary of the parametric studies of Design A1(iii)

Parameters f1 Return loss f2 Return loss
(GHz) (dB) (GHz) (dB)
Slot Radius, r (mm) 0 2.506 -20.28 5.230 -9.47
1 2.500 -12.74 5.326 -16.17
2 2.500 -9.25 5.440 -22.99
Distance between Slots, x 5.75 2.512 -30.34 5.650 -37.51
(mm) 7.50 2.512 -29.40 5.554 -13.45
9.25 2.506 -26.40 5.422 -5.58

7.3 Simulation and Measurement Results

In this section, the result of all dual-band antenna designs are presented and
discussed. Table 7.4 shows the optimized value of design parameters for each
antenna design. While Fig. 7.4 show the fabricated antenna that have been design.
Figures 7.5, 7.6, 7.7 and 7.8 shows the simulation and measurement results of
design antenna.
From the graph in Fig. 7.5, it shows that the most optimum design of Design
A1(i) resulting resonance frequency of 2.386 GHz and 5.068 GHz whereas the
design specification is designing dual-band antenna with resonance frequency
2.4 GHz and 5.2 GHz as f1 and f2 respectively. Because it is hard to achieve the
specification using this design, it has been modified and improved as these next
designs (Design A1(ii) and A1(iii)). From the graph, it also shows that there are
shifting between the simulation and measurement resonance frequencies, f1 and f2
for Design A1(ii). That is may be caused by the dimension of fabricated design
which is slightly inaccurate or not exactly same as the simulation. From the
parametric studies of Design A1(ii) that have been performed earlier, it is noticed
that the f1 is dependence to the dimension of the patch, which is the width, W and
7 Design of Multi-band Antenna for Wireless MIMO Communication Systems 69

Table 7.4 Optimized dimension of Design A1

Parameter A1(i) A1(ii) A1(iii)
Patch width, W 31.5 mm 33.8 mm 36.5 mm
Patch length, L 26.6 mm 26.12 mm 28.45 mm
Slot width, a 0.5 mm 1.5 mm 1.5 mm
Radii of slot, r 6.0 mm 4.2 mm 5 mm
Distance between slots, x 12 mm 5 mm 4.5 mm
Slot radius differences, o NA 1 mm
Patch size changes 8.70 % 3.80 % 13.16 %
(reduce) (reduce) (increases)
Coordinate of centre of circle, (6,4) NA
(X,Y)

Fig. 7.4 Fabricated design of Design A1

Fig. 7.5 Frequency response of Design A1

70 M.M. Saad et al.

Fig. 7.6 Simulation radiation pattern of Design A1(i). a E plane. b H plane

Fig. 7.7 Simulation and measurement radiation pattern of Design A1(ii). a E plane. b H plane

Fig. 7.8 Simulation and measurement radiation pattern of Design A1(iii). a E plane. b H plane

the length, L, while f2 is dependence to the dimension and position of the slots.
Due to lack of equipment, it is very difficult to maintain the accuracy of the
dimension, especially at the slots part which only 1.5 mm width. Same case
happened to Design A1(iii) but the frequency shifting are more greatly especially
on second resonance frequencies, f2. That is also may be due to the inaccuracy of
the slots dimension on the fabricated design. Consistent with the parametric
studies that have been performed earlier which proved that f2 is also dependence
to the dimension and position of the slots.
7 Design of Multi-band Antenna for Wireless MIMO Communication Systems 71

As mentioned before, both Design A1(ii) and A1(iii) are the improvement and
modification of Design A1(i) as it is hard to achieve the desired resonant frequency
which are 2.4 GHz and 5.2 GHz. From Fig. 7.5, it seems like both Design A1(ii)
and A1(iii) gives the similar result in their resonance frequency, but the Three-
shaped Slot Antenna is slightly better than the Pumpkin-shaped Slot Antenna in
terms of the return loss and bandwidth. The main factor that makes the Design
A1(ii) is better is the size of the patch. Based on Table 7.4, Design A1(ii) gives
size reduction by 3.89 percent but Design A1(iii) which is the Pumpkin-shaped
Slot Antenna gives size increment by 13.16 % which are not fulfilling the design
specification that required size reduction at least by 50 % of the basic rectangular
microstrip patch designed earlier.

7.4 Conclusion

This paper present the design of multiband antenna by using C-shape slot tech-
niques. Three design has been proposed based on different combination of C-shape
slot. The design of double C-shape slot can produce the resonant frequency at 2.4
and 5.2 GHz. Besides that, the size of the double C-shape slot antenna can be
reduced up to 50 % compared to the microstrip patch antenna. So, this design can
be used to develop the compact mobile communication devices.

References

1. Daud, Z., Aziz, M., Suaidi, M., Rose, M., Kadir, M., Shah, M.: MIMO channel
characterization and optimization. In: 6th National Conference on Telecommunication
Technologies 2008 and 2008 2nd Malaysia Conference on Photonics, NCTT-MCP 2008.
pp. 132–135 (2008)
2. Stuber, G., Barry, J., McLaughlin, S., Li, Y., Ingram, M., Pratt, T.: Broadband MIMO-OFDM
wireless communications. Proc. IEEE 92:271–294 (2004)
3. Keusgen, W.: On limits of wireless communications when using multiple dual-polarized
antennas. In: 10th International Conference on Telecommunications, ICT 2003, pp. 204–210
(2003)
4. Jungnickel, V., Pohl, V., Nguyen, H., Kruger, U., Haustein, T., von Helmolt, C.: High
capacity antennas for MIMO radio systems. In: The 5th International Symposium on
Wireless Personal Multimedia Communications, vol. 86, pp. 407–411 (2002)
5. Nasr, A., Molina, J., Lienard, M., Degauque, P.: Optimisation of antenna arrays for
communication in tunnels. In: 3rd International Symposium on Wireless Communication
Systems, ISWCS ‘06, pp. 522–524 (2006)
6. Winters, J.: On the capacity of radio communication systems with diversity in a Rayleigh
fading environment. IEEE J. Sel. Areas Commun. 5:871–878 (1987)
7. Golden, G., Foschini, C., Valenzuela, R., Wolniansky, P.: Detection algorithm and initial
laboratory results using V-BLAST space-time communication architecture. Electro. Lett.
35:14–16 (1999)
72 M.M. Saad et al.

8. Ozdemir, M., Arslan, H., Arvas, E.: A mutual coupling model for MIMO systems. IEEE
Topical Conf. Wirel. Commun. Technol. 306–307 (2003)
9. Matsunaga, M., Kakemizu, K., Candotti, M., Matsunaga, T.: An omni-directional multi-
polarization and multi-frequency antenna. In: IEEE International Symposium on Antennas
and Propagation (APSURSI), pp. 2765–2768 (2011)
10. Luo, Q., Salgado, H., Pereira, J.: Printed C-shaped monopole antenna array with high
isolation for MIMO applications.: In: Antennas and Propagation Society International
Symposium (APSURSI), 2010 IEEE, pp. 1–4 (2010)
Chapter 8
Design of Linear Polarization Antenna
for Wireless MIMO Application

K.A.A. Rashid, M.N. Husain, A.R. Othman, M.Z.A. Aziz,

M.M. Saad, M. Senon, M.T. Ahmad and J.S. Hamidon

Abstract This paper present the design of the linear polarized antenna for
wireless MIMO communication system. It is impossible to fulfil the demand of the
wireless communication system due to limitations in channel capacity on single
input single output (SISO) systems. Multiple input multiple output (MIMO) sys-
tem has become a famous research field for the next generation wireless com-
munication system in order to overcome this problem. Since polarization diversity
is effective to avoid the fading loss caused by multipath effects, therefore, polar-
ization diversity becomes one of the most important techniques that can be used to
enhance MIMO system performances. It can be utilized to improve the commu-
nications channel capacity and utilize the frequency spectrum with frequency reuse
technique. Therefore, the development of linear polarized antenna is significant in
order to improve the wireless MIMO system performance based on polarization
diversity technique. Polarization diversity can be utilized to double the frequency
spectrum to realize frequency reuse and improve the communications capacity.
This project is to design an antenna that can provide linear polarization to reduce
the signal losses.

8.1 Introduction

Antenna plays a crucial role in telecommunication field such as satellite

communication, mobile phone and for military use [1]. The growth of mobile
communications results in the increasing demand of smart phones, wireless
internet and other broadband applications. The demand for high data rate and high
capacity are increasing to satisfy the growth of mobile communications [2, 3].

K.A.A. Rashid (&)
M.N. Husain
A.R. Othman
M.Z.A. Aziz
M.M. Saad
M. Senon

M.T. Ahmad
J.S. Hamidon
Faculty of Electronics and Computer Engineering, Universiti Teknikal Malaysia Melaka,
Melaka, Malaysia
e-mail: [email protected]; [email protected]

 Springer International Publishing Switzerland 2015 73

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_8
74 K.A.A. Rashid et al.

The MIMO technology is become popular nowadays as the demand for high
speed, high capacity and high quality transmission in wireless mobile telecommu-
nication are increasing [3–5]. MIMO technology provide a very high spectral effi-
cient and also increase the channel capacity without extra spectrum by using multiple
transmitter and receiver [3, 5–7]. The loss of spectral efficiency occurs due to the
spatial correlation between antennas [6, 7]. In order to decrease the spatial corre-
lation, the spatial diversity is the most common used technology in MIMO system.
The distances between MIMO antennas have to be at least half wavelength apart thus
larger spacing is required for this technique [3, 7]. MIMO capacity increased pro-
portionally with the number of antennas and thus higher spatial distance between
antennas is required [3]. The mutual coupling between antennas occurs due to the
narrow space for MIMO antenna design is the critical problem in MIMO system [1].
As a result, the polarization diversity becomes better solution for MIMO system.
In the modern wireless communication, the products such as smart phones and
radio frequency identification (RFID) tags are designed in light weight and very
compact size. The purpose of compact size and light weight enable the devices to
be carried to everywhere. The antenna needs to be designed in a very small size so
that the antenna is able to fit inside the compact size products. Moreover, the
wireless devices are design to be portable; the antenna designed must be able to fit
into the small device and are practical use. Thus, the size of an antenna is also one
of the main criteria to take into account when designed the antenna [8–10].

8.2 Design Specification

The antenna is design for the frequency of 2.4 GHz, which is useful for the Bluetooth
application. The 2.4 GHz frequency is an unlicensed band and is free for the public.
The antenna is simulating using FR4 board and PEC or copper. FR4 board with
thickness of 1.6 mm, tangent loss of 0.019 S/m and permittivity of 4.4 is used as the
substrate. While the PEC with thickness of 0.035 mm is used as the conductor.
In this project, the antenna design is targeted to have gain from range of 2–4 dB
for each of the polarization state. This is to avoid the problem of one of the
polarization state provide higher gain and become the dominant mode. The range
of the gain is referring to the works of previous researchers. Meanwhile the axial
ratio for a circular polarization antenna must be below -2.5 dB and for linear
polarization antenna must be above -3 dB. For this project, the antenna is design
to fulfill the compact issue in terms of size by combined the radiator for both
polarization states by using 2 ports. The radiation efficiency and the total efficiency
for both polarization states must be below -3 dB to ensure that the antenna is able
to transmit at least 50 % of the radiate power respectively. The matching is affect
by the total efficiency, as the total efficiency getting closer to positive value the
matching is getting better. While for return loss, the result must be below -10 dB
to ensure that the transmission power of the antenna is more than 90 %. The
specification of the design is listed at Table 8.1.
8 Design of Linear Polarization Antenna … 75

Table 8.1 Main specification Specification Value

of the design
Frequency 2.4 GHz
Return loss \-10 dB
Gain 2–4 dB [42, 51, 52]
Radiation efficiency \-3 dB
Total efficiency \-3 dB
Multipolarization Linear polarization
Axial ratio [3 dB for linear polarization

8.2.1 Antenna Design Process

The linear polarization antenna is first design to be a rectangular patch. The patch
is then added double H-shaped slots. Lastly, the return loss of the design is then
improved by adding stair notch to the patch.

8.2.2 Rectangular Patch with Slots

The rectangular patch is first drawn in CST microwave studio by using PEC (grey
colour) and FR4 as the substrate (blue colour). The patch is connected to a CPW
fed line. Then the slots are drawn on top of the patch and then substrate the slots
from the patch as shown in Fig. 8.1. After the drawing of the patch is done, the
setting for the simulation is set as same as the previous design. The effects of the
slots dimension are obtained using parametric study method. From previous
parametric study, the performance of the design is under performance. Thus, the
gap of the ground and CPW fed are studied (Fig. 8.2).
From Table 8.2, the spacing between the CPW fed and the ground has sig-
nificant effect on the return loss of the design. The return loss start to increased
until b = 0.6 mm. In addition, spacing does not affect much on the axial ratio and
gain of the design.
The parameter of c does not affect much on the axial ratio and gain of the
antenna as shown in Table 8.3. However, the axial ratio of the antenna is
decreasing as the c increasing. While for the return loss of the antenna is unstable
when the c is increased (Table 8.4).
Slot4 length, d does not affect much on axial ratio and gain of the design. For
return loss of the design, it did not have a stable trend as the d increased.
From Table 8.5, the return loss of the design do not showed a stable pattern.
The return loss slightly increased after e = 4 mm. Furthermore, axial ratio and
gain of the design do not affect much by the slot2 length. The gain only differs
around 0.2 dB (Table 8.6).
The width of slot3, f does not affect much on the gain of the design. The return
loss of the design decreased when the width getting wider. But the axial ratio
76 K.A.A. Rashid et al.

Fig. 8.1 Front view of the rectangular patch with slots

Fig. 8.2 Back view of the

rectangular patch with slots

Table 8.2 Effects of spacing b (mm) RL (dB) AR (dB) Gain (dB)

(b)
0 -8.57 11.51 0.73
0.1 -9.36 11.57 0.77
0.2 -10.33 11.54 0.66
0.3 -13.98 11.53 0.97
0.4 -15.49 11.55 0.97
0.5 -19.95 11.57 0.96
0.6 -27.74 11.59 0.96
0.7 -17.62 11.63 0.94
8 Design of Linear Polarization Antenna … 77

Table 8.3 Effects of slot3 c (mm) RL (dB) AR (dB) Gain (dB)

length (c)
0 -15.61 11.93 1.25
2 -14.56 11.81 1.02
4 -18.64 11.68 0.87
6 -16.65 11.64 0.94
8 -15.14 11.57 1.01
10 -13.73 11.52 1.05

Table 8.4 Effects of slot4 d (mm) RL (dB) AR (dB) Gain (dB)

length (d)
0 -15.61 11.93 1.25
2 -14.85 11.85 1.08
4 -18.74 11.81 0.91
6 -16.83 11.85 1.02
8 -15.37 11.78 1.09
10 -13.98 11.71 1.11

Table 8.5 Effects of slot2 e (mm) RL (dB) AR (dB) Gain (dB)

length (e)
0 -15.61 11.93 1.25
2 -14.63 11.79 1.02
4 -18.69 11.87 0.86
6 -16.63 12.01 0.98
8 -15.21 11.91 1.01
10 -13.85 11.86 1.09

Table 8.6 Effects of slot3 f (mm) RL (dB) AR (dB) Gain (dB)

width (f)
0 -17.88 11.63 0.93
0.5 -18.36 11.64 0.99
1 -18.59 11.64 1.02
1.5 -18.76 11.64 1.03
2 -18.77 11.63 1.03

remains the same as the slot3 width increasing. The gain is still smaller even the
slot3 width increasing.
As showed in the Table 8.7, width of slot4 did not affect much on the axial
ratio, gain and return loss. But still the gain is getting better as the width
increasing. The axial ratio of the design is still a linearly polarized antenna.
78 K.A.A. Rashid et al.

Table 8.7 Effects of slot4 g (mm) RL (dB) AR (dB) Gain (dB)

width (g)
0 -16.11 11.81 1.06
0.5 -16.54 11.82 1.12
1 -16.76 11.85 1.18
1.5 -16.74 11.86 1.18
2 -16.68 11.87 1.18

From Table 8.8, the results showed that the width of slot2 does not affect much
on the return loss, axial ratio and gain of the design. As the width of slot2 increase,
the gain is only increased from 1.02 to 1.18 dB. The axial ratio is also increased
around 0.26 dB from 11.79 to 12.05 dB (Table 8.9).
As the slot1 length increasing, the return loss is slightly increased. However, the
gain is only decreased a bit from 5.37 to 5.21 dB. In addition, the axial ratio is
increased around 0.2 dB and the antenna is still a linearly polarized antenna
(Table 8.10).
As the slot1 width increased, the return loss is increasing. While for the gain of
the design is decreasing as the width of slot1 increasing. The axial ratio of the
design is increasing from 11.93 to 12.04 dB.

Table 8.8 Effects of slot2 h (mm) RL (dB) AR (dB) Gain (dB)

width (h)
0 -16.47 11.79 1.02
0.5 -16.75 11.84 1.10
1 -17.01 12.02 1.13
1.5 -16.79 12.04 1.17
2 -16.58 12.05 1.18

Table 8.9 Effects of slot1 i (mm) RL (dB) AR (dB) Gain (dB)

length (i)
2 -19.18 9.00 5.19
4 -18.77 8.99 5.23
6 -19.11 8.98 5.18
8 -18.94 9.03 5.21
10 -16.72 8.79 5.31

Table 8.10 Effects of slot1 j (mm) RL (dB) AR (dB) Gain (dB)

width (j)
0 -15.61 11.93 1.25
1 -12.33 11.89 0.79
2 -12.21 11.87 0.95
2.5 -11.83 12.09 1.04
8 Design of Linear Polarization Antenna … 79

8.2.3 Result and Discussion

The final design of linear polarization antenna consists of a rectangular patch with
double H-shaped slots and the design is added with stair notches. Figures 8.3 and
8.4 are the front view and back view of the linear polarization final design antenna.
Table 8.11 is the optimum dimensions of the linear polarizes antenna.

Fig. 8.3 Front view design of final linear polarization antenna

Fig. 8.4 Back view design

of final linear polarization
antenna
80 K.A.A. Rashid et al.

Table 8.11 Optimum Value(mm)

dimension of linear polariza-
tion antenna L1 35
W1 28
L2 19
W2 15.4
A 1
B 0.7
C 5
d 7
e 12
f 1
g 1
h 1
i 11
j 1
k 3.7
l 2.5
m 1.7
n 1.5
o 10.5

Figures 8.5, 8.6 and 8.7 are the comparison graphs of return loss, axial ratio and
gain between the rectangular patch, rectangular patch with slots and rectangular
patch with slots and notches. The return loss is improved with the used of notches
from -8.9 dB at frequency of 2.398 GHz to -38.7 dB at frequency of 2.401 GHz.

Fig. 8.5 Return loss comparison

8 Design of Linear Polarization Antenna … 81

Fig. 8.6 Axial ratio comparison

Fig. 8.7 Gain comparison

However, the gain of the patch with slots and notches is decreased from 2.16 to
1 dB. The axial ratio of these three antennas is all linear polarized antennas. The
return loss is improved with slots and notches, but the gain is degraded.
Figure 8.8 is the comparison of return loss between the measurement and the
simulation for linear polarized antenna. The measured return loss consists of ripple
at the front and the end of the response. The bandwidth of the measured return loss
is much wider than the simulation.
82 K.A.A. Rashid et al.

Fig. 8.8 Return loss comparison for simulation and measurement (linear polarization)

Fig. 8.9 Radiation pattern comparisons for linear polarization

Figure 8.9 is the radiation pattern for linear polarization in terms of vertical and
horizontal orientation. The simulated radiation pattern for vertical is a ‘8’ shape,
however the measured radiation pattern is more like a circle shape. While, the
radiation pattern for horizontal orientation, the shape of the measured radiation
pattern is almost the same with simulated radiation pattern (Fig. 8.10).
8 Design of Linear Polarization Antenna … 83

Fig. 8.10 Gain and directivity

8.3 Conclusion

This paper presents the design of Linear Polarization for wireless MIMO
communication system. Wireless channel capacity seems very important nowa-
days in order to support the broadband applications. Then, the design of linear
polarized antenna has been done. Rectangular patch with slots technique been
discussed. The linear polarized antenna can be used in developing the polarization
diversity in wireless MIMO communication system.

References

1. Chang, L.C., Tsai, C.H., Hsu, P., Liu, C.C.: A polarization diversity MIMO antenna design
for WiMAX dongle application. In: Microwave Conference Proceedings (APMC), 2010
Asia-Pacific, pp. 762–765, Dec. 2010
2. Vergerio, S., Rossi, J.-P., Sabouroux, P.: A two PIFA antenna systems for mobile phone at
2 GHz with MIMO applications. EuCAP 2006, 1–5 (2006)
3. Elliot, P.G., Rosario, E.N., Davis, R.J., Rzhanov, A.E.: MIMO polarization diversity antenna
with ultra-wide bandwidth and small size. In: IEEE International Symposium Phased Array
Systems and Technology (ARRAY), 2010, pp. 559–566, Oct. 2010
4. Wang, X., Chen, W., Feng, Z., Zhang, H.: Compact dual-polarized antenna combining
printed monopole and half-slot antenna for MIMO applications. In: IEEE APSURSI’09,
pp. 1–4, June 2009
5. Hu, S., Pan, J., Qiu, J.: A compact polarization diversity MIMO microstrip patch antenna
array with dual slant polarizations. In: IEEE APSURSI’09, pp. 1–4, June 2009
6. Chae, S.h., Choi, M.G., Park, S.O.: The realization and analysis of polarization diversity in
WiBro MIMO antenna. In: Antennas and Propagation Society International Symposium,
2007 IEEE, pp. 2409–2412, June 2007
84 K.A.A. Rashid et al.

7. Chae, S.h., Choi, M.G., Park S. O.: The realization of pattern/polarization diversity by
applying vertical excitation. In: TENCON 2007, pp.1–4, Nov. 2007
8. Lu, Y.C., Wu, Y.S., Chiu, C.T., Lin, Y.C.: A novel compact design of USB dongle antenna
for bluetooth applications. In: IEEE RWS’09, pp. 31–34, Jan. 2009
9. Lee, C.J., Gummalla, A., Achour, M.: Compact dualband antenna subsystem for MIMO
application. In: iWAT 2009, pp. 1–4, March 2009
10. Liu, X.F., Chen, Y.B., Jiao, Y.C., Zhang, F.S.: Design and optimization of a compact patch
antenna. In: ICMMT’07, pp. 1–3, April 2007
Chapter 9
The Effect of IV Characteristics
on Optical Control of SDR Si IMPATT
Diode

T.S.M. Arshad, M.A. Othman, M.N. Hussain and Y.A. Rahim

Abstract Effect of light incident on the Si SDR IMPATT diode is investigated in

this paper. The authors have used an IMPATT diode which is consists of p+, n+
(contact region), n-well region and p-sub region. Since the n-well region is used to
be the drift region of the structure, the light is shined on the top of the layer
through tiny hole created on the SiO2 layer. The results of the IV characteristics
are compared to the dark current condition which no light will be supply on top of
the structure. The result of the electric field and mobility in those two conditions
are also observed in this paper.

9.1 Introduction

IMPATT diode is known as one of the most powerful millimeter wave sources.
IMPATT stands for ‘Impact Ionization Avalanche Transit Time’ as it employ
impact ionization and transit time properties of its semiconductor structures to
produce negative resistance at microwave frequencies [1]. IMPATT diode has
various structures such as mesa, planer, distributed and so on. Those structures
consist of contact and drift region that are stacked perpendicular to the substrate

T.S.M. Arshad  M.A. Othman (&)  M.N. Hussain  Y.A. Rahim

Microwave Research Group, Faculty of Electronics and Computer Engineering, Centre for
Telecommunication Research and Innovation (CeTRI), Universiti Teknikal Malaysia
Melaka (UTeM), Durian Tunggal, Melaka, Malaysia
e-mail: [email protected]
T.S.M. Arshad
e-mail: [email protected]
M.N. Hussain
e-mail: [email protected]
Y.A. Rahim
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 85

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_9
86 T.S.M. Arshad et al.

surface [2]. Since this paper are focus on the optical control of IMPATT structure,
the lateral structure has been choose since this structure are more suitable in
coupling optical energy into the active region of the device. Compared to those
structures, lateral IMPATT structure has a unique design which is the location of
the contact and the drift region adjacent to the substrate [3].
The optical of microwave semiconductor devices have a large growing interest
because of their very wide bandwidth, the inherent high dc and reverse signal
isolation between the control and RF signals, and it also suitable to use with optical
fiber links [4]. The basic operation of semiconductor devices when it relates to the
direct optical control is the electron hole pairs (EHP) within the active region of
the device when sufficient light photon energy is absorbed. When the dimension of
depletion regions is modified, the main effect is to generate a photocurrent and
change the built-in potential of the junction. Hence, the conductivity of the
semiconductor material increases effect by the photoconductivity [4].
The authors proposed a lateral DDR IMPATT structure which suitable for
millimeter-wave optical interaction [2]. The structure is designed to operate at
94 GHz frequency by using standard 0.18 lm CMOS technology. The structure
consists of p+ and n+ layers (contact layers) and drift layers (p- and n-epitaxial
layers) that adjacent to the substrate. By using this type of structure, light can be
easily coupled to the drift layers through tiny holes created on the SiO2 layer [5].
The paper shows two type of optical control which on p+ layer (to generate
electron dominated photocurrent) and n+ layer (to generated hole dominated
photocurrent). By comparing those two types of optical control, the hole domi-
nated photocurrent has large value of efficiency than electron dominated
photocurrent.
An indium phosphate IMPATT diode under optical illumination at 94 GHz has
been proposed. The effect of the photo-generated predominated electron or hole
components of the leakage current on the admittance and negative properties are
also has been reported in the paper [6]. The result presented in the paper shows that
the frequency shift as well as the admittance and negative resistance characteristics
are more sensitive to hole-dominated photocurrent than to electron-dominated
photocurrent [6]. However, result in the paper [7] shows that the DC and high-
frequency parameters of the device are more sensitive to electron-dominated
photocurrent (Top Mount structure) compared to the hole-dominated photocurrent
(Flip Chip structure). It also observed that the avalanche response time decreases
as more number of photon are incident of the active region of the device [7].
The authors proposed a lateral SDR IMPATT diode in an earlier paper [8]
which will be use in this present paper. The simulation study will carry out to
investigate the optical control performance. This present paper will focus on the
effect of IV characteristics when dark current and has light incident. The authors
also have made an attempt in this paper to study the effect of the electric field and
mobility of the device in these two different conditions.
9 The Effect of IV Characteristics … 87

Fig. 9.1 Structure of SDR Si IMPATT diode

9.2 Design Structure

The structure proposed in this paper is design by using Silvaco TCAD ATHENA
tools. Athena provides a suitable platform for simulating processes used in
semiconductor industry which are ion implantation, diffusion, oxidation and so on
[9]. So, it is a convenient tool to design the proposed IMPATT diode since this
structure follows the standard fabrication process of complementary metal-oxide-
semiconductor (CMOS) technology.
Figure 9.1 show the SDR Si IMPATT diode consists of p+ and n+ layer
(contact layers), n-well region and p-sub region. The base material for this
structure is silicon. This structure is designed for a particular frequency (fd) from
transit time formula [9] given by

Wdep ¼ 0:37 vs =fd ð9:1Þ

where Wdep, vs and fd are total depletion layer width (n or p side), saturation
velocity of electrons/holes and design frequency respectively. At first step, the
doping concentration of n-well and p-sub region is initially chosen according to
88 T.S.M. Arshad et al.

Table 9.1 Design parameter for SDR Si IMPATT diode

Wn (lm) ND (m-3) Nn+, Np+ (m-3)
0.16 5 9 10e17 1 9 10e20

previous paper [2–4]. Later the doping concentration and design parameter for
both regions are adjusted to get the corresponding frequency design. Table 9.1
show the structure design of SDR Si IMPATT diode.

9.3 Simulation Result and Discussion

9.3.1 Dark Current Condition

This paper is focus on the effect of the current-voltage characteristics (IV char-
acteristics) of the device in two different conditions which is in dark current (no
light incident) and in light incident. Since IMPATT diodes are operated in reverse
biased voltage, the simulation process supplies negative voltage to anode contact.
By using the same structure that show in Fig. 9.1, the IV characteristic is obtained
from the simulation process shown in Fig. 9.2. The result is for dark current
condition which is no light incident on the surface of structure. From the graph, it
shows that the breakdown voltage is -5 V and it produce value of current in range
of -6e-14. The small value of current is because it is in reverse biased condition.
Figure 9.3 show the electric field of the structure when there’s no light supplied
on the structure. The value for the electric field is in the range from -83.5679 to -
91272.6 V/cm. The values of electric field are shown in Table 9.2.
The simulation also has carry out the result for e- mobility and h+ mobility of
the structure. The results are shown in Fig. 9.4. From the graph, it shows that the
electron mobility are higher than hole mobility. From here [10], it state that the
mobility of electrons and hole are normally in less or equal to 1,400 and 450 cm2/
Vs respectively. Therefore, the values from the result are in same range since it
used silicon for the structure. The values plotted in Fig. 9.4 are summarized in
Table 9.2. Later, the value in Table 9.2 will be compared to the value in the next
section which is simulated with light incident.

9.3.2 Light Incident Condition

Based on the structure in previous section, the light energy can be easily coupled
into n-well region through tiny holes created on the SiO2 which is located on the
top of the structure. It can be shown in Fig. 9.5. From previous paper [11], there
are two types of optical control which either on p+ layer or n+ layer. If light energy
9 The Effect of IV Characteristics … 89

Fig. 9.2 IV characteristics

for SDR Si IMPATT diode
(dark current)

Fig. 9.3 Electric field for

SDR Si IMPATT diode (dark
current)

Table 9.2 Electric field, e-mobility and h+ mobility for Si IMPATT diode (no light)
Voltage (V) Electric field (V/cm) e- mobility (cm2/Vs) h+ mobility (cm2/Vs)
0 -83.5679 449.435 252.094
-1 -120.611 449.432 251.867
-2 -254.553 449.411 251.049
-3 -535.642 449.317 249.348
-4 -1283.64 448.742 244.934
-5 -3393.47 444.636 233.286
-6 -6979.23 430.117 215.84
-7 -13159.1 390.217 191.198
-8 -27674.0 287.471 150.769
-9 -57083.8 168.137 105.548
-10 -91272.6 109.945 78.2606
90 T.S.M. Arshad et al.

Fig. 9.4 Mobility for SDR

Si IMPATT diode (dark
current)

Light incident

Fig. 9.5 Structure of SDR Si IMPATT diode (light incident)

is supply to the p+ layer, it will generate electron dominated photocurrent and if

the light energy supply to the n+ layer, it will produce hole dominated photo-
current [5].
9 The Effect of IV Characteristics … 91

Fig. 9.6 IV characteristics

for SDR Si IMPATT diode
(light incident)

In this section will be presented on the effect of IV Characteristics in light

incident condition. Since the light will be incident on the top of n-well layer (drift
layer), the photo-generated current will be dominated by hole. From the equation
given below,

kg ¼ hc = Eg ð9:2Þ

where h = 6.625 9 10-34 Js, c = 3 9 108 ms-1, the corresponding cut off
wavelength can be obtained which is kg = 1.118 lm since the bandgap of Si at
300 K is Eg = 1.111 eV [12]. Therefore, the simulations use the wavelength of
light less or equal to 1.118 lm for the absorption of light energy by the Si
structure. This process will generate electron-hole pairs (EHP) in Si structure.
The simulations for light incident condition start with the observation of IV
characteristics. The same value of voltage supply from previous section is used
which is -10 V. From the graph from Fig. 9.6, it shows breakdown voltage of the
structure is -3.7 V. From the observation, the presence of light reduced the
voltage breakdown to produce conductivity.
The presence of light on semiconductor devices is exactly similar to the effect
of temperature or heat on semiconductor devices. Light energy causes electrons to
break their covalent bonds and then generated EHP [13]. The conductivity of the
device is increase since the light energy decrease the resistance of the structure.
Compared to the dark current condition in previous section, the resistance is high
because there are few free electrons in the device. The resistance in this condition
is called dark resistance [13]. So, it takes more times to generate EHP and produce
conductivity (Table 9.3).
Figure 9.7 shows the graph of electric field of the device in presence of light.
The value of electric field is almost similar compared to previous section. This is
because the value of voltage supply uses in these two different conditions is same.
The observation of electron and hole mobility can be shown in Fig. 9.8. There are
also slightly similar values of these two mobility compare to previous section.
92 T.S.M. Arshad et al.

Table 9.3 Electric field, e-mobility and h+ mobility for Si IMPATT diode (with light)
Voltage (V) Electric field (V/cm) e- mobility (cm2/Vs) h+ mobility (cm2/Vs)
0 -79.4343 449.436 252.120
-1 -96.1578 449.434 252.017
-2 -175.775 449.425 251.529
-3 -381.337 449.377 250.279
-4 -812.410 449.159 247.697
-5 -2047.64 447.672 240.584
-6 -4896.17 439.611 225.643
-7 -9674.43 414.355 204.354
-8 -19067.0 346.172 172.383
-9 -41523.6 217.968 125.457
-10 -75877.4 130.518 88.5718

Fig. 9.7 Electric field for

SDR Si IMPATT diode (light
incident)

Fig. 9.8 Mobility for SDR

Si IMPATT diode (light
incident)
9 The Effect of IV Characteristics … 93

There are only slightly increased for mobility in light incident because the light
energy causes the ions vibrating with sufficient movement to impede the move-
ment of electrons. For this cases, the light that supplied on the structure give small
effect to the lattice scattering to increase and the same time it is not affect to the
reduction of mobility.

9.4 Conclusion

In this paper, the authors have made an attempt to investigate the effect of photo
incident on IV characteristics of SDR Si IMPATT diode. This paper also has
shown the result of electric field and mobility of the structure. The results are
compared with dark current condition. From the IV curve, it shows that the
breakdown voltage in light incident condition are faster than dark current condi-
tion. This is because the presence of light is similar to the effect of temperature.
When light is supply on the structure, it causes electron to break their covalent
bonds. With the energy of light, it will generate EHP and increase its conductivity.
So, in light incident condition, it required few voltage supplies to start the con-
duction process. So, for the conclusion, the optical control is affect to the IV
characteristics of the devices.

Acknowledgments The authors wish to thank Dato’ Prof Dr. Mohd Nor Bin Husain for the
professional guidance and useful critiques of this research work. I would also like to extend my
thanks to Mr. Mohd Azlishah Bin Othman for the suggestion and patient guidance of this work.
My thanks must also go to Mr. Yahaya bin Abdul Rahim (PJP/2012/FTMK(56D)S1063) and
Mr. Zul Atfyi Fauzan bin Napiah (PJP/2012/FKEKK(13C)S01116) for their valuable support.

References

1. Srivastava, G.P., Gupta, V.L.: Microwave Devices and Circuit Design, p. 294. PHI Learning
Pvt. Ltd, New Delhi (2006)
2. Acharyya, A., Banerjee, J.P.: A proposed lateral DDR IMPATT structure for better
millimeter-wave optical interaction. In: International Conference on Devices, Circuits and
Systems, March 2012, pp. 599–602
3. Stabile, P.J., Lalevic, B.: Lateral IMPATT diodes. IEEE Electron Device Lett. 10(6),
249–251 (1989)
4. Seeds, A.J., Augusto, A.: Optical control of microwave semiconductor devices. IEEE. Trans.
Microw. Theory Tech. 38(5), 577–585 (1990)
5. Acharyya, A., Banerjee, S., Banerjee, J.P.: Optical control of millimeter-wave lateral double-
drift region silicon IMPATT device. Radioengineering 21(4), 1208–1217 (2012)
6. Banerjee, J.P., Mukherjee, R.: Effect of electron- and hole-dominant photocurrent on the
millimetre wave properties of an indium phosphide IMPATT diode at a 94 GHz window
under optical illumination. Semicond. Sci. Technol. 9, 1690 (1994)
94 T.S.M. Arshad et al.

7. Acharyya, A., Banerjee, J.P.: Dependence of avalanche response time on photon flux incident
on DDR silicon IMPATT devices. In: The 32nd PIERS in Moscow, Russia, pp. 867–872,
August 19–23, 2012
8. Arshad, T.S.M., Othman, M.A., Yasin, N.Y.M., Taib, S.N., Napiah, Z.A.F.M., Hussain,
M.N., Rahim, Y.A., Pee, A.N.C., Ismail, M.M., Misran, M.H., Said, M.A.M., Sulaiman,
H.A., Ramlee, R.A.: Variable junction temperature analysis in silicon IMPATT diode. In:
2013 3rd International Conference on Instrumentation, Communications, Information
Technology, and Biomedical Engineering (ICICI-BME), pp. 76,79, 7–8 Nov. 2013
9. SILVACO TCAD ATHENA, http://www.silvaco.com/products/tcad/process_simulation/
athena
10. Electronic Achieve: New Semiconductor Materials, Characteristics and Properties. http://
www.ioffe.ru/SVA/NSM/Semicond
11. Vyas, H.P., Gutmann, R.J., Borrego, J.M.: The effect of hole versus electron photocurrent on
microwave—Optical interactions in IMPATT oscillators. IEEE Trans. Electron Device.
26(3), 232–234 (1979)
12. Varshni, Y.P.: Temperature dependence of the energy gap in semiconductor. Physica 34,
149–154 (1967)
13. Bakshi, U.A., Godse, A.P.: Semiconductor Devices & Circuits, p. 25. Technical Publications,
India (2008)
Chapter 10
Variable Intrinsic Region in CMOS PIN
Photodiode for I–V Characteristic
Analysis

M.A. Othman, N.Y.M. Yasin, T.S.M. Arshad, Z.A.F.M. Napiah,

M.M. Ismail, H.A. Sulaiman, M.H. Misran, M.A. Meor Said
and R.A. Ramlee

Abstract In this paper presented an investigation on I–V characteristic for CMOS

PIN Photodiode. PIN diodes are widely used in optics and microwave circuits as it
acts as a current controlled resistor at these frequencies. PIN diode performance is
greatly influenced by the geometrical size of the device, especially in the intrinsic
region. Two different I-layer thickness of PIN diode structure has been designed
using Sentaurus Technology Computer Aided Design (TCAD) tools. The I-layer

M.A. Othman
N.Y.M. Yasin (&)
T.S.M. Arshad
Z.A.F.M. Napiah
M.M. Ismail

H.A. Sulaiman
M.H. Misran
M.A. Meor Said
R.A. Ramlee
Faculty of Electronic and Computer Engineering, Centre for Telecommunication Research
and Innovation (CeTRI), Universiti Teknikal Malaysia Melaka, Hang Tuah Jaya,
76100 Durian Tunggal, Melaka, Malaysia
e-mail: [email protected]
M.A. Othman
e-mail: [email protected]
T.S.M. Arshad
e-mail: [email protected]
Z.A.F.M. Napiah
e-mail: [email protected]
M.M. Ismail
e-mail: [email protected]
H.A. Sulaiman
e-mail: [email protected]
M.H. Misran
e-mail: [email protected]
M.A. Meor Said
e-mail: [email protected]
R.A. Ramlee
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 95

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_10
96 M.A. Othman et al.

thickness (or width) is varied from 4 to 8 lm in order to investigate its effects on

the current-voltage (I–V) characteristics. These structures were design based on
CMOS process.

10.1 Introduction

Nowadays a lot of microwave and optical devices have been design by acade-
mician, scientist and engineers. One of two terminal devices that get intention is
PIN diode. PIN diodes name attribute to overall their structure where P is P type
layer, I intrinsic layer and N is N type layer. The intrinsic layer is interesting layer
since this layer makes the PIN diodes application comes to be as attenuators, RF
switches and photodiode. Hence, PIN diodes are used extremely in RF and
microwave applications due to ability to control the magnitude and phase of the
signals [1, 2]. In addition, the ability to control RF and microwave signals while
using a smaller level of dc excitation makes the PIN diode suitable for attenuating,
limiting, phase shifting, modulating and microwave switching.
The I-layer or intrinsic layer is the one that gives a changes in-term of prop-
erties comparing with PN diode. The intrinsic layer comprises of undoped or
virtually undoped semiconductor and in most PIN diodes and very thin between
1 lm up to 200 lm. PIN photodiode is a special case of the PN junction photo-
diode, in which a large intrinsic or lightly doped N semiconductor area is inserted
in between the P and N region as shown in Fig. 10.1.
It is a well-known fact that in any PN diode, the depletion region extends more
into the lightly-dope N region than into the heavily doped P region. This is
because, in the heavily doped region, the number of free charge carriers available
for conduction is quite large compared to that in the lightly doped region [3]. In a
normal operation a reverse-bias voltage is applied across the device so that no free
electrons or holes exist in the intrinsic region (Fig. 10.2).
Electrons in semiconductor materials are allowed to reside in only two specific
energy bands. A forbidden region called energy gap separates the two allowed band.

Fig. 10.1 PIN photodiode

structure
10 Variable Intrinsic Region in CMOS PIN Photodiode … 97

Fig. 10.2 (a) PIN photodiode operated in reverse bias, hv is the energy of radiation, V is the bias,
and RL is the local load resistance (b) IV characteristics for a pin photodiode (1) with no light (2)
with light

The energy difference between the top and bottom bands is referred to as the band
gap energy [4]. In a generic photodiode, light enters the devise through a thin layer
whose absorption typically causes the light intensity to exponentially drop with
penetration depth. For enhanced performance it is often necessary for the device to
have a shallow junction followed by a wide depletion region where most of the
photon absorption and electrons holes generation should take place.
The magnitude of the generated current is proportional to the intensity of the
incident light. It can be used as a photon detector by operating it in the third
quadrant of its electrical current voltage (I–V) characteristics. PIN photodiode is
usually functioning by applying a reverse-bias voltage. The magnitude of the
reverse-bias voltage depends on the photodiode purpose, but naturally is less than
a few volts. While no light is incident on the photodiode, a current is still produce
[5]. The objectives of this paper are to study the theory behind PIN photodiode, to
design and analyze PIN-Photodiode and to determine characteristic of PIN-
Photodiode.

10.2 Design Procedure

The simulations were performed using Silvaco ATLAS. Silvaco ATLAS is a

TCAD product that is physics based on modeling system. This allows TCAD to
predict device performance based upon equations, which describe the physics
within the structure of the device [6].
The structure under consideration of lightly doped with 5.e16 for N-epi doping,
1e20 for N+ doping, 1e20 for P+ doping, sandwiched with I+ doping with con-
centration 1e21. The incident light absorbed in neutral P+ region and neutral
N+ region as well as in the depleted N- region. All the calculations presented in
98 M.A. Othman et al.

Fig. 10.3 Intrinsic with

2 9 8 lm

8µm I+

2µm

Fig. 10.4 Intrinsic with

2.4 9 8 lm

8µm
I+

2.4µm

this work have been obtained with ATLAS software of SILVACO unless stated
otherwise. The intrinsic region size had been varied with 2 different sizes which is
4 and 8 lm in width.
ATLAS enables device technology engineers to simulate the electrical, optical,
and thermal behavior of semiconductor devices. Different models such as conmob,
fldmob, srh, auger and bgn are used for effective implementation of changes
occurring related to radiation damage [7] (Figs. 10.3, 10.4).
10 Variable Intrinsic Region in CMOS PIN Photodiode … 99

Fig. 10.5 Intrinsic with

2.4 9 8 lm

Without light

0.75e-18
With light

As shown in the figure, the intrinsic width for both structures had been
increased to 0.4 lm. By capturing most of the light in the depletion region
(intrinsic layer), it will effect to improve the responsivity of the device. To fully
benefit from the intrinsic layer or depletion region, the PIN Photodiode normally
requires an applied reverse bias that ensures a depletion region extension all the
way through this layer. Figures 10.5, 10.6 and 10.7 shows the simulated I–V
characteristic of Silicon PIN Photodiode for the structure that had been created.
The simulated results of p+, i+, n+ photo detector are obtained by developing a
program in DECKBUILD window, interfaced with ATLAS, simulation software
of SILVACO and TonyPlot.
In a generic photodiode, light enters the device through a thin layer. For
Fig. 10.5, it shows the difference of IV characteristic when supply light and
without supply light for 2.4 9 8 lm. The IV characteristic of a photodiode with no
incident light is similar to a rectifying diode. When the photodiode is forward
biased, there is an exponential increase in the current. When a reverse bias is
applied, a small reverse saturation current appears. The light has the effect of
shifting the IV Characteristics down into the fourth quadrant where power can be
extracted from the photodiode.
Figure 10.6 show the difference of IV characteristic when supply light and
without supply light for 2 9 8 lm. Basically, it use the same theory with the result
show in Fig. 10.5, the difference is the size of depletion use based on the structure.
For IV Characteristic in Fig. 10.5, it uses huge depletion size compared to IV
Characteristics in Fig. 10.6. That is why the effect of shifting of IV Characteristics
100 M.A. Othman et al.

Fig. 10.6 Intrinsic with

2.4 9 8 lm
1e-18

-1.125e-18

Fig. 10.7 Voltage

Breakdown for 2 9 8 lm and
2.4 9 8 lm
Voltage Breakdown for 2.4µm x 8µm

Voltage Breakdown for 2µm x 8µm

-3.5 -3

from the photodiode is smaller compared to IV Characteristic in Fig. 10.5 due to

difference structure used.
Based on Fig. 10.7, it shows the voltage breakdown for 2 9 8 lm and
2.4 9 8 lm. There is a difference for the voltage breakdown between each
structure. Voltage breakdown for intrinsic for 2.4 9 8 lm is faster compared to
Voltage breakdown for intrinsic for 2 9 8 lm. This is due to the depletion region
in the structure since the size of intrinsic is not the same.
10 Variable Intrinsic Region in CMOS PIN Photodiode … 101

Based on graph in Figs. 10.5 and 10.6, the shape of graph had a little bit
difference in their variance. This is due to the size of the mesh that had been varied
since the size of intrinsic layer had been changed. Thus, the mesh for Intrinsic with
2 9 8 lm becomes closer and that is why variance appeared.

10.3 Conclusion

In this study, in order to observe the width of intrinsic or depletion region effect on
the IV Characteristics performance, the two-dimensional silicon PIN photodiode
with varying width has been simulated using the SILVACO TCAD tools. Through
the simulations, it shows that there is a difference occurs during voltage break-
down for difference intrinsic size. PIN Photodiode give effect when light supplied
to the photodiode by shifting the IV Characteristics down. In general, PIN pho-
todiode need small voltage supply to have a better I–V performance (voltage
breakdown).

Acknowledgments Authors would like to thank Centre for Telecommunication, Research and
Innovation (CeTRi), Universiti Teknikal Malaysia Melaka for their equipment, assistance support
and also financing the conference’s paper.

References

1. Yashchyshyn, Y.: Reconfigurable antennas by RF switches technology. In: 5Ih International

Conference on Perspective Technologies and Methods in MEMS Design, pp. 155–157, April
2009
2. Yang, J.G., Kim, M., Yang, K.: An InGaAs PIN-diode based broadband travelling-wave
switch with high-isolation characteristics. In: IEEE International Conference on Indium
Phosphide and Related Materials, pp. 207–209, May 2009
3. Somanathan Nair, B.: Electronic Devices and Applications, 3rd edn, pp. 330, Sept (2006)
4. Keiser, G.: Optical Communications Essentials. McGraw-Hill, NY pp. 108 (2003)
5. Piprek, J.: Optoelectronic Devices: Advanced Simulation and Analysis. Springer Media and
Business, Berlin, pp. 382 (2005)
6. Maiti, C.K., Armstrong, G.A.: TCAD for Si, SiGe and GaAs Integrated Circuits. The Inst. Eng.
Technol. 47 (2007)
7. Vasileska, D., Goodnick, S.M.: Computational Electronics. Morgan and Claypool Publishers,
US pp. 98 (2006)
Chapter 11
Variable Depletion Region in CMOS PN
Photodiode for I–V Characteristic
Analysis

M.A. Othman, T.S.M. Arshad, Z.A.F.M. Napiah, M.M. Ismail,

N.Y.M. Yasin, H.A. Sulaiman, M.H. Misran, M.A. Meor Said
and R.A. Ramlee

Abstract In this paper, CMOS PN photodiode will be design and analyze for the
application at 5 GHz optical communication. The paper will be divided in several
section; the theory of CMOS PN photodiode and design with analysis of I–V
characteristics of PN photodiode. A better understanding of the operation will be
investigated through this. The PN photodiode will be design using Silvaco TCAD

M.A. Othman (&)  T.S.M. Arshad  Z.A.F.M. Napiah  M.M. Ismail  N.Y.M. Yasin 
H.A. Sulaiman  M.H. Misran  M.A. Meor Said  R.A. Ramlee
Faculty of Electronic and Computer Engineering, Centre for Telecommunication Research
and Innovation (CeTRI), Universiti Teknikal Malaysia Melaka, 76100 Hang Tuah Jaya,
Durian Tunggal, Melaka, Malaysia
e-mail: [email protected]
T.S.M. Arshad
e-mail: [email protected]
Z.A.F.M. Napiah
e-mail: [email protected]
M.M. Ismail
e-mail: [email protected]
N.Y.M. Yasin
e-mail: [email protected]
H.A. Sulaiman
e-mail: [email protected]
M.H. Misran
e-mail: [email protected]
M.A. Meor Said
e-mail: [email protected]
R.A. Ramlee
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 103

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_11
104 M.A. Othman et al.

and will be characterize and experimental in I–V Characteristic. The effects of I–V
characteristic will be analyzed in term of changes the width and light. Further
understanding of I–V characteristic will be presents in this paper.

11.1 Introduction

PN junction is one of the most simple of all semiconductor devices because it is

much faster and very small in size and weight. It is the only most important device
in the studies of modern semiconductors. It is the heart of the most photocells,
rectifiers and transistors. It is also cheaper and has greater sensitivity [1].
PN junction forms a diode, and consequently a junction used as a photodetector
is frequently called a photodiode [2]. It is formed by doping donor atoms on one
side (N-side) and doping acceptor atoms on the other (P-side). Figure 11.1 shows
the structure of PN photodiode [3]. In this paper, the authors concentrate on the
geometry of PN photodiode using different size of P+ region and its effects toward
the I–V characteristics PN photodiode. The structure of PN photodiode is simu-
lated with different size of P+ region and the analysis of I–V characteristic, which
is use Silvaco TCAD tools discussed in the next topic.
The PN junction operated in the reverse biased condition. The PN photodiode
operation involving light and the light will be directed fall on the upper top of the
portion of the photodiode. The depth where photons come into the depletion region
depends on the wavelength of incident light. Electron-hole pairs will exist in the
depletion region when the absorbed optical energy is sufficient. Electrons and
holes are now attracted towards the opposite respective terminals of the battery.
EHP that formed outside the depletion region will produce a photocurrent [4].
When a reverse bias voltage is applied to the PN junction, more electrons and
holes are attracted to the contact. As a result, more donor and acceptor ions appear
at the depletion region which in turn increases as well [5] (Figs. 11.2 and 11.3).

11.2 Design Procedure

In this project, SILVACO TCAD tool is used to design the PN photodiode and
simulate the I–V characteristic. ATLAS is a physically based device simulator,
providing general capabilities for 2D and 3D simulation of semiconductor devices
[6]. In the Atlas project, it divided into 2 parts, which is the structure and the I–V
curve simulation. In the structure part, mesh is needed to generate meshes that are
suitable for the design of PN photodiode. There are some methods and model that
used in the second part of the coding. It will use for the simulation of I–V curve.
11 Variable Depletion Region in CMOS PN Photodiode … 105

Fig. 11.1 PN junction

photodiode

Fig. 11.2 I–V Characteristic

of PN photodiode

Fig. 11.3 I–V

Characteristics with different
light level

11.3 Result and Discussion

Figures 11.4 and 11.5 show the structure of the PN photodiode using Atlas tools.
Two heavily doped P region and N region with doping concentration 1e + 20 is
located beside each other. Two contacts for PN photodiode called anode and
106 M.A. Othman et al.

Fig. 11.4 PN Photodiode

with width, W = 1 lm

Fig. 11.5 I–V curve for PN

Photodiode with width,
W = 1 lm

cathode is located at the top of P-type and N-type region. It may be noted that
N-type and P-type region is formed within the n-well region. There is also a p-sub
region which is located below n-well.
The reduction of depletion region width with increased the electric field and
reduced breakdown voltage. The breakdown voltage of the structure with width,
W = 3 lm is 4.9 V while the breakdown voltage for the structure with width,
11 Variable Depletion Region in CMOS PN Photodiode … 107

Fig. 11.6 PN Photodiode

with width, W = 3 lm

W = 1 lm is 4.7 V. The applied electric field creates the depletion region on

either side of the PN junction. Carriers—free electrons and holes—leave the
junction area. The electric field and potential in the depletion region can affect
the breakdown voltage. It can show from the equation below:

Em W 2S Em2
VB ¼ ¼ ðNB Þ1 ð11:1Þ
2 2q

From the Eq. (11.1), VB is proportional to the W. As can be expected, reduce of

depletion region can generate the small voltage breakdown. The structure that only
uses less voltage supply will start the operation of reverse biased faster than the
structure, which uses more voltage supply.
In reverse bias, the number of free carriers in the device depends primarily on
the current in the device and the voltage is clamped at the knee voltage of the
diode. Because of the I–V curve for the photodiode in reverse bias is exponential, a
small change in V has a much larger impact to the device [7].
For the analysis of presence of light, the second structure is chose, shown in
Fig. 11.6. From the analysis, some photons are incident on the surface of PN
photodiode. Since the photon will pass through it and strike the junction. Light that
is absorbed by the photodiode produces current flow through the entire external
circuit (Fig. 11.7).
In the analysis, two different width of PN photodiode have been tested in the
device structure which are 1 and 3 lm. By comparing both structures, the different
of width is change due to the different size of P+. Since the size of P+ reduced, the
width will also change. The I–V Curve is compiled together as it is easier to view
the trend of voltage breakdown versus the width depletion region. Figure 11.8
illustrates the simulation for both structures. From the result, it is shown that both
108 M.A. Othman et al.

Fig. 11.7 I–V curve for PN

Photodiode with width,
W = 3 lm

Fig. 11.8 I–V curve of PN

Photodiode with different
width

photodiode are start from V = 0 V and operated at the different reverse bias
voltage.
The depth at which the photons reach into the depletion region depends on the
incident of light. Figure 11.9 show the I–V curve with no light falling on the PN
photodiode. When no light is falling on the photodiode, a very small current passes
through the photodiode. This current is basically due to the reverse bias applied to
the PN photodiode.
Figure 11.10 show the I–V curve with light falling on the PN photodiode. As
the light is incident on the photodiode, photocurrent is developed. This photo-
current gets increased by increasing the light intensity.
11 Variable Depletion Region in CMOS PN Photodiode … 109

Fig. 11. 9 I–V curve for PN

Photodiode without light

Fig. 11.10 I–V curve of PN

Photodiode with presence of
light

11.4 Conclusion

In this paper, the different width of depletion region effects of the PN photodiode
on its I–V characteristics has been successfully simulated and verified. The
presence of light also affects the I–V curve of the PN photodiode. In order to
perform the I–V curve, the SILVACO TCAD is used. The result is compared
110 M.A. Othman et al.

between the different of structure. It is proved that the smaller width of depletion
region, the better the device. From this paper also can conclude that the photo-
current gets increased by increasing the photon energy or light intensity.

Acknowledgments The authors would like to thanks Universiti Teknikal Malaysia Melaka,
Malaysia (UTeM) for their equipment, financial and assistance support. Last but not list, the
author would like to thanks those involved direct and indirect in completing this project success
fully.

References

1. Chiang, C.Y.-T., Yeow, Y.T.: Inverse modelling of two-dimensional MOSFET dopant profile
via capacitance of the source/drain gated diode. IEEE Trans. Electron Devices 47(7),
1385–1392 (2000)
2. Alexander, S.B.: Optical Communication Receiver Design. SPIE Press, Bellingham (1997)
3. Nair, B.S.: Electronic Devices and Application, p. 329. Prentice-Hall India, New Delhi (2006)
4. Chou, F., Wang, C., Chen, G., Sin, Y.: An 8.7 GHz Si photodiode in standard 0.18-lm CMOS
technology. In: OECC, pp. 826–827 (2010)
5. Mohamad, M., Jubadi, W.M., Tugiman, R., Zinal, N., Zin, R.M.: Comparison on I–V
performances of Silicon PIN diode towards width variations. In: IEEE ICSE, pp. 12–14 (2010)
6. Silvaco: Atlas User’s Manual: Device Simulation Software. Silvaco International, Santa Clara
(2000)
7. Sze, S.M.: Physics of Semiconductor Devices, 3rd edn. Wiley, New York (2007)
Chapter 12
U-Slot Rectangular Patch Antenna
for Dual Band Application

Mohammad Shawkat Habib, I.M. Rafiqul, Khaizuran Abdullah

and M. Jamil Jakpar

Abstract Dual and multi-band rectangular microstrip antennas can be realized by

cutting U-slots inside the patch. In this paper, the length and width of U-slots are
optimized in order to achieve dual-band and multi-band operation. Computer
Simulation Technology (CST) software was used to design, simulate and opti-
mization of antenna. Two resonant frequencies at 1.8 and 2.4 GHz were found
with reasonable gain. Additional resonant frequencies could also be achieved from
2.8 to 3.0 GHz using the similar approach.

Keywords Patch antenna  U-slot  Dual band

12.1 Introduction

Wireless local area network (WLAN) is one of the most important applications of
the advancing wireless communication technology. Developed by the Institute of
Electrical and Electronics Engineers (IEEE) and the 802.11 standard the wireless
local-area network (WLAN) standard is a family of specifications for WLAN
technology [1–3]. Most of the wireless devices are integrated with IEEE WLAN
functionalities [4–6]. The emerging market of wireless devices like the laptops,
tablet pc etc. has set off notable research activities on the design of cost-effective,
multi band yet simple antennas. With the benefits of having low manufacturing

M.S. Habib  I.M. Rafiqul (&)  K. Abdullah  M.J. Jakpar

Department of Electrical and Computer Engineering, International Islamic University
Malaysia, Jalan Gombak, 53100 Kuala Lumpur, Malaysia
e-mail: [email protected]
M.S. Habib
e-mail: [email protected]
K. Abdullah
e-mail: [email protected]

Ó Springer International Publishing Switzerland 2015 111

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_12
112 M.S. Habib et al.

cost and compatible in size, the planar antennas are good choice for the majority of
the wireless LAN stations both on subscriber end and base station side. This paper
presents a patch antenna with U shaped slot resonant at 2.4 GHz for WLAN
application and 1.8 GHz for cognitive radio application.
Dual-band and multi-band rectangular microstrip antennas are realized by
cutting U-slots, V-slots, or a pair of rectangular slots inside the patch. The tech-
nique for designing dual-band microstrip antenna is to cut slots of different shapes
at an appropriate position inside the rectangular patch [7–9]. Since the slots are cut
inside the microstrip antenna, they neither increase the patch size nor significantly
affect the radiation pattern of the antenna. When the slots are cut very close to the
radiating edge of the microstrip antenna, they alter the third-order-mode resonance
frequency of the patch and, along with the fundamental mode; result in a dual-band
response [10]. By integrating four slots inside the patch, a nine-band antenna,
covering various cellular and TV bands, was reported in Ref. [11]. The analysis for
studying the effects of a U-slot on the broadband or the dual-band response in a
rectangular microstrip antenna was reported in Ref. [12].
In most of the design, depending upon where the slot is cut, the slot length is
taken to be equal to either a quarter-wavelength or a half-wave length. However,
these simpler approximations of slot length as a function of frequency do not give
a close match for different slot lengths and their positions inside the patch. The
surface currents and voltage distributions for a dual-band U-slot-cut on rectangular
microstrip antennas are studied over a wide frequency range. It was observed that
the slot does not introduce any mode, but reduces the higher-order orthogonal
mode resonance frequency of the patch and, along with the fundamental mode,
realizes the dual-band response. In this paper, formulation proposed by [13, 14] for
U-slot were utilized and an antenna has been designed for 1.8 and 2.4 GHz dual-
band applications with reasonable gain. The technique has been extended to design
third resonance also at 2.95 GHz. Results are obtained by using Computer Sim-
ulation Technology (CST) software.
Section 12.2 describes the parameter values that were considered in the final
design of the antenna. The simulated results of the final design were produced in
Sect. 12.3. Section 12.4 overviews the fabrication of the antenna and its charac-
teristics which were further contrasted with the simulated results in Sect. 12.5. The
reason for the slight variation of the simulated and the fabricated result are also
discussed in the later part of Sect. 12.5.

12.2 Antenna Design

The length, width, return loss, VSWR of the patch antenna can be calculated from
the Eqs. (12.1)–(12.6) narrated in Ref. [15]. Where L and W are the length and
width of the patch, c is the velocity of light, er is the dielectric constant of sub-
strate, h is the thickness of the substrate, fo. is the target center frequency, ee is the
effective dielectric constant and q is the radiation coefficient.
12 U-Slot Rectangular Patch Antenna for Dual Band Application 113

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
s 
c er þ 1
W¼ ð12:1Þ
2fo 2

c
L¼ pffiffiffiffi  2Dl ð12:2Þ
2fo er
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi
1 1 10h
e e ¼ ð e r þ 1Þ þ ð e r  1Þ 1þ ð12:3Þ
2 2 W
 
ðee þ 0:3Þ Wh þ 0:8
Dl ¼ 0:412h W  ð12:4Þ
ðee  0:258Þ h þ 0:8

1þq
VSWR ¼ ð12:5Þ
1q
 
1
Return Loss ¼ 10 log ð12:6Þ
q2

While designing the antenna some of the parameters were varied and the effects
were observed in order to tune the antenna resonant frequency to the desired
configuration. With different values of Lv and Lh the return loss was observed. The
variation of the vertical slot cut Lv was observed and the variation in return loss is
displayed in Fig. 12.1. Lv was varied from 20 to 50 mm with a step of 10 mm
keeping the horizontal slot cut Lh constant at 30 mm. It was observed that with the
increase in value of Lv the return loss increases till Lv = 40 mm, and at

Fig. 12.1 Variation of return loss (S11) when Lh kept constant at 30 mm and Lv is varied from
20–50 mm
114 M.S. Habib et al.

Fig. 12.2 Front view of

U-slot rectangular microstrip
antenna

Lv = 50 mm the value of return loss is less than -24 dB. Moreover the resonant
frequency near to 2.4 GHz shifts left with the increase of Lv.
The final microstrip antenna has been designed by using a patch of length
80 mm and width of 100 mm whose top view is displayed at Fig. 12.2. The patch
is having a U-slot craved within it, and the dimensions are 31 and 20 mm for
horizontal and vertical slots respectively. The substrate used is of FR-4 with a
dielectric constant of 4.3 and the length and width is twice the dimension of the
copper patch and the thickness is of 1.6 mm.
The bottom of the substrate is ground layered with copper with a thickness of
0:1 mm and the dimension is same as the substrate’s dimension and the assumed
impedance of the line is 50 ohms. The patch antenna with the mentioned design is
simulated with the CST Microwave Studio and required results are obtained. The
detailed parameters are provided in the Table 12.1.

Table 12.1 Designed parameters of U-slot dual band microstrip antenna for 1.8 and 2.4 GHz
Parameter Value (mm)
Length, L 80
Width, W 100
Feed slot cut, Fi 12.5
Horizontal slot cut, Lh 31
Vertical slot cut, Lv 20
Distance of the starting U-slot from the middle, Xf 25
Feed line width, Wf 8.7
Substrate thickness, h 1.6
12 U-Slot Rectangular Patch Antenna for Dual Band Application 115

Fig. 12.3 Simulated return losses for the U-slot microstrip patch antenna

12.3 Simulation Result

Utilizing CST Microwave Studio the proposed design has been simulated and the
obtained results are plotted in Fig. 12.3. The desired resonant frequencies were at
1.8 and 2.4 GHz and was obtained when the design parameters were Lh = 31 mm
and Lv = 20 mm. At the first resonant frequency f1 = 1.8 GHz, the return loss is
equal to -35 dB and at the second resonant frequency f2 = 2.4 GHz, the return
loss is equal to -20 dB.
The radiation patterns of the antenna at 1.8 and 2.4 GHz are presented in
Fig. 12.4a, b respectively. The antenna directivity for both frequencies of 1.8 and
2.4 GHz from the simulation output is presented in Table 12.2. Both of the frequency
bands display effective results in terms of antenna gain which is above 5.0 dBi.

Fig. 12.4 Radiation pattern of a 1.8 GHz and b 2.4 GHz U-slot microstrip antenna
116 M.S. Habib et al.

Table 12.2 Antenna characteristics

Frequency 1.8 GHz 2.4 GHz
Main lobe magnitude 8.0 dBi 5.6 dBi
Main lobe direction 40.0° 0.0°
Angular width (3 dB) 55.7° 45.9°
Side lobe level -9.3 dB -17.3 dB

Fig. 12.5 Current distribution at 1.8 GHz

Current distribution of the proposed antenna is shown in Figs. 12.5, 12.6 and
12.7. Arrow sign is used to indicate current distribution. From the figures, it can be
easily observed that current flow is the maximum to the microstrip line that is used
as a feeding technique and to the insidious part of the U-slot along with the corners
of the patch near to the feed line for the 1.8 GHz resonant frequency in Fig. 12.5.
For the 2.4 GHz resonant frequency in Fig. 12.6, near the feed line connection of
the patch, the current density is found to be the maximum. In 2.95 GHz resonant
frequency, the current distribution is at its maximum inside the U-slot and around
the two slots close to the connection of the feed line displayed in Fig. 12.7.

12.4 Fabrication and Test Result

The antenna being designed in the simulation section has been fabricated. The
artwork design used in order to fabricate the antenna is shown in Fig. 12.8a. After
following different steps for the fabrication, the final fabricated product of U-slot
Microstrip Patch Antenna is shown in Fig. 12.8b.
12 U-Slot Rectangular Patch Antenna for Dual Band Application 117

Fig. 12.6 Current distribution at 2.4 GHz

Fig. 12.7 Current distribution at 2.95 GHz

The fabricated antenna was connected with SMA-Female (Gold Type) con-
nector in order to test the antenna using Vector Network Analyzer (VNA) at the
lab. The VNA test results are presented in Fig. 12.9. Result shows that the oper-
ating frequency band of the antenna has been minutely shifted from 1.8 to
1.839 GHz with a return loss of -13.755 dB. The same shifting is observed in
case of the second resonant frequency band that shifted from 2.4 to 2.48 GHz with
118 M.S. Habib et al.

Fig. 12.8 a Prepared artwork design of the b Fabricated U-slot microstrip antenna

Fig. 12.9 Test result from vector network analyzer (VNA)

return loss of -10.195 dB. This deviation of the practical result with the simulated
one may be due to the fabrication error and the different characteristics of substrate
dielectric constant.
12 U-Slot Rectangular Patch Antenna for Dual Band Application 119

Fig. 12.10 Comparison of simulated and test result of U-slot microstrip antenna

12.5 Comparison Between Simulation and Test Result

The result of the design dimension is L = 80 mm, W = 100 mm, Lh = 31 mm

and Lv = 20 mm. The Length L and width W of the antenna is quite big due to the
formulation that was made in Ref. [14] which enabled the antenna to vary the
resonant frequency from 900 MHz to 7 GHz. With the decrease of the values of
the resonant frequencies, the wavelength of the radiation wave increases. There-
fore in order to deal with such wavelength, the size of the antenna is increased.
Through producing the U-slot in the microstrip patch antenna, the dual band
operational at frequencies of 1.8 and 2.4 GHz was achieved.
The simulated and fabricated test results are plotted in Fig. 12.10 for com-
parison. From Fig. 12.10, it can be observed that the simulation result are almost
identical to the one occurred in that of the fabrication test result. The only dif-
ference is that all the resonances are gradually shifted with different value of return
loss. This shifting is witnessed due to several reasons.
The differences between the result of simulation and fabrication may be
because of the value of substrate constant. In simulation process, the substrate
constant value being used is er = 4.3 which varies slightly in available board at
lab. Then, the dimension of the board and slot cut could not be extremely precise.
The process of cutting the board is by using automated drilling machine in PCB
lab which may also cause the differences of the results.
120 M.S. Habib et al.

12.6 Conclusion

In this paper we successfully designed a microstrip patch antenna with U-slot that
was resonant at 1.8 and 2.4 GHz with gains of 8.0 and 5.6 dBi respectively. It is
also observed that the same structure can be used to design more resonances to
achieve triple or quad band antenna for future works. Designed dual band antenna
was fabricated and tested. The test result is similar to simulated result with slight
shifting of resonant frequencies.

Acknowledgments Authors are grateful to Kulliyyah of Engineering and Research Management

Center, International Islamic University Malaysia to support this research through grants.

References

1. Jordan, R., Abdallah, C.T.: Wireless communica-tions and networking: an overview. IEEE
Trans. Antennas Propagat. Mag. 44(1), 185–193 (2002)
2. Joseph, M., Paul, B., Raj, R.K., Mohanam, P.: Compact wideband antenna for 2.4 GHz
WLAN applications. Electron. Lett. 40, 1460–1461 (2004)
3. Suo, Wei: Internal PIFAs for UMTS/WLAN/WiMAX multi network operation for a USB
dongle. Microw. Opt. Technol. Lett. 48(11), 22492253 (2006)
4. Karaboikis, M., Soras, C., Tsachtsiris, G., Makios, V.: Compact dualprinted inverted-F
antenna diversity systems for portable wireless devices. IEEE Antennas Wirel. Propag. Lett.
3, 9–14 (2004)
5. Eldek, A.A., Elsherbeni, A.Z., Smith, C.E.: Wideband bow-tie slot antennas for radar
applications, 2003 IEEE Topical. In: Conference Wireless Communication Technology,
Honolulu, Hawai, (2003)
6. Jan, J.Y., Tseng, L.C.: Small planar monopole antenna with a shorted parasitic inverted-L
wire for wireless communications in the 2.4-, 5.2-, and 5.8-GHz bands. IEEE Trans.
Antennas Propag. 52(7), 1903–1905 (2004)
7. Deshmukh, A.A., Ray, K.P.: Half U-slot loaded multi-band rectangular microstrip antennas.
Int. J. Microw. Opt. Technol. 2(2), 216–221 (2007)
8. Lee, K.F., Steven Yang, S.L., Kishk, A.A.: Dual and multi band U-slot patch antennas. IEEE
Antennas Wirel. Propag. Lett. 7, 645–647 (2008)
9. Deshmukh, A.A., Kumar, G.: Compact broadband U-slot loaded rectangular microstrip
antennas. Microw. Opt. Technol. Lett. 46(6), 556–559 (2005)
10. Maci, S.: Dual Band Slot Loaded Antenna. IEEE Proc. Microw. Antennas Propag. 142,
225–232 (1995)
11. Boyle, K.R., Massey, P.J.: Nine band antenna system for mobile phones. Electron. Lett.
42(5), 265–266 (2006)
12. Weigand, S., Huff, G.H., Pan, K.H., Bernhard, J.T.: Analysis and design of broadband single
layer rectangular U-slot microstrip patch antenna. IEEE Trans. Antennas Propag. AP-51(3),
457–468 (2003)
13. Lee, K.F., Yang, S.L.S., Kishk, A.A., Luk, K.M.: The versatile U-slot patch. IEEE Antennas
Propag. Mag. 52(1), 71–88 (2010)
14. Deshmukh, Amit A., Ray, K.P.: Formulation of resonance frequencies for Dual-band slotted
rectangular microstrip antennas. IEEE Antennas Propag. Mag. 54(4), 78–97 (2012)
15. Islam, M.M., Islam, M.T., Faruque, M.R.I.: Bandwidth enhancement of a microstrip antenna
for X-band applications. ARPN J. Eng. Appl. Sci. 8(8), 591–594 (2013)
Chapter 13
Analysis of Synthetic Storm Technique
Based on Ku-Band Satellite Beacon
Measurements in Malaysia

Ali K. Lwas, I.M. Rafiqul, Mohamed Hadi Habaebi,

Ahmad F. Ismail, Mandeep Singh, Jalel Chebil,
Al-Hareth Zyoud and Hassan Dao

Abstract Most of the existing rain attenuation prediction models were proposed
based on measurements taken in temperate climates. These models are found not
accurate in tropical regions and were thus modified in order for such models to be
applied in tropical regions. Synthetic Storm Technique (SST) is one of the most
reliable methods to estimate rain attenuation time series in Europe. However, due
to the lack of measured data in the tropical regions of the world, the above-
mentioned method is yet to be validated for those regions. This paper aims to
investigate SST validity in Malaysia by focusing on both rain events and the
overall statistical behavior. Its performance is assessed based on concurrent
measurement of Ku-band satellite beacon and rain rate over University of Science

A.K. Lwas  I.M. Rafiqul (&)  M.H. Habaebi  A.F. Ismail  A.-H. Zyoud  H. Dao
Kulliyyah of Engineering International Islamic University Malaysia, Jalan Gombak,
53100 Kuala Lumpur, Malaysia
e-mail: [email protected]
A.K. Lwas
e-mail: [email protected]
M.H. Habaebi
e-mail: [email protected]
A.F. Ismail
e-mail: [email protected]
A.-H. Zyoud
e-mail: [email protected]
H. Dao
e-mail: [email protected]
M. Singh
Faculty of Engineering, University Kebangsaan, Bangi, Malaysia
e-mail: [email protected]
J. Chebil
Higher Institute of Transport and Logistics, University of Sousse, Sousse, Tunisia
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 121

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_13
122 A.K. Lwas et al.

Malaysia (USM) campus at Tronoh. Preliminary analysis shows that SST is

capable of providing details of time-series of many rain events to reflect the
dynamics of rain fade. However, it is unable to predict the entire range of rain
intensity.

13.1 Introduction

Most severe conditions of propagation impairments such as rain, clouds and

atmospheric gases are occurring frequently in tropical regions. Furthermore, rain-
induced attenuation is the major issue at frequencies above 10 GHz which faces
the signal of satellite communications, more especially in this area which is
subjected to heavy rainfall. To design a reliable earth-to-satellite link, the effects of
these are required to measure and understand clearly. Scarcity of measured data at
higher frequencies is also an issue in tropics. This makes it difficult to test
available propagation prediction models or to develop new ones. Thus, this fact
gives a real reason of why the researchers are still in their researches improve and
update prediction models to give abstract view about the behavior of the rain rate
and attenuation phenomena in tropical regions [1].
The synthetic storm technique is a method which can transform a rainfall rate
time series directly into a rain attenuation time series. This technique developed by
[2] based on rain rate time series recorded in Italia. Compared to nine prediction
models, this model overcomes all of them in the three Italian sites. In addition,
study [3] focused on the applicability of the SST on single rain events affecting V-
band satellite links. It finds that the SST gives specifics of time series of single rain
events as well as matches long-term statistics. Also, in paper [4] performance of
SST is evaluated for Ka bands over a tropical location in India. The analysis
indicated that SST is an appropriate to estimate the fade characteristics from the
rain rate time series measurements over this region. A new prediction model is
proposed in [5] that bases on advantages of SC-EXCELL (stratiform-convective
rain discrimination) and of SST, named SC-SST (Stratiform-Convective SST).
From analysis, it is found to be in good agreement with beacon measurements
collected at Kuala Lumpur from MEASAT-1 satellite.
This paper aims to investigate applicability of SST in Malaysia by comparing
measured rain attenuation with that predicted by SST from rain rate both measured
concurrently. Rain attenuation and rain rate measurements were conducted at
USM campus (4.390 N, 100.980 E) at Tronoh, Perak (200 km from Penang),
which faces very heavy rainfall frequently. The receiver antenna is pointing
towards Superbird-C at 40.1 elevation angle and diameter size of 2.4 m. The
beacon signal frequency is 12.255 GHz. The data logging system has a sampling
rate of one sample per second and the rain gauge is of 1-min integration time.
Section two of the paper provides brief on the SST. Results are introduced and
analyzed in Section three. Section four concludes the contribution of the paper.
13 Analysis of Synthetic Storm Technique … 123

13.2 Synthetic Storm Technique

Synthetic Storm Technique is a physical-mathematical radio propagation method

that can be used to generate reliable rain attenuation time series by converting a
rain rate time-series at a specified site into a rain attenuation time-series. Local
parameters such as rain rate, length of signal path through rain cell and the rain cell
velocity at the site under investigation are required in the SST [6]. Moreover, by
applying the SST method, rain attenuation time series at any frequency and
polarization can be generated, and for any slant path above approximately 10, as
long as the hypothesis of isotropy of the rainfall spatial field holds in the long term.
The SST method is very useful for designing communication satellite systems and
improving their performance [7] for calculating rain attenuation using the SST
over a satellite path. The following are the basic assumptions, which are explained
in Fig. 13.1 [2, 3, 7, 8]. During rain, the vertical structure of the troposphere
separates into two layers. A is the rain layer and B is the melting layer as shown in
Fig. 13.1. The signal attenuation in case of satellite path is obtained from specific
attenuation at a point using the following expression:

ZLA ZLB
aA aB
AðxÞ ¼ KA R ðx0 þ Dx0 ; nÞdn þ KB r RaB ðx0 ; nÞdn ð13:1Þ
0 LA

where A (x) is the attenuation at a specific point and n is the distance measured
along the satellite path. K and a depend on the electromagnetic wave, frequency
and polarization, and raindrop size distribution. They are given by [9] for water
temperature of 20 C, and Parson’s law drop size distribution and for 0 C, are
given by [10]. More details about the SST are explained in [6].
According to [2], the following equation was derived by applying the Fourier
transform theory and some assumptions on Eq. (13.1) of the attenuation time
series

Fig. 13.1 Schematic diagram of rain structure for Synthetic Storm Technique [2]
124 A.K. Lwas et al.

Aðt Þ ¼ KA RaB ðtÞLA þ r aB KB RaB ðtÞðLB  LA Þ ð13:2Þ

where R(t) is rain rate time series. LA and LB are the radio path lengths and are
given by

LA ¼ ðHA  HS Þ=sin ðhÞ ð13:3Þ

LB ¼ ðHB  HS Þ=sin ðhÞ ð13:4Þ

where HS is the height above sea level of the Earth station, HA is the height above
sea level of the upper limit of layer A and HB is the height above sea level of the
upper limit of layer B. According to ITU-R Rec. 839-0 (1992) [11], HB is given by

5; u\23
HB ¼ ð13:5Þ
5  0:075ð/  23Þkm; /  23

The above equation indicates that HB depends on the latitude (;) of the Earth
station. The author in [2] supposes the HA is given by

HA ð/Þ ¼ HB ð/Þ  hðkmÞ ð13:6Þ

where h = 0.4 km is the thickness of the melting layer.

13.3 Results and Discussion

Conversion from rainfall rate to rain attenuation was performed by the Synthetic
Storm Technique method using Eq. (13.2). The radio path lengths LA and LB are
calculated according to Eqs. (13.3) and (13.4) respectively. The specific attenua-
tion of melting region (B) is constructed according to [12] and the values for
coefficients k and a are taken to be 0.2521 and 1.1635 respectively. The k and a
values are related to rainy layer (A) and are selected from [10] to be 0.02514 and
1.24 respectively. As originally proposed in [2] the rain advection velocity is
v = 10 m/s.
In Figs. 13.2 and 13.3, measured beacon rain attenuation and that predicted by
SST for 12.255 GHz are plotted with respect to time using measured rain rate time
series. It is observed that there is a good correlation between the measured rain rate
and the measured rain attenuation. Predicted rain attenuation time series using the
SST method followed the measured rain attenuation time series in both events.
Many peaks are observed in Fig. 13.2. In the first peak, the value of rain attenu-
ation is close to 20 dB by the SST method but the measurement is about 15 dB.
Also, the value of rain attenuation peak observed in the last peak is about 20 dB by
the SST method but the measurement is close to 14 dB. Figure 13.3 shows that the
13 Analysis of Synthetic Storm Technique … 125

Fig. 13.2 Comparison between measured rain attenuation and that converted by SST for a rainy
event on 10/8/2009

Fig. 13.3 Comparison between measured rain attenuation and that converted by SST for a rainy
event on 26/8/2009

highest value of measured attenuation reaches to 30 dB while the SST prediction is

close to 74 dB. One month statistics of measured rain attenuation for August 2009
and predicted rain attenuation are presented in Tables 13.1 and 13.2. The differ-
ence between measured and predicted rain attenuation for rain rate up to 60 mm/hr
is shown in Fig. 13.4. 1 min rain rate at 60, 120 and 180 mm/h are considered for
comparing the peak value of measured rain attenuation respectively. Three values
of 60, 120 and 180 mm/h were measured for rain rate. Therefore, the resolution for
this data was very low.
Figure 13.4 explains the differences between measured peak attenuation and
attenuation predicted by SST for all events where rain rate are 60 mm/h. The per-
centage of error is calculated using Eq. (13.7). More than 100 peaks with 60 mm/hr
rain rate are recorded during August 2009. Average of percentage error is 19 %, in
other words, in most cases SST prediction is higher than measured attenuation.
Since the resolution of measured rain rate is only 1.0 mm/min or 60 mm/h, the
peak value of 120 mm/h can be assumed average peak of 90 mm/h, while the peak
value of 180 mm/h can be assumed average peak of 150 mm/h. According to
126 A.K. Lwas et al.

Table 13.1 Percentage of error between measured attenuation and that predicted by the SST
method for Medium Rain Rate (R \ = 120 mm/h)
Date Ameasur ASST (max) Error (%) ASST (R = 90) Error (%)
1/8/2009 (1) 23.5 45.56 94 32.25 37
1/8/2009 (2) 18.12 45.56 151 32.25 78
5/8/2009 20.11 45.56 126 32.25 60
9/8/2009 27.75 45.56 64 32.25 16
25.32 45.56 80 32.25 27
25.51 45.56 79 32.25 26
26.44 45.56 72 32.25 22
27.46 45.56 66 32.25 17
13/8/2009 28.63 45.56 59 32.25 12.6
25.76 45.56 77 32.25 25.2
27.9 45.56 63 32.25 15.6
17/8/2009 25.5 45.56 79 32.25 26.5
28.19 45.56 62 32.25 14.4
27.56 45.56 65 32.25 17.1
26/8/2009 26.95 45.56 79 32.25 19.7
27/8/2009 25.69 45.56 77 32.25 25.5

Table 13.2 Percentage of error between measured attenuation and that predicted by the SST
method for High Rain Rate (R \ = 180 mm/h)
Date Ameasured ASST (max) Error (%) ASST (R = 150) Error (%)
4/8/2009 28.23 74.2 163 59.59 111
26/8/2009 27.39 74.2 171 59.59 118
26/8/2009 27.145 74.2 173 59.59 120

Table 13.1, where the maximum rain rate is 120 mm/h and the assumed average
peak is 90 mm/h, the differences between measured rain attenuation and the SST
predications are very high for rain rate when it equals 120 mm/h. It varies between
17 and 27 dB, which is 60–150 %. For the assumed average rain rate of 90 mm/h,
the differences are 4–14 dB, which varies from 13 to 78 %. Furthermore,
Table 13.2 shows that the differences between the measured attenuation and the
SST predictions are very divergent for 180 mm/h (46.6 dB), while it is 32 dB for
an average of 150 mm/h. In both cases, the errors are more than 100 %. The
percentage of errors (E) between the measured attenuation (Am) and the predicted
attenuation (Ap) are calculated based on the following equations [13]:
13 Analysis of Synthetic Storm Technique … 127

Fig. 13.4 Percentage of error between measured attenuation and that predicted by the SST
method with respect to number of events

Ap  Am
E¼ % ð13:7Þ
Am

13.4 Conclusion

This paper focuses on the applicability of the SST in Malaysia, which has been
established as an accurate method in Europe. The SST method is used to generate
rain attenuation time series by using rain rate data measured for one month in
Malaysia. The attenuation predicted by using the SST method is compared with
measured rain attenuation at 12.255 GHz satellite beacon recorded at Satellite
Lab, USM. It is found that the attenuation derived by SST method is close to the
measured attenuation in low rain intensity but it is overestimated in medium
rainfall and extremely overestimated in high rainfall rate. Despite these discrep-
ancies, the SST method can be used for generating time-series of rain attenuation
on satellite to earth links in tropical regions with proper modifications. However,
this method is required to be modified based on long term measurements to be
applicable for tropical regions.

Acknowledgments This work is supported by the Research Management Centre (RMC), project
no. EDW B13-038-0923, International Islamic University Malaysia (IIUM).
128 A.K. Lwas et al.

References

1. Emiliani, D., Agudelo, J., Gutierrez, E., Restrepo, J., Mendez, F.: Development of rain
attenuation and rain rate maps for satellite system design in the Ku and Ka bands in
Colombia. IEEE Antennas Propag. Mag. 46, 54–68 (2004)
2. Matricciani, E.: Physical-mathematical model of the dynamics of rain attenuation based on
rain rate time series and a two-layer vertical structure of precipitation. Radio Sci. 31,
281–295 (1996)
3. Sánchez-Lago, I., Fontán, F.P., Mariño, P., Fiebig, U.C.: Validation of the syntheticstorm
technique as part of a time series generator for satellite links. IEEE Antennas Wirel. Propag.
Lett. 6, 372–375 (2007)
4. Shukla, K.A., Das, S., Roy, B.: Rain attenuation measurements using synthetic storm
technique over ahmedabad. In: International Conference on Computers and Devices for
Communication India (2007)
5. Lam, Y.H., Luini, L., Din, J., Capsoni, C., Panagopoulos, A.D.: Investigation of rain
attenuation in equatorial kuala lumpur. IEEE Antennas Wirel. Propag. Lett. 11, 1002–1005
(2012)
6. Lwas, A., Islam, Md., Chebil, J., Habaebi, M., Ismail, A., Zyoud, A., Dao, H.: Rain
attenuation analysis using synthetic storm technique in malaysia. In: International Conference
on Mechatronics (ICOM 2013) Malaysia (2013)
7. Matricciani, E., Riva, C., Castanet, L.: Performance of the synthetic storm technique in a low
elevation 5 slant path at 44.5 GHz in the French, Pyrénées, European Conference Antennas
Propagation (EuCAP 2006) Nice: France (2006)
8. Panagopoulos, A.D., Arapoglou, P.D., MChatzarakis, G.E., Kanellopoulos, J.D., Cottis, P.G.:
A new formula for the prediction of the site diversity improvement factor. Int. J. Infrared
Millimeter Waves 25, 1781–1789 (2004)
9. ITU-R P.838-3: Specific attenuation model for rain for use in prediction methods, ITU-R
Recommendations Geneva (2005)
10. Maggiori, D.: Computed transmission through rain in the 1–400 GHz frequency range for
spherical and elliptical drops and any polarization. Alta Frequenza 50, 262–273 (1981)
11. ITU-R P.839-0: Rain height model for prediction methods. Recommendation P. Series
Fascicle Radio wave propagation International Telecommunication Union Geneva (1992)
12. ITU-R P.839-3: Rain height model for prediction methods: Recommendation P. Series
Fascicle Radio wave propagation International Telecommunication Union Geneva (2001)
13. Mandeep, J.S.: Prediction of signal attenuation due to rain models at TRONOH, Malaysia.
Mapan J. Metrol. Soc. India 28, 105–111 (2013)
Chapter 14
The Evolution of Double Weight Codes
Family in Spectral Amplitude Coding
OCDMA

N. Din Keraf, S.A. Aljunid, A.R. Arief and P. Ehkan

Abstract This paper presents the review of Double Weight (DW) codes family
from perspective of codes evolution since beginning until now. First generation
focuses on one-dimensional (1D) code which spreading has been carried out in
time. Later, the second generation was introduced known as two-dimensional (2D)
code where the encoding of data bit has spread in both wavelength and time
domain. Codes construction and variety of detection techniques scheme used in
DW codes family are also discussed. In this paper we focus on structure of among
DW codes family as all the design of code sequence aim to eliminate Multiple
Access Interference (MAI). Previous papers of DW codes family show a good
quality of transmission and satisfactory the standard quality of Bit Error Rate
(BER) above 10-9.

 
Keywords Double weight Optical code division multiple access Bit error rate

N. Din Keraf  S.A. Aljunid  A.R. Arief  P. Ehkan (&)

School of Computer and Communication Engineering, Universiti Malaysia Perlis,
Pauh Putra Campus, 02600 Arau, Perlis, Malaysia
e-mail: [email protected]; [email protected]
N. Din Keraf
e-mail: [email protected]
S.A. Aljunid
e-mail: [email protected]
A.R. Arief
e-mail: [email protected]
N. Din Keraf
Department of Electrical Engineering, Politeknik Sultan Abdul Halim Mu’adzam Shah,
06000 Jitra, Kedah, Malaysia

 Springer International Publishing Switzerland 2015 129

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_14
130 N. Din Keraf et al.

14.1 Introduction

About last three decades, optical code division multiple access (OCDMA) has
been rapidly growth [1] as a result of exhaustive research and to satisfy the hunger
user with new services and applications especially for multimedia services. In
OCDMA, an optical code represents a user address and signs or each user has its
own codeword. Various types of OCDMA codes family such as optical orthogonal
code (OOC), modified frequency hopping (MFH) and prime codes have been
introduced for spectral amplitude coding [2]. Besides, DW codes family are also
including in this type of coding. Several codes for DW have been proposed
comprising of DW, Modified Double Weight (MDW) [3, 4] and Enhanced Double
Weight (EDW) [5].
The principle source of noise and the major limit factor to system performance
in OCDMA is called multiple-access interference (MAI) [6, 7]. As the number of
simultaneous users increase, the effect of MAI is also increase in incoherent
OCDMA. Spectral Amplitude Coding (SAC) is one of the methods used in
incoherent OCDMA systems to eliminate MAI [2] by using code sequences. Most
of the codes including DW codes family aim to overcome MAI as its effect a poor
BER performance for the overall system. This paper is organized as the evolution
of the codes is detailed in Sect. 14.2, structures of the codes are described in
Sect. 14.3 and finally, conclusion of this paper in Sect. 14.4.

14.2 The Evolution of the Codes

DW codes family was first introduced by Aljunid [3] in year 2004. Initially, DW
codes family are proposed for SAC OCDMA system which has a constant weight
of two. Later, the development of this code family has become aggressively
expanded as illustrated in Fig. 14.1. The first three codes were categorized as one-
dimensional (1D) whereas the last two codes we will refer as two dimensional
(2D) codes. For 1D code, the spreading has been carried out in time while for 2D
encoding performs the frequency spreading in time and wavelength domain
simultaneously.
The constraint of constant weight of two in DW code inspires the development
of MDW code that has a variable weight greater than two. MDW codes are able to
support simultaneous transmissions at different bit rates [4]. Another variation of
DW code has been proposed by Hasoon et al. [5] known as EDW code which has a
variable weight odd number greater than one. The extended of 1D DW codes
family was proposed in 2011 [8] namely 2D MDW with the conscious that all
those 1D codes suffer from a numerous limitations. Moreover, realizing the use of
bandwidth must be optimized, two dimensional (2D) code has proposed to increase
the number of simultaneous users compared to 1D code. Recently, 2D TEDW was
proposed [9] and give more advantages such as lower bandwidth consumption
compared to the conventional 1D EDW.
14 The Evolution of Double Weight Codes Family … 131

Fig. 14.1 The evolution of

double weight codes

2D EDW
2D MDW or
Code known
Enhance
(2011) as TEDW
d Double
Modified Weight Code
Double (EDW) (2012)
Double Weight Code
Weight (MDW) (2006)
(DW) Code Code
(2004) (2004)

Construction techniques used to construct these codes family are matrix con-
struction technique and mapping scheme [3, 10]. The details of code structure in
DW code family will be described in Sect. 14.3. Other than that, selection of
detection techniques has been introduced recent years such as complimentary
subtraction, AND subtraction, NAND subtraction and Modified AND subtraction
[11–13]. For conventional complementary subtraction technique, two different
code sequences are modulated with data and sent to multiplexer. The received
signal is divided into two branches of spectral chips; upper branch and lower
branch [12]. These two branches of spectral signals are sent to a subtractor that
computes the correlation difference [11]. On the other hand, for AND subtraction
the received signal splits into two parts; one to the decoder that has an identical
filter structure with the encoder and the other to the decoder that has the AND filter
structures [11]. A subtractor is then used to subtract the overlapping data from the
intended code. Furthermore, the modified-AND subtraction technique provides a
better performance on higher data rate or a larger number of users [12]. The major
advantage of modified-AND subtraction is to suppress MAI and the impacts of
PIIN by dividing the spectrum of the used code sequence.
Moreover, DW codes family is also used in many applications [12] such as for
instance using MDW code to support triple play services (voice, video, and data)
by utilizing the different detection scheme in SAC-OCDMA. Furthermore, [14]
proposed that MDW code is applicable to use in Wide Area Network (WAN). DW
code was also implemented in local area network (LAN) environment which is
applied in ring network as proposed by [15]. In addition, EDW codes are suitable
to support multiple bit rate transmissions as suggested by [16].
There are three main properties that take into account for design a code [17] i.e.
correlation properties, maximum supported user number and BER. In order to
maintain as many as possible simultaneous users within an acceptable BER i.e.
BER = 10-9, correlation properties must be good enough because they are con-
nected and restricted by each other. Thus for designing the DW is code, most
authors that previously mentioned make a great effort to fulfill the requirement as
much as possible. For instance the value of cross correlation are highly considered
132 N. Din Keraf et al.

in DW family code as the large value of cross correlation affects the system
performance due to MAI. On the other hand, to maximize the number of user, 2D
code was proposed. It is not only increases the flexibility of the code design but
also improves the cardinality of codes dramatically [17].

14.3 Constructions of the Codes

In one dimensional DW code family, three important parameters are denoted as

(N, W, k) where N represents code length, W is code weight and k for correlations.
Three types of code consists in this category, they are DW, MDW and EDW.
However in two dimensional code there are five parameters involved consists of
number of wavelength M, temporal code length N, w is weight, ka and kc are auto
and cross-correlation values are denoted by (M 9 N, W, ka, kc). In designing 2D
code, cross correlation kc and auto correlation, ka are significantly concern as the
cross correlation kc value implies the interference between users. On the contrary,
the auto correlation value implies the signal power and the distinction in the
presence of other users. As above mentioned, the major degrading factors which
affect the overall performance of OCDMA networks are the multiple access
interference (MAI) constraints. Accordingly, codes with minimum cross correla-
tion value are preferred for OCDMA communications to reduce the effect of MAI
[7].

14.3.1 DW Code

Reference [3] described the DW of (N, W, k) family can be constructed by using

the following steps:
Step 1:
The DW code can be represented by using a K 9 N matrix. In DW codes struc-
tures, the matrix K rows and N columns will represent the number of user and the
minimum code length respectively. A basic DW code is given by a 2 9 3 matrix,
as
 
0 1 1

H1 ¼  
1 1 0

Notice that H1 has a chips combination sequence of 1, 2, 1 for the three columns
(i.e. 0 + 1, 1 + 1, 1 + 0).
Step 2:
A simple mapping technique is used to increase the number of codes as
14 The Evolution of Double Weight Codes Family … 133

 
0 0 0 0 1 1   
  
0 0 0 1 1 0   0 H1 
H2 ¼  ¼
0 1 1 0 0 0  H1 0 
 
1 1 0 0 0 0

Note that as the number of user, K increases, the code length, N also increases.
The relationship between the two parameters, K and N is given by
 
3K 1 Kp 2
N¼ þ sin ð14:1Þ
2 2 2

A position of weight is very important to maintain in pairs as it reduces the

number of filters use in the encoder and decoder. This way, a filter with the
bandwidth twice of the chip width can be used, instead of two different filters.

14.3.2 MDW Code

MDW is the modified version of DW code. The MDW code weight can be any
even number that is greater than two. As a family of DW code, MDW can also be
represented by using the K 9 N matrix. Reference [3] described that the basic
MDW can be developed by using the following steps:
Step 1:
The basic matrix for MDW codes also consists of a K 9 N matrix depending on
the value of code weight. The general form of matrix for a MDW code is

A B

C D
where;
w
2 1
P
i. A consists of a 1  3 j matrix of zeros
j¼1
ii. B consists of a 1 9 3n matrix containing the basic matrix of [X2] for every 3
columns (i.e. a 1 9 3n matrix which is n repetition of [X2])
iii. C is the basic code matrix for the next smaller weight, W = 2(n - 1).
iv. D is a matrix n 9 n consisting of basic matrix of [X3] arranged as

 
 000 000 ½X3  

 000 ½X3  000 

 ½X3  000 000 
134 N. Din Keraf et al.

And n = W/2, W = 2, 4, 6,… R where X1, X2 and X3 are the [1 9 3] matrix and
consists of

X2 ¼ ½ 0 1 1
X3 ¼ ½ 1 1 0
X1 ¼ ½ 0 0 0

Step 2:
There are two basic components in basic matrix for MDW codes:
Code length,
W
X
2

NB ¼ 3 j ð14:2Þ
j¼1

Number of user,

W
KB ¼ þ1 ð14:3Þ
2

The Eqs. (14.2) and (14.3) represent the basic matrix for MDW code, where NB is
the column (i.e. its represent basic code length) and KB is the row (its represents
basic number of user). The MDW matrix is consisting of (KB 9 NB). The basic
MDW code with code length 9, weight 4 and an ideal in-phase cross correlation
denoted by (9, 4, 1) is
A B
0 0 0 0 1 1 0 1 1
0 1 1 0 0 0 1 1 0
1 1 0 1 1 0 0 0 0
C D

Notice that similar structure of the basic DW code, H1 is still maintained with a
slight modification, whereby the DW pairs are maintained in a way to allow only
two overlapping chips in every column. Thus, the 1, 2, 1 chips combination is
maintained for every three columns as in the basic DW code. This is important to
maintain k = 1.

14.3.3 EDW Code

Reference [5] described EDW code as the enhancement version of DW code

family. Its code weight can be any odd number that greater than one. The con-
struction of EDW code with the weight of three can be described as
14 The Evolution of Double Weight Codes Family … 135

Step 1:
EDW codes also consists of a K 9 N matrix which K rows and N columns will
represent the number of user and the minimum code length respectively. A basic
EDW code is given by a 3 9 6 matrix, as
   
0 01 10 1
   
H0 ¼  0 1  0 0  1 1 
1 10 10 0

Notice that similar structure of the basic DW code, H1 is still maintained with a
slight modification, whereby the double weight pairs are maintained in a way to
allow only two overlapping chips in every column.
Step 2:
From the basic matrix, a larger number of K can be achieved by using a mapping
technique as
 
0 0 0 0 0 0 0 0 1 1 0 1
 
0 0 0 0 0 0 0 1 0 0 1 1  
 
0 0 0 0 0 0 1 1 0 1 0 0   0 H0 
H1 ¼  ¼
0 0 1 1 0 1 0 0 0 0 0 0   H0 0 
 
0 1 0 0 1 1 0 0 0 0 0 0
 
1 1 0 1 0 0 0 0 0 0 0 0

An EDW code with weight of 3 denoted by (N, 3, 1) for any given code length
N, can be related to the number of user K through
     
4 Kp 2 8 ðK þ 1Þp 2 4 ðK þ 2Þp 2
N ¼ 2K þ sin sin þ sin ð14:4Þ
3 3 3 3 3 3

14.3.4 2D MDW Code

Development of 2-D MDW code is actually from the 1-D MDW code and denoted
by (M 9 N, W, ka, kc) [8]. As mentioned earlier, 2D encoding performs the
frequency spreading in time and wavelength domain simultaneously. Thus for 2D
j j
MDW, matrix of M row vectors dk;N is related to the temporal spreading; d1;N ¼
h i
j j j j
ck;1 ; ck;2 ; . . .ck;N1 ; ck;N while k is the emitted wavelength k 2 f1; . . .M g as
described in Eq. (14.5).
136 N. Din Keraf et al.

2 j 3
d1;N
6 7j
d2;N
6 7
6 7..
j
CM;N 6
¼6 7 ð14:5Þ
7 .
6 j 7
4 dM1;N 5
j
dM;N

The signals rk,N (t) represent the sum of the temporal spreading data of each
user carried on the wavelength, kk. They are expressed as:

X
Fu
rK;N ðtÞ ¼ bij ðtÞdk;n
j
ð14:6Þ
j¼1

j
The M signals rk;N ðtÞ are multiplexed and the total signal RM,N (t) is transmitted on
the optical fiber. It can be represented by a matrix (M 9 N):
2 3
r1;N ðtÞ
6 7
r2;N ðtÞ
6 7
6 7
..
RM;N ðtÞ ¼ 6 7
. ð14:7Þ
6 7
4 rM1;N ðtÞ 5
rM;N ðtÞ
j P j j
The signal rk;N ðtÞ is expressed as Fuj¼1 bi ðtÞdk;n represent sum of the temporal
spreading data user carried the wavelength, t. 2-D MDW network consists of M, N
pairs of transmitters and receivers. Ag,h is the code where g 2 ð1; 2; 3; . . .; M  1Þ
and h 2 ð1; 2; 3; . . .; N  1Þ. Xg is the spectral encoding and Yk is the spatial
encoding. Table 14.1 shows some examples of 2-D MDW code sequences.
The cross correlation of 2-D MDW code can be obtained by introducing the
four characteristic matrices A(d), where d 2 ð0; 1; . . .; 3Þ are defined as

Að0Þ ¼ Y T X ð14:8Þ

Að1Þ ¼ Y T X
 ð14:9Þ

Að2Þ ¼ Y T X ð14:10Þ

Að3Þ ¼ Y T X
 ð14:11Þ

Parameter X  and Yare

 the complementary of X and Y respectively. The cross
correlation of 2D MDW code A(d) and Ag,h is expressed as
14 The Evolution of Double Weight Codes Family … 137

Table 14.1 2-D MDW code for k1 = 4 and k2 = 2 sequences

Xg,h [000011011] [011000110] [110110000] Yk
0 000000000 000000000 000000000
1 000011011 011000110 110110000
1 000011011 011000110 110110000
1 000011011 011000110 110110000
1 000011011 011000110 110110000
0 000000000 000000000 000000000

Table 14.2 Cross correlation of 2D MDW Code

X g,h R(0) (g, h) R(1) (g, h) R(2) (g, h) R(3) (g, h)
g = 0, h = 0 k1k2 0 0 0
g = 0, h 6¼ 0 k1 k1 0 0
g 6¼ 0, h = 0 k2 0 k2 (k1 - 1) 0
g 6¼ 0, h 6¼ 0 1 1 k1 - 1 k1 - 1

X
M 1 X
N 1
ðdÞ
RðdÞ ðg; hÞ ¼ aij aðiþgÞðjþhÞ ð14:12Þ
i¼0 j¼0

ðdÞ
where aij is the (i, j)th of A(d) and aðiþgÞðjþhÞ is the (i, j)th of Ag,h; g 2
ð0; 1; 2; . . .; M  1Þ and h 2 ð0; 1; 2; . . .; N  1Þ. Table 14.2 illustrates the cross
correlation of 2-D MDW code generated from Eq. (14.12).
The derivation of new correlation functions can be expressed as

Rð2Þ ðg; hÞ Rð3Þ ðg; hÞ k1 ; k2 ; for g ¼ 0 and h ¼ 0
Rð0Þ ðg; hÞ  Rð1Þ ðg; hÞ  þ ¼
k1  1 k1  1 0; otherwise
ð14:13Þ

14.3.5 TEDW Code

The generating algorithm steps and equations for TEDW code can be described as
follows [9]:
(i) Base Matrix Generating
A 0’s matrix is generated Bm and the matrix dimensions determined by

Rows ¼ W ð14:14Þ
138 N. Din Keraf et al.

A B D
Users C1 C2 C3 C4 C5 C6
1 1 0 1 1 0 0 3
2 1 1 0 0 1 0 3
3 0 0 1 0 1 1 3
`
2 1 2 1 2 1

Fig. 14.2 EDW code matrix W = 3 [9]

Pw
j¼1 j
Columns ¼ Na ¼ ð14:15Þ
w

for all rows in the base matrix Bm but for the final row a round shift operation is
performed by using the Eq. (14.16).

X
Na
ðBm½j; j ¼ Bm½j; j þ 1Þ ð14:16Þ
j¼1

For Final row we use,

X
Na
Bmn Final ¼ ðBm½n; j ¼ Bm½j; jÞ ð14:17Þ
j¼1

By applying the Eqs. (14.14)–(14.17) we have Part A (Base Matrix) as shown

in Fig. 14.2.
(ii) Shift Level
The number of shifts is determined as

W þ1
Number of Shifts ¼ ð14:18Þ
2

(iii) Full Row shift and Full Base Matrix Shift

The sequence order is determined by S = W and for the purpose of generalization,
assuming X = Bm and X = [] (Row, Columns) = (W, S). To determine part B
for the code from the previews matrix assume that y = X and use the simplified
Eq. (14.19).
( )
X
s X
s1
½X ð1; iÞ ¼ yðw; iÞ ½X ðj þ 1; iÞ ¼ yðj; iÞ ð14:19Þ
i¼1 j¼1
14 The Evolution of Double Weight Codes Family … 139

Next shift Xnext = the matrix from the previews part of the code and y = Xnext
shifted. The full generation of the code is done by applying Eqs. (14.14) till (14.19)
and following the sequence steps.

14.4 Conclusion

The DW codes family proved to be one of the successful codes in the spectral
amplitude coding OCDMA systems to reduce MAI due to the ability of these
codes family to perform the standard acceptable BER (i.e. BER = 10-9). In
addition, these codes family are also designed to obtain the codes with ideal cross
correlation due to the fact that multiuser interference significantly affects the
overall system performance. The 1D DW family codes has properties summarized
as each code sequence has cross correlation, k equal to 1, the weight chips are
always in pairs and there must be relation between number of user and code length.
On the other hand, to increase the cardinality in OCDMA system, various domain
and parameters are combined to create multidimensional codes. However, multi-
dimensional brings a lot of complexity to the system implementation and archi-
tecture and so the DW codes family only focuses on 2D codes. An interesting
future research for DW family codes would be the 3D code which promises the
better performance and support more active users simultaneously.

References

1. Prucnal, P.R.: Optical code division multiple access fundamental and applications. CRC
Press, Florida (2006)
2. Yin, H., Richardson, D.J.: Optical code division multiple access communication networks.
Theory and applications. Tsinghua University Press, Springer-Verlag GmbH, Beijing, Berlin
(2007)
3. Aljunid, S.A., Ismail, M., Ramli, A.R., Ali, B.M., Abdullah, M.K.: A new family of optical
code sequences for spectral-amplitude-coding optical CDMA systems. IEEE Photonics
Technol. Lett. 16(10), 2383–2385 (2004)
4. Aljunid, S.A., Samad, M.D.A., Othman, M., Hisham, M.H., Kasiman, A.H., Abdullah, M.K.:
Development of modified double-weight code and its implementation. In: IEEE International
Conference, pp. 288–192 (2005)
5. Hasoon, F.N., Aljunid, S.A., Abdullah, M.K., Shaari, S.: Spectral amplitude coding OCDMA
systems using enhanced double weight code. J. Eng. Sci. Technol. 1(2), 192–202 (2006)
6. Fouli, K., Maier, M.: OCDMA and optical coding: principles, applications, and challenges.
IEEE Commun. Mag. 45, 27–34 (2007)
7. Ghafouri-Shiraz, H., Massoud Karbassian, M.: Optical CDMA networks: principles, analysis
and applications. Wiley-IEEE Press, New York (2012)
8. Arief, A.R., Aljunid, S.A., Anuar, M.S., Junita, M.N., Ahmad, R.B., Ghani, F.: Enhanced
performance of new family modified double weight codes spectral amplitude coding optical
CDMA system network. In: IEEE International Conference on Control System, Computing
and Engineering, pp. 488–494 (2011)
140 N. Din Keraf et al.

9. Zahid, A.Z.G., Mandeep, J.S., Susthitha Menon, P., Bakarman, H., Hasoon, F.N., Bakar,
A.A.A., Ali, M.A.M.: Performance analysis of multi-weight 2D OCDMA TEDW. In: 3rd
International Conference on Photonics, pp. 204–208 (2012)
10. Mohammed, A., Saad, N.M., Aljunid, S.A., Safar, A.M., Abdullah, M.K.: Optical spectrum
CDMA: a new code construction for double weight code family. In: International Symposium
on Communications and Information Technology, pp. 812–815 (2006)
11. Norazimah, M.Z., Aljunid, S.A., Fadhil, H.A., Md Zain, A.S.: Analytical comparison of
various SAC-OCDMA detection techniques. In: 2nd International Conference on Photonics,
pp. 1–5 (2011)
12. Al-Khafaji, H.M.R., Aljunid, S.A., Amphawan, A., Fadhil, H.A.: Triple-play services using
different detection techniques for SAC-OCDMA systems. In: 3rd International Conference on
Photonics, October 2012, pp. 350–354 (2012)
13. Ahmed, N., Aljunid, S.A., Ahmad, R.B., Rashid, M.A.: Novel OCDMA detection technique
based on modified double weight code for optical access network. Elektron. Ir Elektrotech.
18, 117–121 (2012)
14. Radhi, I.F., Aljunid, S.A., Fadhil, H.A., Al-Khafaji, H.M.R.: Performance evaluation of
spectral amplitude coding signature sequences for OCDMA systems. In: 2nd International
Conference on Photonics, pp. 1–4 (2011)
15. Dayang, H.K., Aljunid, S.A.: Optical code division multiple access (OCDMA) using double
weight (DW) codes for local area network. In: International Conference on Computer and
Communication Engineering (ICCCE 2010) (2010)
16. Hasoon, F.N., Aljunid, S.A., Abdullah, M.K., Shaari, S.: Multi-rate transmissions on SAC-
OCDMA system using new enhancement double-weight (EDW) codes. In: 2nd Information
and Communication Technologies (ICTTA) (2006)
17. Zhang, M.: Design and performance analysis of novel signature code in two-dimensional
optical CDMA systems (2012)
Chapter 15
Performance Evaluation of LTE
Scheduling Techniques for Heterogeneous
Traffic and Different Mobility Scenarios

Lukmanhakim Sukeran, Mohamed Hadi Habaebi,

Al-Hareth Zyoud, Musse Mohamud Ahmad, Shihab Hameed,
Amelia Wong and I.M. Rafiqul
Abstract In this paper five scheduling algorithms were investigated and their
performance was evaluated in terms of Fairness Index, Peak Throughput, Average
Throughput and Edge Cell User Throughput. A system level MATLAB simulator
was used. The simulation takes into account different types of traffic for several
mobility scenarios and propagation channel models. Results indicate that the
scheduling algorithms showed some quality in certain parameter of evaluation but
lack in other terms. While some scheduling algorithm take the moderate path but
still be lacking especially in Edge Cell User Throughput necessitating the use of
Relays or femtocells.

15.1 Introduction

Evolution of Universal Mobile Telecommunications System (UMTS) has not

reached its end even though with the existence of High Speed Packet Access
(HSPA). UMTS Long Term Evolution (LTE) has been introduced in 3rd

L. Sukeran
M.H. Habaebi (&)
A.-H. Zyoud
M.M. Ahmad
S. Hameed
A. Wong

I.M. Rafiqul
Faculty of Engineering, Electrical and Computer Engineering Department,
International Islamic University Malaysia (IIUM), 53100 Gombak, Kuala Lumpur, Malaysia
e-mail: [email protected]; [email protected]
A.-H. Zyoud
e-mail: [email protected]
M.M. Ahmad
e-mail: [email protected]
S. Hameed
e-mail: [email protected]
A. Wong
e-mail: [email protected]
I.M. Rafiqul
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 141

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_15
142 L. Sukeran et al.

Generation Partnership Project (3GPP) Release 8 to guarantee the competitiveness

of UMTS for the next coming years. The rapid grow of mobile data usage in the
recent years such as gaming, mobile channel TV, and other streaming content have
concerned in the (3GPP) leading to motivation on LTE. Therefore the work
towards 3rd Generation Partnership Project (3GPP) Long Term Evaluation started
in 2004 and the targets of LTE standard were set [1].
Orthogonal Frequency Division Multiplexing (OFDM) has been adopted as the
downlink transmission scheme for the 3GPP LTE. OFDM is a multicarrier
transmission scheme since it splits up the transmitted high bit-stream signal into
different sub-streams and sends these over many different sub-channels [2]. OFDM
simply divides the available bandwidth into multiple narrower sub-carries and
transmits the data on these carries in parallel streams. Each sub-carrier is modu-
lated using different modulation scheme, e.g. Quadrature Phase Shift Keying
(QPSK), Quadrature Amplitude Modulation (QAM), 64QAM and an OFDM
symbol is obtained by adding the modulated subcarrier signals [3].
The scheduling algorithm is the radio resource management technique that is
used by the base station to manage and control the available radio resources and
assign them efficiently to the available users to meet their service requirement. The
minimum resources that could be assigned for a user are called Resource Block
(RB). RB includes 12 adjacent OFDM subcarriers. The scheduler task is to assign
these RBs to the users in the network. Many scheduling algorithms have been
proposed in the literature. So far, Different studies have been conducted to
investigate the performance of the proposed algorithms for different scenarios
using several simulation platforms.
The work in [4] investigated the performance of five scheduling algorithms for
video traffic using 3GPP LTE simulator. The results showed that Maximum-
Largest Weighted Delay First (M-LWDF) algorithm performs better than other
algorithms like Round Robin (RR), Exponential/Proportional Fair (EXP/PF),
Maximum Rate (Max-Rate), and Proportional Fair (PF) in terms of throughput,
number of users supported and fairness. In [5], a comparison of different sched-
uling algorithms for downlink channel was performed using NS-3 simulator.
Similarly, NS-3 was used in [6] to evaluate the performance of scheduling algo-
rithms for uplink scenarios. Moreover, LTE-Sim [7] was used in [8] to compare
the performance of three different scheduling algorithms in video traffic scenarios.
Habaebi et al. [9] evaluated three of the most known scheduling algorithms
namely, RR, PF and Best Channel Quality Indicator (BCQI) using LTE system
level simulator [10]. They found that the BCQI outperforms RR and PF in terms of
throughput and Block Error Rate (BLER).
In this paper five types of scheduling are considered which are RR, Proportional
Fair Sun (PFS), BCQI, Resource Fair Maximum Throughput (RF) and Max-Min
Fairness (MaxMin). System level simulations were carried out to compare and
evaluate the previous algorithms. The performance was evaluated in terms of
fairness index, peak throughput, average throughput and cell edge throughput.
15 Performance Evaluation of LTE Scheduling Techniques … 143

15.2 Scheduling Algorithms

The scheduling algorithms that are investigated in this paper are highlighted in the
sub sections below:

15.2.1 Resource Fair Maximum Throughput

Resource Fair Maximum Throughput algorithm integrates the Max-rate and Pro-
portional Fair scheduling. This algorithm efficiently employs available radio
resource as user’s packets are transmitted on a radio resource with a good channel
condition. Users are treated according to the rank and the schedulers are either
Max-rate of Proportional Fair [11].

15.2.2 Proportional Fair Sun

This scheduling algorithm is basically the improvement and less complexity form
of the PF scheduling algorithm. Karush-Kuhn-Tucker (KKT) condition was used
to reduce the complexity of the PF [12].

15.2.3 Max-Min Fairness

The algorithm key parameter is fairness. It distributes the resource block to

achieve optimal fairness. It aims to provide the maximum resource to the mini-
mum date rate for the receiver so that the data rate distribution is fair [11].

15.2.4 Best Channel Quality Indicator

BCQI scheduling policy is to allocate resource blocks to the user with the best
channel condition. In order to perform scheduling, terminals send Channel Quality
Indicator (CQI) to the base station [13].

15.2.5 Round Robin

RR is proposed to solve the problem of fairness that appears in the BCQI and RF
algorithms. It is allocate equal time for each user without priority option. There-
fore, the channel condition has no impact on the user chance. Basically this
algorithm rate the user in term of first come first serve basis [13]. The fairness is
improved, however the throughput is degraded significantly.
144 L. Sukeran et al.

Table 15.1 Simulation parameters for LTE system level simulator

Parameter Value
Frequency 2.14 GHz
Bandwidth 20 MHz
Simulation length 100 TTI
Inter eNodeB distance 500 m
eNodeB TX power 20 dBW
Number of UEs 1, 2, 5, 10, 20, 40 users
Antenna pattern Omni-directional
eNodeB antenna gain 15 dBi
Uplink delay 3 TTIs
UEs speed 3 and 120 km/h

15.3 Simulation Environment

In this paper, 7 hexagonal base stations (eNodeBs) with various number of user
equipments (UEs) are used. The users are randomly located in the eNodeB region
of interest. The LTE system level simulator parameters are given in Table 15.1.
The mobility is considered in all scenarios (pedestrian and vehicular). Five dif-
ferent traffic types have been considered: VoIP, Video, FTP, HTP, and Gaming.

15.4 Simulation Results and Discussions

Using the parameters presented in Sect. 15.3, the results were generated for dif-
ferent traffic types, for each mobility scenario and for various numbers of users.
The figures in this section show the generated result for VoIP service only since
there is no enough space to show all the generated results for all traffic types.
However, all results were discussed at the end of the section. The results for VoIP
traffic for pedestrian UE with speed 3 km/h are presented in Figs. 15.1, 15.2, 15.3,
and 15.4.
In terms of fairness index as in Fig. 15.1, BCQI shows declination rapidly
compares to other scheduling algorithms as the number of users increase. Other
scheduling algorithms have uniform distribution of fairness index ranging in
between the values of 0.64 and 0.87. However, In terms of peak throughput as in
Fig. 15.2, BCQI scores the highest value of peak throughput 160 Mbps and this
value is maintain up to 2 users before it dropped. Other scheduling algorithms
showed decreasing trend as the number of user increase and only has the highest
Peak Throughput at the smallest number of user which is 1.
For average throughput as in Fig. 15.3, all the scheduler showed a somewhat
same behavior. Average throughput decreases as the number of user increases.
BCQI comes to have the highest value in throughout, while other scheduling
15 Performance Evaluation of LTE Scheduling Techniques … 145

Fig. 15.1 Fairness index against number of users

Fig. 15.2 Peak throughput against number of users

146 L. Sukeran et al.

Fig. 15.3 Average throughput against number of users

algorithms are having about the similar values. For the edge users as in Fig. 15.4,
PFS comes out as the best for 1 to 2 users before the MaxMin takes the lead for 5
and above number of users. On the other hand, BCQI only treats the edge user
when there is only 1 user while as the number of user increases to 2, the edge users
are no longer experiencing throughput.
In the second scenario, the speed of the UE was changed to be 120 km/h for
VoIP service. Figure 15.5 shows the fairness as a function of number of users.
MaxMin has the highest value among other schedulers in case of 1 and 2 UEs.
Then Round Robin took the place as the number of users increase from 3 to 40.
Round Robin has the smallest decrement of fairness index compare to other
schedulers. Same as the first scenario, BCQI has the lowest value among the other
schedulers. However, in terms of peak throughput as in Fig. 15.6, BCQI able to
achieve the highest throughput value at 160 Mbps and maintain it until 2 users.
PFS also able to achieve the value of 160 Mbps but dropped sharply as the number
of user increase from 1 to 5 users and above. Other schedulers show almost the
same behavior.
In addition, BCQI shows the best performance in terms of average throughput
as in Fig. 15.7. The other schedulers have about the similar value to each other. All
the schedulers showed the same trend as the value of Average Throughput
decrease exponentially as the number of user increases from 1 to 40 in the cell.
Finally, for the edge users as in Fig. 15.8, RR and RF are able to support more than
5 users at the cell edge. BCQI shows the worst case where only can support 1 user.
15 Performance Evaluation of LTE Scheduling Techniques … 147

Fig. 15.4 Edge user throughput against number of users

Fairness Index
0.9
MaxMin
0.8 ResourceFair
BCQI
0.7
RR
0.6 PropFairSun

0.5

0.4

0.3

0.2

0.1

0
1 2 5 10 20 40
Users

Fig. 15.5 Fairness index against number of users

In all the traffics types’ scenarios, generally, all the results show that the
throughput decreasing as the number of users increasing, same goes for the edge
cell UE. Both of these parameters have inversely proportional relationship to each
other. However in terms of the fairness index, RR algorithm able to maintain the
148 L. Sukeran et al.

Peak Throughput (Mbps)

180

160 MaxMin
ResourceFair
140 BCQI

120 RR
PropFairSun
100

80

60

40

20

0
1 2 5 10 20 40
Users

Fig. 15.6 Peak throughput against number of users

Average Throughput (Mbps)

70
MaxMin
60 ResourceFair
BCQI
50
RR
PropFairSun
40

30

20

10

0
1 2 5 10 20 40
Users

Fig. 15.7 Average throughput against number of users

fairness index value as the number of users increase in several scenarios. In

pedestrian and vehicular scenarios, values of fairness index can be considered as
stagnant and having fixed range except for BCQI algorithm. However, BCQI
performs very well in terms of peak and average throughput.
To sum, BCQI is not suitable for real-time transmission since it seems to cannot
provide the QoS for the VoIP when incomes to cell edge users compare to the
other scheduler especially Round Robin, the reason behind that is the channel
quality for the edge user is worse than the cell center users and using BCQI
scheduler the edge user will not be able to get a good service.
15 Performance Evaluation of LTE Scheduling Techniques … 149

Edge Throughput (Mbps)

16

14
MaxMin
ResourceFair
12
BCQI
RR
10
PropFairSun

0
1 2 5 10 20 40
Users

Fig. 15.8 Edge user throughput against number of users

In terms of efficiency, the BCQI can be seen as the most efficient scheduler
compare to the other scheduler due to the peak throughput achieved has the highest
value show that the spectral efficiency is the highest while, in term of effectiveness,
Round Robin is the most effective scheduler since it still provide throughput to the
edge users and for high number of users, since RR allocates equal time and data
rate for each user in the network. The other scheduler performed moderately.

15.5 Conclusion

All the proposed scheduling algorithms; RR, PFS, RF, MaxMin, and BCQI has
been investigated. Fairness index, peak throughput, average throughput, and edge
users throughput also has been achieved and the entire scheduling algorithm has
been tested for several traffic types. Each one of the scheduling algorithm has
shown performance merit in certain criteria of evaluation. For instance, RR has
shown that it is good for vehicular channel model, while BCQI has shown the best
in achieving peak throughput. However, the best suited scheduling algorithm is
still remains argumentative and in need for extensive and more comprehensive
improvement not just the scheduling algorithm but the system infrastructure as a
whole and also the simulation platform. However, the use of cell edge Relays and
Femtocells is necessary.

Acknowledgments This work was funded by E-Science grant 01-01-08-SF0194 from Malay-
sian ministry of Science, Technology and Innovation (MOSTI).
150 L. Sukeran et al.

References

1. 3GPP: Evolved Universal Terrestrial Radio Access (E-UTRA) and Evolved Universal
Terrestrial Radio Access Network (E-UTRAN); Overall description; Stage 2. TS 36.300, 3rd
Generation Partnership Project (3GPP) (2010)
2. Myung, H.G., Goodman, D.: Single Carrier FDMA New Air Interface for LTE. Wiley, New
York (2008)
3. Goldsmith, A.: Wireless Communications. Cambridge University Press, Cambridge (2005)
4. Ramli, H., Basukala, R., Sandrasegaran, K., Patachaianan, R.: Performance of well known
packet scheduling. In: Proceedings of the IEEE 9th Malaysia International Conference on
Communications, Kuala Lumpur, pp. 815–820, (2009)
5. Zhou, D., Baldo, N., Miozzo, M.: Implementation and validation of LTE downlink schedulers
for ns-3. In: Proceedings of the 6th International ICST Conference on Simulation Tools and
Techniques (SimuTools ‘13). ICST (Institute for Computer Sciences, Social-Informatics and
Telecommunications Engineering), ICST, Brussels, Belgium, Belgium, pp. 211–218 (2013)
6. Safa H., Tohme, K.: LTE uplink scheduling algorithms: performance and challenges. In: 19th
International Conference on Telecommunications (ICT), pp. 1–6, (2012)
7. Piro, G., Grieco, L.A., Boggia, G., Capozzi, F., Camarda, P.: Simulating LTE cellular
systems: an open-source framework. IEEE Trans. Veh. Technol. 60(2), 498–513 (2011)
8. Sahoo, B.: Performance comparison of packet scheduling algorithms for video traffic in LTE
cellular network. Int. J. Mob. Netw. Commun. Telematics (IJMNCT) 3(3), 9–18 (2013)
9. Habaebi, M.H., Chebil, J., Al-Sakkaf, A.G., Dahawi, T.H.: Comparison between scheduling
techniques in long term evolution. IIUM Eng. J. 14(1), 66–75 (2013)
10. Ikuno, J.C., Wrulich, M., Rupp, M.: System level simulation of LTE networks. In:
Proceedings of the 71st IEEE Vehicular Technology Conference, Taipei, Taiwan, May
(2010). http://publik.tuwien.ac.at/files/PubDat_184908.pdf
11. Manor R., Romsey: LTE MAC Scheduler and Radio Bearer QoS (2011)
12. Sun, Z., Yin, C., Yue, G.: Reduced-complexity proportional fair scheduling for OFDMA
systems. In: Proceedings of the IEEE International Conference on Communications, Circuits
and Systems, vol. 2 (2006)
13. Dahlman, E., Parkvall, S., Skold, J., Beming, P.: 3G Evolution HSPA and LTE for Mobile
Broadband, 2nd edn. Elsevier, Amsterdam (2008)
Chapter 16
Design for Energy-Aware IP Over WDM
Networks with Hibernation Mode
and Group-Node Techniques

M.N.M. Warip, Ivan Andonovic, Ivan Glesk, R. Badlishah Ahmad,

P. Ehkan, Mohamed Elshaikh Elobaid Said Ahmed, Shamsul Jamel
Elias and Fazrul Faiz Zakaria

Abstract The focus of the paper is an investigation and evaluation of energy

efficient solutions in IP over WDM core networks using as a foundation, a hier-
archy of hibernation modes implementing different degrees of node groupings and
fibre links establishment that support a sleep state. It seeks to embed this groups-
nodes strategy into an intelligent control plane implementing routing schemes
targeting energy consumption, adaptive signalling and traffic engineering.
A Group-Nodes mechanism is proposed as a function of topology and node dis-
tribution based on a fixed (or geographical) and random (or ownership) principle.
The impact of the proposed technique on energy saving and network performance

M.N.M. Warip  R. Badlishah Ahmad  P. Ehkan (&)  M.E.E.S. Ahmed  S.J. Elias 
F.F. Zakaria
School of Computer and Communication Engineering, Universiti Malaysia Perlis,
Pauh Putra Main Campus, 02600 Arau, Perlis, Malaysia
e-mail: [email protected]
M.N.M. Warip
e-mail: [email protected]
R. Badlishah Ahmad
e-mail: [email protected]
M.E.E.S. Ahmed
e-mail: [email protected]
F.F. Zakaria
e-mail: [email protected]
I. Andonovic  I. Glesk
Department of Electronic and Electrical Engineering, University of Strathclyde,
204 George Street, Glasgow G1 1XW, UK
e-mail: [email protected]
I. Glesk
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 151

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_16
152 M.N.M. Warip et al.

is assessed; results are presented and evaluated for various scenarios. Evaluation of
this methodology indicates potential reduction in power consumption from 7 % up
to 15 % at the expense of reduced network performance.

16.1 Introduction

With regards to zero carbon emission, the issues of green networks technology has
become primary interest among researchers. Recent studies shows that rapid
changes on the information and communication technology (ICT) devices and
deployment of network infrastructures are having a critical effect on carbon
footprint.
For this reason, the power consumption prediction as stated in [1], exhibit that the
worldwide operation of network equipment accounts for 25GW (yearly average) of
the total ICT consumption. Conversely, the joules/bit in telecommunication net-
works is decreasing with time, the joules/user keeps steadily increasing.
So far, the Smart Sleep Mode is the pinnacle of current green networks tech-
nology in which focusing on access networks, offering automated and low power
design mechanisms. However, far too little attention has been paid to core networks.
In this paper, the research centre on the development of energy saving schemes
that support the evolution of greener core IP over Wavelength Division Multi-
plexer (WDM) networks. The cornerstone of the adopted strategy is various
schemes underpinned by the hibernation state implemented through a modification
of the control plane, in particular for transparent network architectures under
different scenarios. The research evaluated the impact and constraints that arise
under this strategy, to provide useful insights on the viability of the approach for
practical energy efficient savings.

16.2 Potential Energy Conservation in Core IP Over WDM

Networks

One of the most momentousness cost minimisation strategies in core IP over

WDM networks (also sometimes known as IP over Optical or optical IP) [2, 3] in
terms of provisioning, operation and maintenance is savings in energy consump-
tion. The major power consumption contributors of equipment in the network are:
• WDM chassis; receiving/transmitting equipment such as transponders modules,
short-reach optical interface transponders.
• Optical Switching (OXC) Chassis, opaque/transparent optical transponders,
Optical-Electrical-Optical (OEO) conversion.
• Core Router: electronic processing, traffic grooming and aggregation in the IP
layer.
• Optical Amplifiers: predominately the Erbium Doped Fibre Amplifier (EDFAs).
16 Design for Energy-Aware IP Over WDM Networks … 153

Fig. 16.1 An energy model

design for IP over WDM Gn LSP Gn+1
networks 10 kW G1 Gn+1

λn
100 W Xn Xn+1

λn λn+1

70 km
120 W Wn A1 An Wn+1

1W

Node N Node N+1

• 3R Regenerators: signal regeneration with re-timing, re-amplification and re-

shaping operations.
• Control Plane: signalling and routing algorithm modification.

16.3 Network Energy Model Design

In our approach in order to evaluate overall network power consumption and

consumed energy per a data bit we used so called equivalent network energy
model (see Fig. 16.1) based on multilayer Internet Protocol/Generalized Multi-
Protocol Label Switching (IP/GMPLS) over optical layers. In this model, a net-
work carrier bandwidth of OC-192 and average energy consumption of 1019 nJ
per bit was assumed following [4, 5, 6].
The parameter Gn denotes a Router’s dissipated power of 10 kW within its
energy consumption of 1000 nJ/bit. Xn represents Optical Cross-Connect (OXC)
dissipating 100 W and consumes 10nJ/bit. Wn denotes Wavelength Division
Multiplexing (WDM) part of the node with dissipating power of 120 W and
energy consumption of 12 nJ/bit. An represents the consumption owing to Erbium-
Doped Fibre Amplifiers (EDFA) within connection spans placed at 70 km inter-
vals, the power consumption is estimated to be 1 W with energy of 0.1 nJ/bit.

16.3.1 Energy Per Bit

Further we define energy per bit consumed by the node as Eb = PT/C where PT
represents the node total power consumption and C is the bandwidth offered by the
network link.
154 M.N.M. Warip et al.

16.3.2 Energy Consumption and Power Consumption

The total power consumption of the link is given by:

2 3 2 3
PTOTAL ¼ 4DpCONTROL PLANE 5 þ 4OXC þ DpWDM þTRANSPONDERS þDpEDFA 5
|fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
IP LAYER OPTICAL LAYER
ð16:1Þ

where DpCONTROL PLANE, DpOXC, DpWDM, DpTRANSPONDERS, and DpEDFA repre-

sent power consumed by the IP/GMPLS router, OXC, WDM, transponders/
transmitters, and EDFA optical amplifier, respectively.
Similarly, the total energy consumption of the link is given by:
2 3 2 3
ETOTAL ¼ 4DeCONTROL PLANE 5 þ 4OXC þ DeWDM þTRANSPONDERS þDeEDFA 5
|fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
IP LAYER OPTICAL LAYER
ð16:2Þ

where DeCONTROL PLANE, DpOXC, DpWDM, DpTRANSPONDERS, and DpEDFA repre-

sent power consumed by the IP/GMPLS router, OXC, WDM, transponders/
transmitters, and EDFA optical amplifier, respectively.
The total energy consumption in IP over WDM networks calculated per data bit
Ebit in order to support the network offered load can be defined as:
2 32 3
nP
1 nP
1 nP
1
X
1 6 CONTROL PLANE 76 DpWDM DpTRANSPONDERS 7
6k¼0 76k¼0 k¼0 7
EBIT ¼ 6 76 þ 7þ
k¼0
4|fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} 5
C 54 C C
IP LAYER OPTICAL LAYER
2 3 ð16:3Þ
2 3
6Xn1 X
n1 7
6 7
þ6 EDFA þ ða þ 1Þ DeOXC 7 þ 4 b 5
4 k¼0 k¼0
5 |{z}
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} CONSTANT
OPTICAL LAYER

where DeEDFA is the energy consumed by EDFA; DeOXC is the energy consumed
by the node’s OXC; b represent the noise factor associated with the Bit Error Rate
(BER) and a heat transfer rate in network equipments; and finally, a is number of
hops.
16 Design for Energy-Aware IP Over WDM Networks … 155

IP Control Plane IP Control Plane IP Control Plane IP Control Plane

1 5 9
2 6 10
OXC 3
OXC 7
OXC 11
OXC
4 8 12

GROUP 1 GROUP 2
Grouping the nodes

Fig. 16.2 Hibernation mode: grouped nodes structure

16.4 Hibernation Mode: Group Nodes Distribution Factors

The Group Node structure in relation to the hibernation mode is illustrated in

Fig. 16.2. In this architecture, nodes consisting of IP Router and OXC are inter-
connected by point-to-point optical fibre links.
The exchange of messages updating energy consumption profiles can be
described by a sequence diagram; Fig. 16.3 illustrates the messages sequence
diagram for the transition from ON to OFF state. For end-to-end provisioning -
from ingress to egress node—after path computation, the wavelength is reserved
by signalling. Node 1 transmits a connection request along the link to reserve the
wavelength and establish an end-to-end channel through the LSP Path/Resv
Message. The LSP setup request is then forwarded to the next node until the
message request is at the egress node. If there is an idle node at an intermediate
node (in this case Node 3), after the nested hold-off timer expires, the hibernation
notification message is propagated back along the path to release the reserved
wavelength. The loopback LSP Resv_Confirm message will be transmitted back
along the link until at the ingress node to request the suspension of the idle node. If
Node 3 receives a LSP setup request message to place the node in Sleep state, it
sends a Resv_Err message to acknowledge the ingress node that Node 3 is in
powering off state. As a result, the network updates the routing table and TED
topology. The ingress node releases a LSP by propagating Path_Tear message and
powers down the connection to the idle node.
Figure 16.4 presents a message sequence diagram for the transition from OFF state
to ON state. Node 3 detects traffic and changes its state to BUSY (active transition
state) and full power operation is resumed, confirmed by sending a notify message to
inform the adjacent node that it is in the process of waking-up. A Resv_Tear message is
sent to the ingress node to notify that node is powering to ON state.

16.4.1 Fixed (Geographical) Nodes Effect

Fixed node or Geographical Node groups are defined as a grouping topology that
contains selected neighbouring nodes and grouped as disjoint clusters.
156 M.N.M. Warip et al.

Link 1 Link 2 Link 3 Link 4

Ingress NODE 1 NODE 2 NODE 3 NODE 4 NODE 5 Egress

Path
Signalling request Path
Path
Path

Resv Signalling acknowledged

Resv
Resv
Resv
Signalling completed
Node 3: Idle

Notify (Hibernate)
Hibernation Mode: Notify (Hibernate)
Group-Nodes
LSP Setup Request
LSP Setup Request

Resv_Confirm Idle
Hibernation Mode Resv_Confirm (Suspend Node) Node
acknowledge (Suspend Node)

LSP Setup Request

LSP Setup Request

Resv_Err (Powering off)

Resv_Err (Powering off)
OFF state
OSPF-TE
Routing updates OSPF-TE
OSPF-TE
OSPF-TE
PathTear (Release) Routing updates
LSP Teardown PathTear
LSP Teardown

Fig. 16.3 Messages sequence diagram for transition from ON state to OFF state of hibernation:
group-nodes

Link 1 Link 2 Link 3 Link 4

Ingress NODE 1 NODE 2 NODE 3 NODE 4 NODE 5 Egress

Node 3: Idle

Notify (Hibernate)
Hibernation Mode: Notify (Hibernate)
Group-Nodes
LSP Setup Request
LSP Setup Request

Resv_Confirm
Idle
Hibernation Mode Resv_Confirm (Suspend Node)
acknowledge
Node
(Suspend Node)

LSP Setup Request

LSP Setup Request

Resv_Err (Powering off)

Resv_Err (Powering off)
Sleep state

Node 3: Busy

Notify (Wake-up)
Notify (Wake-up)
WAIT awake
Path
Path

Resv (Wake-up) Active

Resv (Wake-up) Node

Transition LSP Setup Request

Hibernation Mode LSP Setup Request

Resv_Tear
Resv_Tear (Powering on)
ON state
(Powering on)
Path_Tear
LSP Teardown
(Release) Path_Tear
LSP Teardown
(Release)

Fig. 16.4 Messages sequence diagram for transition from OFF state to ON state of hibernation:
group-nodes
16 Design for Energy-Aware IP Over WDM Networks … 157

16.4.2 Random (Ownership) Nodes Effect

Random node or Ownership-based node groupings are defined as nodes belonging

to the same owner (service provider) e.g. organization that having many entities
under the same company name.

16.5 Simulation Results

The Group-Nodes hibernation mode that invokes cluster based architectures was
evaluated and investigated. By dividing the nodes into several disjoint sets, as well
as providing each node with geographical and ownership topology settings, pro-
duces clusters adopting sleep cycles to reduce power consumption. Figure 16.5
illustrates the network topology which is being utilised by the European Optical
Network (EON network). The EON network has a Full Mesh Network Topology
with 9 nodes and 20 bidirectional fibre links [7, 8]. The IP/GMPLS nodes are
linked by bidirectional pairs of single mode fibres. The EON network topology
was used in our simulations and was based on a discrete event modelling tool
known as OMNet++ (Object Modular Network Tested in C++). It has been
assumed that all links are equal in terms of number of wavelengths (eight), that the
message length is fixed at 256 bytes, and a nodal processing delay is 20 ms. All
EON network Nodes are capable to maintain information on their total power
consumption as well as energy per bit consumed. Wherein, the standard GMPLS
signalling and routing protocols are implemented following the Internet Engi-
neering Task Force (IETF) standard [9, 10].
The performance metrics takes into account the average power consumption,
blocking probability and average request blocking [9–13]. We also assumed that
lightpath requests are uniformly distributed. Note that, the inter-arrival connection
requests are independent Poisson processes with an arrival rate of a and the queue
lengths exponentially distributed with the expected service rate time of 1/l mea-
sured in seconds. Therefore, the network offered load is a/l.
Full Mesh Network Topology: European Optical Network (EON) (Fig. 16.5).
The average power dissipation and energy consumption values assigned to each
node are captured in the network energy model. In this architecture, the power
consumption of nodes comprises the core router (10 kW), OXC (100 W), WDM
(120 W) and EDFAs (1 W) placed at 70 km intervals along links. For example,
the power consumption (reference value) between Node A and Node G (Fig. 16.5)
linked by the single mode optical fibre across a distance of 2090 km at a data rate
of 10 Gb/s is 29 W. Therefore, the total power consumption between Node A and
Node G is 10.249 kW and the energy per bit is 1024.9 nJ (Eqs. 16.1–16.3).
The proposed Hibernation concept was verified on described EON by imple-
menting group-nodes schemes. These nodes/links are then put into ‘‘hibernation’’
or an ‘‘SLEEP state’’, in which nodes have suspended their unused functionalities
158 M.N.M. Warip et al.

10.239kW D 10.242kW
1023.9nJ 1024.2nJ
10.236kW 10.247kW
1023.6nJ 1024.7nJ
10.249kW
A G
1024.9nJ
10.235kW
1023.5nJ 10.236kW
1023.6nJ
10.236kW
10.236kW
1023.6nJ 10.239kW
1023.6nJ
C E 1023.9nJ

10.237kW 10.242kW I
1023.7nJ 1024.2nJ 10.236kW
1023.6nJ 10.239kW
10.239kW
1023.9nJ
1023.9nJ
10.239kW
1023.9nJ F 10.246kW 10.236kW
1024.6nJ 1023.6nJ
B 10.235kW
1023.5nJ
10.245kW H
1024.5nJ

Fig. 16.5 Power consumption for full mesh European Optical Network (EON)

Table 16.1 EON node grouping categorization

Group Geographical Ownership-based
G1 Nodes {A, B} Nodes {B, H}
G2 Nodes {C, D, E} Nodes {D, E, G}
G3 Nodes {F, G, H, I} Nodes {A, C, F, I}

(e.g. unused ports/interfaces, Mux/DeMux capabilities, signalling gates, unused

wavelengths, etc.) and keep only the minimum network operation activities.
The aim is to evaluate the impact of a route to energy saving through hiber-
nation in core optical IP network by performing the Fixed (Geographical) grouping
and Random (Ownership) grouping for full mesh network topologies. Note that,
these results are based on cross-layer optical/IP domains integration and previous
research produced results based in the optical domain only. Therefore, results are
difficult to compare in terms of power savings and network performance.
Using the full mesh EON (Fig. 16.5), hibernation settings are applied to net-
work nodes based on the group membership type (see node grouping categoriza-
tion in Table 16.1).
Figures 16.6 and 16.7 present the average power consumption and average
request blocking for various ‘‘Geographical’’ (adjacent nodes) groupings as a
function of offered network load for the EON network mesh topology. For ‘‘All
Groups HM’’ (case when nodes’ unused functionalities are suspended)
*0.137 kW of power is saved per node’ (Fig. 16.6) but the probability of
blocking is *55 % (Fig. 16.7). For Groups ‘‘G2 and G3 = HM’’ or Groups ‘‘G1
and G2 = HM’’, the power savings of 0.12 kW or 0.10 kW is obtained respec-
tively (Fig. 16.6) with a corresponding blocking probability of 33 % or 10 %,
respectively (Fig. 16.7).
16 Design for Energy-Aware IP Over WDM Networks … 159

Fig. 16.6 Simulation results for different EON Fixed (geographical) node groupings. ‘HM’
stands for hibernation mode

Fig. 16.7 Request blocking for EON Fixed (Geographical) grouping of nodes

‘‘All Groups ON’’ yields the lowest blocking probability but the power savings
per node are minimal. The EON network also becomes congested for network
loads exceeding 8 Erlangs. The trade-off between a reduction in energy con-
sumption and the probability of blocking is evident.
160 M.N.M. Warip et al.

Fig. 16.8 Average power consumption difference per node as a function of network load for
different EON random (Ownership) node grouping

Fig. 16.9 Request blocking probability as a function of network load for EON random
(Ownership) grouping of nodes

Figure 16.8 depicts the average power consumption for different groups with
respect to offered network load and Fig. 16.9 presents the blocking probability for
the EON ownership (random) grouping of nodes. As expected, savings in power
with ownership grouping in the full mesh (EON) topology improves when
16 Design for Energy-Aware IP Over WDM Networks … 161

compared to the partial mesh (NSFnet) topology, particularly in the case of

ownership grouping. In this case, the power savings for grouping ‘‘G2 and G3’’ or
‘‘G1 and G2’’ are most significant, the improvement being *0.135 and *0.11 kW
respectively (Fig. 16.8); the corresponding blocking probability is 29 and 8 %,
respectively (Fig. 16.9). The reason for this is that the optical bypass at interme-
diate nodes reduces the number of required core router electrical ports in the IP
layer and thereby, the energy owing to this electrical equipment is saved. The EON
ownership-based node groupings (Figs. 16.6 and 16.8), ‘‘All Groups = HM’’ and
‘‘All Groups = ON’’ deliver similar results as the geographical node groupings.

16.6 Conclusion

In this paper we have presented an evaluation of an approach to energy saving in

IP over WDM networks-based GMPLS control plane based on hibernation. The
impact of the grouping of nodes following the principles of hibernation on network
performance as a function of energy saving was examined and quantified for two
representative network topologies. The approach was implemented through two
grouping network node strategies—Geographical and Ownership-based. Results
show that hibernation has the potential to deliver energy savings at the expense of
reduced network performance. Results show that the ‘‘Ownership-based node
groupings for ‘‘All Groups = HM’’ (hibernate) and ‘‘All Groups = ON’’ delivers
similar performance as the Geographical nodes groupings.

Acknowledgments The research leading to these results has received funding from the Ministry
of Education Malaysia under grant Fundamental Research Grant Scheme (FRGS).

References

1. Pickavet, M., Vereecken, W., Dameyer, S., Audenaert, P., Vermeulen, B., Develder, Colle,
D., Dhoedt, B., Demeester, P.: World energy needs for ICT: the rise of power-aware
networking. In: Proceedings of the Advanced Networks and Telecommunication Systems,
Bombay, India (2008)
2. Musumecci, F., Tornatore, M., Pattavina, A.: A power consumption analysis for IP-over-
WDM core network architecture. IEEE/OSA J. Opt. Comms. Netw. 4, 108–117 (2012)
3. Rajagopalan, B., Luciani, J.V., Awduche D.O.: IP over optical networks: a framework.
RFC3717 IETF. (2004)
4. Balinga, J., Ayre, R., Hinton, K., Sorin, W.V., Tucker, R.S.: Energy consumption in optical
IP networks. IEEE J. Lightw. Technol. 27(13), 2391–2403 (2009)
5. Bathula, B.G., Alresheedi, M., Elmirghani, J.M.H.: Energy efficient architectures for optical
networks. In: Proceedings of the London Communications Symposium, pp. 1–4. University
College, London (2009)
6. Cisco Systems Data Sheets: http://www.cisco.com (2013)
7. Shen, G., Tucker, R.S.: Energy-minimized design for IP over WDM networks. IEEE/OSA J.
Opt. Comms. Netw. 1(1), 176–186 (2009)
162 M.N.M. Warip et al.

8. Kim, Y., Lee, C., Kevin Rhee, J.K., Lee, S.: IP over WDM cross-layer design for green
optical networking with energy proportionally consideration. IEEE. J. Lightw. Technol.
30(13), 2088–2096 (2012)
9. Albarrak, S.: Failure recovery in distributed GMPLS-based IP-over-optical networks. Ph.D.
thesis, Department of Electronic and Electrical Engineering, University of Strathclyde,
Glasgow, UK (2008)
10. Chabarek, J., Sommers, J., Barford, P., Estan, C., Tsiang, D., Wright, S.: Power awareness in
network design and routing. In: Proceedings of the 27th Conference on Computer
Communications INFOCOM, Phoenix, pp. 1130–1138 (2008)
11. Ben Yoo, S.: Energy efficiency in the future internet: the role of optical packet switching and
optical-label switching. IEEE J. Sel. Quantum Electr. 17(2), 406–418 (2011)
12. Berger, L.: Generalized Multi-Protocol Label Switching (GMPLS) signalling functional
description. RFC3471. IETF (2003)
13. Lang, J.: Generalized Multi-Protocol Label Switching (GMPLS) signalling Resource
ReserVation Protocol-Traffic Engineering (RSVP-TE) extension. RFC3473. IETF (2003)
Chapter 17
Content Based Image Retrieval Using
Color Layout Descriptor and Generic
Fourier Descriptor

Muhammad Imran, Rathiah Hashim and Noor Elaiza

Abstract Image databases have been a significant part of systems or applications

in many fields including education, forensics and medical sciences. However, due
to the increase in the number of digital resources especially images, the demand to
manage these databases such as storing and retrieving the images effectively and
efficiently from these databases is crucial. Content Based Image Retrieval (CBIR)
is one of the best techniques used to retrieve similar images from the image
database. Even though it is a good and well-researched technique, still there are
some issues to be improved. One of the challenges in CBIR is to represent the
image as a feature vector. In this paper, we proposed a new feature vector using
Generic Fourier Descriptor (GFD) and Color Layout Descriptor (CLD) to increase
the accuracy of the retrieval process. We tested our technique with the Coral
Database and compared the results with other CBIR techniques. Proposed method
achieved better results than previous techniques. The proposed technique can be
used by the forensic department for identification of criminal suspect using
forensic database.

Keywords CLD  CBIR  Color layout descriptor  GFD

M. Imran (&)  R. Hashim

University Tun Hussein Onn Malaysia, Batu Pahat, Malaysia
e-mail: [email protected]
R. Hashim
e-mail: [email protected]
N. Elaiza
University Teknologi MARA Malaysia, Shah Alam, Malaysia
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 163

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_17
164 M. Imran et al.

17.1 Introduction

Due to powerful processors and cheap memories, a variety of applications are

using large image databases. Different fields such as medicine, geography,
advertising, architecture and design are using large image databases to help their
users [1]. Therefore, an effective and efficient image retrieval technique is now a
necessity. Two approaches to search the image are available. One is the Keyword
Based Image Retrieval (KBIR) and the other is Content Based Image Retrieval
(CBIR). The KBIR searches the images based on keywords tagged with an image.
KBIR faces many problems—keywords are manually annotated, the person adding
the keyword may have a different perception about the image than the one who is
searching the image. To overcome these drawbacks, researchers introduced CBIR.
In CBIR, searching is performed by the visual features of the images. Therefore,
there is no further need for manual keywords annotation.
CBIR used color, texture and shape features of the image. In CBIR, features of
all images in the image database are extracted and saved as feature database. When
a user input query image to the system, system checks the similarity between the
query image features and image database feature through any similarity measure
such as Euclidean distance, Manhattan distance, earth movers distance etc. and
display ranked result to the user. Due to the high demand for searching from an
image database, CBIR becomes a popular image searching technique. Particle
swarm optimization (PSO) with relevance feedback was used by the Broilo [2] to
enhance the performance of CBIR. Broilo formulates the image retrieval processes
as an optimization problem and applied PSO on CBIR. PSO was proposed by
Kennedy and Eberhart [3] in 1995 and modified by different research works such
as [4–6].
However, CBIR is facing problems and accuracy achieved up to a limit only. In
this paper, we proposed a new signature development approach for CBIR. We used
the color and shape features to search similar images. Color Layout Descriptor
(CLD) is used for the color feature extraction and Generic Fourier Descriptor
(GFD) is used for shape feature extraction. Both feature vectors are combined and
then, similarity measure is performed using Manhattan distance. Section 17.2
discusses about the related work and proposed approach is illustrated in Sect. 17.3.
Sections17.4 and 17.5 are reserved for the results and analysis. Finally, the paper is
concluded in Sect. 17.5.

17.2 Related Works

Bhuravarjula and kumar [7] used color moments to enhance the accuracy of CBIR.
They divided the image into four segments and then extracted the color moments
from all segments and clustered them into four classes. Finally, they calculated the
means moments for each class which was further used for the similarity measure
17 Content Based Image Retrieval Using Color Layout … 165

between the query image and the database images. Coral Database was used to
perform experimental work. Precision and recall were used as performance metrics
and results were compared with the previous CBIR feature extraction techniques.
Imran et al. [8] used color histogram for searching similar images. Center
moment is adopted to describe the histogram. Each image is divided into 4 9 4
sub images and each sub images is divided into HSV components to generate the
histogram. Experiments are performed on Coral Database using precision as
performance metric.
Zhang and Zou [9] used the color feature and edge direction features for CBIR.
For edge direction feature, Edge Histogram Descriptor (EHD) of MPEG-7 was
used. Experiments were performed using Coral Database to assess the performance
of proposed technique. Recall and Precision are calculated as performance metrics.
The combination of color and texture feature was applied by Soman et al. [10].
Discrete Cosine Transforms (DCT) was used for texture feature extraction while
mean, deviation and skewness from color moments were used for the color feature
extraction. The proposed technique was tested on the Coral Database having 1000
images and achieved better accuracy then previous CBIR techniques. Hema-
chandran and Singh [11] combined color moments with Gabor Texture descriptor.
The image is divided into horizontally three non-overlapping regions to extract
first three moments of the color distribution. Performance of proposed technique is
tested using Coral Database, which contains images from 10 different categories
and each category has 100 images.
Abubacker et al. [12] used color, texture and shape features for the image. For
the color feature, they used the spatial based color moments. First they divided the
image into 25 blocks then calculated the Red Green Blue (RGB) values of each
block. RGB values are converted to Hue, Saturation Intensity (HSI). According to
the author the three color moments; mean, variance and skewness are effective and
efficient for the color distribution of images. The formula for mean, variance and
skewness are given below:

1X N
Meanðli Þ ¼ fij ð17:1Þ
N j¼1

!1=2
1X N  2
Varianceðri Þ ¼ fij  li ð17:2Þ
N j¼1

!1=3
1X n  3
SkewnessðSi Þ ¼ fij  li ð17:3Þ
N J¼1

where, fij is the value of the ith color component of the image block j and N is the
number of blocks in the image. For the texture feature, author used the Gabor filter.
Author applied the 2D Gabor function to obtain the set of Gabor filter with dif-
ferent scale and orientation. By using Gabor filter, author performed convolution
166 M. Imran et al.

on the image to obtain the Gabor transform. Invariant shape features were used to
extract the shape features. Following are the steps taken by the author to extract
the shape feature.
1. Based on the threshold value the image is converted to the binary image.
2. Using canny algorithm the edges of the binary image are detected.
3. The centroid of the object is obtained by arranging the pixels in clockwise
order and forming a closed polygon.
4. The centroid distance and complex coordinate function of the edges is found.
5. The farthest points are found and Fourier transform is applied on them.

17.3 Proposed Approach

To represent the image as feature vector we used Shape and Color features. To
extract the shape features, GDF is used. About the detail of GFD, we refer our
readers to Zhang and Lu [13]. GFD containing 36 features reflecting 4 radial
frequencies and 9 angular frequencies are selected to index the shape. The selected
36 features build a feature vector which is combined with the CLD feature vector
to form the final feature vector. Color feature is one of the basic features used to
retrieve the images. It is most intuitive and obvious feature of the image and color
is easily extracted. To describe color information different methods are available.
In this paper, we have used the CLD from MPEG-7 standard to extract the color
information from the image. Following are the steps to extract the color feature
vector from the image.
1. Divide the image into 8 9 8 blocks
2. Single representative color is selected from each block
3. The selection results in a tiny image icon of size 8 9 8
4. The color space is converted from RGB to YCbCr
5. The luminance (Y), blue and red chrominance (Cb and Cr) are transformed by
8 9 8 DCT
6. A zigzag scanning is performed on these three sets of DCT coefficients
7. As a result we obtain three matrixes for each block of Y, Cb and Cr color space
8. Take sum of each matrix to get three feature vectors
9. Finally horizontally concatenate three feature vectors to obtain a final feature
Vector for an image
The feature vector calculated by GFD and CLD are combined in one feature
vector which represent the each image in the database. When user input query
image to the system to search similar images, the similarity between the query
image and the database images is calculated using Manhattan Distance and results
are sorted, ranked and presented to the user.
17 Content Based Image Retrieval Using Color Layout … 167

17.4 Results

To assess the proposed technique, experiments were performed using the Coral
Database. The database contains images from 10 different categories and each
category has 100 images, so a total of 1000 images are available in the Coral
Database. To validate the proposed technique, the results were compared with
Variance Segment Method [7] and Histogram based taken from [14]. The proposed
approach is implemented in Matlab 2010b. For performance evolution precision is
used as performance metric, which is reported by previous work such as Hiremath
et al. [15], Banerjee et al. [14] and Wang et al. [16]. Following expression is used
to calculate the precision.

Number of True Positive

Precision ¼
Number of True Possitive þ False Possitive

In our experiments, we select 10 random images from each category as the

query image and perform retrieval process. The process is repeated for 10 times
and category wise average precision is calculated. Average precision is calculated
for 20, 30 and 40 top retrieval where 20 top retrieval means 20 most similar images
are displayed and this can be represented as P@20. The results achieved by the
proposed feature vector technique are shown in Table 17.1 (Fig. 17.1).

17.4.1 Comparison with Previous Methods

Table 17.2 illustrates the comparison of the proposed technique with Simple Hist
and Variance Segment Method.

Table 17.1 Performance of proposed technique at different top retrieval

Class P@40 P@30 P@20
Africa 0.27 0.32 0.27
Beach 0.44 0.34 0.48
Buildings 0.22 0.26 0.29
Buses 0.37 0.36 0.44
Dinosaurs 1.00 1.00 1.00
Elephant 0.49 0.50 0.64
Flower 0.82 0.83 0.86
Horses 0.62 0.68 0.75
Mountains 0.40 0.19 0.44
Food 0.23 0.29 0.28
Avg 0.48 0.50 0.54
168 M. Imran et al.

Fig. 17.1 Average precision of proposed method at different top retrieval

17.5 Analysis

From Table 17.2 and Fig. 17.2, it is obvious that the proposed method has better
results than Histogram based and variance segment method. The overall average
precision of Histogram based method, Variance Segment and proposed method are
0.36, 0.306 and 0.50 respectively. Histogram based method achieved 36 %, Var-
iance Segment method achieved 30 % while proposed technique achieved 50 %
accuracy rate. Figure 17.2 illustrates the comparison of proposed method with
Simple Hist where average accuracy of proposed method is calculated for top 30
and 20 retrieval results. The graph depicted that the proposed method has better
precision than both previous methods. Category wise performance comparison of
the proposed method with Variance Segment method is presented in Fig. 17.3.

Table 17.2 Comparison of proposed method with previous methods

Class Simple hist Variance segment Proposed method
Africa 0.3 0.13 0.32
Beach 0.3 0.26 0.34
Buildings 0.25 0.11 0.26
Buses 0.26 0.17 0.36
Dinosaurs 0.9 0.96 1.00
Elephant 0.36 0.34 0.5
Flower 0.4 0.49 0.83
Horses 0.38 0.2 0.68
Mountains 0.25 0.25 0.4
Food 0.2 0.15 0.29
Avg 0.36 0.306 0.5
17 Content Based Image Retrieval Using Color Layout … 169

Fig. 17.2 Comparison of the proposed method with simple hist and variance segment

Fig. 17.3 Category-wise comparison of proposed method (P@30) with previous methods

For all categories, proposed method achieved higher average precision than Var-
iance Segment method. For the class dinosaurs, the proposed method has 100 %
retrieval rate as the precision reaches 1.0.

17.6 Conclusion

This paper proposed and implemented a new signature technique to improve the
performance of CBIR. This new signature technique is the combination of the
color and shape features. Color features are extracted using CLD while GFD is
used to extract the shape features. The performance of proposed technique is tested
on coral Database containing the images from 10 different categories. To validate
170 M. Imran et al.

the proposed technique, results were compared with the previous CBIR tech-
niques. The proposed technique increased the performance of CBIR in term of
accuracy. Previous techniques achieved 30 % and 36 % while proposed technique
achieved 50 % accuracy. For future work, the performance of the proposed sys-
tem, texture feature can be combined with color and shape features.

Acknowledgments The researchers would like to thank University Tun Hussein Onn Malaysia
(UTHM) for supporting this project under Project Vote No 1315

References

1. Lee, S.M., Bae, H.J., Jung, S.H.: Efficient content-based image retrieval methods using color
and texture. ETRI J. 20(3), 272–283 (1998)
2. Broilo, M., Natale, F.G.B.D.: A stochastic approach to image retrieval using relevance
feedback and particle swarm optimization. IEEE Trans. Multimedia 12, 11 (2010)
3. Kennedy, J., Eberhart, R.: Particle swarm optimization In: IEEE International Conference on
Neural Networks, pp. 1942–1948 (1995)
4. M. Imran, et al., ‘‘Modified Particle Swarm Optimization with student T mutation (STPSO),’’
in Computer Networks and Information Technology (ICCNIT), 2011 International
Conference on, 2011, pp. 283-286
5. Imran, M., et al.: Particle swarm optimization (PSO) variants with triangular mutation.
J. Eng. Technol. (2013)
6. Imran, M., et al.: Opposition based particle swarm optimization with student T mutation
(OSTPSO). In: Data Mining and Optimization (DMO), 2012 4th Conference on, pp. 80–85
(2012)
7. Bhuravarjula, H., Kumar, V.: A novel content based image retrieval using variance color
moment. Int. J. Comput. Electron. Res. 1(3), 93–99 (2012)
8. Imran, M., et al.: New Approach to Image Retrieval Based on Color Histogram. In: Tan, Y.,
et al. (eds.) Advances in Swarm Intelligence, vol. 7929, pp. 453–462. Springer, Berlin
Heidelberg (2013)
9. Zhang, J., Zou, W.: Content based image retrieval using color and edge direction features. In:
IEEE 2nd International Conference on Advanced Computer Control (ICACC), vol. 5,
pp. 459–462 (2010)
10. Soman, S., Ghorpade, M., Sonone, V., Chavan, S.: Content based image retrieval using
advanced color and texture features. In: International Conference in Computational
Intelligence (ICCIA), vol. 3, (2012)
11. Singh, S.M., Hemachandran, K.: Content-based image retrieval using color moment and
gabor texture feature. Int. J. Comput. Sci. Issues (IJCSI) 9(5) 299 (2012)
12. Abubacker, K., Indumathi, L.: Attribute associated image retrieval and similarity re ranking.
In: International Conference on Communication and Computational Intelligence (INCOCCI),
pp. 235–240 December 2010
13. Zhang, D., Lu, G.: Generic fourier descriptor for shape-based image retrieval, in proceedings.
In: 2002 IEEE International Conference on Multimedia and Expo, pp. 425–428 (2002)
14. Banerjee, M., Kundu, M.K., Maji, P.: Content-based image retrieval using visually significant
point features. Fuzzy Sets Syst. 160(23), 3323–3341 (2009)
15. Hiremath, P., Pujari, J.: Content based image retrieval using color boosted salient points and
shape features of an image. Int. J. Image Process. 2(1), 10–17 (2008)
16. Wang, J., Li, J., Wiederhold, G.: Simplicity: semantics-sensitive integrated matching for
picture libraries. IEEE Trans. Pattern Anal. Mach. Intell. 23(9), 947–963 (2001)
Chapter 18
Pilot Based Pre FFT Signal to Noise
Ratio Estimation for OFDM Systems
in Rayleigh-Fading Channel

A.M. Khan, Varun Jeoti and M. Azman Zakariya

Abstract In design of adaptive orthogonal frequency division multiplexing

(OFDM) transmission, Signal-to-Noise Ratio (SNR) is well-known to be an
informative performance measures. Channel impairments, specifically in wireless
communication degrade the performance of accurate SNR measurement and said
problem becomes more serious in low SNR regime. As a result it reduces the
performance of adaptive transmission in OFDM systems. To alleviate this problem
we develop a pilot based Pre FFT (time domain) Signal-to-Noise Ratio (SNR)
estimator in the presence of frequency selective Rayleigh fading channel. A novel
time domain SNR estimation technique is proposed that accurately measures the
Signal-to-Noise Ratio in low SNR regime, where signal plus noise power is
evaluated by using autocorrelation of received signal and signal power is indirectly
estimated from pilot power using cross correlation. Simulation results show that
proposed method has very less bias and very close to the actual SNR values.

18.1 Introduction

OFDM has been applied widely in wireless communication systems due to its high
data rate transmission capability with high bandwidth efficiency and its robustness
to multipath delay. OFDM introduces Cyclic Prefix (CP), which eliminates the
Inter-Symbol-Interference (ISI) between OFDM symbols [1] and its high data rates
are available without having to pay for extra bandwidth. With these advantages,
OFDM is widely accepted in numerous wireless standards such as Digital Video

A.M. Khan (&)  V. Jeoti  M. Azman Zakariya

Department of Electrical and Electronic Engineeirng, Universiti Technologi PETRONAS,
Tronoh, Malaysia
e-mail: [email protected]
V. Jeoti
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 171

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_18
172 A.M. Khan et al.

Broadcasting (DVB), Digital Audio Broadcasting (DAB), Wireless Metropolitan

Area Network (WMAN) and Wireless Local Area Network (WLAN).
In order to exploit all these advantages and optimize the performance of OFDM
systems; channel state information (CSI) plays a vital role. A Signal-to-noise ratio
(SNR) gives a comprehensive knowledge of CSI, which allows us to decide
whether a transition to higher bit rates would be favorable or not. Similarly, SNR
estimation is required for power control, adaptive modulation, coding and channel
estimation. Therefore major input parameter for fourth generation adaptive system
is a good SNR estimator.
Mainly there are two types of SNR estimation [2], which are as follows: (1)
Data-aided (DA-SNR) estimator. (2) Non-data aided (NDA-SNR) estimator. DA is
further divided into two types: (1) pilot-aided and (2) training sequence. In pilot-
aided scheme, known information is transmitted together with data. While, in
training sequence known information is transmitted over one or more OFDM
symbols without data. In NDA estimator, no known data is transmitted and
therefore SNR is estimated at the receiver blindly. In this work, an estimator of
type DA-SNR (pilot-aided) is proposed because it works well in multipath fading
channel.
One of the most significant current discussions in SNR estimation is Pre FFT
(time domain) and Post FFT (frequency domain) estimators. Several studies have
been produced SNR estimation in frequency domain, but little attention has been
paid to time domain. Popular SNR estimators reported in literature are Boumard’s,
Ren’s and Minimum Mean Square Error (MMSE) [3].
In [4], it was assumed that the channel conditions throughout observation are
same. However, in highly frequency selective multipath channels, the assumption
is not valid, and the performance is degraded greatly. Therefore this problem is
overcome in [5], but its complexity is similar in terms of addition and multipli-
cations. In addition, pilot aided MMSE [6] is investigated in frequency selective
fading channel but it suffers when noise level increases and SNR fluctuates from
its threshold value.
Recently, in [7] Pre FFT pilot aided SNR estimation is proposed where low
MSE is achieved in flat fading channel. Therefore, to the best of our knowledge no
work has been published for pilot aided Pre FFT SNR estimation over frequency
selective Rayleigh fading channel that works well in low SNR environment.
A novel estimator presented here differs from previous methods in a way that it
is pilot aided SNR estimation in time domain, by using autocorrelation of received
signal. In addition, proposed method can accurately measure SNR which has very
less bias and very close to the actual SNR values. It can also be seen that proposed
SNR estimation performs better at low SNR.
This paper has been organized in the following way. In Sect. 18.2, overview of
proposed Pre FFT SNR estimation technique is explained. Section 18.3 addressed
the methodology used by proposed estimator. Section 18.4 described the channel
model used in simulation. Results and analysis are discussed in Sect. 18.5. Con-
clusion of the work is presented in Sect. 18.6.
18 Pilot Based Pre FFT Signal to Noise Ratio Estimation … 173

Fig. 18.1 Typical OFDM baseband transceiver [8]

18.2 Proposed Technique

Pre FFT OFDM system model is shown in Fig. 18.1. It begins with ‘‘signal
mapping’’ block where binary information is mapped according to modulation.
Then serial to parallel conversion takes place. After that, pilot sub-carriers P(k) are
inserted along with data sub-carriers D(k), arrangement of pilots and data sub-
carriers are shown in Fig. 18.2. Therefore S(k) can be written as

SðkÞ ¼ DðkÞ þ PðkÞ ð18:1Þ

Then IFFT is applied on S(k) samples that transform frequency domain samples
S(k) into time domain s(n), which can be shown as

X
N 1 j2Pkn
sðnÞ ¼ IFFTfSðkÞg ¼ SðkÞeNFFT ð18:2Þ
k¼0

where NFFT is no of FFT points and s(n) can also be written as

sðnÞ ¼ dðnÞ þ pðnÞ ð18:3Þ

After passing through multi-path frequency selective fading channel, receiver

input signal is first passed through serial to parallel converter and then guard band
removal takes place. Therefore signal r(n) is obtained, which can be shown as

rðnÞ ¼ hðnÞ  sðnÞ þ wðnÞ ð18:4Þ

where h(n) is channel impulse response and w(n) is additive white Gaussian noise.
The h(n) can be expressed as
174 A.M. Khan et al.

Fig. 18.2 Proposed pilots and data sub carriers arrangement

hðnÞ ¼ ½h0 h1 h2 . . .hT1  ð18:5Þ

where h0, h1…hT are the coefficient of channel impulse response and T is total no
of channel taps. By substituting Eq. (18.3) in Eq. (18.4), r(n) is given by

rðnÞ ¼ hðnÞ  ½dðnÞ þ pðnÞ þ wðnÞ ð18:6Þ

The SNR estimation technique is such that it attempts to estimate the signal
power using a priori known pilots and determines the noise power from it using
measurement of total received power. Towards that end, autocorrelation of the
received signal is performed to measure the total received power which is the
value at lag zero. And in order to estimate the signal power, pilot power is
estimated with the help of cross correlation and indirectly the signal power is
deduced from it. Hence, moving on, we calculate the autocorrelation of complex
valued received signal, written as

X
n¼þ1
Rrr ðlÞ ¼ rðnÞr  ðn þ lÞ ð18:7Þ
n¼1

where
18 Pilot Based Pre FFT Signal to Noise Ratio Estimation … 175

r  ðn þ lÞ ¼ h ðn þ lÞ  ½d ðn þ lÞ þ p ðn þ lÞ þ w ðn þ lÞ ð18:8Þ

By substituting the Eqs. (18.6) and (18.8) in Eq. (18.7), the autocorrelation
Rrr(l) is written as
nX
¼1
Rrr ðlÞ ¼ fhðnÞ  ½dðnÞ þ pðnÞ þ wðnÞ ð18:9Þ
n¼1
 hðn þ lÞ  ½d ðn þ lÞ þ p ðn þ lÞ þ w ðn þ lÞg
nX
¼1  
Rrr ðlÞ ¼ fRhh ðlÞ  Rdd ðnÞ þ Rdp ðnÞ þ hðnÞ  Rdw ðnÞ
n¼1
  ð18:10Þ
þ Rhh ðlÞ  Rpd ðnÞ þ Rpp ðnÞ þ hðnÞ  Rpw ðnÞ
 
þ h ðn þ lÞ  Rwd ðnÞ þ Rwp ðnÞ þ Rww ðnÞg

At zero lag when l = 0

Rdp ð0Þ ¼ Rpd ð0Þ ffi 0 ð18:11Þ

Rpw ð0Þ ¼ Rwp ð0Þ ffi 0 ð18:12Þ

Rdw ð0Þ ¼ Rwd ð0Þ ffi 0 ð18:13Þ

By using Eqs. (18.11), (18.12) and (18.13), Rrr(l)|l=0 is given as


Rrr ðlÞ ¼ Rhh ðlÞ  ½Rdd ðlÞ þ Rpp ðlÞ þ Rww ðlÞl¼0 ð18:14Þ

where Eq. (18.14) represents the received signal power as function of autocorre-
lation. It can also be in written form of K

Rrr ðlÞ ¼ Rhh ðlÞ  ½K þ 1Rpp ðlÞ þ Rww ðlÞl¼0 ð18:15Þ

where

Rdd ðlÞ
K¼ ð18:16Þ
Rpp ðlÞl¼0
0
Therefore total Signal plus Noise Power PSþN can be written as
0 
PSþN ¼ Rhh ðlÞ  ½K þ 1Rpp ðlÞ þ Rww ðlÞl¼0 ð18:17Þ
0
Similarly, the Signal Power PS can be expressed as
176 A.M. Khan et al.

0 
PS ¼ Rhh ðlÞ  ½K þ 1Rpp ðlÞl¼0 ð18:18Þ

To evaluate channel impulse response h(n), the cross correlation of Eq. (18.6) is
determined, with locally generated pilot signal. Therefore Rrp can be written as

Rrp ðlÞ ¼ hðlÞ  ½Rdp ðlÞ þ Rpp ðlÞ þ Rwp ðlÞn¼l ð18:19Þ

where Rrp is the cross correlation between received signal and pilot carriers.
Similarly Rdp and Rwp are cross correlations of data and noise with pilots
respectively. By substitution of Eqs. (18.12) and (18.13) in Eq. (18.19), Rrp(n) is
given by

Rrp ðnÞ ffi hðnÞ  Rpp ðnÞl¼n ð18:20Þ

Equation (18.20) is the convolution of h(n) and Rpp(n), which can be shown in
the form of impulses
hðnÞ ¼ ½h0 dðnÞ þ h1 dðn  1Þ    hT dðn  TÞ; ð18:21Þ

Rpp ðnÞ ¼ ½a0 dðnÞ þ a1 dðn LÞ þ    aN dðn NLÞ ð18:22Þ

where a0…aN are coefficients of Rpp(n). Therefore Rrp(n) can be written as

Rrp ðnÞ ¼ ½a0 h0 dðnÞ þ a0 h1 dðn  1Þ þ    a0 hT dðn  TÞ þ   

a1 h0 dðn LÞ þ a1 h1 dðn L  1Þ þ    a1 hT dðn L  TÞ þ   
aN h0 dðn NLÞ þ aN h1 dðn NL  1Þ þ    aN hT dðn NL  TÞjn¼l
ð18:23Þ

Equation (18.23) shows that the channel coefficients h0, h1…hT are available at
different location on both sides of zero lag (±L), but only zero lag position is used
at (n = 0) due to its high energy. Therefore Eq. (18.23) can be written as

½a0 h0 dð0Þ þ a0 h1 dð1Þ þ    a0 hT dðTÞ
Rrp ð0Þ ¼  ð18:24Þ
a0 n¼0

and it gives

hðnÞ ¼ ½h0 h1 h2 . . . hT1  ð18:25Þ

0
Finally the estimated Signal to Noise Ratio (SNR ) can be written as
0
0 PS
SNR ¼ 0 ð18:26Þ
PN
18 Pilot Based Pre FFT Signal to Noise Ratio Estimation … 177

0
0 PS
SNR ¼ 0 0 ð18:27Þ
PSþN  PS

Rhh ðlÞ  ½K þ 1]Rpp ðlÞ 
SNR0 ¼  ð18:28Þ
Rrr ðlÞ  Rhh ðlÞ  ½K þ 1]Rpp ðlÞl¼0

The Mean Square Error (MSE) is calculated by using following expression

X
I  2 
MSE ¼ SNR0i ðlÞ  SNR  ð18:29Þ
l¼0
i¼1

where (I = 1,000) is the number of iteration used for simulation. SNR0i ðlÞ and SNR
are the estimated and actual SNR respectively.

18.3 Methodology

In proposed algorithm, signal-to-noise ratio (SNR) is estimated in time domain. An

initial step of receiver is to remove the guard band from the receiver input signal
r(n). After the removal of guard band, signal r(n) is provided to autocorrelation
processor, where received signal is autocorrelated Rrr(n). To estimate the Signal
0 0
plus Noise Power PSþN and Signal Power PS , we required the autocorrelation of
channel impulse response Rhh(n). Therefore channel impulse response h(n) is
evaluated by taking cross correlation of r(n) signal with time domain pilot signal
p(n). In final step, time domain SNR0 is estimated by using Signal plus Noise
0 0
Power PSþN and Signal Power PS . OFDM simulation parameters are mentioned in
Table 18.1. It is also assumed that there is perfect synchronization between
transmitter and receiver, so no Inter-Carrier Interference (ICI) and ISI are present
in OFDM symbols. This is justified if the estimation error is sufficiently low at low
SNRs.

18.4 Channel Model

In this work, six multi-paths fading channel model (COST 207 Typical Urban
Reception TU6) for DVB-T applications are utilized in the simulations.
The taps of channel follow the Rayleigh statistics, whose parameters are shown
in Table 18.2. The static case Impulse response of channel can be written as:
178 A.M. Khan et al.

Table 18.1 OFDM simulation parameters

No of FFT points (NFFT) 2048 (2 K-Mode)
Length of cyclic prefix (NG) 512
Total no of subcarriers (NOFDM) 2,560
Total no of symbol (NSYS) 1,000
Pilots spacing (NF) 4
Numbers of pilots (NP) 512
Pilot arrangement Comb type
Pilot and data constellation BPSK
Data per OFDM symbol 1,536
Bandwidth 8 MHz
Data constellation BPSK

Table 18.2 COST 207 typical urban reception (TU6) for DVB-T application [9]
Tap number Average power (dB) Delay(us)
1 -3.0 0
2 0 0.2
3 -2.0 0.5
4 -6.0 1.6
5 -8.0 2.3
6 -10.0 5

" #
0:5dðnÞ þ dðn  3Þ þ 0:63dðn  8Þ
hðnÞ ¼ ð18:30Þ
þ 0:25dðn  25Þ þ 0:15dðn  36Þ þ 0:1dðn  80Þ

18.5 Results and Analysis

Figure 18.3 shows the cross correlation Rdp between pilots and data carriers. Pilots
carriers are inserted into data in such a manner that it produces a value of zero
when (lag = 0). It is also depicted that the value of 1.574e-18 is produced at zero
lag, which is very small as compared to other time instant values and therefore it is
neglected.
Figure 18.4 shows a cross correlation Rwp(n) of noise signal with pilot carriers.
In SNR estimation Rwp(n) plays a very pivotal role because channel MSE is
18 Pilot Based Pre FFT Signal to Noise Ratio Estimation … 179

Fig. 18.3 (Rdp) Cross correlation between data and pilot sub-carriers at 10 dB SNR

Fig. 18.4 (Rwp) Cross Correlation between noise and pilot sub-carriers at 10 dB SNR
180 A.M. Khan et al.

Fig. 18.5 (Rrp) Cross correlation between received signal and pilot sub-carriers

proportional to Rwp(n). It is assumed that Rwp(n) is expected to be relatively small

due the arrangement of pilot carriers, which is also found at zero lag position.
An important feature of the proposed Pre FFT SNR estimation is investigated in
Fig. 18.5 [i.e. Rrp(n)]. It has been shown that Rrp(n) is the convolution of channel
impulse response h(n) and pilots autocorrelation Rpp(n). Equation (18.20) also
reflects this property of Rrp(n). It is found that the channel co-efficient h0, h1…hT
are present at every L number of lags. For illustration purpose only the first
L instants co-efficient values (L = Nfft/4) are shown in Fig. 18.5. Channel co-
efficient is estimated at first L instants because at this location high energy and low
MSE values are present.
In Fig. 18.6 the estimated SNR is plotted verses actual SNR for Rayleigh fading
channel. To improve the estimation in low SNR regime, -8 to 20 dB range is used
for simulation. It is depicted that the proposed SNR estimator has very small
deviation with respect to the actual SNR values. It is also shown that Pre FFT SNR
estimation performs better at low SNR.
In Fig. 18.7, MSE performance is analyzed over Rayleigh fading channel. The
range of SNR used for simulation is [-20 to 20 dB]. Contrary to common sense it
is observed that MSE of estimator is large at high SNR. This can be explained
from the following observation. The SNR estimation is obtained primarily from
signal power estimation. Noise power P0N is estimated from signal power P0S as
shown in Eq. (18.27). So, at large SNR the signal power and total power are nearly
same. Therefore, at large SNR, the noise power, estimated from it, is of the same
order as the signal power estimation error, hence the SNR estimation error
18 Pilot Based Pre FFT Signal to Noise Ratio Estimation … 181

Fig. 18.6 Actual SNR versus estimated SNR for Raleigh fading channel

Fig. 18.7 MSE performance of proposed SNR estimated technique for Raleigh fading channel

increases at large SNR. In order to be statistically accurate; mean is obtained over

1,000 samples. It is shown that proposed estimator performs better with multipath
channel especially at low SNR environment.
182 A.M. Khan et al.

18.6 Conclusion

This paper investigates a novel pilot-based (DA) Pre FFT SNR estimation, where
autocorrelation of received signal is used to estimate the signal plus noise power.
Proposed estimator performs well over frequency-selective Rayleigh fading
channel at low SNR regime. The standardization DVB-T parameters are setup for
OFDM simulation. From the simulation results it is found that a low MSE is
achieved by exploiting proposed estimator. The amount deviation corresponds to
each SNR values is measured in term of MSE and it is observed that estimated
SNR has very less bias and very close to the actual SNR values.

References

1. Ozdemir, M.K., Arslan, H.: Channel estimation for wireless ofdm systems. IEEE Commun.
Surv. Tutor. 9, 18–48 (2007)
2. Wiesel, A., et al.: SNR estimation in time-varying fading channels. IEEE Trans. Commun. 54,
841–848 (2006)
3. Ramchandar Rao, S.R.M., Anil Kumar, P.T., Srinivas Rao, K.: Comparative study of SNR
estimation techniques for Rayleigh and Rician channel. Int. J. Sci. Eng. Technol. Res.
(IJSETR) 2, 2091-2094 (2013)
4. Boumard, S.: Novel noise variance and SNR estimation algorithm for wireless MIMO OFDM
systems. In: Global Telecommunications Conference, GLOBECOM ‘03. IEEE, vol. 3,
pp. 1330–1334, 2003
5. Guangliang, R., et al.: SNR estimation algorithm based on the preamble for OFDM systems in
frequency selective channels. IEEE Trans. Commun. 57, 2230–2234 (2009)
6. Adachi, K.T.a.F.: SNR estimation for pilot-assisted frequency domain MMSE channel
estimation. In: Paper Presented at the IEEE VTS APWCS, Hokkaido, Japan, 2005
7. Gafer, A.Y. et al.: Front-end signal to noise ratio estimation for DVBT fixed reception in flat-
fading channel. In: 2012 4th International Conference on Intelligent and Advanced Systems
(ICIAS), pp. 296–300, 2012
8. Coleri, S., et al.: Channel estimation techniques based on pilot arrangement in OFDM systems.
IEEE Trans. Broadcast. 48, 223–229 (2002)
9. Khan, A.M. et al.: Improved pilot-based LS and MMSE channel estimation using DFT for
DVB-T OFDM systems. In: 2013 IEEE Symposium on Wireless Technology and Applications
(ISWTA), pp. 120–124, 2013
Chapter 19
The Use of Convolutional Code
for Narrowband Interference Suppression
in OFDM-DVBT System

Aizura Abdullah, Muhammad Sobrun Jamil Jamal,

Khaizuran Abdullah, Ahmad Fadzil Ismail and Ani Liza Asnawi

Abstract The problem of mitigating narrowband interference (NBI) due to

coexistence between Digital Video Broadcasting-Terrestrial (DVB-T) and Inter-
national Mobile Telecommunication-Advanced (IMT-A) system is considered. It
is assumed that a spectrum of IMT-A system between 790 and 862 MHz interfere
the spectrum of the OFDM signal in DVB-T band. Two types of convolutional
code (CC) which are non-systematic convolutional code (NSCC) and recursive
systematic convolutional code (RSCC) are proposed to mitigate NBI. The per-
formance of the two techniques is compared under additive white Gaussian noise
(AWGN) channel. It is observed that NSCC has a better bit error rate (BER)
performance than RSCC. The result showed good performance for low SNR
(B5 dB).

Keywords OFDM  Convolutional code  Narrowband interference  DVB-T

A. Abdullah (&)  M.S.J. Jamal  K. Abdullah  A.F. Ismail  A.L. Asnawi

Department of Electrical and Computer Engineering, International Islamic University
(IIUM), Kuala Lumpur, Malaysia
e-mail: [email protected]
M.S.J. Jamal
e-mail: [email protected]
K. Abdullah
e-mail: [email protected]
A.F. Ismail
e-mail: [email protected]
A.L. Asnawi
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 183

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_19
184 A. Abdullah et al.

19.1 Introduction

Orthogonal Frequency Division Multiplexing (OFDM) is a popular multiplexing

scheme used for transmission of high data rates in various communication stan-
dards such as DVB-T, WLANS, WMANs and WiMAX [1, 2]. For certain stan-
dards such as WLANs and WMANs, an OFDM system has the ability to operate in
unlicensed frequency bands. As a result, there is a possibility that they have to
share the same frequency band with other communication systems such as cordless
telephones, garage door openers and baby monitors. This leads to narrowband
interference (NBI) in the systems [3]. Another example of systems sharing the
same frequency band is WiMAX and UWB systems. According to [4], UWB
system is required to modify its spectrum to avoid interference with WiMAX. In
WRC-07 conference, ITU-R has allocated the 790–862 MHz frequency for IMT-A
system. This also suggests that the DVB-T system which operates between 470
and 862 MHz have to share its upper frequency band with the IMT-A system [5].
There are several techniques proposed to mitigate NBI such as using orthogonal
codes, frequency domain cancellation, receiver windowing and excision filtering
[3, 4]. Although orthogonal codes is found to give better performance compared to
error control code (ECC), this method does not comply with the current OFDM
standard such as DVB-T and IEEE [3]. On the other hand, frequency domain
cancellation technique is not suitable to be implemented in broadcasting because
the channel and interference information from the receiver needed to be fed back
to the transmitter periodically for update. A limitation of receiver windowing
method is that it is suitable to be used together with frequency domain cancellation
to reduce the effect of sinc shape side lobes from spreading to adjacent channel
while excision filtering method provides less benefit with quadrature amplitude
modulation (QAM) [4].
ECC is a suitable candidate to mitigate NBI as it is able to protect the data using
a specific code. The data which is corrupted during transmission in the noisy
channel will be recovered by the specific decoding method. Theoretically, ECC
has the capability to lower the bit error rate of an uncoded system by a certain
coding rate [6]. In brief, there are three types of ECC known as block, convolu-
tional and modern codes. In this work, convolutional code (CC) is proposed as it is
suitable to be used in broadcasting, deep space communication, digital speech and
also for Gaussian channel condition [7, 8]. Two types of convolutional code
proposed to mitigate NBI in DVB-T system are non-systematic convolutional code
(NSCC) and recursive systematic convolutional code (RSCC).
Section 19.2 describes the OFDM system and NBI model used. Section 19.3
explains about the proposed ECC techniques. Section 19.4 provides the simulation
results and discussion while Sect. 19.5 concludes this paper.
19 The Use of Convolutional Code … 185

19.2 System Model

The OFDM simulation model of a DVB-T system under narrowband effect is as

shown in Fig. 19.1. It is referred from a MATLAB simulation by [9]. The simu-
lation model is modified by adding ECC as a narrowband mitigation technique and
using different carrier frequency. ECC acts as encoder in the transmitter and
decode the signal back for recovery in the receiver.
At baseband, ECC is applied at the stream of binary data k = {k1 k2 k3 … kn}.
Then, the coded binary data c = {c1 c2 c3 c4 … cn}, is converted into symbols to
be modulated with M number of Quadrature Amplitude Modulation (QAM). Each
serial modulated symbols S = {S1 S2 S3 S4 … Sn}, are mapped into N number of
parallel subcarriers. The modulated symbols X(k), appeared as a complex signal in
frequency domain:

X ðkÞ ¼ RðkÞ þ jI ðkÞ ð19:1Þ

The modulated symbols are passed to inverse fast Fourier transform (IFFT)
processing block to create a time domain OFDM signal for transmission. 2N-IFFT
processing is used to center the subcarriers and processed the discrete signal x(n),

X
N 1
xðnÞ ¼ 1=N XðkÞej2pnk=N ð19:2Þ
k¼0

where n = 0, 1, …, N - 1, k = 0, 1, 2, 3, … N - 1; N being the number of

subcarriers.
An OFDM symbol of N subcarriers is to be transmitted in an OFDM symbol
period duration. The next processing block is to sample the OFDM discrete signal
x(n), within the OFDM symbol period duration. It undergoes filteration process in
digital-to-analog (DAC) converter to obtain continuous time domain signal x(t).
Finally, the signal x(t), is modulated with its RF transmit signal carrier fc ; and
ready for transmission. The receiver system is the reverse process of the trans-
mission system. After demodulation, a decoder recovered the data based on ECC
scheme applied. After the data is recovered, it is compared with the original data
for bit error rate (BER) calculation. From Fig. 19.1, the received signal and the
effect of channel can be written as follows:

r ðtÞ ¼ xðtÞ þ nðtÞ þ iðtÞ ð19:3Þ

where r(t) is the received signal consist of transmitted signal x(t), Gaussian noise
(AWGN) n(t) and narrowband interference i(t).
Figure 19.2 shows the theoretical model of OFDM-DVBT band adopted from
[2] which is used to represent general scenario in this work. For all the channels
that are used for transmission, there are 49 channels in the DVB-T frequency band.
From Eq. (19.3), i(t) has a frequency range fi ; between 790 and 862 MHz
186 A. Abdullah et al.

Fig. 19.1 OFDM simulation model with DVB-T parameters

Fig. 19.2 Theoretical model of OFDM-DVBT band adopted from [2]

interfered with the upper channel in DVB-T band. In this work, the 48th channel in
the DVB-T band which has carrier frequency of 850 MHz is chosen as simulation
parameter with the unwanted NBI signal of frequency 851 MHz.
The narrowband interference (NBI) signal is modeled as sinusoidal signal i(t),

iðtÞ ¼ I cosð2pfi þ hÞ; 0\h\2p ð19:4Þ

where I is the amplitude of the NBI signal and h is the phase angle. The value of h
considered in this work is h ¼ p: Substituting (19.4) into (19.1), the received
signal r(t), is derived as:

r ðtÞ ¼ A cosð2pfc tÞ þ nðtÞ þ I cosð2pfi t þ hÞ ð19:5Þ

where A is the amplitude and fc is the carrier frequency of the OFDM signal. The
performance of an OFDM system is degraded when a strong NBI signal fi , with
carrier frequency close to the OFDM signal’s carrier frequency fc , overlapped,

fi ¼ fc þ Df ð19:6Þ

and that the amplitude of the NBI signal is greater than the amplitude of the
OFDM signal (I [ A). Further details can be found in simulation part 4.
19 The Use of Convolutional Code … 187

19.3 The Proposed Technique

19.3.1 Non-systematic Convolutional Code

Figure 19.3 shows the block diagram of 1/2 rate NSCC encoder which consist of
m number of memory registers. It is used to store previous binary input data. If a
binary data k, enters an encoder, it produces n coded bits at the output with code
rate R = k/n. The code representation is written as (n, k, m). The design of NSCC
can be found in literatures such as [7, 8, 10] with different generator polynomials.
In this work, the generator polynomials used are different compared to the ones
used in example [7, 8, 10] because based on simulation result, it gave BER
performance curve closest to the OFDM system without NBI effect. The generator
polynomials used for 1/2 rate NSCC are g1 = [111] and g2 = [011].
An input bit 1 which entered the encoder will be modulo-2 added with stored
values in memory register to generate the coded bits u1 u2. The generator poly-
nomials determine which stored values in memory register needed to be modulo-2
added with the input bit. Assuming the initial state of memory register is 000, the
output is shown in Table 19.1 below. The input bit is then moved into shift register
m1 with all the bits in the memory register shifted. The oldest stored bit in m3 is
disappeared. The next input bit entered is modulo-2 added with stored values in
memory register which is 100 and the process is repeated.
The system is then extended to 1/3 rate by addition of another generator
polynomial, g3 = [101] as shown in Fig. 19.4. In the case of 1/3 rate encoder, one
bit entered the encoder produced three output bits.
NSCC gave better performance when Viterbi decoder is used [8]. The possible
path that the encoder has undergone is represented in Trellis diagram as shown in
Fig. 19.5. All the memory register’s possible state is written at the first column.
The input bit (in bracket) is written next to the matched output bits referred to
encoder’s truth table. Usually, the process will start at state 000. The branch metric
is calculated by comparing the number of bit agreement with the coded bits. The
process is repeated for all the coded bits. The path which has the highest number of
branch metric is chosen as ‘survivor path’ and the decoded bits are determined.

19.3.2 Recursive Systematic Convolutional Code

RSCC, also known as turbo code is developed from NSCC. Compared to NSCC,
RSCC is formed by concatenating in parallel two RSCCs separated by an inter-
leaver. It is a systematic coding because one of the message bit itself is called
systematic bit and the other two are the parity bits generated by the two RSCC
188 A. Abdullah et al.

Fig. 19.3 1/2 rate convolutional encoder (2, 1, 3)

Table 19.1 Example of truth table for 1/2 rate convolutional encoder (2, 1, 3)
Start state Input End state U1 U2 Output
000 0 000 (0  0  0  0) = 0 (0  0  0) = 0 00
000 1 100 (1  0  0  0) = 1 (1  0  0) = 1 11
100 0 010 (0  1  0  0) = 1 (0  0  0) = 0 10
100 1 110 (1  1  0  0) = 0 (1  0  0) = 1 01

Fig. 19.4 1/3 rate convolutional encoder (3, 1, 3)

Fig. 19.5 Example of Trellis Coded bits 00 01 Branch metric

diagram for 1/2 rate
convolutional code State
000 00(0) 00 (0) 3
001
010 10(0) 0
011
100 11 (1) 11 (1) 3
101
110 01(1) 2
111

encoders. Since the aim of this research is to determine which type of convolu-
tional code performs better in mitigating NBI, the design of RSCC used in this
paper is adopted from [11]. Its coding rate R, is 1/3 with generator polynomial
g1 = [101], g2 = [111], and g3 = [101]. Iterative decoding is used to decode the
message.
19 The Use of Convolutional Code … 189

Welch Power Spectral Density Estimate

Power/frequency (dB/Hz)
-80

-100

-120

-140

-160
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
Frequency (GHz)
Welch Power Spectral Density Estimate
Power/frequency (dB/Hz)

-40

-60

-80

-100

-120
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
Frequency (GHz)

Fig. 19.6 An OFDM signal spectrum appears between 0.4 and 1.25 MHz along the x-axis for
850 MHz carrier frequency, showing OFDM transmitted signal spectrum (top) and OFDM
received signal spectrum with Gaussian noise and the presence of 851 MHz NBI signal (bottom)

19.4 Simulation Results

The performance of convolutional codes in OFDM system under NBI effect is

simulated using MATLAB based on the DVB-T parameters for 2 k mode [9].
Figure 19.6 shows the illustration of the OFDM signal in the presence of Gaussian
effect and unwanted NBI signal.
The BER performance curve of the OFDM system with and without the
presence of NBI is presented in Fig. 19.7. The initial OFDM curve (without
interference) has average signal power value of -10 dB. The value obtained is
calculated based on simulation according to [1]. The NBI signal is added to the
initial OFDM system and simulated in two conditions i.e. with sinusoidal ampli-
tude I = 10 V and I = 20 V. Referring to [6], the NBI signal power Pi ; for the
case of Eq. (19.4) is,
2
Pi ¼ I 2 ð19:7Þ

where I is the amplitude of the narrowband sinusoidal signal.

The NBI signal power for sinusoidal amplitude I = 10 V and I = 20 V are
approximately 17 and 23 dB respectively. For NBI signal power less than 17 dB, it
190 A. Abdullah et al.

0
10

X: 5
Y: 0.1332

-1 X: 5
10 Y: 0.06631
Bit Error Rate

X: 5
Y: 0.03815

-2
10

AWGN 16QAM-OFDM
AWGN 16QAM-OFDM + i(t), I=10
-3 AWGN 16QAM-OFDM + i(t), I=20
10
0 1 2 3 4 5 6 7 8
SNR in dB

Fig. 19.7 BER performance of OFDM system with NBI effects (blue) and initial system without
interference (red). The simulation also shows comparison of the system performance when the
unwanted sinusoidal amplitude is varied (I = 10 V and I = 20 V), having different signal power

has a small effect on the OFDM system. In Fig. 19.7, at SNR = 5 dB, the bit error
rate is 0.06631 for system under NBI signal power of 17 dB and 0.1332 for NBI
signal power of 23 dB compared to 0.03815 for initial system. The difference of
about 0.02816 between system with 17 dB NBI power and initial system is due to
the fact that the system performance is affected by the NBI signal. When the NBI’s
signal amplitude is increased to I = 20, the difference with initial system is
0.09505 which implies further degradation of the system performance due to the
increased in NBI signal power.
Figure 19.8 shows the performance of convolutional coded (CC) OFDM with
the presence of 17 dB NBI signal power. For the case of SNR = 5 dB, the bit error
rate is about 0.04577 for 1/2 rate NSCC which is close to the initial system. As the
SNR increased, the performance of 1/2 rate NSCC did not follow the curve of
initial OFDM system. RSCC obtained BER of 0.09507 at SNR = 5 dB which
showed worst performance compared to the OFDM system with NBI. On the other
hand, 1/3 rate NSCC gave faulty result because it achieved BER of 0.01819 at
SNR = 5 dB which is lower than the BER of initial system.
The convolutional coded OFDM under 23 dB NBI signal power effect is shown
in Fig. 19.9. Based on the figure, 1/3 rate NSCC outperformed 1/2 rate NSCC and
RSCC. At SNR = 5 dB, an error of 0.04401 is obtained by 1/3 rate NSCC which
19 The Use of Convolutional Code … 191

0
10

X: 5
Y: 0.09507
-1
10
Bit Error Rate

X: 5
Y: 0.04577

-2
X: 5
10 Y: 0.01819

AWGN 16QAM-OFDM
AWGN 16QAM-OFDM + i(t), I=10
AWGN 16QAM-OFDM + i(t) + 1/2 rate NSCC
AWGN 16QAM-OFDM + i(t) + 1/3 rate NSCC
-3 AWGN 16QAM-OFDM + i(t)+ RSCC
10
0 1 2 3 4 5 6 7 8
SNR in dB

Fig. 19.8 Performance comparison of convolutional coded OFDM (blue) with the presence of
17 dB NBI signal power with uncoded system (red)

is similar to the initial system. As the SNR increased, the performance of 1/3 rate
NSCC also did not follow the curve of the initial system. 1/2 rate NSCC showed
less performance as it followed the curve of OFDM system under NBI effect
whereas RSCC showed the worst performance with error rate 0.2019 at
SNR = 5 dB.
Based on observations on Figs. 19.8 and 19.9, at low SNRs (B5 dB), the
performance curve of NSCC followed the curves of initial system (without
interference) compared to high SNR ([5 dB). This implies that the performance of
NSCC with Viterbi decoder is different at low and high SNR. According to [7], for
a convolutional code, the error correction and detection capability t, is

t ¼ ðdfree  1Þ=2 ð19:8Þ

where dfree is free distance which is the smallest Hamming distance between all
possible code sequences of the code. At high SNR, the performance is limited by
the capability of Viterbi decoder to correct more than t number of errors in n bits.
Power, bandwidth constraint and nature of noise in the channel can also affect the
performance of the coding scheme. Based on the simulation results, the proposed
192 A. Abdullah et al.

0
10

X: 5
Y: 0.2019

-1
10 X: 5
Y: 0.1467
Bit Error Rate

X: 5
Y: 0.04401

-2
10

AWGN 16QAM-OFDM
AWGN 16QAM-OFDM + i(t), I=20
AWGN 16QAM-OFDM + i(t) + 1/2 NSCC
AWGN 16QAM-OFDM + i(t) + 1/3 NSCC
-3 AWGN 16QAM-OFDM + i(t) + RSCC
10
0 1 2 3 4 5 6 7 8
SNR in dB

Fig. 19.9 Performance comparison of convolutional coded OFDM (blue) with the presence of
23 dB NBI signal power with uncoded system (red)

NSCC has good performance for low SNR (B5 dB) to mitigate NBI compared to
RSCC. The performance of this code also showed considerable result at low SNR
when compared with time windowing method for NBI mitigation under DVB-T
system [9].

19.5 Conclusion

In this paper, a narrowband mitigation technique is proposed using convolutional

code. NSCC and RSCC are presented as two different ECC techniques suitable for
NBI mitigation for DVB-T transmission under Gaussian channel. The interference
is assumed coming from IMT-A services affected the upper channel of the DVB-T
band. The simulation showed that 1/2 rate NSCC can mitigate the 17 dB NBI
signal power while 1/3 rate NSCC suited for 23 dB NBI signal power at low SNR
(B5 dB). The performance result for RSCC showed that it is not effective in
mitigating the NBI for this work.

Acknowledgments The authors would like to thank the Malaysian government, Ministry of
Higher Education (MOHE) for sponsoring this work under FRGS grant no. FRGS13-026-0267.
19 The Use of Convolutional Code … 193

References

1. Cho, Y.S., Kim, J., Yang, W.Y., Kang, C.-G.: MIMO-OFDM wireless communications with
MATLAB. Wiley, Singapore (2010)
2. E. B. Union: Etsi en 301 701 (2000)
3. Coulson, A.J.: Bit error rate performance of OFDM in narrowband interference with excision
filtering. IEEE Trans. Wirel. Commun. 5(9), 2484–2492 (2006)
4. Batra, A., Zeidler, J.R.: Narrowband interference mitigation in OFDM systems. In: Military
Communications Conference MILCOM 2008, pp. 1–7 (2008)
5. Shamsan, Z.A., Rahman, T.A., Al-hetar, A.M.: Point-point fixed wireless and broadcasting
services coexistence with IMT-advanced system. Prog. Electromagn. Res. 122, 537–555
(2012)
6. Lathi, B.P.: Modern digital and analog communication systems. Oxford University Press,
New York (1998)
7. Jiang, Y.: A practical guide to error control coding using MATLAB. Artech House, Boston
(2010)
8. Sweeney, P.: Error control coding from theory to practice. Wiley, England (2004)
9. Abdullah, K.: Interference mitigation techniques for wireless OFDM. Ph.D. dissertation,
RMIT, Melbourne (2009)
10. Arshad, N., Basit, A.: Implementation and analysis of convolutional codes. Int.
J. Multidiscip. Sci. Eng. 3(8), 9–12 (2012)
11. Arshad, N., Jamal, M.A.: Implementation and analysis of turbo codes using MATLAB.
J. Expert Syst. (JES) 2(1), 115–118 (2013)
Chapter 20
Two-Elements Crescent Shaped Printed
Antenna for Wireless Applications

Wan Noor Najwa Wan Marzudi, Zuhairiah Zainal Abidin,

Ma Yue and Raed A. Abd-Alhameed

Abstract This study presents an investigation of the mutual coupling between two
printed elements antenna for a multiple-input-multiple-output (MIMO) antenna
performance. It consists of two crescent shaped radiators placed symmetrically,
and a neutralization line is applied to improve the mutual coupling. Theoretical
and experimental characteristics are presented and compared. The antenna yields
an achieved impedance bandwidth of 18.67 % (over 2.04–2.46 GHz) with a
reflection coefficient \-10 dB and mutual coupling minimization of \-20 dB in
addition to a reasonable and stable radiation pattern and envelope correlation.

Keywords Multiple-input-multiple-output (MIMO)  Neutralization line 

Mutual coupling

W.N.N.W. Marzudi (&)  Z.Z. Abidin

Research Center of Applied Electromagnetic, Universiti Tun Hussein Onn Malaysia, Johor,
Malaysia
e-mail: [email protected]
Z.Z. Abidin
e-mail: [email protected]
M. Yue
National Astronomical Observatories (NAOC), Chinese Academy of Sciences,
Beijing, China
R.A. Abd-Alhameed
Mobile and Satellite Communication Research Centre, University of Bradford,
Bradford, UK

 Springer International Publishing Switzerland 2015 195

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_20
196 W.N.N.W. Marzudi et al.

20.1 Introduction

The potential for MIMO antenna systems to improve reliability and enhance
channel capacity in wireless mobile communications has generated great interest
[1]. A major consideration in MIMO antenna design is to reduce correlation
between the multiple elements, and in particular the mutual-coupling electro-
magnetic interactions that exist between multiple elements are significant, because
at the receiver end this effect could largely determine the performance of the
system. Lower mutual coupling can result in higher antenna efficiencies and lower
correlation coefficients [2]. The effect of mutual coupling on antenna diversity
performance of the MIMO antenna array has been reported in [3–10]. In [3], a
shorting strip and isolation stub was used to reduce mutual coupling for portable
wireless devices with isolation values lower than 225 dB at 2.45 GHz. Authors in
[4] introduced I-shaped conductor in a modified ground plane that reached
214 dB of mutual coupling across 1.6–2.8 GHz. Inserting stubs is one of the
method used to enhanced the isolation and minimize mutual coupling between
elements [5]. In addition, by inserting neutralization line technique between
antenna elements is was also promising method [6, 7]. Low mutual coupling can
also be achieved through defected ground structure [8]. Other methods to reduce
mutual coupling and enhanced isolation of the MIMO antenna, such as protrude
branch and T-slot etched on ground plane [9] and inserting slits into ground plane
[10].
In this paper, two-element crescent shaped MIMO antenna presented for the
purpose of wireless applications. The MIMO antenna consists of two crescent
shaped radiators placed symmetrically with respect to ground plane with neu-
tralization line (NL) connected in between of the two antennas. The total
dimensions of this antenna are 100 9 45 9 1.6 mm3. Both simulated and mea-
sured result of the fabricated prototype details reported and discussed.

20.2 Antenna Design Concept

The proposed antenna geometry is illustrated in Fig. 20.1. The antenna system
comprises with two crescent shaped radiators that is similar to that in [11]
deployed on an economically FR-4 substrate with relative permittivity of 4.4 and a
thickness of 1.6 mm operating at 2.4 GHz. The radiators are separated by 0.147 k
(18.36 mm) for the minimization of mutual coupling. While, 83 9 45 mm2
ground plane placed on the other side of the substrate as shown in Fig. 20.1b. The
overall dimensions of the proposed antenna are 100 9 45 9 1.6 mm3 which is
suitable for wireless application such as a network card or mobile device.
20 Two-Elements Crescent Shaped Printed Antenna … 197

Fig. 20.1 Geometry of the proposed antenna (in mm) a Top view, and b Bottom view

To minimize the mutual coupling and enhanced the isolation of the antenna,
neutralization line was inserted in between of the radiating element. In general, the
neutralization line transfer the signal from the first antenna element and the second
antenna element receive the signal in order to cancel out existing coupling between
two elements.

20.3 Parametric Study

To clarify the effectiveness of the neutralization line (NL) of the proposed antenna,
the parametric study of the location of the neutralization line, NLd was carried out
with the width of the NL is kept at 0.5 mm. From Fig. 20.2, it can be observed that
the optimal distance of NLd is at 9.79 mm which gives the lowest mutual coupling
at 2.4 GHz.
198 W.N.N.W. Marzudi et al.

Fig. 20.2 Simulated transmission coefficient, S21 of the various distance location of neutraliza-
tion line

20.4 Simulated and Measured Performance

Figures 20.3 and 20.4 show the simulated and measured S-parameters output for
the proposed antenna with and without neutralization line, respectively. As can
been observed, the measured return loss, |S11| and mutual coupling, |S21| for both
figures (Figs. 20.3 and 20.4) are reasonably good agreement with the simulated
results. The resonance frequency is slightly shifted between the simulated and
measured results and this is probably due to the discrepancy of SMA connector
and fabrication tolerance. It is apparently seen that by implementing the neutral-
ization lines, the mutual coupling, |S21| of the proposed antenna can be improved.
To validate the simulated results, the physical prototypes of the proposed antenna
with and without the neutralization line were fabricated and tested, as shown in
Fig. 20.5. The S-parameters of the antenna were measured by Vector Network
Analyser 8722ET (VNA). The measured return loss |S11| and mutual coupling |S21|
are plotted in Fig. 20.6. As can be seen, when the neutralization line was inserted, the
mutual coupling has been reduced around 7.14 dB (from 214.63 to 221.77 dB) with
an impedance bandwidth of 18.67 % (over 2.04–2.46 GHz). The bandwidth
achieved fully covered the wireless application such as network card at 2.4 GHz.
To evaluate the capabilities of MIMO/diversity antenna, the envelope corre-
lation coefficient (ECC) is an important criterion to be presented. Basically,
envelope correlation can be computed by using S-parameters or radiation pattern
of the antenna. The envelope correlation of the MIMO antenna system can be
expressed by using the following expression [12]:
  
S S12 þ S S22 2
11 21
qe ¼    ð20:1Þ
1jS11 j2 jS21 j2 1jS22 j2 jS12 j2
20 Two-Elements Crescent Shaped Printed Antenna … 199

Fig. 20.3 Comparative plot of S-parameters output for simulated and measured results for the
proposed antenna without neutralization line

Fig. 20.4 Comparative plot of S-parameters output for simulated and measured results for the
proposed antenna with neutralization line

The simulated envelope correlation coefficient of the proposed antenna with

and without neutralization showed in Fig. 20.7. An improvement of the ECC can
be seen after the neutralization line was inserted and it fulfils the characteristic of
diversity pe \ 0.5 [13]. Therefore, the proposed antenna is suitable candidate for
MIMO application.
The simulated and measured radiation patterns of the proposed antenna in the
X–Z plane (E-plane) and Y–Z plane (H-plane) with port 1 excited while port 2
terminated with 50 X load plotted in Figs. 20.8 and 20.9, respectively. The
antenna shows a stable omnidirectional pattern in the E-plane and H-plane over the
operating frequency of 2.4 GHz.
200 W.N.N.W. Marzudi et al.

Fig. 20.5 Practical prototype of the proposed antenna a with neutralization line b without the
neutralization line

Fig. 20.6 Measured S-parameters of the proposed antenna with and without the neutralization
line
20 Two-Elements Crescent Shaped Printed Antenna … 201

Fig. 20.7 Simulated envelope correlation coefficient for the proposed antenna with and without
neutralization line

Fig. 20.8 Simulated radiation patterns for the proposed antenna for two planes at 2.4 GHz; a X–
Z plane. ‘‘xxxx’’ (blue) simulated co-polarization, ‘‘oooo’’ (red) simulated cross-polarization.
b Y–Z plane. ‘‘xxxx’’ (blue) simulated cross-polarization, ‘‘oooo’’ (red) simulated co-polarization
port 1 (left) excited and port 2 (right) terminated in 50 X (colour figure online)
202 W.N.N.W. Marzudi et al.

Fig. 20.9 Measured radiation patterns for the proposed antenna for two planes at 2.4 GHz; a X–
Z plane. ‘‘xxxx’’ (blue) simulated co-polarization, ‘‘oooo’’ (red) simulated cross-polarization.
b Y–Z plane. ‘‘xxxx’’ (blue) simulated cross-polarization, ‘‘oooo’’ (red) simulated co-polarization
port 1 (left) excited and port 2 (right) terminated in 50 X (colour figure online)

20.5 Conclusion

A two elements crescent shaped printed MIMO antenna for covering 2.4 GHz
wireless applications is presented. Neutralization line is applied to meet the
requirement of MIMO in term of low mutual coupling parameter. Simulated and
measured results show that the antenna achieves an impedance bandwidth of
18.67 % (over 2.04–2.46 GHz) with a reflection coefficient \-10 dB and mutual
coupling minimization of \-20 dB which is suitable for wireless applications.

Acknowledgement The authors of this paper wish to acknowledge the funding of this project by
Universiti Tun Hussein Onn Malaysia under short term grant Vot 0992.

References

1. Foschini, G.J., Gans, M.J.: On limits of wireless communications in a fading environment

when using multiple antennas. Wireless Pers. Commun. 6, 311–335 (1998)
2. Thaysen, J., Jakobsen, K.B.: Design considerations for low antenna correlation and mutual
coupling reduction in multi antenna terminals. Eur. Trans. Telecommun. 18, 319–326 (2007)
3. Ling, X., Li, R.: A novel dual-band MIMO antenna array with low mutual coupling for
portable wireless devices. IEEE Antennas Wirel. Propag. Lett. 10, 1039–1042 (2011)
20 Two-Elements Crescent Shaped Printed Antenna … 203

4. See, C., Abd-Alhameed, R., McEwan, N., Jones, S., Asif, R., Excell, P.: Design of a printed
MIMO/diversity monopole antenna for future generation handheld devices. Int. J. RF
Microw. Comput. Aided Eng. 24, 348–359 (2013)
5. Liu, L., Cheung, S., Yuk, T., Wu, D.: A compact ultrawideband MIMO antenna. In: 2013 7th
European Conference on Antennas and Propagation (EuCAP), pp. 2108–2111 (2013)
6. See, C.H., Abd-Alhameed, R.A., Abidin, Z.Z., McEwan, N.J., Excell, P.S.: Wideband printed
MIMO/diversity monopole antenna for WiFi/WiMAX applications. IEEE Trans. Antennas
Propag. 60, 2028–2035 (2012)
7. Wang, Y., Du, Z.: A wideband printed dual-antenna system with a novel neutralization line
for mobile terminals. IEEE Antennas Wirel. Propag. Lett. 12, 1428–1431 (2013)
8. Prasanna, K.M., Behera, S.: A hexagonal MIMO antenna system with defected ground
structure to enhance bandwidth and isolation (2013)
9. Xia, X.-X., Chu, Q.-X., Li, J.-F.: Design of a compact wideband MIMO antenna for mobile
terminals. Prog. Electromagn. Res. C 41, 163–174 (2013)
10. Wu, Y.-T., Chu, Q.-X., Yao, S.-J.: A dual-band printed slot diversity antenna for wireless
communication terminals. In: 2013 IEEE International Wireless Symposium (IWS), pp. 1–3
(2013)
11. See, C.H., Abd-Alhameed, R.A., Zhou, D., Lee, T.H., Excell, P.S.: A crescent-shaped
multiband planar monopole antenna for mobile wireless applications. IEEE Antennas Wirel.
Propag. Lett. 9, 152–155 (2010)
12. Xiong, L., Gao, P.: Compact dual-band printed diversity antenna for Wimax/WLAN
applications. Prog. Electromagn. Res. C 32, 151–165 (2012)
13. Li, J.-F., Chu, Q.-X., Huang, T.-G.: A compact wideband MIMO antenna with two novel bent
slits. IEEE Trans. Antennas Propag. 60, 482–489 (2012)
Chapter 21
Wideband Linearly Polarized Printed
Monopole Antenna for C-Band

Touhidul Alam, Mohammad Rashed Iqbal Faruque

and Mohammad Tariqul Islam

Abstract This paper presents a printed wideband elliptical patch antenna on FR-4
substrate for C-band applications. High-frequency structural simulator (HFSS)
based on the finite element method (FEM) and Computer simulation technology
(CST) based on the finite difference time domain (FDTD) softwares have been
used in this research and a wide bandwidth of 4.34 GHz (3.38 to 7.72 GHz) was
achieved. The parametric study and equivalent impedance matching circuit of the
proposed antenna has been investigated.

21.1 Introduction

The elliptical patch antenna has been playing an important role in communication
system due to its some special characteristics like ultra wideband, easy fabrication
and compact in size. However, the conventional antenna exhibits narrow bandwidth
[1–6], but the present communication system demands compact antenna with wider
bandwidth. To meet these demands, designers are looking for methods to increase
the bandwidth of the antenna. For example, in [7], Behdad et al. proposed a new
method to increase wideband by using slot. And Deshmukh et al. used proximity-fed
in rectangular microstrip antenna for achieving wider bandwidth in [8]. Moreover, to
improve bandwidth Rafi et al. proposed a V-shaped slot antenna in [9]. Parasitic
elements have been used to increase the bandwidth in [10, 11].

T. Alam (&)
M.R.I. Faruque

Space Science Center (ANGKASA), Research Centre Building, Universiti Kebangsan
Malaysia, 43600UKM Bangi, Selangor, Malaysia
e-mail: [email protected]
M.T. Islam
Department of Electrical Electronic and System Engineering, Universiti Kebangsan
Malaysia, 43600UKM Bangi, Selangor, Malaysia

 Springer International Publishing Switzerland 2015 205

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_21
206 T. Alam et al.

Luis et al. proposed elliptical antenna array for wider bandwidth in [12], but the
antenna size was 98 9 28 mm for operating frequency band of 5.3 to 6.4 GHz.
In this paper, we presented a wideband elliptical patch antenna for C-band
applications. We achieved a wide bandwidth of 3.43 GHz, from 3.38 to 7.72 GHz,
with the presented antenna size of 30 9 34 9 1.6 mm3. Parametric analysis has
been performed and analysed LC matching circuit of the presented antenna.
This paper is presented as follows: In Sect. 21.2, detail in geometry of the
presented antenna, in Sect. 21.3, parametric study of the presented antenna will be
discussed. In Sect. 21.4, LC matching circuit of the antenna and in Sect. 21.5,
result and discussions will be presented. Finally in Sect. 21.6, conclusion will be
presented.

21.2 Antenna Design

The design layout of the proposed antenna is given in Fig. 21.1. The proposed
antenna has been printed on 1.6 mm thick FR-4 substrate materials with material
the properties of relative permittivity of 4.6, relative permeability of 1 and
dielectric loss tangent of 0.02. The dimensions of the antenna are shown in
Table 21.1.

21.3 Parametric Studies

The effects of the proposed antenna parameters have been investigated. The
parametric study on surface current distribution and reflection coefficient has been
performed.

21.3.1 For Different Substrate Materials

The effect of substrate material properties of reflection coefficient has been

observed, which is shown in Fig. 21.2. The material properties of different types of
substrate material are shown in the Table 21.2.

21.3.2 Substrate Thickness

The optimum value of substrate thickness for the desired frequency band has been
obtained at 1.6 mm, which is shown in Fig. 21.3.
21 Wideband Linearly Polarized Printed … 207

Fig. 21.1 Design layout of

the presented antenna

Table 21.1 Antenna design Parameters Values (mm)

specifications
Substrate length (L) 34
Substrate height, H 1.6
Substrate width (W) 30
Patch radius R1 8
Patch radius R2 11
Feed gap (Wg) 0383
Feed line length (Lf) 16.383
Feed line width (Wf) 1.885
Ground plane length (Lg) 16
Rectangular slot length (L1) 7
Rectangular slot width (W1) 1
Circular slot width (W2) 0.6

21.3.3 The Effect of Slotting in the Patch

The effect of slotting in the elliptical patch has been investigated. The surface
current of the antenna for three different frequencies without slotting and with
circular slotting are shown in Figs. 21.4 and 21.5, respectively.
208 T. Alam et al.

Fig. 21.2 Reflection

coefficient for different types
of substrate materials

Table 21.2 Material properties of different substrates

Substrate name Permitivity Loss tangent Substrate thickness
FR4 4.6 0.02 1.6
RT 5870 2.33 0.0012 1.6
RT 6010 10.2 0.002 1.6

Fig. 21.3 Reflection

coefficient for different
thickness of FR-4 substrate
material
21 Wideband Linearly Polarized Printed … 209

Fig. 21.4 Surface current distribution without slotting—a at 4 GHz, b at 5 GHz, and c at 7 GHz

Fig. 21.5 Surface current distribution with circular slotting—a at 4 GHz, b at 5 GHz, c at
7 GHz

Fig. 21.6 LC matching

circuit of the proposed
antenna
210 T. Alam et al.

Fig. 21.7 Reflection

coefficient of the presented
antenna

21.4 LC Matching Circuit

Input impedance matching is an important factor in antenna desigining. If the

antenna is not properly matched, then the input impedance will fluctuate with the
transmission line length. That’s why sufficient power will not deliveried to the
antenna. LC matching of the presented antenna is shown in Fig. 21.6. The value of
the capacitor and inducor are 0.229 pF and 0.00184 uH, respectively.

21.5 Results and Discussions

The reflection coefficient of the presented antenna has been investigated by using
CST microwave studio and Ansoft HFSS simulation softwares and compared two
results, which is presented in Fig. 21.7. Moreover, the farfield directivity of the
antenna has been analyzed at 4 GHz, 5 GHz, 6 GHz and 7 GHz, with is presented
in Fig. 21.8.

21.6 Conclusions

A wideband circularly polarized elliptical patch antenna for C-band application

has been realised. A bandwidth of the presented antenna is 4.34 GHz, which is
78.19 % of fractional bandwidth with respect to centre frequency of 5.55 GHz.
21 Wideband Linearly Polarized Printed … 211

Fig. 21.8 Farfield directivity of the presented antenna—a at 4 GHz, b at 5 GHz, c at 6 GHz and
d at 7 GHz

The input impedance matching of the antenna has been analysed and found
approximately 50X of input impedance. So, the proposed antenna can be a suitable
one to implement in modern communication system for C-band applications.

References

1. James, J.R., Hall, P.S.: In: Handbook of Microstrip Antennas, pp. 825–849. Peregrinus,
London (1989)
2. Islam, M.M., Islam, M.T., Faruque, M.R.I.: Dual-band operation of a microstrip patch
antenna on a duroid 5870 substrate for ku- and k-bands. Sci. World J. 2013, 1–10 (2013)
(Article ID 378420)
3. Faruque, M.R.I., Islam, M.T.: Novel design of triangular metamaterial for electromagnetic
absorption in human head. Prog. Electromagnet. Res. PIER 141, 463–478 (2013)
212 T. Alam et al.

4. Faruque, M.R.I., Islam, M.T., Misran, N.: Design analysis of new metamaterial for EM
absorption reduction. Prog. Electromagnet. Res. PIER 124, 119–135 (2012)
5. Islam, M.T., Faruque, M.R.I., Misran, N.: Specific absorption rate analysis using metal
attachment. Informacije MIDEM 40(3), 238–240 (2010)
6. Faruque, M.R.I., Islam, M.T., Misran, N.: Evaluation of specific absorption rate (SAR)
reduction for PIFA antenna using metamaterials. Frequenz J. 64(7/8), 144–149 (2010)
7. Behdad, N., Sarahandi, K.: A multi-resonant single element wideband slot antenna. IEEE
Antennas Wirel. Propag. Lett. 3(1), 5–8 (2004)
8. Deshmukh, A.A., Ray, K.P.: Broadband proximity-fed modified rectangular microstrip
antennas. IEEE Antennas and Propag. Mag. 53(5), 41–56 (2011)
9. Rafi, Gh.Z., Shafai, L.: Wideband V-slotted diamond-shaped microstrip patch antenna.
Electron. Lett. 40(19) 1166–1167 (2004)
10. Wong, K.L.: Compact and Broadband Microstrip Antennas. Wiley, New York (2002)
11. Priyashman, V., Jamlos, M.F., Lago, H., Jusoh, M., Ahmad, Z.A., Romli, M.A., Salimi,
M.N.: Effects of parasitic ring on the performance of an elliptical shaped antenna. In: IEEE
Symposium on Wireless Technology and Applications (ISWTA), 2012
12. Brás, L., Carvalho, N.B., Pinho, P.: Circular polarized planar elliptical antenna array. In:
Antennas and Propagation (EuCAP), 7th European Conference, pp 891–893 2013
Chapter 22
A Novel Anti-collision Protocol
for Optimization of Remote Sensing
in Dense Reader Network

Faiza Nawaz and Varun Jeoti

Abstract Passive Radio Frequency Identification (RFID) network with several

reader placed densely and close to each other are susceptible to reader collision
problem. In this paper, a novel and efficient RFID reader’s anti-collision protocol
is proposed based on Neighbor Friendly Reader Anti-collision (NFRA) mechanism
by revising its contention procedure to provide higher throughput in dense reader
network. The behavior of the algorithm is evaluated through a set of simulation
experiments which demonstrates that the algorithm is 15 % efficient than NFRA. It
also has higher fairness as compared to NFRA and other state-of-the-art proposals.

Keywords Radio frequency identification  Anti-collision protocol  NFRA 

Dense reader environment

22.1 Introduction

The Radio Frequency Identification (RFID) [1] part of the Automatic Identification
and Data Capture (AIDC) group has revolutionized many applications including
the industrial environment monitoring. In such time critical environments, reliable
and energy efficient monitoring is often required. The RFID tagging and sensing
technology is becoming popular in a variety of fields, such as health care [2]
warehouse inventories [3], object tracking [4] sports [5], food traceability [6] chain
management [7] etc. An RFID network consists of RFID tags/sensor, readers, air
interface, and backend servers. The reader communicates with the tags by means

F. Nawaz (&)  V. Jeoti

Department of Electrical and Electronic Engineering, Universiti Teknologi Petronas,
31750 Tronoh, Perak, Malaysia
e-mail: [email protected]
V. Jeoti
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 213

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_22
214 F. Nawaz and V. Jeoti

of radio signals, in a finite space, known as the interrogation zone. A passive tags
have no battery, and they acquire the necessary power from the electromagnetic
field of the reader. The basic idea of functioning is backscatter. These tags are cost
effective and durable than active battery operated tags, but have a shorter inter-
rogation range and limited processing capabilities. A Dense Reader Environments
(DRE) consists of several interrogators closely placed in an area to be monitored
using passive tags/sensors. In DRE scenarios, readers collect tags information from
time to time by using a reader-to-tag identification protocol like ETSI EN 302 208,
ISO/IEC 18000 and EPCglobal Class-1 Gen-2 [8]. EPCglobal is the most widely
accepted standard for RFID network in which the information collected by the
reader is sent to a central server (CS) by mean of a wired or wireless link using
Low Level Reader Protocol (LLRP), whereas for reader to tag communication
slotted ALOHA is used. However EPCglobal is not designed to deal explicitly
with reader to reader collision problem that is a major issue in DRE which negates
the network performance.
The collisions in DRE affect the throughput and efficiency of the system. Three
types of collisions can happen in a DRE. Tag to tag collision (TTC) occurs when
multiple tags respond simultaneously to same reader. Reader to Tag collision
(RTC) occurs when two or more readers try to read the same tag because of an
overlap in their read ranges. Reader-to-Reader Collision (RRC) happens when the
signal generated by one reader interferes with the reception system of other readers
[9]. Figure 22.1 shows a RRC scenario; reader 1 (R1) attempts to read data from
tag 1 (T1) and reader 2 (R2) is trying to read data from tags tag 2 (T2) using the
same channel. The weak response signal of T1 is degraded because of the strong
interference signal of R2.
Numerous approaches have been proposed in the literature to minimize RFID
network collisions [9]; however the solutions to this problem are limited. This
research work proposes a new anti-collision algorithm for DRE and its operational
performance is illustrated using a set of simulation iterations.
The rest of the paper is organized as follows: in Sect. 22.2, related work in the
field of DRE anti-collision is discussed. The approaches are classified as distrib-
uted and centralized. This section also gives a detailed overview of Neighbor

Fig. 22.1 Reader to reader

interference
22 A Novel Anti-collision Protocol for Optimization of Remote Sensing … 215

Friendly Reader Anti-collision (NFRA) approach. In Sect. 22.3, the proposed

algorithm is described. Section 22.4 evaluates the results of proposed algorithm
under different scenarios. Section 22.5 presents a conclusion of this work.

22.2 Related Work for DRE Anti-collision Management

Implementation of reader to reader anti-collision mechanisms is very important in

DRE. Two broad classifications of RRC algorithm are centralized algorithms and
distributed algorithms. In distributed schemes every reader communicates directly
with its neighbors and do not rely on a centralized device for network resources
allocation. The most common distributed technique for RRC, based on carrier
sense multiple access (CSMA) or listen before talk (LBT) is Pulse [10]. In Pulse
random back off time is introduced, to avoid simultaneous transmission by more
than one reader. Pulse protocol has less overhead but when compared with Time
division Multiple Access (TDMA) based approaches it lacks in efficiency and
throughput. An extension of Pulse protocol is [11] in which Slot Occupied
Probability (SOP) is used for reducing the number of collisions. The algorithm
effectively mitigates reader collisions in dense reader mode. Distributed tag access
with Collision avoidance (DiCa) is another distributed algorithm based on listen
before talk and focusing both on RTC and RRC reduction [12]. It is an energy
saving system, however readers’ energy consumption has a minor impact on
operational cost, and high performance and low complexity are more desirable
features in RFID reader networks. Reader Anti-Collision Algorithm for Multi-
channel Mobile RFID Networks (RAC-Multi) [13] has separate data and control
channels for communication. Adjacent channel interference is avoided by intro-
ducing even and odd numbered data channels. Distributed Color Scheme (DCS)
protocol uses a single frequency for all the readers, with same frame sizes. Time is
divided in predetermined identification cycles, which are subdivided into time-
slots named as colors [14]. Probabilistic DCS (PDCS) [15] is the extension of this
work intended for increasing the low performance of DCS. In PDCS a probability
is considered for choosing new colors for collided readers. It reduces the number
of collisions and also the number of readers that change color with time. Color-
wave [14] is proposed with the aim of improving the low performance of DCS. In
this approach each identification cycle has a variable number of colors and the
number of colors per cycle increases when RRC are extremely high. Colorwave
performs better than ALOHA algorithms in less dense networks and slightly worse
in highly loaded networks. Expowave [16] is also a novel approach that outper-
forms Colorwave and DCS.
In centralized RRC mechanism, server is the acting agent for all communica-
tions. All readers are connected to a server that stores both the communications
and the readers information. Among the most notable, and high performance
centralized RRC avoidance algorithms is Neighbor Friendly Reader Anti-collision
(NFRA) [17]. An extension of NFRA with emphasis on high fairness and less RRC
216 F. Nawaz and V. Jeoti

is proposed in [18], but the protocol is not compatible with EPCglobal Class-1
Gen-2, and its implementation in real world requires the use of extra wireless
network at 433 MHz. Hierarchical Qlearning algorithm (HiQ) is intended for
finding dynamic solutions to the reader collision problem by mapping collision
patterns among readers [19]. The main shortcoming of this approach is that readers
have to manage a huge amount of data, also the final result depends on the quality
of the neural network training. Resource allocation based on genetic algorithm
(RA-GA) [20] is also a FDMA-TDMA technique based on a HiQ algorithm. It
uses the SNR constraint of each reader to appropriately assign spectral and tem-
poral resources. However, there is no description given about how it satisfies the
requirements of standards and regulations. Distributed adaptive power control
(DAPC) [21] is a novel solution which used a back off algorithm to improve
coverage. Distributed color no cooperative selection (DCNS) is a high throughput
solutions for static RFID networks [22]. It uses the killer configuration and
dynamic priority management for improving the performance of RFID readers as
compare to other state of the art reader to reader anti-collision protocols. Dis-
tributed color no cooperative selection (DCNS) protocol is based on Colorwave
[23], with additive killer configuration. It reduces the unused time slots to increase
throughput of the network. It does not require any prior deployment knowledge,
and it is appropriate for low cost RFID readers. It has reduced channel control
overhead compare to Colorware by employing a new color update mechanism.
DCNS provides 16 percent higher throughput than NFRA. Geometric distribution
reader anti-collision (GDRA) [24] is a new centralized scheduler which exploits
the Sift geometric probability distribution function to minimize reader collision
problems.
The results presented in literature shows that centralized algorithms are more
efficient in term of throughput as compared to distributed techniques. The work
proposed in this paper is based on NFRA [17] in which a polling server is des-
ignated to divides the time into identification rounds. Every round begins by an
arrangement command AC (random numbers from 1 to maximum number, MN)
broadcasted to all the readers. The readers that receive the AC, generate its own
random number. The server then issues an ordering command (OC); the readers
then compare their random numbers with the value in the OC. If both values are
same, the readers broadcasts their beacons signals to determine whether a collision
occurs or not. If a readers does not detect any collisions, it send overriding frame
(OF) to the neighboring readers. The OF prevents the neighboring readers from
receiving the next OC from the server. The neighboring readers which do not
identify the next OC due to the OF or which detect a collision of beacons do not
actively operate. Communications between a Reader and Tags (CRT) is performed
by successful readers only. The throughput of NFRA is affected by the random
selection of MN. In many situations a collision detected at the beginning of the
round could no longer be valid at the end. Next section describes the details of
proposed algorithm based on modified NFRA contention phase.
22 A Novel Anti-collision Protocol for Optimization of Remote Sensing … 217

22.3 Proposed NFRA-C Algorithm

This Section illustrates the proposed RRC avoidance algorithm. This anti-collision
algorithm extends NFRA, using counters, in dense reader networks. Each reader
maintains the history of his successful communication with tags in the form of
counters. The counter is incremented one time for each successful communication.
Counters are exchanged along with beacons to detect the presence of other readers
within the range. Algorithm 1 describes the pseudo code of the proposed mech-
anism. Whenever a collision of beacons is detected, counter of both colliding
readers are compared and the reader with a lower value of counter is permitted to
participate in that particular round.
Consider the scenario shown in Fig. 22.2. After a collision is detected between
reader 1 and 2, counters are compared and the reader with lower value of counter
participates in that round, unlike NFRA which does not permit any reader to
communicate with tags until a new AC is received. Each identification round
begins by the server broadcasting AC to all the readers. Upon receiving the AC
readers generate a random number between 1-MN, and wait for OC from server.
As soon as the OC is received each reader compares the OC with its random
number and if the generated random number is equal to OC number reader
broadcasts beacons to its neighbors confirming its intension to participate in that
round. If a single collision of beacon is detected at a reader then counters are
compared. If the value of counter broadcasted by the reader is less than the
received value; the reader broadcasts an OF to its neighbors and conduct

Fig. 22.2 NFRA network scenario

218 F. Nawaz and V. Jeoti

identification of tags during CRT time. Else if multiple collisions at a reader are
detected than that reader has to wait until next AC from the server is received.

Algorithm 1 NFRA-C anti-collision algorithm

1: Loop
2: While a signal is not received from the server do
3: No operation
4: End while
5: If a reader receives signal = AC then
6: Generate a random number r among [1,MN]
7: Wait for OC from the server
8: Decode every received OC to extract number j
9: If ( j = r ) then
10: Broadcast a beacon to neighbor readers
11: If a single beacon collision is detected then
12 Compare counter
13: If send counter <received counter
14: Broadcast OF to neighbor readers
15: Conduct identification during CRT
16: Else if send counter > received counter send counter=received counter
17: Wait until next AC signal from server
18: End if
19: Else if multiple collisions of beacons is detected then
20: Wait until next AC from server
21: End if
22: Else
23: Broadcast OF to neighbor readers
24: Conduct identification during CRT
25: End
26: Else If OF is received
27: Wait until the next AC from server
28: Else
29: Wait until the next OC from server is received
30: End
31: End
32: End Loop_______________________________________________________

The proposed algorithm is evaluated using a number of simulation scenarios.

The first measurable parameter investigated in the algorithm is system efficiency
which depends on the successful queries performed by the readers. It is determined
as (22.1)

Total successful queries

Efficiency =  100 ð22:1Þ
Total no: of queries
22 A Novel Anti-collision Protocol for Optimization of Remote Sensing … 219

Until, total number of queries = total number of readers

Throughput measurement is defined as the number of successful reader trans-
missions per second. The throughput is evaluated by using Eq. (22.2)

Total successful queries

System Throughput = ð22:2Þ
Total Time

The Jain’s fairness index [25] is used to rate the fairness of n readers. It
evaluates the fairness of throughput distribution among the readers. The fairness
index ranges from 0 (min fair) to 1 (max fair). It is given in Eq. (22.3)

P
n
ð x i Þ2
i¼1
j(x1 ; x2 ; . . .; xn Þ ¼ ð22:3Þ
P
n
n x2i
i¼1

where xi is the throughput of the ith reader, and n is the quantity of readers.

22.4 Results and Performance Evaluation

The proposed algorithm is analyzed and compared with the state-of-the-art anti-
collision protocols using various performance evaluation scenarios. OMNET++ is
used to evaluate the performance of 100 readers, placed in a grid of 500 9 500 m.
Efficiency of proposed algorithm is demonstrated in Fig. 22.3, with different values
of MN, i.e. 16, 32 and 64. The performance with MN = 32 is highest. With
MN = 64, delay becomes huge hence a lower efficiency is achieved. The perfor-
mance of MN = 64 got better when number of readers were increased from 100.
NFRA-C shows higher percentage efficiency than NFRA and other state of the
art anti-collision protocols as shown in Fig. 22.4. For 100 readers it has a 15 %
increased efficiency as compare to NFRA, 21 % increased efficiency as compare to
PULSE and 43 % increased efficiency as compare to Colorware. The collision rate
for different RRC mechanism are compared in Fig. 22.5.
NFRA-C achieved a maximum fairness value for number of readers [75, also
providing a significant increase in throughput among readers. In Pulse the per-
formance of readers is dramatically reduced by two neighbors that take turns with
each other at querying tags. In Colorwave, each reader varies the number of colors
used in each round; hence there is a significant difference of the frequency used by
the readers to query tags, resulting in a severe impact on fairness. The fairness of
NFRA is affected by the random number and no history maintenance of successful
rounds as shown in Fig. 22.6.
220 F. Nawaz and V. Jeoti

Fig. 22.3 Efficiency of NFRA-C with MN = 16, 32, 64

Fig. 22.4 Efficiency comparison of NFRA-C with MN = 16

Fig. 22.5 Collision Rate comparison of NFRA-C

22 A Novel Anti-collision Protocol for Optimization of Remote Sensing … 221

Fig. 22.6 Fairness comparison of NFRA-C

22.5 Conclusion

NFRA-C is a novel reader to reader anti-collision approach that provides high

throughput and fairness in dense RFID reader environment. Among the existing
protocols, NFRA achieves higher network throughput however, its performance is
dependent on number of neighbors and probability of collision. The proposed
NFRA-C ensures high throughput and better fairness by introducing counters in
the contention process. Readers that have not participated in tags identification
process for a long time gets a higher priority. NFRA-C has been compared with
NFRA, Colorware, PULSE and Expoware by analyzing collisions, efficiency, and
fairness. The analysis shows that counters maintained by each reader helps in
increasing the performance and fairness of the network, and providing a
throughput better than state of the art DRE anti-collision approaches. The fairness
reaches to a value of 1(maximum) for number of readers [75 showing that all
readers receive almost same allocation and the efficiency is 15% increased as
compare to NFRA.

References

1. López, T.S.: RFID and sensor integration standards: State and future prospects. Comput.
Stand. Interfaces 33(3), 207–213 (2011)
2. Chena, Y.Y., Tsaib, M.L.: An RFID solution for enhancing inpatient medication safety with
real-time verifiable grouping-proof. Int. J. Med. Inform. 83(1), 70–81 (2014)
3. Vitaz, J., Buerkle, A., Sallin, M., Sarabandi, K.: Enhanced detection of on-metal retro-
reflective tags in cluttered environments using a polarimetric technique. IEEE Trans.
Antennas Propag. 60(8), 3727–3735 (2012)
4. Geng, L., Bugallo, M., Athalye, A., Djuric, P.: Indoor tracking with RFID systems. IEEE J.
Sel. Top. Signal Process. 8(1), 96–105 (2014)
5. Bialkowski, A., Lucey, P., Carr, P., Denman, S. Matthews, I., Sridharan, S.: Recognising
team activities from noisy data. In: IEEE Conference on Computer Vision and Pattern
Recognition Workshops (CVPRW) (2013)
222 F. Nawaz and V. Jeoti

6. Exposito, I., Gay-Fernandez, J.A., Cuinas, I.: A complete traceability system for a wine
supply chain using radio-frequency identification and wireless sensor networks [wireless
corner]. IEEE Antennas Propag. Mag. 55(2), 255–267 (2013)
7. Chena, J.C., Chengb, C.-H., Huangb, P.B.: Supply chain management with lean production
and RFID application: a case study. Expert Syst. Appl. 40(9), 3389–3397 (2013)
8. EPC Radio-Frequency Identity Protocols Class-1 Generation-2 UHF RFID Protocol for
Communications at 860 MHz–960 MHz, Jan 2005
9. Nawaz, F., Jeoti, V., Awang, A., Drieberg, M.: Reader to reader anticollision protocols in
dense and passive RFID environment. In: 11 IEEE Malaysian International Conference on
Communication, Kuala Lumpur, Malaysia, Nov 2013
10. Birari, S.M., Iyer, S.: Mitigating the reader collision problem in RFID networks with mobile
readers. In: Proceedings of the 13th IEEE International Conference on Networks, Nov 2005
11. Song, I., Hong, S., Chang, K.: An improved reader anti-collision algorithm based on pulse
protocol with slot occupied probability in dense reader mode. In: IEEE 69th Vehicular
Technology Conference, VTC Spring 2009, April 2009
12. Kwang-il, H., Kyung-tae, K., Doo-seop, E.: Distributed tag access with collision avoidance
among mobile RFID readers. In: International Conference on Computational Science and
Engineering, Vancouver, Canada (2009)
13. Shin, K., Song, Q.: RAC-multi: reader anti-collision algorithm for multichannel mobile RFID
networks. Sensors 10, 84–96 (2009)
14. Waldrop, J., Engels, D.W., Sarma, S.E.: Colorwave: an anticollision algorithm for the reader
collision problem. In: IEEE International Conference on Communications (2002)
15. Gandino, F., Ferrero, R., Montrucchio, B., Rebaudengo, M.: Probabilistic DCS: an RFID
reader to reader anti-collision protocol. J. Netw. Comput. Appl. 34(3), 821–832 (2011)
16. Konstantinou, N.: Expowave: an RFID anti-collision algorithm for dense and lively
environments. IEEE Trans. Commun. 60(2), 352–356 (2012)
17. Eom J.; Yim S.; Lee T.: An efficient reader anti-collision algorithm in dense RFID
networks with mobile RFID readers. IEEE Trans. Ind. Electron. 56(7), 2326–2336 (2009)
18. Ferrero, R., Gandino, F., Motrucchio, B., Rebaudengo, M.: Fair anti-collision protocol in
dense RFID networks. In: Third International EURASIP Conference on RFID Technology,
Spain (2010)
19. Ho, J., Engels, D.W., Sarma, S.E.: HiQ: a hierarchical Qlearning algorithm to solve the
reader collision problem. In: International Symposium on Application and the Internet
Workshops (2006)
20. Seo, H., Lee, C.: A new GA-based resource allocation scheme for a reader-to-reader
interference problem in RFID systems. In: IEEE ICC, Cape Town, South Africa (2010)
21. Cha, K., Jagannathan, S.: Adaptive power control protocol with hardware implementation for
wireless sensor and RFID reader networks. IEEE Syst. J. 1(2), 145,159 (2007)
22. GandinGandino, F., Ferrero, R., Montrucchio, B., Rebaudengo, M.: DCNS: an adaptable high
throughput RFID reader-to-reader anticollision protocol. IEEE Trans. Parallel Distrib. Syst.
24(5), 893–905 (2013)
23. Waldrop, J., Engels, D., Sarma, S.: Colorwave: an anticollision algorithm for the reader
collision problem. In: IEEE International Conference on Communications (2002)
24. Bueno-Delgado, M.V., Ferrero, R., Gandino, F., Pavon-Marino, P.: A geometric distribution
reader anti-collision protocol for RFID dense reader environments. IEEE Trans. Autom. Sci.
Eng. 10(2), 296–306 (2013)
25. Jain, R., Chiu, D., Hawe, W.: A quantitative measure of fairness and discrimination for
resource allocation in shared computer systems. Technical Report DEC-TR-301, Digital
Equipment Corporation, Maynard, Mass, USA (1984)
Chapter 23
Double Square Loop Frequency Selective
Surface (FSS) for GSM Shielding

Nur Khalida Binti Abdul Khalid and Fauziahanim Binti Che Seman

Abstract This paper proposes the deployment of Frequency Selective Surface

(FSS) in any close building that aims to block the signals from the mobile phones
without disrupting other types of communication. The proposed FSS is designed as
a band-stop filter to attenuate the Global System for Mobile Communications
(GSM) frequency bands operating at 900 and 1,800 MHz. The structure consists of
a periodic array of double square loop elements etched on FR-4 substrate. The FSS
shows a stable frequency response for the angles of incidence ranging from 0–60
with an attenuation of at least 16.7 dB. The measured results are shown to be in a
very good agreement with the simulated results.

23.1 Introduction

The increasing use of wireless communication systems can lead to an electro-

magnetic interference. This includes an interference of the widely used GSM
signals with the highly sensitive equipments employed in the buildings such as
hospitals, airports and military camps. This interference not only can degrade the
system’s performance but also can harm the security of the system [1]. In order to
reduce the interference level, an environment without the radiation from the GSM
sources needs to be provided. This can be done by using few approaches available
such as the mobile phone jammer and the shielding paint techniques [2, 3].
Practically, the mobile phone jammer can be used in any location since it can be
easily carried due to the size that is fairly small. However, this electronic device

N.K.B.A. Khalid (&)  F.B.C. Seman

Faculty of Electrical and Electronic Engineering, University Tun Hussein Onn Malaysia,
86400 Batu Pahat, Johor, Malaysia
e-mail: [email protected]
F.B.C. Seman
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 223

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_23
224 N.K.B.A. Khalid and F.B.C. Seman

requires power supply to operate [2]. Moving towards the environmental friendly
approach, the needs for power supply can be eliminated by using the shielding
paint technique. However, this technique is against the operation of the jammer as
it blocks all other microwave signals that might be useful for other types of
communication [3].
Therefore, in this study, a double square loop FSS that acts as a spatial filter
with the ability to attenuate the GSM signals is proposed to overcome the limi-
tations of the previous shielding techniques [4]. In Malaysia, the commercially
available GSM frequency bands are GSM900, GSM1800 and IMT2000 [5]. In this
paper, a prototype of the FSS is designed to reflect and block the incident signals at
GSM900 and GSM1800 only. GSM is the second generation (2G) of mobile
communication standard used for cellular network and introduced in 1990s. Since
then, few newer standards are implemented such as the Universal Mobile Tele-
communication System (3G) and the Long Term Evolution (LTE) Advanced (4G).
However, the operating frequency of GSM900 and GSM1800 are still applicable.
The bandwidth requirements provided by Malaysian Communications and Mul-
timedia Commission (MCMC) are taken into account when designing the FSS.
The FSS is widely used as antenna radomes [6], spatial filters [1], dichroic
reflectors [7] and many other applications. In order to design the FSS, there are
various types of element that can be used such as square loop [1], circular loop [8],
cross dipole [9] or convoluted [10]. The application of the proposed FSS is similar
to [4], where the FSS is designed as a dual-band to attenuate GSM frequency
bands and transmit other microwave signals. However, in [4], the authors intro-
duced a new dual-band element of the square loop type and did not emphasize the
stability of the FSS under various angles of incidence.
This paper presents the performance of double square loop FSS for normal and
oblique angles of incidence up to 60. The design details, including the dimensions
of square loop FSS are shown in Sect. 23.2. The simulation by using commercially
available Computer Simulation Technology (CST) software and discussion are
described in Sect. 23.3. The fabrication and the measurement setup conducted in
an anechoic chamber are discussed in Sect. 23.4. Lastly, in Sect. 23.5 the con-
clusion of the study is presented.

23.2 FSS Design Considerations

The square loop element is chosen due to its superior performance in terms of the
stability under various angles of incidence, cross-polarization, bandwidth and level
of band seperation [11]. The unit cell dimensions of square loop FSS are shown in
Fig. 23.1a. The square loop element is printed on dielectric substrate with the
dielectric permittivity, er = 4.3 and thickness, t = 1.6 mm. The effective dielec-
tric permittivity is calculated as eeff = 0.5(er + 1) [12]. The resonance frequency
23 Double Square Loop Frequency Selective Surface … 225

(a)
g1 (b)
w1

L1 L2
l2 l1
Z0 Z0

C1 C2
w2
g2
ls

Fig. 23.1 a Unit cell of square loop FSS, ls = 46.6 mm, l1 = 44.6 mm, l2 = 38.0 mm,
w1 = 1.0 mm, w2 = 2.5 mm, g1 = 1 mm, g2 = 3.3 mm b An equivalent-circuit model for
double square loop FSS

will decrease as the dielectric permittivity increases due to the loading effect of the
dielectric [4]. Therefore, it is very crucial to take into account the dielectric
properties of the substrate when designing the FSS.
Another important design parameters to consider is the dimensions of both
square loops FSS. Generally, a double square loop FSS can be represented by a
parallel equivalent circuit with a capacitive component in series with an inductive
component [1] as illustrated in Fig. 23.1b. The resonance frequency of the FSS is
pffiffiffiffiffiffi
inversely proportional to LC in which contributed by the periodicity, length and
width of the conductor loop. In CST, the optimization for dimensions of the outer
and inner square loops FSS are performed separately. Note that the outer and inner
square loops provide the resonance frequency at 900 and 1,800 MHz respectively.
Further optimization is required after combining the inner and outer square loops
in a single unit cell due to the effect of mutual coupling between both elements.
A higher packing density FSS per unit area is preferable due to the insensitiveness
to oblique angles of incidence.
The length and width of the outer square loop element is optimised to 44.6 and
1.0 mm respectively to tune the resonance frequency at 900 MHz. On the other
hand, the inner square loop that contributes to a higher resonance frequency at
1,800 MHz is optimised to the length of 38 mm and width of 2.5 mm. Both of the
square loops are designed to provide sufficient attenuation at the GSM operating
bands from 880 to 960 MHz and 1,710 to 1,880 MHz [5] in which the bandwidth
requirements are 80 and 170 MHz respectively.
226 N.K.B.A. Khalid and F.B.C. Seman

23.3 Simulated Results and Discussion

The proposed double square loop FSS as shown in Fig. 23.1 is simulated by using
commercially available CST Microwave Studio software. The frequency domain
solver is chosen to emulate an infinite size and highly resonant structures. Since
the FSS can be seen as an infinite periodic structure, therefore, it is only necessary
to analyse a single unit cell of the FSS.
Figure 23.2 shows the simulated transmission frequency response that is
obtained for TE polarization under various angles of incidence. At a normal
incidence, the FSS provides a maximum attenuation of 34.1 and 37.7 dB at the
resonance frequencies of 920 and 1,830 MHz respectively. The -10 dB band-
width for GSM900 frequency band is 290 MHz which is from 745 to 1,035 MHz.
On the other hand, the -10 dB bandwidth for the second frequency band,
GSM1800 is 660 MHz which is from 1,590 to 2,250 MHz. As stated in the pre-
vious section, the bandwidth requirements for GSM900 and GSM1800 are 80 and
170 MHz respectively. These indicate that the proposed FSS is capable to provide
adequate attenuation for the GSM signals far beyond the requirements. By refer-
ring to the MCMC operating band requirement, the attenuation varies between

Fig. 23.2 Simulated and measured transmission frequency response a at normal incidence
b h = 20 c h = 40 and d h = 60
23 Double Square Loop Frequency Selective Surface … 227

Table 23.1 The attenuation of the FSS for TE polarization under various angles of incidence
Frequency band provided Attenuation
by MCMC 0 20 40 60
GSM900 (880–960 MHz) 19.6–34.1 dB 20.3–34.1 dB 21.5–35.1 dB 21.8–37.5 dB
GSM1800 16.7–37.7 dB 17.2–37.7 dB 17.8–38.7 dB 21.1–40.8 dB
(1,710–1,880 MHz)

19.6 and 34.1 dB for 880–960 MHz while 16.7–37.4 dB for 1,710–1,880 MHz as
tabulated in Table 23.1.
As the angle of incidence increases to 20, both resonance frequencies as well
as the attenuation remain the same. However, this trend slightly changes as the
angle of incidence varies to 40 and 60. At 40, the resonance frequency for the
lower band increases to 930 MHz with an attenuation of 35.1 dB. The upper
resonance frequency remains the same as 1,830 MHz. However, the maximum
attenuation at that resonance frequency improves to 38.7 dB. As the angle of
incidence increases up to 60, the lower resonance frequency shifts right by
2.8 %–946 MHz. In contrast, the upper resonance frequency shifts left by 0.5 %–
1,820 MHz. The attenuation for both resonance frequencies, 946 and 1,820 MHz
at the oblique angle, 60 greatly improves to 37.5 and 40.8 dB respectively. For
the GSM900 frequency band, the -10 dB bandwidth performance increases to
453 MHz which is from 6,641 to 1,114 MHz. On the other hand, for the GSM1800
frequency band, the -10 dB bandwidth slightly decreases to 636 MHz which is
from 1,518 to 2,154 MHz as the angle of incidence varies to 60. By referring to
the MCMC operating band requirements, at 60, the attenuation varies between
21.8 and 37.5 dB for 880–960 MHz while 21.1–40.8 dB for 1,710–1,880 MHz as
shown in Table 23.1. These results show that the proposed FSS manages to cover
the required bands provided by the MCMC under various angles of incidence.

23.4 Fabrication and Measurement

The square loop FSS was etched on FR-4 dielectric subtrate to emulate the
required dielectric permittivity of 4.3. The FR-4 substrate was used due to
the availability of the material in the laboratory although it is expected that the
dielectric substrate with lower dielectric permittivity offers a better attenuation.
Since the FSS does not require a ground plane, a single layer PCB, where the
copper is only mounted on one side of the PCB was used. The fabricated square
loop FSS is shown in Fig. 23.3. Since the size of the fabricated FSS was small due
to the limitation of the UV exposure machine, several pieces of the FSS were
fabricated. All the fabricated FSSs were properly aligned on the plywood to
immitate an infinite size sheet as defined in the computer model.
The bi-static measurement technique was employed in order to measure the
fabricated FSS. The measurement setup consists of two horn antennas with the
228 N.K.B.A. Khalid and F.B.C. Seman

Outer loop
Inner loop

Fig. 23.3 The fabricated square loop FSS

Fig. 23.4 The measurement

set up with the FSS placed in FSS board
between the transmitter and
the receiver

Receiver
Transmitter

gain of 15 dB for transmitting and receiving, with both were connected to the
network analyzer by using the coaxial cable. These two horn antennas were sep-
arated about 1 m away to ensure that the rule to operate in far-field region was
obeyed. This is based on the equation, dfarfield C 2D2/k where D is the horn
antennas’ maximum dimension and k is the wavelength [13]. The frequency range
of the network analyzer was set up to operate from 800 to 2,000 MHz. The
fabricated FSS was positioned in the middle between the transmitter and the
receiver as shown in Fig. 23.4.
During the measurement, the loss due to the propagation path was taken out in
order to ensure that the attenuation of the microwave signal was contributed
merely by the FSS. This was done by subtracting the receiving signal without the
FSS with the receiving signal with the FSS. The comparison between the simu-
lated and the measured results is shown in Fig. 23.3. The fluctuation appears in the
measured results is expected due to the scattering of the microwave signals from
the surrounding. In general, the measured results showed a good agreement with
the simulated results.
23 Double Square Loop Frequency Selective Surface … 229

23.5 Conclusions

The performance of the FSS with the double square loop elements is demonstrated.
The proposed FSS shows a good shielding performance in terms of the bandwidth
and the sensitivity towards different angles of incidence ranging from 0 to 60 for
TE polarization. A maximum reflection at 900 and 1,800 MHz with an attenuation
of at least 16.7 dB is achieved. The measured results are shown to be in a very
good agreement with the simulated results. Our future work will focus on the
performance of the FSS for TM polarization.

Acknowledgments The authors would like to thank the Ministry of Education Malaysia for
supporting this study under the Exploratory Research Grant Scheme (ERGS/1/2012/TK06/
UTHM/02/1/E005). Much appreciation also goes to the Research Center for Applied Electro-
magnetics, Universiti Tun Hussein Onn Malaysia for providing the measurement facilities.

References

1. Sung, G.H., et al.: A frequency-selective wall for interference reduction in wireless indoor
environments. IEEE Antennas Propag. Mag. 48(5), 29–37 (2006)
2. Mishra, N.K.: Development of GSM—900 mobile jammer: an approach to overcome existing
limitation of jammer. In: Proceedings of the Fifth IEEE Conference Wireless Communication
and Sensor Networks (WCSN), pp. 1–4. (2009)
3. Y-shield EMR-Protection, Shielding Paints [Online]. Available: http://www.yshield.com/
shielding-paints.html
4. Kiermeier, W., Biebl, E.: New dual-band frequency selective surfaces for GSM frequency
shielding. In: Proceedings. 37th Eur. Microwave Conference, pp. 222–225. (2007)
5. Malaysian Communications and Multimedia Commission, Spectrum Allocation [Online].
Available: http://www.skmm.gov.my/Spectrum/Spectrum-Allocation-List/Spectrum-Allocation.
aspx
6. Zhao, J., Xu, X.: Study of the effect of a finite FSS radome on a horn antenna. In: IEEE
International Conference on Microwave Technology and Computational Electromagnetics
(ICMTCE), pp. 74–76. (2011)
7. Pasian, M., et al.: Accurate modeling of dichroic mirrors in beam-waveguide antennas. IEEE
Trans. Antennas Propag. 61(4), 1931–1938 (2013)
8. Taylor, P.S., et al.: A passively switched dual-band circular FSS slot array. In: Proceedings of
the IEEE-APS Topical Conference Antennas and Propagation in Wireless Communications
(APWC), pp. 648–651. (2011)
9. Kiani, G.I., et al.: Cross-dipole bandpass frequency selective surface for energy-saving glass
used in buildings. IEEE Trans. Antennas Propag. 59(2), 520–525 (2011)
10. Parker, E.A., et al.: Frequency selectively screened office incorporating convoluted FSS
window. Electron. Lett. 46(5), 317–318 (2010)
11. Wu, T.K.: Frequency Selective Surface and Grid Array. Wiley, New York (1995)
12. Munk, B.A.: Frequency Selective Surfaces: Theory and Design. Wiley, New York (2000)
13. Raspopoulos, M., Stavrou, S.: Frequency selective buildings through frequency selective
surfaces. IEEE Trans. Antennas Propag. 59(8), 2998–3005 (2011)
Chapter 24
Analysis of the Active Region
of Archimedean Spiral Antenna

Abdirahman Mohamoud Shire and Fauziahanim Che Seman

Abstract The paper elaborates the current distribution of the Archimedean Spiral
Antenna (ASA), demonstrating the concept of frequency dependent active region
and this determines the effective radiation area on the spiral arm. The band theory
is used to explain the theoretical principles of the operation of the spiral antenna.
In this paper, a two arm Archimedean spiral antenna is designed using Computer
Simulation Technology, (CST). The properties of the active region of the spiral
antenna are analyzed in different types of dielectric substrates. The calculated and
simulations results show the position of the active region is very dependent to the
operating frequency and dielectric permittivity of the substrates. The maximum
surface current induced to the spiral arm increases as the operating frequency and
permittivity of the dielectric substrates reduces.

24.1 Introduction

Frequency independent antennas currently receive huge interest [1, 2] due to their
special properties in wideband applications. The term frequency-independent (FI)
is reserved for antennas with electrical characteristics that vary insignificantly over
an extremely wide operating frequency range [3]. This properties is valid on its
own and an improper insertion of the ground plane (GP) behind the antenna might
distrupt the FI belonginess [4]. In this paper, the key important parameters of the
FI of an Archimedean Spiral Antenna backed by a metalic plate has been analysed
accordingly. In ASA the FI property is contributed by the position of the active

A.M. Shire (&)  F.C. Seman

Wireless and Radio Science Centre (WARAS), Faculty of Electrical and Electronic
Engineering, University Tun Hussein Onn Malaysia, 86400 Parit Raja, Johor, Malaysia
e-mail: [email protected]
F.C. Seman
e-mail: [email protected]

Ó Springer International Publishing Switzerland 2015 231

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_24
232 A.M. Shire and F.C. Seman

region on the spiral arm where the effective radiation pattern of the ASA take
places. Then the GP is carefully inserted in order to provide a unidirectional
radiation pattern and a higher gain [5] and later, the possible change of the active
region indicated by the current distribution on the spiral arm is analysed. Fur-
thermore, the variation of the magnitude of the current induced to the spiral arm
with operating frequency range and dielectric substrates is investigated.
The organization of the paper is described as follows; in the theoretical oper-
ation section of the ASA, the current distribution along the spiral arm is analyzed
numerically. Later, the maximum current induced on the spiral arm associated by
the position of the active region is analyzed based on the numerical and the
simulation results.

24.2 Theoretical Operation for Spiral Antenna

Considering a two-arm spiral antenna operating in Mode 1, which is fed from its
center at ports X and X’ as illustrated in Fig. 24.1. The band theory (radiating ring
theory) is applied to explain the theoretical principles behind the operation of the
antenna. The current distribution on the antenna is divided into three areas, which
are feeding region, transition region and decay region [6].

C= λ

Z’ Z
Y Y’
− (π)
X’ X
+ (0)

Fig. 24.1 Active region of spiral antenna based on Radiation band theory
24 Analysis of the Active Region of Archimedean Spiral Antenna 233

The spiral antenna is fed at r +(0) and r -(0), which presents the time harmonic
excitation of the spiral. Then, the current at the point r ±(h) and the phase of the
current on the spiral antenna is expressed by the travel time from the feeding is
defined in Eqs. (24.1) and (24.2) respectively [3, 7, 8]:

I ðh; tÞ ¼ i ðhÞejxt : ð24:1Þ


td ¼ s ðhÞ vp : ð24:2Þ

Where the real current of the spiral is the real part of I±(h, t), i±(h) is a complex
of h, vp is the phase velocity and s±(h) is the distance along the spiral arms and it is
found as :
  sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
    rffiffiffiffiffiffiffiffiffiffiffiffiffi
dr ðhÞ
Zh Zh drðhÞ 2 dh  ah  1
  
s  ð hÞ  s  ð hÞ ¼  dh ¼ þ r dh ¼ e  1 1 þ 2 :
0
 dh  0
dh dh a
ð24:3Þ

The phase of the current is

\I ðh; tÞ ¼ \I ðh; t  td Þ ¼ \I ðhÞejxt þ \ejxt

2pf s  ðhÞ 2pf s  ðhÞ ð24:4Þ
¼ \I ðh; tÞ  ¼ \I ðh; tÞ  :
vp k

In the Feeding region, the spiral antenna is fed in a balanced feeding at points X
and X’ as illustrated in Fig. 24.1, and the feeding diameter is very small compared
to the wavelength, due to that, the current these points are anti-phase and it is
calculated as:

\Iþ ðp; tÞ  \I ð0; tÞ  \Iþ ð0; tÞ  \Iþ ðp; tÞ ¼ p: ð24:5Þ

The simplified mathematical formulation of active region of spiral antenna is

analyzed and presented:

r ðhÞ ¼ eahþb : ð24:6Þ

The spiral arm rotates accordingly which can be represented as a curve [7, 8]
and in polar coordinate, the constants a determines the rate of wrapping and scales
the curve, constant b scales the curve, while h is angle of the phase and calculated
as:
!
1 k
h ¼ log pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : ð24:7Þ
a 2ðeap  1Þ 1 þ a2
234 A.M. Shire and F.C. Seman

Spiral with tight wrapping can facilitate to describe the active region in simple
way, so letting b tend to zero; it is obtained an approximate radiation condition
from Eqs. (24.6) and (24.7):

ak ak
r ðhÞ ¼ eah ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffi   k=ð2pÞ;
ap
2ð1 þ e Þ 1 þ a 2 2 ð 1 þ 1 þ ap þ . . .Þ1
2pr ðhÞ ¼ k;
ð24:8Þ

Based on ‘‘radiation band’’ theory, Eq. (24.8) clearly elaborates the Mode 1
radiation predominantly occurs from the area whose circumference is approxi-
mately one of wavelength (C = 1k) [8–10] as indicated in Fig. 24.1 at points Y
and Y’, which belong to neigbouring arms, because the current flowing in con-
tiguous arms is in phase leading to coherent or constructive radiation in the far-
field. On the other side, the same conditions occured diametrically opposite points
Z and Z’. This area defines the position of the active region, where most of the
radiation takes place due to the decaying of the in-phase current.
Outside these regions the current is not in- phase and, therefore, the radiated
field interferes destructively [8–10]. The nonradiated traveling wave currents will
flow past this region, and if the size of the spiral permits, radiate in the next
properly phased section in which the in-phase current conditions will show up at
odd wavelength circumferences of the spiral and higher order modes will radiate
(this will occur at a circumference equal to three wavelengths; mode 3 for a two-
arm spiral). If the spiral is not large enough, the currents will reach the end of the
spiral arms where they are either absorbed or reflected back toward the spiral’s
center.The position of the active region vary accordingly with the changes of the
dielectric permittivity as defined by Eq. (24.9).

keff k0
reff ¼ ¼ pffiffiffiffiffiffi : ð24:9Þ
2p 2p eeff

24.3 Archimedean Spiral Antenna Design

The ASA is designed using CST MWS where the operating frequency range is
chosen to be in the 2–12 GHz range and this frequency range is available for UWB
applications. The arm width (w), arm spacing (s), inner radius (r1) and the outer
radius (r2) of the ASA are calculated using equations in [5]. The r2 is optimized to
be 35 mm while the other parameters are optimized to be 1.3 mm. The number of
turns of each arm is selected as 8 turns and in computer model, the ASA is fed at
the center using discrete port. The spiral arm is cascaded to an electrically k/4
thick dielectric substrate backing by a metallic plate.
24 Analysis of the Active Region of Archimedean Spiral Antenna 235

24.4 Results and Analysis

Figure 24.1 shows the current distribution on the spiral arm in which indicating the
current behavior in the feeding region, transition region and decay region. In the
feeding region, the current at neighbor arms is anti-phase as proven in Eq. (24.5)
thus there is no radiation takes place and this is illustrated in Fig. 24.1a. The
calculated phase at Eq. (24.5) is 180°, so since the phase difference at points X and
X’ is 180° out phase, the radiation is negligible because the two current waves of
the two neighbor arms are cancelling each other due to the anti-phase condition.
According to the current distribution theory of spiral antenna, the current reaches
its maximum at the end of the feeding region which is clearly observed in the
simulated current on the spiral antenna, in which the maximum current magnitude
is 82.57 A/m. The current starts to decay to 69.24 A/m at the transition region as
shown in Fig. 24.2b, but still no effective radiation takes place in this region since
the currents at the neighbor arms are anti-phase.
Identification of the position of the active region in decay region is done based
on the (i) calculation (See Table 24.1, third column) by referring to Eq. (24.8) and
(ii) current distribution of the CST simulation results (Refer to Table 24.1, fourth
column). Note that the spiral arm is attached on the free space so the dielectric
permittivity, er is 1. Figure 24.1 shows that at the lower operating frequency, the
active region of the spiral is located around the perimeter (see Fig. 24.2a) of the
antenna and as the frequency increases the position of the active region moves
inwards the center feeding (see Fig. 24.2c). The simulated active regions of the
active region are shown in Fig. 24.2. The calculated and simulated locations of the
active region are tabulated in Table 24.1 and the comparisons of the two results are
shown in Fig. 24.3.
The ASA spiral antenna is simulated on four different substrates permittivity, er
in order to examine the effect of the dielectric substrate on the characteristics of the
active region of the ASA. It is clearly seen in Table 24.2 that as er increases from 1
to 2.33, the maximum current induced at the active region reduces from 82.57 to
70.41 A/m. The maximum current reduces 12.7 % as er increases from 4.3 to 10.2.
This is due as the currents propagate along the spiral arm, a higher dielectric
permittivity assimilates more input power fed to the antenna and therefore leads to
reduction of the current distribution on the spiral arms [10]. The current magnitude
also degrades as the operating frequency of the ASA increases because spiral
antenna behaves like inductive circuit at higher frequencies, so the impedance of
the inductive circuit increases as the frequency increases which leads to the
reduction of the current magnitude [11].
The location of the active region of the ASA changes as the dielectric per-
mittivity increases from 1 to 10.2. Numerically, based on Eq. (24.9), as the sub-
strate permittivity increases, the location of the active region shrinks towards the
236 A.M. Shire and F.C. Seman

Fig. 24.2 Current distribution of two arms Archimedean spiral antenna is shown as black band
for different frequencies a 3.1 GHz, b 6.85 GHz and c 10.6 GHz

center feeding of the ASA. This is proven at the operating frequency of 2 GHz
when er = 2.33 the active region is positioned 18.5 mm from the center fed and
this significantly reduces 10.1 mm when er = 10.2 as demonstrated in Fig. 24.4.
As the operating frequency increases to 12 GHz, the position of the active region
occurs 1.68 and 3.08 mm away from the feeding point for er = 10.2 and er = 2.33
respectively. Again, the higher dielectric substrate slows the traveling wave in
which leads to shrinkage of the active region area and increases the coupling
between the neighboring arms [5]. Both calculation and simulation results dem-
onstrates excellent agreement as shown in Fig. 24.4.
24 Analysis of the Active Region of Archimedean Spiral Antenna 237

Table 24.1 Details position of the active region

Frequency (GHz) k (mm) r ðhÞðmmÞ Calculation r ðhÞðmmÞ Simulation
2 150 23.87 25
3 100 15.92 16
4 75 11.94 12.5
5 60 9.4 9.5
6 50 7.96 8.2
7 42.86 6.82 8
8 37.5 5.96 7.5
9 33.33 5.31 6
10 30 4.78 5.5
11 27.27 4.34 4.8
12 25 3.98 4

Fig. 24.3 Calculation and 12

simulation results of position 11 Cal
of the active region of ASA 10
Sim
9
Freq (GHz)

8
7
6
5
4
3
2
1
0
0 3 6 9 12 15 18 21 24
λ/2π (mm)

Table 24.2 The maximum current distribution (A/m) at the active region
Substrate Permittivity (er) 3.1 GHz 6.85 GHz 10.6 GHz
(A/m) (A/m) (A/m)
1 82.57 69.24 60.75
2.33 70.41 62.38 56.3
4.3 65.75 57.83 52.93
10.2 57.39 53.21 49.51
238 A.M. Shire and F.C. Seman

12
11 er=2.33 (cal)
10 er=4.3 (cal)
Frequency (GHz) 9 er=10.2 (cal)
er=2.33 (sim)
8
er=4.3 (sim)
7
er=10.2 (sim)
6
5
4
3
2
1
0
0 3 6 9 12 15 18 21 24
λ /2π (mm
(

Fig. 24.4 Calculated and simulated results of the position of the active region of ASA on
different dielectric substrate

24.5 Conclusion

The characteristics of the active region based calculation and simulation results are
presented in this paper. The ASA is designed in CST to operate between 2 and
12 GHz for UWB applications. It demonstrated that the position of the active
region depends on both operating frequency and dielectric substrate. As the
operating frequency and the substrate permittivity increases, the locality of the
active region shrinks inward the antenna.

References

1. Dyson, J.D.: Frequency-independent antennas (survey of development), Electronics, 1-01 35,

39–44 (1962)
2. Jordan, E.C., Deschamps, G.A., Dyson, J.D., Mayes, P.E.: Developments in broadband
antennas. IEEE Spectr. 1, 58–71 (1964)
3. McFadden, M.: Analysis of equiangular spiral antenna. Ph.D. Dissertation, Georgia Institute
of Technology (2009)
4. Nakano, H., Nogami, K.: A spiral antenna backed by a conducting plane reflector. IEEE
Trans. Antennas Propag. AP-34, 1417–1423 (2007)
5. Shire, A.M., Seman, F.C.: Effects of dielectric substrate on the performance of UWB
Archimedean spiral antenna. In: Proceeding of the 2013 IEEE International Conference on
Space Science and Communication (IconSpace), pp. 412–415. (2012)
6. Yeh, Y.S., Mei, K.K.: Theory of conical equiangular spiral antennas, part ii- current
distributions and input impedances. IEEE Trans. Antennas Propag. 16, 14–21 (1964)
7. Rumsey, V.H.: Frequency independent antennas. IRE National Convention Record 5,
114–118 (1957)
24 Analysis of the Active Region of Archimedean Spiral Antenna 239

8. Kaiser, J.A.: The Archimedean two-wire spiral antenna. IRET Trans. Antennas Propag. AP-
8, 312–323 (1960)
9. Bawer, R., Wolfe, J.J.: The spiral antenna. IRE Int. Convention Rec. 8, 84–95 (1960)
10. Burdine, B.H.: The spiral antenna Massachusetts Institute of Technology, Research
Laboratory Technical Report, (1955)
11. Rosu, I.: Microstrip, Strip Line and CPW Design RF Technical Artical. 2014. http://www.qsl.
net/va3iul
Chapter 25
Optimization of BER Performance
in the MIMO-OFDMA System for Mobile
WiMAX System Using Different
Equalization Algorithm

Azlina Idris, Norhayati Abdullah, Nor Azlizan Hussein and D.M. Ali

Abstract Combination of Multiple Input Multiple Output (MIMO) and Orthog-

onal Frequency Division Multiple Access (OFDMA) is implemented to offer a
simple and high performance system as to increase channel capacity and serve
high data rate. Even though the OFDMA concept is simple in its basic principle,
but it suffers one of the most challenging issues, which is synchronization error
that introduces the inter-symbol interference (ISI), thus degrades the signal per-
formance. The goal of this paper is to provide a method to mitigate this ISI by
employing the equalizers at the receiver end and using Space Time Frequency
Block Codes (STFBC) to improve the Bit error rate (BER) performance and to
achieve a maximum diversity order in MIMO-OFDMA by using simulation based
on the platforms of MATLAB software As a result, the BER performance is
improved when implementing equalizers at the receiver with STFBC outperforms
the conventional system without equalizer with a maximum diversity order and an
efficient bandwidth in the Mobile WiMAX system.

Keywords Inter-symbol interference (ISI) 

Multiple input multiple output

(MIMO) Orthogonal frequency division multiple access (OFDMA) Bit error 
 
rate (BER) Space time block codes (STFBC) Worldwide interoperability for
microwave access (WiMAX)

A. Idris  N. Abdullah (&)  N.A. Hussein  D.M. Ali

Faculty of Electrical Engineering, Universiti Teknologi MARA,
40450 Shah Alam, Selangor, Malaysia
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 241

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_25
242 A. Idris et al.

25.1 Introduction

Successful deployment of wireless voice communication systems promises a bright

future for wireless high data rate services such as internet access or multimedia
applications [1]. Orthogonal Frequency Division Multiplexing (OFDM) provides
such high data rate services and considered as a good choice of this matter due to its
ability to overcome multipath fading. Currently, there is a strong interest in
extending the OFDM concept to multiuser communication scenarios. A prominent
example of this trend is orthogonal frequency division multiple access (OFDMA)
technology, which results from the combination of OFDM with a frequency divi-
sion multiple access (FDMA) protocol [2]. Many broadband wireless networks
have now included the MIMO option in their protocols. In principle, OFDMA and
MIMO can be combined to offer the benefits of simplicity, high performance
system [3] and exploitation of the multipath diversity which increases the
achievable rate and enhances link reliability [4]. The proposed method is to test the
diversity performance of this system by using the Alamouti code technique which is
the simplest compared to the others. Basically, in this OFDMA system, there are
two basic diversity order systems which are the Space-Time Block Codes (STBC)
and Space-Frequency Block Codes (SFBC), while the Space-Time-Frequency
Block Codes (STFBC) is the combination of both. STFBC can offer spatial, tem-
poral and frequency diversity MIMO channels. The coding distributes symbols
along transmit antennas, time slots and at different frequencies. This STFBC may
contain several OFDM symbols which can increase diversity order [5].
Even OFDMA has a lot of advantages, yet there are still some disadvantages
exist in this system. For instance, different users share available subcarriers in
OFDMA thus, synchronization becomes a difficult task. The receiver must esti-
mate a number of parameters and need to compensate inter-symbol (ISI) inter-
ference [6]. The cyclic prefix (CP) can be added to overcome this matter but ISI
may still exist if channel delay spread is larger than the CP and this will severely
affect the system performance. Thus, by adapting the ideal equalizer at the
receiver, performance degradation and ISI can be reduced. There were several
work done previously for instance, in [5], the researcher investigated BER of
system performance using STFBC with intercarrier interference self-cancellation
scheme (ICI-SC) without equalizer to reduce ICI only but not ISI.
Besides, in [7], the author introduced a diversity technique, but it is applied for
MIMO-OFDM system using a new ICI-SC technique subcarrier mapping scheme
without equalizers. However, it is difficult to obtain frequency diversity gain and
suffers ICI and ISI. In [8], the researcher studied the performance evaluation of
BER using STFBC MIMO-OFDM using equalization algorithm. However, the
system could not achieve the maximum diversity order and an efficient bandwidth.
In [9], the researcher evaluates system performance using pair-wise error proba-
bility with specific subcarrier mapping and a linear equalizer for MIMO-OFDM
system. So far, there is no literature on performance evaluation of BER using
equalization and diversity order technique (STFBC) in MIMO-OFDMA system.
25 Optimization of BER Performance in the MIMO-OFDMA System … 243

Therefore, this research paper is proposed. The objectives of this paper are to
simulate the BER performance of MIMO-OFDMA using MATLAB software, to
mitigate the inter-symbol interference (ISI) in the OFDMA system, to implement
the equalizers at the receiver and to evaluate the different type of diversity order in
this system.

25.2 System Model

An OFDMA system is defined as one in which each terminal occupies a subset of

subcarriers (termed an OFDMA traffic channel), and each traffic channel is
assigned exclusively to one user at any time [10]. In OFDMA, users are not
overlapped in frequency domain at any given time [11]. However, the frequency
bands assigned to a particular user may change over the time.
In order to mitigate the presence of ISI, the implementation of equalization at
the receiver can be made in frequency or time domain to diminish such interfer-
ence. This paper will focus on implementing three different types of equalizers
which are Zero-Forcing Equalizer (ZF), Minimum Mean Square Error (MMSE)
Equalizer and Maximum Likelihood Sequence Estimation (MLSE) Equalizer. In
this paper, the BER performance will be compared through with and without
equalization at the receiver OFDMA system. Figure 25.1 above shows an example
of a baseband model of OFDMA system. The block diagram basically comprising
of three major parts namely transmitter, channel and receiver. The data input at the
transmitter side are random data which are being produced within the MATLAB
command language. Then, the random data will be generated in serial format to
perform serial to parallel conversion. The serial data stream represents the data
information to be transmitted.
The parameters such as number of subcarriers used and the FFT size are using
Mobile WiMAX wireless communication standards in OFDMA technology sys-
tem. Quadrature Amplitude Modulation (QAM) is being used to perform modu-
lation on parallel stream. Each symbol is presented by complex number in phase
and quadrature phase vector [12]. The selection of modulation scheme applied to
each sub-channel depends solely on the compromise between the data rate
requirement and transmission robustness [12]. The samples of the transmitted
OFDM signal can be obtained by performing an IFFT operation on the group of
data symbols to be sent on orthogonal sub-carriers [13]. The IFFT is used to
convert the frequency domain data into time domain signal while maintaining the
orthogonality of subcarriers [14]. Cyclic prefix consists of a block of redundant
samples at the beginning of each transmitted frame and it is also a cyclic extension
of the symbol to eliminate ISI effects on original symbols.
Additive white Gaussian noise (AWGN) channel is the most common channel
model but it does not work well due to multipath propagation. It is static in real
environment and applied in simulation in the MATLAB software. Practical
channel that is being used is Rayleigh channel which may introduce a different
244 A. Idris et al.

Fig. 25.1 OFDMA system block diagram with equalizer

phase, amplitude attenuation, delay and Doppler shift to the signal. At the receiver
part, serial input data is converted to parallel form and the symbol transformation
is performed by FFT [12]. The output from the FFT will contain interferences or
distortions including ISI. Equalization is implemented to mitigate this ISI. The
equalized output is demapped, deinterleaved and then convolutionaly decoded to
get back original data words [12]. To get the original message information, the
data words will then be multiplexed.

25.2.1 Zero Forcing Equalizer

Zero Forcing Equalizer is a form of linear equalization algorithm which applies

the inverse of the channel frequency response to the received signal, to restore the
signal after the channel in communication system [10]. This form of equalizer was
first proposed by Robert Lucky. It has many useful applications. For example, it is
applied for IEEE 802.16e (Mobile WiMAX) in MIMO, where knowing the
channel allows recovery of the two or more streams which will be received on top
of each other on each antenna. The name Zero-Forcing corresponds to bringing
down the intersymbol interference (ISI) to zero and will be useful when ISI is
significant compared to noise [11].
Figure 25.2 above shows an example to show that there are different parameters
used in Zero-Forcing equalization. Let CZF(k) be the equalizing circuit filter. The
LTI filter with transfer function, CZF(k) is considered to be the ZF equalizer, that
can be realized by multiplying the OFDMA received signal as an Eq. (25.1) with
the vector 1/H(k) which produces the Eq. (25.2). In this case, the equalizer filter
compensates for the channel induced ISI as well as the ISI, but this is not elimi-
nating all ISI because the filter is of finite length.

yðkÞ ¼ xðkÞH ðkÞ þ wðkÞ þ IðkÞ ð25:1Þ

1
Czf ðkÞ ¼ ð25:2Þ
HðkÞ
25 Optimization of BER Performance in the MIMO-OFDMA System … 245

Fig. 25.2 Block diagram of zero-forcing equalizer

where y(k) is complex coefficient envelope of FFT, x(k), H(k) and w(k) are
frequency domain equivalents of input data, channel impulse response and
AWGN noise respectively.

25.2.2 Minimum Means Square Error Equalizer

A more balanced linear equalizer in this case is the minimum mean-square error
equalizer, which does not usually eliminate ISI completely but instead minimizes
the total power of the noise and ISI components in the output.
Since this scheme can minimize the mean square error (MSE) between the
desired equalizer output and the actual equalizer output, it is found that adaptive
MMSE equalizer is an effective and feasible method to mitigate the serious effects
of multipath dispersion.
Figure 25.3 above shows a block diagram whereby a few parameters involved
which are, H(k) is the channel impulse response, Xp(k) is the actual output while
the Yp(k) is the desired output. The ek is the error between these two output of the
system .In this type of equalizer, the tap weights are chosen as they minimize the
mean-square-error (MSE) of all the ISI terms and the noise power at the output of
the equalizer. MMSE is the expected value of the squared difference between the
desired data symbol and the estimated data symbol. Error between desired and
actual output is given by [12].

e k ¼ xp ðkÞ  Yp ðkÞT Wk ð25:3Þ

where Wk is the weight vector of filter. Mean square error is the square of
Eq. (25.3) which produces Equation as followed,
 2
MSE ¼ E Xp ðkÞ þWk RWk  2PT Wk ð25:4Þ

R and P are the correlation and auto-correlation matrices. To minimize ISI we have
to find filter weights which minimized when R = P Wk. The vector Wk corre-
sponds to the number of taps in the equalizing filter. The equalizer correction term
246 A. Idris et al.

Fig. 25.3 Block diagram of MMSE equalizer

which is the inverse of channel response multiplying with the Eq. (25.1) produces
the equalizer output as below,

1
CMMSE ¼ 1=½H ðkÞþNO  ð25:5Þ
H ðkÞ þ NO

where the H(k) is the channel impulse response, while the No is the noise in the
system.

25.2.3 Maximum Likelihood Sequence Estimator

Among of all those equalizers, MLSE exhibits the strongest capability in com-
pensating ISI. However, its main problem is the complexity that increases expo-
nentially with the memory length of the channel. The MLSE Equalizer block uses
the Viterbi algorithm to equalize a linearly modulated signal through a dispersive
channel. The block processes input frames and outputs the maximum likelihood
sequence estimate (MLSE) of the signal, using an estimate of the channel modeled
as a finite input response (FIR) filter.
Figure 25.4 above shows the architecture of the MLSE equalizer. The receiver
compares the time response with the actual received signal and determines the
most likely signals. The problem to be solved is to use the observations {r(k)} to
create a good estimation of {x(k)}. In this system, the r(k) is received signal, h(k)
denotes the overall channel response of the system, whereas z(k) is the output of
the received signal passed to the match filter, and the Viterbi algorithm is obtained
by computing the recursive relation iteratively and produce the estimated
sequence, ^xðkÞ which is defined to be sequence of values which maximize the
functional [12].

CMLSE ¼ pðrjxÞ ð25:6Þ

where p(r|x) denotes the conditional joint probability density function of the
observed series r(k) given that the underlying series has the values x(k).
25 Optimization of BER Performance in the MIMO-OFDMA System … 247

Fig. 25.4 Block diagram of MMSE equalizer

25.3 Simulation Results

One of the important parameter in wireless communication for quality measure-

ment of recovered data is the performance of BER. It is observed that different
equalization techniques can give low BER performance under multipath fading
environment, but with a slight changes in BER the quality changes many folds
[13]. Simulation results are plotted for bit error rate (BER) performance of MIMO-
OFDMA system and being compared with and without the implementation of
three different equalizers. Besides, BER performance also can be observed when
the three different diversity methods used in the system. Table 25.1 shows the
system parameters for MIMO-OFDMA for Mobile WiMAX system that have been
used in this project [14].

25.3.1 Zero-Forcing, MMSE, MLSE Equalizer

Figure 25.5 above shows that BER performance can be compared between dif-
ferent equalizers. From the simulation, it can be observed that the MLSE equalizer
gives the best performance which produces the less interference (ISI) in the system
compared to MMSE and Zero-Forcing equalizer. So, as the interference is
decreased, the BER is decreased as well but SNR is increased. This is because of
MLSE evaluates a sequence of received data samples to determine the most likely
correct transmitted sequence. This is proven in the Eq. (25.6), as the most likely
transmitted signal, x(k) is increased, the equalizer output, CMLSE is also increased.
So, it can successfully minimize the interference in the signal. That is why changes
in BER can be seen clearly at high SNR = 17 dB. On the other hand, the MMSE
equalizer gives better performance compared to ZF equalizer because it is not only
equalizing the channel but also suppressing the noise as in Eq. (25.5) which proves
that when the channel impulse response, H(k) and noise, No is decreased, the
output equalizer, CMMSE will increased. Besides, by applying the ZF equalizer in
the system, the performance is improved too as in Eq. (25.2) whereby the channel
impulse response, H(k) is decreased, the output equalizer CZF will increased which
tends to produce an increment of SNR, as the interference and BER is reduced.
248 A. Idris et al.

Table 25.1 MIMO-OFDMA parameter [14]

System bandwidth (MHz) 1.25 2.5 5 10 20
Sampling frequency (MHz) 1.4 2.8 5.6 11.2 22.4
FFT size 128 256 512 1024 2048
Subcarrier spacing (kHz) 10.94
OFDM symbol duration (ls) 102.86
Useful symbol time (ls) 91.43
Cyclic prefix (ls) 11.43

Fig. 25.5 BER comparison using ZF equalizer, MMSE equalizer, MLSE equalizer and without
equalizer

But it does not perform as well as the MLSE and MMSE equalizer since this
equalizer forces the ISI to zero when only the ISI is significant but does not
consider the noise.

25.3.2 STF, ST, SF Diversity

Figure 25.6 above shows the BER performance comparison when using different
diversity order system when implementing the MLSE equalizer since this is the
best equalizer among the others. It can be seen that by using space, time, frequency
diversity (STF), it offers the maximum diversity order compared to space, time
25 Optimization of BER Performance in the MIMO-OFDMA System … 249

Fig. 25.6 BER performance with three different diversity methods

(ST) diversity and space, frequency (SF) diversity because it can transmit data in
different time and frequency slot which improved the performance of the system as
the SNR is increased while BER and ISI is decreased. It means that, there is least
amount of ISI when STF diversity order is applied in the system. At higher
SNR = 17 dB, BER value is at 7 9 10-3, which gives the least amount of ISI.
This is because in the Eq. (25.6), as the most likely transmitted signal, x(k) is
increased, the equalizer output, CMLSE is also increased. So, it can successfully
minimize the interference in the signal.

25.4 Conclusion

The multipath propagation causes fading of received signal power which leads to
ISI. The equalizers are used to improve the distorted received signal caused by ISI.
This paper compares the performance of equalized system with the unequalized
system to observe which equalizer is the best among three different equalizers that
can generate the least ISI in the system and at the same time, it will improve the
BER performance within the OFDMA system. From the simulation results, it can
be proved that with the equalization system, the ISI can be mitigated, the maxi-
mum diversity order can be achieved and the BER performance also can be
improved.
250 A. Idris et al.

Future Work
This project basically applies the basic concept in Mobile WiMAX system. In
future works, Quadrature Phase Shift Keying (QPSK) as the modulation scheme
can be used instead of using QAM. Moreover, the BER system performance also
can be investigated for other types of equalizer such as Decision Feedback
Equalizer, Blind Equalizer, and Linear Equalizer.

Acknowledgments This work was supported in part by the Research Management Institute
University Teknologi MARA under Excellent fund (Research Intensive Faculty) grant number
600-RMI/DANA 5/3/RIF(86/2012).

References

1. Rhee, W., Cioffi, J.M.: Increase in capacity of multiuser OFDM system using dynamic
subchannel allocation. In: Proceedings of the IEEE 51st Vehicular Technology Conference,
2000 (VTC 2000-Spring), Tokyo, pp. 1085–1089 (2000)
2. Morelli, M., Kuo, C.-C., Pun, M.-O.: Synchronization techniques for orthogonal frequency
division multiple access (OFDMA): a tutorial review. Proc. IEEE 95, 1394–1427 (2007)
3. Xu, J., Kim, J., Paik, W., Seo, J.-S.: Adaptive resource allocation algorithm with fairness for
MIMO-OFDMA system. In: Proceedings of the IEEE 63rd Vehicular Technology
Conference, 2006 (VTC 2006-Spring), pp. 1585–1589 (2006)
4. Zheng, L., Tse, D.N.C.: Diversity and multiplexing: a fundamental tradeoff in multiple-
antenna channels. IEEE Trans. Inf. Theory 49, 1073–1096 (2003)
5. Idris, A., Syed Yusof, S.K.: Performance evaluation of intercarrier interference self-
cancellation schemes for space times frequency block codes MIMO-OFDM system. IEEE J.
(2008)
6. Love, D.J., Heath Jr, R.W.: Diversity performance of precoded orthogonal space-time block
codes using limited feedback. IEEE Commun. Lett. 8, 305–307 (2004)
7. Idris, A., Dimyati, K., Syed Yusof, S.K., Ali, D.: Diversity technique for multiple input
multiple output orthogonal frequency division multiplexing (MIMO-OFDM) system using a
new subcarrier mapping scheme. Int. J. Phys. Sci. 6(16), 3879–3884 (2012)
8. Idris, A., Jais, S.M., Salih, N.M., Yusof, A.L., Ali, D.M.: Performance evaluation of STFBC
MIMO-OFDM system using different equalization algorithm. J. Eng. Technol. 2(2), 62–67
(2012). ISSN 2231-8798
9. Idris, A., Abdullah, N., Ali, D., Yaacob, N., Mohamad, H.: Reduction of interference with
linear equalizer using quarter subcarrier mapping scheme. Paper presented at the IEEE
Symposium on Wireless Technology and Applications (ISWTA), (2013)
10. Morelli, M.: Timing and frequency synchronization for the uplink of an OFDMA system.
IEEE Trans. Commun. 52, 296–306 (2004)
11. Mark, J., Zhuang, A.W.: Wireless Communications and Networking. Prentice Hall, New
Jersey (2003)
12. Sharma, P.: Performance analysis of zero-forcing equalizer for ISI reduction in wireless
channels. Int. J. Eng. Res. Technol. (IJERT) 1, 3 (2012)
13. Gupta, M., Nema, R., Mishra, R.S., Gour, P.: Bit error rate performance in OFDM system
using MMSE & MLSE equalizer over Rayleigh fading channel through the BPSK, QPSK, 4
QAM & 16 QAM modulation technique
14. Rajesh Goel M.S., Bansal, P.K.: On importance of bit error rate in wireless communication
15. Wang, F., Ghosh, A., Sankaran, C., Fleming, P., Hsieh, F., Benes, S.: Mobile WiMAX
systems: performance and evolution. IEEE Commun. Mag. 46, 41–49 (2008)
Chapter 26
Performance Analysis of Polling Delay
in Transparent and Non-transparent
Multi-hop Relay WiMAX Network

Mohd Daud A. Hassan, Habibah Hashim and D.M. Ali

Abstract The relay task group has extended the IEEE 802.16e-2005 to a new
standard known as IEEE 802.16j Mobile Multi-hop Relay (MMR). The deploy-
ment of relay station (RS) in WiMAX network will overcome the increase in cost
and improve the economic viability of the system. Furthermore, the MMR stan-
dard also addresses the problem of limited spectrum, low SINR at the cell edge and
coverage hole of the previous standard. However, the introduction of relay may
cause longer delays and degrade the performance of resource efficiency. In this
paper, we investigate the performance analysis of polling delay for transparent and
non-transparent relay modes in MMR WiMAX. We derived and analyzed the
average polling delay of both modes using M/M/1 and tandem queue respectively.
Numerical result has shown that the average polling delay decreases when the
number of poll increases for both relay modes. Moreover, the delay for non-
transparent relay mode is higher due to the signals overhead.

26.1 Introduction

The IEEE 802.16j standard was introduced to support multi-hop operation and
relay station (RS) relay packets between a base station (BS) and subscriber station/
mobile station (SS/MS). In mobile multi-hop relay (MMR) WiMAX system, the

M.D.A. Hassan (&)

Faculty of Electrical Engineering, Universiti Teknologi MARA, 13500 Penang, Malaysia
e-mail: [email protected]
H. Hashim  D.M. Ali
Faculty of Electrical Engineering, Universiti Teknologi MARA, 40450 Shah Alam,
Malaysia
e-mail: [email protected]
D.M. Ali
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 251

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_26
252 M.D.A. Hassan et al.

framing information of MS is allowed to route through intermediate RSs to reach

the BS, which differs from the single hop. There are benefits of the introduction of
relay in WiMAX network as compared to conventional repeater in a cellular
system such as easy network deployment, and reduction in infrastructure cost. A
relay does not need any connection to the access network wired backhaul hence
can be deployed to a location where the wired connection does not exist or when it
is complicated to install. Furthermore, throughput and capacity enhancement can
be achieved. The relay that is used in the IEEE 802.16j standard is also inad-
vertently compatible with the IEEE 802.16e for SS/MS.
The history of IEEE 802.16j was created in March 2006 [1] and the first
technical contribution was made 8 months later. In order to support mobile multi-
hop relay specification, mesh mode was removed from the IEEE 802.16 -2009
standard. The specification is an amendment of the IEEE 802.16e standard for
achieving throughput enhancement and coverage extension.
A study on the performance of the IEEE 802.16j network has been conducted
from various aspects. An overview of the relay based technology which focuses on
modes of operation, framing structure and network entry procedure is described in
[2]. In [3], the paper investigates the impact of the location of the RS and the
possibility of RS to increase the network capacity. In addition, the author analyzed
the signaling that arises within such system and compared it with a single hop
implementation. In [4], an adaptive polling approach is proposed for the real-time
Polling System (rtPS) in order to improve on the average delay. However, the
author has performed the simulation in a standard conventional WiMAX envi-
ronment without considering the multi-hop relay. The performance analysis of
both relay modes in MMR is demonstrated in [5], and yet the analysis is limited to
sustain the rate of uplink transmission. The adaptation to the traffic pattern was
conducted by Nie et al. [6] which satisfied the delay constraints of most real-time
applications. However this approach only performed the multi-hop bandwidth
request mechanism in terms of overall spectrum efficiency. In [7], the authors
investigated the usage of RS in improving the per-user throughput. The investi-
gation of the Non-Transparent relay mode has also been discussed in [8] and the
simulation results have shown an improvement in the transmission delay. To the
best of the author‘s knowledge, there is no current research that specifically
examines the queuing analysis of polling based on multi-hop WiMAX for both
relay operation modes.
We present the analysis of the polling based on MMR WiMAX using Markov
chain which focuses on the delay centralized scheduling for transparent and non-
transparent modes. We investigated the polling delay affected by the numbers of
poll (NOP) and the total numbers of MS and only considered unicast polling mode
in the analysis.
26 Performance Analysis of Polling Delay … 253

26.2 Relay Technologies

The standard of IEEE 802.16j, defines 2 operation modes of RS which differ in

terms of the usage. These modes are known as the transparent and non-transparent
relay modes. The difference between both modes is basically on how the framing
information is transmitted. Generally, in transparent mode operation, the relay
does not transmit/forward frame header information. The frame header contains
essential scheduling information (such as preamble, frame control header (FCH),
and media access protocol (MAP) and the nodes used to determine when they can
transmit and receive information [2]. However in the non-transparent mode the RS
generates their own framing information and forward those provided by the BS
(depending on scheduling approach).
There are also two different modes of resource allocation which are centralized
and distributed. For centralized, BS determines the scheduler and the scheduling
packets are transmitted in a collision freeway within the scheduling control sub-
frame and the transparent relay mode operates using this type of scheduling. While
in distributed mode, RSs have some autonomy and can make scheduling decision
for the nodes which they communicate with. Thus, the non-transparent relay has a
choice between distributed or centralized modes to operate based on the sched-
uling approach.
The frame structure of multi-hop relay consists of downlink (DL) and uplink
(UL) sub-frame, where it includes at least one access zone (AZ) and may include
one (transparent zone, TZ) or more relay zone (RZ). This structure enables RS to
operate in either transmit or receive mode. For the polling purpose, BS polls
individual MSs by using UL-MAP in the DL sub-frame.
Figure 26.1 shows the frame structure of a two-hop transparent relay which
contains the DL sub-frame (AZ and TZ) on the left side, and UL sub-frame
(AZ and RZ) on right side of the frame. The RS serving can decode the control
information of the BS and is not required to transmit control information them-
selves. This type of serving can be utilized to decode the central information in
order to achieve higher throughput. Figure 26.2 shows the non-transparent relay
frame structure which for the non-transparent mode contains DL sub-frame
(AZ and RZ), and UL sub-frame (AZ and RZ). Meanwhile, due to the fact that the
relay serving MSs cannot decode the BS control information, they must transmit
control information themselves.
In transparent relay mode, AZ for DL sub-frame, BS sent out data burst to RS
and MS, respectively. Similarly in TZ, RS forward data burst received from BS to
MSs. The DL sub-frame starts with preamble which is used by the PHY layer for
synchronization and equalization followed by the frame control and data section.
The UL transmission sub-layer frame operation is identical to that of DL. The
description of the frame structure in transparent and non-transparent relay can be
referred in [9]. We focus on the UL sub-frame which is divided into 3 parts: initial
ranging, bandwidth request and data transmission period [10].
254 M.D.A. Hassan et al.

Fig. 26.1 A transparent frames structure for both the BS and RS

Fig. 26.2 A non-transparent frames structure for the BS and RS

Assumed that for zone of UL (AZ and RZ) are combined as depicted in
Fig. 26.3. In standard WiMAX network, each MS that has data to be sent out must
request for bandwidth which is provided by the BS. The MS will request the
bandwidth based on the corresponding service class (e.g., data, voice, video or real
time services) by replying to the polling message from the BS before the data
uplink transmission. The BS will accept the bandwidth request if resources are
sufficient enough. The information about polling control is stored in the UL-MAP
within the downlink sub-frame which includes uplink channel identifier, uplink
channel descriptor, number of information elements to map, allocation start time
and map information elements. Therefore in multi-hop relay WiMAX network
polling plays an important role to accomplish an efficient performance particularly
those that involve the transmission of data and video.
26 Performance Analysis of Polling Delay … 255

Fig. 26.3 Generic frame structure

The analysis uses the following notations;

Tf frame duration time
TfBS base station frame duration time
TfRS relay station frame duration time
TDL downlink sub-frame duration time
TUL uplink sub-frame duration time
TIR initial ranging period for uplink sub-frame
TBW bandwidth request period for uplink sub-frame
TP polling period for MSs to send bandwidth request
N1 and N2 numbers of MS in the two system in the steady state
Tave average polling delay
R data rate
Pt packet transmission probability
/ no. of polls within a frame period

26.3 Network Model Analysis

This part presents the analysis of a transparent and non-transparent polling for
centralized scheduling. The number of MSs served in the network is the parameter
that affects the network performance. As afore mentioned, the signal from BS are
not forwarded to the RS in the transparent mode. Thus, the UL-MAP will be
transmitted directly to MS. We model the polling based of the multi-hop relay
WIMAX network using the Markov chain model as in [10], where the number of
served MSs is the state of model. The analysis of the model is a birth-death process
as shown in Fig. 26.4, with a birth rate of kn and death rate of ln.
By using the global balance equations, derived for steady-state solution of an
irreducible, homogeneous continuous Markov chain;
256 M.D.A. Hassan et al.

Fig. 26.4 State transition diagram of the number of MSs in the MMR network

X
pk qkj  qj pj ¼ 0 ð26:1Þ
k6¼j

and combine with the equation

X
pj ¼ 1 ð26:2Þ
j

where pj is the steady-state limiting probability of the system being in state j and
qkj and qj are the transition rates. Then applying this to birth-death
 
 kj þ lj pj þ pjþ1 ljþ1 þ pj1 kj1 ¼ 0 ð26:3Þ

k0 p0 þ l1 p1 ¼ 0 ð26:4Þ

Equation (26.4) is the balance equation derived for the birth-death as illustrated in
Fig. 26.4. Assumed that the arrival rate is the same and denoted as k and the
service rate is also the same and denoted as l. The number of MSs in a system of
an M/M/1 queue is a homogeneous, irreducible birth-death in which ki = k for 8i
and li = l for 8i . Furthermore, traffic intensity of q = k/l and q \ 1 is defined
for a stable system. The mean number of MSs in the system in the steady-state

X
1
E½ N  ¼ npn
n¼0
X1
¼ nqn ð1  qÞ
n¼0
X
1
¼ qð1  qÞ nqn1 ð26:5Þ
n¼0
1
¼ qð1  qÞ
ð1  qÞ2
q
¼
ð1  q Þ

The average polling delay of multi-hop WiMAX depends on the number of polls
(NOP), ;. Every MS is polled by the BS in order for the MS to send its packet
26 Performance Analysis of Polling Delay … 257

which is then, mapped in the UL-MAP. Subsequently after receiving the poll, MS
will request for bandwidth through the slot in the uplink sub-frame. Thus, the
average polling delay of MMR WiMAX for transparent mode is

1
Tave ¼ Tf   E½N 
;
1
¼ ðTDL þ TUL Þ   E½ N  ð26:6Þ
;
 
1 q
¼ ðTDL þ TIR þ TBW þ ; Tp þ TdataÞ  
; 1q

For the non-transparent relay mode, the signal received from BS is forwarded by
the RS to MS. Therefore we model the polling based of a multi-hop relay WiMAX
as a two-stage tandem queue in queuing system. This tandem refers to an
arrangement of objects in which they are lined up one behind the other, all facing
the same direction and multiple job classes, one after another and an arriving
packet undergoes each job class before leaving the system [11]. Figure 26.5
depicts the two stage queue in a tandem system.
The arrival of the first stage is define as k while in the second stages, it‘s arrival
is the of the departure of the MS from the first stage. The state of the first stage
(N1 = n1) is independent from the previous sequence of departure and it is these
departure that defines the arrivals to the second system. Therefore if N1 and N2
represent the numbers of served MSs in the two systems in then steady state

pðn1 ; n2 Þ ¼ PrðN1 ¼ n1 ; N2 ¼ n2 Þ ¼ PrðN1 ¼ n1 Þ PrðN2 ¼ n2 Þ ð26:7Þ

thus from result for the M/M/1 queue,

pn ¼ ð1  qÞqn
ð26:8Þ
pðn1 ; n2 Þ ¼ ð1  q1 Þqn1 1 ð1  q2 Þqn2 2 ; n1 ; n2  0

where q1 ¼ k=l1 and q2 ¼ k=l2 and 0 \ q1 ; q2 \ 1

by summing over n1, obtained that
q2
PrðN2 ¼ n2 Þ ¼ ð1  q2 Þqn2 2 and E½N2  ¼ ð26:9Þ
1  q2

Similarly,
q1
PrðN1 ¼ n1 Þ ¼ ð1  q1 Þqn1 1 and E½N1  ¼ ð26:10Þ
1  q1

and then E [N1 + N2]

258 M.D.A. Hassan et al.

Fig. 26.5 Two stages Markovian queue in tandem

XX
¼ ðn1 þ n2 Þð1  q1 Þqn11 ð1  q2 Þqn22
n1 n2
X X X X
¼ n1 ð1  q1 Þqn11 : ð1  q2 Þqn22 þ n2 ð1  q2 Þqn22 : ð1  q1 Þqn11
n1 n2 n2 n1
q1 q2
¼ þ
1  q1 1  q 2
¼ E½N1  þ E½N2 
ð26:11Þ

As described in Sect. 26.2, the RS received the signal from the BS and forward
it to MS. Therefore the equation is derived based on the operation of the trans-
parent mode by doubling the time frame to represent the two hops. Thus, the
equation for an average polling delay for a non-transparent network is

1
Tave ¼ Tf   E½N1 þ N2 
;
  ð26:12Þ
1 q1 q2
¼ ðTfBS þ TfRS Þ   þ
; 1  q 1 1  q2

Both average polling delays show the inverse to the number of polls. This will
result in, reduced in the average polling delay when the number of polls in the
system is increased. However, it will increase the bandwidth waste of polling if the
MS has no packet to send.

26.4 Numerical Result

The number of served MS in MMR WiMAX is a part of the important parameter

that affects the system. The BS polls the MS in every frame period and this will
cause an increase in polling delay.
In this analysis, the MSs are assumed to arrive at the MMR WiMAX network
using Poisson distribution with arrival rate of k = 0.3. The ratio for the DL:UL is
2:1. Table 26.1 summarizes the parameters for the analysis. Figures 26.6, 26.7, and
26.8 illustrates the result for MMR WiMAX based on the M/M/1 and tandem queue.
Figure 26.6 shows the average polling delay for transparent and non-transparent
mode when the numbers of polls (NOP) (ranging from 1 to 11) are varied.
26 Performance Analysis of Polling Delay … 259

Table 26.1 Parameter for analysis

Parameter Values
Number of polls (NOP), ; 1 * 20
Numbers of mobile station (MS) 1 * 100
Frame periods, Tf 5 ms
Data rates, R 50 Mbps
Total number of slots in frame, Lf 630
Number of slots in DL sub-frame, LDLf 420
Number of slots in UL sub-frame, LULf 210
Polling bandwidth slots 3
Required bandwidth slots 32
MS mean service time, l 0.5

Fig. 26.6 Average polling 1.2

delay versus no. of polls
1.0 Transparent mode
Non-transparent mode
Average polling delay

0.8

0.6

0.4

0.2

0.0

0 2 4 6 8 10 12
No. of polls

Fig. 26.7 Average polling 0.30

delay versus no. of MS (for NOP=1
transparent) 0.25
Average polling delay

NOP=3
NOP=5
NOP=7
0.20 NOP=9

0.15

0.10

0.05

0.00

0 20 40 60 80 100 120
No. of MSs
260 M.D.A. Hassan et al.

Fig. 26.8 Average polling 1.2

delay versus no. MS (for non- NOP=1
transparent) 1.0 NOP=3
NOP=5

Average polling delay

NOP=7
0.8 NOP=9

0.6

0.4

0.2

0.0

0 20 40 60 80 100 120
No. of MSs

The average polling delay shows a decrease trend when the NOP increases for both
modes. The transparent mode has less delay than the non-transparent one. This is
because the decision is only done by the BS whereas the non-transparent mode has
to undergo two decisions which are BS and RS. The average polling delay of
transparent mode and non-transparent mode with increasing number of MS are
shown in Figs. 26.7 and 26.8 respectively. The delay will vary in line with the
increased of the NOP and the no. of MS. When NOP is increased, the average
polling delay for the transparent mode will decrease and the same goes for non-
transparent mode.
However, the delay for the non-transparent mode is higher than the transparent
mode due to the signaling messages received by the RS from the BS which is then
forwarded to the MS. This is proven by Eq. (26.12). Although the average polling
delay increases with the number of MS, it decreased significantly for both modes
when the NOP is increased from 1 to 3. For instance, when MS = 100, the average
polling delay decreased approximately to 70 %. However there is a slight changes
to the average polling delay when the NOP is increased.
This paper presents a queuing model to evaluate the delay in the polling based
operation of MMR WiMAX IEEE802.16j network. The average polling delay for
both modes show the inversed to the number of polls. As a result, the polling delay
is reduced if more polls in the system are created. However, this will result in
waste of bandwidth if the MS has no packet to send. For future direction we will
validate the mathematical analysis derive with the results of the simulation. We
will also focus on optimizing the number of polls by the BS in the network while
reducing the average polling delay and increasing the throughput of the system.

Acknowledgments This research was supported by the Universiti Teknologi MARA (UiTM)
under Exploratory Research Grant Scheme (ERGS).
26 Performance Analysis of Polling Delay … 261

References

1. Peters, S.W., Heath, R.W.: The future of WiMAX: multihop relaying with IEEE 802.16j.
IEEE Commun. Mag. 47(1), 104–111 (2009)
2. Genc, V., Murphy, S., Yu, Y., Murphy, J.: IEEE 802.16j relay-based wireless access
networks: an overview. IEEE wirel. Commun. 15(5), 56–63 (2008)
3. Genc, V., Murphy, S., Murphy, J.: Performance analysis of transparent relays in 802.16j
MMR networks. In: 6th International Symposium on Modeling and Optimization in Mobile,
Adhoc and Wireless Network and Workshops, pp. 273–281. (2008)
4. Chang, B.J., Chou, C.M.: Adaptive polling algorithm for reducing polling delay and
increasing utilization for high density subscribers in WiMAX wireless networks. In: 10th
IEEE Singapore International Conference on Communication Systems, pp. 1–5. (2006)
5. Yusoff, R., Baba, M.D., Rahman, R.A., Ibrahim, M., Mat Isa, N.: Performance analysis of
transparent and non-transparent relays in MMR WiMAX networks, In: IEEE Symposium on
Industrial Electronics and Application (ISIEA), pp. 237–240. (2011)
6. Nie, C., Korakis, T., Panwar, S.: A multi-hop polling service with bandwidth request
aggregation in IEEE 802.16j networks, In: IEEE Conference on Vehicular Technology,
pp. 2172–2176. (2008)
7. Izza, W.N., Baba, M.D., Ali, D.M.: Performance study on relay station usage in IEEE 802.16j
mobile multi-hop relay network. In: IEEE Symposium on Computer Application and
Industrial Electronics (ISCAIE), pp. 218–223. (2012)
8. Sayenko, A., Alanen, O., Martikainen, H.: Analysis of the non-transparent in-band relays in
the IEEE 802.16 multi-hop system. In: IEEE Wireless Communication and Networking
Conference, pp. 1–6. (2010)
9. Mohd Daud, A.H., Habibah, H., Darmawaty M.A., NurHidayat A.: A queueing analysis of
polling based for mobile multihop relay WiMAX networks. In: IEEE International
Conference on Computational Intelligence, Communication Systems and Networks
(CICSyN), pp. 322–327. (2013)
10. Chang, B.J., Chou, C.M., Liang, Y.H.: Markov chain analysis of uplink subframe in polling-
based WiMAX networks. Comput. Commun. J. 31(10), 2381–2390 (2008)
11. Ng, C.H., Soong B.H.: Queueing Modeling Fundamentals with Application in
Communication Network, 2nd edn. Wiley, New York (2008)
Chapter 27
Blind Source Computer Device
Identification from Recorded Calls

Mehdi Jahanirad, Ainuddin Wahid Abdul Wahab

and Nor Badrul Anuar

Abstract This study investigates the use of blind source computer device iden-
tification for forensic investigation of the recorded VoIP call. It was found that a
combination of mel-frequency cepstrum coefficients (MFCCs) and entropy as an
intrinsic audio feature captures the specific frequency response due to the tolerance
in the nominal values of the electronic components associated to individual
computer device. By applying the supervised learning techniques such as naïve
Bayesian, linear logistic regression, neural networks (NN), support vector
machines (SVM) and sequential minimal optimization (SMO) classifier to the
Entropy-MFCC features, state-of-the-art identification accuracy of above 99.8 %
has been achieved on a set of 5 iMacs and 5 desktop PCs from the same model.

27.1 Introduction

Audio forensics mostly focus on situations that impose trust in authenticity and
integrity of audio signals. An example for these scenarios is forensic acquisition,
analysis and evaluation of admissible audio recording as crime evidence in court.
Authenticity of audio evidence is important as part of a civil and criminal law
enforcement investigation or as part of an official inquiry into an accident or other
civil incidents. In these processes, authenticity analysis determines whether the
recorded information is original, contains alterations, or has discontinuities

M. Jahanirad (&)  A.W.A. Wahab  N.B. Anuar

Faculty of Computer Science and Information Technology, University of Malaya,
50603 Kuala Lumpur, Malaysia
e-mail: [email protected]
A.W.A. Wahab
e-mail: [email protected]
N.B. Anuar
e-mail: [email protected]

Ó Springer International Publishing Switzerland 2015 263

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_27
264 M. Jahanirad et al.

attributed to recorder stops and starts. Authenticity evaluations by using the

device-based techniques are attracting widespread interest in fields such as: (a)
identification of computer-generated audio from the original audio recording file
[1]; (b) identification of the source brand or model of the recording devices, such
as telephone handsets, microphones [2, 3], and cell phones [4]; (c) identification of
the speech codecs [5]. The most related works to this approach are the recording
source forensics.
Kraetzer et al. [6] published the first practical evaluation to identify a device.
They adopted the statistical pattern recognition technique to determine the source
microphone and its recording environments. Buchholz et al. [7] focused on
microphone classification through the histogram of Fourier coefficients and
extracted the coefficients from near-silent frames to capture microphone proper-
ties. With the use of a suitable context model for microphone forensics, Kraetzer
et al. [8] extended the works in [6] by reevaluating their sample data separately.
Kraetzer et al. [9] extended the proposed context model in [8] toward better
generalization and constructed a new application scenario model for microphone
forensic investigations with the aim of detecting playback recordings. In addition
to microphone identification, Garcia–Romero and Espy–Wilson [2] extended
device identification by identifying landline telephone handsets. A similar landline
telephone handset identification method proposed by Panagakis and Kotropoulos
[3] improved the accuracy of the identification by using sparse representation
classifier (SRC). Hanilçi et al. [4] proposed a cell phone identification method that
uses SVM by identifying the audio source based on the cell phone brand and
model. However, these approaches lack sufficient study to eliminate convolution
by speech context. Moreover, they only focused on identifying source recording
devices.
This paper focuses on the novel idea of recognizing the communicating
acquisition devices based on the recorded call. We present a case study that
records calls by using stationary Notebook that makes VoIP calls to computer
devices of the same model. The motivation for this case study is based on the fact
that the combination of anonymity, ease of access and free offerings of VoIP
services provide fertile ground for criminal activity. Thus, identifying the com-
puter devices used in VoIP call can help the forensic investigator to reveal useful
information in addition with authenticity of the recorded call. This study identifies
source brand and model of the computer devices from recorded calls based on
entropy-mel-frequency cepstrum coefficient (MFCC) features. Extracting the
entropy of MFCCs adds an advantage to the application of MFCCs in [4] by
eliminating the effects of speech contents. The method evaluates the feasibility of
entropy-MFCC features and its robustness against speech signals by using clas-
sifier benchmarking.
The remainder of this paper is organized as follows: Sect. 27.2 discusses an
overview of the methodology. Section 27.3 outlines the recording setup. Sec-
tion 27.4 describes the experiments and evaluates the performance of the proposed
method. Finally, Sect. 27.5 discusses further implications of the practical study, its
limitations, and future applications.
27 Blind Source Computer Device Identification from Recorded Calls 265

27.2 Source Computer Device Identification Scheme

In our implementation we followed [10] by using audio mining techniques: (a)

creates blocks through preprocessing the recorded samples, (b) determines
intrinsic computer device fingerprint through feature extraction and (c) uses
supervised learning techniques known as classification.

27.2.1 Preprocessing

Preprocessing method uses two different approaches as illustrated by Fig. 27.1 to

create blocks from: (a) original speech signal, and (b) near-silent segments. This is
to justify the robustness of the proposed method against speech signals.
Original Speech Signal. The pre-processing stage includes sampling, framing,
windowing and cleaning the signal. The signals become more distinct when noise
is eliminated. Thus, we adopted cleaning to remove the noise generated by
environment. The algorithm splits clean signals into overlapping audio frames of
length 40 samples. In other words the output is a matrix with 40 columns, when
each column represents one block for feature extraction.

Sampling
WAV Files
Audio Frames Audio Frames

Create Blocks from Create Blocks from

Original Signal Near-Silent Segments

Clean Audio Audio

Near-Silent
Cleaning Windowing Frames/ Block Windowing Frames/ Block Cleaning
Detection
Window Window

Clean Audio Frames

Compute MFCCs

Windowing [FFT] Filterbank INV[DCT]

MFCC Output

Entropy/
Normalising
window

Fig. 27.1 Flow chart of the proposed feature extraction

266 M. Jahanirad et al.

Short time 0.2 Short time energy (original)

energy Short time energy (filtered)
0.15
0.1
0.05

0 500 1000 1500 2000 2500 3000 3500 4000

Frequency

Spectral Centroid (original)

Centroid
Spectral

0.6
Spectral Centroid (filtered)
0.4
0.2
0
0 500 1000 1500 2000 2500 3000 3500 4000
Frequency
Amplitude

0.2
0
-0.2

0 2 4 6 8 10 12 14 16 18 20
Time(ms)

Fig. 27.2 Visualization of Near-Silent Detection algorithm. Top and middle plots represent the
histogram of short time energy and spectral centroids of the signal, respectively. The horizontal
line shows the estimated threshold. The bottom plot shows the spectrum of the original signal,
when red color determines the near-silent segments

Near-Silent Segments. This approach uses simple segmentation of the recor-

ded signal in order to extract the near-silent segments as demonstrated in Fig. 27.2.
We implemented the near-silent detection algorithm based on the silent removal
approach in [11], however our objective is to select the silent segments instead of
removing them. This method uses two audio features known as signal energy and
spectral centroid [12]. This algorithm includes following steps: (a) extracts two
feature sets from the original recorded signal, (b) computes the histogram of the
feature series values, (c) applies the smoothing filter on histograms, (d) estimates
two thresholds for each histogram, (e) applies a simple thresholding criterion to the
feature sets, (f) and detects silent segments based on the assigned criterion. Finally,
the algorithm assigns equal number of samples from near silent segments into
blocks, and then performs cleaning on each block prior to feature extraction.

27.2.2 Intrinsic Computer Device Fingerprints

Assuming that the computer device is a linear time-invariant system, its influence
on the recorded call is modeled by the convolution of its impulse response and the
original speech. The convolution means the spectrum of any recorded speech
segment is the product of the spectrum of the original speech signal and the device
frequency response. Well known MFCC features are selected to capture the device
frequency response because of the fact that the convolution in time domain is
27 Blind Source Computer Device Identification from Recorded Calls 267

represented by the summation in cepstrum domain. Thus, MFCCs produce

inherent invariance toward linear spectral distortions. Furthermore, the entropy of
MFCCs reduces the dimensionality of the feature space. According to information
theory, this value increases for silent segments that contain uncertainty and
reduces for the speech segments that contain information [13]. As a result this
paper computes Entropy-MFCC features as the intrinsic computer device finger-
prints through three stages: (a) computes the MFCCs, (b) computes the entropy of
MFCCs, and (c) normalizes the output.
MFCCs. The complex cepstrum of the signal is defined as the Fourier trans-
form of the log of the signal spectrum. The signal was transferred to a linear
frequency scale by using fast Fourier transform and was converted to the mel
frequency scale by using a filter bank. Finally, the MFCCs were determined based
on the analysis of the short-time mel frequency log spectrum. This analysis
includes computing the inverse discrete cosine transform (DCT) of the log spec-
trum of the signal. The mel-cepstrum output consists of N frames and 12
coefficients.
Entropy. The entropy of the MFCC vectors was computed in two stages. First,
the spectrum was normalized into the probability mass function (PMF), where l cd
is the mel-cepstrum coefficient d in frame l and pðl cd Þ ¼l pd is the PMF of the
signal. In the second stage, the entropy H was computed by using

X
N
Hd ¼  l pd log2 l pd : ð27:1Þ
l¼1

Overall, 12 entropy-MFCC features were extracted by using MATLAB

functions.
Normalization. The last and one of the most important steps in preparing the
features is normalization. This step reduces large differences between the maxi-
mum and minimum data values.

27.2.3 Supervised Learning Techniques

Supervised learning techniques known as classification problems can be imple-

mented through different learning algorithms, such as statistical modeling, linear,
non-linear, and ensemble learning models. However, to determine which approach
is the most efficient for a particular problem, evaluation metrics are required. We
selected five simple classifiers based on the fact that the simplicity-first method-
ology is the best choice for analyzing practical datasets [14]. Furthermore, these
algorithms demonstrated high performances for similar works in the literature.
Table 27.1 details the employed classifiers in the experiment and their
specifications.
268 M. Jahanirad et al.

Table 27.1 Classification algorithms used in the experiment

Classification Specifications
algorithms
Naïve Bayesian Works based on Bayes’ rule of conditional probability that assumes
independence
Linear logistic Measures goodness of fit by using the log-likelihood of the model
regression
Neural network Primarily learns the network structure and the connection weights by
fixing the network structure to determine the weights
Support vector Uses the LibSVM wrapper that implement a multi-class SVM
machine classifier with a radial basis function (RBF) kernel [15]
Sequential minimal- Implements the sequential minimal-optimization algorithm for
optimization training a SVM classifier using a Polynomial kernel

27.3 Recording Setup

The implemented setup enabled recording of VoIP communication between

computer devices and the single stationary inside faculty building. The stationary
user made Skype call to computer user, then recorded signals in mono with WAV
format by using Pamela for Skype-Version 4.8 [16]. The setup collected record-
ings with respect to computer devices including: (a) five iMacs of identical model
located in Multimedia Research Lab, (b) five desktop PCs of identical model
located in Micro Lab. At the same time the stationary was located inside the open
corridor. All conversations were conducted between the same male and female
over the experiment. Table 27.2 indicates the specifications for computer devices
that employed in the setup. All iMacs were 21.5-inch, Late 2012 model built with
3.1GHz Intel Core i7 processor, stereo speakers, and dual microphones. All
desktop PCs were Lenovo ThinkCentre M81 7518 model with 3.3 GHz Intel Core
i3 processor that attached to the same external microphone. The desktop PCs use
32 bit, Windows 7 professional Service Pack 1 as operating system.

Table 27.2 iMac and desktop PC devices and class names

iMac Desktop PC
Serial no. Software no. Class Serial no. Class
C02JWXXJDXXX OS X 10.8.4 (12E55) i1 XXXXXXXXXXXX3D6B PC1
C02JWXXDDXXX OS X 10.8.5 (12F45) i2 XXXXXXXXXXXX3DMK PC2
C02JWXXGDXXX OS X 10.8.2 i3 XXXXXXXXXXXX3AME PC3
(12C3103)
C02JWXXEDXXX OS X 10.8.3 (12D78) i4 XXXXXXXXXXXX3CGM PC4
C02JWXXKDXXX OS X 10.8.2 i5 XXXXXXXXXXXX2ZRB PC5
(12C3103)
27 Blind Source Computer Device Identification from Recorded Calls 269

27.4 Experiments and Results

The experiments evaluated the feasibility of the source computer device identifi-
cation method through features that extracted from original speech signal against
features that extracted from near silent segments. Moreover the performance of the
proposed scheme was evaluated for inter and intra-model device identification
using five classification algorithms that implemented in data mining tool Weka
Version 3.6 [17].

27.4.1 Experiment on Original Speech Sample

The total of 1,320 and 1,400 blocks were extracted from original speech samples
collected from each iMac and desktop PC devices, respectively. The first part of
the experiment used the histogram of all 12 features to visualize the discrimina-
tory effect of entropy-MFCC features among five identical iMacs and desktop PCs,
as demonstrated in Fig. 27.3. For further investigation we selected three different
pair of iMacs and examined the squared Euclidean distance between their entropy-
MFCC feature vectors in Fig. 27.4. The result of this measurement indicates the
considerable distances between feature vectors corresponding to pair of iMacs,
therefore justifying the effectiveness of entropy-MFCC features in differentiating
individual iMac devices. The second part of the experiment employed classifica-
tion benchmarking to evaluate the classification performance among both desktop
PCs and iMac devices of identical models.
Table 27.3 indicates the results of three sets of experiments on original speech
signal with all five classifiers, 10-fold cross-validation and default parameters. For

1 1
entropy-MFCC Values

1
entropy-MFCC Values

entropy-MFCC Values

0.8 0.8 0.8

0.6 0.6 0.6

0.4 0.4 0.4

0.2 0.2 0.2 PC3

PC1 PC2
iMac1 iMac2 iMac3
0 0 0
0 2 4 6 8 10 12 0 2 4 6 8 10 12 0 2 4 6 8 10 12
entropy-MFCC Index entropy-MFCC Index entropy-MFCC Index

1
entropy-MFCC Values

entropy-MFCC Values

1
0.8 0.8
0.6 0.6
0.4 0.4

0.2 0.2 PC5

PC4
iMac4 iMac5
0 0
0 2 4 6 8 10 12 0 2 4 6 8 10 12
entropy-MFCC Index entropy-MFCC Index

Fig. 27.3 Histogram of entropy-MFCC features for each iMac and desktop PC devices
270 M. Jahanirad et al.

iMac1-iMac2 iMac2-iMac3 iMac3-iMac4

30 25 35
Silent Silent Silent
Original Original Original

Average squered Euclidean distance

30
Average squered Euclidean distance

Average squered Euclidean distance

25
20
25
20
15
20
15
15
10
10
10
5
5
5

0 0 0
0 2 4 6 8 10 12 0 2 4 6 8 10 12 0 2 4 6 8 10 12
entropy-MFCC Index entropy-MFCC Index entropy-MFCC Index

Fig. 27.4 Average squared Euclidean distances of each entropy-MFCC features on three
different iMac pairs

Table 27.3 Performance on intra-model identification using original speech signal

Intra-model identification based on iMacs
Classification algorithms TCB FCB RMSE ACC (%) Elapsed time (s)
Naïve Bayesian 6,600 0 0 100 2.6
Linear logistic regression 6,588 12 0.0218 99.82 35.1
NN 6,591 9 0.0202 99.86 72.7
LibSVM 6,593 7 0.0206 99.89 5.5
SMO 6,596 4 0.3157 99.94 41.5
Intra-model identification based on desktop PCs
Naïve Bayesian 7,000 0 0 100 0.51
Linear logistic regression 6,983 17 0.029 99.76 36.53
NN 6,987 13 0.0224 99.81 140.81
LibSVM 6,993 7 0.02 99.90 6.0
SMO 6,989 11 0.3163 99.84 1.18
Source brand/model Identification Based on all computer devices
Naïve Bayesian 13,600 0 0 100 1.11
Linear Logistic Regression 13,567 33 0.0194 99.76 146.63
NN 13,570 30 0.0183 99.78 450.22
LibSVM 13,579 21 0.0176 99.85 15.81
SMO 13,575 25 0.2719 99.82 3.03
27 Blind Source Computer Device Identification from Recorded Calls 271

Table 27.4 Confusion matrix of linear logistic regression classifier based on iMacs
ACC = 99.82 % Predicted
i1 i2 i3 i4 i5
Actual i1 1,318 0 0 1 1
i2 1 1,316 1 2 0
i3 1 1 1,318 0 0
i4 1 1 1 1,317 0
i5 0 1 0 0 1,319

Table 27.5 Confusion matrix of linear logistic regression classifier based on PCs
ACC = 99.76 % Predicted
PC1 PC2 PC3 PC4 PC5
Actual PC1 1,397 2 0 1 1
PC2 1 1,396 0 3 0
PC3 0 0 1,400 0 0
PC4 1 3 0 1,395 1
PC5 2 0 0 3 1,395

intra-model identification based on iMacs, the overall results proved the feasibility
of the entropy-MFCC features with high classification accuracies (ACC)
(99.82–100 %) and minimal root mean square error (RMSE). Naïve Bayesian
classifier performed higher classification accuracy and computational efficiency.
SMO classifier outperformed LibSVM classifier with respect to classification
accuracy, but produced lager RMSE. Table 27.4 shows the confusion matrix for
the linear logistic regression classifier, with a total of 12 falsely classified blocks
for intra-model device identification among five iMacs, where the truly and falsely
classified blocks (TCB and FCB) are in diagonal and non-diagonal cells, respec-
tively. In Table 27.3 the performance for intra-model identification based on
desktop PCs strongly confirmed the previous result, even though the sound was
transferred through the same headset microphone for all PCs. Table 27.5 shows
the confusion matrix for the linear logistic regression classifier, with a total of 17
falsely classified blocks for intra-model device identification. Finally in Table
27.3 the results for source brand/model identification based on all computer
devices showed that increasing the number of blocks amplifies the computation
time, however the classification accuracy maintained in the same range. More-
over, we achieved lower RMSE values because increasing the number of blocks
reduces the average classification error. Table 27.6 shows the confusion matrix for
the linear logistic regression classifier, with a total of 33 falsely classified blocks
for source brand/model device identification based on all devices.
 272

Table 27.6 Confusion matrix of linear logistic regression classifier based on all devices
ACC = 99.82 % Predicted
i1 i2 i3 i4 i5 PC1 PC2 PC3 PC4 PC5
Actual i1 1,317 0 0 0 0 0 1 0 1 1
i2 0 1,317 0 1 0 1 0 0 0 1
i3 1 0 1,318 0 0 0 1 0 0 0
i4 1 0 0 1,316 0 1 1 0 0 1
i5 0 1 0 0 1,317 0 0 1 1 0
PC1 0 0 0 1 0 1,396 1 0 1 1
PC2 1 0 2 0 0 1 1,396 0 0 0
PC3 1 0 0 0 0 0 1 1,398 0 0
PC4 3 0 0 0 0 1 0 0 1,395 1
PC5 0 0 0 0 1 0 0 0 2 1,397
M. Jahanirad et al.
27 Blind Source Computer Device Identification from Recorded Calls 273

27.4.2 Experiment on Near-Silent Segments

This experiment aims to evaluate the performance of the proposed scheme without
the interference of the speech signal. The experiment extracted the near-silent
segments according to the algorithm that discussed in Sect. 1.1, and created a total
of 1,319 and 1,400 blocks with respect to each iMac and desktop PC devices. At
first the average squared Euclidean distances were calculated as in previous
experiment for the same pair of iMacs to allow comparison, as in Fig. 27.4. The
second part of the experiment repeated the evaluation experiments in previous
Sub-Section using blocks created from near-silent segments. In overall the results
in Table 27.7 shows slight improvements in classification accuracy, computation
time and RMSR with comparison to Table 27.3. However, for intra-model iden-
tification based on iMacs through NN classifier the computation time was
noticeably longer with compare to the results for the original speech samples. It is
plausible that creating the blocks from near-silent segments is a practical option
for eliminating the effects of signal variations by different speakers in real-time
scenarios. However, with adaptation of the entropy-MFCC features the perfor-
mance that obtained for source computer device identification based on original
speech signal was in good agreement with near-silent segments. This justified our
hypothesis in Sect. 27.2.2 for selecting entropy-MFCC features.

Table 27.7 Performance on proposed entropy-MFCC Features using near-silent segments

Intra-model identification based on iMacs
Classification algorithms TCB FCB RMSE ACC (%) Elapsed time (s)
Naïve Bayesian 6,595 0 0 100 0.54
Linear logistic regression 6,583 12 0.0218 99.82 36.28
NN 6,587 8 0.0202 99.88 116.8
LibSVM 6,591 4 0.0206 99.94 5.6
SMO 6,593 2 0.3157 99.97 0.87
Intra-model identification based on desktop PCs
Naïve Bayesian 7,000 0 0 100 0.51
Linear logistic regression 6,994 6 0.0164 99.91 33.15
NN 6,995 5 0.0158 99.93 139.33
LibSVM 6,996 4 0.0151 99.94 5.30
SMO 6,995 5 0.3163 99.93 1.09
Source brand/model identification based on all computer devices
Naïve Bayesian 13,595 0 0 100 1.17
Linear logistic regression 13,567 19 0.0152 99.86 144.80
NN 13,574 21 0.0183 99.85 450.60
LibSVM 13,578 17 0.0158 99.88 14.10
SMO 13,578 17 0.272 99.88 2.10
274 M. Jahanirad et al.

27.5 Conclusions

A blind source computer device identification scheme is developed based on the

entropy-MFCC feature set and recorded VoIP calls. MFCC and entropy features
identify the distinguishing pattern amongst individual computer of the same model
as well as computer devices of different brands. This feature exhibited high per-
formance to capture characteristics of the transfer function of the computer devices
even with the existence of speech signal’s transfer function. The naïve Bayesian
classifier always achieved the highest classification accuracy of 100 % for blind
source computer device identification. However, the scheme was set up in control
condition and further studies are required to implement this approach on more real
case forensic scenarios including large number of devices.

Acknowledgments This work is fully funded by the Malaysian Ministry of Higher Education
under the University of Malaya High Impact Research Grant UM.C/625/1/HIR/MOHE/FCSIT/17.

References

1. Keonig, B.E., Lacey, D.S.: Forensic authenticity analysis of the header data in re-encoded
WMA files from small Olympus audio recorders. J. Audio Eng. Soc. 60, 255–265 (2012)
2. Garcia-Romero, D., Epsy-Wilson, C.Y.: Automatic acquisition device identification from
speech recordings. In: Proceedings of ICASSP, pp. 1806–1809. Dallas, Texas (2010)
3. Panagakis, Y., Kotropoulos, C.: Telephone handset identification by feature selection and
sparse representations. In: Proceedings of WIFS, pp. 73–78. Tenerife (2012)
4. Hanilçi, C., Ertasß, F., Ertasß, T., Eskidere, Ö.: Recognition of brand and models of cell-phones
from recorded speech signals. IEEE Trans. Forensics Secur. 7, 635–634 (2012)
5. Jenner, F.: Non-intrusive identification of speech codecs in digital audio signals. ProQuest
(2011)
6. Kraetzer, C., Oermann, A., Dittmann, J., Lang, A.: Acm: digital audio forensics: a first
practical evaluation on microphone and environment classification. In: Proceedings of the
Multimedia & Security Workshop 2007 Mm&Sec’07, pp. 63–73 (2007)
7. Buchholz, R., Kraetzer, C., Dittman, J.: Microphone Classification Using Fourier
Coefficients. Information Hiding, LNCS, vol. 5806, pp. 235–246. Springer, Berlin (2009)
8. Kraetzer, C., Qian, K., Schott, M., Dittmann, J.: A context model for microphone forensics
and its application in evaluations. In: Proceedings of SPIE-IS&T, San Francisco, CA (2011)
9. Kraetzer, C., Qian, K., Dittmann, J.: Extending a context model for microphone forensics. In:
Proceedings of SPIE, vol. 8303, p. 83030S, 83012, Burlingame, CA (2012)
10. Bhatt, C.A., Kankanhalli, M.S.: Multimedia data mining: state of the art and challenges.
Multimed. Tools Appl. 51, 35–76 (2011)
11. Giannakopoulos, T.: A method for silence removal and segmentation of speech signals,
implemented in Matlab, pp. 1–3. University of Athens, Athens (2010)
12. Giannakopoulos, T.: Study and application of acoustic information for the detection of
harmful content, and fusion with visual information. Department of Informatics and
Telecommunications, vol. PhD. University of Athens, Greece (2009)
13. Beigi, H.: Fundamentals of Speaker Recognition. Springer, New York (2011)
14. Witten, I.H., Frank, E., Hall, M.A.: Data Mining: Practical Machine Learning Tools and
Techniques, 3rd edn. Morgan Kaufmann, Boston (2011)
27 Blind Source Computer Device Identification from Recorded Calls 275

15. Chang, C.-C., Lin, C.-J.: LIBSVM : a library for support vector machines. ACM Trans. Intell.
Syst. Technol. 2, 1–27 (2011)
16. Pamela for Skype—Professional Edition 4.8, http://www.pamela.biz/en/
17. Hall, M., Frank, E., Holmes, G., Pfahringer, B., Reutemann, P., Witten, I.H.: The WEKA data
mining software: an update. SIGKDD Explor. 11, 10–18 (2009)
Chapter 28
Visibility for Network Security
Enhancement in Internet Protocol
Over Ethernet Networks

W.K. Alzubaidi, Longzheng Cai, Shaymaa A. Alyawer

and Erika Siebert-Cole

Abstract This study addresses the naming architecture security problems arising
in Internet protocol (IP) over Ethernet networks. These problems arise because of
the compatibility issue between two different compositional protocols: the IP and
Ethernet protocol. The findings of this study have given rise to proposals for
modifications. A reduction in the current naming architecture design is advocated,
which led to utilize Ethernet frame to provide a visibility for IP over Ethernet
networks. The use of the IP address, as one flat address for the naming architec-
ture, is proposed instead of using both the IP and Media Access Control (MAC)
addresses. The proposed architecture has shown a promising in network security
enhancement.

28.1 Introduction

IP over Ethernet network has become the primary network used by the Internet. In
this network, the data link layer (Layer 2) security problems, has not yet been
adequately addressed. The motivation behind addressing the compatibility prob-
lem is to improve the security of networks by studying the link compatibility
between the Ethernet and IP protocols. The Ethernet was not established to work
with a specific network layer (Layer 3) protocol. Likewise, the IP protocol was not

W.K. Alzubaidi (&)
E. Siebert-Cole

Information Technology Department, University of Tun Abdul Razak, Selangor, Malaysia
e-mail: [email protected]
L. Cai
International University, Selangor, Malaysia
e-mail: [email protected]
S.A. Alyawer
Computer Science Department, Baghdad College, Baghdad 645, Iraq
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 277

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_28
278 W.K. Alzubaidi et al.

designed to work with a specific Layer 2 protocol [1]. This scenario clearly
indicates that the relationship between the IP and Ethernet protocols is not fully
compatible because such networks are not dedicated to each other, which can give
rise to numerous security problems. Resolving the IP to Media Access Control
(MAC) address and the encapsulation of the IP packet into the Ethernet frame are
some requirements to link the IP and Ethernet protocols. Therefore, flat IP address
was proposed to represent the naming architecture [2]. The use of the IP address,
as one flat address for the naming architecture, is proposed instead of using both
the IP and MAC addresses.
In this study, a new concept is introduced to reveal the origin of the private
LAN addresses or of an undeclared Layer 2 address. Internet visibility is not
possible with the current naming architecture in the IP over Ethernet networks [3];
thus, a new solution to current security problems is introduced in this study.
The visibility in the network provides new methods to handle the various
network security threats. For instance, a client can connect to and request services
from a Web server in the WAN without revealing private IP and MAC addresses
[4] (even with a public IP with a proxy case) in the current architecture. As a
result, several types of network attacks can be accomplished without disclosing the
source of the attack. The IP address from Layer 3 is not detected because of the
existence of a NAT in the subsequent router that replaces the original private IP
address with the router’s public IP address. The Layer 2 MAC address cannot be
revealed in any side point because the destination and the source MAC addresses
are replaced with new addresses in each hop. Determining the source of the IP
address, in case it is a public IP address, is possible in Layer 3; however, even this
IP type, which has network proxies, cannot be revealed. This study introduces the
main procedures for the visibility concept proposed for the networks and focuses
on the factors that enable this concept. The dependency and the requirements of
the visibility mechanism are also revealed, and the Visibility Merging Address
(VMA) is also explained in the next sections.
This paper is organized as follows. Next, in Sect. 28.2, we describe utilizing
source hardware address in Ethernet frame header. Section 28.3 presents the
visibility merging address VMA. Section 28.4 describes the VMA construction
mechanism. Section 28.5 describe the visibility architecture and Sect. 28.6 the
conclusions.

28.2 Utilizing Source Hardware Address in Ethernet

Frame Header

In a LAN, the ARP is used to map the IP address to the MAC address. The
destination of the MAC address should be obtained by a source machine using a
destination IP address to construct and transmit the Ethernet frame in IP over
Ethernet networks [5]. This task is performed by the ARP through a broadcast
28 Visibility for Network Security Enhancement … 279

request for mapping the IP to the MAC address and through a stored reply in a
memory space called an ARP cache table. ARP is work as follows: An application
attempts to send data to an IP address of a machine. The IP packet is created by the
network stack and then encapsulated into the Ethernet frame. The destination of
the MAC address is required to transmit this frame [6]. Therefore, the network
stack verifies the IP in the ARP cache table to locate the destination of the MAC
address. If such information is not there, the broadcast ARP request is sent in the
network. Each machine in the network examines the ARP request and checks if the
requested IP is owned. The machine that owns this IP will create an ARP reply
containing the MAC address. A unicast reply would then be sent to the originator
of this request. The originator uses this address in the destination MAC address
field to complete and transmit the frame. This simple protocol does not have any
type of security to bind the IP to the MAC and may result in serious breaches in
security. For instance, the ARP poisoning attack uses unsolicited ARP reply
messages. Network devices cannot verify the ARP sender and whether the mes-
sage comes from the correct device. The ARP does not provide any security
measures, but is based on broadcast messages on the LAN.

28.3 Network Visibility Concept

Providing an address that represents the source of the IP and the MAC addresses
on the Internet clearly informs the end point regarding the origin of the delivered
information. Thus, the two ends of the connection to obtain precise information
about each other. If the VMA concept is utilized, the server clearly identifies the
location of a client in the client server architecture and vice versa. The role of the
source hardware address field in the Ethernet frame in this approach is to provide
this representative source address. Therefore, the transfer of the source MAC
address and the source IP address is proposed to provide and enable visibility
between the two ends of the connection. The two addresses are represented in the
merged address form concept under the condition that 48 bits of the source
hardware address field must be fitted [7].
The VMA address is created by combining the least significant bits of each
source IP address and MAC address. An original MAC address and an IP address
are transferred to the outer side of the LAN, which provide precise source
addresses for the data that are being transmitted. The transparency and visibility
between the two points of the connection help to prevent unauthorized modifi-
cations of the source MAC and IP address pairs and prevents anonymous attacks.
For instance, a DoS attack such example defines the source of the starting attack
points by providing the original source MAC address and the private IP address
[8]. The link between the original MAC source and the private IP address source is
critical in providing security and aids in the clear identification of the other party’s
connection. The DHCP server may assign a different Layer 3 IP address to a single
network node each time it is attached to the network [9], even though the node and
280 W.K. Alzubaidi et al.

its MAC address remain the same. Visibility plays a main role in security, as even
the dynamic IP address may change and may not be guided to the right attack
node. Visibility can provide a Layer 2 source MAC address directed at the precise
network machine that began the event. Thus, visibility can identify the source that
is using the addresses of the two layers.
The proposed VMA address uses the existing size of the MAC address 48-bit to
fit the source hardware address field and to avoid break the standard. The VMA
address also provides advantages such as avoiding the creation of a non-standard
Ethernet frame format. Avoiding the difficulties of adopting a new scheme entails
fresh efforts and additional costs. Each router in the network path must follow the
new procedure when re-encapsulating the arrived frame to resend it instead of
following the current procedure included in a current router’s MAC address. This
procedure maintains the travel of the VMA through the networks until it arrives at
the destination node.

28.4 The VMA Construction Mechanism

The VMA is entered into the 48-bit source hardware address field. The six-byte
length of the MAC address is divided into two three-byte lengths. The first three
least significant bytes from the source IP address replace the first three least
significant bytes in the source hardware address field, whereas the first three least
significant bytes from the source MAC address replace the last three most sig-
nificant bytes in the source hardware address field. In other words, the 48-bit VMA
is constructed when the first three least significant bytes from the source IP address
replace the first three least significant bytes. The last three most significant bytes
from the 48-bit VMA replace the first three least significant bytes in the source
MAC address (as shown in Fig. 28.1).
However, revealing the source’s private IP address and MAC address may not
be desirable in most cases. Thus, the sending process can proceed without enabling
the VMA and providing the source hardware address containing a VMA in the
frame header; the process may maintain the proposed scheme by providing the
source IP address in the source hardware address field. In the destination MAC
address, the most significant bits are occasionally used to indicate broadcast sta-
tion. A broadcast station that checks or uses the source MAC address in the source
hardware address field is available; in other words, a broadcast address is not used
in the source hardware field of the Ethernet frame header. The VMA address in the
proposed scheme uses the source hardware address field in the Ethernet frame as a
carrier. To recognize whether the VMA address has been utilized or not, the last
most significant byte included in the broadcast bits in the VMA address is
examined. If this byte is used, the VMA address was utilized.
28 Visibility for Network Security Enhancement … 281

Fig. 28.1 Generation mechanism for the Visibility Merging Address (VMA)

After converting the source IP address into the hexadecimal form, take the first
three octets (the least significant bits) to be the first three octets from the VMA
address. Taking the last three octets from the original MAC address (the most
significant bits) to take the last three octets position from the VMA address that is
want to be construct. The VMA address will carry in the destination hardware
address field in Ethernet frame header (Fig. 28.2).

28.5 Visibility Architecture

The transmission of a Layer 2 MAC address to the outer side of the LAN to reach
the second end of the connection on the WAN side is proposed. This proposal
considers the necessary level of Internet visibility. The transfer of the MAC
address outside the LAN has not been proposed previously because the MAC
address is typically used to guide Ethernet frame travel within the LAN; thus, the
MAC address does not exit its LAN. The transfer of the original MAC address
from one network to another using the source hardware address field in the
Ethernet frame is proposed in this study. The source hardware address field no
longer plays a main role given the previous description of the new Layer 2
transmission mechanism. Therefore, a new concept is introduced to utilize the
282 W.K. Alzubaidi et al.

Fig. 28.2 Ethernet frame with VMA for the proposed visibility architecture

source hardware address field to carry the information regarding the original MAC
and IP address source. In the current scheme, the online destination IP address
travels through the networks in a packet until it arrives at the target network node.
The source IP address in the packet does not change if it is a public IP address
type. If the source IP is a private IP address, it changes at the next router in the
NAT process.
In this approach, the source hardware address in the Ethernet frame remains
fixed until it reaches the end of the connection along with the destination IP
address fields in the packet. As the destination MAC address in the Ethernet frame
header changes, each hop obtains the value of the next destination. The destination
IP address is used as a destination address for Layer 2 in the next hop instead of the
destination MAC address in the current scheme, as previously described, on the
condition that the content of the source hardware address is maintained without
changing the visibility of the merging VMA with each intermediate node. This
process has a significant effect on the private network, especially on the NAT
function.
In terms of security, the visibility helps to identify attacks. Therefore, the
visibility concept has a significant effect on the enhancement of network security,
especially with elements, such as the MAC address and the original source private
IP address, that are not revealed to the outside LAN. Visibility does not introduce
any additional complexity to the naming system. The only amendments made are
to the contents of the Ethernet frame header fields. This mechanism includes
encapsulation with the new VMA concept. The size of the frame header field is
also unaffected because the proposed VMA fits the 48-bit source hardware address.
28 Visibility for Network Security Enhancement … 283

In the proposed scheme, the source hardware address field in the Ethernet frame
header is proposed as a carrier of information to the WAN side to provide the
original source MAC and IP address. This information is viewed as a VMA that
consists of a combination of part of the source MAC address and part of the IP
address that conveys the 48-bit standard source hardware address field. All of the
information for the original source MAC address is removed at the first router
gateway and a new Layer 2 header is added because the Ethernet frame is used
within the LAN. The IP and MAC addresses in the outgoing packet/frame from the
router do not provide any information regarding the original source MAC address
used in the frame. The Layer 2 encapsulation typically carries information only
within the LAN. In the network node procedure, the Ethernet frame in each
delivered node is de-capsulated and re-encapsulated with the new control infor-
mation. In the current procedure, the routers do not allow the source MAC address
to cross to another communication end point. The routers also do not allow the end
points to know the original source of the private IP and MAC addresses. Each
intermediate node removes the Layer 2 information and re-capsulate the infor-
mation with its MAC address as a source address. As shown in Fig. 28.6, how a
sending and receiving in the visibility was enabled with a Private IP Address and
NATing in the proposed naming architecture. In private IP networks, the router has
a NAT that removes the original source private IP address in the packet. Therefore,
the maintenance of the source hardware address field in the Ethernet frame header,
which contains an address generated by the merging of parts of the MAC and IP
addresses in the source node, is proposed. In this approach, the intermediate
routers do not change the parameters of the source hardware address field. In the
intermediate router, the Ethernet frame is de-capsulated and then re-encapsulated
on the condition that the VMA address remains unchanged and is resent to the next
node.
As a result of this approach, visibility on the Internet is provided to the original
Layer 2 device address and the Layer 3 private IP address. Therefore, the original
MAC and IP address link is provided with protection to avoid unauthorized
changes.
The destination hardware address field is examined to determine whether the
delivered frame is for the right node or not because the IP address is used as a
source and destination address instead of the MAC address in Layer 2, as described
in the proposal. The flowcharts describing sending and receiving procedures are
shown in Figs. 28.3 and 28.4.
In this approach, the reply of the network node to the source of the delivered
frames is dependent on the source IP address field in the packet to construct the
frame using the MAC address that was generated from the source IP address. This
mechanism traditionally enters the source MAC address into the source hardware
address field in the frame, and the reply uses the source hardware address as the
destination address in the new frame. The algorithm used to generate the Layer 2
MAC address based on the source IP address in the header of the delivered packet
shown as flowcharts as following.
284 W.K. Alzubaidi et al.

Fig. 28.3 Flowchart sending

data with visibility enabled

The generated MAC address is used as a Layer 2 destination address and

includes the constructed frame to be sent. As previously mentioned the procedure
does not use the ARP cache table but depends only on the Layer 3 IP address as a
flat address to guide the frame in Layer 2 transmission [12]. The source hardware
address field in the Ethernet frame header does not play a role at present because
28 Visibility for Network Security Enhancement … 285

Fig. 28.4 Flowchart

receiving data with visibility
enabled

the source IP address is used as a reference in case the reply message requires
knowledge regarding the source of the delivered frame.
Visibility also provides security advantages. In the event that a server is sus-
ceptible to any type of attack (such as Distributed Denial of Service or DDoS), the
exact source of each delivered packet is easily determined. Thus, the identity of
any anonymous attacker can be revealed as seen the description for the visibility
architecture in Fig. 28.5 (Fig. 28.6).
286 W.K. Alzubaidi et al.

Fig. 28.5 Ethernet frame with VMA for the proposed visibility architecture

(a) DMA C SMAC SIP DIP DMAC SMAC SIP DIP DMAC SMA C SIP DIP DMAC SMAC SIP DIP
C1 IP’ A VMA A IP F IP
NAT D1 IP’ A VMA C2 IP F IP E1IP’ A VMA C2 IP F IP F IP’ A VMA C2 IP F IP

C2 D1 D2 E1
C1 E2 F
Host A Router D Server
Gateway Gateway
A IP
A MAC

Ethernet Switch Ethernet Switch

Host B

(b) DMAC SMAC SIP DIP DMAC SMAC SIP DIP DMAC SMAC SIP DIP DMAC SMAC SIP DIP
AIP’ C1IP’ F1IP AIP NAT C2IP’ D1IP’ FIP C2IP D2IP’ E1IP’ FIP C2IP E2IP’ FIP’ FIP C2IP

C2 D1 D2 E1
C1 E2 F
Host A Router D Server
Gateway Gateway
AIP
AMAC

Ethernet Switch Ethernet Switch

Host B
Visibility Merging Address (VMA) = First three leased significant bytes of MAC address + First three leased significant bytes from IP address
DMAC: Destination Hardware Address Field in the frame header SIP: Source IP address field in the Packet Header
SMAC: Source Hardware Address Field in the frame header DIP: Destination IP Address Field in the Packet Header
IP’: The 48bits Address that derived from the Origen 32bit IP Address from Layer 3 and used Instead MAC address

Fig. 28.6 Visibility enabled with a private IP address and NATing in the proposed naming
architecture. a Sending, b reply message
28 Visibility for Network Security Enhancement … 287

28.6 Conclusions

A new Internet visibility architecture is presented to enhance the security level by

raising the Ethernet working level to match the WAN side. A visibility mechanism
is proposed to provide the private IP and MAC addresses to the WAN side of the
network. The private IP address is generally hidden behind the NAT in the router.
In the proposed architecture, the private IP address is revealed to the second end
node in the communication. This visibility causes the Internet to become more
visible and to be easily tracked. The visibility also eliminates Internet attacks such
as anonymous and DDoS attacks. The limitation is in the scope of this study, the
current visibility concept is presented under the Ethernet protocol. This concept
can also be applied to Layer 2 WAN protocols and provides visibility to for all of
the architectures in the LAN and WAN networks. The application of the visibility
concept to ATM and frame relay as WAN protocol examples may require further
study. The influence of the proposed architecture and its effectiveness in terms of
Internet visibility are theoretical evaluated.

References

1. Postel, J.: Internet protocol (1981)

2. Alzubaidi, W.K., Cai, L., & Alyawer, S.A.: Enhance the security and performance of IP over
ethernet networks by reduction the naming System Design. Int. J. Comput. Netw. (IJCN) 4(5)
(2012)
3. Wulf, V., Hartmann, A.: The ambivalence of network visibility in an organizational context.
In: Paper presented at the NetWORKing (1993)
4. Forouzan, B.A.: TCP/IP Protocol Suite. McGraw-Hill, New York (2002)
5. Plummer, D.C.: RFC 826: An ethernet address resolution protocol. InterNet Network
Working Group (1982)
6. Alzubaidi, W.K., Cai, L., Alyawer, S.A.: A new verification method to prevent security
threads of unsolicited message in IP over ethernet networks. Int. J. 4(6) (2012)
7. Spurgeon, C.E.: Ethernet: The Definitive Guide. O’Reilly, Sebastopol (2009)
8. Mirkovic, Jelena, Dietrich, Sven, Dittrich, David, Reiher, Peter: Internet Denial of Service:
Attack and Defense Mechanisms (Radia Perlman Computer Networking and Security).
Prentice Hall PTR, New Jersey (2004)
9. Strebe, Matthew: Windows 2000 Server 24seven. Wiley, New York (2006)
10. Hayawi, Kadhim, Al Braiki, Arwa, Mathew, Sujith Samuel: Network Attacks and Defenses: a
Hands-on Approach. CRC Press, New York (2012)
11. Kiravuo, T., Sarela, M., Manner, J.: A Survey of Ethernet LAN Security
12. Alzubaidi, W.K., Cai, L., Alyawer, S.A.: A framework for optimizing IP over ethernet
naming system. Int. J. Comput. Sci. Issues (IJCSI) 9(6) (2012)
Chapter 29
Comparative Analysis of Different Single
Cell Metamaterial

Pankaj Rameshchandra Katiyar

and Wan Nor Liza Binti Wan Mahadi

Abstract Artificially constructed metamaterial have become of considerable

interest due to their negative properties. Under normal circumstance the electro-
magnetic energy inside metamaterial flows in the reverse direction. In this
research, we compare seven different metamaterial structures against their nega-
tive frequency, simulation time, memory requirement and mesh cell. We also
present a comparative analysis on tunability of single cell metamaterial structure to
a different frequency band of interest. To come up with proper analysis we have
restricted cell size of 2.5 mm 9 2.5 mm and simulated structure from 0 Hz to
20 GHz. SRR structure has a negative frequency response in a lower band while
Jerusalem cross takes less simulation time. S structure has lowest memory and
mesh cell requirement, but it is hard to tune. However, CLS loaded split ring
resonator has advantages over all the parameters of comparison and is found best.

29.1 Introduction

V.G. Vesslago in the negative direction and hence also called as Left handed
Material. J.B. Pendry in 1968 introduced the concept of metamaterial and
described their distinct negative properties [15]. Under normal condition the
electromagnetic energy in metamaterial flows fabricated first metamaterial by
arranging metal rod into an array [12]. This array when exposed to electromagnetic
energy was capable of focusing energy into a narrow beam. J.B. Pendry also

P.R. Katiyar (&)
W.N.L.B.W. Mahadi

EMRD, Department of Electrical Engineering, Faculty of Engineering,
University of Malaya, 50603 Kuala Lumpur, Malaysia
e-mail: [email protected]
W.N.L.B.W. Mahadi
e-mail: [email protected]

Ó Springer International Publishing Switzerland 2015 289

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_29
290 P.R. Katiyar and W.N.L.B.W. Mahadi

showed that negative permittivity and permeability can be obtained using Split
Ring Resonator [13].
Due to their unique negative property Metamaterial find many applications in
optimizing electromagnetic waves. One of the first application as proposed by
Pendry was to create an electromagnetic lens of sub wavelength resolution. Ran
[14, 16] carried out several experiments on metamaterial such as Power trans-
mission experiment, Prism refraction, Beam Shifting experiment, and focusing
experiment. These experiments form the basis of all the application areas of
metamaterial [1, 4, 5, 7, 17]. Specific experiments have been conducted to enhance
gain of antenna by using metamaterial in different forms and places surrounding
the antenna.
We present comparison of 7 types of metamaterial structures. The comparison
is done against their negative frequency range by keeping substrate size and its
properties constant. Simulation of each model is performed using CST and cor-
responding 2 port S-Parameter is extracted. NRW method is used to extract
constitutive parameter from simulated S-Parameter. Nicholson-Ross-Weir (NRW)
[10] method provides a direct calculation of both permittivity and permeability
from the s-parameter. However, the method diverges for low loss material at
frequencies corresponding to integer multiples of one-half wavelength in the
sample which is due to the phase ambiguity, Hence it is restricted to an optimum
sample thickness of kg/4 and used preferably for short samples. The comparison is
also extended to analyze tunability, size as compared to resonance frequency,
memory consumption while simulating and meshing, simulation time and number
of mesh cell required. This comparison is helpful in determining the type of
metamaterial cell for desired application.

29.2 Metamaterial Design and Simulation

This paper shows a comparison between following four metamaterial structures.

A. Split ring resonator structure
B. Omega structure
C. S structure
D. Symmetrical ring structure
E. Jerusalem cross
F. Offset fed split ring [DSRR]
G. CLS loaded SRR
To come up with fair comparison, several constrain on simulation and modeling
single cell structure is enforced. The modeling constraints such as substrate size
should not exceed 2.5 mm 9 2.5 mm, substrate is 0.25 mm thick regular FR4. The
copper pattern used on top and bottom layer of PCB has 0.017 mm thickness.
Regular FR4 has permittivity (e) of 4.35 and loss tangent (tan d) of 0.025 at
20 MHz. All the single cell metamaterial structures are made to fit in the above set
29 Comparative Analysis of Different Single Cell Metamaterial 291

Fig. 29.1 Single cell split

ring resonator structure

Fig. 29.2 Extracted negative 30

Eps_re Mu_re
permittivity and permeability 20
10
0
-10
-20
-30
0.0
1.1
2.1
3.2
4.3
5.3
6.4
7.4
8.5
9.6
10.6
11.7
12.7
13.8
14.9
15.9
17.0
18.0
19.1
Frequency in GHz

physical parameter. Simulation constrains such as the space surrounding meta-

material is constant and is vacuum, the boundary condition is kept same.

29.2.1 Split Ring Resonator Structure

Split Ring resonator has two rings on top layer and one strip at the back acting as a
rod. The outer ring has opening on top side while the opening of inner ring is
rotated by 180°. The outer ring has length is 2.2 mm and the inner ring length is
1.5 mm. The trace width of both rings is 0.2 mm. Inner ring and outer ring are
separated by 0.15 mm. The gap between inner ring and outer ring is 0.3 mm. The
trace on bottom layer of PCB is of 2.5 mm length and the width of trace is 0.2 mm.
The inductance is provided by the ring pattern formed on top layer while the
capacitance is provided by the separation between inner ring and outer ring.
Additionally the strip on the bottom side of the PCB provides some more
inductance and capacitance (Figs. 29.1 and 29.2).

29.2.2 Omega Structure

Omega structure [6] consists of an omega shape structure on top and bottom surface
of the PCB. The ring traces width is 0.2 mm with inner ring radius of 0.9 mm.
292 P.R. Katiyar and W.N.L.B.W. Mahadi

Fig. 29.3 Single cell omega

structure

Fig. 29.4 Simulated results 10

of Omega structure 0
-10
Eps_re Mu_re
-20
-30
-40
-50
0.0
1.1
2.1
3.2
4.3
5.3
6.4
7.4
8.5
9.6
10.6
11.7
12.7
13.8
14.9
15.9
17.0
18.0
19.1
Frequency in GHz

Fig. 29.5 Single cell

S structure

The trace at the lower section of omega is also 0.2 mm wide. The gap at the lower
center of the structure is 0.4 mm wide. The ring of Omega structure provides
inductance. Since the ring is printed on both the side of PCB they couple to provide
capacitance. Additional capacitance is provided by the 0.4 mm split in the ring
(Figs. 29.3, 29.4).

29.2.3 S Structure

Figure 29.5 shows the simulated single cell S-Structure [2]. The trace width used to
create S-Structure is 0.2 mm. S-Structure is placed in substrate of 3 mm 9 2.5 mm.
The width of the substrate used here is 0.25 mm. It is very difficult to fit S-Structure
in 2.5 mm 9 2.5 mm substrate and get a negative response below 20 GHz.
29 Comparative Analysis of Different Single Cell Metamaterial 293

Fig. 29.6 Simulated results 15

of S structure Eps_re Mu_re
10
5
0
-5
-10
-15

18.0
18.1
18.2
18.4
18.5
18.6
18.7
18.8
18.9
19.1
19.2
19.3
19.4
19.5
19.7
19.8
19.9
Frequency in GHz

Fig. 29.7 Symmetrical ring

structure

The inductance is provided by the trace forming S pattern while capacitance is

provided by coupling between S pattern on top and bottom layer. In this case, the
mutual coupling is limited due to limited overlap of pattern between top and
bottom layer, hence the negative permeability response is weak. The inductance
providing negative permittivity response is weaker as compared to Omega pattern
(Fig. 29.6).

29.2.4 Symmetrical Ring Structure

Another type of metamaterial structure which exhibits negative permittivity and

permeability is symmetrical ring structure as shown in Fig. 29.7. This structure is
made to fit in the physical constrain set earlier [11].
Trace width 0.2 mm is used to construct a symmetrical ring. Two symmetrical
ring of 1.1 mm in length are placed facing each other at a gap of 0.1 mm. The rod
on the bottom layer of the PCB is 0.4 and 2.5 mm in length.
The major part of the inductance of symmetrical ring structure is due to two
ring structure on top while the capacitance is due to the coupling between one side
of both the rings on top layer. The split on both the ring provides additional
capacitance. Additionally the strip on the back has the same effect as to strip in
SRR structure (Fig. 29.8).
294 P.R. Katiyar and W.N.L.B.W. Mahadi

Fig. 29.8 Simulated results 15

of symmetrical ring structure Eps_re Mu_re
10
5
0
-5
-10
-15

14.0
14.3
14.6
15.0
15.3
15.6
15.9
16.2
16.5
16.9
17.2
17.5
17.8
18.1
18.5
18.8
19.1
19.4
19.7
Frequency in GHz

Fig. 29.9 Jerusalem cross

structure

29.2.5 Jerusalem Cross

Jerusalem cross metamaterial is very similar to Jerusalem cross and hence the name
[3]. The design used for our simulation is a single layer design with 0.1 mm wide
trace constructing cross. Each arm of the cross is 1.9 mm long as shown in
Fig. 29.9. The simulated permittivity and permeability is as shown in Fig. 29.10.
Jerusalem cross is a complex structure in analyzing capacitance and inductance.
The 0.1 mm traces forming part of the Jerusalem cross structure provides induc-
tance while the capacitance is provided by the coupling formed on four corners of
the Jerusalem cross. If the width and height of the trace are increasing the capac-
itance will also increase. Since Jerusalem cross has pattern on top layer it has
limited capacitance and inductance. If the Jerusalem cross is used in array the
capacitance will increase exponentially due to coupling with outer arm of the cross.

29.2.6 Offset Fed Diamond Shape Split Ring Resonator

[DSRR]

DSRR is similar to split ring resonator but rotated by 45° angle anti clockwise [8].
Due to limitation applied on substrate size the dimensions of the rings are changed.
The trace is 0.16 mm wide. The outer ring is 1.72 mm long while the inner ring is
29 Comparative Analysis of Different Single Cell Metamaterial 295

Fig. 29.10 Simulated results 10 4000

of Jerusalem cross Mu_re Eps_re 3000
5 2000
1000
0
0
-5 -1000
-2000
-10 -3000

0.0
1.2
2.4
3.6
4.7
5.9
7.1
8.3
9.5
10.6
11.8
13.0
14.2
15.4
16.5
17.7
18.9
Frequency in GHz

1.17 mm long. The gap between inner and outer ring is 0.2 mm as shown in
Fig. 29.11. Simulated permittivity and permeability is shown in Fig. 29.12.

29.2.7 Capacitive Loaded Strip with Split Ring Resonator

As the name suggests capacitive loaded strip SRR has two additional strips on
sides of SRR [9]. The capacitive strip is I in shape with 1.32 mm in length and
1.15 mm wide. The gap between two capacitive strips is 0.05 mm as shown in
Fig. 29.13. The simulated permittivity and permeability is shown in Fig. 29.14.

Fig. 29.11 DSRR

metamaterial

Fig. 29.12 Simulated results 5 400

of DSRR metamaterial Mu_re Eps_re
3
200
1
Eps
Mu

0
-1
-200
-3

-5 -400
12.0
12.3
12.6
12.9
13.2
13.4
13.7
14.0
14.3
14.6
14.9
15.2
15.5
15.7

Frequency in GHz
296 P.R. Katiyar and W.N.L.B.W. Mahadi

Fig. 29.13 Capacitive loaded strip with SRR

400
Eps_re Mu_re
200
0
-200
-400
0.0
1.1
2.1
3.2
4.3
5.3
6.4
7.4
8.5
9.6
10.6
11.7
12.7
13.8
14.9
15.9
17.0
18.0
19.1
Frequency in GHz

Fig. 29.14 Simulated results of capacitive loaded strip with SRR

CLS loaded SRR has the advantage of SRR and I pattern. Additional I pattern
increased the amount of capacitance exponentially. The coupling of I pattern to I
pattern and both I pattern to outer ring brings the resonance frequency down. The
SRR structure is 3.5 times smaller but the negative response and resonance are
close to SRR.

29.3 Comparison and Observation

To have fair comparison all the structure is constrained to be designed for 2.5
mm 9 2.5 mm 9 0.25 mm FR4 substrate. The only exception here is S-Structure
which was about 2.5 mm 9 3.0 mm 9 0.25 mm. We are comparing the simula-
tion result against their resonance frequency, negative permittivity and perme-
ability frequency and ease of tuning structure to operate at the desired frequency
band. We are also listing the amount of memory consumed while simulating along
with number of mesh cell generated and simulation time. While simulating care is
taken to maintain structure orientation and the space surrounding structure
constant.
29 Comparative Analysis of Different Single Cell Metamaterial 297

Table 29.1 Comparison of simulated parameters

Resonance Negative Negative Tunability
frequency (GHz) permittivity permeability
(GHz) (GHz)
Split ring 10.4 B10 C10 Easy
resonator
Omega structure 16.8 Entire band C15.7 Hard
S structure 19 C15.7 15.7 Hard
Symmetrical 15.5 and 19.1 C14.3 C14.3 Hard
ring structure
CLS loaded SRR Low loss, entire B10.2 B10.2 Easy
bandwidth
DSRR 14.4 13.6 * 13.7 13.6 * 14.2 Hard
Jerusalem cross No clear C18.3 GHz 18.3 * 18.4 Moderate
resonance

Table 29.2 Comparison of simulation time and memory requirement

Simulation time (Sec) RAM memory (MB) Mesh cell
Split ring resonator 121 78 195,250
Omega structure 367 71 213,120
S structure 111 50 107,712
Symmetrical ring structure 92 57 150,672
CLS loaded SRR 88 54 149,124
DSRR 680 88 248,400
Jerusalem cross 72 59 110,400

The two most important parameters for antenna design are simulation time and
size. In terms of simulation time CLS loaded SRR and Jerusalem cross takes the
least amount of time. The amount of memory consumed by these structures is also
low. However, Jerusalem cross has negative permittivity and permeability at
higher frequencies that CLS. In terms of negative parameters SRR and CLS loaded
SRR provides negative permittivity and permeability at the lowest frequency, so
has a huge size advantage. Omega structure, S Structure and Symmetrical structure
are complex to tune and shows negative parameter at higher frequency
(Table 29.1).
Simulation time of DSRR is highest. Symmetrical ring has 2 resonance fre-
quency once at 15.5 GHz and second at 19.1 GHz but it is hard to tune and
simulate. CLS loaded SRR and Jerusalem cross has very wide resonance frequency
and hence no clear peak is observed. CLS-SRR has low loss over the entire
frequency band (Table 29.2).
298 P.R. Katiyar and W.N.L.B.W. Mahadi

29.4 Conclusion

CLS loaded SRR offers the best solution for small size and lowest simulation time.
The next alternative is the Split Ring Resonator which is easily tunable, but takes
33 s more to simulate. DSRR on the other side takes 680 s to simulate and has
higher memory requirement. The negative response of DSRR is also on the higher
side.

References

1. Buell, K., Mosallaei, H., Sarabandi, K.: A substrate for small patch antennas providing
tunable miniaturization factors. IEEE Trans. Microw. Theory Tech. 54(1), 135–146 (2006)
2. Chen, H., Ran, L., Huangfu, J., Zhang, X., Chen, K., Grzegorczyk, T.M., Kong, J.A.: Left-
handed materials composed of only S-shaped resonators. Phys. Rev. E 70(5), 057605 (2004)
3. Dongying, L., Szabo, Z., Xianming, Q., Li, E.P., Zhi Ning, C.: A high gain antenna with an
optimized metamaterial inspired superstrate. IEEE Trans. Antennas Propag. 60(12),
6018–6023 (2012). doi:10.1109/TAP.2012.2213231
4. Faruque, M., Islam, M., Misran, N.: Analysis of electromagnetic absorption in mobile phones
using metamaterials. Electromagnetics 31(3), 215–232 (2011)
5. Faruque, M.R.I., Islam, M.T., Misran, N.: Design analysis of new metamaterial for EM
absorption reduction. Prog. Electromagnet. Res. 124, 119–135 (2012)
6. Huangfu, J., Ran, L., Chen, H., Zhang, X-m, Chen, K., Grzegorczyk, T.M., Kong, J.A.:
Experimental confirmation of negative refractive index of a metamaterial composed of X-like
metallic patterns. Appl. Phys. Lett. 84(9), 1537–1539 (2004)
7. Islam, M., Faruque, M., Misran, N.: SAR reduction in a muscle cube with metamaterial
attachment. Appl. Phys. A 103(2), 367–372 (2011)
8. Joshi, J., Pattnaik, S., Devi, S., Lohokare, M., Vidyasagar, C.: Offset fed diamond shaped
split ring (DSSR) planar metamaterial antenna. In: IEEE Applied Electromagnetics
Conference (AEMC), pp. 1–4 (2009)
9. Majid, H.A., Abd Rahim, M.K., Masri, T.: Microstrip antenna’s gain enhancement using left-
handed metamaterial structure. Prog. Electromagnet. Res. M 8, 235–247 (2009)
10. Nicolson, A., Ross, G.: Measurement of the intrinsic properties of materials by time-domain
techniques. IEEE Trans. Instrum. Meas. 19(4), 377–382 (1970)
11. O’brien, S., Pendry, J.: Magnetic activity at infrared frequencies in structured metallic
photonic crystals. J. Phys.: Condens. Matter 14(25), 6383 (2002)
12. Pendry, J.B.: Negative refraction makes a perfect lens. Phys. Rev. Lett. 85(18), 3966 (2000)
13. Pendry, J.B., Holden, A.J., Robbins, D., Stewart, W.: Magnetism from conductors and
enhanced nonlinear phenomena. IEEE Trans. Microw. Theory Tech. 47(11), 2075–2084
(1999)
14. Ran, L.-X., Huang-Fu, J.T., Chen, H., Zhang, X.-M., Chen, K.S., Grzegorczyk, T.M., Kong,
J.A.: Experimental study on several left-handed matamaterials. Prog. Electromagnet. Res. 51,
249–279 (2005)
15. Vesslago, V.: The electrodynamics of substances with simultaneously negative values of e
and l’. Sov Phys Usp 10, 509–514 (1968)
16. Wu, B.-I., Wang, W., Pacheco, J., Chen, X., Grzegorczyk, T.M., Kong, J.A.: A study of using
metamaterials as antenna substrate to enhance gain. Prog. Electromagnet. Res. 51, 295–328
(2005)
17. Ziolkowski, R.W.: Metamaterial-based antennas: Research and developments. IEICE Trans.
Electron. 89(9), 1267–1275 (2006)
Chapter 30
Distributed Video Coding with Frame
Estimation at Decoder

Kin Honn Chiam and Mohd Fadzli Mohd Salleh

Abstract In distributed video coding, input video stream is split into group of
pictures of various lengths. Longer group of pictures is preferred as there is more
temporal redundancy. However, the system time might increase as more frames
need to be transmitted and be stored in the buffer. In this paper, we try to reduce
the system time by proposing a model where only the odd-numbered Wyner-Ziv
frames shall be transmitted from the encoder. At the decoder, the missing even-
numbered Wyner-Ziv frames shall be estimated for full reconstruction of the video
sequence. The simulation results show that the proposed model is more efficient as
the output video sequence could be obtained in a shorter time. The estimated
missing frames at the decoder are also of acceptable quality compared to the
original frames at the encoder.

30.1 Introduction

Distributed video coding (DVC) is a new video coding paradigm for video
transmission, based on the Slepian-Wolf and Wyner-Ziv information theoretic
results. In the conventional video compression techniques, decoder is of low-
complexity. The primary objective of DVC is to achieve low-complexity encod-
ing, by shifting the bulk computation to the decoder [8]. There are couple of early
architectures and implementations of DVC as stated in the publications [1, 3, 7].
Based on the cumulative motion crossing a pre-defined threshold, the incoming
video sequence is sampled into consecutive frames in groups. The number of

K.H. Chiam (&)  M.F.M. Salleh

School of Electrical and Electronic Engineering, Universiti Sains Malaysia, Seri Ampangan,
14300 Nibong Tebal, Pulau Pinang, Malaysia
e-mail: [email protected]
M.F.M. Salleh
e-mail: [email protected]

Ó Springer International Publishing Switzerland 2015 299

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_30
300 K.H. Chiam and M.F.M. Salleh

frames in each group is called group of pictures (GOP). The first frame in a GOP is
known as the key frame while the remaining frames are called the Wyner-Ziv
(WZ) frames. Depending on the input video, the GOP can be of various sizes.
Higher GOP size increases the number of WZ frames between key frames.
However, this reduces the data rate [4].
In this study, we proposed a model where only certain frames in the GOP will
be transmitted from the encoder. An interpolator is used at the decoder to estimate
the missing nontransmitted frames. The quality of the estimated frames is mea-
sured by comparing the frames with the corresponding input frames.
Hence, together with the estimated frames, there is still a complete GOP at the
decoder. All frames will then be used to reconstruct the output video sequence.
The efficiency of the model is determined by measuring the time needed by the
system to complete processing and transmitting a video sequence.
Prior to transmission, we utilized the Reed-Solomon (RS) codes as the forward
error correction (FEC) technique to protect the message. These codes are word
oriented rather than bit oriented [9]. This stems the codes the capacity to correct
burst errors, where a series of bits in the codeword are received in error. The burst
error is relatively common in wireless communication due to fading. A bit oriented
code would treat this situation as many independent single-bit errors. However, for
RS codes, a single error means any or all incorrect bits within a single word [6].
The Sect. 30.2 highlights the basic concepts of DVC, RS codes, and linear
interpolation method. In Sect. 30.3, the details of the proposed DVC codec model,
methodology, and implementation are presented. The results with the C++ codes
implementation are in the Sect. 30.4. Finally, the Sect. 30.5 states the conclusions
and further work.

30.2 Foundation of DVC, RS Codes, and Linear

Interpolation

30.2.1 Concepts of Distributed Video Coding

During the encoding process of DVC, incoming video sequence is first split into
the consecutive frames, to form the GOP. First frame in the group, the key frame,
is intra-coded with the conventional way, providing side information (SI) at the
decoder. Other frames, the WZ frames, are interframe encoded, with only the
parity bits sent across the network.
In contrast, the DVC decoding process is relatively more complicated. The
received key frame will first be decoded for reconstruction of the SI.
The SI is an estimation of the WZ frame, which is only available at the encoder.
Together with the parity bits, SI is used to complete the decoding process. If the
decoding fails, more parity bits will be requested via the feedback channel. This is
iterated until successful decoding is obtained. The detailed operation of the DVC
could be found in other publications [2, 4].
30 Distributed Video Coding with Frame Estimation at Decoder 301

Fig. 30.1 Typical Structure of RS Codeword [6]

30.2.2 Reed-Solomon Codes

The RS codes are a subset of BCH codes and class of linear block codes. Normally,
RS codes are specified with the notation, RS(n, k) with m-bits symbols. A RS
encoder takes k-blocks of digital data and adds extra redundant symbols, t to create
n-blocks of codeword [6]. The RS decoder processes each codeword block and
attempts to correct the errors occurred during transmission, recovering the original
data. The number and type of errors that can be corrected depends on the charac-
teristics of that particular RS codes and the number and distribution of errors.
(1) If 2s + r \ (n – k) (s-errors, r-erasures), then the original transmitted
codeword will always be recovered.
(2) The decoder is not able to recover the original codeword and indicate this
fact.
(3) The decoder will mis-decode and recover an incorrect codeword without any
indication.
Figure 30.1 shows a typical RS codeword. There are more details about the RS
coding scheme in the publication [9].

30.2.3 Linear Interpolation

Linear interpolation can be used as a method to estimate the missing values of a

function by curve fitting or assuming that it is a straight line between two values.
Generally, the linear interpolation function is given by (30.1)

ðx  x 1 Þ
f ð xÞ ¼ y1 þ ðy2  y1 Þ: ð30:1Þ
ðx 2  x 1 Þ

where x is an unknown point on a line with ðx1 ; y1 Þ and ðx2 ; y2 Þ are two other
points on the same line.
On a regular two dimensional grid, bilinear interpolation is used to interpolate a
bi-variate function. Linear interpolation is first performed in one direction, and
then again in the other direction. The estimated value of the desired point, f(x, y)
on a xy-plane is given by (30.2)
302 K.H. Chiam and M.F.M. Salleh

WZ Encoder WZ Decoder
Decoded
Input RS RS Output WZ
Buffer Encoder Decoder Buffer Frames

Request
Bits

Quantizer Inverse
Quantizer

Input WZ
Frames

Input Video Interpolator and

Video Splitter Reconstructor

Input Key
Frames

Conventional Conventional Decoded

Intraframe Intraframe Key
Encoder Decoder Frames

Fig. 30.2 Proposed distributed video coding model

1
f ðx; yÞ ¼ ½ f ðx1 ; y1 Þðx2  xÞðy2  yÞf ðx2 ; y1 Þðx  x1 Þðy2  yÞ
ðx2  x1 Þðy2  y1 Þ
þ f ðx1 ; y2 Þðx2  xÞðy  y1 Þ þ f ðx2 ; y2 Þðx  x1 Þðy  y1 Þ:
ð30:2Þ

where ðx1 ; y1 Þ; ðx1 ; y2 Þ; ðx2 ; y1 Þ and ðx2 ; y2 Þ are four other points on the same plane.

30.3 Methodology

In this section, the proposed model as shown in Fig. 30.2 and the detailed meth-
odology are presented. Figure 30.3 describes the operational flow of the model.
Firstly, the input video sequence is split into a GOP of eight frames with the video
splitter. All WZ frames will be stored in the input buffer. Only the odd-numbered
WZ frames will be sent for quantization and encoded with the RS codes. At the
decoder, the received frames will be first decoded and stored in the output buffer.
The missing even-numbered WZ frames will be estimated by using the
30 Distributed Video Coding with Frame Estimation at Decoder 303

Start

Input
video

Split video to GOP

No Intraframe encode
WZ frame

Yes

Odd-numbered No Next frame Transmit to

frames decoder

Yes

Quantization Intraframe decode

RS encode

Transmit to
decoder

RS decode

Yes Estimate even-

Decode success numbered WZ frames

No

Request bits from Reconstruction

encoder

Output
video

End

Fig. 30.3 Flowchart for the proposed distributed video coding model
304 K.H. Chiam and M.F.M. Salleh

Table 30.1 Rs Parameter

Parameter Meaning Value or range
m Number of bits per symbol Integer between 3 and 16
n Number of symbols per codeword Integer between 3 and 2m-1
k Number of symbols per message Positive integer\n, where (n-k) is even
t Error correction capability of the ðnkÞ
2
codes

interpolation function. Each decoded or interpolated output frame will be stored in

the output buffer and compared with the corresponding input frame. Based on this
comparison, the peak signal to noise ratio (PSNR) is calculated. The quality of the
output frame is considered good if the value of the PSNR is above 30 dB.
On the other hand, the key frame is encoded using the conventional intraframe
encoding method. The chosen technique in this proposed model is the H.264
encoding scheme. The model is tested with a video sequence of low motion
activity.
The performance of the model is measured by calculating the time required by
the model to finish processing the GOP of eight frames. As a reference, the system
will first transmit all WZ frames. Then, the time taken for the system to complete
processing the GOP till the complete reconstruction of the output video sequence
is recorded. After that, the same procedure is repeated with the system is now
transmitting only the odd-numbered WZ frames. To ensure that the results are
compared fairly, the same video sequence shall be used for both scenarios.
Prior to transmission from the encoder, all frames in the GOP shall be indexed
with a number. The first frame (key frame), F0 shall always be sent. On the other
hand, the p-th WZ frame, Fp will only be transmitted if and only if p is an odd
number. Therefore, p will first be verified using the (30.3).

q ¼ p % 2: ð30:3Þ

The Fp is only transmitted if and only if q 6¼ 0.

The RS function, RS (n, k, m) has the following parameters: message length, k,
codeword length, n, and number of bits per symbol, m. Table 30.1 summarizes the
meaning of each parameter and the allowable positive integers of the RS codes.
In this work, eight bits are chosen for each symbol, m = 8. The data rate, r is 15/
31, with n = 31 and k = 15. For Galois field of the form of GFð2m Þ, the primitive
element, b is 2. The method used to encode the message using the RS (n, k, m) is by
varying the primitive polynomial of the Galois field that contains that symbol, using
an input argument in Galois field as given by the Matlab function in (30.4).

encode ¼ rsencðmsg; n; kÞ: ð30:4Þ

This function encodes the message, msg with the RS (n, k) codes using a narrow
sense generator polynomial [5]. The message is Galois array of symbol having
30 Distributed Video Coding with Frame Estimation at Decoder 305

m-bits each. Each k-element row of msg represents a message word, with the
leftmost symbol as the most significant symbol. At most, the codeword length,
n ¼ 2m  1. However, if n is not as exact as 2m  1, shortened RS codes will be
used. Parity symbols are added at the end of each word in the output Galois codes.
Another Matlab function (30.5),

decode ¼ rsdecðmsg; n; kÞ: ð30:5Þ

will decode the received message, msg with the narrow sense generator polyno-
mial [5]. The codes used at the decoder are a Galois array of symbols with m-bits
each. Each n-element row of codes represents a corrupted systematic codeword,
with the leftmost symbol as the most significant symbol.
If there are more thanðnk Þ
2 errors detected in a row of the received message, a
decoding failure occurs. In this case, the function in (30.5) forms a corresponding
row of decoded codeword by merely removing (n – k) symbols from the end of the
row of the received message.
To determine whether decoding is successful, the Hamming distance, between
the received message and the SI generated using the previously decoded code-
word, is calculated. If the Hamming distance is non-zero, then the decoder pro-
ceeds to the next iteration to request more parity bits from the encoder via the
feedback channel. On the other hand, if the Hamming distance is zero, the
decoding operation will then be verified with the 8-bits cyclic redundancy check
(CRC) sum. Should the CRC sum computed on the decoded plane match the value
received from the encoder, decoding is considered successful and the decoded
codeword is sent for reconstruction.
The missing even-numbered WZ frames are estimated with the interpolation
method. An even-numbered WZ frame must exist between two consecutive odd-
numbered WZ frames which had been decoded successfully previously. Each pixel
value in the even-numbered frames is interpolated from the same points of the two
dimensional grid of the consecutive

odd-numbered WZ
frames.
The pixel value, f xp ; yp at the coordinate xp ; yp , of a p-th WZ frame can be
estimated using the linear interpolation equation in (30.6).

1


f xp ; yp ¼ f xp1 ; yp1 þ f xpþ1 ; ypþ1  f xp1 ; yp1 : ð30:6Þ
2

where 2 B p B 6 and p is even number. This function is iterated for a number of

times equal to a frame size to reconstruct the whole frame.

30.4 Simulation Result

The DVC model proposed in section III is completely implemented with the
C++ programming language. Both encoding and decoding functions of RS codes
are imported from the Matlab library and converted to the C++ programming
306 K.H. Chiam and M.F.M. Salleh

Table 30.2 PSNR Values fOR Each Frame

Frame number PSNR (dB)
Transmits only odd WZ frames Transmits all frames
0 36.1473 36.1477
1 36.3032 36.3022
2 29.4138 36.3240
3 36.2582 36.2575
4 30.9052 36.2931
5 36.2789 36.2795
6 28.9724 36.2923
7 36.2521 36.2524

language. All calculations to determine the odd or even frame number and to
estimate the missing even-numbered frames follow (30.3) and (30.6) respectively,
and are written in the C++ programming language.
The model is evaluated with the Hall monitor QCIF video sequence. This video
sequence has very low amount of motion activity.
Each reconstructed output frame in the GOP is compared with the original input
frame. Based on the comparison, the PSNR is calculated using (30.7).

2552
PSNR ¼ 10 log10 : ð30:7Þ
d

where d is the distortion between the two frames. Table 30.2 and Fig. 30.4 present
the simulation results.
From the data, it can be seen that if all frames are transmitted, the PSNR value
for each frame is above 36 dB. This also proves that there is very good decoding of
the output frames at the decoder although there is no prior information available at
the decoder. This finding matches the results in [1].
For the simulation with only odd-numbered WZ frames transmission, the odd-
numbered WZ frames those are transmitted are decoded perfectly with the similar
PSNR values like the simulation with all frames transmission. However, those
nontransmitted even-numbered WZ frames which are estimated at the decoder
show poorer PNSR values. This shows that the linear interpolation at the decoder
is still unable to estimate the frames perfectly like the conventional inter-frame
decoding.
Nevertheless, the PSNR values of the estimated frames are still close to the
acceptable threshold values of 30 dB. As this input video sequence is of low
motion activity, the temporal redundancy between the frames is more. Hence, the
linear interpolation method can be a good approximation of the missing frames.
The simulation, where all frames are transmitted, needs an average time of
3447 s to finish processing the video sequence. On the other hand, the simulation,
30 Distributed Video Coding with Frame Estimation at Decoder 307

Fig. 30.4 PSNR values for 40

each frame
30
Transmits

PSNR (dB)
only odd
20 WZ frames

Transmits
10 all frames

0
0 1 2 3 4 5 6 7
Frame Number

where only odd-numbered WZ frames are transmitted, only needs 3286 s to finish
processing the same video sequence. There is a reduction of about 4.90 % pro-
cessing time.
Longer processing time might due to more frames need to be transmitted and be
put waiting in the buffer. Moreover, the inter-frame decoding requires the parity
bits request from the encoder and also the SI from decoded key frames. Hence, the
more WZ frames are to be decoded, the longer the waiting time for the parity bits
request.
Contrary, interpolation method only needs two consecutive odd-numbered
frames to estimate the missing even-numbered frames. All these information is
readily available at the decoder and the decoder does not need to wait any
information from the encoder.

30.5 Conclusion and Further Work

In this paper, the model is designed such that only the odd-numbered WZ frames
in the GOP are transmitted across the network. The missing even-numbered WZ
frames are recovered at the decoder with estimation using the interpolation
method. Although this algorithm is less efficient in estimating the frames com-
pared to the conventional inter-frame decoding method, the proposed model
involves less complicated calculation and is able to produce the result in a shorter
time. Nevertheless, the estimated frames at the decoder are also of acceptable
quality compared to the corresponding input frames.
To improve the quality of the estimation of the missing even-numbered WZ
frames, interpolation with motion estimation can be applied at the decoder. In [4],
the authors did research about the 8 9 8 bidirectional motion estimation. This
method can achieve better rate distortion (RD). The algorithm shall also be included
in the proposed model to investigate whether there is improvement to the RD.
The proposed model shall also be tested with video sequences of different
motion activity for RD comparison. This is because the RD performance of the
308 K.H. Chiam and M.F.M. Salleh

DVC is inconsistent with different video streams, as stated in [10]. Lower or

medium activity video sequences give better RD performance, contrast to video
sequences of significant motion activity. For this reason, the low motion activity
video sequence of Hall sequence is selected for this work.

References

1. Aaron, A., Thang, R., Girod, B.: Wyner-Ziv coding of motion video. In: Proceedings of
Asilomar Conference on Signals and Systems, Pacific Grove, CA, pp. 240–244 (2002)
2. Aaron, A., Setton, E., Girod, B.: Towards practical Wyner-Ziv coding of video. In:
Proceedings of IEEE International Conference on Image Processing ICIP, Barcelona, Spain,
pp. 869–872 (2003)
3. Artigas, X., Ascenso, J., Dalai, M., Klomp, S., Kubasov, D., Ouaret, M.: The discover codec:
architecture, techniques, and evaluation. In: Proceedings of Picture Coding Symposium
(PCS). doi: 10.1.1.131.1402
4. Girod, B., Aaron, A., Rane, S., Rebollo-Monedero, D.: Distributed video coding. In:
Proceedings of IEEE, Special Issue on Video Coding and Delivery, pp. 71–83 (2005) Invited
Paper
5. Matlab Help, Matlab v7.4.0.287, The Math Works, Inc (R2007a)
6. Maggs, B.: Decoding Reed-Solomon codes [Online]. Available: http://www.cs.duke.edu/
courses/spring10/cps296.3/decoding_rs_scribe.pdf Accessed 24 Oct 2000
7. Puri, R., Ramchandran, K.: PRISM: a new robust video coding architecture based on
distributed compression principles. In: Proceedings of Allerton Conference on
Communication, Control, and Computing, Monticello, IL (2002)
8. Riccardo, B., Roberto, R., Pamela, Z.: Distributed video coding: principals and evaluation of
wavelet-based schemes. In: Sudhakar R (ed.) Effective Video Coding for Multimedia
Applications. Intech, Rijeka, Crotia, pp. 111–115 (2011)
9. Shu, L., Daniel, J., Error Control Coding: Fundamentals and Applications. Prentice Hall,
Eaglewood Cliffs (1983)
10. Varodayan, D., Aaron, A., Girod, B.: Rate-adaptive codes for distributed source coding.
J. Sig. Proc. 86, 3123–3130 (2006)
Chapter 31
Performance Analysis of an OCDMA
System Based on SPD Detection Utilizing
Different Type of Optical Filters
for Access Networks

Sarah G. Adbulqader, Hilal A. Fadhil, S.A. Aljunid

and Anuar Mat Safar

Abstract This paper aims to study the performance of spectral-amplitude coding

optical code-division multiple-access (SAC-OCDMA) systems based on single
photo-diode (SPD) detection under various types of optical filters decoder schemes
(Optical Gaussian filter, Thin film and Fiber Bragg Grating (FBG)). In our work,
we utilized Modified Double weight (MDW) code as one of the SAC codes; the
proposed system is performed by using single photodiode for each user. The
results characterizing the bit-error-rate (BER) of 10-12 with respect to the data rate
show that the SDP offers a significant improved performance for long haul
applications (wide area network (WAN)) over other types of SAC-OCDMA
detection techniques. Furthermore, the FBG filters have higher dispersion than
Gaussian and thin film filters. The comparison states that the optical Gaussian
filters have better filter amplitude response leading to lower dispersion bandwidth
reduction, while FBG filters have higher dispersion which could reduce the goal of
622 Mbps channels in SPD detection scheme.

31.1 Introduction

There has been a huge interest in applying Optical Code Division Multiple Access
(OCDMA) techniques to fiber optic communication access networks. This tech-
nique is one of the multiple access schemes that are becoming popular because
their advantages such as the flexibility in the allocation of channels, ability to
operate asynchronously, enhanced privacy and increased capacity in bursty
networks [1–3]. Moreover, single photo-diode (SPD) detection based on OCDMA

S.G. Adbulqader (&)  H.A. Fadhil  S.A. Aljunid  A.M. Safar

School of Computer and Communications Engineering, University Malaysia Perlis,
Kangar, Perlis, Malayasia
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 309

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_31
310 S.G. Adbulqader et al.

systems have been investigated recently to apply for high speed Wide Area
Network (WAN) as multiple users allow to access network simultaneously [1–3].
Recently, many detection techniques have been proposed for SAC-OCDMA
systems such us AND-detection technique, modified AND, and spectral direct
detection techniques [3–6]. However, among all SAC-OCDMA detection tech-
niques and based on the previously published papers at [3, 5]. The SPD detection
technique has been proven as a good solution for ultra-high speed transmission,
Multi-Access Interference (MAI) suppression and cost-effective [3]. In an optical
access networks, the OCDMA signals pass through a number of optical comple-
ments such as multiplexer, de-multiplexer, add-drop multiplexer, and passive
optical router. However, each of these components include optical filter to separate
and select the wavelengths. Choosing a proper optical filter for an OCDMA system
can increase the efficiency of the system by reducing the effective bandwidth
making the system less susceptible to noise, filter passband misalignment and
dispersion effects. Therefore, in this study it is desirable to investigate the different
type of optical filters (Fiber Bragg Grating (FBG), thin film and optical Gaussian
filters) in order to be applicable with the various access networks such as MAN,
and WAN. This paper is organized as follows: Sect. 31.2 provides a detailed
explanation of our proposed system. Next, in Sect. 31.3, we focus on performance
analysis followed by the discussion of the results. Finally, Sect. 31.4 summaries
our conclusions.

31.2 System Design

31.2.1 Modified Double Weight Code

The code structure is based on Modified Double –Weight (MDW) code families for
SAC-OCDMA systems. The MDW codes have a large number of weight can be
developed based on double weight (DW) code of weight two, the MDW code is the
modified version of DW code [6]. The MDW code possesses ideal cross-correlation
properties and exists for every natural number [5, 6]. However, the MDW code
weight can be any even number that is greater than 2. Moreover, the MDW codes
can also be represented by using a (K 9 N) matrix as shown in Fig. 31.1. The
details of code structure and code parameters have been presented in [6].
Figure 31.1 shows that we can increase the number of user from 1 to 3 while the
weight is still fixed at 4. An MDW code with weight of 4 denoted by (N, 4, 1) for
any given code length N, can be related to the number of user K through:
  2
3 kp
N ¼ 3K þ sin ð31:1Þ
8 3
31 Performance Analysis of an OCDMA System … 311

Fig. 31.1 The basic MDW code with code length 9, weight 4, and an ideal in-phase cross
correlation

31.2.2 SPD Detection Based Gaussian Optical Filters

The proposed SAC-OCDMA receiver diagram of this technique is shown in

Fig. 31.2. The received optical signal is decoded by the decoder, which has an
identical spectral response to the intended encoder for the data to be received. The
remainder of the signal from the decoder is then transmitted to the subtractive
decoder (s-Decoder) to cancel out signals with mismatched signatures, i.e.,
interferers. The output from the s-Decoder is either zero power unit for active user
or cross-correlation power unit for interferers. The proposed technique can be
performed using inexpensive optical Gaussian filter to decode the received signal.
Moreover, other types of filters such us Fiber Bragg Grating (FBG) and thin film
filters are also used as the main part in the corresponding SPD implementations.

Fig. 31.2 SPD detection based on MWD code

312 S.G. Adbulqader et al.

After optical subtraction, the output is either code weight power unit for active
user or zero power units for interferers. This implies that the interference signals
are suppressed in the optical domain before the conversion of the signals to the
electrical domain, as a result, the proposed SPD scheme alleviates both PIIN and
MAI in the optical domain [3]. Moreover, the two interference signals at the
optical subtractor are assumed to be equal and cancel each other out. However,
practically, the interference signals differ slightly at the optical subtractor and
results in a small amount of optical power to reach the photodiode. The main
advantage of using the SPD is that the cancellation of the interference signals in
the optical domain allows the use of only a single photodiode rather than two
photodiodes as in typical subtraction detection schemes [7]. This reduces the
amount of optical-to-electrical conversion and shot noise generated at the receiver
part. The proposed detection technique can also be implemented with any fixed in-
phase cross-correlation SAC codes with differ spectral chips distribution of the s-
Decoder, depending on the structure of the SAC codes itself. Finally, after the
desired signal is detected by a photodiode, the data-carrying electrical signal is low
pass-filtered by a Bessel-Thompson filter [8].
The signal-to-noise (SNR) for an electrical signal is defined as the average
signal to noise power SNR = [I2/i2t ], where i2t is the variance of noise source (note:
the effect of the receiver’s dark current and nonlinear noises are neglected in the
analysis of the proposed system), given by
     
r2 ¼ i2shot þ i2PIIN þ i2thermal ; ð31:2Þ

Equation (31.2) can be expressed as

4Kb Tn B
r2 ¼ 2eBI þ I 2 BsC þ ; ð31:3Þ
RL

where the symbols used in Eq. (31.3) bear the following meaning.
e Electron charge;
I Average photocurrent;
I2 The power spectral density for I;
B Electrical bandwidth;
Kb Boltzmann Constant;
Tn Absolute receiver noise temperature;
RL Receiver load resistor.

The formula used to calculate the bit-error-rate (BER) with Gaussian approx-
imation can be expressed as [7, 8]
rffiffiffiffiffiffiffiffiffi!
1 SNR
BER ¼ Pe ¼ erfc ; ð31:4Þ
2 8

where erfc is the complementary error function

31 Performance Analysis of an OCDMA System … 313

31.2.3 Different Decoders Architecture Schemes

Different decoder architecture scheme are based on different type of filters used in
the SPD implementation. In this sub-section only three type of filter will be
investigated and analysed which are the Optical Gaussian filter, thin film filter, and
FBG filter based Gaussian apodization profile. It is well known that the apodized
and linearly chirped Bragg gratings reduced the side lobe level in the reflectivity
response and also the group delay response ripple [9]. The FBG based Gaussian
apodization profile can be presented as [9]:


z 2
TðzÞ ¼ exp G ð31:5Þ
l

The parameters z and l are used to control the apodization sharpness parameter.
Also, the optical Gaussian filter is based on Gaussian distribution expressed as:
 
1 1 2
TðzÞ ¼ exp z ð31:6Þ
2p 2

The thin film filters are known to have a Butterworth response and are com-
monly modeled as 3rd order Butterworth filters [10]. However, the different
decoder architecture schemes have the different impact; it would be very useful to
investigate and to analyse the different filter functions to find out most significant
one for dispersion compensation (long haul applications) and access network
categories as well. Table 31.1 shows the system parameters used in the simulation
model (OptisystemTM), these parameters based on the previously published papers
at Hamza et al., Norazimah et al., and Ghafouri-Shiraz [3, 5, 8].

31.3 Simulation Results

Figure 31.3 shows the average BER for three users against the fiber length at data
rates of 622 Mbps and 155 Mbps. It can be seen that the average BER value
increases with the increasing of the transmission length. Further, for a fiber length
of 55 km the BER are 1 9 10-12 and 1 9 10-18 for the data rate of 622 and
155 Mbps, respectively. Moreover, the FBG based Gaussian apodization of 622
Mbps allows short transmission in the fiber length as compared with data rate of
155 Mbps. However, this type of filter possesses easily controllable dual-wave-
length narrow transmission peaks. It provides a simple and low cost approach of
achieving the dual-wavelength fiber laser operation [9].
Figure 31.4 shows average BER for three users against the data rate at trans-
mission distance of 30 and 50 km. It can be seen that the average BER value
increases with the increasing of data rate for both distances. Moreover, for a given
314 S.G. Adbulqader et al.

Table 31.1 System parameters used in the simulation

Wavelength 1,550 nm
Thermal noise 1.8 9 10-23 W/Hz
Dark current 5 nA
Responsively 1 A/W
Optical bandwidth B0 = 3.75 THz
Attenuation 0.25 dB/km
No. of user 3
Code weight 4
SPD filter FBG, thin film, and Gaussian

Fig. 31.3 BER versus fiber 10

-5

length at data rate 622,155 155Mbps

622 Mbps
Mbps for FBG based
-10
Gaussian adopization 10

-15
10
BER

-20
10

-25
10

-30
10
10 15 20 25 30 35 40 45 50 55 60
fiber length (km)

Fig. 31.4 BER versus data 0

10
rate at different fiber length 30 km
50 km
for FBG based Gaussian -5
10
adopization
-10
10
BER

-15
10

-20
10

-25
10

-30
10
1 2 3 4 5 6 7 8 9
8
data rate (Mbps) x 10
31 Performance Analysis of an OCDMA System … 315

Fig. 31.5 BER versus fiber (a) 10-10

length for thin film filter and GAUSSIAN FILTER
THIN FILM FILTER
Gaussian filter at data rate of -15
a 155 Mbps; b 622 Mbps 10

-20
10

BER
-25
10

-30
10

-35
10

-40
10
60 70 80 90 100 110 120 130 140 150 160
fiber length (km)
(b) 10
-5
GAUSSIAN FILTER
THIN FILM FILTER

-10
10

-15
10
BER

-20
10

-25
10

-30
10
60 70 80 90 100 110 120 130 140 150
fiber length (km)

data rate at transmission distance of 30 km the proposed system offers better

performance in terms of BER than at 50 km. This result indicates that there is a
significant improvement in performance or, for a fixed BER, could accommodate a
higher data rate for greater capacity.
Figure 31.5 shows average BER against the fiber length at different data rate of
155 and 622 Mbps. It can be seen that the average BER value increases with the
increasing of the transmission length. Further, for a BER of 1 9 10-12 with data
rate of 155 Mbps, optical Gaussian filter allows extra transmission of 20 km in the
fiber length and better performance as compared with thin film filter, whereas for a
BER of 1 9 10-12 with data rate of 622 Mbps, Gaussian filter allows longer
transmission of 12 km in the fiber length as compared with thin film filter. This
improvement can be explained as the optical Gaussian filters offers passband sharp
which exhibits the lowest loss at the peak compared with thin film filter.
316 S.G. Adbulqader et al.

Fig. 31.6 Comparison (a) 10-5

between thin film filter and GAUSSIAN FILTER
THIN FILM FILTER
Gaussian filter at fiber length 10
-10

a 70 km; b 120 km
-15
10

-20
10

BER
-25
10

-30
10

-35
10

-40
10
1 2 3 4 5 6 7 8
data rate (Mbps) x 10 8

(b) 10-5 GAUSSIAN FILTER

THIN FILM FILTER
-10
10

10-15

10-20
BER

10-25

10-30

10-35

10-40
1 2 3 4 5 6 7
data rate (Mbps) x 108

Figure 31.6 shows average BER against the data rate at different fiber length of
70 and 120 km. It can be seen that the average BER value increases with the
increasing of transmission data rate. Further, for a BER of 1.5 9 10 - 12 with a
fiber length of 70 km allows higher transmission rate than a fiber length of 120 km.
However, the optical Gaussian filter is considered an ideal solution to achieve long
haul transmission (120 km) for SAC-OCDMA system with minimized dispersion
affects compared with thin film filter. Moreover, an SAC-OCDMA system is
considered an ideal solution for different transmission rates on the access links [11].
The Eye pattern diagrams for the various proposed filters are shown in
Fig. 31.7. Figure 31.7 clearly shows the dispersion effects by FBG based Gaussian
adopization compared with other filters.
31 Performance Analysis of an OCDMA System … 317

Fig. 31.7 Eye diagram for the data rate of 622 Mbp at fiber length of 60 km of one of the: a FBG
based Gaussian adpoziation BER = 1.1 9 10-10; b Optical Gaussian filter, BER = 4.5 9 10-15;
c thin film filter BER = 4.7 9 10-9

Table 31.2 Applications of proposed system based different decoder architecture

Application Decoder type Fiber length (km) BER
WAN Gaussian filter 140 &10-12
WAN Thin film filter 130 &10-12
MAN FBG Gaussian 55 &10-12

Table 31.2 depicts that our proposed system based on different decoder
architecture schemes can be applied to interconnection ranging from localized
links within an equipments rack to links that span continents or Fiber-to-Building
(FTTB) applications. As Table 31.2 illustrates, networks are divided into two
broad categories based on filter types and transmission distance (WAN and MAN).
318 S.G. Adbulqader et al.

31.4 Conclusions

In this study, we have tested optical Gaussian, FBG and thin film filters to assess
the suitability of these filter types as decoding devices in OCDMA networks using
SPD detection. The FBG filters have higher dispersion than Gaussian or thin film
filters (Thin film filters and Gaussian filter both have low dispersion). Thin film
and Gaussian filters have better filter amplitude response leading to lower band-
width reduction, while FBG filters have higher dispersion which could reduce the
goal of 622 Mbps channels in SPD detection scheme. The properties of this system
are described and discussed with the related equations. However, based on the
simulation results, these filter can be implemented in the SPD decoder architecture
at different access categories (WAN, MAN) as a specific BER of 10-12 is required
in order to achieve different quality-of-service (QoS) requirements. Furthermore,
the proposed study is employed for access network applications could be an
excellent candidate for applying in the next generation OCDMA networks.

References

1. Tarhuni, N.G., Korhonen, T.O., Mutafungwa, E., Elmusrati, M.S.: Multiclass optical
orthogonal codes for multiservice optical CDMA networks. J. Lightwave Technol. 24(2),
694–704 (2006)
2. Kwong, W.C., Guu-Chang, Y.: Design of multilength optical orthogonal codes for optical
CDMA multimedia networks. IEEE Trans. Commun. 50(8), 1258–1265 (2002)
3. Hamza, M.R., Al-Khafaji, : Reducing BER of spectral-amplitude coding optical code
division multiple-access systems by single photodiode detection technique. Europ. Opt. Soc.
Rap. Public. 8, 13022 (2013)
4. Fadhil, H., Aljunid, S., Badlisha, R.: Triple-play services using random diagonal code for
spectral amplitude coding OCDMA systems. J. Opt. Commun. 30(3), 155–159 (2009)
5. Norazimah, M.Z., et al.: Performance comparison of different detection techniques in Long-
Haul Fiber SAC-OCDMA systems. In: 3rd International Conference on Photonics, Malaysia
(2012)
6. Aljunid, S.A., et al.: A new code For optical code division multiple access systems. Malays.
J. Comput. Sci. 17(2), 30–39 (2004)
7. Prucnal, P.: Optical Code Division Multiple Access: Fundamentals and Applications. Taylor
and Francis, Boca Raton (2006)
8. Ghafouri-Shiraz, H.: Optical CDMA Networks, Principle, Analysis and Applications. Wiley,
Chichester (2012)
9. Sher Shermin, A., et al.: Determination of the best apodization function and grating length of
linearly fiber bragg grating for dispersion compensation. J. Commun. 7(11), 840–846 (2012)
10. Fadhil, H., Aljunid, S., Ahmed, B.: Performance of OCDMA Systems using random diagonal
code for different decoders architecture schemes. Int. Arab J. Inf. Technol. 7(1), 7 (2010)
31 Performance Analysis of an OCDMA System … 319

11. Fadhil, H.A., et al.: Multi-rate transmissions on spectral amplitude coding optical code
division multiple access system using random diagonal codes. Optica Applicata XXXIX(2)
(2009)
12. Khaleghi, S., Pakravan, M.R.: Quality of Service Provisioning in Optical CDMA Packet
Networks. J. Opt. Commun. Netw. 2(5), 283–292 (2010)
Chapter 32
Deployment of Optimized Algorithm
for MPEG-4 Data Over Wireless
Multimedia Sensor Network

Norlezah Hashim, Sharifah Hafizah Syed Ariffin, Farizah Yunus,

Fakrulradzi Idris and Norsheila Fisal

Abstract Typical Wireless Sensor Network (WSN) always deals with scalar data
such as temperature. These types of data are suitable for low rate networking
technology such as IEEE802.15.4. Transmitting a video for IEEE 802.15.4 raised
other challenges for bandwidth limited sensor networks like WSN. The optimi-
zation method able to determine the optimal limit for quantization scale, group of
picture and frame per second in order to suit the WSN environment. Changing
these parameters affects the bandwidth requirement and video quality in term of
Peak Signal to Noise Ratio (PSNR). This project aims to create an embedded code
in TelG mote according to this concept. Results from experiment show
improvements in packet delivery ratio of 28 % for Akiyo qcif file, 27 % for
Foreman qcif file and 9 % for Mobile qcif file. This work proved that the method
has successfully increased the network performance.

32.1 Introduction

Analysis in Wireless sensor network (WSN) often includes the study of network
properties of wireless communication with small battery powered sensors. Now-
adays some of WSN applications include disaster prevention, environmental
monitoring, and logistics tracking. Most WSN uses IEEE 802.15.4 standard, is one

N. Hashim (&)
Faculty of Engineering Technology, Universiti Teknikal Malaysia Melaka,
Hang Tuah Jaya, 76100 Melaka, Malaysia
e-mail: [email protected]
S.H. Syed Ariffin
F. Yunus
N. Fisal
UTM-MIMOS Center for Telecommunication Technology, UTM, 81310 Skudai, Malaysia
F. Idris
Faculty of Electronics and Computer Engineering, Universiti Teknikal Malaysia Melaka,
Hang Tuah Jaya, 76100 Melaka, Malaysia

 Springer International Publishing Switzerland 2015 321

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_32
322 N. Hashim et al.

of the interest due to its capability in producing low cost with low complexity but
still maintaining the good quality of video. However, transferring a video is more
challenging since it requires bigger bandwidth to carry bigger data and WSN only
allow the maximum bandwidth of 250 kbps. Therefore a compression method such
as MPEG-4 is needed before a video being transferred through WSN to reduce the
requirement of its bandwidth.
The researchers in [1] have developed the benchmark for the settings of video
encoding which can provide better video quality suitable for WSN application.
They mentioned three parameters, namely quantization scale, group of picture, and
frame per second that are important in determining the quality of a video received.
The researchers used Network Simulator 2 (NS2) as an approach to the study.
Even though simulation is a reliable approach to study the network properties, the
real scenario of WSN is much more complicated. Therefore, a real world testbed
implementation is needed to justify the results obtained using simulation. In this
paper, the design, deployment and evaluation of WSN testbed using optimized
algorithm to transmit MPEG-4 data will be discussed. The deployment will be
done using TelG mote and WiseOS operating system. Results on packet delivery
ratio and total packet loss at different distances are presented. Moreover, com-
parisons of results between MPEG-4 data with optimized algorithm and MPEG-4
data with non-optimized algorithm are further elaborated. Lastly, comparisons of
results between experiment and simulation are presented to the readers.
The rest of this paper is organized as follows; Sect. 32.2 briefly describes the
design concept of optimized parameter for MPEG-4 data, followed by testbed
implementation in Sect. 32.3. In Sect. 32.4, the results obtained are elaborated and
finally the conclusions are given in Sect. 32.5.

32.2 Design Concept

The standard for video compression which are MPEG-1 [2] and MPEG-2 [3], are
not versatile enough to efficiently address the requirement of multimedia appli-
cation. Therefore, Moving Picture Expert Group (MPEG) developed MPEG-4
standard that provides a platform for a wide range of multimedia applications [4].
In [5, 6], MPEG-4 video is already proven suitable to be transmitted over IEEE
802.15.4 standard. Many researchers focused research on MPEG-4 transmitted for
IEEE802.11 standard that can support high bandwidth, such as in [7, 8], and only a
few researches done for MPEG-4 over IEEE802.15.4.
In optimized method, the video need to be encoded without quantization scale
with a default value of 30 frames per second and 30 groups of frame. Next, the
bandwidth of video is checked and if it does not meet the requirement condition of
WSN which is 250 kbps; the video will be quantized with the increasing value of
the quantization until 31. If the bandwidth still does not meet the requirement, the
value for frame per second will be reduced until it meets the bandwidth require-
ment. Next the PSNR is checked to ensure the video is still in good quality after
32 Deployment of Optimized Algorithm for MPEG-4 Data … 323

Fig. 32.1 TelG system architecture

several adjustments have been made to the video. In order to get better video
quality, the PSNR values need to be increased by decreasing the group of frame
value while maintaining the bandwidth requirement. This step will be repeated
until the optimal value for group of frame, frame per second, and quantization
scale is obtained but still within the bandwidth requirement of WSN and good
PSNR value.

32.2.1 TelG System Architecture

Figure 32.1 shows TelG system architecture used in testbed deployment. TelG
hardware system architecture consists of sensing unit, communication unit, pro-
cessing unit, and power unit [9]. TelG is particularly designed where program-
ming, computation and communication are integrated onto a single device. Each
TelG can be reprogrammed on the board using ISP.
In sensing unit, the electrically erasable programmable read only memory
(EEPROM) is used to store MPEG-4 data before transmitting. ATmega644PV
microcontroller from Atmel Corporation is used in TelG as the processing unit.
ATmega644PV has its own advantage in term of low power consumption used
[10]. In the communication unit, the wireless device is using the XBee module
324 N. Hashim et al.

Fig. 32.2 TelG device

Reset and
User
USART connector Buttons Power connector

LED

Programming XBee
connector

from MaxStream Inc. XBee is an IEEE 802.15.4 compliant radio based on Carrier
Sense Multiple Access CSMA. TelG is battery powered, with a cutoff voltage of
1.8 V [10].
WiseOS was developed by a group of researchers from Telekom Research
Group (TRG) of UTM. The programming language used for WiseOS is called
nesC, which is C with some addition language features for components and
concurrency. WiseOS is an event-driven operating system. Figure 32.2 shows
TelG device used in experiment.

32.2.2 Experiment Setup

The experiment location is chosen at UTM block P19 as shown in Fig. 32.3.
Figure 32.4 shows the configuration setup which consists of three TelG points
spaced equally among them. TelG node A acts as the source node, TelG node B is
the intermediate node and base station behaves as the sink node. Besides, base
station also collects all data and sent them to a Personal Computer for analysis
purposes.

32.3 Methodology

The video frame format used in the testbed is known as Quarter Common Inter-
mediate Format (qcif) where this format has a resolution of 176 9 144 pixels that
represents different color spaces [1]. Three types of video samples used in the
testbed are Akiyo, Foreman and Mobile. These video samples appear as low,
medium and high motion and scene complexity respectively. These video samples
are commonly used among researchers. Table 32.1 shows the summary of the data
used in this research where six different sets of MPEG-4 data were conducted in
32 Deployment of Optimized Algorithm for MPEG-4 Data … 325

Fig. 32.3 Experiment setup at P19, UTM

Laptop
Base TelG TelG
Station node B node A

Two hops

Fig. 32.4 Configuration setup

Table 32.1 Summary of the data used in experiment

Type of motion Quantization Frame per Group of Bandwidth
scale second frame
Optimized Akiyo 2 30 20 248
Foreman 4 15 10 229
Mobile 4 5 10 249
Non- Akiyo 2 30 10 293.5
optimized Foreman 3 15 10 310.7
Mobile 3 5 10 324.8

the testbed. It can be seen when any value from the optimal value is varied, will
result in the change of bandwidth requirement.
Before making any measurement, each TelG needs to be programmed
according to the algorithm to test the proposed method. Figure 32.5 shows the
overall flowchart of the proposed method. The flow chart is divided two parts.
326 N. Hashim et al.

st Start
1 part

Encode the video without

quantization scale with default value
(fps=30 and GOP=30)

BW< NO Increase quantization scale

250kbps (maintain fps and GOP)

YES
YES BW<
250kbps

(PSNR> 37=excellent), NO
(PSNR, 31-37=good)
Maintain GOP=30
and decrease fps < 30

YES
NO

Increase PSNR, decrease GOP

and maintain the BW <
250kbps

nd
2 part Set Frame sequence {x} where (x=1,2,...n)
Set Frame Type {kx} where (x=1,2,..n)
Set Frame Size {yx} where (x=1,2,..n)
Set Encoding Time {zx} where (x=1,2,..n)
Set [count]=0

Set [packet] =0
Total Packet = Frame Size[count]/70

Send video [packet]

NO
Succeed receive
[packet] = [packet]+1
packet?

YES

YES
[packet]<Total
Packet?
NO

[count]=[count]+1

NO

[count]> n ?

YES

End

Fig. 32.5 Project flowchart

32 Deployment of Optimized Algorithm for MPEG-4 Data … 327

The first part is to select the optimal values for frame per second, quantization
scale and group of frame while the second part is to transfer the video packets
through WSN channel.

32.4 Result and Discussions

Figure 32.6 shows the packet delivery ratio for three samples namely Akiyo,
Foreman, and Mobile qcif data. In general, the packet delivery ratio was showing a
decreased pattern when the distances between nodes are increased. As the dis-
tances are increased, it is reasonable to experience more losses because the
probability a data to loss is increased as the signal strength is decreased. From the
graph, packet delivery ratio for Akiyo optimized data improves an average of
about 28 % compared to non-optimized data, 27 % differences for Foreman and
9 % differences for Mobile.
It is observed from the graph, the improvement for Foreman is smaller than
Akiyo as Foreman contains more packets which is twice the Akiyo’s total packets.
This will increases the probability of packets dropping since the network is getting
busy with packets transferring. From Table 32.1, it can be seen that in order to
maintain the network bandwidth requirement not more than 250 kbps and to
maintain the video quality in term of PSNR, Foreman’s file contains a smaller
group of frame compared to Akiyo’s generated from the optimized method.
Smaller group of frame increased the quantity of I frames, which constitutes to
bigger file size and lower packet delivery ratio.
For Mobile, the optimized data improves the network of about 9 % in averages
compared to non-optimized data. Compared to Foreman and Akiyo, Mobile has
the smallest improvement in term of packet delivery ratio. Mobile file represents a
high motion and scene complexity of a horizontally moving toy train with a
calendar moving vertically in the background. Therefore, Mobile file’s size is
bigger than Akiyo and Foreman which results in lowest packet delivery ratio.
In order to maintain the bandwidth requirement of Mobile below the maximum
allowable rates of WSN, the value of quantization scale is increased to 4 (refer to
Table 32.1). As we know quantization scale represents how much compression is
done to the video. Increasing quantization scale will reduce the requirement of
bandwidth but will introduce a lower video quality. However, by using the opti-
mized method, the quality of the video is still maintained at a good PSNR value
above 31 according to Mean Opinion Score (MOS) [1].
The packet loss versus distance graphs for Akiyo, Foreman and Mobile were
shown in Fig. 32.7. In overall, packet loss was showing an increased pattern when
the distance between nodes are increased. From this figure, it is clearly observed
that optimized MPEG-4 data improved the overall network performance for
Akiyo, Foreman and Mobile by reducing the total packet loss.
328 N. Hashim et al.

Fig. 32.6 Packet delivery 1.2

ratio for Akiyo, Foreman, and

Packet Delivery Ratio

mobile qcif data 1

0.8
akiyo optimized
0.6 akiyo non-optimized
foreman optimized
foreman non-optimized
0.4
mobile optimized
mobile non-optimized
0.2

0
0 5 10 15 20
Distance (meter)

Fig. 32.7 Packet loss for

Akiyo, Foreman, and mobile
qcif data

Akiyo qcif shows the highest improvement with the ability to reduce the packet
loss to an average of 2 % compared to non-optimized data where the average
packet loss is 31 %. The average packet loss for Foreman qcif is 12 % compared to
the average packet loss for non-optimized data which is 38 %. For Mobile qcif, the
average packet loss is 46 % compared to non-optimized data which is 55 %. There
are higher possibilities to decode optimized Akiyo and optimized Foreman com-
pared to optimized Mobile since the average packet loss for both files is smaller.
When packet loss is small, the probability of receiving most of I frame which
serves the most important information are higher. However, for non-optimized
data, it is almost impossible to decode the data as the packet losses are very high.
Packet losses are more critical as the distances are increased.
Figure 32.8 shows the comparison of results between simulation using NS2 and
experiment. From the graph, it can be seen the results observed from experiments
are more or less the same with simulation. The average differences between
experiment and simulation results are 2.3 % for Akiyo, 11.5 % for Foreman, and
10.3 % for Mobile qcif.
32 Deployment of Optimized Algorithm for MPEG-4 Data … 329

Fig. 32.8 Simulation versus

experiment packet delivery
ratio for Akiyo, Foreman and
mobile qcif data

32.5 Conclusions and Future Works

In this work, we have deployed two hops wireless sensor network testbed and
investigated its performance. The results show that the optimized algorithm for
MPEG-4 data improves packet delivery ratio and reduces packet losses in WSN.
However, among the three MPEG-4 qcif data, the highest improvement is
observed for Akiyo, followed by Foreman and Mobile.
In future, the optimized method is suggested to be tested on TelG using
802.11 g WiFi module. The maximum allowable bandwidth in WiFi is bigger in
802.11 g compared to 802.15.4 which will further improve the network
performance.

Acknowledgments The authors would like to thank Ministry of Higher Education (MOHE),
UTM-MIMOS Center for Telecommunication Technology, Universiti Teknologi Malaysia
(UTM) and Research Management Center (UTM-RMC) for the financial support of this project
under GUP research grant no. Q.J130000.2523.04H39. The first author would also like to thank
Universiti Teknikal Malaysia Melaka (UTeM) for providing the scholarship.

References

1. Farizah, Y., Sharifah, H.S.A., Sharifah, K.S.Y., Nor Syahidatul N.I., Abdul Hadi, F.A.H.,
Norsheila, F.: Optimum parameters for MPEG-4 data over wireless sensor network. Int.
J. Eng. Technol. (0975-4024), 5(5), 13p, (2013) 4501-4513
2. Information technology—Coding of Moving Pictures and Associated Audio for Digital
Storage Media at up to about 1.5 Mbit/s—Part 3: Audio, International Standard ISO/IEC
11172–3:1993 (1993)
3. Information Technology—Generic Coding of Moving Pictures and Associated Audio: Part
2—Video, International Standard ISO/IEC 13818–2:1995 (1995)
4. Information technology—Coding of audio-visual objects—Part 4: Conformance Testing,
International Standard ISO/IEC 14496-4:2004 (2004)
330 N. Hashim et al.

5. Antonio Javier, G.S., Felipe Garcia, S., John Garcia, H.: Feasibility study of MPEG-4
transmission on IEEE 802.15.4 networks. In: International Conference on Wireless and
Mobile Computing, Networking and Communication (2008)
6. Antonio Javier, G.S., Felipe Garcia, S., John Garcia, H.: A Cross-layer solution for enabling
real-time video transmission over IEEE 802.15.4 Networks. In: Multimedia Tools and
Application (2010)
7. Pilgyu, S., Kwangsue, C.: A Cross-layer based rate control scheme for MPEG-4 Video
transmission by using efficient bandwidth estimation in IEEE 802.11e. In: International
Conference Information Networking ICOIN (2008)
8. Zemin, M., Jinhe, Z., Tonghai, W.: A cross-layer QoS scheme for MPEG-4 streams. In:
Wireless Communications, Networking and Information Security (WCNIS), 2010 IEEE
International Conference, pp. 392–396 (2010)
9. Rozeha, A.R., Norsheila, F., Abdul Hadi, F.A.H.: Wireless multimedia sensor network
platform for low rate image/video streaming. Jurnal Teknologi. Special Edition
(Instrumentation and Sensor Technology) 54, 1–12 (2011)
10. Abdul Hadi Fikri, F.A.H., Rozeha, A.R., Norsheila, F., Sharifah, K.S.Y., Sharifah, H.S.A.:
Development of IEEE802.15.4 based wireless sensor network platform for image
transmission. Int. J. Eng. Technol. 9(10), 112–118 (2009)
Chapter 33
Partially Compensated Power Control
Technique for LTE-A Macro-Femto
Networks

Sawsan Ali Saad, Mahamod Ismail and Rosdiadee Nordin

Abstract Femtocells are expected to remarkably enhance network capacity,

indoor coverage and introduce brand-new services. However, in co-channel
deployment scenarios, and with the femtocell being applied as a closed subscriber
group (CSG) access, femtocells can cause severe interference to the neighbouring
cells and derive users to the outage. In this paper a downlink power control scheme
without additional signalling exchange is proposed, which adjusts the transmit
power subject to user’s pathloss measurements. However the minimum level of
power transmission is constrained to the acquired target SINR of femtocell user
that could be set according to the QoS requirements. The performance is confirmed
via system level simulations. The results show that, the proposed scheme reduces
the outage probability of the macrocell users, and the spectral efficiency of fem-
tocell users is improved. Furthermore, reducing the transmit power level helps in
saving more power towards green femtocell networks.

33.1 Introduction

The concept of femtocell has attracted much attention in the wireless communi-
cation industry; as a solution for indoor coverage problems. Femtocell also known
as Home enhanced NodeB (HeNB) in the third generation partnership project
(3GPP) long term evolution-advanced (LTE-A) standard; has many attractive

S.A. Saad (&)  M. Ismail  R. Nordin

Department of Electrical, Electronic and Systems Engineering,
National University of Malaysia, 43600 Bangi, Malaysia
e-mail: [email protected]
M. Ismail
e-mail: [email protected]
R. Nordin
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 331

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_33
332 S.A. Saad et al.

features of plug-and-play, low deployment cost, traffic offload from macrocell and
providing enhanced indoor coverage with high data rates [1].
The transmit power level of a femtocell base station affects its coverage range and
the amount of interference it generates in the network. Properly selecting the fem-
tocell base station transmit power level can help manage the interference from the
femtocells to the macro-users, while maintaining femtocells performance. Although
femtocells are low power base stations, the massive deployment of them will result
in increasing the total power consumption of the network. Therefore, efficient
methods, such as cell zooming, sleep mode and power control are required to reduce
the power consumption while maintaining the performance of femtocells [2].
A method that discriminates between indoor and outdoor is presented in [3].
The location information gained from the discrimination procedure together with
the required SINR for the HUE is used to perform power control for the HeNB.
The proposed scheme outperforms the random power scheme by providing higher
SINR for the home user equipment (HUEs). However the performance gap
between the proposed scheme and the random method has decreased in environ-
ments with higher shadowing effect. This is due to the decreased accuracy of
positional state discrimination.
The power control problem is formulated in [4] considering the worst case sce-
nario, assuming no dominating interferer. A heuristic distributed algorithm is exe-
cuted to determine the optimum power level, if no feasible solution for a sub-channel
the algorithm determines the admissible subset of users on the given sub-channel by
eliminating the interferers causing excessive interference level. The scheme delivers
suboptimal results which are shown to be close to the optimal. However the effect of
the user mobility is neglected as stationary network is assumed.
The co-channel deployment of femtocells with macrocells is investigated in [5].
A macro user (MUE) assisted HeNB power control scheme is proposed; that
adjusts the transmit power of the HeNB when receiving an interference message
from an MUE. Two timers are used to control the decrease and increase of the
transmit power. The scheme can reduce the outage probability for the victim
MUEs and also avoid unnecessary throughput loss of the HeNBs. However the
MUE needs to send interference messages to HeNB, yet there is no direct con-
nection between HeNB and MUE, which implies possible delay and low reliability
in transmitting the control information.
This paper proposes a technique to mitigate the downlink interference in
co-channel multi-tier macro-femto networks, while maintaining the performance of
the HUEs. The scheme allows the implementation of the concept: coverage follows
HUE; without requiring HUE positioning technique and hence saving the trans-
mission of such information. Instead the position of the HUE is estimated using the
pathloss between the HUE and the HeNB; the pathloss is partially compensated,
which helps more in mitigating the interference to the neighbouring cells. However,
the minimum level of power transmission is set adaptively to guarantee the required
target SINR of the HUE which in consequence conserves the QoS requirements.
The performance of the scheme is evaluated via system level simulation for the
metrics of SINR, outage probability and the spectral efficiency.
33 Partially Compensated Power Control Technique … 333

The rest of this paper is organized as follows: Sect. 33.2 presents the system
model. Section 33.3 provides details about the proposed power control scheme and
describes the simulation scenario. Section 33.4 is dedicated to the performance
evaluation results and discussion. Concluding remarks are given in Sect. 33.5.

33.2 System Model

The system model is used to study the interference impact on the performance
MUEs and indoor HUEs, in terms of the SINR, the outage probability of MUEs
and the spectral efficiency, due to the introduction of the power control approach
for femtocell base stations.

33.2.1 Interference Scenario

In the downlink, the received signal at the user equipment (UE) contains the
OFDMA transmitted symbols of the serving base station plus the interference
induced by nearby femto and macro base stations. Figure 33.1 shows the inter-
ference scenario used in this study. Two types of interference are considered: (i)
Cross-tier interference between macrocell and femtocell, in which the cell-edge
MUE may suffer from severe interference. (ii) Co-tier interference between the
femtocells.

Fig. 33.1 Interference scenario

334 S.A. Saad et al.

33.2.2 Propagation Models and SINR Computing

In this study we adopt the 3GPP LTE-Advanced path loss models for urban
deployments [6], in which the path loss between the eNB and the UE is calculated
as follows:
(
15:3 þ 37:6log10 R(outdoor UE)
PLmacro ðdBÞ ¼ ð33:1Þ
15:3 þ 37:6log10 R þ Low (indoor UE)

where R is the distance between the UE and eNB in meters and Low is the pene-
tration loss of an outdoor wall, which is 20 dB.
While the path loss between HeNB and UE within or outside an apartment for
5 9 5 grid scenario, is formulated as:

PLfemto ðdBÞ ¼ 127 þ 30log10 (R=1000) ð33:2Þ

The received SINR of the MUE in the downlink can be expressed as [7]:

PTm gmj
SINRMUE ¼ P P ð33:3Þ
PTk gkj þ PTf gfj þ Nth
f 2F
m 6¼ k
k2M

where PTm, PTk and PTf are the transmit power of serving and interfering MeNB
and HeNB respectively, M and F is the set of eNBs and HeNBs respectively, gmj,
gkj and gfj are the link gain between MUE j and the eNB and HeNB. Nth is the
thermal noise.
The downlink received SINR of the HUE is given by [7]:

PTf gfh
SINRHUE ¼ P P ð33:4Þ
PTk gkh þ PTm gmh þ Nth
k2F
k 6¼ f

where PTf and PTk are the transmit power of the serving and interferer HeNBs
respectively.
33 Partially Compensated Power Control Technique … 335

33.2.3 Spectral Efficiency Calculations

The spectral efficiency of a channel can be approximated by the attenuated and

truncated form of Shannon bound [8]. Given a particular SINR, the spectral effi-
ciency can be determined by the following:
8
< Thr ¼ 0SINR\SINRmin
>
Throughput; Thr ½bps=Hz ¼ Thr ¼ a:SðSINRÞSINRmin \SINR\SINRmax
>
:
Thr ¼ Thrmax SINR [ SINRmax
ð33:5Þ

where S(SINR) = log2(1 + SINR) is Shannon bound, a is the attenuation factor,

SINRmin and SINRmax are the minimum and maximum SINRs supported by the
available AMC scheme. Thrmax is the maximum spectral efficiency. The
parameters a, SINRmin, SINRmax and Thrmax are set to 0.6, -10dB, 23 dB and
4.4 bps/Hz in this study [8].

33.3 Proposed Power Control Scheme

The basic requirement for a HeNB is to provide a strong enough signal for its
HUEs. On the other hand the transmit power should not be too large, as to create
strong interference to the neighbouring femtocells or MUEs. In the fixed power
setting scheme (i.e. no power control) all the HeNBs transmit at their highest
available transmit power without considering any surrounding information.

33.3.1 Analytical Model

In this paper, we propose a distributed HUE-assisted based power control scheme.

In which the transmit power (Ptx) is adjusted according to the distance between the
HUE and the HeNB as in (33.6):
(  K ) !
d
Ptx ¼ min max Ptmax  ; Ptmin ; Ptmax ð33:6Þ
R

where d is the distance between the HUE and HeNB. R is the radius of the
femtocell. Ptmax, Ptmin are the maximum and minimum transmitting power of
HeNB respectively. Ptmin depends on the SINR target related to the QoS
requirements of the HUE. K is an exponent that controls the dynamic range of
power control.
336 S.A. Saad et al.

However, it is more convenient and practical to estimate the parameter d from

the reported path loss between HUE and the corresponding HeNB, without
acquiring HUE position information. The path loss can be given as follows:

PL ¼ xðf Þ þ Nlog10 ðdÞ þ xr ð33:7Þ

where x(f) represents the dependence of the path loss on the frequency. N is a
coefficient related to the type of environment. xr denotes the shadow fading, is a
Gaussian random variable with zero mean r2 variance.
The distance (d) can be expressed as an exponential function:
1
d ¼ 10N ðPLxðf Þxr Þ ð33:8Þ

This can be written as:

1
d ¼ 10N PUE ð33:9Þ

Considering (33.8), the formula (33.5) can be modified in the following

manner:
  h 1 iK  
ðPUE Pmax Þ
Ptx ¼ min max Ptmax  10 N ; Ptmin ; Ptmax ð33:10Þ

The Ptmin is calculated so as to guarantee the target SINR of the HUE. Fur-
thermore the pathloss is partially compensated [9] as in (33.11):

Ptmin ¼ maxðminðSINRtar þ Iserv þ aPLserv ; Ptmax HeNB Þ; Ptmin HeNB Þ ð33:11Þ

where Iserv is the average link interference, PLserv is the the pathloss between HUE
and the serving HeNB, a is a fractional compensation parameter between 0 and 1
note that if a = 1, then we have full pathloss compensation. Ptmin_HeNB is the
minimum allowed power that could be transmitted by HeNB which is set to 0 dBm
as in 3GPP standard.

33.3.2 Algorithm Description

The proposed scheme makes use of local status information of the HeNB; obtained
measurement reports from connected HUEs. The HeNB utilizes this information
and based on the defined algorithm routines, makes a local decision.
Each HUE has a target SINR (SINRtar) that satisfied the quality of service
(QoS) for him. The minimum transmit power of the HeNB is set so as to guarantee
SINRtar for the HUE. To eliminate the burden of frequent Ptx update according to
33 Partially Compensated Power Control Technique … 337

the mobility of all the attached HUEs, Ptx update is associated with the arrival of
new HUE. However, there is also a periodical update if the is no new user for a
dedicated period of time. Two routines are used to calculate the transmit power: (i)
Cal_PWR_SINRtar, in which Ptx is calculated according to (33.10). (ii) Dec_-
power, in which the Ptx is decreased by a step D Ptx that is equal to the difference
between the current SINR of the HUE and his SINRtar. The algorithm for the
proposed scheme is shown in the following pseudo code.
338 S.A. Saad et al.

Table 33.1 Simulation parameters

Parameter Macrocell Femtocell
Cell radius 500 m 10 m
HeNB transmitter power 46 dBm 20 dBm
HeNB antenna gain 14 dBi 5 dBi
Log normal shadowing standard deviation 8 dB 10 dB
Carrier frequency 2.0 GHz
Bandwidth 10 MHz
Antenna pattern Omnidirectional
Thermal noise -174 dBm/Hz

33.3.3 Simulation Scenario

The downlink macrocell/femtocell scenario as in Fig. 33.1 is considered with

seven eNB at the centre of the macrocell. A typical 5 9 5 grid scenario of a dense
urban area for femtocells with 25 houses each house has the size of 10 m 9 10 m
is located at the edge of the centred macrocell. Twenty HeNBs are located ran-
domly and uniformly in the houses. The closed access method is considered for
HeNBs. This means only closed subscriber group (CSG) is allowed to connect to
the femtocell. HUEs are dropped randomly and uniformly in the femtocells. The
SINR target range for HUEs is -4 * 20 dB. Table 33.1 shows the main simu-
lation parameters confirmed with 3GPP TR 36.814 [10].

33.4 Results and Discussion

The outage probability of MUEs is defined as the ratio of MUEs whose SINR
bellow -6 dB to the total number of MUEs [12]. Figure 33.2 shows the average
outage probability of MUEs when conducting the simulation with different sce-
narios of no power control is applied and with different values of a for power
control. For a = 1 (i.e. full compensation for pathloss) the reduction in outage is
not significant, however, with small (a) a clear improvement in MUE outage is
obtained. As we consider the closed access mode for femtocells; the less the HeNB
transmit power, the more the interference reduction could be obtained and in
consequence the MUE outage is minimized.
The proposed scheme not only reduced the cross-tier interference, but fur-
thermore the co-tier interference between femtocells is mitigated and that would
explain the enhancement in the spectral efficiency (SE) of the HUEs despite the
reduction of the transmit power as in Fig. 33.3. For small (a) the HUE SE is still
better than no power control scenario, but it is less than when a = 1.
33 Partially Compensated Power Control Technique … 339

Fig. 33.2 Average outage

probability of MUEs

Fig. 33.3 Average spectral

efficiency of HUE

As a result of mitigating the cross-tier interference, the signal quality is

improved for MUEs as shown in Fig. 33.4. Higher MUE SINR is obtained with a
= 0.5. Improving SINR for MUEs indicates the ability to enhance the MUE
throughput.
As the transmit power of the base station represents the major part that con-
tributes to the power consumption. Applying the proposed scheme provides power
consumption reduction of 17 % with a = 1 compared to no power control scenario
where all the HeNBs transmit at their maximum power. While a power reduction
of 86.5 % is obtained with a = 0.5. These savings are necessary for deploying the
future femtocells in a cost effective and green way (Fig. 33.5).
340 S.A. Saad et al.

Fig. 33.4 Average SINR of

MUEs

Fig. 33.5 Average HeNB

transmit power

33.5 Conclusion

In this paper a power control scheme based on HUE measurements is proposed. In

this scheme, the transmit power of the HeNB is adjusted based on the location of
the HUE that is estimated from the path loss measurements for simplicity. Fur-
thermore, the minimum level of power transmission is constrained to the target
SINR that could be set according to the QoS requirements. In addition the mini-
mum transmit power level is calculated based on partial compensation for the
33 Partially Compensated Power Control Technique … 341

pathloss which helps more in mitigating the interference to the neighbouring cells.
The system level simulation results show that the proposed scheme has reduced the
outage probability for the MUEs. Moreover, minimizing the transmit power,
according to the position of HUE helps to reduce the power consumption for
greener deployment of femtocells while maintaining good performance for HUEs
in terms of spectral efficiency. Our next work will focus on finding the optimal
value of a that compromise between the outage of the MUEs and the SE of the
HUEs.

Acknowledgments This research was supported by The National University of Malaysia under
grant DPP-2013-006 and Ministry of Science, Technology and Innovation Malaysia under grant
01-01-02-SF0788.

References

1. Chandrasekhar, V., Andrews, J.G., Gatherer, A.: Femtocell networks: a survey. IEEE
Commun. Mag. 46(9), 59–67 (2008)
2. Saquib, N., Hossian, E., Lee, L.B., Kim, D.I.: Interference management in OFDMA femtocell
networks: issues and approaches. IEEE Wireless Comm. Mag. 19(3), 86–95 (2012)
3. Cho, K.-T., Kim, J., Jeon, G., Ryu, B.H., Park, N.: Femtocell power control by discrimination
of indoor and outdoor users. Proceedings of the IEEE Wireless Telecommunications
Symposium (WTS), pp. 1–6 (2011)
4. Akbudak, T., Czylwik, A.: Distributed power control and scheduling for decentralized
OFDMA networks. In: Proceedings of International ITG Workshop on Smart Antenna,
pp. 59–65 (2010)
5. Wang, Z., Xiong, W., Dong, C., Wang, J., Li, S.: A Novel downlink power control scheme in
LTE heterogeneous network. In: Proceedings of IEEE International Conference on Compu-
tational Problem-Solving (ICCP), pp. 241–245 (2011)
6. GPP tech. rep. 36.814. v.9.0.0. Mar. 2010
7. Wang, M., Zhu, X., Zeng, Z., Wan, S., Li., W.: System performance analysis of OFDMA-
based femtocell networks. In: Proceedings of IET International Conference on Communication
Technology and Application (ICCTA), pp. 405–410 (2011)
8. GPP tech. rep. 36.942. v.11.0.0. Sept. 2012
9. Rao, A.M.: Reverse link power control for managing inter-cell interference in orthogonal
multiple access systems. In: Proceedings of IEEE 66th Vehicular Technology Conference
(VTC), pp. 1837–1841 (2007)
Chapter 34
Design and Development of the Visible
Light Communication System

Anuar Musa, Mazlaini Yahya, Nazaruddin Omar,

Mohd Kamarulzamin Salleh and Noor Aisyah Mohd Akib

Abstract This paper presents the development of a Visible Light Communications

(VLC) system for a local area network (LAN) based on the IEEE 802.15.7 standard.
The proposed VLC system enables new indoor applications by providing simul-
taneous illumination and networking capability between the optical wireless
devices. The VLC system uses an AVAGO Technologies HSDL-4230 LED as the
transmitter and an OSRAM SFH203FA photodiode as the receiver. The distance in
which the optical link was measured was up to 1 meter and achieved about 9.2 Mb/s
throughput. Therefore, it is suitable for most of the applications for the home or
office environment such as Internet communication, multimedia streaming, etc.

34.1 Introduction

In recent years, the interest in optical wireless as a favorable complementary

technology for the radio frequency (RF) communication has gained significant
momentum with substantial deployments using solid state lighting technology and
the released of the IEEE 802.15.7 draft standard [1]. The IEEE 802.15.7 compliant
networks consist of two types of nodes, namely, a coordinator which initializes and

A. Musa (&)
M. Yahya
N. Omar
M.K. Salleh
N.A.M. Akib

TM Innovation Centre, TM Reserach and Development Sdn. Bhd., Lingkaran Teknokrat
Timur, 63000 Cyberjaya, Selangor, Malaysia
e-mail: [email protected]
M. Yahya
e-mail: [email protected]
N. Omar
e-mail: [email protected]
M.K. Salleh
e-mail: [email protected]
N.A.M. Akib
e-mail: [email protected]

Ó Springer International Publishing Switzerland 2015 343

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_34
344 A. Musa et al.

peer-to-peer star broadcast

coordinator

device devices

Fig. 34.1 IEEE 802.15.7 standard defines three types of topologies namely peer-to-peer, star,
and broadcast

manages the network and a remote device which communicates with each other
via the coordinator.
The standard defines three types of topologies consisting peer-to-peer, star, and
broadcast as shown in Fig. 34.1. The standard also defines the first two layers of
ISO/OSI stack protocol that includes Physical and Medium Access Control (MAC)
layers. IEEE 802.15.7 MAC is a Carrier Sense Medium Access/Collision
Avoidance (CSMA/CA) based protocol.
The basic concept of the VLC is simple. It uses a visible light medium to carry
data through the air or space. Likewise, the VLC link architectures are very similar
to an optical fiber communication links, without the optical fibers, lights are
deployed as a transmission medium. The VLC is also similar to the RF wireless
transmission with the radio waves replaced with a light-wave medium and the
antennas are replaced with the free-space optical transceivers.
Despite this superficial similarity between the VLC and the RF links, the VLC
exhibits several appealing attributes when compared to the RF system. The VLC
links are inherently broadband. The optical frequencies are in the light-wave
infrared and visible spectrums that are neither regulated nor required license. The
optical components are also cheaper and consume less electrical power than the
high-speed RF components. In fact, the VLC links do not suffer from severe
multipath fading and interferences like the RF transmission since the transmission
beam is focused and requires direct line-of-sight (LOS) transmission. These
advantages do not, however, imply that the VLC is a universal replacement for the
RF communications system.
The applications of the VLC systems are quite limited when considering area
coverage and user mobility where the RF technologies are proven invaluable. In
addition, the VLC systems operate under strict eye safety regulations. The VLC
receiver is less sensitive than their RF counterparts because of their photo-electric
conversion mechanisms and the impact of ambient light noise sources.
Substantial researches have been conducted into the VLC technology and have
created a number of test beds presented in the journals [2–4]. Many of these test
34 Design and Development of the Visible Light Communication System 345

beds do not follow IEEE 802.15.7 standard. In this paper, we propose to transmit a
standard LAN 10/100BASE-T (10/100 Mb/s) Ethernet signal over a VLC link as a
proof-of-concept for future homes, offices, hospitals as well as for providing an
optical wireless LAN (VLC-LAN) for the end users.
This paper focuses on the design and the development of the IEEE 802.15.7
system by combining both lighting and communication attributes. This paper also
will outline the requirements to achieve an effective VLC-LAN system and doc-
uments the properties of the proposed system. The rest of the paper is organized as
follows. In Sect. 34.2, we present the system design and implementation. System
performance is discussed in Sect. 34.3. Finally, we present our conclusions in
Sect. 34.4.

34.2 System Design and Implementation

The proposed proof-of-concept (POC) system architecture is shown in Fig. 34.2.

The VLC system uses an AVAGO Technologies HSDL-4230 [5] LED as the
transmitter and an OSRAM SFH203FA [6] photodiode as the receiver.
The proposed transceiver block diagram is outlined in Fig. 34.3. At the trans-
mitter end, the output of the laptop passes the packets through a modem, as an
interface between the laptop and the optical transceiver and to the amplifier. To
prevent flickering and brightness changes due to the transmission of other mes-
sages than the idle pattern, such as data messages, we use Manchester Coding [7]
to distribute the occurrences of ‘1’ and ‘0’ more evenly; the human observer will
perceive an always-on optical signal unnoticeable of very high speed flickering.

Fig. 34.2 High level system architecture for the full-duplex visible light communication (VLC)
system test bed
346 A. Musa et al.

Fig. 34.3 Block diagram of the receiver and the transmitter for the uplink. Similar block
diagram for the downlink with minor different in LED and Photodiode due to different
wavelength used to avoid interference. Tables 34.1 and 34.2 provide the key implementation
parameters

The amplifier amplifies the signal before sending the signal to the LED driver.
The LED converts the electrical signal into optical signal for transmission. At the
receiver, a photodiode (PD) followed by a transimpedance amplifier (TIA) and an
amplifier are used to recover the input signal. The PD shown in the receiver circuit
(Fig. 34.3) is OSRAM SFH203FA. These devices offer a fast response and a large
7 mm active area useful for home environment where the practical distance is
more than a few centimeters. At the end of the signal recovery, a limiting amplifier
limits the receiver output signal from clipping effect before entering into the
modem.
Tables 34.1 and 34.2 provide the specifications of the hardware used for the
implementations of the transmitter and the receiver for the uplink and downlink
channels, respectively.

Table 34.1 Keys Parameters of the Uplink VLC System

Device Property Value
LED Infrared LED HSDL 4230 [5]
Power 75 m/sr
Beam width 17°
Voltage 2.5 V
Current 100 mA
Size 3 mm
Photo diode Photo sensitivity 0.59 A/W
Spectral response range 800-1100 nm
Peak sensitivity wavelength 900 nm
Active Area 1 mm
34 Design and Development of the Visible Light Communication System 347

Table 34.2 Keys Parameters of the downlink VLC System

Device Property Value
LED Pure white STAR LED [8]
Power 75 m/sr
Beam width 127°
Voltage 3.25 V
Current 700 mA
Size 8 mm
Photo diode Photo sensitivity 0.46 A/W
Spectral response range 340–1040 nm
Peak sensitivity wavelength 760 nm
Active area 7 mm

34.3 System Evaluation

This paper provides the design and implementation of the VLC system for the desk
lamp environment. When the system is deployed to the end-user, the information
source can be as far away as 100 cm (desk lamp to receiving plane) in range. The
actual system setup is shown in Fig. 34.4.
The UDP throughputs as function of packet size for the VLC system are shown
in Fig. 34.5. The UDP throughput increases with packet size and reaches the
maximum rate at 9.2 Mb/s. This is expected as the payload increase the UDP/IP
overhead decreases; thus, the average throughput will also increase.

Fig. 34.4 Visible Light

Communications (VLC) test
bed for the desk lamp
environment that is capable
of transmitting and receiving
a HD quality video
348 A. Musa et al.

Fig. 34.5 Measured UDP

throughput per packet size for
VLC system in a standard
office environment with
different distances

34.4 Conclusion

This paper has presented a POC for a VLC LAN link. The distance achieved is one
meter, which makes possible for a LAN internet link from an LED based desk
lamp to the PC connection as a complementary technology to Wi-Fi. The link is
successfully achieved by using a VLC driving circuit that consists of a series of
amplifiers and feeds directly to the HSDL-4230 LED device. The incident light
passes through the SFH203FA photodiode device, allowing 9.2 Mb/s communi-
cations. Future work will extend the POC with the design and development of the
MAC layer to support multi-users for the communication system for practical
usage in the homes and the offices environment.

References

1. Wireless Medium Access Control (MAC) and Physical Layer (PHY) specifications, IEEE
802.15.7-2011 draft8 standard
2. Elgala, H., Mesleh, R., Haas, H.: Indoor broadcasting via white LEDs and OFDM. IEEE Trans.
Consum. Electron. 55(3), 1127–1134 (2009)
3. Omega: The home gigabit access project. http://www.ict-omega.eu/, viewed 12-01-2012
4. Schmid, S., Gorlatova, M., Giustiniano, D., Vukadinovic, V., Mangold, S.: Networking Smart
Toys with Wireless ToyBridge and ToyTalk. In: Poster Session, Infocom (2011)
5. HSDL-4230 datasheet. http://www.farnell.com/datasheets/95278.pdf
6. SFH203FA datasheet. http://www.farnell.com/datasheets/1672046.pdf
7. Proakis, J.G.: Digital Communications. McGraw-Hill Series in Electrical and Computer
Engineering, New York (2001)
8. STAR-LED datasheet. http://www.farnell.com/datasheets/1444587.pdf
Chapter 35
The Embroidered Antenna on Bending
Performances for UWB Application

M.S. Shakhirul, A. Sahadah, M. Jusoh, A.H. Ismail

and Hasliza A. Rahim

Abstract The performances of embroidered UWB wearable textile antenna due to

the effect in bend conditions is discussed in this paper. The radiating element of the
antenna is made of silver plated nylon thread by using embroidery technique,
while the antenna substrate is cotton fabric and Nora dell as a ground plane. This
UWB antenna is designed for operating frequency from 3.1 up to 10.6 GHz with a
center frequency of 6.85 GHz. This research demonstrates the antenna bending
effect at seven different angles in terms of the simulated and measured reflection
coefficient.

35.1 Introduction

Nowadays, a lot of investigations on technology textile antenna to carry out the

wearable monitoring system. The monitoring is a necessary for risky activities or
environment such as mining, diving, mountain climbing as well as another sort of
military applications. Moreover, the wearable antennas will be attached on the
body or into clothing or may be worn as a button antenna. Hence, all these broad

M.S. Shakhirul (&)  A. Sahadah  M. Jusoh  A.H. Ismail  H.A. Rahim

School of Computer and Communication Engineering, Universiti Malaysia Perlis
(UniMAP), Kampus Pauh Putra, 02600 Arau, Perlis, Malaysia
e-mail: [email protected]
A. Sahadah
e-mail: [email protected]
M. Jusoh
e-mail: [email protected]
A.H. Ismail
e-mail: [email protected]
H.A. Rahim
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 349

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_35
350 M.S. Shakhirul et al.

applications of monitoring with data transmission function can be achieved by

using wearable antennas that do not force the wearer to abandon the comfort zone
with such compact and durable materials [1].
The commercial use of frequency bands from 3.1 to 10.6 GHz for ultra-wide-
band (UWB) system has been approved by the FCC in 2002 [2]. With the
advantage of life long battery, the UWB reduces the wearable device size since it
uses low power signal to transmit data [3]. Therefore, this research applies the
UWB textile technology with an embroidery technique for WBAN application.
The worn antenna needs to be low profile and hidden for the convenience of the
user. Furthermore it is impossible to keep the wearable antennas in flat condition
all the time because the antennas will bend quite often due to human body
structure and movement. The bending could affect the performance characteristics
of the antennas as its resonant length altered [4]. Therefore, an investigation of the
effects antenna performance characteristics due to bending condition carried out in
this paper.

35.2 Materials and Methods

The cotton fabric is used as substrate materials in this project instead of using rigid
circuit boards. While the patch antenna is made of silver plated nylon thread and
the conductive fabric ‘‘Nora dell’’ is used as a ground plane. The designing process
of the antenna required the specification of the substrate materials such as
dielectric constant, er and the thickness, while the conductive required electrical
permittivity. Table 35.1 shows the details of specification materials used in this
paper [4].

35.2.1 Circular UWB Patch Antenna

A circular shape antenna is mostly suited to the specification of UWB application

which provides several advantages such as omni-directional, low transmit power
and large channel capacity [5, 6, 7]. Hence, this research proposed a circular
antenna designed by Eq. (35.1), where a is radius of the circular patch antenna in
millimeter; fr is the resonance frequency and er is the relative permittivity of the

Table 35.1 Specification of materials used

Materials Thickness (mm) Values
Cotton fabric 0.5 er = 1.6
Conductive fabric ‘‘Nora dell’’ 0.13 1.538e + 6
Silver plated nylon thread 0.105 –
35 The Embroidered Antenna on Bending Performances … 351

Fig. 35.1 Optimum dimension of antenna designed (The left sight is a front view while the right
side is a back view of the antenna)

textile substrate material [1]. The resonance frequency and dielectric constant of
this paper is 6.85 GHz and 1.6 respectively.

87:94
a¼ ð35:1Þ
frðpffiffiffi
erÞ

All designs and simulations have been carried out using the CST simulation tool,
however some modification has been made to suit the requirements. Figure 35.1
shows the geometry and the dimension for the proposed patch antenna.

35.2.2 Antenna Design in Bend Conditions

In this project, the bending effect of UWB textile antenna is investigated for seven
different angles; 20, 30, 40, 50, 60, 70 and 80. The cylinder is used as
reference structure of the antenna in order to be bent. Hence the Eq. (35.2) is used
to find the outer radius of the cylinder [7].

S ¼ rh ð35:2Þ

Where: S = Arc length,

r = Radius of the circle,
h = Measure of the central angle in radians.
The width of the substrate antenna was used as the arc length S is 60 mm.
Based on the Eq. (35.2), the cylinder radius value of the intended angle was
calculated in Table 35.2.
352 M.S. Shakhirul et al.

Table 35.2 The UWB antenna calculation for bending perpose

20 bending

h ¼ 20p
180 ¼ 0:349 (degree in radian)
r ¼ Sh ¼ 0:349
60
¼ 171:92 mm (radius of cylinder)
30 bending 40 bending

r = 114.59 mm r = 85.94 mm
50 bending 60 bending

r = 68.75 mm r = 57.29 mm
70 bending 80 bending

r = 49.11 mm r = 42.97 mm

35.2.3 Antenna Fabrication

The embroidered UWB textile antenna is fabricated using a computerized sewing

machine. The particular machine has sewn the silver plated nylon thread on the
cotton fabric. In embroidery technique, thread is sewn on the substrate will
be penetrated at the back of the substrate. Hence, another layer of cotton will be
inserted as to separate between the conductive thread and Nora dell as ground
plane which means there are two layers of cotton fabric.
The process is continued by sewn the second layer of cotton with the ground
plane denotes as ‘‘Nora dell’’ fabric. The first layer cotton is the embroidery
circular patch antenna while the second layer cotton is sewn with the ground plane
of ‘‘Nora dell’’. The special glue called silver conductive epoxy is used to solder
35 The Embroidered Antenna on Bending Performances … 353

Fig. 35.2 Prototype of an embroidered UWB antenna. a Front view b Back view

the SMA connector with the embroidery antenna. Figure 35.2 shows the prototype
of UWB embroidered antenna.

35.3 Results and Discussions

All measurements are done in free space using the Programmable Network
Analyzer functioned between 100 and 20 GHz. The antenna reflection coefficient
performance is evaluated based on the bending condition.

35.3.1 Simulated Result of UWB Antenna in Bending

Conditions

The simulated reflection coefficient is done between 2.8 and 11 GHz under seven
different bending angles; 20, 30, 40, 50, 60, 70 and 80. Figure 35.3 shows

Fig. 35.3 The simulated

comparison of S-Parameter
result for antenna in bends
condition
354 M.S. Shakhirul et al.

Fig. 35.4 The measured comparison of S-Parameter result for antenna in bends condition

the simulated comparison of S11 (dB) result for antenna in bends condition. The
antenna shows the similar pattern of return loss at several degrees of antenna
bends. At 70 degree bending, the antenna achieved a significant reflection coef-
ficient of -46.75 dB at frequency 9.65 GHz. The antenna at 70 also has a lower
resonance where the reflection coefficient is -12.53 dB at frequency 7.7 GHz.

35.3.2 Measured Result of UWB Antenna in Bending

Conditions

Figure 35.4 shows the measured comparison of S-parameter result for antenna in
bends conditions. By referring to the measured S-parameter result of antenna in
bend condition, the antenna bends at 60 and 70 have a similar pattern of reflection
coefficient follow by antenna bend at 20. The antenna in these bends conditions
which is 20 (green line), 60 (red line) and 70 (pink line), shown the better
reflection coefficient matching. While the others bend angles shared a similar
reflection coefficient pattern. However, the proposed antenna still can performed
under tolerable reflection coefficient of less than -10 dB.

35.4 Conclusion

This paper presents the development, fabrication and analysis of the embroidered
UWB antenna performance between frequency ranges of 3.1 GHz to 10.6 GHz.
The investigation focused on the bending effect towards the antenna reflection
coefficient result. There are seven bending angles considered; 20, 30, 40, 50,
60, 70 and 80. The result shows that the antenna has different performances
35 The Embroidered Antenna on Bending Performances … 355

according to the bending angles. Regardless of the bending angles, both simulated
and measured results have slightly difference in reflection coefficient. The higher
the bending angles, the better the impedance matching of the proposed antenna.

Acknowledgments Acknowledgements are expressed to Research and Development (RND)

Centre, School of Computer and Communication Engineering and University of Malaysia Perlis
for providing the lab facilities and short term grant which enabled the publication of this article.

References

1. Osman, M.A.R., Rahim, M.K.A., Samsuri, N.A., Elbasheer, M.K., Ali, M.E.: Textile UWB
Antenna bending and wet performances. Int. J. Antennas Propag. 2012(251682) 12 (2012)
2. Sanz-Izquierdo, B., Huang, F., Batchelor, J.C.: Convert dual-band wearable button antenna.
Electron. Lett. 42(12668–12670) 2006
3. Osman, M.A.R., Rahim, M.K.A., Azfar, M., Kamardin, K., Zubir, F., Samsuri, N.A.: Design
and analysis UWB wearable textile antenan. In: Proceedings of the 5th European Conference
on Antenna and Propagation (EUCAP) (2011)
4. Shieldex website: http://itp.nyu.edu/*kh928/sensorreport/ShieldexNoraDell.pdf
5. Klemm, M., Troster, G.: Textile UWB antenna for on-body communications. In: Proceedings
of ‘EUCAP 2006, Nice, France. 6–10 Nov 2006
6. Soh, P.J., Vandenbosch, G.A.E., Javier H.-O.: Design and evaluation of flexible CPW-fed
Ultra Wide Band (UWB) Textile antennas. In: IEEE International RF and Microwave
Conference (RFM 2011), 12–14 Dec 2011
7. Dey, S., Saha, N., Biswas S.: Design and Performance analysis of UWB Circular disc
monopole textile antenna and bending consequences. In: Proceedings of the 5th European
Conference on Antennas and Propagation (EUCAP), 11–15 Apr 2011
Chapter 36
The Embroidered Wearable Antenna
for UWB Application

M.S. Shakhirul, A. Sahadah, M. Jusoh, A.H. Ismail, C.M. Nor

and F.S. Munirah

Abstract This paper discusses the design of a wearable textile antenna for Ultra
Wide Band (UWB) application. This embroidered antenna addresses the issues of
miniature size, wide bandwidth and low power consumption. A textile cotton has
been chosen as a substrate and silver nylon plated yarn as a conductive element
embroidered on cotton substrate. Simulated and measured results show that the
proposed antenna design meets the requirements of wide working bandwidth with
compact size and flexible material. The performances in terms of reflection
coefficient, impedance bandwidth, current distribution as well as gain and antenna
efficiency are compared between simulations and measurements and good agree-
ment was observed.

M.S. Shakhirul (&)  A. Sahadah  M. Jusoh  A.H. Ismail  C.M. Nor

School of Computer and Communication Engineering, Universiti Malaysia Perlis
(UniMAP), Kampus Pauh Putra, 02600 Arau, Perlis, Malaysia
e-mail: [email protected]
A. Sahadah
e-mail: [email protected]
M. Jusoh
e-mail: [email protected]
A.H. Ismail
e-mail: [email protected]
C.M. Nor
e-mail: [email protected]
F.S. Munirah
Fakulti Kejuruteraan Elektrik, Universiti Teknologi Mara (UiTM), 13500 Permatang Pauh,
Penang, Malaysia
e-mail: [email protected]

Ó Springer International Publishing Switzerland 2015 357

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_36
358 M.S. Shakhirul et al.

36.1 Introduction

The development of wearable textile antenna becomes popular due to wide

application of personal communication, wireless sensor and medical application.
Other advantages that textiles antenna offer are, washable, flexibility, light weight,
low cost and reliability. Generally, there are two manufacturing techniques to
produce wearable textile antenna [1]; (i) Using conductive fabric fixed on the non
conductive textile fabric and (ii) Using conductive textile yarns to weave, knit or
embroider the conductive pattern of the antenna on the textile fabric.
The designing wearable textile antenna is quite challenging due to the effect of
lossy environment such as humidity and temperature. Dimension wise, the
selection of a proper textile in terms of dielectric permittivity, conductivity and
loss tangent have to be determined [2, 3].
The performance of the embroidered UWB textile antenna is studied at a fre-
quency range between 3.1 and 10.6 GHz. The embroidery technique being pre-
ferred due to its robustness and esthetic value. Our study focuses on the
performance effectiveness cause by the T-slot. The antenna embroidered circular
antenna with and without T-slots performance is verified by comparing mea-
surement and simulation result.

36.2 Materials and Method

Instead of using rigid circuit boards, cotton fabric is used as substrate, conductive
fabric ‘‘Nora dell’’ as a ground plane and silver plated nylon thread as patch
antenna. The properties of selected conductive fabrics may optimize the charac-
teristics of the designed textile antenna in a specific application. Some of the
electro-textiles properties they are flexible for deformation when worn, low
electrical resistance to minimize losses, lightweight and comfortable. Therefore,
the ‘‘Nora dell’’ has been selected. There are three Nora dell elements which are
nickel, copper and nylon silver. Nora dell proposed a highly protective from
galvanic corrosion and extremely flexible to the harsh environment of 90 °C
temperature [3].
Regarding fabrication techniques for textile antenna, a conductive thread need
to be used in the embroidery techniques. Silver platted nylon thread is chosen as
conducting materials that provide high quality conducting thread. According to the
manufacturer specifications, this conducting material provides superior strength,
ability to resist the normal conditions of use such as multiple deformations for
wearable applications. In fact, the conducting thread can be washed with the
ability to resist temperature up to 150 °C [4]. In embroidery technique, thread is
sewn on the substrate will be penetrated at the back of the substrate. Hence,
another layer of cotton will be inserted as to separate between the conductive
thread and Nora dell as ground plane which means there are two layers of cotton
fabric. Table 36.1 depicted the material used and their dimensions.
36 The Embroidered Wearable Antenna for UWB Application 359

Table 36.1 Detailed of material used

Antenna Material Dimension
components (mm)
Circular patch Silver plated nylon thread embroidered throughout the Radius = 15
antenna pattern on a cotton material
Substrate Cotton Thickness,
h = 0.5
Ground plane Conductive fabric—‘‘Nora dell’’ Thickness,
h = 0.13
Microstrip feed Silver plated nylon thread 30 9 2
line

Table 36.2 Dimension of Parameter Value

Modified Patch Antenna
Patch width, W 60 mm
Effective dielectric constant, eeff 1.5025
Patch length extension, DL 0.3 mm
Patch length, L 63 mm
Effective patch length, Le 17.91 mm

It is important to determine the relative permittivity and thickness of the textile

materials. The device used to measure dielectric constant is E8362B Portable
Network Analyzer (PNA) operates from 10 MHz to 20 GHz while a vernier cal-
liper is used to measure the thickness of the textile materials.
A circular shape is mostly suited to UWB antenna which provides several
advantages such as omni-directional, low transmit power and large channel
capacity that suited the specification of UWB characteristic [5–7]. In this paper,
two prototypes are simulated and their performances are compared.
A CST simulation tool has been used in designing the UWB antennas, however
some modification has been made to suit the requirements. Initially a circular
shape patch antenna was designed using cotton as a substrate with dielectric
constant er of 1.6. The resonance frequency, fr is 6.85 GHz. The radius is calcu-
lated by using Eq. (36.1) while the patch antenna dimensions are calculated using
basic equations for microstrip patch antenna [8]. Table 36.2 shows the summa-
rized dimension for patch antenna and Fig. 36.1 shows the geometry of a modified
design antenna for first and second designs respectively.

87:94
a ¼ pffiffiffiffi ð36:1Þ
fr er
360 M.S. Shakhirul et al.

Fig. 36.1 Optimum antenna dimension. a First design the left sight is a front view while at the
right sight is a back view of the antenna, b second design the left sight is a front view while at the
right sight is a back view of the antenna

36.3 Results and Discussions

All designs and simulations of both antennas are simulated by using CST. The
performance of the antenna is evaluated based on the reflection coefficient, gain,
radiation pattern and the impedance bandwidth.
Figure 36.2 shows a reflection coefficient comparison between UWB embroi-
dered antenna without T-slot and with T-slot. The simulated result is plotted from
2 to 11 GHz. It shows fluctuates pattern with a narrowband under the acceptable
reflection coefficient of less than -10 dB. Moreover, the impedance bandwidth is
2.47 GHz.
Gain is one of the significant parameters that can determine the performance of
the antenna. Table 36.3 shows the variation of frequencies against directivity, gain
and efficiency for both antennas. First design indicates the lowest gain of
3.342 dBi at 6 GHz while the highest gain is 5.384 dBi at 8 GHz. The highest
efficiency is 93 % at 4 GHz while at 10 GHz, the efficiency recorded the lowest
36 The Embroidered Wearable Antenna for UWB Application 361

Fig. 36.2 Comparison of simulated reflection coefficient (S11), between with T-slot and without
T-slot

Table 36.3 Comparison of gain, directivitiy and % efficiency

Frequency (GHz) Gain (dB) Directivity (dBi) % Efficiency
First design of UWB embroidered antenna without T-slots
3 4.079 4.075 73
4 4.042 4.051 93
5 3.624 3.661 90
6 3.342 3.436 63
7 4.778 4.899 57
8 5.384 5.479 92
9 5.219 5.336 89
10 5.347 5.515 52
Second design of UWB embroidered antenna with T-slot
3 3.753 3.793 93
4 3.719 3.748 92
5 3.507 3.581 93
6 3.053 3.118 97
7 4.778 4.597 88
8 4.234 4.310 92
9 4.006 4.068 89
10 4.535 4.608 96
362 M.S. Shakhirul et al.

Table 36.4 Current distribution for first and second design

Frequency First design: UWB embroidered antenna Second design: UWB embroidered
(GHz) without T-slots antenna with T-slot
3

value of 52 %. Generally, some of the gains are more than 5 dBi that depicted
some frequencies using high power consumption. Therefore, some modification of
the first design should be introduced to achieve low gain with large bandwidth and
low power consumption that can cover up the entire frequency of UWB appli-
cations. The enhancement considered on the port width, partial ground and T-slot.
Antenna with T-slot achieved frequency range between 3 and 10 GHz. Maxi-
mum gain of 4.778 dBi is obtained at 7 GHz while the minimum gain is 3.053 dBi
at 6 GHz. The second design has 97 % of highest antenna efficiency at 6 GHz
while the lowest efficiency is 88 % at 7 GHz. Overall, the average gain is less than
5 dBi throughout the frequencies ranges from 3 to 10 GHz. Thus, the antenna has
low power consumption and meets the specifications of the design.
Table 36.4 presented the current flows for both antennas designed at different
frequencies; 3 GHz as low frequency, 6 GHz as medium frequency and 9 GHz as
high frequency. High strength of current is radiating along the transmission line and
the boundary of the patch after applying T-slot at the circular patch antenna.
Moreover, the boundary of a partial ground plane also is a significant radiation area.
Hence, the partial ground plane provides greater impedance bandwidth and used as
low power consumption that suited the UWB applications. The UWB is achieved
by considering suitable gap between partial ground planes and the antenna.
36 The Embroidered Wearable Antenna for UWB Application 363

36.4 Conclusion

This paper presents the development, fabrication and evaluation of the antenna’s
performance between frequency ranges of 3.1–10.6 GHz. Two embroidered UWB
antennas using cotton as substrate were considered. The investigation focussed on
the performances of both antennas with and without T-slots. Based on the com-
parison of both designs, it shows that the antenna with T-slot achieved a better
impedance matching with a wider bandwidth compared to the antenna without
T-slot. In terms of gain and efficiency, the second design provides low gain with
better performance. The gain is less than 5 dBi and basically used low power
consumption. Hence, the modified second design provides low gain, omni-direc-
tional pattern and used low power which suited the performance of the UWB
specifications. Moreover, the antenna with T-slot has a high efficiency of data rate.
Further work has been done in another conference paper that focusing more on the
bending issue.

Acknowledgments Acknowledgements are expressed to Research and Development (RND)

Centre, School of Computer and Communication Engineering and University of Malaysia Perlis
for providing the lab facilities and short term grant which enabled the publication of this article.

References

1. Maleszka, T., Kabacik, P.: Bandwidth properties of embroidered loop antenna for wearable
applications. In: Proceedings, 3rd European Wireless Technology Conference, 2010
2. Van Langenhove, L.: Smart Textiles for Medicine and Health Care. CRC Press, Cambridge
(2007)
3. Rahmat-Samii, Y.: Wearable and implantable antennas in body centric communications. Los
Angeles, CA 90095, USA, 2010
4. Shieldex website: http://itp.nyu.edu/*kh928/sensorreport/ShieldexNoraDell.pdf
5. Osman, M.A.R., Rahim, M.K.A., Samsuri, N.A., Salim, H.A.M., Ali, M.F.: Embroidered fully
textile wearable antenna for medical monitoring applications. Prog. Electromagn. Res. 117,
321–337 (2011)
6. Sanz-Izquierdo, B., Batchelor, J.C., Sobhy, M.I.: UWB Wearable Button Antenna. Department
of Kent, The University of Kent, Canterbury, Kent, 2006
7. Hsu, S.-H., Chang, K.: Ultra-thin CPW-fed rectangular slot antenna for UWB applications. In:
Proceeding, IEEE International Symposium Antennas and Propagation Society, 2006
8. Shobanasree, R., Radha, S.: Ultra wideband wearable textile antenna. In: Proceedings,
International Conference on Computing and Control Engineering, 2012
Chapter 37
Bowtie Shaped Substrate Integrated
Waveguide Bandpass Filter

Z. Baharudin, M.Z.U. Rehman, M.A. Zakariya, M.H.M. Khir,

M.T. Khan and J.J. Adz

Abstract A planar bandpass filter based on a technique that utilizes substrate

integrated waveguide (SIW) is presented. The SIW based bandpass filter is
implemented using a bowtie shaped resonator structure. The bowtie shaped filter
exhibits similar performance as found in rectangular and circular shaped SIW
based bandpass filters. This concept reduces the circuit foot print of SIW, along
with miniaturization high quality factor is maintained by the structure. The design
for single-pole resonator structure is presented; and by coupling the resonators a
two-pole bowtie shaped SIW bandpass filter is achieved for the frequency of
7 GHz. The simulation result reveals that the filter’s insertion loss is better than
1 dB and return loss is less than 30 dB.

Z. Baharudin (&)  M.Z.U. Rehman  M.A. Zakariya  M.H.M. Khir  M.T. Khan  J.J. Adz
Communication Cluster, Electrical and Electronics Engineering Department,
Universiti Teknologi PETRONAS, Tronoh, Perak, Malaysia
e-mail: [email protected]
M.Z.U. Rehman
e-mail: [email protected]
M.A. Zakariya
e-mail: [email protected]
M.H.M. Khir
e-mail: [email protected]
M.T. Khan
e-mail: [email protected]
J.J. Adz
e-mail: [email protected]

Ó Springer International Publishing Switzerland 2015 365

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_37
366 Z. Baharudin et al.

37.1 Introduction

Filters have received a particular attention with the advent of various wireless
systems, this interest has dramatically increased with the introduction and devel-
opment of new millimeter waves applications over the past decade. Various
applications have been recently proposed including wireless local area networks
[1], radars [2], intelligent transportation systems [3] and imaging sensors [4].
Efficient filters demand has also increased with the development of chip-sets
operating at 60 GHz or even higher frequencies by a number of semiconductor
industries [5].
Filters based on Substrate Integrated Waveguide (SIW) structures are achieved
through incorporating the rectangular waveguide structure into the microstrip
substrate [6]. SIWs are dielectric filled and are formed from the substrate material
utilizing two rows of conducting vias connecting bottom and top metal plates,
these vias are embedded in dielectric filled substrate; hence providing easy com-
bination with other planar circuits and a reduction in size. The size reduction along
with involving dielectric filled substrate instead of air-filled reduces the quality
factor (Q), but the entire circuitry including waveguide and microstrip transitions
can be realized by using printed circuit board (PCB) technology or other tech-
niques, like LCP [8] and LTCC [9].
The design of an SIW bandpass filter can either utilize a design methodology
based on coupling matrix method, or it can also follow a methodology used for
designing air filled waveguide filters. The design of an SIW filter based on the
methodology adopted in a rectangular waveguide, a shunt inductive coupling
realization is adopted. Vias of irregular diameters placed in the center of the cavity
may possibly occur in an inductive post filter; which is based on a requirement of
control couplings. Large couplings might occur in the use of a small diameter. The
utilization of shunt inductive vias at the couplings of the filter realizes a shunt
inductive coupling filter as depicted in r (a) or an iris (aperture) coupling post as
shown in Fig. 37.1b. A detailed literature on the development of SIW filters has
been reported in [11].
A three pole structure of a SIW bandpass filter based on shunt inductive vias is
shown in Fig. 37.1a. It utilizes four coupling vias placed in the center of the cavity

(a) (b)

Fig. 37.1 a Shunt inductive coupling post filter b Iris coupling based filter
37 Bowtie Shaped Substrate Integrated Waveguide Bandpass Filter 367

(b)

(a)

Fig. 37.2 Cavity filters with a circular cavities b rectangular cavities

of the filter, the small two vias are provide facilitation in input and output coupling
while the large vias provide coupling between the resonators. The transitions from
SIW to microstrip are placed at both the input and the output of the SIW bandpass
filters.
An SIW bandpass filter based on iris coupling posts is shown Fig. 37.1b; the
apertures form three resonators. The filter’s structure is such that the three cavities
of half wavelength are formed in the center while SIW to microstrip transition are
on the two edges of the filter; such a filter operational at 60 GHz has been pre-
sented in [12].
Cavity filters with circular [13] as presented in Fig. 37.2a and rectangular
cavities [14] as depicted in Fig. 37.2b has been observed in literature. These
variants of SIW allow more design variations and transmission zeros are also
introduced due to cross coupling, better selectivity is also presented by these
designs.
Various SIW filters structures have been proposed in the literature; however
there still exists a need to further miniaturize the structure. Furthermore the cav-
ities are only either in circular or rectangular shape. In this paper a triangular
resonators based bandpass filter is presented. This resonator structure is a minia-
turized form of SIW cavity, and exhibits similar low-loss and high quality char-
acteristics as found in its other counterparts. The proposed bandpass filter
configuration is suitable for integration with planar devices and its small footprint
area allows other devices to be easily integrated on a single board.

37.2 Substrate Integrated Waveguide

Design and Implementation of SIW filters are being performed through defined
Practical methods so far. The most common technique is to form the SIW cavity
through metallic sidewalls [10] as shown in Fig. 37.3. A dielectric substrate having
width of h forms the resonator and the resonator is of length L. The bottom and top
of the resonator are constructed through placing metallic plates and conducting
posts/vias going through the substrate connects the top and bottom plates; hence
forming the sidewalls of the cavity. The vias are of diameter d and the separation
368 Z. Baharudin et al.

Fig. 37.3 Substrate

integrated waveguide s
structure geometry p ost
cting
n du
Co
d

between two neighboring vias is given as s. The choice of diameter and separation
between the two vias forms the basis of the SIW filters, therefore these should be
selected in a manner that minimum radiation loss is exhibited.
The Dealandes and Wu [7] study reveals two primary design rules for SIW
structures as given in Eq. (37.1); these rules are followed in order to ensure same
design and modeling methodology adopted for rectangular waveguides. These
rules pertain to the diameter d of the via posts and the via post spacing s:

kg
d\ ð37:1Þ
5
s  4d ð37:2Þ

In our design d and s are chosen to be 0.8 and 2 mm respectively, these values
ensure less radiation losses and the SIW cavity acts closely to a rectangular
waveguide. For the TE101 mode, the dimensions of the SIW resonator structure are
calculated by using the relation in Eq. (37.3) [7].
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
   2ffi
c 1 2 1
fTE101 ¼ pffiffiffiffiffiffiffiffi þ ð37:3Þ
2 lr er Weff Leff

Weff and Leff denote the effective width and length of the SIW resonator,
respectively, and are given as:

d2 d2
Weff ¼ W  ; Leff ¼ L  ð37:4Þ
0:95s 0:95s

where W and L are the real width and length of the SIW resonator, c is the velocity
of light in free space. In this design the width and length of the triangular resonator
37 Bowtie Shaped Substrate Integrated Waveguide Bandpass Filter 369

Fig. 37.4 Initial triangular

SIW structure

Table 37.1 Design specifi- Notation Values

cations of the bandpass filter
Passband centre frequency 7 GHz
Passband return loss, S11 \-25 dB
Passband insertions loss, S21 [-2 db
Passband bandwidth at -3 dB [300 MHz
Stopband rejection [30 dB

structure is computed using Eqs. (37.3) and (37.4) as shown in Fig. 37.4. Utilizing
this method the cavity is designed for the specifications laid out in Table 37.1.
The triangular shaped resonator and subsequent bowtie shaped two pole filter is
designed using Roger RT/Duriod 5880 material substrate having dielectric con-
stant of 2.2 and substrate height of 787 lm. Theoretically, the resonance frequency
does not depend on the thickness of the substrate. However, it has been observed in
literature that it does play a role on the loss (mainly on radiation loss). The thicker
the substrate the lower is the loss or higher Q. It has been shown that slight
increase or decrease in the substrate thickness changes the unloaded quality factor.
To accomplish two pole bandpass SIW filter design, once the triangular reso-
nator is created for a specific resonant mode, the design methodology closely
resembles conventional simulation-based microstrip filter design [7]. Two single
cavity resonators are coupled together through capacitive coupling along the RF
input and output microstrip-to-SIW transition areas and a capacitive coupling
located in-between two SIW cavities as can be seen in Fig. 37.5 (Table 37.2).
370 Z. Baharudin et al.

Fig. 37.5 Bowtie shaped

two pole bandpass filter
dimensions

Table 37.2 Bandpass filter Notation Value (mm) Description

parameters dimensions and
descriptions Wt 36.0 Bowtie filter full width
Lt 70.0 Bowtie filter full length
L 30.0 Resonator’s length
W 35.0 Resonator’s width
Wa 12.0 I/O coupling
Wb 15.9 Coupling opening width
Wc 0.5 Inter-resonator coupling
Wx 2.0 Transition width
Lc 30.0 Coupling opening length

External quality factor and coupling coefficients are calculated from derived
expressions based on lowpass prototype parameters [10].

FBW g0 g1 g2 g3
M1;2 ¼ pffiffiffiffiffiffiffiffiffi ; Qe1 ¼ ; Qe2 ¼ ð37:5Þ
g1 g2 FBW FBW

These values are then compared to the simulated extracted external quality
factors and coupling coefficients for a particular cavity geometry.

f0
Qext ¼ ð37:6Þ
Df3 dB

Iterations and adjustments to the dimensions of the coupling areas of the filter
are performed until the calculated values match the extracted values from full-
wave simulation, providing the desired filter.
37 Bowtie Shaped Substrate Integrated Waveguide Bandpass Filter 371

37.3 Results and Discussions

The desired fixed filter described in the specifications in Table 37.1 and its cor-
responding designs structure shown in Fig. 37.5 are realized with the responses
shown in Figs. 37.6 and 37.7. The simulation to obtain the filter responses from the
designed structure shown in Fig. 37.5 is conducted using ANSYS High Frequency
Structure Simulator (HFSS). In addition Agilent Vector Network Analyzer (VNA)
is utilized for the measurements of the fabricated filters.
Figure 37.6 shows the response of the simulated two pole filter designed for
7 GHz. Response of the filter is obtained through realizing the resonator structure
shown in Fig. 37.5, the resonator designed based on the requirements using the
equations presented in Sect. 37.2. The simulated S21 and S11 response of the
bowtie shaped bandpass filter reveals that the S11 value at the center frequency of
6.9 GHz is less than -30 dB, whereas the S21 response is greater than -0.5 dB
and the passband bandwidth at -3 dB is greater than 500 MHz.

Fig. 37.6 Simulated

response of the bowtie shaped
two pole bandpass filter

Fig. 37.7 Measured

response of the bowtie shaped
two pole bandpass filter
372 Z. Baharudin et al.

Table 37.3 Summary of the bandpass filter performance

Key parameters Simulated Measured
Passband centre frequency (GHz) 6.9 6.93
Passband return loss, S11 (dB) \30 \25
Passband insertions loss, S21 (dB) [-0.50 [-1.7
Stopband rejection (dB) [30 [30
Passband bandwidth at -3 dB level 0.5 GHz 0.35 GHz

Fig. 37.8 Fabricated design of the bowtie shaped two pole filter

Measured S21 and S11 response of the two pole bandpass filter is presented in
Fig. 37.7. The S21 response at the centre frequency of 5.9 GHz is better than
-1.7 dB and its corresponding S11 response at the centre frequency is less than
-25 dB. The passband bandwidth at -3 dB is greater than 350 MHz, whereas the
lower and upper stopband rejections are better than -25 dB. These responses are
summarized in Table 37.3.
The performance of measured response in terms of the insertion loss is due to
the quality of the SMA connectors and wires used for recording readings. The
upper transmission zero produced due to cancellation of the TE101 and TE201
modes is distanced enough from the filter passband, hence a pure chebyshev
response is observed. However high order resonant modes presents spurious at
9.5 GHz. Consequently suppression of spurious is achievable by employing
lowpass filters at the input and output of the resonators. Therefore the cut-off for
the lowpass filters has to be matched with the upper transmission zero of the
bandpass filter (Fig. 37.8).
37 Bowtie Shaped Substrate Integrated Waveguide Bandpass Filter 373

37.4 Conclusion

A 7 GHz bowtie shaped bandpass filter based on triangular structure SIW is

proposed in this paper, the filter has presented good performance and a minia-
turized version of the SIW structure is exploited in the design process. The filter
has also presented broad bandwidth at the -3 dB level. This filter presents good
performance in its small size, and it can be readily integrated with planar circuits
and devices.

References

1. James, J., Shen, P., Nkansah, A., Liang, X., Gomes, N.J.: Millimeter-wave wireless local area
network over multimode fiber system demonstration. IEEE Photonics Technol. Lett. 22,
601–603 (2010)
2. Wang, L., Glisic, S., Borngraeber, J., Winkler, W., Scheytt, J.C.: A single-ended fully
integrated SiGe 77/79 GHz receiver for automotive radar. IEEE J. Solid-State Circuits 43,
1897–1908 (2008)
3. Yan, X., Zhang, H., Wu, C.: Research and development of intelligent transportation systems.
In: 11th International Symposium on Distributed Computing and Applications to Business
Engineering and Science (DCABES), vol. 19–22, pp. 321–327, October 2012
4. Wilson, J.P., Schuetz, C.A., Martin, R., Dillon, T.E., Yao, P., Prather, D.W.: Polarization
sensitive millimeter-wave imaging sensor based on optical up-conversion scaled to a
distributed aperture. In: 37th International Conference on Infrared, Millimeter, and Terahertz
Waves (IRMMW-THz), vol. 1, pp. 23–28, September 2012
5. Niknejad, A.M., Hashemi, H.: Millimetre-Wave Silicon Technology: 60 GHz and Beyond.
Springer, Berlin (2008)
6. Hirokawa, J., Ando, M.: Single-layer feed waveguide consisting of posts for plane TEM wave
excitation in parallel plates. IEEE Trans. Antennas Propag. 46(5), 625–630 (1998)
7. Deslandes, D., Wu, K.: Accurate modeling, wave mechanisms, and design considerations of a
substrate integrated waveguide. IEEE Trans. Microw. Theory Tech. 54(6), 2516–2526 (2006)
8. Yang, K.S., Pinel, S., Kwon, K., Laskar, J.: Low-loss integrated-waveguide passive circuits
using liquid-crystal polymer system-on-package (SOP) technology for millimeter-wave
applications. IEEE Trans. Microw. Theory Tech. 54(12), 4572–4579 (2006)
9. Xu, J., Chen, Z.N., Qing, X., Hong, W.: 140-GHz planar broadband LTCC SIW slot antenna
array. IEEE Trans. Antennas Propag. 60(6), 3025–3028 (2012)
10. Tang, H.J., Hong, W., Hao, Z.C., Chen, J.X., Wu, K.: Optimal design of compact millimetre-
wave SIW circular cavity filters. Electron. Lett. 41(19), 1068–1069 (2006)
11. Ur Rehman, M.Z., Baharudin, Z., Zakariya, M., Khir, M., Khan, M., Weng, P.W.: Recent
advances in miniaturization of substrate integrated waveguide bandpass filters and its
applications in tunable filters. In: Business Engineering and Industrial Applications
Colloquium (BEIAC), 2013 IEEE, pp. 109–114, 2013
12. Hirokawa, J., Ando, M.: Efficiency of 76-GHz post-wall waveguide-fed parallel-plate slot
arrays. IEEE Trans. Antennas Propag. 48(11), 1742–1745 (2000)
13. Tang, H.J., Hong, W., Hao, Z.C., Chen, J.X., Wu, K.: Optimal design of compact millimetre-
wave SIW circular cavity filters. Electron. Lett. 41(19), 1068–1069 (2005)
14. Chen, X.-P., Wu, K.: Substrate integrated waveguide cross-coupled filter with negative
coupling structure. IEEE Trans. Microw. Theory Tech. 56(1), 142–149 (2008)
Chapter 38
Logical Topology Design with Low Power
Consumption and Reconfiguration
Overhead in IP-over-WDM Networks

Bingbing Li and Young-Chon Kim

Abstract Reducing the power consumption of networks has drawn a lot of

attention in recent years. To improve the energy efficiency, selectively turning off
network components during light traffic periods could be an effective method.
However, turning off/on network devices adapting to traffic load may result in the
reconfiguration of logical topology in optical wavelength routed networks. In this
paper, we study the logical topology design problem under multi-period traffic
load in IP-over-WDM networks. To solve the problem, a mixed integer linear
programming (MILP) model is proposed with the objective to minimize the overall
network power consumption and the reconfiguration overhead between consecu-
tive time periods. The proposed model is evaluated and compared with conven-
tional schemes in terms of power consumption and reconfiguration overhead
through a case study.

 
Keywords Energy efficiency IP-over-WDM Logical topology reconfiguration 
MILP

38.1 Introduction

With the exponential growth of end users and the emergence of large-capacity
required services, the Internet traffic has increased by 50–100 times during the last
decade. It is expected that this increment speed will be kept in the near future.

B. Li (&)  Y.-C. Kim

Department of Computer Engineering, Chonbuk National University, Jeonju, Korea
e-mail: [email protected]
Y.-C. Kim
e-mail: [email protected]
Y.-C. Kim
Smart Grid Research Center, Jeonju 561-756, Korea

 Springer International Publishing Switzerland 2015 375

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_38
376 B. Li and Y.-C. Kim

To meet the traffic demand, network should be deployed with more transmission
and switching equipment with higher capacity, consequently causing more power
consumption. The network infrastructures are estimated to account for 12 % of
total Internet power consumption at present and this portion will increase to 20 %
by 2020 [1]. Hence, improving the energy efficiency of the Internet becomes a
challenging issue nowadays.
Due to the high speed, huge capacity, low signal attenuation and other
advantages of optical fibers, they are now widely used as the transmitting infra-
structure in communication networks. It has been proven that optical component
and equipment are much more energy efficient, comparing to their electronic
counterparts. In particular, with the development of wavelength division multi-
plexing (WDM) technology, transmitting IP packets directly over WDM channel
(IP-over-WDM) is considered as a promising paradigm. IP-over-WDM network
can be implemented in different ways: namely IP with no Bypass, Transparent IP
with Bypass and Grooming, Opaque IP with Bypass and Grooming [2], etc.
Among these schemes, Transparent IP with Bypass and Grooming is the most
energy efficient solution since the wavelengths can bypass at some intermediate
nodes and low demand traffic flows can be groomed onto high-speed wavelength
channels and transmitted integrally. As a result, the electronic processing at some
nodes is avoided and the utilization of wavelength channels is improved. In such
wavelength routed network, traffic demands are serviced by connection-oriented
optical circuits, which are called lightpaths. Lightpaths can be established based on
given network physical topology and corresponding traffic matrix, constructing a
logical topology. Thus, designing an energy efficient IP-over-WDM network can
be translated into logical topology optimization problem [3, 4]. There has been
several researches focusing on IP-over-WDM network design with minimum
power consumption [2, 5, 6]. However, they consider power consumption as the
only parameter to be optimized. No or little attention is given to the changes of
logical topology.
Following the user behavior over different periods in 1 day, the Internet traffic
pattern can be approximated to sinusoidal function [7]. Figure 38.1 shows the
traffic load of 2 days (from 8:00 23rd to 8:00 25th June, 2013), monitored from
Amsterdam Internet Exchange. The peak time occurs at around 21:00. Usually, the
logical topology is designed with capability to provision the heaviest network load.
To reduce the power consumption, we can switch off some network elements when
traffic load is light, like in deep night or early morning [8]. Consequently, the
logical topology needs to be reconfigured adapting to immediate network status.
Reconfiguration may include tearing down existing lightpaths and setting up new
ones. The change from one logical topology to another will result in disruption of
network, introducing data loss or delay [9]. Even though network resource (in our
case, the power consumption) is optimized to fit the changing traffic load, the
quality of service (QoS) in network is deteriorated. Hence, the traditional recon-
figuration optimization solutions try to minimize the average number of hops
encountered by a packet, total number of lightpaths, or total number of physical
links used [3, 4, 9]. Reference [10] proposed a model with the objective to
38 Logical Topology Design with Low Power Consumption … 377

Fig. 38.1 Daily traffic load (from AMS-IX)

minimize power consumption while the reconfiguration is restricted via

constraints. Nevertheless, none of these optimization models considers power
consumption as well as reconfiguration at the same time.
In this paper, we consider a wavelength-routed optical backbone network based
on the transparent IP-over-WDM network architecture. When network load is low,
lightpaths need tearing down and corresponding elements can be turned off to save
power. On the contrary, when network load increases, network elements are
activated and new lightpaths should be established to provision increased traffic. In
any case of network status change, influenced flows need rerouting and traffic
interruption may occur. To solve the problem, we propose a MILP formulation
with the objective to minimize the sum of total power consumption and the
reconfiguration overhead multiplied with a weighting factor. This formulation
considers minimizing both the power consumption in current status and the dif-
ference between two logical topologies of current and previous network status.
The rest of this paper is organized as follows: In Sect. 38.2, the mathematical
formulation is presented and explained. In Sect. 38.3, the MILP models are
evaluated and compared by illustrative examples; and the numerical results will be
analyzed. Finally, we conclude the paper in Sect. 38.4.

38.2 Proposed Mathematical Model

38.2.1 Network Assumption

IP-over-WDM network consists of two layers: electrical layer and optical layer. IP
routers are deployed at network nodes and constitute the IP layer (electrical layer).
The functions of IP router is to generate (as a source node), process (as a grooming
node) and drop (as a destination node) IP services. They are connected with an
378 B. Li and Y.-C. Kim

optical cross-connect (OXC) via transponders which are used to emit and termi-
nate lightpaths. Two adjacent OXCs are connected by an optical fiber and
responsible for switching lightpaths. Each optical fiber can support many wave-
length channels. All the OXCs and optical fibers construct the WDM layer (optical
layer). IP packets are groomed at IP layer and then transmitted directly on optical
WDM channels. Based on the transparent node architecture with bypass and
grooming function, the power contributors are: (1) IP routers, for electronically
processing traffic when grooming is needed; (2) Transponders, for establishing
lightpaths; (3) OXCs, for optical switching wavelengths. The traffic processed at
source node is not considered since it is fixed for a given traffic matrix. Note that
the electronic processing is dependent with traffic amount while the power con-
sumed by optically switching a lightpath is fixed, independent with the amount of
traffic traveling on that lightpath. The power consumption of a transponder is also
constant if it is activated, no matter the established lightpath is busy or idle.
On the other hand, we assume that the less reconfiguration happens to the
logical topology, the less traffic needs rerouting, which leads to less data loss or
delay, i.e. the less QoS degradation. The reconfiguration overhead is defined as the
number of changes in physical links involving in forming a new logical topology
comparing to the previous network status. Note that the change of wavelength
assignment is also considered, which means that even if the route of a lightpath
traverses the same physical links, different wavelength assignment decision is also
viewed as a change and can affect those traversed links.

38.2.2 MILP Model

According to the network assumption mentioned above, we propose a mixed

integer linear programming (MILP) formulation. The proposed MILP formulation
deals with a static network design problem. The physical topology and corre-
sponding estimated traffic matrix should be given in advance. To reflect traffic
variation, 1 day can be divided into several periods. In each period, MILP model is
run to find optimal solution with minimal power consumption for current period
and least logical topology change from previous period. For variable indexing rule,
s and d index the source and destination nodes of a traffic demand, m and n rep-
resent the nodes in physical topology, and i and j indicate the nodes in logical
topology. The notations are summarized in Table 38.1.
According to the notations defined, the objective function can be written as:

Minimize PC þ a  RO ð38:1Þ
38 Logical Topology Design with Low Power Consumption … 379

Table 38.1 Summary of notations

Parameter Meaning
G(V, E) Network physical topology, consisting of node set V and edge set E
ksd Amount of traffic demand from source s to destination node d, s, d [ V
T = [ksd] Traffic matrix, set of traffic demands, s, d [ V
C Capacity of each wavelength
W Set of wavelengths on a fiber
Cep Routing capacity of an IP router
Ti Maximum number of transmitters at node i
Ri Maximum number of receivers at node i
Ptr Power consumption of a transponder
Pos Power consumption for optical switching one wavelength
Pep Power consumption for electronic processing per traffic unit (in Gbps)
PC Power consumption of whole network
0
Nij;w The number of lightpaths between node i and j, using wavelength w (in previous
period)
P0ij;w
mn The number of lightpaths between node i and j, being routed through physical link
(m, n), using wavelength w (in previous period)
Variable Meaning
fijsd Traffic amount of ksd that travels on lightpath lij
Nij;w The number of lightpaths between node i and j, using wavelength w
Nij The number of lightpaths between node i and j
Pij;w
mn The number of lightpaths between node i and j, being routed through physical link
(m, n), using wavelength w

where
0 1
XXX X XX
PC ¼ Pos  @ ij;w
Pmn þ Nij;w A þ 2  Ptr  Nij þ Pep
w i j;j6¼i i j;j6¼i
X X X X ðm;nÞ
 fijsd
i j s;i6¼s d

ð38:2Þ
XXXXX 
RO ¼ Pij;w  P0ij;w  ð38:3Þ
mn mn
i j m n w

subject to the following constraints:

8 sd
X X <k ; if i ¼ s
fijsd  fjisd ¼ ksd ; if i ¼ d 8i; s; d 2 V ð38:4Þ
j j
:
0; otherwise
380 B. Li and Y.-C. Kim

8
X X < Nij;w ; if m ¼ i
ij;w ij;w
Pmn  Pnm ¼ Nij;w ; if m ¼ j 8i; j; m 2 V; 8w 2 W ð38:5Þ
:
n n 0; otherwise
XX
fijsd  C  Nij ; 8i; j 2 V ð38:6Þ
s d
XXX
fijsd  Cep ; 8i 2 V ð38:7Þ
j s;i6¼s d

XX
Nij;w  Ti ; 8i 2 V ð38:8Þ
w j

XX
Nji;w  Ri ; 8i 2 V ð38:9Þ
w j

X
Nij;w ¼ Nij 8i; j 2 V ð38:10Þ
w
XX
ij;w
Pmn 1 8ðm; nÞ 2 E; 8w 2 W ð38:11Þ
i j

In the MILP formulation, Eq. (38.1) gives the objective function, which min-
imizes the total power consumption (PC) as well as the reconfiguration overhead
(RO). The power is consumed by OXCs and transponders in optical domain and
routers in electrical domain of the IP-over-WDM networks. The RO is defined as
the total number of physical links involved in lightpath establishment different
from that in previous logical topology. A weight, a, is assigned to RO so as to
make two factors mutually comparable. Equations (38.4) and (38.5) guarantee the
flow balancing in network in the view of traffic flow and physical link, respec-
tively. Constraint (38.6) limits that the total traffic amount transmitting on all
lightpaths cannot beyond total capacity they offer. Constraint (38.7) guarantees
that all traffic electronically processed at a node is restricted by the maximum
capacity of IP router. The number of usable transmitters and receivers are limited
by constraints (38.8) and (38.9), respectively. Equation (38.10) calculates the
number of lightpaths between node i and node j. Constraint (38.11) ensures that
each wavelength on a physical link can be used to establish at most one lightpath.
To compare the proposed model with conventional schemes, other two MILP
models are presented. First one tries to minimize the total PC of network (shortly,
Min PC), while the other one considers to minimize the RO from previous logical
topology (Min RO). Both comparing models can share the same constraints with
our model. Because our model considers minimizing both PC and RO, it is rep-
resented shortly as ‘‘Hybrid’’ in the following part.
38 Logical Topology Design with Low Power Consumption … 381

38.3 Numerical Results

The numerical results will be shown and analyzed in this section. To evaluate the
performance of Hybrid and compare it with Min PC and Min RO, we apply three
schemes to a case study. Our results are obtained via optimization software IBM
ILOG CPLEX Optimization Studio Version 12.5 on the computer with Intel Core
2 (TM) i5-2500 CPU (3.30 GHz) and 8 GB RAM.

38.3.1 Network Topology and Traffic

The case study is implemented in Pan-European COST239 network, shown as

Fig. 38.2. Physical topology consists of 11 nodes and 26 links. Nodes are con-
nected by bi-directional links, one fiber on each direction. One fiber can maximally
support 16 wavelength channels, each with capacity 40 Gbps. Hence, the network
can supply total capacity 33.28 Tbps. Based on the network topology, we assume
the traffic matrix at peak time as shown in Table 38.2. The unit of traffic demand
for each source-destination pair is Gbps and the total traffic amount is 1 Tbps.
To reflect realistic network load, we refer to the data from AMS-IX (24-h from
0:00 24th to 0:00 25th June, 2013) as shown between two dotted lines in Fig. 38.1.
Considering the variation of traffic amount in 1 day, we divide 24 h into 12 time
periods. During each 2-h period, the traffic amount is constant, while the amount is
different from one period to another. We generate non-peak time traffic as a
fraction of traffic at peak time. By multiplying each entry in peak time traffic
matrix with a factor [0.72, 0.44, 0.31, 0.42, 0.61, 0.72, 0.77, 0.82, 0.87, 0.95, 1.0,
0.98], we can represent an approximated daily traffic pattern which is shown as
Fig. 38.3.

Fig. 38.2 Pan-European

COST239 network topology
382 B. Li and Y.-C. Kim

Table 38.2 Traffic matrix of peak load

Node 1 2 3 4 5 6 7 8 9 10 11
1 0 1 1 3 1 1 1 35 1 1 1
2 1 0 5 14 40 1 1 10 3 2 3
3 1 5 0 16 24 1 1 5 3 1 2
4 3 14 16 0 6 2 2 21 81 9 9
5 1 40 24 6 0 1 11 6 11 1 2
6 1 1 1 2 1 0 1 1 1 1 1
7 1 1 1 2 11 1 0 1 1 1 1
8 35 10 5 21 6 1 1 0 6 2 5
9 1 3 3 81 11 1 1 6 0 51 6
10 1 2 1 9 1 1 1 2 51 0 81
11 1 3 2 9 2 1 1 5 6 81 0

Fig. 38.3 Approximation of daily traffic amount

38.3.2 Power Consumption and Other Parameters

The PC considered is given in Table 38.3, referring to some literatures and data
sheets of commercial products [6, 10, 11]. For IP router, Cisco CRS 16-Slot Carrier
Routing System is considered. The total routing capacity per chassis is
38 Logical Topology Design with Low Power Consumption … 383

Table 38.3 Power consumption

Device Power Consumption
40 Gbps transponder 175 W
Optical switching 4 W per one 40 Gbps wavelength
Electronic processing 14.5 W (per 1 Gbps)

Cep = 4480 Gbps. For OXC, the MEMS-based optical switch is considered. At
each node, maximum number of transmitters/receivers is 16 (Ti = 16, Ri = 16). To
investigate the influence resulting from the weight of RO, we implement Hybrid
with different values of a (a = 1, 10, 100, 350, 1000), shortly represented as
‘‘Hybrid_1’’, ‘‘Hybrid_10’’, ‘‘Hybrid_100’’, ‘‘Hybrid_350’’ and ‘‘Hybrid_1000’’.

38.3.3 Results

Figure 38.4 shows the reconfiguration overhead according to different time peri-
ods. Among all cases, Min RO needs smallest change of the logical topology when
traffic varies from a period to another. In particular, the logical topology has no
change during six periods. Hybrid_1000 shows similar feature with Min RO. With
reducing the importance of RO factor, more changes are needed. In the view of the
cumulative changes in whole day, Min RO makes 43 changes of physical links
involved in lightpath establishment; Hybrid_1000 and Hybrid_350 perform sim-
ilarly to Min RO and need 45 and 50 changes, respectively; Hybrid_100,

Fig. 38.4 Reconfiguration overhead versus time period

384 B. Li and Y.-C. Kim

Hybrid_10 and Hybrid_1 result in 76, 261 and 266 changes, respectively; while
Min PC results in 865 changes. To limit the RO, the rate of change (ROC) can be
defined as the ratio between the number of wavelengths involving in reconfigu-
ration to the number of total wavelengths occupied for constructing the initial
logical topology in the first period. The value of ROC can indicate the strictness of
QoS. In our case study, it is ruled that the ROC cannot be larger than 0.2. Then
Hyrid_100, Hyrid_350, Hyrid_1000 and Min RO can satisfy this requirement
among all tested cases.
As another important metric, the total PC of network in different time periods is
given in Fig. 38.5. Min PC achieves the lowest PC, because the model essentially
tries to minimize the PC factor. In the view of total PC in whole day, Hybrid_1,
Hybrid_10, Hybrid_100, Hybrid_350 and Hybrid_1000 consume 0.8, 1.4, 4.4, 10.4
and 19.5 % more power than Min PC, respectively. Min RO achieves the highest
PC, 1.43 times of the power consumed by Min PC. This is resulted from the
extreme effort of Min RO on minimize RO. To guarantee logical topology
unchanged, many lightpaths cannot be torn down and the corresponding tran-
sponders should be kept in ‘‘on’’ state when traffic load is light, for instance, from
second to fifth period.
Obviously, even though Min PC or Min RO can reach the best performance in
the view of single objective, they both perform the worst if the other metric is
evaluated. Min RO model leads to large power waste: at first, many single-hop
lightpaths are established; after network initialization, logical topology keeps
nearly no change and transponders have to be turned on in light load periods. On
the other hand, Min PC configures the logical topology for each period, inde-
pendent with previous network status. The wavelengths used to establish lightpath

Fig. 38.5 Power consumption versus time period

38 Logical Topology Design with Low Power Consumption … 385

need to be assigned again, making the overhead to configure a brand new logical
topology considerably large. Different from the schemes mentioned above, the
proposed model can obtain substantial reduction of PC and limit RO. However, the
threshold of ROC could strongly influence the choice of a. By limiting
ROC B 0.2, Hybrid_100, Hybrid_350, Hybrid_1000 and Min RO can satisfy the
requirement, among which Hyrid_100 achieves the lowest PC. Hence, a = 100 is
reasonable suggestion for the case study.

38.4 Conclusion

Considering the traffic variation, turning off some network elements in low-load
period can effectively reduce power consumption. Based on this load-adaptive
scheme, logical topology needs to be reconfigured. However, reconfiguration may
cause data delay and loss resulted from traffic interruption and rerouting. To solve
such problem, we proposed a MILP model (Hybrid) for designing logical topology
with low PC and RO in IP-over-WDM networks. Then the proposed model was
evaluated and compared to conventional schemes with univocal objective (Min PC
and Min RO) via illustrative case study. In addition, different values of a were
investigated. With the definition of ROC, a = 100 was suggested based on the
tested network because it achieved the least power consumption without breaking
the rule that ROC should be less than 0.2. The numerical results showed that our
proposed model could achieve low reconfiguration overhead while keeping power
consumption low.

Acknowledgments This work was supported by the National Research Foundation of Korea
(NRF) funded by the Korea government (MSIP) (2010-0028509).

References

1. The Climate Group, http://www.smart2020.org/_assets/files/02_Smart2020Report.pdf

2. Musumeci, F., Vismara, F., Grkovic, V., Tornatore, M., Pattavina, A.: On the energy
efficiency of optical transport with time driven switching. In: IEEE International Conference
on Communications, pp. 1–5. Kyoto (2011)
3. Banerjee, D., Mukherjee, B.: Wavelength-routed optical networks: linear formulation,
resource budgeting tradeoffs, and a reconfiguration study. IEEE/ACM Trans. Netw. 8(5),
598–607 (2000)
4. Almeida, R.T.R., Calmon, L.C., Oliveira, E., Segatto, M.E.V.: Design of virtual topologies
for large optical networks through an efficient MILP formulation. Opt. Switch. Netw. 3(1),
2–10 (2006)
5. Shen, G., Tucker, R.S.: Energy-minimized design for IP over WDM networks. J. Opt.
Commun. Netw. IEEE/OSA 1(1), 176–186 (2009)
386 B. Li and Y.-C. Kim

6. Idzikowski, F., Luca Chiaraviglio, L., Portoso, F.: Optimal design of green multi-layer core
networks. In: Proceedings of the 3rd International Conference on Future Energy Systems:
Where Energy, Computing and Communication Meet. ACM (2012)
7. Chiaraviglio, L., Mellia, M., Neri, F.: Energy-aware backbone networks: a case study. In:
First International Workshop on Green Communications. Dresden, Germany, June 2009
8. Yayimli, A., Cavdar, C.: Energy-aware virtual topology reconfiguration under dynamic
traffic. In: 14th International Conference on Transparent Optical Networks, pp. 1–4.
Coventry, UK (2012)
9. Ramamurthy B., Ramakrishnan, A.: Virtual topology reconfiguration of wavelength-routed
optical WDM networks. In: IEEE Global Telecommunications Conference, vol. 2,
pp. 1269–1275. San Francisco (2000)
10. Zhang, Y., Tornatore, M., Chowdhury, P., Mukherjee, B.: Energy Optimization in IP-over-
WDM Networks. Opt. Switc. Netw. (OSN) 8(3), 171–180 (2011)
11. Idzikowski, F.: Power consumption of network elements in IP over WDM networks. TKN
Technical Report, TKN-09-006 (2009)
 Part II
Computer
Chapter 39
Systematic Analysis on Mobile Botnet
Detection Techniques Using Genetic
Algorithm

M.Z.A. Rahman and Madihah Mohd Saudi

Abstract Nowadays smart phone has been used all over the world and has
become as one of the most targeted platforms of mobile botnet to steal confidential
information especially related with online banking. It is seen as one of the most
dangerous cyber threat. Therefore in this research paper, a systematic analysis on
mobile botnet detection techniques is further investigated and evaluated. A case
study was carried out to reverse engineering the mobile botnet codes. Based on the
findings, this mobile botnet has successfully posed itself as a fake anti-virus and
has the capability to steal important data such as username and password from the
Android-based devices. Furthermore, this paper also discusses the challenges and
the potential research for future work with relate of the genetic algorithm. This
research paper can be used as a reference and guidance for further study on mobile
botnet detection techniques.

Keywords Mobile botnet  Genetic algorithm  Reverse engineering  Android

39.1 Introduction

Statistics taken from F-Secure [1] as displayed in Fig. 39.1 show that in every five
malware threats, there is one mobile botnet threat. Botnet is seen as a new emerged
cyber threat that attacks mobile devices, where all confidential information such as
bank account number, username and password for online banking, credit card
number, are kept in the smart phone. This kind of information makes mobile

M.Z.A. Rahman (&)  M.M. Saudi

Faculty of Science and Technology, Universiti Sains Islam Malaysia (USIM),
71800 Bandar Baru Nilai, Negeri Sembilan, Malaysia
e-mail: [email protected]
M.M. Saudi
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 389

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_39
390 M.Z.A. Rahman and M.M. Saudi

Fig. 39.1 Mobile bot threat

2013. [Adapted from F-
Secure mobile threat report
(Q3 2013)]

devices now targeted by attacker out there. There are several major mobile
operating systems such as iOS, Android and Windows phone. Among them,
Android is the most popular in term of attack as it is an open source operating
system [2]. Earlier Google Play acts as an official market for Android platform and
the main source for the users but nowadays there are several third-party market
arise as an alternative such as Yandex market [3].
Usually, botnet infects victim’s device by installing untrusted third-party
application that acts as a trojan horse, entering the device and dropping botnet in it.
By executing botnet such as Zeus, FakeDefender or Waledac, every data such as
banking credential, email and SMS in victim’s mobile device can be accessed
within a second. Surprisingly, Google Play does not have firm control over its
content where in June 2013, malwares reported have been hosted in the market.
F-Secure reported there are applications that imitate the original application but
repacked with malicious code and changed permission and has been downloaded
between 100 and 22,000 times [4].
Therefore based on the botnet implication, this research is formed. The
objectives of this research paper are to investigate and to evaluate mobile botnet
detection techniques and to improve the gaps identified in the previous work. To
achieve these objectives, a reverse engineering was conducted to the mobile botnet
codes to see how the mobile botnet works. The findings from the evaluation and
the reverse engineering are used as the basis to develop the mobile botnet archi-
tecture and detection algorithm. Later, these are used as the basis to produce an
effective model to detect the mobile botnet attacks, but will not be discussed in this
paper.
This paper is organised as follows. Section 39.2 presents the related works with
mobile botnet. Section 39.3 explains the methodology used in this research paper
which consists of static and dynamic analyses. Then follows by the architecture of
the controlled laboratory architecture for the mobile botnet testing. Section 39.4
presents the research finding which consists of proof of concept of mobile botnet
implication. Section 39.5 concludes and summarises the challenges and future
work of this research paper.
39 Systematic Analysis on Mobile Botnet Detection Techniques … 391

39.2 Previous Works

A botnet is a set of computers that are infected by a specific malware or software

which gives an attacker known as attacker the ability to remotely control the
infected computers [5]. The infected computers are used by the attacker to launch
the cyber-attack, such as sending spam messages, interruption, distributed denial
of services (DDoS) and collecting sensitive information.
The term mobile botnet refers to a group of infected mobile devices and
remotely controlled by the attacker via Command and Control (C&C) channel to
do malicious tasks [6]. In 2004, the earliest case was reported as Cabir was
detected and identified infecting Symbian OS [7] and this proves that mobile
devices can be infected by malware. However it can only perform limited mali-
cious tasks and spreads to other mobile devices using Bluetooth medium. As in
2009, SymbOS.Exy.C [8] and Ikee.B [9] show how mobile malware has evolved
as it has the capability to connect back to C&C server and steal valuable infor-
mation from the infected device. In 2011, DroidKungFu was discovered attacking
Android smartphones and has the capability of rooting vulnerable Android phone
and manage to evade the mobile antivirus detection [10]. It spreads via repack-
aging the mobile application, which has been downloaded from third party android
market. It collects automatically OS version, IMEI number and phone model. Then
it will contact the attacker by sending HTTP Post.
In the past few years, many studies have been done on mobile botnets such as
by [11–13]. Traynor et al. [13] studied the possibility of using bluetooth as the
command and control (C&C) channel of a botnet. As for Mulliner, he proposed
SMS-HTTP command and control system in which the attacker created command
and then the command was sent to bots via SMS [12]. The command is then being
uploaded to a designated website in an encrypted file. Then, each bot will
download, decrypt the file, and send out the commands to other bots via SMS.
While Zeng and his colleagues designed a SMS-P2P hybrid botnet which uses
SMS as the C&C channel, and the peer-to-peer (P2P) network as the underlying
structure [11]. Botnet communicates by obtaining commands in a P2P fashion by
sending and receiving SMS messages.
Yusoff and Jantan [14] implemented Genetic Algorithm (GA) to improve
classification and the accuracy rate of PE file that failed to be classified by decision
tree classifier. While work done by [15], implemented (GA) to the layered system
to detect and filter http botnet attack and has succeeded provide less false positive
rate. Monther and Rami also used (GA) to expand the training dataset through
mutations. It managed to reduce training time and increased the detection from 77
to 80 % [16]. Most of the existing works discussed above were not focusing on
mobile botnet and targeted for Windows platform only.
In this research paper, the mobile botnet sample was being reverse engineered
using static and dynamic analyses, prior the formation of new mobile malware
detection model. GA is chosen for classification as it simulates the natural pro-
cesses. It uses selection, recombination and mutation acting on a genotype that is
392 M.Z.A. Rahman and M.M. Saudi

Fig. 39.2 Mobile malware controlled lab architecture

decoded and evaluated for fitness [17]. Furthermore, it is suitable for optimization
of complex systems, robust and gives good result in real-time dynamic environ-
ment compared to other artificial intelligence such as Artificial Neural Network
[18]. For this new model, it is expected to produce a better accuracy rate and lower
false negative rates, but will not be discussed in this research paper.

39.3 Methodology

For reverse engineering, an isolated lab was set up, with no outgoing connection to
internet as displayed in Fig. 39.2. It is a controlled lab environment and almost all
software used in this experiment are an open source or available on free basis as
displayed in Table 39.1. The training dataset in this research consists of different
types of mobile malware and was downloaded from Android Malware Genome

Table 39.1 Software used in the testing

Software Function
VMWare/VirtualBox To build up virtual operating systems in a computer
ApkTool To conduct the static analysis
Dex2Jar To conduct the static analysis
JD-GUI To conduct the static analysis
Notepad+++ To conduct the static analysis
AndroidSDK To conduct the dynamic analysis
Wireshark To monitor the network traffic generated from the infected computer
7zip To unzip compressed file
39 Systematic Analysis on Mobile Botnet Detection Techniques … 393

Fig. 39.3 KDD processes integrated with genetic algorithm

Project [19, 20]. As for the testing and evaluation, this experiment used dataset from
Contagio [21]. The static and dynamic analyses were used in this isolated envi-
ronment. As for static analysis, the files and its content associated with the mobile
botnet were monitored without running the code or the program. While dynamic
analysis includes executing the mobile botnet in a sandbox and observes its actions.
Knowledge Data Discovery (KDD) is an overall process of discovering useful
knowledge from data and it has been applied in this research as well [22]. Fig-
ure 39.3 displays the KDD that has been improved by integrating an improved
genetic algorithm to classify and to detect mobile botnet more efficiently with
higher positive rate and lower false positive rate.
The basic Genetic Algorithm equation is defined based on the f function and the
binary strings of length l and it is known as string fitness (refer to Eq. 39.1). A new
string is created from the current population. The probability that a parent string Hj
will be selected from N strings H1, H2, to HN as displayed in Eq. (39.1).
 
  f Hj
p H j ¼ PN ð39:1Þ
n¼1 f ðHnÞ

The string with greater fitness will be selected. The average fitness of all string
in the population fl as displayed in Eq. (39.2). As for mobile botnet, the mobile
botnet is transformed into chromosome containing the representation of the mobile
botnet class, which is based on its target and operation behavior. In the mobile
botnet analysis, a comparison based on different mobile malware features, which
are the infection, behavior, payload, activation, operating algorithms and propa-
gation were conducted. Mobile botnet then being categorized based on these
features. Further, the data matching process was becoming easier when payload is
used as the unique key identification. Once the data matching process completed,
the eradication steps are identified for each botnet.

1X N
fl ¼ f ðH i Þ ð39:2Þ
N i¼1
394 M.Z.A. Rahman and M.M. Saudi

Based on Eqs. (39.1) and (39.2), it has been integrated and mapped into mobile
botnet classification as displayed in Eq. (39.3).
1 
X 
classscore
f ð xÞ ¼  W1 ð39:3Þ
n¼1
maxscore

where:
classscore Botnet attribute based on botnet classification (Payload, infection,
activation, propagation, operation algorithm)
maxscore Total classscore (In this case, payload + infection + activa-
tion + propagation + operation algorithm = 5)
W1 Weight of group of selected individuals/class, to get W1 , equation
used: avg (botnet_class *4).

Then for selection method, Roulette wheel is used to create new population
with better fitness value. The formula in Eq. (39.4) is used to find the probability
of the selected chromosome. Genetic Algorithm (GA) is a heuristic search that
simulates the process of natural evolution.

fi
pi ¼ PN ð39:4Þ
j¼1 fi

where:
fi The fitness
PN The Sum of all fitness.
fi
j¼1

39.4 Findings

In this section a proof of concept on how the mobile botnet works is discussed. The
Android.Fakedefender code was downloaded from Contagio [21] and it is a type of
fake antivirus software [23]. Once installed, the application displays an icon as
Android Defender and it will keep asking the victim to grant Device Admin
Privilege. Once the Device Admin Privilege is granted, the mobile botnet will
monitor and control the victim’s smartphone and the reverse engineered codes as
displayed in Fig. 39.4.
It started by displaying fake security status of a smartphone and tried to con-
vince the victim to purchase a full version of the software, which is used to clean
non-existing infection or fake infection. The message keeps on pop-up until the
payment is made or until the infection is removed. Until payment is made, it will
39 Systematic Analysis on Mobile Botnet Detection Techniques … 395

Fig. 39.4 Reverse engineered code

prohibit victim from doing anything on the phone. Next, it collected information
such as SMS messages in the device’s inbox, phone number, OS version, device
manufacturer, location, and sends it to Command and Control (C&C) server.
Based on the experiment conducted, it can be concluded that in order to produce
an effective mobile botnet detection algorithm, all the mobile botnet behaviors
should be further investigated and evaluated to identify the pattern used in existing
mobile botnet for future prediction of the new mobile botnet.

39.5 Conclusions and Future Work

This paper presented a systematic analysis for mobile botnet detection techniques
where gaps in the existing works have been identified for further improvement.
Based on the findings for further improvement of the mobile botnet systematic
analysis and experiment conducted, it can be concluded that mobile botnet has it
owns pattern and mechanism to spread and infect the victim smartphone. There-
fore, further investigation needs to be carried out and this research paper has
proposed an improved genetic algorithm to be integrated for a better solution of
the mobile botnet classification and detection mechanism. This paper is part of
larger project to design mobile botnet self-destruction model.

Acknowledgments The authors would like to express their gratitude to Universiti Sains Islam
Malaysia (USIM) for the support and facilities provided. This research paper is supported by
Universiti Sains Islam Malaysia (USIM) grant [PPP/FST/SKTS/30/12712] and [PPP/GP/FST/
SKTS/30/11912].
396 M.Z.A. Rahman and M.M. Saudi

References

1. F-Secure: F-Secure Mobile Threat Report Q3. http://www.f-secure.com/static/doc/labs_

global/Research/Mobile_Threat_Report_Q3_2013.pdf (2014). Accessed 24th Feb 2014
2. Teufl, P., Ferk, M., Fitzek, A., Hein, D., Kraxberger, S., Orthacker, C.: Malware detection by
applying knowledge discovery processes to application metadata on the Android market
(Google Play). Secur. Commun. Netw. doi:10.1002/sec.675 http://dx.doi.org/10.1002/sec.675
(2013). Accessed 24th Feb 2014
3. Yandex: Yandex | Android apps market: download free and paid Android application. http://
market.yandex.ru/ (2014). Accessed 24th Feb 2014
4. Sullivan, S.: Bad bad piggies on google play. http://www.f-secure.com/weblog/archives/
00002566.html (2013). Accessed 24th Feb 2014
5. Grizzard, J.B., Sharma, V., Nunnery, C., ByungHoon Kang, B., Dagon, D.: Peer-to-peer
botnets: overview and case study. In: Proceedings of First Workshop on Hot Topics in
Understanding Botnets (2007)
6. Xiang, C., Binxing, F., Lihua, Y.: Andbot: towards advanced mobile botnets. In: 4th Usenix
Workshop on Large-scale Exploits and Emergent Threats (2011)
7. Lee, W.K., Wang, C., Dagon, D.: Botnet Detection: Countering the Largest Security Threat.
Springer, New York (2007)
8. Irfan, A.: Could sexy space be the birth of the sms botnet? http://www.symantec.com/
connect/blogs/could-sexyspace-be-birth-sms-botnet (2013). Accessed 24th Feb 2014
9. Porras, P.A., Saidi, H., Yegneswaran, V.: An analysis of the iKee.B iPhone botnet. In:
Proceedings of the 2nd International ICST Conference on Security and Privacy on Mobile
Information and Communications Systems (Mobisec) (2010)
10. Security Alert: New sophisticated Android malware DroidKungFu found in alternative
Chinese app market. http://www.csc.ncsu.edu/faculty/jiang/DroidKungFu.html. Accessed
24th Feb 2014
11. Zeng, Y., Hu, X., Shin, K.G.: Design of SMS commanded-and-controlled and P2P-structured
mobile botnets. University of Michigan Technical Report (2010)
12. Mulliner, C., Seifert, J.P.: Rise of the iBots: owning a telco network. In: Proceedings of
MALWARE 2010, France, pp. 71–80 (2010)
13. Traynor, P., Lin, M., Ongtang, M.: On cellular botnets: measuring the impact of malicious
devices on a cellular network core. In: Proceedings of CCS 2009, Chicago, USA (2009)
14. Yusoff, M.N., Jantan, A.: Optimizing decision tree in malware classification system by using
genetic algorithm. Int. J. New Comput. Arch. Their Appl. (IJNCAA) 1(3), 694–713 (2011)
15. Mathew, S.E., Ali, A., Stephen, J.: Genetic algorithm based layered detection and defense of
HTTP botnet. ACEEE Int. J. Netw. Secur. 5(1), 50–61 (2014)
16. Monther, A., Rami, A.: MALURLS: a lightweight malicious website classification based on
URL features. J. Emerg. Technol. Web Intell. 4(2) (2012)
17. Noreen, S., Murtaza, S., Zubair Shafiq, M., Farooq, M.: Evolvable malware. In: GECCO ‘09
Proceedings of the 11th Annual Conference on Genetic and Evolutionary Computation
(2009)
18. Lei, L., Wang, H., Wu, Q.: Improved genetic algorithms based path planning of mobile robot
under dynamic unknown environment, mechatronics and automation. In: Proceedings of the
2006 IEEE International Conference (2006)
19. Zhou, Y., Jiang, X.: Dissecting Android malware: characterization and evolution. In:
Proceedings of the 33rd IEEE Symposium on Security and Privacy (Oakland 2012), San
Francisco, CA, May 2012
20. Zhou, Y., Jiang, X.: Android malware genome project. http://www.malgenomeproject.org/
(2012). Accessed 24th Feb 2014
39 Systematic Analysis on Mobile Botnet Detection Techniques … 397

21. Contagio Mobile: Mobile malware mini dump. http://contagiominidump.blogspot.com/

(2013). Accessed 24th Feb 2014
22. Dunham, M.H.: Data mining: Introductory and Advanced Topics. Prentice Hall, New Jersey
(2002)
23. Symantec Corporation: Android.Fakedefender. http://www.symantec.com/security_response/
writeup.jsp?docid=2013-060301-4418-99 (2013). Accessed 24th Feb 2014
Chapter 40
An Empirical Study of the Evolution
of PHP MVC Framework

Rashidah F. Olanrewaju, Thouhedul Islam and N. Ali

Abstract Commercial, social and educational importance of web technology has

tremendously increased research activities in web programming/scripting. Several
methods for writing PHP codes such as Object Oriented Programming (OOP),
Procedural PHP coding and Model View Controller (MVC) pattern have been
proposed. Model View Controller (MVC) which is one of the most powerful
method for developing PHP application has many variant such Laravel, Symfony,
CodeIgniter, CakePHP etc. However, selection of best MVC framework among
the variants is of concern to the programmers as well as project managers, espe-
cially when managing big applications. Hence, performance evaluation criterions
are required. This paper discusses the MVC based most famous PHP frameworks,
evaluate their performance and it was found that Laravel outperforms other MVC
framework, hence Laravel is proposed as the most suitable PHP framework for
future web technology.

Keywords MVC
Laravel
PHP framework
CakePHP
CodeIgniter

Symfony

R.F. Olanrewaju (&)
T. Islam

Department of Electrical and Computer, Kulliyyah of Engineering, International Islamic
University Malaysia, P.O. Box 10, 50728 Kuala Lumpur, Malaysia
e-mail: [email protected]
T. Islam
e-mail: [email protected]
N. Ali
College of Information Technology, Universiti Tenaga Nasional Malaysia, Selangor,
Malaysia
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 399

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_40
400 R.F. Olanrewaju et al.

40.1 Introduction

The rapid development of internet for web based application indicates a higher
demand of reliability, scalability, security and maintainability of coding method-
ology. PHP, a scripting tool for web that enable dynamic interactive web devel-
opment such intuitive, compiled fast, cross platform, open source, flexibility as
well as required minimal setup [1]. This became one of the important web
development language thus, PHP is one of the most powerful programming lan-
guage in the web world. Several developers choose to deploy application based on
PHP putting all the issues such as data access, business logic, and data repre-
sentation layer together [2]. This in turn create development problems especially
for big projects. To solve this problem, MVC design pattern brings an effective
ways to separate code in layers from each other based on each layer activities.
MVC design pattern is a proven effective way to develop application such as
CakePHP, CodeIgniter, Laravel, Symfony. The main methods of MVC to spilt an
application into separate layer that can work separately and produce same result.
The advantage of using MVC pattern are:
• Standard, consistency and predictability
• Software components or building-blocks so that developers can share and reuse
code [3]
• A model or standard architecture that allows easy visualization of how the
entire system works [4]
• Reusable and thoroughly tested code in the libraries, classes and functions [5].
• Well-structured code using architectural pattern [6].
• Security, interoperability and Maintenance.
Although MVC based framework (CakePHP, Laravel, CodeIgniter) has number
of advantages [7], however, selecting of best PHP framework is still a concern.
This is because all of the framework does not cover all aspect of web applications.
This study evaluate most famous PHP frameworks based on MVC design model
and it performance as well as proposed the best efficient PHP MVC framework for
future web development.

40.2 Materials and Methods

The operation of Model View Controller (MVC) method is to spilt or separate the
different parts of code into layers such as view, data access, controlling user’s
requests and forward request to relevant layers [8].
The MVC pattern’s title is a collation of three core parts: Model, View, and
Controller. A visual representation of a complete and correct MVC pattern looks
like the following in Fig. 40.1.
40 An Empirical Study of the Evolution of PHP MVC Framework 401

Fig. 40.1 MVC mechanism

MODEL
MA
NIP
TES UL
AT
DA ES
UP

VIEW CONTROLLER

SE
ES
ES
US
USER

Figure 40.1 shows the pattern and interaction with the user and the application
itself. It is a single flow layout of data, how it’s passed between each component,
and finally how the relationship between each component works.

40.2.1 Model

The Model is the name given to the permanent storage of the data used in the
overall design. It must allow access for the data to be viewed, or collected and
written to, and is the bridge between the View component and the Controller
component in the overall pattern [9].
One important aspect of the Model is that it’s technically ‘‘blind’’—by this, the
model has no connection or knowledge of what happens to the data when it is
passed to the View or Controller components [10]. It neither calls nor seeks a
response from the other parts of the component; its main purpose is to process data
into its permanent storage, seek and prepare data to be passed along to the other
parts.
The Model cannot simply be assumed as a database toolkit only, or a gateway
to another system which handles the data process [11]. The Model represents a
gatekeeper to the data itself, asking no questions but accept all requests which
comes its way. Often this most complex part of the MVC system, the Model
component is also the pinnacle of the whole system since without it there will be
no connection between the Controller and the View.
402 R.F. Olanrewaju et al.

40.2.2 View

The View is a module where data, requested from the Model is viewed and its final
output is determined. Traditionally in web application use MVC for development,
the View is the part of the system where the HTML is generated and displayed.
The View also ignites reactions from the user, who then goes on to interact with
the Controller. The basic example of this is a button generated by the View, which
a user clicks and triggers an action in the Controller.

40.2.2.1 Misconceptions About View

There are some misconceptions held about View components, particularly by web
developers using the MVC pattern to build their application. For example, many
mistaken the View as having no connection whatsoever to the Model and that all
of the data displayed by the View is passed from the Controller. In reality, this
flow disregards the theory behind the MVC pattern completely. Fabio Cevasco’s
article, The CakePHP Framework: Your First Bite [1] demonstrates this confused
approach to MVC in the CakePHP framework.
In order to correctly apply the MVC architecture, there must be no interaction
between models and views: all the logic is handled by controllers [13]. Further-
more, the description of Views as a template file is inaccurate. The View is really
much more than just a template, the modern MVC inspired frameworks have
bastardised the view almost to the point that no one really cares whether or not a
framework actually adheres to the correct MVC pattern or not. It’s also important
to mention that the View part is never given data by the Controller. There is no
direct relationship between the View and the Controller without the Model in
between them.

40.2.3 Controller

The third component of the triad is the Controller. Its job is to handle data the user
submits as well as update the Model accordingly. The Controller can be summed
up as a collector of information, which then passes it on to the Model to be
organized for storage, and does not contain any logic other than collecting user
input [9]. The Controller is also only connected to a single View and to a single
Model, making it a one way data flow system, with handshakes and signoffs at
each point of data exchange. Controller is only given tasks to perform when the
user interacts with the View first, and that each Controller function is a trigger, set
off by the user’s interaction with the View [14]. The most common mistake made
by developers is confusing the Controller as a gateway, and ultimately assigning it
functions and responsibilities that the View should do (this is normally a result
of the same developer confusing the View component as a template).
40 An Empirical Study of the Evolution of PHP MVC Framework 403

Additionally, it’s a common mistake to assign the Controller functions that gives it
the sole responsibility of crunching, passing, and processing data from the Model
to the View. Nonetheless, the MVC pattern relationship should be kept between
the Model and the View.

40.3 MVC Design Frameworks

40.3.1 CodeIgniter

CodeIgniter is an open source rapid development web application framework, for

use in building dynamic websites with PHP. Its goal is to enable to develop
projects much faster than writing code from scratch, by providing a rich set of
libraries for commonly needed tasks, as well as a simple interface and logical
structure to access these libraries. The first public version of CodeIgniter was
released on February 28, 2006, and the latest stable version 2.1.4 was released July
8, 2013 [15].
CodeIgniter is loosely based on the popular Model-View-Controller develop-
ment pattern. While view and controller classes are a necessary part of develop-
ment under CodeIgniter, models are optional. CodeIgniter is most often noted for
its speed when compared to other PHP frameworks.

40.3.2 CakePHP

CakePHP is an open source web application framework. It follows the Model-

View-Controller (MVC) approach and is written in PHP, modeled after the con-
cepts of Ruby on Rails, and distributed under the MIT License [16].
CakePHP uses well-known software engineering concepts and software design
patterns, as Convention over configuration, Model-View-Controller, Active
Record, Association Data Mapping, and Front Controller.

40.3.3 Symfony

Symfony is a PHP web application framework for MVC applications. Symfony is

free software and released under the MIT license. The symfony-project.com
website was launched on October 18, 2005 [17].
404 R.F. Olanrewaju et al.

40.3.4 Laravel

Laravel is a free, open source PHP web application framework, designed for the
development of MVC web applications. Laravel is released under the MIT license,
with its source code hosted on GitHub.
The key design points of Laravel are:
• Bundles provide Laravel with a modular packaging system, and numerous
bundled features are already available for easy addition to applications.
• Eloquent ORM is an advanced PHP implementation of the active record pat-
tern, providing internal methods for enforcing constraints to the relationships
between database objects [18].
• Application logic is part of developed applications, either by using controllers,
or as part of route declarations. Syntax used for definitions is similar to the one
used by Sinatra framework.
• Reverse routing defines a relationship between links and routes, making it
possible for later changes to routes to be automatically propagated into relevant
links. When links are created by using names of existing routes, appropriate
URIs are automatically created by Laravel [19].
• Restful controllers provide an optional way for separating the logic behind
serving HTTP GET and POST requests.
• Class auto loading provides automated loading of PHP classes, without the
need for manual maintenance of inclusion paths. On-demand loading prevents
loading of unnecessary components; loaded are only those components which
are actually used [20].
• View composers are logical code units that can be executed when a view is
loaded.
• Migrations provide a version control system for database schemas, making it
possible to associate changes in the application’s code base and required changes
in the database layout, easing deployment and updating of applications [21].
• Unit testing plays an important role in Laravel, which itself has a large number
of tests for detecting and preventing regressions. Unit tests can be run through
the artisan command-line utility [22].
• Automatic pagination simplifies the task of implementing pagination, replacing
the usual manual implementation approaches with automated methods inte-
grated into Laravel [23].

40.4 Benchmarking

To evaluate the performance of four mentioned PHP frameworks; CodeIgniter

(CI), Symfony, CakePHP and Laravel. The best way to do the benchmarking is
by applying several criteria such as request per second, system load average,
40 An Empirical Study of the Evolution of PHP MVC Framework 405

memory usage, number of function calls and number of files required on each of
the MVC. To evaluate these four frameworks, web design which contains ‘‘hello
word’’ was carried out on apache (ab –c 200 –n 50,000), below are results of each
evaluation criteria.

40.4.1 Request per Second

This benchmarking is based on apache (ab –c 200 –n 50,000). Figure 40.2 shows
the performance comparison among four MVC: CI, CakePHP symphony and
Laravel. It can be seen that Laravel outperforms other MVC in terms of request
person. It was able to handle 3,000 request per second compare to others. In this
case, bigger output indicates perfect result hence, it denotes best performance.

40.4.2 System Average Load

The system average load is measure in relation to time. that is, in 1 min when
Apache Benchmark is complete, the smaller is better in the condition of (ab –c 200
–n 50,000). Figure40.3 shows the comparison of the four MVC in terms of system
load within 1 min. In this graph, Laravel contain lowest times (0.98) where
CakePHP contain maximum load time 5.1 per minute to load system. Based on
average time, lowest average time is better to run MVC application.

40.4.3 Memory Usage

This benchmarking checks how much memory is used in a one word ‘hello world’
page display. The smaller number of KB the better the memory usage. Figure 40.4
indicated that Laravel is about 518 KB compare to CI which is which 725 Kb

Fig. 40.2 Request per

second among PHP
framework
406 R.F. Olanrewaju et al.

Fig. 40.3 System average

load time

Fig. 40.4 Memory usage

follow by symphony with memory usage of about 1,711 then CakePHP with
2,824 KB. Hence, it can be concluded that Laravel memory usage is efficient.

40.4.4 Response Time

The time of page request to response from framework is one of the most important
criteria to evaluate MVC performance. It is calculated in millisecond. The lower
number of millisecond calculated the better performance. Figure 40.5 depicts the
result of all the four MVC used. Among all, Laravel came out to be with the least
response time, 4.46 ms compare to CI with 7.2 followed by symphony with 12
then CakePHP with about 14 ms.
40 An Empirical Study of the Evolution of PHP MVC Framework 407

Fig. 40.5 Response time for

various MVC

Fig. 40.6 Numbers of

functions called

40.4.5 Number of Function Calls

This test checks how many functions are calling for one ‘‘hello world’’ pages in
terms of Facebook XHProf [24]. Here smaller number of functions is most
effective for PHP framework evaluation. It can be seen from Fig. 40.7, Laravel
outperforms other MVC framework with the minimal number of function calls
with 238 calls compare to CakePHP 834 and others (Fig. 40.6).

40.4.6 Number of Files

The number of files included or required in one ‘hello world’ page. Less amount of
required files represent that such framework will be loaded first in-terms of file
408 R.F. Olanrewaju et al.

Fig. 40.7 Numbers of file

are required by various MVC

running. Smaller numbers of required files are highly appreciable. As shown in

Fig. 40.7, CI comes up with the least file of 22, then symphony with 15 files
followed by Laravel with 26, while CakaPHP loaded 37 files. Here CI outperforms
other frameworks

40.5 Results and Discussion

Based on different core criteria of PHP framework such as benchmark, pattern,

database access, field database, session, cache and library, this research compare
among four top PHP framework (Laravel, CakePHP, CodeIgniter and Symfony)
and their performance. Table 40.1 shows comparisons the frameworks in terms of
facilities.
From the Table 40.1, it is assumable that, Laravel has advantage over other
MVC PHP framework.
The four frameworks was also measured based on current trends for the future
PHP framework that will lead the next generation of web. Figure 40.2 shows the
details.
From Fig. 40.8, it is shown clearly that, Laravel took almost 26 % place in
world web development in 2013 by MVC pattern framework. Based on this
analysis, it is understandable that, Laravel is going to be one of the most famous
MVC pattern framework in PHP world with huge flexibility of deployment as well
as manage.
40 An Empirical Study of the Evolution of PHP MVC Framework 409

Table 40.1 Comparism among four MVC framework

Specification CakePHP CodeIgniter Laravel Symfony

MVC
DB
Fielddb UK
Auth
Validate
Session
Cache
Ajax UK
Db Eloquent
MVC type PMVC UMVC HMVC UMVC
MVC db AR AR ORM ORM
Upload
Form Objects Procedural Objects Objects
Xml
library UK

Fig. 40.8 Compare among

Laravel
all MVC-PHP framework
based on current trends 13.52% Phalcon
4.51%
Symfony
1.50% 25.85%
Codeigniter

7.52% Yii
Cake PHP
4.51% Zend
16.73%
7.62% Pear
Aura
7.62%
10.62% Others

40.6 Conclusion

An empirical study on major MVC pattern for PHP framework has been evaluated
in this paper.
The results obtained from evaluating the four frameworks: CodeIgniter (CI),
Symfony, CakePHP and Laravel using criteria such as request per second, system
load average, memory usage, number of function calls and number of files required
as well as available facilities in each framework. It shows that the Laravel out-
performs other MVC framework. A request per second of as high as 3,000 was
recorded for Laravel compare to others like CakePHP with as low as 750 request
per second. The results obtained for all other parameters such has storage, function
calls, number of files, response time etc. indicated that Laravel has huge flexibility
410 R.F. Olanrewaju et al.

of development of web application, it has more facilities for programmers that

makes it acceptable to all web programmer in terms of different criteria such as
intuitive, compiled fast, cross platform, open source and flexibility. It enable easy
migration, enriched library, template system, eloquent ORM and wide range of
community support that helps to develop application smoothly. All the criteria and
facilities of Laravel prove that, it is of opinion that Laravel would be the best
choice to deploy next generation PHP based web application.

References

1. Bergmann, S., Kniesel, G.: GAP: generic aspects for PHP. In: Proceedings of EWAS’06
(2006)
2. Bakken, S.S., Aublach, A., Schmid, E., et al.: PHP manual (The PHP Documentation Group).
https://php.net/manual/en/index.php, Accessed 10 March 2014
3. Nakajima, S., Hokamura, K., Ubayashi, N.: Aspect-oriented development of PHP-based web
applications, 34th annual IEEE computer software and applications conference workshops
(2010)
4. Veglis, A., Leclercq, M., Quema, V.: PHP and SQL made simple distributed systems online,
Volume 6 Issue 8, August 2005, Page 4
5. http://webcoderpro.com/blog/top-6-most-popular-php-frameworks-of-2013/. Accessed 21
Jan 2014
6. http://www.catswhocode.com/blog/top-10-php-frameworks-for-2014. Accessed 23 Jan 2014
7. http://www.sitepoint.com/best-php-frameworks-2014/. Accessed 20 Jan 2014
8. Ricca, F., Tonella, P.: Analysis and testing of web applications. In: Proceedings of 23rd
ICSE, pp. 25–34 (2001)
9. http://www.tonymarston.net/php-mysql/model-view-controller.html. Accessed 13 Dec 2013
10. Cui, W., Huang, L., Liang, L.J., Li, J.: The research of PHP development framework based on
MVC pattern, 4th international conference on computer sciences and convergence
information technology (2009)
11. http://www.sitepoint.com/the-mvc-pattern-and-php-1/. Accessed 15 Jan 2014
12. http://www.sitepoint.com/application-development-cakephp/. Accessed 17 Jan 2014
13. Enderlin, I., Giorgetti, A., Bouquet, F.: A constraint solver for PHP array, 6th international
conference on software testing, verification and validation workshops (2013)
14. http://matrix.include-once.org/framework/simplese. Accessed 19 Jan 2014
15. http://en.wikipedia.org/wiki/CodeIgniter. Accessed 10 Mar 2014
16. http://en.wikipedia.org/wiki/MIT_License. Accessed 28 Feb 2014
17. http://en.wikipedia.org/wiki/Symfony. Accessed 11 Mar 2014
18. http://www.developed.be/2013/07/16/php-frameworks-which-to-choose/. Accessed 16 Jan
2014
19. http://brianretterer.com/why-laravel-is-the-best-php-framework/. Accessed 15 Jan 2014
20. http://www.ruilog.com/blog/view/b6f0e42cf705.html. Accessed 8 Mar 2014
21. http://www.webdesignermag.co.uk/features/laravel-a-modern-php-framework/. Accessed 17
Jan 2014
22. Merlo, E., Letarte, D., Antoniol, G.: Automated protection of PHP applications against SQL-
injection attacks, 11th European conference on software maintenance and reengineering
(2007)
23. http://en.wikipedia.org/wiki/Laravel_(framework). Accessed 9 Mar 2014
24. http://www.php.net/manual/en/intro.xhprof.php. Accessed 10 Mar 2014
Chapter 41
Evolutionary Approach of General
System Theory Applied on Web
Applications Analysis

Aneta Bartuskova, Ondrej Krejcar and Kamil Kuca

Abstract This paper reviews evolution stages of websites and presents framework
for websites analysis, based on the evolutionary approach of general systems
theory. Development of websites and web-based applications is discussed,
according to their historical emergence, usage, increasing complexity and inte-
gration of new aspects and principles. Resulting individual stages of this devel-
opment are suggested in accordance with the evolutionary approach. Framework
for websites analysis and evaluation is then presented in a form of criteria list for
each defined stage. Website’s maturity (in the meaning of internet’s evolution) can
be then determined by confronting these criteria.

Keywords Web evolution
Websites evaluation
General systems theory

41.1 Introduction

The internet belongs to major providers of information and has a significant impact
on our lifestyle. It functions also as a platform for communication, used by a wide
range of users [1–4]. Considering an important role and rapid evolution of the
internet, it is useful to organize our knowledge of this evolvement. By identifying
individual stages of complexity in the field of website development, we can evaluate
any website or web-based application according to their evolvement or maturity.

A. Bartuskova (&)
O. Krejcar
K. Kuca

Faculty of Informatics and Management, Center for Basic and Applied Research, University
of Hradec Kralove, Rokitanskeho 62, 50003 Hradec Kralove, Czech Republic
e-mail: [email protected]
O. Krejcar
e-mail: [email protected]
K. Kuca
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 411

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_41
412 A. Bartuskova et al.

Framework for such evaluation, presented in this paper, is based on the evolutionary
approach of general systems theory. Use of these principles for websites is justified
by Yourdon’s application of general systems theory to the information technology
and systems [5].
General systems theory was invented by biologist Ludwig von Bertalanffy, who
identified similar principles across many fields of knowledge, such as biology,
social sciences, engineering or management [6]. Purpose of his theory was iden-
tification of laws pertaining these many branches and creation of suitable condi-
tions for their collective development [7]. One of the key aspects of this approach
is investigating systems as organizational units. Boulding defined two possible
approaches to general systems theory, which are more complementary than
competitive [8]. First approach relies on picking out general phenomena across
various disciplines and create theoretical models. Key idea of the second approach
is an arrangement of relevant constructs or empirical fields in a hierarchy, which
defines organization of individual units within the system. The second approach
will be used in this study and used for defining a hierarchy of evolution stages of
websites and web-based applications.

41.2 The Evolutionary Approach of General Systems

Theory

The second approach towards general systems theory was defined by Boulding as a
systematic approach leading to system of systems. Rapoport specified it as the
‘‘evolutionary approach’’, since levels of abstraction are increasingly complex,
marking the evolution of knowledge [9]. Each of these levels can be also defined
by input, output, throughput or process, feedback, control, environment and goal or
purpose, known as common elements of a system, which originated from Berta-
lanffy’s types of finality [10, 11, 12]. The input means an energy or a material
which is transformed by the system through some process, resulting in an output as
a product of system’s processing. The feedback is also a product of the process,
which returns to the system as an input. An evaluation of the input, process and
input is encapsulated in a control element, an environment denotes the area around
a system and a goal is a purpose of the system. Individual levels are described in
Table 41.1.

41.3 The Evolution of Websites: Review of Stages

We can distinguish individual stages of development in the field of websites and

web-based applications. The stages were identified by authors in accordance with
the evolutionary approach and appropriately to their complexity, usage and
41 Evolutionary Approach of General System Theory … 413

Table 41.1 Arrangement of levels by Vossen [8], further specified by Bertalanffy [9]
Level Name Included entities
1 The structure level Static structure, framework, arrangement
2 The simple dynamic Predetermined motions, simple machines
level
3 The control mechanism Transmission and interpretation of information
4 The open system Self-maintenance, self-reproduction
5 The genetic social level/ Division of labour, differentiated parts
Level of the cell
6 The animal level Increased mobility, teleological behaviour, self-
awareness, specialized information-receptors
7 The human level Self consciousness, self-reflexivity, speech
8 Human society Social organizations, units as a role in society
9 The transcendental level Ultimates and absolutes and the unknowables

historical appearance. Numbering of stages follows complexity of web develop-

ment and comply with an arrangement of levels in general systems theory by
Vossen and Bertalanffy [8, 9]. Individual stages will be reviewed in this section as
a basis for next research part of the paper. Descriptive figures are appended to
illustrate processing of the website on particular levels of complexity.

41.3.1 Static Website

The internet had started its existence with basic functionality—requesting and
displaying HTML files. Its proper functioning was ensured as a collaboration of
three fundamental technologies, specified by Penhaker [13]. These are: HTML,
URI and HTTP. A markup language, such as HTML or XML, forms a structure of
a web document. Such website is considered static from a view of a user, as it does
not change once loaded, until a user clicks on a next hypertext link [14, 15].
The first of evolutional levels is defined by Boulding as a static structure, also a
framework or an arrangement. Static website fulfils this concept and therefore it
was identified as the first evolutional level of websites.

41.3.2 Interactive Website

We consider an interactive website as a static website with capability of client-side

interaction. It is usually powered by a combination of HTML, CSS and JavaScript.
CSS is primarily used for styling HTML documents, but it can also convey
interaction like a hover effect. The latest specification of CSS—CSS3—also
brought a wide range of interactive features. It is essential that this interaction does
not change the website as it does not involve communication with server.
414 A. Bartuskova et al.

Temporary change of appearance as a result of user interaction is only visible on

this particular user’s browser [16, 17].
The evolutionary approach describes the second stage as a simple dynamic
level with predetermined motions. This level is represented by simple machines. In
relation to web applications, the second level corresponds with an interactive
website. Interaction here proceeds as a reaction of a website to a predetermined
event on the client side, followed by predetermined effect [18].

41.3.3 Dynamic Website

Dynamic website is defined by server-side scripting, which is often accompanied

by a database system. Server-side scripting requires server-side language, such as
PHP, ASP, Java or Perl. As a database system is usually used relational database
MySQL or MSSQL. Additional value of dynamic websites is possibility to use the
same page structure and design for dynamically loaded content, and also to use
different parts of structure and functionality depending on the request. This made
possible content management, search engines and variety of applications with
preserving and maintaining user data like webmail or online stores [19].
The third evolutional level is defined as the control mechanism, with a purpose
of transmission and interpretation of information. The dynamic website corre-
sponds with this description, as it enables inserting, updating and a retrieval of
information. It dynamically delivers information and it also provides mechanisms
for data validation. Content management system is a great example of a control
mechanism, which is a principal description of the third evolutional level in the
Boulding’s hierarchy (Fig. 41.1).

41.3.4 Social Web Application

In this stage, internet has evolved from provider of information to a socialization

platform, where every user can be a contributor to its content. Socialization of the
internet is covered by term Web 2.0. It is technologically associated with AJAX,
enabling rich functionality. Other feature are mash-ups, joining multiple data sources
or services to create a new service. Core aspects of Web 2.0 are then data (mash-ups),
functionality (AJAX) and socialization (community) [20, 21]. This approach leads
to a functionality dependence and information redundancy, as the same data occur
in many variations across the web, with little or none unified content organization.
The open system is a fourth stage according to Boulding, defined by an ability
of self-maintenance and self-reproduction. A Web 2.0 application can be perceived
as such an open system. Every user can be a contributor to this kind of application,
so its content is growing and is maintained without central interventions.
Social networks and wiki sites are great examples on self-maintenance. Activity of
41 Evolutionary Approach of General System Theory … 415

Fig. 41.1 Schema of first three levels of website development

sharing, a social aspect with none or loose terms, is typical for this stage. Self-
reproduction can be viewed e.g. in the form of mash-ups. Boulding also defined
this stage as a level at which life begins to differentiate itself from not-life [8]. In
accordance with this, social networks create living systems, which are changing
our social behaviour [20].

41.3.5 Semantic Web Application

Fifth level of websites can be represented by an aspiration for semantic web, which
aims to implement a logical structure with help of taxonomies and ontologies. The
term Web 3.0 is emerging as a possibility of combining today’s web with semantic
architectures [22]. Embedding web content in a logical structure enables not only
416 A. Bartuskova et al.

Fig. 41.2 Schema of the fourth level (on the left) and the fifth level (on the right)

machine-readable data, but also machine-understandable data [23]. Semantic web

aspires for providing information models and languages that embed semantic
contexts and metadata to enable automated processing of data [20].
The genetic social level is the fifth level of the evolutionary approach, defined
by a division of labour and differentiated and mutually dependent parts of a system
[8]. The semantic web application can represent such a model. By integrating and
encapsulating data, they can be handled differently and these differentiated parts of
a system enable a division of labour. Another aspect of the genetic social level and
also semantic web is a collaboration, as of a community, which needs to be active
in order to implement linguistic and structural concepts on the web. A collabo-
ration here represents a more elaborate social aspect with established terms and
structure (Fig. 41.2).

41.3.6 Adaptive Web Application

Adaptive web applications are suggested as the sixth level. Considering expansion
of mobile devices with internet access, an adaptation of web is necessary. This can
41 Evolutionary Approach of General System Theory … 417

be relevant to visual appearance, since desktops, notebooks, tablets, mobile phones

etc. have different range of screen dimensions and control possibilities [24].
Adaptation is also providing different functionalities according to capabilities of
chosen device or personalized content. New techniques are emerging to deal with
these requirements, such as HTML5 API. Evolution of websites is connected to a
development in ambient intelligence, ubiquitous computing and intelligent user
interfaces [25]. Web applications are becoming context-aware systems with three
basic functionalities—sensing, thinking and acting [26]. A research on sensors is
also closely connected with this stage, e.g. sensing movement, light, location,
proximity or biological signals [27].
The sixth level as the animal level is characterized by an increased mobility,
teleological behaviour, self-awareness and specialized information-receptors.
Adaptive web applications correspond with this stage as context-aware systems
[26]. Mobility can be understood as an ability to function appropriately in different
environments. Teleological behaviour ensures different functionalities according
to capabilities of chosen device. Adaptive web application is aware of its capa-
bilities, and of the relevant environment, which determines use of these capabil-
ities (Fig. 41.3).

41.4 Summary of the Web Evolutional Stages

Evolutional stages of websites and web-based applications were reviewed in

Sect. 41.3. A summary of conclusions, completed with remaining stages of
Bouldinǵs hierarchy (but yet expected in web evolution) is presented in
Table 41.2.
To the authors best knowledge, the current state of the internet can be placed
past levels 1–3, in level 4 and in the beginning of both levels 5 and 6. The
Boulding’s hierarchy has three more levels, 7 the human level, 8 human society
and 9 the transcendental level. According to their features and a position in the
hierarchy, we can roughly predict associated future stages of web applications. The
seventh level as the autonomous web systems, which encapsulate wide range of
functionality and are capable of complex decisions by their expert systems. The
eighth level as the cooperative web systems, capable of communication among
autonomous applications and delivering desirable performance without human
intervention.

41.5 Framework Proposal for Websites Analysis

On the basis of reviewed evolutional stages, list of criteria can be defined, which
are typical for particular stage. Some of the criteria are of course applicable also in
earlier stages, but they have been deliberately allocated to the stages, where their
418 A. Bartuskova et al.

Fig. 41.3 Schema of the sixth level of website development

effect is most pronounced or it can be completely performed by means available in

this particular stage. By confronting these criteria with factual state of evaluated
website, we can determine its maturity in the sense of the internet evolution.
These criteria should be in theory hierarchically based, with one level as nec-
essary requirement for the next level. This is not usually true with real web
applications, as there is no enforcement on the internet regarding quality of
development and poorly designed sites get as much space as standards-complying
websites. Also to a certain extent, it is possible to ignore possibilities of particular
stage and target more recent and complex issues. Finally, requirements may differ
for specific web projects (Table 41.3).
At the first level of static website, three main components are responsible for
result—structure, visual design and content. Structure is regarded here from pure
technical aspect via HTML. Visual design considers layout, colors, fonts etc.
ensured by CSS. Content has a quality perspective, especially its relevancy,
41 Evolutionary Approach of General System Theory … 419

Table 41.2 Stages of web applications in relation to the Boulding’s hierarchy

Level Type of website/ Essential feature/input
output
1 Static website Markup language, creating a static structure
2 Interactive website Scripting client-side language, styling language (hover),
creating simple dynamics
3 Dynamic website Scripting server-side language and a database, creating a control
mechanism
4 Social web Creating and sharing content in a community, forming a living
application system
5 Semantic web Established logical structure, ensuring differentiation of the
application content types
6 Adaptive web Adaptation and context-awareness—ability to function correctly
application in any environment
7 Autonomous web Encapsulating functionality and decision-making with internal
systems expert systems
8 Cooperative web Communication among autonomous web systems without
systems human intervention

Table 41.3 List of criteria, which apply to individual evolutional stages of websites
Level Type of website Principal criteria
(technologies/concepts)
1 Static website Structure (technical aspect)
(HTML, CSS) Visual design
Content
2 Interactive website User-website interaction
(CSS, JavaScript) Performance issues
Strategy of fallback
3 Dynamic website Content management
(PHP/ASP/.., MySQL/MSSQL) Navigation and usability
Search engine optimization
4 Social web application Customization
(social aspect—sharing) Communication
Rights and security issues
5 Semantic web application Structure (logical semantic aspect)
(social aspect—collaboration) Machine-understandable
6 Adaptive web application Responsiveness to device
(HTML5 API, JavaScript) Context-awareness (sensors)

correctness and amount. These three criteria remain essential throughout the whole
hierarchy at every level.
The second level adds CSS effects and JavaScript to create interaction between
a user and website. Additional criteria to consider in this stage are connected to
this interaction.
The third stage, dynamic website, uses server-side scripting language and
database primarily to ensure content (or functionality) management. With this
420 A. Bartuskova et al.

approach comes increased need for viable navigation, usability and search engine
optimization.
At the fourth level, the development is more of social nature than technological.
Website complying with this evolutional stage should facilitate customization i.e.
support for personalization and user content, and some form of communication
and feedback. Important back-end criteria are rights management and security
issues, which were already important in the third stage, but reach increasing
importance here, considering that users have much greater access to web appli-
cation than before [28].
The fifth stage, semantic web application, is defined by more complex social
aspect in the form of collaboration. This social aspect takes place on back-end part
of the internet in order to create underlying structure and rules [29].
Adaptive web applications on the sixth stage are, in current state, enabled
primarily by HTML5 API and JavaScript, along with infrastructure equipment as
sensors. General criteria for this evolutional stage are responsiveness to device
parameters and capabilities and context-awareness (which is on client-side
delivered by sensors).

41.6 Conclusions

This paper reviewed evolution stages of websites and web-based applications

according to their historical emergence, usage, complexity and integration of new
aspects. The authors believe that this classification clarified the evolvement of
internet from simple static websites to complex web applications. Individual stages
of this development were discussed with connection to the evolutionary approach
of general systems theory. Framework for websites analysis and evaluation was
then proposed in a form of criteria list for each defined stage. By confronting these
criteria with factual state of the evaluated website, we can define its maturity in the
meaning of internet’s evolution and available possibilities and necessities, which
are connected with individual evolution stages. Elaboration of this framework
along with concrete computation of websites evolution index is planned to be
subject of further studies.

Acknowledgment This work and the contribution were supported by project ‘‘SP/2014/05—
Smart Solutions for Ubiquitous Computing Environments’’ from University of Hradec Kralove.

References

1. Shneiderman, B.: Universal design. Commun. ACM 43, 84–91 (2010)

2. Yourdon, E.: Modern Structured Analysis. Yourdon Press, Prentice-Hall International,
Englewood Cliffs, New Jersey. ISBN 978-0135986240 (1989)
41 Evolutionary Approach of General System Theory … 421

3. Weinreich, H., Obendorf, H., Herder, E., Mayer, M.: Not quite the average: an empirical
study of web use. ACM Trans. Web, 2(1) (Article 1) (2008)
4. Boulding, K.: General Systems theory—the skeleton of science. Manag. Sci. vol. 2 No. 3,
197–208 Apr 1956 Reprinted in E:CO 6(1-2), 127–139 (1956/2004)
5. World Wide Web Foundation.: History of the Web. http://www.webfoundation.org/vision/
history-of-the-web/ (2008–2013)
6. Liou, C.Y., Cheng, W.C.: Manifold construction by local neighborhood preservation. In
Springer Lect. Notes Comput. Sci. 4985, 683–692 (2007)
7. Chen, A.Q., Harper, S.: Web Evolution: Method and Materials. Technical Report, University
of Manchester. http://wel-eprints.cs.man.ac.uk/74 (2008)
8. Hofkirchner, W.: Ludwig von bertalanffy. forerunner of evolutionary systems Theory. In: Gu,
J., Chroust, G. (eds.) The New Role of Systems Sciences For a Knowledge-based Society,
Proceedings of the First World Congress of the International Federation for Systems
Research, Kobe, Japan, 6 (2005)
9. Loke, S.: Context-Aware Pervasive Systems: Architectures for a New Breed of Applications.
Taylor and Francis Group (2007). ISBN 978-0849372551
10. Gillies, D. A.: Understanding General Systems Theory. Nursing Management A Systems
Approach, pp. 56–74. W. B. Saunders Company, Philadelphia (1982)
11. Doyle, M.: Beginning PHP 5.3. Wiley, Indianapolis, Indiana. ISBN 978-8126527977 (2010)
12. Bertalanffy.: An outline of general system theory. Br J Philos Sci 1(2), 134–165 (1950)
13. Hofkirchner, W.: General System Theory. The origins of General System Theory (GST).
http://www.hofkirchner.uti.at/wp-content/uploads/2010/10/GSTcombined.pdf (2010)
14. Hercik, R., Slaby, R., Machacek, Z., Koziorek, J. Correlation methods of OCR algorithm for
traffic sign detection implementable in microcontrollers. Adv. Int. Syst. Comput. 189,
381–389 (2013). ISSN: 2194-5357
15. Krawiec, J., Penhaker, M., Krejcar, O., Novak, V., Bridzik, R., Web system for
electrophysiological data management. In: Proceedings of 2010 Second International
Conference on Computer Engineering and Applications ICCEA 2010, 19–21 Mar 2010,
p. 404–407. Bali Island, Indonesia, vol. 1 (2010)
16. Machaj, J., Brida, P.: Performance comparison of similarity measurements for database
correlation localization method. Lect. Notes Comput. Sci. 6592, 452–461, (2011) ISBN 978-
3-642-20041-0
17. BRIDA, P., MACHAJ, J.: A Novel enhanced positioning trilateration algorithm implemented
for medical implant in-body localization. Int. J. Antennas Propag. 2013, 10 (2013) Article ID
819695, ISSN: 1687-5877
18. Krejcar, O., Penhaker, M., Janckulik, D., Motalova, L.: Performance test of multiplatform
real time processing of biomedical signals. In: Proceedings of 8th IEEE International
Conference on Industrial Informatics, INDIN 2010, 13–16 July 2010, pp. 825–839. Osaka,
Japan (2010). doi:10.1109/INDIN.2010.5549635
19. Berners-Lee, T., Hendler, J., Lassila, O.: The Semantic web—a new form of web content that
is meaningful to computers will unleash a revolution of new possibilities. Sci. Am. 284(5),
35–43 (2001)
20. Ankolekar, A., Krötzsch, M., Tran, T., Vrandecic, D.: The two cultures: mashing up web 2.0
and the semantic web. In: Proceedings of the 16th International Conference on World Wide
Web, 08–12 May 2007, Banff, Alberta, Canada (2007)
21. Penhaker, M., Cerny, M.: Sensitivity Analysis and Application of Transducers, In: 5th
International Summer School and Symposium on Medical Devices and Biosensors, 01–03
Jun 2008, pp. 103-106. Hong Kong, P.R.China, (2008) doi:10.1109/ISSMDBS.2008.4575028
22. Rapoport, A.: General systems theory. Int. Encycl. Soc. Sci. 15, 452–458 (1968)
23. Marzano, S., Aarts, E.: The New Everyday View on Ambient Intelligence. Uitgeverij 010
Publishers (2003). ISBN 978-9064505027
24. Behan, M., Krejcar, O.: Modern smart device-based concept of sensoric networks. EURASIP
J. Wirel. Commun. Net. 2013(155) (2013)
422 A. Bartuskova et al.

25. Schmidt, A.: Ubiquitous computing—computing in context. Ph.D. thesis, Computing

Department, Lancaster University (2002)
26. Frain, B.: Responsive Web Design with HTML5 and CSS3. Packt Publishing, UK (2012).
ISBN 978-9350237885
27. Longo, L., Kane, B.: A novel methodology for evaluating user interfaces in health care. In:
24th IEEE International Symposium on Computer-Based Medical Systems CBMS 2011,
Bristol, England, June 27–30 (2011)
28. Vossen, G., Hagemann, S.: Unleashing Web 2.0, Morgan Kaufmann, San Francisco,
pp. 1–68, ISBN 9780123740342, doi:10.1016/B978-012374034-2.50002-2 (2007)
29. Wahlster, W., Dengel A. (eds.): Web 3.0: Convergence of Web 2.0 and the Semantic Web.
Deutsche Telekom Laboratories, Technology Radar Feature Paper, Edition II/2006, June,
pp. 1–23 (2006)
Chapter 42
A Novel Distributed Image Steganography
Method Based on Block-DCT

Rosemary Koikara, Dip Jyoti Deka, Mitali Gogoi and Rig Das

Abstract Distributed Image Steganography (DIS) is a method of hiding secret

information in multiple carrier images, making it more difficult to trace than
conventional steganographic techniques, and requiring a collection of affected
images for the retrieval of the secret data. In this paper we concentrate on per-
forming DIS on grayscale images using Block-DCT (Discrete Cosine Transfor-
mation). Distributed Image Steganography using Block-DCT adds to the security
of DIS by embedding the secret data in the Frequency Domain. This makes the
carrier images more immune to various steganalysis attacks as the secret data is
more evenly distributed amongst the pixels of the carrier images making it more
difficult to determine its existence. We use parity check in order to compensate for
round-off errors that are typically associated with DCT.

Keywords Steganography  Distributed steganography  Block-DCT  PSNR

R. Koikara  D.J. Deka  M. Gogoi

Department of Computer Science and Engineering and Information Technology, Don Bosco
College of Engineering and Technology, Guwahati 781017, Assam, India
e-mail: [email protected]
D.J. Deka
e-mail: [email protected]
M. Gogoi
e-mail: [email protected]
R. Das (&)
Department of Computer Science and Engineering, National Institute of Technology,
Rourkela 769008, Orissa, India
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 423

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_42
424 R. Koikara et al.

42.1 Introduction

Steganography is the art of hiding information in ways that prevent the detection
of hidden messages [1, 2]. It includes a vast array of secret communication
methods that conceal the message’s very existence. Some of the more common
methods include invisible inks, microdots, character arrangement, digital signa-
tures, covert channels, and spread spectrum communications [3]. Many conven-
tional steganographic schemes hide the secret data in a single host image. These
techniques include least significant bit (LSB) insertion or frequency domain
embedding using the Discrete Cosine Transform (DCT), Discrete Fourier Trans-
form (DFT) or Wavelet Transform (WT) [2, 4].
However, a common weakness of these techniques is that the secret data are all
in a single information-carrier, and the secret data cannot be revealed completely
[5], if the information-carrier is lost or crippled. Use of many duplicates may
overcome the weakness but increase the danger of security exposure. Moreover
conventional steganographic methods have restricted data hiding capacity.
Distributed Image Steganography overcomes these shortcomings by using a
(k, n) threshold based image secret sharing technique for k B n and allows large
information payload embedding by generating n steganographic images. DIS
allows (i) k or more steganographic images to reconstruct the secret image, and
(ii) (k - 1) or fewer images cannot reveal the secret image [1]. Figure 42.1 shows

Fig. 42.1 The block diagram of a simple distributed steganographic system with (3, 5) threshold
scheme
42 A Novel Distributed Image Steganography Method … 425

the block diagram of a simple DIS system with (3, 5) threshold scheme that means
the secret image will be distributed within 5 stego images and any 3 stego images
are enough to regenerate the secret image.
Distributed Image Steganography in Block-DCT is refinement of the existing
DIS technique in order to enhance the security of the secret data against various
steganalysis attacks. In this paper, we have implemented Distributed Image Ste-
ganography by means of Block-DCT. We take the cover images and embed the
shares of the secret data into the transformation domain of the cover images.
During extraction, the shares need to be retrieved from the transformation domain
of the stego images in order to recreate the original secret data. This whole process
comprises of a series of computationally intensive operations and hence is
impractical to deploy steganalysis for DIS in a massive scale.
This paper is organized as follows. Section 42.2 explains some related works of
Distributed Steganography. The proposed novel embedding and extraction algo-
rithms for DIS using Block-DCT are explained in Sect. 42.3. All experimental
results are shown in Sects. 42.4 and 42.5 concludes.

42.2 Related Work

Different researchers employed different techniques for the purpose of distributing

secret image over a set of stego images. Shamir’s Secret Sharing Scheme and
Thien and Lin’s Secret Image Sharing Scheme are the two essential processes to
protect secret image in DIS. Following are some of the related works carried out
by some of the researchers.

42.2.1 Thien and Lin’s Secret Image Sharing Scheme [6]

Suppose we want to divide the secret image D into n shadow images (D1, …,Dn),
and the secret image D cannot be revealed without r or more shadow images. In the
proposed method, they generate the r - 1 degree polynomial, by letting the r
coefficients be the gray values of r pixels. Therefore, the major difference between
Thien and Lin’s method and Shamir’s [7] is that they use no random coefficient.
Because the gray value of a pixel is between 0 and 255, they took the prime number
p be 251 which is the greatest prime number not larger than 255. To apply the
method, it must truncate all the gray values 251–255 of the secret image to 250 so
that all gray values are in the range 0–250. The image is divided into several
sections. Each section has r pixels, and each pixel of the image belongs to one and
only one section. For each section j, define the r - 1 degree polynomial as:
qj(x) = (a0 + a1x + _ + ar-1 xr-1) mod 251, Where a0, …,ar-1 are the r pixels of
the section, and then evaluate qj(1), qj(2), …,qj(n), The n output pixels qj(1) - qj(n)
of this section j are sequentially assigned to the n shadow images. Since for each
426 R. Koikara et al.

given section (of r pixels) of the secret image, each shadow image receives one of
the generated pixels; the size of each shadow image is 1 / r of the secret image. The
reveal phase uses any r (of the n) shadow images, and the Lagrange’s interpolation
to extract the secret image.

42.2.2 An Estimation Approach to Extract Multimedia

Information in Distributed Steganographic Images [1]

In this paper, a blind steganalysis technique is been proposed to attack DIS in

which no host image is required for detecting and extracting hidden information.
To develop this counter-measure for DIS they have put two assumptions: (i) One
hidden image in a set of unsuspected steganographic images, (ii) Threshold value
k is known.
The counter-measure process consists of three modules:
(i) Detection Module (DM) is responsible for detecting possible stegano-
graphic images
(ii) Estimation Module (EM) is responsible for extracting image shares
embedded in steganographic images
(iii) Reconstruction Module (RM) is responsible for combining quantized image
shares to reconstruct the secret image

42.3 Proposed Novel Method for Distributed Image

Steganography Based on Block-DCT

As Thien and Lin’s method is primarily based on Spatial Domain and there is also
a Steganalysis method to counter the LSB based DIS [1], a novel method for DIS
based on Block-DCT is been proposed in this paper which will add much more
security to the secret image as the secret information is being embedded in fre-
quency domain and extraction of the secret information is much more difficult than
spatial domain based techniques.
DIS using Block-DCT is a refinement of DIS. Here we use a polynomial
equation to create shares of the secret image and embed these shares into multiple
cover images that have been transformed using Block-DCT. The schematic/block
diagram of the whole process is given in Figs. 42.2 and 42.3.
Our novel algorithm for DIS based on Block-DCT is based on (k, n) Threshold
Scheme and has two parts, one for Embedding the Secret Image inside n–Cover
Images and another for Extracting the Secret Image from k–Stego Image. The
Secret Image is divided into several sections. Each section has k pixels, and each
pixel of the secret image belongs to one and only one section. For each section j,
we define the following k - 1 degree polynomial:
42 A Novel Distributed Image Steganography Method … 427

Fig. 42.2 Insertion of a secret image inside n–cover images

Fig. 42.3 Extraction of secret image from k–stego images

428 R. Koikara et al.

 
pj ðxÞ ¼ a0 þ a1 x þ    þ ak1 xk1 mod 256 ð42:1Þ
  
qj ðxÞ ¼ floor a0 þ a1 x þ    þ ak1 xk1 =256 ð42:2Þ

where, value of x ranges from 1 to n and a0 to ak - 1 are Secret Image’s k number

of pixel’s intensity values. These intensity values changes sequentially e.g., first
we take 1–8 pixels’ intensity values then 9–16 pixels’ intensity values and so on.
Modulus and Floor operations are performed using 256 as divisor because a grey
level image has 256 different intensity levels. Equation (42.1) finds the Remainder
Value after performing Modulus operation and (42.2) finds the Quotient value after
performing Floor operation. This method consists of several phases as explained
below.

42.3.1 Block-DCT

We have created blocks of size 8 9 8 of the cover images and perform DCT
operation on each of these blocks. This makes sure that the secret image is evenly
distributed amongst all the pixels on the 8 9 8 blocks.

42.3.2 Polynomial Equation

A polynomial equation is used to divide the secret image into its shares. The
polynomial equation used:

Sx ði; jÞ ¼ Iði  k þ 1; jÞ þ Iði  k þ 2; jÞx þ    þ Iði  k þ k; jÞxk1 ð42:3Þ

where, Sx is the share of the secret image I and j denote the pixel positions, I is
the secret image, k is the threshold of the secret sharing scheme. We calculate both
the remainder and the quotient of this equation as follows:
 
Mx ði; jÞ ¼ Iði  k þ 1; jÞ þ Iði  k þ 2; jÞx þ    þ Iði  k þ k; jÞxk1 mod 256
ð42:4Þ
 
Qx ði; jÞ ¼ floor ðIði  k þ 1; jÞ þ Iði  k þ 2; jÞx þ    þ Iði  k þ k; jÞxk1 =256Þ
ð42:5Þ

where, Mx is the remainder of the polynomial equation, Qx is the quotient of the

polynomial equation, mod is the operation that calculates the remainder floor gives
the nearest integer that is less than or equal to the real value. We need the values of
both the remainder and the quotient to reconstruct the image during extraction.
42 A Novel Distributed Image Steganography Method … 429

42.3.3 Lagrange’s Interpolation

Lagrange’s interpolation is the Nth degree polynomial approximation formula to

the function P(x), which is known at discrete points xi, i = 0, 1, 2, …, Nth

ðx  x2 Þðx  x3 Þ. . .ðx  xn Þ ðx  x1 Þðx  x3 Þ. . .ðx  xn Þ

PðxÞ ¼ y þ y
ðx1  x2 Þðx1  x3 Þ. . .ðx1  xn Þ 1 ðx2  x1 Þðx2  x3 Þ. . .ðx2  xn Þ 2
ðx  x1 Þðx  x2 Þ. . .ðx  xn1 Þ
þ  þ y
ðxn  x1 Þðxn  x2 Þ. . .ðxn  xn1 Þ n
ð42:6Þ

42.3.4 Parity Check

This is an error detection and correction code used to eliminate the rounded-off
errors that may arise due to transformation into the frequency domain. Before
performing Block-DCT over the cover image, the cover image is changed into
DOUBLE format. After performing Block-DCT the pixels intensity values gets
changed into frequency domain which has negative fraction values. Now as the
LSB of these negative fraction values can’t be modified to insert the Secret
Image’s data, it should have to be changed into positive integer number. After
modifying the LSB of the cover images and performing Inverse 2-D DCT, to
generate the stego image it is required to convert these pixels intensity value into
UINT8 format from DOUBLE format (as the original cover image was in UINT8
format). Now at the time of change from DOUBLE to UINT8 format the fraction
value gets rounded off. But the modification on LSB has been done in those
fraction part itself. For example if a pixel’s intensity is 169.78 then it gets rounded
off into 170 and if it is 152.21 then rounded off to 152. So the value of every LSB
gets changed. Thus a small amount of change can change the actual Secret Image
into havoc [8]. To reduce this loss of information an algorithm is devised to detect
and correct this error. The algorithm we have formulated is a modified form of
parity check. This method does not completely eliminate rounded-off error; it just
reduces to a certain extent. Following is the devised Parity Checking Algorithm.

42.3.4.1 Algorithm for Parity Check During the Insertion of Secret

Image

Input: Stego-image
Output: Stego-image with parity check information embedded into its 1st and
2nd LSB Position

Step-1: Calculate even parity and embed it into the 1st LSB position
Step-2: Calculate odd parity and embed it into the 2nd LSB position
430 R. Koikara et al.

42.3.4.2 Algorithm for Parity Check During the Extraction of Secret

Image

Input: Stego Image

Output: Stego Image with Corrected pixel value

Step-1: Let x = Pixel value of the Stego Image.

Step-2: Calculate even parity of x and embed it in x’s 1st LSB
Step-3: Calculate odd parity of x and embed it in x’s 2nd LSB
Step-4: If x is equal to the Pixel value then it is correct or else we need to
increment or decrement the pixel value respectively.
Step-5: Repeat Step-2 and Step-3 till x becomes equal to the pixel value

42.3.5 Proposed Novel Algorithm for DIS Based on Block-

DCT

Embedding Algorithm
Input: n number of M 9 N Carrier Images and a P 9 Q Secret message/
Image
Output: n number of M 9 N Stego-Images

Step-1: Read Secret Image and Cover Images

Step-2: Divide the Cover Image into non overlapping blocks of size 8 9 8 and
apply 2-D DCT on each of the blocks of the cover image.
Step-3: Sequentially take k–number of not-shared-yet pixels of the Secret Image
and use (42.1) and (42.2) to find the n number of Remainder and
n number of Quotient value.
Step-4: Change the 3rd LSBs of DCT transformed n Cover Images to insert
each set of Remainder and Quotient values found in Step-4 in each of
the Cover Images.
Step-5: Add parity bit information into 1st and 2nd LSBs of every pixel of n-
Cover Images using the algorithm proposed in Sect. 42.3.4.1.
Step-6: Repeat Step-3, Step-4 and Step-5 until all the pixels of the Secret Image
are embedded into n-Cover Images.
Step-7: Write all the n Stego Images into the disk.
Extraction Algorithm
Input: k number of M 9 N Stego-Images
Output: A P 9 Q Secret Image
42 A Novel Distributed Image Steganography Method … 431

Step-1: Read k number of Stego Images

Step-2: Divide the Stego Image into non overlapping blocks of size 8 9 8 and
apply 2-D DCT on each of the blocks of the Stego image.
Step-3: Extract k number of Remainders and k number of Quotients from the k
Stego Images by extracting the 3rd LSBs of the pixels.
Step-4: Use Parity bit Checking as described in Sect. 42.3.4.2 to reduce the
rounded off error (as described in Sect. 42.3.4).
Step-5: Use Lagrange’s Interpolation to retrieve k number of pixel’s intensity.
Step-6: Repeat Step-3 and Step-4 until the total number of pixels of the Secret
Image are processed.
Step-7: Write the Extracted Secret Image into the disk.

42.4 Simulation and Results

In this section, some experiments are carried out on our proposed algorithm for
Distributed Image Steganography (DIS).The measurement of the quality between
the cover image f and stego-image g of sizes M 9 N is done using PSNR (Peak
Signal to Noise Ratio) value and the PSNR is defined as:
 
PSNR ¼ 10  log 2552 =MSE ð42:7Þ

where,

N 1 X
X N 1
MSE ¼ ðf ðx; yÞgðx; yÞÞ2 =ðM  NÞ ð42:8Þ
x¼0 y¼0

f (x, y) and g (x, y) means the pixel intensity value at position (x, y) in the cover-
image and the corresponding stego-image respectively. The PSNR is expressed in
dB. The larger PSNR indicates the higher the image quality i.e., there is only little
difference between the cover-image and the stego-image. On the other hand, a
smaller PSNR means there is huge distortion between the cover-image and the
stego image.
All the simulation has been done using the MATLAB 7 program on Windows
XP platform. Three different sets of 8-bit grayscale TIFF images of size
1024 91024 and 256 9 256 are used as the cover-images and secret image
respectively to form the stego-images. Three different sets of cover images are
considered to evaluate our results. Each set consists of five cover images. A single
secret image was used to embed into all the three sets. Figure 42.4(1)–(5) shows
the first set of original cover (carrier) images, Fig. 42.5(1)–(5) shows second set of
cover images. Figure 42.6(1)–(5) shows third set of cover images of the proposed
DIS method based on (4, 5) threshold scheme. Figure 42.7 shows the Original
Secret Image and Fig. 42.8 shows three extracted secret images from Set 1, Set 2
and Set 3 of the cover images.
432 R. Koikara et al.

Fig. 42.4 Cover images set 1, (1)–(5) five cover images of proposed DIS method

Fig. 42.5 Cover images set 2, (1)–(5) five cover images of proposed DIS method
42 A Novel Distributed Image Steganography Method … 433

Fig. 42.6 Cover images set 3, (1)–(5) five cover images of proposed DIS method

Fig. 42.7 Original secret

images

Table 42.1 exhibit the PSNR comparison of Stego Images with their corre-
sponding Cover Images for (4, 5) threshold scheme for all three sets of images.
From Table 42.1 it is observed that for threshold scheme (4, 5) PSNR is greater
than 40 dB for all the cases, so the quality of the stego image is quite acceptable
and the deterioration in quality due to embedding of secret image cannot be
distinguished by naked eye. Best result is achieved in case of Set 2 of Images, as
434 R. Koikara et al.

Fig. 42.8 Secret image extracted from cover images set 1, 2 and 3

Table 42.1 PSNR comparison of cover images and stego images for (4, 5) threshold scheme for
three different sets of cover images and their corresponding stego images
Cover PSNR (DB) between cover image and stego image
Image Threshold scheme (4, 5)
Set
Cover Cover Cover Cover Cover Original secret
image-1 image-2 image-3 image-4 image-5 image and
and stego and stego and stego and stego and stego extracted
image-1 image-2 image-3 image-4 image-5 secret image
Set 1 +48.92 +44.86 +50.43 +47.79 +48.50 +40.37
Set 2 +46.64 +48.53 +48.96 +46.01 +47.41 +45.08
Set 3 +47.59 +48.36 +47.27 +47.99 +47.21 +42.97

the PSNR is greater than +45 dB for the extracted secret image. If we pass the
extracted secret image through a median filter then there will be further
improvement of the extracted secret image quality.

42.5 Conclusion

Our proposed novel Distributed Image Steganographic method based on Block-

DCT which uses (k, n) Threshold Scheme improves the security and the quality of
the Stego Images. According to the simulation results the Stego Images of our
proposed algorithm are very difficult to distinguish from the Cover Images. We
have achieved a quite satisfactory quality of the extracted Secret Image for any set
of chosen Stego Images for (4, 5) threshold scheme. Distributing the Secret Image
among n–number of Cover Images keeps the Secret Image away from stealing;
destroying by any unintended users and Block-DCT adds to the security of the
secret image as the secret image embedding is done in frequency domain. Hence
the proposed method may be more robust against brute force attack.
42 A Novel Distributed Image Steganography Method … 435

References

1. Bai, L., Biswas, S., Blasch, P.E.: An estimation approach to extract multimedia information in
distributed steganographic images. In: Proceedings of the 10th International Conference on
Information Fusion, Quebec, Canada, 9–12 July 2007
2. Jhonson, F.N., Jajodia, S.: Exploring steganography: seeing the unseen. In: Proceedings of the
IEEE paper of Feb 1998
3. Cheddad, A., Condell, J., Curran, K., Kevitt, M.P.: Digital image steganography: survey and
analysis of current methods. J. Sign. Proces. 90, 727–752 (2010)
4. Li, B., He, J., Huang, J., Shi, Y.Q.: A survey on image steganography and steganalysis. J. Inf.
Hiding Multimedia Sign. Proces. 2(2), 142–172 (2011)
5. Provos, N., Honeyman, P.: Hide and seek: an introduction to steganography. IEEE Secur. Priv.
1(3), 32–44 (2003)
6. Thien, C.C., Lin, J.C.: Secret image sharing. J. Comput. Graph. 26(5), 765–770 (2002)
7. Shamir, A.: How to share a secret. Commun. ACM. 22(11), 612–613, (1979)
8. Das, R., Tuithung, T.: A review on ‘‘A novel technique for image steganography based on
block-DCT and Huffman encoding’’. In: Proceedings of the 4th International Conference on
Computer Graphics and Image Processing, ICGIP-2012, 6–7 Oct, Singapore, SPIE 2012
Chapter 43
An Improved History-Based Test
Prioritization Technique Technique Using
Code Coverage

Avinash Gupta, Nayneesh Mishra, Aprna Tripathi, Manu Vardhan

and Dharmender Singh Kushwaha

Abstract Prioritization of test cases provides a way to run test cases with the
highest priority earliest. Numerous empirical studies have shown that prioritization
can improve a test suite’s rate of fault detection. Software testers prioritize test
cases, to reduce the cost of regression testing. History Based Approach is one of
the methods to prioritize the test cases. This approach takes into account the
history of each of the test cases of the test suite such as fault detection, number of
executions and other such factors to prioritize the test cases in the coming sessions.
This paper extends the above approach to the modified lines. The modified lines
are being prioritized first and subsequently followed by the concerned test cases.
The proposed approach has been able to detect fault faster than the previous
approach with less effort in comparison to the previous approach.

Keywords Prioritization  History based  Fault detection  Test suit

A. Gupta (&)  N. Mishra  A. Tripathi  D.S. Kushwaha

MNNIT, Allahabad, India
e-mail: [email protected]
N. Mishra
e-mail: [email protected]
A. Tripathi
e-mail: [email protected]
D.S. Kushwaha
e-mail: [email protected]
M. Vardhan
National Institute of Technology, Raipur, India
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 437

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_43
438 A. Gupta et al.

43.1 Introduction

Since test development is costly, software testers often save the test suites they
develop, so that they can reuse those test suites later as software evolves. Such test
suite reuse, in the form of regression testing, is pervasive in the software industry
and, together with other regression testing activities, has been estimated to account
for as much as one-half of the cost of software maintenance. Running all of the test
cases in a test suite, however, can require a large amount of effort. Hence,
researchers have considered various techniques for reducing the cost of regression
testing. Three different techniques have, therefore, been proposed for test suite
reduction as—prioritization, selection and minimization of test suite.
A test suite minimization technique lowers the cost by reducing a test suite to a
minimal subset that maintains equivalent coverage of original set with respect to
particular test adequacy criterion [14]. Test suite minimization techniques [5],
however, can have some drawbacks. Although one clan of researchers think that,
in certain cases there is little or no loss in the ability of a minimized test suite to
reveal faults [2] in comparison to its un-minimized original, the other clan thinks
otherwise [18]. The fault detection ability of test suites can be severely compro-
mised by minimization.
Test case prioritization is the process of scheduling test cases in an order to
meet some performance goal [16]. Prioritization gives priority to test cases based
on criteria. The criteria may be code coverage etc. However, one of the limitations
in this process is that the fault detection efficiency of the test suite may be com-
promised. The test suite may contain test cases on higher priority which may not
be able to detect the errors [8]. Hence, several techniques have been proposed for
prioritizing the existing test cases to accelerate the rate of fault detection in
regression testing. Some of these approaches are Coverage-based Prioritization
[16], Interaction Testing, Distribution-based Approach [10], Requirement-based
Approach, and the Probabilistic Approach [9].
All these approaches apart from probabilistic approach referred above consider
prioritization as an unordered, independent and one-time model. They do not take
into account the performance of test cases in the previous regression test sessions,
such as the number of times a test case revealed faults [19]. History Based
Approach (HBA) has been applied to increase the fault detection ability of the test
suite. Kim and Porter [9] considered the problem of prioritization of test cases as a
probabilistic approach and defined the history-based test case prioritization. Alireza
et al. [8] proposed an extension of history-based prioritization proposed in [9],
and modifies the equation given by Kim and Porter [9], to have dynamic coeffi-
cients. The priority is calculated using the mathematical equation by computing
the coefficients of the equation from the historical performance data.
In this paper, we propose a new approach which is an extension of the history
based approach in [9]. Unlike in [9], where the prioritization equation has been
applied on each test case, we apply the approach on each modified line of the code.
The application of the prioritization at the code level makes the selected test suite
43 An Improved History-Based Test Prioritization Technique … 439

more effective in terms of fault detection effectiveness in comparison to that

obtained by applying the approach in [9].
The rest of the paper is organized as follows. Section 43.2 reviews related
literature. In Sect. 43.3, we present the proposed approach and implementation.
Section 43.4 describes performance analysis and comparison results. We conclude
the paper and discuss future work in Sect. 43.5.

43.2 Related Research Work

Regression Testing is the process of validating modified software to provide

confidence that the changed parts of the software behave as intended and that the
un-changed parts of the software have not been adversely affected by the modi-
fication [7].
In [15], Rothermel showed that because of time and resource constraints, it is
never feasible to re-run the entire test-suite for regression testing, which may be
very large in case of software of large size. Hence there is a need for prioritization,
selection, minimization of test suite. Horgan and London applied linear pro-
gramming to the test case minimization problem in their implementation of a data-
flow based testing tool, ATAC [6]. Chen and Lau [2] applied GE and GRE heu-
ristics. The GE and GRE heuristics can be thought of as variations of the greedy
algorithm that is known to be an effective heuristic for the set cover problem [13].
Offutt et al. [12] also treated the test suite minimization problem as the dual of the
minimal hitting set problem, i.e., the set cover problem [13]. Marre and Bertolino
formulated test suite minimization as a problem of finding a spanning set over a
graph [11]. Tallam and Gupta developed the greedy approach further by intro-
ducing the delayed greedy approach, which is based on the Formal Concept
Analysis of the relation between test cases and testing requirements [18]. Harder
et al. approached test suite minimisation using operational abstraction [4]. Harder
et al. use the widely studied Daikon dynamic invariant detector [3] to obtain
operational abstractions.
Test case prioritization is the process of scheduling test cases in an order to
meet some performance goal [16, 17]. Agrawal et al. [1] in his work considered
version specific test case prioritization [1] instead of general test case prioritization
[1] so that the final test suite may contain test cases relevant as per the modifi-
cations. However, his approach required too many iterations to arrive at the final
test suite and hence not very efficient. However, as discussed in [9], most of the
prioritizing algorithms are unordered, independent and one-time model. In [14], it
was stated that feedback mechanism makes the fault detection effectiveness of the
test suite better.
From the discussion above, it can therefore be easily deciphered that those
approaches which consider history to prioritize the test cases would certainly be
better in terms of fault detection effectiveness than those that did not. HBA
approach takes into account the history of test cases while prioritizing them in the
440 A. Gupta et al.

present session to increase the fault detection effectiveness of the test suite. The
technique by Alireza et al. directly computes the priority of each test case using
the historical information of the test case, such as the number of executions, the
number of times it exposed a fault and other relevant data. In this approach, the
prioritization has nothing to do with the source code. The test cases are prioritized
simply on the basis of their history.

43.3 Proposed Approach

In our proposed approach, we have extended the previous approach [8] by pri-
oritizing the modified lines. The history is kept for each of the modified lines
which act as feedback for the next session. In our proposed approach, the test cases
are selected for each modified line such that the test case is having the maximum
coverage among all the test cases which contain the modified line. By the phrase
‘containing a modified line’ we mean that the test case executes the modified line
in its line of execution.
Our proposed approach has been implemented in a ‘C’ program and the history
is being stored in text format in text files. The history contains all the test cases and
parameters such as number of executions of test case, number of times fault
detected by test case, number of times each line has been delayed execution. The
test cases contain the number of all the lines traversed along the line of execution
of the test case. The parameters have been stored in the form of arrays, where each
index represents a line in the code. The proposed approach in this paper includes
the steps shown in Fig. 43.1.

Step 1: Extract History from Database

In this phase all the parameter values of the previous session are extracted from the
database by the program into arrays for use in the present session. The following
parameters are extracted from the database:
eck[] is an array to store no. of times a line has executed till the session k. fck[] is
an array to store no. of times a line has been detected as faulty till the session k.
hk[] is the value is set to 0 if the line is executed in the last session else keeps on
increasing from 0 by one in each session till it does not execute. mod_locode[] is
an index is set to 1 if the corresponding line has been modified else to 0.
del_locode[] is an index is set to 1 if the corresponding line has been deleted else
0. PRk-1[] is the value at an index indicates the priority value of the corresponding
line in last session. PRk[] is the value at an index indicates the priority value of the
corresponding line in present session.
Step 2: Input Modified Lines The modified lines are taken as input from the user
through a well defined interface of the program. Any new test cases are also
entered through this interface.
Step 3: Find max coverage and ax coverage test case If the modified line say, mi
is present in a test case Ti, then max coverage of line mi = max (no. of lines in
43 An Improved History-Based Test Prioritization Technique … 441

Fig. 43.1 Flow chart of

proposed algorithm

Ti/total number of lines) * 100 for all Ti in which mi is found. Ti is the max
coverage test case for mi, if Ti has max code coverage.
Step 4: Calculate Priority Value for Modified Value For each modified line, the
priority value is calculated using Eq. (43.1) [11].

PRk ¼ ða:hk þ b:PRk1 Þ=k ð43:1Þ

 h
a ¼ 1  ððfck þ 1Þ=ðeck þ 1ÞÞ2 ð43:2Þ
k

b ¼ ððfck þ 1Þ=ðeck þ 1ÞÞx ð43:3Þ

In Eq. (43.1), 0 B a, b \ 1, k C 1.
In Eq. (43.3), x = 1 if the test case has revealed some fault in the previous
session and x = 2 if the test case has not revealed any fault in the previous session.
In Eq. (43.1), hk is the test cases execution history. eck denotes the total number of
executions done by a test case till the session k. fck denotes the total number of
faults detected by a test case till the session k. In Eq. (43.1), PR0 is defined for
each test case as the percentage of code coverage of the test case. The presence of
PR0 will be helpful in refining the ordering of the test cases in the first session.
442 A. Gupta et al.

Step 1: Prioritize Modified Lines

Modified lines are prioritized by their corresponding PRk, in descending order. If
the modified lines m1, m2 and m3 have PRk values as—PRk [m1] = 10.56 PRk
[m2] = 54.56 PRk [m3] = 9.64. Hence ordering would be—m2, m1, m3.
Step 2: Prioritize Test cases in order of modified lines
Max coverage test case for, say m1 = T2 m2 = T1 m3 = T3
Hence ordering of test cases in order of the respective modified line m2, m1 and m3
would be—T1, T2, T3.
Step 3: Output prioritized the test cases
The final output for session k is T1, T2, T3. After prioritizing, the test cases are
executed. Let us assume that only 40 % of all the test cases prioritized are able to
get executed. Out of all the test cases executed, there are certain test cases which
detect fault, and after debugging a fault is detected. Then, the parameters are
updated in the following manner:
• For each executed line i, in the present session k, increment the value of
parameter eck by 1 and set the value of hk to 0 for the line i.
• For the rest of the lines which did not execute in the present session k,
increment the value of hk by 1.
• For each faulty line i, detected in session k, increment the value of parameter
fck by 1 for the line i.
The database is updated with all these modifications.

43.4 Implementation

The proposed approach has been implemented using ‘C’ program. The database to
keep the history and all the test cases has been kept in two text files: ‘textcases.txt’
and ‘Parameters.txt’. The file ‘testcases.txt’ contains the test cases in the form of
traces of each of the test case. This means that ‘testcases.txt’ contains the lines
which will be covered by each test cases once they execute. There is another file
called ‘Parameters.txt’ which keeps the history of all the parameter values which
will be used to calculate the priority value PRk for each modified line. The C
program, itself has namely two sections:
• An interface:
This is the interface provided to insert all the test cases. This interface is also
used to insert the line numbers of modified, deleted and added lines. It is also
used to insert the line number of faulty line of the previous session.
• Prioritizing section:
This section calculates the priority of each modified line and outputs them in
order of priority. This section also outputs all the test cases in order of priority.
43 An Improved History-Based Test Prioritization Technique … 443

Table 43.1 Changes in the sample program

Line no. Original line Modified line
8 a[0 a\0
26 (1/(b + 2)) (1/(b-1))
30 (1/(x + 4)) (1/(x-4))
35 (1/(x + 5)) (1/(x-5))
40 (1/(x + 5)) (1/(x-5))

Fig. 43.2 Control flow graph

for program in Tables 43.2

Case Study
The proposed approach is demonstrated with an example here. We are considering
a small ‘C’ program as that have 43 lines and that calculates the value of math-
ematical equations. The equation consists of variables x and y whose value
depends on the value of a, b, c, d and e.
The program is modified at line numbers 8, 26, 30, 35 and 40. The changes in
each of the modified lines are shown in Table 43.1. These modifications in the
program will introduce divide by zero error in the program.
The Control Flow Graph (CFG) for the sample program and the modified
sample program is same since no any change in branch condition and none of the
line is added that is given in Fig. 43.2. The CFG has been developed using
444 A. Gupta et al.

SourceCode Visualizer Tool [2] which is a plug-in of Eclipse. Based on branch

coverage, we have the following test cases:
T1-8 9 10 11 20 38 39 40 41 42 43; T2-8 12 13 14 15 17 18 19 20 21 22 33 34
35 36 37 42 43; T3-8 12 13 14 15 16 19 20 21 22 23 24 25 26 27 42 43; T4-8 12 13
14 15 17 18 19 20 21 22 23 24 25 26 27 42 43; T5-8 9 10 11 38 39 40 41 42 43;
T6-8 12 13 14 15 16 19 38 39 40 41 42 43; T7-8 12 13 14 17 18 19 20 21 22 23 28
29 30 31 32 37 42;
These test cases are kept in the ‘testcases.txt’ file.
Now, for session k = 1:
Step 1: Extract history from database The test cases are extracted from the
‘testcases.txt’ file into the program which is developed to implement the proposed
algorithm. Before the first session, for each modified line, the value of parameters,
namely hk, eck, fck, prk and prk 1, is set to 0. For each modified line the value of
parameter x is set to 2.
Step 2: Input line numbers of modified lines As is mentioned in Fig. 43.3, the
line numbers of modified lines are 8, 26, 30, 35 and 40. The line numbers of
modified lines are entered into the program via the input interface of the program.
Step 3: Find max coverage and max coverage test case Line 26 is found in test
cases: T3, T4. Since, Code coverage of a test case T = (no. of lines in the test case
/total number of lines in the program) * 100
So, Code coverage of test case T3 = (17 /50) * 100 = 34 %
Code coverage of test case T4 = (18 /50) * 100 = 36 %
Hence, max code coverage for line 26 = 36 %
Since, Test case T4 is having the max code coverage value of 36 among all the
test cases containing the line 26. So, max code coverage test case for line 26 = T4.
As we calculated the value of max code coverage and found out the max code
coverage test case for line number 26, we can similarly compute the value of max
code coverage and find out the max code coverage test case for line number 8, 30,
35 and 40. The results are:
Max code coverage test case (MCCTC) for line 8 = T2, and Max Code Cov-
erage (MCC) = (18 /50) * 100 = 36 %
MCCTC for line 30, 35 and 40 are T7, T2 and T6 respectively. The MCC for
line 30, 35 and 40 are 36 %, 36 % and 26 % respectively.
Step 4: Calculate priority values for modified lines For session k = 1, the status
of all the parameters is as shown in the Table 43.2. Calculating
Now, substituting the values of hk, eck, fck, x and PRk-1 in Eqs. (43.1), (43.2)
and (43.3) and from Table 43.2, to calculate the corresponding value of PRk for
lines 8, 26, 30, 35 and 40. We get: PRk [8] = 36, PRk [26] = 36, PRk [30] = 36,
PRk [35] = 36, PRk [40] = 26.
Step 5: Prioritize the modified lines in order of priority value Based on the
priority values calculated in step 4, modified lines in order of priority as per PRk
values are: 8, 26, 30, 35, 40.
43 An Improved History-Based Test Prioritization Technique … 445

Table 43.2 Status of parameters before session k = 1

Line no. hk eck fck x PRk PRk-1
8 0 0 0 2 0 36
26 0 0 0 2 0 36
30 0 0 0 2 0 36
35 0 0 0 2 0 36
40 0 0 0 2 0 26

Table 43.3 Status of parameters before session k = 2

Line No. hk eck fck x PRk PRk-1
8 0 1 1 1 0 36
26 0 1 1 1 0 36
30 0 1 1 1 0 36
35 0 1 1 1 0 36
40 1 0 0 2 0 26

Step 6: Prioritize the test cases in order of modified lines Modified lines in
order of priority as per PRk value are: 8, 26, 30, 35, 40. Max code coverage test
case for Line 8 is T2, Line 26 is T4, Line 30 is T7, Line 35 is T2, Line 40 is T6. So
ordering the test cases in the same order as their respective modified lines are, we
get: T2, T4, T7, T2, T6. Removing the repeated test case T2 from the fourth place,
we get T2, T4, T7, T6.
Step 7: Output prioritized the test cases The final output for session k = 1 is T2,
T4, T7, T6. After the end of session 1, 40 % of all the test cases are executed i.e.,
test cases T2, T4, T7 are executed in the order of priority as given by the final
output of session 1. After the execution of the test cases it is found that all the test
cases fail and subsequently debugging is carried out. As a result of debugging,
fault is found at line number 8, 26, 30, and 35.
Session 2: (For session k = 2) Lines to be prioritized in session k = 2 are 26, 30,
35, 40 and 8. The status of parameters shown in Table 43.3 will be used in session
2. All the seven steps of the proposed approach are followed for k = 2.
The final output for session k = 2 is: T2, T4, T7, T6. We then used 40 % of the
total test cases to execute. The test cases T2, T4, T7 are executed in the given
order. T6 is left out this time. No new faults found as all the faults have already
been revealed in session k = 1.
Session 3: (For session k = 3) The status of parameters shown in Table 43.4 will
be used in session 3.
The final output for session k = 3 is: T6, T2, T4, T7. T6, T2, T4, T7 was
executed in order. T6 fails and reveals the error at line number 40. Thus all the
errors were revealed in 3 sessions. Hence, we can conclude that the proposed
approach has a better rate of fault detection than the Alireza approach.
446 A. Gupta et al.

Table 43.4 Status of parameters before session k = 3

Line no. hk eck fck x PRk PRk-1
8 0 2 1 2 0 18
26 0 2 1 2 0 18
30 0 2 1 2 0 18
35 0 2 1 2 0 18
40 2 0 0 2 0 13

Table 43.5 Proposed Approach Results

Program Line no. Session Faulty line Faulty line No. of No. of
modified No. detected by detected sessions sessions
Proposed by Alireza in in Alireza
approach approach Proposed approach
approach
Branch 26, 30, S1,S2 26, 30, 35, 26, 30, 35, 2 2
coverage 35, 40, 8 40, 8 40
Sample
program
Bank 6, 17, 22, S1, S2, 17, 22, 24, 17, 22, 24, 3 3
account 24, 27 S3 27, 6 27, 6
Library 199, 172, S1, S2, 199, 172, 199, 172, 2 3
management 223, 143, S3 223, 143, 223, 143,
126 126 126
Kruskal 97, 93, S1, S2 97, 106, 97, 106, 3 3
algorithm 106, 47, 110, 47, 93 110, 47, 93
110
Payroll 70, 118, S1, S2, 124, 118, 124, 118, 2 3
management 124, 207, S3 231, 207, 231, 70,
system 231 70 207
Heap sort 11, 12, S1, S2, 12, 19, 30, 12, 19, 30, 2 3
19, 27, 30 S3 27, 11 27, 11
Airline 27,43, S1, S2, 27,43, 27,43, 2 3
reservation 64,97,126 S3 64,97,126 64,97,126
system
Linked list 26, 78, S1, S2, 280, 147, 280, 147, 2 3
280, 54, S3 78, 54, 26 78, 54, 26
147
Tic-tac toe 159, 164, S1, S2, 159, 164, 159, 164, 2 3
191, 195, S3 191, 195, 191, 195,
261 261 261
Student 24, 27, S1, S2, 24, 27, 66, 24, 27, 66, 3 3
records 66, 69, 80 S3 69, 80 69, 80
management
System
43 An Improved History-Based Test Prioritization Technique … 447

43.5 Performance Analysis

Proposed approach is illustrated by ten programs of Java, C and C++ based

platforms. After that, we compared our proposed approach with Alireza et al.
approach. For comparison, we applied our proposed approach as well as Alireza
approach on ten example programs and five faults were seeded in each of the
programs with multiple sessions of regression test. Results in Table 43.5 shows the
faults detected in each session by our proposed approach and Alireza approach.
This is followed by a comparison of the total number of sessions to find all the
faults in proposed approach and Alireza approach as shown in Table 43.5. After
analyzing the results, we find that the number of fault detected per session is more
than or equal to in our proposed approach than in comparison to Alireza approach.
At the same time the number of sessions to discover all faults in our proposed
approach is less than or equal to Alireza approach. Hence we can say that, our
approach is more efficient in terms of fault detection effectiveness.

43.6 Conclusion and Future Work

Proposed approach is helpful in early detection of fault. The proposed approach

brings out those test cases to the front which are relevant to the modifications
made. Also, the proposed approach is able to prioritize test cases in less number of
iterations. This is because only those test cases are being processed which are
having the modified line in their line of execution, instead of calculating the
priority value for each of the test case as is done in Alireza approach. In the
proposed approach we have used different factors in the history of each modified
line. In future work, we may consider other factors such as severity of fault
detected, which may help to refine the process of prioritization.

References

1. Aggrawal, K., Singh, Y., Kaur, A.: Code coverage based technique for prioritizing test cases
for regression testing. ACM SIGSOFT Softw. Eng. Notes 29(5), 1–4, (2004)
2. Chen, T.Y., Lau, M.F.: Dividing strategies for the optimization of a test suite. Inf. Proces.
Lett. 60(3), 135–141 (1996)
3. Ernst, M.D., Cockrell, J., Griswold, W.G., Notkin, D.: Dynamically discovering likely
program invariants to support program evolution. IEEE Trans. Softw. Eng. 27(2), 99–123
(2001)
4. Harder, M., Mellen, J., Ernst, M.D.: Improving test suites via operational abstraction. In:
Proceedings of the 25th IEEE International Conference on Software Engineering, pp. 60–71
(2003)
5. Harrold, M.J., Gupta, R., Soffa, M.L.: A methodology for controlling the size of a test suite.
ACM Trans. Softw. Eng. Methodol. (TOSEM) 2(3), 270–285 (1993)
448 A. Gupta et al.

6. Horgan, J.R,. London, S.: A data flow coverage testing tool for C. In: IEEE Proceedings of
the Second Symposium on Assessment of Quality Software Development Tools, pp. 2–10
(1992)
7. Hsu, H.-Y., Orso, A. Mints: a general framework and tool for supporting testsuite
minimization. In: Proceedings of IEEE 31st International Conference on Software
Engineering. ICSE, pp. 419–429 (2009)
8. Khalilian, M., Azgomi, A., Fazlalizadeh, Y.: An improved method for test case prioritization
by incorporating historical test case data. Sci. Comput. Program. 78(1), 93–116 (2012)
9. Kim, J.-M., Porter, A.: A history-based test prioritization technique for regression testing in
resource constrained environments. In: Proceedings of the 24th IEEE International
Conference on Software Engineering. ICSE, pp. 119–129 (2002)
10. Leon, D., Podgurski, A.: A comparison of coverage-based and distribution-based techniques
for filtering and prioritizing test cases. In: 14th IEEE International Symposium on Software
Reliability Engineering. ISSRE, pp. 442–453 (2003)
11. Marŕe, M., Bertolino, A.: Using spanning sets for coverage testing. IEEE Trans. Softw. Eng.
9(11), 974–984 (2003)
12. Offutt, J., Pan, J., Voas, J.M.: Procedures for reducing the size of coverage based test sets. In:
Proceedings of the 12th International Conference Testing Computer Software. Citeseer,
pp. 111–123 (1995)
13. Papadimitriou, H., Steiglitz, K.: Combinatorial optimization: algorithms and complexity.
Courier Dover, New York (1998)
14. Rothermel, G., Harrold, M.J., Ostrin, J., Hong, C.: An empirical study of the effects of
minimization on the fault detection capabilities of test suites. In: Proceedings of the IEEE
International Conference on Software Maintenance, pp. 34–43 (1998)
15. Rothermel, G., Harrold, M.J., von Ronne, J., Hong, C.: Empirical studies of test-suite
reduction. Softw. Test. Verification Reliab. 12(4), 219–249 (2002)
16. Rothermel, G., Untch, R.H., Chu, C., Harrold, M.J.: Prioritizing test cases for regression
testing. IEEE Trans. Softw. Eng. 27(10), 929–948 (2001)
17. Tallam, S. Gupta, N.: A concept analysis inspired greedy algorithm for test suite
minimization. ACM SIGSOFT Softw. Eng. Notes. 31, 35–42 (2005)
18. Wong, W.E., Horgan, J.R., Mathur, A.P., Pasquini, A.: Test set size minimization and fault
detection effectiveness: a case study in a space application. J. Syst. Softw. 48(2), 79–89
(1999)
19. Yoo, S., Harman, M.: Regression testing minimization, selection and prioritization: a survey.
Softw. Test. Verification Reliab. 22(2), 67–120 (2012)
Chapter 44
Local Pricewatch Information Solicitation
and Sharing Model Using Mobile
Crowdsourcing

Hazleen Aris

Abstract Shopping, in particular groceries shopping, is a necessity and has

become part and parcel of our life, be it daily, weekly or monthly. Tied to this
activity is the concern over the ever increasing price of goods. Ideally, one would
want to be able to compare the price of an item between one shop and another
before getting the best deal. However, this is impracticable if not impossible due to
a number of factors, especially the time constraint. Nevertheless, advances in the
internet technology, combined with the widespread use of the internet-ready
devices that are often packaged with mobile internet plans have enabled rapid and
timely information sharing, hence, creating an opportunity that can overcome the
impossibility. People who go to different shops everyday can share information on
the prices of items at the shops that they visit, which will be useful for those who
are planning to buy similar items. What is missing is the mechanism to enable such
information sharing. This paper therefore explains about such mechanism that we
call the local pricewatch information solicitation and sharing (LoPrice) model that
uses voluntary crowdsourcing technique to solicit the information from the public.
Prior to that, justification on the need for the model is presented. Judging from the
success of its counterparts in other areas of social networking, it is expected that
the model can quickly gain interest from the public, hence helping them to save on
their expenditures.

Keywords Mobile crowdsourcing Price comparison
Crowdsourcing model



Open crowdsourcing Voluntary crowdsourcing

H. Aris (&)
College of Information Technology, Universiti Tenaga Nasional,
43000 Kajang, Selangor, Malaysia
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 449

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_44
450 H. Aris

44.1 Introduction

Were we ever be in a situation where the price of good that we had just bought
from a shop is found cheaper at another nearby shop? We would have bought the
item at the latter shop should we knew it, right? But how could we possibly know
about it unless somebody else shares the information with us? And how can we
know the information before it is too late? In this modern society, groceries
shopping is certainly inevitable and has become the main source of obtaining food
and other supplies. It is no longer an option, but a necessity. Gone were the days
where raw food can be obtained just by plucking them from the backyard plants or
from the livestock reared around the house.
When one thinks about shopping, one cannot help but to also think about the
expenses incurred by it. The ever increasing price of goods, that is becoming more
and more significant lately, has therefore become the main concern of many
people. Latest report on household expenditure survey by the Department of
Statistics Malaysia [1], which is done once every 5 years, showed that the average
monthly household expenditure had increased by 12.1 % from 2004/05 to 2009/10.
An independent survey [2] also showed that increment on prices of items even
reached 43 % within a six-month period, which was certainly burdensome as the
prices hike was far beyond the yearly income increment. Making life affordable
and less burdensome is so important that it has now become the concern of our
(Malaysian) government. A number of initiatives are seen being implemented by
the government to help easing the citizens with their expenditures such as the cash
aid scheme [3].
In our opinion, solution to this problem can be approached from at least three
directions. First, is by raising the living standard of low-income household, which
happens to be one of the national key result areas (NKRAs) of our government [4].
Second, is by keeping the goods prices low and third, is by being more selective in
buying the goods. While the first approach is a long term approach, the second
approach may require strict and close monitoring from the authorised bodies.
Therefore, it is the third approach that seems most amenable to all of us and hence,
the approach that we are interested in.
As we are aware, and probably as part of the marketing strategies, prices for the
same goods differ between one shop and another. This difference can be signifi-
cant, amounting to as much as RM10 per item, if, for example, the other shop is
having a sale. Most of the time, we have to depend on luck in order to get the best
price as the price changes from time to time and the pattern of change is unknown
to us. In an ideal situation, we would want to be able to compare the price between
one shop and another for each item that we want to buy. Unfortunately, it is time
consuming to move from one shop to another just to compare prices and then to
return to the shop that offered the lowest price to buy it. Let alone to do it within a
reasonable timeframe. It will also incur additional cost, as we need transportation
to move about. Not to mention the difficulty in getting a parking space and et
cetera, which render this practice infeasible. To overcome these problems, the
44 Local Pricewatch Information Solicitation and Sharing Model … 451

local pricewatch information solicitation and sharing (LoPrice) model is therefore

proposed. The aim is to help the users in finding the needed items at the most
reasonable, if not lowest price and at nearby stores. The difficulty in comparing
prices of items to be bought within an acceptable period of time is therefore the
problem statement to be addressed by the research.

44.2 Background

Recent advances in the internet and mobile technology have shed some light
towards overcoming the stated problem above. The widespread use of internet-
ready mobile devices combined with the availability of more and more affordable
mobile data plans and packages has enabled the users to share information on just
about anything at anytime. One type of information that they can share, and should
be sharing is on the prices of items. People who go to different shops everyday can
share information on the prices of items at the shops that they visit with others
using their mobile devices. This information will be useful to those who are
planning to buy similar items. In this way, the task of visiting each different shop
can be distributed amongst them and price comparison can be done by just tapping
and scrolling the mobile devices. This idea of distributing the potential solutions to
the public in order to obtain the best one is what Surowiecki [5] termed as ‘the
wisdom of crowds’ and has become amongst the seminal work in crowdsourcing.
The term ‘crowdsourcing’ was later formally coined by Howe [6] to mean the act
of outsourcing tasks to an undefined, large group of people or community (the
crowd) through an open call [7]. Crowdsourcing process generally comprises the
following steps [7].
1. Crowdsourcer proposes tasks and make it known to the public, with clear
requirements and reward if any.
2. The public submit potential solutions.
3. The public investigate, or evaluate the solutions and choose the best one.
4. Crowdsourcer determines the best solution and reward the winner.
5. Crowdsourcer gets and owns the best solution.
Two prerequisites of crowdsourcing are (1) the open call format and (2) the
large network of potential labourers [6]. Figure 44.1 shows the steps in a
crowdsourcing process in a slightly different arrangement, but maintaining the
same composition.
Motivations to participate in a crowdsourcing exercise vary. More often than
not, the strongest motivation is monetary reward, as discovered by Brabham [8],
although a more recent study showed otherwise [8]. As a result, crowdsourcing has
been successfully applied in business or other profit making sectors such as
Threadless and iStockphoto. The potential of crowdsourcing has also been
explored in auction [8].
452 H. Aris

Fig. 44.1 The eight steps in a crowdsourcing process

However, the presence of crowdsourcing and widespread availability of mobile

devices per se are not sufficient to solve the stated problem. A model is needed for
the crowdsourcing to be successfully applied in information sharing of items prices
and for the benefits to be optimally reaped, which prompts the need for this
research. Furthermore, Brabham [9] argued that the benefits of crowdsourcing
should also be enjoyed by the non-profit making sectors. Though examples on the
use of crowdsourcing for non-profit purposes are already seen [10–12], rooms for
research in this area are still plenty.
With regard to price comparison application, a number of applications were
found as a result of our exhaustive search, noticeably in the US [13–18] and France
[19, 20]. In the context of Malaysia, only three such applications were found as
shown in Table 44.1. However, they were all web-based applications, which
means that the information sharing was not done using dedicated mobile appli-
cations. Furthermore, the information on these websites was also outdated. In one
particular instance, we found that the price of an item was last updated in 2011!

44.3 The LoPrice Model

Taking into consideration (1) the need to have a means to properly share the
information on prices of items, (2) the widespread use of mobile devices, (3) the
research opportunity on the use of crowdsourcing for non-profit making sectors
44 Local Pricewatch Information Solicitation and Sharing Model … 453

Table 44.1 Price comparison applications in Malaysia

Name Website Source of Coverage (items) Coverage (stores)
information
Yellavia http://www. Users 21 categories of Not specified
yellavia. items, groceries is
com/ one of them
1pengguna http://www. Administrator Groceries Not specified
1pengguna.
com/
PriceChecker.my http://www. Inactive – the site is not working MYDIN,
pricechecker. Carrefour, Tesco,
my/ Giant, Lazada

The need to have a means

to share information on
prices of items

The scant use of

Widespread use of mobile The LoPrice crowdsourcing in non-profit
devices model making sectors

The missing of price

comparison application in
Malaysia

Fig. 44.2 Motivations for the research

that is still wide and (4) the missing of such applications in Malaysia, the LoPrice
model is proposed with the four considerations above being the motivations
towards the construction of the model as shown in Fig. 44.2.
The conceptual view of the LoPrice model is shown in Fig. 44.3. As can be
seen from the figure, information comes from the public users through their mobile
devices. Information for each item includes its price, location of the store where it
can be bought from, its category, and the validity period of the stated price, if
known. Application on the server side captures the information and makes it
available to other users whenever requested. The existing crowdsourcing process
are modified to suit the non-profit making purpose and subsequently applied in
454 H. Aris

Users share
Information on item price,
items are category, Wireless
shared with location, and infrastructure
users upon validity period of
request prices of items

Mobile devices with data plan

HTTP
Investigation areas :
• Mechanisms to ensure participation and
‘
‘
• Geotagging the information sent based on
location
• Algorithm for timely update of new and existing
information

TCP/ IP

Application server stores the

information provided and
purges the information when
expires

Fig. 44.3 Conceptual view of the LoPrice model and the areas of investigation

obtaining the information on prices of items from the public. The LoPrice model
will be developed in line with the characteristics of social perspective based on the
taxonomy of network communities proposed by Albors et al. [21]. To realise the
LoPrice model, a price information sharing application will be developed, which
can be used as a point of reference for users to find the location of the needed items
at the best price.

44.3.1 Investigation Areas

As explained above, in the context of Malaysia, applications that can be used to

compare prices of items were once developed and in use. However, their use could
not be sustained over time. Information was not updated, hence, defeating their
44 Local Pricewatch Information Solicitation and Sharing Model … 455

intended purpose. The preliminary part of the construction of the LoPrice model is
therefore a study that investigates the factors that influence users participation and
their ‘stickiness’ to such application. Findings from this investigation are taken
into consideration in constructing the LoPrice model. Secondly, to enable quick
updates on items information, users are expected to provide as minimum infor-
mation as possible. In other words, the LoPrice model should automate as much
information as possible and one particular information that can potentially be
automatically obtained is the location of the item, assuming that a user provides
the information while he or she is still in the shop. For this reason, mechanism to
geographically tag (geotag) the information sent based on location sent is inves-
tigated. Finally, an algorithm for timely update of new and existing information is
also developed. The algorithm, among others, ensures that obsolete information is
being purged from the application server. These areas of investigation are also
shown in Fig. 44.3.

44.4 Issues and Challenges

Two main issues regarding crowdsourcing, which are the misuse of cheap labour
and the reliability of information provided [22] are noted. With regard to cheap
labour, it was argued that the crowds get rewarded and they are not forced to do
the tasks. With regard to quality, Surowiecki [5] discovered that groups, albeit
amateurs are often smarter than the smartest people in them, under the right
circumstances. Also, a study reported on the use of crowdsourcing in transcribing
showed 74 % accuracy, which is viable compared to 88.5 % with professional
transcription [23]. Not to mention that measures to overcome the issues are already
in the pipeline [22]. Nevertheless, at this point, we did not anticipate these issues to
become hindrances to this research because no direct profit in monetary form will
be generated. The benefit of using the LoPrice model will return to its users, hence
they are not the labourers in any way. Since they are benefitting from the use of the
model, we can account on them to provide accurate information.

44.5 Conclusion and Further Work

This paper provides justification on the need for the LoPrice model and briefly
explains about the model. Judging from the success of its counterparts in other
areas of social networking, it is expected that the application developed from the
LoPrice model can quickly gain interest from the public. The expected benefit
gained by the users is still in the form of monetary reward, albeit indirectly,
through the reduced amount of expenditures. Future work includes the develop-
ment of the prototype system to be used to evaluate the LoPrice model. At least
456 H. Aris

two types of evaluation will be performed; experimental and empirical. Experi-

mental evaluation will assess the efficiency of the LoPrice model such as the
timeliness of the information shared and the empirical results obtained will be
quantitatively analysed using appropriate statistical analyses to determine its
effectiveness. From the analysis results, the research contribution and the scope of
applicability of the LoPrice model will be determined. Additionally, further work
will also be made to look at the scalability of the LoPrice model, that is, the
potential of extending its coverage beyond local community.

References

1. Department of Statistics Malaysia. http://www.statistics.gov.my/portal/index.php?option=

com_content&view=article&id=767&Itemid=111&lang=en#2
2. Kenaikan harga barang keperluan isirumah. http://jamal7478.blogspot.com/2011/08/
kenaikan-kos-harga-barang-keperluan.html
3. BR1M (Bantuan Rakyat 1Malaysia) Cash Aid Scheme Info. http://www.br1m.info
4. Enam Bidang Keberhasilan Utama Negara. http://pmr.penerangan.gov.my/index.php/
component/content/article/466-pengenalan-nkra/
4808-pointers-6-bidang-keberhasilan-utama-negara-nkra.html
5. Surowiecki, J.: The Wisdom of Crowds: Why the Many Are Smarter Than the Few and How
Collective Wisdom Shapes Business, Economies, Societies and Nations. Doubleday, New
York (2004)
6. Howe, J.: Crowdsourcing: a definition. http://crowdsourcing.typepad.com/cs/2006/06/
crowdsourcing_a.html
7. Zhang, L., Zhang, H.: Research of crowdsourcing model based on case study. In: Proceedings
of the 8th International Conference on Service Systems and Service Management, pp. 1–5
(2011)
8. Liu, Y., Alexandrova, T., Nakajima, T., Lehdonvirta, V.: Mobile image search via local
crowd: a user study. In: Proceedings of the 17th IEEE International Conference on Embedded
and Real-Time Computing Systems and Applications, pp. 109–112, IEEE (2011)
9. Brabham, D.C.: Crowdsourcing as a model for problem solving an introduction and cases, in
convergence. Int. J. Res. New Media Technol. 14(1), 75–90 (Sage) (2008)
10. Armstrong, A.W., Harskamp, C.T., Cheeney, S., Wu J., Schupp, C.W.: Power of
crowdsourcing: novel methods of data collection in psoriasis and psoriatic arthritis. J. Am.
Acad. Dermatol. 67(6), 1273–1281 (Elsevier) (2012)
11. Rovere, A., Raymo, M.E., O’Leary, M.J., Hearty, P.J.: Crowdsourcing in the quaternary sea
level community: insights from the Pliocene. Quarternary Sci. Rev. 56, 164–166 (Elsevier)
(2012)
12. Heipke, C.: Crowdsourcing geospatial data. ISPRS J. Photogrammetry Remote Sens. 65(6),
550–557 (Elsevier) (2010)
13. HuMuch. http://www.humuch.com
14. CatCrunch. https://play.google.com/store/apps/details?id=com.saverr
15. ShopSavvy. http://shopsavvy.mobi
16. Smoopa Price Checker. http://techcrunch.com/2012/04/19/smoopa, Smoopa
17. Price Comparison. http://www.appszoom.com/android_applications/shopping/price-comparison_
bbexw.html
18. Grocery King Shopping List. https://play.google.com/store/apps/details?id=com.groceryking
&hl=en
19. Grocery Price Watch. https://itunes.apple.com/us/app/grocery-price-watch
44 Local Pricewatch Information Solicitation and Sharing Model … 457

20. Leclerc. http://shopperculture.integer.com/2011/08/index.html

21. Albors, J., Ramos, J.C., Hervas, J.L.: New learning network paradigms: communities of
objectives, crowdsourcing, wikis and open source. Int. J. Inf. Manage. 28, 194–202 (2008)
(Elsevier)
22. Hodson, H.: Crowdsourcing grows up as online workers unite. NewScientist 2903, 22–23
(2013)
23. Aron, J.: Crowdsourcing serves up the subtitles to your life. NewScientist 2874, 21 (2012)
24. Brabham, D.C.: Moving the crowd at threadless: motivations for participation in a
crowdsourcing application. Inf. Commun. Soc. 13(8), 1122–1145 (Taylor & Francis Online)
(2010)
25. The crowdsourcing process in eight steps. http://en.wikipedia.org/wiki/File:Crowdsourcing_
process2.jpg
26. Satzger, B., Psaier, H., Schall, D., Dustdar, S.: Auction-based crowdsourcing supporting skill
management. Inf. Syst. 38(4), 547–560 (2012)
Chapter 45
Enhancement of Nurse Scheduling Steps
Using Particle Swarm Optimization

Norhayati Mohd Rasip, A.S.H. Basari, Nuzulha Khilwani Ibrahim

and Burairah Hussin

Abstract Allocating of working schedule, especially for shift approach is hard to

ensure its fairness among them. In the case of nurse scheduling, to set up the time
table for each available nurse is time consuming and complicated, which consider
many factors including rules, regulation and human factor. Moreover, most nurses
are women, which have personnel constraints and maternity leave factors. The
undesirable schedule can affect the nurse productivity, social life and the absen-
teeism can significantly as well affect the patient’s life. This paper aimed to
enhance the scheduling process by utilizing the particle swarm optimization in
order to generate an intelligent nurse schedule. The result shows that the multiple
initial schedules can be generated and can be selected with the lowest cost of
constraint violation.

45.1 Introduction

Staff scheduling is the process of constructing work timetables encoding for staff
in order to satisfy the demand of services. The ability to develop a good staff
schedule is a crucial process if the services demand round-the-clock and com-
plicated balancing act between an organization’s need and the legal contractual

N. Mohd Rasip (&)  A.S.H. Basari  N.K. Ibrahim  B. Hussin

Faculty of Information and Communication Technology, Universiti Teknikal Malaysia
Melaka, 76100 Durian Tunggal, Melaka, Malaysia
e-mail: [email protected]
A.S.H. Basari
e-mail: [email protected]
N.K. Ibrahim
e-mail: [email protected]
B. Hussin
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 459

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_45
460 N. Mohd Rasip et al.

obligations to its staff. Thus, all such factors should be considered when designing
a staff shift schedule, leading to an extremely complex problem for which finding a
better solution and within a reasonable time frame, can be difficult. Additional
working shift can have a significant impact on the health and lives of staff while
can in turn affect a person’s productivity at work. Differences between industries,
their goals and restrictions, often mean that specific models and algorithms must
be developed for each of them [1].
This paper focuses on the hospital’s staff and nurse is the proposed staff to be
scheduled. Usually hospital wards must be staffed 24 h for 7 days a week by a
limited number of staff, which very difficult to satisfy everybody. However, as well
as the usual legal and cost constraints are met, high patient care services are the
goal. Consequently, the risk is encountered when the nurse work on an undesirable
schedule that may affect the patients’ lives and also nurse performance since the
job is beyond their expertise. Nevertheless, with shortages in qualified nurses
reported regularly, good schedules are important to provide satisfactory patient
care and potentially improve nurse retention.
Most hospitals face the nurse scheduling problem in its primary form, which is
to allocate shifts and days to nurses so as to satisfy a given set of hard and soft
constraints. The unique of combination of constraints depending on the pertinent
policies that usually differ from one hospital to another. Once the number of
resources defining the success of each schedule is determined, the problem may
become complex, which is more constraints occur with feasible schedule to satisfy
and more conflicting objectives are identified.
The class of nurse scheduling problem is well known by rapid growth in the
number of potential solutions and the modest growth of the resources to be
scheduled is known as NP-hard problem. Thus, the importance of finding the
potential solution has led to research initiative in industries including airlines, call
centers and hospital with a range of solution methods by utilizing such as math-
ematical programming, heuristic technique and artificial intelligence [2].

45.1.1 Related Work

The advance of information technology is rising to help researchers in developing

a number of approaches to solve nurse rostering problem in a reasonable amount of
time. As pinpointed in the varies depth literature reviews in [1, 3–5]. These four
papers have provided details about problem formulation on scheduling. Basically,
nurse scheduling formulated by using mathematical programming, artificial
intelligence and meta-heuristics methods. In [6] is presented an integer mathe-
matical programming (IP) to solve the nurse scheduling problem with two phase
strategy.
The strategy is to effectively deal with the conflicting requirement based on
nurses contracts and preferences. In [7] formulated mathematical programming with
the multi objective to construct the schedule of medical resident based on seniority.
45 Enhancement of Nurse Scheduling Steps Using Particle Swarm … 461

They aims to maximize the quality of shift allocation by considering the shift
coverage requirements, seniority-based workload rules and resident work prefer-
ences. The multi objective model with sequential method and weighted method
based on the seniority level has promised the optimal solution and less computa-
tional time.
One of the aims of nurse scheduling is to meet the balancing between schedule
feasibility and quality. The quality of the timetable is normally measured as a
maximize the fulfillment of soft constraints. In [7–9], dealing soft constraints with
hierarchical oriented. These approaches are generally achieved the optimal solu-
tion based on individual preferences, but hard to solve real world problems that
have high numbers of constraints and not relatively static over the time. Then the
calculation of penalties is considered as pessimistic because the past penalty value
is assumed that a roster for a week before is empty. Glass and Night proposed in
[10] a continuous rostering environment for handling constraints and preferences
arising. This solution approach is calculated the roster across a number of whole
weeks, including the transition week from the current rostering period to the next.
The result shows that the penalties are more optimistic.
Recently many researchers turn focus to population-based stochastic search
methods such as genetic algorithms, ant colony optimization and particle swarm
optimization (PSO), which able to reach near optimal solution within acceptable
computation time. The PSO based algorithms have been successful designed and
applied in many research area such as in university timetabling [11], job shop
scheduling [12], manufacturing scheduling [13]. For nurse scheduling problem,
PSO is utilized to solve unfair assignment among the nurses using a new evalu-
ation function [14]. This new evaluation function allows a better discrimination
between candidate solutions and is in accordance with the idea of the original
function, which searched for assignment fairness. Wu et al. applied in [15] the
basic PSO to the nurse scheduling problem. Improving work stretch pattern during
initial solution using PSO had proved that using PSO produce optimal solution in a
very efficient manner. The advantages of PSO are that this method not needed the
calculation of derivatives due to the result of good solution is retained by all
particles and that particle in the swarm share information between them. The
objective function can be used for stochastic objective function and can escape
from local minima. In order to be programmed, PSO has few parameters to adjust,
more effectively memory capability than the GA which is every particle to
remember their own previous best value as well as the neighborhood and more
efficient in maintaining the diversity of the swarm [17].
The main contribution of this paper is to simplify the actual process of allo-
cating shift to nurse by utilizing the advantages of the particle swarm algorithm
with the same objective. Although there are plenty of PSO algorithms applied to
the scheduling problems in the literature, to the best of our knowledge, applied to
the nurse scheduling problem there is still significant room for improvements in
this area [16]. This was one of the main motivations in order to design and apply a
PSO based algorithm so as to solve effectively and efficiently the nurse scheduling
problem with the real world problem case at Malaysia public educational hospital.
462 N. Mohd Rasip et al.

This paper is organized as follows: Sect. 45.2 discusses the nurse rostering
problem while the particle swarm optimization approaches to the Nurse Rostering
Problem is described in Sect. 45.3. Section 45.4 discusses the experimental results
in conclusion and possible research directions are provided in Sect. 45.5.

45.2 Nurse Scheduling Problem

Nurse scheduling is the process of allocating, subject to constraints, the available

resources into slot in a pattern within scheduling period. The process need a lot of
time to produce a good roster. In this work, we consider a real nurse scheduling
problem (NRP) one of the public hospital in Malaysia [17].
There have the four shifts a day that are morning shift (M), evening shift (E),
night (N) or day off (O). For any holiday is included as days off. There have a two
level of nurses either senior or junior. For each shift must be at least one senior.
The challenges of assigning process are considered subject to hard and soft
constraints. All the constraint is presented in Sect. 45.2.1. The details step show as
below:

Step 1: Nurse request: Each nurse fill in the log book manually and submit their
preference at least 1 day before the head nurse starts to construct the
schedule. regarding the nurse’s requests, the head nurse will consider
some of the criteria before accept or reject the request.
Step 2: After the nurse has the number of available nurses, the head nurse will
construct the initial of schedule. In this stage the head nurse will allocate
the night shift which is an unpopular shift until the number of nurses for
night shift is satisfied. Then the head nurse continues to allocating the
morning and evening shift (popular shift). This process is repeated until
all the hard constraint and soft constraint is accepted.
Step 3: The initial schedule will send to the department of nursing and after it is
approved the schedule will print and distribute to all nurses and get their
feedback.

The actual scheduling is ‘‘person based’’ process which is based on head nurse’s
knowledge and relies on her capabilities to create the best schedule for her unit.
When assigning shift, the head nurse has to verify if all assignments fulfill to the
quota requirement for each shift. The head nurse always changes the spreadsheet
to find out whether it is respected or not, and memorize it while moving to the next
step. This ‘‘searching process’’ is very time-consuming, not efficient and certainly
not optimized. Figure 45.1 illustrates the mapping of the process.
The standard rules are when all the hard constraint is satisfied, then the feasible
nurse scheduling is constructed, that can actually be used by the ward it was made
for. Nevertheless, the number of soft constraint is satisfied the main factor that
affects the quality of a nurse scheduling. The final goal, indeed, is to create feasible
45 Enhancement of Nurse Scheduling Steps Using Particle Swarm … 463

Fig. 45.1 Scheduling process map

schedule while maximizing its quality, means to create nurse scheduling which
satisfied all the hard constraints and fulfilled as many as possible number of soft
constraints.

45.2.1 Hard Constraints

In public hospital the rules are known as constraints. The hard constraint is the
rules which must be fulfilled; otherwise the schedule is considered to be infeasible
and unacceptable. The constraints are set based on current practices of the hospital
studied, which taking into account for regulations, contractual requirements,
operational requirements for all nurses. Most of these rules agree with published
hospital policies.
464 N. Mohd Rasip et al.

The hard constraints detail as shown below:

(H1) Each shift has to meet the minimum coverage demanded of nurses
(H2) Each nurse cannot work more than one shift a day
(H3) Each nurse must have at least 2 days off during the scheduling period
(H4) Each shift must have at least one senior nurse
(H5) Each nurse must not be scheduled for an isolated working day
(H6) This constraint presents the minimum number of consecutive days that any
nurse should work
(H7) Each nurse must be working at least 10 days and not more than 12 days
(H8) Each nurse must work in consecutive working days is 4 days
(H9) Each nurse that has four consecutive night shifts must follow 2 days off.

45.2.2 Soft Constraints

The soft constraints taken into consideration which is not a compulsory to be

fulfilled, but satisfying as much as possible represent the degree of a high quality
result. There are other soft constraints that are normally created by an agreement
between the head nurse and the individual nurses.
The hospital rules for soft constraints are as shown below:
(S1) The total number of working days and day off should be uniformly dis-
tributed to all nurses.
(S2) The nurse should have at least 1 day off in a weekend during the scheduling
period.
(S3) The nurse should have 1 day off after four consecutive days shift (morning
and evening).

45.3 Particle Swarm Optimization for NSP

Particle swarm optimization (PSO) is a population-based optimization algorithm

that was originally proposed by Keneddy and Eberhart (1995). PSO is an evolu-
tionary optimization technique that is a population-based search algorithm. Each
particle flies through the search space with velocities adjusted dynamically
according to their previous behavior to look for the optimal solution. In PSO
particles continuously adjust their position on their own and the experience of their
neighboring particles. When compared with other evolutionary optimization
techniques, PSO perform better in terms of success rate and solution quality. The
movement of a particle is influenced by its inertia, its personal best position, and
the global best position. Its velocity, its personal best value, that is the best
45 Enhancement of Nurse Scheduling Steps Using Particle Swarm … 465

objective value the particle ever experienced, and its personal best position, that is
the position at which the personal best value has been found.
The particle’s data could be anything. In the flocking birds, the data would be
the X, Y, Z coordinates of each bird. The individual coordinates of each bird
would try to move closer to the coordinates of the bird which is closer to the food’s
coordinates (global best value).
To start the PSO approach, the algorithm starts with random initialization of a
particle’s position and velocity. In this section, this paper describes all the com-
ponents of the algorithm. The algorithm works to solve a problem with discrete
domains, thus in order to apply PSO to this problem, a careful about the meaning
of both the position and the velocity are very crucial.

45.3.1 Solution Representation

Step 1 Initialization: The assignment of shift to each nurse is randomly generated

and for each nurse representing as one particle (solution). The position of particle
represents the type of shift. The search space, represent as a matrix X of size N*D,
where N represents the number of particles and D represents the total number for
schedule period. Each particle Xti has a position, xtid is a discrete number repre-
sented the type of assignments (M, E, N, O) for each pair (nurse, day). If the
xtid ¼ 1 represent the morning shift (M),number 2 represents the evening shift (E),
number 3 represents the night shift (N) and the number 4 represent the day off (O).
If there are 5 nurses, N and the schedule period is 14 days, the particle position
vector, then has the value of 5 9 14 = 60 elements. Elements 1–14 stand for the
nurse schedule of the first nurse, 15–28 for the second nurse, 29–42 for the third
nurse, and so on as shown in Table 45.1. First, each particle is initialized and the
range of the initial position is set between 1 and 4.
Step 2 Evaluate the fitness value: All the particle’s fitness value is obtained by
the fitness function. It checks all violations of hard and soft constraints in every
particle. The fitness value for each particle is obtained by using weighted sum
objective function MinWS using formula (45.1). Let c [ C be the set of soft
constraints then, wc is the weight associated with constraint c and nc the number of
violations of c. The penalty weight as assigning by head nurse based on the
importance rate as Table 45.2. The MinWS consists of two parts: Cvio associated
with the costs of coverage nurse to a given shift and Hvio with the costs of
assigning a nurse to a given shift. Thus the objective is to find a roster with the
lowest overall penalty, denoted as MinWS.
X
Cvio ¼ w c nc ð45:1Þ
0  c  jCj
 466

Table 45.1 A sample matrix of initial position for PSO

Particle in the swarm (1, 1) (1, 2) … (1, 13) (1, 14) (2, 15) (n, d) … (2, 28) … (N, N*D)
Particle 1 M M ... E E E O ... M ... E
: M M ... M E E M ... P ... E
Particle 5 E O ... M M M N ... O ... M
: E E ... M M M M ... E ... M
Particle i M M ... O M N M ... M ... M
: N N ... E E N E ... M ... O
Particle N O N ... E N O E ... E ... E
N. Mohd Rasip et al.
45 Enhancement of Nurse Scheduling Steps Using Particle Swarm … 467

Table 45.2 The constraint weighting

Soft constraints Penalty
Each shift has to meet the minimum coverage demanded of nurses 100
Assign morning shift following night shift 100
Assign each nurse at least 1 day off on the weekends 100
Assigns four consecutive morning shifts followed by 1 day off (O) or an evening (E) 10
shift after 2 days off that follow the night (N) shift pattern
Assign four consecutive evening shifts followed by 1 day off 10
Assigns either a day off (O) or an evening (E) shift after 2 days off that follow the 10
night (N) shift pattern, i.e., (NNNNOOE) or (NNNNOOO)

X
MinWS ¼ Cvio þ Hvio ð45:2Þ
0  c  jCj

Step 3 Updating particle’s position: After the particle position is initialized, each
particle is assigned a rate of position change or velocity randomly. Just in the way
the particles adjust its flight direction through referring to its own experience and
population’s experience. The particles update the velocity and position according
to the following two equations:

Vtþ1
i ¼ W  Vti þ c1  r1  ðPi best  xti Þ þ c2  r2  ðGbest  xti Þ ð45:3Þ

xtþ1
i ¼ xti þ Vtþ1
i ð45:4Þ

where Xti represents the position of the particles, Vti represents the velocity of the
particles, w represents the inertia weight, which is the inertia weight value setting
will have an effect on the velocity of particle position update and also indirectly on
the next position the particle will move to. If the w value setting is inappropriate, it
is impossible to explore unknown areas and the search for the personal best and
global best will also be affected. The c1 and c2 are the learning rates, which is a
positive constant parameter called acceleration coefficients (which control the
maximum step size of a particle) and how to approach the new position either for
close to the individual experience position or expecting close to the global
experience position. The r1 and r2 are two random numbers independently gen-
erated, which are between 0 and 1. Pi_best is a particle’s position at which the best
fitness so far has been achieved. Gbest is the population global (or local neigh-
borhood, in the neighborhood version of the algorithm) position at which the best
fitness so far has been achieved. The first part of Eq. (45.1) is the ‘‘inertia’’,
determined by a particle’s previous velocity; the second part is the ‘‘individual
cognition’’, represents a particle’s own experience, which guides a particle to
evolve into its ‘‘Pi_best’’; the third part is the ‘‘social cognition’’, denotes the
cooperation and information sharing among the particles, which guides a particle
468 N. Mohd Rasip et al.

Table 45.3 Result of experiment

Particle in the MinP MaxP Avg. hard Avg. soft Fitness value
swarm constraint constraint (MaxP-MinP)
Particle 1 100 400 180 240 420
Particle 2 100 700 260 370 630
Particle 3 10 600 170 157 327
Particle 4 10 310 180 98 278
Particle 5 110 510 130 303 433

to evolve into the population’s ‘‘Gbest’’. Just in this way, each particle adjusts its
flight (evolution) direction through referring to its own experience and the pop-
ulation’s experience.

45.4 Preliminary Results

In this study, the proposed PSO algorithm based technique has been developed in
order to make it suitable for solving nurse constrained optimization problems. A
procedure is imposed to check the feasibility of the candidate solutions in all
stages of the search process. The proposed technique has been implemented on
personal computer a Intel Core i5 CPU and 4 GB RAM. To complete the fol-
lowing experiments, the PSO-based intelligent mechanism for scheduling is pro-
grammed by using MATLAB. The computational flowchart of the proposed PSO
algorithm is depicted in Fig. 45.1.
For this stage the result shows the fitness value of each schedule generated (par-
ticles) with hard and soft constraints. The advantages are by using intelligent nature
inspired optimization; the head nurse can generate more than one schedule for the
initial schedule and can select the best schedule which has the best fitness value. The
guided randomized assignment of shift to available nurse can reduce time consuming
and be more flexible to use because the user can change the few parameters based
on their need. Table 45.3 shows the preliminary result of the experiment.

45.5 Conclusion

The result shows that the effectiveness and efficiency of constructing the nurse
schedule can be achieved by using particle swarm optimization. The advantages of
particle swarm optimization which are the setting of parameters is reasonably
simple and implementation time to find a proper schedule is fast can be able to use
for all the Malaysian Hospital. The finding suggests that the proposed solution
using particle swarm optimization is capable to enhance the nurse scheduling by
utilizing the intelligent process in order to obtain the intelligent nurse schedule.
45 Enhancement of Nurse Scheduling Steps Using Particle Swarm … 469

For further research could be extended with other scheduling scenario and
variant particle swarm optimization.

Acknowledgments This paper is part of Master by research in Information and Communication

Technology and funded under the Ministry of Education research grant RAGS/2012/FTMK/
TK06/1 B00018.

References

1. Ernst, A., Jiang, H., Krishnamoorthy, M., Sier, D.: Staff scheduling and rostering: a review of
applications, methods and models. Eur. J. Oper. Res. 153(1), 3–27 (2004)
2. Hussin, B., Basari, A.S.H., Shibghatullah, A.S., Asmai, S.A., Othman, N.S.: Exam
timetabling using graph colouring approach. In: Proceedings of the 2011 IEEE Conference
on Open Systems, pp. 133–138, Sept. 2011
3. Brucker, P., Qu, R., Burke, E.: Personnel scheduling: models and complexity. Eur. J. Oper.
Res. 210(3), 467–473 (2011)
4. Burke, E.K., de Causmaecker, P., Berghe, G.V., van Landeghem, H.: The state of the art of
nurse rostering. J. Sched. 7(6), 441–499 (2004)
5. Cheang, B., Li, H., Lim, A., Rodrigues, B.: Nurse rostering problems—a bibliographic
survey. Eur. J. Oper. Res. 151(3), 447–460 (2003)
6. Valouxis, C., Gogos, C., Goulas, G., Alefragis, P., Housos, E.: A systematic two phase
approach for the nurse rostering problem. Eur. J. Oper. Res. 219(2), 425–433 (2012)
7. Topaloglu, S.: A shift scheduling model for employees with different seniority levels and an
application in healthcare. Eur. J. Oper. Res. 198(3), 943–957 (2009)
8. Ziarati, K., Akbari, R., Zeighami, V.: On the performance of bee algorithms for resource-
constrained project scheduling problem. Appl. Soft. Comput. 11(4), 3720–3733 (2011)
9. Brucker, P., Burke, E.K., Curtois, T., Qu, R., Berghe, G.V.: A shift sequence based approach
for nurse scheduling and a new benchmark dataset. J. Heuristics 16(4), 559–573 (2008)
10. Glass, C.A., Knight, R.A.: The nurse rostering problem: a critical appraisal of the problem
structure. Eur. J. Oper. Res. 202(2), 379–389 (2010)
11. Tassopoulos, I.X., Beligiannis, G.N.: A hybrid particle swarm optimization based algorithm
for high school timetabling problems. Appl. Soft. Comput. 12(11), 3472–3489 (2012)
12. Ping, Y., Minghai, J.: An improved PSO search method for the job shop scheduling problem.
In: Proceedings of the 2011 Chinese Control Decision Conference CCDC, vol. 2,
pp. 1619–1623 (2011)
13. Chou, F.-D.: Particle swarm optimization with cocktail decoding method for hybrid flow
shop scheduling problems with multiprocessor tasks. Int. J. Prod. Econ. 141(1), 137–145
(2013)
14. Altamirano, L., Riff, M.-C., Trilling, L.: A PSO algorithm to solve a real anaesthesiology
nurse scheduling problem. In: Proceedings of the 2010 International Conference of Soft
Computing and Pattern Recognition, pp. 139–144 (2010)
15. Wu, T., Yeh, J., Lee, Y.: A particle swarm optimization approach for nurse rostering
problem. In: Proceedings of the International Conference on Business and Information,
pp. 737–752 (2013)
16. Rasip, N.M., Ibrahim, N.K., Basari, A.S.H.: An investigation of intelligent search techniques
for nurse scheduling improvement in healthcare organization. In: Proceedings of the EHealth
Symposium 2013, pp. 1–7 (2013)
17. Abobaker, R., Ayob, M., Hadwan, M.: Greedy constructive heuristic and local search
algorithm for solving nurse rostering problems. In: Proceedings of the 2011 3rd Conference
on Data Mining and Optimization, pp. 28–29, June 2011
Chapter 46
Hardware Implementation
of MFCC-Based Feature Extraction
for Speaker Recognition

P. Ehkan, F.F. Zakaria, M.N.M. Warip, Z. Sauli and M. Elshaikh

Abstract The most important issues in the field of speech recognition and rep-
resentative of the speech is a feature extraction. Feature extraction based Mel
Frequency Cepstral Coefficient (MFCC) is one the most important features
required among various kinds of speech application. In this paper, FPGA-based for
speech features extraction MFCC algorithm is proposed. The complexities of
computational as well as the requirement of memory usage are characterized,
analyzed, and improved. Look-up table (LUT) scheme is used to deal with the
elementary function value in the MFCC algorithm and fixed-point arithmetic is
implemented to reduce the cost under accuracy study. The final feature extraction
design is implemented effectively into the FPGA-Xilinx Virtex2 XC2V6000
FF1157-4 chip.

Keywords Speaker recognition  Mel frequency cepstral coefficients  Field

programmable gate array

P. Ehkan (&)  F.F. Zakaria  M.N.M. Warip  M. Elshaikh

School of Computer and Communication Engineering, Universiti Malaysia Perlis,
Pauh Putra Campus, 02600 Arau, Perlis, Malaysia
e-mail: [email protected]
F.F. Zakaria
e-mail: [email protected]
M.N.M. Warip
e-mail: [email protected]
M. Elshaikh
e-mail: [email protected]
Z. Sauli
School of Microelectronic Engineering, Universiti Malaysia Perlis, Pauh Putra Campus,
02600 Arau, Perlis, Malaysia
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 471

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_46
472 P. Ehkan et al.

46.1 Introduction

Biometric systems are the automated method of verifying or recognizing the

identity of the person on the basis of some physiological characteristic, such as a
finger print, face pattern and human voice [1]. The human voice conveys infor-
mation about the language being spoken and the emotion, gender and identity of
the speaker. Speaker recognition is the computing task of validating a user’s
claimed identity using characteristics extracted from their voices [2, 3]. Voice of a
person has many prominent characteristics like pitch, tone which can be used to
distinguish a person from the other. It has a history dating back some four decades
[4] where the output of several analog filters was averaged over time for matching.
Speaker recognition uses the acoustic features of speech that have been found to
differ between individuals. These acoustic patterns reflect both anatomy (size and
shape of the throat and mouth) and learned behavioral patterns such as voice pitch
and speaking style. This incorporation of learned patterns into the voice templates
has earned speaker recognition its classification as a ‘‘behavioral biometric’’ [5].
Such systems extract features from speech, model them and use them to recognize
the human voice.
Speaker recognition system basically involves two main phases namely the
training stage and the testing stage [6]. These phases involve two main parts called
feature extraction and pattern classification. At the time of training, speech sample
is acquired in a controlled and supervised manner from the user. The speaker
recognition system has to process the speech signal in order to extract speaker
discriminatory information from it and this will form the speaker model, which is a
process of enrolment speaker data. At the time of testing a speech sample is
acquired from the user. The speaker recognition system has to extract the features
from this sample and compare it against the models been stored beforehand. This
is a pattern matching or classification task.
The improvement in FPGA technology as well as design tools has introduced a
new option for Digital Signal Processing (DSP) applications recently. FPGA is a
type of semiconductor device that contain programmable logic and interconnec-
tions which mostly used in logic or digital electronic circuits. The programmable
logic components or logic blocks as they are known may consist of anything from
logic gates, through to memory elements or blocks of memories, or almost any
element. FPGA supports thousands of gates and popular for prototyping integrated
circuit (IC) designs. Once a design is set, hardwired chips will be produced to
faster performance. FPGA chip is programmable and reprogrammable which is
considered as an advantage of it. In this way, it becomes a large logic circuit that
can be configured according to a design, but if changes are required it can be
reprogrammed with an update. Thus, if circuit board is manufactured and contains
an FPGA as part of the circuit, this is programmed during the manufacturing
process, but can be reprogrammed to reflect any changes. The user programma-
bility gives the user access to complex ICs without the high engineering costs
associated with ASICs.
46 Hardware Implementation of MFCC-Based Feature Extraction … 473

FPGA contains many identical logic cells that can be viewed as standard
components. Each design is implemented by specifying the simple logic function
for each cell and selectivity closing the switches in the interconnect matrix. The
array logic cells and interconnects form a basic building blocks for logic circuits.
Complex designs are created by combining these basic blocks to create the desired
circuit. The logic cell architecture varies between different device families. This
logic cells can be configured via high levels design tools or hardware description
languages (HDLs) to implement virtually any digital system. Few years ago these
chips required programming using HDLs or weak design tools, making imple-
mentation of complex DSP operations, such as FFT and DCT very tedious. The
newest generation of design tools offers libraries of common DSP functions,
enabling developers to implement something as complex as FFT [7]. FPGA have
been used in many areas to accelerate algorithms that can make use of massive
parallelism, and in every case allowing them to improve the flexibility, cost
reduction and time to market. A great promising application area the use of FPGA
is that it can exploit the pipeline ability and parallelism with the solution algo-
rithms in a much more thorough way that can be done with parallel computers
using general-purpose microprocessors or a single standard processor.
This paper presents results for the speaker recognition front-end processing
using MFCC approach on the platform consisting of a Xilinx Virtex-II XC2V6000
FPGA. The paper is organized the sections as the speaker recognition system,
front-end processing, implementations, results and conclusion.

46.2 Speaker Recognition System

A block diagram shown in Fig. 46.1 is the basic structure system designed to
implement speaker recognition. The input speech is sampled and converted into
digital format. The vectors from the speech consisting of the MFCCs are extracted.
The system then branches into two separate phases namely training and classifi-
cation. In the training phase, each registered speaker has to provide samples of
their speech so that the system can train a reference models for that speaker while
in the classification phase, the input speech is matched with the stored reference
models and then a recognition decision is made. The speaker associated with the
most likely, or higher scoring model is selected as the recognized speaker. This is
simply a maximum likelihood classifier.
The more general problem for speaker recognition is that out of a total popu-
lation of N speakers, find that speaker whose reference pattern is most similar to
the sample pattern of an unknown speaker. Since the sample pattern is compared to
each of the N reference patterns and also there is a finite probability of an incorrect
decision for each comparison, it is apparent that the overall probability of an
incorrect decision must be a monotonically increasing function of N [8].
474 P. Ehkan et al.

Speaker Database

Stored speaker models

Training
(Speaker Speaker #1…..
Speaker #n
Modelling)

Training Phase
Input Feature Extraction
Speeches (MFCCs)
Recognition Phase
Speaker #1

Speaker #2 Max
Pattern scored Result:
Classification . Speaker
. #ID
Compare
Speaker #n

Speaker Database

Fig. 46.1 Basic structure of speaker recognition system

46.3 Front-End Processing

This part is the most important issue in the field of speech processing. It is
converting the speech waveform to a set of features for further analysis. Speech
front-end processing consists of transforming the speech signal to a set of feature
vectors [9]. The aim of this process is to obtain a new representation which is more
compact, less redundant, and more suitable for statistical modelling [10]. Feature
extraction is the key to front-end process.
According to Stolcke et al. [11], the purpose of feature extraction is to convert
the speech waveform to some type of parametric representation at a considerably
lower information rate. A wide range of possibilities exist for parametrically
representing the speech signal for the speaker recognition task, such as LPC [12],
MFCC [13], PLP [14] and others. The MFCC and PLP are among the most popular
acoustic features used in speaker recognition. It usually depends on the task which
of the two methods leads to a better performance. Currently, researchers are
focusing on improving these two cepstral features [10] or appending new features
on them [15]. In fact, it is commonly believed that the spectrum smoothing done
by MFCC and PLP has some sort of speaker normalization effect.
The selection of feature determines the separability of the speaker, and it also
has large influence on the classification step, since the classifier must be turned to
the given feature space. Thus, the selection of the features should be carefully
considered in designing a system. Several analyses have been done for feature
extraction technique in order to observe the best technique for transforming the
speech signal. Davis and Mermelstein [16] reviewed the literature of a few feature
extraction methods and compared them in a syllable-oriented speaker dependent
speech recognition system. The experiments were made in a clean environment
46 Hardware Implementation of MFCC-Based Feature Extraction … 475

and the segmentation was done manually. They found out that features derived
using cepstrum analysis outperformed those that do not use it and that filter bank
methods outperformed LPC method (PLP methods were not included). This means
that the best performance was achieved by using MFCC.
A comparable result of the use of PLP and MFCC was reported by John and
Wendy [17] that MFCC-based feature extraction method seems to be performing
well in most studies. Besides, a theoretical comparison of MFCC and PLP analysis
was given by Milner [18]. The theoretical comparison continues with a practical
implementation. The spectral analysis is followed by channel normalization (both
RASTA and CMN) and extraction of dynamic features. The best results were
reported for MFCC with RASTA filtration. Similarly, Schmidt and Thomas [19]
has pointed out the state-of-the-art speaker recognition systems typically
employing the MFCC as representative acoustic feature and GMM as pattern
classification method has achieved very good performance which even better than
recognition by human. Chakroborty et al. [20] listed three reasons why MFCC
method has become so dominant for speaker identification system. First, MFCC is
less vulnerable to noise perturbation, it gives little session variability and is easier
to extract. Moreover, a calculation of MFCC is based on the human auditory
system aiming for artificial implementation of the ear physiology assuming that
the human ear can be a good speaker recognizer too. Furthermore, computation of
MFCC involves averaging the low frequency region of the energy spectrum
(approximately demarcated by the upper limit of 1 kHz) by closely spaced over-
lapping triangular filters while smaller number of less closely spaced filters with
similar shape are used to average the high frequency region. Thus, MFCC can
represent the low frequency region more accurately than the high frequency region
and hence, it can capture formants which lie in the low frequency range and which
characterize the vocal tract resonances [21, 22].
All these facts suggest that any speaker recognition system based on MFCC can
possibly be improved. Based on above studied, this project has been decided to
employ MFCC method as the feature extraction part in order to obtain most
advantageous feature vector to be adapted in the propose of hardware implemented
pattern classification model.

46.3.1 Mel Frequency Cepstral Coefficients

Mel Frequency Cepstral Coefficients (MFCC)s are well known features used to
describe speech signal. They are based on the known variation of the human ear’s
critical bandwidths with frequency; filters spaced linearly at low frequencies and
logarithmically at high frequencies have been used to capture the phonetically
important characteristics of speech. The processes to obtain the MFCC can be
summarized as in Fig. 46.2.
The speech input is typically recorded at a sampling rate above a frequency of
10 kHz. This sampling frequency was chosen to minimize the effects of aliasing in
476 P. Ehkan et al.

Continuous Frame
Speech Blocking
Frame

Windowing
Windowed Frames

FFT
Magnitude Spectrum

Mel-freq
warping
Mel Spectrum

Cepstrum Mel
Cepstrum

Fig. 46.2 MFCC block module

the analogue-to-digital (ADC) conversion. These sampled signals are able to

capture all frequencies up to 5 kHz which cover most energy of sounds that are
generated by humans. The main purpose of the MFCC processor is to mimic the
behavior of human ears. Besides, rather than the speech waveforms themselves,
MFCCs are shown to be less susceptible to the mentioned variations.
Frame Blocking. The continuous speech signal is blocked into frames of N
samples, with adjacent frames being separated by M (M \ N). The first frame
consists of the first N samples. The second frame begins M samples after the first
frame, and overlaps it by (N–M) samples as shown in Fig. 46.3. Similarly, the
third frame begins 2 M samples after the first frame (or M samples after the second
frame) and overlaps it by (N–2M) samples. This process continues until all the
speech is accounted for within one or more frames. Typical values for N and M are
N = 256 (which is equivalent to *30 ms windowing and facilitate the fast radix-2
FFT) and M = 100, respectively.
Windowing. This step is windowing each individual frame so as to minimize
the signal discontinuities at the beginning and end of each frame. The concept here
is to minimize the spectral distortion by using the window to taper the signal to
zero at the beginning and end of each frame. If the window is defined (where N is
the number of samples in each frame) then the result of windowing is the signal of

y1ðnÞ ¼ x1ðnÞwðnÞ; 0  n  N  1: ð46:1Þ

The Hamming window is typically used in a form of

 
2pn
wðnÞ ¼ 0:54  0:46 cos ; 0nN  1 ð46:2Þ
N1

where, N is the samples number of each window and n is the sample being evaluated.
46 Hardware Implementation of MFCC-Based Feature Extraction … 477

Frame1 Frame2 Frame3 Frame N

Window
function

Signal processing
Signal processing

Signal processing
Signal processing

Frame length

Frame overlap
x11 x21 x31 xN1
x12 x22 x32 xN2
x1 = . x2 = . x1 = . xN = .
. . . Feature vector .
x1d x2d x3d xNd

Fig. 46.3 Frame blocking short term analysis

Fast Fourier Transform (FFT). The FFT is a fast algorithm to implement the
DFT converts each frame of N samples from time domain into frequency domain.
It can be defined on the set of N samples {xn} as shown

X
N 1
Xk ¼ xn ej2pkn=N ; k ¼ 0; 1; 2; . . .; N  1: ð46:3Þ
n¼0

Generally, the values of Xk are complex numbers and in this case the frequency
magnitudes which are their absolute values are considered.
Mel-Frequency Wrapping. Psychophysical studies have shown that the human
perception of the frequency consists of sounds for speech signal does not follow
the linear scale. Therefore, for each tone with an actual frequency, f, measured in
Hertz, a subjective pitch is measured on the scale called the ‘Mel’ scale. The
frequency scale is then converted from Hertz to Mel. The Mel-frequency scale is
linear frequency spacing below 1 kHz and logarithm spacing above 1 kHz. As a
reference point, the pitch of a 1 kHz tone, 40 dB above the perceptual hearing
threshold, is defined as 1,000 mels. Therefore, it is possible to use the following
approximate formula to compute the mels for a given frequency f in Hz:

Melðf Þ ¼ 2595  log10 ð1 þ f =700Þ: ð46:4Þ

478 P. Ehkan et al.

Cepstrum. The final step is converting the log mel spectrum back to time
domain. The result is called the MFCC. The cepstral representation of the speech
spectrum provides a good representation of the local spectral properties of the
signal for the given frame analysis. Mel spectrum coefficients are real numbers.
Therefore, is possible to convert them to the time domain using the DCT.

X
K
Cn ¼ ðlog Sk Þcoz½nðk  1=2Þðp=KÞ; n ¼ 1; 2; . . .; K ð46:5Þ
k¼1

where, Sk is the mel scaled signal had after wrapping and Cn is the cepstral
coefficient.

46.4 Implementations

The front-end processing features extraction MFCC-based speaker recognition

system was implemented into Xilinx-based FPGA Virtex-II XC2V6000 FF1152-4
platform. This device is a mid-range FPGA, member of Virtex-II family has
76,032 logic cells, 2592 k bits block RAM, 144 18 bit multipliers, 824 user I/O
and a speed grade of 4. The specifications for the system implemented in the
hardware used the first 17 MFCCs and their respective delta values, population
size of 20 and 5 s of test utterances. The speech samples were taken from the
TIMIT database.

46.5 Results

Hardware Resources Requirement. The breakdowns of hardware utilization for

MFCC module are presented in Table 46.1.

Table 46.1 Hardware resources usage

Item Used Available % Utilization
Slices 23251 33792 68.8
Slice FFs 35104 67584 51.9
4 Input LUTs 39452 67584 58.4
Bonded IOBs 295 824 35.8
BRAM 14 144 9.7
MULT18 9 18 33 144 22.9
GCLKs 1 16 6.26
46 Hardware Implementation of MFCC-Based Feature Extraction … 479

Fig. 46.4 Datapath Block in MFCC Module vs. Clock Cycles

Breakdown 700
600

Clock Cycles
596
500
400 378
300
200 169
100
0 26 5 4
FFT Mel Logarithm DCT Mean Delta
Warping Removal
Blocks in MFCC Module

Simulation Results. The entire datapath taken by the MFCC module is 1,178
clock cycles. Figure 46.4 shows the breakdown of clock cycles used by each block
of MFCC module.
The FFT block is the most complexity sub-block in MFCC module and it
required the most clock cycles which is 596 followed by the DCT block which
occupied 378 clock cycles.

46.6 Conclusion

As the conclusion this paper aims to highlight the implementation of speech signal
MFCC features extraction algorithm on hardware-based FPGA. The hardware
solution has taken the advantage of parallelism and pipelining throughout the
entire signal processing compared to conventional approach which is mostly
software-based. The FPGA-based hardware solution can be an optimal and
alternative solution to the conventional DSPs for the speech signal processing
technology.

References

1. Kung, S.Y., Mak, M.W., Lin, S.H.: Biometric Authentication: a Machine Learning Approach,
1st edn. Prentice Hall, New Jersey, USA (2005)
2. Campbell, J.P.: Speaker recognition: a tutorial. Proc. IEEE. 85(9), 1437–1462 (1997)
3. Sadaoki, F.: Fifty years of progress in speech and speaker recognition. J. Acoust. Soc. Am
116(4), 2497–2498 (2004)
4. Atal, B.S.: Automatic recognition of speakers from their voices. Proc. IEEE 64, 460–475
(1976)
5. Furui, S.: An overview of speaker recognition technology, ESCA Workshop on Automatic
Speaker Recognition, Identification and Verification, 1–9 (1994)
6. Richard, D.P., Daryl, H.G.: An introduction to speech and speaker recognition. IEEE
Comput. Soc. Press 23(8), 26–33 (1990)
480 P. Ehkan et al.

7. ‘‘System Generator for DSP’’ Xilinx Inc. (2006). http://www.xilinx.com/ise/optional_prod/

system_generator.htm
8. Rosenberg, A.E., Soong, F.K.: Recent research in automatic speaker recognition. In: Sadaoki,
F. (ed.) Advances in Speech Signal Processing, 701–738 (1992)
9. Moretto, P.: Mapping of speech front-end signal processing to high performance vector
architectures. Technical report, International Computer Science Institute (1995)
10. Premakanthan, P., and Mikhad, W. B.: Speaker verification/recognition and the importance of
selective feature extraction: review. In: Proceedings of the 44th IEEE 2001 Midwest
Symposium on Circuits and Systems, 1(1), 57–61 (2001)
11. Stolcke, A., Shriberg, E., Ferrer, L., Kajarekar, S., Sonmez, K., and Tur, G.: Speech
recognition as feature extraction for speaker recognition. In: IEEE Workshop on Signal
Processing Applications for Public Security and Forensics 1–5 (2007)
12. Atal, B.S., Hanauer, L.S.: Speech analysis and synthesis by linear prediction of the speech
wave. J. Acoust. Soc. Am. 50, 637–655 (1971)
13. Davis, S. B., Mermelstein, P.: Comparison of parametric representations for monosyllabic
word recognition in continuously spoken sentences. IEEE Trans. Acoust Speech Sig. Process.
28(4), 357–366 (1980)
14. Hermansky, H.: Perceptual Linear Predictive Analysis of Speech. J. Acoust. Soc. Am. 87(4),
1738–1752 (1990)
15. Waleed, H.A.: Robust speaker modeling using perceptually motivated feature. Elsevier Sci.
Pattern Recogn. Lett. 28(11), 1333–1342 (2007)
16. Davis, S. B., Mermelstein, P.: Comparison of parametric representations for monosyllabic
word recognition in continuously spoken sentences. IEEE Trans. Acoust. Speech Sig.
Process. 28(4), 357–366 (1980)
17. John, H., Wendy, H.: Speech Synthesis and Recognition, 2nd edn. Taylor & Francis Inc,
Bristol, USA (2002)
18. Milner, B.: A Comparison of Front-End Configurations for Robust Speech Recognition.
Proceeding of ICASSP ’2002, 1(1), 797–800 (2002)
19. Schmidt, N.A., Thomas, H.C.: Speaker verification by human listeners: experiments
comparing human and machine performance using the NIST1998 speaker evaluation data.
J. Digit. Sig. Process. 10(1–3), 249–266 (2000)
20. Chakroborty, S., Roy, A., Saha, G.: Improved closed set text-independent speaker
identification by combining MFCC with evidence from flipped filter banks. Int. J. Sig.
Process. 4(2), 114–122 (2008)
21. Rabiner, L., Juang, B.H.: Fundamentals of Speech Recognition, 2nd. ed. Pearson Education,
USA (2003)
22. Ben, G., Nelson, M.: Speech and Audio Signal Processing, 2nd edn. Wiley, USA (2002)
Chapter 47
Parallel ASIP Based Design of Turbo
Decoder

F.F. Zakaria, P. Ehkan, M.N.M. Warip and M. Elshaikh

Abstract Application Specific Instruction-set Processor (ASIP) has a general-

purpose architecture that can be modified and used in a variety of applications.
However, this increases the power and memory utilization and affects the func-
tionality and efficiency of ASIP. This paper is defining the flexibility of ASIP for
Turbo decoding in term of its functionality and architecture for specific applica-
tions such as DVB-RCS, 3GPP. The proposed architecture has a dedicated SIMD
(Single Instruction Set Multiple Data), coupled with distributed memory based
ASIP. It has been concluded in this paper that ASIP facilitates parallelism at
different levels, thereby, increasing the efficiency, power consumption, and pro-
cessing time.

Keywords Application-specific instruction-set processor  Bahl-Cocke-Jelinek-


Raviv (BCJR) Turbo decoding recognition

47.1 Introduction

The technology development in field of telecommunication system nowadays had

changed it’s from analog systems to digital systems exclusively. One of the major
reasons is digital communications offer possibility of channel coding (line, and

F.F. Zakaria (&)  P. Ehkan  M.N.M. Warip  M. Elshaikh

School of Computer and Communication Engineering, Universiti Malaysia Perlis,
Pauh Putra Campus, 02600 Arau, Perlis, Malaysia
e-mail: [email protected]
P. Ehkan
e-mail: [email protected]
M.N.M. Warip
e-mail: [email protected]
M. Elshaikh
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 481

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_47
482 F.F. Zakaria et al.

error control, coding) to minimize effects of noise and interference. Since the
introduction of Turbo Codes in 1993 by Berrou and Glavieux [1], several standards
and protocols had been adapted to it as it serves astonishing performance. Over the
years it has seen the emergence of an increasing number of wireless protocols such
as UMTS, LTE and WiMAX. The growth of these protocols is to meet the
technologies needs of now-days hand-held wireless devices, which convergence
with many disparate features, including wireless communication, real-time mul-
timedia and interactive applications, into a single platform. Thus the flexibility of a
decoder becomes a dominant aspect for each wireless device. Today’s 3GPP
standard limits the throughput to 2 Mbit/s, while future systems are predicted to
require above 10 Mbit/s. Scalable and energy efficient architecture are required to
catch up with the challenge and to serves time to market constrains. Therefore,
insisting on a programmable solution, ASIPs are becoming the architecture of the
first choice.
Before the flexibility been taken into account in implementation of turbo
decoder, ASIC had been exclusively the choice of turbo decoder implementer.
Some of these implementations succeeded in archiving high throughput for spe-
cific standards with a fully dedicated architecture for example, in [2] a dedicated
high performance turbo decoding for 3GPP implemented on ASIC, in [3], a new
class of turbo codes more suitable for high throughput implementation is proposed.
However, such implementations do not consider the flexibility issues and imple-
mentation of ASICs becomes extremely expensive because their high cost and low
design productivity. Only high manufacturing volumes can allow manufacturing.
Application specific Instruction-set Processor (ASIPs) solve a part of these prob-
lems [4], a higher volume of unit is demanded because there are more applications
that fit on it.
This paper presents the architecture of ASIP models for turbo decoding to show
their flexibility and performance. The rest of the paper is organized as Sect. 47.2
presents the turbo decoding algorithm and the parallelism technique, Sect. 47.3
presents a decoder requirement regarding the different standards and channel
codes, Sect. 47.4 presents the architecture of ASIPs Turbo decoding, Sect. 47.5
results of the proposed architecture has been shown and lastly Sect. 47.6 sum-
marizes the result obtained to conclude the paper.

47.2 Turbo Decoding and Parallelism

Turbo code is an efficient error correction code, using a concatenation of two (or
more) simple constituent codes, which are iteratively decoded using soft decision.
This work implements Max Log MAP algorithm for the application.
47 Parallel ASIP Based Design of Turbo Decoder 483

47.2.1 Maximum Log Maximum-a-Posteriori Algorithm

The max log-MAP algorithm is an approximation of the MAP algorithm that

operates in the log-domain, allowing multiplications in the original MAP algo-
rithms to be implemented by additions. The max log-MAP algorithm, with a pair
of Viterbi algorithm, sweeps through the trellis in forward and reverse direction,
hence, the MAP and the logarithmic variants are called the forward-backward
algorithm. The branch metric of max log-MAP algorithm with the addition of a
priori log-likelihood ratio (LLR) is same as that branch metric that is used in
Viterbi algorithm.
If rk is the trellis state value at time k, then the likelihood values (Lk ) at time k,
is defined as

Lk ¼ ak1 ðrk1 Þ þ ck ðrk1 ; rk Þ þ bk ðrk Þ ð47:1Þ

where ak1 ðrk1 Þ is the alpha metric that calculates the probability of the current
state based on the input values before time k, ck ðrk1 ; rk Þ is the branch metric that
calculates the probability of the current state transition, bk ðrk Þ is the beta metric
that calculates the probability of the current state given the future state values after
time k. Both a and b are called the state metrics and c is the branch metric.
Furthermore the a and b computations are the forward and backward recursive,
respectively. They are given as

ak ðrk Þ ¼ maxðak1 ðrk1 Þ þ ck ðrk1 ; rk ÞÞ ð47:2Þ

and

bk ðrk Þ ¼ maxðbkþ1 ðrkþ1 Þ þ ck ðrk ; rkþ1 ÞÞ ð47:3Þ

The branch metric c is calculated from ysystematic ; yapriori and yredundancy as

ysystematic þ yapriori yredundacy

c¼ þ extrinsic ð47:4Þ
yintrinsic y

If s1 and s0 were the 1-branch and 0-branch of the trellis state transitions, the
soft output value, Log-likelihood Calculation (LLC) at time k, is defined as

LLCk ¼ maxðLk Þs1  maxðLk Þs0 : ð47:5Þ

It requires large computation and storage, forward and backward recursion

results in long decoding delay as the drawbacks.
484 F.F. Zakaria et al.

47.2.2 Parallel Processing Levels

In turbo decoding, parallelism is applied at metric level, soft-in soft-out (SISO)

decoder level and turbo-decoder level. The Bahl-Cocke-Jelinek-Raviv (BCJR)
metric level parallelism concerns the processing of all metrics involved in the
decoding of each received symbol. Trellis transition and BCJR Computation
parallelism is exploited by the metric level.
Parallelism of trellis transitions. For the complete frame, each transition pairs
has the same trellis computation technique. The trellis computation can be divided
into Branch-Metric Calculation (BMC) and Add-Compare-Select (ACS) stages.
Since the same operations are repeated in every pair so parallelism can be
extracted from trellis structure. This results in equaling to half the number of
transitions per trellis section and constitutes the upper bound of the trellis-tran-
sition parallelism degree. It facilitates low area overhead as only the ACS units
have to be duplicated, eliminates the use of extra memory as all the parallelized
operations are executed on the same trellis section, and in consequence on the
same data.
Parallelism of BJCR computations. The forward-backward scheme permits
parallel execution of backward recursion and A posteriori Probability (APP). The
butterfly scheme permits doubling this parallelism degree only by duplicating the
BCJR resource without addition of memory. It helps optimizing area efficiency.
The parallelism degree is limited by the decoding algorithm and the code structure.
Thus, achieving higher parallelism degree implies exploring higher processing
levels.
Parallelism of SISO decoder. SISO decoder parallelism consists of the use of
multiple SISO decoders, each executing the BCJR algorithm and processing a sub-
block of the same frame in one of the two interleaving orders.
Sub-block parallelism. The frames are divided into sub-blocks to be processed
on a BCJR-SISO decoder. Duplication of decoders can result in communication
conflicts and hence use of conflict management mechanism and finally leads to
long and variable communication time. The parallelism, though intended to reach
high throughput, has a reduced throughput since resolving the initialization issue
implies a computational overhead following an Amdahl’s law (due to acquisition
length or additional iteration).
Decoder component parallelism. Here, all the decoder’s components are
executed in parallel and the extrinsic information is exchanged as generated. The
shuffled decoding technique performs decoding and interleaving concurrently
thereby reducing iteration period but requires techniques to avoid memory con-
flicts. Shuffled decoding is more efficient than sub-block parallelism for high
throughput.
47 Parallel ASIP Based Design of Turbo Decoder 485

Table 47.1 Selection of standards and channel codes

Standard Codes Rates Throughput (Mbit/s)
LTE bTC 1/3 …100
UMTS bTC 1/3 …2
DVB-RCT dbTC 1/2,3/4 …31
IEEE802.16(WiMAX) dbTC 1/3…7/8 …20

47.3 Flexible Decoder Requirement

An analysis from [5] has concluded that most of communication standards are
using binary convolutional and binary and duo-binary turbo codes. Commonalities
between convolutional and turbo decoding can be find although there are different
in block size polynomials, and coding rates.
Table 47.1 shows summarize between different communication standards
which using turbo code scheme for channel encoder. The differences in parameters
such as constraint lengths, generator polynomials and interleaving patterns can be
seen. Thus to implement a flexible decoder, ASIP has to support these wide range
of coding parameters.
Through investigation of standards in Table 47.1, the following specifications
were derived which have to be provided by ASIP to fulfill the flexibilities
requirement for turbo decoding systems:
• supports of N = 4…16 states for bTC (binary turbo codes),
• supports of N = 8…16 states for dbTC (double-binary turbo codes),
• arbitrary feedback polynomials,
• arbitrary generator polynomials,
• fast reconfigurable trellis structure,
• High throughput and low latency.

47.4 Turbo Decoder Architecture

The top level view of the architecture as Fig. 47.1, are composes of operative,
control, communication interface and memories to store data. Inside the operative
part consists of two identical BJCR units, corresponding to forward and back ward
processing in the MAP algorithm. These units will processes a window of 64
symbols and produces recursion metrics and extrinsic information. The memories
units are divided by two, internal and external. Two recursion metric storage
(cross) implements under internal memories, it’s will be used by each forward and
backward unit. Another internal memory (config) contains up to 256 trellis
descriptions. The ASIP architecture can be configured for the correspondent
standard depending on the information store in the config. On the other side the
external memories consists of input_data and info_ext to store the incoming data
486 F.F. Zakaria et al.

Fig. 47.1 ASIP architecture of turbo decoder

Fig. 47.2 a BCJR computation unit. b Adder node

systematic and redundant, also future and past used to initialize state metric value
for beginning and the end of each window.
BCJR computation unit. It also known as recursion unit, in order to support
trellis parallelism (up to 32 adder nodes) this unit using Single Instruction Multiple
Data (SIMD) architecture. As seen in Fig. 47.2a, this could be decomposed into
maximum 8 states incorporated each other by the maximum of 4 decisions per
47 Parallel ASIP Based Design of Turbo Decoder 487

Fig. 47.3 a BCJR computation unit. b Adder node

decoding time. It can be view as a processing matrix. For a 16 state simple binary
codes, transition with ending states 0–7 are mapped on matrix nodes of row 0, if the
transition bit decision is 0, or matrix node of row 0, or matrix nodes of row 1, if
transition bit decision is 1, whereas states 8–15 are mapped on nodes of row 2 and 3.
An adder node (Fig. 47.2b) contains one adder optimized in [7] compare to two
adders in [6], multiplexers, one register for configuration (RT) and an output
register (RADD). This adder node function to support addition required in
recursion between state metric (RMC) and a branch metric (RG) and also the
addition required in information generation since it can accumulate the previous
result with the state metric of the other recursion coming from register bank RC.
To find the max between recursion a max node contains three max operators is
place between four nodes. With this structure it make possible to perform either 4-
input maxim or two two-input maximum. Resulted are stored either in first col-
umns or rows of RADD matrix or in RMC bank to archive recursion computation.
Control unit. ASIP design methodologies allow the turbo decoder’s control
unit to use the pipeline strategies. These six-stage pipelines consist of Fetch,
Decode, Operand Fetch, Branch Metric (added stage), Execute and Store. The
pipeline was kept short to preserve some flexibility for further extensions. Branch
metric stage was added to reduce the processing cycle in Execute in order to
increase the throughput buy increase the clock frequency of the ASIP.
Instruction set. Each instruction set is 16 bit wide, to perform basic turbo
decoding, the 30 instruction line shows in Fig. 47.3 are compulsory. It performs
488 F.F. Zakaria et al.

the basic MAP algorithm operation such as control, operative and IO. The con-
figuration and initialization or recursion metric are performs in the first six
instructions-line, then it follow by initialize the butterfly loop by ZOLB instruction
[7]. Compared to [6], the initialization of the butterfly are done by part, several
ZOL instruction are needed. Since the example is to decode duo-binary turbo
codes 2 max operator are required (max 2 m in [7] and two max m in [6]). The last
5 instructions are to compute the extrinsic information for 8 states and the program
branched to beginning of butterfly.
Concerning of the execution time the enhanced version [7] showed improve-
ment, 7*N/2 cycles (3.5 cycles roughly) are needed to process N symbols of sub-
block while the original version [6], 15*N/2 (7.5 cycles roughly) cycles are needed
to process N symbols of sub-block.

47.5 Results

The architecture was implemented with LISATek tool set, and the generated
VHDL model was synthesized with Synopsys Design Compiler in in ST 0.09 lm
(resp. 0.18 lm) ASIC technology under worst-case conditions. The synthesized
ASIP has a maximum clock frequency of 335 (resp. 180) MHz and occupies about
97 (resp. 93) KGates (equivalent). This means that a single processor, running the
two SISO of 6 turbo iterations over a DVB-RCS code, will potentially have a
throughput of 7.4 (resp. 4) Mbit/s.

47.6 Conclusion

An ASIP for Turbo decoding has been described in this article. As we can con-
clude that application specific flexibility is mandatory to meet the flexibility and
performance requirement of the emerging communication standards nowadays. It
can be archived by Application Specific Instruction set Processor with designed
pipeline topology, architecture and memory infrastructure.

References

1. Berrou, C., Glavieux, A., Thitimasjhima, P.: Near Shannon limit error-correcting coding and
decoding: turbo codes, ICC Geneva, (1993)
2. Prescher, G., Gemmeke, T., Noll, T.: A parameterizable low-power high-throughput turbo
decoder. In proceedings Of ICASSP vol. 05, pp. 25–28 Pennsylvania (2005)
3. Gnaëdig, D., Boutillon, E., Jezequel, M., Gaudet, V., Gulak, G.: ‘‘On multiple slice turbo
code’’ The 3rd International Symposium on Turbo Codes and Related Topics, pp.343–346
Brest, France (2003)
47 Parallel ASIP Based Design of Turbo Decoder 489

4. Keutzer, K., Malik, S., Newton, A.R.: ‘‘From ASIC to ASIP- next design discontinuity’’. In:
IEEE International Conference on Computer Design: VLSI in computers and processors,
pp. 84–90, IEEE Computer Society, Washington (2002)
5. Vogt,T., Wehn, N.: ‘‘A reconfigurable application specific instruction set processor for viterbi
and Log-MAP decoding’’. In: IEEE Workshop on Signal Processing Systems Design and
Implementation, SIPS ‘06 pp. 142–147, Banff, Canada (2006)
6. Muller, O., Baghdadi, A., Jezequel, M.:‘‘ASIP-based multiprocessor SoC design for simple
and double binary turbo decoding’’. In: Proceedings Design, Automation And Test In Europe,
pp. 1330–1335, Munich (2006)
7. Muller, O., Baghdadi, A., Jezequel, M.:‘‘From application to ASIP-based FPGA prototype: a
case study on turbo decoding’’. In: IEEE/IFIP International Symposium on Rapid System
Prototyping, Monterey, (2008)
Chapter 48
A Comparative Study of Web Application
Testing and Mobile Application Testing

Maryam Ahmed and Rosziati Ibrahim

Abstract Web application have gained increased acceptance over the years in
companies and organization as the world move to a global village. Software
developers have also grown interest in developing web applications compared to
stand-alone application because of the immense benefits it offers such as ubiquity,
platform dependence, low cost of support and maintenance, better speed and
performance, piracy proof etc. As mobile application emerged in the last decade,
attention has been focused on mobile applications by organizations and businesses
in order to maximize their profits as much as possible. There has been a rapid
increase of software release in the mobile applications store. As the growth of both
web application and mobile application increase, the question of quality assurance
remains a concern. A comparative study of software testing techniques can be
performed to improve the standard of testing of both web and mobile application.
This paper therefore reviews the similarity and difference in the testing
mechanism.

Keywords Software testing Mobile application testing
Web application test-



ing Mobile applications Web applications

48.1 Introduction

The idea of quality assurance has been in existence even before the existence of
software hence quality of manufactured products including software has been a
subject of keen interest from past till present. Quality has been an issue as long as

M. Ahmed (&)
R. Ibrahim

Computer Science and Informtion Technology, Universiti Tun Hussein Onn Malaysia
(UTHM), 86400 Parit Raja, Batu Pahat, Johor, Malaysia
e-mail: [email protected]
R. Ibrahim
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 491

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_48
492 M. Ahmed and R. Ibrahim

human has been producing product. Software testing is an important and major
area of Software Quality Assurance (SQA). Software testing is the process of
executing a program with the purpose of finding faults. Testing include effort to
find defects but does not include getting solution to fixing the defects. This is the
difference between testing and quality assurance as quality assurance is not limited
to developing the test plan but also include testing, preventing and fixing the faults
found during the process of testing. However, testing can cover different areas
such as Specification testing, design testing and implementation testing. Imple-
mentation testing which deals with the working system of the software is most
time referred to as software testing [1]. Testing cannot be complete or perfect.
Having a flawless testing is not realistic [1], as it could take more than a life time
to complete even the simple software. Saleh explained the unrealistic nature of
complete testing by using a small program of a user filling a text field of 20
characters. This program test will be complete by testing all possible input values.
If an assumption of 80 characters is done, then number of possible combinations is
2080. A computer that takes nanoseconds to process one combination, testing will
take 1011 years to complete all possible combination which is very impossible and
unrealistic. Since it is impossible to guarantee an 100 % error-free software, errors
that are not detected at the time of development before deployment may be dis-
covered and reported by the end user or in the process of testing for a subsequent
release of the software. Despite the imperfection of software testing, testers need to
put their best effort in improving the reliability and efficiency of the software as it
can affect the well-being of human and their safety.
The current surge in both number and demand for mobile applications has
called the attention of test engineers to the testing of mobile application as users
are more concerned about the quality and functionality of the software. However,
the mobile applications testing has been an issue given the different platforms the
applications work on such as the iOS, Windows mobile, and the Google android;
whose market has continued to grow rapidly over the years [2, 3]. Also, Hardware
complexity and diversity is another challenge in mobile application testing. Unlike
the PC environment that operates with one central processing unit (986-micro-
processor) on which applications running on the system must be tested (whether
voice or data services), mobile phones are built with different types of processors
that run at different speed and with different memory usage. Testing web appli-
cation involve looking into the application and attention should be paid to the
browser being used (considering factors like backward and forward buttons) [4].
Connectivity is another issue with the smart mobile device. A mobile device could
use the 3G technology, wifi, GPRS all with different level of signal strength [5].
Therefore, testing mobile application requires a testing technique that will be
compatible with the software and hardware complexity of the mobile phone.
48 A Comparative Study of Web Application Testing … 493

48.2 Web Application Testing

The advancement of web application is following an evolution similar to that of

software systems. Web applications tend to evolve rapidly and pass through fre-
quent modifications as a result of new technological and business opportunities as
well as feedback from users [6]. Web application testing can however be more
complex than testing standalone or traditional programs due to their synchronized,
dispersed and platform-independent environment on which it is been run. Several
researches have been going on since the last decade when web application started
becoming common around the globe. In 2000, research was carried out to extend
data flow testing technique to web application testing. An approach to supporting
data flow testing for web application was presented [7]. The approach is such that
the white box test artifacts of an application were captured in a Web Application
Test Model (WATM) there by treating each component of the web application as
an object, while the elements of an HTML or XML were considered as code
variables within an object. However, there is need for review of this technique and
other similar approach as this is based on the structural testing.
Qian [8] developed a tool for testing web application and to support analysis of
the application. They came up with ReWeb and TestWeb. The ReWeb views could
be useful in understanding the system organization in terms of navigation path and
variable usage while the TestWeb generator and executor of test cases of an
application can be used to explore the system to a satisfactory level. Improvement
is needed in reducing the manual activities in the process. Automation is required
in the area of state unrolling and merging. Bin Zhu, Huaikou Miao and Lizhi Cai
(2009) proposed an approach to generating test cases for web application testing
using a navigation tree considering the web browser history mechanisms and the
user interface facilities such as the back, forward and refresh tool [9]. Another
work on web application testing worth mentioning is the development of a cross-
browser web application testing tool (Shauvik Roy, 2010). This tool is to find the
difference between corresponding elements of a web page opened in different
browsers [10]. The tool was evaluated on nine real world applications and result
shows that the tool is effective in finding cross-browser issues while keeping the
false positive low. Qian Zhongsheng also came with an approach to reducing and
optimizing the test case generation from user request traces [8]. A large volume of
meaningful user sessions were obtained after purging their irrelevant information
by analyzing user logs on the web server. This approach evaluates test cases
considering the test coverage ratio only. Many factors need to be considered aside
from the test coverage ratio such as the running cost of each test case (CPU time),
loading time, time to save the test case and the influence of different test criteria on
a test case. The Table 48.1 summarizes the state of the art of web application
testing.
494 M. Ahmed and R. Ibrahim

Table 48.1 Web application testing techniques

Author Topic Methodology/ Comment
Contribution
Chien-Hung Liu Structural testing of Web application test This approach was
David C. Kung web applications model (WATM ) based on Structural
Pei Hsia Chih- or white box testing
Tung Hsu [7]
Fillipo Ricca Analysis and testing of ReWeb and TestWeb Involves manual
Pallo Tonella [6] web applications processes, therefore
need to review to
automate processes
Bin Zhu Huaikou Testing a web Navigation tree and Other factors affect
Miao Lizhi Cai application involving put into consideration the testing of web
[4] web browser web browser history aside from these two
interaction and user interface factors
Shauvik Roy A cross-browser web Tool to find the From the result, it
Choudhary application testing tool difference between two was effective in
Husayn Versee different browsers finding cross
Alessandro Orso browser difference
[10]
Qian Test case generation To reduce and More factors need to
Zhongsheng [8] and optimization for optimize test cases be considered
user session-based web generation from user
application testing request traces

48.3 Mobile Application Testing

Mobile application also known as Mobile Applications can be used to define the
applications that are developed for handheld smart devices such as the mobile
phones, etc. These applications can be pre-installed applications that come with
the phone straight from the manufacturer or it can be downloaded from the web or
app store to enhance better functionality of the smart device. These devices work
with powerful application for user consumption and for this reason increases its
complexity. The complexity of the mobile applications comes with a cost [9].
Mobile application testing can be very challenging. Many questions need to be
addressed when testing a mobile app. Different framework have been proposed for
mobile application testing. A cloud based approach was proposed by Baride and
Dutta. This approach was proposed to solve the problem of homogeneous testing
of mobile application of different platforms [11]. The success of this work was
explained in terms of it automation, adaptability to different mobile environment
and actual devices and the complexity of the testing which include performance
testing, security testing and synchronization testing. Another mobile testing
framework was proposed by Ping et al. This framework was based on the V-model
and cloud test mechanism. The criteria to be considered in designing a testing
model for mobile application was mentioned to include testing scope in terms of
the network connection been used, emulators used in testing and devices used [12].
48 A Comparative Study of Web Application Testing … 495

The developed model is however yet to be implemented. The work of Muhammad

Karami (2013) presented an automatic analysis approach for security inspection of
android applications. This research was inspired by the scrutiny that some malware
examples are only triggered on user behavior or action [13]. Hu Cuixiong proposed
a novel technique of bringing android specific bug to light and shows how to
construct an effective test automation approach for addressing the bugs detected,
hence ensuring the reliability of application running on android platform [5].
Different tools have also been developed to aid mobile application testing.
MobileTest (2007) is an automatic testing tool for black box mobile devices. This
tool can develop refined, maintainable and reusable test case library for testing
system level and application level software on different smart devices [14]. The
tool was validated by comparing with the result of the TestQuest. Hermes (2009) is
another tool for testing smart mobile applications. The motivation of development
of this tool comes from the heterogeneity of the mobile device [15]. Hermes can be
used for test writing and a distributed run time for automating test execution and
reporting. There is need to look into the cost/benefit ratio to improve the evaluation
of this tool. Adaptive Random Testing (ART) 2010: ART was also developed from
a black box view of testing mobile application [16]. The motivation was from the
fact that mobile application has to deal with user input and constant changing in
the device or user environment. MZoltar (2013): Mzoltar offers a vigorous analysis
of mobile applications which shows a diagnostic report that makes it easy to
comprehend [17]. This approach helps in localizing the bugs in android mobile
applications by relying on the Spectrum-based fault localization (SFL). Table 48.2
shows a summary of the state of the art of mobile application testing.

48.4 Difference of Web and Mobile Application Testing

With the high increase of mobile users and the rate at which the internet is being
accessed, there is need to bridge the gap in the quality of mobile and web
applications. There are different factors to be considered in understanding the
adaptability of the two testing techniques. These factors affect the way at which
these applications are developed and tested. Such factors include and not limited
to: Life conformance or event trigger mode, GUI (GUI of the Input and Output
system), Data Synchronization or Network Management, Power Management and
Memory management.

48.4.1 Event Context/Life Conformance

In web application running on a desktop, application life cycle is dependent on the

operating system. The operating system handles the states of the event the appli-
cation is passing through and it safeguards the precise comportment of the
496 M. Ahmed and R. Ibrahim

Table 48.2 Mobile application testing techniques

Author Topic Methodology/ Comment
Contribution
SrikanthBaride A cloud based Cloud based This approach did
Kamlesh Dutta [11] software testing approach to solve not cover the issue
paradigm for mobile the problem of of testing
applications heterogeneity of connectivity
mobile devices
Tan Ping Ping Hamizan Designing a mobile V-model and cloud An ongoing
Sharbini1 Wee Bui Lin application testing base mechanism research on mobile
[12] model testing framework
Mohammad Karami Behavioral analysis An automated There is need for
Mohamed Elsabagh of android dynamic analysis support for more
ParnianNajafiborazjani applications using approach for complex interfaces
Angelos Stavrou [13] automated Security inspection such as OpenGl and
instrumentation of android gesture views
applications
Cuixiong Hu Iulian Automating GUI An approach to test There were
Neamtiu [5] testing for android android application exceptions in the
applications with focus on the result due to bugs
GUI that do not fall
under activity/
event/type
categories
Jiang Bo Long Xiang MobileTest: A tool MobileTest took to This helps in
Gao Xiaopeng [14] supporting event based reducing the
automatic black box approach to simplify complexity of smart
test for software on generation of test mobile testing while
smart mobile cases still being effective
devices
Sakura She Sasindran Hermes: a tool for Hermes an More research to
Sivapalan Ian Warren testing mobile automated test validate and refine
[15] device applications execution and the tool especially
reporting tool in the area of
reporting
Zhifang Liu Xiaopeng Adaptive random Adaptive random The experimental
Gao Xiang Long [16] testing of mobile test (ART ) result of this tool
application shows that it
reduces time taken
to find first defect
and reduces number
of test cases
generated.
Pedro Machado José MZoltar: automatic MZoltar localizes This approach is
Campos Rui Abreu debugging of the bugs in android unique in
[17] android applications mobile applications identifying potential
by relying on the defects quickly
Spectrum-based through diagnostic
fault localization report
(SFL)
48 A Comparative Study of Web Application Testing … 497

application at all statuses. This is not the case in smart mobile operating system
such as iOS, Android. Due to limited resources and complexity of the system, the
operating systems cannot retain the comprehensive state of an application at any
time when there is a change in the state of the life cycle. The application therefore
takes care of itself to avoid data loss in case the application is paused or even
killed. This makes it a requirement for a mobile application to be life conformance
in its design. Testing the life conformance of an application life implies it is
responding reasonably to change of state in the operating system such that events
like low memory or low battery do not lead to loss of data in the mobile device
[18–20].

48.4.2 GUI

Testing GUIs on a general note is a tough and demanding task for so many
reasons: Firstly, the input space has a countless and potentially unlimited number
of combinations of inputs and states of system outputs, generating test cases then
becomes tougher. Secondly, even simple GUIs possess an enormous number of
states which are also due to interaction with the inputs. Also, many complex
dependencies may hold between different states of the GUI system, and/or between
its states and inputs [21]. Web application GUI testing still seems easier. Users of
desktop GUI applications might be expected to refer to documentation or lessons
to completely understand the usage of the applications. Unlike mobile device,
mobile applications are expected to have a simple and intuitive user interface
where most, if not all, usage scenarios of an application should be evident to the
average user from the GUI [22].

48.4.3 Network Management

The emphasis of present research is tending towards planning and building of

network and infrastructures and applications for mobile system. Testing software
in relation to network connection now remains a concern. This is becoming a
hindrance in the evolution of mobile computing since the development of smart
mobile application is challenging owing to the scarcity in computational resources.
The change in location and network can also lead to data loss and poor func-
tionality. This implies that a change in network or location can mean a small shift
away from the servers that is being used toward a new one [23].
498 M. Ahmed and R. Ibrahim

Table 48.3 Adaptability of mobile and web application testing

Features Mobile Web References
GUI Input: touch screen and key Input: key board Hu [5], She et al
board [15], Chang et al
Output: small screen Output: larger [17], Belli [20],
screen Yang et al. [26]
Life cycle Life cycle is dependent on Operating system Amalfitano et. al.
application since system takes care of the [8], Bo et al. [14],
resources are limited life cycle of an Franke et al. [21],
application She et al. [15]
Performance Mobility allow easy change Network Zhu et al. [4]
(Memory, in location and therefore connectivity not
storage and could result in network strip frequently cut-off
network off for short distance
management) network
Power Most mobile applications Power Zhang et al. [24],
management tend to consume much management Ismail et al. [27]
power, hence need for under control
power management mobile
software

48.4.4 Power Management

As users tend to use vital and complex applications, low power consumption
becomes a requirement in building mobile applications. Developers of smart
mobile device (hardware and software) have considered power-saving features,
such that applications can vigorously regulate their power ingestions depending on
the required functionality and performance. Software developers therefore need to
understand the implication of building a high power-hungry application and should
put into consideration the in-built feature of power management in mobile device
[24] (Table 48.3).

48.5 Conclusion

While the web has taken over businesses and individual life, the mobile is however
anticipated to surpass as the world platform for local and internet applications in
the nearest future. As the functionality of the mobile and web applications increase
so is the complexity and hence the complexity of the testing technique. There is
need to fill the gap in the testing of these two important applications area. The
difference in performance and testing could be viewed from the event context, GUI
interface, network management, power and memory management. There are
present works that have been looking into these factors as highlighted in this paper;
48 A Comparative Study of Web Application Testing … 499

however, there is a strong need for more research in building more effective testing
technique that can handle both the web and mobile application testing.

Acknowledgments This research is supported under the Graduate Research Incentive Grants
(GIPS), vote 1256, Universiti Tun Hussein Onn Malaysia.

References

1. Saleh, K.A.: Software engineering. J. Ross Publishing Inc. p. 244

2. Franke, D., Weise, C.: Providing a software quality framework for testing of mobile
applications. 2011 fourth IEEE international conference on software testing, verification and
validation, 431–434 (2011). doi:10.1109/ICST.2011.18
3. Wasserman, A. I.: Software engineering issues for mobile application development. In:
Proceedings of the FSE/SDP workshop on future of software engineering research – FoSER-
10, p. 397. (2010). doi:10.1145/1882362.1882443
4. Zhu, B., Miao, H., and Cai, L.: Testing a web application involving web browser interaction.
2009 10th ACIS international conference on software engineering, artificial intelligences,
networking and parallel/distributed Computing, 589–594 (2009). doi:10.1109/SNPD.2009.59
5. Hu, C.: Automating GUI testing for android applications, (Section 4), pp. 77–83
6. Ricca, F., Tonella, P.: Analysis and testing of web applications. In: Proceedings of the 23rd
international conference on software engineering. ICSE 2001, 25–34 (2001). doi:10.1109/
ICSE.2001.919078
7. Kung, C. L. D. C., Box, P. O.: Structural testing of web applications Chih-Tung Hsu, 84–96
8. Qian, Z.: Test Case Generation and Optimization for User Session-based Web Application
Testing. J. Comput 5(11), 1655–1662 (2010). doi:10.4304/jcp.5.11.1655-1662
9. Zhifang, L., Bin, L.: Test automation on mobile device (2007), 1–7 (2010)
10. Roy Choudhary, S., Versee, H., Orso, A.: A cross-browser web application testing tool. 2010
IEEE international conference on software maintenance, 1–6 (2010). doi:10.1109/ICSM.
2010.5609728
11. Baride, S., Dutta, K.: A cloud based software testing paradigm for mobile applications. ACM
SIGSOFT Softw. Eng. Notes 36(3), 1 (2011). doi:10.1145/1968587.1968601
12. Ping, T. P., Sharbini, H., Lin, W. B., Tan, V., Mun, W., Julaihi, A. A.: Designing a mobile
application testing model, 255–260
13. Karami, M., Elsabagh, M., Najafiborazjani, P., Stavrou, A.: behavioral analysis of android
applications using automated instrumentation
14. Bo, J., Xiang, L., Xiaopeng, G.: MobileTest: a tool supporting automatic black box test for
software on smart mobile devices. Second international workshop on automation of software
test (AST ’07), 8–8. (2007). doi:10.1109/AST.2007.9
15. She, S., Sivapalan, S., Warren, I.: Hermes: a Tool for Testing Mobile Device Applications.
Australian Softw. Eng. Conf. 2009, 121–130 (2009). doi:10.1109/ASWEC.2009.17
16. Liu, Z., Liu, B., Gao, X.: SOA based mobile application software test framework. 2009 8th
international conference on reliability, maintainability and safety, 765–769 (2009). doi:10.
1109/ICRMS.2009.5270087
17. Machado, P., Campos, J., & Abreu, R.: MZoltar: automatic debugging of Android
applications. In: Proceedings of the 2013 international workshop on software development
lifecycle for mobile - demobile 2013, 9–16 (2013). doi:10.1145/2501553.2501556
18. Amalfitano, D., Fasolino, A. R., Tramontana, P.: A GUI crawling-based technique for android
mobile application Testing. 2011 IEEE fourth international conference on software testing,
verification and validation workshops, 252–261. (2011). doi:10.1109/ICSTW.2011.77
500 M. Ahmed and R. Ibrahim

19. Amalfitano, D., Fasolino, A. R., Tramontana, P., Amatucci, N.: Considering Context Events
in Event-Based Testing of Mobile Applications. 2013 IEEE sixth international conference on
software testing, verification and validation workshops, 126–133. (2013). doi:10.1109/
ICSTW.2013.22
20. Belli, F.: Finite state testing and analysis of graphical user interfaces. In: Proceedings 12th
international symposium on software reliability engineering, 34–43 (2001). doi:10.1109/
ISSRE.2001.989456
21. Franke, D., Kowalewski, S.: Testing conformance of life cycle dependent properties of
mobile applications, (2012). doi:10.1109/ICST.2012.36
22. Yang, W., Prasad, M. R., Xie, T.: A grey-box approach for automated gui-model generation
of mobile applications
23. Satoh, I., Society, I. C.: A testing framework for mobile computing software, 29(12),
1112–1121 (2003)
24. Zhang, L., Dick, R. P., Mao, Z. M., Wang, Z., Arbor, A.: Accurate online power estimation
and automatic battery behavior based power model generation for smartphones, 105–114
25. Chang, T., Yeh, T., Miller, R. C.: GUI testing using computer vision, (Figure 1), 1535–1544
(2010)
26. Yang, Y. J., Kim, S. Y., Choi, G. J., Cho, E. S., Kim, C. J., Kim, S. D.: A UML-based object-
oriented framework development methodology. In: Proceedings 1998 asia pacific software
engineering conference (Cat. No.98EX240), 211–218. doi:10.1109/APSEC.1998.733722
27. Ismail, M.N., Ibrahim, R., MdFudzee, M.F.: A survey on content adaptation systems towards
energy consumption awareness. Adv. Multimedia 2013, 3 (2013)
Chapter 49
Multi-objective Functions in Grid
Scheduling

Zafril Rizal M. Azmi, M.A. Ameedeen

and Imran Edzereiq Kamarudin

Abstract In order to fully utilize the Grid resources, an implementation of a good

scheduling algorithm is greatly important. However, for a complex scheduler that
aims to achieve high performance for more than one performance metrics, a
suitable objective function should be carefully considered. This paper shows that a
different objective function will have different affect to the Grid performance.

49.1 Introduction

Recently, Grid computing have been attracting interest from researchers in the
field of mathematics [8], biosciences [10], engineering [17] and others important
fields. Furthermore, the increasing need for network, storage and computing
resources is projected to double every 9, 12 and 18 months respectively [7]. As the
increase in data size, processing complexity as well as communication technology
is becoming more challenging in the Grid computing environment, issues related
to performance becomes increasingly important. Achieving high-performance
Grid computing requires techniques to efficiently and adaptively allocate jobs to
available resources in a large scale, highly heterogeneous and dynamic environ-
ment. Therefore, it is very important for a Grid system to have a scheduler that
meets both administrator and user performance expectation.

Z.R.M. Azmi (&)  M.A. Ameedeen  I.E. Kamarudin

Faculty of Computer System and Software Engineering, Universiti Malaysia Pahang,
Lebuhraya Tun Razak, Gambang, 26300 Pahang, Kuantan, Malaysia
e-mail: [email protected]; [email protected]
M.A. Ameedeen
e-mail: [email protected]
I.E. Kamarudin
e-mail: [email protected]

Ó Springer International Publishing Switzerland 2015 501

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_49
502 Z.R.M. Azmi et al.

49.2 Single Objective and Multi-objective Function

Although scheduler is a very important element in a Grid based application, a good

scheduler always comes paired with an efficient objective function. In a Grid
environment, a feasible good solution is a must and an optimal solution is a goal.
Commonly used Priority Rule (PR) scheduling procedure such as Shortest Job
First (SJF), First Come First Serve (FCFS) and Earliest Deadline First (EDF) can
quickly create feasible solution. However, as problem complexity increases, the
solution may be far from optimal. In order to tackle this problem, more complex
scheduling procedure based on approximate and heuristic have been introduced.
Schedulers that apply standard PR procedure has no need for mechanisms to filter
the solution since the PR generate solution based on the job characteristics such as
job length and deadline, and the solution is final, but heuristics based scheduler
usually generate a set of solutions that needs a filtering mechanism in order to
select the solution that meets the objective. The purpose of objective function is to
conduct the filtering process by evaluating the quality of the solutions generated by
the scheduler.
From the perspective of this paper, objective function in general can be divided
to two categories: direct objective function and indirect objective function. Direct
objective function involves certain calculation implemented in the schedulers
specifically targeting the main objective function. For example in [11], in order to
achieve minimum makespan, the new generated solution by Genetic-Based
scheduler must have lower value of makespan compared to previous solution to be
accepted. Since the scheduler only accepts much lower makespan solution, the
overall makespan for the Grid system can be guaranteed to be minimum. On the
other hand, indirect objective function is the outcome or side effect of the direct
objective function. For example, by using Shortest Job on Fastest Node (SJFN)
[19] which targets to minimize the makespan, the side effect is the increasing of
flowtime and machine usage. The indirect objective function always reflects the
tradeoffs made by the scheduler to the objective function, for example, minimizing
objective A is subject to maximizing objective B. The tradeoff is often a main
topic for the multi-objective scheduling which has to be properly considered when
developing a new scheduler.
There are two types of direct objective functions implemented in scheduling
algorithms. The first one is just using a single objective function and another is
combining two or several objective functions to make a decision which also called
multi-objective function. Generally, makespan minimization is often used as the
criterion of optimization in a single objective function. Pandey in [21] have stated
that makespan is the most common objective function used in scheduling.
Makespan is the time taken to finish the overall jobs. Since it is very important to
minimize the overall time, many researchers used makespan as the objective
function in their work [9, 13, 24, 25].
49 Multi-objective Functions in Grid Scheduling 503

However using just a single objective function to schedule a set of jobs can
decrease the overall performance [21, 26]. Moreover, users’ objectives such as the
need for much faster completion of jobs and resource provider objective such as
maximizing resource utilization always conflict with each other [20, 28]. Obvi-
ously, it is not realistic to satisfy a single objective, but to search for a tradeoff,
which is nondominated solution. In addition, Farzi in [12] have stated that
scheduling in Grids computing is a multi-objective optimization problem, hence
objectives other then makespan must also be taken into consideration. To address
this problem and to increase other performance criteria’s, multi-objective function
has been introduced. Moreover, using more than single objective function is
believed to improve and also balance the performance. Multi-objective function
can have two or more objective functions working together to calculate the
decision. For a multi-objective function with k sub objectives, the objective can be
Pk
defined as f ð xÞ ¼ min wi fi ð xÞði ¼ 1; 2; . . .kÞ, and wi C 0 is the weight of the ith
i¼1
P
k
sub objective, in general, wi ¼ 1 [2, 27].
i¼1
Since multi objective function is very important, many researchers, in their
work have combined several single objective functions to make a decision for their
scheduler. For example [1] used the combination of makespan, machine usage and
acceptable queues to make a decision for a Genetic Algorithms (GA) based
scheduler. Although any combination of two criteria’s is possible, some combi-
nations are more obvious than the other. Nevertheless, most multi objective based
scheduler pairs makespan with other objective functions, but there is no concrete
reason why they are paired together. Leung in [18] have stated, a schedule that
simultaneously minimizes both makespan and flow time is called an ideal sche-
dule. This statement is agreed by [6] that focusing on minimizing makespan and
flow time by using the GA based scheduler.
Table 49.1 summarize the different between single-objective function and
multi-objective function. Selecting the best objective function for a specific
scheduler will determine the achievement of the scheduler objective. Hence, this
paper will further discuss the interaction between the Grid scheduler with objective
function in Sect. 49.3 and provide experimentation results that will show the
scheduler performance using different objective functions in Sect. 49.4.

49.3 Multi-objective Function in SH-PR-GSA Scheduler

In this paper, various single and multi-objective functions have been considered to
be tested. In order to do this, a set of PR schedulers optimize by Backfilling and
Simulated Annealing called Smallest Hole-Priority Rule-Guided Simulated
504 Z.R.M. Azmi et al.

Table 49.1 Comparison of single objective function with multi-objective function

Criteria Single objective functions Multi-objective functions
Decision making Decision whether to accept or reject Decision whether to accept or reject
process a schedule is based on one a schedule is based on more than
performance criteria one performance criteria
Target users A group of user that aim to achieve A group of user that have different
similar single objective objectives to achieve
Advantage Performance criteria used as an Most of the performance criteria
objective function will have higher will have average performance,
performance compared to other where no criteria is obviously lower
criteria’s than the other criteria’s
Disadvantage Some criteria performance are Criteria previously used as single
obviously lower than the criteria objective function will has a
used as an objective function dropdown in performance after
combined with other objective
functions
Key objective Makespan Makespan + nondelayed jobs
functions Nondelayed jobs Makespan + machine usage
analyze in this
Machine usage Makespan + tardiness
paper
Tardiness Machine usage + nondelayed jobs
Machine usage + tardiness
Tardiness + nondelayed jobs

Annealing (SH-PR-GSA) integrated with the selected objective functions have

been used as a test bed. Five PR algorithms [First Come First Serve (FCFS),
Shortest Job First (SJF), Longest Job First (LJF), Earliest Deadline First (EDF) and
Minimum Time To Deadline (MTTD)] have been used separately. The main
objective of SH-PR-GSA schedulers is to produce schedule aimed to achieve
minimum makespan, minimum flowtime, minimum total tardiness, minimum
delayed jobs and maximum machine utilization. In order to achieve this, objective
function is one of the components that should be carefully evaluated. This is
because, the decision of whether a new schedule should be accepted or rejected
depend on the objective function used in the scheduling algorithms. In SH-PR-
GSA, objective function has been integrated to the scheduler in SH-PR and GSA
modules separately as shown in grey box in Fig. 49.1. Although separately exe-
cuted, these two modules are using the same objective function in order to
maintain the consistency of the schedule. For example, in order to achieve mini-
mum possible makespan for the Grid, if SH-PR module implements a Makespan
Objective Function, GSA module must also implement the same objective func-
tion to evaluate the schedule.
49 Multi-objective Functions in Grid Scheduling 505

Fig. 49.1 Multi objective function in SH-PR-GSA

In this section objective function referred as ‘‘Multi Objective Function’’ if

there is more than one objective function integrated with SH-PR-GSA scheduler.
In order to effectively execute SH-PR-GSA to achieve highest possible perfor-
mance, this section has considered four main objective functions which are total
non-delayed jobs (Nondelayed), Makespan, Tardiness and Machine Usage. These
objective function have been chosen because each one of them, theoretically
representing the targeted performance metrics. For example, by accepting a
schedule with lower Nondelayed Objective Function, lowest possible number of
delayed jobs can be achieved. On the other hand, Makespan Objective Function
aims to achieve the best makespan performance for all SH-PR-GSA schedulers,
and as already discussed, higher makespan usually results in lower flowtime.
Tardiness Objective Function will ensure that the total tardiness metric will be the
lowest metric and finally Machine Usage Objective Function makes sure the
scheduler highly utilizes the resources as well as keeping the machine usage metric
optimum for SH-PR-GSA. Since objective function play an important part in
determining performance of a scheduler, it is very important to conduct a com-
prehensive study on this matter. During the development process of SH-PR-GSA,
apart from the main objective functions explained earlier, there are another six
multi objective functions analyzed, each one incorporated by two single objective
functions as shown in Table 49.2.
The weight is calculated to recognize whether the new schedule is better or worse
compared to current schedule by using Eq. 49.1 for objectives to be minimized such
as makespan and Eq. 49.2 for objective to be maximized such as machine usage.
506 Z.R.M. Azmi et al.

Table 49.2 Objective functions tested with SH-PR-GSA

Number Type of objective function/multi-objective function
1. Makespan objective function
2. Tardiness objective function
3. Usage objective function
4. Nondelayed jobs objective function
5. Makespan + nondelayed jobs multi-objective function
6. Makespan + usage multi-objective function
7. Makespan + tardiness multi-objective function
8. Usage + nondelayed multi-objective function
9. Usage + tardiness multi-objective function
10. Tardiness + nondelayed multi-objective function

weightobjective ¼ ðcurrentObjective  expectedObjectiveÞ=currentObjective

ð49:1Þ

weightobjective ¼ ðexpectedObjective  currentObjectiveÞ=currentObjective

ð49:2Þ

Decision based on single objective is done by referring to single Weightobjective.

However, by combining more than one objective functions (multi-objective
function) to make a decision over whether to accept or reject a new schedule, a
better performance is aimed to be achieved. Instead of focusing on a single
objective such as makespan, multi objective such as Makespan + Tardiness Multi-
Objective Function decide the solution based on two objectives and in this
example they are makespan and total tardiness objective function. Originally, the
SH-PR-GSA scheduler applied the combination of makespan and nondelayed jobs
weight. The total weight (totalWeight) that consists of weightmakespan and
weightnondelayed are calculated based on the Eq. 49.3 as suggested by Klusacek in
[16]. In order to analyze the effects of objective function to the Grid scheduler, this
paper has added six multi-objective functions as listed in No. 6–10 in Table 49.2
and defined in Eqs. 49.3–49.8.
Makespan and Nondelayed Jobs (Makespan + Nondelayed Jobs) Multi-
Objective Function:

wm ¼ ðmc  me Þ=mc
wn ¼ ðne  nc Þ=nc ð49:3Þ
X
w ¼ wm þ wn
49 Multi-objective Functions in Grid Scheduling 507

where
w Total weight (totalWeight)
wm Makespan weight (weightmakespan)
wn Nondelayed jobs weight (weightnondelayed)
mc Current value of makespan (currentMakespan)
me Expected value of makespan (expectedMakespan)
ne Expected value of nondelayed jobs (expectedNondelayed)
nc Current value of nondelayed jobs (currentNondelayed)

Makespan and Machine Usage (Makespan + Machine Usage) Multi-Objective

Function:

wm ¼ ðmc  me Þ=mc
wu ¼ ðue  uc Þ=uc ð49:4Þ
X
w ¼ wm þ wu

where
w Total weight (totalWeight)
wm Makespan weight (weightmakespan)
wu Machine usage weight (weightusage)
mc Current value of makespan (currentMakespan)
me Expected value of makespan (expectedMakespan)
ue Expected value of machine usage (expectedUsage)
uc Current value of machine usage (currentUsage)

Makespan and Tardiness (Makespan + Tardiness) Multi-Objective Function:

wm ¼ ðmc  me Þ=mc
wt ¼ ðtc  te Þ=tc ð49:5Þ
X
w ¼ wm þ wt

where
w Total weight (totalWeight)
wm Makespan weight (weightmakespan)
wt Tardiness weight (weighttardiness)
mc Current value of makespan (currentMakespan)
me Expected value of makespan (expectedMakespan)
te Expected value of tardiness (expectedTardiness)
tc Current value of tardiness (currentTardiness)
508 Z.R.M. Azmi et al.

Machine Usage and Nondelayed Jobs (Machine Usage + Nondelayed) Multi-

Objective Function:

wu ¼ ðue  uc Þ=uc
wn ¼ ðne  nc Þ=nc ð49:6Þ
X
w ¼ wu þ wn

where
w Total weight (totalWeight)
wu Machine usage weight (weightusage)
wn Nondelayed jobs weight (weightnondelayed)
ue Expected value of machine usage (expectedUsage)
uc Current value of machine usage (currentUsage)
ne Expected value of nondelayed jobs (expectedNondelayed)
nc Current value of nondelayed jobs (currentNondelayed)

Machine Usage and Tardiness (Machine Usage + Tardiness) Multi-Objective

Function:

wu ¼ ðue  uc Þ=uc
wt ¼ ðtc  te Þ=tc ð49:7Þ
X
w ¼ wu þ wt

where
w Total weight (totalWeight)
wu Machine Usage weight (weightusage)
wt Tardiness weight (weighttardiness)
ue Expected value of machine usage (expectedUsage)
uc Current value of machine usage (currentUsage)
te Expected value of tardiness (expectedTardiness)
tc Current value of tardiness (currentTardiness)

Tardiness and Nondelayed Jobs (Tardiness + Nondelayed Jobs) Multi-Objec-

tive Function:

wt ¼ ðtc  te Þ=tc
wn ¼ ðne  nc Þ=nc ð49:8Þ
X
w ¼ wt þ wn
49 Multi-objective Functions in Grid Scheduling 509

where
w Total weight (totalWeight)
wt Tardiness weight (weighttardiness)
wn Nondelayed jobs weight (weightnondelayed)
te Expected value of tardiness (expectedTardiness)
tc Current value of tardiness (currentTardiness)
ne Expected value of nondelayed jobs (expectedNondelayed)
nc Current value of nondelayed jobs (currentNondelayed)

A positive weightmakespan indicates the new generated schedule has lower

makespan than the current. It goes the same for positive weightnondelayed which
means the new schedule expected to have less number of delayed jobs compared to
current schedule. Similarly, when weighttardiness and weightusage are positive, it
means that the new schedule has lower tardiness and higher machine usage than the
current schedule. However, in case of multi–objective function that have clashing
values, in example Makespan + Tardiness Multi Objective Function with positive
value 0.5 of weightmakespan and negative value -0.2 of weighttardiness. The value for
totalWeight will be 0.3 which happen to be positive and also means the new
schedule will be accepted with the probability of total tardiness to be slightly high
from the original schedule. Suppose the next cycle weightmakespan has a negative
value -0.2 and weighttardiness carry a positive value 0.5. This time the new schedule
is also been accepted since the value of totalWeight is positive, with probability of
new schedule makespan to be slightly high compared to previous schedule. From
this example, the value of final makespan metric of Makespan + Tardiness Multi
Objective Function might be slightly higher from single Makespan Objective
Functions. Similarly, total tardiness metric of Makespan + Tardiness Multi-
Objective Function might be higher from single Tardiness Objective Function since
the first cycle accepted schedule with higher tardiness and second cycle accepted
schedule with higher makespan. However, in terms of overall performance, the
multi-objective function performance will be much better compared to single
objective function because multi-objective consider more than one objective in
making the decision and for the example given, the makespan metric and total
tardiness metric for Makespan + Tardiness Multi-objective function is much lower
compared to makespan metric of single Tardiness Objective Function and total
tardiness metric of single Makespan Objective Function.

49.4 Performance Metrics

Total Tardiness: One of the main objectives of the scheduling procedure is the
completion of all jobs before their agreed due dates. Failure to keep that promise
has negative effects on the credibility of the service provider. The lateness of job
510 Z.R.M. Azmi et al.

j can be defined as the difference between its completion time Cj and the corre-
sponding due date dj. This metric is known as the tardiness of job and is calculated
using the following expression:
 
Tj ¼ max 0; Cj  dj ð49:9Þ

On the other hand, for total tardiness involving a set of n jobs, which are to be
processed each in a single machine, the formula can be described as follows:
X
n  
T¼ max 0; Cj  dj ð49:10Þ
j¼1

Cj ¼ Sj þ pj ð49:11Þ

where Cj is the completion time of job j and Sj is the start time of job j in machine m.
Each job j has a processing time pj, and a due date dj.
Makespan: Makespan is a standard performance metric to evaluate scheduling
algorithms. A small value of makespan means the scheduler is providing good and
efficient planning of jobs to resources. The makespan of a schedule can be defined
as the time it takes from the instance the first task begins execution to the instance
at which the last task completes execution [25]. Makespan can be represented by
the following equation:
Cmax ¼ max Cj ð49:12Þ
1jn

In simple terms, makespan is the time it takes to finish the last job.
Flowtime: Flowtime, or also known as response time is the sum of completion
times Cj of all the jobs [25]. Mathematically, flowtime can be formulated as:
X
n
F¼ Cj ð49:13Þ
j¼1

Flowtime and makespan always become two major objectives to be minimized

in researches involving scheduling. However, minimization of makespan always
results in the maximization of response time [23].
Delayed Jobs: Delayed Jobs stands for jobs that fail to meet their deadline or
due date. Deadline is a period of time in which a job must be completed [5]. The
goal typically in such problems is to complete the maximum number of jobs by
their deadlines [3]. A higher machine usage fulfils resource owner’s expectations,
while a higher number of non delayed jobs guarantees a higher Quality of Service
49 Multi-objective Functions in Grid Scheduling 511

(QoS) provided to the users [14, 16]. By reducing the number of delayed jobs, QoS
for the system that uses the proposed scheduling technique will also be improved.
Delayed jobs D is measured by:
X
n
D¼ Dj ð49:14Þ
j¼1

 
1 if ðCj [ djÞ
where Dj ¼ ð49:15Þ
0 otherwise

Dj is equal to 1 if job j is late (Cj [ dj). Otherwise Dj is equal to 0.

Machine Usage: Maximizing resource utilization or machine usage in the grid
system is another important performance metrics. Utilization is the percentage of
time that a resource is actually occupied, as compared with the total time that the
resource is available for use. Low utilization means a resource is idle and wasted.
Throughout this paper resource utilization will be referred as machine usage.
According to [4, 22], Machine usage (MU) is computed as:

CPUactive
MU ¼   ð16Þ
min CPUavialable ; CPUrequired

where CPUactive denotes the numbers of non idle CPU, CPUavialable denotes the
number of available CPU, and CPUrequired represent the number of required CPU.

49.5 Results and Discussions

This section will present and discuss results obtained from the experimentation
conducted to SH-PR-GSA schedulers based on different objective functions.
However this experimentation only focuses on the situation where the Grid system
has to manage a heavy load data since optimization is more efficient in this kind of
environment. Simulation had been conducted using Alea [15] simulator with total
number of 3,000 jobs and 150 machines.
Figures 49.2, 49.3, 49.4, 49.5 and 49.6 provide the performance results of
SH-PR-GSA schedulers using different single and multi objective functions.
In order to get a clear picture of differentiation between each objective functions,
the results have been regrouped based on the objective function, and the per-
centage of difference for each objective function over makespan objective function
512 Z.R.M. Azmi et al.

Fig. 49.2 Delayed jobs of

SH-PR-GSA using different
objective functions

Fig. 49.3 Flowtime of

SH-PR-GSA using different
objective functions
49 Multi-objective Functions in Grid Scheduling 513

Fig. 49.4 Makespan of

SH-PR-GSA using different
objective functions

Fig. 49.5 Machine usage of

SH-PR-GSA using different
objective functions
514 Z.R.M. Azmi et al.

Fig. 49.6 Total tardiness of

SH-PR-GSA using different
objective functions

Fig. 49.7 SH-PR-GSA

performance using
nondelayed objective
function: percentage of
different compared with
makespan objective function
49 Multi-objective Functions in Grid Scheduling 515

Fig. 49.8 SH-PR-GSA

performance using machine
usage objective function:
percentage of different
compared with makespan
objective function

have been calculated. Results for single objective functions can be seen in
Figs. 49.7, 49.8 and 49.9. Positive graph means the criteria for that particular
objective function is better than makespan objective function while negative
means it is worse than makespan objective function. From these results, it can be
clearly noticed that if the applied objective function focuses on only one particular
metric (i.e. tardiness), the benefit it gives is only to that particular criteria. Other
metrics most probably will show significantly bad performances.
516 Z.R.M. Azmi et al.

Fig. 49.9 SH-PR-GSA

performance using tardiness
objective function:
percentage of different
compared with makespan
objective function

For instant, Fig. 49.9 which applied tardiness objective function in SH-PR-GSA
show 36–55 % better than makespan objective functions in term of total tardiness
metric. However, it suffers from a great disadvantages for other metrics especially
makespan metric which is -1,900 to -3,300 % lower compared to makespan
objective function. This is because by focusing on the tardiness, schedulers only
approve new solutions if the new solution appears to be satisfying the tardiness
objective without considering other metrics. By doing this, the total tardiness time
can be reduced but the scheduler might accept more jobs that have earlier deadline
49 Multi-objective Functions in Grid Scheduling 517

Fig. 49.10 SH-PR-GSA

performance using
makespan + nondelayed
multi-objective function:
percentage of different
compared with makespan
objective function

and reject jobs with much later deadline. This situation might lead to the blocking
of many other jobs especially jobs that does not have specific deadlines (30 % of
them in datasets). Hence, the very low percentage in makespan metrics justify that
it is contributed mainly by jobs that have the longest processing time as well as
jobs that does not have deadline. This situation is the same for all single objective
functions including makespan objective functions. If makespan objective function
518 Z.R.M. Azmi et al.

Fig. 49.11 SH-PR-GSA

performance using
makespan + machine usage
multi-objective function:
percentage of different
compared with makespan
objective function

were applied, except for makespan, other matrices might not show their best
performance.
When multi-objective is adapted to the scheduler (Figs. 49.10, 49.11, 49.12,
49.13, 49.14 and 49.15), this situation changes dramatically. The multi-objective
function will try to balance the performance between two metrics and give a much
better tradeoff to the others compared to single objectives. However an example
such as Fig. 49.12 that uses multi objective function does not show as good results
as expected. For multi-objective function that combines makespan and tardiness
criteria, there are significant drops of performance for flowtime, makespan and
machine usage metrics compared to Makespan Objective Function. However, if
compared to the results of the single Tardiness Objective Function (previously
49 Multi-objective Functions in Grid Scheduling 519

Fig. 49.12 SH-PR-GSA

performance using
makespan + tardiness multi-
objective function:
percentage of different
compared with makespan
objective function

discussed Fig. 49.6), actually the performance of Makespan + Tardiness Multi

Objective function has recorded significant improvement. For example, with the
Tardiness Objective Function, makespan for SH-FCFS-SA is -1,962 % worse
than the makespan recorded for makespan objective function. After integrating
tardiness with makespan in a multi-objective function (Fig. 49.12), the SH-FCFS-
GSA makespan is now -450 %, which is 1,512 % better compared to a single
Tardiness Objective Function while at the same time also still preserving the total
tardiness metrics at 34 %. Comparing between multi objective functions, as for
Makespan + Machine Usage Multi Objective Function (Fig. 49.11), some of the
performance metrics is slightly better than Makespan + Tardiness Multi Objective
520 Z.R.M. Azmi et al.

Fig. 49.13 SH-PR-GSA

performance using machine
usage + nondelayed multi-
objective function:
percentage of different
compared with makespan
objective function

Function, but it is not really stable for all the schedulers and the improvement
made is too little (below 0.2 %). The most perfect combination is Make-
span + Nondelayed Multi Objective Function as shown in Fig. 49.10 which also
implemented by EG-EDF-TABU in [16]. Three out of five performance metrics
show significant improvement. The tradeoff for the other two, makespan and
machine usage is acceptably low. It is also stable compared to the previously
explained combinations.
49 Multi-objective Functions in Grid Scheduling 521

Fig. 49.14 SH-PR-GSA

performance using machine
usage + tardiness multi-
objective function:
percentage of different
compared with makespan
objective function
522 Z.R.M. Azmi et al.

Fig. 49.15 SH-PR-GSA

performance using
tardiness + nondelayed
multi-objective function:
percentage of different
compared with makespan
objective function

49.6 Conclusions

In this paper, analysis on the effects of objective functions has been made by
testing the Grid scheduler with various single and multi-objective functions. This
analysis is important to show that a bad choice of objective function can extremely
affect the scheduler. The best objective function recognized after completing the
experimentation is Makespan + Nondelayed Multi Objective Function. This
objective function provides the scheduler with a tolerance tradeoff between five
criteria’s; makespan, flowtime, machine usage, delayed jobs and tardiness.

Acknowledgement This research was supported by Universiti Malaysia Pahang Research Grant
(RDU1203116)
49 Multi-objective Functions in Grid Scheduling 523

References

1. Albert, Y.Z.: Observations on using genetic algorithms for dynamic load-balancing. IEEE
Trans. Parallel Distrib. Syst. 12, 899–911 (2001)
2. Altiparmak, F., Gen, M., Lin, L., Paksoy, T.: A genetic algorithm approach for multi-
objective optimization of supply chain networks. Comput. Ind. Eng. 51(1), 196–215 (2006)
3. Bansal, N., Chan, H.-L., Lam, T.-W., Lee, L.-K.: Scheduling for speed bounded processors.
In: Aceto, L., Damgård, I., Goldberg, L., Halldórsson, M., Ingólfsdóttir, A., Walukiewicz, I.
(eds.) Automata, Languages and Programming, pp. 409–420. Springer, Berlin (2008)
4. Baraglia, R., Dazzi, P., Capannini, G., Pagano, G.: A multi-criteria job scheduling framework
for large computing farms. In: 2010 IEEE 10th International Conference on Computer and
Information Technology (CIT) (2010)
5. Brucker, P.: Scheduling Algorithms, 5th edn. Springer, Berlin (2007)
6. Carretero, J., Xhafa, F.: Using genetic algorithms for scheduling jobs in large scale grid
applications. J. Technol. Econ. Dev. 12, 11–17 (2006)
7. Casanova, H.: Distributed computing research issues in grid computing. SIGACT News
33(3), 50–70 (2002)
8. Collignon, T.P., van Gijzen, M.B.: Minimizing synchronization in IDR (s). Numer. Linear
Algebra Appl. 18, 805–825 (2011)
9. Cooper, K., Dasgupta, A., Kennedy, K., Koelbel, C., Mandal, A., Marin, G., Mazina, M.,
Mellor-Crummey, J., Berman, F., Casanova, H., Chien, A., Dail, H., Liu, X., Olugbile, A.,
Sievert, O., Xia, H., Johnsson, L., Liu, B., Patel, M., Reed, D., Deng, W., Mendes, C., Shi, Z.,
YarKhan, A., Dongarra, J.: New grid scheduling and rescheduling methods in the GrADS
project. In: Proceedings of 18th International Parallel and Distributed Processing Symposium
(2004)
10. Dickmann, F., Falkner, J., Gunia, W., Hampe, J., Hausmann, M., Herrmann, A., Kepper, N.,
Knoch, T.A., Lauterbach, S., Lippert, J., Peter, K., Schmitt, E., Schwardmann, U., Solodenko,
J., Sommerfeld, D., Steinke, T., Weisbecker, A., Sax, U.: Solutions for biomedical grid
computing–Case studies from the D-Grid project Services@MediGRID. J. Comput. Sci. In
Press, Corrected Proof (2011)
11. Entezari-Maleki, R., Movaghar, A.: A genetic-based scheduling algorithm to minimize the
makespan of the grid applications, in grid and distributed computing, control and automation.
In: Kim, T.-h., Yau, S., Gervasi, O., Kang, B.-H., Stoica, A., Śle˛zak, D. (eds.), pp. 22–31.
Springer, Berlin (2010)
12. Farzi, S.: Efficient job scheduling in grid computing with modified artificial fish swarm
algorithm. Int. J. Comput. Theory Eng. 1(1), 1793–8201 (2009)
13. Izakian, H., Abraham, A., Snášel, V.: Metaheuristic based scheduling meta-tasks in
distributed heterogeneous computing systems. Sensors 9(7), 5339–5350 (2009)
14. Klusacek, D., Rudova, H.: Improving QoS in computational grids through schedule-based
approach. In: Scheduling and Planning Applications Workshop at the Eighteenth
International Conference on Automated Planning and Scheduling (ICAPS 2008): Sydney,
Australia (2008)
15. Klusacek, D., Rudova, H.: Alea 2: job scheduling simulator. In: Proceedings of the 3rd
International ICST Conference on Simulation Tools and Techniques. ICST (Institute for
Computer Sciences, Social-Informatics and Telecommunications Engineering):
Torremolinos, Malaga, Spain. pp. 1–10 (2010)
16. Klusacek, D., Rudová, H., Baraglia, R., Pasquali, M., Capannini, G.: Comparison of multi-
criteria scheduling techniques. In: Gorlatch, S., Fragopoulou, P., Priol, T. (eds.) Grid
Computing, pp. 173–184. Springer, US (2008)
17. Komisarczuk, P., Welch, I.: Internet sensor grid: experiences with passive and active
instruments. In: Pont, A., Pujolle, G., Raghavan, S. (eds.) Communications: Wireless in
Developing Countries and Networks of the Future, pp. 132–145. Springer, Boston (2010)
524 Z.R.M. Azmi et al.

18. Leung, J.Y.-T.: Handbook of Scheduling: Algorithms, Models and Performance Analysis.
CRC Press, Boca Raton (2004)
19. Liu, H., Abraham, A., Hassanien, A.E.: Scheduling jobs on computational grids using a fuzzy
particle swarm optimization algorithm. Future Gener. Comput. Syst. 26(8), 1336–1343
(2010)
20. Oluwatope, A., Iyanda, D., Aderounmu, G., Adagunodo, R.: Computational modeling of
collaborative resources sharing in grid system. In: Dua, S., Sahni, S., Goyal, D.P. (eds.)
Information Intelligence, Systems, Technology and Management, pp. 311–321. Springer,
Berlin (2011)
21. Pandey, S., Buyya, R.: Scheduling of scientific workflows on data grids. In: 8th IEEE
International Symposium on Cluster Computing and the Grid, 2008. CCGRID ‘08 (2008)
22. Pasquali, M., Baraglia, R., Capannini, G., Ricci, L., Laforenza, D.: A multi-level scheduler
for batch jobs on grids. J. Supercomput 57(1), 81–98 (2011)
23. Subashini, G., Bhuvaneswar, M.C.: Non dominated particle swarm optimization for
scheduling independent tasks on heterogeneous distributed environments. Int. J. Adv. Soft
Comput. Appl. 3(1), 1–17 (2011)
24. Vazquez, M., Whitley, D.: A comparison of genetic algorithms for the static job shop
scheduling problem. In: Schoenauer, M., Deb, K., Rudolph, G., Yao, X., Lutton, E., Merelo,
J., Schwefel, H.-P. (eds.) Parallel Problem Solving from Nature PPSN VI, pp. 303–312.
Springer, Berlin (2000)
25. Xhafa, F., Abraham, A.: Meta-heuristics for grid scheduling problems. In: Xhafa, F.,
Abraham, A. (eds.) Metaheuristics for Scheduling in Distributed Computing Environments,
pp. 1–37. Springer, Berlin (2008)
26. Xiao-Juan, W., Chao-Yong, Z., Liang, G., Pei-Gen, L.: A survey and future trend of study on
multi-objective scheduling. In: Fourth International Conference on Natural Computation,
2008. ICNC ‘08 (2008)
27. Xue, X.D., Cheng, K.W.E., Ng, T.W., Cheung, N.C.: Multi-objective optimization design of
in-wheel switched reluctance motors in electric vehicles. IEEE Trans. Ind. Electron. 57(9),
2980–2987 (2010)
28. Yang, Y., Wu, G., Chen, J., Dai, W.: Multi-objective optimization based on ant colony
optimization in grid over optical burst switching networks. Expert Syst. Appl. 37(2),
1769–1775 (2010)
Chapter 50
Experimental Analysis on Available
Bandwidth Estimation Tools for Wireless
Mesh Network

Imran Edzereiq Kamarudin, M.A. Ameedeen

and Zafril Rizal M. Azmi

Abstract Measurement of available bandwidth in a wireless mesh network

(WMN) environment has always been a topic of great interest. Several active and
passive-based tools has been tested and proposed in previous research. However,
the performance of these tools was never tested extensively in terms of the con-
dition of the WMN such as varying bandwidth across the network and external
traffic factors. In this work, we perform an extensive experimental analysis study
on both active and passive available bandwidth tools by looking at the accuracy,
failure rate and consistency of each tool. We also investigate the effects of varying
the WMN bandwidth and external traffic on the performance of these tools. Our
results indicate that all tools performance was affected by the WMN testing
environment. In term of accuracy, failure rate and consistency, Pathload was the
most favorable tool in these conditions.

Keywords WMN  ASSOLO  Pathchirp  IGI/PTR  Pathload  Pathrate 

Wbest

50.1 Introduction

Wireless mesh network (WMN) has been a standard deployment spanning from
large enterprise network to small home network. From the beginning with 802.11a
which only support up to 2 Mbps of data stream, up to the latest 802.11ac which

I.E. Kamarudin (&)  M.A. Ameedeen  Z.R.M. Azmi

Universiti Malaysia Pahang, Lebuhraya Tun Razak, 26300 Gambang, Kuantan, Pahang,
Malaysia
e-mail: [email protected]
M.A. Ameedeen
e-mail: [email protected]
Z.R.M. Azmi
e-mail: [email protected]; [email protected]

 Springer International Publishing Switzerland 2015 525

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_50
526 I.E. Kamarudin et al.

supports up to 1 Gbps, the rapid growth of data stream was achieved in less than
5 years’ time frame only. Each new standard defined in the 802.11 protocol show
significant improvement in terms of the maximum bandwidth supported. For
example, 802.11b supports up to 11 Mbps whereas the later improve 802.11 g
supports up to 54 Mbps.
A common mistake done by user is to assume that the actual bandwidth sup-
ported by the WMN is based on the protocol itself. For example, in 802.11 g
wireless network, it supports constantly at 54 Mbps. As a result, this assumption is
wrong and may lead to underestimate the total bandwidth require by the network.
In order to correctly measure the actual bandwidth, the concept of available
bandwidth is introduced.
As mentioned before, bandwidth capacity refers to the maximum data or
throughput that can be transmitted on a link or a medium. It is important to
understand and identify the maximum throughput of a link in network planning to
cater the needs of the end user or end nodes. The available bandwidth (ABW) at a
link is its unused capacity. Since, at any time, a link is either idle or transmitting
packets at the maximum speed, the definition of the available bandwidth ought to
look at the average unused bandwidth over some time interval T. Thus,

1 Tþt
Z
Aiðt; TÞ ¼ ðCi  kiðtÞÞdt ð50:1Þ
T t

where Ai (t; T) is the available bandwidth at link i at time t, Ci is the link’s

capacity, and ki is its traffic. The available bandwidth along a path is the minimum
available bandwidth of all traversed links.
To measure the AWB in WMN, there are two techniques to estimate—passive
and active measurement. Passive measurement is performed by observing the
traffic without intruding the network. Active measurement on the other hand, will
probe the network by generating packet traffic into the network to perform the
measurement. Basically, all available bandwidth tools are created using either one
technique. Several recent works was done on both type of AWB but mainly
focusing on only one type of technique [1, 2]. As a result, comparison between
active and passive tools is not clear. Furthermore, testing was done purely on the
network without any external traffic to look at the effect of AWB when other traffic
is generated in the WMN [3–5].
The objective of our study is to evaluate existing active and passive AWB tools
and study their performance for 802.11-based WMN. We evaluate these tools by
varying various parameters such as physical data rate and the existing of other
traffic in the network. Our main contributions are highlighted as below:
• We select three AWB tools from both active and passive technique, and
evaluate their performance in WMN.
• We perform extensive testing in terms of the WMN physical rate and the
condition of the WMN (with and without external traffic other than the AWB
tools)
50 Experimental Analysis on Available Bandwidth Estimation Tools … 527

• We evaluate the tools based on: (1) Accuracy: This will measure the accuracy
of the tool to estimate the available bandwidth whether it will over estimate or
under estimate the available bandwidth. (2) Failure patterns: This attribute will
monitor and measure the reliability of the tool’s failure or error prone to
estimate the bandwidth throughout the testing cycle. (3) Consistency of mea-
surement: This attribute will measuring the consistency of the measurement of
the tool as whether it will fluctuate of over estimating or under estimating the
bandwidth.
The rest of the paper is organized as follows. Section 50.2 lists the previous
works and explains the foundation and motivation behind our work. Section 50.3
outlined the details of our tools, testbed settings and WMN environment for testing
and evaluation methodology. In Sect. 50.4, we present the results of the com-
parison study for both WMN environments. Section 50.5 concludes the paper.

50.2 Foundation

In this chapter, preliminary information regarding the technologies and the testbed
environment used to undertake this research is introduce in order for the reader to
easily understand the contents of this paper.

50.2.1 Active-Based AWB Tools

Active-based AWB tool is based on the idea of induced congestion, in which probe
packets are sent at increasing rates. At the receiver, the delays of the probe packets
are measured to determine the point at which they start increasing in a consistent
basis. The available bandwidth is then estimated by looking at the probe packet
rate utilized when the turning point is found. PathChirp [3], ASSOLO [6], and IGI/
PTR [5] are examples of tools utilizing this approach.
PathChirp [3] sends a variable bit-rate stream consisting of exponentially
spaced packets. The actual unused capacity is inferred from the rate responsible for
increasing delays at the receiver side. During probing process, Pathchirp increases
the probing rate within each chirp (having a variable number of packets) in an
exponential manner. By doing that, it captures delay correlation information using
a smaller number of probing packets. Pathchirp also uses variable size probe
packets of minimum 40 bytes.
ASSOLO [3] is a tool based on the same principle, but it features a different
probing traffic profile and uses a filter to improve the accuracy and stability of
results. It is based on the concept of ‘‘self-induced congestion’’. It features a
528 I.E. Kamarudin et al.

replacement inquisitor traffic profile known as REACH (Reflected Exponential

Chirp) which tests a large range of rates being a lot of accurate in the center of the
inquisitor interval. Moreover, the tool runs inside a real-time operating system and
uses some de-noising techniques to improve the activity method.
IGI/PTR [5, 7] obtain the most accurate measurement when the packet-train
sending rate at source equals its inbound rate at destination, where the initial
packet combine gap that provides a high correlation between the packet gap
changes and also the competing traffic throughput on the tight link. Both IGI and
PTR share the probing procedure, the main difference between them is that IGI
focuses on calculating background traffic load, whereas PTR directly calculates
packet transmission rate, to estimate end-to-end available bandwidth.

50.2.2 Passive-Based AWB Tools

In passive-based AWB tool also known as Probe Gap Model (PGM), packet pairs
(or packet trains) are sent to the path at a single rate. This probing rate is set to the
capacity of the tight link, and so it is larger than (or equal to) the available
bandwidth of the path. Pathload [5, 8], Pathrate [9] and WBest [10] are examples
of tools utilizing this approach.
Pathload [5, 8] estimate the available bandwidth of an end-to-end path from a
host S (sender) to a host R (receiver). The available bandwidth is the maximum IP-
layer throughput that a flow can get in the path from S to R, without reducing the
rate of the rest of the traffic in the path. The basic idea in Pathload is that the one-
way delays of a periodic packet stream show increasing trend, when the stream is
larger than the available bandwidth. The measurement algorithm is iterative and it
requires the cooperation of both Sender and Receiver. Pathload consists of a
process running at S and a process running at R. S sends periodic streams of UDP
packets from S to R at a certain rate.
Pathrate [9] uses packet pairs and packet-trains, in conjunction with statistical
techniques, to estimate the capacity of the narrow link in the path. It uses UDP for
transferring probing packets. Additionally, Pathrate establishes a TCP connection,
referred to as control channel, between the sender and the receiver. The control
channel acknowledges each correctly received packet pair or train, and it also
transfers commands sent from the receiver to the sender regarding the probing
packet size, train length, number of measurements, train spacing, etc.
WBest [10] is a two stage algorithm: (1) a packet pair technique estimates the
effective capacity over a flow path where the last hop is a wireless LAN (WLAN);
and (2) a packet train technique estimates achievable throughput to infer the
available bandwidth. WBest parameters are optimized given the tradeoffs of
accuracy, intrusiveness and convergence time.
50 Experimental Analysis on Available Bandwidth Estimation Tools … 529

50.3 Comparison of Existing Tools

In this section, we describe the WMN testbed environment used and the validation
methodology.

50.3.1 Tools and WMN Testbed Environment

We selected three tools each from active: ASSOLO, IGI/PTR, Pathchirp and
passive; Pathload, Pathrate, WBest probe base tools. For evaluation, we created a
testbed in a close environment consisting of 802.11 g wireless devices and access
point (AP). This was to make sure that no other wireless network signals can
interfere with out test environment. Two wireless nodes consist of two Linux-
based laptops were setup in order to run the tools. To create the 802.11 g WMN
environment, two APs were used. The settings for APs were set as default except
for data rate.
In summary, there were two conditions in WMN environments for testing: (1)
WMN without any external traffic (2) WMN with external traffic. For the first
condition, the tools were tested in an optimum environment where the only traffic
was from the tools itself. For the second condition, FTP traffic was generated on
top of the tools itself. Both condition were then tested in three different settings
(Fig. 50.1).
• One AP with two nodes—Tools was tested across a single AP as shown in
Fig. 50.2.
• Two AP with two nodes—Tools was tested across two AP via bridging.
Bandwidth between APs remain the same as shown Fig. 50.3.
• Two AP with two nodes and variable bandwidth—Similar with above except
that the bandwidth between APs was lower compare to the nodes. This setting
helps to look at the impact of changing of bandwidth across the network
towards AWB tools.

Fig. 50.1 One AP with two nodes

530 I.E. Kamarudin et al.

Fig. 50.2 Two APs with two nodes

100%

80%

60%

40%

20%

0%
ASSOLO IGI/PTR PATHCHIRP PATHLOAD PATHRATE WBEST
One Access Point(AP) Two AP (Same Bandwidth)

Two AP (Different Bandwidth) One Access Point(AP) + FTP

Two AP (Same Bandwidth) + FTP Two AP (Different Bandwidth) + FTP

Fig. 50.3 Accuracy comparisons

50.3.2 Testing and Evaluation Methodology

For each of the selected tools, we ran a few experiments just to make sure that all
tools has the same default parameter such as probe-size and number of probes.
This was an important step to make sure that all tools were tested at the same rate.
We observed that some tools where having higher numbers of probe. If some
minor changers, we managed to leverage all tools into the same settings in terms of
probe size and number of probes.
Each tool was then tested in terms of accuracy, failure pattern and consistency
across the two condition WMN environment with three different settings. For
sampling purposes, 20 reading were taken for each of the tools. The summary of
test conducted as below:
• WMN environment without external traffic: (1) One AP with two nodes (2)
Two AP with two nodes (same bandwidth) (3) Two AP with two nodes (var-
iable bandwidth)
50 Experimental Analysis on Available Bandwidth Estimation Tools … 531

• WMN environment with external traffic: (1) One AP with two nodes (2) Two
AP with two nodes (same bandwidth) (3) Two AP with two nodes (variable
bandwidth)

50.4 Experimental Results

50.4.1 Accuracy

For this experiment, we measured the accuracy of the tool to estimate the available
bandwidth whether it will over estimate or under estimate the available bandwidth.
For the available bandwidth, although the capacity is 54 Mbps, the achievable
available bandwidth will not reach to that level. It is fairly common for 802.11 g
connections to run at 36, 24 Mbps, or even lower [11]. Recommended data rate
might be close to 45 Mbps [12]. Hence, in our assessment, the benchmark accu-
racy of available bandwidth was set at the range of 8–45 Mbps. So the reading that
falls into this range will be considered as accurate.
However, we took two other factors into consideration: (1) The under and
overestimation of the tool to estimate the bandwidth. Overestimation in this case
means that the tool overestimated the bandwidth of the 802.11 g bandwidth which
is 54 Mbps. (2) For simulation with external traffic, any reading that falls below
the average of the same simulation without external traffic were omitted

Number of times within range

Accuracy ½within range 8  54 Mbps ¼  100 %
20
ð50:2Þ

Based on our findings on WMN environment without external traffic, ASSOLO

was the only tool able to produce 100 % accurate reading within the benchmark
bandwidth (8–54 Mbps) for all three settings. In contrast, WBest had the worst
accuracy rate at below 35 %. We found that WBest had a very high rate of failure
in detecting available bandwidth.
When testing on WMN environment with external traffic, Pathload were able to
produce 100 % accurate reading within the benchmark for all three settings. In
contrast, Pathchirp had the lowest percentage at 12 %.
In summary when looking at the average accuracy rate across all environments
and settings, Pathload provided the most consistent accuracy of 94 % while
Pathchirp had the worst consistency rate at 54 %. We also found that other than
Pathload, the rest of the tools tend to give inaccurate readings especially when
tested with external traffic and variable bandwidth settings. Figure 50.4 shows the
accuracy rate for all tools with the respective testing environment.
532 I.E. Kamarudin et al.

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
ASSOLO IGI/PTR PATHCHIRP PATHLOAD PATHRATE WBEST
One Access Point(AP) Two AP (Same Bandwidth)
Two AP (Different Bandwidth) One Access Point(AP) + FTP
Two AP (Same Bandwidth) + FTP Two AP (Different Bandwidth) + FTP

Fig. 50.4 Failure rate comparison

50.4.2 Failure Patterns

The second attributes which is the failure pattern, where we measured and eval-
uated the reliability of the tool to measure the bandwidth as whether it will succeed
to measure the bandwidth under the 802.11 g wireless environment, or otherwise
fail to measure. The reading ‘0’ or negative value indicated that the tool failed to
estimate the bandwidth. The percentage will be calculated and compared among
the tools.

number of 0 readings
Failure pattern ¼  100 % ð50:3Þ
20

Based on our findings on WMN environment without external traffic, WBest

had the highest failure rate at 65 %. ASSOLO, Pathrate and Pathload recorded 0 %
failure rate. When we tested on WMN environment with externall traffic, WBest
failure rate was at 90 %. The rests of the tools were below 40 %. In summary,
WBest had the highest average failure rate across all environment and settings
with 65 % failure rate while ASSOLO, Pathrate, Pathchirp and Pathload recorded
less than 2 % average in failure rate. In addition, we found that apart from Pathrate
and Pathload, the rest of the tools were affected when tested with WMN envi-
ronment with external traffic and variable bandwidth settings. Figure 50.4 shows
the accuracy rate for all tools with the respective testing environment.
50 Experimental Analysis on Available Bandwidth Estimation Tools … 533

Table 50.1 Standard devition for all tools

Consistancy ASSOLO IGI/ PATHCHIR PATHLOAI PATHRATE WBEST
PTR
One access point (AP) 0.0432 1.3982 1.2806 0.6401 0.2421 4.0357
One acess point 1.2818 0.6821 0.6577 1.3035 0.7891 0.0000
(AP) + FTP
Two AP (Same 1.2187 0.4480 0.3593 0.6621 0.5174 3.8013
bandwidth)
Two AP (Same 0.4510 2.5232 0.6448 0.6654 0.8035 0.9292
bandwidth) + FTP
Two AP (Different 0.4396 0.4005 0.4656 0.4859 0.7891 0.6540
bandwidth)
Two AP (Different 1.4857 0.5861 0.3656 0.3677 1.0867 1.1005
bandwidth + FTP)

50.4.3 Consistency

In this assessment, a calculation of mean and standard deviation was done. The
calculation was done using the Eqs. 50.4 and 50.5 below. The lower the value of
standard deviation indicates that the tool’s consistency in measuring the bandwidth
is better. In order to calculate the standard deviation, the mean value needs to be
obtained first.
Pn
k0 xk
l¼ ð50:4Þ
n
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn 2
k1 ðXk  lÞ
r¼ ð50:5Þ
n

Based on our findings on WMN environment without external traffic, we found

that ASSOLO recorded the lowest value at 0.0432. However, when comparing the
average value for all three settings, Pathrate were able to produce the lowest value
at 0.5162. Wbest has the highest value at an average of 2.8. When tested on WMN
environment with external traffic, we found that Pathchirp had the lowest value at
0.3656. Pathchirp also gave the lowest average value at 0.566 for all three settings.
ASSOLO and Wbest produce the highest value at 1.072 and 1.012 respectively.
In summary, we found that the effect of external traffic was great especially on
ASSOLO, IGI/PTR and Pathrate. Their readings were mostly overestimated. The
worst effect was on WBest, which recorded majority with zero readings. In the
other hand, Pathload and Pathrate were able to give a more consistence reading
throughout the testing as shown in Table 50.1. Figure 50.5 shows the consistency
comparison. The space constraint allows us to display only one graph for com-
parison of consistency between the tools. For a more complete set of results, the
readers may contact the lead author.
534 I.E. Kamarudin et al.

4.50
4.00
3.50
3.00
2.50
2.00
1.50
1.00
0.50
0.00
ASSOLO IGI/PTR PATHCHIRP PATHLOAD PATHRATE WBEST

One Access Point(AP) Two AP (Same Bandwidth)

Two AP (Different Bandwidth) One Access Point(AP) + FTP

Two AP (Same Bandwidth) + FTP Two AP (Different Bandwidth) + FTP

Fig. 50.5 Consistency comparison

50.5 Conclusion

In this paper, we have tackled the problem of estimating available bandwidth in

WMN environment. Several available bandwidth tools have been proposed pre-
viously but were lacking in certain testing environment. Some of the recent works
have proposed of using passive technique for estimation of bandwidth but what not
tested particularly on WMN network with the presents of external traffic and
varying bandwidth rate across the WMN. We implement our testing approach and
look at the performance in terms of accuracy, failure rate and consistency.
The results from our experiments suggest that Pathload is the most favorable
tool to estimate available bandwidth across WMN. This is because even when
external traffic is present in the WMN, Pathload was able to provide higher
accuracy, lower failure rate and consistence reading throughout our testing. It was
also observed that using any tools when tested over WMN with the presence of
external traffic and variable bandwidth causes further degradation in their
performance.

Acknowledgments This research was supported by Universiti Malaysia Pahang Research Grant
(RDU120388).
50 Experimental Analysis on Available Bandwidth Estimation Tools … 535

References

1. Gupta, D., et al.: Experimental comparison of bandwidth estimation tools for wireless mesh
networks. IEEE (2009)
2. Montesino-Pouzols, F.: Comparative analysis of active bandwidth estimation tools passive
and active network measurement. In: Barakat, C., Pratt, I. (eds.) pp. 175–184. Springer
Berlin, Heidelberg (2004)
3. Guerrero, C.D., Labrador, M.A.: On the applicability of available bandwidth estimation
techniques and tools. Comput. Commun. 33(1), 11–22 (2010)
4. Botta, A., Pescape, A., Ventre, G.: On the performance of bandwidth estimation tools. IEEE
(2005)
5. Johnsson, A., Bjorkman, M., Melander, B.: An analysis of active end-to-end bandwidth
measurements in wireless networks. IEEE (2006)
6. Goldoni, E., Rossi, G., Torelli, A.: Assolo, a new method for available bandwidth estimation.
pp. 130–136 (2009)
7. Kayange, Daniel S., Ramadhani Sinde, A.S.: available bandwidth estimation techniques
(ABETS) For An Efficient Telemedicine Content Transport Network. Int. J. Eng. Res.
Technol (IJERT), 2(7) (2013)
8. Kayange, Daniel S., Ramadhani Sinde, A.S.: Pathload for available bandwidth estimation
techniques (ABETs) for an efficient telemedicine content transport network. Int.
J. Advancements in Res. Technol. 2(8), 6 (2013)
9. Prasad, R., Dovrolis, C., G.I.o.T.: Bandwidth estimation: metrics, measurement techniques,
and tools. Network IEEE. 17(6), 27–35 (2003)
10. Mingzhe, L., Claypool, M., Kinicki. R.: WBest: a bandwidth estimation tool for IEEE 802.11
wireless networks. IEEE (2008)
11. Seide, R.: Capacity, coverage, and deployment considerations for IEEE 802.11 g. cisco
Systems white paper, San Jose (2003)
12. Prasad, R., et al.: Bandwidth estimation: metrics, measurement techniques, and tools. IEEE
Network 17(6), 27–35 (2003)
Chapter 51
A Survey of Petri Net Tools

Weng Jie Thong and M.A. Ameedeen

Abstract Petri net is a mathematical modeling language used to describe a system

graphically. It is a strong language that can be used to represent parallel or con-
current activities in a system. With a Petri net tool, users can view the overall system
graphically and edit it with the editor. A Petri net tool can also be used to analyze the
performance of the system, generate code, simulate the system and perform model
checking on it. A review on twenty Petri net tools in this paper will give the readers
an idea on what is a Petri net tool and the main functions of a Petri net tool. This
paper can serve as an introduction of twenty Petri net tools to the reader. However, to
date, there are many Petri net tools available to be downloaded online. This survey
paper aims to compare twenty Petri net tools in different aspect. This is important as
users will not have to read the reviews of Petri net tools online one by one. Just be
having a look at the discussion provided, readers can determined the best recom-
mended Petri net tools to be used based on their operating systems and the types of
Petri net tool to be analyzed. The main purpose of this survey paper is to recommend
Petri net tools based on the operating system and the types of Petri net to be analyzed.

Keywords Petri net
Petri net tools

51.1 Introduction

Petri net were introduced by Petri in 1962 [1]. Petri designed a sequence of
modules, with each module containing a single data element and communicating
with its two neighbors [2]. Petri net can be applied informally to any system that

W.J. Thong (&)
M.A. Ameedeen

Fakulti Komputer Sistem and Kejuruteraan Perisian, Universiti Malaysia Pahang,
Kuantan, Pahang, Malaysia
e-mail: [email protected]
M.A. Ameedeen
e-mail: [email protected]

Ó Springer International Publishing Switzerland 2015 537

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_51
538 W.J. Thong and M.A. Ameedeen

can be described graphically like flow charts and that needs some means of rep-
resenting parallel or concurrent activities [3]. Since Petri net can be applied in
most system to characterize it graphically, a lot of Petri net tools had been
developed for this purpose. Using Petri net tools, users can represent their system
in details and analyses the performance of the system. Users can also use the Petri
net tools as a graphical editor and code generator. Some Petri net tools can also be
used to simulate the system and provide model checking for it. To date, there are
many different types of Petri net tools for different environments and purposes.
However, there are no published papers on the recommendation of Petri net tools
to the users based on the users’ operating system and Petri net to be analyzed. This
survey paper aims to compare twenty different types of Petri net tools in different
aspects. At the end of the paper, a discussion will be drawn to recommend different
types of Petri net tools to users with different operating systems and depending on
what type of Petri net the users wanted to analyze.

51.2 Preliminaries

Petri net can be defined as a formal modeling language that can be represented
graphically with a strong mathematical foundation [4]. It is represented graphically
in the sense that it serves as a visual communication aid to model the system
behavior. It is based on a mathematical foundation in the sense that it represents
the equations, algebraic equations and algorithms in the system. Petri net are used
to model control flow in a system and is capable of modeling concurrency and
synchronization in distributed systems.
A Petri net consists of three types of components: places (circles), transitions
(rectangles) and arcs (arrows). Places represent different states of the system.
Transitions represent events or actions which cause the change of a state. An arc
connects a place with a transition or a transition with a place. Another element in
Petri net is the token. The movement of a token from place to place indicates a
change of state. The movement of a token is also known as firing.
Figure 51.1 represents an example of a Petri net with six places (states), four
transitions (change of states) and nine connecting arcs.

Fig. 51.1 Example of petri

net
51 A Survey of Petri Net Tools 539

The token is at s1 for starters. A token travels to the next state via transition. The
direction of the token’s movement is represented by the arrow head of the arcs.
The token can be fired into s2 via t1. The token is now present in s2 after leaving s1.
The firing continues with the token bring fired to s4 and s5 via t3. This is also known
as a concurrency or parallel relationship where one token is split into two
(depending on the number of concurrent nodes). After this process, both the s4 and
s5 states contain a token each. The s1 and s2 states are now without any tokens.
For another example, the token is s1 can also be fired into s3 state via t2. This
result in s3 contains a token and s1 being empty. The process continues with t4
firing the token from s3 into s6.
In 1995, Monika Trompedeller proposed a classification of Petri net based on a
survey by Bernardinello and De Cindio from 1992 [5]. The classifications have not
been updated since then but still it is useful for getting a quick overview of the main
differences between various kinds of Petri net. Petri net can be classified into 3 levels;
in level 1, Petri net is characterized by places (states) which can represent Boolean
value, for example, a place is marked by at most one unstructured token. Examples of
level 1 Petri net are Condition/Event (C/E) Systems, Elementary Net (EN) Systems
and 1-safe systems. An example of Petri net tool that supports analysis of level 1 Petri
net is the Environment for Action and State based Equivalences (EASE).
For level 2, Petri net is characterized by places (states) which can represent
integer values, for example, a place is marked by a number of unstructured tokens.
An example of level 2 Petri net is the Place/Transition (P/T) Nets. Petri net tools
that support analysis of level 2 Petri net include CPN/AMI, MOBY, POSES, PNS
and WINPETRI.
In level 3, Petri net is characterized by places (states) which can represent high-
level values, for example, a place is marked by a multi-set of structured tokens.
Examples of level 3 Petri net include High-Level Petri net with Abstract Data
Types (HL + ADT), Environment Relationship (ER) Nets, Traditional High-
Level Petri net and Well-Formed (Colored) Nets (WN). An example of Petri net
tools that support level 3 Petri net analyzing is the Cabernet which is used to
analyze real time systems based on Petri net augmented with data, predicates,
actions, and temporal information.
Judging by the classifications of Petri net and their complexity, different Petri
net analyzing tools can only support different level of Petri net. There is not any
best or worst tool in analyzing Petri net. Different categories of Petri net require
different Petri net tools to analyze. For example a level 2 P/T Petri net will require
tools such as CPN/AMI to analyze it instead of a level 3 Petri net tool such as
Cabernet.

51.3 Survey

In this paper, several Petri net tools are compared.

540 W.J. Thong and M.A. Ameedeen

51.3.1 AlPiNA

AlPiNA (Algebraic Petri net Analyzer) is a model checker for Algebraic Petri net
created by the SMV (Software Modeling and Verification) Group at the University
of Geneva [6]. AlPiNA is fully written in Java and it is available under the terms of
the GNU general public license. AlPiNA provides a user-friendly user interface
that was built with the latest metamodeling techniques on the eclipse platform. An
Algebraic Petri net is categorized as a level 3 High-Level Petri net with Abstract
Data Types (HL + ADT). Hence, AlPiNA is a Petri net tool which is capable of
analyzing a level 3 Petri net.

51.3.2 CoopnBuilder

CO-OPN stands for Concurrent Object-Oriented Petri net. CoopnBuilder is an

environment composed of a set of tools destinated to the support of concurrent
software development based on the CO-OPN language [7]. CoopnBuilder is also a
research project from the SMV Group. CO-OPN is an object-oriented specification
language based on synchronized Algebraic Petri net. CO-OPN allows the defini-
tion of active concurrent objects, and includes facilities for sub-typing, sub-
classing and genericity.

51.3.3 GreatSPN

GreatSPN2.0 is a software package for modeling, validation and performance

evaluation of a distributed system using Generalized Stochastic Petri net and their
colored extension: Stochastic Well-formed Nets [8]. It provides a friendly
framework to experiment with timed Petri net based modeling techniques.
GreatSPN implements analysis algorithms to allow its use on complex application.

51.3.4 LoLA

LoLA is a low level Petri net analyzer [9]. It is able to verify a broad variety of
behavioral properties on a given Petri net. It is based on explicit state space
verification, for example, an exhaustive enumeration of reachability state. In the
context of service-technology.org, LoLA can be used to verify compatibility and
for the validation of formal semantics.
51 A Survey of Petri Net Tools 541

51.3.5 PEP

PEP (Programming Environment based on Petri net) is a comprehensive set of

modeling, compilation, simulation and verification components, linked together
within a Tcl/Tk-based graphical user interface [10]. PEP’s modeling components
facilitate the design of parallel systems by parallel program, the compiler then
generate Petri net from such models. The PEP tool is considered as an open
platform. Further algorithms can be integrated in the user interface.

51.3.6 Snoopy

Snoopy is software tool to design and animate hierarchical graphs, among others
Petri net [11]. The tool has been developed and is still under development at the
University of Technology in Cottbus, Dep. of Computer Science, ‘‘Data Structures
and Software Dependability’’. The tool is used for the verification of technical
systems, typically software-based systems, as well as for the validation of natural
systems such as biochemical networks as metabolic, signal transduction etc.

51.3.7 Marcie

Marcie (Model Checking and Reachability analysis done efficiently) is a tool for
qualitative and quantitative analysis of a Generalized Stochastic Petri net with
extended arcs [12]. The tool consists of four engines with each engine carrying
their own unique function. It is possible to use Marcie with a graphical user
interface provided by the tool Charlie.

51.3.8 Charlie

Charlie is a software tool used to analyze a level 2 Petri net in particularly a Place/
Transition Net [13]. The tool has been developed and is still under development at
the University of Technology in Cottbus, Dep. of Computer Science, ‘‘Data
Structures and Software Dependability’’. The main features of the Petri net tool
include structural properties analysis, invariant based analysis, reachability graph
based analysis, reachability/coverability graph visualization using the JUNG
library and plugin support. Charlie is able to read Place/Transition nets which have
been created by the tool Snoopy, or P/T nets which are given in the Abstract Petri
net Notation and also P/T nets that are given in the INA file format.
542 W.J. Thong and M.A. Ameedeen

51.3.9 JSARP

JSARP (Simulator and Analyzer Petri net in Java) is a Petri net tool that describes
and verifies Petri net with the support of a graphical interface [14]. JSARP is
developed in Java and employs modern object oriented techniques and design pat-
terns. JSARP is able to edit the Petri net with graphics and also works as a simulator.

51.3.10 MIST

MIST is a safety checker for Petri net and other extensions [15]. It implements
several algorithms that solve the coverability problem for monotonic extensions of
Petri net. The tool implements several algorithms: forward/backward search of the
state using a symbolic data structure (the IST library). The tool also implements
abstraction-refinement techniques or efficient traversal techniques to tackle the
state explosion problem.

51.3.11 Petruchio

Petruchio is a tool used for computing Petri net translations of dynamic networks
modeled in terms of Pi-calculus processes [16]. It provides a mean to further
analyze nets. Petruchio is able to simulate GSPN (Generalized Stochastic Petri net)
which is a high level Petri net, check coverability for low-level Petri net and
application of reduction techniques for low-level Petri net.

51.3.12 PNEditor

PNEditor is an open-source Petri net editor [17]. It offers usual features of a

graphical editor for the design of Place/Transition nets. In addition, the tool offers
features like saving the net to a file, definition of roles, definition of subnets
(nested nets), saving of predefined subnets to files and their reuse as reusable
components, replacement of subnets, definition of static places etc. It requires
JAVA SE 6+ to operate.

51.3.13 Yasper

Yasper (Yet Another Smart Process Editor) is a tool for modeling and simulating
stepwise process [18]. It uses extended Petri net as its modeling technique. Yasper
offers easy editing, token gameplay and performance analysis with randomized
51 A Survey of Petri Net Tools 543

automatic simulation for basic Place/Transition nets with some extension,

including case-specific vs. inter-case token flow, reset and inhibitor arcs, decision
nodes with parameterized probabilities of alternatives etc.

51.3.14 PAPETRI

PAPETRI is a general and integrated environment for editing and analyzing Petri
net [19]. It allows users to work with difference classes of nets. Several analysis
tools are available for each of these classes. PAPETRI aims to provide a friendly
editing environment and to afford a greater deal of analysis tools for different
classes of Petri net.

51.3.15 Xpetri

XPetri is a graphical simulator of Petri net [20]. XPetri is an Xwindows application

designed to be portable across UNIX platform. It supports Place/Transition nets
with weighted arcs and a strict firing rule. It also allows a minimum and maximum
number of tokens to be specified for each place. Xpetri also supports temporary
disabling of transitions for a single fire or until re-enabled.

51.3.16 PROD

PROD is a reachability analysis tool for Predicate Transition Nets [21]. The
stubborn set method for reduced state space generation has been implemented in
PROD. PROD also has a rich query language for inspecting the generated state
space. It is available for download and free of charge.

51.3.17 ARP

ARP (Petri net Analyzer) is a software tool for Petri net analysis and simulation
developed by Maziero [22]. The interface is simple and the ARP tool accepts
Place/Transition Nets, Timed Nets and Extended Timed Nets. The features of ARP
include accessibility analysis, invariant analysis, equivalence analysis, perfor-
mance evaluation and manual simulation.
544 W.J. Thong and M.A. Ameedeen

51.3.18 JPetriNet

JPetriNet is software that is used to model, analysis conventional Petri net and to
simulate Timed Petri net [23]. The project is a Petri net modeling, analysis and
simulation tool made in Java Programming Language. The tool is created for
educational purpose and also to be used in any other purpose involving concurrent
system.

51.3.19 Petri .NET Simulator

Petri .NET Simulator is a tool used for modeling and simulation of Petri net and
analysis of their behavior [24]. It can be used to simulate flexible manufacturing
systems and also be used for discrete event system. The tool is free for download.

51.3.20 QPME

QPME (Queuing Petri net Modeling Environment) is an open-source tool for

stochastic modeling and analysis based on the QPN (Queuing Petri net) modeling
formalism [25]. Queuing Petri net is a combination of conventional queuing net-
works and stochastic Petri net which provides improved expressiveness and thus
making it possible to model systems at a higher degree of accuracy. QPME is
made of two components, which is QPE (QPN Editor) and SimQPN (Simulator for
QPN).

51.4 Discussion and Conclusion

In this section, the survey is tabulated in Table 51.1. Analysis is done based on five
criteria, which are Petri net supported, Component, Analysis, Environments and
whether it is Free of Charge.
The table above is the comparison between the Petri net tools surveyed in
Sect. 51.3 based on five main criteria. The first group of criteria is the Petri net
supported. In this category, the Petri net tools are compared in terms of what type
of Petri net is supported. Majority of the tools support Place/Transition Petri net
with some supporting high-level Petri net (i.e. AlPiNA, CoopnBuilder, PROD, and
QPME). However, QPME stands out in this category as it supports Queuing Petri
net (a combination of Queuing Network and Petri net). For Continuous Petri net,
only Snoopy and Charlie support it.
Table 51.1 Comparison between petri net tools
51

Petri net tool Petri net supported Component

High-level Object- Stochastic Petri nets Place/ Continuos Transfer Queueing Graphical State Condensed
petri nets oriented petri petri nets with time transition petri nets petri nets petri nets editor spaces state spaces
nets nets
AlPiNA 9 9 9 9
CoopnBuilder 9 9 9
GreatSPN 9 9 9 9 9 9
LoLA 9 9 9 9
A Survey of Petri Net Tools

PEP 9 9 9 9 9 9
SNOOPY 9 9 9 9 9
MARCIE 9 9
CHARLIE 9 9 9 9
JSARP 9 9
MIST 9 9 9
PETRUCHIO 9 9 9 9 9 9 9
PNEditor 9 9
Yasper 9 9 9 9
PAPETRI 9 9 9
Xpetri 9 9 9
PROD 9 9 9 9
ARP 9 9 9
JPetriNet 9 9 9
(continued)
545
Table 51.1 (continued)
546

Petri net tool Petri net supported Component

High-level Object- Stochastic Petri nets Place/ Continuos Transfer Queueing Graphical State Condensed
petri nets oriented petri petri nets with time transition petri nets petri nets petri nets editor spaces state spaces
nets nets
Petri .NET 9 9 9
Simulator
QPME 9 9 9 9 9

Component Analysis
Code Token Fast Place Transition Net Model Petri net Interchange Simple Structural Advance
generation game simulation invariants invariants reduction checking generator file format performance analysis performance
animation analysis analysis
9 9
9 9 9
9 9 9 9 9 9
9
9 9 9 9 9 9 9 9
9 9
9
9 9
9 9 9
9 9 9
9 9 9 9 9 9
9
(continued)
W.J. Thong and M.A. Ameedeen
 51

Table 51.1 (continued)

Component Analysis
Code Token Fast Place Transition Net Model Petri net Interchange Simple Structural Advance
generation game simulation invariants invariants reduction checking generator file format performance analysis performance
animation analysis analysis
9 9 9 9 9
9 9
9 9
9
A Survey of Petri Net Tools

9 9 9 9 9
9
9 9 9
9 9 9

Analysis Environments
Reachability graph based Invariant based Java Linux Sun HP, HP- Silicon graphics, MS Windows Macintosh UNIX Free of
analysis analysis UX IRIX DOS charge
9 9
9 9
9 9 9
9 9 9 9 9 9 9
9 9 9
9 9 9 9
9 9 9
(continued)
547
Table 51.1 (continued)
548

Analysis Environments
Reachability graph based Invariant based Java Linux Sun HP, HP- Silicon graphics, MS Windows Macintosh UNIX Free of
analysis analysis UX IRIX DOS charge
9 9 9 9 9 9
9 9
9 9
9 9 9 9 9 9 9
9 9
9 9
9 9
9 9 9
9 9 9 9 9 9
9 -9
9 9
9
9 9 9 9 9 9
W.J. Thong and M.A. Ameedeen
51 A Survey of Petri Net Tools 549

The second criteria of comparison are the components in each tool. Most of the
tools provide a graphic editor and a fast simulation on Petri net. Tools that provide
graphic editor and fast simulation on Petri net can be good teaching materials.
Users will be able to edit the Petri net and simulate different Petri net to aid them
in understanding Petri net. PEP has the highest number of components which
includes graphical editor, state spaces, condensed state spaces, token game ani-
mation, place invariants, transition invariants, net reduction, model checking, Petri
net generator and interchange file format. Users will be able to experience more on
PEP compared to the other tools.
The next category of comparison is the analysis of Petri net. Some of the tools
surveyed provide simple performance analysis, while tools such as GreatSPN,
PEP, Charlie, JSARP, Xpetri, ARP and JPetriNet provide structural analysis.
GreatSPN and QPME are also able to carry out advance performance analysis. The
reachability graph based analysis is however only able to be carried out by
MARCIE.
The next criterion used to compare between the Petri net tools is the envi-
ronment. LoLA, PETRUCHIO and QPME have the highest amount of environ-
ment supported for their tools with six environments for each of them. Tools like
AlPiNA, CoopnBuilder, MIST, Yasper, PAPETRI, ARP and Petri .NET Simulator
is very environment specific with each of them only supporting one specific
environment to be run on.
The final criterion of comparison is the pricing of the Petri net tools. All of the
Petri net tools are either free of charge or free of charge for academic purpose to be
downloaded.
In this section, Table 51.2 is presented to recommend Petri net tool for a few
scenarios which includes the users’ operating systems and the types of Petri net to
be analyzed.
Table 51.2 summarizes on the recommended Petri net tools for different
environments and different types of Petri net to be analyzed. Overall, users running
on the MAC OS X are recommended to use the tool PETRUCHIO which supports
most of the different types of Petri net. For Continuous Petri net, users will need to
use the tool SNOOPY; while for Queuing Petri net, users are recommended to use
the tool QPME. For Windows users, PETRUCHIO is recommended too as it
supports most of the Petri net types. However, analysis of Continuous Petri net and
Queueing Petri net requires users to use the tool CHARLIE and QPME respec-
tively. For those running on Linux, there isn’t any specific tool overall that can
analyze most of the different types of Petri net. For High-level Petri net, Petri net
with time and Place/Transition Petri net, users are recommended to use the tool
PEP. For Stochastic Petri net, users are recommended to use GreatSPN or PE-
TRUCHIO. CHARLIE is recommended for the analysis of Continuous Petri net
while QPME is recommended for the analysis of Queueing Petri net.
Overall, the PEP tool offers the most components and analysis types amongst
the twenty Petri net tools compared. However, due to the lack of supporting
environment, PEP is not user friendly. Users will need to have Linux or Sun
operating system to support the PEP tool. Where else PETRUCHIO supports most
550 W.J. Thong and M.A. Ameedeen

Table 51.2 Recommended petri net tools for different usage

Operating system Type of petri nets Recommended tool
MAC OS X High-level petri nets PETRUCHIO
Object-oriented petri nets Not supported
Stochastic petri nets PETRUCHIO
Petri nets with time PETRUCHIO
Place/transition petri nets PETRUCHIO
Continuous petri nets SNOOPY
Transfer petri nets PETRUCHIO
Queueing petri nets QPME
Windows High-level petri nets PETRUCHIO
Object-oriented petri nets Not supported
Stochastic petri nets PETRUCHIO
Petri nets with time PETRUCHIO
Place/transition petri nets PETRUCHIO
Continuous petri nets CHARLIE
Transfer petri nets PETRUCHIO
Queueing petri nets QPME
Linux High-level petri nets PEP
Object-oriented petri nets Not supported
Stochastic petri nets GreatSPN/PETRUCHIO
Petri nets with time PEP
Place/transition petri nets PEP
Continuous petri nets CHARLIE
Transfer petri nets PETRUCHIO
Queueing petri nets QPME

of the major environments (MAC OS X, Windows, Linux), it is also the tool with
the second highest component and analysis types. Based on Table 51.2, each
operating system has at least two types of Petri net that is able to be analyzed by
PETRUCHIO. From this deduction, we can conclude that PETRUCHIO is the best
tool as it supports most of the operating system and it has an adequate amount of
features in it.

References

1. Petri, C.A.: Communication with automata (1966)

2. Brauer, W., Reisig, W.: Carl adam petri and ’’petri nets’’. Fundam. Concepts Comput.
Sci. 3(5), 129 (2009)
3. Murata, T.: Petri nets: properties, analysis and applications. Proc. IEEE 77(4), 541–580
(1989)
51 A Survey of Petri Net Tools 551

4. Ameedeen, M.A.: A model driven approach to analysis and synthesis of sequence diagrams.
Diss. University of Birmingham (2012)
5. Trompedeller, M.: A Classification of Petri Nets (1995)
6. Hostettler, S., Marechal, A., Linard, A., Risoldi, M., Buchs, D.: High-level petri net model
checking with AlPiNA. Fundam. Inf. 113(3), 229–264 (2011)
7. Al-Shabibi, A., Buchs, D., Buffo, M., Chachkov, S., Chen, A., Hurzeler, D.: Prototyping
object oriented specifications. In: Applications and Theory of Petri Nets 2003, pp. 473–482.
Springer, Heidelberg (2003)
8. Chiola, G., Franceschinis, G., Gaeta, R., Ribaudo, M.: GreatSPN 1.7: graphical editor and
analyzer for timed and stochastic petri nets. Perform. Eval. 24(1), 47–68 (1995)
9. Schmidt, K.: Lola a low level analyser. In: Application and Theory of Petri Nets 2000,
pp. 465–474. Springer, Berlin (2000)
10. Grahlmann, B., Best, E.: PEP—more than a petri net tool. In: Tools and Algorithms for the
Construction and Analysis of Systems, pp. 397–401. Springer, Berlin (1996)
11. Heiner, M., Herajy, M., Liu, F., Rohr, C., Schwarick, M.: Snoopy–a unifying petri net tool. In
Application and Theory of Petri Nets, pp. 398–407. Springer, Berlin (2012)
12. Schwarick, M., Heiner, M., Rohr, C.: Marcie-model checking and reachability analysis done
efficiently. In: Proceedings of the IEEE, 8th International Conference on Quantitative
Evaluation of Systems (QEST), pp. 91–100 (2011)
13. Wegener, J., Schwarick, M., Heiner, M.: A plugin system for charlie. In: Proceedings of the
CS&P, pp. 531–554 (2011)
14. Oliveira Lino, F.G., Sztajnberg, A.: JSARP: Simulator and analyzer petri net in java, a final
project of the undergraduate course of computer and information technology, University of
Rio de Janeiro (2006) (Unpublished manuscript)
15. Ganty, P., Van Begin, L., Delzanno, G., Raskin, J-F.: Coverability checkers included in mist.
Online, URL: https://github.com/pierreganty/mist/wiki
16. Meyer, R., Strazny, T.: Petruchio: From dynamic networks to nets. In: Computer Aided
Verification, pp. 175–179. Springer, Berlin (2010)
17. Riesz, M., Seckár, M., Juhás, G.: PetriFlow: A petri net based framework for modelling and
control of workflow processes. In: Proceedings of the ACSD/Petri Nets Workshops,
pp. 191–205 (2010)
18. van Hee, K., Oanea, O., Post, R., Somers, L., van der Werf, J.M.: Yasper: a tool for workflow
modeling and analysis. In: Proceedings of the IEEE, 6th International Conference on
Application of Concurrency to System Design (ACSD 2006), pp. 279–282 (2006)
19. Berthelot, G., Johnen, C., Petrucci, L.: PAPETRI: environment for the analysis of petri nets.
In: Computer-Aided Verification, pp. 13–22. Springer, Berlin (1991)
20. Geist, R., Crane, D., Daniel, S., Suggs, D.: Systems modeling with xpetri. In: Proceedings of
the 26th Conference on Winter simulation, Society for Computer Simulation International,
pp. 611–618 (1994)
21. Varpaaniemi, K., Heljanko, K., Lilius, J.: Prod 3.2: an advanced tool for efficient reachability
analysis. In: Computer Aided Verification, pp. 472–475. Springer, Berlin (1997)
22. Maziero, C.A.: ARP: Petri net analyzer. Control and Microinformatic Lab, Federal
University, Santa Catarina (1990)
23. Azevedo, M.: JPetriNet. Sapucai Valley University, Brazil (2004) Online, URL: http://
jpetrinet.sourceforge.net/
24. Genter, G.: Petri .NET simulator. Online, URL: http://www.petrinetsimulator.com/
25. Kounev, S., Dutz, C.: QPME: a performance modeling tool based on queueing petri nets.
ACM SIGMETRICS Perform. Eval. Rev. 36(4), 46–51 (2009)
Chapter 52
Towards a Exceptional Distributed
Database Model for Multi DBMS

Mohammad Hasan Ali and Mohd Azlishah Othman

Abstract This paper reported on the current issues distributed transaction faces in
the multi database management system. We looked at how the distributed trans-
action in database differs from other distributed processing. After highlighting the
major issues in distributing transition in the multi DBMS, concurrency control and
recovery such as (Site failure, network failure, time failure, and so on) were
considered in the design of our model. We proposed an exceptional distributed
transaction model for processing the transaction queries in the multi DBMS by
keeping in mind the point-to-point transaction processes. The expected results
were described in this paper.

Keywords Distributed transaction
DBMS
Transaction network

52.1 Introduction

Generally, distributed transaction defined as a transaction that updates data on two

or more networked computer systems. However, different conceptual structure for
processing the transactions elements are extend for fitting a certain needs for the
applications that must update distributed data in term of peer to peer connection.
Developing a transaction model is difficult because most of these techniques are
customized to be occurring in a single transaction point attach with a multiple
failures such as (Client’s failure, server failure, and the network connection).

M.H. Ali (&)
M.A. Othman

Centre for Telecommunication Research and Innovation, Faculty of Electronic and
Computer Engineering, Universiti Teknikal Malaysia Melaka, Hang Tuah Jaya,
76100 Durian Tunggal, Melaka, Malaysia
e-mail: [email protected]
M.A. Othman
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 553

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_52
554 M.H. Ali and M.A. Othman

Furthermore, connecting different distributed points require an advance network

security technology to achieve the best performance for detecting and recovering
these failures over a network connection [1].
Recent transaction applications are optimizing various security integrations for
securing the transaction between client and server that detect and manage the
change action (such as a database update) before the transaction can occur. The
transaction process over the network can be determined a complicated process
because of the multi security issues in these networks. The performance of
transaction between single and multi database management systems is coordinate
to other network parts for the components involved and acts as a transaction
manager for each computer that manages transactions.
Algorithms were developed to serve a certain needs, which begin used in
banking, investment environments, enterprise applications, etc. [1]. During the
communication between computers a transaction manager sends prepare, commit,
and abort messages to all its subordinate transaction managers [2]. In addition, the
common steps for obtaining a distributed transaction are:
• The client starts a transaction by calling the transaction manager. This appli-
cation allows for the transaction to be deployed.
• While the transaction manager processing the client request, an update will be
saved for the deployed transaction to commit the process based on the client
details.
• After processing the client request, the transaction manager keeps a sequential
transaction log so that it’s committing or abort decisions will be durable.
– If all components are prepared,
– If any component cannot prepare,
– While a component remains prepared but not committed or aborted.

52.1.1 Transaction Problems

During the database transaction over multi database management systems, prob-
lems were addressed while dealing in the concurrency control and recovery, most
of these problems occurred in the distributed databases [3]. However, the main
problems can be simply classified into the following:
• Site’s failure: this action happened when one or more sites in a DDBMS fail.
This type of failure will corroborate with the application manager for restoring
the transaction updates,
• Network Problems: this action occurred during the communication network
fails for unclear transaction, this type of network failure indicates one or more
sites to be cut off from the rest of the site in the distributed database,
52 Towards a Exceptional Distributed Database Model for Multi DBMS 555

• Data Duplication: this problem occurred when multiple copies of the database
are unable to be tracked accordingly for maintaining consistency,
• Distributed Transaction: this problem arises when a distributed transaction
happened in multi sites,
• Distributed Deadlocks: during the database transaction deadlocks could be
launched in any single or multi accessing for the site contents.
An additional various problems been reported in the transaction management
among the distributed environments is ensuring the consistency of the data in the
presence of site and network failure. However, various transaction protocols were
developed for providing a stable transaction over a network. Moreover, different
problems were addressed in securing the transaction elements over a network

52.1.2 Related Works

The importance of database transaction has brought the needs to develop and adopt
new transaction technologies. A study by Sami and Habib [4] explored and
developed a new transaction mechanism based on the using of agent systems for
providing a high transferring reliable among different distributed algorithms.
However, the study developed a transaction algorithm for solving DCSP issues
based on the agent systems by classifying it into different groups: Variables’ and
Controller Agents, these two groups helps to reformulating of the detected agent
communication algorithm. The proposed model can be used also in order to treat
non-binary constraints and managing the transaction according to its original
destination. The instantiation of these variables can be done by negotiation in order
to separate the sub problems into totally independent ones. Figure 52.1 presents
the proposed classification based on the agent systems [4].
However, another study by Qiming and Umesh [5] described the transaction
issues in the E-Commerce applications. The study customized the computing
environment for distributing a limit transaction through the database management
systems within a large number of the autonomous service requester. The study
suggested the using of transaction agents that cooperate to perform business
transaction activities. The study also reported the transaction requirements among
these e-commerce applications that require to provide a peer-to-peer protocols
based on distributed communication.
Hence, Qiming and Umesh developed cooperative multi-agent transaction
model that includes peer-to-peer protocols for committing control and failure
recovery. This model could help to facilitate the urgent tasks for transferring the
data elements through channel query. Figure 52.2 shows the proposed model
structure with the database components [5].
556 M.H. Ali and M.A. Othman

Fig. 52.1 Constraint networks based agent [4]

Fig. 52.2 The proposed model structure with the database components [5]
52 Towards a Exceptional Distributed Database Model for Multi DBMS 557

52.1.3 SQL Distributed Transaction Example

The current distributed computing development has generated a new retrieving and
representing techniques for the data elements over the grid. SQL presents one of
these techniques which obtain a high level processing of client queries for a set of
processes that seek to achieve some form of cooperation in the database. More-
over, executing a different conceptual representation simplify the processes of a
distributed system for performing some actions by crashing or being disconnected.
SQL server provides a sustainable database processing which starts with the
first SQL statement from beginning the transaction. This transaction carries out a
different meaning of the grid contents when it is committed or rolled back, the
main statements that SQL starts with are COMMIT or ROLLBACK statement.
The next step is to extract the transaction concept by considering a certain
procedure for obtaining a successful transaction along with the database elements.
Example, when one or many customers transfer a certain amount into other
accounts, it will be directed to the saving section to the checking account.
There are important operations for obtaining that transaction such as:
• Decrement the amount of the savings account.
• Add in other elements of the checking account.
• Record the transaction steps by saving its actions.
SQL database allows a huge number of transactions to be performed when all
the SQL operations can be applied to maintain the accounts with the adopted
balance. Meanwhile, in case of any leaks of the transaction such as transaction
invalid number, software and hardware failure, which leads to end the transaction
without processing the amounts into another account?
Figure 52.3 descries the transaction mode in the SQL database.
This kind of transferring can be provided by the internet services or wireless
technology (Fig. 52.4).
It is desirable to start a new development for adopting a new mechanism to
facilitate the distributed transaction process over database management systems
which require specific requirements. Below are presented some recommendation
for achieving a successful transaction over SQL database components.
• To allow a stable transaction for a certain elements over SQL statements, SQL
transaction statements must initially agree to the client request by classifying
the transaction details into several points with common formats for repro-
cessing the transactions.
• Obtaining a successful data exchanging can be easily achieved by determining
the require actions for the unknown transactions into the grid with several
alternative plans of analyzing.
• Providing a collaborative transaction can also be accepted for starting a new
distributed transaction based on the SQL statements for the cooperating
processes.
558 M.H. Ali and M.A. Othman

Fig. 52.3 The transaction mode in the SQL database

Fig. 52.4 SQL transaction statement’s performance

• This kind of transactions my require not only to agree on the requested actions,
but also its requiring to execute on the transaction query by which these actions
need to be executed among the database.
52 Towards a Exceptional Distributed Database Model for Multi DBMS 559

Deployment
Consumer manager
Transaction
balancer

Caching proxy Application

cluster cluster
internet

Application
database

Policy server
Data
Provider

Load balancer
Firewall

Data base
Registry database
Web Services Gateway
cluster

KEY
Internal flows
Consumer-Mediator flows LOAP- UDDI cluster
Mediator-Supplier flows

Fig. 52.5 Proposed exceptional distributed database model

52.1.4 Proposed Technique

Based on the problems that have been discussed above in the distributed trans-
action applications, and based on the related and previous works, we proposed an
exceptional distributed transaction over multi database management systems. This
model provides a sustainable communication between client and server to perform
the client queries in more secure and reliable way. However, the proposed model
could reduce the number of errors occurred during the transaction process, which
optimize a multi transaction point among servers. Additionally, this model adopted
an advance network communication for reducing the network failure which occurs
560 M.H. Ali and M.A. Othman

during the transaction. From the Fig. 52.5 can see that the distributed communi-
cation indirectly depends on the database query which gives high privileges for the
transactions.

52.1.5 Expected Result

The expected result of performing the proposed model can be summarized into the
following:
• Addressed the problem of transaction directly by looking into the history of
transactions,
• Discovering dependencies among services from historical transactions,
• Execution time period data can be determined for multi transactions among
DBMS,
• Monitoring instrumentation over multi DBMS.

52.2 Conclusion

This study illustrated the major issues in processing transactions in multi database
management systems along with the techniques for processing databases in dif-
ferent fragments. Moreover, this study adopted a new way for resolving the site
and network failure occurred during the database transaction in a multi DBMS.
The study proposed an exceptional distributed transaction model for performing
and managing the transaction details more efficiently. We have considered the
related studies in developing the proposed model.

References

1. Gupta, M., Neogi, A., Agarwal, M.K., Kar, G.: Discovering dynamic dependencies in
enterprise environments for problem determination. In: Proceedings of the 14th IFIP/IEEE
International Workshop on Distributed Systems: Operations and Management. Lecture Notes
in Computer Science, vol. 2867, pp. 221–232. Springer, Oct 2003
2. Bessiere, C., Brito, I.: Asynchronous Backtracking without Adding Links: A New Member in
the ABT Family. Artif. Intell. 161(1–2), 7–24 (2005)
3. Al-Maqtari, S., Abdulrab, H., Nosary, A.: Constraint programming and multi-agent system
mixing approach for agricultural decision support system. In: Emergent Properties in Natural
and Artificial Dynamical Systems, pp. 199–213 (2006)
4. Sami, M., Habib, A.: Controller-agent based approach for solving distributed constraint
problem. LITIS Lab—INSA of Rouen, France (2006)
5. Qiming, C., Umesh, D.: Multi-Agent Cooperative Transactions for E-Commerce, vol 2,
pp. 33–37, USA (2007)
Chapter 53
Semantic Search Engine Using Natural
Language Processing

Sudhakar Pandiarajan, V.M. Yazhmozhi and P. Praveen kumar

Abstract The World Wide Web has become colossal and its growth is also
dynamic. Most of the people rely on the search engines to retrieve and share
information from various resources. All the results returned by search engines are
not always relevant as it is retrieved from heterogeneous data sources. Moreover a
naive user finds it difficult to confirm that the retrieved results are significant to the
user query. Therefore semantic web plays a major role in interpreting the rele-
vancy of search results. In this work, a novel algorithm is proposed for retrieving
relevant documents using semantic web based on the concept of Natural Language
processing (NLP). In this proposed system, NLP is used to analyse the user query
in terms of Parts of Speech. The extracted terms are compared to the domain
dictionary to identify the relevant domain of the user interest. On the other hand,
the retrieved documents of the user query are investigated with the help Natural
language processing to identify the relevant domain. Now the documents are
ranked as per the relevancy of the contents against user query. The experimental
result of the proposed algorithm indicates that the accuracy of the retrieved doc-
ument is 97 %.

Keywords Semantic search engine  Natural language processing

S. Pandiarajan (&)  V.M. Yazhmozhi  P. Praveen kumar

Kamaraj College of Engineering and Technology, Virudhunagar 626101, India
e-mail: [email protected]
V.M. Yazhmozhi
e-mail: [email protected]
P. Praveen kumar
e-mail: [email protected]

Ó Springer International Publishing Switzerland 2015 561

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_53
562 S. Pandiarajan et al.

53.1 Introduction

With the increasing use of World Wide Web as an essential source of information,
there is a need to work with the Semantic Web [1], in order to reduce the irrelevant
data obtained during the search. The existing search engines [2], [3] do not provide
domain specific search and they simply perform keyword matching. Those search
engines cannot understand the negative senses, Example: ‘‘I do not want cluster-
ing’’. The result set of the existing search engines for such a query would be
related to clustering as they merely perform keyword matching and don’t analyse
the actual meaning of the user query. Other example of queries with negative
senses includes ‘‘I want clustering algorithms except cobweb’’. The pro-posed
approach using semantic search algorithm overcomes all the above stated pitfalls.
Semantic Web provides the users a comfort zone and reduces the wastage of
time. The proposed work on Semantic search is accomplished by POS (Parts Of
Speech) tagging using Natural Language Processing. Using POS Tagging, the
proposed semantic search algorithm can understand what the user query conveys
and hence it provides more relevant results to the user. The query entered by the
user is POS tagged using Stanford Parser, and the tags for each word in the user
query are obtained. For each tag obtained, if it is a noun it is added to the noun list
(NL), and all other tags correspondingly. Each word in the NL is now compared
with the word dictionary of each document that is obtained during pre-processing
the document. If a match is found, then the word weight is incremented, and is
added to the word weight list (WWL). The documents are sorted based on the
WWL and added to the result set. Similarly, each sentence in the document is
detected using OpenNLP, and its weight is incremented if all the nouns in the NL
are matched. This weight of each sentence is added to the sentence weight list
(SWL). The documents are then sorted based on the SWL and added to the result
set. If there is any occurrence of negative word (e.g., not, except, NEITHER-NOR),
then all the nouns in the NL are skipped and compared with the word dictionary of
each document if verbs of possession occurs before those negative words. These
documents are also added to the result set. The ranked documents from the result set
are retrieved to the user. By this approach of Semantic search [4], [5], [6], the search
results that are more relevant to the user’s interest are provided. In the proposed
work, pdf, word and html documents have been considered for analysis.

53.1.1 Outline of the Paper

Section 53.2 presents the various works on semantic web and natural language
processing supportive to the proposed research. Section 53.3 describes the
Architectural design of the proposed scheme. Section 53.4 illustrates the experi-
mental results and Performance Evaluation. Section 53.5 depicts the conclusion
and future work.
53 Semantic Search Engine Using Natural Language Processing 563

53.2 Related Works

In [7, 8], Mukhopadhyay et.al has ogy which is made effective by mapping the
instances and the classes. Even thought the results of this research prove that the
performance are good than the regular search engine, the results are proved within
the domain only. In [9], Cafarella et al. developed a search engine using natural
language processing in which it out performs well in terms of producing relevant
information using natural language processing. In [10], Karpagam had proposed a
framework which is based on ontology to build the semantic search engine. The
author finds the relevant document for the user query using the techniques like
word stemming, ontology matching, weight assignment, rank calculation. If the
approach is extended to a larger data set, the weight assignment and rank calcu-
lation will become tedious. In [11], Jiang et al. developed a semantic search engine
that overcomes the problem of knowledge overhead by a query interface. In [12],
Lei et al. proposed a semantic web portal that has been designed to ensure the
quality of the extracted metadata and it also facilitates for data querying. In [13],
Kruse et al. had used WordNet, a lexical database to find the word senses in order
to achieve semantic. But WordNet does not provide a classification of word senses
in technical terms. In [14], Lara et al. had proposed a hybrid searching technology
which is the combination of ontology and traditional keyword based matching in
which the drawbacks of keyword based search like stop words removal in the user
query was reflected. In [15], Madhu had presented a survey on the search engine’s
generation and role of the search engines in the intelligent web in which it insists
the necessity to build semantic search engines. In [16], Kalaivani had proposed a
question answering system based on the semantic searching terminology and
natural language processing technique. In [11], Jiang et al. developed a full text
search engine has been designed to exploit ontological knowledge for document
retrieval. In [17], Kerschberg et al. recommended a methodology has been
developed to capture the semantics of the users search intent into target search
queries of the existing search engine.

53.3 Architectural Design

In the proposed work, Structured and Unstructured documents are maintained

separately. Whenever a user query arises, it is given to Stanford Parser for POS
tagging. Based on POS Tagging [18] NN (Noun, Singular), NNS (Noun, Plural),
NNP (Proper Noun, Singular), JJ (Adjective) and IN (Preposition) are extracted
from user query and stored in a term list table. Decisions are taken based on the
accompanying diagram. If the user query belongs to only one domain then all
564 S. Pandiarajan et al.

Fig. 53.1 Domain classification based on term list

the documents in the repository are compared against the domain dictionary and
the results are computed. On the other hand, if the user query belongs to more than
one domain then the dominant domain is computed based on Fig. 53.1. There are 2
possible cases can occur in selecting dominant domain

Case 1: If Most of the Words belong to one domain then the same domain is
taken as dominant domain.
Case 2: If Equal number of Matches found with two domains then the choice are
given to the user to confirm the dominant domain.

Once the dominant domain is selected, each document is split into sentences
and each sentence is further divided into words. Each word is compared with term
list and domain dictionary for matching. If a match is found then the ‘‘sentence
weight’’ is incremented. The same process is continued for the whole document
and cumulative document weight is calculated. If the user query contains negative
words, then a negative ag is set to the user query and the process will request the
53 Semantic Search Engine Using Natural Language Processing 565

Fig. 53.2 Architectural design of the proposed approach

user to re-enter the direct query for better retrieval of results. In this way the
accuracy of the retrieval is brought out. Once all the documents weight is calcu-
lated the highest weight of the document is ranked first and given to user as a good
match in the retrieved results (Fig. 53.2).
566 S. Pandiarajan et al.

Algorithm 1 - POS Tagging using NLP

INPUT : User query
OUTPUT : Resultant Document Set
METHOD : POS Tagging IN NLP

Step 1 : Initialize the Resultant document set RES = { }

Step 2 : Initialize nounlist NL= { }
Step 3 : Initialize sentencelist SL={ }
Step 4 : Initialize negativelist NGL= { }
Step 5 : Initialize NEGLIST= { not, no,neither, nor, except... }
Step 6 : Initialize neg ag=0, word weight=0, sentence weight=0;
Step 7 : POS Tag the user query using stanford parser.
Step 8 : foreach tag obtained
Step 8a : If tag is NN, NNS, NNP, NNPS, JJ, VBG then add the appropriate word in
the query to NL
Step 8b: If tag is RB,DT,CC and the word is in NEGLIST then set neg ag=1
Step 9 : If NL.count=1 and PL.count=0 and neg ag=0 then nd the concept under which
the single noun in NL occurs in Domain Dictionary.
Step 10 : If the single noun occurs under more than one concept, display all concepts.
Step 11 : As per the user selection, add that selected concept too in NL.
Step 12 : Detect sentences in the web documents using OpenNLP and store in SL
Step 13 : foreach sentence sent in SL
Step 14 : If all nouns in NL occur within a sentence
Step 15 : Increment weight for each match.
Step 16 : Store sentence weight of each document into SentenceWeightList SWL Step
17 : Sort the documents according to sentence weight and store it in the RES.
Step 18 : Compare each entry in the NL with word dictionary of each document
Step 18a: If a match is found then increment word weight with the count attribute's
value.
Step 19 : Store word weight of each document into WordWeightList WWL.
Step 20 : Sort the documents according to word weight and if that document is not in
RES store it in RES
Step 21 : If neg ag=1 then skip all nouns in NL and compare with the word dictionary of
each document
Step 21a : Store such documents into RES
Step 22 : Retrieve the ranked documents fD1,.Dng in RES to the user

53.4 Experimental Results and Discussion

The experiment was conducted using American national corpus (anc) [19, 20] by
in-creasing the number of documents in the repository gradually. The main factor
considered for evaluation is accuracy rather than the speed. Three different
scenarios are considered in evaluating the performance of the proposed system.
53 Semantic Search Engine Using Natural Language Processing 567

A confusion matrix is created with various combinations of Relevant documents

(RD) and Conflicting documents (CD) to analyse the system. Performance eval-
uation of the proposed approach is done based on the classification context sce-
nario. Precision, Recall, Accuracy and F-measure are the major measures used for
classification based performance. Precision is the probability that measure a
retrieved document is relevant to the context. Precision is calculated based on the
formula

TP
Precision ¼
TP þ FP

where
TP is true positive (Correctly retrieved)
TN is true negative (Correctly rejected)
FP is false positive (Incorrectly retrieved)
FN is false negative (Incorrectly rejected).

Recall is the probability that calculates a relevant document is retrieved in a

search process. Recall is calculated based on the formula

TP
Recall ¼
TP þ FN

Accuracy is calculated based on the formula

TP þ TN
Accuracy ¼
TP þ TN þ FP þ FN

First scenario creates a large number of Relevant Documents against the small
number of Conflicting documents. The Second scenario creates an equal number
of Relevant and Conflicting documents. The Third scenario creates a small number
of Relevant Documents against with large number of Conflicting documents. All
the three cases, the document set contents are increased gradually (Tables 53.1,
53.2, 53.3).
The accuracy and recall of the system with respect to the above stated 3 cases is
plotted in Figs. 53.3 and 53.4 respectively.
All the performance measures of the system clearly highlights that the results
produced by the system in terms of all aspects are better compared with existing
systems.
568 S. Pandiarajan et al.

Table 53.1 Relevant documents are higher than the conflicting documents
No. of documents Accuracy F-measure Recall Precision
100 0.978 0.965 0.974 0.962
200 0.975 0.963 0.972 0.96
300 0.974 0.964 0.973 0.961
400 0.974 0.962 0.971 0.957
500 0.97 0.963 0.973 0.961
600 0.967 0.962 0.974 0.964
700 0.966 0.96 0.972 0.961
800 0.968 0.962 0.973 0.962
900 0.97 0.96 0.972 0.961
1000 0.968 0.952 0.965 0.95
1500 0.964 0.958 0.962 0.951
2000 0.962 0.956 0.96 0.95
3000 0.963 0.956 0.961 0.951
5000 0.962 0.955 0.959 0.95

Table 53.2 Relevant documents are equal to the conflicting documents

No. of documents Accuracy F-measure Recall Precision
100 0.975 0.963 0.972 0.96
200 0.974 0.961 0.97 0.958
300 0.972 0.963 0.967 0.951
400 0.974 0.96 0.962 0.955
500 0.972 0.962 0.963 0.952
600 0.97 0.963 0.965 0.952
700 0.965 0.961 0.962 0.949
800 0.968 0.962 0.964 0.95
900 0.964 0.96 0.96 0.948
1000 0.964 0.952 0.962 0.95
1500 0.962 0.95 0.96 0.949
2000 0.96 0.955 0.958 0.948
3000 0.96 0.952 0.955 0.95
5000 0.961 0.951 0.954 0.95
53 Semantic Search Engine Using Natural Language Processing 569

Table 53.3 Relevant documents are smaller than the conflicting documents
No. of documents Accuracy F-measure Recall Precision
100 0.972 0.958 0.965 0.954
200 0.97 0.957 0.967 0.953
300 0.968 0.953 0.962 0.95
400 0.973 0.952 0.961 0.952
500 0.97 0.954 0.964 0.952
600 0.965 0.951 0.96 0.95
700 0.964 0.948 0.961 0.951
800 0.965 0.95 0.962 0.953
900 0.965 0.948 0.962 0.951
1000 0.96 0.952 0.958 0.95
1500 0.961 0.947 0.956 0.945
2000 0.958 0.952 0.957 0.946
3000 0.959 0.955 0.958 0.95
5000 0.96 0.954 0.958 0.951

Fig. 53.3 Shows that the

accuracy of the proposed
systems is maintained
between 96 and 98%
irrespective of the number of
documents
570 S. Pandiarajan et al.

Fig. 53.4 Presents the recall

measure of the proposed
system against various
numbers of document sets.
Irrespective of the documents
combinations the proposed
system outperforms well in
between 95.5 and 97.5%

53.5 Conclusion

In this study a new approach is proposed based on Natural language processing to

understand and classify the documents based on the user query. The experimental
results of the proposed systems point out that the accuracy of this system is vary in
between 95 and 97 % in terms of relevancy. However the time taken to classify a
document is little high compared with existing search engines. In future, our
system will address the existing drawbacks to compete with other search engines
in terms of the time factor.

References

1. http://swoogle.umbc.edu/
2. http://hakia.com/
3. https://duckduckgo.com/
4. Yu, L.: A Developers Guide to the Semantic Web. Springer, Berlin (2011). ISBN:
9783642159695
5. Antoniou, G., Groth, P., van Harmelen, F., Hoekstra, R.: A Semantic Web Primer, 3rd edn.
ISBN: 0262018284
6. Segaran, T., Taylor, J., Evans, C., O’Reilly (2009) Programming the Semantic Web (2009).
ISBN: 9780596802066
7. Mukhopadhyay, D., Banik, A., Mukherjee, S., Bhat-tacharya, J., Kim, Y.-C.: A Domain
Specific Ontology Based Semantic Web Search Engine. In: Proceedings of the 7th
International Workshop (MSPT), p. 8189, Feb 5 2007. ISSN 1975-5635, 89-8801-90-0
8. FinKelstein, L., Gabrilovich, E., Matias, Y.: Placing search in context: the concept revisited.
In: Proceedings of the 10th International Conference on World Wide Web, pp. 406–414
(2001)
9. Cafarella, M.J., Etzioni, O.: A search engine for natural language processing. In: Proceedings
of the International World Wide Web Conference Committee (IW3C2), ACM 1595930469/
05/0005 (2005)
53 Semantic Search Engine Using Natural Language Processing 571

10. Karpagam, G.R., Uma Maheswari, J.: A conceptual framework for ontology based
information retrieval. Int. J. Eng. Sci. Technol. 2(10), 5679–5688 (2010)
11. Jiang X., Tan, A.-H.: OntoSearch: a full-text search engine for the semantic web
12. Lei, Y., Uren, V., Motta, E.: SemSearch: a search engine for the semantic web. In:
Proceedings of the International Conference on Knowledge Engineering and Knowledge
Management Managing Knowledge in a World of Networks, pp. 238–245. Springer, Berlin
(2006)
13. Kruse, P.M., Naujoks, A., Rsner, D., Kunze, M.: Clever search: a WordNet based wrapper for
internet search engines, computing research repository CORR, vol. Abs/cs/050 (2005)
14. Lara, R., Han, S.-K., Lausen, H., Stollberg, M., Ding, Y., Fensel, D.: An evaluation of
semantic web portals. In: Proceedings of the IADIS Applied Computing International
Conference 2004, Lisabon, Mar 23–26 2004
15. Madhu, G., Govardhan, A., Rajinikanth, T.V.: Intelligent semantic web search engines: a
brief survey. Int. J. Web Semant. Technol. 2(1), 34–42 (2011)
16. Kalaivani, S., Duraiswamy, K.: Personalized semantic search based intelligent question
answering system using semantic web and domain ontology. In: Proceedings of the
International Conference on Advanced Computer Technology (IJCA), pp. 15–17 (2011)
17. Kerschberg, L., Kim, W., Scime, A.: Intelligent web search via personalizable Meta—search
agents
18. Stanford Parser: http://nlp.stanford.edu/software/index.shtml
19. http://www.anc.org/data/masc/downloads/data-download/
20. Lei, Y., Lopez, V., Motta, E.: An infrastructure for building se-mantic web portals. In:
Proceedings of the International Workshop on Web Information Systems Modeling (WISM),
pp. 283–308 (2006)
Chapter 54
Integration of Mobile Based Learning
Model Through Augmented Reality Book
by Incorporating Students Attention
Elements

Zarwina Yusoff, Halina Mohamed Dahlan and Norris Syed Abdullah

Abstract The limitation of current e-learning technology has caused a lack of

student attention in educational environment. Therefore, this study describes the
integration of mobile based learning through Augmented Reality Environment to
incorporate the student attention elements by computer-generated content. To
incorporate student attention element, this study propose an integration model of
mobile learning by utilizing Augmented Reality Environment. To validate the
integration model, this study has developed the AR prototype called AF-LAR
(Animal Fun Learning—Augmented Reality) through smart phone technology as
mobile based learning device. (AF-LAR) is an AR Book that adapts learning
concept via mobile devices and enables student to bring and access the learning
content anywhere and anyplace. AF-LAR has been developed using Metaio and
Junaio channel while the design is incorporated with visual learner styles from
Visual Auditory Kinesthetic (VAK) model with information visualization
approach. The result of this study was evaluated by using attention element in
Keller’s Motivation ARCS Model to prove the student attention attribute from the
prototype based on the five experts. Descriptive statistics was chosen as a tech-
nique to evaluate the mean and reliability of attention attribute; perceptual arousal
(PA), inquiry arousal (IA) and variability (V) inside the proposed integration
model to achieve the student attention in education environment.

Z. Yusoff (&)
H.M. Dahlan
N.S. Abdullah

Department of Information System, Faculty of Computing, Universiti Teknologi Malaysia,
Skudai, 81310 Johor Bahru, Malaysia
e-mail: [email protected]
H.M. Dahlan
e-mail: [email protected]
N.S. Abdullah
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 573

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_54
574 Z. Yusoff et al.

Keywords Augmented reality
Education
Mobile learning
Student attention

54.1 Introduction

As far as the progress of e-learning technology as a medium in learning process,

many researchers try to improve the learning process by utilizing technology in
particular, towards in enhancing students attention. Today, many mobile devices
such as cell phones, notebooks and tablet computer are gaining more relevance to
learning environments and education. The recent progress of Augmented Reality
(AR) has brought this technology to the mobile computing area to generate more
user experience on it. Hence, further facilitates the growth of user experience with
AR environment. Furthermore, the emerging of technological concepts in educa-
tion was generating the rapid development of education where it was bringing it
into mobile based learning technology. AR is an effective ways to visualize the
learning content, researchers was proven that AR supported particular learning
activities such as problem solving, in a highly interactive and memorable fashion
[1]. This emerging technology can also be used in any fields to facilitate user in
visualizing the object or information in any particular situations.
The main objective of this paper is to embed AR into mobile learning to
improve the learning technology in classroom using the AR. In order to actualize
this aim, there are other objectives that need to be achieved which are:
i. To propose the model integration of mobile learning through Augmented
Reality Book by incorporating student attention elements.
ii. To evaluate the prototypes of mobile based learning application through
Augmented Reality Book.
Apart from this objective, we were chosen the concept of AR Book to imple-
menting the model integration to be tested in enhancing the student attention.

54.2 Augmented Reality Book for Education

The implementation of AR for mobile learning purposes has been discussed and
also presented in a number of studies. In order to achieve the objective, this paper
is searching on current model integration of mobile learning to enhance the lim-
itation through achieving the objective in this study. The visualization in AR must
be embedded in mobile learning architecture to ensure the mobile learning through
AR are fluently can use in learning environment to improve the learning process.
54 Integration of Mobile Based Learning Model … 575

In this research, the process of embedding of educational elements in the mobile

AR for learning system derived from AR Magic Books that was develop by
Billinghurst et al. [2] in enhance the traditional normal text books through Aug-
mented Reality Concept [2].
The integration of AR through mobile also successfully was done by Ramdas
et al. [3] which was developed the AR Eco System where the concepts of modules
are based on four categories; player, context awareness, type of learning and mode
of interaction [3]. To get the effective ways in enhancing learning experience
among learners; we proposed that, the integration of AR in mobile learning will be
implementing through AR Book where this will enhance the concept of Magic
Book that proposed by [2].
In order to actualize the objective, image based tracking techniques is chosen to
make the mobile phones camera can track the virtual object through the AR book.
Besides that, Oh and Woo [4] was developing the AR Gardening System that
implements the personalized pedagogical agent through the animated pedagogical
agents to improve student’s learning experience in educational system [4]. The
gardening system that consist the one garden and include flower, blue bird, scoop
and sprinkle water to learn the process of growing flower in garden. Each of the
objects is representing by a marker to make the tracking process.

54.3 Methodologies

To facilitate the evaluation process, the model was realized in a form of mobile
based learning application. The mobile based learning application incorporates all
the components of the models. In ensure this research successfully conducted, this
section briefly explain the participants, data collection and data analysis that
support the evaluation process.

54.3.1 Participants

Five participants of experts from five criteria were chosen to get the analysis in the
limitation of current e-learning system in educational environment which con-
ducted through questionnaire. The questionnaire is design to identify the limitation
of current e learning technology in educational environment to achieve student
attention during learning process which the criteria in the questionnaire will be
map with the model integration and will be evaluated through the prototypes
implementation. This questionnaire is distributed into five experts as a targeted
respondent to identify the problem and limitation in the usage of technological
element in learning environment in order to improve student attention.
576 Z. Yusoff et al.

Table 54.1 Finding for data collection

Limitation of Problem analysis Factors to achieve student
element in e-learning attention through mobile
technology [5] learning model [6]
Technological and 1. The limitation of traditional Technology
hardware limitation method in learning process
2. How the technological tools can
help educational environment
3. The current limitation of e-learning
courseware technology
4. The limitation of current
educational application software
Design limitation 1. The appropriate combination color Pedagogy
to attract student attention
2. The effective and efficient ways in
conveying learning content through
visualization approach
3. The visualization approach to
maintain the student attention in
classroom
4. The suitable subject to be adapting
in implementing learning through
mobile
Personal issues 1. Factor that causes student lack of People
attention in learning and teaching
process.
2. The suitable approach to improve
student attention in a classroom
3. The factor of mobile based
learning should to be implemented

54.3.2 Data Collection

In this research study, the data collection about the current limitation of e-learning
technology was get from the five experts through the questionnaire. The Question
for the data collection is based on the three main limitations in e-learning tech-
nology that identify Mahanta [5] which consist by technology and hardware
limitation, design issues and personal issues [5]. Based on that, eleven question
was derives from the factor to analysis the current limitation of e-learning tech-
nology. Table 54.1 listed the data collection and analysis.
54 Integration of Mobile Based Learning Model … 577

Fig. 54.1 Integration model of mobile learning based on augmented reality

54.3.3 Data Analysis

Statistical Package for Social Science (SPSS) was used to measure the mean
resulted from the evaluation. The descriptive statistics was used to measure the
prototypes in improving the student attention.
The process development of this prototype is start with designing the system
flowchart where the design consists of two parts, the Book interface and the AR
system. Analyzing the user is the first step in the development of AF-LAR Book
(Fig. 54.1).

54.4 Findings

The conceptual of mobile learning model is derived from Prasertsilp [6] where
mobile learning was based on two factors, mobile learning environment and
learning outcomes [6]. This research identified the model to be integrated with
mobile learning system so that it can contribute in improving the limitation of
current mobile learning system because it covers the three aspects of mobile
learning environment; users, technology and pedagogy to be implementing
through AR. Figure 54.2 presented the model integration.
Mobile Learning dimension consists of three main attributes; technology,
people and pedagogical elements while educational dimension consists of type of
578 Z. Yusoff et al.

Tracking Registration Display

Fig. 54.2 Processes in augmented reality technology

Fig. 54.3 The cover of

AF-LAR

learning and mode of interaction. The model of mobile learning is derives from
Pratersilp [6], where it consist of from three main factors; technology, pedagogy
and people and the attributes inside of mobile learning are derives from Laouris
and Eteokleous [7] which consists of space (s), time (t), method (m), prior
knowledge (k), attention and learning environment (LE) [6, 7].
People are the target of user who will use the mobile learning application which
consists by student in any related subject to enhance their learning experiences
during class. The learning environment (LE) is a people which conducting the
learning session successfully. In these factors, it consists by learner as a student
who wants to gain their knowledge, using the AR Books to adding their infor-
mation. Besides that, people in learning environment (LE) is also consists by
teacher in give instruction to students for completing their task. Based on the
figure, the attribute of student attention is support by element attention in ARCS
model to evaluate the model integration.
Variability is used to evaluate the model integration through the AR Book
prototypes. In variability aspects, it emphasized the variety tactics to maintain the
attention through the mobile technology, teacher’s instruction, and integrated of
learning content.
In order to emphasize the model integration, Vision based tracking was used to
visualize the AR object. The main of Augmented Reality process are consists of
three main phases; tracking, registration and display the AR content.
54 Integration of Mobile Based Learning Model … 579

Fig. 54.4 Image based marker

Fig. 54.5 Interactive animation and video

Fig. 54.6 Interactive website about animation

Fig. 54.7 Animal song through video

580 Z. Yusoff et al.

Table 54.2 AF-LAR design

Animal fun learning book Results
(a) Book cover that shows the adaptation of visual learner concept to generate Fig. 54.3
student interest in explore the animal world through the features below:
i. The colorful graphics
ii. The enjoyable concept through the images and pictures
iii. The text styles is suitable for primary schools student

Based on the model integration, the mobile learning attributes are strongly
support the student attention when realizing it through prototypes implementation.
The embedding of educational elements in the mobile AR for learning system
derived from Ramdas et al. [3] developed the AR Eco System where the concepts
of modules are based on four categories; player, context awareness, type of
learning and mode of interaction [3]. To get the effective ways in enhancing
learning experience among learners; we proposed that, the integration of AR in
mobile learning will be implementing through AR Book. The book shows it dis-
playing information visualization through the AR Book, where its focus on
visualization and interactivity to improve student learning experience in educa-
tional systems. In order to actualize this research objective, the concept of AR
Book was chosen to embed the model integration and to validate the conceptual
framework of mobile through AR in support student attention to perform suc-
cessfully. Based on that, this study was proposed the AR Book through mobile
display that called AF-LAR (Animal Fun Learning through AR) that embedding
with visual learner styles for designing and visualization approach through Aug-
mented Reality Technology in order to improve student attention. Table 54.2
shows the results of AF-LAR design.
Based on the Table 54.2, the concept of designing AF-LAR is very interesting
to get student attention in order to provide platform to tracking the AR object
based on image based marker in this book through mobile. In aligning with pri-
mary schools characteristics, the concept of thematic stories and short stories are
used to be adapt with the level of student in reading the information about the
animal, in this case the concept of people in mobile learning model are emphasize
which is, student must have their teacher or facilitator to help them in learning
session. In order to construct the AR element to be integrated with the mobile,
Table 54.3 illustrates the concept of Mobile AR and the explanation results in AF-
LAR in achieve student attention.
The process of evaluation is implemented in objective to achieve the model
integration that supports student attention through AF-LAR. This evaluation
process has implemented in targeted respondent, which is consist by five experts
that have some criteria using Samsung Galaxy Tab 3 10.1 and AF-LAR Tangible
book as a marker based tracking. The evaluation process is based on the Attention
element that derives from ARCS Motivation Model that support by Keller 2000.
Table 54.4 presented the explanation of attention criteria that used to be evaluated
in AF-LAR.
54 Integration of Mobile Based Learning Model … 581

Table 54.3 Displaying learning content through mobile AR

Mobile AR Results
Image based marker to display an AR object using mobile camera detection Fig. 54.4
(a) The picture background is based on jungle situations to make student feel more
real with the animal world
(b) During image based marker displays 3D object, the video about the animal Fig. 54.5
presented
(c) Visual mode presents the interactive 3D graphics and auditory mode present the
video
(d) Learning through interactive web based also provides in showing learner the Fig. 54.6
animal website with the real environment
(e) Learning through animal song, more interactive and attract student attention to Fig. 54.7
sing together through the lyrics

Table 54.4 Explanation about attention criteria

Attention Explanation
Capture interest (Perceptual What I can do to capture their interest?
arousal)
Stimulate inquiry (Inquiry arousal) How I can stimulate attitude of inquiry?
Maintain attention (Variability) How I can use a variety of tactics to maintain the
attention?

Based on the perceptual arousal, we can capture the student interest through the
colorful images, picture, graphic element in the book, and usage of mobile and AR
as a part of technology. To stimulate of an attitude of inquiry by learners expe-
riences, it need the basic knowledge about the learning information and to achieve
the variability elements, it can implemented through the mobile technology,
teachers instructions and integrated of learning content. In order to full fill the
element of attention in Table 54.4, we identify eleven questions that representative
as Perceptual Arousal (PA), Inquiry Arousal (IA) and Variability (V) to evaluate
the prototypes. In achieve the research contribution; Table 54.5 below indicates
the details of descriptive statistic for the prototype testing results. Based on that
table, descriptive statistics on mean was performed to describe the level of student
attention on five experts’ participation.
Perceptual Arousal (PAI) in Table 54.5 represents ‘‘visualization based
approach is the best method in enhance student attention’’ has a spread of 0.447
over 4.80 mean with the maximum score is 5. Perceptual Arousal (PA2) in
Table 54.5 represents ‘‘Smartphone’s is the flexible ways to be used in educational
environments’’ has a spread of 0.548 over 4.40 mean with the maximum score is 5.
Perceptual Arousal (PA3) in Table 54.5 represents ‘‘The features of AR that
allows multimedia element such as graphic, audio, video, and interactivity can able
to attract student attention during learning session’’ has a spread of 0.548 over 4.60
mean with the maximum score is 5. Perceptual Arousal (PA4) in Table 54.5
represents ‘‘The usage of multimedia element can help the student attention in
582 Z. Yusoff et al.

Table 54.5 Results of prototype testing

Item Likert scale points
Strongly Disagree Neutral Agree Strongly Mean Standard
disagree agree deviation
PA 1 4 4.80 0.447
1
PA 3 2 4.40 0.548
2
PA 2 3 4.60 0.548
3
PA 2 3 4.60 0.548
4
IA 1 1 4 4.80 0.447
IA 2 2 3 4.80 0.447
IA 3 1 2 2 4.20 0.837
V1 2 3 4.60 0.548
V2 2 3 4.60 0.548
V3 1 1 3 4.40 0.894
V4 1 4 4.80 0.447

mobile learning’’ has a spread of 0.548 over 4.60 mean with the maximum score is
5. Based on the results of Perceptual Arousal, all the mean is [2.5 that indicates
the element of perceptual arousal in the proposed model is strongly emphasized
the technology element is support student attention attributes.
The attributes; are spaces (s), real time (t) and AR (m) a strong combination
technology that used in enhances student attention. Spaces are provided through
mobile devices and AR is a method that applying objects visualization in the
learning content. Apart from that, real time (t) is represent by technology of AR
and the prototypes is mobile is run in a real time to providing the spaces for
students to conduct their learning activity. In AR, the element of technology also
integrates with vision based tracking technique that conducted to produce AR
content in the mobile devices. Based on the results, it proves the model integration
of mobile learning and AR through technological element is strongly emphasized
can improve student attention.
Inquiry Arousal (IA1) in Table 54.5 represents ‘‘The lack of student attention in
mobile learning can be avoid through visualization approach’’ has a spread of
0.447 over 4.80 mean with the maximum score is 5. Inquiry Arousal (IA2) in
Table 54.5 represents ‘‘The concept in Augmented Reality Technology is very
appropriate to be adapted in Science Subject’’ has a spread of 0.447 over 4.80
mean with the maximum score is 5. Inquiry Arousal (IA3) in Table 54.5 represents
‘‘Visualization can give easier to student for remember the learning content
because the conveying of object and information graphically to student’’ has a
spread of 0.837 over 4.20 mean with the maximum score is 5.
54 Integration of Mobile Based Learning Model … 583

Based on the results of Inquiry Arousal (IA), all the mean is [2.5 that indicates
the entire element in inquiry arousal are support of attention element in peda-
gogical of mobile learning model. The attributes of student attention in peda-
gogical element are prior knowledge (p) and attention (a) is achieved to support
the proposed model integration that proposed in Chap. 5. Apart from that, the
result proves the element of attention attributes is embedding in the prototypes
successfully to achieve student attention.
Variety of tactics 1 (V1) in Table 54.5 represents ‘‘The features of Smart-
phone’s that ubiquitous, easy to bring anywhere and can be access the internet is
the main factor to encourage learning through mobile’’ has a spread of 0.548 over
4.60 mean with the maximum score is 5. Variety of tactics 2 (V2) in Table 54.5
represents ‘‘Learning through mobile is the new methods which must to introduce
and monitored by teachers’’. Has a spread of 0.548 over 4.60 mean with the
maximum score is 5. Variety of tactics 3 (V3) in Table 54.5 represents ‘‘Mobile
technologies can help teachers in get student attention’’ has a spread of 0.894 over
4.40 mean with the maximum score is 5.
Variety of tactics 4 (V4) in Table 54.5 represents ‘‘Augmented Reality tech-
nology is very appropriate to be integrated with mobile technologies’’ has a spread
of 0.447 over 4.80 mean with the maximum score is 5.
Based on the variability results, it strongly supports the people element in the
model integration because all the variability analysis is [2.5. The people element
in model integration are consists by teacher, learner and syllabus (c) as a learning
environment (LE). Variability result was proves the people element has a strong
relation with learning concept through visualization approach in Augmented
Reality.
Learner and teacher is interdependencies attributes to ensure the learning ses-
sion is fluency. Besides that, syllabus is needed to be embedding into prototypes
development. Prototypes evaluation is implemented as the ends phase in the
research methodology and to measure the model validation by incorporate student
attention elements. To validate the accuracy of data, the calculations of mean and
standard deviation are used to extract the results in order to evaluate the prototypes
and the Cronbach’s Alpha also is implemented to found the reliability statistics
among the question. As a conclusion, the mean of the prototypes is [2.50, that
means the prototypes can improve student attention through the model integration
of mobile learning through Augmented Reality.

54.5 Conclusion

This paper solved limitation of current e-learning technology by improving student

attention in educational environment. This paper proposed the integration of
mobile based learning through Augmented Reality Environment by incorporating
the student attention elements. The objective of this paper was achieved when the
elements of student attention that embedded in the AR Book can improve student
584 Z. Yusoff et al.

attention based on expert perspective. The content of AR Book that display

through mobile devices was proves the new model integration can be used in order
to improve student attention in educational environment. Based on the finding, it
was proven that the utilizing of technology can enhance the student attention
through the AR Book concept. However, the mobile devices used must have the
internet connection to ensure the right channel can be loaded to visualize the
learning content. For the future work, a robust AR application is proposed in
displaying the AR content during the tracking process.

References

1. Luckin, R., Stanton Fraser, D.: Limitless or pointless? an evaluation of augmented reality
technology in the school and home. Int. J. Technol. Enhanced Learn. 3(5), 510–524 (2011)
2. Billinghurst, M., Kato, H., Poupyrev, I.: The magic book: a transitional AR interface. Comput.
Graph. 25(5), 745–753 (2001)
3. Ramdas, C.V., Parimal, N., Utkrash, M., Sumit, S., Ramya, K., Smitha, B.P.: Application of
sensors in augmented reality based interactive learning environments. In: Proceedings of the
IEEE , 6th International Conference on Sensing Technology (ICST) (2012)
4. Seijin, O.H. Woontack, W.O.O, Augmented Garderning System with Personalized Pedagog-
ical Agents.International Symposium on Ubiqitous Virtual Reality (ISUVR), CEUR-WS
Proceedings, 62, 17–18, (2007)
5. Mahanta, D., Ahmed, M.: E-learning objectives, methodologies, tools and its limitation. 1,
46–51 (2012)
6. Prasertsilp, P.: Mobile learning: designing a socio-technical model to empower learning in
higher education mobile learning: designing a socio-technical model to empower learning in
higher education. 2(1) (2013)
7. Laouris Y.: Eteokleous, N.: We need an educationally relevant definition of mobile learning.
In: Proceedings of the 4th World Conference on Mobile Learning, mLearn, Cape Town, South
Africa, Oct 25–28 2005
8. Keller J.M.: How to Integrate learner motivation planning into lesson planning.The ARCS
model approach.Paper presented at VIISemanario, Santiago, Cuba (2000)
Chapter 55
Law Reckoner for Indian Judiciary:
An Android Application for Retrieving
Law Information Using Data Mining
Methods

S. Poonkuzhali, R. Kishore Kumar and Ciddarth Viswanathan

Abstract Data mining is an emanating area not only in the field of computer
science but also contributes more to knowledge mining and information man-
agement. This analysis mainly focuses on organizing the Indian judicial system
which enables the lawyers and law students to access the constitutions details,
code for civil and criminal procedure in an easier and faster way. In this work
android is used as a medium for this analysis because smart phones are being used
by 9 persons out of 10 in the legal society. The user of this application can search
using keywords and all sections from the constitution relating to the keyword are
displayed along with review of related procedures either civil procedure or
criminal procedure as mentioned above. The major advantage of this application is
that it reduces the time for the lawyers and law students to search using android
phones rather than to do survey on books and materials related to their case. In this
proposed work all the legal data are classified using ID3 classification algorithm
which results in efficient searching.


Keywords Classification Civil procedure  Criminal procedure  Indian con-

stitution Keyword search

S. Poonkuzhali (&)  R. Kishore Kumar  C. Viswanathan

Rajalakshmi Engineering College, 602105 Chennai, India
e-mail: [email protected]
R. Kishore Kumar
e-mail: [email protected]
C. Viswanathan
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 585

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_55
586 S. Poonkuzhali et al.

55.1 Introduction

Data mining, the extraction of concealed analytical information from large

databases, is a powerful new technology with great potential to focus on the
most important information in their data warehouses. Data mining tools predict
future trends and behaviors which help to make proactive, knowledge-driven
decisions. Data mining techniques can be implemented rapidly on existing
software and hardware platforms to enhance the value of existing information
resources. Knowledge management (KM) is an effort to increase useful knowl-
edge within the organization.
Though vast researches are being carried out in the area of knowledge man-
agement, yet it requires a wider acceptance from people from various fields for
data/pattern analysis. As most of the judicial data contains large number of
attributes, data mining techniques such as feature relevance analysis and classi-
fication are required to perform efficient searches. Among the data mining meth-
ods, classification plays a major role in analyzing the clinical data. Some of the
frequently used classifiers in this domain are ID3, Naive Bayes, C4.5, CART,
Multi layer perceptron, KNN, J48, Random forest and Support Vector Machines.
As the Indian judiciary is the largest democracy of the world, this research is
carried on the attributes of the Indian constitution and more significantly in the
civil and criminal procedures. Moreover it carries out many critical functions such
as classifying the Indian judiciary based on civil and criminal domains. The most
common judicial terms are the organization of the state, how a state and unions are
formed, what is the legal procedure for civil cases and criminal cases.

55.2 Related Works

Silwattananusarn and Tuamsuk [6] presented about data mining and knowledge
management process. The Knowledge management rationale and major knowl-
edge management tools integrated in the knowledge management cycle are
described. Finally the applications of the data mining techniques are summarized.
The limitations in this paper are it just provides generalization of the knowledge
management process and does not provide any specific details and moreover data
integration is not addressed. Kumar Varma [4] presented the use of data mining
concepts to match data from multiple sources in order to enrich the data and
improve its quality on the whole. It is also used for the Integration of the databases
on-line negotiations and interactions. The major drawback on this system is that it
causes bottlenecks in distributed environment. Srivatsan et al. [7] introduced how
to use the contents of a shopping cart to predict what else the customer is likely to
55 Law Reckoner for Indian Judiciary: An Android Application … 587

buy. Association rule mining, boolean vectors and prediction techniques are used
in this paper. The major drawbacks of this system are a lot of complex algorithms
are used which may lead to loss in efficiency and speed. And also it predicts based
on the existing items on the cart and does not adapt dynamically. Mohamed et al.
[5] explained, record Linkage is the computation of the associations among
records of multiple databases. In a framework where entities are unwilling to share
data with each other, the problem of carrying linkage computation without full
data exchange is called private record linkages. This paper provides improvements
when compared with previous techniques such that (i) they make no use of a third
party and (ii) they achieve much better performance in terms of execution time and
quality of output. The major drawbacks of this system are it cannot work in
distributed systems and it does not use creative techniques for evaluation. Kishore
Kumar et al. [3] uses various data mining techniques to classify spam dataset.

55.3 Methods and Materials

In this proposed system an android application named Law Reckoner was

developed to classify and search Law Information quickly. An Android mobile
service is made available for the user to log into the server. The information
provided by the end user travels through the network to the server. Then the server
validates the request and sends back the response to the android mobile. If the
person is a registered user it allows the person to search through the constitutions
or else it shows a error message and ask the user to register. In registration phase
the user needs to submit user name, password and other details to the server. All
information is stored in the database via server for future purpose. Then the user
will start to search for the law details via choosing the category that they want to
search. Once they choose the category, they have to provide the other information
like chapter number, article number and part number and submit the request to the
server. The general architecture of the proposed work is shown in Fig. 55.1.
This is implemented by creating an application using android and installing the
application in the user Smart phone. By using this application the user can com-
municate with the server via GPRS connectivity as in Fig. 55.2. Once the user
sends the request to the server, the server will validates the request and display the
law details in the users Android Smart phone as shown in Fig. 55.3. Once the
server receives the user’s request, it will search for the requested data in the law
database and classify them using decision tree induction (ID3) algorithm and
display the relevant data to the user’s Smart phone itself. ID3 algorithm builds the
tree in a top down manner starting from a set of objects and a specification of
588 S. Poonkuzhali et al.

properties. At each node of the tree, a property is tested and the results used to
partition the object set using information gain as a splitting criteria based on
entropy measure. This process is recursively done till it contains objects belonging
to the same category. ID3 can handle high-cardinality predictor variables. ID3
doesn’t apply any pruning procedure. It doesn’t handle numeric attributes or
missing values. Figures 55.4, 55.5, 55.6, 55.7 and 55.8 shows the results of this
application.

Fig. 55.1 Architecture of the proposed system

Fig. 55.2 User registration

55 Law Reckoner for Indian Judiciary: An Android Application … 589

Fig. 55.3 Main menu to choose the category

Fig. 55.4 Sub menu to constitution

590 S. Poonkuzhali et al.

Fig. 55.5 Search menu of the proposed system

Fig. 55.6 On click the main menu chapters will be listed

55 Law Reckoner for Indian Judiciary: An Android Application … 591

Fig. 55.7 List of Punishment under criminal category

Fig. 55.8 Punishment under criminal category

592 S. Poonkuzhali et al.

55.4 Pseudo for ID3 Classification Algorithm

Input: Law data.

Output: Decision Rules.
Step 1: Extract the input training dataset (TD).
Step 2: Let IA be input attributes & TA be the Target Attributes.
Step 3: For a base class construct the tree as follows:
Step 3A: If TD is empty then
Return a single node with the value Failure
Step 3B: If all instance from TD have the same class value for TA
then
Return the single node with that class value.
Step 3C: If IA is empty then
Return a single node with the most frequent value of TA in
TD.
Else proceed to step 4
Step 4: Compute the information gain for all IA relative to TD based on
Entropy measure.

X
m
InfoðDÞ ¼  ðpi logðpi ÞÞ ð55:1Þ
i¼1

X
v   
EðAÞ ¼  Dj  jDj  InfoðDÞ ð55:2Þ
j¼1

Where |Dj|/|D| is weight of the jth partitionGainðAÞ ¼ Info ðD  EðAÞÞ

pi is the probability that arbitrary instance in TD
E(A) is the entropy of the attribute A
Gain(A) is the information gain of attribute A
Info(D) expected information needed to classify an instance in TD
Step 5: Let X be the split attribute with largest information gain among IA
Step 5A: let Xj|j = 1, 2….m be the value of X.
Step 5B: Let Tj|j = 1, 2….m be the subsets of T when T is partitioned.
Step 6: Partition the TD according to the value of X
Step 7: Return trees with the root node labeled X and arcs labeled X1, X2….Xm
Step 8: Repeat step 4 to 6 until it contains instants belonging to the same
category.
55 Law Reckoner for Indian Judiciary: An Android Application … 593

55.5 Conclusion

This research work presents an efficient application for lawyers and law students
across the country which gives them mobile access for law related data such as
Indian constitution, code of civil procedure and code of criminal procedure. In
addition to information retrieval, an efficient search mechanism is incorporated
with which they can perform precise and accurate searches using article number,
section number, chapter number, part number etc. Future works can be carried out
for:
• To provide more content like the Indian penal code and a database of Indian
origin.
• To create a law dictionary to search the meanings of certain legal terms
• To reveal a case history for similar occurrences of the articles and its actions.

References

1. Gupta, G.K.: Introduction to Data Mining with Case Studies. Prentice-Hall India, New Delhi
(2006)
2. Han, J., Kamber, M.: Data Mining: Concepts and Techniques. Morgan Kaufmann Publishers,
San Jose (2001)
3. Kishore Kumar, R., Poonkuzhali, G., Sudhakar, P.: Comparative study on email spam classifier
using data mining techniques. In: Proceeding of the International Multiconference on
Engineers and computer scientists, Vol I (2012)
4. Kumar Varma, M.V.K.: Efficient techniques for online record linkage. Int. J. Comput. Trends
Technol. 3(3), 321–324 (2012)
5. Mohamed, Y., Mikhail, J.A., Ahmed, E.: Efficient Private Record Linkages. Proceedings of the
2009 IEEE International Conference on Data Engineering, IEEE Computer Society, Vol. 4,
pp. 1283–1286 (2009)
6. Silwattananusarn, T., Tuamsuk, K.: Data mining and its applications for knowledge
management: a literature review from 2007 to 2012. Int. J. Data Min. Knowl. Management
Process. 2(5), 13–24 (2012)
7. Srivatsan, M., Sunil Kumar, M., Vijaya Shankar, V., Leela Rani, P.: Predicting missing items
in shopping carts using fast algorithm. Int. J. Comput. Appl. 21(5), 35–41 (2011)
Chapter 56
Enhancing the Efficiency of Software
Reliability by Detection and Elimination
of Software Failures Through Univariate
Outlier Mining

S. Poonkuzhali, R. Kishore Kumar and R. Kumar

Abstract In this competitive market, there is always a growing demand for

providing high quality software in all aspects of societal services: education,
industry, defense, travel, health, retail, telecommunications and so on. As the size
and complexity of software grows drastically, the demand for highly complex
software systems also increases rapidly which in turn raises the number of soft-
ware failures. These software failures not only degrade the business performance
but also cause more economic damages. Therefore, detection and elimination of
these software failures becomes a vital importance for better software reliability
which is a measure of quality. In this proposed work, identification of software
failures is done through outlier mining which mainly focuses on rare and infre-
quent patterns like faults, noise, irrelevant and redundant data. After the removal
of these software defects, classification algorithms are applied to provide a better
and more reliable result.

  
Keywords Classifiers Error rate Defects Outlier mining Quality Software 
reliability

56.1 Introduction

In recent years, more number of software was used for decision making process in
various fields. As software becomes voluminous, sophisticated in complexity, and
originated by integration of multiple components, it is an increasingly challenging

S. Poonkuzhali (&)  R. Kishore Kumar  R. Kumar

Rajalakshmi Engineering College, Chennai 602105, India
e-mail: [email protected]
R. Kishore Kumar
e-mail: [email protected]
R. Kumar
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 595

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_56
596 S. Poonkuzhali et al.

task to ensure software reliability. Software Reliability is the application of sta-

tistical techniques to data collected during system development and operation to
specify, predict, estimate, and assess the reliability of software-based systems.
From a knowledge perspective, the analysis of executions of a buggy program is
essentially a data mining process which traces the data generated during program
executions containing both relevant patterns and outliers that may help the dis-
covery of software bugs. In general, the design for reliability techniques includes:
fault avoidance, fault detection, masking redundancy, and dynamic redundancy.
An efficient data mining technique can improve the software reliability by ana-
lyzing, predicting and removing software faults in advance.
Data mining is a process of extracting hidden and useful information from the
data and the knowledge discovered by data mining is previously unknown,
potentially useful, valid and of high quality. Finding outliers is an important task in
data mining. Outlier detection as a branch of data mining has many important
applications and deserves more attention from data mining community [1]. Early
detection of outliers not only reduces the risk of making poor decisions based on
erroneous data but also aids in identifying, preventing, and repairing the effects of
malicious or faulty behavior. Outliers are observations that deviate so much from
other observations to arouse suspicions that they might have been generated using
a different mechanism. Outliers may also reflect the true properties of data from
rare and interesting events which may contain more valuable information than
normal data. Outlier mining is dedicated to find data objects which differ signif-
icantly from the rest of data.
In the proposed system Univariate outlier detection technique is used to detect
and remove the sample that contains software defects. The classification algo-
rithms are applied to the original dataset as well as after applying outlier detection
techniques. Then the error rate of various classifiers is compared. From the
accuracy of various classifiers, it is proved that the classification results after
applying outlier detection technique gives quality software products which in turn
improves software reliability.

56.2 Related Works

56.2.1 Software Reliability

Go et al. [2] have proposed that there is a need to model an approach that is
capable of considering the architecture and evaluating the reliability by taking into
account the interaction between components, reliability of components and reli-
ability of interfaces with other components. Xie et al. [3] have proposed practical
method to predict the software reliability using test data. Two major parameters
are used to evaluate reliability while testing one is total number of initial faults and
fault detection rate in testing. Hence the software reliability of the new system
56 Enhancing the Efficiency of Software Reliability by Detection … 597

could be predicted from the test data of similar systems. Huang and Lin [4]
classified faults into two types, dependent faults and independent faults. Mutual
independent faults are those that can be removed directly. Dependent faults could
be removed only if leading faults are removed. Higher proportion of dependent
faults affects the reliability of the system. As the reliability requirement increases,
time taken to achieve the reliability also increases hence increases the cost
involved to achieve reliability. Robert et al. [5] stated that reliability increases as
the software is improved. The concept of visit counts is used to represent the
criticality of the component based on the combination of depth of the component,
number of user functions called and number of test cases written for that com-
ponent. System reliability is calculated by considering the individual reliability of
the component along with the number of visits to each component. Podgorelec [6]
applied on a real-world software reliability analysis dataset based on evolutionary
induced decision trees to identify the outliers.

56.2.2 Outlier Mining

The distribution based methods use a statistics or probability model (e.g. a normal
distribution or Poisson distribution) for the given data set and then identifies
outliers using a discordancy test. A discordancy test verifies whether an object Oi,
is significantly large (or small) in relation to the standard distribution model [1].
Depth based techniques represent every data in a k-d space, and assign depth to
each object. Here the data objects with smaller depths are declared as outliers [7].
Roousseeuw [8] propose robust regression to provide resistant results in the
presence of outliers. ISODEPTH and Fast Depth Contours (FDC) are two depth
based algorithms based on depth contours. Knorr and Raymond [9] proposed a
method based on distance calculation to identify outliers. Ramaswamy et al. [10]
presented new definition for outliers and propose a novel formulation for distance-
based outliers that is based on the distance of a point from its kth nearest neighbour
and developed a highly efficient partition-based algorithm for mining outliers.
Peng and Biao [11] defined the correlation matrix considering the importance and
relationship of attributes to detect outlier in large dataset. Ali et al. [12] presented
an overview of the major developments in the area of detection of Outliers in
numerical datasets. These include projection pursuit approaches as well as Ma-
halanobis distance-based procedures. They also discuss principal component-
based methods, which is applicable for high dimensional data. Aleksandar and
Vipin [13] presented a novel feature bagging approach for detecting outliers in
high dimensional and noisy databases by combining multiple outlier detection
algorithms. Breunig et al. [14] introduced a new method for finding outliers in a
multidimensional dataset through density based which uses a local outlier factor
(LOF) for each object in the dataset, indicating its degree of outliers. Podgorelec
et al. [15] introduced the class confusion score metric based on the classification
results for predicting outlier in medical dataset. Gongxian Cheng [16] presents a
598 S. Poonkuzhali et al.

novel attempt in automating the use of domain knowledge in helping distinguish

between different types of outliers. Jin at al. [17] proposed cluster based algorithm
that restricts LOF value computation to selected clusters that constitute the can-
didate outlier set. It computes the lower and upper limits for each cluster based on
their local reachability densities for detecting outliers. The replicator neutral
network (RNN) is employed to detect outliers by Hawkins et al. [18].

56.3 Framework of the Proposed System

The system design proposed in this paper comprises of two data mining techniques
and it’s framework is presented in Fig. 56.1. The first design focuses on outlier
mining of the KC1 dataset. The second design focuses on the classification of the
dataset. This phase also includes the comparison of classifier results before and
after Outlier detection and removal. Outlier detection is performed based on Sigma
Rule. After the error rates of each classification algorithm are recorded, the
algorithm with the least error-rate is considered as the efficient classifier. A test
data is used to test and validate the accuracy of this efficient classifier.

Fig. 56.1 Architecture of the proposed system

56 Enhancing the Efficiency of Software Reliability by Detection … 599

56.3.1 Training Dataset

The training dataset is downloaded from the NASA Promise software defect data
sets [19] and is described by 2096 instances of KC1 data. This data comprises of
21 input attributes and one target attribute. The target class has two distinct values
yes, no whether it is defective or not. These attributes are categorized as: 5 dif-
ferent lines of code measure, 3 McCabe metrics, 4 base Halstead measures, 8
derived Halstead measures, 1 branch-count, and 1 class field. The detailed
description of the KC1 data is given in Table 56.1. After the data is pre-processed
through outlier detection algorithm 633 outliers are removed. Then 16 classifi-
cation algorithms are executed on the original dataset as well as on filtered dataset.

56.3.2 Outlier Detection

Outlier mining has been studied extensively by statistics community. Data objects
that show significantly different characteristics from the remaining data are
referred as outliers [20]. Outliers may occur as a result of mechanical faults,
changes in system behavior, fraudulent behavior or through natural deviations in
population. Outlier detection approaches focus on discovering patterns that are

Table 56.1 Error rate of classification algorithms on KCI dataset

S.No Classification algorithm Before outlier detection After outlier detection
1 C4.5 0.0692 0.0560
2 C-PLS 0.3216 0.2871
3 C-RT 0.1551 0.0868
4 CS-CRT 0.1551 0.0868
5 CS-MC4 0.1345 0.0868
6 C-SV4 0.1498 0.0868
7 ID3 0.1446 0.0868
8 K-NN 0.1126 0.0752
9 Log-reg TRIRLS 0.1527 0.0882
10 Multilayer perceptron 0.1360 0.0868
11 Multinomial logistic regression 0.1403 0.0848
12 Naive Bayes continuous 0.1937 0.1962
13 PLS-DA 0.1455 0.0868
14 PLS-LDA 0.1441 0.0957
15 Rnd tree 0.0148 0.0157
16 SVM 0.1512 0.0868
600 S. Poonkuzhali et al.

rare or infrequent, while remaining data mining techniques attempt to find patterns
that are frequent. As the removal of outliers from a dataset leads to specific model
or algorithm to succeed, the outliers detection strategies are used for data cleaning
before any traditional mining algorithm is applied to the data. In this proposed
work, Univariate Statistical Outlier Detection technique is applied using TANA-
GRA tool [21] to detect and remove outliers in the KC1 dataset.

56.3.3 Univariate Outlier Detection

Most of the earliest univariate methods for outlier detection rely on the assumption
of an underlying known distribution of the data, which is assumed to be identically
and independently distributed. Moreover, many discordance tests for detecting
univariate outliers further assume that the distribution parameters and the type of
expected outliers are also known [22]. Needless to say, in real world data-mining
applications these assumptions are often violated.
Given a data set of n observations of a variable x, let be the mean and let be
standard deviation of the data distribution. One observation is declared as an
outlier if lies outside of the interval.

ðl  ka; l þ kaÞ ð56:1Þ

where the value of k is usually taken as 2 or 3. The justification of these values

relies on the fact that assuming normal distribution one expects to have a 95 %
(99 %, respectively) of the data on the interval centered in the mean with a semi-
length equal to two (three, respectively) standard deviation.
A total of 633 instances in the KC1 dataset are detected as outliers by this
univariate outlier detection method. Then these outliers were removed from further
analysis. Next various classification algorithms are applied on these filtered dataset
to enhance the accuracy of the classifier.

56.3.4 Classification

The sixteen classification algorithms were executed and its misclassification rate
was recorded for the KC1 data from NASA Promise software defect data sets.
Classification is a supervised learning method which predicts a class label to a set
of data values whose class label is unknown. Except Rand Tree and Naive Bayes
Continuous algorithm all other classifiers have improved accuracy after the out-
liers are removed from the KC1 dataset. The classifier error-rates before and after
outlier detection on the KC1 data are compared and the best performing classifier
is based upon the accuracy in classification. Here, the Rand Tree classifier
gives 98.4 % accuracy and is considered as the best classifier for this dataset.
56 Enhancing the Efficiency of Software Reliability by Detection … 601

The accuracy of this best classifier is verified by testing the rules generated with a
KC1 test data. The classifier reported the class label for each test data with high
level of precision.

56.4 Experimental Results

The experimental results are dealt with two sections. In the first section, the results
of Univariate Outlier detection algorithm are provided. The second section gives
the comparative analysis of the error rate of the classification algorithms before
and after outlier detection. Finally, performance analysis of various classifiers
before and after outlier detection in terms of accuracy is represented graphically.

56.4.1 Outlier Analysis

The results obtained using the data mining tool, TANAGRA on execution of the
Univariate Outlier Detection Algorithm is presented in Fig. 56.2. Nearly 633
instances are removed from further analysis on the KC1 dataset using Sigma rule.

56.4.2 Classification Results of KC1 Dataset

The performance of the 16 classification algorithms was executed on the KC1 data
of NASA Promise software defects dataset. The error rate is reduced for all the
classifiers listed in Table 56.1. The accuracy of Rand tree classifier is 98.4 %, C4.5
is 94.0 % and the remaining classifiers except C-PLS and Naive Bayes continuous
all other classifiers gives above 90 % accuracy after performing outlier detection.
The accuracy of all the classifiers before as well as after outlier detection is
presented in Fig. 56.3. Since almost all the classifiers shows improved perfor-
mance after outlier detection, it is proved that the quality of the results gets
improved which in turn gives better reliability.

56.5 Conclusion

The proposed research work mainly aims at highlighting the impact of outlier
mining prior to classification of software defect dataset will leads to the overall
improvement of software reliability. The comparison of classification techniques
on predicting the defect will certainly assist in providing better quality of software
products. As the presence of outliers degrades the performance of the software
602 S. Poonkuzhali et al.

Fig. 56.2 Results of outlier detection algorithm on KC1 data using sigma rule

Fig. 56.3 Accuracy of classification algorithm

56 Enhancing the Efficiency of Software Reliability by Detection … 603

products, it needs special attention. In this work, the impact of a Univariate outlier
detection method is analyzed on KC1 dataset to show its dependency on improving
the software reliability.

References

1. Han, J., Kamber, M.: Data Mining: Concepts and Techniques, Morgan Kaufmann Publishers,
San Jose (2001)
2. Go, K., seva-Popstojanova, Trivedi, K.S.: Architecture-based approach to reliability
assessment of software systems. Perform. Eval. 45(2/3), 179204 (2001)
3. Xie, M., Hong, G.Y., Wohlin, C.: Software reliability prediction incorporating information
from similar projects. J. Softw. Syst. 49(1), 43–48 (1999)
4. Huang, C.Y., Lin, C.T.: Software reliability analysis by considering fault dependency and
debugging time lag. IEEE Trans. Reliab. 55 (3), 436450 (2006)
5. Roberto, R.S., Trivedi, K. S.: Software reliability and testing time allocation: an architecture
based approach. IEEE Trans. Reliab. 36(3), 322–337 (2010)
6. Podgorelec,V.: Improved mining of software complexity data on evolutionary filtered
training sets. WSEAS Trans. Inf. Sci. Appli. 6(11), 1751–1760 (2009)
7. Ruts, I., Rousseuw, P.: Computing depth contours of bivariate points cloud. Comput. Stat.
Data Anal. 23, 153–160 (1996)
8. Roousseeuw, P., Leroy, A.: Robust Regression and Outlier Detection. Wiley, New York
(2003)
9. Knorr, E.M., Ng, R.T.: A unified notion of outliers: properties and computation. In:
Proceedings of KDD 97, pp. 219–222 (1997)
10. Ramaswamy, S., Rastogi, R., Shim, K.: Efficient algorithm for mining outliers from large
datasets. In: Proceedings of ACM SIGMOD, pp. 127–138 (2000)
11. Peng, Y., Biao, H.: An efficient outlier mining algorithm for large dataset. In: International
Conference on Information Management, Innovation Management and Industrial
Engineering, IEEE Computer Society, pp. 199–202 (2008)
12. Ali, H.S., Rahmatullah Imon, A.H.M., Mark, W.: Detection of outliers overview. Wiley
Interdisc. Rev. Comput. Stat. 1(1), 57–70 (2009)
13. Aleksandar, L., Vipin, K.: Feature Bagging for Outlier Detection, KDD05, Chicago, Illinois,
USA, August 2124 (2005)
14. Breunig, V., Kriegel, M.M., Ng, R.T., Sander, J.: LOF: identifying density based local
outliers, Proc. of ACM SIGMOD (2000) (Dallas, TX)
15. Podgorelec, V., Hericko, M., Rozman, I.: Improving mining of medical data by outliers
prediction. In: Proceedings of the 18th IEEE Symposium on Computer-based Medical
Systems CBMS’2005 (2005)
16. Gongxian Cheng, J.: Outlier Management in Intelligent Data Analysis. Department of
Computer Science, Birkbeck College, University of London, London (2000)
17. Jin Tung, W.A.K.H., Han, J.: Mining top-n local outliers in large databases. In: Proceedings
of KDD, San Francisco, CA, USA, (2001)
18. Hawkins, S., He Willams, H.G.J., Baster, R.A.: Outlier detection using replicator neural
networks. In: Proceedings of the DaWaK02, pp. 170–180 (2002)
19. Promise http://promise.site.uottawa.ca/SERepository/datasets-page.htm
20. Hawkins, D.: Identification of Outliers. Chapman and Hall, London (1980)
21. Tanagra-Data Mining tutorials http://data-mining-tutorials.blogspot.com
22. Barnett, V., Lewis, T.: Outliers in Statistical Data. Wiley, New York (1998)
604 S. Poonkuzhali et al.

Author Biographies

Dr. S. Poonkuzhali received B.E degree in Computer Science and Engineering from University
of Madras, Chennai, India, in 1998, and the M.E degree in Computer Science and Engineering
from Sathyabama University, Chennai, India, in 2005, and PhD in the Faculty of Information and
Communication Engineering at Anna University Chennai, in 2012. Currently she is working as a
Professor and Head in the Department of Information Technology in Rajalakshmi Engineering
College, Chennai. Her research interests include Data Mining, Neural Networks, Web Mining
and Knowledge Management. She has organized and chaired many Workshops, Seminars and
Conferences in national and International level. She has presented and published more than 25
research papers in international conferences & journals and authored 5 books. She is a life
member of ISTE (Indian Society for Technical Education), IAENG (International Association of
Engineers), ISCSIT and CSI (Computer Society of India).

R. Kishore Kumar received B.E degree in Computer Science and Engineering from Rajalakshmi
Engineering College, Anna University, Chennai, India in 2011 and M.E degree in Computer
Science and Engineering in SSN College of Engineering, Anna University Chennai, India in
2013. Currently he is working as Assistant Professor in Department of Information Technology in
Rajalakshmi Engineering College, Chennai. He has presented 11 papers in International con-
ferences and published 6 research papers in international journals and 3 papers in national
journals. One of his papers has been selected as the Best Paper. He also received the Certificate of
Merit Award for the paper presented in IAENG Conference. He is also the member of Computer
Society of India and IAENG.

R. Kumar is an Associate Professor in Rajalakshmi Engineering College, Chennai. He holds

M. Tech degree in Computer Science and Engineering and is pursuing his PhD in The Anna
University, Chennai as a part time scholar. His research interests include Knowledge Engineering
and Management, Data Mining and Software Engineering. He is a member of Computer Society
of India (CSI).
Chapter 57
A Survey on the Application of Robotic
Teacher in Malaysia

Noraidah Blar and Fairul Azni Jafar

Abstract Robot application in Malaysia is just a couple of number these days

contrast with different nations particularly in education. Robotic teacher applica-
tion is likewise not extremely commonplace around Malaysian students. The
survey is intended to distinguish the sentiment around Malaysian technical insti-
tutes about utilization of robotic teacher in their organization. An overview was
developed and dispersed by utilizing web interface that is Google Form applica-
tion. The result demonstrates that the greater part of students who completed the
survey do not consent to utilize a robot as a teacher. Numerous Malaysian indi-
viduals finished not think about the genuine proficiency of a robotic teacher. A
further study about this theme will be led after this investigation.

57.1 Introduction

As a human, there are likewise various types of teaching method. There are always
questions highlighted that it is, is adequate or not to make students completely
comprehend what they attempt to educate. What’s more yes, a few students are
neglected to comprehend what they take in because of numerous factors, one of
them is teaching method conveyed by their teacher. Living in a world that
undeniably propelled, constrained human to find some other elective to make the
instruction more viable. One of the finding is utilizing robots as a part of teaching
session.

N. Blar (&)
Center for Graduate Student, Universiti Teknikal Malaysia Melaka, Melaka, Malaysia
e-mail: [email protected]
F.A. Jafar
Faculty of Manufacturing Engineering, Universiti Teknikal Malaysia Melaka, Melaka,
Malaysia
e-mail: [email protected]

Ó Springer International Publishing Switzerland 2015 605

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_57
606 N. Blar and F.A. Jafar

According to Berk [1], there are twelve potential sources of evidence to mea-
sure teaching effectiveness that are critically reviewed; student ratings, peer rat-
ings, self-evaluation, videos, student interviews, alumni ratings, employer ratings,
administrator ratings, teaching scholarship, teaching awards, learning outcome
measures, and teaching portfolios. From [2], it is stated that the strategies to
convey learning are; practical examples, show and tell, case studies, guided design
projects, open-ended labs, the flowchart technique, open ended quizzes, brain-
storming, question-and-answer method, software, teaching improvement, and fast
feedback form.
In order to understand about robot teacher, basic information about robot must be
learned first. A robot is a mechanical device that can perform preprogrammed
physical tasks. It may act under the direct control of a human or automatically under
the control of a pre-programmed computer. Robot can be classified by several types.
There is industrial, mobile, service, military, humanoid and other type of robot.
In a research conducted by Chang et al. [3], the research is aim to find the
possibilities of using humanoid robots as instructional tools for teaching second
language in primary school. They found that the absolute majority of the students
actively participated in learning activities throughout the lesson and interacted
with the robot-teachers with great interest.
Numerous researches about robot teacher have been carried out in other nation
particularly at the advancing countries, for example, Japan and United State. It is
accepted that the level of competencies and thinking in instruction for Malaysian’s
students is much distinctive contrast with different countries. Consequently, a
considerable measure of exploration is required to know the suitability of utilizing
robots within teaching Malaysian individuals. The focus of this paper is to know the
feedback of Malaysian technical students about the application of robotic teacher.

57.2 Survey

Main subject of this survey is the feedback from the respondents. Respondents are
consisting of Malaysian technical students from technical universities (Universiti
Teknikal Malaysia Melaka, Universiti Malaysia Pahang, Universiti Tun Hussein
Onn Malaysia, and Universiti Malaysia Perlis). The survey’s questions was created
and it then distributed by using Google Form. The link of this form was shared at
social webpages of the technical universities that listed. The survey’s question is
constructed as follow;
1. Is there any robot used in your university?
2. If yes, what they are used for?
3. What is your level of exposed with robot?
4. Have you ever experience a teaching delivered by a ROBOT?
5. Just imagine, what will you feel when a robot is teaching your subject in class?
57 A Survey on the Application of Robotic Teacher in Malaysia 607

Fig. 57.1 Survey’s question using google form (minimized)

608 N. Blar and F.A. Jafar

6. In your experience, is teaching method delivered by a human teacher is

effective enough?
7. In your opinion, what will affect the effectiveness of delivering lesson?
8. In your prediction, will robot teach students more effective than human teacher
did?
9. Opinion.
The purpose of the first and the second question is to know the alertness of
respondent on the existing robot at their place. The third and the fourth question
are constructed to know the exposure condition among the respondents. While, the
other questions are asked in order to know about respondent’s feedback on the
application of the robot teacher (Fig. 57.1).

57.3 Results

There are a total of 67 respondents. 45 out of them are male and 22 persons are
female. Total of respondents according to each universities are 30 persons (45 %)
from Universiti Teknikal Malaysia Melaka, 12 people (18 %) from Universiti Tun
Hussein Onn Malaysia, 15 people (23 %) from Universiti Malaysia Pahang, 8
people (12 %) from Universiti Malaysia Perlis, and a person (2 %) from other
institution. According to the survey, there are 59 students, 3 lecturers, and 4 others.
Below are the graphs of results of each question.

57.4 Discussion

Figure 57.2 shows the result for the first question. The question asks about the
existing robot in their university. It is shown that 54 respondents state that there is
robot(s) in used in their university. While the other 12 respondents stated that robot
is not in used in their institute. It is shown that some of them are not alert enough
about the robot. This situation may due to they did not use robot in their study. The
second question (Fig. 57.3) asked about the function of robot in their university if

Fig. 57.2 Result of

question 1
57 A Survey on the Application of Robotic Teacher in Malaysia 609

Fig. 57.3 Result of

question 2

Fig. 57.4 Result of

question 3

there is any robot in used. From the list of function suggested, the highest number
is 45 which is laboratory equipment. The second highest usage of robot is for
student’s project with 41 respondents, followed by teaching tool, contest, display,
and other with values of 19, 18, 14, and 6 subsequently. The next graph (Fig. 57.4)
shows the result for the third question. The question is asking about the level of
exposure of respondents with robot. In the question, there are 10 level of exposure.
Which is Level 1 is the fewest, while Level 10 is the most frequent. From the
result, it is shown that most of the respondents are level 5 for the question. The
highest value is 13 respondents with 19 % of the graph, followed by the second
highest with value of 10 respondents (15 %) which is level 6. The third highest
value of respondent is level 1 which has the number of respondent of 8 (12 %).
Level 3, 4, and 7 have the same number of respondent that is 7 (10 %).
610 N. Blar and F.A. Jafar

Meanwhile, level 2 and 8 share the same value of 6 respondents (9 %). Level 9 and
10 has the smallest value with 2 (3 %) and 1 (1 %) respectively. It is shows that
most of the respondents have average level of experience.
The next question (Fig. 57.5) is about the experience of respondents in teaching
by robot. Most of them say no with the number of 62 respondents, while the other
5 respondents have experience learning from robot. Respondents do not have
experience yet in having a lesson delivered by a robot. The graph in Fig. 57.6
shows the statistical result of feelings by respondents. There are 17 feelings that
listed in the question. Respondents can choose more than one feeling to answer the
question. Most of the respondents feel excited when they imagine getting lesson
from a robot. There are 56 people that feel that way with the highest percentage of
16 %. The second highest is impressed feeling with 50 people, followed by
inspired expression with 35 people. The lowest number of results in the graph is

Fig. 57.5 Result of question 4

Fig. 57.6 Result of question 5

57 A Survey on the Application of Robotic Teacher in Malaysia 611

Fig. 57.7 Result of

question 6

Fig. 57.8 Result of

question 7

confident with 5 respondents. Most of the respondents responded to be excited as

because they never imagine robot cannot do such things.
The graph in Fig. 57.7 shows the opinion from the respondents on the effec-
tiveness of teaching by human teacher. It is shown that 35 people are agreed that
human teacher already teach in an effective way, while the other 32 feel that it is
not good enough. In Fig. 57.8, there are 52 people choose on interaction factor that
will affect the effectiveness on delivering lesson. This is the highest number of
factor selected among the other factor. Meanwhile, communication factor become
the second highest of option from the respondents with value of 44. This is
followed by teaching method and teaching skills that shared the same value of
number of choice from the respondents. There are least people that choose
facilities as the factor. Interaction factor become the highest choice from the
respondents because the might be known that, to achieve a full understanding from
each other, interaction is important to each other.
The graph of Fig. 57.9 shows the result on the prediction of respondents about
comparison of effectiveness of teaching by both human and robot. Majority of
them, with 45 number of people, disagree with the statement that robot will teach
more effective than human teacher did. The rest of the respondent, with 22
numbers of respondent, have a positive feedback that believe robot will teach
better than the human teacher did.
612 N. Blar and F.A. Jafar

Fig. 57.9 Result of

question 8

Table 57.1 List and summary of respondent’s opinion

Positive opinion Negative opinion
Both robot and human can be a good Robot did not have feeling like human
teacher
Robot is fun A teacher is a human that know how their student
perform. They can approach that kind of student
Robot can attract people to learn Teacher will lose their job
better
Useful for future study Beware of 3 Asimov law of robot
High technology robot can replace Robot is dangerous to human being
human in the future
Everything have their own Robot is heartless, human is close with human heart
weakness, both human and robot

Below is the list and summary of the last survey’s question, that asking about
the respondent’s opinion about robotic teacher. Overall of the result shows that
mostly human still did not have believed in robot in delivering lesson (Table 57.1).

57.5 Conclusion

Through the result that we get by this survey, we can see that the implementation
of robot teacher have many challenges. This is because there are many respondents
that still not believe about the effectiveness of robot teacher (Fig. 57.8). Although
when we ask about their feeling when robot give them a teaching lesson, most of
them feel excited and impressed (Fig. 57.5).
So, we have decided to further this study in order to identify the effectiveness of
robot teacher in delivering technical lesson to human. The study will involve a
robot who will teach the respondents and then the respondents will have to fill up a
survey form.
57 A Survey on the Application of Robotic Teacher in Malaysia 613

References

1. Berk, R.A.: Survey of 12 strategies to measure teaching effectiveness. Int. J. Teach. Learn.
High. Educ. 17, 48–62 (2005) (John Hopkins University, USA)
2. Couter, S., Balaraman, P., Lacey, J., Hochgraf, C.: Strategies for Effective Teaching: a
Handbook for Teaching Assistants. Teaching Assistant Fellows, University of Wisconsin,
Madison, College of Engineering, Madison (1995)
3. Chang, C.W., Lee, J.H., Chao, P.Y., Wang, C.Y., Chen, G.D.: Possibility of using humanoid
robots as instructional tools for teaching a second language in primary school. Educ. Technol.
Soc. 13(2), 13–24 (2010)
Chapter 58
A Novel Method for Distributed Image
Steganography

Bismita Choudhury, Rig Das and Themrichon Tuithung

Abstract Distributed Image Steganography (DIS) is a new method of concealing

secret information in several host images, leaving smaller traces than conventional
steganographic techniques and requiring a collection of affected images for secret
information retrieval. DIS, compared to other conventional steganographic tech-
niques, can improve security and information hiding capacity because DIS leaves
reduced signatures of hidden information in host images. This paper presents a
novel technique for Distributed Image Steganography based on Shamir’s (k, n)
threshold based secret sharing scheme. An 8-bit, gray level, Secret Image of size
P 9 Q is divided into several sections and embedded inside n-Cover Images of
size M 9 N and only k-Stego Images (k B n) are required to retrieve the secret
image. The size of the secret image is also embedded inside every cover images, so
that the stego images become standalone information to the receiver. Experimental
result shows that the proposed novel method has high capacity and good
invisibility.


Keywords Steganography Distributed image steganography PSNR Shamir’s  
secret sharing scheme

B. Choudhury (&)
Department of Computer Science and Engineering and Information Technology, Don Bosco
College of Engineering and Technology, Guwahati 781017, Assam, India
e-mail: [email protected]
R. Das
Department of Computer Science and Engineering, National Institute of Technology,
Rourkela 769008, Orissa, India
e-mail: [email protected]
T. Tuithung
Department of Computer Science and Engineering, North Eastern Regional Institute of
Science and Technology (Deemed University), Itanagar 791109, Arunachal Pradesh, India
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 615

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_58
616 B. Choudhury et al.

58.1 Introduction

Image Steganography is the art of hiding secret data or images into a Cover
Image [1]. In recent years, many techniques to increase the security of the secret
image were proposed. However a common weakness of these techniques is that the
secret data are all in a single information-carrier and the secret data cannot be
revealed completely if the information-carrier is lost or crippled [2, 3]. If we use
many duplicates to overcome the weakness, the danger of security exposure will
also increase [2]. To solve this dilemma, Distributed Image Steganography (DIS)
might be one of the possible solutions. DIS is a new method of concealing secret
information in several host images, leaving smaller traces than conventional ste-
ganographic techniques and requiring a collection of affected images for secret
information retrieval [4]. DIS is a secret image sharing method derived from the
Shamir’s Secret Sharing Scheme [5]. Shamir developed the idea of a (k, n)
threshold-based secret sharing technique where k B n and for this we will require:
1. A secret image that will be used to generate n stego images.
2. Any k or more stego images can be used to reconstruct the secret image.
3. Any k – 1 or less stego images cannot get sufficient information to reveal the
secret image [6].
By using a (k, n) threshold scheme with n = 2k - 1 we get a very robust key
management scheme: We can recover the original key even when bn=2c ¼ k  1 of
the n pieces are destroyed, but our opponents cannot reconstruct the key even when
security breaches expose bn=2c ¼ k  1 of the remaining k pieces [5]. The main
objective of Distributed Image Steganography is to communicate securely in such a
way that the true message is not visible to the observer i.e. any unwanted parties
should not be able to distinguish between cover-images (images not containing any
secret message) and stego-images (modified cover-images that containing secret
message). Thus the stego-images should not deviate much from original cover-
images. Figure 58.1 shows the block diagram of a simple DIS system with (3, 5)
threshold scheme that means the secret image will be distributed within 5 stego
images and any 3 stego images are enough to regenerate the secret image.
This paper organized as follows. Section 58.2 reviews some of the related
works performed by the researchers. Proposed novel method is presented in
Sect. 58.3 along with the block diagram and the algorithm. Experimental results
are shown in Sect. 58.4 and finally the conclusion is being stated in Sect. 58.5.

58.2 Related Work

Different researchers employed different techniques for the purpose of distributing

secret image over a set of stego images. Shamir’s Secret Sharing Scheme and Thien
and Lin’s Secret Image Sharing Scheme are the two essential processes to protect
secret image in DIS. Following are the related works carried out by these groups:
58 A Novel Method for Distributed Image Steganography 617

Fig. 58.1 The block diagram of a simple distributed steganographic system with (3, 5)

58.2.1 Shamir’s Secret Sharing Scheme [5]

Shamir’s scheme is based on polynomial interpolation: given k points in the

2-dimensional plane (x1, y1),…,(xk, yk) with distinct xi’s, there is one and only one
polynomial q(x) of degree k - 1 such that q(xi) = yi for all i. It is assumed that the
data D is (or can be made) a number. To divide it into pieces Di, pick a random
k - 1 degree polynomial q(x) = a0 + a1x + _ ak-1xk-1 in which a0 = D, and
evaluate: D1 = q(1), …, Di = q(i), …, Dn = q(n). Given any subset of k of these
Di values (together with their identifying indices), the coefficients of q(x) can be
found by interpolation, and then evaluating D = q(0). Knowledge of just k - 1 of
these values, on the other hand, does not suffice in order to calculate D.
The set of integers modulo a prime number p forms a field in which interpo-
lation is possible. Given an integer valued data D, we pick a prime p which is
bigger than both D and n. The coefficients a1,…,ak-1 in q(x) are randomly chosen
from a uniform distribution over the integers in [0, p), and the values D1,…,Dn are
computed modulo p. Now assuming that k – 1 of these n pieces are revealed to an
opponent. For each candidate value D0 in [0, p) he can construct one and only one
polynomial q0 (x) of degree k - 1 such that q0 (0) ¼ D0 and q0 (i) ¼ Di for the k - 1
618 B. Choudhury et al.

given arguments. By construction, these p possible polynomials are equally likely,

and thus there is absolutely nothing the opponent can deduce about the real value
of D.

58.2.2 Thien and Lin’s Secret Image Sharing Scheme [6]

Suppose we want to divide the secret image D into n shadow images (D1,…,Dn),
and the secret image D cannot be revealed without r or more shadow images. In
the proposed method, they generate the r - 1 degree polynomial, by letting the
r coefficients be the gray values of r pixels. Therefore, the major difference
between Thien and Lin’s method and Shamir’s is that they use no random coef-
ficient. Because the gray value of a pixel is between 0 and 255, they took the prime
number p be 251 which is the greatest prime number not larger than 255. To apply
the method, it must truncate all the gray values 251–255 of the secret image to 250
so that all gray values are in the range 0–250. The image is divided into several
sections. Each section has r pixels, and each pixel of the image belongs to one and
only one section. For each section j, define the following r - 1 degree polynomial
as qj(x) = (a0 + a1x + _ + ar-1xr-1) mod 251, Where a0,…,ar-1 are the r pixels
of the section, and then evaluate qj(1), qj(2), …,qj(n), The n output pixels qj(1) –
qj(n) of this section j are sequentially assigned to the n shadow images. Since for
each given section (of r pixels) of the secret image, each shadow image receives
one of the generated pixels; the size of each shadow image is 1 / r of the secret
image. The reveal phase uses any r (of the n) shadow images, and the Lagrange’s
interpolation to extract the secret image.

58.3 Proposed Novel Method for Distributed Image

Steganography

In this paper we have proposed a spatial distributed steganographic technique

based on Shamir’s secret sharing scheme for hiding a large amount of data with
high security, a good invisibility and no loss of secret message. The schematic/
block diagram of the whole process is given in Figs. 58.2 and 58.3.
Our proposed algorithm for DIS is based on (k, n) Threshold Scheme and has
two parts, one for Embedding the Secret Image inside n–Cover Images and another
for Extracting the Secret Image from k–Stego Image. The Secret Image is divided
into several sections. Each section has k pixels, and each pixel of the secret image
belongs to one and only section. For each section j, we define the following k - 1
degree polynomial:
58 A Novel Method for Distributed Image Steganography 619

Fig. 58.2 Insertion of a secret image inside n–Cover images

Fig. 58.3 Retrieval of secret image from k–Stego images

620 B. Choudhury et al.

 
pj ðxÞ ¼ a0 þ a1 x þ    þ ak1 xk1 mod 256 ð58:1Þ
  
qj ðxÞ ¼ floor a0 þ a1 x þ    þ ak1 xk1 =256 ð58:2Þ

Where, value of x ranges from 1 to n and a0 to ak-1 are Secret

Image’s k number of pixel’s intensity values. These intensity values changes
sequentially e.g. first we take 1–8 pixels’ intensity values then 9–16 pixels’
intensity values and so on. Modulus and Floor operations are performed using 256
as divisor because a grey level image has 256 different intensity levels. Equation
(58.1) finds the Remainder Value after performing Modulus operation and (58.2)
finds the Quotient value after performing Floor operation.
The main objective in here is to develop a procedure which will provide a better
security to the secret image without compromising on the quality of the stego
image. Our algorithm has two main parts. First, it embeds the size of the secret
image in first 8 9 8 block of every cover image. Second, it embeds each set of
pj(x) and qj(x) that means the remainder and the quotient value sequentially to
every cover image.

58.3.1 Four Tier Storage Procedure of ‘‘Size of Secret

Image’’ [7]

The size of the Secret Image multiplied by 8 (for 8 bit images) should be less than
the size of the Cover Image. Because every pixel of secret image has 8 bit (for 8 bit
images) and it requires 8 pixels of the cover image to get embedded. E.g. if the size
of the Secret Image is 256 9 256 = 65536 then after multiplying it by 8 it
becomes 523288. This is lesser than the size of Cover Image i.e.
1024 9 1024 = 1048576. So this secret image can be embedded. But if the size of
the secret image is 512 9 512 = 262144 then after multiplying it by 8 it becomes
2097152 which is more than the number of pixels in the cover image of size
1024 9 1024. So a secret image of size 512 9 512 can’t be embedded inside the
cover image of size 1024 9 1024.
The size of the Secret Image needs to be embedded inside the cover image to let
the decoder know which pixel’s LSB holds the Secret Image’s information. Now
the question is how will we get the size of the ‘‘Size of Secret Image’’? If we store
this size of the ‘‘Size of Secret Image’’ then also the same question comes
recurrently like what is the size of this size? As a permanent answer to this
question we have used a four tier storage procedure to store the ‘‘Size of Secret
Image’’ which had been described in [7].
58 A Novel Method for Distributed Image Steganography 621

58.3.2 Proposed Novel Algorithm for Distributed Image

Steganography

Embedding Algorithm
Input: n number of M 9 N Carrier Images and a P 9 Q Secret message/Image.
Output: n number of M 9 N Stego-Images.
Step-1: Read all the n–Cover Images and the Secret Image and store their
intensity values of different pixels in different arrays.
Step-2: Calculate size of Secret Image. The size of the Secret Image multiplied
by 8 (for 8 bit images) should be less than the size of the Cover Image.
E.g. if the size of the Secret Image is 256 X 256 = 65536 then after
multiplying it by 8 it becomes 523288. This is lesser than the size of
Cover Image i.e. 1024 X 1024 = 1048576.
Step-3: Store the size found in Step-2 using the four tier storage procedure
described above by modifying the LSB of first 8 9 8 block of pixels of
all the cover images.
Step-4: Sequentially take k–number of not-shared-yet pixels of the Secret Image
and use (58.1) and (58.2) to find the n number of Remainder and
n number of Quotient value.
Step-5: Change the LSBs of n Cover Images to insert each set of Remainder and
Quotient values found in Step-4 in each of the Cover Images excluding
the first 8 9 8 block.
Step-6: Repeat Step-4 and Step-5 until all the pixels of the Secret Image are
processed.
Step-7: Write all the n Stego Images into the disk.
Extraction Algorithm
Input: k number of M 9 N Stego-Images.
Output: A P 9 Q Secret Image.
Step-1: Read k number of Stego Images
Step-2: Read the Stego Image and extract the size of the Secret Image from the
first 8 9 8 block of any Cover Image by extracting the LSB of the
pixels using the procedure described in the four tier storage method like
variable D to C to B to A.
Step-3: Extract k number of Remainders and k number of Quotients from the k
Stego Images by extracting the LSBs of the pixels.
Step-4: Use Lagrange’s Interpolation to retrieve k number of pixel’s intensity.
Step-5: Repeat Step-3 and Step-4 until the total number of pixels of the Secret
Image are processed.
Step-6: Regenerate the original Secret Image from the pixel intensities found in
Step-4 and Step-5.
622 B. Choudhury et al.

58.4 Simulation Results

In this section, some experiments are carried out on our proposed algorithm for
(DIS).The measurement of the quality between the cover image f and stego-image
g of sizes M 9 N is done using PSNR (Peak Signal to Noise Ratio) value and the
PSNR [8, 9] is defined as:
 
PSNR ¼ 10  log 2552 =MSE ð58:3Þ

N 1 X
X N 1
Where; MSE ¼ ðf ðx; yÞ  gðx; yÞÞ2 =N 2 ð58:4Þ
x¼0 y¼0

f (x, y) and g (x, y) means the pixel intensity value at position (x, y) in the cover-
image and the corresponding stego-image respectively. The PSNR is expressed in
dB. The larger PSNR indicates the higher the image quality i.e. there is only little
difference between the cover-image and the stego-image. On the other hand, a
smaller PSNR means there is huge distortion between the cover-image and the
stego image.
All the simulation has been done using the MATLAB 7 program on Windows
XP platform. A set of 8-bit grayscale TIFF images of size 1024 9 1024 and
256 9 256 are used as the cover-image and secret image respectively to form the
stego-image. Figure 58.4a–e shows the original cover (carrier) images and
Fig. 58.4f shows the original secret message. Figure 58.5a–e shows the stego
images of proposed DIS method based on (4, 5) threshold scheme and Fig. 58.5f
shows the Secret Image retrieved from the any four Stego Images. Figure 58.6a–e
shows another sets of the original cover (carrier) images and Fig. 58.6f shows the
original secret message. We have tested for both the sets of cover images and

Fig. 58.4 a–e Five cover images for simulations, f original secret image

Fig. 58.5 a–e Five stego images of proposed DIS method, f extracted secret image
58 A Novel Method for Distributed Image Steganography 623

Fig. 58.6 a–e Another five sets of cover images for simulations, f original secret image

Table 58.1 PSNR comparison of cover images and stego images for different threshold scheme
Cover image PSNR (dB) between cover image and stego image
Threshold scheme
(1, 5) (2, 5) (3, 5) (4, 5) (5, 5)
Lenna +56.88 +55.49 +57.41 +58.69 +55.85
Baboon +56.88 +55.65 +57.55 +59.05 +56.06
Airplane +56.89 +55.67 +57.73 +59.41 +56.23
Boat +56.89 +55.77 +58.02 +59.76 +56.42
Coin +56.42 +55.40 +57.52 +59.43 +55.88

Table 58.1 exhibit the PSNR comparison of first set of Ste-go Images with their
corresponding Cover Images for different threshold scheme.
From Table 58.1 it is observed that for threshold schemes (1, 5) to (5, 5) PSNR
is greater than 55 dB, so the quality of the stego image is very high and the
deterioration in quality due to embedding of secret image cannot be distinguished
by naked eye. Best result is achieved in case of (4, 5) threshold scheme, as the
PSNR is greater than +58 dB for all the stego images.
In case of threshold scheme (1, 5), Secret Image is extracted from any one of
the five Stego Images, which is identical with the original Secret Image. Although
(1, 5) threshold scheme is nothing but alike normal LSB based steganography, as
the Secret Image is not distributed over the five Cover Images and any of the
5(five) stego images can be used to extract the secret image. Similarly for
threshold scheme (2, 5), (3, 5), (4, 5) and (5, 5), the Secret Image is extracted from
any chosen 2(two), 3(three), 4(four) or 5(five) (respectively) Stego Images out of
five stego images. Extracted Secret Image is identical to the original Secret Image
as the PSNR between original secret image and extracted secret image is infinite
for all the threshold schemes from (1, 5) to (5, 5). So 100 % recovery of the secret
image is being achieved for all the cases.
Shamir’s (k, n) threshold scheme says k B n so (2, 5), (3, 5), (4, 5) should be the
ideal threshold based secret sharing scheme. But to justify proposed novel method,
simulation is also done for (1, 5) and (5, 5) threshold scheme and for all possible
threshold scheme proposed novel method extracts exactly identical secret Image.
According to the simulation result our proposed method gives best result when
threshold scheme (4, 5) is used, as the PSNR values are mostly over +59 dB.
624 B. Choudhury et al.

Table 58.2 PSNR compari- Cover image PSNR (dB) between cover image
son of second set of cover and stego image
images and stego images for
(4, 5) threshold scheme Threshold scheme (4, 5)
Pout +58.68
MRI +74.07
Cell +59.41
Tire +59.65
Moon +59.85

Table 58.3 PSNR compari- Cover PSNR (dB) between cover image and stego image
son between steganography image
based on Huffman Steganography algorithm
encoding [7] and Based on Huffman DIS with (4, 5)
proposed DIS algorithm encoding threshold scheme
Lenna +57.43 +58.69
Baboon +57.46 +59.05
Airplane +57.46 +59.41
Boat +57.46 +59.76

Table 58.2 provides the PSNR value comparison for second set of cover images
(shown in Fig. 58.6) with their corresponding stego images for the threshold
scheme (4, 5) only. As it is seen from Table 58.1, best results are achieved in case
of threshold scheme (4, 5) we have only chosen (4, 5) threshold scheme for
Table 58.2. The PSNR values are mostly over +59 dB. For these set of images also
the PSNR value is infinite between the extracted secret image and original secret
image. That means Identical Secret Image is being recovered. Same experiments
also being performed over some more sets of cover images and for each one of
those sets identical secret image is being recovered. So it can be claimed that
100 % recovery is possible by our proposed method.
Table 58.3 shows the PSNR and Stego Image Quality of proposed DIS algo-
rithm is better than the one proposed in [7].

58.5 Conclusion

Our proposed novel Distributed Image Steganographic method based on (k, n)

Threshold Scheme improves the security and the quality of the stego images.
According to the simulation results the Stego Images of our proposed algorithm
are almost identical to the Cover Images and it is very difficult to differentiate
between them. We have achieved 100 % recovery of the Secret Image for any set
58 A Novel Method for Distributed Image Steganography 625

of chosen Stego Images. That means original and extracted Secret Images are
identical. Distributing the Secret Image among n–number of Cover Images keeps
the Secret Image away from stealing; destroying by any unintended users hence
the proposed method may be more robust against brute force attack. Apart from
images the proposed DIS method can also be used for textual information hiding,
audio information hiding etc. We think that the proposed method can become an
excellent intelligence embedding and transferring mechanism for law enforce-
ments and military analysts.

References

1. Jhonson, N.F., Jajodia, S.: Exploring steganography: seeing the unseen. In: IEEE Paper of
February (1998)
2. Provos, N., Honeyman, P.: Hide and seek: an introduction to steganography. IEEE Secur. Priv.
1(3), 32–44 (2003)
3. Filler, T., Fridrich, J.: Gibbs construction in steganography. In: IEEE Transactions on
Information Forensics and Security (2010)
4. Bai, L., Biswas, S., Blasch, E.P.: An estimation approach to extract multimedia information in
distributed steganographic images. In: 10th International Conference on Information Fusion, 9-
12 July 2007, Quebec, Canada (2007)
5. Shamir, A.: How to share a secret. Commun. ACM 22(11), 612–613 (1979)
6. Thien, C.C., Lin, J.C.: Secret Image Sharing. J. Comput. Graph. 26(5), 765–770 (2002)
7. Das, R., Tuithung, T.: A novel steganography method for image based on Huffman encoding.
In: 3rd IEEE National Conference on Emerging Trends and Applications in Computer Science
(NCETACS), Shillong, India (2012)
8. Cheddad, A., Condell, J., Curran, K., Mc Kevitt, P.: Digital image steganography: survey and
analysis of current methods. J. Sign. Process. 90, 727–752 (2010)
9. Li, B., He, J., Huang, J., Shi, Y.Q.: A survey on image steganography and steganalysis. J. Inf.
Hiding Multimed. Sign. Process. 2(2), 142–172 (2011)
Chapter 59
An Efficient Beam Scanning Algorithm
for Hidden Node Collision Avoidance
in Wireless Sensor Networks

Moorthy Sujatha and Raghuvel Subramaniam Bhuvaneswaran

Abstract Wireless sensor networks are characterized by an assembly of low power

nodes that collect information about the environmental conditions and report to a
base station in general. The hidden node collision problem is one major problem
faced by the wireless sensor networks (WSNs). Many solutions have been proposed
and implemented with an aim to mitigate to the effect of the same. H-NAMe is one
such scheme that fulfills the quality-of-service (QoS) requirements imposed by the
applications of the WSNs. It relies on a grouping strategy that splits each cluster of a
WSN into disjoint groups of non-hidden nodes that scales to multiple clusters that
guarantees no interference between overlapping clusters. A design weakness
identified in this scheme, energy consumption of nodes, when a new node tries to
join a group, has been eliminated by the proposed work. In this paper, an efficient
Intra cluster grouping scheme (IC-GS) for a new node to be added into the WSN
with and without beam scanning is proposed and simulated using network simu-
lator. The IC-GS with beam scanning is different from the IC-GS scheme as it
proposes a beam scanning process at every fixed angle to determine a cluster head
using directional antennas before it communicated with the determined cluster
head. The simulation results are provided to prove that IC-GS with beam scanning
as an energy efficient method for hidden node collision avoidance in WSNs.

59.1 Introduction

Wireless Sensor Networks have found their place in various industrial applications
and intelligent systems, where intensive sensing operations take place and the
sensed information is continuously reported to either a base station or a server. In a

M. Sujatha (&)
Sathyabama University, Jeppiaar Nagar, Rajiv Gandhi Salai,
Chennai 600119, TamilNadu, India
e-mail: [email protected]
R.S. Bhuvaneswaran
Anna University, Chennai, India

 Springer International Publishing Switzerland 2015 627

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_59
628 M. Sujatha and R.S. Bhuvaneswaran

WSN, collisions may happen when a receiver is within the transmission range of
two transmitters that are transmitting simultaneously, so that the receiver captures
neither frame [1]. As each collision represents unnecessary energy dissipation,
reducing collisions should be the main design objective in any method. Charac-
teristics of the physical environment lead to a major source of QoS degradation in
WSNs—the ‘‘hidden node problem’’. This problem greatly impacts network
throughput, energy-efficiency and message transfer delays, and the problem dra-
matically increases with the number of nodes.
A lot of research has been done to avoid hidden node collision problems and
many schemes have emerged. Many research works have proposed the solutions
for eliminating or reducing the impact of the hidden-node problem in wireless
networks, roughly categorized as: (1) Request-To-Send/Clear-To-Send (RTS/CTS)
mechanisms; and (2) node grouping mechanisms [2]. Grouping mechanisms are
effective compared to the other mechanisms mentioned above. In this work, we
propose two grouping mechanisms, Intra-cluster grouping scheme (IC-GS) and IC-
GS with beam scanning, for the efficient group joining process of a new node in a
WSN. The IC-GS with beam scanning is different from the other in the fact that a
beam scanning process takes place at every fixed angle to determine a cluster head
and communicate with it. These two schemes are energy efficient and produce
lesser delay than previous methods. The remainder of this paper is organized as
follows. Section 59.2 gives the existing works relevant to this paper. Section 59.3
describes the proposed work followed by the simulation and analysis in Sect. 59.4.

59.2 Related Work

The existing methods have various design issues. The Carrier Sense Multiple Access
(CSMA) was first introduced by Kleinrock and Tobagi [3] in which the hidden node
problem was later identified in [4]. The Busy Tone Mechanisms as in [4–7] consists
of a node that is currently hearing an ongoing transmission sends a busy tone to its
neighbors (on a narrow-band radio channel) for preventing them from transmitting
during channel use. The reasons why these mechanisms are not energy efficient are
that they need additional hardware to be able to provide separate radio channels and
they are not cost-effective. The RTS-CTS handshake mechanisms were also pro-
posed [8–14] with an aim to avoid hidden node collision problems. But clearly these
mechanisms do not completely solve the problem as stated in [15] because the
energy consumption is still high and the RTS/CTS handshake signals are smaller
frames and the probability of the hidden node problem reduction is not optimal.
Carrier Sense Tuning Mechanism consists of tuning the receiver sensitivity
threshold of the transceiver, which represents the minimum energy level that
indicates channel activity, to have extended radio coverage. Higher receiver sen-
sitivities enable a node to detect the transmissions of nodes farther away, thus
allowing it to defer its transmission (to avoid overlapping). The works that use this
mechanism were proposed in [16–19]. The limitation of this mechanism is that
59 An Efficient Beam Scanning Algorithm … 629

increasing receiver sensitivity directly leads to more energy consumption, which

might not be acceptable for most WSN applications. Even in situations where
energy consumption may not be a major concern, it is not possible to indefinitely
increase the carrier sense range due to hardware physical limitations [2].
A Grouping Strategy for Solving Hidden Node Problem in IEEE 802.15.4 LR-
WPAN was proposed in [20] that solves the hidden node problem quite well. The
grouping strategy enhances the IEEE 802.15.4 protocol to solve the HNP by pre-
cisely finding HNC and collect the hidden relationships among nodes using a four-
phase process. After grouping, the coordinator periodically allocates each group the
bandwidth according to the group size among all groups. The reserved channel
period for grouping access is named as hidden avoidance guaranteed time slots
(HA-GTS) and nodes access their own HA-GTS by applying the standard modified
CSMA/CA [20]. By controlling the transmissions of nodes, the grouping strategy
can significantly relieve the contentions and improve the network performance. The
limitations of this grouping process are: Complexity is high because it needs O (n2)
messages to collect information; grouping is performed at each time when a new
node arrives and it does not take number of nodes in each group into consideration.
H-NAMe [2] proposed an efficient, practical and scalable approach for syn-
chronized cluster-based WSNs—H-NAMe. Importantly, it showed how to inte-
grate the approach in the IEEE 802.15.4/ZigBee protocols with only minor add-
ons and fully respecting backward compatibility which is better than [11]. First, H-
NAMe requires no hidden-node detection since it relies on a proactive approach
(grouping strategy is node-initiated) rather than a reactive approach to the hidden
node problem. Second, the complexity of the group join process was drastically
reduced. In this approach, for each group assignment, only the requesting node and
its neighbors will be subject to the group join procedure and not all cluster nodes,
resulting in a simpler, more energy-efficient and scalable mechanism, especially
appealing for more densely deployed clusters.
However, a design issue has been identified in the design of H-NAMe that
arouses the following drawbacks: Hidden node relationship table construction is
unreliable due to the fact that the neighbor nodes not able to hear group join
request from the new node because they would be in the sleep mode. Therefore
HNC table would not be reliable and this leads to a wrong group formation and
results in collision. Complexity is high because it needs O(n2) messages to collect
information, grouping is performed at each time when a new node arrives and it
does not take number of nodes in each group into consideration.

59.3 Proposed Work

Two methods of grouping mechanisms are proposed to address the design issues in
H-NAMe. Intra-Cluster Grouping Scheme (IC-GS) aims at eliminating the unre-
liability grouping in H-NAMe and the IC-GS with beam scanning proposes a novel
beam scanning algorithm that proves to be better than IC-GS.
630 M. Sujatha and R.S. Bhuvaneswaran

59.3.1 Intra-Cluster Grouping Scheme

The Intra-cluster grouping scheme is an efficient and reliable grouping scheme

initiated by node. This is a proactive approach where, the process does not wait for
a hidden node collision problem to occur, but tries to avoid such a problem before
occurrence. Coordinator sends beacon signal to all the nodes in coverage area in
order to wakeup nodes. Thereby it meets a reliable information collection process.
It consists of four phases:
Hidden node information collection phase. In this phase all the nodes collect
information of the nodes hidden and non-hidden to it. The incoming node sends a group
joining request to the coordinator as a first step. Coordinator sends Beacon_2 to all
nodes. Beacon_2 contains two important messages: (1) Beacon broadcaster role will
performed by new node; and (2) The time, when new node would broadcast Beacon_3 is
also a part of the information in Beacon_2. Then the coordinator sends Beacon_2_data
information signal to new node, the new node needs to acknowledge it. Beacon_2_data
information carries necessary information to be broadcasted and the time when new
node can transmit Beacon_3. After new node receives Beacon_2_data information
without error it sends Ack signal back to the coordinator. The third step in the hidden
node collection phase is the Neighbor notification phase. New Node sends the Beacon_3
and triggers the timer which is used to wait for coordinator collecting responses
(Beacon_2 ack) from nodes. Beacon_3 is like the normal beacon but broadcaster here is
the new node and the receiving nodes sends ack signal to coordinator (Fig. 59.1).
Grouping Phase. The grouping phase is the stage at which the coordinator
assigns a new group to the newly joining node based on the hidden node

Fig. 59.1 An example scenario where node 8 is the new node and CH is the coordinator
59 An Efficient Beam Scanning Algorithm … 631

information table. Coordinator allocates a group to a new node, if all nodes in

group are not hidden to the new node. It gives priority to the group having
minimum of nodes. If none of a group existed with non hidden nodes, then a new
group is assigned to the new node.
The grouping algorithm used in this grouping phase is given in the following
section. A hidden node information table of the new coming nodes (T) and the
number of the existed group (K) are the inputs given to the system. The expected
output is that the new node is allocated to one of the group sets Gi.

group set G={G1, G2,……Gg}, g K;

while |T|>0 do
Pick a node X having the maximal number of
hidden nodes from T;
if no existed group then
K=1;
construct group set Gk={X};
T = T - {X};
return Gk;
else
while |G|>0 do
pick a group G’ having the minimal
number of nodes from G;
if node X has no edge to any nodes in G’
then
G’ = G’ + {X}; // join into group G’
T = T - {X};
return G’;
else
G= G- G’;
end if
end
K++;
T = T - {X};
construct group set Gk = {X};
return Gk;
end
end
632 M. Sujatha and R.S. Bhuvaneswaran

Fig. 59.2 Group slots in the IC-GS scheme

Group access period allocation. The group access period allocation has to be
determined for the efficient functioning of this algorithm which is given by the
following formula.

Ibi ðNi Þ  Ni
TGi ðNi Þ ¼ L
Pk
Ibi ðNi Þ  Ni
i¼0
" # ð59:1Þ
X 3
j
X
j1
Ibi ðNi Þ ¼ Pi f1  Pi g Ij;s þ Ix;f
j¼0 x¼0

Assume that the active period excluding the minimal contention access period
(CAP) and the Contention Free Period (CFP) is L in Unit Backoff Period (UBP)
and there are total M nodes which can be separated into K groups. Groupi has Ni
devices, access period is TGi, and the probability of a frame successfully trans-
mitted is Pi, for i = 1, 2…, k. Let wj be the size of the contention window (CW)
for a frame transmission in the jth backoff retries, and the basic CW unit is UBP.
  Ni
Pi ¼ ek , where Ni is the total number of nodes in group i and k is the
frame arrival rate in the system. Since we want to balance the transmission
opportunity of all nodes, we let each group have the same transmission times and
let each node has the same probability of transmission. And the sum of group
access period is, (Fig. 59.2)

TG1 þ TG2 þ TG3 þ . . .TGk ¼ L: ð59:2Þ

In this phase new node is informed about its group by the coordinator.
Therefore, new node can contend to transmit its data in a particular group access
period. This is the final phase of the IC-GS scheme and the node is admitted as part
of the WSN under a particular coordinator in a suitable group. There exists a
limitation that has been identified in this method. When a new node requests to
join a WSN by sending a request signal to all the coordinators present in the
network, it is possible to still encounter a collision there. And there is a good
amount of energy wastage because of more than one cluster head communicating
to the new node.
59 An Efficient Beam Scanning Algorithm … 633

Fig. 59.3 Beam scanning in the IC-GS scheme

59.3.2 IC-GS with Beam Scanning

To eliminate the design issue occurred in the IC-GS scheme, a beam scanning
method is used to scan the area at a fixed angle of 60 to find a coordinator that
allocates a group in the cluster as a member in its groups (Fig. 59.3).
The new node first broadcasts (360) the beacon signal to collect the infor-
mation about available cluster heads and their respective weight (based on mini-
mum number of nodes and maximum energy). After that it creates a cluster head
list. It then selects a cluster head from the cluster head list. Next it steers antenna to
60(beam scanning) and checks the location of selected cluster head by trans-
mitting a beacon signal. If the new node finds the selected cluster head, then it
starts the group joining process with the selected cluster head.

59.4 Simulation and Analysis

Simulation is performed by using the simulator NS2. Network simulator is a

discrete event time driven simulator that has been extensively used for research
purposes, especially for simulation of network scenarios and structures.
The IC-GS scheme and IC-GS with beam scanning are simulated using NS-2.28
with the parameters shown in Table 59.1. The existing H-NAMe scheme serves as
a standard for comparison of the two schemes proposed here. Figure 59.4 is a
634 M. Sujatha and R.S. Bhuvaneswaran

Table 59.1 Simulation parameters for the simulation of IC-GS-with beam scanning
S.No Parameter Value
1 Simulation area 1000 9 1000
2 Number of nodes 50
3 Min. no. of nodes in a cluster 10
4 Simulation time 60 ms
5 Radio propagation model Two ray ground type
6 Antenna type Omni antenna
7 Initial energy 100 KJ

Fig. 59.4 New node joins into cluster group

screenshot of the Network animator (NAM) output obtained during simulation of

the proposed schemes. It shows that a new node enters into a wireless sensor
network, where all nodes of the same color form a cluster, with the cluster heads
highlighted using a hexagon mark. Grouping has been done according to the
59 An Efficient Beam Scanning Algorithm … 635

grouping algorithm proposed in the earlier sections. Then it sends the Beacon_2
signal to all cluster heads in that network. An acknowledgement is received from
all the nodes in response to the beacon signal. It shows that the new node joins the
cluster group according to the proposed scheme. Then it will start functioning as a
normal member of the cluster belonging to one of its subgroups.

59.4.1 Packet Receive Ratio

See Fig. 59.5.

59.4.2 Throughput

See Fig. 59.6.

59.4.3 End-to-End Delay

In Fig. 59.7, the total time consumed in the process by which a node admitted into
the group is calculated by the delay occurred in the system. The graph that follows
shows that the overall delay caused by IC-GS with beam scanning is lesser than the
IC-GS scheme. IC-GS with beam scanning is the better among all the three

Fig. 59.5 Packet receive Packet Recieved Ratio

ratio 3500
Packets Recieved (Packets)

3000

2500

2000

1500

1000

500

0
0
2.7
5.4
8.1
10.8
13.5
16.2
18.9
21.6
24.3
27
29.7
32.4
35.1
37.8

Time (ms)
IC-GS with beam scanning
IC-GS H-NAMe
636 M. Sujatha and R.S. Bhuvaneswaran

Fig. 59.6 Throughput Throughput

3500

Throughput (packets)
3000

2500

2000

1500

1000

500

0
2.3
4.6
6.9
9.2
11.5
13.8
16.1
18.4
20.7
23
25.3
27.6
29.9
32.2
34.5
Time (ms)

IC-GS with beam scanning

IC-GS H-NAMe

Fig. 59.7 End-to-end delay Delay

25

20
Delay (ms)

15

10

0
0
3
6
9
12
15
18
21
24
27
30
33
36
39

Time (ms)

IC-GS with beam scanning

IC-GS H-NAMe

schemes for delay reduction. Also, it is significant to mention that the IC-GS
scheme finds a solution for a new node to join a group faster than the
other schemes. Hence the delay is comparatively less for the IG-CS scheme with
beam scanning.
59 An Efficient Beam Scanning Algorithm … 637

59.4.4 Energy Analysis

The energy remaining in a node during network operations is called as residual

energy. The graph in Fig. 59.8 shows that the residual energy of the IC-GS scheme
remains greater compared to the other schemes. This is due to the fact that energy
consumption is less during the beam scanning phase of the new node. Energy
reduction takes place for the nodes that perform sensing, transmit and receive
operations including those that perform routing operations. Even though the
experiment was run for around 50 runs, for clarity, the screenshot of the initial run
is presented in Fig. 59.8.

Fig. 59.8 Residual Energy

638 M. Sujatha and R.S. Bhuvaneswaran

59.5 Conclusion

Two schemes, IC-GS and IC-GS with beam scanning have been proposed, sim-
ulated and analyzed. To minimize the energy consumption during the new node
joining process, a beam scanning mechanism was introduced in the IC-GS scheme,
which outperforms the IC-GS scheme proposed and also the existing H-NAMe
method. The simulation results have proved that IC-GS with beam scanning offers
improved and better Quality of service.
In future, mobility will be taken into consideration for the IC-GS and the IC-GS
with beam scanning schemes.

References

1. Bachir, A., Barthel, D., Heusse, M., Duda, A.: Hidden nodes avoidance in wireless sensor
networks. In: Proceedings of the International Conference on Wireless Networks,
Communications and Mobile Computing, 2005
2. Koubâa, A., Severino, R., Alves, M., Tovar, E.: Improving quality-of-service in wireless
sensor networks by mitigating hidden-node collisions. IEEE Trans. Ind. Inform. 5(3),
299–313
3. Kleinrock, L., Tobagi, F. A.: Packet switching in radio channels: part 1—carrier sense
multiple-access modes and their throughput-delay characteristics. IEEE Trans. Commun.
COM-23(12), 1400–1416 (1975)
4. Tobagi, F.A., Kleinrock, L.: Packet switching in radio channels: Part II—the hidden terminal
problem in carrier sense multiple-access and the busy-tone solution. IEEE Trans. Commun.
23, 1417–1433 (1975)
5. Michalewicz, Z.: Genetic algorithms + Data structures = Evolution programs, 3rd edn.
Springer, Berlin (1996)
6. Haas, Z.J., Deng, J.: Dual busy tone multiple access (DBTMA)—a multiple access control
scheme for ad hoc networks. IEEE Trans. Commun. 50, 975–985 (2002)
7. Chandra, A., Gummalla, V., Limb, J.O.: Wireless collision detect (WCD): multiple access
with receiver initiated feedback and carrier detect signal. In: Proceedings of the IEEE ICC,
pp. 397–401 (2000)
8. Ji, B.: Asynchronous wireless collision detection with acknowledgement for wireless mesh
networks. In: Procedings of the IEEE Vehicular Technology Conference, vol. 2, pp. 700–704,
Sept 2005
9. Tobagi, F.A., Kleinrock, L.: Packet switching in radio channels: Part III—polling and
(dynamic) split channel reservation multiple access. IEEE Trans. Comput. 24(7), 832–845
(1976)
10. Karn, P.: MACA—A new channel access method for packet radio. In: Proceedings of the 9th
ARRL/CRRL Amateur Radio Computer Network Conference, pp. 134–140 (1990)
11. Bharghavan, V., Demers, A., Shenker, S., Zhang, L.: MACAW: a media access protocol for
wireless LAN’s. In: Proceedings of the ACM SIGCOMM, pp. 212–225, London, U.K., Aug
1994
12. Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY)
Specifications, ISO/IEC IEEE-802-11, IEEE Standard for Information Technology (1999)
13. Fullmer, C.L., Garcia-Luna-Aceves, J.J.: Solutions to hidden terminal problems in wireless
networks. In: Proceedings of the ACM SIGCOMM, pp. 39–49, Cannes, France, Sept 1997
59 An Efficient Beam Scanning Algorithm … 639

14. Yang, Y., Huang, F., Ge, X., Zhang, X., Gu, X., Guizani, M., Chen, H.: Double sense
multiple access for wireless ad hoc networks. Int. J. Comput. Telecomm. Netw. 51(14),
3978–3988 (2007)
15. Xu, K., Gerla, M., Bae, S.: How effective is the IEEE 802.11 RTS/CTS handshake in ad hoc
networks. In: Proceedings of the Global Telecommunication Conference (GLOBECOM’02),
vol. 1, pp. 72–76 (2002)
16. Deng, J., Liang, B., Varshney, P.K.: Tuning the carrier sensing range of IEEE 802.11 MAC.
In: Proceedings of the IEEE Global Telecommunication Conference (GLOBECOM), vol. 5,
pp. 2987–2991 (2004)
17. Ye, F., Yi, S., Sikdar, B.: Improving spatial reuse of IEEE 802.11 based ad hoc networks. In:
Proceedings of the IEEE Global Telecommunication Conference (GLOBECOM’03), vol. 2,
pp. 1013–1017, San Francisco, CA, Dec 2003
18. Zhai, H., Fang, Y.: Physical carrier sensing and spatial reuse in multirate and multihop
wireless ad hoc networks. In: Proceedings of the IEEE INFOCOM, pp. 1–12, Apr 2006
19. Ho, I., Liew, S.: Impact of power control on performance of IEEE 802.11 wireless networks.
IEEE Trans. Mob. Comput. 6(11), 1245–1258 (2007)
20. Hwang, L.-J., Sheu, S.-T., Shih, Y.-Y., Cheng, Y.-C.: Grouping strategy for solving hidden
node problem in IEEE 802.15.4 LR-WPAN. In: IEEE, Proceedings of the First International
Conference on Wireless Internet (WICON’05), 2005
Chapter 60
Evaluation of Stereo Matching Algorithms
and Dynamic Programming for 3D
Triangulation

Teo Chee Huat and N.A. Manap

Abstract A good result of triangulation or known as Three-Dimensional (3D) is

depending on the smoothness of the disparity depth map that obtained from the
stereo matching algorithms. The smoother the disparity depth map, the better
results of triangulation can be achieved. This paper presents the evaluation of the
existing stereo matching algorithms in the aspects of the speed of computational on
depth map obtained. The stereo matching algorithms that we applied for experi-
mental purpose are basic block matching, sub-pixel accuracy and dynamic pro-
gramming. The dataset of stereo images that used for the experimental purpose are
obtained from Middlebury Stereo Datasets. This research is to provide an idea on
choosing the better stereo matching algorithms to work on the disparity depth map
for the purpose of 3D triangulation applications, as the good result of 3D trian-
gulation is depending on how smooth is the disparity depth map can be obtained.

Keywords Dynamic programming  3D triangulation  Stereo matching algo-


rithms Depth map

60.1 Introduction

Stereo matching algorithms are the procedure to obtain a disparity depth map from
the corresponding pixels from the pair of stereo images. Many existing stereo
matching algorithms developed by researchers on image processing field as it is an
important function in computer vision technology to analyze the two dimensions,
(2D) and three dimensions, (3D) of output based on stereo images. In developing

T.C. Huat (&)  N.A. Manap

Faculty of Electronics and Computer Engineering, Department of Computer Engineering,
Universiti Teknikal Malaysia Melaka, Melaka, Malaysia
e-mail: [email protected]
N.A. Manap
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 641

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_60
642 T.C. Huat and N.A. Manap

an algorithm of stereo matching, the accuracy and speed are conditions that need to
be concerned to produce a precise output of computer vision. As to get better
results of disparity on stereo images, there are many stereo matching methods
evaluated on the stereo datasets for this research such as dynamic programming,
basic block matching, sub-pixel estimation, and image pyramiding. The datasets
used for the experimental on the stereo matching algorithms are from the Mid-
dlebury Stereo Vision Page which is an open source of datasets and evaluation on
stereo vision algorithms.
The existing stereo matching algorithms mostly will go through the standard
steps like the matching cost computation, cost aggregation, disparity optimization
and disparity refinement. There are some equations can be included to find the
pixel-based matching cost, which are the Sum of Absolute Differences (SAD),
Sum of Squared Difference (SSD), Sum of Truncated Absolute Differences
(STAD), Normalized Cross Correlation (NCC) and Zero Mean Normalized Cross
Correlation (ZMNCC). In this particular research, the SAD equation is used to get
the corresponding points between the original and target of stereo images. The cost
of the correspondence between the original and target data set can be represented
by Disparity Space Image (DSI) [1–3]. The DSI is formed from the matching cost
values from the disparities and pixels when summing up the cost values on each
matching image data set [4].
In order to find the disparity depth map, the basic blocking matching is used as it
is the initial step to find the absolute difference between the pixel intensities by
using the Sum of Absolute Differences (SAD) to compare the correspondence
pixels on the left and right of stereo images [1, 2, 4]. One of the examples of dataset
used for stereo matching experimental is the Tsukuba stereo pairs, taken from
Middlebury Stereo Datasets. The disparity map is calculated using three sizes of
window which are 3 by 3, 5 by 5 and 7 by 7 pixel of block around the original of
image data. In basic block matching, it locates the pixels in the range of ±15 over
columns sequence [5]. On the step of cost aggregation, there are two types of
support region which are the two dimensional, 2D fixed disparity and the three
dimensional, 3D with the variables of x, y and d as in Eq. 60.1. The 2D aggregation
is by using square windows, shiftable windows and windows with adaptive sizes.
Furthermore, the disparity optimization can be categorized into two approaches
which are the local approaches which use the Winner Takes All (WTA) by picking
each pixel where the disparity is correlated to the minimum cost value in order to
increase the signal to noise ratio (SNR) while to reduce the ambiguity. Another
approach which is the global approach, it is a frame work to find for disparity
function, d that minimize an energy function or global energy over the disparity
computation phase like the pixel-based matching cost [1],

Eðd Þ ¼ Edata ðd Þ þ Esmooth ð60:1Þ

where Edata(d) represent the disparity function with the input of data set which
minimized through the pixels on the correspondence of disparity map, d if there is
similarity in intensities and maximized when the disparity map putting the pixels
60 Evaluation of Stereo Matching Algorithms … 643

in correspondence which slightly differ in intensities. The Esmooth represent the

conjecture of the smoothness made from the algorithm which measured from the
disparity between the pixels on pixel grid [1, 2, 4].
On disparity refinement, there are many approaches to improve the smoothness
of the disparity map. In this research, the method of dynamic programming is
chosen as the approach to improve the output obtained from the basic block
matching stereo correspondence algorithm. Dynamic programming is able to find
the global minimum of independent scanlines when in polynomial time for
smoothness functions [1]. It is chosen based on its accuracy when dealing at the
areas of depth borders and uniform regions [6, 7].

60.2 Algorithm Outlines

60.2.1 Basic Block Matching

Block matching is a technique that used to find the corresponding pixels in stereo
matching datasets. The basic block matching is used in this research as the initial
step to find corresponding pixels as of the dataset that used throughout the
experimental of stereo matching. During the experimental phase, the pixel value of
the target image is predicted as the corresponding pixel in the reference image
where the displacement of the corresponding pixels or as the motion vector to be
approximated using the block matching [8]. The block matching is used to reduce
the matching errors between the block at position of (x, y) in the target image, It
while for the position of the reference image, It-1 will be (x + u, y + v) where
u and v is the motion vector. These variable defined can be reviewed as sum of
absolute difference (SAD) [8],

S1 X
X S1
ðu; vÞ  jIt ðx þ i; y þ jÞ  It1 ðx þ u þ i; y þ v þ jÞj ð60:2Þ
j¼0 i¼0

where S is the block size, S 9 S, i and j representing the pixels. As to minimize the
SAD(X, Y)(u, v), (a, b) is defined as the motion vector estimation to compare and
obtain the SAD of each position, ðx þ u; y þ vÞ for the dataset. The equation shows
as [8],

ða; bÞ  arg min SADðx;yÞ ðu; vÞ ð60:3Þ

ðu;vÞ2Z

where Z = {(u, v)| - B B u, v B B and ðx þ u; y þ vÞ represent the suitable

position of pixel in the reference image, It-1} while B is an integer to seek for
range. From the SAD equation, the global minimum of matching error can be
attained. Block matching algorithm able to save more computations to obtain a
depth map but it still have its disadvantage which is its disability to guarantee the
644 T.C. Huat and N.A. Manap

global matching error. Besides that, the main issue with block matching algorithm
is that the prediction on the position of the images sequence is not accurate enough
and this may cause on the minimum matching error will be larger. In addition, the
most critical issue is that the matching errors are decreasing directly proportionally
to the coming order of the search positions.

60.2.2 Sub-Pixel Accuracy

Most of the stereo matching algorithms apply sub-pixel refinement or the sub-pixel
accuracy after attaining the correspondence points of the stereo datasets. Sub-pixel
accuracy is a technique that goes through isolated disparity levels that from the
matching cost and also the iterative gradient descent [9]. The major point of the
sub-pixel accuracy method is to enhance the resolution of the stereo matching
algorithm output from the stereo datasets. This method is able to smooth the
alteration between the regions from different disparity that cause contouring effect
on the images on a depth map. Through the progression of sub-pixel accuracy, it
will focus on the minimum cost and the neighboring cost values to acquire the sub-
pixel alteration. In applying the technique of sub-pixel accuracy, the Normalized
Cross Correlation (NCC) is used to work out the essential stereo images where the
cross correlations at the sub-pixel location of the stereo images can be figured
proficiently and the equation used for computation is shown as following where
the NCC (x, y, u, v) is equal to [9],
P
ði;jÞ2w I1 ðx
þ i; y þ jÞ  I2 ðx þ u þ i; y þ v þ jÞ
pffi ð60:4Þ
½2I12 ðx; yÞ  I22 ðx þ u; y þ vÞ

where the NCC can be defined as left image window at the position (x, y) while for
the right image window position at (x + u, y + v). The I1 and I2 are represented as
target image and reference image. From the NCC equation, sub-pixel accuracy can
be computed with NCC through integral images as it can be substituted by using
integral stereo images which with squared of pixel values. There are limitations
can be found from sub-pixel accuracy algorithm as it unable to determine the local
ambiguities successfully. Another issue on sub-pixel accuracy is it causes the right
matching unrecognized to be count in for local maximum and this bring failures
for the usage of local maxima [9].

60.2.3 Dynamic Programming

For the disparity optimization phase, the dynamic programming is selected as

global optimization, which is optimizes energy function to be non-deterministic
polynomial-time hard (NP-hard) for smoothness purpose [1, 6]. There are two
60 Evaluation of Stereo Matching Algorithms … 645

categories of global optimization such as one dimension and two-dimension

optimization methods. One dimension optimization is traditional technique, where
its evaluation on the disparity is focusing on a pixel that depending on other pixels
on the same scanline, but independent on disparity that focus on other scanlines.
One dimension is not considered as a truly global optimization as its smoothness
technique is only focus on horizontal direction. However, one dimension opti-
mization is still being used by some of the researchers due to its simple imple-
mentation and its usefulness on the disparity maps outputs.
Two-dimension optimization approach is smoothing the stereo images in the
vertical and horizontal direction by using simulated annealing, continuation
methods and mean-field annealing [10–12]. However, these methods are not well-
organized enough to optimize the equation in (60.1). There are two methods which
companionable in optimizing the Eq. (60.1), the graph-cuts and belief propagation
as these two methods able to attain better results accordingly to ground truth data
from stereo matching algorithm [13–15] In this paper, the dynamic programming
used for the experimental results is the dynamic programming on tree due to its
competence as the one dimension optimization. The tree graph of the dynamic
programming can represent as T (V, E) where V as vertices and E as edges. The
effectiveness of the dynamic programming on tree begins with its optimization on
the energy function [6],
X X
E ðd Þ ¼ mðda Þ þ k sðda ; db Þ ð60:5Þ
a2V ða;bÞ2E

where a is the pixel in the left image and the da as the value of disparity map, d at
the pixel of a. Assuming mðda Þ is the matching consequence of relating da to the
pixel of a where it is the absolute difference between the pixel, a in the left image
and the a pixel which shifted to the right image and can be summarized as
P
a2V mðda Þ. Meanwhile, assume sðda ; db Þ as the smoothness consequence for the
disparity
P of da and db to the pixel a and b and the variables can be summarized as
ða;bÞ2E sðda ; db Þ.
As to get the minimum energy of Eq. (60.5), let h as the root vertex of tree,
h [ V and assume z [ V as the number of edges the root of distance between h and
z. Each node of z belongs a parent as a(z) and the depth is equally to the depth of
z-1 while if it is not a root, the minimum value of the energy in Eq. (60.5) have a
sub-tree rooted at z and the edge in the middle of z and a(z) can be summarized as
daðzÞ [6],
    X
Ez daðzÞ ¼ minðmðdz Þ þ s dz ; daðzÞ þ Ew ðdz Þ ð60:6Þ
dz 2D
w2Cz
646 T.C. Huat and N.A. Manap

where Cz as the children set of z. while for the optimal disparity for the root node
h can be represented as [6],
X
Lh ¼ arg minðmðdh Þ þ Ew ðdh ÞÞ ð60:7Þ
dh 2D
w2Ch

where if z is a node that without children then Cz is empty and the function of Ez
and Lz can be estimated directly. h* shows that the variable can be replaced with
z and w. Let take J as the maximum depth in the tree, the energy function of
Eq. (60.5) is optimized by appraising the functions Ez and Lz for each node z at the
depth, J. After estimation on the functions, proceed with the estimation on the
same functions for all the nodes at depth of J - 1 due to any child w has the depth
of J, this is the evaluation on Ew and Lw. Next step is to keep appraising the
function of Ez and Lz in decreasing order for the depth till it reach to the root for
disparity assignment optimal computation use. Dynamic programming on a tree is
an algorithm that simple to be implemented and efficient as well on the results
obtained compared to most of the stereo matching algorithms but there still a
disadvantage of it which is its speed is slower due to its tree traversal is less
capable than its array traversal [6]. Therefore, based on the characteristics of the
three stereo matching algorithms described, dynamic programming has been
selected as the main stereo matching algorithm due to its accuracy and its capa-
bility in smoothing most of noise in the datasets of stereo images compared to the
other two algorithms, basic block matching and sub-pixel accuracy.

60.3 Experimental Results

There are three existing stereo matching algorithms evaluated in this paper which
are the basic block matching, sub-pixel accuracy, and dynamic programming. From
the three stereo matching algorithms, dynamic programming is the well-organized
algorithm among the other algorithms in smoothing the disparity depth map. The
helpfulness of dynamic programming in smoothing the depth map is also depending
on the appropriate disparity range (DR) of the stereo pair of images used. The stereo
pairs of images that used in this research are Tsukuba, Teddy, Sawtooth and Venus
which these stereo pairs of images are chosen from Middlebury Stereo Pages. The
results obtained are shown as Table 60.1, Figs. 60.1 and 60.2.
From Table 60.1, the disparity range of each stereo images pair is determined
based on the smoothness of the basic depth map that obtained from the stereo
matching algorithms. The biggest number of the disparity range is most preferable
to be used for stereo matching algorithms process, for example Sawtooth stereo
pairs sample from Table 60.1, the suitable and the most preferable disparity range
to obtain a good result is by using 30, while for Tsukuba is by using 16. The
interval range that provided as results in Table 60.1 are to show that few number
of disparity range on the range can obtain slightly good results.
60 Evaluation of Stereo Matching Algorithms … 647

Table 60.1 Disparity range

Stereo images Tsukuba Teddy Sawtooth Venus
Disparity range 15–16 45 27–30 30

500
Teddy
450
Venus
400
Sawtooth
350
Tsukuba
300
250
200
150
100
50
0
Basic Block Sub-pixel accuracy Dynamic
Matching Programming

Fig. 60.1 Time taken (in second) for different sets of stereo image applied on each type of stereo
matching algorithm. For the dataset of Teddy used the disparity range of 45, Venus used disparity
range of 30, Sawtooth used 29 and Tsukuba used 16 as the disparity range

From the evaluation on the results of three stereo matching algorithms, the
precision which depends on smoothness in rising order is starting from sub-pixel
accuracy followed by basic block matching and the smoothest results are obtained
by the dynamic programming algorithm as revealed in Fig. 60.2. The hardware
used to run the simulation of the three stereo matching algorithms is the portable
computer with integrated of processor of 2.5 gigahertz (GHz) and three gigabytes
(GB) of installed memory. Table 60.1 shows on the disparity range for each
dataset of stereo image where the disparity range are achieved from the experiment
by applying stereo matching algorithms and observing on the output accuracy for
different disparity range applied on every dataset coding. Figure 60.1 shows the
results of the time taken in second for each stereo matching algorithm on different
sets of stereo images, where the computation competence of the dynamic pro-
gramming is the lowest among the stereo matching algorithms. Meanwhile the
highest computation effectiveness of stereo matching algorithms is the basic block
matching where its average time taken for all the stereo images datasets is more
rapidly than the average time taken of sub-pixel accuracy algorithm. Besides that,
from the results acquired in Fig. 60.1, it also shown that the higher disparity ranges
of stereo images datasets, the longer time taken to run on the stereo matching
algorithms.
648 T.C. Huat and N.A. Manap

Stereo Basic block matching Sub - pixel accuracy Dynamic Programming

images
Tsukuba
(disparity
range =
16)

Teddy
(disparity
range =
45)

Sawtooth
(disparity
range =
29)

Venus
(disparity
range =
30)

Fig. 60.2 Results achieved from stereo matching algorithm

60.4 Discussion

From observation on each stereo matching algorithms, it is found that each of it

have the issues faced or the imperfect of the algorithms developed. Therefore, for
each issue that been through, we will try to overcome it by modifying the equa-
tions used on each stereo matching algorithm. For example, in block matching
algorithm we can attempt to combine some other equations such as the two
dimension optimization to search for the corresponding pixels in horizontal and
vertical so that the prediction on the position of images sequence are much more
accurate than before. Besides that, for sub-pixel accuracy as it is used to improve
the resolution, it can be added some part of algorithm in improving the contouring
effect by applying the suitable disparity range. Therefore, by pairing with the
algorithm with sub-pixel accuracy, the dataset can be doubled up in smoothing the
ambiguities. For dynamic programming, as the results obtained are the most
60 Evaluation of Stereo Matching Algorithms … 649

accurate among the other stereo matching algorithms, the only issue that we
observed is the speed is slower than the other stereo matching techniques. How-
ever, the time taken is a minor issue for the scope of this research.

60.5 Conclusion

As a conclusion, the evaluation between the three stereo matching algorithms can
clearly shows that dynamic programming is the most capable technique between
the others in smoothing the depth map. It is depending on the appropriate disparity
for diverse content of stereo pair of images used. Besides that, from the experi-
mental outcomes by using the Middlebury datasets shows that dynamic pro-
gramming algorithm able to reduce the matching errors and acquire an improved
stereo matching result compare to the basic block matching and sub-pixel accuracy
algorithm. Therefore, based on the comparison among the three stereo matching
algorithms, it is suggested that dynamic programming could helps to obtain more
agreeable effect of depth map especially in removing the apparent defective
stripes. The depth map that obtained from the dynamic programming algorithm is
effective for most applications compared to basic block matching and sub-pixel
algorithm such as appliances on 3D intentions.

Acknowledgements This research is funded by the grant from Faculty of Electronic and
Computer Engineering, Universiti Teknikal Malaysia Melaka (UTeM) with project number of
PJP/2013/FKEKK (16C)/S01203.

References

1. Scharstein, D., Szeliski, R.: A taxonomy and evaluation of dense two-frame stereo
correspondence algorithms. Technical Report MSR-TR-2001-81, Microsoft Research (2001)
2. Bleyer, M., Gelautz, M.: A layered stereo matching algorithm using image segmentation and
global visibility constraints. J. Photogram. Remote Sens. 59(3), 128–150 (2005)
3. Di Stefano, L., Mattoccia, S., Mola, M.: An efficient algorithm for exhaustive template
matching based on normalized cross correlation, IEEE Comput. Soc. 2, 322–327 (2003)
4. Tombari, F., et al.: A 3D Reconstruction System Based on Improved Spacetime Stereo. In:
IEEE, pp. 7–10 (2010)
5. Birchfield, S., Tomasi, C.: Depth discontinuities by pixel- to- pixel stereo. In: ICCV,
pp 270–293 (1998)
6. Veksler, O.: Stereo correspondence by dynamic programming on a tree. CVPR 2, 384–390
(2005)
7. Olofsson, A.: Modern stereo correspondence algorithms: investigation and evaluation. PhD
thesis, Linkoping University, Sweden, 5–86, June 2010
8. Chen, Y., Hung, Y., et al.: Fast block matching algorithm based on the winner update
strategy. IEEE 10(8), 1212–1222 (2001)
9. Donate, A., Wang, Y., et al.: Efficient and accurate subpixel path based stereo matching.
IEEE 8(6), 1–4 (2008)
650 T.C. Huat and N.A. Manap

10. Geiger, D., Girosi, F.: Parallel and deterministic algorithms from MRF’s: surface
reconstruction. IEEE Trans. Pattern Anal. Mach. Intell. 13(5), 401–412 (1991)
11. Geman, S., Geman, D.: Stochastic relaxation, Gibbs distributions, and the Bayesian
restoration of images. IEEE Trans. Pattern Anal. Mach. Intell. 6, 721–741 (1984)
12. Blake, A., Zisserman, A.: Visual Reconstruction. MIT Press, Cambridge (1987)
13. Kolmogorov, V., Zabih, R.: Multi-camera scene reconstruction via graph cuts. In: ECCV02,
p. III: 82 ff. (2002)
14. Felzenszwalb, P., Huttenlocher, D.: Efficient belief propagation for early vision. In: CVPR04,
p. I: 261–268
15. Sundstr, O., Guzzella, L.: A generic dynamic programming matlab function. In: IEEE
International Conference on Control Applications.1(7), 1625–1630 (2009)
Chapter 61
Image Enhancement Filter Evaluation
on Corrosion Visual Inspection

Syahril Anuar Idris and Fairul Azni Jafar

Abstract This project is focusing on corrosion inspection using image. Inspection

which have particularly challenging environmental conditions and characteristics,
increase the complexity of the inspection operation. By using software image filter
to enhance the image data, it is believe that the object recognition technique will
be able to analyse the image data accurately. A few software filters have been
identified in this works based on textural feature and colour progression factor that
are the characteristic of image corrosion. Therefore, in order to obtain suitable
software image filter, neural network is use for optimization. The experiment result
shows among those identified image enhancement filters for visual corrosion
inspection, Wavelet De-noising gives desirable result in terms of Mean Square
Error, Peak Signal to Noise Ratio and Neural Network optimization.

61.1 Introduction

These days, utilization of camera as inspection tools has been expanding. The
flexibility functions of camera, fit to get different sorts of information, for example,
position, speed, rate of development and others from single apparatus. However, in
visual inspection the quality of raw image affect the accuracy of inspection result.
The quality of raw image can be enhanced externally or internally, by externally
means to control the inspection environment. Unfortunately for corrosion
inspection, inspection environment is hard to be controlled. Therefore to obtain

S.A. Idris (&)

Center of Graduate Studies, Universiti Teknikal Malaysia Melaka, Melaka, Malaysia
e-mail: [email protected]; [email protected]
F.A. Jafar
Faculty of Manufacturing Engineering, Universiti Teknikal Malaysia Melaka,
Melaka, Malaysia
e-mail: [email protected]

Ó Springer International Publishing Switzerland 2015 651

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_61
652 S.A. Idris and F.A. Jafar

good quality image for visual inspection, the inspection system need to be able to
enhance raw image internally, after the image acquisition process.
There are lot of methods for image enhancement process. One of the popular
enhancement method used for inspection is filtering [1]. In selecting suitable fil-
ters, an understanding of target inspection is required. Common question arise for
the selection of suitable image enhancement filter is ‘‘what kind of data needed to
be extract out from target?’’ By knowing the type of data required in inspection,
one can clustered the possible filter to be used for enhancement process.
This research work is focusing on selecting best image enhancement filter use
for visual corrosion inspection. The end-result should present the comparison of
performance between each recognized image enhancement filter, and identified the
best filter for enhancement corrosion image.

61.2 Visual Corrosion Inspection

One of the primary requisitions of vision system is application in inspection

system. By utilizing vision system as part of inspection system, the accuracy and
reliability of value could be setup during the predetermine range. Visual inspection
systems are utilized to check parts for dimensional accuracy and geometrical
integrity. Regardless of the fact that human able to perform image inspection using
their own eyes, the detail repetition of some inspection tasks is simply beyond the
capability of humans.
Remote Visual Inspection or RVI is one method of non-destructive testing
(NDT) use in corrosion inspection [2]. RVI is a visual examination method that
aids in acquisition of visual information by utilizing visual equipment but not
limited to video pan/tilt/zoom cameras, borescopes, push cameras, or automated
crawlers. It is frequently used where distance, angle of view and restricted lighting
may hinder direct visual examination or where access is constrained by time,
financial constraints or atmospheric hazards.
There are lot of different experimental and analysis methods used to identify
corrosion damage for inspection and monitoring purposes. One of the methods is
mechanical measurements (weight loss, chemical analysis, and visual inspections).
In visual inspections, the corrosion level identification requires expert who can
clearly determine the corrosion based on experience as well as types of corrosion,
with red rust as a common experience. Usually, the corrosion process produces
rough surfaces, and image analysis based on textural features can be used for
quantification and discriminate corrosion extent and type [3, 4]. Additionally to
textural features, colour progressions of metallic surfaces are also used for the
detection of corrosion because of different metal oxides and other corrosion
products [5].
61 Image Enhancement Filter Evaluation on Corrosion Visual Inspection 653

61.3 Image Enhancement Filter

As explain in previous topic, image analysis for corrosion can be identified by

textural feature and colour progression. Thus in enhancement image process,
selected filter must be able to preserve both characteristics of the image. The
image enhancement filters identified for this study emphasizes on preserving these
characteristics. Table 61.1 lists the image enhancement filter based on ‘‘How they
do it?’’, and ‘‘What do they enhance?’’
There are several methods in selecting best filter for image enhancement, and of
them is by calculating the image error measurement. Image error can be measure
by calculating Mean Square Error (MSE) and Peak Signal Noise-to-Ratio (PSNR).
However, for corrosion visual inspection, selecting one suitable filter only
based on image error measurement is not accurate. This is because, image error
measurement does not consider on textural feature and colour progression of
corrosion image. Therefore, neural network is used to determine the best filter for
corrosion visual inspection by taking consideration on preserving textural feature
and colour progression of corrosion image.

61.3.1 Image Error Measurement

The Mean Square Error (MSE) and the Peak Signal to Noise Ratio (PSNR) are the
two error metrics used to compare image compression quality. The MSE repre-
sents the cumulative squared error between the compressed and the original image,
whereas PSNR represents a measure of the peak error. The lower value of MSE
shows that the image error is low [12].
The PSNR block computes the peak signal-to-noise ratio, in decibels, between
two images. This ratio is often used as a quality measurement between the original
and a compressed image. The higher the PSNR value, the better the quality of the
compressed or reconstructed image [12].
To compute the PSNR, the block first calculates the mean-squared error using
the following equation:
P
 I2 ðm; nÞ2
M;N ½I1 ðm; nÞ
MSE ¼ ð61:1Þ
MN

In the Eq. (61.1), M and N are the number of rows and columns in the input
images, respectively. Then the block computes the PSNR using the following
equation:
 
R2
PSNR ¼ 10log10 ð61:2Þ
MSE
654 S.A. Idris and F.A. Jafar

Table 61.1 Image enhancement filter characteristics

Homomorphic Bayer Wavelet Gaussian [9] Linear [10] Anisotropic
[6] [7] de-noising diffusion [11]
[8]
Image Filter Reduces Give no Improve Reducing
enhancement mosaic, high overshoot to images in image noise
and correction colour frequency a step many ways: without
filter noise function sharpening removing
array input while edges, significant
(CFA) minimizing reducing parts of the
the rise and random noise, image content
fall time correcting
unequal
illumination,
etc.
Simultaneously Arrange Direct Impulse Based on the Image
normalizes the RGB wavelet response on same two generates a
brightness colour transform Gaussian techniques as parameterized
across an image filters on is function conventional family of
and increases a square computed DSP: successively
contrast grid of from the convolution more and
photo original and Fourier more blurred
sensors image analysis images based
on a diffusion
process

In the Eq. (61.2), R is the maximum fluctuation in the input image data type.
For example, if the input image has a double-precision floating-point data type,
then R is 1. If it has an 8-bit unsigned integer data type, R is 255.

61.3.2 Optimization: Neural Network

In this research, artificial neural network will be used for optimizing image
enhancement filter algorithm. As network representation provides such powerful
visual and conceptual aid for portraying the relationship between the components
or tools of systems that it is used in virtually every field of scientific, social, and
economic endeavor [13]. In artificial neural network, the elements called as neu-
rons or node, process the information. The signals are transmitted by means of
connection links, and these links possess an associated weight which is multiplied
along with the incoming signal (input) for any typical neural network. Finally the
output signal is obtained by applying activations to the network input [14].
Figure 61.1 represent neural network used for the optimization of the enhance-
ment filter selection. Each filter will be denoted by each single node on hidden layer.
Raw image will be the input and final selected filter image would be the output for
this network. The weight for each node is determined as shown in Table 61.2.
61 Image Enhancement Filter Evaluation on Corrosion Visual Inspection 655

Fig. 61.1 Neural network Hidden Layer

map for optimizing image
enhancement filter used on A
visual corrosion inspection

B
Input Layer Output Layer

Filter
Raw Image C Image

A : Homomorphic Filter B : Bayer Filter C : Gaussian Filter

D : Kalman Filter E : Wavelet De-noising F: Anisotropic Diffusion

From Table 61.2, the weight for Linear Filter is 0.29, Bayer Filter is 0.07,
Homomorphic Filter is 0.14, Gaussian Filter is 0.07, Wavelet De-Noising is 0.29
and lastly Anisotropic Diffusion is 0.14. The PSNR values obtain from each filter
will be multiplied with the weight of each filter. Highest value will be taken as best
filter for corrosion visual inspection

61.4 Experiment Result

The experiment of finding suitable image enhancement filter for corrosion visual
inspection has been tested on 3 corrosion images (corrosion 01, corrosion 02, and
corrosion 03) as in Fig. 61.2. The sample image use in the experiment is RGB
image and the size is converted to become square image for ease of filter instal-
lation on the image. Figure 61.3 shows the result of filtered image for each sample
image. In addition Table 61.3 shows the RMSE and PSNR value for each image
enhancement filter. Blue highlighted block in the table, show the maximum value
of PSNR compared to other filters. While red highlighted blocks show the highest
value for each PSNR value multiplied with weight that determined earlier. From
the Table 61.3, the suitable image enhancement filter for corrosion visual
inspection is Wavelet De-noising.
 656

Table 61.2 Weight determination for image enhancement filters

Filters Edge preserving Colour restoration De-noising Fix unequal illumination Sharpen image Total Weight, wi
Linear – 1 1 1 1 4 0.29
Bayer – 1 – – – 1 0.07
Homomorphic 1 – – 1 – 2 0.14
Gaussian – – 1 – – 1 0.07
Wavelet De-noising 1 1 1 – 1 4 0.29
Anisotropic Diffusion 1 – 1 – – 2 0.14
S.A. Idris and F.A. Jafar
61 Image Enhancement Filter Evaluation on Corrosion Visual Inspection 657

Fig. 61.2 Raw images (input)

Fig. 61.3 Filtered images (output)

61.5 Discussion

From the result, the selected suitable image enhancement filter for corrosion visual
inspection is Wavelet De-noising. In Table 61.3, each corrosion image tested in
the experiment, Wavelet De-noising gives the highest value compare to other
filters. Although in image corrosion 03, the highest PSNR value obtains by
Gaussian filter. However, due to the weight identified earlier for each filter,
Wavelet De-noising has the highest value and selected as best filter for corrosion
visual inspection.
Figure 61.4 shows the result for filtered images on image corrosion 03. Even
though the Gaussian filter image obtains highest PSNR, the textural feature of
corrosion is tempered, thus affect the corrosion extraction data, resulting for
possible un-accurate inspection result.
 658

Table 61.3 Result of image enhancement filter optimization on corrosion image

Filter Corrosion 01 Corrosion 02 Corrosion 03
PSNR RMSE Weight, Total PSNR RMSE Weight, Total PSNR RMSE Weight, Total
wi wi wi
Linear 61.232 0.22 0.29 17.757 57.435 0.34 0.29 16.656 60.438 0.24 0.29 17.527
Gaussian 62.197 0.20 0.07 4.354 58.158 0.32 0.07 4.071 60.757 0.23 0.07 4.253
Wavelet DN 62.217 0.20 0.29 18.043 58.295 0.31 0.29 16.905 60.443 0.24 0.29 17.529
Bayer 60.440 0.24 0.07 4.231 57.897 0.32 0.07 4.053 60.296 0.25 0.07 4.221
Homomorphic 60.672 0.24 0.14 8.494 57.195 0.35 0.14 8.007 60.690 0.24 0.14 8.497
Anisotropic 61.555 0.21 0.14 8.618 56.631 0.38 0.14 7.928 60.745 0.23 0.14 8.504
DF
Max Min Max Min Max Min
PSNR RMSE PSNR RMSE PSNR RMSE
62.217 0.20 58.295 0.31 60.757 0.23
S.A. Idris and F.A. Jafar
61 Image Enhancement Filter Evaluation on Corrosion Visual Inspection 659

Fig. 61.4 Filtered image for image corrosion 03

61.6 Conclusion

This paper identified several image enhancement filters, applicable in corrosion

inspection. The image enhancement filters emphasize on textural features and
colour progression that are the characteristics of surface defects created due to
corrosion. The performances of the filter are compared and analysed by PSNR and
MSE values for corrosion images.
Using Neural Network to incorporate textural features and colour progression
that are characteristic of corrosion, from the result, the most suitable image
enhancement filter to be use on corrosion visual inspection is Wavelet De-noising.
Though the Gaussian filter shows higher PSNR value in images corrosion 03, due
to the neural network weight identified based on image corrosion characteristic,
wavelet De-noising was selected as the best suitable filter for corrosion visual
inspection. The selected filters are to be analysed in term of the possibility for any
combination, which may produce a better image enhancement quality.

References

1. He, K., Sun, J., Tang, X.: Guided image filtering. In: Computer Vision–ECCV 2010.
pp. 1–14, Springer, Berlin, Heidelberg (2010)
2. Remote Visual Inspection. Advantech Alliance Sdn Bhd. http:// www.advantech.net.my.
Accessed 16 Dec 2013
3. Livens, S., et al.: A texture analysis approach to corrosion image classification. Microscopy
microanalysis microstructures 7.2, p. 143 (1996)
4. Pidaparti, R.M., Hinderliter, B., Maskey, D.: Evaluation of corrosion growth on SS304 based
on textural and color features from image analysis. In: ISRN Corrosion, vol. 2013, Article ID
376823, 7 p (2013). doi:10.1155/2013/376823
660 S.A. Idris and F.A. Jafar

5. Medeiros, F.N.S., et al.: On the evaluation of texture and color features for nondestructive
corrosion detection. In: EURASIP Journal on Advances in Signal Processing 2010, p. 7
(2010)
6. Hamblin, J.D.: Oceanographers and the Cold War: Disciples of Marine Science, University of
Washington Press, Seattle (2005)
7. Compton, J., Hamilton, J.: Color Filter Array 2.0., A Thousand Nerds: A Kodak blog. http://
archive.today/vqZt4 (2013). Accessed Dec 2013
8. Larson, D.R.: Unitary systems and wavelet sets. In: Wavelet Analysis and Applications.
Appl. Numer. Harmon. Anal. Birkhäuser, pp. 143–171 (2007)
9. Williams, D.B., Vijay, M.: The Digital Signal Processing Handbook, Second Edition. CRC
Press, p. 438. ISBN: 978-1-4200-4606-9 (2009)
10. Gonzalez, C., Woods, E., Eddins, L.: Digital Image Processing Using MATLABÒ, Pearson
Education Inc., p. 155. ISBN 81-7758-898-2 (2007)
11. Sapiro, G.: Geometric Partial Differential Equations and Image Analysis. Cambridge
University Press, p. 223 (2001). ISBN 978-0-521-79075-8
12. Padmavathi, G., Subashini, P., Muthu, M., Suresh, T.: Comparison of filters used for
underwater image pre-processing. IJCSNS Int. J. Comput. Sci. Netw. Secur. 10(1):58 (2010)
13. Hillier, S., Lieberman, J.: Introduction to Operation Research. Tata McGraw Hill Education
PLT. p. 368 (2010). ISBN 978-007-126767-0
14. Sivanandam, S.N., Sumathi, S., Deepa S.N.: Introduction to neural networks using MATLAB
6.0. Tata McGraw Hill Education PLT. p. 11 (2011). ISBN-10: 0-07-059112-1
Chapter 62
A Framework for Sharing Communication
Media in Supporting Creative Task
in Collaborative Workspace

Norzilah Musa, Siti Z.Z. Abidin and Nasiroh Omar

Abstract In networked collaborative working environment, users are involved

with either repeatable or exclusive task. These variety tasks required the collab-
orative system to support it dynamic and creative solutions especially in sharing
the communication media. Most of the related research are objectively focusing on
the alignment of user behavior in shared activities. There is not much research
conducted concerning on sharing communication media components in creative
collaborative activity in details. In this paper, we analyze collaborative work
application systems focusing on the communication module in four common areas;
business, health, education and manufacturing. Based on the analysis, communi-
cation media sharing elements are formed into three layers. In each layer, media
sharing element and its functions are identified. Then, we proposed a general
software framework for communication media sharing for creative collaborative
activities. The framework is functioning as a platform for users to manage their
dynamic creative solutions to accomplish complex task.

62.1 Introduction

Globalization and changed in market trends emerged from the evolving of internet
and communication technology. These trends contribute to the creating of new
organizational structure and business sector to expand in remote areas. Currently,

N. Musa (&)
S.Z.Z. Abidin
N. Omar

Faculty of Computer and Mathematical Sciences, UiTM Shah Alam, Selangor, Malaysia
e-mail: [email protected]
S.Z.Z. Abidin
e-mail: [email protected]
N. Omar
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 661

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_62
662 N. Musa et al.

workers in Collaborative Working Environments (CWE) are composed of

dynamic assemble groups from diverse professional skills that work together
within shared Collaborative Working Environments (CWE) [1]. This new structure
makes possible for dispersed experts matters and workers be connected to col-
laborate and accomplish business projects [2]. Hence, collaborative work breaks
through the common concept of computer application by providing collective
cooperative workspace for users.
Collaborative workspaces are also designed for unpredictable problems that
need current workflow systems adapt to the changing environment [3]. In this
situation, users need a system that can support their creative solution activities
such as ad hoc communications and allow them to react accordingly. This kind of
activities needs the system efficiently control the services between sustaining the
space for user creativity and coordinating the resources [4]. Moreover, the content
of resources involved in such activities might be changed. Hence, provisioning of
new knowledge and data transformation need to be coordinated efficiently. Fur-
thermore, dispersed users in such collaborative activities are not sharing the same
physical work environment. They are outside of sensory range of each other. Thus,
communication technology plays a major role in supporting the information
exchange and interaction between remote group of users by using available
communication media (text, video, voice and etc.).
Since the communication is a central activity in any collaborative work, issues
in communication became one of the ongoing challenges encounter by the group
of users [5]. Moreover, most of the problems highlighted by researchers are
associated to project-related task coordination, such as workflow management,
access control and conflict management [6, 7]. Moreover, the solutions proposed
by the researchers are to adhere the need of user behavior alignment on shared
activities and resources. In this context, communication media became utility tools
with the predefined functions and fixed parameters. Hence, it is important for the
systems to provide users with flexibilities to facilitate the sharing of communi-
cation media to support creative activities.

62.2 Collaborative Workspace System Architecture

In designing the high-level system application architecture, most of the collabo-

rative application systems embraced the layered design method. The ideology of
this process is adopted from the network reference model; The Open Systems
Interconnection (OSI) Model [8]. This model consists of seven layers. From the
top to bottom are; application, presentation, session, transport, network, data-link
and physical layers. Each layer has its own functions that categorized by a set of
standard protocols. It receives information from the above layer and pass it to the
next layer based on their hierarchy.
62 A Framework for Sharing Communication Media in Supporting … 663

62.2.1 Classification of Layered Architecture Reference

Model

There is three reference model that used by the researcher in designing the high-level
collaborative application system architecture; client/server model, N-level of dis-
tributed computing and web based application model [9]. These reference models
are the backbone of any communication between networked computers; mainly in
collaborative work application systems. The approach in designing collaborative
frameworks that adopt these reference models can be classified into three categories:
service composition oriented, rule-based [10] and network technology driven.
Service Composition Oriented This approach utilizes services provided by the
web services. The system allows multi-user to access central information and
communicate each other through web services. The communication media
between users are through chat tools, whiteboards, audio conferences and webc-
ams. The web services language, Service Oriented Architecture Protocol (SOAP)
and Web Services Description Language (WSDL) make the configuration of the
communication capabilities more flexible. In order to support variety and rigid
collaborative activities, services such as service abstraction, discovery and selec-
tion, placement and binding services [11] based on Service Oriented Architecture
(SOA) is applied. Most of these services are associated in the middleware of
the framework [12, 13]. However, these useful services have to comply with the
well-defined rules in a specific domain problem.
Rule-based This approach defines rules to control the behavior of services,
users, resources and application in a collaborative environment [10]. This tech-
nique allows the system environment to provide flexible features for configuration,
operation and creation of supporting services. In the context of communication
media, synchronization rules are enforced to maintain the harmony of the col-
laborative environment.
Network Technology Driven This approach deals with the components and
logical entities of the network. It utilizes the advantages features and services
provided by the technology. The approach is suitable for problem with well-
defined problem scope and knowledge about tools supported by the network. This
enables developers to reuse current techniques for dissimilar problems. Activity
such as interaction among users is supported by the pre-defined and established
communication techniques of the chosen technology.
In summary, most of the reviewed architectures and framework of collaborative
working application systems have four common elements in their communication
module. The four elements are user interface, collaborative activities, software
design method and network technology. Hence, based on the layered architecture
design method, Fig. 62.1 shows the general view of communication layers in the
collaborative system. The communication is the main element in creating collab-
orative culture. Its allow user to create, update, change and delete shared artefact.
As users are not sharing the same physical work, exchanging and sharing resources
and information are infeasible without sharing the communication media.
664 N. Musa et al.

Fig. 62.1 Summarization of communication layers [14]

In the next section, sharing communication media in three contextual domains

will be explore in order to define the sharing components.

62.3 Communication Media Sharing Entities

There are three uses of communication in collaborative work activities [1]. First, to
assists groups of users in integrating and coordinating their work activities by
deliberating their past, current and future work activities. Secondly, serves as a
platform for user to share and exchange information. Finally, it is one of the tools
to simulate and encourage new knowledge among the users.
Communication is not only about interaction among a group of users. It is also
include several types of objects that need to be shared through various platforms
and user interfaces. Hence, user interaction and object sharing in a networked
collaborative working system are analyzed. Based on the four common areas
(business, education, health and manufacturing), seven components involve in
communication in collaborative work activities have been studied. The compo-
nents are user, communication channel, resources, software component, work
activity, coordination method and shared services. Table 62.1 shows the sharing
communication media components in the existing system.
Each domain has its own approach of disseminating information to its end-
users. This dislocated and dispersed users establish various kinds of communi-
cation relationships during collaborative activities. The relationship can be one-to-
one, one-to-many and many-to-many [14] that can determine the level of com-
munication between the users. Furthermore, user’s profile; such as role, privacy
setting and own objects to share are important features in coordinating rules on
handling the sharing process.
 62

Table 62.1 Sharing communication media components in selected domains

Domain Comm Type of Co ordination Shared resources Shared Context profile System component
channel interaction services
Business [10], Audio, One to Event-based Documents, Paint Roles, actions Mediator—service
(Belen video, chat, many, mechanism, agent- multimedia (context control, composable
Pelegrina et al. email many to interaction protocol contents awareness), service, reusable and
2010; Cheaib many, one (agreed-on, request— coordinate extended component
et al. 2009) to one response), rule-based
Education (El Whiteboard, One to JXTA protocol suit, Web browser, Web Role, name, access JXTA architecture,
Saddik et al. chat, audio, many, distributed client- server telepointer, browser level adaptive bitrate
2008; Bijlani video, many to module video, (IE, streaming
et al. 2011) message many, one documents, Mozilla) methodology
to one graphics
Health (Ciampi Text, video One to Rule paradigm, Multimedia Nil Name, medical Triggers
et al. 2010; one interaction protocol objects, patients record mechanism, agent-
Dube et al. information, based infrastructure
2005) imaging
Manufacturing Text, chat, One to Product Process Designer Nil Name, design API. Web based
(Sadeghi et al. email one, one Organisation Module drawing model, copies, company- technology
A Framework for Sharing Communication Media in Supporting …

2010), tomany and meta rules, Model- Planning name, production

(Mourtzis 2010) view- controller pattern Information, status, products
company availability
schedule
665
666 N. Musa et al.

The medium of communication in any work activities will determine the scope
of the task supported by the system. The user interface and sharing method would
become more complex when the activity used various type of communication
media. Most of the systems pre-defined their communication media into standard
operation. The credibility of the coordination techniques affects by this condition.
In order to collaborate effectively, the systems should dynamically adapt changes
in task-related activity environment. This is due to the nature of works in certain
domains that required changes of communication media during the interaction.
Furthermore, not all work activities in the domains are repeatable workflow
(create, edit, delete, etc.). Some activities need user creativity in using the com-
munication media to solve ad hoc and important problems such as exception
handling problem. This can be well supported if appropriate communication media
can match with the user requirements dynamically.
Most of the work activities in the studied domains are supported by various
kinds of shared resources. The resources can be report document, slides presen-
tation, proposal, architectural design, image, audio or video files. Moreover, some
of the objects are associated with special shared features such as special access
control that tagged with selected users. The features are set by different methods
that tailored to the solutions strategy. In fact, the success of the sharing process is
depending on the interaction and synchronization of user, resources and media
communication elements.
Different organizations will have different objectives and activities with dif-
ferent input devices and network communication model. Therefore, in dealing with
this heterogeneous communication architecture; software component act as an
intermediary between the user and network devices play a vital role in giving
seamless communication in the collaboration process. Application program
interface (API), embedded software components, special purpose program or
custom made, and toolkits are common software components that used by shared
media applications. This component helps developers to create tools to ease users
in their tasks especially group formation along with sharing activities that include
managing access and security.
In summary, the formation of communication media sharing is made possible
by five related entities; context profile, sharing method, algorithm strategy,
interaction protocol and shared objects. These entities will embed in the software
framework of the systems, particularly in the communication module. General
software based frameworks for collaborative work system focusing on the com-
munication module is depicted in Fig. 62.2. The framework was mapped with
common communication structure as shown in Fig. 62.1 which divide the structure
into three layers: front-end, intermediate and back-end.
The front-end layer act as an interface for user to interact with the systems; and
a place where communication relationship establishes among the group of users. It
also provides an information about the user profile, communication media and
work activity. In contrary, the intermediate layer is a place where the coordination
method, interaction protocol and shared services work together to create the
seamless sharing communication process. It is the place where the request of the
62 A Framework for Sharing Communication Media in Supporting … 667

Fig. 62.2 General communication framework in collaborative work

user or systems are entertained and served harmoniously by the system. Then, the
back-end layer is the place where the communication channel transforms the data
technically to be transported to the assigned users.

62.4 Conceptual Framework

62.4.1 Dynamic and Creative Collaborative Activities

In any software development, developers programmed solutions to all foresee

exception that could happen during the process. This is where the testing phase is
crucial in software development methodology. All the possible errors are captured
and user requirements are hard coded into the system after the user acceptance test
phase is done. In collaborative work, it is crucial for the systems to provide a space
for creative and knowledge intense processes like online meeting and report doc-
umentation [15]. The work activities always involve ad hoc and short-based unique
project that needs a different approach which cannot be mapped with a repeatable
workflow solution. All the dynamic changes required during the runtime should be
supported and adopt by the system. Therefore, the users should allow to commu-
nicate and shared data in creative manners to support various kinds of activities by
the groups, subgroups or individual in the collaborative work environments.
668 N. Musa et al.

Fig. 62.3 Communication media sharing framework for creative collaborative activity

62.4.2 Proposed Framework

The proposed framework (Fig. 62.3) was conformed to the general communication
framework for collaborative work structure (Fig. 62.2) for creative activities with
media sharing elements. Each user (client) has three logical layers: front-end layer,
immediate layer and back-end layer that adapts to client/server reference model for
the layered architecture. The front-end layer host the collaborative applications
communication media such as chat, shared browser and whiteboard. It provides
awareness and inform intercepting events (local and remote) to the next layer. It
sends messages to the next layer once any work activity is initiated by the users.
The intermediate layer is the cornerstone of any collaborative work application,
as it is a central of many important functions such as controlling consistency
among users, detecting and overcome any arise conflict and controlling informa-
tion exchange. The event manager is in charge of intercepting events including the
objects and users involved. It passed the information related to the event to the
session module to initiate and administer the session.
A session manager handles sessions, which represent shared workspace for
users to collaborate. It responsible to create, maintain and terminate of any ses-
sions among users. It maintains the user context profile and identified the session
owner or moderator. The moderator invoked appropriate tools to coordinate
activities and users in the session. All used objects in the session are handled by
the data manager. The creative trigger mechanism will invoke special rules to
allow users override administrator-defined policy especially regarding sharing
policy. The communication manager will receive requests and information about
the activities and transport it to the server for permission and further process.
62 A Framework for Sharing Communication Media in Supporting … 669

Since the proposed framework is based on the client/server reference model; the
server is acting like a central processor while clients are dealing with the users
request. Hence, components in the server are mirrored for clients intermediate
layer components except the event manager that only have in the clients. It con-
tains communication manager that responsible to receive any request from clients
and in charge of the application’s context profile. All messages received will be
translated and pass to server session manager to take further action. The session
manager will control and facilitate all services related to the activities include the
shared objects, protocol and invoke composer engine to cater any creative col-
laborative activities. The composer engine will process user self-manager sharing
permission policy and prepare a platform for users to manage the shared media and
objects flexibly.

62.5 Conclusion

In this paper, we give detailed explanations about the high-level design archi-
tecture for collaborative work application systems that embrace the layered-design
method which adopted from the OSI model. We also discussed the three
approaches used in designing communication module in the system framework:
service composition oriented, rule-based and network technology. Three main
domains have been selected based on their active researcher contribution in this
area. Based on these domains, seven important components have been identified:
user, communication channel, resources, software component, work activity,
coordination method and shared services. We also explain the importance of
handling creative activity in the workspace and highlight the issues when dealing
with the scattered users. Currently, most research proposed new approaches and
techniques in their collaborative working environment. However, not much
research interested in detail on communication media sharing components in
creative collaborative activity. Hence, based on the analysis, we proposed a
framework which represents the communication media sharing activity in the
collaborative work. The software framework is functioning as a platform for users
to dynamically manage the creative work activities. We left the media sharing
compose engine functionality attributes and its constraints as one of future work
for this research.

Acknowledgments The authors would like to thank Universiti Teknologi MARA and Ministry
of Education, Malaysia for the financial support.
670 N. Musa et al.

References

1. Pallot, M., Bergmann, U., Kühnle, H., Pawar, K.S., Riedel, J.C.K.H.: Collaborative working
environments: distance factors affecting collaboration. In: Proceedings of the 16th
International Conference on Concurrent Enterprising, ICE’2010 (2010)
2. Xue, X., Shen, Q., Fan, H., Li, H., Fan, S.: IT supported collaborative work in A/E/C projects:
A ten-year review. Autom. Constr. 21, 1–9 (2012)
3. Kammer P.J., Taylor, R.N.: Techniques for supporting dynamic and adaptive workflow.
pp. 1–19 (2002)
4. Schuster, N., Zirpins, C., Tai, S., Battle, S., Heuer, N.: A service-oriented approach to
document-centric situational collaboration processes. In: 2009 18th IEEE International
Workshops on Enabling Technologies: Infrastructures for Collaborative Enterprises,
pp. 221–226 (2009)
5. Bassanino, M., Fernando, T., Wu, K.-C.: Can virtual workspaces enhance team
communication and collaboration in design review meetings? Architect. Eng. Des. Manag.
(May 2013) pp. 1–18, Mar 2013
6. Sun, D., Xia, S., Sun, C., Chen, D.: Operational transformation for collaborative word
processing. In: Proceedings of the 2004 ACM conference on Computer supported cooperative
work, vol. 6, no. 3, pp. 437–446 (2004)
7. Jeffery, C., Dabholkar, A., Tachtevrenidis, K., Kim, Y.: A framework for prototyping
collaborative virtual environments. In: Groupware: Design, Implementation, and Use,
pp. 17–32 (2005)
8. Bielstein, B.: Computer networks and protocol—the OSI reference model. http://nsgn.net/
osi_reference_model/the_osi_reference_model.htm (2006). Accessed 16 Jul 2013
9. Sun, Q., Qiu, Y., Ma, W., Gu, Y.: Collaborative development of network application system
based on layered conceptual model. In: 2010 2nd International Work Database Technology
Applications, pp. 1–4, Nov 2010
10. Han, S.W., Kim, J.: A service composition oriented framework for configuring SMeet
multiparty collaboration environments. Multimedia Tools Appl. (2012)
11. Jerstad, I., Dustdar, S., Thanh, D.V.: A service oriented architecture framework for
collaborative service. In: 14th IEEE International Workshops on Enabling Technologies:
Infrastructure for Collaborative Enterprise 2005, pp. 121–125 (2005)
12. Galatopoullos, D.G., Kalofonos, D.N., Manolakos, E.S.: A P2P SOA enabling group
collaboration through service composition. In: Proceedings of 5th International Conference
on Pervasive Services—ICPS ’08, p. 111 (2008)
13. Kim, N., Kim, J., Uram, T.: A hybrid multicast connectivity solution for multi-party
collaborative environments. Multimedia Tools Appl. 44(1), 17–37 (2009)
14. Musa, N., Abidin, S.Z.Z., Omar, N.: Towards flexible media sharing: control and
coordination issues in network collaborative virtual environment. In: 2012 IEEE
Colloquium on Humanities, Science and Engineering, no. Chuser, pp. 154–158, Dec 2012
15. Schuster, N., Zirpins, C., Scholten, U.: How to balance flexibility and coordination? Service-
oriented model and architecture for document-based collaboration on the Web. In: IEEE
International Conference on Service-Oriented Computing and Applications (SOCA), pp. 1–9
(2011)
Chapter 63
Joint Torque Estimation Model of sEMG
Signal for Arm Rehabilitation Device
Using Artificial Neural Network
Techniques

M.H. Jali, T.A. Izzuddin, Z.H. Bohari, H. Sarkawi, M.F. Sulaima,

M.F. Baharom and W.M. Bukhari

Abstract Rehabilitation device is used as an exoskeleton for peoples who had

failure of their limb. Arm rehabilitation device may help the rehab program to
whom suffered with arm disability. The device is used to facilitate the tasks of the
program and minimize the mental effort of the user. Electromyography (EMG) is
the techniques to analyze the presence of electrical activity in musculoskeletal
systems. The electrical activity in muscles of disable person is failed to contract
the muscle for movements. To minimize the used of mental forced for disable
patients, the rehabilitation device can be utilize by analyzing the surface EMG
signal of normal people that can be implemented to the device. The objective of
this work is to model the muscle EMG signal to torque for a motor control of the
arm rehabilitation device using Artificial Neural Network (ANN) technique.

M.H. Jali (&)  T.A. Izzuddin  Z.H. Bohari  M.F. Sulaima  M.F. Baharom  W.M. Bukhari
Faculty of Electrical Engineering, Universiti Teknikal Malaysia Melaka, Malacca, Malaysia
e-mail: [email protected]
T.A. Izzuddin
e-mail: [email protected]
Z.H. Bohari
e-mail: [email protected]
M.F. Sulaima
e-mail: [email protected]
M.F. Baharom
e-mail: [email protected]
W.M. Bukhari
e-mail: [email protected]
H. Sarkawi
Faculty of Electronics and Computer Engineering, Universiti Teknikal Malaysia Melaka,
Malacca, Malaysia
e-mail: [email protected]

Ó Springer International Publishing Switzerland 2015 671

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_63
672 M.H. Jali et al.

The EMG signal is collected from Biceps Brachii muscles to estimate the elbow
joint torque. A two layer feed-forward network is trained using Back Propagation
Neural Network (BPNN) to model the EMG signal to torque value. The perfor-
mance result of the network is measured based on the Mean Squared Error (MSE)
of the training data and Regression (R) between the target outputs and the network
outputs. The experimental results show that ANN can well represent EMG-torque
relationship for arm rehabilitation device control.

63.1 Introduction

Human support system is endoskeleton. Endoskeleton plays a role as a framework

of the body which is bone. Our daily movements are fully depends on the func-
tionality of our complex systems in the body. The disability one or more of the
systems in our body will reduce our physical movements. The assistive device is a
need for rehab as an exoskeleton. The functionality of the rehabilitation device has
to smooth as the physical movement of normal human.
The rehabilitation programs provide the suitable program for conducting the
nerve and stimulate the muscles. People who have temporary physical disability
have the chances to recover. Nowadays, rehabilitation program are using exo-
skeleton device in their tasks. The functionality of exoskeleton depends on muscle
contraction. Electromyogram studies help to facilitate the effectiveness of the
rehabilitation device by analysing the signal transmitted from the muscle.
The rehabilitation device is a tool that used to help the movements for daily life
activities of the patients who suffer from the failure of muscle contractions, due to
the failure of the muscles contractions the movements is limited. The ability of the
patients to do the tasks in the rehabilitation programs need to be measured. The
rehabilitation programs have to assure whether the tasks will cause effective or
bring harm to the patients [1].
Historically, the rehabilitation tasks have been avoided due to a belief that it
would increase spasticity [2]. In this research, the analysis of the data will be
focusing on upper limb muscles contraction consisting of biceps muscles only. The
experiment is limited to the certain of upper limb movements that use in training.
EMG is a division of bio signal; the bio signal analysis is the most complex
analysis. Thus, the signal analysis is a complicated process that has to be through
many phases of analysis [3].
EMG signal function as a control signal for the arm rehabilitation device. A
system needs a model to estimate relationship between EMG and torque [4]. EMG
signal based control could increase the social acceptance of the disabled and aged
people by improving their quality of life. The joint torque is estimated from EMG
signals using Artificial Neural Network [5, 6]. The Back Propagation Neural
Network (BPNN) is used to find a solution for EMG-joint torque mapping. The
EMG signal of the biceps brachii muscle act as the input of the ANN model whiles
63 Joint Torque Estimation Model of sEMG Signal ... 673

the desired torque act as the ideal output of the model. Hence the EMG signals
considered the ‘intent’ of the system while the joint torque is the ‘controlled’
variable for the arm rehabilitation device [7]. The network is evaluated based on
the best linear regression between the actual joint torque and the estimated joint
torque [4]. The experiment results shows that the model can well represent the
relationship between EMG signals and elbow joint torque by producing MSE of
0.13807 and average regression of 0.999.
This paper is organized as follows. Section 63.1 explains brief introduction about
this research work. Section 63.2 describes all the related works of this study. Sec-
tion 63.3 demonstrates the method implemented for this work that covered the
experimental setup, EMG data processing, desired torque determination and ANN
technique. Section 63.4 presents the experimental results as well as the discussion of
each result obtained. Finally the paper ends with conclusion and recommendations.

63.2 Related Works

There have been several studies that have applied ANN for modeling the muscle
activity to joint relationship. Kent et al. [7] proposed the ANN model to measure
the ankle EMG-joint torque relationship at a full range of torque under isometric,
supine conditions by inserting EMG signal from 6 muscle sites into the model as in
the input, while the measured torque is entered into the model as the ideal output.
The learning process occur approximately 16,000 iterations resulting error that is
less than 6 %.
More recent study has been conducted by Li et al. [4] to predict the elbow joint
angle based on EMG signals using ANN. The three layer BPNN was constructed
by using the RMS of the raw EMG signal form biceps and triceps. The result from
40 group EMG signals when subjects do bowing and extending elbow joint action
reveal that the prediction output from the trained network was very close with the
target angle. According to Morita et al. [6], it is quite difficult to know the elbow
shoulder joint torque on the natural condition. The learning method which is based
on the feedback error learning schema is proposed by modifying the ANN with the
torque error which calculated from the desired angle and measured angle.
Studies of muscle force models have been carried out by Naeem et al. [8]. The
model was estimated based on a rectified smoothed EMG signal using the BPANN
method to predict the muscle force. The proposed model can efficiently extract
muscle force features from (EMG) signals in a fast and easy method. The results
showed that the regression of the ANN model exceeded 99 %. However among all
the previous studies related to EMG based ANN model been conducted, none of
them really emphasize on the data processing phase which involve EMG data
processing and desired torque determination. Therefore this paper will explain in
detail this phase.
674 M.H. Jali et al.

Experimental setup
EMG data
for sEMG data Joint torque
processing
acquisition estimation
ANN
model
Arm rehabilitation Determine the
device desired torque

Fig. 63.1 Research methods

63.3 Methods

Figure 63.1 shows a block diagram of our research methodology that consists of two
major phases. First phase is EMG data processing and desired torque determination.
Second phase is the ANN construction and testing. The data collection from the first
phase is used to validate and teach the ANN algorithm in second phase.

63.3.1 Experimental Setup

Implementation of arm rehabilitation device based on movement is recorded from

the EMG signal of healthy subjects. From the human anatomy studies, different
angle movements of upper limb with elbow as the reference is depends on relation
of agonist and antagonist. In this study is focusing on the behaviour of biceps
muscle as agonist and the triceps as the antagonist respectively. Muscle that
involved in this movement is biceps and triceps, however in this study to under-
stand the electrical activity during muscle contraction, the biceps is the only
muscle that taking into account. The movements’ ranges in between position of
arm flexion until arm fully extend.
The environment is in a room with low lighting especially the fluorescent light,
any electromagnetic devices is away from the experiment equipment and the
environment is in silent room. Then, the experimental is set up with the subject sit
on the chair while the hand is on the table. The subject has to complete the task of
lift up the dumbbell with 2.268 kg of weight as shown in Fig. 63.2b. Normally, the
appearance of EMG signal is chaos and noisy depends on the type of electrodes
also the noise factor. To simplify the difference of amplitude response for the
motion, the dumbbell is functioned to amplify the amplitude in analysing the
electrical activity during rest and contract. The rehabilitation devices (white in
color on Fig. 63.2a) helps to keep the position of the elbow joint and the wrist joint
in line Mostly, the EMG signal is obtained after several trials of the movements.
These movements are specified from angle of 0° (arm in rest position), up to 120°
(arm is fully flexion). Data was collected from two subjects by 5 repetitions of
each flexion movements [9].
Prior of data collection process, the skin needs a preparation. The preparation of
skin is ruled by the Surface Electromyography for the Non-Invasive Assessment of
63 Joint Torque Estimation Model of sEMG Signal ... 675

Fig. 63.2 Subject is set-up with arm rehabilitation assistive device for experiment (a).
Simulation of subject’s to lift up the dumbbell 2.268 kg of weight (b)

Fig. 63.3 The biceps brachii (a) (b)

muscles for electrode
positions (a), The electrode
placements on subject skin
(b)

Biceps

Muscles (SENIAM) procedure for non-invasive methods. The subject’s skin has to
be shaved by using small electrical shaver and cleaned with sterile alcohol swabs
saturated with 70 % Isopropyl Alcohol. This step is to be taken for minimizing the
noise and to have a good contact with the electrodes of the skin by decreasing the
impedance of the skin. The skin has to be clean from any contamination of body
oil, body salt, hair and the dead cells. The preparation of skin can be done by
wiping the alcohol swab into the area of skin that electrode placement to be
applied. The placements of the electrode have to be at the belly of the muscles not
in the tendon or motor unit. This ensured the detecting surface intersects most of
the same muscle on subject as in Fig. 63.3a at the biceps brachii, and as a result, an
improved superimposed signal is observed. Reference electrode has to be at the
bone as the ground, for this experiment it placed at elbow joint as shown in
Fig. 63.3b. These electrodes are connected to the combination of hardware Olimex
EKG-EMG-PA and Arduino Mega for data collection.
676 M.H. Jali et al.

Determine cut off

Raw EMG Filtering & Signal
frequency range for
signal rectify smoothing
filtering using FFT

Fig. 63.4 EMG data processing

Fig. 63.5 Raw EMG signal (a), Power Spectral Density (b), Rectified signal (c), Smooth signal (d)

63.3.2 EMG Data Processing

Figure 63.4 shows the EMG data processing block diagram. After obtained satis-
factory EMG signal as shown in Fig. 63.5a, Fast Fourier Transform (FFT) is per-
formed to the signal to analyses the frequency content of the signal. The EMG signal
is break into its frequency component and it is presented as function of probability of
their occurrence. In order to observe the variation of signal in different frequency
components, the FFT signal is represented by Power Spectral Density (PSD). From
the PSD we can describes how the signal energy or power is distributed across
frequency. Figure 63.5b shows that most of the power is in the range of below
10 Hz, therefore the EMG signal should be filtered in the range of above 10 Hz as a
63 Joint Torque Estimation Model of sEMG Signal ... 677

Fig. 63.6 Arm rehabilitation device position (a), Desired torque characteristic (b)

cut off frequency for low pass filtered. After decide the cut off frequency for filtering,
the DC offset of the EMG signal is removed and is rectified to obtain its absolute
value as shown in Fig. 63.5c. Finally, the signal was smoothed and normalized
passing it through a 5th order Butterworth type low-pass filter with cut off frequency
10 Hz and the smooth signal is illustrate in Fig. 63.5d [8, 9].

63.3.3 Desired Torque

Desired torque of the elbow joint is used as target data for our ANN techniques as
well as act as output signal for muscle [10]. The data is collected throughout the
angle from 0° to 120° angle with increment of 0.0619° each step to align with the
sample number of EMG signal. Figure 63.6a shows the arm rehabilitation device
position for torque calculation. The desired torque for elbow joint is determined by
applying standard torque equation:

s ¼ ðrload Fload þ rarm Farm Þcos h ð63:1Þ

where rload is distance from the elbow joint to the load, Fload is force due to load,
Farm is the force due to the mass of the lever arm and rarm is evenly distributed
distance of mass of arm distance which is half of rload. The angle h between r and
F is drawn from the same origin. A applied load is 2.268 kg dumbbell and the
distance from the elbow joint is 0.25 m while the mass of the lever arm is 0.1 kg.
Figure 63.6b shows the desired torque characteristic for the elbow joint.

63.3.4 Artificial Neural Network Techniques

ANN is a computing paradigm that is loosely modeled after cortical structures of

the brain. It consists of interconnected processing elements called nodes or neuron
678 M.H. Jali et al.

that work together to produce an output function. It capable to map a data set of
numeric inputs with a set of numeric outputs. It is also the most widely applied
training network which has input layer, hidden layer and output layer. The neurons
on each layer need to be considered carefully to produce high accuracy network.
The number of hidden neurons could affect the performance of the network. The
network performance not always been improved if the hidden layer and its neurons
is increased [4]. Therefore the number of hidden neurons is tested to achieve the
optimized network. However there is constraint in determining the number of
neurons. If the numbers of hidden neurons is too large, the network requires more
memory and the network become more complicated while if the number of hidden
neurons is too small, the network would face difficulty to adjust the weigh properly
and could cause over fitting which is problem where the network cannot be
generalized with slightly different inputs [11].
The input signal is propagated forward through network layer using back
propagation algorithm. An array of predetermined input is compared with the
desired output response to compute the value of error function. This error is
propagated back through the network in opposite direction of synaptic connec-
tions. This will adjust the synaptic weight so that the actual response value of the
network moved closer to the desired response [12]. BPNN has two-layer feed-
forward network with hidden neurons and linear output neurons. The function used
in the hidden layer of network is sigmoid function that generates values in range of
-1 to 1 [9]. There are layers of hidden processing units in between the input and
output neurons. For each epoch of data presented to the neural network, the
weights (connections between the neurons) and biases are updated in the con-
nections to the output, and the learned error between the predicted and expected
output, the deltas, is propagated back through the network [13].
A Lavenberg-Marquardt training back propagation algorithm is implemented
for this work to model the EMG to torque signal. This algorithm is chosen due to
the facts that it outperforms simple gradient descent and other conjugate gradient
methods in a wide variety problem such as convergence problems. Single input
EMG signal that act as a training data while single output which is the desired
torque act as a target data. The network is optimized for 20 hidden neurons as
shown in Fig. 63.7 [4]. The network was trained using 1839 sets of EMG data for
arm flexion motion from 0° angles to 120° angle. It also has output data which is
torque of correspondent arm motion. The training process was iteratively adjusted
to minimize the error and increased the rate of network performance [11]. The
training has been done by dividing the input data of 70 % for training, 15 % for
validation and 15 % for testing [8].
The performance evaluation of the network is based on the Mean Squared Error
(MSE) of the training data and Regression (R) between the target outputs and the
network outputs as well as the characteristics of the training, validation, and
testing errors. The network is consider has the best performance if it has lowest
MSE and highest R while exhibit similar error characteristics among the training,
validation and testing. However even if the MSE shows very good result but the
validation and testing vary greatly during the training process, the network
63 Joint Torque Estimation Model of sEMG Signal ... 679

Fig. 63.7 BPNN model

Fig. 63.8 Best validation

performance

structure is still considered unsatisfactory because the network is not generalized.

Therefore further tuning and training need to be conducted in order to improve the
network performance [11].

63.4 Experimental Results

In order to optimize the network performance, different number of hidden neurons is

simulated for several times until achieved the satisfactory results [11]. The network
is trained using Lavemberg-Marquardt algorithm and the performance of the net-
work is measured using MSE and R. The best validation performance of the network
is 0.13807 at epoch 46 as shown in Fig. 63.8. It is an acceptable result since the test
set error and the validations set error have similar behavior. It shows that the MSE
has decreased rapidly along the epochs during training. The regression also produces
a good curve fitness for training, test and validation data around 0.999 which give an
optimal value for our model as shown in Fig. 63.9 [8, 14].
680 M.H. Jali et al.

Fig. 63.9 Regression of the trained model

Fig. 63.10 Error histogram of the trained model (a), Target versus trained network output (b)
63 Joint Torque Estimation Model of sEMG Signal ... 681

Error sizes are well distributed since most error approaching zero values that
make the trained model perform better as shown in Fig. 63.10a. The network has
maximum instance around 450 of MSE distributed around the zero line of the error
histogram [14]. Figure 63.10b shows a comparison between prediction output
from the trained network and target torque. The prediction output has fairly good
agreement with the characteristics of the target data.

63.5 Conclusions

Based on the result, it can be concluded that the ANN model with 20 hidden
neurons produce MSE of 0.13807 and average regression of 0.999. It is consider a
good performance as it shows that this neural network model can well represent
the relationship between EMG signals and elbow joint torque. Hence this model
can be used for motor torque control of the arm rehabilitation devices. The model
can be further improved by applying other artificial intelligence training algorithm
such as genetic algorithm and particle swarm optimization to produce better mean
squared error and regression performance result.

Acknowledgements The authors would like to thanks Universiti Teknikal Malaysia Melaka
(UTeM) and Ministry of Education, Malaysia for the financial supports given through Research
Grant.

References

1. Louise Ada, S.D.C.G.C.: Strengthening interventions increase strength and improve activity
after stroke: a systematic review. Aust. J. Physiotherapy 52, 241–248 (2006)
2. Adult Hemiplegia, B.B.: Evaluation and treatment. Butterworth-Heinemann, Oxford (1990)
3. Muthuswamy, J.: Biomedical Signal Analysis in Standard Handbook of Biomedical
Engineering and Design, pp. 18.1–18.30. McGraw-Hill, New York (2004)
4. Li, D., Zhang, Y.: Artificial neural network prediction of angle based on surface
electromyography. In: International Conference on Control, Automation and Systems
Engineering (CASE), pp. 1–3 (2011)
5. Reaz, M.B.I., Hussain, M.S., Mohd-Yasin, F.: Techniques of EMG signal analysis: detection,
processing, classification and applications. Biol. Proced. Online 8, 11–35 (2006)
6. Morita, S., Kondo, T., Ito, K.: Estimation of forearm movement from emg signal and
application to prosthetic hand control. In: IEEE International Conference on Robotics and
Automation (ICRA), vol. 4, pp. 3692–36972 (2001)
7. Kent, L.M., Siegler, S., Guez, A., Freedman, W.: Modelling of muscle EMG to torque by the
neural network model of back propagation. In: Proceedings of the Twelfth Annual IEEE
International Conference of the Engineering in Medicine and Biology Society,
pp. 1477–1478 (1990)
8. Naeem, U.J., Abdullah, A.A., Caihua X.: Estimating human arm’s muscle force using
artificial neural network. In: Proceedings of IEEE International Symposium on Medical
Measurements and Applications (MeMeA), pp. 1–6 (2012)
682 M.H. Jali et al.

9. Favieiro, G.W., Balbinot, A., Barreto, M.M.G.: decoding arm movements by myoeletric
signals and artificial neural networks. In: Conference of Biosignals and Biorobotics (BRC),
pp. 1–6 (2011)
10. Jali, M.H., Sulaima, M.F., Izzuddin, T.A., Bukhari, W.M., Baharom, M.F.: Comparative
study of EMG based joint torque estimation ANN models for arm rehabilitation device. Int.
J. Appl. Eng. Res (IJAER) 9(10), 1289–1301 (2014)
11. Ahsan, M.R., Ibrahimy, M.I., Khalifa, O.O.: EMG motion pattern classification through
design and optimization of neural network. In: International Conference on Biomedical
Engineering (ICoBE), pp 175–179 (2012)
12. Hudgins, B., Parker, P., Scott, R.: A new strategy for multifunction myoelectric control. IEEE
Trans. Biomed. Eng. 40(1), 82–94 (1993)
13. Mars, P., Chen J.R., Nambiar, R.: Learning Algorithms.: Theory and Applications in Signal
Processing, Control and Communications. CRC Press , Boca Raton (1996)
14. Supeni, E.E., Epaarachchi, J.A., Islam, M.M., Lau, K.T.: Development of artificial neural
network model in predicting performance of the smart wind turbine blade. In: 3rd Malaysian
Postgraduate Conference (MPC 2013), pp. 4–5. Sydney, Australia (2013)
Chapter 64
Enhancement of RSA Key Generation
Using Identity

Norhidayah Muhammad, Jasni Mohamad Zain,

M.Y.M. Saman and Mohd Fadhil Ramle

Abstract The purpose of this paper is to enhance previous algorithm called

Tripathi algorithm. The Tripathi algorithm proposes an RSA based algorithm to
generate cryptographic keys using user identity such as email address of a person.
This algorithm used user’s identity to replace the numbers that are used as a public
key in the RSA algorithm. However, the Tripathi algorithm cannot use all of the
users’ email addresses as a public key. This is because, there are two reasons why
it is unable to use all email addresses: (i) this algorithm use the same modulo value
for every email, if the email is not related prime to modulo value, the new email
should be entered. (ii) Entered email is composed of odd and even number. If the
email is even number, then it cannot be the public key. Therefore the Tripathi
algorithm needs to be improved. Proposed algorithm called CLB-RSA has been
implemented. This algorithm can used all user emails as a public key, and this
achievement is after two experiments are done on this study.

64.1 Introduction

Information security or also known as computer security is an approach to protect

information from unauthorized access, disruption, modification or manipulation of
information [1]. Sensitive data need to protect from unauthorized by any unrelated

N. Muhammad (&)
J.M. Zain

University Malaysia Pahang, 26600 Pahang, Malaysia
e-mail: [email protected]
J.M. Zain
e-mail: [email protected]
M.Y.M. Saman
University Malaysia Terengganu, 21300 Terengganu, Malaysia
e-mail: [email protected]
M.F. Ramle
Kolej Komuniti Kuala Terengganu, Terengganu, Malaysia
e-mail: [email protected]

Ó Springer International Publishing Switzerland 2015 683

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_64
684 N. Muhammad et al.

person. As done by [2, 3] to protect medical image. Cryptographic system or

cryptosystem is a form of encryption, decryption algorithm, and the key generation
[4]. Key generation become the biggest problem for cryptography. This is because
the key generation process to determine the public key and private key for use during
encryption and decryption process. The safety of a cryptography algorithm depends
on the complexity of a cryptography keys. RSA algorithm security is commonly
known can be a tool for good security. The actual RSA cryptographic process is
generally a complicated mathematical formulation, the more complex of keys, the
more difficult to break the cipher text, and more secure, the disadvantage of RSA
algorithm is the RSA algorithm using a key consisting of a row of numbers and also
requires large storage space and is only suitable for use on large device memory.
Several techniques had already been proposed for distribution of public keys.
Out of them one was public key certificates. The basic idea was to use trusted third
party called Certificate Authority (CA) to provide trusted public key to the various
participants on demand. To setup the hierarchical infrastructure for numerous
CA’s extra overhead was required. In 1984, Shamir [5] proposed public key
encryption scheme in which the public key can be an arbitrary string. Shamir’s
original motivation for identity based encryption was to simplify the certificate
management. Several proposals for IBE schemes [6–9]. Hence the Tripathi
algorithm [10] presents an RSA based algorithm to generate the cryptographic
keys required by participants for secure communication using their identities
which is similar to identity based encryption scheme (IBE).
However the main problem in Tripathi algorithm is cannot use all of the user
email addresses as a public key. This is because, there are two reasons why it is
unable to use all email addresses: (i) this algorithm use the same modulo value for
every email, if the email is not related prime to modulo value, the new email
should be entered. (ii) Entered email is composed of odd and even number. If
email users cannot be used as a public key, then the user must enter a new email
and the process is repeated until the user’s email can be used as a public key.
Therefore the Tripathi algorithm needs to be improved. In this study, two times of
experiments that were carried out to produce satisfactory results and make sure all
emails entered by user can be a public key. In a first experiment, the algorithm that
has been improved from the Tripathi algorithm and it’s called LB-RSA.
Looping process has been added to this algorithm to produce a new modulo
value and helps to produce more email addresses that can be used as a public key
and LB-RSA have shown good results, that is, 50 % of the total number of emails,
can be used as a public key. This number was increased compared to the amount
generated by Tripathi algorithm that is 25 %. The resulting decision issued by LB-
RSA algorithms in experiment 1 do not reach 100 %, the second experiment
carried out to provide a better algorithm and can achieve 100 % of emails that can
be used as a public key. The classification process added in LB-RSA algorithm is
to determine whether the decimal is even or odd number, this process helps to
make all email can be a public key. Before emails are tested in these algorithms,
emails will be converting to decimal value.
64 Enhancement of RSA Key Generation Using Identity 685

64.2 Related Work

64.2.1 Tripathi Algorithm

RSA algorithm was gone through several phases of change towards, a variety of
improvements have been made in the original RSA algorithm, including algo-
rithms that have been developed by Sachin Tripathi [11]. This algorithm also has
several advantages compared with the original RSA algorithm. (1) Tripathi
algorithm used user string identity as a public key such as an email address. (2)
Need small device memory to store the public key. The algorithm presents the
RSA based algorithm to generate the cryptographic keys required by participants
for secure communication using their identities, which is similar to identity based
encryption (IBE) scheme. Hence the authentication of the public key in IBE is a
big challenge and for which public key certificates, provided by a Certification
Authority (CA), are used. At large scale communication to set up numerous CA’s
is a major overhead of public key cryptography.
Hence this algorithm attempts to avoid the use of public key certificates and
proposes an RSA based algorithm to generate the cryptographic keys using
identity such as an email identity of a person. Figure 64.1 shows the key gener-
ation algorithm of the Tripathi flow algorithm. The email address entered will be
converted into a fixed—size string using CRC32 hash function and then converted
to a decimal value before test in this algorithm. After that the decimal value h (id)
will be tested whether it is relative prime to modulo phi (n) or not, h (id) is relative
prime to the modulo phi (n), then email address can be used as a public key, but if
the email is not relative prime to the modulo, then the user must enter a new email
as a public key. This is an example of email address convert to form of decimal
using CRC32 hash function. However the problem in Tripathi algorithm is unable
to use all user email as a public key. There are two reasons why it is unable to use
all email addresses: (i) this algorithm use the same modulo value for every email,
if the email is not relative prime to modulo value, the new email should be entered.
(ii) Entered email is composed of odd and even number.

64.3 Proposed Algorithm (CLB-RSA)

The methodology proposed will be named CLB-RSA algorithm. The character of

‘‘CLB’’ is standing for classification and loop based process. A representation of
‘‘C’’ is standing for classification and ‘‘LB’’ is standing for loop based. The
advantage of this algorithm is able to use all user emails address as a public key.
There are two enhancement process added in this proposes algorithms are: looping
and classification process. Advantages of using user identity abridged in order to
facilitate the user to remember public key. Example of user identity is including user
name, user email, nickname and so on. Improvements done on this algorithm has
686 N. Muhammad et al.

Fig. 64.1 Tripathi algorithm

flowchart Start

Select prime
number p and q

e= valid email ID

h (id) = hash(e)

phi (n) = (p -1)*(q -1)

k = GCD (h(id),phi(n)

Is
k= 1?

e = public key

End

improved the performance of CLB-RSA, the performance on the number of emails

that can be used as a public key is increase and reach of hundred percent results.
The improvement was made to ensure that all email address can be used as a
public key. As discussed in related work, Tripathi algorithm cannot use all the
emails as a public key, and only certain email that can be used as a public key.
Comparison of the performance can be seen between the Tripathi algorithms and
CLB-RSA algorithms. Comparison between these algorithms is in terms of the
number of emails that can be made as a public key email based on twenty samples
listed in Table 64.1. CLB-RSA is second algorithm evaluate from first experiment
by Muhammad [12].

64.3.1 LB-RSA Algorithm

In this study, two times of experiments that was carried out to produce an algo-
rithm that can produce a satisfied result. In a first experiment, the algorithm that
64 Enhancement of RSA Key Generation Using Identity 687

Table 64.1 Result LB-RSA and CLB-RSA

No Email Decimal LB-RSA CLB-RSA
1 [email protected] 786745081 Yes Yes
2 [email protected] 2730798235 Yes Yes
3 [email protected] 1812161612 No Yes
4 [email protected] 3238888356 No Yes
5 [email protected] 795509258 No Yes
6 [email protected] 3700340025 Yes Yes
7 [email protected] 4067366447 Yes Yes
8 [email protected] 1199321065 Yes Yes
9 [email protected] 3074048744 No Yes
10 [email protected] 2205045334 No Yes
11 [email protected] 3009834837 Yes Yes
12 [email protected] 2654073728 No Yes
13 [email protected] 3765641505 Yes Yes
14 [email protected] 296517678 No Yes
15 [email protected] 2386521674 No Yes
16 [email protected] 2962995952 No Yes
17 [email protected] 3843922542 No Yes
18 [email protected] 2489219047 Yes Yes
19 [email protected] 3211800885 Yes Yes
20 [email protected] 2388137343 Yes Yes

has been improved from the Tripathi algorithm and it’s called LB-RSA [12].
Looping process is added in LB-RSA to generate the new parameters of modulo
phi (n). In Tripathi algorithm, looping process will be run on email address, it’s
mean, if email cannot be a public key, so users need to enter new email address, so
the loop process is happen on email address. However in LB-RSA algorithm, the
looping process is running modulo value phi (n), if the email address cannot be a
public key, so new modulo value will be created. When the new parameters of phi
(n) are produced, then the probability of the value of h (id) and phi (n) is relative
prime is higher, and it is usually referred to equation k = GCD (h (id), phi (n)). If
k is equal to 1, so this it meant h (id) is relative prime to phi (n), and the email’s
address can be used as public key. In Tripathi algorithm, modulo phi (n) generated
only once and if the value of k is not equal to 1, then the email entered is declared
as a non-public key, and the user should enter a new email address. These pro-
cesses write:

If GCD (h(id),phi(n))=1;
h(id)=public key;
else
Looping process until k=1;
688 N. Muhammad et al.

Start Start

Select prime
Select prime
number p and
number p and q

e = valid email
e= valid email ID

h(id) = hash(e)
h (id) = hash(e)

phi (n) = (p-1)* (q -1) phi (n) = (p-1)*(q-1)

k = GCD (h(id),phi(n) k = GCD (h(id),phi(n)

Is is
k= 1? k= 1?

e = public key e = public key

End End

Tripathi Algorithm LB-RSA Algorithm

Fig. 64.2 Comparison between Tripathi and LB-RSA Key Generation Algorithm

Figure 64.2 shows the flowchart of LB-RSA and Tripathi key generation algo-
rithm. This comparison between two algorithms: Tripathi algorithm and LB-RSA
algorithm. As shown in Fig. 64.2, the main differences between these algorithms
are, in Tripathi algorithm, looping process is running on email address. When the
email entered by the users, can’t be a public key, so users need to enter the new email
address until email address can be used as a public key. So it’s mean reenter the new
emails is the looping process in Tripathi algorithm. However in LB-RSA algorithm,
the looping process occurred on modulo value, if the email address can’t be a public
key, thus a new modulo value phi (n) will be continuously created until the email can
be a public key. So it’s mean the looping process is done in modulo value and not on
email address. This process will be fully effective because the email addresses that
can be a public key are increasing in LB-RSA algorithm.

64.3.2 CLB-RSA Algorithm

The mathematical calculations that are used in CLB-RSA key generation algo-
rithm describe to show detail how it’s calculated in table 64.2. Steps 6 and 7 are
64 Enhancement of RSA Key Generation Using Identity 689

Start

Start
Select prime
number p and q

Select prime
number p and e= valid email ID

h (id) = hash(e)
e= valid email

phi (n) = (p-1)* (q-1)

h(id) = hash(e)

phi (n) = (p-1)*(q-1) Is

h(id)= odd? h (id) – 1

k = GCD (h(id),phi(n)

k = GCD (h(id),phi(n)
Y
is
k= 1?
Is
k = 1?

e = public key
e = public key

End
End

LB-RSA Algorithm CLB-RSA Algorithm

Fig. 64.3 Comparison betweenCLB-RSA Algorithm and LB-RSA Algorithm

the additional steps of this algorithm different with Tripathi algorithm. Small
number used in example in table 64.2 to give simple calculation, and also used
even decimal value as a public key to show how this algorithm make a classifying
process to determine either the decimal entered is even or odd number. The even
number of decimal value is converted to odd number by reducing the value as
shown in step 6. After emails tested, and the result for entered email is an odd
number, then the next process can be carried out and an email address can be used
as a public key, even maybe the looping process requires more than a one-time
loop. However, if the decimal value is even, then the value should be converted to
the odd before proceed to the nest process. Therefore, when the value is odd
umber, then the email can be used as a public key. This classifying process will be
solving the second problem (even decimal value can’t be a public key) in Tripathi
algorithm that causes this algorithm cannot used all email as a public key.
Therefore LB-RSA can solve the first problem that causes the Tripathi algorithm
690 N. Muhammad et al.

cannot used all email as a public key using looping process hence create the new
modulo value until email an modulo value is related prime.
To convert an even decimal value to odd decimal value, decimal value should
be minus 1 or plus 1. But for this algorithm, the decimal value will be minus 1 to
produce odd decimal value. Once the email is turned to the odd number, and then
the next process can be implemented. This is the difference process between LB-
RSA and CLB-RSA, and this improvements step can complement this algorithm
because CLB-RSA can make all email entered can be used as a public key. The
addition process of this algorithm is to make tests on the decimal to classify the
decimal value, whether it is an odd number, or even number. This equation is used
for this process. If h (id) = odd, then continue step 7, else h (id)-1, continue step 7.
Figure 64.3 shows the flowchart of the steps in CLB-RSA key generation algo-
rithm, the process begins with the declaration of two random numbers as a
parameter p and q and lastly public key and private key are determined.
Figure 64.3 shows the comparison between two key generation algorithms, LB-
RSA and CLB-RSA. As shown in Fig. 64.3, the main difference is the classifi-
cation process. This process is added in this algorithm to make classifying on
decimal value, this process is very important because if the decimal value is even
number, so the email address entered cannot be a public key. Classification process
added in this algorithm is to make a classifying of decimal value to odd or even
number. If the decimal value is an odd number, so there is nothing happened to the
decimal value and can continue to the next step and make is as a public key.
However, if the decimal value is even, so this decimal value must be converted to
odd number first before proceeds as a public key. However the looping process is
still available in this algorithm because it will be useful when the decimal value is
an even number. If a decimal value is an odd number, but decimal value and
modulo value is not a prime number, so new modulo value will be created and this
process is fully working when the email that can be a public key is increasing.
Table 64.1 shows the result produced by LB-RSA and CLB-RSA. LB-RSA
algorithm is tested in experiment 1 and CLB-RSA algorithm is tested in algorithm
2. Both of these tests using the same twenty sample email listed in Table 64.1. In
LB-RSA, half of data can be used as a public key, and a half of the data cannot be
a public key, but in CLB-RSA all data can be used as a public key. This proves the
effectiveness of CLB-RSA algorithm and the addition step is performed on this
algorithm have a significant impact on this algorithm, successfully making all the
emails as a public key.
64 Enhancement of RSA Key Generation Using Identity 691

Table 64.2 CLB-RSA Key Generation Culculation

1. Select two parameters, namely p and q, these two parameters must be a

prime number.
p= 3, q= 7
2. Compute n by using the following formula.
n=p*q
n= 21
3. Compute phi (n).
phi (n) = (p – 1) * (q – 1)
phi(n) = 12
4. Input the user valid email Id.
e = valid email id
e = [email protected]
5. Convert email id to hash function.
h(id) = hash(e)
h(id) = 1812161612
6. Classified decimal value.
if h(id) = odd number?
Continue step 7
else
h(id) – 1
Continue step 7
1812161612== odd?= NO
else
1812161612-1=1812161611
h(id)==1812161611
Continue step 7
7. Choose the exponent k if :
If GCD (h(id), phi (n)) = 1
Continue step 8
Else, loop step 1 the process until k=1
GCD(1812161611, 12) = 1
8. Compute private key exponent d through following formula
d = e ^ -1 mod (phi (n))
d=7
9. Thus the public key consists of public key exponent e and n. And the private
key consists of private key exponent d and n.
Public key: (n, e)
Public key: (n, d)
Public key: (21, 1812161611)
Public key: (21, 7)
692 N. Muhammad et al.

64.4 Conclusion

The algorithm has been improved and called CLB-RSA has achieved the objectives
of improving the RSA algorithm using the email address as the public key, and make
sure all the email entered by the user can be used as a public key. The email addresses
used in this study consisted of a variety of different domains. The results produced by
CLB-RSA are very gratified that 100 % of the email can be public key. These results
are better than the previous algorithm (Tripathi algorithm). Problems faced by
Tripathi algorithm can be solve by CLB-RSA algorithm using loop based and
classification process. However in future, this algorithm can be improved better than
CLB-RSA such as private key can be replaced with private user identity.

Acknowledgments This study is funded by Skim Latihan Akademik IPTA (SLAI) under the
Malaysia Ministry of Higher Education (MOHE) and Universiti Sultan Zainal Abidin (UniSZA)
Malaysia. The authors would like to acknowledge all contributors, and others who have helped
and greatly assisted in the completion of the study.

References

1. Diffie, W., Hellman, M.E.: New directions in cryptography. IEEE Trans. Inf. Theory 22,
644–654 (1976)
2. Zain, J.M., Fauzi, A.R.: Medical image watermarking with tamper detection and recovery. In:
28th Annual International Conference of the IEEE on Engineering in Medicine and Biology
Society (EMBS’06) (2006)
3. Zain, J.M., Fauzi, A.R.: Evaluation of medical image watermarking with tamper detection
and recovery (AW-TDR). In: 29th Annual International Conference of the IEEE on
Engineering in Medicine and Biology Society (EMBS 2007) (2007)
4. Wenbo, M.: Modern cryptography: theory and practice. Publisher: Prentice Hall PTR,
Hewlett Packard (2004)
5. Shamir, A.: Identity-based cryptosystems and signature schemes. In: Advances in Cryptology
(1985)
6. Desmedt, Y., Quisquater, J.-J.: Public-key systems based on the difficulty of tampering (Is there
a difference between DES and RSA?). In: Advances in Cryptology—CRYPTO’86 (1987)
7. Maurer, U.M., Yacobi, Y.: Non-interactive public-key cryptography. In: Advances in
Cryptology—EUROCRYPT’91 (1991)
8. Hühnlein, D., Michael, Jr J., Weber, D.: Towards practical non-interactive public-key
cryptosystems using non-maximal imaginary quadratic orders. Des. Codes Crypt. 30,
281–299 (2003)
9. Boneh, D., Franklin, M.: Identity-based encryption from the Weil pairing. J. Comput. SIAM
32, 586–615 (2003)
10. Rivest, R.L., Shamir, A., Adleman, L.: A method for obtaining digital signatures and public-
key cryptosystems. Commun. ACM 26, 96–99 (1983)
11. Tripathi, S., Biswas, G.P., Kisan, S.: Cryptographic keys generation using identity. In: 3rd
International Conference on Advances in Recent Technologies in Communication and
Computing (ARTCom 2011) (2011)
12. Muhammad, N., Zain, J.M., Mohd Saman, M.Y.: Loop-based RSA key generation algorithm
using string identity. In: 13th International Conference on Control, Automation and Systems
(ICCAS) (2013)
Chapter 65
Rules Mining Based on Clustering
of Inbound Tourists in Thailand

Wirot Yotsawat and Anongnart Srivihok

Abstract Tourism industries are growing up rapidly with more competition. So,
travel agencies or tourism organizations must have a good planning and provide
campaign for tourist’s needs. This study proposes the usage of data mining for
tourism industries in Thailand. Data clustering and association rule mining were
chosen as the data mining methods in order to discover useful knowledge. Two-level
clustering with decision tree bagging was applied to construct the segments of tourist.
Apriori algorithm was then used to find the rules on each cluster. The experimental
results indicated that the tourists data was separated into eleven differently segments
and decision tree bagging for attributes weighting can enhance the quality of clusters.
The eleven segments were analyzed in order to identify tourists’ behavior patterns
and their preferences. Association rule mining was applied to each segment in order
to find the relationship among the features of tourist data. The rules were filtered
again by experts. The clustering and association rule results can be served to tourism
organization in order to support their strategic and market planning.

65.1 Introduction

Tourism industries are growing up rapidly in many countries. There are many
supporting policy and developing plans from their governments. Activities and
destinations are promoted for the attraction of tourists and tourism investors from
foreign countries. Tourists receive more choices for selecting the best interesting
places. So, travel agencies or tourism organizations must have a good planning and
provide campaigns for tourist’s needs. They have to know tourists’ behavior

W. Yotsawat (&)  A. Srivihok

Department of Computer Science, Faculty of Science, Kasetsart University,
Bangkok, Thailand
e-mail: [email protected]
A. Srivihok
e-mail: [email protected]

Ó Springer International Publishing Switzerland 2015 693

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_65
694 W. Yotsawat and A. Srivihok

patterns and their preferences. This paper proposes data mining techniques on
inbound tourists in Thailand by using segmentation and association rule tech-
niques. Two-level clustering with factors weighting by Decision Tree bagging,
was a methodology for applying cluster analysis to explore the tourists patterns for
market planning, promotion and package design for each group. Association rule
mining was then applied to each segment. The results of association rule can be
applied for tourists’ recommendation system. Thus, this study can serve as useful
knowledge for travel agencies and other tourism organizations.

65.2 Literature Review

Data mining is the process of automatically discovering useful information in large

data base [1]. Since data mining was introduced, it has developed by many
researchers. Data mining technique has been applied in many fields such as
accounting, medicine, law, and so forth. Some researchers focused on the
implementation of data mining for tourism such as Wong, Chen, Chung, and Kao
[2], they proposed the usage of three data mining techniques to analyze the travel
patterns of Northern Taiwan tourists. RFM (Recency, Frequency and Monetary
Value) was applied to identify valuable travelers, C4.5 decision tree and associ-
ation rule were then applied to discover the traveling pattern and rule. Gul Gokay
Emel, Cagatan Taskin and Omer Akat [3] applied Apriori rule mining to profile the
domestic tourists of Bursa, Turkey and provided suggestions for relevant tourism
enterprises. Moreover, Brida et al. [4] implemented two-level approach to conduct
cluster analysis based on Italian Christmas Market visitors.
Some researchers used the Decision Tree weighting for improving the accuracy
of classifier such as Kaewchinporn et al. [5] used Decision Tree bagging to weight
features for improve the predictive performance of K-Means. Hall [6] applied
Decision Tree bagging to weight attributes for improve the performance of Naive
Bayes classifier. This study, the researchers enhanced the quality of cluster by
applying the Decision Tree bagging weighted to the features of data. After clusters
were constructed, tree bagging and attributes weighting were applied. K-Means
and Self Organizing Map (SOM) algorithm were then applied to refine the clusters.
Useful knowledge can be served to tourism enterprises.

65.3 Related Algorithms

65.3.1 Two-Level Clustering

There were some researchers who used two-level clustering for market segmen-
tation. Conventional approach of two-level clustering is the combination of
Hierarchical and non-Hierarchical techniques. The advantages of first level are
65 Rules Mining Based on Clustering of Inbound Tourists in Thailand 695

solved the limitation of algorithm on the second level. For example, Punj and
Steward [7] presented two-step clustering by binding of Hierarchical Clustering
(HAC) and K-Means approach. The HAC was used to find the number of cluster
and initial seeds for the input of K-Means algorithm. However, HAC cannot
handle the large data set. Moreover, once a decision is made to combine two
clusters, it cannot be undone. Kuo [8] proposed two-stage clustering by combining
of SOM and K-Means algorithms. In the first stage, the researcher replaced HAC
with SOM for solve the limitation of HAC. His proposed method is slightly better
than the conventional two-stage method.
In this present, we implement SOM and K-Means approach to segment tourist’s
dataset. SOM algorithm was discovered by Kohonen. The SOM algorithm is used
to find the optimum number of cluster by calculating two criterions consist of
RMSSTD and RS [9] which defined as.
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uP P ij
u i¼1::nc nk¼1 ðxk  xj Þ2
u j¼1::d
RMSSTD ¼ u
t P ð65:1Þ
i¼1::nc ðnij  1Þ
j¼1::d

Pd Pnj P Pnij
j¼1 k¼1 ðxk  xj Þ2  i¼1::c k¼1 ðxk  xj Þ2
j¼1::d
RS ¼ Pd Pnj ð65:2Þ
j¼1 k¼1 ðxk  xj Þ2

c is the number of cluster.

d is the number of dimension.
xj is the mean value of jth dimension.
nij is the number of sample in ith cluster, jth dimension.

After the optimum number of cluster was found, the initial seeds were deter-
mined. They were used to input to K-Means method in the second step. For further
calculation of SOM and basic K-Means algorithm are available in Tan [1].

65.3.2 Decision Tree Bagging and Features Weighting

This study, the researchers used Decision Tree bagging for features weighting
because of the factors may have different importance. This method may improve
the quality of cluster. Tree bagging uses a Decision Tree algorithm to construct
n models based on a different diversity of training data. The bagging algorithm was
shown as following [5].
696 W. Yotsawat and A. Srivihok

Algorithm: Bagging
Input: D: data set;
n: the number of models;
- a learning scheme (e.g., decision tree)
Output: A composite model, M*.
Method:
(1) For i=1 to n do
(2) create bootstrap sample, Di, by sampling D with
replacement
(3) use Di to derive a model, Mi;
(4) End for

The attributes weighting technique uses the features which appear in the trees.
After decision trees were constructed, attributes which appeared in each trees were
selected and computed the weight. The weight of attribute was varying on a size of
tree and the position of that attribute appearing in that tree. The computation of
weight for each attribute was defined as [5].

wðk;iÞ ¼ ðheight Mi  j þ 1Þ=ðheight Mi þ 1Þ ð65:3Þ

where wk,i is the weight of attribute k in tree i, height_Mi is the height of tree i and
j is the level of attribute node k in tree i.
All attribute weights from each tree models were calculated for an average
weight by the following equation.

wk ¼ ðwk;1 þ wk;2þþ wk;n Þ=n ð65:4Þ

where wk is the total weight of attribute k and n is the number of tree models.

65.3.3 Apriori Algorithm

Association rule was discovered by Agrawal et al. [10]. It is used in the recom-
mendation systems such as www.amazon.com. Apriori was well-known algorithm
and popular usage in market basket data analysis. It was used to find the rela-
tionship between two or more attributes in the large database. There were two
standard measurements such as minimum support (Minsup) and minimum confi-
dence (Minconf). Support was used to evaluate the statistical importance of a set of
transactions in database such as Sup(X, D) represented the rate of transactions in D
containing the item set X. Confidence represented the rate of transactions in D that
contain item set X and also item set Y. It was defined as Conf(X ? Y) = Sup(X \
Y)/Sup(X, D). The first step of Apriori algorithm was to detect a large item set
with greater than minimum support and the second step was to generate associ-
ation rules with greater than minimum confidence. Moreover, lift value was used
to reduce the possible biases when used the support and confidence values. Lift
was defined as Lift = Conf(X ? Y)/Sup(Y) [11]. The association rules were
65 Rules Mining Based on Clustering of Inbound Tourists in Thailand 697

useful for many applications such as tourism marketing [3], tourism recommen-
dation systems [12, 13], new product development and customer relationship
management [14].

65.4 Methodology

65.4.1 Data Collection

This study, the researchers used secondary data accumulated from the Department
of Tourism, Ministry of Tourism and Sports, Thailand. The total number of
inbound tourist in the data files is 83,402 who arrived to Thailand in a period from
2008 through 2010. Data pre-processing were applied by removing unreliable,
missing values and outliers. So, the total number of observation in the sample was
79,473. The attributes of data are the length of stay (days), age (years), gender,
occupations, annual incomes (US Dollars), average expenditures (Baht per day),
purpose of visit, types of accommodation, tourist origins, places of residence and
types of transportation.

65.4.2 Study Framework

The specification of this study required the applying of data mining technique to
partition the data into segments and to discovered the relationship among the
features of tourist in each segment, respectively. Using cluster analysis and
association rule, this study analyzed tourist behaviors and then extracted knowl-
edge to explore useful information for tourism businesses. The useful knowledge
can be helpful for tourism organizations in order to understand the tourist needs
and their preferences.
So, the research design was divided into two phases included data clustering
phase and association rule mining phase. Preprocessing methods were used to
clean the data. Clustering phase was then conducted by two-level clustering and
weighting features by decision tree bagging. SOM algorithm was performed to find
the optimum number of cluster. The appropriate number of cluster was defined by
the computation of RMSSTD and RS. The cluster labels were assigned to build
decision tree bagging for attributes weighting. Attributes weighting by decision
tree bagging included four cases. Case one, case two and case three consisted of
ten trees which each tree was constructed by 25, 50 and 75 % of random with
replacement tourist data, respectively. Case four was built by overall data for one
tree. The features set were weighted and the data set was refined by the clustering
algorithms. Each cluster was analyzed when the clustering phase was finished.
After some attribute was preprocessed in order to qualify the requirement of
698 W. Yotsawat and A. Srivihok

Tourists Data Set Data Preprocessing

Clustering phase

Rule mining phase Step 1: Finding the optimum

SOM number of cluster and initial
Apriori Rules Mining seeds.

Step 2: Clustering and weighting

Rules Evaluation

K-Means Tree Bagging and

SOM
Features weighting

Conclusion and Discussion

Fig. 65.1 Study framework

Apriori algorithm, association rule mining was performed for each segment. The
study framework was illustrated in Fig. 65.1.

65.5 Experimental Results

65.5.1 Clustering Results

Finding the optimum number of cluster, Table 65.1, eleven clusters are obtained
from SOM method by the computation of RMSSTD and RS. After cluster labels
were assigned to each record, decision tree bagging was constructed for calculate
the weight of each attribute. After that, tourists’ data was recomputed and com-
pared the RMSSTD and RS between data with weighted attributes and data
without weighted attributes. The experimental result indicated that data with
weighted attributes give a better quality. It was illustrated on Table 65.2.
The result of clustering phase was shown on Table 65.3. The differences of
attribute among segments were illustrated on Table 65.3. The result showed no
significant difference on gender, age and occupation. Segment 1 was relatively
dominant among the eleven clusters. It comprised over 31 % (n = 24,777) of
overall tourists (N = 79,473) and can be considers as a homogeneous cluster. The
majority of visitor in Cluster 9 traveled to Thailand for business purpose. The
expenditure per day of Cluster 9 was highest around 4,747 Baht by average. The
tourists of cluster 11 stayed in Thailand for a longest time when compare with the
tourists in other clusters. Thus, they used a variety of transportation’ types. Cluster
3 chose domestic airplane for transportation. Moreover, clustering results
65 Rules Mining Based on Clustering of Inbound Tourists in Thailand 699

Table 65.1 Finding the optimum number of cluster by SOM clustering

#Cluster RMSSTD RS #Cluster RMSSTD RS
2 2.7653 0.1092 9 2.4865 0.2798
3 2.6827 0.1616 10 2.4487 0.3015
4 2.6279 0.1955 11 2.4279 0.3134
5 2.5873 0.2202 12 2.4320 0.3111
6 2.5678 0.2319 13 2.4523 0.2995
7 2.5173 0.2618 14 2.4359 0.3089
8 2.4880 0.2790

Table 65.2 The comparison of RMSSTD and RS between SOM and K-Means for data with
weighted and not weighted attributes (number of cluster = 11)
Algorithms Weighting RMSSTD RS
SOM No 2.4279 0.3134
K-Means No 2.4277 0.3135
SOM Yes 1.6933 0.4395
K-Means Yes 1.7001 0.4350

suggested that the expenditure per day was inverse variation with the length of
stay. In other word, the expenditure of cluster 11 was the least but cluster 11 stayed
in Thailand for the longest term, the expenditure of cluster 9 was the highest but
cluster 9 stayed in Thailand for a short time. Cluster analysis results shown that the
tourism organizations can apply the useful knowledge for market planning, pro-
motion design and other related tourism developments.

65.5.2 Association Rule Mining Results

Clustering methods provided the distinct characteristics of segment but cannot

show the association among the features of data. Thus, Apriori rule mining was
applied to each tourist cluster. The rules with lift value ranging from 1.00 to 1.34
are obtained with minimum rule support (Minsup) of 25–35 % and minimum rule
confidence (Minconf) of 80–85 %. Table 65.4 described a part of association rules
which discovered from each tourist segment. Four significant rules were demon-
strated on Table 65.4.
The experimental results indicated that the tourist segments provided associa-
tion rules which were related to the characteristics of each segment. The rules
could be transformed to if-then clause. Finally, the rules were evaluated by tourism
Table 65.3 The significant characteristics of cluster
700

Factors Clusters
1 2 3 4 5 6 7 8 9 10 11
Sample size (%) 31.18 2.50 5.6 15.47 5.65 3.59 4.41 7.5 10.33 10.14 3.61
Stay’ length (days) 6 12 13 8 9 9 11 11 6 8 17
Income $US (%)
\20 k 43 40 25 44 50 34 36 32 32 32 42
20–40 k 36 29 34 33 30 31 34 34 31 36 33
40–60 k 12 14 24 14 12 18 17 18 17 22 14
Expenditure (THB) 4,439 3,242 4,551 4,174 3,322 4,268 3,568 4,147 4,747 4,215 2,815
Purpose (%)
Holiday 100 85 91 92 94 85 93 90 92 92
Business 3 2 3 1 6 2 2 55 2 1
Other 12 7 5 5 9 5 8 45 6 7
Accommodation (%)
Hotel 100 94 100 100 91 99
Resort 35 60 32 1 89
Guesthouse 38 18 57 100 1
Apartment 23 5 12 7 4 1 9
Zone (%)
America 5 14 11 5 10 12 10 9 5 7 16
East Asia 28 21 11 30 30 22 29 20 21 19 8
Europe 13 27 46 20 24 25 34 38 10 30 55
Oceania 4 7 11 3 6 9 4 12 4 6 8
(continued)
W. Yotsawat and A. Srivihok
 65

Table 65.3 (continued)

Factors Clusters
1 2 3 4 5 6 7 8 9 10 11
South Asia 1 5 4 12 6 3 4 3 17 3 1
SEA 33 17 6 21 19 20 12 11 34 26 5
Destination (%)
Central 100 100 97
Southern 96 91 94
Northern 88 100 95
Eastern 100 100
Transportation (%)
Plane 6 36 100 4 15 29 10 41 5 0 53
Bus 11 37 23 15 45 14 38 25 10 12 68
Train 6 16 10 3 21 6 8 10 6 4 30
Ferry 7 36 9 26 28 30 5 9 62
Other 65 60 72 45 76 48 63 54 80 42 71
n = 79,473
Rules Mining Based on Clustering of Inbound Tourists in Thailand
701
702 W. Yotsawat and A. Srivihok

Table 65.4 Four significant rules for each tourist cluster with Minsup = 25–35 %, Min-
conf = 80–85 % and lift = 1.00–1.34
# Antecedents Consequent
R11 Pd = Holiday, A = Hotel, Ec = 4,514–6,832 Baht/ Db = Central
day
R12 P = Holiday, Occupation = Professional Aa = Hotel, D = Central
R13 A = Hotel, Age = 25–34 years P = Holiday, D = Central
R14 P = Holiday, Zone = East Asia A = Hotel, D = Central
R21 A = Guesthouse D = Northern, P = Holiday
R22 D = Northern, Income = less than 20,000 $US P = Holiday
R23 Income = less than 20,000 $US D = Northern
R24 Te = Plane D = Northern
R31 D = Southern, Zone = Europe T = Plane, P = Holiday
R32 T = Plane, Age = 25–34 years D = Southern
R33 T = Plane, P = Holiday, Gender = Male D = Southern, A = Hotel
R34 D = Southern, Age = 25–34 years A = Hotel
R41 D = Eastern, Gender = Male, T = Other A = Hotel
R42 D = Eastern, Income = 20,000–39,999 $US A = Hotel, P = Holiday
R43 D = Eastern, A = Hotel, Age = 25–34 years P = Holiday
R44 P = Holiday, Income = less than 20,000 $US D = Eastern, A = Hotel
R51 P = Holiday, Age = 15–24 years D = Central
R52 Income = less than 20,000 $US D = Central, P = Holiday
R53 D = Central, P = Holiday A = Resort
R54 P = Holiday, T = Other, Income = less than 20,000 D = Central
$US
R61 D = Northern, Gender = Female A = Hotel, P = Holiday
R62 A = Hotel, P = Holiday, Gender = Female D = Northern
R63 D = Northern, Income = 20,000–39,999 $US A = Hotel
R64 Occupation = Professional D = Northern, A = Hotel
R71 Zone = Europe D = Eastern, P = Holiday
R72 D = Eastern, T = Bus P = Holiday
R73 D = Eastern, A = Guesthouse P = Holiday
R74 Age = 25–34 years D = Eastern
R81 T = Ferry A = Guesthouse,
D = Southern
R82 Occupation = Professional A = Guesthouse
R83 A = Guesthouse, Income = 20,000–39,999 $US P = Holiday
R84 A = Guesthouse, Age = 25–34 years D = Southern
R91 A = Hotel, P = Business D = Central
R92 T = Other, Gender = Male, P = Business D = Central, A = Hotel
R93 D = Central, Occupation = Professional A = Hotel
R94 D = Central, Zone = South East Asia A = Hotel
R101 P = Holiday, D = Southern, Gender = Female A = Hotel
(continued)
65 Rules Mining Based on Clustering of Inbound Tourists in Thailand 703

Table 65.4 (continued)

# Antecedents Consequent
R102 Income = 20,000–39,999 $US A = Hotel
R103 P = Holiday, Gender = Female D = Southern
R104 D = Southern, Gender = Male A = Hotel, P = Holiday
R111 D = Southern, T = Plane P = Holiday, A = Resort
R112 P = Holiday, Zone = Europe D = Southern, A = Resort
R113 D = Southern, P = Holiday, T = Ferry A = Resort
R114 D = Southern, Income = less than 20,000 $US A = Resort
a
A = Accommodation
b
D = Destination
c
E = Expenditure
d
P = Purpose
e
T = Transportation

Table 65.5 Example of implementation of clustering and association rules mining results
Knowledge founding Actionable activities which should be focused
on
Most of the extracted rules on cluster 1 belong Tour agencies design package tours as a base
to ‘‘Holiday’’ purpose, ‘‘Hotel’’ for segmentation., Tourism recommendation
accommodation and ‘‘Central’’ destination systems
Cluster 11 is a ‘‘price-conscious’’ segment Marketing managers should be focused on
pricing strategies and planning
Cluster 9 is a ‘‘hi-end’’ segment Marketing managers should be focused on the
quality of services and products
For cluster 3, if ‘‘Destination = Southern’’ and Information center should be focused on
‘‘Zone = Europe’’ then tourist attraction and transportation for
‘‘Transportation = Plane’’ and holiday on Southern part of Thailand
‘‘Purpose = Holiday’’
For cluster 9, if ‘‘Destination = Central’’ and Hotel at Central part of Thailand should be
‘‘Purpose = Business’’ then prepared their location and relationship
‘‘Accommodation = Hotel’’ between business activities

professionals. Some rule was accepted by minimum criterions but it was rejected
by experts filtering. However, the most difficult of this study was the translation of
the segmentation and association rules to the suggestion of tourism management.
Table 65.5 shown the activity which tourism organizations should be focused.

65.6 Conclusion

Data mining can be extracted hidden information and patterns on the inbound
tourist data. This study focused on the role of data mining for tourism industries in
Thailand. Data clustering and association rule mining were chose as the data
704 W. Yotsawat and A. Srivihok

mining methods in order to discover hidden knowledge. The clustering results

indicated that inbound tourists in Thailand consisted of various segments with
different profiles. Tourist data was segmented into eleven clusters by two-level
clustering. Decision tree bagging method was then applied in order to enhance the
quality of cluster. The eleven segments were analyzed to identify tourists’
behavior patterns and their preferences. The experimental results indicated that the
distinct characteristics among clusters can be helpful for tourism organizations in
order to define the market planning or strategic making such as tourism man-
agement or package tour designing. After eleven segments were analyzed, asso-
ciation rule mining was applied in order to discover the relationship among the
features of tourist in each segment. Minimum support and minimum confidence
were set to filter the rules which generated by Apriori algorithm. Finally, the rules
were evaluated by experts. Association rules mining results indicated that the
tourist segments provide association rules which related to the characteristics of
each segment. The rules can be implemented on recommendation systems or
marketing action which should be focused on the rules found.
Future studies of data mining on tourism can focus on other related features
such as cultural, socio-economic variables and values added from incompletely
data such as RFM analysis. Finding a good way to weight features is very
important for optimizing clustering results of many real world dataset. Moreover,
the researchers should be suggested the data collector to collect more information
for data analysis and researching.

Acknowledgments This work was funded by Faculty of Science and the Graduate School of
Kasetsart University.

References

1. Tan, P.N., Steinbach, M., Kumar, V.: Introduction to Data Mining. Pearson Education Inc,
San Francisco (2006)
2. Wong, J., Chen, H., Chung, P., Kao, N.: Identifying valuable travelers and their next foreign
destination by the application of data mining techniques. Asia Pac. J. Tourism Res. 11(4),
355–373 (2006)
3. Emel, G.G., Taskin, C., Akat, O.: Profiling a domestic tourism market by means of
association rule mining. Anatolia: Int. J. Tourism Hospitality Res. 18(2), 334–342 (2007)
4. Brida, J.G., Disegna, M., Osti, L.: Segmenting visitors of cultural events by motivation: a
sequential non-linear clustering analysis of Italian Christmas market visitors. Expert Syst.
Appl. 39(13), 11349–11356 (2012)
5. Kaewchinporn, C., Nattakan,V., Vongsuchoto, S.: A combination of decision tree learning
and clustering for data classification. In: Proceedings of 2011 Eight International Join
Conference on Computer Science and Software Engineering (JCSSE), pp. 11–13 (2011)
6. Hall, M.: A decision tree-based attribute weighting filter for Naïve Bayes. Knowl.-Based
Syst. 20(2), 120–126 (2007)
7. Punj, G., Steward, D.W.: Cluster analysis in marketing research: review and suggestions for
applications. J. Mark. Res. 20(2), 134–148 (1983)
65 Rules Mining Based on Clustering of Inbound Tourists in Thailand 705

8. Kuo, R.J., Ho, L.M., Hu, C.M.: Integration of self-organizing feature map and K-means
algorithm for market segmentation. Comput. Oper. Res. 29(11), 1475–1493 (2002)
9. Kovacs, F., Legany, C., Babos, A.: Cluster Validity Measurement Techniques. In: World
Scientific and Engineering Academy and Society (WSEAS), pp. 388–393 (2006)
10. Agrawal, R., Imilienski,T., Swami, A.: Mining association rules between sets of items in
large databases. In: Proceedings of the 1993 ACM SIGMOD International Conference on
Management of Data, pp. 207–216. ACM Press, New York, NY, USA (1993)
11. Wang, Y.F., Chuang, Y.L., Hsu, M.H., Keh, H.C.: A personalized recommender system for
the cosmetic business. Expert Syst. Appl. 26(3), 427–434 (2004)
12. Aghdam A. R., Mostafa, K., Dong, C. et al.: Finding interesting places at Malaysia: a data
mining perspective. In:Math. Comput. Contemp. Sci., pp. 89–93 (2013)
13. Juwattanasamrn, P., Supattranuwong, S., Sinthupinyo, S.: Applying data mining to analyze
travel pattern in searching travel destination choices. Int. J. Eng. Sci. (IJES) 2(4), 38–43
(2013)
14. Liao, S., Chen, Y., Deng, M.: Mining customer knowledge for tourism new product
development and customer relationship management. Expert Syst. Appl. 37(6), 4212–4223
(2010)
Chapter 66
Designing a New Model for Worm
Response Using Security Metrics

Madihah Mohd Saudi and Bachok M. Taib

Abstract Nowadays, worms are becoming more sophisticated, intelligent and

hard to be detected and responded than before and it becomes as one of the main
issues in cyber security. It caused loss millions of money and productivities in
many organizations and users all over the world. Currently, there are many works
related with worm detection techniques but not much research is focusing on worm
response. Therefore, in this research paper, a new model to respond to the worms
attack efficiently is built. This worm response model is called as eZSiber, inspired
by apoptosis or also known as cell-programmed death. It is a concept borrowed
from human immunology system (HIS), where it has been mapped into network
security environment. Once the user’s computer detects any indication of the
worm attacks, the apoptosis is triggered. In order to trigger the apoptosis, security
metrics plays a very important role in identifying the weight and the severity of the
worm attacks. In this model, the static and dynamic analyses were conducted and
the machine learning algorithms were applied to optimize the performance. Based
on the experiment conducted, it produced an overall accuracy rate of 99.38 %
using Sequential Minimal Optimization (SMO) algorithm. This performance cri-
teria result indicated that this model is an efficient worm response model.


Keywords Worm response Security metrics Apoptosis   Static analysis 

Dynamic analysis Sequential minimal optimization (SMO)

M.M. Saudi (&)  B.M. Taib

Faculty of Science and Technology, Universiti Sains Islam Malaysia (USIM),
71800 Bandar Baru Nilai, Negeri Sembilan, Malaysia
e-mail: [email protected]
B.M. Taib
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 707

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_66
708 M.M. Saudi and B.M. Taib

66.1 Introduction

Australia Computer Emergency Response Centre (AuSCERT) reported that 7,962

cases of the compromised Australia web sites were serving malwares in the year
2013 [1]. While statistics from Malaysia Computer Emergency Response Team
(MyCERT) show that malwares contributes as one of the major cyber threats in
Malaysia [2]. In this research, a worm is defined as a malicious program that can
replicate itself, moving from one computer to another or can propagate via a
network without human intervention or an owner’s consent. One of the examples
of worm that has caused chaos all over world is called as the Stuxnet worm. It has
infected many computers at Iran’s nuclear station and moreover it has also infected
computers all over the world primarily at Indonesia, India and United States of
America [3]. Furthermore, earlier in the year 2012, Ramnit worm has succeeded
stole 45,000 Facebook passwords and after a year the worm characteristics have
been improved and it is capable to steal bank password and account details [4, 5].
If a comparison between current trends and those of 10 years ago is made, these
historical worm attacks and infections ensured the reputation of the attacker and
thus gaining respect from other attackers or hackers was paramount [6]. In con-
trast, the motivation for cyber-attacks in the past 5 years is based on profit.
When dealing with worm attacks, one of the hardest jobs to do is to identify
either the attack is genuine or not. Then once the attack has been identified, the
next step is to decide the severity level. If the attack caused damage to the
infrastructure and stole confidential information from the organization or user, it
can be categorized as high severity. Prior assigning the severity level, a proper
procedure or mechanism should be used as guidance. As for this research, security
metrics has been identified as one of the efficient way how to assign severity level
for worm attacks which then leads to the worm response mechanism. Therefore,
the objectives of this research paper are to produce a new model called as eZSiber
for worm response by applying security metrics and apoptosis concept, where the
apoptosis is mapped from human immunology system (HIS) into computer
security perspective and then to evaluate the proposed model. Prior to that, the
existing worm response techniques are investigated and evaluated. As for security
metrics, it is very important method on how to assign the worm attacks severity
level for the apoptosis to be triggered.
The apoptosis that is also known as the cell-programmed death is applied to
avoid the worm from propagating to other computers in the same network or via
the Internet. The apoptosis will only be triggered based on five (66.5) main
characteristics of a worm, which are the payload, operating algorithm, infection,
activation and propagation. These five main characteristics of a worm are further
refined and reused by assigning it with a weight and severity, where the security
metrics has been applied to trigger the apoptosis.
66 Designing a New Model for Worm Response Using Security Metrics 709

As mentioned earlier, security metrics is very important method to assign the

worm severity level. Security metrics is a method that helps to quantify, classify
and measure information on security operations. In security metrics, the studied
threats are defined, and then the threats are transformed into metrics or repre-
sentations that can easily be measured. Security metrics can also be measured
based on the perimeter defence, control and coverage, availability and reliability
and application risks. All these measurements were already taken into consider-
ation when the worm analysis was conducted. Therefore, as a result, the weight
and severity performance and value are tested based on data criticality level,
infrastructure availability and loss of productivity. The main reason why security
metrics method has been chosen in this research is due to its capabilities to make
the job of defining, understanding, identifying and measuring information security
efficient, accurate, measurable and reliable. This is also supported by Atzeni and
Lioy [7], where they state that work can be more profitable if it is enhanced using
the security metrics and is more efficient if it is measurable.
For eZSiber worm response model, the security metrics is applied based on data
criticality, infrastructure availability and loss of productivity. Then these 3 factors
were used as the basis for assigning a weight and severity value. The above 5 main
characteristics of a worm were extracted from STAKCERT worm classification,
which were used as the input for this worm response model. The STAKCERT
worm classification is not discussed in this research paper and can be referred in
paper by Saudi et al. [8].
This paper is organised as follows. Section 66.2 presents the related works and
the challenges in applying weight, severity and security metrics into the worm
response model. Section 66.3 explains the methodology used in this research paper
which consists of static and dynamic analyses and the architecture of the controlled
laboratory environment. Section 66.4 presents the research findings which consists
the results of the testing and evaluation of the proposed worm response model.
Section 66.5 concludes and summarises the future work of this research paper.

66.2 Related Works

In applying apoptosis in worm response model, the main challenges, which should
be considered thoroughly is the method of assigning the apoptosis. Therefore,
based on the experiments and analysis conducted in this research, weight and
severity (assigned using security metrics) are identified as two important factors
that trigger apoptosis. There are a few studies, which have considered weight as
part of their work. Examples are those of Su [9], who built a real time anomaly
detection system for denial of service (DoS) attacks using weighted k-nearest
neighbour classifiers, Siddique and Maqbol [10], who used weighting in software
clustering, Kim et al. [11], who used weight as part of the log analysis of incident
response in a DSS system, Fisch et al. [12], who used weight to optimise radial
basis function neural networks for an intrusion detection system and Middlemiss
710 M.M. Saudi and B.M. Taib

and Dick [13], who used weighted feature extraction using a genetic algorithm for
an intrusion detection system. Based on these works, it can be concluded that there
is no standard way of assigning weight, which has been seen as an important
feature in increasing the accuracy or optimizing the performance of different works
in different fields. Therefore, weight has been integrated in the eZSiber model,
where security metrics and frequency analysis have been used to retrieve the rank
and the value of the weights. Then, the weights are used for assigning the level of
severity which triggers the apoptosis.
In a study conducted by Miles [14], he assigned a severity incident into three
categories, which are high, medium and low. The high severity involves incidents
with long term effects to the business or critical system for examples root access,
denial of service (DoS) and it also involves with unauthorized privilege (root),
limited access (user), unsuccessful attempt, utilisation of services and probe, poor
security practices, malicious logic, hardware, software or infrastructure failure and
espionage. Medium severity involves non-critical system and detection of initial
attack and low severity involves detection on reconnaissance, threats of future
attacks and rumours of security incidents. On the other hand, Reese [15] defined
high severity as posing a threat to an entire autonomous system, such as a uni-
versity network; that is a threat to the operation of critical network systems that
threatens one or more applications that are integral to daily university functions.
Medium severity involves a risk to isolated and non-production university systems
and low severity involves minimal exposure of threats. By referring to the previous
studies conducted in assigning severity, it can be concluded that severity must
consider the data criticality, infrastructure availability and loss of productivity
where these have been mapped in security metrics.

66.3 Methodology

In order to produce a new worm response technique, the researchers’ had con-
ducted few experiments and researches. A controlled laboratory environment is
created to conduct the experiment as illustrated in Fig. 66.1. It is a controlled

Fig. 66.1 Controlled

laboratory architecture
66 Designing a New Model for Worm Response Using Security Metrics 711

laboratory environment and almost 80 % of the software used in this testing is an

open source or available on a free basis. No outgoing network connection is
allowed for this architecture. The static and dynamic analyses were conducted.
The dataset used for this research are from VX Heaven [16] and Offensive
Computing [17]. Many studies have used these dataset in their experiment [18,
19]. In these datasets there are many variant of malwares and benign files. The
Knowledge Discovery and Data Mining (KDD) Processes as displayed in
Fig. 66.2, were applied to these datasets. As a result, 160 datasets which consist of
variants of the Windows worm and benign executable have been used for this
research.
For this research, enhancements have been made to the KDD data pre-pro-
cessing and pattern extraction process. Under the data pre-processing process, the
static and dynamic analyses are implemented using the incident response standard
operating procedures (SOP). While under the pattern extraction process, statistical
methods comprising Chi-square and symmetric measure and security metrics are
also introduced, as illustrated in Fig. 66.2. The details how the security metrics is
applied in eZSiber worm response model can be referred in Table 66.1.
Furthermore, to retrieve the exact number of values for each of the worm’s
attributes, relative frequency is used. Then these values were further tested with
different algorithms to identify the best overall accuracy value. The equation used
for relative frequency is shown in Eq. 66.1.
r
rfn ðEÞ ¼ ð66:1Þ
n

Iterative

If security metrics
not involve

Chi-square and
Data
Input data symmetric Security metrics Data mining Postprocessing Knowledge
preprocessing
measure

-Feature selection using static -Determine -Data criticality, -Similarity -Pattern

and dynamic analysis, relationship between availability and between interpretation
statistical analysis & incident worms characteristics productivity attributes/objects
response -Quantify relationship -Clustering
-Cleaning data to remove strength -Classification
noise and duplication
-Data transformation

Fig. 66.2 KDD processes integrated with security metrics

712 M.M. Saudi and B.M. Taib

Table 66.1 Security Metrics in eZSiber worm response model

Security metrics processes Applying security metrics in eZSiber
Define worm threats Yes
Represents worm threats into metrics Yes. Worm data is represented based on
payload, infection, activation, propagation and
operating algorithm
Understand and identify the vulnerability, Yes
flaw, problem, weakness and damage to • Run the static and dynamic analysis
security infrastructure • Identify the need to assign weight and
severity value to assign the countermeasure
process
Check the performance of the existing Yes
countermeasures • Integrate and run data mining using JAVA-
WEKA to check the accuracy rate of weight
and severity assigned
Recommend any technology or Yes
countermeasure process for improvement Apoptosis to isolate the most severe worm
attacks

where: rfn = relative frequency, E = number of events, n = total number of

experiments conducted, r = number of times an event occurs.
Relative frequency is another term for proportion. It is the value calculated by
dividing the number of times an event occurs by the total number of times an
experiment is carried out. Since the cases involved a long run relative frequency,
probability was seen as the best way to calculate the weight. It was in the range of
0–1. The equation is simplified in Eq. 66.2. Based on the frequency analysis, the
worm’s attributes are ranked.

PðEÞ ¼ lim rfn ðEÞ ð66:2Þ

n!1

where, P(E) = number of outcomes corresponding to event E/total number of

outcomes, rfn = relative frequency.
There are several performance parameters that need to be measured during the
experiment which are as in the Eqs. 66.3–66.5. TPR represents the true positive
rate, FPR represents as false positive rate, TP represent as true positive, FN rep-
resents as false negative, FP represents as false positive and TN represents as true
negative.

TPR ¼ TP=ðTP þ FN Þ ð66:3Þ

66 Designing a New Model for Worm Response Using Security Metrics 713

FPR ¼ FP=ðFP þ TN Þ ð66:4Þ

Accuracy ¼ ðTP þ TN Þ=ðTP þ TN þ FP þ FNÞ ð66:5Þ

66.4 Findings

Two important attributes that are being measured when conducting this research
are weight and severity, where the security metrics has been applied. The data
criticality, infrastructure availability and loss of productivity are used as a basis
and guide for assigning weight and severity value. While, frequency analysis was
conducted to support the fact the effectiveness of the weight and severity formed.
Furthermore, based on the frequency analysis testing results in Table 66.2, the
ranking of worm attributes was identified as follows: (1) Payload; (2) Infection; (3)
Activation; (4) Propagation and (5) Operating algorithm.
160 cases were tested from the datasets. Then, the accuracy of each case with an
assigned severity value was further tested using different data mining algorithms
by means of WEKA (Java-open source software) [20]. Table 66.3 presents the
results where five different algorithms were tested. These were the Multilayer
Perceptron, SMO, Naïve Bayes, J48 and IBk. The testing was carried out to
identify the overall accuracy for the assigned class of severity. Each of the worm’s
attributes, which were payload, infection, activation, propagation and operating
algorithm, were assigned with either a high, medium or low weight value. Based
on the testing conducted, the average overall accuracy for each algorithm was
more than 98.13 % with SMO had the highest overall accuracy of 99.38 %.
The above results have been used as the input for the eZSiber worm response
model. The pseudocodes of the model are simplified in Fig. 66.3. All of these
pseudocodes explain how the weight and severity was assigned for each attribute
of the worm. The attributes were the payload, infection, activation, propagation
and operating algorithm. Covering algorithm, also known as the separate-and-
conquer algorithm is used to form the eZSiber worm response model.

Table 66.2 Frequency analysis results

Security Payload Infection Activation Propagation Operating Severity
metrics algorithm
processes
High 150(0.938) 129(0.806) 121(0.756) 22(0.138) 8(0.05) 430(0.538)
Medium 9(0.056) 26(0.163) 33(0.206) 0 152(0.95) 220(0.275)
Low 1(0.006) 5(0.031) 6(0.038) 138(0.863) 0 150(0.188)
Total 160(1) 160(1) 160(1) 160(1) 160(1) 160(1)
 714

Table 66.3 Severity Results Using Different Algorithms

Classifier Multilayer Perceptron SMO Naïve Bayes J48 IBk
Severity in (%) H M L H M L H M L H M L H M L
TPR 100 83.3 0 100 100 0 99.4 100 0 98.7 83.3 0 100 83.3 0
FPR 16.7 0 0 0 0 0 0 0.6 0 16.7 1.3 0 16.7 0 0
FNR 0 16.7 0 0 0 0 0.6 0 0 1.3 16.7 0 0 16.7 0
OA 99.38 99.38 0 0 0 0 99.38 99.38 0 98.1 98.13 0 99.38 99.38 0
TPR = true positive rate, FPR = false positive rate, FNR = false negative rate, OA = overall accuracy, H = high, M = medium, L = low
M.M. Saudi and B.M. Taib
66 Designing a New Model for Worm Response Using Security Metrics 715

Given:
- Set security metrics.
- Set worm attributes: {payload, infection, activation, operating algo-
rithm and propagation}.
- Set frequency analysis.

Output:
- Weight ranks.
- Severity ranks.
- Weight values.
- Severity values.
- Triggers or halts Apoptosis.

Algorithms:
1) Apply security metrics to worm attributes.
a) Go to Weight_cases to determine the weight ranks.
b) Go to Severity_cases to determine the severity ranks.
2) Apply frequency analysis to worm attributes.
a) Go to Frequency_cases to compute the weight and severity values.
3) Apply apoptosis to Severity_cases.
a) Go to Apoptosis_cases to trigger the apoptosis.

Fig. 66.3 eZSiber worm response model pseudocodes integrated with security metrics

66.5 Conclusions and Future Works

As conclusion, using the security metrics, the severity level of the worm attacks can
be easily assigned by using weight. Once the severity level is high, the apoptosis is
triggered to avoid the worm from further propagate. The eZSiber worm response
model was developed by using the security metrics to assign the weight and the
severity of the worm attacks. This model can be used as a reference and comparison
by other researchers with the same interests. For future work, different human
immunology system (HIS) concept will be tested to the dataset produced from this
research. This paper is part of a larger project to build up an automated malware
response model. Ongoing research will include other malware classification and the
development of software to automate the malware dataset cleanup.

Acknowledgments The authors would like to express their gratitude to Universiti Sains Islam
Malaysia (USIM) for the support and facilities provided. This research paper is supported by
Universiti Sains Islam Malaysia (USIM) grant [PPP/GP/FST/SKTS/30/11912], [PPP/FST/SKTS/
30/12812], [PPP/FST/SKTS/30/12812] and Research Management Centre, Universiti Sains Islam
Malaysia (USIM).

References

1. Australia Computer Emergency Response Team (AusCERT).: AusCERT Incident Metrics.

https://www.auscert.org.au/render.html?it=17856 (2013). June 2013
2. Malaysia Computer Emergency Response Team (MyCERT).: MyCERT Incident Statistics.
http://www.mycert.org.my/en/services/statistic/mycert/2013/main/detail/914/index.html
(2013)
716 M.M. Saudi and B.M. Taib

3. Telgraph Media Group Limited.: Computer worm infects Iran’s nuclear station. http://www.
telegraph.co.uk/news/worldnews/middleeast/iran/8026284/
Computer-worm-infects-Irans-nuclear-station.html (2010)
4. Warwick, A.: Ramnit worm steals 45,000 Facebook passwords. http://www.computerweekly.
com/news/2240113383/Ramnit-worm-steals-45000-Facebook-passwords (2012)
5. Press Trust of India.: Bank details stealing Ramnit malware hits Indian cyberspace. http://
gadgets.ndtv.com/internet/news/
bank-details-stealing-ramnit-malware-hits-indian-cyberspace-358719 (2013)
6. Whitty, B.: Why do people create computer Viruses?. Technibble.com. http://www.
technibble.com/why-do-people-create-computer-viruses/ (2007) Accessed 27 Feb 2014
7. Atzeni, A., Lioy, A.: Why to adopt a security metric? A brief survey. In: Gollmann, D.,
Massacci, F., Yautsiukhin, A. (eds.) Quality of Protection Security Measurements and
Metrics, pp. 1–12. Springer, Berlin (2006)
8. Saudi, M.M., Cullen, A.J. Woodward, M.: Efficient STAKCERT KDD processes in worm
detection, world academy of science. Eng. Technol. J. 79, 453–457 (2011) (pISSN 2010-
376X, eISSN 2010-3778)
9. Su, M.-Y.: Real-time anomaly detection systems for Denial-of-Service attacks by weighted
k-nearest-neighbor classifiers. Expert Syst. Appl. 38(4), 3492–3498 (2011)
10. Siddique, F., Maqbool, O.: Analyzing term weighting schemes for labeling software clusters.
In: Proceedings of 15th European Conference on Software Maintenance and Reengineering,
pp. 85–88 (2011)
11. Kim, H.K., Im, K.H., Park, S.C.: DSS for computer security incident response applying CBR
and collaborative response. Expert Syst. Appl. 37, 852–870 (2010)
12. Fisch, D., Hofmann, A., Sick, B.: On the versatility of radial basis function neural networks: a
case study in the field of intrusion detection. Inf. Sci. 180(12), 2421–2439 (2010)
13. Middlemiss, M.J., Dick, G.: Weighted feature extraction using a genetic algorithm for
intrusion detection. Proc. Evol. Comput. 3, 1669–1675 (2003)
14. Miles, S.G.: Incident response part #2: identification. http://www.securityhorizon.com/
whitepapersTechnical/IncidentResponsepart2.pdf (2001)
15. Reese, R.L.R.: Incident handling: an orderly response to unexpected events. In: Proceedings
of the 31st Annual ACM SIGUCCS Conference on User Services (SIGUCCS 03), pp. 21–24.
Texas, USA, Sept 2003, ACM, New York, pp. 97–102 (2003)
16. Heaven, VX.: Computer virus collection. http://vxheaven.org/vl.php (2014)
17. Offensive Computing.: Malware search. http://www.offensivecomputing.net (2014)
18. Schultz, M.G., Eskin, E., Zadok, E., Stolfo, S.J.: Data mining methods for detection of new
malicious Executables. In: Proceedings of the 2001 IEEE Symposium on Security and
Privacy, IEEE Computer Society, pp. 38 (2001)
19. Henchiri, O., Japkowicz, N.: A feature Selection and evaluation scheme for computer virus
detection. In: Proceedings of the Sixth International Conference on Data Mining, 2006
(ICDM ‘06), pp. 891–895. IEEE Xplore, Hong Kong (2006)
20. Mark, H., Eibe, F., Geoffrey, H., Bernhard, P., Peter, R., Ian, H.W.: The WEKA data mining
software: an update. SIGKDD Explor. 11(1), 10–18 (2009)
Chapter 67
Neural Network Training Algorithm
for Carbon Dioxide Emissions Forecast:
A Performance Comparison

Herrini Mohd Pauzi and Lazim Abdullah

Abstract Artificial neural network with many types of algorithms is known as an

efficient tool in forecasting as it is able to handle nonlinearity behaviour of data. This
paper investigates the performances of Levenberg-Marquardt and gradient descent
algorithms of back propagation neural networks carbon dioxide emissions forecast.
The inputs for the model were selected and the ANNs were trained using the Malaysian
data of energy use, gross domestic product per capita, population density, combustible
renewable and waste and carbon dioxide intensity. The forecasting performances were
measured using coefficient of determination, root means square error, mean absolute
error, mean absolute percentage error, number of epoch and elapsed time. Comparison
between these algorithms show that the Levenberg-Marquardt was outperformed the
gradient descent in carbon dioxide emissions forecast.

67.1 Introduction

It is a well-known fact that carbon dioxide (CO2) emissions are largely responsible
for global warming problems. CO2 emissions happen when the gas is released into
the atmosphere over a specific area and of time either through natural processes or
human activities [1]. Due to overzealous of the growing of development over the
world, continuing growth of the world population and acceleration of industrial-
ization and urbanization, there are continual increment of CO2 emissions over
these few decades. According to Jana et al. [2], there are two major processes that
could reduce the CO2 emissions: by reducing anthropogenic emissions of CO2 and
by creating and/or enhancing carbon sink in the biosphere.
A considerable amount of literatures have been published to explain the rela-
tionship between CO2 emissions and socio economic variables. One of the possible

H.M. Pauzi  L. Abdullah (&)

School of Informatics and Applied Mathematics, Universiti Malaysia Terengganu,
21030 Kuala Terengganu, Terengganu, Malaysia
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 717

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_67
718 H.M. Pauzi and L. Abdullah

methods in estimating the relationship is through forecasting modeling. For exam-

ple, Brondfield et al. [3] applied a linear model regression downscaling to model on-
road CO2 emissions at Boston, Massachusetts and tested the approach with surface-
level CO2 observations. Another study applied linear regression analysis to deter-
mine the strength of correlation between methane (CH4) and CO2 concentrations and
barometric pressure [4]. On the other hand, Shih and Tsokos [5] have developed two
statistical models for predicting the CO2 emissions and the atmosphere in the United
States. A statistical modeling of CO2 emissions in Malaysia and Thailand con-
structed by Hui et al. [6], where a few statistical tests were carried out to ensure that
the best regression model will be selected for further analysis. However, a major
problem with forecasting model is the linearity of data. It has been admitted that the
forecasting of the air quality such as CO2 emissions are sometimes experienced the
problems of non-linearity of the selected data.
Apart from linear models, another category of exploring the relationship between
CO2 emissions and causal variables is non linear model. Artificial neural network
(ANN) is one of the most widely used models in this category. ANN is a modelling
technique that can determine non-linear relationships between variables in input
datasets and variables in output datasets [7]. By its learning capability, the required
information could be extracted directly from the data [8]. It is an artificial intelli-
gence method which facilitates nonlinear state space through function approxima-
tions in constructing a dynamic model. Its ability to handle nonlinearity data could
provide better accuracy of forecasting. Numerous studies have attempted to explain
the success of ANN with nonlinearity data. The networks have been applied in
several areas such as speech recognition [9–11], dynamic modelling [12–14],
forecasting in financial field [15–17], monitoring water quality [18], just to name a
few. The networks also have been applied in several of fields including gas emis-
sions. Since data of air pollution emissions usually show its non-linearity behaviour,
many studies have used the model to forecast air pollutant emissions [7, 19–21].
Literature reviews shown that the gradient descent learning algorithm is the
most frequently used as the ANN training algorithm [22–25]. The algorithm lets
the network learns the relationship between stipulated input-output data pairs in
supervised manner. Although many different types of training algorithms appear in
recent neural network literature, it is difficult to know which algorithm works best,
in terms of convergence speed and accuracy. There many influential factors of
ANN that determine an accuracy of the network. The factors including complexity
of problem, number of datasets used in training, number of weights and biases in
network, error goal seem to have influence [26]. There are many substantial
research have been carried out to alleviate this problem and to date, comparison
study is still one of the possible mechanisms for ascertaining which algorithm
works best under certain specific cases or problems.
Tongyu et al. [25] conducted a study to recognize transient power based on
back propagation (BP) neural network theory. They had applied three different
algorithms, the gradient descent, gradient descent with momentum and the LM
algorithm to train the BP network. Obviously, they found that the LM outper-
formed the other two algorithms.
67 Neural Network Training Algorithm for Carbon Dioxide Emissions … 719

Another researchers, presented multi-layer perceptron model with a novel hybrid

training method to perform the forecasting of ozone layer [27]. The training method
synergistically couples a stochastic particle swarm optimization algorithm and a
deterministic Levenberg–Marquardt (LM) algorithm, which aimed at exploiting the
advantage of both. From their comparison of error analysis, they concluded that the
multi-layer perceptron with hybrid algorithm was the best training method followed
by LM algorithm and particle swarm optimization algorithm.
On the other hand, Ghaffari et al. [28] carried out a study where five different
training algorithms belonging to three classes; gradient descent, LM and genetic
algorithm were used to train ANN containing a single hidden layer of four nodes.
Besides of trying to model the effect of two causal factors for bimodal drug
delivery, the researchers were trying to compare the performance of the afore-
mentioned neural network training algorithms. The result of the study showed that
back propagation (gradient descent) outperformed the others followed by LM,
quick propagation and genetic algorithm.
A study by Piotrowski and Napiorkowski [29], investigated on eight different
Evolutionary Computation optimization methods, mostly from Differential Evolu-
tion family, to multilayer perceptron neural network training for daily rainfall-runoff
forecasting. The overall performance of the LM algorithm and the Differential
Evolution with Global and Local Neighbours method for neural networks training
turns out to be superior to other Evolutionary Computation-based algorithms.
Based on the mentioned comparison studies, it is observed that the LM optimi-
zation must be considered as one of the most efficient algorithms due to its con-
vergence speed. On the other hand, the gradient descent is the most popular and basic
training algorithm discussed by researchers. It has been suggested that LM algorithm
has better convergence properties than the gradient descent algorithm based on an
investigation case of emissions from biodiesel fuelled transit buses [30]. The
investigation relies too heavily on coefficient of determinations to compare the
performance of conjugate gradient, gradient descent and LM in prediction of bio-
diesel emissions.
However, the performances of the LM algorithm against the gradient descent
algorithm specifically in a case of relationship between CO2 emissions data and its
socio economic and demographic variables are not fully explored. Owing to the facts of
the LM and gradient descent algorithm above, this paper seeks to address the issue of
forecasting performance between these two ANN algorithms. Performances between
these two algorithms will be compared as to decide which algorithm could provide a
good forecasting tool of ANN and prevent loss of predictive power for the forecasting.

67.2 Simulation Data

ANNs need a considerable amount of historical data to be trained; upon the

satisfactory training an ANN should be able to provide output for previously
‘‘unseen’’ inputs [31]. The selection of input variables for an ANN forecasting
720 H.M. Pauzi and L. Abdullah

model is a key issue, since irrelevant or noisy variables may have negative effects
on the training process, resulting to unnecessarily complex model structure [32].
This study collects annual historical data for Malaysia from 1980 to 2008. The
data obtained from the website of World Bank’s World Development Indicators.
Gross domestic product per capita (GDP) with the unit constant 2000 US$, energy
use in kg of oil equivalent per capita, population density (people per sq. km of land
area), combustible renewable and waste (% of total energy), CO2 intensity (kg per
kg of oil equivalent energy use) were selected as the input variables for CO2
emission model [33].
Construction of the neural networks (NNs) required three processes that are
training, validation and testing. Collected data were divided into three datasets.
These datasets are defined as training dataset, validation data set and test dataset.
The data from 1980 to 2003 was used as the training and validation datasets and
the data from 2003 to 2008 was used as testing dataset.

67.3 Design of the Artificial Neural Network Architecture

Two models of ANN were developed in this study. Generally, the architecture of
the ANNs are designed based on the input, output, hidden layer and number of
node the respective layer. Both of the two models were constructed and designed
with the same architecture as to perform a better comparison of performances. A
tool for numerical computing known as MATLAB was used to implement the
networks. Matlab command line operations were applied to develop the model
instead of using Matlab graphic user interfaces (GUIs). Figure 67.1 illustrates the
architectures of the two designed models.

67.3.1 Input Layer

The influential variables of CO2 emissions that have been chosen in this study are
fed to the input layer of the ANNs. Hence, the present study was constructed with
five nodes in the input layer represented the five variables.

67.3.2 Hidden Layer

According to Ayat et al. [34] higher number of neurons in the hidden layer may
derive better learning capability by the network. However, in this research, the
authors decided to implement only one hidden layer for the constructed ANNs. The
decision was made in order to prevent high computational cost of the networks.
67 Neural Network Training Algorithm for Carbon Dioxide Emissions … 721

Fig. 67.1 Architecture of neural network models

The authors decided to choose twelve nodes in the hidden layer. The decision is
made after considering three different numbers of nodes, 8, 12 and 15. The three
different numbers of nodes were randomly picked by the authors and eight num-
bers of nodes shown the best result among them.
Another important element in this layer is activation function. The function is
well known as a transfer function as it transmitted information from one layer to
another layer. Tan-sigmoid function was chosen in this study, in which can be
expressed as below:

2
a ¼ Tan  sigðnÞ ¼ 1 ð67:1Þ
ð1 þ expð2  nÞÞ

67.3.3 Output Forms

Output layer consists of one neuron represents our main goal. This study aims to
search the best learning algorithm for neural network in CO2 emissions forecast
Therefore, the CO2 emissions were considered as the single variable in this layer.

67.4 Results

Results of the performance comparison of testing dataset are presented into two
subsections.

67.4.1 Error Analyses

Precision of the forecasting model is examined by calculating three different error

measures. The measures are root mean square error (RMSE), mean absolute error
(MAE), and mean absolute percentage error (MAPE). All these measures are
expressed as follows:
722 H.M. Pauzi and L. Abdullah

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X n
ðx  ^xÞ2
RMSE ¼
i¼1
n
X
n
jx  ^xj
MAE ¼ ð67:2Þ
n
l¼1
 
100 x  ^x
MAPE ¼  :
n ^x 

where x = actual value, ^x = predicted value of x, n = total number of points.

By using the three error measures and two efficiency indices for each training
algorithm, the forecasting performance of LM algorithm against the gradient
descent algorithm of NN can be measured. Table 67.1 summarises the perfor-
mances based on the five measures.
As can be seen in Table 67.1, better precision of prediction were seen with LM
as compared with gradient descent. As mentioned previously, the training was
terminated when the minimum error of the data was attained. Small values of
RMSE, MAE, and MAPE indicate the accuracy of the ANN.
It was found that these values from LM algorithm were about two to three-fold
smaller than the gradient descent algorithm (training set: 0.3901 vs. 10.9289;
testing set 1.7296 vs. 4.0176). Therefore, the LM algorithm outperformed the
gradient descent algorithm.
The number of training epochs and time elapsed for total epochs will reflect the
terminated point of training dataset. Number of epochs and CPU time elapsed at
the end of the training for gradient descent was obtained as 9 epochs within 30 s.
Meanwhile, for LM algorithm, it took 10 epochs within a very short time, 0 s. As
can be seen in Figs. 67.2 and 67.3, the gradient descent algorithm needed 6 val-
idation checks and the LM needed only 3 validation checks to stop the training.
Based on these values, it is worth mentioning that LM algorithm showed better
performance than the gradient descent algorithm.

67.4.2 Regression Analyses

Regression analysis of outputs and targets for LM and gradient descent algorithms
during training and testing phase are shown in Fig. 67.4.
The best fit lines in Fig. 67.4 illustrate the relationship between the predicted
value (calculated) and observed value (CO2 emissions). Generally, for evaluation
of robustness of a modelling technique, R2 of a test data set should be computed
and R2 greater than 0.9 can be regarded as a good overall fit [35]. It can be
interpreted that the value of R2 that approaches to 1 shows that the actual values
and the output value from ANN model have very high correlation between each
other while 0 shows random relationship.
67 Neural Network Training Algorithm for Carbon Dioxide Emissions … 723

Table 67.1 Error measures and comparison efficiency index of the training algorithm
Performance index Training algorithms
Gradient descent Levenberg-Marquadt
Training dataset
RMSE 0.4063 0.0185
MAE 0.3070 0.0102
MAPE (%) 10.9289 0.3901
Testing dataset
RMSE 0.1182 0.0532
MAE 0.1139 0.0492
MAPE (%) 4.0176 1.7296
Number of epoch at the end of training 9 9
CPU time elapsed at the end of training (s) 30 1

Fig. 67.2 Results of NN model with gradient descent algorithm

From the Fig. 67.4 it can be seen that the R2 value of NN model with LM
algorithm are fitter compared to gradient descent algorithm (0.8921 vs. 0.9783).
The gradient descent algorithm achieved lower value of R2 which is means that
this algorithm is lack of ability to train data for the model compared to LM
algorithm. Thus higher value of R2 by LM algorithm indicated that it was suc-
cessfully trained the data until it can fit the network that has been built. Overall,
this suggested that forecasting ability of the NN using LM algorithm for training
the data was much better than NN with gradient descent training algorithm in the
case of CO2 emissions in Malaysia.
724 H.M. Pauzi and L. Abdullah

Fig. 67.3 Results of NN model with LM algorithm

Fig. 67.4 Regression analyses of outputs and targets during training and testing for a gradient
descent algorithm, b LM algorithm

67.5 Conclusions

This paper presented the development of two NN models in application of fore-

casting annual CO2 emissions using the socio-economic and demographic param-
eters. MATLAB tools were implemented to construct the models. The performance
of the model using the Levenberg-Marquardt algorithm shows that it can provide
better agreement with the collected data of CO2 emissions than those trained by
gradient descent algorithm. This can be concluded based on the observation of error
analysis. The RMSE, MAE, MAPE from Levenberg-Marquardt algorithm are
smaller than the values from gradient descent algorithm.
67 Neural Network Training Algorithm for Carbon Dioxide Emissions … 725

The presented ANN models not only can be used for CO2 emissions fore-
casting, but also for simulating various scenarios of CO2 emissions by changing
the parameters for the input variables. The results obtained from this research
could be useful for the governments and authorities to take any initiatives or
preventions in order to keep the sustainability of air quality. Deciding on the
appropriate algorithm with higher efficiency is a worth effort to better dealt with
environmental assessment. Another featured environmental quality indicators
could be considered in further researches. Improvement of the forecasting models
using other computational intelligence tools such as adaptive neuro-fuzzy infer-
ence system could be explored in future research. Malaysia.

Acknowledgments The authors are grateful to the Malaysian Ministry of Higher Education and
University Malaysia Terengganu for financial support under the FRGS grant number 59243.

References

1. Hui, T.S., Rahman, S.A., Labadin, J.: An empirical study on CO2 emissions in ASEAN
countries. In: International Conference on Statistics in Science, Business, and Engineering
(ICSSBE), pp. 1–6 (2012)
2. Jana, B.K., Biswas, S., Majumder, M., Roy, P., Mazumdar, M.: Estimation of Carbon
Dioxide Emission Contributing GHG Level in Ambient Air of a Metro City: A Case Study
for Kolkata, Impact of Climate Change on Natural Resource Management, (pp. 3–18).
Springer, Netherlands (2010)
3. Brondfield, M.N., Hutyra, L.R., Gately, C.K., Raciti, S.M., Peterson, A.: Modeling and
validation of on-road CO2 emissions inventories at the urban regional scale. Environ. Pollut.
70, 123–133 (2012)
4. Nwachukwu, A.N., Anonye, D.: The effect of atmospheric pressure on CH4 and CO2
emission from a closed landfill site in Manchester, UK. Environ. Monit. Assess. 185,
5729–5735 (2012)
5. Shih, S.H., Tsokos, C.P.: Prediction models for carbon dioxide emissions and atmosphere.
J. Neural Parallel Sci. Comput. 16, 165–178 (2008)
6. Hui, T.S., Rahman, S.A., Labadin, J.: Statistical modeling of CO2 emissions in Malaysia and
Thailand. Int. J. Adv. Sci. Eng. Inf. Technol. 2, 10–15 (2012)
7. Antanasijević, D.Z., Pocajt, V.V., Povrenović, D.S., Ristić, M.Ð., Perić-Grujić, A.A.: PM10
emission forecasting using artificial neural networks and genetic algorithm input variable
optimization. Sci. Total Environ. 443, 511–519 (2012)
8. Manjunatha, R., Narayana, P.B., Redy, K.H.C.: Application of artificial neural networks for
emission modeling of biodiesels for a C.I engine under varying operating conditions. Mod.
Appl. Sci. 4, 77–89 (2010)
9. Dede, G., Sazlı, M.H.: Speech recognition with artificial neural networks. Digit. Signal Proc.
20, 763–768 (2010)
10. Sabato, M.S., Yu, D., Deng, L., Lee, C.H.: Exploiting deep neural networks for detection-
based speech recognition. Neurocomputing 106, 148–157 (2013)
11. Ting, H.N., Yong, B.F., Mirhassani, S.M.: Self-adjustable neural network for speech
recognition. Eng. Appl. Artif. Intell. 26(9), 2022–2027 (2013)
12. Lamrini, B., Valle, G.D., Trelea, I.C., Perrot, N., Trystram, G.: A new method for dynamic
modelling of bread dough kneading based on artificial neural network. Food Control 26(2),
512–524 (2012)
726 H.M. Pauzi and L. Abdullah

13. Li, N., Xia, L., Shiming, D., Xu, X., Chan, M.Y.: Dynamic modeling and control of a direct
expansion air conditioning system using artificial neural network. Appl. Energy 91, 290–300 (2012)
14. Timothy Hong, Y.S.: Dynamic nonlinear state-space model with a neural network via
improved sequential learning algorithm for an online real-time hydrological modeling.
J. Hydrol. 468(469), 11–21 (2012)
15. Wang, J.J., Wang, J.Z., Zhang, Z.G., Guo, S.P.: Stock index forecasting based on a hybrid
model. Omega 40, 758–766 (2012)
16. Venkatesh, K., Ravi, V., Prinzie, A., Van den Poel, D.: Cash demand forecasting in ATMs by
clustering and neural networks. Eur. J. Oper. Res. 232(2), 383–392 (2014)
17. Yu, L., Wang, S., Lai, K.K.: A multiscale neural network learning paradigm for financial
crisis forecasting. Neurocomputing 73, 716–725 (2010)
18. Xu, L., Liu, S.: Study of short-term water quality prediction model based on wavelet neural
network. Math. Comput. Model. 58(3–4), 807–813 (2013)
19. Kurt, A., Oktay, A.B.: Forecasting air pollutant indicator levels with geographic models
3 days in advance using neural networks. Expert Syst. Appl. 37, 7986–7992 (2010)
20. S
ß ahin, Ü.A., Bayat, C., Uçan, O.N.: Application of cellular neural network (CNN) to the
prediction of missing air pollutant data. Atmos. Res. 101(1–2), 314–326 (2011)
21. Fernando, H.J.S., Mammarella, M.C., Grandoni, G., Fedele, P., Di Marco, R., Dimitrova, R.,
Hyde, P.: Forecasting PM10 in metropolitan areas: efficacy of neural networks. Environ.
Pollut. 163, 62–67 (2012)
22. Yaghini, M., Khoshraftar, M.M., Fallahi, M.: A hybrid algorithm for artificial neural network
training. Eng. Appl. Artif. Intell. 26, 293–301 (2013)
23. Zazoun, R.S.: Fracture density estimation from core and conventional well logs data using
artificial neural networks: the Cambro-Ordovician reservoir of Mesdar oil field, Algeria.
J. Afr. Earth Sci. 83, 55–73 (2013)
24. Nakama, T.: Theoretical analysis of batch and on-line training for gradient descent learning
in neural networks. Neurocomputing 73(1–3), 151–159 (2009)
25. Tongyu, X., Wei, Z., Peng, S., Qin, Z.: Transient power quality recognition based on BP
neural network theory. Energy Procedia 16, Part B, 1386–1392 (2012)
26. Coskun, N., Yildrim, T.: The effects of training algorithm in MLP network on image
classification. In: Proceedings of International Joint Conference on Neural Network
27. Wang, D., Lu, W.Z.: Forecasting of ozone level in time series using MLP model with a novel
hybrid training algorithm. Atmos. Environ. 40, 913–924 (2006)
28. Ghaffari, A., Abdollahi, H., Khoshayand, M.R., Bozchalooi, I.S., Dadgar, A., Tehrani, M.R.:
Performance comparison of neural network training algorithms in modeling of bimodal drug
delivery. Int. J. Pharm. 327(1–2), 126–138 (2006)
29. Piotrowski, A.P., Napiorkowski, J.J.: Optimizing neural networks for river flow forecasting—
Evolutionary Computation methods versus the Levenberg–Marquardt approach. J. Hydrol.
407, 12–2 (2011)
30. Mudgal, A., Gopalakrishnan, K., Hallmark, S.: Prediction of emissions from biodiesel fueled
transit buses using artificial neural networks. Int. J. Traffic Transp. Eng. 1, 115–131 (2011)
31. Palani, S., Liong, S.Y., Tkalich, P.: An ANN application for water quality forecasting. Mar.
Pollut. Bull. 56, 1586–1597 (2008)
32. Voukantsis, D., Karatzas, K., Kukkonen, J., Rasinen, T., Karppinen, A., Kolehmainen, M.:
Intercomparison of air quality data using principal component analysis and forecasting of
PM10 and PM2.5 concentrations using artificial neural networks, in Thessaloniki and
Helsinki. Sci. Total Environ. 409, 1266–1276 (2011)
33. World Bank: World Development Indicators. http://data.worldbank.org/country/malaysia
34. Ayat, S., Farahani, H.A., Aghamohamadi, M., Alian, M., Aghamohamadi, S., Kazemi, Z.: A
comparison of artificial neural networks learning algorithms in predicting tendency for
suicide. Neural Comput. Appl.23, 1381-1386 (2013)
35. Bourquin, J., Schmidli, H., Hoogevest, P., Leuenberger, H.: Application of artificial neural
networks (ANNs) in the development of solid dosage forms. Pharm. Dev. Technol. 2,
111–121 (1997)
Chapter 68
Theorem Prover Based Static Analyzer:
Comparison Analysis Between ESC/Java2
and KeY

Aneesa Saeed and S.H.A. Hamid

Abstract Software developers utilize static analyzers to discover the defects in

the software source code. One of the static analyzer categories is based on theorem
prover. Due to the strength of the theorem prover in proving the programs cor-
rectness and soundness without producing false warnings, analysis on the verifi-
cation tools is important to assess the performance. The objective of this paper is to
analyze the performance of the open source theorem prover based static analyzers
for Java. The analysis is done by comparing two static analyzers namely KeY and
ESC/Java2 using four evaluation metrics for 20 test cases developed based on
Common Weakness Enumeration (CWE). The result shows the performance of
KeY is better than ESC/Java2 especially on Detection Coverage. The analysis also
presents that the performances of theorem prover based static analyzers are
effective on small developed benchmark only.

Keywords Static analysis  Theorem prover  Verification  KeY  ESC/java2 

Software testing

68.1 Introduction

Testing becomes an important step in software engineering lifecycle because it

typically consumes 40–50 % from development efforts [1]. Static analysis is one
of testing strategies that detects defects automatically for assessing code quality
without running the code. Static analysis tools inspect automatically software

A. Saeed (&)  S.H.A. Hamid

Faculty of Computer Science and Information Technology, University of Malaya,
Kuala Lumpur, Malaysia
e-mail: [email protected]
S.H.A. Hamid
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 727

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_68
728 A. Saeed and S.H.A. Hamid

vulnerabilities such as redundant code, unsuitable use of variables, uncaught

runtime exceptions and division by zero. Static analyzers automatically identify
the software defects by parsing and analyzing the source code. The purpose is to
search for a fixed number of patterns in the code. Static analyzers analyze the
source code using control flow analysis, interface analysis, information flow
analysis, data flow analysis and path analysis. The use of the static analyzers are
increased currently in the software development process as a fault detector [2]. The
powerful of the static analyzers come from their ability to detect the defects in the
code which they are hardly discovered during execution [3]. There are some
programming errors that hardly caught by static analyzers [4], for example, Basic
cross-site scripting (XSS) vulnerability [2]. Static analyzers have six categories
based on internal design schemes as discussed more in Sect. 68.3. The categories
are type checking, style checking, program understanding, program verification
(theorem prover) and property checking, bug finder and security review. Verifi-
cation tools have progressed noticeably during the last years due to the advances in
verification aspects which are: the specification, program verification and theorem
prover methods.
Theorem prover based verification tools take the programs with their specifi-
cations as an input and prove the codes are matched with the specification. The
general framework is described in Fig. 68.1. Several verifiers have been developed
rely on theorem prover for Java, C, and C#. Examples for these static analyzers
are: for Java; ESC/Java2 [5], KeY [6], Jive [7], LOOP [8] and Jstar [9], for C;
veriFast [10] and Frama-C [11], and for C#.Net; Boogie [12]. Due to the strength
of the theorem prover, analysis on the verification tools is important to assess the
performance. The ESC/Java2 and KeY are common open source static analyzers
based on theorem prover that have reported successful results in [13, 14]. KeY was
used to verify the correctness of such resource guarantees. The results show that
the KeY caught all the bugs except one bug [13]. Results of using ESC/Java2 to
verify tally subsystem of the Dutch Internet voting system were presented well in
[12]. In few comparison studies [15–17], they focused on the KeY and ESC/Java2
separately with insufficient results, only searching on proofing time with KeY and
on reduced small set of vulnerabilities with ESC/Java2.
The objective of this study is to analyze the performance of the open source
theorem prover based static analyzers for Java. The analysis is done by comparing
two static analyzers namely KeY and ESC/Java2. We developed a small bench-
mark based on the CWE. The objective evaluation is used to assess the perfor-
mance of the selected static analyzers. The objective evaluation is measures the
Detection Coverage, the number of defects that are correctly and incorrectly
detected (True and False positives respectively) and the F-measure.
The rest of this paper is organized as follows: Sect. 68.2 presents the previous
related work. Section 68.3 provides a background on static analyzers with their
classification. The methodology that we follow in this research is discussed in
Sect. 68.4. Section 68.5 discuses our experiments and finding. Finally, conclusion
and future work are presented in Sect. 68.6.
68 Theorem Prover Based Static Analyzer: Comparison … 729

Fig. 68.1 Framework of

verification tools based on Program+ Specification
theorem prover

Translation

Extract proof obligation

Theorem prover

Output (proof/ Warrnings)

68.2 Related Work

This section presents the previous works related to the comparison studies on the
performance of the static analyzers. Beckert et al. [18] studied two verification
tools which are VCC and KeY. The evaluation on the two static analyzers is
conducted by using test suites provided by each tool developers. A new coverage
criterion has been investigated to test the tool correctness depends on the axi-
omatization. Using the test suites that are written by the tool developers them-
selves is not enough to the evaluation.
Study reported by Collavizza et al. [15] presented a comparison between CPBVP
with the ESC/Java, CBMC, Blast, EUREKA and Why verification tools. The
analysis was done against six small programs. However, the small program is not
efficient to evaluate the tools performance especially current systems are scalable.
Rutar et al. [17] analyzed five open source tools (Bandera, ESC/Java2, Find-
Bugs, JLint, and PMD) against small test suite that involved common for Java. The
authors investigated a meta-tool to allow developers to recognize the different
classes of vulnerabilities. The experimental results showed that the tools find non-
overlapping vulnerabilities.
In educational view, Feinerer and Salzer [16] compared four tools (namely the
Frege Program Prover, the KeY system, Perfect Developer, and the Prototype
Verification System (Pvs)) to check their suitability for teaching formal software
verification. They evaluated the tools against a suite of small programs, which are
typical of courses dealing with Hoare-style verification, weakest preconditions, or
dynamic logic. Finally they report their experiences by Perfect Developer in class.
In other work in [2], they have compared the performance of nine static ana-
lyzers in term of detecting security vulnerabilities. The tools are CBMC, K8-
Insight, PC-lint, Prevent, Satabs, SCA, Goanna, Cx-enterprise and Codesonar.
Seven of them are commercial tools but having different designs. The tools are
based on model checking and syntactic analysis. They conducted the experiments
730 A. Saeed and S.H.A. Hamid

against two dataset test suites (45, and 46) from SAMATE Reference Dataset for C
language. One includes test cases (TC) with known vulnerabilities and the other
one is designed with specific vulnerabilities fixed. The results are objectively
assessed by using a set of well known evaluation metrics (Recall, Precision and F-
measure). Some recommendations for improving the reliability and usefulness of
static analyzers and the process of benchmarking are recommended.
In [19] 11 software contract technologies were explored namely JML, ICon-
tract, Contract Java, Handshake, Jass, JContractor, JMsassert, Spec#, Code Con-
tracts, Eiffel and OCL. They presented the similarities with the areas of significant
disagreement and highlight the shortcomings of existing technologies. They briefly
introduced PACT, a software contract tool under development, explaining its
approach to various aspects of software contracts.
The previous related works concerned on the tools based on verification, lexical
and syntactic analysis against small programs except the study by Díaz and
Bermejo [2]. They analyzed static analyzers for C against standard benchmark. To
our best knowledge, this research is the first comparison studies that focusing on
KeY theorem prover static analyzer as fault detector. Moreover, well-known
evaluation metrics in terms of Recall, Precision and F-measure are used for the
true performance assessment.

68.3 Static Analyzers

68.3.1 Background

All static analyzers follow the same steps for analyzing source code as the following:
(1) transforming the code into set of data structures called program model, (2)
analyzing the model using different rules and/or properties, (3) showing the results to
the static analyzer. Code transforming is done via a mixture of various techniques as
lexical and semantic analysis, abstract syntax and parsing. Intraprocedural (local)
and interprocedural (global) analysis are used for analyzing the program model to
assess the individual functions and their interactions by tracing control flow and data
flow, pointer aliasing, etc... The rules can be fixed or extended based on the selected
tool. Finally the results of the selected tools are concise, without more information
about the possible existed defects, or more detailed. The differences in the infor-
mation of the output lead to hard comparison among the different tools.

68.3.2 Static Analyzers Categories

The static analyzers are classified based on different schemes. According to [20], the
categorization can be based on the language or the defects types. The schema presented
by Chess and West [3] utilized general purpose of the tools. This schema is the most
relevant one to our work. In next paragraphs, we discuss each category in the schema.
68 Theorem Prover Based Static Analyzer: Comparison … 731

Type checking tools Type checking is used by programmers to remove whole

categories of programming mistakes. For example, it prevents assigning integral
values into other variables types. Because of detecting errors at compile stage, the
type checking based tools avoid runtime errors. However, these tools suffer from
the false positive (FP) and false negative (FN).
Style checking tools This kind of tools achieves checking based on lexical and
syntactic analysis to detect the code problems as inconsistencies in function calls
such as calling the function with changeable number of attributes. Although, the
produced errors often affect the readability and the maintainability of the code,
these tools do not refer that the specific defect will happen in the execution time.
Therefore, these tools have limitations compared with others due to they perform
the analysis without simulating what happens in runtime. One Example of style
checking based static analyzers is the PMD tool for Java [21].
Program Understanding Tools These tools are involved in several Integrated
Development Environments and are designed to support programmers to gain
vision about how the program works. The reviewers use these tools to understand
the large code and detect the security vulnerabilities. Nevertheless, using these
methods is time consuming. One example of such tools is Fujaba tool [22].
Bug Finding Tools Bug Finding tools are not concern on formatting issues as
style checker nor are doing matching between the code and its specification. These
tools basically notify about places in the code where the program will act in a
different way that it is not desired by the programmer. Bug Finding tools have two
major features. The first feature is easy to use due to these tools have a set of rules
describing the patterns in the code that usually specify security vulnerabilities. The
second feature is scability where they are usually used for analyzing large appli-
cations code. Less number of is produced by using bug finders. FindBugs [23] and
Coverity [24] are examples of Find Bug tools.
Security Review Tools These tools designed as a hybrid of property checkers
and bug finding technique. However, they concern on identifying specific security
vulnerabilities. Due to the many security properties can be expressed as program
properties. The thoughtful side of the balance between the number of FP and
FN motivates the researchers to use these tools. Furthermore, many points in the
code are shown that will be manually reviewed after applying the tools. These
tools producing more FP compared with bug finder. Ounce 6 and SCA are the two
most relevant examples of this category.
Theorem Proven Verification and Property Checking Tools Using these kinds
of tools requires the code and the specification. The verification tools (based on
theorem prover) attempt to prove that the code meets the specification. Verification
tools sometimes go by the name property checking when they check the code
against a fractional specification that details only part of the program behavior.
The majority of property checking tools work either by applying logical inference
or by performing model checking. The programmers rarely insert the specification
that is detailed enough to be used for equivalence checking. Writing specification
needs more time and effort than writing the code, this is the main limitation. These
732 A. Saeed and S.H.A. Hamid

tools never suffer from FP. Numbers of verifiers are available, for examples
CBMC, KeY, ESC/Java2, or Satabs tools.
Unfortunately, there are some FP or FN produced by all static analyzers.
Fortunately, some of theorem prover verification tools do not suffer from false
positives.

68.4 Methodology

In this section, we present the methodology that used to conduct the analysis study.

68.4.1 Select Tools

In this study, we selected KeY and ESC/Java2 which are well-known verification
tools for program static analysis. The performance evaluation is important to give
overall insights to programmers to choose suitable verification tools. Both tools are
open source that their complete code and documentation are available. This can
make us understand better their limitations. They need specifications written in
Java Modeling language (JML).
KeY ([25]): KeY is a formal software development tool that aims to integrate
design, implementation, formal specification, and formal verification of object-
oriented software as seamlessly as possible. At the core of the system is a novel
theorem prover for the first-order dynamic logic for Java with a user-friendly
graphical interface. KeY is sound and complete. The framework of KeY is shown
in Fig. 68.2.
ESC/Java2 ([26]): Extended Static Checker for Java (ESC/Java2) is a pro-
gramming tool that aims to detect common run-time errors in JML-annotated Java
programs by static analysis of the program code and its formal specifications. The
main limitation of this tool is it is neither sound nor complete. Therefore, false
positive and negatives may occur. Figure 68.3 shows the frame work of the ESC/
Java2.

68.4.2 Select Test Suites

The best approach to compare and evaluate the static analyzers is by analyzing the
same standard benchmark. The input of the theorem based tools must be the source
code and the specification. Finding suitable benchmark for conducting experiments
was complicated. This is because there are no standard benchmark test suites
containing the code in Java with the JML specification. In this study, we developed
our benchmark depending on the common code defects that described in
68 Theorem Prover Based Static Analyzer: Comparison … 733

Fig. 68.2 KeY framework

[6] Translation to DL

Specification
Java Proof obligation JML, OCL
Program

Interactive/Auto.prover

Auxiliary prover(simplify, SMT,..)

Show proof details

Fig. 68.3 ESC/Java2 Annotated Java

framework [27]

Front End

Abstract Syntax
Type-
specific Translator
Background AST
predicate Guarded Commands

VC Generator GCs
Verification Conditions

Theorem Prover
VCs
Prover Results
Postprocessor
Universal
Background
Predicate (UBP) Output to User

Table 68.1. We used and modified some TC (test cases) that developed by KeY
developers, Samate Reference Dataset and the TC used in [15] study. Table 68.1
illustrates the code defects based on CWE from MITRE corporation [20, 28]. The
defects names, number of the TC under each defect and small descriptions on the
defects are presented in Table 68.1.
734 A. Saeed and S.H.A. Hamid

Table 68.1 Common defects in the source code [20, 28]

Common defects TC Description
Divide by zero 2 An input divides by zero can lead undefined behavior
Heap overflow 2 Input is used in an argument to the creation or copying of blocks of
data beyond the fixed memory boundary of a buffer in the heap
portion of memory
Unchecked loop 2 No checking for loop conditions that potentially leading to a denial
condition of service because of excessive looping
Infinite loop 2 The number of loop repetition is infinite where the exit condition
are not be reached
Integer overflow 2 A numeric value that is too large to be represented within the
available storage space
Null dereferences 2 A pointer with a value of NULL is used as though it pointed to a
valid memory area
Array index out of 2 Try to read or write data into an index of the array that doesnot
bound exist
Unused variable 2 A variable is created without using it. It is subsequently referenced
in the program, causing potential undefined behavior or denial or
service
Uninitialized 2 A variable is created without assigning it a value. It is subsequently
variable referenced in the program, causing potential undefined behavior or
denial or service
Unchecked error 2 No action is taken after an error or exception occur
condition

68.4.3 Evaluation Metrics

Using only the defects Detection Coverage percentage is not enough for real
assessment for the tools performance. Due to the static analyzers suffer from FP
and FN. FN and FP are not covered in Detection Coverage percentage. Therefore,
our evaluation covers the F-measure [29]. The F-measure represents the harmonic
mean of Recall and Precision measurements which can be calculated as the
following:
Precision is the ratio of the number of correct discovered defects to the number
of all detected defects and can be calculated as:

TP
P¼ ð68:1Þ
TP þ FP

Recall is the ratio of the number of correct detected defects to the number of the
known defects and can be calculated as:

TP
R¼ ð68:2Þ
TP þ FN
68 Theorem Prover Based Static Analyzer: Comparison … 735

F-measure is the harmonic mean of Recall and Precision and can be calculated
as

2PR
F-measure ¼ ð68:3Þ
PþR

Defects Detection Coverage is the ratio of the number of detected defects to the
number of all defects in the benchmark and can be calculated as:

TP þ FP
Defects Detection Coverage ¼ ð68:4Þ
Number of all defects

where TP (true positive): is the number of true detected defects, FP is the number
of detected defects that in fact do not exist and FN is the total number of defects
not detected in the code.

68.5 Experiments

In this experiment, we have run the two static analyzers as Eclipse plug-in in Intel
Pentium 4 CPU 3.20 GHz 3.19 GHz, 4.0 GB ram and 32-bit operating system. We
used our benchmark with 20 TC written in Java language. The complexity of the
TC ranges from small Java programs to more complex programs (included many
complex properties like inheritance). Each kind of defects in the benchmark has
two TC. KeY 2.0.1 is the latest stable version released on June 20th, 2013 and
ESC/Java2 2.0.5 is the latest version released on 8 November 2008 were used to
conduct the experiments.
We calculated the detection percentage of each static analyzer. Table 68.2
illustrates the summary of the execution results of the two static analyzers against
20 TC. TP means the analyzer correctly detect this number of defects. Fails (FN)
means the analyser could not detect the existing defects in the benchmark.
FP refers to the number of detected defects but they are not real defects. The static
analyzers suffer from FP so depending on the Detection Coverage is not enough to
assess correctly the static analyzers performance. Furthermore, we calculate the
Recall, Precision and F-measure for each static analyzer. The recall rate refers to
the percentage of all the known defects in the benchmark are detected and the
percentage calculated by (1-recall)*100 from the defects are uncovered. Precision
expresses the chance of the defects to be detected correctly.
We conclude from the Table 68.2 that the performance of the KeY is more
effective than the ESC/Java2 in terms of all evaluation metrics. The effective per-
formance of KeY means it catches the known defects without producing any FP. All
the defects are detected by KeY except divide by zero and unused variable. The KeY
obtains value of 1 compare to ESC/Java2 which only receives 0.4. The difference of
0.6 result value shows that all defects are detected and correctly covered in KeY.
736 A. Saeed and S.H.A. Hamid

Table 68.2 Results of KeY and ESC/Java2 execution

Static Result summary
analyzers
Good FP Fails Recall Precision Detection F-
(TP) (FN) coverage measure
ESC/Java2 6 9 5 0.546 0.4 30 0.4617
KeY 15 0 5 0.75 1 75 0.8571

The variation between the obtained recall values is 0.204 that indicates less number
of defects is missed by KeY compared to ESC/Java2. The KeY does not produce FP
in all the TC. Comparing the number of true positives (6) is less than the number of
false warnings (9) by ESC/Java2.ESC/Java2 detects less number of defects with
producing many FP. The ESC/Java2 needs less time and easy to trace for the warning
signs compared to KeY which needs more time for tracing the errors and warnings.
Using the KeY for verification complex programs will prove correctly the programs
correctness and soundness but it takes long time that consumes the developing cost
(in term of time). In our opinion, using one static analyzer is not enough to analyze
the code and discover all the defects. Analyzing the code with many static analyzers
guarantees the most defects in the code will be discovered.

68.6 Conclusions and Future Work

In this study, we presented a comparison study between KeY and ESC/Java2 as fault
detectors on our benchmark. An objective evaluation has been conducted in terms of
Recall, Precision and F-measure to assess correctly the static analyzers performance.
Using KeY verification tool as fault detector obtained better Detection Coverage
without producing FP rather than ESC/Java2. We believe that using theorem prover
based static analyzers are a promising way in checking statically the code vulner-
abilities. For future, we plan to study the performance of all the theorem prover
based static analyzers against big benchmarks that covered more common defects.

Acknowledgments The authors acknowledge the support provided by University of Malaya

Research Grant, RG10612ICT.

References

1. Luo, L.: Software Testing Techniques. Institute for software research international Carnegie
Mellon University, Pittsburgh, 19 p (2001)
2. Díaz, G., Bermejo, J.R.: Static analysis of source code security: assessment of tools against
SAMATE tests. Inf.Softw. Technol. 55(8), 1462–1476 (2013)
3. Chess, B., West, J.: Secure Programming with Static Analysis. Pearson Education, Boston
(2007)
68 Theorem Prover Based Static Analyzer: Comparison … 737

4. Young, M., Taylor, R.N.: Rethinking the taxonomy of fault detection techniques. In:
Proceedings of 11th international conference on Software engineering, pp. 53–62. ACM
(1989)
5. Cok, D.R., Kiniry, J.R.: Esc/java2: Uniting esc/java and jml. Construction and Analysis of
Safe, Secure, and Interoperable Smart Devices, 3362,pp. 108–128. Springer (2005)
6. Beckert, B., Hähnle, R., Schmitt, P.H.: Verification of Object-Oriented Software: The KeY
Approach. Springer, Berlin, Heidelberg (2007)
7. Meyer, J., Poetzsch-Heffter, A.: An architecture for interactive program provers. Tools and
Algorithms for the Construction and Analysis of Systems,1785, pp. 63–77. Springer (2000)
8. Jacobs, B., Poll, E.: Java program verification at Nijmegen: Developments and perspective.
Software Security-Theories and Systems, 3233, pp. 134–153. Springer (2004)
9. Distefano, D., Parkinson, J., jStar, M.J.: Towards practical verification for Java. ACM
SIGPLAN Not. 43, 213–226 (2008)
10. Jacobs, B., Smans, J., Piessens, F.: VeriFast: Imperative programs as proofs. In: VSTTE
Workshop on Tools and Experiments. (2010)
11. Cuoq, P., Kirchner, F., Kosmatov, N., Prevosto, V., Signoles, J., Yakobowski, B.: Frama-C.
Software Engineering and Formal Methods, pp. 233–247. Springer (2012)
12. Barnett, M., Chang, B.-Y.E., DeLine, R., Jacobs, B., Leino, K.R.M.: Boogie: a modular
reusable verifier for object-oriented programs. In: Formal Methods for Components and
Objects, 7504, pp. 364–387. Springer (2006)
13. Cok, D., Kiniry, J.: ESC/Java2: Uniting ESC/Java and JML. In: Barthe, G., Burdy, L.,
Huisman, M., Lanet, J.-L., Muntean, T. (eds.) Construction and Analysis of Safe, Secure, and
Interoperable Smart Devices, vol. 3362, pp. 108–128. Springer, Berlin Heidelberg (2005)
14. Albert, E., Bubel, R., Genaim, S., Hähnle, R., Puebla, G., Román-Díez, G.: Verified resource
guarantees using COSTA and KeY. In: Proceedings of the 20th ACM SIGPLAN Workshop
on Partial Evaluation and Program Manipulation. ACM, pp. 73–76 (2011)
15. Collavizza, H., Rueher, M., Van Hentenryck, P.: Comparison between CPBPV, ESC/JAVA,
CBMC, BLAST, EUREKA and WHY for bounded program verification. arXiv preprint
arXiv:0808. 1508 (2008)
16. Feinerer, I., Salzer, G.: A comparison of tools for teaching formal software verification.
Formal Aspects Comput. 21, 293–301 (2009)
17. Rutar, N., Almazan, C.B., Foster, J.S.: A comparison of bug finding tools for Java. In:
Software Reliability Engineering. 5th IEEE International Symposium on ISSRE, pp. 245–256
(2004)
18. Beckert, B., Bormer, T., Wagner, M.: A Metric for Testing Program Verification
Systems, 7942, pp. 56–75. Tests and Proofs. Springer (2013)
19. Voigt, J., Irwin, W., Churcher, N.: Comparing and Evaluating Existing Software Contract Tools.
Evaluation of Novel Approaches to Software Engineering, 275, pp. 49–63. Springer (2013)
20. Zheng, J., Williams, L., Nagappan, N., Snipes, W., Hudepohl, J.P., Vouk, M.A.: On the value
of static analysis for fault detection in software. IEEE Trans. Softw. Eng. 32, 240–253 (2006)
21. Copeland, T.: PMD Applied. Centennial Books, San Francisco (2005)
22. Burmester, S., Giese, H., Niere, J., Tichy, M., Wadsack, J.P., Wagner, R., Wendehals, L.,
Zündorf, A.: Tool integration at the meta-model level: the Fujaba approach. Int. J. Softw.
Tools Technol. Transfer 6, 203–218 (2004)
23. Hovemeyer, D., Pugh, W.: Finding bugs is easy. ACM SIGPLAN Not. 39, 92–106 (2004)
24. Almossawi, A., Lim, K., Sinha, T.: Analysis Tool Evaluation: Coverity Prevent. Carnegie
Mellon University, Pittsburgh (2006)
25. http://www.key-project.org/
26. http://www.kindsoftware.com/products/opensource/ESCJava2/download.html
27. Flanagan, C., Leino, K.R.M., Lillibridge, M., Nelson, G., Saxe, J.B., Stata, R.: Extended
static checking for Java, 37(5), pp. 234–245. ACM SIGPLAN Notices. ACM (2002)
28. http://cwe.mitre.org
29. Manning, C.D., Raghavan, P., Schütze, H.: Introduction to Information Retrieval. Cambridge
University Press, Cambridge (2008)
Chapter 69
Designing a New Model for Trojan Horse
Detection Using Sequential Minimal
Optimization

Madihah Mohd Saudi, Areej Mustafa Abuzaid, Bachok M. Taib

and Zul Hilmi Abdullah

Abstract Malwares attack such as by the worm, virus, trojan horse and botnet
have caused lots of troublesome for many organisations and users which lead to
the cybercrime. Living in a cyber world, being infected by these malwares
becoming more common. Nowadays the malwares attack especially by the trojan
horse is becoming more sophisticated and intelligent, makes it is harder to be
detected than before. Therefore, in this research paper, a new model called Effi-
cient Trojan Detection Model (ETDMo) is built to detect trojan horse attacks more
efficiently. In this model, the static, dynamic and automated analyses were con-
ducted and the machine learning algorithms were applied to optimize the perfor-
mance. Based on the experiment conducted, the Sequential Minimal Optimization
(SMO) algorithm has outperformed other machine learning algorithms with
98.2 % of true positive rate and with 1.7 % of false positive rate.

Keywords Malwares 
Trojan horse 
Detection 
Automated analysis 
 
Sequential minimal optimization (SMO) True positive rate False positive rate 
Machine learning

M. Mohd Saudi (&)  A.M. Abuzaid  B.M. Taib  Z.H. Abdullah

Faculty of Science and Technology, Universiti Sains Islam Malaysia (USIM),
71800 Bandar Baru Nilai, Negeri Sembilan, Malaysia
e-mail: [email protected]
A.M. Abuzaid
e-mail: [email protected]
B.M. Taib
e-mail: [email protected]
Z.H. Abdullah
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 739

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_69
740 M. Mohd Saudi et al.

69.1 Introduction

Trojans have been improved and ranges of sophisticated techniques have been
integrated, which make the detection processes much harder and longer than in the
past. Lack of understanding and knowledge and proper procedures of the trojan
analysis have led to money loss, reduced productivity and tarnishing of organi-
zation’s reputations [1]. Nowadays there are so much information about vulner-
abilities and exploitation in operating systems and applications, attacks and
defense technologies. Though information regarding these widely distributed and
different defense technologies has been implemented, many organizations con-
stantly being attacked and exploited by malwares such as worm and trojan horse.
Nowadays, the computer network systems are more threatened by trojan horse.
This is mostly true for Windows system because there are large numbers of trojan
horses designed to enable attacks upon running Windows platforms. Furthermore,
new trojan horses emerge almost every day. Trojan horse is defined as a malicious
program, that must be executed on a victim’s computer and once it is installed, it
can control the victim’s computer remotely and steal any confidential information
from it. It is different compared to worm and virus, since it has the capability to
control the victim’s computer remotely and it does not replicate itself [2].
On May 2012, the Flame trojan has infected thousands of computers all over the
world and it has been described as one of the most complex threats ever discov-
ered. It has the capabilities to take screenshots secretly, record audio and sends this
information to its creator via encrypted channel. It caused chaos and loss of money
and productivity [3]. In the year of 2013, the Australia Computer Emergency
Response Centre (AuSCERT) reported that 7,962 cases of the compromised
Australia web sites were serving malwares [4]. The implications of the trojan horse
show that a good detection technique is crucial in detecting the trojan horse.
The objectives of this research paper are to investigate and evaluate existing
trojan horse detection techniques, to develop a new model trojan horse detection
technique and to evaluate the proposed model. Knowledge Discovery and Data
Mining (KDD) has been used as part of the methods in this research. Based on the
gaps and challenges found in the existing works, the pre-processing in the KDD
processes has been improved. This includes the improvement of the feature
selection where the static, dynamic and automated analyses have been integrated.
As a result, prior forming the trojan horse detection technique, a new trojan horse
classification has been developed. This classification is used as the input for the
trojan horse detection technique. This trojan horse classification is called as an
Efficient Trojan Classification (ETC) and is not discussed in this research paper
and can be referred in paper by Saudi and her colleagues [5]. Furthermore, bigger
dataset were used in the experiment and different machine learning algorithms
have been integrated with the proposed model to optimise the trojan horse
detection technique. Moreover, from the experiment conducted, SMO algorithm
has been proven as the best machine learning algorithm to optimise the trojan
horse detection performance.
69 Designing a New Model for Trojan Horse Detection Using Sequential … 741

This paper is organised as follows. Section 69.2 presents the related works with
trojan horse detection techniques. Section 69.3 explains the methodology used in
this research paper which consists of static, dynamic and automated analyses and
the architecture of the controlled laboratory environment. Section 69.4 presents
the research findings which consists the results of the testing and evaluation of the
proposed trojan horse detection. Section 69.5 concludes and summarises the future
work of this research paper.

69.2 Related Works

Though the trojan horse study was started by Thimbleby et al. [6], only after
10 years later, more studies were carried out such as by [7–10]. However, this
works more focusing on trojan horse hardware taxonomy and hardware detection
techniques instead. Each of these works has it owns strengths and gaps that can be
further improved. Zhang and colleagues used timestamp-based data stream clus-
tering algorithm to detect trojan horse theft activity [11]. The researchers used
clusters to compress trojan horse communication data stream information and
extracted clusters characteristics for the detection processes. Based on the
experiment conducted, it produced 90 % an accuracy rate and lower false negative
rate. However this work is only focusing on trojan horse with theft capability.
Apart from that, Tang presented a new trojan horse detecting method, based on
Portable Executable (PE) file static attributes [12]. An intelligent information
processing technique is used to analyze those static attributes in the PE files. The
experiment result showed the test pass rate is 63.90 %. The result can be further
improved if the experiment involves bigger volume of dataset.
While Liu and colleagues used data mining to detect the trojan horse in Win-
dows environment [13]. This study shows that the accuracy of classification can be
increased when the more relevant features are used in the data mining processes
and reduces the consumption of time space. However, the more features are
selected, the more time building classification cost, it responds slower in real time
and it needs bigger dataset from real network environment. As for work by Dai and
colleagues, they presented a novel malicious code detection approach by mining
dynamic instruction sequences [14]. Their result showed that their approach is
accurate, reliable and efficient. But they used dynamic analyses only and when
conducting their experiments, the method was not able to detect any malicious
code hooked in the remaining part of the executable code. Improvement can be
done if their experiment combining both static and dynamic analysis.
Based on all the previous works discussed above, the main challenges which should
be considered thoroughly are the dataset types and volume, analysis and detection
techniques and feature selection to detect the trojan horse efficiently. Therefore, in this
research, a new trojan detection technique is developed by integrating static, dynamic
and automated analyses and by using bigger and standard dataset and an improved
feature selection, which is further explained in Sects. 69.3 and 69.4.
742 M. Mohd Saudi et al.

69.3 Methodology

In order to produce a new trojan horse detection technique, the researchers’ had
conducted few experiments and researches. A controlled laboratory environment is
created to conduct the experiment as illustrated in Fig. 69.1. It is a controlled
laboratory environment and almost 80 % of the software used in this testing is an
open source or available on a free basis. No outgoing network connection is
allowed for this architecture. The static, dynamic and automatic analyses were
conducted. The cuckoo sandbox (for automated analyses) has been integrated in
the laboratory architecture as well.
The dataset used for this research are from VX Heaven [15] and Offensive
Computing [16]. Many studies have used these dataset in their experiment [14, 17–
20]. In these datasets there are many variant of malwares and benign files. A total
amount of 1640 trojan horse datasets have been tested in this lab. These dataset
were categorized into different types which are the Clicker, DDOS, BAT, AOL,
BOOT, PHP and PWS. For this research, Knowledge Discovery and Data Mining
(KDD) is used as a technique to identify the trojan horse patterns in the datasets.
KDD is defined as several stages of process that are experienced by the data to
determine the patterns, which must be important, potentially useful and under-
standable. Enhancements have been made to the KDD data pre-processing and
pattern extraction process. Under the data pre-processing process, the static,
dynamic and automated analyses are implemented as illustrated in Fig. 69.2.
The whole processes involved in the experiment are summarized in the flow-
chart in Fig. 69.3. After the static, dynamic and automated analyses are conducted
successfully (refer Fig. 69.3), a pattern of useful Trojan horse characteristics is
identified and is transformed into nominal data to be mined in the WEKA soft-
ware. It is an open source JAVA software [21]. In the WEKA, five different
classification algorithms were executed in order to study the type of classifier
suitable to deal with huge amount of features extracted from the static, dynamic
and automated analyses processes. The different classifiers that were executed are
the Multilayer Perceptron (MLP), Sequential Minimal Optimization (SMO), IBk,
Naïve Bayes and J48 tree algorithm.
After the above data analyses are conducted successfully, a pattern or features
of useful trojan horse characteristics is identified, where the ETC Classification has
been formed where it consists of the trojan horse infection, activation, payload and
operating algorithm [5]. Later, these four trojan horse characteristics are used to
represent all the dataset used for the experiments. Then, these dataset were used in
data mining (using WEKA), the trojan horse characteristics are transformed into
nominal data with a certain number representation.
As for the evaluation, 10-fold cross validation is used to divide the dataset into
testing and training dataset. The reasons why the 10-fold cross validation is used
are firstly, it uses as much data as possible for training and testing and secondly,
the better accuracy of its findings. All analysis of the trojan horse, were docu-
mented and recorded properly. This record is useful in understanding on how the
69 Designing a New Model for Trojan Horse Detection Using Sequential … 743

Fig. 69.1 Controlled laboratory architecture

Fig. 69.2 ETDMo KDD processes

trojan horse works. There are several performance parameters that need to be
measured during the experiment which are as in the Eqs. 69.1 and 69.2. TPR
represents the true positive rate, FPR represents as false positive rate, TP represent
as true positive, FN represents as false negative, FP represents as false positive and
TN represents as true negative.

TPR ¼ TP = ðTP þ FNÞ ð69:1Þ

FPR ¼ FP = ðFP þ TNÞ ð69:2Þ

69.4 Findings

This section presents the finding results of the machine learning algorithms. The
dataset have been classified using the WEKA. The MLP, SMO, IBk, Naïve Bayes
and J48 algorithms were chosen to classify the dataset and the result as displays in
744 M. Mohd Saudi et al.

Fig. 69.3 Summarization of the ETDMO research processes

Table 69.1. Table 69.1 illustrates the number of true positive rates (TPR) for five
different algorithms that are used in the analysis were conducted. The SMO
algorithm got the highest true positive rate with 98.2 % accuracy rate, followed by
MLP algorithm with 98.0 % accuracy rate, J48 algorithm with 94 % accuracy rate
and IBK algorithm with 90.7 % accuracy rate. The lowest TPR was 86.2 %
accuracy rate for Naive Bayes algorithm. The False positive Rate are (FPR): the
Naïve Bayes with 13.8 % accuracy rate has the highest FP accuracy, followed by
IBk with 9.3 %, J48 with 6.0, and 2.0 % for Multilayer Perceptron and SMO with
1.7 % accuracy rate. Based on the experiment conducted, SMO algorithm has the
69 Designing a New Model for Trojan Horse Detection Using Sequential … 745

Table 69.1 Machine learning algorithm results

Classifier ETDMO results (%)
TPR FPR
SMO 98.2 1.7
MLP 98.0 2.0
J48 94.0 6.0
IBk 90.7 9.3
Naïve Bayes 86.2 13.8
TPR represents: True Positive Rate, FPR represents: False Positive Rate

Table 69.2 Comparison with the existing work

Classifier ETDMO results (%) Existing work
TPR FPR TPR FPR
SMO 98.2 1.7 93.2 9.6

highest of the True Positive Rate (TPR) with 98.2 % and the lowest of the False
Positive Rate (FPR) of 1.7 %.
A comparison with similar work by Dai and colleagues [14] is carried out (refer
Table 69.2). This work used the same source of the dataset. True positive rate
(TPR) and false positive rate (FPR) are used during the experiment. The experi-
ment was conducted using the WEKA software. The experiment result after the
compared with related existing work shows that, the ETDMo model has the higher
accuracy rate for trojan horse detection. For this research the TP rate was 98.2 %
with FP rate 1.7 %, compared with Dai and colleagues work [14], the TP rate was
91.9 and 9.6 % for FP rate. After the comparison with the existing work found
that, this research results got a better performance, which could be due to the
development made by the integration of static, dynamic and automated analysis
analyses which are part of whole ETDMo KDD process.

69.5 Conclusions and Future Works

As a conclusion, this research has managed to provide a better TPR, which is

98.2 % that outperformed the existing work. This result can be used as a reference
and comparison by other researchers with the same interests. For future work,
different machine learning algorithms will be tested to the dataset produced from
this research. This paper is part of a larger project to build up an automated
malware clean up model. Ongoing research will include other malware classifi-
cation and the development of software to automate the malware dataset cleanup.
746 M. Mohd Saudi et al.

Acknowledgments The authors would like to express their gratitude to Universiti Sains Islam
Malaysia (USIM) for the support and facilities provided. This research paper is supported by
Universiti Sains Islam Malaysia (USIM) grant [PPP/GP/FST/SKTS/30/11912], [PPP/FST/SKTS/
30/12812] and Research Management Centre, Universiti Sains Islam Malaysia (USIM).

References

1. Mitropoulos, S., Patsos, D., Douligeris, C.: On incident handling and response: a state-of-the-
art approach. Comput. Secur. 25(5), 351–370 (2006)
2. Saudi, M.M.: User awareness in handling computer viruses incident for windows platform.
Jurnal Teknologi Maklumat dan Multimedia 4, 53–72 (2007). ISSN: 1823-0113
3. Kaspesky Lab ZAO: What is flame? http://www.kaspersky.com/flame (2012)
4. Australia Computer Emergency Response Team (AusCERT): AusCERT incident metrics
June 2013. https://www.auscert.org.au/render.html?it=17856 (2013)
5. Saudi, M.M., Abuzaid, A.M., Taib, B.M., Abdullah, Z.H.: An efficient trojan horse
classification (ETC). IJCSI Int. J. Comput. Sci. Issues 10(2), 3. March 2013, ISSN (Print):
1694-0814 | ISSN (Online): 1694-0784
6. Thimbleby, H., Anderson, S., Cairns, P.: A framework for modelling trojan horse s and
computer virus infection. Comput. J. 41(7), 444–458 (1998)
7. Chakraborty, R.S., Narasimhan, S., Bhunia, S.: Hardware trojan horse : threats and emerging
solutions. In: Proceedings IEEE International High Level Design Validation and Test
Workshop, San Francisco, CA, pp. 166–171 (2009)
8. Karri, R., Rajendran, J., Rosenfeld, K., Tehranipoor, M.: Trustworhty hardware: identifying
and classifying hardware trojan horse s. Computer 43(10), 39–46 (2010)
9. Tehranipoor, M., Koushanfar, F.: A survey of hardware trojan horse taxonomy and detection.
Des. Test of Comput. IEEE 27(1), 10–25 (2010)
10. Karri, R., Rajendran, J., Rosenfeld, K.: Trojan horse taxonomy. In: Tehranipoor, M., Wang, C.
(eds.) Introduction to Hardware and Security Trust, pp. 325–338. Springer, New York (2011)
11. Zhang, X., Liu, S., Meng, L., Shi, Y.: Trojan horse detection based on network flow
clustering. In: Proceedings of the Multimedia Information Networking and Security
Conference. IEEE, pp. 947–950 (2012)
12. Tang, S.: The detection of Trojan horse based on the data mining. In: Fuzzy Systems and
Knowledge Discovery International Conference. IEEE, vol. 1, pp. 311–314 (2009)
13. Liu, Y., Zhang, L., Liang, J., Qu, S., Ni, Z.: Detecting Trojan horses based on system
behavior using machine learning method. In: 2010 Machine Learning and Cybernetics
Conference. IEEE, vol. 2, pp. 855–860 (2010)
14. Dai, J., Guha, R., Lee, J.: Efficient virus detection using dynamic instruction sequences.
J. Comput. 4(5), 405–414 (2009)
15. VX Heaven: Computer virus collection. http://vxheaven.org/vl.php (2014)
16. Offensive Computing: Malware search. http://www.offensivecomputing.net (2014)
17. Schultz, E.E., Shumway, Russell: Incident Response: a Strategic Guide to Handling System
and Network Security Breaches, 1st edn. New Riders Publishing, USA (2001)
18. Henchiri, O., Japkowicz, N.: A feature selection and evaluation scheme for computer virus
detection. In: Proceedings of the Sixth International Conference on Data Mining, 2006.
ICDM ‘06. IEEE Xplore, Hong Kong, p. 891 (2006)
19. Moskovitch, R., Stopel, D., Feher, C., Nissim, N., Japkowicz, N., Elovici, Y.: Unknown
malcode detection and the imbalance problem. J. Comput. Virol. 5(4), 295–308 (2008).
doi:10.1007/s11416-009-0122-8
20. Khan, H., Mirza, F., Khayam, S.A.: Determining malicious executable distinguishing
attributes and low-complexity detection. J. Comput. Virol. 7(2), 95–105 (2010)
21. Hall, M., Frank, E., Holmes, G., Pfahringer, B., Reutemann, P., Witten, I.H.: The WEKA data
mining software: an update. SIGKDD Explor. 11(1), 10–18 (2009)
Chapter 70
An Access Control Framework
in an Ad Hoc Network Infrastructure

Tanya Koohpayeh Araghi, Mazdak Zamani, A.A. Manaf

and Sagheb Kohpayeh Araghi

Abstract The union of an arbitrary topology constitutes of self-configuring mobile

routers and associated hosts connecting by wireless links is called a mobile ad hoc
network (MANET). In this paper, we propose an access control framework for an ad
hoc network infrastructure. Compared with the wired and semi wireless network
infrastructures in which users should be authenticated and given specific rules for
access to the network resources, in an ad hoc network infrastructure some trusted
nodes will be required to play the roles of authentication servers and policy
enforcement rule servers when the previous servers leave the network. In order to
reach to this purpose firstly, a model for Dynamic Monitoring in Ad hoc Networks is
described which is going to be used as the trust based part of our proposed model,
after that an access control Framework for semi-infrastructure ad hoc networks will
be explained. Then, we will introduce our proposed framework to develop the
mentioned semi ad hoc framework in a fully ad hoc network environment.

T.K. Araghi (&)  M. Zamani  A.A. Manaf

Advanced Informatics School, Universiti Teknologi Malaysia,
54100 Kuala Lumpur, Malaysia
e-mail: [email protected]
M. Zamani
e-mail: [email protected]; [email protected]
A.A. Manaf
e-mail: [email protected]
S.K. Araghi
Faculty of Engineering, Multimedia University Malaysia,
63100 Cyberjaya, Selangor, Malaysia
e-mail: [email protected]

Ó Springer International Publishing Switzerland 2015 747

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_70
748 T.K. Araghi et al.

70.1 Introduction

Setting up a MANET is usually in case of emergency for temporary operations or

when there are no resources to set up an elaborate network. Routing the packets
between nodes in ad hoc networks relies on implicit trust relationships in one hand,
and on the other hand, the ad hoc routing protocols must address either general
security objectives like authentication, confidentiality, integrity, availability and
non-repudiation or, location confidentiality, cooperation fairness and absence of
traffic diversion [1–5].
Besides of security threats and vulnerabilities including lack of proper
authentication, insecure routing and complex mobility and identity management,
the most important challenges in trust management and security caused by self
organizing in mobile ad hoc networks. Trust establishment becomes complicated
while nodes participating in the network do not have any prior knowledge from
each other [6–9].
Protecting the network resources from unauthorized accesses makes policy
based security as a major need for such networks. Although, this purpose is being
very complex because of the diversity in security loop holes in different network
layers and uncontrolled media access to change topology specifications [10–13].
Managing permissions and roles are usually manual tasks that impose a huge
administrative burden and it is error-prone if a large number of permissions and
roles are frequently changed. Enforcing policies for security is in the primary
stages in MANETs. In general, to facilitate retrieval of data management, data
access in a wireless ad hoc network requires a service channel to boost cache
recovery, application logging and role base access logging [14–17].
In this paper we developed an access control framework in a fully ad hoc
environment which had been introduced for semi-ad hoc environment by [6]. In
the proposed framework firstly, based on the dynamic trust monitoring the key
nodes which are going to play the roles of authentication servers and role servers
will be specified. Then, considering the proposed framework, the policies will be
defined to let the routing protocol for every routing decision.

70.2 Literature Review and Related Works

Memon summarized the database access concerns in order to highlight access

constraints and the pertinent issues surface after an ad hoc network of devices is
formed. These constraints are divided to organizational policy and access criteria
including policies defined for users, groups and roles in which the users access
depends on their organizational placement, application contents that enables
identification of applications regardless of used ports, standard ACLs like those
worked in layers 2, 3, 4 which defined based on the source MAC address, source
and destination IP Address and protocol and either type, periodic and absolute time
70 An Access Control Framework in an Ad Hoc Network Infrastructure 749

such as policies defined for a specific time or during particular time of days,
location in which Access policy can be defined based on the direction of traffic that
is a function of the source and destination locations and finally access control
based on the user’s roles in which once a session is established, a subset of roles is
activated, which contains functional roles include permissions that a user need
during a session based on his role in organization [15, 18, 19].
Alicherry et al. and Zhao et al. proposed frameworks to implement the
mechanisms of access control similar to a service on the resources of the network,
but the effect of these access control mechanisms is not specified on routing
schemes [20, 21, 13].
In the following some related works about the concept of trust in the MANET
and also a framework for implementing access control policies which are the basis
of our proposed model will be investigated.

70.2.1 Dynamic Monitoring for Trust in Ad Hoc Networks

Shohreh et al. proposed a trust based clustering model setting up based on a

computed trust value of the nodes participating in the network with regards to
previous experience of the nodes in forwarding or receiving the packets in order to
supervise the behavior of the nodes in each cluster. For this purpose a head cluster
and an agent for it will be selected in every cluster. The cluster head monitors and
supervises intra cluster transactions considering interactions among its cluster. If
the cluster head leaves the network its agent compensates its role in the network,
so cluster head and its agent maintain the information of all nodes in their cluster.
The highest trust value will be assigned to master head which supervises the inter
cluster transactions and monitors all cluster heads in the network. Direct trust
value is computed based on the rate of successful packet and total packet which
transmitted with nodes. T is total packets, s is success packets and n is the number
of nodes, p is the number of recorded previous interactions and p B (n/3).
!
S XP
1
T V¼ þ TV : ð70:1Þ
T N1 3

Based on this formula, last one-third experiences of each node are preserved in
a table. For example when the number of nodes is 15, 5 last experiences will be
maintained in the table as trust levels. The range of trust value is from 0 to 1,
which represents uncertain, distrust and trustable nodes, so three levels as A, B and
C considered as Table 70.1 shows. Nodes within a range of 0.7 to 1 are called as
the most trustable nodes and can be selected for particular roles and responsibil-
ities [22, 23].
750 T.K. Araghi et al.

Table 70.1 The different trust levels

Trust level Min range Max range
Level A 0.7 1
Level B 0.3 0.7
Level C 0 0.3

70.2.2 An Access Control Framework

in Semi-Infrastructures Ad Hoc Network Model

This framework designed for an implemented semi ad hoc network connected to a

backbone network. The functionality of backbone depends on the global security
policies enforced on LAN or WAN, but in ad hoc section these policies cannot be
implemented because of the mobility of nodes. Hence, the framework is imple-
mented to consider the security policies in both backbone and ad hoc sections.
Backbone network includes one or more authentication server(s) and a Global Policy
Management Server (GPMS). The GPMS is used in the framework for managing the
global policies appropriately. On the other hand, several mobile nodes will be
selected in MANET by the backbone network as Policy Enforcing Node (PEN).
These nodes are responsible for running security rules in MANET. The security rules
will be distributed through a network message named Policy Enforcing Message
(PEM). A network connectivity matrix is also needed in each time interval in order to
distribute global policies among nodes connected to the network. The connectivity
matrix represents connectivity between different nodes in ad hoc network. The
process of selecting PEN and distributing the global policy rules will be done by the
framework after every time interval using the network information given by the
authentication and role servers. When PEN is selected, GPMS sends the global policy
rules to the policy enforcing node. Receiving the global policy rules, the PEN dis-
tributes the rules to different MANET nodes. The GPMS selects the PENs. The policy
rules will be bound by each node to its own Ad hoc interface and the routing decisions
will be made by the nodes based on the policy rules. For the new nodes joining to the
network, they should be authenticated with the backbone. Message Authentication
Code (MAC) is used in order to authenticate PEMs. These policies can be specified
by the role of the MANET nodes, which let enforcing the Role Based Access Control
(RBAC) policies in MANET. The message exchanged among servers, PENs and
nodes include the sequence and TTL fields [6, 24, 25].

70.3 Proposed Model

The mentioned model for semi ad hoc network can be implemented for the whole
of ad hoc network environment, if the role of authentication and role servers and
also global policy management servers (GPMS) are distributed among nodes in the
70 An Access Control Framework in an Ad Hoc Network Infrastructure 751

network. For this purpose, trusted nodes within the radio range of each zone is
required to have a vast capacity using as repositories to maintain the role infor-
mation and global policy rules. Also a master head will be selected based on the
Table 70.1, which is a node with the highest value in the range A as an authen-
tication server.
Selecting nodes as GPMS is based on the proposed formula in the literature, so
the nodes with the highest trust level (placed in level A) will be chosen as GPMS,
also there is an agent for each GPMS.
Regarding the mobility of nodes, when a GPMS node leaves its zone its agent
selected as a new GPMS in a short delay time, and then the supervising among
zone is preserved dynamically.
Since authenticity of the policy enforcing message is vital, nodes can confirm
the source of the PEM by Message Authentication Code (MAC) to over- come the
threat of impersonation. In MAC, the hash value of the whole message body is
calculated and sent with the header. The hash key will be negotiated at the time of
authentication of the node [26–28].
Policy enforcing messages are transmitted between MANET nodes and GPMS.
In comparison to the previous mentioned framework, since the new framework
has developed in an ad hoc environment, the two parts of ‘‘GPMS at backbone’’
and ‘‘PENs’’ will be eliminated. GPMS will be selected based on the highest trust
level and also the agent node is prepared to compensate the role of GPMS in case
of leaving the network. GPMS constitutes of two parts, role assigning and
enforcing the policy rules to the nodes.

70.3.1 Functional Steps

Step 1: When a node wants to join to the MANET, it will be authenticated by
the master head as an authentication server.
Step 2: A role request will be sent to GPMS. The roles will be assigned and the
access level for the specified role will be granted to the node.
Step 3: Afterwards, policy rules will be sent through policy enforcing messages
(PEM) to the nodes. Each PEM will be enforced based on the roles that
each node acquired on step 2.
Step 4: Based on the enforced rules from step 3, the decision will be made for
routing that whether a route request is allowed to be performed based on
the policies or not. The flow of operational framework is shown in
Fig. 70.1.
The rules contain a five tuple. For example, Deny A * B80 says, node A cannot
access B via port 80. These rules are acted like the Access Lists (ACL) in order to
enforce the filtering on input and output traffics and protect the resources of
networks from unauthorized access.
752 T.K. Araghi et al.

Fig. 70.1 Ad hoc flow of operational framework

70.4 Conclusion and Future Work

This framework is designed for small organizations with a limited number of

nodes. So, the future work is to develop this model for large organizations.

References

1. Bhatia, R., et al.: Security issues prtaining to ad hoc networks-a survey, IJCSMS Int.
J. Comput. Sci. Manage. Stud. 12 (2012)
2. Araghi, T.K., Zamani, M., Manaf, A.A., Abdullah, S.M., Bojnord, H.S., Araghi, S.K.: A
survey for prevention of black hole attacks in wireless mobile adhoc networks using IDS
agents. 12th International Conference on Applied Computer and Applied Computational
Science (ACACOS ‘13), Kuala Lumpur, Malaysia, 2–4 April 2013
3. Ghazizadeh, E., Zamani, M.: Cloud computing and single sign on: a literature review. Sci.
World J. ISSN: 1537-744X. IF = 1.73. Q1 (Hindawi Publishing Corporation (UR)) (2013)
4. Janbeglou, M., Zamani, M., Ibrahim, S.: Redirecting network traffic toward a fake DNS
server on a LAN. 3rd IEEE International Conference on Computer Science and Information
Technology, pp. 429–433, Chengdu, China, 9–11 July 2010
5. Janbeglou, M., Zamani, M., Ibrahim, S.: Redirecting Outgoing DNS requests toward a fake
DNS server in a LAN. IEEE International Conference on Software Engineering and Service
Science, pp. 29–32. Beijing, China, 16–18 July 2010
6. S. Maity, et al., An access control framework for semi-infrastructured ad hoc networks, in
Computer Technology and Development (ICCTD), 2010. 2nd International Conference on,
2010, pp. 708–712
7. Araghi, T.K., Zamani, M., Manaf, A.A., Abdullah, S.M., Bojnord, H.S., Araghi, S.K.: A
secure model for prevention of black hole attack in wireless mobile ad hoc networks. 12th
International Conference on Applied Computer and Applied Computational Science
(ACACOS ‘13), Kuala Lumpur, Malaysia, 2–4 April 2013
8. Ghazizadeh, E., Zamani, M., Manan, J.A., Alizadeh, M.: Trusted Computing Strengthens
Cloud Authentication. Sci. World J. 2014, Article ID 260187, 17 (2014). doi:10.1155/2014/
260187 (IF: 1.730)
70 An Access Control Framework in an Ad Hoc Network Infrastructure 753

9. Araghi, T.K., Zamani, M., Manaf, A.A.: Performance analysis in reactive routing protocols in
wireless mobile ad hoc networks using DSR, AODV and AOMDV. International Conference
on Informatics and Creative Multimedia 2013 (ICICM2013), pp. 76–79. Kuala Lumpur, 3–6
September 2013
10. Savola, R.M.: Node level security management and authentication in mobile ad hoc
networks, in mobile data management: systems, services and middleware, 2009. MDM’09.
Tenth International Conference on, 2009, pp. 449–458 (2009)
11. Araghi, T.K., Zamani, M., Manaf, A.A., Abdullah, S.M., Bojnord, H.S., Araghi, S.K.: A
survey for prevention of black hole attack in wireless mobile ad hoc networks using
cryptographic techniques. 12th International Conference on Applied Computer and Applied
Computational Science (ACACOS ‘13), Kuala Lumpur, Malaysia, 2–4 April 2013
12. Mohebbi, K., Ibrahim, S., Zamani, M.: UltiMatch-NL: A web service matchmaker based on
multiple semantic filters. PLOS ONE J. Publ. Libr. Sci. eISSN: 1932-6203. (UR). Impact
factor (2012) = 3.73. Q1 (2013)
13. Beigzadeh, S., Zamani, M., Ibrahim, S.: Development of a web-based community
management information system. The Fourth International Conference on Information and
Computing (ICIC2011). pp. 3–6. Phuket, Thailand, 25–27 April 2011
14. Araghi, T.K., Zamani, M., Manaf, A.A., Abdullah, S.M., Bojnord, H.S., Araghi, S.K.: A
survey for prevention of black hole attack in wireless mobile ad hoc networks using trusted
neighbor nodes. 12th International Conference on Applied Computer and Applied
Computational Science (ACACOS ‘13), Kuala Lumpur, Malaysia, 2–4 April 2013
15. Memon, Q.A.: Implementing role based access in healthcare ad hoc networks. J. Netw. 4,
192–199 (2009)
16. Sadeghian, A., Zamani, M.: Detecting and preventing DDoS attacks in botnets by the help of
self triggered black holes. Asia-Pacific Conference on Computer Aided System Engineering
(APCASE). Bali, Indonesia, 10–12 Feb 2014
17. Yazdanpanah, S., Chaeikar, S.S., Zamani, M., Kourdi, R.: Security features comparison of
master key and ikm cryptographic key management for researchers and developers. 2011 3rd
International Conference on Software Technology and Engineering (ICSTE 2011),
pp. 365–369. Kuala Lumpur, Malaysia, 12–13 Aug 2011
18. Janbeglou, M., Zamani, M., Ibrahim, S.: Improving the security of protected wireless internet
access from insider attacks. AISS: Adv. Inform. Sci. Serv. Sci. 4(12), 170–181. ISSN: 2233-
9345 (2012)
19. Ghazizadeh, E., Zamani, M., Manan, J.A., Pashang, A.: A survey on security issues of
federated identity in the cloud computing. IEEE International Conference on Cloud
Computing Technology and Science, pp. 562–565. Taiwan, 3–6 Dec 2012
20. Alicherry, M., et al.: Deny-by-default distributed security policy enforcement in mobile ad
hoc networks, In: Security and Privacy in Communication Networks, pp. 41–50 (2009)
21. Zhang, H., et al.: Bootstrapping deny-by-default access control for mobile ad hoc networks,
In: Military Communications Conference, 2008. MILCOM 2008. IEEE, pp. 1–7 (2008)
22. Shohreh, H., et al.: Dynamic monitoring in ad hoc network. Appl. Mech. Mater. 229,
1481–1486 (2012)
23. Honarbakhsh, S., Zamani, M., Honarbakhsh, R.: Dynamic Monitoring in Ad Hoc Network.
Applied mechanics and materials, vols. 22–231, pp. 1481–1486. Trans Tech Publications,
Switzerland. ISSN: 1660–9336 (2012)
24. Zamani, M. Manaf, A.A., Daruis, R.: Azizah technique for efficiency measurement in
steganography. ICIDT 2012, 8th International Conference on Information Science and
Digital Content Technology 3, pp. 480–484. Jeju, Korea, 26–28 June 2012
25. Ghazizadeh, E., Zamani, M., Manan, J.A., Khaleghparast, R., Taherian, A.: A trust based
model for federated identity architecture to mitigate identity theft. International Conference
for Internet Technology and Secured Transactions. London. 10–12 Dec 2012
26. Min, Z., Jiliu, Z.: Cooperative black hole attack prevention for mobile ad hoc networks,
pp. 26–30 (2009)
754 T.K. Araghi et al.

27. Chaeikar, S.S., Razak, S.A., Honarbakhsh, S., Zeidanloo, H.R., Zamani, M., Jaryani, F.:
Interpretative key management (IKM), a novel framework. 2010 International Conference on
Computer Research and Development, pp. 265–269. Kuala Lumpur, Malaysia 7–9 May 2010
28. Nikbakhsh, S., Zamani, M., Manaf, A.A., Janbeglou, M.: A novel approach for rogue access
point detection on the client-side. 26th IEEE International Conference on Advanced
Information Networking and Applications. Japan, 26–29 March 2012
Chapter 71
Enhancement of Medical Image
Compression by Using Threshold
Predicting Wavelet-Based Algorithm

N.S.A.M. Taujuddin and Rosziati Ibrahim

Abstract In recent decades with the rapid development in biomedical engineer-

ing, digital medical images have been becoming increasingly important in hos-
pitals and clinical environment. Apparently, traversing medical images between
hospitals need a complicated process. Many techniques have been developed to
resolve these problems. Compressing an image will reduce the amount of redun-
dant data with the good quality of the reproduced image sufficiently high,
depending on the application. In the case of medical images, it is important to
reproduce the image close to the original image so that even the smallest details
are readable. The aim of this paper is to propose a new compression algorithm by
using the threshold values. It started by segmenting the image area into Region of
Interest (ROI) and Region of Background (ROB) and use the special features
provide by wavelet algorithm to produce efficient coefficients. These coefficients
are then will be used as threshold value in our new proposed thresholding pre-
dicting for compression algorithm. The new compression algorithm is expected to
produce a fast compression algorithm besides decreasing the image size without
tolerating with the precision of image quality.

Keywords Wavelet  Hard threshold  Soft threshold  Image compression

N.S.A.M. Taujuddin (&)

Faculty of Electrical and Electronic Engineering, Universiti Tun Hussein Onn Malaysia,
86400 Parit Raja, Batu Pahat, Johor, Malaysia
e-mail: [email protected]; [email protected]
R. Ibrahim
Faculty of Computer Science and Information Technology, Universiti Tun Hussein Onn
Malaysia, 86400 Parit Raja, Batu Pahat, Johor, Malaysia
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 755

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_71
756 N.S.A.M. Taujuddin and R. Ibrahim

71.1 Introduction

Advances over the past decade in many aspects of digital technology especially
devices for image acquisition, data storage, and bitmapped printing and display have
brought about many applications of digital imaging. However, problems involving
storage space and network bandwidth requirements arise when large volumes of
images are to be stored or transmitted, as is the case with medical images.
From the diagnostic imaging point of view, the challenge is how to deliver
clinically critical information in the shortest time possible. A solution to this
problem is through image compression. The main objective of this compression is
to reduce redundancy of the data image in order to be able to stored or transmit
data in an efficient form [1–5].
Among various algorithms proposed by researchers, wavelet gains a high
popularities in compression domain because of its distinctive features. Wavelet is
well known because of its energy compactness in the frequency domain. Thres-
holding is one of the processes in wavelet. Tay et al. says that the selection of
threshold(s) is/are the key performance to an effective compression [6].
The rest of the paper is organized as follows: the principles of wavelet is
detailed in Sect. 71.2. The project’s methodology is presented in Sect. 71.3.
Section 71.4 contains the experimental result and analysis. While Sect. 71.5
conclude the entire paper.

71.2 The Wavelet

Wavelet is a flexible tool with rich mathematic content and has enormous potential
in many applications and greatly being used in the field of digital images. Wave-
let algorithm work as signal processing in such a way like the human vision do. It
provides a much more precisely in digital image, movies, color image and signal.
It also has widely used in data compression, fingerprint encoding and also
image. There are three properties of wavelets, (a) separability, scalability and
translatability (b) multiresolution compatibility and (c) orthogonality [7].
One of the popular wavelet is Discrete Wavelet Transform (DWT). The term
discrete in DWT refer to the separation of transformation kernel as well as sep-
aration in fundamental nature and function.
Basically, DWT is a transformation process that produces the minimum number
of coefficients that sufficient enough for reconstruction of the transform data
accurately to the original signal. DWT is usually presented in term of its recovery
transformation:

X
1 X
1
k
xðtÞ ¼ dðk; ‘Þ22 Wð2k t  ‘Þ ð71:1Þ
k¼1 ‘¼1
71 Enhancement of Medical Image Compression by Using … 757

dðk; ‘Þ is sampling of W(a, b) at discrete points k and ‘. While k is referring to a as

a is a = 2 k and b is referring to ‘ as b ¼ 2k ‘.
The DWT introduce the scaling function or sometimes referred as smoothing
function. It use dilation or two-scale difference equation:

X
1 pffiffiffiffiffi
/ðtÞ ¼ 2cðnÞ/ð2t  nÞ ð71:2Þ
n¼1

In this equation, c(n) is a series of scalars describing specific scaling function.

Wavelet in DWT itself can be define from these scaling function:

X
1 pffiffiffiffiffi
WðtÞ ¼ 2d ðnÞ/ð2t  nÞ ð71:3Þ
n¼1

Here, d(n) is a series of scalar that define the discrete wavelet in terms of
scaling function. DWT can well implemented in the above equation as well as
using filter bank technique. Typically, wavelet use two filters, namely analysis
filter and synthesis filter. The analysis filter is used to split the original signal to
several spectral components called subband.
Firstly, the signal will pass a low pass filter for approximation coefficients
outputs. Then, it will pass through the high pass filter resulting the detail
coefficients.
In the analysis filter, some points need to be eliminated. This operation is called
downsamping or quantization process and usually illustrated as ;2. The process is
done to maximizing the amount of necessitate detail and ignoring ‘not-so-wanted’
details.
Here, some coefficient values for pixel in image are thrown out or set to zero.
This is called as the thresholding process and it will give some smoothing effect to
the image.
In order to compress the image, Wavelet analysis can be used to divide the
information of an image into approximation and detail sub-signals. The approxi-
mation sub-signal shows the common trend of pixel values, and three detail sub-
signals show the horizontal, vertical and diagonal details or changes in the image.
If these details are very small then they can be set to zero without significantly
changing the image. The details under the fixed threshold represent a small enough
detail and it can be set to zero. The greater the number of zeros leads to the greater
compression.
In inverse of analysis bank, the synthesis bank will do the upsampling (:2) to
reconstruct the original fine scale coefficient by combining the scale and wavelet
coefficients at lower coarser scale. During upsampling the value of zero will be
inserted between 2 coefficients because during the downsampling, the every sec-
ond coefficient is thrown away.
The process of upsampling and downsampling is illustrated as in Fig. 71.1.
758 N.S.A.M. Taujuddin and R. Ibrahim

Fig. 71.1 Two-band multirate analysis/synthesis system [3]

71.2.1 Concept of Thresholding

In the wavelet transform, the noise energy is distribute in all wavelet coefficients,
while the original signal energy is found in some of the coefficients. Therefore, the
signal energy is found much larger than noise energy. So, small coefficients can be
considered as caused by noise while large coefficients are triggered by significant
signal features.
Based on this idea, thresholding process is proposed. Thresholding is a process
of shrinking the small absolute coefficients value while retaining the large absolute
coefficient value. It will produce finer reconstruct signal. Threshold also can be
define as the Peak Absolute Error (PAE) accepted for image reconstruction [8].
Hard and soft threshold are the common operator used in conjunction with
DWT. Donoho is the person who first introducing the word ‘de-noising’ to explain
the process of noise reduction in threshold [9].
The wavelet coefficient for hard threshold, hh, is performed as follows:

yðiÞ; jyðiÞj [ k
hh ðiÞ ¼ ð71:4Þ
0; others

where y(i) is the wavelet coefficients, k is the specified threshold. While, for soft
threshold, hs, the coefficients is expressed as below:

sgnðyðiÞÞ½jyðiÞj  k; jyðiÞj  k
hs ðiÞ ¼ ð71:5Þ
0; others

The elements with absolute value is lower than the threshold value will be set to
zero and then the other coefficient will be shrunk. sgn(*) is a sign function.

1 n[0
sgnðnÞ ¼ ð71:6Þ
1 n\0

The hard threshold zeros all the coefficients valued below than threshold value
and retain the rest unchanged. Whereas, the soft threshold scaled the coefficients in
continuous form with the center of zero [10]. These two techniques have their own
71 Enhancement of Medical Image Compression by Using … 759

Fig. 71.2 a Hard threshold function. b Soft threshold function

strength and weakness. Hard threshold known as good in preserving edges but bad
in de-noising while soft threshold is contradict as can be seen in Fig. 71.2.
There are two types of thresholding; global and level dependent threshold.
Global threshold imply single threshold value globally to all wavelet coefficient
while level dependent threshold use different threshold value at different level.
Threshold value estimation is very crucial. If the threshold value set too small,
it will adopt noise into the signal. While, if the threshold value is too high, the
important coefficients value will be screened out leading to deviation condition.
There are some threshold estimation techniques exists such as [11]:
(a) VisuShirnk
Donoho is first introducing the wavelet threshold shrinkage method where he
suggest that the coefficients have a tendency be to set as zero when the
coefficients are greater than threshold. The threshold must meet the specifi-
pffiffiffiffiffiffiffiffiffiffiffiffi
cation of k ¼ rn : 2 ln N where rn is the noise variance and N is the length of
the signal.
(b) SureShrink
SureShrink is a soft estimator, where for the given k, the probability is
collected while reducing the non-probability.
(c) HeurSure
HeurSure will choose the best predictor variable threshold.
(d) Minmax
By using the fixed threshold value, it will generate a minimum mean-square
error based on minimax criterion.
As an example for wavelet threshold de-noise calculation, the signal x(t),
contain of impulse and noise can be expressed as:

xðnÞ ¼ pðtÞ þ nðtÞ ð71:7Þ

where n(t) refer to Gaussian noise with mean zero and standard deviation r while
p(t) refer to impulse. The signal x(t) is then being transferred to time-scale plane.
760 N.S.A.M. Taujuddin and R. Ibrahim

The threshold, k, is measured by using the existence rules to shrink the wavelet
coefficients. Here, universal threshold is applied:
pffiffiffiffiffiffiffiffiffiffiffiffi
k ¼ r 2 ln N ð71:8Þ

For unknown r, one can be replace by MAD/0.6745, where MAD is median

absolute value of the finest scale wavelet coefficients and N is the number of data
samples in measured signal. Inverse wavelet transform use the shrunken coeffi-
cient, and the series retrieve represent the estimation of impulse p(t) [12].

71.2.2 Recent Research on Wavelet Threshold Predicting

Algorithm in Image Compression

The wavelet threshold uses the ordinary multiresolution analysis where the dis-
crete detail coefficients and discrete approximation coefficients are attained by
multilevel wavelet decomposition.
To find the best threshold is hard because it requires the knowledge of original
data. Beside the contemporary hard threshold, the soft-threshold can be used since
it is closer to the optimum minimal rate and protect signal regularity beside
reducing the gap between preserve and discarded coefficient for a better recovery
[13]. Another solution is by using adaptive soft-threshold or fix the threshold for
each wavelet sub-band.
Chen et al. [14] are using adaptive prediction technique in solving multicol-
linearity problem. Besides, he suggest to adjust predictor variable based on image
properties so that more accurate prediction is archived.
Tree-Structured Edge-Directed Orthogonal Wavelet Packet Transform (TS-E-
DOWPT) proposed in [15] decomposes image using edge-directed orthogonal and
calculate the cost function. This technique improves the PSNR value and visual
quality. Besides, trimming the structure into quad-tree may reduce the stain.
To get a better compression ratio, Hosseini et al. [16] recommending a tech-
nique that separating the Region of Interest (ROI) and Background (BG) using
growing segment and then encode both of the segment using Contextual Quanti-
zation. Different weightage is used for different region, where higher rate for ROI
and lower rate for BG.
Pogam et al. in [17] give different view which combining the wavelet transform
with curvelet transform and incorporating it with the local adaptive analysis
thresholding. This technique contributes to an efficient denoising while pre-serving
as much as possible the original quantitative and structural information and ROI of
image.
71 Enhancement of Medical Image Compression by Using … 761

71.3 Methodology

Although the forgoing techniques can make an effective approximation of

threshold value, but they got no idea on the relationship between threshold value
and quantization step [18]. Besides, the usage of hard threshold in quantizing the
coefficients will lead to blocky artifact on medical image [19].
Targeting on this problems, this research is done to develop an efficient
threshold prediction algorithm by using the wavelet features to produce a fast
compression algorithm besides decreasing the image size without tolerating with
the precision of image quality.
Medical community also raise a high intention to produce a low computational
cost algorithm with high speed compression and decompression to assist the
existence network bandwidth capability while reducing the image size to upkeep
the limited storage size.
As in the literature, the wavelet coefficient is predicted based on fix location and
variables. But, medical images have its own statistical distribution and have dif-
ferent properties on different subbands. So, to get more precise prediction, the
amount of predictor variable must be adjusted based on the image’s properties.
Figure 71.3 shows the general image compression system with examples of
algorithms used in each process. The compression process starts with transforming
the image into coefficient where it usually done by using Discrete Cosine Trans-
form (DCT), Discrete Wavelet Transform (DWT) or Fast Fourier Transform
(FFT).
Then, the retaining coefficients are quantized by using hard, soft or semi-soft
threshold resulting the stream of symbol. This is where information lost occurs.
The entropy coding or compression process will use efficient lossy or lossless
compression algorithm for example the EZW, SPHIT, Huffman, Bitplane
Encoding and many more to produce the bit streams representing the compressed
image.
By using the same general image compression system, we propose a new
compression algorithm with some extended features. Below are the proposed
algorithm steps and it is illustrated in Fig. 71.4:
1. The original image is segmented to Region of Interest (ROI) and Region of
Background (ROB).
2. DWT is used to produce sequence of wavelet coefficient and separate it to low
frequency and high frequency subband.
3. The correlation between adjacent wavelet coefficients are analyzed to get the
best suit coefficient relationship.
4. Resulting wavelet coefficient are thresholded by using efficient prediction
scheme to get the best truncated threshold. Then the prediction equation is
applied for thresholding process to get the significant predicted wavelet
coefficient.
762 N.S.A.M. Taujuddin and R. Ibrahim

Fig. 71.3 General image compression system

Fig. 71.4 The proposed algorithm steps

71.4 Experimental Result and Analysis

The standard gray-scale 512 9 512 sized Barbara image is used in this testing to
evaluate the popular existing threshold algorithms; the hard and soft threshold. The
image is added with Gaussian white noise to facilitate the comparison.
To evaluate the efficiency of the compression algorithm, Peak Signal to Noise
Ratio (PSNR) and Mean Square Error (MSE) are used to evaluate the quality of
compression. Higher the PSNR value representing a higher compression quality
and vice versa [16]. While lower MSE value representing better image quality vice
versa.
The PSNR can be defined as:
 
2552
PSNR ¼ 10 log10 dB
MSE

While MSE is define as:

1 X M X N  
f ðx; yÞ  ^f ðx; yÞ2
MSE ¼
M  N x¼1 y¼1
71 Enhancement of Medical Image Compression by Using … 763

Fig. 71.5 a Original image. b Noisy image. c Hard thresholded image. d Soft thresholded image

M.N is the size of the image while f ðx; yÞ is the pixel value of original image
and ^f ðx; yÞ is the pixel value of the reconstructed image.
Figure 71.5 shows the resulting Barbara original, noisy, hard thresholded and
soft thresholded image. It is clearly can be seen that the quality of image is
significantly affected with different threshold algorithm. The soft thresholded
image is better in terms of noise overwhelmed while protecting the image edge.
But soft threshold method shows the degradation value of PSNR as can be seen in
Table 71.1.

71.5 Conclusion

This paper has discussed a new compression algorithm by putting an efficient

threshold prediction algorithm using the wavelet features to produce a fast com-
pression algorithm. The purpose of this new compression algorithm is to decrease
the image size without tolerating with the precision of image quality. As the
project is still under go, the preliminary analysis done in Sect. 71.4 shows the need
764 N.S.A.M. Taujuddin and R. Ibrahim

Table 71.1 The PSNR value for different de-noising method

De-noising method PSNR
r = 25 r = 50 r = 100
Noisy image 28.8167 27.8777 27.4628
Hard threshold 23.0161 21.3947 19.9306
Soft threshold 21.7055 20.3142 19.1190

to develop an effective thresholding algorithm that can provide the improvement

preservation details at low bit rates while increasing the PSNR value. Protecting
details at edges is very crucial especially for sensitive data such as the medical
images. Therefore, modification on prediction procedure in threshold step as
proposed in our algorithm should be performed in order to abolish the blocking
and edge effect while increasing the effectiveness and reliability of the compressed
image.

Acknowledgment The authors would like to thanks the Universiti Tun Hussein Onn Malaysia
(UTHM), Office of Research, Innovation, Commercialization and Consultancy Management
(ORICC) and Malaysian Ministry of Education for facilitating this research activity.

References

1. Strintzis, M.G.: A review of compression methods for medical images in PACS. Int. J. Med.
Inform. 52(1–3), 159–165 (1998)
2. Burak, S., Carlo, G., Bernd, T., Chris, G.: Medical image compression based on region of
interest, with application to colon CT Images. In: 23rd Annual EMBS International
Conference, 2001, pp. 2453–2456 (2001)
3. Kofidis, E., Kolokotronis, N., Vassilarakou, A., Theodoridis, S., Cavouras, D.: Wavelet-based
medical image compression. Futur. Gener. Comput. Syst. 15(2), 223–243 (1999)
4. Janaki, R.: Enhanced ROI (region of interest algorithms) for medical image compression. Int.
J. Comput. Appl. 38(2), 38–44 (2012)
5. Celik, M.U., Sharma, G., Tekalp, A.M.: Gray-level-embedded lossless image compression.
Signal Process. Image Commun. 18(6), 443–454 (2003)
6. Tay, P.C., Acton, S.T., Hossack, J.A.: Computerized medical imaging and graphics a wavelet
thresholding method to reduce ultrasound artifacts. Comput. Med. Imaging Graph. 35(1),
42–50 (2011)
7. Gonzalez, R.C., Woods, R.E.: Digital Image Processing, Second, p. 43. Prentice Hall, Upper
Saddler River (2002)
8. Baligar, V.P.: Low complexity, and high fidelity image compression using fixed threshold
method. J. Inf. Sci. 176, 664–675 (2006)
9. Donoho, D.L.: De-noising by soft-thresholding. IEEE Trans. Inf. Theory 41(3), 613–627
(1995)
10. Li Jianmin, X.S.: Analysis and application of modified methods of wavelet threshold
functions. In: Asia-Pasific Conference on Computational Intelligence and Industrial
Application, 2009, pp. 150–153 (2009)
11. Zhen, C., Su, Y.: Research on wavelet image threshold de-noising. In: 2010 International
Conference on Future Power Energy Engineering, pp. 3–6, Jun 2010
71 Enhancement of Medical Image Compression by Using … 765

12. Zang, H., Wang, Z., Zheng, Y.: Analysis of signal de-noising method based on an improved
wavelet thresholding. In: 2009 9th International Conference on Electronic Measurement and
Instruments, vol. 1, pp. 1–987, 1–990, Aug 2009
13. Bruni, V., Vitulano, D.A.: Combined image compression and denoising using wavelets.
Signal Process. Image Commun. 22, 86–101 (2007)
14. Chen, Y.T., Tseng, D.C.: Wavelet-based medical image compression with adaptive
prediction. Comput. Med. Imaging Graph. 31(1), 1–8 (2007)
15. Huang, J., Cheng, G., Liu, Z., Zhu, C., Xiu, B.: Synthetic aperture radar image compression
using tree-structured edge-directed orthogonal wavelet packet transform. AEUE—Int.
J. Electron. Commun. 66(3), 195–203 (2012)
16. Hosseini, S.M., Naghsh-Nilchi, A.R.: Medical ultrasound image compression using
contextual vector quantization. Comput. Biol. Med. 42(7), 743–750 (2012)
17. Le Pogam, A., Hanzouli, H., Hatt, M., Le, C.C., Visvikis, D.: Denoising of PET images by
combining wavelets and curvelets for improved preservation of resolution and quantitation.
Med. Image Anal. 17(8), 877–891 (2013)
18. Manikandan, M.S., Dandapat, S.: Wavelet-threshold based ECG compression with smooth
retrieved quality for telecardiology. In: 2006 Fourth International Conference on Intelligent
Sensing and Information Processing, pp. 138–143, Dec 2006
19. Prudhvi, N.V., Venkateswarlu, T.: Denoising of medical images using dual tree complex
wavelet transform. Procedia Technol. 4, 238–244 (2012)
Chapter 72
Detection and Revocation of Misbehaving
Vehicles from VANET

Atanu Mondal and Sulata Mitra

Abstract The present work is the detection and revocation of misbehaving vehicles
in vehicular ad hoc network. In the present work vehicles are within the coverage
area of base stations and the base stations are within the coverage area of certifying
authority. Each vehicle detects misbehaving vehicles from its neighbors, creates a
certificate revocation list by mentioning the identification of the misbehaving
vehicles and sends this list to its parent base station. Each base station creates a
certificate revocation list after receiving the certificate revocation lists from the
vehicles within its coverage area and sends it to the certifying authority. The cer-
tifying authority creates a final certificate revocation list after receiving the certifi-
cate revocation lists from the base stations within its coverage area and broadcasts it
among the vehicles within its coverage area. The qualitative and quantitative per-
formance of the proposed scheme outperforms the existing schemes.

72.1 Introduction

The vehicular ad hoc network (VANET) consists of a group of independent

vehicles which are moving throughout the wireless network freely. The potential
threat and road accident are increasing due to high velocity of vehicles in VANET.
Several types of messages are exchanged among vehicles such as traffic infor-
mation, emergency incident notifications and road conditions to avoid road acci-
dents and congestion. It is important to forward message correctly in VANET.

A. Mondal (&)
Department of Computer Science and Engineering,
Camellia Institute of Technology, Kolkata, India
e-mail: [email protected]
S. Mitra
Department of Computer Science and Technology,
Bengal Engineering and Science University, Shibpur, India
e-mail: [email protected]; [email protected]

Ó Springer International Publishing Switzerland 2015 767

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_72
768 A. Mondal and S. Mitra

However, attacker nodes or misbehaving vehicles may damage the messages. The
misbehaving vehicles are authentic vehicles but their behavior deviates from the
required standard behavior. These vehicles may jeopardize the safety of other
vehicles, drivers, passengers as well as the efficiency of the transportation system.
Hence the critical part of any security mechanism in VANET is to identify and
revoke misbehaving vehicles.
Several revocation schemes have been proposed so far. In [1] the service
provider issues certificates for the vehicles with a limited temporal/spatial scope.
These certificates are usable by the vehicles within a particular geographic area or
within a certain time or both. The certificates are not tied to the vehicle’s regis-
tration and can be changed periodically during one service period. But it has a lot
of overhead for the creation and revocation of certificates for each new coming and
leaving vehicle.
The certificate revocation techniques and privacy-protection techniques are
proposed in [2, 3]. A security architecture for vehicular communication systems is
developed in [2]. The authors identify threats and models of adversarial behavior
as well as security and privacy requirements that are relevant to the vehicular
communication context. A SeVeCom baseline architecture is presented in [3]. The
various implementation and deployment-specific aspects such as flexible integra-
tion in existing communication-stacks, use of a hardware security module and
secure connections of vehicular communication on board units to in-vehicle bus
systems are also highlighted in [3]. But the authors discussed the concept of
certificate revocation list (CRL) distribution without considering high vehicle
traffic densities.
In [4] the scalable methods of distributing CRL and other large files over
VANET using vehicle to vehicle (V2V) and vehicle to infrastructure (V2I)
communications while taking advantage of the multi-channel operations in IEEE
1609.4 is proposed. The size of the CRL increases with vehicle density. Hence it is
difficult to distribute the CRL over the entire network with minimum delay and
network congestions.
In [5] the authors consider the distribution of CRL across a large-scale and
multi-domain vehicular communication system in a timely manner. A collabora-
tion scheme between regional certificate authorities (CAs) that allows CRLs to
contain only regional revocation information is proposed in [5]. Moreover it uses
erasure codes to enhance the robustness of the CRL distribution. The scheme does
not require any communication and cooperation between road side units (RSUs)
during CRL distribution. It minimizes CA-RSU and vehicle-CA-RSU interactions
thus limiting congestion in the network. The CRL size is reduced by using regional
CA and short lived certificates for travelling vehicles. But the distribution time of
CRL is tens of minutes which is too long for a high and dense network like
VANET.
A flexible, simple and scalable design for VANET certificates and new methods
for efficient certificate management is proposed in [6]. It reduces channel overhead
by eliminating the use of CRL. It also protects the system from adversary vehicles
by distributing information about adversary vehicles among the whole network
72 Detection and Revocation of Misbehaving Vehicles from VANET 769

and by revoking the certificates of malicious nodes. But the performance of the
scheme is not evaluated on the basis of overhead and delay.
An adaptable method to detect packet forwarding misbehavior based on the
principal of flow conservation of messages and the application of policy-based
management [7] is proposed in [8]. The message is forwarded to those vehicles
which are moving towards the location of the event. In the proposed scheme the
different tasks are assigned to a vehicle at different instant of time depending upon
the certain policy for identifying the misbehaving vehicles which increases the
complexity of each vehicle.
The present work is the detection and revocation of misbehaving vehicles in
VANET. The proposed VANET is a hierarchy having CA at the root level, base
stations (BSs) at the intermediate level and vehicles at the leaf level.
Each vehicle has an electronic license plate (ELP) in which its vehicle iden-
tification number (VIN) is embedded in encrypted (E_VIN) form. The ELP of a
vehicle broadcasts the E_VIN after entering into the coverage area of a new BS.
The BS verifies the authentication of the vehicle after receiving its E_VIN and
assigns a digital signature (D_Sig) to the vehicle if it is authentic [9]. The CA
maintains a CRL (CA_CRL) to store the E_VINs of the misbehaving vehicles and
broadcasts it for the vehicles within its coverage area.
Each vehicle receives beacon message periodically and service message
(S_MSG) after the occurrence of an event from the vehicles which are within its
coverage area. The steganography method is used to protect S_MSGs from the
access of intruder. The S_MSG consists of D_Sig of the sender vehicle, hidden
message (H_MSG), E_VIN of the sender vehicle and the operational part
(O_MSG) to retrieve the message (MESS) from H_MSG.
Each vehicle switches on a timer. It receives a S_MSG from a vehicle within its
coverage area, and searches CA_CRL for the value of E_VIN field of S_MSG. If
found it discards S_MSG. Otherwise it stores the S_MSG in a FIFO queue, triggers
ALGO_MESS to retrieve MESS from H_MSG field of S_MSG and inserts a
record in the form (E_VIN, MESS, REMARK) in a data table (DT). It repeats the
same steps of operation for the other received S_MSGs till the timer expires.
Initially the value of the REMARK field of all the records in DT is null. Each
vehicle triggers ALGO_V algorithm to detect the misbehaving vehicles using the
records in the DT after the expiry of the timer and inserts the E_VIN(s) of the
misbehaving vehicle(s) in a CRL. The algorithm repeats the same steps of oper-
ation for all the records in the DT and finally sends the CRL to its parent BS.
Each BS switches on a timer and triggers ALGO_B algorithm. The algorithm
receives CRLs from the vehicles within its coverage area and stores them in a
FIFO queue. It performs union operation among the received CRLs in the queue
for generating a new CRL. It repeats the same steps of operation till the timer
expires and sends the new CRL to CA after the expiry of the timer.
The CA switches on a timer and triggers ALGO_C algorithm. The algorithm
receives CRLs from the BSs under it and stores them in a FIFO queue. It performs
union operation among the received CRLs in the queue for generating CA_CRL.
770 A. Mondal and S. Mitra

It repeats the same steps of operation till the timer expires and broadcasts
CA_CRL among the vehicles within its coverage area after the expiry of the timer.
Unlike [1, 4, 6] the present work uses steganography to protect S_MSGs from
intruder in VANET. The use of encryption/decryption algorithm for secure mes-
sage transmission consumes a considerable amount of time due to their compu-
tational hardness. Moreover the length of message may increase during encryption
which increases storage overhead along with communication overhead. In the
present work CA_CRL consists of E_VIN of the misbehaving vehicles instead of
their IP address like [5]. So only the E_VIN of a vehicle is sufficient to identify,
detect and revoke misbehaving vehicles from VANET. Unlike [5] the present work
performs well in high density of vehicles. In [5] CA distributes pieces of CRL
among vehicles via RSUs. So in the worst case a vehicle needs to encounter all the
RSUs for collecting all the pieces of CRL. Hence the performance of CRL gen-
eration degrades if congestion occurs, if vehicles move slowly and if the distance
between RSUs increase. The misbehaving vehicles are identified by their E_VINs
during V2V communication in the present work. So no extra task is required to
assign to a vehicle for identifying misbehaving vehicle like [8]. There is no
requirement of reporting the reason of revocation to CA in the present work like
[10] which helps to reduce the delay in revocation.

72.2 Present Work

In this section the function of vehicle, BS and CA are elaborated for vth vehicle
(Vv) within the coverage area of Bth BS (BSB) under CA. The number of BSs
under CA and the number of vehicles under BSB are assumed as NO_OF_BS and
NO_OF_VB respectively.

72.2.1 Function of Vv

The function of Vv is elaborated for jth S_MSG (S_MSGj) that it receives from jth
vehicle within its coverage area. S_MSGj is of the form of (D_Sigj, E_VINj,
H_MSGj, O_MSGj). D_Sigj is the digital signature of the jth vehicle, E_VINj is the
E_VIN of the jth vehicle, H_MSGj is the hudden message and O_MSGj is the
operational part to retrieve the message (MESSj) from H_MSGj. O_MSGj field
contains the bit pattern corresponding to the different bitwise operators (AND, OR,
XOR). Each bitwise operator is represented by Size_BWO number of bits and the
size of O_MSGj (Size_O_MSGj) is exact multiple of Size _BWO. The ALGO_-
MESS algorithm is used by Vv to retrieve MESSj from H_MSGj by performing
(Size_O_MSGj/Size_BWO) number of bitwise operations (Number_BWOj)
among H_MSGj and E_VINj.
72 Detection and Revocation of Misbehaving Vehicles from VANET 771

Vv switches on a timer and initializes it to sv. It receives S_MSGj, increases a

counter (NS_MSGv) by 1 and searches CA_CRL for E_VINj. If E_VINj is in
CA_CRL or if (Size_O_MSGj%Size_BWO6¼0) it discards S_MSGj. Otherwise it
stores S_MSGj in a FIFO queue (Qv), triggers ALGO_MESS algorithm to retrieve
MESSj from H_MSGj and inserts a record (Rj) in the form (E_VINj, MESSj,
REMARKj) in a data table (DTv) for S_MSGj.

/*Retrieval of MESSj from H_MSGj*/ {MESSj / H_MSGj _ E_VINj

{i / 1 Go to L1}
k/1 else if (O_MSGj (i, i + Size_BWO-1)
while (k \ Number_BWOj){ represents XOR bit wise operator)
if (O_MSGj (i, i + Size_BWO-1) represents {MESSj / H_MSGj  E_VINj
AND bit wise operator) Go to L1}
{MESSj / H_MSGj ^ E_VINj else
Go to L1} Go to L1
else if (O_MSGj (i, i + Size_BWO-1) L1: {i / i + Size_BWO
represents OR bit wise operator) k / k + 1}}}

It repeats the same steps of operation for the other received S_MSGs till the
timer expires. The maximum number of S_MSGs in Qv and the maximum number
of records in DTv is NS_MSGv.
It triggers ALGO_V algorithm to detect the misbehaving vehicle after the
expiry of the timer. The function of this algorithm is elaborated for S_MSGj in Qv.
It compares MESSj of Rj with the value of the MESS attribute field of all other
records in DTv. In case of match it increases Matchj counter and in case of
mismatch it increases Mismatchj counter by 1 after each comparison. If
Matchj = Mismatchj it ignores Rj. If Matchj [ Mismatchj the Algo_V inserts
TRUE in the REMARKj attribute field of Rj. Otherwise it inserts FALSE in the
REMARKj attribute field of Rj, inserts E_VINj in a CRL (CRLv) and increases a
counter (NEVIN_CRLv) by 1. The Algo_V repeats the same steps of operation for
all the records in DTv to generate CRLv. Vv sends CRLv to its parent BS. The
maximum number of E_VINs in CRLv is NEVIN_CRLv.

72.2.2 Function of BSB

BSB triggers ALGO_B algorithm for generating a CRL (CRLB) and for sending
CRLB to CA. The algorithm switches on a timer and initializes it to sB. It receives
CRLs from the vehicles within its coverage area of BSB, stores them in a FIFO
queue (QB), increases a counter (NB_CRL) after receiving each CRL by 1 and
assigns the first received CRL to CRLB. It starts to update CRLB by performing
union operation among the existing CRLB and the received CRLs in QB as soon as
NB_CRL becomes equal to 2 till the timer expires. The ALGO_B algorithm sends
CRLB to CA after the expiry of sB. The maximum value of NB_CRL is NO_OF_VB
and the maximum number of E_VINs in CRLB is assumed as NEVIN_CRLB.
772 A. Mondal and S. Mitra

72.2.3 Function of CA

CA triggers ALGO_C algorithm for generating CA_CRL and for broadcasting

CA_CRL among the vehicles within its coverage area. The algorithm switches on
a timer and initializes it to sC. It receives CRLs from the BSs within the coverage
area of CA, stores them in a FIFO queue (QC), increases a counter (NC_CRL) after
receiving each CRL by 1 and assigns the first received CRL to CA_CRL. It starts
to update CA_CRL by performing union operation among the existing CA_CRL
and the received CRLs in QC as soon as NC_CRL becomes equal to 2 till the timer
expires. The ALGO_C algorithm broadcasts CA_CRL among the vehicles within
the coverage area of CA after the expiry of sC. The maximum value of NC_CRL is
NO_OF_BS and the maximum number of E_VINs in CA_CRL is assumed as
NEVIN_CA_CRL.

72.3 Simulation

The performance of the proposed scheme is evaluated qualitatively and quanti-

tatively. In this section the simulation parameters, qualitative analysis and quan-
titative analysis of the proposed scheme are considered for discussion.

72.3.1 Simulation Parameters

The size of E_VIN (Size_E_VIN) is 17 characters [9] and the size of each character
is assumed as 8 bits (extended ASCII format) in the proposed scheme. Hence
Size_E_VIN is 136 bits. The bit wise operation is performed among H_MSG and
E_VIN to retrieve MESS from H_MSG. Hence the size of H_MSG (Size_H_MSG)
is 136 bits. The size of D_Sig (Size_D_Sig) is 160 bits [9]. The size of S_MSG
(Size_SMSG) is (Size_D_Sig + Size_E_VIN + Size_H_MSG + Size_O_MSG)
bits. The value of sv, sB, sC, NO_OF_BS, NEVIN_CRLv, NEVIN_CRLB, NE-
VIN_CA_CRL, NO_OF_VB, NS_MSGv and Size_SMSG are assumed as 30, 20,
20 s, 3, 400, 400, 1200, 144, 400, and 3,232 bits. The data transmission rate
(Data_TR) is assumed as 6 Mb/s [11].

72.3.2 Qualitative Analysis

The qualitative analysis on the basis of communication overhead (COMM_OH),

storage overhead (STO_OH) and computation overhead (COMP_OH) is reported
in this section. The qualitative performance is evaluated by considering the
72 Detection and Revocation of Misbehaving Vehicles from VANET 773

maximum length of Qv, QB, and QC i.e. Qv has NS_MSGv number of S_MSGs, QB
has NO_OF_VB number of CRLs and QC has NO_OF_BS number of CRLs. The
qualitative performance of the proposed scheme is compared with [12] on the basis
of STO_OH.
Communication Overhead. The COMM_OH of the proposed scheme is
PNO OF BS
B¼1 ðCOMM OHB Þ/Data sec where COMM_OHB is the communication
overhead of BSB in bits.
Computation of COMM_OHB. BSB receives CRLs from NO_OF_VB number of
vehicles within its coverage area. The size of CRLv is (NEVIN_CRLv 9
Size_E_VIN) bits. Hence COMM_OHB due to the reception of NO_OF_VB
P
number of CRLs is NO v¼1
OF VB
(NEVIN_CRLv 9 Size_E_VIN) bits.
BSB sends CRLB to CA. The size of CRLB is NEVIN_CRLB 9 Size_E_VIN
bits. Hence COMM_OHB due to the transmission of CRLB is NEVIN_CRLB 9
Size_E_VIN bits.
BSB receives CA_CRL from CA. The size of CA_CRL is NEVIN_CA_CRL 9
Size_E_VIN bits. Hence the COMM_OHB due to the reception of CA_CRL is
NEVIN_CA_CRL 9 Size_E_VIN bits.
BSB sends CA_CRL to NO_OF_VB number of vehicles within its coverage
area. Hence COMM_OHB due to the transmission of CA_CRL is
NO_OF_VB 9 NEVIN_CA_CRL 9 Size_E_VIN bits.
P
Hence COMM_OHB = NO v¼1
OF VB
(NEVIN_CRLV 9 Size_E_VIN) +
Size_E_VIN 9 (NEVIN_CRLB + NEVIN_CA_CRL (1 + NO_OF_VB)) bits.
Storage Overhead. The STO_OH of the proposed scheme is the sum of STO_OH
of CA (STO_OH_CA) and STO_OH of NO_OF_BS number of BSs under CA
(STO_OH_BS).
STO_OH_CA is due to the maintenance of NO_OF_BS number of CRLB in QC.
P
Hence STO_OH_CA = NO OF BS
NEVIN_CRLB 9 Size_E_VIN bits.bits.
PNO OFB¼1BS
STO_OH_BS is B¼1 ðSTO OHB Þ bits where STO_OHB is STO_OH of
BSB.
STO_OHB is due to the maintenance of QB and NO_OF_VB number of Qv.
Now Q has NO_OF_VB number of CRLs and hence size of QB (Size_QB) is
PNO OF BVB
v¼1 ðNEVIN CRLv Þ  Size E VIN bits.
Qv has NS_MSGv number of S_MSGs and hence the size of Qv (Size_Qv) is
PNS MSGv
j¼1 Size SMSGj bits, where Size_SMSGj is the size of S_MSGj.
P
Hence STO OHB ¼ Size QB þ NO v¼1
OF VB
ðSize Qv Þ bits.
Computation Overhead. The COMP_OH of the proposed scheme is the sum of
COMP_OH of CA (COMP_OH_CA) and COMP_OH of NO_OF_BS number of
BSs under CA (COMP_OH_BS).
COMP_OH_CA is for updating NC_CRL and for performing union operation
among the received CRLs from NO_OF_BS number of BSs.
Now COMP_OH of updating NC_CRL for NO_OF_BS times is
O(NO_OF_BS).
774 A. Mondal and S. Mitra

The COMP_OH of performing union operation among NO_OF_BS number of

CRLs is O(dlog (NO_OF_BS)e).
P 
NO OF BS
COMP_OH_BS is O B¼1 ðCOMP OH B Þ where COMP_OHB is the
sum of COMP_OH of BSB and NO_OF_VB number of vehicles under it.
COMP_OH of BSB is for updating NB_CRL and for performing union opera-
tion among the received CRLs from NO_OF_VB number of vehicles.
Now COMP_OH of updating NB_CRL for NO_OF_VB times is O(NO_
OF_VB) and of performing union operation among NO_OF_VB number of CRLs
is O(dlog (NO_OF_VB)e).
The COMP_OH of NO_OF_VB number of vehicles is the sum of searching
overhead of CA_CRL for E_VIN, execution overhead of ALGO_MESS and
ALGO_V.
Now COMP_OH of Vv for searching CA_CRL for NS_MSGv number of
E_VINs corresponding to NS_MSGv number of received S_MSGs in Qv is O(NS_
MSGv 9 NEVIN_CA_CRL), for executing ALGO_MESS algorithm for NS_
MSGv number of S_MSGs is O(Number_BWO 9 Size_H_MSG 9 NS_MSGv)
and for executing ALGO_V algorithm for NS_MSGv number of records in DTv is
O(NS_MSG2v).
Hence COMP_OH of NO_OF_VB number of vehicles under BSB is
P
Oð NO OF VB
ðððNS MSGv Þ  NEVIN CA CRLÞ þ ðNumber BWO  Size H
v¼1  
MSG  NS MSGv Þ þ NS MSG2v ÞÞ.
Figures 72.1, 72.2 and 72.3 show the plot of COMM_OH, STO_OH and
COMP_OH versus number of vehicles in VANET. STO_OH of the present
scheme is less than that in [12] as observed from Fig. 72.2. The number of vehicles
per BS increases with the number of vehicles in VANET which in turn increases
COMM_OH, STO_OH and COMP_OH as observed from Figs. 72.1, 72.2 and
72.3 respectively.

72.3.3 Quantitative Analysis

The performance of the proposed scheme is studied quantitatively on the basis of

CA_CRL distribution time (CA_CRL_DT) and it is compared with [3, 4]. The
CA_CRL_DT is the time which is required to broadcast CA_CRL by CA among the
vehicles within its coverage area. It is determined during simulation. The quanti-
tative performance of the proposed scheme is also studied on the basis of delay in
detection of misbehaving vehicles (Delay_MV). Delay_MV for Vv (Delay_MVv) is
computed as the sum of waiting time of S_MSGs in Qv, searching time of CA_CRL
for E_VINs corresponding to the received S_MSGs in Qv, time to execute
P
ALGO_MESS and ALGO_V. So, Delay MV ¼ NO v¼1
OF VB
Delay MVv .
72 Detection and Revocation of Misbehaving Vehicles from VANET 775

Fig. 72.1 COMM_OH 3.5

COMM_OH(sec)
versus number of vehicles 3
2.5
2
1.5
1
0.5
0
10 20 50 100 150
Number of vehicles

Figure 72.4 shows the plot of CA_CRL_DT versus Size_CA_CRL. The

Size_CA_CRL is the size of CA_CRL and it is computed as NE-
VIN_CA_CRL 9 Size_E_VIN bits. It can be observed from Fig. 72.4 that
CA_CRL_DT increases with Size_CA_CRL both in the present scheme and in [4].
In [4] Size_CA_CRL is 1 Mbyte and CA_CRL_DT is 300 s. The curve is a
straight line as Size_CA_CRL is constant to 1 Mbyte. In the present work
Size_CA_CRL varies dynamically with NEVIN_CA_CRL and hence
CA_CRL_DT increases slowly with Size_CA_CRL. In the present work
CA_CRL_DT is about 1 s.
Figure 72.5 shows the plot of CA_CRL_DT versus number of vehicles in
VANET. CA_CRL_DT depends upon the size of CA_CRL which in turn depends
upon the number of misbehaving vehicles in VANET. Hence CA_CRL_DT

250000
STO_OH(bit)

200000
150000
100000
50000
0
50 100 150 200 250 300 350 400
Number of vehicles
[12] Present Scheme

Fig. 72.2 STO_OH versus number of vehicles

3.5
3
COMP_OH

2.5
2
1.5
1
0.5
0
10 25 50 75 100 125 150
Number of vehicles

Fig. 72.3 COMP_OH versus number of vehicles

776 A. Mondal and S. Mitra

CA_CRL_DT (sec)
Fig. 72.4 CA_CRL_DT 400
versus Size_CA_CRL
300
200
100
0
136 256 1024 8192
Size_CA_CRL (bit)
[4] Present Scheme

Fig. 72.5 CA_CRL_DT 3500

versus number of vehicles 3000
CA_CRL_DT (sec) 2500
2000
1500
1000
500
0
100 150 200 250 300 350 400
Number of vehicles
[3] Present Scheme

Fig. 72.6 Delay_MV versus 600

Delay_MV (ms)

number of vehicles 500

400
300
200
100
0
50 100 150 200 250 300 350 400
Number of vehicles

increases slowly with the number of vehicles as observed from Fig. 72.5. In [3]
CA_CRL_DT is approximately 3,000 s whereas in the present work CA_CRL_DT
is almost 3 s.
Figure 72.6 shows the plot of Delay_MV versus number of vehicles in VANET.
It can be observed from Fig. 72.6 that Delay_MV increases with the number of
vehicles as per its definition.
72 Detection and Revocation of Misbehaving Vehicles from VANET 777

72.4 Conclusion

Each vehicle detects and revokes misbehaving vehicles from VANET during V2V
communication. The CA_CRL is created by CA after collecting the CRLs from the
BSs within its coverage area. Each BS creates the CRL after collecting the CRLs
from the vehicles within its coverage area.
The CA may verify the revocation decision of the vehicles before creating
CA_CRL. The performance of the proposed scheme may be studied by varying the
value of the other parameters and by incorporating the fuzzy logic concept.

References

1. Aslam, B., Zou, C.C.: Distributed certificate architecture for VANET. Sigcomm (2009)
2. Papadimitratos, P., Buttyan, L., Holczer, T., Schoch, E., Freudiger, J.: Secure vehicular
communication systems: design and architecture. IEEE Commun. Mag. 46, 100–109 (2008)
3. Kargl, F., Papadimitrators, P., Buttyan, L., Muter, M., Wiedersheim, B., Schoch, E., Thong,
T.V., Calandriello, G., Held, A., Kung, A., Hubaux, J.P.: Secure vehicular communication
systems: implementation, performance and research challenges. IEEE Commun. Mag. 46,
110–118 (2008)
4. Nowatkowski, M.E., Owen, H.L.: Scalable certificate revocation list distribution in vehicular
ad hoc networks. IEEE Globecom, Workshop on Seamless Wireless Mobility (2010)
5. Papadimitratos, P., Mezzour, G., Hubaux, J.P.: Certificate revocation list distribution in
vehicular communication systems. Proceedings of the Fifth ACM International Workshop on
Vehicular Internetworking, pp. 86–87 (2008)
6. Samara, G., Al-Salihy, W.A.H., Sures, R.: Efficient certificate management in VANET.
Second International Conference on Future Computer and Communication, vol. 3,
pp. 750–754 (2010)
7. Boutaba, R., Aib, I.: Policy-based management: a historical perspective. J. Netw. Syst.
Manage. 15, 447–480 (2007)
8. Duque, O.F.G., Hadjiantonis, A.M., Pavlou, G., Howarth M.: adaptable misbehavior
detection and isolation in wireless ad hoc networks using policies. IFIP/IEEE International
Symposium on Integrated Network Management (2009)
9. Mondal, A., Mitra, S.: Identification, authentication and tracking algorithm for vehicles using
VIN in centralized VANET. International Conference on Advances in Communication,
Network, and Computing, Springer LNICST, vol. 108 (2012)
10. Kherani, A., Rao, A.: Performance of node-eviction schemes in vehicular networks. IEEE
Trans. Veh. Technol. 59, 550–558 (2010)
11. Towards Effective Vehicle Identification: The NMVTRC’s strategic framework for
improving the identification of vehicles and components (2004)
12. Huang, J.L., Yeh, L.Y., Chien, H.Y.: ABAKA: an anonymous batch authenticated and key
agreement scheme for value-added services in vehicular ad hoc networks. IEEE Trans. Veh.
Technol. 60, 248–262 (2011)
Chapter 73
A Novel Steganalysis Method Based
on Histogram Analysis

Bismita Choudhury, Rig Das and Arup Baruah

Abstract Steganalysis is the art of detecting hidden messages embedded inside

Steganographic Images. Steganalysis involves detection of steganography, esti-
mation of message length and its extraction. Recently Steganalysis receives great
deal of attention from the researchers due to the evolution of new, advanced and
much secured steganographic methods for communicating secret information. This
paper presents a universal steganalysis method for blocking recent steganographic
techniques in spatial domain. The novel method analyses histograms of both the
cover and suspicious image and based on the histogram difference it gives decision
on the suspicious image of being stego or normal image. This method for steg-
analysis extracts a special pattern from the histogram difference of the cover and .
By finding that specific pattern from the histogram difference of the suspicious and
cover image it detects the presence of hidden message. The proposed steganalysis
method has been experimented on a set of stego images where different stega-
nographic techniques are used and it successfully detects all those stego images.

Keywords Steganalysis
Steganography
Histogram
PSNR

B. Choudhury (&)
A. Baruah

Department of Computer Science and Engineering and Information Technology,
Don Bosco College of Engineering and Technology, Guwahati 781017, Assam, India
e-mail: [email protected]
A. Baruah
e-mail: [email protected]
R. Das
Department of Computer Science and Engineering, National Institute of Technology,
Rourkela 769008, Orissa, India
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 779

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_73
780 B. Choudhury et al.

73.1 Introduction

The battle between Steganography and Steganalysis never ends. For hiding secret
message or information, Steganography provides a very secure way by embedding
them in unsuspicious cover media such as image, text or video [1]. As a counter
action Steganalysis is emerging out as a process of detection of steganography.
Steganalysis refers to the science of discrimination between stego-object and
cover-object. Steganalysis detects the presence of hidden information without
having any knowledge of secret key or algorithm used for embedding the secret
message into the cover image [2].
In the general process of steganalysis, steganalyzer simply blocks the stego
image and sometimes try to extract the hidden message. Figure 73.1 shows the
block diagram of the generic steganalysis process. Generally, Steganalysis tech-
niques are classified into two broad categories: specific and universal blind steg-
analysis. The targeted steganalysis process is designed for some specific
steganographic methods where all features of that particular steganographic
method are well known. On the other hand, universal blind steganalysis process
uses combination of features to detect arbitrary steganographic methods [3, 4].
Steganalysis can be achieved by applying various image processing techniques
like image filtering, rotating, cropping etc. Also it can be achieved by coding a
program that examines the stego-image structure and measures its statistical
properties, e.g., first order statistics (histograms) or second order statistics (cor-
relations between pixels, distance, direction) [4].
This paper, presents a novel steganalysis method which uses histogram dif-
ference for detection of steganography in spatial domain. Here a special pattern in
the histogram difference of suspicious image and cover image is utilized for the
detection purpose.
This paper is organized as follows. Section 73.2 reviews some previous work
done in steganalysis. The proposed novel steganalysis method is explained in
Sect. 73.3. Simulation and results are shown in Sects. 73.4 and 73.5 concludes.

73.2 Related Work

Many research works have been carried out on steganalysis till now. Based on the
domain of message embedding (Spatial or Frequency domain) different methods
are employed to detect presence of steganography. Some of them are as follows.

73.2.1 RS Steganalysis

Fridrich et al. described a reliable and accurate method for detecting Least Sig-
nificant Bit (LSB) based steganography [5]. For performing RS Steganalysis they
73 A Novel Steganalysis Method Based on Histogram Analysis 781

Fig. 73.1 Block diagram of Steganalysis

divided the image pixels into three groups—Regular, Singular and Unchanged
group. In normal image number of regular groups is greater than that of singular
group. But after embedding any data in the image, Regular and Singular group of
pixels have a tendency of becoming equal. Based on this characteristic they
proposed RS steganalysis technique for attacking steganography. Here detection is
more accurate for messages that are randomly scattered in the stego-image than for
messages concentrated in a localized area of the image.

73.2.2 Breaking F5 Algorithm

Fridrich et al. presented a steganalysis method to reliably detect messages (and

estimate their size) hidden in JPEG images using the steganographic algorithm F5
[6]. The estimation of the cover-image histogram from the stego-image is the key
point. This is done by decompressing the stego-image, cropping it by four pixels in
both directions to remove the quantization in the frequency domain, and recom-
pressing it using the same quality factor as the stego-image. The number of relative
changes introduced by F5 is determined using the least square fit by comparing the
estimated histograms of selected DCT coefficients with those of the stego-image.

73.2.3 Histogram Estimation Scheme for Defeating Pixel

Value Differencing Steganography Using Modulus
Function

In this paper Jeong-Chun Joo Kyung-Su Kim and Heung-Kyu Lee presented a
specific steganalysis method to defeat the modulus Pixel Value Differencing
782 B. Choudhury et al.

(PVD) steganography [7]. By analyzing the embedding process they provided

three blind Support Machines (SMs) for the steganalysis and each are used for
checking three different features. SM1: the fluctuations around the border of the
sub range, SM2: the asymmetry of the stego PVD histogram, and SM3: the
abnormal increase of the histogram value. The Support Vector Machine (SVM)
classifier is applied for the classification of the cover and stego images. Here
Original histogram is estimated from the suspicious image using two novel his-
togram estimation schemes (HES): a curve-fitting method and a histogram reverse-
tracing method those work without the cover image.

73.2.4 Steganalysis by Subtractive Pixel Adjacency Matrix

Tomas Pevny and Patrick Bas and Jessica Fridrich presented a method for detection
of steganographic method LSB matching [8]. By modeling the differences between
adjacent pixels in natural images, the method identifies some deviations those occur
due to steganographic embedding. For steganalysis a filter is used for suppressing
the image content and exposing the stego noise. Dependences between neighboring
pixels of the filtered image are modeled as a higher-order Markov chain. The sample
transition probability matrix is then used as a vector feature for a feature-based
steganalyzer implemented using machine learning algorithms.

73.3 A Novel Method for Steganalysis Using Histogram

Analysis

In this paper we proposed a novel steganalysis technique for detection of stega-

nography in spatial domain based on the histogram analysis of the cover and the
suspicious image. The schematic diagram of the whole process is given in Fig. 73.2.
The main goal in here is to develop a steganalysis method which is able to
block most of the recently developed steganographic algorithms with a good
accuracy. The novel algorithm first finds the histograms of both the cover and
suspicious image. Then it uses difference values of both the histograms to detect
the stego-image.

73.3.1 Histogram Difference

Image histogram proves to be one of a good feature for analyzing the difference
between cover image and stego image. In general, histograms of cover image and
stego image have some significant differences that help in discriminating between
73 A Novel Steganalysis Method Based on Histogram Analysis 783

Fig. 73.2 Block diagram of proposed steganalysis method

cover and stego image. In steganography, while embedding secret data in a cover
image by modifying the Least Significant Bits (LSBs) of the cover image, some of
the pixel values of the cover image get changed and thereby the histogram of the
stego image acquires some variations from that of the cover image. If we find the
histogram difference of both the cover and stego image we can observe that some
of the difference values possess same magnitude to their adjacent values but of
different signs (For e.g. 2, -2; -35, 35; … etc.). But this kind of pattern is not
found in the histogram difference between cover and noisy image or any processed
image.
The Table 73.1 shows the histogram difference values of the cover image with
stego image (LSB embedding) and noisy image introduced with Gaussian noise
tested on the Lenna image. From the table we can see that the most of the adjacent
difference values are having same magnitude but with different sign only in case of
stego image (For e.g. -2, 2; -48, 48; -132, 132), but not in case of noisy image. In
this way the steganalysis method tries to find out such pairs in the histogram
difference of the cover and the suspicious image and based on this characteristic
stego images are detected.
784 B. Choudhury et al.

Table 73.1 Histogram Histogram difference of Histogram difference of cover and

difference of cover image cover and stego image noisy (Gaussian noise) image
with stego image and
noisy image -2 -37,079
2 -2,101
-9 -2,180
9 -2,204
-48 -2,179
48 -2,048
-58 -1,747
58 -1,662
-152 -1,454
152 -1,088
-132 -711
132 -383
-266 120
266 601

73.3.2 Proposed Novel Algorithm for Steganalysis

Algorithm
Input: M 9 N Suspicious Image and M 9 N Cover Image.
Output: Decision whether the Suspicious Image is a Stego Image or not.
Step-1: Read both the Cover and Suspicious Image and store their intensity
values of different pixels in two different arrays.
Step-2: Find histograms of both the Cover and Suspicious Image.
Step-3: Plot both the histograms in a single plot and find the difference.
Step-4: In the different values, if there are adjacent values those are same in
magnitude but different in sign then increment a counter.
Step-5: Repeat Step 4 until all the difference values are checked and the counter
incremented accordingly.
Step-6: Set a threshold value of the counter and if the counter value goes
beyond the threshold value then detect the Suspicious Image as the
Stego Image else as the Normal Image.
Step-7: End.

73.4 Simulation and Results

Some experiments are carried out to check the capability and efficiency of the
novel steganalysis process. This method is capable of detecting stego image where
most of the newly developed steganographic algorithms are used. The proposed
73 A Novel Steganalysis Method Based on Histogram Analysis 785

Fig. 73.3 a–d Four cover images for training, e Secret image/message. a Lenna, b Baboon, c
Airplane, d Boat, e Cameraman

Fig. 73.4 a Histogram of cover image of Lenna, b Histogram of stego image using LSB
replacement, c Histogram difference of cover and stego image

steganalysis algorithm is tested on six steganographic algorithms in spatial

domain, viz. Least Significant Bit (LSB) replacement, LSB matching, Steganog-
raphy based on Huffman Encoding, Wavelet Obtained Weight (WOW), Universal
Wavelet Relative Distortion for spatial domain (S_UNIWARD) and HUGO.
For the testing purpose, all the simulation has been done in MATLAB 2012 on
Windows 7 platform. A set of 8-bit grayscale images of size 1024 9 1024 are used
as cover-image and image of size 256 9 256 are used as the secret image to form
the stego-image. Figure 73.3a–d shows the four original cover images (Here test
results are shown only for Lenna Image) and Fig. 73.3e shows the secret image
used to embed using LSB replacement [8], LSB matching [8] and Steganography
based on Huffman Encoding [9]. For the steganographic algorithms S_UNIWARD
[10], WOW [10] and HUGO [10] randomly generated message bits are used to
create stego-image. The histogram of the cover image is used to compare with the
histogram of the stego image created for testing the proposed steganalysis method.
The novel steganalysis algorithm successfully detects the stego-image by ana-
lyzing the histogram difference of both suspicious and cover image.
Figure 73.4a shows the histogram of Lenna image, Fig. 73.4b shows histogram
of Lenna image after using LSB replacement steganography in which LSBs of
786 B. Choudhury et al.

Fig. 73.5 a Histogram of cover image of Lenna, b Histogram of stego image using LSB
matching, c Histogram difference of cover and stego image

Fig. 73.6 a Histogram of cover image of Lenna, b Histogram of stego image created by
steganography based on Huffman encoding, c Histogram difference of cover and stego image

individual cover elements are replaced with message bits [8], Fig. 73.4c shows
histogram difference of the cover and the stego image.
Figure 73.5a shows the histogram of Lenna image, Fig. 73.5b the histogram of
Lenna image after using LSB matching steganography which randomly increases
or decreases pixel values by one to match the LSBs with the communicated
message bits [8], Fig. 73.5c shows histogram difference of cover and stego image.
The recent Steganographic method based on Huffman encoding proposed by Das
and Tuithung [9] is also a very much secured method and very few specific
73 A Novel Steganalysis Method Based on Histogram Analysis 787

Fig. 73.7 a Histogram of cover image of Lenna, b Histogram of stego image created using
S_UNIWARD method, c Histogram difference of cover and stego image

Fig. 73.8 a Histogram of cover image of Lenna, b Histogram of stego image using
steganographic method WOW, c Histogram difference of cover and stego image

patterns can be observed in the histogram difference. However, our proposed

steganalysis algorithm is able to block it (Fig. 73.6a–c).
Three very recent and secure steganographic algorithms S_UNIWARD [11]
(Fig. 73.7a–c), WOW [12] (Fig. 73.8a–c) and HUGO [13] (Fig. 73.9a–c), pro-
posed by Fridrich et al., make a few modifications in the cover image to embed
randomly generated message bits. The novel steganalysis method successfully
detects those stego images even though they possess few artifacts.
788 B. Choudhury et al.

Fig. 73.9 a Histogram of cover image of Lenna, b Histogram of stego image using
steganographic method HUGO, c Histogram difference of cover and stego image

Table 73.2 PSNR between the cover and the stego image
Steganographic algorithms PSNR value between the cover
and the stego image (dB)
LSB embedding +56.88
LSB matching +56.88
Steganography based on Huffman encoding +57.43
WOW +62.69
S_UNIWARD +62.21
HUGO +61.92

From the Peak Signal to Noise Ratio (PSNR) values, shown in Table 73.2, it
can be seen that the most of the used steganographic methods have done less
modification to the cover image which is very difficult to get noticed. However, the
proposed steganalysis method successfully blocks the stego images where these
steganographic techniques are applied.

73.5 Conclusion

In this paper, we have proposed a universal steganalysis method that checks the
histogram difference of the suspicious image with that of the cover image to get
adjacent difference values having same magnitude but of different sign. This method
73 A Novel Steganalysis Method Based on Histogram Analysis 789

has a great capability of detecting stego images even though very small changes are
done in the cover image. Experimental results show that it can block from generic
LSB modification techniques to much secured recent steganographic methods. The
PSNR values, shown in the Table 73.2, for tested stego images using different
steganographic methods depicts that the tested steganographic methods are efficient
methods.
Most of the steganalysis algorithms are targeted methods to attack specific
steganographic techniques. So in the small group of the universal blind steg-
analysis this novel algorithm provides a new addition. In future we will work on
the steganalysis of the steganography in frequency domain. Then we would like to
develop a universal steganalysis method to detect stego images irrespective of the
data embedding domain.

References

1. Johnson, F.N., Jajodia, S.: Exploring steganography: seeing the unseen. IEEE Computer
Society Press. 31(2), 26–34 (1998)
2. Fridrich, J., Goljan, M.: Practical steganalysis of digital images—state of the art. In:
Proceedings of Electronic Imaging, SPIE, vol. 4675, pp. 1–13 (2002)
3. Lou, D.C., Hu, C.H., Chiu, C.C.: Steganalysis of histogram modification reversible data
hiding scheme by histogram feature Coding. Int. J. Innov. Comput. Inf. Control 7, 11 (2011)
4. Cheddad, A., Condell, J., Curran, K., Kevitt, M.P.: Digital image steganography: survey and
analysis of current methods. Elsevier Signal Process. 90, 727–752 (2010)
5. Fridrich, J., Goljan, M., Du, R.: Reliable detection of LSB steganography in grayscale and
color images. In: Proceedings of ACM, Special Session on Multimedia Security and
Watermarking, Ottawa, Canada, October 5 (2001)
6. Fridrich, J., Goljan, M., Hogea, D.: Steganalysis of jpeg images: breaking the F5 algorithm.
In: Proceedings of the 5th Information Hiding Workshop, Springer, vol. 2578, pp. 310–323
(2002)
7. Joo, C.J., Kim, S.K., Lee, K.H.: Histogram estimation-scheme-based steganalysis defeating
the steganography using pixel-value differencing and modulus function. Opt. Eng. 49,
077001 (2010)
8. Pevny, T., Ba, P., Fridrich, J.: Steganalysis by subtractive pixel adjacency matrix. In: ACM
Multimedia and Security Workshop, Princeton, NJ, September 7–8, pp. 75–84 (2009)
9. Das, R., Tuithung, T.: A novel steganography method for image based on huffman encoding.
In: 3rd IEEE National Conference on Emerging Trends and Applications in Computer
Science (NCETACS—2012), pp. 14–18 (2012)
10. Steganography codes for windows. http://dde.binghamton.edu/download/stego_algorithms/
11. Holub, V., Fridrich, J.: Digital image steganography using universal distortion. In: ACM
Workshop on Information Hiding and Multimedia Security, June (2013)
12. Holub, V., Fridrich, J.: Designing steganographic distortion using directional filters. In: IEEE
Workshop on Information Forensic and Security (WIFS), Tenerife, Spain, December (2012)
13. Filler, T., Fridrich, J.: Gibbs construction in steganography. IEEE Trans. Inf. Forensics
and Security. 5(4), 705–720 (2010)
Chapter 74
Pattern Recognition Techniques: Studies
on Appropriate Classifications

Sasan Karamizadeh, Shahidan M. Abdullah, Mazdak Zamani

and Atabak Kherikhah

Abstract Pattern recognition techniques are divided into categories of supervised,

unsupervised and semi supervised. Supervised pattern recognition methods are
utilized in the examination of various sources’ chemical data such as sensor
measurements, spectroscopy, and chromatography. The unsupervised classification
techniques use algorithms to classify and analyze huge amounts of raster cells.
Semi-Supervised Learning is an approach that is in the middle ground between
supervised and unsupervised learning and guarantees to be better at classification
by involving data that is unlabeled. In this paper, we tried to categories pattern
recognition methods and explain about each of them and we compared supervised
method with unsupervised method in terms of types and location of features.

74.1 Introduction

Pattern recognition techniques are divided into categories of supervised, unsu-

pervised and semi supervised. This is dependent on the analyst’s intention of the
information that needs to be utilized or that is available regarding the samples
comprising of the data matrix. In the supervised methods, or the classification
method, prior description is made on the classes as the concept or the attribute used

S. Karamizadeh
S.M. Abdullah (&)
M. Zamani
A. Kherikhah

Advanced Informatics School (AIS), Universiti Teknologi Malaysia,
54100 Kuala Lumpur, Malaysia
e-mail: [email protected]
S. Karamizadeh
e-mail: [email protected]
M. Zamani
e-mail: [email protected]
A. Kherikhah
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 791

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_74
792 S. Karamizadeh et al.

to classify the samples into subsets are already known [1]. In the unsupervised
method, the classification is removed by considering only the variations and
resemblances among the samples, without utilizing any of their details. The semi-
supervised method is in the middle ground between the supervised and unsuper-
vised analysis and assures to be a better classification using the non-labeled details
[2]. Figure 74.1 Classification of Pattern-Recognition Techniques.

74.2 Supervised Methods

Supervised pattern recognition methods are utilized in the examination of various

sources’ chemical data such as sensor measurements, spectroscopy, and chroma-
tography. Various supervised techniques exist which have been widely utilized in
the analytical chemistry [2]. In all the cases, the most suitable method is reliant on
the problem that needs to be addressed since the criteria and bases of the tech-
niques differ significantly according to the problems faced. As revealed in
Fig. 74.1, different criteria can be utilized to apply the supervised methods. Sev-
eral of the common methods are elaborated below.

74.2.1 Parametric and Non-parametric Methods

Metric methods utilize the mathematical models that have parameters that can be
adjusted to perform classification of samples. These methods involve Soft Inde-
pendent Modeling of Class Analogy (SIMCA), Linear Discriminate Analysis
(LDA), Discriminate Analysis (DA), and Support Victor Machine (SVM) [3].
Non-parametric techniques do not utilize the parameters according to the
mathematical model for sample classifications. Some of the popularly utilized non-
parametric techniques include Artificial Neural Networks (ANN), and k-Nearest
Neighbors (kNN) [3].

74.2.2 Discriminate and Class Modeling Assessment

Supervised pattern recognition methods differentiate the variables’ hyperspace that

distinguish the samples into various classifications. Utilizing the discriminant
methods, when a new sample is put into the hyperspace classifications, it is
identified with that classification, however, when it is put outside, this does not
happen. There is a lack of an in-between or middle ground [4]. The techniques in
use here are KNN, LDA, ANN, and DA. The analysis on class modeling considers
the samples that fit the model as part of the class, whereas rejected non-members
are the objects that do not fit. In the event of modeling more than one class, three
74 Pattern Recognition Techniques: Studies on Appropriate Classifications 793

Method basses on factor models

Artificial Neural Networks

Unsupervised
Technique
Cluster analysis

Hybrid

Generative

Self-training

Co training and multi view

Pattern Semi
Recognition Supervised Graph-Based Methods

Avoiding Changes in Dense

Hybrid

Deterministic/Probabilistic

Discriminant/Class modeling

Supervised Parametric

Non parametric

Hybrid

Fig. 74.1 Classification of pattern-recognition techniques

various circumstances can be identified; for example every sample can be desig-
nated into a single classification, or more than one classification or none of the
classifications [1].

74.2.3 Deterministic/Probabilistic Methods

No statement is made regarding the reliability of the decision when a deterministic

system is utilized to designate each sample’s class. Probabilistic methods, how-
ever, do measure the classification’s reliability. Deterministic methods are namely
KNN, and the probabilistic methods are namely DA, LDA and ANN [1].
794 S. Karamizadeh et al.

74.3 Unsupervised Methods

The unsupervised classification techniques use algorithms to classify and analyze

huge amounts of raster cells. These procedures need set values for several of the
operating parameters; however, the classifying method goes on without any
intervention from the users. The efficacy of the unsupervised techniques is
dependent on the basis that the input raster dataset has natural statistical groups of
spectral patterns that consist of specific forms of physical characteristics [1]. The
entire unsupervised classification techniques, aside from the Simple One-Pass
Clustering, utilize the interactive procedure to examine a set of sample input cells
and decide on the set of class centers and other related statistical features [5]. All
the input raster set is then analyzed, and a classification rule is utilized to designate
each raster cell to a defined class. The techniques that are normally utilized can be
classified into four significant groups as revealed in Fig. 74.1.

74.3.1 Cluster Analysis

Cluster Analysis or CA has been the most commonly utilized used method of
pattern recognition until several years ago. This particular method designates
samples to similar clusters based on the level of similarity in the variables
(characteristics) which have been utilized to identify the objects, and, simulta-
neously designate samples that are dissimilar to various other clusters. It is
commonly utilized to design a new category of samples being studied study,
however, it can also be used to confirm an existing group. Researchers have
introduced a complete monograph of cluster analysis based on analytical chemical
data [6].

74.3.2 Artificial Neural Networks

ANNs are mathematical methods that follow the workings of the nervous system
in humans, by making up pattern recognition models. They are normally very
successful in addressing the challenges often faced during the process of classi-
fication. ANNs begin from a data training set, that contains characteristics such as
spectra or concentration levels that measure samples that are indifferent, to mea-
sure probabilities of samples that are a class member (output variables) [3]. ANNs
are utilized in both supervised and unsupervised pattern identification, however,
since their usage not as simple as CA, their usages are limited somewhat [4].
74 Pattern Recognition Techniques: Studies on Appropriate Classifications 795

74.3.3 Techniques According to Factor Models

The aim of these techniques are to constrain the n-dimensional information about
objects to a limited and more inclusive aspect. In this way, all the samples can be
depicted graphically in a two or three-dimensional (2D or 3D) arena, simplifying
identification of the major characteristics. Principal Component Analysis (PCA)
are some of the commonly used methods that operate the 2D data tables and multi-
set methods [5].

74.4 Semi Supervised

Semi-Supervised Learning (SSL) is an approach that is in the middle ground

between supervised and unsupervised learning and guarantees to be better at
classification by involving data that is unlabeled. Since getting labeled data is
costly and complicated, by causing unlabeled data to be less expensive to get in
many applications [6], SSL tries to gain better classification function by utilizing
both unlabeled and labeled data. The self-training technique is one of the first
algorithms suggested for utilizing the unlabeled data. Another two significant
methods include the transductive S3VM and co-training.

74.4.1 Generative

One of the oldest semi-supervised learning techniques is the generative models.

The model’s assumption is that p(x, y) = p(y) p (x|y) where p(x|y) is a distribution
of identifiable mixture; for instance the Gaussian mixture models. Having a huge
sum of unlabeled data, the mixture elements can be recognized; and typically, only
one labeled example per element is needed to completely decide on the distri-
bution of the mixture [7].

74.4.2 Self-Training

Self-training or decision-directed or self-labeling learning, is the easiest and fre-

quently used SSL approach. This wrapper algorithm utilizes the forecasting of a
supervised learning technique to label the unlabeled data. The classifier utilizes its
own forecasts to teach itself. Initially, it begins by training a separated hyper plane
with only the labeled data. At every stage, the algorithm chooses a portion of the
unlabeled samples for labeling, based on the target or a decision task. After that,
the technique adds on these objects to the set of training. Lastly, the classifier
retrains itself and the procedure is repeats once more [8].
796 S. Karamizadeh et al.

The self-learning algorithm is easy and can be utilized as an algorithm for meta-
learning. However, it depends on the goodness- of-fit of the obtained classifier,
taking into consideration that errors tend to strengthen themselves. One other
drawback of self-learning is the complexity of examining it generally, but there
have been several researches on the convergence of particular base learners [9].
Self-training will be used as one of the semi-supervised tactics to develop the
models on credit scoring.

74.4.3 Co-training and Multi View Learning

Co-training techniques depend on three assumptions. Firstly, it is stated that must be

a natural variables split in two of the subsets. Secondly, every subset must be large
enough to train a good classifier. Lastly, the technique presumes that both of the
subsets are conditionally independent considering the class. This technique trains
two various classifiers; one for each subset and uses just the labeled data. After that,
each of the classification tasks categorizes a portion of the unlabeled data and trains
the other classifier. Both classifiers will be retrained using this new labeled data
handed out by the other classifier (cross information) in an iterative method [10].

74.4.4 Graph-Based Methods

Semi-supervised techniques that are graph-based refer to a graph where the nodes
consist of both the labeled and unlabeled samples in the dataset, and edges (may be
weighted) show the samples’ similarity. These techniques normally assume that
there is smoothness of label across the graph. Graph techniques are non-para-
metric, transductive and discriminative [11, 12].

74.4.5 Avoiding Changes in Dense Areas

Discriminative techniques function directly on p(y|x). This causes the danger of

leaving p(x) outside the parameter’s estimation loop, if p(x) and p(y|x) do not have
similar parameters. Normally, p(x) is all that can be retrieved from the unlabeled
data. It is suggested that if p(x) and p(y|x) do not have similar parameters, semi-
supervised learning is unable to assist emphasizes this fact [13].
A method of binary classification that locates the optimal linear decision sur-
face between two classifications is known as the Support Vector Machine. The
decision surface is a weighted mixture of the supported vectors. The SVM in these
utilizations need to be imputed with an individual’s images, which will contain one
class and the other class will contain images of other individuals besides the first
individual. The SVM will then create a linear decision surface [14–16].
74 Pattern Recognition Techniques: Studies on Appropriate Classifications 797

74.5 Hybrid

The hybrid models are defined as the models for credit scoring that have been
developed by integrating two or more existing models. The benefit of these models
is that the creditor can gain from having two or more models aside from reducing
the weakness of the model by combining them with other models. However, these
techniques are difficult to plan and execute in comparison to other methods that are
easier [14] claim that the hybrid method faces faster compared to the traditional
concept of neural networks. Several successful credit scoring prototypes of hybrid
techniques have also been developed in current years [13]. Examine the hybrid
model empirically by implementing two real groups of domain information.

74.6 Comparison Technique Between Supervised,

Unsupervised and Semi Supervised

Unsupervised classification techniques use algorithms to classify and analyze huge

amounts of raster cells. These procedures need set values for several of the
operating parameters; however, the classifying method goes on without any
intervention from the users. The efficacy of the unsupervised techniques is
dependent on the basis that the input raster dataset has natural statistical groups of
spectral patterns that consist of specific forms of physical characteristics. The
entire unsupervised classification techniques, aside from the Simple One-Pass
Clustering, utilize the interactive procedure to examine a set of sample input cells
and decide on the set of class centers and other related statistical features. All the
input raster set is then analyzed, and a classification rule is utilized to designate
each raster cell to a defined class [17].
The supervised methods of classification are carried out according to the user-
defined classes and subsequent representative sample sets. The training raster data
sets specify the sample sets, which must be developed before imputing the
Automatic Classification procedure. The activation of the Training Data button is
carried out when a supervised classification technique is selected, which shows
that selection of training is needed and to set the raster. The Feature Mapping
procedure offers the tools required to develop a training raster, as shown in the
segment known as designing the Training Set Raster. The training sectors are first
examined to decide on the statistical characteristics of each classification. In the
last stage of classification, every cell in the input raster set is designated to each of
the training classes by utilizing a suitable decision rule [17].
Supervised techniques result in superior outcomes when the classification idea
is translated into specific groupings that are represented well by training sections
and suitable for input raster’s. In the Table 74.1 shows advantage and disadvan-
tage of techniques in pattern recognition.
798 S. Karamizadeh et al.

Table 74.1 Advantage and disadvantage of pattern recognition techniques

Advantage Disadvantage Reference
PCA • It is used to reduce the • It is very time consuming [18, 19]
dimension of the data • High order dependencies still
• It gives high accuracy and low exist in PCA analysis
computational cost
• PCA gave better results for
varying poses
BTC • The algorithm is independent of • Larger size of the feature vector [18]
the size of a face image at BTC level 4 compare with other
• Simple image coding technique levels
DCT • DCT is used to reduce image • DCT based features are sensitive [18]
information redundancy to changes in the illumination
• DCT has been implemented in a direction
single integrated circuit because of
input independency
• DCT packing the most
information into the fewest
coefficients for most natural
images, and Minimizing block like
appearance
LDA • More efficient if model correct, • Bias if model is incorrect [20]
borrows strength from p(x)
SVM • Produce very accurate classifiers • SVM is a binary classifier. To do [21]
• Less over fitting, robust to noise a multi-class classification, pair-
• SVM is defined by a convex wise classifications can be used
optimization problems (no local (one class against all others, for all
minima) for which there are classes)
efficient methods • Computationally expensive, thus
runs slow

74.7 Conclusion

In this paper, we have divided pattern recognition techniques in three categories in

order to supervised, unsupervised and semi supervised. We have elaborated each
category and finally we compare supervised and unsupervised and comparing of
methods show that unsupervised method is better than supervised when we do not
have good knowledge of the surface, set of training classes are not involved all
significantly distinctive types of surface materials, and each training area is not
representative of its intended class.

Acknowledgments The work we presented in this paper has been supported by the Universiti
Teknologi Malaysia.
74 Pattern Recognition Techniques: Studies on Appropriate Classifications 799

References

1. Karamizadeh, S., Abdullah, S.M., Manaf, A.A., Zamani, M., Hooman, A.: An overview of
principal component analysis. J. Sig. Inf. Process. 4, 173 (2013)
2. Karamizadeh, S., Abdullah, S.M., Zamani, M.: An overview of holistic face recognition.
IJRCCT 2, 738–741 (2013)
3. Pandya, J.M., Rathod, D., Jadav, J.J.: A survey of face recognition approach. Int. J. Eng. Res.
Appl. (IJERA) 3, 632–635 (2013)
4. Delac, K., Grgic, M.: A survey of biometric recognition methods. In: Proceedings of 46th
International Symposium Electronics in Marine 2004, Elmar 2004, pp. 184–193 (2004)
5. Price, J.R., Gee, T.F.: Face recognition using direct, weighted linear discriminant analysis
and modular subspaces. Pattern Recogn. 38, 209–219 (2005)
6. Zhu, X.: Semi-supervised learning. In: Sammut, C., Webb, G.I. (eds.) Encyclopedia of
Machine Learning, pp. 892–897. Springer, Berlin (2010)
7. Vandewalle, V., Biernacki, C., Celeux, G., Govaert, G.: A predictive deviance criterion for
selecting a generative model in semi-supervised classification. Computat. Stat. Data Anal. 24,
220–236 (2013)
8. Carlson, A., Betteridge, J., Wang, R.C., Hruschka Jr, E.R., Mitchell, T.M.: Coupled semi-
supervised learning for information extraction. In: Proceedings of the Third ACM
International Conference on Web Search and Data Mining, 2010, pp. 101–110 (2010)
9. Rosset, S., Zhu, J., Zou, H., Hastie, T.J.: A method for inferring label sampling mechanisms
in semi-supervised learning. In Advances in Neural Information Processing Systems, 2004,
pp. 1161–1168 (2004)
10. Zhou, Z.-H., Li, M.: Semi-supervised regression with co-training. In: IJCAI, 2005,
pp. 908–916 (2005)
11. Narayanan, H., Belkin, M., Niyogi, P.: On the relation between low density separation,
spectral clustering and graph cuts. In: Advances in Neural Information Processing Systems,
2006, pp. 1025–1032 (2006)
12. Abokhdair, N.O., Manaf, A.A., Zamani, M.: Integration of chaotic map and confusion
technique for color medical image encryption. In: International Conference on Digital
Content, Multimedia Technology and Its Applications, 16 August 2010, pp. 20–23. Korea
(2010)
13. Seeger, M.: Learning with labeled and unlabeled data. Technical report, University of
Edinburgh (2001)
14. Pradhan, A.: SUPPORT VECTOR MACHINE-A Survey (2012)
15. Abdullah, S.M., Manaf, A.A., Zamani, M.: Capacity and quality improvement in reversible
image watermarking approach. In: IEEE Networked Computing and Advanced Information
Management (NCM), pp. 81–85 (2010)
16. Abdullah, S.M., Manaf, A.A.: Multiple layer reversible images watermarking using
enhancement of difference expansion techniques. In: Networked Digital Technologies,
pp. 333–342. Springer, Berlin, Heidelberg (2010)
17. Long, F., Xu, H.: Data mining technique of Acoustic Emission signals under supervised and
unsupervised mode. In: 2011 Seventh International Conference on Natural Computation
(ICNC), 2011, pp. 752–755 (2011)
18. Sonal, G., Tanuja, S.S.: A review of feature extraction techniques BTC, DCT, walsh and PCA
with FDM and BDM for face recognition (2013)
19. Bhatt, B.G., Shah, Z.H.: Face feature extraction techniques: a survey. In: National
Conference on Recent Trends in Engineering & Technology, 2011, pp. 13–14. B.V.M.
Engineering College, V.V.Nagar, Gujarat, India (2011)
20. Hastie, T., Tibshirani, R., Friedman, J.: Linear methods for classification. In: The Elements of
Statistical Learning, pp. 101–137. Springer, New York (2009)
21. Bhavsar, H., Ganatra, A.: Variations of support vector machine classification technique: a
survey. Int. J. Adv. Comput. Res. 2 (2012)
Chapter 75
Environmental Noise Analysis for Robust
Automatic Speech Recognition

N. Sai Bala Kishore, M. Rao Venkata and M. Nagamani

Abstract Most of the speech communication applications viz. telephony, hands-

free communication, voice recording, automatic speech recognition, interactive
voice response system, human-machine interfaces, etc. that require at least one
microphone, desired speech signal is usually contaminated by background noise
and reverberation. As a result, the speech signal has to be ‘‘cleaned’’ with digital
signal processing tools before it is played out, transmitted, or stored. The noise
estimation and reduction techniques will help to clean and attenuate the noise
component in speech data, known as Speech Enhancement. In this paper, we
recorded the speech in different environmental conditions and estimated the noise
signal/background noise distribution in speech. Now the speech is enhanced by
using the compliment of Weiner-Hopf optimal filter. And this enhanced speech
signal is given for training and testing the Automatic Speech Recognition (ASR)
system, which will improve the word accuracy. The analysis of speech and results
presented in this paper are produced using MATLAB.


Keywords Speech recognition Speech enhancement  Noise filtering  Auto-
matic speech recognition (ASR)

N.S.B. Kishore  M.R. Venkata (&)

Vignan University, Vadlamudi, India
e-mail: [email protected]
N.S.B. Kishore
e-mail: [email protected]
M. Nagamani
University of Hyderabad, Hyderbad, India
e-mail: [email protected]

Ó Springer International Publishing Switzerland 2015 801

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_75
802 N.S.B. Kishore et al.

75.1 Introduction

Speech Enhancement [1] means the improvement in the value or quality of speech
signal. When applied to speech, this simply means the improvement in intelligi-
bility and/or quality of a degraded speech signal [2] by using speech processing
technique. Speech enhancement is a very difficult problem for two reasons. First
the nature and characteristics of the noise signals can change dramatically in time
and application to application. It is therefore laborious to find versatile algorithms
that really work in different practical environments. Second the performance
measure can also be defined differently for each application. Two perceptual
criteria are widely used to measure the performance: quality and intelligibility.
While the former is subjective (it reflects individual preferences of listeners), the
latter is objective (it gives the percentage of words that could be correctly iden-
tified by listeners). It is very hard to satisfy both at the same time. In fact, it can
easily be shown that in the single-channel (one microphone) case and when the
degradation is due to the uncorrelated additive noise, noise reduction (quality
improvement) is possible at the expense of speech distortion (intelligibility
reduction).

75.2 Noise Analysis and Estimation

Environmental robustness [3] is an important area of research in speech recog-

nition. Mismatch between trained speech models and actual speech to be recog-
nized is due to factors like background noise mingled [4] with unvoiced sound. It
can cause severe degradation of performance in the context of speech recognition.
So analysis and estimating of noise plays a vital in speech recognition perfor-
mance. Here we estimate noise to signal [5] to the speech recorded wave and do
noise reduction from speech wave and this noise reduced wave can be given as
input to the speech recognition system. And this will improvise the accuracy of
speech recognition of recorded speech. In the application perspective we use this
in mobile transmission channel to hear speech without any noise.
The optimal estimate of the clean speech can be achieved by optimizing some
criterion, such as the mean-squared error (MSE) [6] between the clean speech and
its estimate, the signal-to-noise ratio (SNR), the a posteriori probability of the
clean speech given its noisy observations, etc.

75.3 Noise Reduction

The below diagram is the general block diagram of noise reduction system
(Fig. 75.1).
75 Environmental Noise Analysis for Robust Automatic Speech Recognition 803

Fig. 75.1 General block diagram of noise reduction system

Additive Noise n[k]

Microphone Weiner –
signal u[k] hopf
Speech
Signal + Optimal
S[k] Filter
h[k]

Fig. 75.2 Architecture of proposed algorithm

The model in the above figure begins with a clean speech signal (from a desired
speaker), x(k), being corrupted with an unwanted additive noise, v(k). The noisy
signal, y(k), which is a superposition of x(k) and v(k) is first processed to determine
whether the desired speech is currently present or absent. The noise and speech
statistics such as the covariance matrices and power spectral densities are esti-
mated based on the detection results and the input noisy speech. These statistics
will be used to estimate a noise reduction filter. This filter can be optimal in the
sense that it optimizes some error criterion (e.g., MSE) [7]. It may be suboptimal,
where parameters are introduced to better control the quality of the output speech.
The estimated filter is applied to the noisy speech to filter out the noise signal [8],
thereby producing an output signal, z(k), which is supposed to be an estimate of the
clean speech, x(k) (Fig. 75.2).
Coming to the proposed algorithm and we are doing compliment of wei-
ner_Hopf optimal filter algorithm [9]. The below is famous weiner-hopf equation
has been used for the cross-correlation and auto-correlation.

X
M2
Ckxy ¼ xx
hj Ckj k ¼ M1; . . .; M2
j¼M1

Equation: Weiner-Hopf Equation

Where the shorthand notation
804 N.S.B. Kishore et al.

X X
Ckxy ¼ xi yi þ k Clxx ¼ xi xiþl
i i

Cxy xx
k is the cross correlation of x, y and Cl is the auto correlation of x.

Algorithm
1. Read speech utterance wave.
2. Calculate length of speech utterance denoted by M.
3. Next calculate cross correlation of signal denoted by Css.
4. Similarly I do cross correlation of white noise denoted by Cww.
5. function [h, e] = WH(Css, Cww, M).
while abs(e0 - e1) [ 1e - 6
N = N + 1;
e0 = e1;
Cxs = Css(M:M + N - 1);
Cxx = Cww(M:M + N - 1) + Css(M:M + N - 1);
C_xx = zeros(N);
for j = 1:N
for n = 1:N
C_xx(j, n) = Cxx(abs(j - n) + 1);
end
end
h = inv(C_xx)*Cxs’;
e1 = Css(M) - h’*Cxs’;
end, N
6. e = e1;
It returns estimation of noise between filtered signal [10] and original signal.

75.4 The Proposed System Architecture

Here First we take input speech utterance can be given to proposed algorithm and here
it removes the noise from the speech signal and this clean speech can be given as input
to ASR system and from the ASR system can get recognized text as output (Fig. 75.3).

75.5 Analysis and Results

The implementation process of proposed Environmental noise analysis for robust

ASR system with the different data conditions, recognition accuracy is tested. The
data is collected in the classrooms and laboratory of U.G and P.G students of
75 Environmental Noise Analysis for Robust Automatic Speech Recognition 805

Input Proposed ASR Recognized

Speech Algorithm SYSTEM Text

Fig. 75.3 The proposed total system architecture

Department of Computer science engineering with A/C noise, and fan noise
conditions, Open space environment in live mode and command modes of ASR
system. Here we are specifying the signal which contains background noise, which
is considered as a silence and represented with a symbol(*) as shown in REF. and
corresponding HYP.

REF: ******* ******** ****** ****************** ********** *******

****** NAXMAXSKAARAXMUH (s10)
HYP: PIXTTAX PAXTNAXM IXCCIX DHAXSAXLUHAMDUHNUH
VAXKKAXDIA KAXNNAA PAALUH UHAMTAARUH (s10)
SENTENCE 1 (s10)
Correct = 0.0 % 0 (0)
Errors = 800.0 % 8 (8)
SC NAXMAXSKAARAXMUH ==> PAALUH UHAMTAARUH

=============================================================
===============

REF: ****** ******* ******* THXAILUHGUHVAARAXNDHAXRAXMUH

********* THXAILUHGIA MAATLAADUHDHAAM (s11)
HYP: PUHLIX PAXDHAX KAXNCAI THXAAGAXDAXM
AIKKAXDAX UHAMTAARUH PUHLIX (s11)

SENTENCE 2 (s11)
Correct = 0.0 % 0 (0)
Errors = 233.3 % 7 (15)
SC THXAILUHGUHVAARAXNDHAXRAXMUH ==> KAXNCAI
THXAAGAXDAXM
SC THXAILUHGUHVAARAXNDHAXRAXMUH ==> THXAAGAXDAXM
AIKKAXDAX
SC THXAILUHGIA ==> AIKKAXDAX UHAMTAARUH

Experiment 1:
Here in this first we observe manually how the speech signal vary to noise [11] and
how speech signal parameters can vary from utterance to utterance and for this
purpose we recorded telugu aksharas and observe speech utterances manually by
using praat and the observed results can be shown in below table. In the Below
table we analyzing the background noise (indicated by silence) in telugu
(Table 75.1).
806 N.S.B. Kishore et al.

Table 75.1 Telugu Aksharas silence duration, pitch, and standard deviation values
Telugu Sound Silence Duration Silence Mean Standard
literals before (s) (s) after (s) pitch(Hz) deviation
A 0.333.712 0.287773 0.40275 111.659 7.27
Aa 0.40275 0.43363 0.438229 112.543 13.394
I 0.438229 0.335078 0.294343 124.86 12.231
Ii 0.294343 0.385668 0.476336 110.601 8.641
U 0.476336 0.306826 0.356102 118.031 5.351
uu 0.356102 0.513786 0.519699 108.09 6.528
rx 0.519699 0.269376 0.279888 122.599 3.58
rx* 0.279888 0.466481 0.486848 108.196 8.062
lx 0.486848 0.342305 0.291058 123.191 6.398
lx* 0.291058 0.574231 0.443485 110.263 7.949
e 0.443485 0.312082 0.536124 117.085 7.55
ei 0.536124 0.468452 0.487505 106.711 8.831
ai 0.487505 0.312739 0.542037 119.433 6.033
o 0.542037 0.43363 0.545322 112.89 9.917
oo 0.545322 0.501959 0.590656 110.334 9.969
ah 0.590656 0.327193 0.487505 121.822 7.891
n’ 0.487505 0.237839 0.335078 123.072 8.314
: 0.335078 0.354131 0.579487 102.986 4.553
k 0.579487 0.236525 0.603797 112.467 2.623
kh 0.603797 0.296314 0.472394 112.155 4.516
g 0.472394 0.37647 0.545322 111.659 7.987
gh 0.545322 0.27069 0.495389 109.946 5.172
ng 0.495389 0.507215 0.505901 116.89 12.802
ch 0.505901 0.319966 0.442828 108.303 3.898
chh 0.442828 0.35216 0.429688 107.088 3.571
j 0.429688 0.496046 0.489476 107.88 5.873
jh 0.489476 0.438229 0.52824 106.362 5.193
nj 0.52824 0.412605 0.492104 112.888 7.395
t’ 0.492104 0.255579 0.577516 108.029 3.065
th’ 0.577516 0.287116 0.451369 106.922 5.512
d’ 0.451369 0.378441 0.488819 109.762 7.249
dh’ 0.488819 0.362015 0.394209 105.849 5.119
nd’ 0.394209 0.403407 0.50853 110.177 5.202
(continued)
75 Environmental Noise Analysis for Robust Automatic Speech Recognition 807

Table 75.1 (continued)

Telugu Sound Silence Duration Silence Mean Standard
literals before (s) (s) after (s) pitch(Hz) deviation
t 0.50853 0.28843 0.496703 109.322 6.188
th 0.496703 0.385668 0.545979 108.185 7.091
d 0.545979 0.388296 0.429688 108.86 6.628
dh 0.429688 0.434287 0.429031 108.35 7.549
n 0.429031 0.408006 0.568318 109.335 4.356
p 0.568318 0.27726 0.557149 107.476 2.524
ph 0.557149 0.346247 0.499331 109.387 7.081
b 0.499331 0.412605 0.484877 108.325 4.649
bh 0.484877 0.423775 0.489476 108.34 5.189
m 0.489476 0.353474 0.467138 112.049 3.002
y 0.467138 0.394209 0.462538 106.83 7.472
R 0.462538 0.367271 0.539409 110.831 8.079
L 0.539409 0.427717 0.513786 108.453 3.137
V 0.513786 0.3653 0.470423 107.352 4.041
Sh 0.470423 0.475022 0.502616 109.004 7.41
shh 0.502616 0.423118 0.422461 108.921 7.035
S 0.422461 0.478307 0.448084 106.463 5.282
H 0.448084 0.417861 0.461881 107.729 4.934
l’ 0.461881 0.394209 0.505901 106.295 5.918
kshh 0.505901 0.331136 0.542037 110.547 10.407
r’ 0.542037 0.310768 0.48811 112.801 10.438

Experiment 2:
Here in this experiment how the utterance can be formed from the literals and the
formants of these literals can vary and based on these analysis we understand what
differentiates noise from speech utterance. And these analyzed results can be
shown in the following table (Table 75.2).
 808

Table 75.2 Telugu words analysis based on formants

Telugu Words Literals Start (s) End (s) Duration (s) Pitch (Hz) F1 (Hz) F2 (Hz)
Adi
a 103.312087 103.471702 0.159614 154.668 624.422568 1685.96567
di 103.487438 103.716744 0.229305 152.011 328.070656 2254.1126
Illu
i 172.380741 172.524619 0.143878 173.63 373.177863 2172.44666
llu 172.533611 172.774157 0.240546 176.3 585.846423 1800.05142
Aame
aa 551.546091 551.721442 0.175351 147.424 844.84012 1295.31013
me 551.757411 551.959739 0.202328 164.44 558.923428 2023.92283
Daakt’ar
Daa 504.453758 504.631357 0.177599 141.164 558.482715 1649.39412
kt’ar 504.862911 505.049502 0.186592 134.192 389.985309 1554.98185
profesar
pro 1256.661 1256.8296 0.168607 135.143 714.400342 1533.15624
fe 1256.83588 1256.97076 0.134885 192.012 760.921128 1926.15001
sar 1256.991 1257.19557 0.204576 150.149 652.577183 1747.85237
N.S.B. Kishore et al.
75 Environmental Noise Analysis for Robust Automatic Speech Recognition 809

2000

1500

1000

500

0
1 2 3 4 5 6 7 8 9 10
speaker4 speaker1
speaker3 NAMES
speaker2

Fig. 75.4 Comparative results diagram of different speakers

Fig. 75.5 Source signal conversion to filter signal

Experiment 3:
In this experiment, we recorded utterances with different speakers and estimate the
noise from the original signal, filtered signal [12] and filter the noise from original
signal and we specify this analyzed result in the following table (Figs. 75.4, 75.5
and 75.6) (Table 75.3).
810 N.S.B. Kishore et al.

Fig. 75.6 The Noise

estimation to source signal

Table 75.3 Noise estimation values of different speakers with accuracy before and after noise
reduction (NR)
Estimation of noise between filtered signal and original signal
Recorded utterances Speaker1 (s) Speaker2 (s) Speaker3 (s) Speaker4 (s)
1 79.8852 157.8153 681.2529 700.0783
2 50.953 405.217 338.6157 179.7575
3 150.4635 219.2007 555.8127 213.5948
4 223.981 557.311 210.1028 57.0407
5 217.9731 155.5096 660.6022 134.9966
6 151.8165 311.5184 315.4281 178.0364
7 128.1514 202.1541 340.905 131.0829
8 201.2992 196.0372 317.9495 185.5135
9 163.5398 285.7568 200.374 186.6847
10 431.5037 362.5521 120.1197 158.4786
Accuracy before NR (%) 50 40 20 60
Accuracy after NR (%) 70 80 30 90

75.6 Conclusion

The compliment of Weiner-Hopf optimum filter algorithm is giving better results

for lab environmental conditions. When it is complex environmental situation like
outdoor environments the performance of the system is not up to the level. The
system works well for small amount of data.
75 Environmental Noise Analysis for Robust Automatic Speech Recognition 811

When the environmental condition is complex situation, the system has to train
in such a way that the complex noise structure should be identified and need to be
removed from speech utterance. The different environment noise conditions are to
be estimated and appropriate filter design analysis has to be studied to clean the
speech signal in real time ASR applications.

References

1. Veisi, H., Sameti, H.: Hidden-Markov-model-based voice activity detector with high speech
detection rate for speech enhancement. IET Sig. Process. 6(1), 54–63 (2012)
2. Kashiwagi, Y., Suzuki, M., Minematsu, N., Hirose, K.: Audio-Visual feature interation based
on piecewise linear transformation for noise robust automatic speech recognition. In:
ICASSP, pp. 149–152. IEEE (2010)
3. Ghaemmaghami, H., Dean, D., Sridharan, S., McCown, I.: Noise robust voice activity
detection using normal testing and time-domain histogram analysis. In: ICASSP,
pp. 4470–4473. IEEE (2010)
4. Dhananjaya, N., Yegnanarayana, B., Senior Member, IEEE: Voiced/nonvoiced detection
based on robustness of voiced epochs. IEEE Sig. Process. Lett. 17(3) (2010)
5. Benesty, J., Chen, J., Huang, Y., Cohen, I.: Noise reduction in speech processing. Springer,
Heidelberg (2009)
6. Kim, D.Y., Un, C.K., Kim. N.S.: Speech recognition in noisy environments using first-order
vector Taylor series. Speech Commun. 24, 39–49 (1998)
7. Ramirez, J., Yelamos, P., Gorriz, J.M., Segura, J.C.: SVM-based speech endpoint detection
using contextual speech features. Electron. Lett. 42(7), 426–428 (2006)
8. Sebastian Seung, H.: Wiener-Hopf equations. Convolution and correlation in continuous
time, 9.29 Lecture 3: February 11, 2003 (2003)
9. Verma, A.R., Singh, R.K., Kumar, A., Ranjeet, K.: An improved method for speech
enhancement based on 2D-DWT using hybrid weiner filtering. In: 2012 IEEE International
Conference on Computational Intelligence and Computing Research (2012)
10. Samudravijaya, K., Barot, M.: A comparison of public domain software tools for speech
recognition. Workshop on spoken language processing, pp. 125–131 (2003)
11. Fukane, A.R., Sahare, S.L.: Role of noise estimation in enhancement of noisy speech signals
for hearing aids. In: Computational Intelligence and Communication Networks (CICN),
pp.648–652. IEEE (2011)
12. Ling, G., Yamada, T., Makino, S., Kitawaki, N.: Performance estimation of noisy speech
recognition using spectral distortion and snr of noise-reduced speech. In: TENCON 2013,
IEEE (2013)
Chapter 76
Performance Comparison of Selected
Classification Algorithms Based on Fuzzy
Soft Set for Medical Data

Saima Anwar Lashari and Rosziati Ibrahim

Abstract Medical data is heterogeneous in nature and associated with uncer-

tainties. For that reason, data mining has been assisting physicians in decision
making and to cope with the information overload. A considerable amount of
literature has been available on medical data classification based on data mining
techniques to automate or facilitating the delineation of images. However, from
image formation to the final analysis, medical imaging is still facing challenges.
New imaging procedures for classification could overcome the inefficiencies and
provide more reliable information to the medical experts. Therefore, this paper
assesses the performance of selected classification algorithms based on fuzzy soft
set for classification of medical data. There are two concepts that underlie the
classification in the fuzzy soft set theory namely: classification based on decision
making problem and classification based on similarity between two fuzzy soft set.
The selected algorithms are evaluated based on two criteria: accuracy and com-
putational time. Moreover, the conducted experiments demonstrated the effec-
tiveness of fuzzy soft set for medical data categorization.

  
Keywords Medical data Classification Data mining Soft set Fuzzy soft set 

76.1 Introduction

The primary objective of medical data classification is not only to achieve good
accuracy but to understand which parts of anatomy are affected by the disease to
help clinicians in early diagnosis of the pathology and in learning the progression

S.A. Lashari (&)  R. Ibrahim

Faculty of Computer Science and Information Technology,
Universiti Tun Hussein Onn Malaysia, Parit Raja, 86400 Batu Pahat, Johor, Malaysia
e-mail: [email protected]; [email protected]
R. Ibrahim
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 813

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_76
814 S.A. Lashari and R. Ibrahim

of a disease. Thus, the concept of data mining was created and to present rising levels
of computerization in the information creation process, subsisting a vast amount of
time burly human activity with regular techniques that improves accuracy [1]. Most
of recent efforts reviewed are more related to the development where ideas are
motivated from concepts of pattern recognition, image processing, and computer
vision. However, with this in mind, it is important to realize that medical data
analysis efforts are heavily influenced by the particular image datasets being utilized
and the clinical or biological tasks that underlie the need for medical data analysis.
Thus, for medical data various classification algorithms have been put forward
in articles and scientific writings, including bayesian rule [2], nearest- neighbor
methods [3, 4], decision tree induction [5], error back propagation [6]. Every one
of them has contributed to address problems in data mining. However, no single
method has been found to be superior over all others for all datasets [7].
In 1999 Molodtsov [8] introduced a new mathematical tool named soft set
theory to deal with uncertain problems. The initial description of the object has an
approximate nature and there is no need to introduce the notation of exact solution.
The applications of this theory boom in recent years and is extended to, among
others, data analysis [9], soft decision making [10], texture classification [11],
musical instrument classification [12].
Soft set can work well on the parameters that having binary numbers but still
difficult to work with parameters that have a real number. There are many issues in
the classification involving real numbers. To overcome this problem Maji et al. [13]
offered a more general concept namely fuzzy soft set which can handle parameters
in the form of real numbers. Since medical data is associated with uncertainty and
most of data is available in the form of real numbers. This motivated us to see the
viability of fuzzy soft set theory to perform medical data classification and to see
how effective this theory is to handle uncertainty nature of medical data.
The rest of the paper is organized as follows: theoretical background of clas-
sification algorithms is detailed in Sect. 76.2. The modeling process presented in
Sect. 76.3. Section 76.4 contains the conclusion of this study.

76.2 Classification Algorithms

Classification algorithm normally known as classifier is a method of finding a

classification model. The role of classifier is significant as it has to analyse and
extract information from numerical vectors into implicit form. However, despite
the fact that there are many classification algorithms, in certain circumstances, one
classification algorithm may perform better than the other [14]. Many different
classification strategies were applied such as nearest neighbor-based approach [4],
decision trees [15] as well as support vector machine [16]. Analogous to feature
combination, classifier combination has also been a popular way to improve
classification performance [4]. The entire algorithm array belongs to the category
of supervised learning methods.
76 Performance Comparison of Selected Classification … 815

Therefore, this seems to indicate that data mining techniques is beginning to be

applied widely in the detection and differential diagnosis of many different types of
abnormalities in medical data obtained in various examinations by use of different
imaging modalities [15].

76.2.1 Fuzzy Soft Set

The concept of fuzzy set was introduced by Zadeh in 1965 [17] to allow elements
to belong to a set in a gradual rather than an abrupt way (i.e. permitting mem-
berships valued in the interval [0, 1]) instead of in the set {0, 1}. Ever since then,
applications and developments based on this simple concept have evolved to such
an extent that it is practically impossible nowadays to encounter any area or
problem where applications, developments, products etc. are not based on fuzzy
sets [18]. Later, Maji et al. [12] have studied a general concept, namely theory of
fuzzy soft set.
There are two important concepts underlying the application of the theory of
fuzzy soft set in numerical classification problems.
• concept of decision making problems
• concept of measuring similarity
Based on an application of soft set in a decision making problem presented by
Maji et al. [19], Mushrif et al. [10] presented a novel method for classification of
natural textures using the notions of soft set theory, all features on the natural
textures consist of a numeric (real) data type, have a value between [0, 1] and the
algorithm used to classify the natural texture is very similar to the algorithm used by
Roy and Maji [13] in the decision making problems. The algorithm was success-
fully classify natural texture with very high accuracy when compared with con-
ventional classification methods such as Bayes classifier and a minimum distance
classifier based on Euclidean distance. He has also proved that the computation
time for classification is much less as compared to with ayes classification method.
Later, Lashari et al. [11] applied soft set theory to classify musical instruments
sounds and their results revealed that soft set theory can be successfully used for
the classification of musical instruments.
Measuring similarity between two entities is a key step for several data mining
tasks, such as classification and clustering. Similarity measures quantify the extent to
which different patterns, signals, images or sets are alike. The studies on measuring
the similarity between soft set have been carried out. They extended their research
to measure the similarity of fuzzy soft set and describe how it can be applied to
medical diagnosis to detect whether a person is suffering from a certain disease [8].
This paper investigates two existing classification approaches based on fuzzy
soft set theory, one is fuzzy soft set based on decision making problems (comprises
of comparison table [13]) and other is based on similarity measurement between
two fuzzy soft set [20].
816 S.A. Lashari and R. Ibrahim

76.2.2 Preliminaries

In this subsection, the basic definition and results of soft set theory and fuzzy soft
set which would be useful for subsequent discussion. Most of the definitions and
results presented in this section may be found in [8, 21].

Definition 1 Let U be an initial universe set and E be a set of parameters. Let

~ ðU Þ denote the power set of U and A  E
P
 
A pair F;~ E a fuzzy soft set over U where F is a mapping given by F
~!P ~ ðU Þ.

In the above definition, fuzzy subsets in the universe U are used as substitutes
for the crisp subsets of U. Therefore, it is well known that notion of fuzzy sets
provides convenient tool for representing vague concepts by allowing partial
memberships. Every fuzzy soft set can be viewed as fuzzy soft set information
system and be represented by a data table with entries belongs to the interval ½0; 1.
Table 76.1 represents both approaches based on fuzzy soft set theory. The
comparison table classifier works by calculating the average value of each
parameter from all objects with same class/label. Then construct a comparison
table in the manner as the preparation of comparison table in the case of decision
making problem where optimal decision is taken from maximum score computed
from the comparison table. The next step is to calculate score to determine class
label for the test data. Whereas, the approach based on similarity measure have the
same learning phase with comparison table only the classification method is
different.

76.3 Modeling Process

In this section, the modeling process of this study is presented. The modeling
process consists of three phases which are data collection, data partitioning and
validation measure. Each of the steps is discussed in detail in the following sub-
sections.

76.3.1 Data Collection

Dataset is one of the crucial elements for designing and developing successful
classification algorithms. Data collection has been done using University of
California at Irvine (UCI) machine learning repository. Since UCI is a public
repository that makes easier for data collection. Dataset includes breast-cancer-
wisconsin (wdbc & wpdc), breast tissue, heart, dermatology, liver disorder, hep-
atitis, pima Indians diabetes and indian liver dataset.
76 Performance Comparison of Selected Classification … 817

Table 76.1 Fuzzy soft set based classification

FSSCT: Fuzzy soft set based on decision FSSSM: Fuzzy soft set based on similarity
making problems [13] measure [20]
Training phase Training phase
1. Input fuzzy soft sets 1. Given N samples obtained from the data
ðF; ~ BÞ and ðH;
~ AÞ; ðG; ~ CÞ class w
2. Input parameters set P as observed by the 2. Calculate the cluster center vector
observer P
N
Ewi ; i ¼ 1; 2; ::; N. Ew ¼ N1 Ewi
i¼1
Obtain Fuzzy soft set model (F, E), which is
W  D table of cluster centers in which an
element of the table is gwd , where w ¼
1; 2; . . .; W and d ¼ 1; 2; . . .; D and a row gwd
is a cluster centre vector for every class
w having D features
3. Compute corresponding resultant fuzzy soft 3. Repeat the process for all W classes
set ð~S; PÞ from fuzzy soft sets
~ AÞ; ðG;
ðF; ~ BÞ; ðH;
~ CÞ and place it tabular form

Classification phase Classification phase

1. Obtain the unknown class data 1. Obtain the unknown class data
2. Compute comparison table of fuzzy soft 2. Obtain a fuzzy soft sets model for unknown
 
class data G;~ E and compute similarity
ð~S; PÞ and compute ri using equation ri ¼    
between G; ~ E and F ~ w ; E for each w using
P
n P
n
equation
cij and ti using equation tj ¼ cij Pn
    jF~ij G~ ij j
i¼1 j¼1
S Fq ; Gd ¼ Mi F; ~ ¼1 P
~ G n
j¼1

ðF~ij þG~ ij Þ
j¼1

3. Compute the score vector S using equation 3. Assign the unknown data to class w if
si ¼ ri  ti If k has more than one value then similarity is maximum
any one of ok may be chosen w ¼ arg½maxW ~ ~
w¼1 SðG; Fw Þ

Datasets undergo for pre-processing treatments. Data pre-processing phase

involve fuzzification technique to make sure that data range lies between ½0; 1.
To classify numerical data with these algorithms, the second step is replaced
with fuzzification process (refer to [10]) having similar algorithm with [13], which
is like counting the normalization so that all parameters have a value between
½0; 1. For example, if the classification algorithm is applied to Breast Tissue
dataset. Fuzzification [20] can be done by dividing each attributes value with the
largest value at each attributes, efi ¼ maxei ðei Þ where ei ; i ¼ 1; 2; . . .; n is the old
attribute and efi is attribute with new value between ½0; 1.
Table 76.2 provides description of all dataset. Most of the datasets having real
numerical features and some contains multiclass labels.
818 S.A. Lashari and R. Ibrahim

Table 76.2 Dataset description

No. Dataset Description
1. Breast tissue i:106, f:10, c:6
2. Statlog (Heart) i:270, f:13, c:2
3. Dermatology i:366, f:33, c:6
4. Liver disorder i:345, f:7, c:2
5. Hepatitis i:155, f:19, c:2
6. Pima Indians diabetes i:768, f:8, c:2
7. Lung cancer i:32, f:56, c:3
8. Lymphography i:148, f:18, c:4
9. Indian liver patient dataset (ILPD) i:583, f:10, c:2
i instance, f features, c class

76.3.2 Data Partitioning

For data partition, a general course in data mining is to split into training and
testing sets. Each dataset divided into two parts: 70 % for training and 30 % for
testing and data were selected randomly for every experiment.

76.3.3 Validation Measure

For classification problems, the major source of performance measurement is

coincidence matrix. However, when the classification problem is not binary,
performance evolution becomes limited to overall classifier accuracy. Therefore,
in order to quantify the performance of classification method, the performance
metrics: overall classifier accuracy (OCA) and computational time have been used
to access the performance of both classifiers:

76.4 Results and Discussion

Table 76.3 provides the results obtained from different datasets. In general both
classifiers can do numerical classification however, highest achievement occurs in
lung cancer dataset where FSSSM accuracy 97.05 and FSSCT accuracy 93.67. The
reason that FSSSM gives better results because it does not need to build com-
parison table therefore it can work faster.
76 Performance Comparison of Selected Classification … 819

Table 76.3 Performance analysis of classification approaches for medical data

Classification algorithms based on fuzzy soft set
Accuracy measures
Classification algorithms FSSCT FSSSM
UCI datasets Accuracy Cpu time Accuracy Cpu time
(%) (s) (%) (s)
Breast tissue 56.13 0.0542 63.87 0.0033
Statlog (Heart) 82.72 0.0050 77.04 0.0049
Dermatology 82.97 0.0114 97.03 0.0098
Liver disorder 51.17 0.0065 53.01 0.0057
Hepatitis 81.91 0.0036 83.62 0.0034
Pima Indians diabetes 70.35 0.0154 70.22 0.0098
Indian liver patient dataset 64.19 0.0033 78.52 0.0050
(ILPD)
Lung cancer 93.67 0.0096 97.05 0.0102
Lymphography 58.06 0.0031 78.89 0.0050

76.5 Conclusion

Current research in medical data classification mainly focuses on the use of effi-
cient data mining algorithms and visualization techniques. Meanwhile, the major
objective of current studies strives towards improving the classification accuracy,
precision and computational speeds of classification algorithms, as well as
reducing the amount of manual interaction.
Therefore, in this paper, we investigated existing classification algorithms based
on fuzzy soft set for medical data. Nine datasets from UCI were used to test the
accuracy and computational time of both classifiers. In general, both can do the
classification of numerical datasets. It is experimentally demonstrated that both
classification algorithms yields better accuracy. From these evidences on medical
data classification, it can be seen that there is still much room for further
improvement over current medical data classification task. More research, how-
ever, is needed to identify and reduce uncertainties in medical data classification to
improve classification accuracy. For future work we will design a new classifi-
cation algorithm based on fuzzy soft set using medical images.

Acknowledgments The authors would like to thank office for Research, Innovation, Commer-
cialization and Consultancy Management (ORICC) and Universiti Tun Hussein Onn Malaysia for
supporting this research under vote no 1255.
820 S.A. Lashari and R. Ibrahim

References

1. Antonie, M.L., Zaiane, O.R., Coman, A.: Application of data mining techniques for medical
image classification. In: MDM/KDD, pp. 94–101 (2001)
2. Fesharaki, N.J., Pourghassem, H.: Medical X-ray images classification based on shape
features and Bayesian rule. In: 2012 Fourth International Conference on Computational
Intelligence and Communication Networks (CICN), pp. 369–373. IEEE (2012)
3. Latifoglu, F., Polat, K., Kara, S., Gunes, S.: Medical diagnosis of atherosclerosis from carotid
artery Doppler signals using principal component analysis (PCA), k-NN based weighting pre-
processing and Artificial Immune Recognition System (AIRS). J. Biomed. Inform. 41, 15–23
(2008)
4. Suguna, N., Thanushkodi, K.: An improved k-nearest neighbour classification using genetic
algorithm. Int. J. Comput. Sci. Issues (IJCSI) 7(4, 2), 18 (2010)
5. Rajendran, P., Madheswaran, M., Naganandhini, K.: An improved pre-processing technique
with image mining approach for the medical image classification. In: 2010 International
Conference on Computing Communication and Networking Technologies (ICCCNT),
pp. 1–7. IEEE (2010)
6. Hadidi, M.R.A.A., Gawagzeh, M.Y., Alsaaidah, B.: Solving mammography problems of
breast cancer detection using artificial neural networks and image processing techniques.
Indian J. Sci. Technol. 5(4), 2520–2528 (2012)
7. Ali, S., Smith, K.A.: On learning algorithms selection for classification. Appl. Soft Comput.
6, 119–138 (2006)
8. Majumdar, P., Samantra, S.K.: Generalized fuzzy soft set. J. Comput. Appl. Math. 58,
1279–1286 (2010)
9. Ali, M.I.: A note on soft sets, rough sets and fuzzy soft sets. Appl. Soft Comput. 11,
3329–3332 (2011)
10. Mushrif, M.M., Sengupta, S., Ray, A.K.: Texture classification using a novel soft set theory
based classification algorithm. In: LNCS, vol. 3851, pp.246–254. Springer, Heidelberg
(2006)
11. Lashari, S.A., Ibrahim, R., Senan, N.: Soft set theory for automatic classification of
traditional Pakistani musical instruments sounds. In: 2012 International Conference on
Computer & Information Science (ICCIS), vol. 1, pp. 94–99. IEEE (2012)
12. Maji, P.K., Biswas, R., Roy, A.R.: Fuzzy soft sets. J. Fuzzy Math. 9(3), 589–602 (2001)
13. Roy, A.R., Maji, P.K.: A fuzzy soft set theoretic approach to decision making problems.
J. Comput. Appl. Math. 203(2), 412–418 (2007)
14. Kotsiantis, S.B.: Supervised machine learning: a review of classification techniques.
Informatica 31, 249–268 (2007)
15. Tu, M.C., Shin, D., Shin, D.: A comparative study of medical data classification methods
based on decision tree and bagging algorithms. In: Eighth IEEE International Conference on
Dependable, Autonomic and Secure Computing, DASC’09, pp. 183–187. IEEE (2009)
16. Xing, Y., Wang, J., Zhao, Z., Gao, Y.: Combination data mining methods with new medical
data to predicting outcome of coronary heart disease. In: International Conference on
Convergence Information Technology, 2007, pp. 868–872. IEEE (2007)
17. Zadeh, L.A.: Fuzzy sets. Inf. Control 8(3), 338–353 (1965)
18. Zimmermann, H.J.: Fuzzy set theory-and its applications. Springer, New York (2001)
19. Maji, P.K., Roy, A.R., Biswas, R.: An application of soft sets in decision making problem.
Comput. Math. Appl. 44, 1077–1083 (2002)
20. Handaga, B., Deris, M.M.: Similarity approach on fuzzy soft set based numerical data
classification. Commun. Comput. Inf. Sci. 180(6), 575–589 (2011)
21. Molodtsov, D.: Soft set theory–first results. Comput. Math. Appl. 37(4–5), 19–31 (1999)
Chapter 77
A Hybrid Selection Method Based
on HCELFS and SVM for the Diagnosis
of Oral Cancer Staging

Fatihah Mohd, Zainab Abu Bakar, Noor Maizura Mohamad Noor,

Zainul Ahmad Rajion and Norkhafizah Saddki

Abstract A diagnostic model based on Support Vector Machines (SVM) with a

proposed hybrid feature selection method is developed to diagnose the stage of
oral cancer in patients. The hybrid feature selection method, named Hybrid Cor-
relation Evaluator and Linear Forward Selection (HCELFS), combines the
advantages of filters and wrappers to select the optimal feature subset from the
original feature set. In HCELFS, Correlation Attribute Evaluator acts as filters to
remove redundant features and Linear Forward Selection with SVM acts as the
wrappers to select the ideal feature subset from the remaining features. This study
conducted experiments in WEKA with ten fold cross validation. The experimental
results with oral cancer data sets demonstrate that our proposed model has a better
performance than well-known feature selection algorithms.

F. Mohd (&)  N.M.M. Noor

School of Informatics and Applied Mathematics, Universiti Malaysia Terengganu,
21030 K. Terengganu, Terengganu, Malaysia
e-mail: [email protected]
N.M.M. Noor
e-mail: [email protected]
Z.A. Bakar
Faculty of Computer and Mathematical Sciences, Universiti Teknology MARA (UiTM),
Selangor 40450 Shah Alam, Malaysia
e-mail: [email protected]
Z.A. Rajion  N. Saddki
School of Dental Sciences, Universiti Sains Malaysia (USM), Kubang Kerian 16150,
Kelantan, Malaysia
e-mail: [email protected]
N. Saddki
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 821

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_77
822 F. Mohd et al.

77.1 Introduction

Feature selection (FS) as preprocessing steps to machine learning in real world

data, is very useful in reducing dimensionality, removing irrelevant data and noise
to improve result. It could directly reduce and remove irrelevant number of the
original features by selecting a subset that contributes to the optimum information
for classification. The FS algorithms are divided into two categories: the filter
methods and wrappers methods [1]. Both methods have their own abilities and
advantages. The filter method contributes high computational efficiency compared
to the wrapper method. The wrapper could achieve better results than the filter
approach. In this study, we combined both methods to propose a hybrid algorithm
to gain the optimum selected features.
Head and neck (HNC) cancer is one of the major cancers worldwide. One part
of HNC is oral cancer with the incidence rising in every country. Early clinical
cancer diagnosis is seen as an important element in reducing the mortality rate of
this deadly disease. The process of clinical diagnosis begins with information
gathering or eliciting data from a patient’s history. It includes data collection from
patient’s primary report of symptoms, past medical history, family history, and
social history. In this process, sometimes decision making can be done, where the
clinician can start the procedure of formulating a list of possible diagnoses [2].
Then, by doing a physical examination, the physician detects abnormalities by
looking at, feeling, and listening to all parts of body. However, the patient’s record
is a collection of features and data that leads to problems for the diagnosis process.
The challenge of applying computational solution to the data collected is in the
conversion of data into an appropriate form, suitable for the diagnosis process [3].
Because of this, FS method is applied to reduce the irrelevant data and finally
select the optimum features to diagnose the stage of oral cancer. This paper
explains the development of a diagnostic model based on Support Vector
Machines (SVM) with a proposed hybrid feature selection method in diagnosing
the stage of oral cancer.
The remaining of the paper is organized as follows: related work is given in
Sect. 77.2, while Sect. 77.3 gives a brief description about the FSA algorithms—
Correlation Attribute Evaluator and LFS, SVM algorithm. Section 77.4 discusses
the diagnostic model for oral cancer and Sect. 77.5 reports the results and dis-
cussion. The concluding remarks are given in Sect. 77.6 to address further research
issues.

77.2 Related Work

Support Vector Machine (SVM) is an affective algorithm used in medical diag-

nosis for pattern recognition, machine learning and data mining. In the literature,
there are some works related to medical diagnosis. Aruna et al. compared the
77 A Hybrid Selection Method Based on HCELFS and SVM … 823

performance criterion of supervised learning classifiers such as Naïve Bayes, SVM

RBF kernel, RBF neural networks, Decision trees J48 and Simple CART. The
experiments conducted were found that SVM RBF Kernel produced highest result
than other classifiers with respect to accuracy, sensitivity, specificity and precision
[4]. Jaganathan et al. have proposed a feature selection method with improved
F-score and SVM for breast cancer diagnosis and produced a classification
accuracy of 95.565 %. This result is better than RBF Network (95.278 %) [5]. In
other field, SVM is also applied in cyber-security. Maldonado and L’Huillier
proposed an embedded approach for feature selection using SVM in phishing and
spam classification. It outperforms other techniques in terms of classification
accuracy by removing the features that affect on the generalization of the classifier
by optimizing the Kernel function [6].
This related work is also focusing on the diagnosis of head and neck cancer using
machine learning and data mining algorithm. For instance, Kawazu et al. [7] used
neural network (NN) to predict lymph node metastasis of patients with oral cancer.
They utilized histopathological data set of lymph nodes which saw an accuracy of
93.6 % in diagnosing patients. Boronti et al. produced four different results with
three different methods such as SVM, Decision Trees (DTs), XCS and NN with
accuracies of 75.5, 76.5, 79.2 and 71.3 % respectively [8]. In another study, they
continued with other methods. They produced a classification result with DTs
(70 %), XCS (79 %) and NN (78 %) [9]. Besides this, Exarchos et al. [10] employed
a feature selection algorithm, Correlation-based Feature Subset selection (CFS) and
the wrapper algorithm in order to omit redundant or possible irrelevant features and
maintain the most informative and discriminatory ones. With the applications of
Bayesian Networks, Artificial Neural Networks, SVM, DTs and Random Forests,
the study produced an accuracy of (69.6 %), (66.1 %), (69.6 %), (66.1 %) and
(58.9 %) respectively. However, with a hybrid model of Relief F-GA-ANFIS,
Chang et al. produced a better classification accuracy with 93.81 % [11].

77.3 Materials and Methods

77.3.1 Features Selection Algorithm

In this study, feature selection for high-dimensional data are conducted in WEKA
with tenfold cross validation. The main idea of feature selection functions are used
to find the most significant attributes by removing features with little or no pre-
dictive information. The functions used for attribute evaluation (feature selection)
within this study are as follows:
Correlation Attribute Evaluator. This algorithm evaluates the worth of an
attribute by measuring the correlation between it and the class. Nominal attributes
are considered on a value by value basis by treating each value as an indicator. An
overall correlation for a nominal attribute is arrived at via a weighted average.
824 F. Mohd et al.

CFS Subset Evaluator. This algorithm evaluates the worth of a subset of

attributes by considering the individual predictive ability of each feature along
with the degree of redundancy between them.
The Correlation Feature Selection (CFS) is a simple filter algorithm that ranks
feature subsets according to a correlation based heuristic evaluation function. It
measures subsets of features on the basis of the hypothesis, ‘‘A good feature subset
is one that contains features highly correlated with (predictive of) the class, yet
uncorrelated with (not predictive of) each other’’. The following equation gives
the merit of a feature subset S consisting of k features:

krcf
Meritsk ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð1Þ
k þ kðk  1Þrff

where MeritSk is the heuristic ‘‘merit’’ of a feature subset S containing k features,

rcf is the average value of all feature-classification correlations (f [ S), and rff is the
average value of all feature-feature correlations. The numerator of (1) can be
thought of as providing an indication of how predictive of the class a set of features
are; the denominator of how much redundancy there is among the features [12].
All the attributes were searched using these algorithms:
Ranker. Ranks attributes by their individual evaluations. Use in conjunction
with attribute evaluators (ReliefF, GainRatio, Entropy etc.).
Linear Forward Selection (forward). Linear Forward Selection, a technique to
reduce the number of attributes expansions in each forward selection step. This
function is extension of Best First. It takes a restricted number of k attributes into
account. Fixed-set selects a fixed number k of attributes, whereas k is increased in
each step when fixed-width is selected. The search uses either the initial ordering
to select the top k attributes, or performs a ranking (with the same evaluator the
search uses later on). This algorithm starting from the empty set, sequentially add
the feature x+ that results in the highest objective function J(Yk + x+) when
combined with the features Yk that have already been selected.
1. Start with the empty set Y0 = (Ø)
2. Select the next best feature X + = arg max [J(Yk + x)]x62Yk
3. Update Yk+1 = Yk + x+; k = k + 1
4. Go to 2

77.3.2 Oral Cancer Dataset

The study obtained a record review of oral cancer patients from the Otorhino-
laryngology Clinic at Hospital Universiti Sains Malaysia (HUSM) in Kelantan
respectively. The dataset is made up of 27 parameters and a primary tumor stage as
attributes for the diagnosis of the patients. The study was conducted after the
77 A Hybrid Selection Method Based on HCELFS and SVM … 825

Table 77.1 Description of Attributes no. Attributes name

the datasets
1. Age
2. Gender
3. Ethnicity
4. Smoking
5. Chewing betel quid
6. Alcohol
7. S1
8. S2
9. S3
10. S4
11. S5
12. S6
13. S7
14. S8
15. S9
16. S10
17. S11
18. Site
19. Size
20. Lymph node
21. Histological
22. SCC
23. T
24. N
25. M
26. Stage (class label): Stage I,
Stage II, Stage III, Stage IV

obtainment of the required approvals from the Research and Ethics Committee
(Human), Universiti Sains Malaysia, No.236.4.(4.4) [13]. Number of instances was
210, and 27 features with patient_id was named as label and stage was named as
class label. The numerical variables were analysed through the corresponding
ranges of their values. Age was divided into five groups (group 1: below 30 years
old; group 2: 30–39 years old; group 3: 40–49 years old: group 4: 50–59 years old
and group 5: 60 years old and above) [14, 15]. The oral cancer regions included in
this study were the tongue, buccal mucosa, palate, floor of mouth, maxilla, lip,
cheek, mandible, tonsil, parotid gland, oropharynx and other unspecified parts. The
details of the attributes found in this dataset for features selection listed in
Table 77.1.
826 F. Mohd et al.

1. Data set with all features (25 attributes)

2. Features ranking using Correlation Attributes Evaluator to rank

relevant features highest to lowest

3. Remove features with ranking rate=0

4. Reduce features subset to remove redundant and irrelevant

5. Selection of most relevant features using LFS

6. Selected optimum features

Fig. 77.1 Stage in hybrid correlation evaluator and linear forward selection (HCELFS)
algorithm

77.4 Diagnostic Model: Hybrid Correlation Evaluator

and Linear Forward Selection

A hybrid feature selection method, named HCELFS is proposed in this study. This
FS method, hybrid Correlation Attribute Evaluator with ranker and CFS Subset
Evaluator with Linear Forward Selection was applied for oral cancer diagnosis. It
combines the advantages of both methods to select the optimal features subset
from the original feature set. In the diagnostic model, the first step included the
Correlation Attribute Evaluator method filtering the relevant features which
resulted in reduced features subset. From this subset CFS Subset Evaluator with
Linear Forward Selection (LFS) searched for the most relevant features resulting
in optimum feature set used for diagnosing the cancer stage. Figure 77.1 shows the
stage in the algorithm.

77.5 Results and Discussion

The experiments of features selection against oral cancer data set are conducted in
WEKA with tenfold cross validation. Algorithm started with 25 features and 210
instances. With Correlation Ranking Filter, the algorithm ranked 25 features
namely 20, 23, 21, 22, 16, 19, 24, 8, 2, 15, 7, 17, 3, 18, 5, 1, 13, 9, 11, 6, 25, 10, 14,
4 and 12 (see Table 77.2). We removed 1 feature namely 12 with ranking rate 0.
With the resultant 24 features, the subset method then remove the redundant and
77 A Hybrid Selection Method Based on HCELFS and SVM … 827

Table 77.2 Correlation Correlation ranking filter Ranked attributes (%)

ranking filter for oral cancer
data set with 25 attributes 20: Lymph node 0.4602
23: T 0.4319
21: Histological 0.3715
22: SCC 0.3611
16: S10 0.3561
19: Size 0.3558
24: N 0.3349
8: S2 0.3203
2: Gender 0.2751
15: S9 0.2660
7: S1 0.2644
17: S11 0.2345
3: Ethnicity 0.2323
18: Site 0.2297
5: Betel quid 0.1966
1: Age 0.1884
13: S7 0.1810
9: S3 0.1496
11: S5 0.1465
6: Alcohol 0.1042
25: M 0.1042
10: S4 0.0905
14: S8 0.0599
4: Smoking 0.0455
12: S6 0
Selected 25 attributes:
20, 23, 21, 22, 16, 19, 24, 8, 2, 15, 7, 17, 3, 18, 5, 1, 13, 9, 11, 6,
25, 10, 14, 4 and 12

irrelevant 10 features namely, 7, 5, 1, 13, 11, 6, 25, 10, 14 and 4. This algorithm
ended with 14 features namely 20, 23, 21, 22, 16, 19, 24, 8, 2, 15, 17, 3, 18 and 9
as optimum features set.
Table 77.3 summarize hybrid the features selection methods experimented in
this study. It started with no features selection (FS0), Correlation Attribute
Evaluator with Ranker (FS1), and then combined FS1 with CfsSubset Evaluator
with Linear Forward Selection (FS2).
828 F. Mohd et al.

Table 77.3 Selected attributes with hybrid feature selection methods

FS Method Selected attributes
FS0 No features selection All attributes
FS1 CorrelationAttributeEval Ranked attributes: 20, 23, 21, 22, 16, 19, 24, 8, 2, 15, 7, 17,
3, 18, 5, 1, 13, 9, 11, 6, 25, 10, 14, 4, 12 (25 attributes)
Ranker Remove ranting value = 0 (attribute 12)
FS2 CfsSubsetEval Remove irrelevant attributes = 10 attributes 7, 5, 1, 13, 11,
6, 25, 10, 14, 4
LinearForwardSelection Optimum features: 14 attributes 20, 23, 21, 22, 16, 19, 24, 8,
(forward) 2, 15, 17, 3, 18, 9

Table 77.4 Accuracy performance for classification algorithms on oral cancer data set
Algorithm FS0 FS1 FS2
Updateable Naïve Bayes 91.9048 91.9048 94.7619
8.0952 8.0952 5.2381
MLP 94.2857 93.8095 95.2381
5.7143 6.1905 4.7619
Lazy-IBK 86.1905 86.1905 91.4286
13.8095 13.8095 8.5714
SMO- poly kernel (E-1.0) 93.3333 93.3333 96.1905
6.6667 6.6667 3.8095

In order to evaluate the efficiency of the FS methods, performance measure of

accuracy were considered. The measures are compiled by the Classification
Accuracy (%) = (TP + TN)/(TP + FP + FN + TN). In this study, four different
machine learning algorithms were used to classify the oral cancer data set with
three features selection methods and optimum features selected by the proposed
hybrid algorithm, Naive Bayes (NB), Multilayer Perceptron (MLP), K-Nearest
neighbors (KNN), and SVM. NB classifier using estimator classes. MLP classifier
uses backpropagation to classify instances. This network can be built by hand,
created by an algorithm or both. The network can also be monitored and modified
during training time. K-Nearest neighbor classifier (lazy.IBk) can select appro-
priate value of K based on cross-validation. It can also do distance weighting.
SVM or SMO-Poly Kernel (E-1.0) implemented globally replaces all missing
values and transforms nominal attributes into binary ones. It also normalizes all
attributes by default. Table 77.4 shows the results for the classifier. The empirical
comparison shows that the features selected by the hybrid algorithm also improved
the accuracy of the entire classifier algorithm used for the oral cancer data set.
Table 77.5 shows the classification accuracies of our method and other classifiers
from literature for the head and neck data set.
77 A Hybrid Selection Method Based on HCELFS and SVM … 829

Table 77.5 Classification accuracies of this study method and other classifiers from literature
Author, year Method Accuracy
(%)
Kawazu et al. 2003 [7] Neural networks (NN) 93.6
Baronti et al. 2005 [8] Support vector machines (SVM) 75.5
Decision trees (C4.5) 76.5
XCS 79.2
NN 71.3
Tung and Quek 2005 FS based on wrapper Above
[16] Monte Carlo evaluative selection (MCES) 90 %
Classification using
SVM with polynomial kernel, K-Nearest
Neighbor (K-NN) classifier, artificial neural
network (ANN) and the GenSoFNN-TVR(S) network
Baronti and Starita, Naive Bayes (NB) 69.4
2007 [9] C4.5 70
NN 78
XCS 79
Hypothesis classifier systems (HCS) 83.8
Xie et al. 2010 [17] Improved F-score and sequential forward floating 100 (best)
search (IFSFFS)
SVM 97.58
(avg)
Exarchos et al. 2011 Bayesian networks (BNs) 69.6
[10] ANN 66.1
SVM 69.6
Decision trees (DTs) 66.1
Random forests (RFs) 58.9
Chang et al. 2013 [11] 1. Pearson’s correlation coefficient (CC) and 93.81 %
Relief-F as the filter approach
2. Genetic algorithm (GA) as the wrapper approach
3. CC-GA and ReliefF-GA as the hybrid approach.
Hybrid model of ReliefF-GA-ANFIS
Calle-Alonso Combines pairwise comparison, Bayesian 97.74
et al. 2013 [18] regression and K-NN
This study FS on erythemato squamous disease data set, combine 98.64
1. CorrelationAttributeEval and Ranker
2. CfsSubsetEval and LinearForwardSelection
(forward)
3. SMO- Poly Kernel (E-1.0)
This study FS on oral cancer data set, combine 96.19
1. CorrelationAttributeEval and Ranker
2. CfsSubsetEval and LinearForwardSelection
(forward)
3. SMO- Poly Kernel (E-1.0)
830 F. Mohd et al.

77.6 Conclusion

In this study, it is noted that a diagnostic model based on Support Vector Machines
(SVM) with a proposed hybrid feature selection method to diagnose the stage of
oral cancer showed an increased of classification accuracy. The hybrid feature
selection method, named HCELFS, combines Correlation Attribute Evaluator
which acts as a filter and SBFS which acts as the wrapper to select the ideal feature
subset from the remaining features.
The experimental results with oral cancer data sets demonstrate that the new
hybrid feature selection method has a better performance than well-known feature
selection algorithms. It obtained optimal classification accuracy with 14 features
from a set of 25 features. The optimal feature subset obtained were then trained
with various data mining algorithms such as Naive Bayes (NB), Multilayer Per-
ceptron (MLP), K-Nearest neighbors (KNN), and SVM to diagnose the stage of
oral cancer. One direction for future studies is to consider proposing a hybrid
algorithm with various dataset and other data mining classifier.

Acknowledgments This study has been supported in part the Exploratory Research Grant
Scheme (ERGS) 600_RMI/ERGS 5/3 (3/2011) under the Malaysia Ministry of Higher Education
(MOHE) and Universiti Teknologi MARA (UiTM) Malaysia. The authors would like to
acknowledge all contributors, technical members at Hospital Universiti Sains Malaysia (HUSM)
who have helped and greatly assisted in the completion of the study. Dr. Zailani Abdullah, data
mining expertise from SIAM, UMT and anonymous reviewers of this paper. Their useful com-
ments have played a significant role in improving the quality of this work.

References

1. Alpaydin, E.: Introduction to machine learning. MIT Press, Cambridge (2004)

2. Neville, B., Damm, D., Allen, C., Bouguot, J.: Oral and maxillofacial pathology. Saunders
Elsevier, St. Louis (2009)
3. Poolsawad, N., Kambhampati, C., Cleland, J.G.F.: Feature selection approaches with missing
values handling for data mining—a case study of heart failure dataset. World Acad. Sci. Eng.
Technol. 60, 828–837 (2011)
4. Aruna, S., Rajagopalan, S.P., Nandakishore, L.V.: Knowledge based analysis of various
statistical tools in detecting breast cancer. In: First International Conference on Computer
Science, Engineering and Applications (CCSEA), pp. 37–45. Chennai, India (2011)
5. Jaganathan, P., Rajkumar, N., Kuppuchamy, R.: A comparative study of improved F-score
with support vector machine and RBF network for breast cancer classification. Int. J. Mach.
Learn. Comput. 2, 741–745 (2012)
6. Maldonado, S., L’Huillier, G.: SVM-based feature selection and classification for email
filtering. In: Latorre Carmona, P., Sánchez, J.S., Fred A.L.N. (eds.) Pattern Recognition—
Applications and Methods, pp. 135–148. Springer, Heidelberg (2013)
7. Kawazu, T., Araki, K., Yoshiura, K., Nakayama, E., Kanda, S.: Application of neural
networks to the prediction of lymph node metastasis in oral cancer. Oral Radiol. 19, 35–40
(2003)
8. Baronti, F., Colla, F., Maggini, V., Micheli, A., Passaro, A., Rossi, A.M.: Experimental
comparison of machine learning approaches to medical domains: a case study of genotype
77 A Hybrid Selection Method Based on HCELFS and SVM … 831

influence on oral cancer development. In: European Conference on Emergent Aspects in

Clinical Data Analysis (EACDA), pp. 81–86. Pisa, Italy (2005)
9. Baronti, F., Starita, A.: Hypothesis testing with classifier systems for rule-based risk
prediction evolutionary computation. Mach. Learn. Data Min. Bioinf. 4447, 24–34 (2007)
10. Exarchos, K., Goletsis, Y., Fotiadis, D.: Multiparametric decision support system for the
prediction of oral cancer reoccurrence. IEEE Trans. Inf. Technol. Biomed. 16(6), 1127–1134
(2011)
11. Chang, S.W., Abdul Kareem, S., Merican, A., Zain, R.: Oral cancer prognosis based on
clinicopathologic and genomic markers using a hybrid of feature selection and machine
learning methods. BMC Bioinf. 14, 170 (2013)
12. Hall, M.A.: Correlation-Based Feature Selection for Machine Learning. The University of
Waikato, Hamilton, New Zealand (1999)
13. Bakar, Z.A., Mohd, F., Noor, N.M.M., Rajion, Z.A.: Demographic profile of oral cancer
patients in east coast of peninsular Malaysia. Int. Med. J. 20, 362–364 (2013)
14. Mohd, F., Bakar, Z.A., Noor, N.M.M., Rajion, Z.A.: Data preparation for pre-processing on
oral cancer dataset. In: 13th International Conference on Control, Automation and Systems
(ICCAS), p. 324. Gwangju, Korea (2013)
15. Razak, A.A., Saddki, N., Naing, N.N., Abdullah, N.: Oral cancer presentation among Malay
patients in hospital Universiti Sains Malaysia, Kelantan. Asian Pac. J. Cancer Prev 10,
1131–1136 (2009)
16. Tung, W.L., Quek, C.: GenSo-FDSS: a neural-fuzzy decision support system for pediatric all
cancer subtype identification using gene expression data. Artif. Intell. Med. 33, 61–88 (2005)
17. Xie, J., Xie, W., Wang, C., Gao, X.: A novel hybrid feature selection method based on
IFSFFS and SVM for the diagnosis of erythemato-squamous diseases. J. Mach. Learn. Res.
Workshop Conf. Proc. 11, 142–151 (2010)
18. Calle-Alonso, F., Pérez, C.J., Arias-Nicolás, J.P., Martín, J.: Computer-aided diagnosis
system: a Bayesian hybrid classification method. Comput. Methods Programs Biomed. 112,
104–113 (2013)
Chapter 78
A Linear Assignment Method of Simple
Additive Weighting System in Linear
Programming Approach Under Interval
Type-2 Fuzzy Set Concepts for MCDM
Problem

Nurnadiah Zamri and Lazim Abdullah

Abstract The ranking phase is valuable to examines the final alternative rankings
of decision making problems. Based on simple additive weighting (SAW) and
linear programming (LP) within the context of interval type-2 fuzzy sets (IT2 FSs),
we develop a linear assignment method to produce the final ranking order of all
alternatives for interval type-2 fuzzy TOPSIS (IT2 FTOPSIS) method. A numer-
ical example is used to check the efficiency and applicability of the proposed
method. The results shows consistent outcomes of the decision making process.
Thus, the proposed method offers an alternative, user-friendly method that is
robust in the decision making framework.

78.1 Introduction

Ranking phase is the step to examines the results of decision making problems.
The interpretation of multiple attribute decision making (MADM) results can show
the differences in the rankings of the alternatives. It was extensively applied and
strengthened the theoretical part of aggregating phase by many authors. A few of
them were; Gao et al. [1] developed a fuzzy approach based on the Technique for
Order Preference by Similarity to Ideal Solution (TOPSIS), where in the ranking
phase, the distances of each alternative from the fuzzy positive ideal solutions
(PIS) and the fuzzy negative ideal solutions (NIS) are computed respectively with

N. Zamri (&)  L. Abdullah

School of Informatics and Applied Mathematics, University Malaysia Terengganu,
21030 Kuala Terengganu, Terengganu, Malaysia
e-mail: [email protected]
L. Abdullah
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 833

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_78
834 N. Zamri and L. Abdullah

a vertex method. Then, a closeness coefficient is obtained to rank order of all

alternatives. Li [2] constructed nonlinear-programming models on the basis of the
concepts of the relative-closeness coefficient and the weighted-Euclidean distance.
Simpler auxiliary nonlinear-programming models were further deduced to calcu-
late relative-closeness of intuitionistic fuzzy (IF) sets of alternatives to the inter-
val-valued intuitionistic fuzzy-positive ideal solutions (IVIF-PIS), which can be
used to generate the ranking order of alternatives. Jolai et al. [3], proposed the goal
programming (GP) technique, and constructed a multi-objective mixed integer
linear programming (MOMILP) model to determine the order quantities of each
selected supplier for each product in each period.
Furthermore, for interval type-2 (IT2) fuzzy approach based on the TOPSIS,
Chen and Lee [4] proposed a ranking value method to cumulative all the collective
decisions and obtained the relative closeness through the traditional TOPSIS
method computing process. However, the used of standard deviation in the ranking
value method is believed influenced by extreme scores and the method is depended
only on the dispersion’s data. It is proved by Chen et al. [5], that the ranking value
proposed by Chen and Lee [4] was difficult and higher in computational volume.
Since that, various authors discussed on the ranking phase of IT2 FTOPSIS
method. For example, Chen et al. [5] proposed a new method MADM based on the
proposed ranking method of IT2FSs. Wang et al. [6] developed IT2 fuzzy
weighted arithmetic averaging operator to aggregate all individual IT2 fuzzy
decision matrices provided by the decision-makers (DMs) into the collective IT2
fuzzy decision matrix, then utilized the ranking-value measure to calculate the
ranking value of each attribute value and constructed the ranking-value matrix of
the collective IT2 fuzzy decision matrix. Chen and Wang [7] presented a new
fuzzy ranking method based on the a-cuts of interval type-2 fuzzy sets (IT2 FSs).
Chen [8] developed a new linear assignment method to produce an optimal
preference ranking of the alternatives in accordance with a set of criterion-wise
rankings and a set of criterion importance within the context of interval type-2
trapezoidal fuzzy numbers (IT2TrFNs) for MADM problems. However, little
research has been conducted on the simple additive weighting (SAW) and linear
programming (LP) for coping with IT2FSs. This linear assignment method (SAW
and LP with IT2FSs method) is developed to handle the ranking phase of IT2
FTOPSIS method.
Thus, the purpose of this paper is to extend the SAW and LP methods in IT2FSs
approach for ranking phase of IT2FTOPSIS. This paper proposes the linear
assignment method with the identification of the SAW and LP methods to
determine the final ranking orders respectively, for each pair of alternatives. The
feasibility and the applicability of the proposed methods are illustrated using the
MADM examples of Chen [9].
This paper is illustrated as follows. Section 78.2 discusses the concept of
weighted average with linear programming. Section 78.3 proposes a linear
assignment method based on SAW and LP methods in IT2 FSs concept.
Section 78.4 illustrates a numerical example in order the check the efficiency of
the proposed method. Finally, Sect. 78.5 presents the conclusions.
78 A Linear Assignment Method of Simple Additive Weighting System … 835

78.2 Weighted Average With Linear Programming

In the following, we recall basic notations and definitions of weighted average

with linear programming.
Definition 78.1 [10, 11] The minimum and maximum for the fuzzy weighted
average for each given aj can be obtained by solving the following two fractional
programming problems:

w1 a1 þ w2 a2 þ    þ wn an
min fL ¼
w1 þ w2 þ    þ wn ð78:1Þ
s:t: ci  wi  di ; i ¼ 1; 2; . . .; n;

w1 b1 þ w2 b2 þ    þ wn bn
max fU ¼
w1 þ w2 þ    þ wn ð78:2Þ
s:t: ci  wi  di ; i ¼ 1; 2; . . .; n;

where ci and di are the two end points of the interval wi for a given aj level cut.

The Charnes and Cooper’s linear transformation is summarized in the follow-

ing. Consider the following simple fractional programming problem:
px
min
qx ð78:3Þ
s:t: Ax  b; x  0;

where p and q are two n-dimensional constant vectors, x is the n-dimensional

variable, A is an m  n matrix, and b is an m-dimensional constant vector.
To transform the above fractional programming problem into a linear problem,
let

1
z¼ and zx ¼ y; ð78:4Þ
qx

where we assume that qx 6¼ 0: Multiplying both the objective function and the
constraints by z and using the definitions given in Eq. 78.5, we obtain

min py
ð78:5Þ
s:t: Ay  bz; qy ¼ 1; y  0; z  0;

which is a linear programming problem.

This weighted average with linear programming is being used in defining a
linear assignment method. Thus, the development of the proposed model is
described in Sect. 78.3.
836 N. Zamri and L. Abdullah

78.3 The Proposed Method

Original SAW method and LP are modified into an IT2 FS manner. Modifications
are made to accommodate the objective of the research and also to simplify the
computational procedure without losing the novelty of SAW and LP. The proposed
method is then applied into a linear assignment method for IT2 FTOPSIS (MCDM
method) to get the optimal preference ranking. This proposed method is believed
to be more flexible rather than the existed model due to the fact that it is used the
IT2 FS. On the other hand, this model is more suitable to represent uncertainties
because it is involve end-users into the whole weighting process. Thus, suppose an
IT2 FTOPSIS has n alternatives ðA1 ; . . .; An Þ and m decision criteria/attributes
ðC1 ; . . .; Cm Þ. Each alternative is evaluated with respect to the m criteria/attributes.
All the values/ratings assigned to the alternatives with respect to each criterion
from a decision matrix, denoted by S ¼ ðyij Þnm , and the relative weight vector
Pm
about the criteria, denoted by W ¼ ðw1 ; . . .; wm Þ, that satisfying j¼1 wj ¼ 1:
Therefore, the rest of the general process of this proposed method is listed as
follows:
Rating State: In this state, all the matrices are transformed into the IT2 FS
concept.
Step 1: Establish a decision matrix and weight matrix
Establish an IT2 decision matrix and IT2 fuzzy weight matrix.
Step 2: Comparable Scale
Construct a comparable scale for all elements in the decision matrix. The
comparable scale is used to divide the outcome of a certain criterion by its
maximum value, provided that the criteria are defined as benefit criteria. There-
fore, the comparable scale is represented as follows:
.   h i
~~r ij ¼ 1 FOU ~~r ij ¼ ~~r ij ; ~~r ij ð78:6Þ
02 3 2 3 1
~~f ~~
~  @ 4 ij 5 4f ij 5A
For positive criteria ~r ij ¼ ;  ð78:7Þ
~~f  ~~
ij f ij
02 min 3 2 31
~~f ~~min
B6 ij 7 4 ij 5Cf
For negative criteria ~~r ij ¼ @4 5; ~ A ð78:8Þ
~
~f ~f
ij ij
78 A Linear Assignment Method of Simple Additive Weighting System … 837

Then the decision matrix can be expressed as follows:

2 C1 C2 . . . Cn 3
x1 ~~r 11 ~~r 12    ~~r 1n
6 ~~r 21 ~~r 22    ~~r 2n 7
D ¼ x2 6 7 ð78:9Þ
.. 6 .. .. .. .. 7
. 4 . . . . 5
xm ~~r m1 ~~r m2    ~~r mn

where ~
~r mn is the comparable scale value in the decision matrix.
Weighting State: Modification of the existed SAW method with ‘‘modified
SAW in IT2 FS concepts’’.
Step 3: Weight of attributes of SAW
Construct the weighting matrix Wp using the SAW formulae of the attributes of

the decision-maker and construct the pth average weighting matrix W.
Step 4: Weighted decision matrix
Construct the weighted decision matrix.
Aggregation State: Upgrading the calculation of separations of each alter-
native with linear programming concepts.
Step 5: Positive ideal solution and negative ideal solution
Determine the matrices that include positive and negative ideal solutions.
Step 6: Construct the separation of each alternative of SAW by linear pro-
gramming approach
Calculate the separation of each alternative from the positive ideal solution I 
and negative ideal solution I  using the formulae as follows:
Pn h ~ 
~~j 1  d~
i
~~ j ~bij þ q ~ij
  j¼1 w
Ci bij ; dij ; s ¼ P h  i P h    i þ sl þ su
n ~ ~
~
~ j bij þ q ~ ~
~
~j 1  d ij þ j¼1 w n ~ ~
~ ~j d~
~
~ j 1  bij þ q ~ij
j¼1 w
Pn h   i
~~ j ~~bij þ q ~~j 1  d~~ij
j¼1 w
¼ Pn   þ sl þ su
w~~ j þ q~~j
j¼1

ð78:10Þ

and let

1
z ¼ Pn   ð78:11Þ
~~j
~~ j þ q
j¼1 w
838 N. Zamri and L. Abdullah

Assigned the value for

~~tj ¼ zw
~~ j ð78:12Þ

and

~~j
~~yj ¼ zq ðj ¼ 1; 2; . . .; nÞ ð78:13Þ

Since

1 1 Xn  ~ ~~j

z ¼ Pn   ) ¼ ~j þ q
w ð78:14Þ
w ~~j
~~ j þ q z j¼1
j¼1

and

~~
~~tj ¼ zw
~~ j ) ~~ j ¼ tj
w ð78:15Þ
z
~~yj
~~j
~~yj ¼ zq ) ~~j ¼
q ð78:16Þ
z

Thus, based on the above Charnes and Cooper’s transformations [12],

Eq. 78.10 can be transformed into the equivalent linear programming models as
follows:
  nXn   o
Ciu ~bij ; d~
~ ~ij ; s ¼ max
j¼1
~~ti bu þ ~~y 1  d l þ sl þ su
ij j ij
8 l
> zw ~ ~
~
~ j  t j  zw~ u
~ j ðj ¼ 1; 2; . . .; nÞ
>
>
>
>
>
> ~ l ~
~j  ~yj  zq ~~u ðj ¼ 1; 2; . . .; nÞ
> zq
> j
>
< Pn ~  ð78:17Þ
~
t j þ ~~yj ¼ 1
s:t: j¼1
>
>
>
> z0
>
>
>
> l
s ¼ n ðn ¼ 0; . . .; 1Þ
>
>
:
su ¼ n ðn ¼ 0; . . .; 1Þ

and
78 A Linear Assignment Method of Simple Additive Weighting System … 839

  nXn   o
Cil ~~
bij ; d~
~ij ; s ¼ max ~~ti bl þ ~~y 1  du þ sl þ su
j¼1 ij j ij
8 l
> zw ~ ~
~  ~t  zw ~~ ðj ¼ 1; 2; . . .; nÞ
u
>
> j j j
>
>
>
> z ~
q
~ l
 ~~yj  zq~~u ðj ¼ 1; 2; . . .; nÞ
>
> j j
>
< Pn ~  ð78:18Þ
s:t: ~ ~yj ¼ 1
j¼1 tj þ ~
>
>
>
> z0
>
>
>
> s l
¼ n ð n ¼ 0; . . .; 1Þ
>
>
: u
 s ¼ n ðn ¼ 0; . . .; 1Þ  
where Ci bij ; dij ; s is an IT2 FS, denoted by Cil ; Ciu .
Step 7: Define the closeness coefficient
Calculate the relative degree of closeness to the ideal solution for each
alternative.
Ranking State:
Step 8: Rank all alternatives
Sort the values of ðCC Þi in a descending sequence, where 1  j  n: The
larger the value of ðCC Þi , the higher the preference of the alternative for ðCC Þi .
In this section, we have successfully introduced a new concept of linear
assignment method. In order to check the efficiency of the proposed method, a
numerical example is provided in Sect. 78.4 to illustrate the proposed method.

78.4 Illustrative Example

In this section, we used an example from Chen [9] to illustrate the proposed
method. This numerical example is used to test the ability of the proposed method
to handle the IT2 MCDM problems in many areas. All the relative importance
weights in this numerical example are described using the linguistic variables
which are defined in Table 78.1.

Table 78.1 Linguistic Linguistic variable Interval type-2 fuzzy number

variables for the relative (IT2FN)
importance weights
of criteria Very low (VL) ((0,0.1;1), (0,0.5;1))
Low (L) ((0,0.3;1), (0.05,0.2;1))
Medium low (ML) ((0.1,0.5;1), (0.2,0.4;1))
Medium (M) ((0.3,0.7;1), (0.4,0.6;1))
Medium high (MH) ((0.5,0.9;1), (0.6,0.8;1))
High (H) ((0.7,1.0;1), (0.8, 0.95;1))
Very high (VH) ((0.9,1.0;1), (0.95,1.0,1))
840 N. Zamri and L. Abdullah

Table 78.2 Linguistic Linguistic variable Interval type-2 fuzzy number

variables for the ratings (IT2FN)
of criteria
Very poor (VP) ((0,1;1), (0,0.5;1))
Poor (P) ((0,3;1), (0.5,2;1))
Medium poor (MP) ((1,5;1), (2,4;1))
Medium (M)/fair (F) ((3,7;1), (4,6;1))
Medium good (MG) ((5,9;1), (6,8;1))
Good (G) ((7,10;1), (8,9.5;1))
Very good (VG) ((9,10;1), (9.5,10,1))

Table 78.3 Weights of the Attributes Decision-makers

attributes evaluated
by decision-makers D1 D2 D3
C1 H VH MH
C2 VH VH VH
C3 VH H H
C4 VH VH VH
C5 M MH MH

Moreover, all the relative importance ratings (i.e. the criteria values) in this
numerical example are described using the linguistic variables which are defined
in Table 78.2.
Assume that there are three decision-makers, D1, D2, and D3 of a software
company to hire a system analysis engineer and assume that there are three
alternatives x1, x2, x3 and five attributes ‘‘Emotional Steadiness’’ (denoted by C1),
‘‘Oral Communication Skill’’ (denoted by C2), ‘‘Personality’’ (denoted by C3),
‘‘Past Experience’’ (denoted by C4), ‘‘Self-Confidence’’ (denoted by C5). Let X be
the set of alternatives, where X ¼ fx1 ; x2 ; x3 g, and let F be the set of attributes,
where F = {Emotional Steadiness, Oral Communication Skill, Personality, Past
Experience, Self-Confidence}. Assume that there are three decision-makers D1,
D2, and D3 used the linguistic terms shown in Table 78.1 to represent the weights
of the four attributes, respectively, as shown in Table 78.3.
Then theses three decision-makers D1, D2, and D3 used the linguistic terms
shown in Table 78.2 to represent the evaluating values of the alternatives with
respect to different attributes, respectively, as shown in Table 78.4.
Using the linguistic scales from Tables 78.3 and 78.4, and the eight steps of the
proposed method (in Sect. 78.3), results for Chen [9] example is shown in
Table 78.5. Table 78.5 shows the min value and max value from Step 6 and
calculates the closeness coefficient ðCC Þi for each of alternatives.
78 A Linear Assignment Method of Simple Additive Weighting System … 841

Table 78.4 Linguistic Attributes Alternatives Decision-makers

of decision matrix
D1 D2 D3
C1 x1 MG G MG
x2 G G MG
x3 VG G F
C2 x1 G MG F
x2 VG VG VG
x3 MG G VG
C3 x1 F G G
x2 VG VG G
x3 G MG VG
C4 x1 VG G VG
x2 VG VG VG
x3 G VG MG
C5 x1 F F F
x2 VG MG G
x3 G G MG

Table 78.5 Final ranking Min Max Closeness coefficient,

order ðCC Þi
x1 2.3575 2.6364 0.4721
x2 2.3835 2.5944 0.4788
x3 2.3668 2.6216 0.4745

As shown in Table 78.5, results for the relative closeness of Chen’s method [9] for
three alternatives are 0.4721 for x1, 0.4788 for x2 and 0.4745 for x3; which lead to the
ranking of x2 [ x3 [ x1. Chen [9]’s result coincides with the proposed results.

78.5 Conclusion

This paper distributed a linear assignment method which consisted with the SAW
and LP method for IT2 FTOPSIS. This proposed method is able to produce an
optimal ranking order of the alternatives. Besides, we provided a numerical
example to analyze the applicability of the proposed method. The proposed
method can capture the imprecise and uncertain decision information instead of
the optimal ranking orders. Furthermore, the proposed method offers an alternative
ways of ranking phase for IT2 FTOPSIS method.
842 N. Zamri and L. Abdullah

Acknowledgments This research is supported by MyBrain15 scholarship and Fundamental

Research Grant, no. 59243. This support is gratefully acknowledged.

References

1. Gao, P., Feng, J., Yang, L.: Fuzzy TOPSIS algorithm for multiple criteria decision making
with an application in information systems project selection. In: Proceeding of IEEE WiCOM
‘08. 4th International Conference on Wireless Communications, Networking and Mobile
Computing, pp. 1–4 (2008)
2. Li, D.: TOPSIS-based nonlinear-programming methodology for multiattribute decision
making with interval-valued intuitionistic fuzzy sets. IEEE Trans. Fuzzy Syst. 18, 299–311
(2010)
3. Jolai, F., Yazdian, S.A., Shahanaghi, K., Khojasteh, M.A.: Integrating fuzzy TOPSIS and
multi-period goal programming for purchasing multiple products from multiple suppliers.
J. Purchas. Suppl. Manage. 17, 42–53 (2012)
4. Chen, S.-M., Lee, L.-W.: Fuzzy multiple attributes group decision-making based on the
interval type-2 TOPSIS method. Expert Syst. Appl. 37, 2790–2798 (2010)
5. Chen, S.-M., Yang, M.-W., Lee, L.-W., Yang, S.-W.: Fuzzy multiple attributes group
decision-making based on ranking interval type-2 fuzzy sets. Expert Syst. Appl. 39,
5295–5308 (2012)
6. Wang, W., Liu, X., Qin, Y.: Multi-attribute group decision making models under interval
type-2 fuzzy environment. Knowl.-Based Syst. 30, 121–128 (2012)
7. Chen, S.-M., Wang, C.-Y.: Fuzzy decision making systems based on interval type-2 fuzzy
sets. Inf. Sci. 242, 1–21 (2013)
8. Chen, T.-Y.: A linear assignment method for multiple-criteria decision analysis with interval
type-2 fuzzy sets. Appl. Soft Comput. 13, 2735–2748 (2013)
9. Chen, C.T.: Extension of the TOPSIS for group decision making under fuzzy environment.
Fuzzy Sets Syst. 114, 1–9 (2000)
10. Charness, A.A., Cooper, W.W.: Programming with linear fractional functionals. J. Naval Res.
Logist. Quart. 9, 181–185 (1962)
11. Charness, A.A., Cooper, W.W.: An explicit general solution in linear fractional
programming. J. Naval Res. Logist. Quart. 20, 181–185 (1973)
12. Guh, Y.-Y., Hon, C.-C., Lee, E.S.: Fuzzy weighted average: the linear programming
approach via charnes and cooper’s rule. J. Fuzzy Sets Syst. 117, 157–160 (2001)
Chapter 79
Hybridization Denoising Method
for Digital Image in Low-Light Condition

Suhaila Sari, Sharifah Zahidah Hasan Al Fakkri, Hazli Roslan

and Zarina Tukiran

Abstract A good noise reduction is a method that can reduce the noise level and
preserve the details of the image. This paper proposes a denoising method through
hybridization of bilateral filters and wavelet thresholding for digital image in low-
light condition. The proposed method is experimented on selected night vision
images and the performances are evaluated in terms of Mean Squared Error (MSE),
Peak Signal to Noise Ratio (PSNR) and visual effects. Results demonstrate that the
proposed denoising method has improved the PSNR and MSE of average perfor-
mance of bilateral filter by 0.97 dB and 1.33 respectively and the average perfor-
mance of wavelet thresholding has improved by 0.98 dB and 1.19 respectively.


Keywords Image denoising Edge preservation  Low-light condition  Poisson
 
noise Bilateral filter Wavelet thresholding

79.1 Introduction

Low-light noise is a significant problem in photography. Most consumers’ cameras

have poor low-light characteristics, which typically result in images with noticeable
noise artifacts. For example, digital images taken by consumer’s digital camera

S. Sari (&)  S.Z.H. Al Fakkri  H. Roslan  Z. Tukiran

Artificial Intelligence and Computer Vision Focus Group (AICOV), Faculty of Electrical
and Electronic Engineering, Universiti Tun Hussein Onn Malaysia, Batu Pahat, Malaysia
e-mail: [email protected]
S.Z.H. Al Fakkri
e-mail: [email protected]
H. Roslan
e-mail: [email protected]
Z. Tukiran
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 843

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_79
844 S. Sari et al.

may suffer from thermal sensor noise demosaicing noise and quantization noise [1].
Although camera technology has been improved extensively, noise still could not
be eliminated completely [2].
In image denoising process, noise is undesired information that contaminates
the image [3]. The information about the type of noise corrupting in the original
image plays a significant role to determine the denoising technique. Typical
images are corrupted with noise modeled as either or combination of the Gaussian
noise, Salt and Pepper noise, Speckle noise, or Poisson noise.
Many denoising methods have been proposed over the years, such as the
Wiener filter, wavelet thresholding [4, 5], anisotropic filtering [6], bilateral filter
[7], total variation method [8], and non-local means methods [9]. Among these,
wavelet thresholding has been reported to be a highly successful method. The
wavelet thresholding crucial task is selection of threshold value. The wavelet
decomposed a signal into low-frequency and high-frequency subbands, and the
coefficients in the detail subbands are processed via hard or soft thresholding [10].
Despite its success, this method, however, experienced a problem on preserving
the edge.
The bilateral filter was proposed by Tomasi and Manduchi [7] as an alternative
to wavelet thresholding. It applies spatially weighted averaging. This is achieved
by combining two Gaussian filters; one filter works in spatial domain, the other in
the intensity domain. Therefore, not only the spatial distance, but the intensity
distance is also important for the determination of weights [11]. Hence, these types
of filters can remove the noise in an image effectively [12].
Combination of bilateral filter and wavelet thresholding was applied by Zhang
and Guntruk to eliminate noise in real noisy images [12]. The similar method is
also used by Roy, Sinha and Sen in order to eliminate noise in medical image [13].
The noise modeled in [12, 13] is Gaussian noise. Through these researches, the
method has proven significant improvement in noise reduction for images cor-
rupted by Gaussian noise.
Since the Gaussian noise distribution has some similar characteristic with
Poisson noise [14], the method is expected to provide effective noise reduction for
Poisson noise as well. Therefore, in this work, the method is proposed to reduce
the Poisson noise on digital image captured in low-light condition.
The proposed method is based on bilateral filter and wavelet thresholding which
exploits features of both bilateral filter and wavelet thresholding at the same time
the limitations of both are overcome. To our knowledge, such studies in hybrid-
ization of bilateral filtering and wavelet thresholding for Poisson noise reduction in
digital image have not yet been done.
The paper is organised as follows. The proposed hybrid denoising method is
described in Sect. 79.2. Results are discussed in Sect. 79.3. Finally, concluding
remarks are drawn in Sect. 79.4.
79 Hybridization Denoising Method for Digital Image … 845

79.2 Proposed Method

In this work, hybridization of bilateral filter and wavelet thresholding is proposed to

find the best solution in image denoising especially for digital image in low light
condition. This proposed method utilizes the advantages of both bilateral filter and
wavelet thresholding. Therefore the proposed method is expected to contribute a
better preservation of fine details and edges in the image. While applying the
threshold rule, the important features like edges, curves and textures can be identified.

79.2.1 Wavelet Thresholding

The wavelets play a major role in image compression and image denoising.
Wavelet coefficients calculated by a wavelet transform represent change in the
time series at a particular resolution. It is then possible to filter out noise by
considering the time series at various resolutions. The term wavelet thresholding is
explained as decomposition of the data or the image into wavelet coefficients,
comparing the detail coefficients with a given threshold value, and shrinking these
coefficients close to zero to take away the effect of noise in the data.
During thresholding, a wavelet coefficient is compared with a given threshold and
is set to zero if its magnitude is less than the threshold; otherwise, it is retained or
modified depending on the threshold rule. Thresholding distinguishes between the
coefficients due to noise and the ones consisting of important signal information.
There are two general categories of thresholding which are hard thresholding
and soft thresholding. Hard thresholding is a ‘‘keep or kill’’ procedure and is more
intuitively appealing while soft thresholding shrinks coefficients above the
threshold in absolute value. Wavelet threshold is effective in reducing certain
amount of noise while preserving fine details and edges.

79.2.2 Bilateral Filter

The bilateral filter is a nonlinear weighted averaging filter. The weights depend on
both the spatial distance and the intensity distance with respect to the center pixel.
It smoothes images while preserving edges with the nonlinear combination of
nearby pixel values. This can be achieved by combining two Gaussian filters; one
filter works in spatial domain, the other filter works in intensity domain. Therefore,
not only the spatial distance but the intensity distance is also important for the
determination of weights.
The weakness of the bilateral filter is its inability to remove salt and pepper type
of noise. It could not access to the different frequency components of a signal
which fails to remove low frequency noise. Besides that, there is no study on the
846 S. Sari et al.

optimal values of rd and rr, which are the parameters that control the behavior of
the bilateral filter. Bilateral filter is known to have good performance in noise
reduction, but tend to over smooth the fine details and edges.

79.2.3 Hybridization of Bilateral and Wavelet Thresholding

This work proposed hybridization of bilateral and wavelet thresholding (db4,

Bayes soft thresholding, level of decomposition = 1) for digital image in low-light
condition. The process flow for the proposed method is shown in Fig. 79.1.

79.3 Results and Discussion

We have conducted some tests to the selected night vision images of well-known
landmark in Malaysia named as Kuala Lumpur Conventional Center in Kuala
Lumpur (klcc), Dataran Bandaraya Johor Clock Tower in Johor (jbTower) and
Dewan Undangan Negeri Kuching in Sarawak (kuchingHall). The image size of
these images is 250 9 250. In order to conduct the testing, these images are
corrupted with Poisson noise. All of these images are tested with conventional
methods which are bilateral filtering and wavelet threshold, as well as proposed
hybrid denoising method.
To do a quantitative comparison, MATLAB software is used to simulate the test
images under different Poisson distribution, k = 0.00001, k = 0.001, and k = 0.1
which represents low, medium and high noise, respectively. From the test, there
are three evaluation methods that we utilized, which are (a) Mean Squared Error
(MSE), (b) Peak Signal to Noise Ratio (PSNR) and (c) visual comparison.
Equation (79.1) is used to calculate the MSE where A represents the original
image while B represents the reconstructed image. Number of rows and columns in
the input images are represented by M and N respectively.

1 M X1 X
N 1
MSE ¼ ½Aði; jÞ  Bði; jÞ2 ð79:1Þ
M  N i¼0 j¼0

Meanwhile Eq. (79.2) is used to calculate the PSNR where MAX is the maxi-
mum possible pixel value of the image.
  
PSNR ¼ 10  log10 MAX 2 MSE ð79:2Þ

The MSE and PSNR values obtained for all methods using all images stated in
this section are shown in Tables 79.1 and 79.2 respectively.
79 Hybridization Denoising Method for Digital Image … 847

Bilateral Inverse Discrete

filter Wavelet Transform

Discrete Apply Thresholding

Wavelet
Noisy Image Denoised Image
Transform

Fig. 79.1 Block diagram of image denoising using hybridization of bilateral filter and wavelet
thresholding

Table 79.1 MSE evaluation Test images Bilateral Wavelet Proposed

for high Poisson noise filter filter method
(k = 0.1)
klcc 0.11 0.10 0.08
jbTower 0.88 0.07 0.60
kuchingHall 0.10 0.09 0.08

Table 79.2 PSNR evaluation Test images Bilateral Wavelet Proposed

for high Poisson noise filter filter method
(k = 0.1)
klcc 67.14 68.32 70.04
jbTower 70.36 71.07 72.26
kuchingHall 68.62 69.42 70.66

MSE evaluation for high Poisson noise (λ =0.1)

Bilateral Filter Wavelet Filter Proposed Method
0.11
0.10 0.10
0.09
0.08 0.08
0.07 0.08
0.06

klcc jbtower kuchinghall

Fig. 79.2 Performance comparisons in terms of MSE

Figures 79.2 and 79.3 depict the graphical comparative performance in terms of
MSE and PSNR. The results show that the proposed method has significantly
better denoising performance in comparison to bilateral filter alone and wavelet
thresholding alone.
848 S. Sari et al.

PSNR evaluation for high Poisson noise (λ =0.1)

Bilateral Filter Wavelet Filter Proposed Method
72.76

71.07
70.36 70.66
70.04
69.42
68.32 68.61
67.14

klcc jbtower kuchinghall

Fig. 79.3 Performance comparisons in terms of PSNR

Original Noisy Image Bilateral filter Proposed Method

Fig. 79.4 The visual effects comparison of Kuching Hall image for (top row) low Poisson noise
level with k = 0.00001, (middle row) medium Poisson noise level with k = 0.001 and (bottom
row) high Poisson noise level with k = 0.1

Figure 79.4 shows the visual effects comparison of denoising result for different
night vision images corrupted by low, medium and high Poisson noise level.
Figure 79.5 shows the visual effects in comparison of all methods using all images
with high Poisson noise.
79 Hybridization Denoising Method for Digital Image … 849

Fig. 79.5 The visual effects comparison of klcc image (above) and jbTower image (bottom) for,
a Original image, b Noisy image with high Poisson noise level with k = 0.1, c Denoised image
using bilateral filter, d Denoised image using wavelet thresholding and e Denoised image using
hybridization of bilateral filter and wavelet thresholding

The visual effects results show that although the output images from bilateral
filter looks smooth and clear, but most of the fine details and edges are overs-
moothed and the image blurred. Therefore, by applying wavelet thresholding, it
multiplies the adjacent wavelet sub bands and strengthens the significant features
in the image which indicates that our proposed method is better in image
denoising.

79.4 Conclusion

In this paper, image denoising using hybridization, which integrates bilateral filter
and wavelet thresholding is performed. From the result, we can conclude that the
proposed method has provided better denoising for images in low-light condition
corrupted by high level Poisson noise. The performance of the proposed method
has improved the average performances of bilateral filter and wavelet thresholding
in terms of the PSNR by 0.98 dB and 0.97 dB, respectively. On the other hand, the
performance of the proposed method has improved the average performances of
wavelet thresholding and bilateral filter in terms of the MSE by 1.19 and 1.33,
respectively. As for the visual effects, the proposed method provides better
denoising while preserving the fine details and edges.

Acknowledgements The authors would like to thank Universiti Tun Hussein Onn Malaysia
(UTHM) (Grant vote: 1088) and Malaysia Government for the support and sponsor of this study.
850 S. Sari et al.

References

1. Roy, S., Sinha, N., Sen, A.K.: A new hybrid image denoising method. Int. J. Inf. Technol.
Knowl. Manag. 2(2), 491–497 (2010)
2. Chatterjee, P., Joshi, N., Kang, S.B., Matsushita, Y.: Noise suppression in low-light images
joint denoising and demosaicing. In: Proceedings of IEEE Conference on Computer Vision &
Pattern Recognition (CVPR), pp. 321–328 (2011)
3. Teimouri, M., Vahedi, E., Avanaki, A.N., Shahi, Z.H.: An efficient denoising method for
color images. In: Proceedings of IEEE Conference on Signal Processing and Its Applications,
pp. 1–4 (2007)
4. Donoho, D.L., Johnstone, I.M.: Ideal spatial adaptation by wavelet shrinkage. Biometrika
81(3), 425–455 (1994)
5. Ogawa, K., Sakata, M., Li, Y.: Adaptive noise reduction of scintigrams with a wavelet
transform. Int. J. Biomed. Imaging. 2012 (2012). http://dx.doi.org/10.1155/2012/130482
6. Perona, P., Malik, J.: Scale-space and Edge detection using anisotropic diffusion. Proc. IEEE
Trans. Pattern Anal. Mach. Intell. 12(7), 629–639 (1990)
7. Tomasi, C., Manduchi, R.: Bilateral filtering for gray and color images. In: Proceedings of
International Conference on Computer Vision, pp. 839–846 (1998)
8. Rudin, L.I., Osher, S., Fatemi, E.: Nonlinear total variation based noise removal algorithms.
Physica D 60, 259–268 (1992)
9. Buades, A., Coll, B., Morel, J.: Neighborhood filters and PDE’s. Numer. Math. 105(1), 1–34
(2006)
10. Donoho, D.L., Johnstone, I.M.: Adapting to unknown smoothness via wavelet shrinkage.
J. Am. Stat. Assoc. 90(432), 1200–1224 (1995)
11. Chang, S.G., Yu, B., Vetterli, M.: Adaptive wavelet thresholding for image denoising and
compression. Proc. IEEE Trans. Image Process. 9(9), 1532–1546 (2000)
12. Zhang, M., Gunturk, B.: A new image denoising method based on the bilateral filter. In:
ICASSP, IEEE, pp. 929–932 (2008)
13. Roy, S., Sinha, N., Sen, A.K.: An efficient denoising model based on wavelet and bilateral
filters. Int. J. Comput. Appl. 53(10), 0975–8887 (2012)
14. Portilla, J., Strela, V., Wainwright, M., Simoncelli, E.P.: Image denoising using scale
mixtures of gaussians in the wavelet domain. Proc. IEEE Trans. Image Process. 12(2),
1338–1351 (2003)
Chapter 80
The Improved Models of Internet Pricing
Scheme of Multi Service Multi Link
Networks with Various Capacity Links

Fitri Maya Puspita, Kamaruzzaman Seman and Bachok M. Taib

Abstract Internet Service Providers (ISPs) nowadays deal with high demand to
promote good quality information. However, the knowledge to develop new
pricing scheme that serve both customers and supplier is known, but only a few
pricing plans involve QoS networks. This study will seek new proposed pricing
plans are offered with multi service multi link networks involved. The multi ser-
vice multi link Networks scheme is solved as an optimization model by comparing
two models in multi QoS networks. The results showed that by fixing the base
price and varying the quality premium or varying the base price and quality and
setting up the equal capacity link values, ISP achieved the goal to maximize the
profit.

80.1 Introduction

Recent works on multiple QoS networks are due to [1–4]. They described the
pricing scheme based auction to allocate QoS and maximize ISP’s revenue. The
auction pricing scheme is actually scalability, efficiency and fairness in sharing
resources. The solution of the optimization problem goes from single bottleneck
link in the network and then she generalized into multiple bottleneck links using
heuristic method. In this paper, she used only single QoS parameter-bandwidth,
while in networks, there are many parameters affect QoS that can be considered.

F.M. Puspita (&)  K. Seman  B.M. Taib

Faculty of Science and Technology, Universiti Sains Islam Malaysia,
71800 Nilai, Negeri Sembilan Darul Khusus, Malaysia
e-mail: [email protected]
K. Seman
e-mail: [email protected]
B.M. Taib
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 851

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_80
852 F.M. Puspita et al.

Although QoS mechanisms are available in some researches, there are few
practical QoS network. Even recently a work in this QoS network [5], it only
applies simple network involving one single route from source to destination.
In previous discussion works on multiservice network proposed by [6–8], we
work on single and multiple link networks to solve the internet pricing scheme.
This work on multiple link multiservice networks can also be improved by con-
sidering all cases of capacity needed in the networks. The results show our
improved method results in better optimal solution than previously conducted by
other research.
So, we intend to improve the mathematical formulation of [5, 9] to be simpler
formulation by taking into consideration the utility function, base price as fixed
price or varies, quality premium, index performance, capacity and also bandwidth
required by looking at all possibility of capacity needed in the network. Next we
consider the problem of internet charging scheme as Mixed Integer Nonlinear
Programming (MINLP). The solver LINGO 13.0 [10] were applied to assist the
nonlinear programming solution to obtain the optimal solution.

80.2 Related Works

A number of related works has been listed for differentiated pricing scheme that
works on multi QoS networks. In paper proposed by [11], they discussed about
pricing scheme that is based on QoS level in different allocation to control con-
gestion and load balance. Multiple class networks require differentiated pricing
scheme to have allocation of different level of service traffic. The investigation of
the connection between QoS characteristics at network with requirement of quality
for users applying the network is proposed by [12]. The results of their findings
mainly are that predictability, consistency of QoS is crucial, the pricing scheme is
crucial to have QoS to be predicted, having reliable service protocols and new
integrated service mechanism is to present alternative in solving the problem.
There exist direct connection between QoS profile application identification in
packet and users requesting the QoS. Models proposed by [13] viewed the rela-
tionship between congestion control, routing and scheduling of wired network as
fair resource allocation. In the research explained in [14], they discussed the flat
fee pricing scheme, and as the simplicity price is maximum revenue. The draw-
back of the rule is due to nonlinearity and does not reflect the price observed in
reality.
Models proposed by [15] stated that in network, it is assumed that n users can
be split into k categories. Each category can apply the same service offered by
application server in a shared link with total bandwidth Ctot but has different
demand framework and also difference price sensitivity. Alderson et al. [16]
discussed issues related to ISP problems dealing with topology of the networks
such as link cost, router technology which impact on availability of topology to
80 The Improved Models of Internet Pricing Scheme … 853

network creator then dealing with equipment of routing adopted to tackle network
traffic flow.
Other research proposed by [17] stated about ways to solve internet optimi-
zation that includes system definition as an interest function and view it as dif-
ferent via points and system mathematical definition. Problem proposed by [18]
focused on problem of arrangement of web services and explained about model of
multi-dimension QoS. Framework presented by [19] analyzed the pricing problem
in integrated service network having guaranteed QoS. The Method proposed by
[20] define terms for performance prediction of service-base system that consist of
performance showing the how fast completion time to finish a service request, time
interval showing the time period to complete service request, dependability
showing the capability of web service to conduct conditional required function,
price setting up by ISP and reputation showing that user perception to the service.
The method proposed by [21] explained about monopoly in pricing model strategy
based on pay-per-volume and pay-per-time of network. They conclude that ISP
will gain more benefit by providing pricing scheme based volume since this
scheme is an alternative to numerous users and scheme of pay-per volume will
benefit the network provider and can prevent from bursting the networks.
Study on multiservice was investigated by [9] which discussed problem of
pricing of internet by considering network share, availability capacity in each
service, the number of users available for the service and the QoS level. They
solved the internet pricing by transforming the model into optimization model and
solved using Cplex software. Recent works conducted by [22–27] also discussed
internet charging scheme under multiple class QoS networks by comparing two
models that involve base price as a fixed and variable set up by ISP. The model
created by setting up base price as fixed price will yield higher optimal solution if
ISP intended to recover the cost. But if ISP would like to compete in market, then
the choice of model involving base price as variable price would be the best option
to choose.
The work discussed on the botnet attacks detection by using nepenthes hon-
eypots [28] is also crucial to be issued for problems increasing if we are dealing
with security problem in multiservice networks.

80.3 Research Method

We attempt to apply optimization techniques in solving the problem in this paper.

We also consider the optimization problem as MBINLP that can be solved by
using optimization tools, LINGO 13.0. We transform the problem of pricing the
internet in multi service networks into optimization model and attempt to solve it
to get optimal solution. This solution will help us interpreting the current issues
involving pricing, network share, base price, quality premium and also QoS level.
854 F.M. Puspita et al.

80.3.1 Model Formulation

The idea basically generates from [5, 8, 9] and is improved in multilink multi
service networks by considering various cases where we can set up requirements
for the capacity link.

80.3.2 Assumptions

Assume that there is only one single network from source to destination since
concentrate on service pricing scheme. Assume that the routing schemes are
already set up by the ISP. As [1–4] pointed out, we have 2 parts of utility function
namely, base cost which does not depend on resource consumption and cost which
depends on resource consumption. The parameters, decision variables and the
models are adopted in [7, 8] and are described as follows.
The parameters are as follows.
aj base price for class j, can be fixed or variables
bj quality premium of class j that has Ij service performance
Cl total capacity available in link l
pil price a user willing to pay for full QoS level service of i in link l

The decision variables are as follows.

xil number of users of service i in link l
ail reserved share of total capacity available for service i in link l
Ii quality index of class i

Formulation when we assign a and b fixed is as follows.

L X
X S
Max ða þ bIi Þpil xil ð80:1Þ
l¼1 i¼1

Such that

Ii dil xil  ail Cl ; i ¼ 1; . . .S; l ¼ 1; . . .; L ð80:2Þ

X
L X
S
Ii dil xil  Cl ; i ¼ 1; . . . S; l ¼ 1; . . .; L ð80:3Þ
l¼1 i¼1

X
L
ail ¼ 1; i ¼ 1; . . .; S ð80:4Þ
l¼1
80 The Improved Models of Internet Pricing Scheme … 855

0  ail  1; i ¼ 1; :::; S; l ¼ 1; :::; L ð80:5Þ

mi  Ii  1; i ¼ 1; :::; S ð80:6Þ

0  xil  ni ; i ¼ 1; . . .; S; l ¼ 1; . . .; L ð80:7Þ

With mi and ni are prescribed positive integer numbers.

fxil ginteger ð80:8Þ

Formulation when we assign a fixed and b vary is as follows.

L X
X S
Max ða þ bi Ii Þpil xil ð80:9Þ
l¼1 i¼1

Subject to Eqs. (80.2–80.8) with additional constraints as follows.

bi Ii  bi1 Ii1 ; i [ 1; i ¼ 1; . . .; S ð80:10Þ

k  bi  q; ½k; q 2 ½0; 1 ð80:11Þ

Formulation we have when a and b vary

L X
X S
Max ðai þ bi Ii Þpil xil ð80:12Þ
l¼1 i¼1

Subject to Eqs. (80.2–80.8) and (80.10) with additional constraints

ai þ bi Ii  ai1 þ bi1 Ii1 ; i [ 1; i ¼ 1; . . .; S ð80:13Þ

y  ai  z; ½y; z 2 ½0; 1 ð80:14Þ

Formulation when we have a vary and b fixed

L X
X S
Max ðai þ bIi Þpil xil ð80:15Þ
l¼1 i¼1

Subject to Eqs. (80.2–80.8) and (80.13–80.14).

Since ISP wants to get revenue maximization by setting up the prices charge-
able for a, b and QoS level to recover cost and to enable the users to choose
services based on their preferences like stated in Eq. (80.1). Equation (80.2) shows
that the required capacity of service does not exceed the network capacity
reserved. Equation (80.3) explains that required capacity cannot be greater than
the network capacity C in link l. Equation (80.4) guarantee that network capacity
856 F.M. Puspita et al.

has different location for each service that lies between 0 and 1 (80.5). Equa-
tion (80.6) explains that QoS level for each service is between the prescribed
ranges set up by ISP. Equation (80.7) shows that users applying the service are
nonnegative and cannot be greater than the highest possible users determined by
ISP. Equation (80.8) states that the number of users should be positive integers.
Equation (80.9) explains that ISP wants to get revenue maximization by setting up
the prices chargeable for a, b and QoS level to recover cost and to enable the users
to choose services based on their preferences. Equation (80.10) explains that b has
different level for each service which is at least the same level or lower level.
Equation (80.11) states that value of b lies between two prescribed values. ISP
wants to get revenue maximization by setting up the prices chargeable for a, b and
QoS level to recover cost and to enable the users to choose services based on their
preferences like stated in Eq. (80.12). Equation (80.13) explains that the sum-
mation of a and b has different level for each service which is at least the same
level or lower level. Equation (80.14) shows that the base price should lie between
predetermined a set up by ISP. ISP wants to get revenue maximization by setting
up the prices chargeable for a, b and QoS level to recover cost and to enable the
users to choose services based on their preferences as stated in objective function
(80.15).

80.4 Optimal Solution

Tables 80.1 and 80.3 below describes the solver status of model formulation in
LINGO when considering base price to be fixed and model formulation when
considering base price to be varied. Generated Memory Used (GMU) shows that
how much the amount of memory use for generating a model. The total time used
so far to generate and solve the model. In Table 80.1, the highest optimal solution
of 811.2 is achieved when ISP sets up a to be fixed and vary b and equals the
capacity link values with GMU = 32 K and ER = 1 s. In Table 80.3, the highest
optimal solution of 912.6 is achieved when ISP varies a, b and equals the capacity
link values with GMU = 33 K and ER = 1 s.
Tables 80.2 and 80.4 depict the optimal solutions of the formulation by setting
up the base price value to be fixed or vary the base price. In Table 80.2, when the
formulation of a to be fixed and vary b, the QoS level of 0.8 is achieved by service
1 and service 2 with C1 = C2. But only 10 users apply the service in service 2 on
link 1 with 100 % network is reserved and 9 users on link 2 with 90 % network is
reserved.
 80

Table 80.1 Solver status and extended solver state of the models by considering a to be fixed for three capacity link cases
Solver status a and b fixed a fixed and b vary
C1 < C2 C1 > C2 C1 =C2 C1 < C2 C1 > C2 C1 = C2
Model class INLP INLP INLP INLP INLP INLP
State Local optimal Local optimal Local optimal Local optimal Local optimal Local optimal
Objective 492.151 477.4 569.831 686.4 667.2 811.2
Infeasibility 2.7 9 10-13 0.3 9 10-3 1.8 9 10-7 1.1 9 10-16 0.1 9 10-15 0.9 9 10-12
Iterations 2,186 259 818 614 375 181
Solver type B&B B&B B&B B&B B&B B&B
Best Objective 492.151 477.4 569.831 686.4 667.2 811.2
Objective Bound 492.151 477.4 569.831 686.4 667.2 811.2
The Improved Models of Internet Pricing Scheme …

Steps 51 9 20 26 8 6
Active 0 0 0 0 0 0
Update interval 2 2 2 2 2 2
GMU (K) 30 30 30 32 31 32
ER (sec) 2 0 1 1 0 1
857
858 F.M. Puspita et al.

Table 80.2 Solutions of Models by considering a to be fixed for three capacity link cases
Var a = 0.5 and b = 0.01(fixed) a = 0.5 (fixed) and b vary
C1 \ C2 C1 [ C2 C1 = C2 C1 \ C2 C1 [ C2 C1 = C2
a1 – – – – – –
a2 – – – – – –
a3 – – – – – –
b1 – – – 0.375 0.375 0.375
b2 – – – 0.375 0.375 0.375
b3 – – – 0.3 0.3 0.3
I1 0.8 0.83 0.83 0.8 0.8 0.8
I2 0.8 0.8 0.8 0.8 0.8 0.8
I3 0.5 0.5 0.5 1 1 1
a11 0.024 0.025 0.025 0.096 0 0
a21 0.6 0.7 0.7 0.904 1 1
a31 0.375 0.275 0.275 0 0 0
x11 2 3 3 8 0 0
x21 4 7 7 6 10 10
x31 9 10 10 0 0 0
a12 0.025 0.12 0.025 0.08 0.1 0.08
a22 0.7 0.46 0.7 0.92 0.9 0.9
a32 0.275 0.4 0.275 0 0 0.02
x12 3 10 3 10 8 10
x22 7 3 7 9 6 9
x32 10 10 10 0 0 0

The Highest QoS level of 1 is achieved by service 3 but no users apply the
service. When the formulation of varying a and b, as stated in Table 80.4, QoS
level of 0.8 is achieved by service 1 and 2 also of 1 in service 3 but only in service
2, 10 users apply the service in link 1 and 9 users in link 2 with 100 % network
reserved for service 2 in link 1 and 90 % network reserved in link 2.
To sum up, the objective of ISP to achieve the maximum profit will be reached
if ISP set up the base price to be fixed and vary the quality premium or to vary the
base price and quality premium with setting up the equal capacity link values
(C1 = C2).
 80

Table 80.3 Solver status and extended solver state of the models by considering a being varied for three capacity link cases
Solver status a vary and b fixed a and b vary
C1 \ C2 C1 [ C2 C1 = C2 C1 \ C2 C1 [ C2 C1 = C2
Model class INLP INLP INLP INLP INLP INLP
State Local optimal Local optimal Local optimal Local optimal Local optimal Local optimal
Objective 585.144 563.2 677.421 772.2 750.6 912.6
Infeasibility 0.0015 0.3 9 10-3 3.7 9 10-5 8.7 9 10-13 0 1.7 9 10-17
Iterations 1,517 348 736 1,622 462 188
Solver type B&B B&B B&B B&B B&B B&B
Best Objective 585.144 563.2 677.421 772.2 750.6 912.6
Objective Bound 585.144 563.2 677.421 772.2 750.6 912.6
The Improved Models of Internet Pricing Scheme …

Steps 26 16 12 25 11 6
Active 0 0 0 0 2 0
Update interval 2 2 2 2 0 2
GMU (K) 32 32 32 33 33 33
ER (sec) 1 0 2 1 0 1
859
860 F.M. Puspita et al.

Table 80.4 Solutions of models by considering a being varied for three capacity link cases
Var a vary and b = 0.01 (fixed) a and b vary
C1 \ C2 C1 [ C2 C1 = C2 C1 \ C2 C1 [ C2 C1 = C2
a1 0.5 0.5 0.5 0.25 0.26 0.26
a2 0.597 0.59 0.59 0.53 0.53 0.63
a3 0.6 0.6 0.6 0.6 0.6 0.6
b1 – – – 0.8 0.8 0.8
b2 – – – 0.45 0.45 0.34
b3 – – – 0.3 0.3 0.3
I1 0.8 0.83 0.8 0.8 0.8 0.8
I2 0.8 0.8 0.8 0.8 0.8 0.8
I3 0.5 0.5 0.5 1 1 1
a11 0.024 0.025 0.024 0.096 0 0
a21 0.602 0.7 0.7 0.9 1 1
a31 0.373 0.275 0.276 0.003 0 0
x11 2 3 3 8 0 0
x21 4 7 7 6 10 10
x31 9 10 10 0 0 0
a12 0.024 0.12 0.024 0.08 0.1 0.08
a22 0.7 0.46 0.7 0.91 0.9 0.9
a32 0.276 0.4 0.276 0.012 0 0.02
x12 3 10 3 10 8 10
x22 7 3 7 9 6 9
x32 10 10 10 0 0 0

80.5 Conclusion

The paper [9] be more upgraded by using our new approach using other tools. We
obtain slightly increasing profit in several solutions we proposed. We also save
human resources by only applying few users to apply the service and also we can
save energy by only promote one service rather than two services. Our solutions
show better profit with less idle time and number of users applied the services.
We have shown that by considering new parameters, more decision variables
and constraints, we obtain better revenue maximization. The cases shown above
basically are ISP strategy to vary its preference to achieve their goals. ISP is able to
adopt the cases to suit their goals. But again, like stated in since it is more theo-
retical point of view and assumptions, we limit our result only static result in data
changes, and cost preference is just based on our discrete data. Further research
should address more generalization of the model to also consider numerous services
offered or generalization of more services.
80 The Improved Models of Internet Pricing Scheme … 861

Acknowledgments The research leading to this study was financially supported by Ministry of
Higher Education Malaysia for support through Fundamental Research Grant Scheme (FRGS)
2011, Research Code: USIM/FRGS-FST-5-50811.

References

1. Yang, W., Owen, H.L., Blough, D.M., Guan, Y.: An auction pricing strategy for
differentiated service network. In: Proceedings of the IEEE Global Telecommunications
Conference. IEEE (2003)
2. Yang, W.: Pricing Network resources in differentiated service networks. In: School of
electrical and computer engineering. Phd Thesis. Georgia Institute of Technology. pp. 1–111
(2004)
3. Yang, W., Owen H., Blough, D.M.: A Comparison of auction and flat pricing for
differentiated service networks. In: Proceedings of the IEEE International Conference on
Communications (2004)
4. Yang, W., Owen, H.L., Blough, D.M.: Determining Differentiated services network pricing
through auctions. In: Networking-ICN 2005, 4th International Conference on Networking
April 2005 Proceedings, Part I. Reunion Island, France, Springer-Verlag, Berlin Heidelberg
(2005)
5. Byun, J., Chatterjee, S.: A strategic pricing for quality of service (QoS) network business. In
Proceedings of the Tenth Americas Conference on Information Systems, New York (2004)
6. Puspita, F.M., Seman, K., Taib, B.M., Shafii, Z.: A new approach of optimization model on
internet charging scheme in multi service networks. Int. J. Sci. Tech. 2(6), 391–394 (2012)
7. Puspita, F.M., Seman, K., Taib, B.M., Shafii, Z.: An improved optimization model of internet
charging scheme in multi service networks. TELKOMNIKA 10(3), 592–598 (2012)
8. Puspita, F.M., Seman, K., Taib, B.M., Shafii, Z.: An Improved model of internet pricing
scheme of multi service network in multiple link QoS networks. In: The 2013 International
Conference on Computer Science and Information Technology (CSIT-2013). Universitas
Teknologi Yogyakarta (2013)
9. Sain, S., Herpers, S.: Profit Maximisation in multi service networks- an optimisation model.
In: Proceedings of the 11th European Conference on Information Systems ECIS 2003.
Naples, Italy (2003)
10. LINGO: LINGO 13.0.2.14. LINDO Systems, Inc: Chicago (2011)
11. Gu, C., Zhuang, S., Sun, Y.: Pricing incentive mechanism based on multistages traffic
classification methodology for QoS-enabled networks. J. Netw. 6(1), 163–171 (2011)
12. Bouch, A., Sasse, M.A.: Network quality of service—an integrated perspective. In:
Proceedings of RTA’s’99. Vancouver, 1–3 June 1999
13. Shakkottai, S., Srikant, R.: Network optimization and control. Found. Trends Networking
2(3), 271–379 (2007)
14. Shakkottai, S., Srikant, R., Ozdaglar, A., Acemoglu, D.: The price of simplicity. IEEE J. Sel.
Areas Commun. 26(7), 1269–1276 (2008)
15. Eltarjaman, W., Ashibani, M., El-Jabu, B.: Towards optimized QoS based—charging model.
In Southern African Telecommunication Networks and Applications Conference(SATNAC
2007). Sugar Beach Resort, Mauritius (2007)
16. Alderson, D., Willinger, W., Li, L., Doyle, J.: An Optimization-Based Approach to
Modelling Internet Topology. Telecommun. Plann. Innovations Pricing Netw. Des. Manage.
Oper. Res./Comput. Sci. Interfaces Ser. 33, 101–136 (2006)
17. Garcell, M.A.G., Delgado, L.Y.M., Torres, L.A., Isaac, A.C.: Identifying and solving
optimization problems on internet. Becnybr TUTE. Tov 14. § 2. Transactions TSTU,
pp. 392–404 (2008)
862 F.M. Puspita et al.

18. Wu, M., Xiong, X., Ying, J., Jin, C., Yu, C.: A web services composition model for QoS
global optimization. In Proceedings of the Second Symposium International Computer
Science and Computational Technology(ISCSCT ’09). P. R. China, Huangshan, 26–28 Dec
2009
19. Wang, Q., Peha, J.M., Sirbu, M.A.: Optimal Pricing for integrated-services networks with
guaranteed quality of service. In: Bailey, J., McKnight, L. (eds.) Internet Economics. MIT
Press, Cambridge (1996)
20. Marzolla, M., Mirandola, R.: QoS Analysis for Web Service Applications: A Survey of
Performance-Oriented Approaches from an Architectural Viewpoint. Department of
Computer Science University of Bologna, Bologna, Italy (2010)
21. Tektas, B., Kasap, N.: Time and volume based optimal pricing strategies for telecommunication
networks. In: 17th International Conference on Management of Technology (IAMOT 2008).
UAE, Dubai (2008)
22. Puspita, F.M., Seman, K., Taib, B.M., Shafii, Z.: The Improved formulation models of
internet pricing scheme of multiple bottleneck link QoS Networks with various link capacity
cases. In: Seminar Hasil Penyelidikan Sektor Pengajian Tinggi Kementerian Pendidikan
Malaysia ke-3. Universiti Utara Malaysia (2013)
23. Puspita, F.M., Seman, K., Sanugi, B.: Internet Charging scheme under multiple QoS networks.
In: The International Conference on Numerical Analysis & Optimization (ICeMATH 2011).
Yogyakarta, Indonesia, Universita Ahmad dahlan, Yogyakarta, 6–8 June 2011
24. Puspita, F.M., Seman, K., Taib B.M.: A Comparison of optimization of charging scheme in
multiple QoS networks. In: 1st AKEPT 1st Annual Young Reseachers International
Conference and Exhibition (AYRC X3 2011) Beyond 2020: Today’s Young Reseacher
Tomorrow’s Leader 19–20 DECEMBER 2011. 2011. PWTC, Kuala Lumpur (2011)
25. Puspita, F.M., Seman, K., Taib, B.M., Shafii, Z.: Models of Internet charging scheme under
multiple QoS networks. In: International Conferences on Mathematical Sciences and
Computer Engineering 29–30 November 2012. Kuala Lumpur, Malaysia (2012)
26. Puspita, F.M., Seman, K., Taib, B.M., Shafii, Z.: Improved Models of internet charging
scheme of single bottleneck link in multi QoS networks. J. Appl. Sci. 13(4), 572–579 (2013)
27. Puspita, F.M., Seman, K., Taib, B.M., Shafii, Z.: Improved models of internet charging
scheme of multi bottleneck links in multi QoS networks. Aust. J. Basic Appl. Sci. 7(7),
928–937 (2013)
28. Kumar, S., Sehgal, R., Singh, P., Chaudhhary, A.: Nepenthes Honeypots based Botnet
detection. J. Adv. Inf. Technol. 3(4), 215–221 (2012)
Chapter 81
Improving the Models of Internet
Charging in Single Link Multiple Class
QoS Networks

Irmeilyana Saidi Ahmad, Indrawati, Fitri Maya Puspita

and Lisma Herdayana

Abstract In this paper, an improved internet charging scheme in multiple QoS

networks will be discussed. The objective is to obtain better solution than previous
results conducted by previous research. ISPs need a new charging scheme to
maximize the revenue and provide better services to customers. The model is set
up by fixing the fixed base price, varying the quality premium and fixing the
sensitivity price for user in each class. The model is considered as Mixed Integer
Nonlinear Programming (MINLP) and that can be solved by LINGO 11.0 to obtain
the optimal solutions. We compare three cases of original, modified one and
modified two models depending with the fixing or varying parameters or variables.
The results show that by improving the pricing scheme model, the user’ sensitivity
price in modified two cases will yield maximum profit for ISPs.

81.1 Introduction

Previous works on pricing scheme of QoS networks is due to [1–4]. They

described the pricing scheme based auction to allocate QoS and maximize ISP’s
revenue. The solution of the optimization problem goes from single bottleneck link
in the network and then they generalized into multiple bottleneck links using

I.S. Ahmad  Indrawati  F.M. Puspita (&)  L. Herdayana

Faculty of Mathematics and Natural Sciences, Universitas Sriwijaya,
Jln. Raya Palembang-Prabumulih, Indralaya, Ogan Ilir Sumatera Selatan 30662, Indonesia
e-mail: [email protected]
I.S. Ahmad
e-mail: [email protected]
Indrawati
e-mail: [email protected]
L. Herdayana
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 863

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_81
864 I.S. Ahmad et al.

heuristic method. In their study, they used single QoS parameter-bandwidth. In

their discussion, they focus on auction algorithm to find the optimal solution.
Based on their idea, it is attempted to improve their mathematical formulation and
combine it with mathematical formulation discussed by Byun and Chatterjee [5]
(see in [6–11]) to show that by improving the models, ISP improve the profit, with
the advantages of availability of base price, quality premium and quality premium
to be measured.
Recent studies have also been conducted to address problem of multiple service
network, other kind of pricing scheme in network. Sain and Herpers [12] discussed
problem of pricing in multiple service networks. They solve the internet pricing by
transforming the model into optimization model and solved using Cplex software.
Also, [13–15] discussed the new approach and new improved model of and got
better results in getting profit maximization of ISP.
Although QoS mechanisms are available in some researches, there are few
practical QoS network. Even recently a work in this QoS network proposed by [1–4],
it only applies simple network involving one single route from source to destination.
So, the contribution is created by improving the mathematical formulation of to
be simpler formulation in single link by taking into consideration the utility
function, base price as fixed price or variable, quality premium as fixed prices and
variable, index performance, capacity in one link, bandwidth required and also the
user price sensitivity. The problem of internet charging scheme is considered as
Mixed Integer Nonlinear Programming (MINLP) to obtain optimal solution by
using LINGO 11.0 software. In this part, the comparison of two models is con-
ducted in which whether decision variable is to be fixed of user admission to the
class or not. This study focuses to fix the user’s price sensitivity in each class. We
consider cases of base price to be fixed and the quality premium to be fixed or vary
depends on what target ISP would achieve. The Objective of ISP is to obtain
maximum profit.

81.2 Research Method

The idea basically generates from [1–5] and are improved in single link multi class
QoS networks. We attempt to improve the models when we consider the cases to
fix the user price sensitivity in each class.
The steps are taken as follows.
1. Determine the parameters and decision variables for original and modified
models.
2. Determine the constraints for the models.
3. Determine the model formulation of Steps 1 and 2.
4. Form the model formulation of base price and quality premium as the constant
value and base price as the constant and quality premium as the variable.
5. Analyze the results and conclude the results.
81 Improving the Models of Internet Charging … 865

81.3 Results and Discussion

81.3.1 Assumptions

Assume that there is only one single network from source to destination since
concentrate on service pricing scheme. Assume that the routing schemes are
already set up by the ISP. As [2] pointed out, we have 2 parts of utility function
namely, base cost which does not depend on resource consumption and cost which
depends on resource consumption. The parameters and decision variables we set
up are presented in Tables 81.1 and 81.2.

81.3.2 Model Formulation

The model formulation follows from [10] except for W ~ ij and Wj we modify by
varying or fixing the prices, for each case of original, modified and modified 1 with
additional constraints if we set up W ~ ij and Wj as the parameters as follows.

~ ij ¼ k; k 2 R
W ð81:1Þ

Wj ¼ 1 ð81:2Þ

81.3.3 The Solution for Original Model

In Table 81.3, the values of decision variables are given for original model. Final
bandwidth (X ^ ij ) of user i is 1.234568. Minimum bandwidth for LM1 and LM2 is
1.234568 kbps. Sensitivity prices for class 1 and class 2, respectively (W1 dan W2)
are 1.234568. We varies the base price 0.2/kbps and 0.3/kbps for all cases to
promote the ISP goal to compete in market. Table 81.4 presents the solver status of
the solver. Best objective is reached on value of 1.
Generated Memory Used (GMU) in Table 81.4 shows that the number of
allocated memory used to run the solver. For original model, we have GMU of
28 K. Elapsed Runtime (ER) explains that the total time needed to solve the
models and is affected by ether application running in the system. ER is 1 s in 4
iterations. We got total profit of only 1 unit price.

81.3.4 The Solutions for Modified Model with a and b Fixed

We modified the models into 3 groups when Wj and W ~ ij as parameters, Wj as and

~ ij as variable and lastly when Wj as variable with W
W ~ ij as parameter. Table 81.5
explains the variable values for modified models. We obtain final bandwidth (X ^ ij )
866 I.S. Ahmad et al.

Table 81.1 Parameters for each case of internet charging scheme

Parameter for original model
Q Total bandwidth
Vi Minimum bandwidth needed by user i
aj Base price for class j
Parameter for model modified 1 (a b constants)
aj Base price for class j
bj Premium quality having service performance Ij
Q Total bandwidth
Vi Minimum bandwidth needed by user i
cj Upper bound value for user i sensitivity price in class j
dj Upper bound for quality index in class j
Parameter for model modified 2 (a constant, b variable)
Q Total bandwidth
Vi Minimum bandwidth needed by user i
aj Base price for class j
cj Upper bound value for user i sensitivity price in class j
dj Upper bound value for quality index in class j
fi Lower bound for premium quality in class j
gi Upper bound for premium quality in class j

for each user is 400.346 kbps. Premium quality for user 1 is 0.01 and for user 2 of
0.02. Minimum bandwidth LM1 and LM2 is 0.01 kbps. The price sensitivity for class
1 and 2 respectively (W1 and W2) are 13 when Wj as variable and W ~ ij as parameter.
We obtain W1 = 10 and W2 = 12 when Wj and Wij as parameters also Wj as
variable and Wij as parameter. It means that ISP can vary the base price of 0.2 unit
price/kbps and 0.3 unit price/kbps, for all cases to compete in the market.
The highest GMU in this model is 29 K for each case as stated in Table 81.6.
ER is 1 s when Wj and Wij as parameter. Also, ER = 0 s for Wj as variable and W ~ ij
as parameter, and last case when Wj as variable and Wij as parameter. ESS is 0
since the solver applies the branch and bound solver. We can see that ISP can gain
the maximum profit of 551.62 unit price if ISP sets Wj as variable and W ~ ij as
parameter to enable ISP to recover cost.

81.3.5 The Solutions for Modified Model with a Fixed

and b Vary

When we set up the modified model with a fixed and b vary, we group it into three
categories namely when Wj and W ~ ij as parameter, Wj as parameter and W ~ ij as
variable, dan lastly when Wj as variable and W~ ij as variable. Table 81.7 shows the
result of decision variables when we set up the modified model with a fixed and b
81 Improving the Models of Internet Charging … 867

Table 81.2 Decision variables for each case of internet charging scheme

Variable for original model 1; user i is admitted to class j
Zij :
0; otherwise
~ ij : Final bandwidth obtained by user i in class j
X
LMj: Minimum bandwidth for class j
Wj: Sensitivity price for class j
Xj: Final bandwidth for class j
~ ij : Sensitivity price for user iin class j
W

Variable for modified model with a and b 1; user i is admitted to class j
parameters Zij :
0; otherwise
~ ij : Final bandwidth obtained by user i in class j
X
LMj: Minimum bandwidth for class j
Wj: Sensitivity price for class j
Xj: Final bandwidth for class j
~ ij : Sensitivity price for user i in class
W
Ij: Quality index of class j

Variable for modified model with a 1; user i is admitted to class j
parameter and b variable Zij :
0; otherwise
~ ij : Final bandwidth obtained by user i in class j
X
LMj: Minimum bandwidth for class j
Wj: Sensitivity price for class j
Xj: Final bandwidth for class j
~ ij : Sensitivity price for user i in class
W
Ij: Quality index of class j
bj: Premium quality of class j having service
performance of Ij

Table 81.3 Decision variable values for original model proposed by [2]
Decision variables values
a1 0.2 ^ 11
X 1.234568
a2 0.3 ^ 12
X 1.234568
b1 – ^ 21
X 1.234568
b2 – ^ 22
X 1.234568
Z11 1 LM1 1.234568
Z12 1 LM1 1.234568
Z21 1 X1 1.234568
Z22 1 X2 1.234568
~ 11
W 1.234568 I1 –
~ 12
W 1.234568 I2 –
~ 21
W 1.234568 W1 1.234568
~ 22
W 1.234568 W2 1.234568
868 I.S. Ahmad et al.

Table 81.4 Solver status of original model proposed by [2]

Solver status Model class INLP
State Local optimal
Infeasibility 0
Iterations 4
Extended solver state Solver type Branch and bound
Active 0
Update interval 2
GMU (K) 28
ER (sec) 1
Best objective 1
Objective bound 1
ESS 0
TSI 4

Table 81.5 Decision variable values for modified model with a and b fixed
Variable Modified model with a and b fixed
Wj Par Wj Par Wj Var
~ ij Par
W ~ ij Var
W ~ ij Par
W
a1 0.2 0.2 0.2
a2 0.3 0.3 0.3
b1 0.01 0.01 0.01
b2 0.02 0.02 0.02
Z11 1 1 0
Z12 1 1 0
Z21 1 1 1
Z22 1 1 1
W1 10 10 13
W2 12 12 13
~ 11
W 12 10 12
~ 12
W 12 12 12
~ 21
W 15 10 15
~ 22
W 15 12 15
^
X11 400.346 400.346 400.346
^ 12
X 400.346 400.346 400.346
^ 21
X 400.346 400.346 400.346
^ 22
X 400.346 400.346 400.346
LM1 0.01 0.01 0.01
LM2 0.01 0.01 0.01
X1 400.346 400.346 400.346
X2 400.346 400.346 400.346
I1 0.9 0.9 0.9
I2 0.8 0.8 0.8
81 Improving the Models of Internet Charging … 869

Table 81.6 Solver status for modified model with a and b fixed
Solver status Wj Par and Wj Par Wj Var
~ ij Par
W ~ ij Var
W ~ ij Par
W
Model class INLP
State Local optimal
Infeasibility 7.38964e-012 7.38964e-012 0
Iterations 5 5 7
Solver type Branch and bound
Active 0 0 0
Update interval 2 2 2
GMU (K) 29 29 29
ER (sec) 1 0 0
Best objective 467.34 467.34 551.62
Objective bound 467.34 467.34 551.62
ESS 0 0 0
TSI 5 5 7

Table 81.7 Decision variables for modified model with a fixed and b vary
Var Modified Model with a Fixed and b Vary
Wj Par Wj Par Wj Var
~ ij Par
W ~ ij Var
W ~ ij Par
W
a1 0.2 0.2 0.2
a2 0.3 0.3 0.3
b1 0.04 0.04 0.04
b2 0.03 0.03 0.03
Z11 1 1 0
Z12 1 1 0
Z21 1 1 1
Z22 1 1 1
W1 10 10 13
W2 12 12 13
~ 11
W 12 10 12
~ 12
W 12 12 12
~ 21
W 15 10 15
~ 22
W 15 12 15
^ 11
X 400.346 400.346 400.346
^ 12
X 400.346 400.346 400.346
^ 21
X 400.346 400.346 400.346
^ 22
X 400.346 400.346 400.346
LM1 0.01 0,01 0.01
LM2 0.01 0,01 0.01
X1 400.346 400,346 400.346
X2 400.346 400,346 400.346
I1 0.9 0.9 0.9
I2 0.8 0.8 0.8
870 I.S. Ahmad et al.

Table 81.8 Solver status of modified model with a fixed and b vary
Solver status Wj Par Wj Par Wj Var
~ ij Par
W ~ ij Var
W ~ ij Par
W
Model Class INLP
State Local optimal
Infeasibility 7.38964e-012 7.38964e-012 0
Iterations 5 5 7
Extended solver state
Solver type Branch and bound
Active 0 0 0
Update interval 2 2 2
GMU (K) 29 30 29
ER (sec) 0 1 1
Best objective 467.41 467.41 551.69
Objective bound 467.41 467.41 551.69
ESS 0 0 0
TSI 5 5 7

vary. Final bandwidth (X ^ ij ) obtained by the users is 400.346 kbps. Premium quality
for user 1 is 0.04 and for user 2 is 0.03. Minimum bandwidth for LM1 and LM2 is
0.01 kbps. Price sensitivity for class 1 and 2 (W1 dan W2) respectively is 13 when
Wj as variable and W~ ij as parameter while W1 = 10 dan W2 = 12 when Wj and Wij
as parameter and also when Wj as variable and Wij as parameter. ISP enables to
vary the base into 0.2/kbps dan 0.3/kbps to promote the available classes.
The highest GMU presented in Table 81.8 is 30 K when Wj as parameter and
Wij as variable, meanwhile GMU = 29 K when Wj and W ~ ij as parameter, also
when Wj as variable and Wij as parameter. ER is 0 s for the case when Wj and Wij
as parameter, while ER = 1 s for Wj as variable and W ~ ij as parameter, also Wj as
variable and Wij as parameter. ESS is 0 for all cases. ISP can obtain maximum
profit of 551.69 unit price when ISP sets up the case when Wj as variable and W ~ ij
as parameter to enable ISP to recover cost.
From Tables 81.4, 81.6 and 81.8 we can check that the best objective is
achieved when ISP sets up either to fix the base price and quality premium to
recover cost and to let user to choose the class; or to fix the base price and vary
quality premium to recover cost and ISP can promote certain services by adding
the condition to the models by setting up the sensitivity price for user j to be fixed
and the sensitivity price for user i in class j to be varied.
81 Improving the Models of Internet Charging … 871

81.4 Conclusion

From the above discussion, we can see that by considering the new parameters,
decision variables and the constraints, we can obtain the better maximum profit.
ISP can adopt either the model of modified by fixing a and b; or fixing a and
~ ij as parameter to attain maximum value
varying b for Wj as variable and fixing W
of 323.78 bps for each file and web traffic data.

Acknowledgments The research leading to this study was financially supported by Directorate
of Higher Education Indonesia (DIKTI) for support through Hibah Bersaing Tahun II, 2014.

References

1. Yang, W., Owen, H.L., Blough, D.M., Guan, Y.: An auction pricing strategy for
differentiated service network. In: Proceedings of the IEEE Global Telecommunications
Conference, IEEE (2003)
2. Yang, W.: Pricing network resources in differentiated service networks, in school of electrical
and computer engineering. Ph.d. Thesis, Georgia Institute of Technology, pp. 1–111 (2004)
3. Yang, W., Owen, H., Blough, D.M.: A comparison of auction and flat pricing for
differentiated service networks. In: Proceedings of the IEEE International Conference on
Communications (2004)
4. Yang, W., Owen, H.L., Blough, D.M.: Determining differentiated services network pricing
through auctions. In: Networking-ICN 2005, 4th International Conference on Networking
April 2005 Proceedings, Part I. Reunion Island, France, Springer-Verlag Berlin Heidelberg
(2005)
5. Byun, J., Chatterjee, S.: A strategic pricing for quality of service (QoS) network business. In:
Proceedings of the Tenth Americas Conference on Information Systems, New York (2004)
6. Puspita, F.M., Seman, K., Sanugi, B.: Internet charging scheme under multiple QoS
networks. In: The International Conference on Numerical Analysis & Optimization
(ICeMATH 2011) 6–8 June 2011. Yogyakarta, Indonesia: Universita Ahmad dahlan,
Yogyakarta (2011)
7. Puspita, F.M., Seman, K., Taib, B.M.: A comparison of optimization of charging scheme in
multiple QoS networks. In: 1st AKEPT 1st Annual Young Reseachers International
Conference and Exhibition (AYRC X3 2011) Beyond 2020: Today’s Young Reseacher
Tomorrow’s Leader 19–20 December 2011. PWTC, Kuala Lumpur (2011)
8. Puspita, F.M., Seman, K., Taib, B.M., Shafii, Z.: Models of internet charging scheme under
multiple QoS networks. In: International Conferences on Mathematical Sciences and
Computer Engineering 29-30 November 2012. Kuala Lumpur, Malaysia (2012)
9. Puspita, F.M., Seman, K., Taib, B.M., Shafii, Z.: The improved formulation models of
internet pricing scheme of multiple bottleneck link QoS networks with various link capacity
cases. In: Seminar Hasil Penyelidikan Sektor Pengajian Tinggi Kementerian Pendidikan
Malaysia ke-3 2013: Universiti Utara Malaysia
10. Puspita, F.M., Seman, K., Taib, B.M., Shafii, Z.: Improved models of internet charging
scheme of single bottleneck link in multi QoS networks. J. Appl. Sci. 13(4), 572–579 (2013)
11. Puspita, F.M., Seman, K., Taib, B.M., Shafii, Z.: Improved models of internet charging
scheme of multi bottleneck links in multi QoS networks. Aust. J. Basic Appl. Sci. 7(7),
928–937 (2013)
872 I.S. Ahmad et al.

12. Sain, S., Herpers, S.: Profit maximisation in multi service networks—An optimisation model.
In: Proceedings of the 11th European Conference on Information Systems ECIS 2003.
Naples, Italy (2003)
13. Puspita, F.M., Seman, K., Taib, B.M., Shafii, Z.: A new approach of optimization model on
internet charging scheme in multi service networks. Int. J. Sci. Technol. 2(6), 391–394 (2012)
14. Puspita, F.M., Seman, K., Taib, B.M., Shafii, Z.: An improved optimization model of internet
charging scheme in multi service networks. TELKOMNIKA 10(3), 592–598 (2012)
15. Puspita, F.M., Seman, K., Taib, B.M., Shafii, Z.: An improved model of internet pricing
scheme of multi service network in multiple link QoS networks. In: The 2013 International
Conference on Computer Science and Information Technology (CSIT-2013). Universitas
Teknologi Yogyakarta (2013)
Chapter 82
A New Aggregating Phase for Interval
Type-2 Fuzzy TOPSIS Using
the ELECTRE I Method

Nurnadiah Zamri and Lazim Abdullah

Abstract Aggregating phase is considered as one of the important steps in

interval type-2 fuzzy TOPSIS (IT2 FT) instead of ratings of alternatives under
criteria and the importance weights of criteria and ranking of alternatives. How-
ever, some problems occur in aggregating phase of IT2 FT when it is have a large
computational procedure due to the hardly defined in the second membership
function. Therefore, we offer a more easier and practical in defining the new
aggregating phase. Our proposed method is to establish a new aggregating phase
for IT2 FT using the ELECTRE I method in the interval type-2 fuzzy set (IT2FS)
concept. A numerical example is constructed to show the practicality and effec-
tiveness of the proposed method.

82.1 Introduction

Interval type-2 fuzzy TOPSIS (IT2 FTOPSIS) is one of the multi-criteria decision
making (MCDM) method and was first established by Chen and Lee [1] in the year
of 2010. IT2 FTOPSIS is an extension from the original TOPSIS [2] and fuzzy
TOPSIS [3]. IT2 FTOPSIS is believed can give more flexibility room due to the
fact that it is uses interval type-2 fuzzy sets (IT2FSs) rather than type-1 fuzzy sets
(T1FSs) to represent the uncertainties.
IT2 FTOPSIS can be divided into four main phases; rating phase, weighting
phase, aggregating phase and ranking phase. Chen and Lee [1] introduced the
concept on interval type-2 fuzzy sets [4] in the rating phase of IT2 FTOPSIS method.

N. Zamri (&)  L. Abdullah

School of Informatics and Applied Mathematics, University Malaysia Terengganu,
21030 Kuala Terengganu, Terengganu, Malaysia
e-mail: [email protected]
L. Abdullah
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 873

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_82
874 N. Zamri and L. Abdullah

The concept of IT2FSs [4] with subjective weight [2] were proposed in the weighting
phase. The concept of ranking values was developed as a reduction method before
the aggregating phase. Using the concept of positive-ideal solutions and negative-
ideal solutions, optimal values of ranking was produced in the ranking phase. Unlike
the IT2 FTOPSIS [1], our paper focuses on aggregating phase. We propose a new
ELimination Et Choix Traduisant la REalite (ELimination and Choice Expressing
the REality) (ELECTRE) with IT2FS concept in the aggregating phase.
Several papers have discussed on IT2 FTOPSIS’ aggregating phase. For
example, Chen et al. [5] proposed a new aggregating judgment method for ranking
interval type-2 fuzzy sets. Besides, Chen [6] developed a new linear assignment
method to produce an optimal preference ranking of the alternatives in accordance
with a set of criterion-wise rankings and a set of criterion importance within the
context of interval type-2 trapezoidal fuzzy numbers. Chen and Wang [7] estab-
lished a new interval type-2 fuzzy TOPSIS (IT2 FT) with a new fuzzy ranking
method based on the a-cuts of interval type-2 fuzzy sets. However, some problems
occur in aggregating phase of IT2 FT when it is have a large computational
procedure due to the hardly defined in the second membership function. Therefore,
we propose a new ELECTRE I with IT2FS concept in the aggregating phase.
ELECTRE I was developed by Roy [8] as the first outranking method. Since
that, it has been applied in various types of decision-making situations, including
energy [9], environment management [10], project selection [11], and decision
analysis [12]. Type-1 fuzzy ELECTRE was the extended method of ELECTRE. It
can easily convert the linguistic preferences into the type-1 fuzzy numbers [13].
Various authors have applied type-1 fuzzy ELECTRE to the decision-making
method. For example, Wu and Chen [14] used the ELECTRE method to solve the
decision-making problems with intuitionistic fuzzy information. Botti and Peyp-
och [15] implemented an application of ELECTRE I in decision-making concept
to tourism destinations. Hatami-Marbini and Tavana [16] proposed an alternative
fuzzy outranking method by extending the ELECTRE I method to the Fuzzy
TOPSIS (FT) method.
Our proposed work is to develop a new aggregating phase method for IT2
FTOPSIS [1] using the IT2 ELECTRE I. The rest of the paper is organized as
follows. Section 82.2 reviews basic concepts related to ELECTRE I. In Sect. 82.3
we propose a new integrated ELECTRE I with IT2FS in IT2 FTOPSIS approaches.
Section 82.4 illustrates the proposed method by applying it to a numerical
example. Section 82.5 summarizes this study and presents future challenges.

82.2 Background of ELECTRE I

In this section, the basic decision of ELECTRE I [16], is briefly introduced. Then
the proposed new ELECTRE I with IT2FS for IT2 FTOPSIS is presented in
Sect. 82.3.
82 A New Aggregating Phase for Interval … 875

Fig. 82.1 The pictorial Proposed Method

representation of the Roughanian et al. (2010)
proposed method and the 6
Roughanian et al. [17]
rankings 5

Action
3

0
A1 A2 A3 A4 A5
Rank

Definition 1 Preference in ELECTRE I method is modelled by using binary

outranking relations, S, whose meaning is ‘‘at least as good as’’. Considering two
actions x and y, four situations may arise:
i. xS y and not y S x, i.e., x P y (x is strictly preferred to y),
ii. yS x and not x S y, i.e., y P x (y is strictly preferred to x),
iii. xS y and y S x, i.e., x I y (x is indifferent to y), and
iv. not x S y and not y S x, i.e., x I y (x is incomparable to y).

Note that the incomparability preference is a useful relation to account for

situations in which decision-makers (DMs) are not able to compare two actions.
Definition 2 According to ELECTRE I method, given two actions x and y, an
outranking relation is based on two major concepts; the concordance and the
discordance. The following statements provide insights into these concepts:
• The concordance concept: For an outranking x S y to be validated, a sufficient
majority of the criteria should be in favor of this assertion.
• The discordance concept: When the concordance condition holds, none of the
criteria in the majority should oppose too strongly to the assertion x S y.
These two circumstances must be implemented for validating the assertion x S y.
We offer a new idea of aggregation phase for IT2 FTOPSIS by using the basic
definition of ELECTRE I in Sect. 82.2. Therefore, the development of an IT2
ELECTRE I integrated IT2 FTOPSIS is shown in Sect. 82.3.
876 N. Zamri and L. Abdullah

82.3 The Proposed Method

The proposed IT2 ELECTRE I integrated IT2 FTOPSIS starts with the constructed
a decision matrix and weighting process using the IT2 linguistic scale as in Step 1
and Step 2. Then, the weighted decision matrix is constructed using the Step 1 and
Step 2. Then the IT2 ELECTRE I is integrated with IT2 FTOPSIS as shown in
Step 3 until Step 8. Full steps of the proposed method is shown as follows,
Assume that there is a set X of alternatives, where X ¼ fx1 ; x2 ; . . .; xn g; and
assume that there is a set F attributes, where F ¼ ff1 ; f 2 ; . . .; fm g: Assume that
there are k decision-makers D1 ; D2 ; . . .; and Dk. The set F of attributes can be
divided into two sets F1 and F2, F1 where denotes the set of benefit attributes, F2
denotes the set of cost attributes. Below are the steps of IT2 ELECTRE I integrated
IT2 FTOPSIS.
Step 1: Establish a decision matrix
Construct the design matrix Yp of the pth decision-maker and construct the
average decision matrix respectively.
Step 2: Calculate the weighting process
Construct the weighting matrix W ~~ of the attributes of the decision-maker and
p
~~
construct the pth average weighting matrix W.
Step 3: Construct the weighted DMs’ matrix.
Construct the weighted decision matrix V: ~~
Step 4 until Step 8 are the development of an ELECTRE I with IT2FS.
Step 4: Calculate the concordance matrix
Next, the concordance and discordance matrices   are calculated using the
weighted normalized IT2 fuzzy decision matrix V ~~ and the average decision

matrix by DMs. Considering two actions Ag and Aj, the concordance set can be
 
J~c ¼ j~~vgj  ~~vfj where ~~J c is the index of all criteria belonging to the
defined as ~
concordance coalition with the outranking relation Ag S Aj .
 
The discordance set can be defined as ~~J D ¼ j~~vgj  ~~vfj where ~~J D is the index
of all criteria belonging to the discordance coalition and it is against the assertion
‘‘Ag is at least as good as Af.’’
The concordance matrix is defined as follows,
2 3
    b1f    b1ðm1Þ b1m
6 .. .. .. .. .. .. 7
6 . . . . . . 7
6 7
B¼6
6 bg1    bgf    bgðm1Þ bgm 77 ð82:1Þ
6 . .. .. .. .. .. 7
4 .. . . . . . 5
bm1    bmf    bmðm1Þ 
82 A New Aggregating Phase for Interval … 877

where
!
  X X X X X
~cgf ¼ ~
~ ~cpgf ; ~~cqgf ; ~~cugf ¼
~clgf ; ~ ~~ ¼
W ~~ l ;
w ~~ p ;
w ~~ q ;
w ~~ u
w
j2J j j j j j
c
j2Jc j2Jc j2Jc j2Jc

In other words, the elements of concordance matrix are determined as the IT2
fuzzy summation of the IT2 fuzzy weights of all criteria in the concordance set.
Step 5: Calculate the discordance matrix
The discordance matrix is defined as
2 3
    d~~1f    d~~1ðm1Þ d~~1m
6 . .. .. .. .. .. 7
6 . 7
6 . . . . . . 7
~~ ¼ 6 ~~ 7
D 6 dg1    d~~gf    d~~gðm1Þ d~~gm 7 ð82:2Þ
6 7
6 .. .. .. .. .. .. 7
4 . . . . . . 5
d~~m1    d~~mf    d~~mðm1Þ 

where
   
...     
max ~v ~~vfj  maxd~~ max ~~vgj ; ~~vfj ; ~~vfj 
~gf ¼ j2JD gj
d~
j2J
¼ D    
~ ~  
max ~vgj  ~vfj
j
maxd~~ max ~~vgj ; ~~vfj ; ~~vfj 
j

Note that there are prominent differences between the elements of C ~~ and D.
~~ The

concordance matrix C ~~ reflects weights of the concordance criteria and the asym-

metric discordance matrix D ~~ reflects most relative differences according to the

discordance criteria. Both concordance
  and discordance indices have to be cal-
culated for every pair of actions ~g~; ~~f , where ~~g 6¼ ~~f .
Step 6: Calculate the Boolean matrix for concordance
Now, we evaluate the value of the concordance matrix elements according to
~~ ~l ~p ~q ~u 
the concordance level. The concordance level, C ¼ ~c ; ~c ; ~c ; ~c , can be gf gf gf gf
defined as the average of the elements in the concordance matrix, represented by
Pm Pm l . Pm Pm p . P
~c ¼ f ¼1 g¼1 ~cgf mðm  1Þ; ~c ¼ f ¼1 g¼1 ~cgf mðm  1Þ; ~~cq ¼ m
~ l ~ ~p ~
f ¼1
Pm q . P m P m
.
~c mðm  1Þ; and ~~cu ¼ f ¼1 g¼1 ~~cugf mðm  1Þ: It is most desirable
g¼1 ~ gf
that the DMs achieve a consensus on the definition of the concordance level. If
there is a disagreement among the DMs, then, the average value should be used for
the definition.
~~ is formed according to the minimum concordance
Next, the Boolean matrix B
~
 as
level, C,
878 N. Zamri and L. Abdullah

2 3
    ~~b1f    ~~b1ðm1Þ ~~b
1m
6 . .. .. .. .. .. 7
6 . 7
6 . . . . . . 7
~~ ¼ 6 ~~ ~~b 7
B 6 bg1    ~~bgf    ~~bgðm1Þ gm 7 ð82:3Þ
6 7
6 .. .. .. .. .. .. 7
4 . . . . . . 5
~~b    ~~bmf    ~~bmðm1Þ 
m1

where
(
~~
~~cgf  C , ~~bgf ¼ 1
~~
~~cgf  C , ~~bgf ¼ 0

Step 7: Calculate the Boolean matrix for discordance

Similarly, the elements of the discordance matrix are measured by a discor-
P Pm ~ .
dance level. The discordance level, D ~~ ¼ m d~ mðm  1Þ; can be
f ¼1 g¼1 gf

defined as the average of the elements in discordance matrix. The Boolean matrix
~
~ is measured by a minimum discordance level as
H
2 3
  ~~h  ~~h ~~h
1f 1ðm1Þ 1m
6 . .. .. .. .. .. 7
6 . 7
6 . . . . . . 7
~ 6 ~~h 7
H ¼ 6 ~~hg1
~  ~~h
gf  ~~h
gðm1Þ gm 7 ð82:4Þ
6 7
6 .. .. .. .. .. .. 7
4 . . . . . . 5
~~h  ~~h  ~~h 
m1 mf mðm1Þ

where
(
~~
d~~gf  D , ~~hgf ¼ 1
d~~gf  D ~~ , ~~h ¼ 0
gf

The elements of this matrix measures the power of the discordant coalition,
~~
meaning that if its element value surpasses a given level, D; the assertion is no
~~
longer valid. Discordant coalition exerts no power whenever d~~gf \D: In other
~
~
words, the elements of matrix H with the value of 1 show the dominance relations
among the actions.
Step 8: Construct the global matrix
Next, the global matrix Z~~ is calculated by peer to peer multiplication of the
~~ and H
elements of the matrices B ~~ as follows:
82 A New Aggregating Phase for Interval … 879

~~
~~  H
Z~~ ¼ B ð82:5Þ
 
where each element ~~zgf of matrix Z~~ is obtained as

~~zgf ¼ ~~bgf ~~f gf ð82:6Þ

The final step of this procedure consists of exploitation of the above outranking
~
~ in order to identify as small as possible a subset of actions, from
relation (matrix Z)
which the best compromise action could be selected.
This section describes the flow of a new IT2 ELECTRE I method (Step 4 till
Step 8) is implemented to the existed IT2 FTOPSIS (starts from Step 1). With an
attempt to consider the IT2FS framework, it would anticipate that the IT2
ELECTRE I integrated IT2 FTOPSIS makes a more comprehensive look.

82.4 Numerical Illustration

This numerical illustration is adapted from Roghanian et al. [17] is used to

illustrate the procedures and feasibility of the proposed IT2 ELECTRE I integrated
IT2 FTOPSIS framework. A company desired to select a suitable material supplier
to purchase the key components of new products. After preliminary screening, five
candidates (A1; A2; A3; A4, and A5) remain for further evaluation. A committee
of three decision makers: D1, D2 and D3, has been formed to select the most
suitable supplier.
Table 82.1 shows the linguistic terms for the rating of alternatives with seven
scales ‘‘Very Poor (VP)’’, ‘‘Poor (P)’’, ‘‘Medium Poor (MP)’’, ‘‘Fair (F)’’,
‘‘Medium Good (MG)’’, ‘‘Good (G)’’ and ‘‘Very Good (VG)’’.
Moreover, Table 82.2 shows the linguistic terms for the weights of attributes
also with seven scales ‘‘Very Low (VL)’’, ‘‘Low (L)’’, ‘‘Medium Low (ML)’’,
‘‘Medium (M)’’, ‘‘Medium High (MH)’’, ‘‘High (H)’’ and ‘‘Very High (VH)’’.

Table 82.1 Linguistic terms for the ratings and their corresponding interval type-2 fuzzy sets [1]
Linguistic terms Interval type-1 fuzzy sets
Very poor (VP) ((0, 0, 0, 1; 1,1), (0, 0, 0, 1; 1, 1))
Poor (P) ((0, 1, 1, 3; 1,1), (0, 1, 1, 3; 1,1))
Medium poor (MP) ((1, 3, 3, 5; 1,1), (1, 3, 3, 5; 1, 1))
Fair (F) ((3, 5, 5, 7; 1,1), (3, 5, 5, 7; 1, 1))
Medium good (MG) ((5, 7, 7, 9; 1,1), (5, 7, 7, 9; 1, 1))
Good (G) ((7, 9, 0.9, 1; 1,1), (7, 9, 9, 1; 1, 1))
Very good (VG) ((9, 1, 1, 1; 1,1), (9, 1, 1, 1; 1, 1))
880 N. Zamri and L. Abdullah

Table 82.2 Linguistic terms of weights of the attributes and their corresponding interval type-2
fuzzy sets [1]
Linguistic terms Interval type-1 fuzzy sets
Very low (VL) ((0, 0, 0, 0.1; 1, 1), (0, 0, 0, 0.1; 1, 1))
Low (L) ((0, 0.1, 0.1, 0.3; 1, 1), (0, 0.1, 0.1, 0.3; 1,1))
Medium low (ML) ((0.1, 0.3, 0.3, 0.5; 1, 1), (0.1, 0.3, 0.3, 0.5; 1, 1))
Medium (M) ((0.3, 0.5, 0.5, 0.7; 1, 1), (0.3, 0.5, 0.5, 0.7; 1, 1))
Medium high (MH) ((0.5, 0.7, 0.7, 0.9; 1, 1), (0.5, 0.7, 0.7, 0.9; 1, 1))
High (H) ((0.7, 0.9, 0.9, 1; 1, 1), (0.7, 0.9, 0.9, 1; 1, 1))
Very high (VH) ((0.9, 1, 1, 1; 1, 1), (0.9, 1, 1, 1; 1, 1))

Table 82.3 The results comparison

Rank Proposed method Roghanian et al. [17]
A1 3 3
A2 1 2
A3 1 1
A4 4 5
A5 4 4

Using the linguistic scales from Tables 82.1 and 82.2, and the eight steps of the
proposed method (in Sect. 82.3), result for Roghanian et al. example is shown in
Table 82.3, as follows:
Result for Roghanian et al. example is graphed in pictorial representation a
shown in Fig. 82.1, as follows:
Based on Roughanian et al. example, the derived Table 82.3 and graphed
pictorial representation (Fig. 82.1) shows that there is a similarity between our
proposed method and the results provided by Rouhghanian et al. [17]. Here, A2
and A3 are categorized in the first rank. Next, A1 is categorized in the second rank.
The last prioritization belongs to A4 and A5.

82.5 Conclusions and Future Research

In this study, a new aggregating phase for IT2 FTOPSIS was developed by using
the ELECTRE 1 method in terms of IT2FS. This proposed method is seen to ease
the computational volume due to its ability to take ordinal scales into account
without converting the original scales into the abstract ones by using the con-
cordance and discordance steps. Besides, it is at the same time maintains the
original verbal meaning of the decision problems [18]. A numerical example was
used in order to test the efficiency of the proposed approach. In summary, the
proposed approach produced some similar results with Roughanian et al. [17]. The
efficiency of using new method was proven with a straight forward computation in
82 A New Aggregating Phase for Interval … 881

the case study. This approach is seen to provide a new perspective in aggregating
phase of IT2 FTOPSIS. Moreover, they offer a practical, effective and low risk
computation to produce a comprehensive judgment. A stream of future research
can extend our algorithms to other variations of the ELECTRE methods such as
ELECTRE II, III, IV, ELECTRE IS and ELECTRE TRI.

Acknowledgments This research is supported by MyBrain15 scholarship and Fundamental

Research Grant, no. 59243. This support is gratefully acknowledged.

References

1. Chen, S.-M., Lee, L.-W.: Fuzzy multiple attributes group decision-making based on the
interval type-2 TOPSIS method. J. Expert Syst. Appl. 37, 2790–2798 (2010)
2. Chen, C.T.: Extension of the TOPSIS for group decision making under fuzzy environment.
Fuzzy Sets Syst. 114, 1–9 (2000)
3. Hwang, C.L., Yoon, K.: Manufacturing plant location analysis by multiple attribute decision
making: single-plant strategy. Int. J. Prod. Res. 23, 345–359 (1985)
4. Mendel, J.M., John, R.I., Liu, F.L.: Interval type-2 fuzzy logical systems made simple. IEEE
Trans. Fuzzy Syst. 14, 808–821 (2006)
5. Chen, S.-M., Yang, M.-W., Lee, L.-W., Yang, S.-W.: Fuzzy multiple attributes group
decision-making based on ranking interval type-2 fuzzy sets. Expert Syst. Appl. 39,
5295–5308 (2012)
6. Chen, T.-Y.: A linear assignment method for multiple-criteria decision analysis with interval
type-2 fuzzy sets. Appl. Soft Comput. 13, 2735–2748 (2013)
7. Chen, S.-M., Wang, C.-Y.: Fuzzy decision making systems based on interval type-2 fuzzy
sets. Inf. Sci. 242, 1–21 (2013)
8. Roy, B.: Classement et Choix en Présence de Points de vue Multiples (Laméthode
ELECTRE). RIRO. 8, 57–75 (1968)
9. Cavallaro, F.: A comparative assessment of thin-film photovoltaic production processes using
the ELECTRE III method. Energy Policy 38, 463–474 (2010)
10. Hokkanen, J., Salminen, P., Ettala, M.: The choice of a solid waste management system using
the electre II decision-aid method. Waste Manage. Res. 13, 175–193 (1995)
11. Figueira, J., Mousseau, V., Roy, B.: Electre methods. In: Figueira, J., Greco, S., Ehrogott, M.
(eds.) Multiple Criteria Decision Analysis. State of the Art Surveys, pp. 133–162. Springer,
New York (2005)
12. Karagiannidis, A., Perkoulidis, G.: A multi-criteria ranking of different technologies for the
anaerobic digestion for energy recovery of the organic fraction of municipal solid wastes.
Bioresour. Technol. 100, 2355–2360 (2009)
13. Kaya, T., Kahraman, C.: An integrated fuzzy AHP–ELECTRE methodology for
environmental impact assessment. Expert Syst. Appl. 38, 8553–8562 (2011)
14. Wu, M.-C., Chen, T.-Y.: The ELECTRE multicriteria analysis approach based on
Atanassov’s intuitionistic fuzzy sets. Expert Syst. Appl. 38, 12318–12327 (2011)
15. Botti, L., Peypoch, N.: Multi-criteria ELECTRE method and destination competitiveness.
Tourism Manage. Perspect. 6, 108–113 (2013)
16. Hatami-Marbini, A., Tavana, M.: An extension of the electre I method for group decision-
making under a fuzzy environment. Omega 39, 373–386 (2011)
17. Roghanian, E., Rahimi, J., Ansari, A.: Comparison of first aggregation and last aggregation in
fuzzy group TOPSIS. Appl. Math. Model. 34, 754–3766 (2010)
18. Greco, S., Matarazzo, B., Sowinski, R.: Rough sets theory for multi criteria decision analysis.
Eur. J. Oper. Res. 129, 1–47 (2001)
Chapter 83
The Role of Green IT and IT for Green
Within Green Supply Chain Management:
A Preliminary Finding from ISO14001
Companies in Malaysia

K.S. Savita, P.D.D. Dominic and Kalai Anand Ratnam

Abstract The purpose of this study is to investigate the role of Information

Technology and Information System for Environmental Sustainability in Green
Supply Chain Management within the context of manufacturing firms in Malaysia.
More specifically, the pilot findings reveal that Green IS related activities among
manufacturing firms in Malaysia are present, but measuring the extent of overall
effectiveness of Green IS practices is not being carried out adequately. The data
collection instrument used for the pilot study was distributed to the management
team of 50 organizations and only 32 responses were received. With a response
rate of 64 % for the pilot study, the results of the preliminary analysis indicates
that manufacturing organizations in Malaysia are practicing green in Malaysia, but
have yet to track and audit the exact energy efficiency level.

83.1 Introduction

Organizations today are obliged to respond to an increasing rate of change since

product and technology life cycles are getting shorter with product differentiations
[1]. Countries that are vigorously developing such as Asia is the main contributor
to environmental issues. These developing countries are negatively impacting the

K.S. Savita (&)
P.D.D. Dominic

Computer and Information Sciences Department, Universiti Teknologi Petronas,
31750, Ipoh, Perak, Malaysia
e-mail: [email protected]
P.D.D. Dominic
e-mail: [email protected]
K.A. Ratnam
Asia Pacific University of Technology and Innovation (APU), Technology Park Malaysia,
Bukit Jalil, Kuala Lumpur 57000, Malaysia
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 883

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_83
884 K.S. Savita et al.

ecological sustainability as compared to the developed countries [2]. The increased

growth of CO2 and Greenhouse Gas (GHG) emissions are significantly impacting
the global climate change. This phenomena is in response to higher rates of
economic growth and growing urban populations [3]. Manufacturing sectors is the
key for the modernization of a country that differentiate the developed countries
from the developing one [4]. The phenomena of globalization, industrialization
and urbanization have associated impacts to worldwide ecological crisis. Never-
theless, companies are recognizing that environmental management is a key
strategic movement that has a strong influence in enhancing organizational per-
formance and gaining competitive edge in the industry [5].
Being a country that has emerged with multi-sector economy, Malaysia offers a
cost-competitive location to manufacture advanced technological products for both
regional and international markets. In 2011, The World Competitiveness Yearbook
2011 Report released by the Institute for Management Development (IMD) con-
tinued to rank Malaysia as among the top five most competitive nations in the
Asia-Pacific region [6]. Manufacturing sector still remains a significant contributor
to the growth of the Malaysia’s economy [7]. Malaysia is making considerable
progression in environmental management compared to other countries [6].
Despite that, Malaysia is facing growing challenge on energy consumption, air
pollution from industrial emissions and poor waste management that leads to
Greenhouse Gas (GHG) emissions that significantly impacting Malaysia’s ecology
[6, 8].
The impacts to the environment is largely through a product life cycle which is
the primary source of today’s environmental issues [2]. Manufacturing companies
need an effective shift in its environmental management from end-of-the-pipe
control and treatment of waste [9] to clean up and control at every stage of a
product life cycle [10]. The supply chains of many modern-day firms cut across
country boundaries, whereby the production and upstream of many supply chains
exists in emerging and developing countries but the downstream may be located in
developed countries [3]. Both the environmental management and supply chain
management has its own root that complement each and must not be disregarded
[11]. Therefore, adding the ‘green’ component to supply chain management
focuses on activities that aims at minimizing ecological impacts of a product
throughout its entire life cycle, namely environmental pollutions (air, water and
land), waste of resources (energy, materials and products) as well as final use and
disposal of products [5].
The implementation of Information System (IS) and Information Technology
(IT) in Supply Chain Management (SCM) is becoming apparently important in an
increasingly globalized and competitive economy. The advancement in the tech-
nology with high degree of automation in business processes is offering more
opportunities than previously. IT as an infrastructure and a solution plays a critical
factor in the improvement of SCM in within, downstream and upstream of the
organization [12]. IT has leads to better performance in the supply chain with its
power to provide timely, accurate, and reliable information, in which enabled real-
time integration of SCM activities [12]. IT facilitates SCM by improving
83 The Role of Green IT and IT for Green … 885

integration and coordination of physical flow as wells as the various information

flow in the supply chain [13]. These highlight the importance of IT in functioning
supply chain. The use of IT is considered as a prerequisite for an effective control
of today’s complex supply chain [14]. It is concluded that ‘‘IT is not an actual
source of competitiveness but a source of competitive necessity’’, in which
explained as IT implementation has become a necessity, not a choice [12]. The
terms IS and IT are often used exchangeable, but both IT and IS contribute to the
ecological differently and contrarily to an extent [15]. Therefore, in line with prior
literature, this research differentiates ‘‘IT’’ and ‘‘IS’’ as two key elements of
‘environmental sustainability in driving green practices in SCM.
The existing studies are lacking in assessing the role IT and IS in supporting
organization’s actions towards environmental and carbon compliance [16]. This is
due to the general lack of urgency of the impacts of IT and IS have on environ-
mental footprints [16]. Therefore, IT and IS are often not given the right ownership
and ignored by organizations in their assessment on organization’s environmental
footprints [17]. Although, environmental conscious practices play a bigger role
towards worldwide ecological sustainability, yet the role of IT and IS are disre-
garded in such undertakings, and much still remains unexplored [15, 18, 19]. Many
of the research conducted in Malaysia are focusing either on green practices in
SCM or green practices in IT and IS, and very few looks at integrating and
incorporating both fields towards environmental and organizational sustainability.
To the best of our knowledge, the research on IT and IS for environmental sus-
tainability in supply chain in Malaysia is scant. Therefore, this study explores the
role of IT and IS for environmental sustainability in driving the green practices in
supply chain management within ISO 14001 manufacturing firms in Malaysia, and
a preliminary finding from pilot study is presented.

83.2 Literature Review and Hypothesis Development

83.2.1 Green Supply Chain Management (GSCM)

In the early days, companies use typical supply chain to achieve economy of scale,
efficiency, lower operating costs and mass produced products with very little
concern on environmental consequence and ecological impacts [20, 21]. GSCM
evolved significantly in the recent years in responding to the growing concern of
environmental sustainability and compliance. GSCM covers activities such as
‘green design’, ‘green sourcing/procurement’, ‘green operations’ or ‘green manu-
facturing’, ‘green distribution’, ‘logistics/marketing’ and ‘reverse logistics’ [5].
The GSCM is categorized into inbound logistics, production or the internal supply
chain, outbound logistics and reverse logistics that cover all supply chain activities,
from green purchasing to the integration of life cycle management, through the
manufacturer and customer, to closing the loop with reverse logistics [22]. Simi-
larly, GSCM practices reflect the whole system of upstream, within, downstream
886 K.S. Savita et al.

and transformation that focus on environmental operations management, suppliers

collaboration (assessments and education), green procurement and logistics [23].
Other than that, GSCM practices covers internal environmental management, green
purchasing, cooperation with customers, eco design and investment recovery [11].
The previous research that investigated on GSCM implementation in South East
Region (Philippines, Indonesia, Malaysia, Thailand and Singapore), found that
green practices exist but still at infancy stage [22]. Looking at Malaysia context, the
fully owned Malaysian companies have a lower level of adoption of GSCM as
compared to foreign based companies and Multinational Company (MNC) [24].
Also, [25] discovered that manufacturing industry in Malaysia has shown a strong
awareness of environmental issues. However, very few companies were actually
involved in the implementation plans, whereby, the larger firms, mostly multi-
national companies, are implementing environmental initiatives, although still
limited and at early stage. This indicate the gaps to investigate in-depth the adoption
of GSCM in developing countries, like Malaysia since the current findings is
inconclusive as different sectors in different countries may be facing different
pressure at various level and mode of implementation of GSCM practices [22, 26].
This further support [11] belief that, ‘‘study on Green SCM is timely and necessary
to better aid the organization and have yet to be fully investigated’’.

83.2.2 Green Information Technology

The impacts IT have on the environment are explained into direct or first-order
effect known as Green IT and indirect and second-order effect refers to Green IS
(IT for Green). Green IT is mainly focused on energy efficiency, carbon footprint
and equipment utilization in contrast Green IS refers to the design and imple-
mentation of information systems that contribute to sustainable business processes
[16, 27, 28]. Although, Green IT and Green IS are inter-related, they each have a
different focus and purpose [29]. Green IT is conceptualized in various ways
depending on its context and scope. Elliot and Binney [30] define Green IT in
terms of design, production, operation and disposal of IT and non-IT-enabled
products/services, which is not detrimental and beneficial to the environment
during its entire life cycle. Murugesan [31] refers Green IT as environmentally
sound IT, whereby it is a study and practice of designing, manufacturing, using
and disposing of computers, servers and its associated subsystems with minimal or
no impact on the environment (environmental sustainability), energy efficiency
and total cost of ownership. Therefore, IT is considered as one of the key roles that
transform supply chain management function in manufacturing organization, in
which it has shorten the product life cycle [22]. Moreover, replacement of tradi-
tional technologies with IT will minimize the amount of resources used in man-
ufacturing [28]. We therefore propose to contribute to the literature by exploring
whether there are positive relationships Green IT in driving the Green Supply
83 The Role of Green IT and IT for Green … 887

Chain Management (GSCM) within ISO 14001 manufacturing firms in Malaysia.

The following hypotheses are therefore proposed:
Hypotheses 1 (H 1) Green IT positively influences the implementation of green
practices in SCM among ISO 14001 manufacturing firms in Malaysia

83.2.3 Green Information System (IS)

Recently, IS literature also began to realize the importance of sustainability, and

proposed the concept of ‘‘Green IS’’ to better understand the role of IS in dealing
with sustainability [18, 32]. Green IS focused not only on reducing the impact of
IT but on the ways information systems can be used to help firms reduce their
carbon footprint through automating and transforming products, business pro-
cesses, business relationships and practices [19, 32]. Nowadays, companies
increasingly integrate their business processes without realizing that Green IS
initiatives are not be limited within companies, but extended to the externalities
that need collaboration from the companies, partners, and even customers [33].
Green IS plays a significant role in making both business processes and the
products they create environmentally sustainable [34]. Since, environmental sus-
tainability is a supply chain commitment, therefore, Green IS are essential for
creation, maintenance, and survival of environmental conscious practices in supply
chain [35]. The information systems have the ability to enable interconnectedness,
realign and reinvent business processes in support of productivity and efficiency
towards ecological improvements [36]. Green IS encourages the application of
information systems thinking and skills to initiatives across all functions of the
organization, from logistics, to waste management, to communicating consump-
tion information to customers [15]. Green IS is view as an enabler in inducing
changes within and among business processes that decreases the environmental
impacts through integration and coordination of IS throughout the SCM [34, 35].
These design and development of information systems represent the backbone of
environmental management efforts that supports the firm’s environmental man-
agement systems [26]. As emphasized by [32], a much in-depth research is
required to determine to what extent IS might improve sustainability in the realm
of supply chains and logistics. Hence, the proposed hypothesis:
Hypothesis 2 (H 2) Green IS (IT for Green) positively influences the imple-
mentation of green practices in SCM among ISO 14001 manufacturing firms in
Malaysia
888 K.S. Savita et al.

83.3 Research Methodology

The chosen samples are from ISO 14001 manufacturing companies in Malaysia as
they have been expected to have embarked on green practices in their operation.
The companies are selected based on purposive sampling method. According to
this method, the elements in the sample are selected for a specific purpose and in
the best position to provide the information required by the researcher since the
subjects has the expertise/knowledge on the topic being investigated [37]. The
information on the companies is obtained from Federation of Malaysia Manu-
facturers Perak Directory of year 2012. The ISO 14001 certification detail is cross
checked from SIRIM QAS website and company’s website to ensure that the
chosen companies are practicing green activities. For this preliminary research, the
researchers’ have chosen Ipoh and Kulim for the pilot testing in obtaining the
initial findings. In both locations, the ISO 14001 manufacturing firms include of
large enterprises and SMEs. The invitation to complete the questionnaire survey
was send out using official email to the Human and Resources Department of the
participating companies. The target respondents were company’s senior manage-
ment whom their scope of work includes managing the company’s operation and
production. After two weeks, 50 companies responded and agreed to participate in
the survey. Subsequent emails were then sent to the identified personnel’s which
includes the link of the survey. The reason for adopting online survey is due to its
potentially quicker response time with wider magnitude of coverage and higher
cost savings compared to mail or direct survey [37]. A period of 2 months was
given for the companies to respond to the survey. During that period, follow-up
reminders via email and telephone calls are utilized in order to increase the par-
ticipation rate of the survey. After 2 months, a total of 37 responses were received,
whereby five of them were incomplete. For the pilot study, the final total of 32
questionnaires was used as the dataset to test the associated hypotheses.

83.4 Analysis

83.4.1 Profile of Respondents

The demographic profile of the participants is shown in Table 83.1, which indi-
cates that the majority of the respondents were large firms, while the remainder
were Small and Medium Enterprises (SMEs). The majority of the companies that
participated in the survey were from Electrical and Electronics as well as Rubber
and Plastics sectors. Most of the respondents are attached to operation department,
where most of them hold top management posts.
83 The Role of Green IT and IT for Green … 889

Table 83.1 Assessments of organizations

Frequency Percentage
Position
Director 3 9.4
Manager 16 50.0
Executive 13 40.6
Sector
Electrical and electronics 8 25.0
Chemicals and chemical Products 3 9.4
Rubber and plastic products 7 21.9
Basic metal and fabricated metal products 5 15.6
Machinery and equipment 4 12.5
Others 5 15.6
Size
Small and medium Enterprise (SME) 9 28.1
Large enterprise 23 71.9

83.4.2 Data Analysis and Discussion

This paper provides a preliminary insight on the role of Green IT and Green IS (IT
for Green) in driving environmental sustainability in Green Supply Chain Man-
agement (GSCM). According to [18], Green IS (IT for Green) has a bigger
influence than Green IT since it tackles a larger scope in making the entire systems
more sustainable. Therefore, this paper choose to categorize the effects of IT into
direct (Green IT) and indirect (Green IS/IT for Green) in order to obtain con-
ceptual and practical understanding in influencing the Green SCM
implementation.
Figure 83.1, illustrates the extent to which factors of Green IT that influences
the Green SCM implementation. The findings obtained are largely consistent with
both hypotheses. The environmentally friendly disposal of electronic wastes car-
ried the highest significant percentage (84.4 %), and followed by procurement of
IT hardware and equipment (56.3 %), print optimization (46.9 %), PC power
management (34.4 %) and rightsizing IT equipment (28.1 %). The results obtained
is consistent with study conducted among 143 organizations from Australia, New
Zealand and the USA that indicated that disposal of IT in an environmentally
friendly manner is the most adopted Green IT practice [38]. In addition, manu-
facturing companies in Malaysia is starting to recognize the importance on
improving energy efficiency since use of virtualization technology is gaining its
popularity. However, in data center, the design and use of energy efficient IT
infrastructures are still lacking because most the large data center are based in the
890 K.S. Savita et al.

Fig. 83.1 To what extent do the following factors influence your firm in implementing
environmentally conscious practices in supply chain management?

headquarters which are located in other countries. The use of energy efficient lights
and other technologies in production floor is only considers among large enter-
prises, due to its rising cost and intangible returns. This is because, most these
companies fails to measure the amount of the IT and its energy usage contribute in
the electrical bills as well as actual cost savings from implementing energy effi-
cient practices and technologies. The similar situation is being faced by 54 % of
respondents (60 % Australian, 50 % New Zealand and 38 % US) that were una-
ware on the actual cost and savings from implementing Green IT [38]. As men-
tioned by [39], exact energy consumption and measurement of desktops, monitors,
notebooks, networks, printers and communication equipment can only be esti-
mated, and still remain a huge challenge for many organizations. Therefore, these
preliminary results concluded that technologies that improve the energy efficiency
of IT and ecological footprint from IT are not as widely adopted as expected
among these ISO 14001 manufacturing companies.
As presented in Fig. 83.2, the highest scored Green IS activities is 81.3 %
which is the application of IS in enhancing green practices within downstream,
production and upstream of supply chain in integrating and coordinating physical
and information flows by means of automation of business workflows and appli-
cations development. These enable the internal customers to get access to infor-
mation anytime anywhere in a paperless environment followed by the use of video
conferencing/telecommuting (68.8 %) and use of online groupware/collaboration
tools (59.4 %) that facilitate more effective internal communication, apart from
83 The Role of Green IT and IT for Green … 891

Fig. 83.2 To what extent do the following factors influence your firm in implementing
environmentally conscious practices in supply chain management?

promote communication and collaboration seamlessly without having the need to

travel, thus reducing carbon emissions enterprise-wide.
Green IS provide the information needed to coordinate within, and extended
relationships with customers and suppliers that are necessary to make decisions
about eco-design, production, packaging, transportation, reuse, recycling and
recovery of materials [35]. This findings are consistent with [15] that categorized
green IS practice into (1) use of systems for pollution reduction in business pro-
cess; (2) use of system to enhance environmental friendliness at each stage supply
chain management and (3) use of system for learning and innovation, such as
knowledge management systems for sustainable business operations transforma-
tion. This trends suggested that the companies have been, using the applications
from category 1 and 2 since few years ago without realizing is part of environ-
mentally-friendly practice that are ecologically driven. However, the use of soft-
ware to monitor, assess and report environmental indicators such as GHG
emissions, waste, toxic and hazardous materials use is less significant to most of
the companies. The lacks of its importance explained that companies are still
lacking of proper governance and evaluation on overall environmental situation
contributed by information technologies and systems. Butler [40] explained that
between 65–80 % of companies lack an integrated IS infrastructure to track, audit
and manage issues around product and process compliance to environmental cri-
teria, which also mentioned in [39].
For technological context, the respondents were enquired on Green IT which
covers the aspect of core technology followed by IT For Green (Green IS) which
covers the aspect of software and applications. Table 83.1 presents the mean
892 K.S. Savita et al.

Table 83.2 Technological context

Dimension Mean Std. Dev. Cronbach’s Alpha
Green IT 3.81 0.785 0.812
IT for Green 4.39 0.452 0.887

values and standard deviation for both Green IT and IT for Green. Based on the 5-
point Likert-type scale, the mean and standard deviation were 3.81 and 0.785
respective for Green IT, followed by IT for Green with 4.39 and 0.452 respec-
tively. Both Mean and standard deviation were higher for IT for Green as com-
pared to Green IT, this inevitably means that manufacturing firms in Malaysia are
inclined towards paying attention towards the green aspect of software systems as
compared to hardware aspect (Table 83.2).

83.5 Conclusion

Most of the previous studies on GSCM, the key drivers that influence its imple-
mentation are organizational and environmental factors, which covers top man-
agement supports, regulations, customer pressures, expected business benefits and
social responsibility. Despite from these drivers, Green Information Technologies
and Systems have huge potential for addressing broader environmental issues in
organizations, including supply chains and logistics. Information Technologies
and systems are embedded in supply chain business processes, in which the effects
of greening will impact many areas and overall sustainability of a business is
remain unclear. The existing Green IT and Green IS research are fragmented and
very little evaluation on the potential of Green IT and Green IS as the enablers for
successful implementation of green practices in supply chains. In this study, the
researchers combine the both Green IT and IT for Green (Green IS) in investi-
gating their influence in driving the implementation of Green SCM within
ISO14001 manufacturing firms in Malaysia. The results presented are at the pre-
liminary stage, whereby it indicates that Green IS has greater influences over
Green IT. This could be understood from the context of investments needed for
such initiatives versus the tangible returns. Nevertheless, the companies showed
receptive response towards the implementation of Green IT and IT for Green in
enabling the green practices in supply chain activities. As such, a full scale study is
been carried to determine to what extent Green IT and Green IS might improve
environmental and business sustainability in the realm of supply chain manage-
ment in Malaysia context.
83 The Role of Green IT and IT for Green … 893

References

1. Azevedo, S.G., Carvalho, H., Cruz Machado, V.: The influence of green practices on supply
chain performance: a case study approach. Transp. Res. Part E Logist. Transp. Rev. 47(6),
850–871 (2011)
2. Markandya, A., Halsnaes, K.: Developing countries and climate change. In: Owen, A.D.,
Hanley, N. (eds.) The Economics of Climate Change, pp. 239–258. Routledge, Taylor &
Francis Group, London and New York (2004)
3. Gholami, R., Sulaiman, A.B., Ramayah, T., Molla, A.: Senior managers’ perception on green
information systems (IS) adoption and environmental performance: results from a field
survey. Inf. Manag. 50(7), 431–438 (2013)
4. El-Khasawneh, B.S.: Challenges and remedies of manufacturing enterprises in developing
countries: Jordan as a case study. J. Manuf. Technol. Manag. 23(3), 328–350 (2012)
5. Srivastava, S.K.: Green supply-chain management: a state-of-the-art literature review. Int.
J. Manag. Rev. 9(1), 53–80 (2007)
6. MGCC: ‘Market watch 2012’ the environmental sector in Malaysia (2012)
7. MIDA: Malaysia investment performance 2012: investment for transformation (2012)
8. Abidin, Z.Z., Jelani, N.: The impact of climate change and the Need for official statistics in
Malaysia, pp. 1–19. 2011
9. Handfield, R., Sroufe, R., Walton, S.: Integrating environmental management and supply
chain strategies. Bus. Strateg. Environ. 14(1), 1–19 (2005)
10. Matos, S., Hall, J.: Integrating sustainable development in the supply chain: the case of life
cycle assessment in oil and gas and agricultural biotechnology. J. Oper. Manag. 25(6),
1083–1102 (2007)
11. Zhu, Q., Sarkis, J., Lai, K.: Confirmation of a measurement model for green supply chain
management practices implementation. Int. J. Prod. Econ. 111(2), 261–273 (2008)
12. Li, G., Yang, H., Sun, L., Sohal, A.S.: The impact of IT implementation on supply chain
integration and performance. Int. J. Prod. Econ. 120(1), 125–138 (2009)
13. Omar, R., Ramayah, T., May-Chuin, L., Sang, T.Y., Siron, R.: Information sharing,
information quality and usage of information technology (IT) tools in Malaysian
organizations. Afr. J. Bus. Manag. 4(12), 2486–2499 (2010)
14. Fasanghari, M., Roudsari, F.H., Chaharsooghi, S.K.: Assessing the impact of information
technology on supply chain management. World Appl. Sci. J. 4(1), 87–93 (2008)
15. Chen, A.J., Watson, R.T., Boudreau, M.-C., Karahanna, E.: An institutional perspective on
the adoption of green IS. Aust. J. Inf. Syst. 17(1), 23–45 (2010)
16. Jenkin, T.A., Webster, J., McShane, L.: An agenda for ‘Green’ information technology and
systems research. Inf. Organ. 21(1), 17–40 (2011)
17. Huang, A.H.: A model for environmentally sustainable information systems development. In:
Pacific Asia Conference on Information Systems (PACIS), 2008
18. Watson, R., Boudreau, M., Chen, A., Huber, M.: Green IS: building sustainable business
practices. In: Watson, R.T. (ed.) Information Systems, pp. 1–15. Global Text Project, Athens,
GA (2008)
19. Chen, A.J.W., Boudreau, M.-C., Watson, R.T.: Information systems and ecological
sustainability. J. Syst. Inf. Technol. 10(3), 186–201 (2008)
20. Sarkis, J., Zhu, Q., Lai, K.: An organizational theoretic review of green supply chain
management literature. Int. J. Prod. Econ. 130(1), 1–15 (2011)
21. Nelson, D.M., Marsillac, E., Rao, S.S.: Antecedents and Evolution of the Green Supply
Chain. J. Oper. Supply Chain Manag. 5(Special Issue on Sustainability), 29–43 (2012)
22. Rao, P., Holt, D.: Do green supply chains lead to competitiveness and economic
performance? Int. J. Oper. Prod. Manag. 25(9), 898–916 (2005)
23. Holt, D., Ghobadian, A.: An empirical study of green supply chain management practices
amongst UK manufacturers. J. Manuf. Technol. Manag. 20(7), 933–956 (2009)
894 K.S. Savita et al.

24. Eltayeb, T.K., Zailani, S.: Going green through green supply chain initiatives towards
environmental sustainability. Oper. Supply Manag. 2(2), 93–110 (2009)
25. Taha, Z., Sakundarini, N., Ariffin, R., Ghazila, R., Gonzales, J.: Eco design in Malaysian
industries: challenges and opportunities. J. Appl. Sci. 6(12):2143–2150 (2010)
26. Cai, S., Chen, X., Bose, I.: Exploring the role of IT for environmental sustainability in China:
an empirical analysis. Int. J. Prod. Econ. 146(2), 491–500 (2013)
27. Dedrick, J.: Green IS: concepts and issues for information systems research. Commun. Assoc.
Inf. Syst. 27(1), 173–184 (2010)
28. Faucheux, S., Nicolaï, I.: IT for green and green IT: a proposed typology of eco-innovation.
Ecol. Econ. 70(11), 2020–2027 (2011)
29. Molla, A.: Identifying IT sustainability performance drivers: instrument development and
validation. Inf. Syst. Front. 15(5):705–723 (2013)
30. Elliot, S., Binney, D.: Environmentally sustainable ICT: developing corporate capabilities
and an industry-relevant is research agenda. In: Proceedings of the Pacific Asia Conference
on Information Systems (PACIS), Suzhou, China (2008)
31. Murugesan, San: Harnessing green IT: principles and practices. IT Prof. 10(1), 24–33 (2008)
32. Melville, N.P.: Information systems innovation for environmental sustainability. MIS
Quaterly 34(1), 1–21 (2010)
33. Brooks, S., Wang, X., Sarker, S.: Unpacking green IS: a review of the existing literature and
directions for the future. In: vom Brocke, J., Seidel, S., Recker, J. (eds.) Green Business
Process Management Towards the Sustainable Enterprise, pp. 15–37. Springer, Berlin (2012)
34. Boudreau, M.C., Watson, R.T., Chen, A.: From green IT to green IS. In: Biros, B. et al. (ed.)
The Organizational Benefits of Green IT, pp. 79–91. Cutter Information LLC, Arlington, MA
(2008)
35. Green, K.W., Zelbst, P.J., Meacham, J., Bhadauria, V.S.: Green supply chain management
practices: impact on performance. Supply Chain Manag. An Int. J. 17(3):290–305 (2012)
36. Dwyer, C., Hasan, H.: Emergent solutions for global climate change: lessons from green IS
research. Int. J. Soc. Organ. Dyn. IT 2(2), 18–33 (2012)
37. Kumar, M., Salim, A.T., Ramayah, T.: Business Research Methods, pp. 1–417. Oxford Fajar,
Oxford University Press (2013)
38. Molla, A., Pittayachawan, S., Corbitt, B.: Green IT diffusion: an international comparison.
Green IT Working Paper Series, no. 1. pp. 1–15, 2009
39. Erek, K.: From green IT to sustainable information systems management: managing and
measuring sustainability in IT organisations. In European, Mediterranean & Middle Eastern
Conference on Information Systems, vol. 2011, pp. 766–781 (2011)
40. Butler, T.: Compliance with institutional imperatives on environmental sustainability :
building theory on the role of Green IS. J. Strateg. Inf. Syst. 20(1), 6–26 (2011)
41. Zhang, X., Van Donk, D.P., Van Der Vaart, T.: Does ICT influence supply chain
management and performance?: a review of survey-based research. Int. J. Oper. Prod. Manag.
31(11), 1215–1247 (2011)
Chapter 84
Integrating e-Learning with Radio
Frequency Identification (RFID)
for Learning Disabilities: A Preliminary
Study

Wan Fatin Fatihah Yahya, Noor Maizura Mohamad Noor,

Mohd Pouzi Hamzah, Mohamad Nor Hassan,
Nur Fadila Akma Mamat and Mohd Arizal Shamsil Mat Rifin

Abstract Integrating learning styles in adaptive educational systems are a

growing trend in technology enhanced learning. Children have different learning
styles, abilities, preferences that focus on different types of information and pro-
cess new information in different ways. Providing adaptively based on learning
styles can promote interest for learners and make learning easier for them. The
purpose of our research is to adopt an e-Learning approach Radio Frequency
Identification (RFID) technology in order to model the Visual, Auditory and
Kinesthetic (VAK) learning style focused on Learning Disabilities (LD) children.
Today’s technology offers great chances to assist students with disabilities to live
freely and learn more easily. Developing the learning environments assisted by
technology is a new way in making their learning processes successful.

W.F.F. Yahya (&)
N.M.M. Noor
M.P. Hamzah
M.N. Hassan
N.F.A. Mamat

M.A.S.M. Rifin
School of Informatics and Applied Mathematics, Universiti Malaysia Terengganu,
21030 Kuala Terengganu, Terengganu, Malaysia
e-mail: [email protected]
N.M.M. Noor
e-mail: [email protected]
M.P. Hamzah
e-mail: [email protected]
M.N. Hassan
e-mail: [email protected]
N.F.A. Mamat
e-mail: [email protected]
M.A.S.M. Rifin
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 895

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_84
896 W.F.F. Yahya et al.

Keywords e-Learning Learning styles Learning disabilities Radio frequency
identification

84.1 Introduction

Computers have been applied to develop electronic learning (e-Learning) systems

since 1980. The e-Learning market has developed and grown worldwide as the
new paradigm of modern education. e-Learning can be defined as an activity to
sustenance a learning practice by either developing or applying information and
communication technology (ICT) and also involves the use of a number of tech-
nological tools that can be applied in various contexts [1, 2]. e-Learning is one
important opportunity for supporting greater access for all children. Children have
different learning styles, abilities, preferences that focus on different types of
information and process new information in different ways [3, 4]. The concept of
e-Learning is to provide children with learning disabilities access to great
opportunity to improve the quality of their learning process. These children are a
special minority group in the society. Therefore, setting [3] down the needs of the
students with a learning disability is an important issue to support and it gives
great chances for them to live freely and learn more easily.
Using learning styles in e-Learning environment is quite a new trend in tech-
nology enhanced learning. Providing adaptively based on learning styles can
sustenance learners and make learning easier for children with Learning Disabil-
ities (LD). e-Learning environments taking into learning style are more efficient
than traditional e-Learning environment [5]. There are some problems with e-
Learning materials, they do not strictly fulfill the conventional multi-sensory
approaches to help Learning Disability children [6]. The element of touch, for
example, is usually not available. A technique that has shown potential in creating
the element of touch is by leveraging the state of the art Radio Frequency Iden-
tification (RFID) technology.

84.2 Related Works

84.2.1 Review Stage Conventional Learning Versus

e-Learning

Conventional learning to conventional teaching and learning contrasts with e-

Learning [5]. There is a disagreement that conventional learning is the best way of
sustaining a learning process. However, there is no finding to support this dis-
agreement, and research shows that technology supported models are at least as
good as conventional learning [1]. Conventional learning assumes students to learn
84 Integrating e-Learning with Radio Frequency Identification (RFID) … 897

Table 84.1 Comparison between conventional learning and e-Learning

Conventional Learning e-Learning
Assumes students to learn skills at the same rate Provide personalized student need
Book based Use technological tools in learning
Teacher centered Student centered
Passive learning Active learning
Single media Multi-sensory and media/interactive
Time and location constraints Time and location flexibility

skills at the same rate. However, one student from another may learn in different
ways from listening, watching, questioning and doing [1]. While both students
have the ability to learn, conventional learning does not recognize this. Students
who are quick learners often becoming uninterested or troublesome. Based on the
result, it shows that both learners are impotent to meet their full potential.
Table 84.1 show the comparison between conventional learning and e-Learning.
With the progress of ICT development, e-Learning is emerging as the paradigm
of modern education. e-Learning means the delivery of information for education,
learning or training program assisted by ICT [7, 8]. e-Learning involves the use of
a number of technological tools that can be applied in various contexts [8]. Content
is delivered via the Internet, intranet/extranet, audio or video tape, satellite TV,
and CD-ROM. It can be self-paced or instructor led and includes media in the form
of text, image, animation, streaming video and audio [1]. A common theme
explored in e-Learning literature is on how technology plays the role in changing
the learning paradigm, which resulted the process of acquiring knowledge
becomes faster and more efficient [9].
Many studies have exposed that technology can play an important role in any
work with specific disadvantaged groups [10–12]. Studies have also explored how
information and communications technologies (ICT) can influence the education
of students with LD and have shown that this technology can play an important
and useful role in the learning process [10, 12, 13]. One particular problem with e-
Learning materials is that it may contain multimedia elements, they do not strictly
comply with conventional multi-sensory approaches to help learning disabilities.
The element of touch, for example, is usually not available [12]. As an alternative,
one of the technologies that have been successful bridging connections between
the physical and virtual environment is RFID [7]. e-Learning can be viewed as a
new style for delivering well designed, learner-centered, interactive, attractive,
flexible, meaningful and facilitated learning environment [14]. One of the most
best and complete comprehensive theoretical e-Learning models is Badrul e-
Learning framework [15]. Figure 84.1 shows the Badrul e-Learning framework
898 W.F.F. Yahya et al.

Fig. 84.1 Badrul e-Learning framework

Fig. 84.2 RFID system component

84.2.2 Radio Frequency Identification

RFID describes systems that use radio waves to transmit an object’s identity.
There are several methods of identifying objects by using RFID. The most com-
mon is store an ID or a serial number that identifies a specific product along with
other information, on a tag, which is a small microchip that is attached to an
antenna. The antenna enables the chip transmit whatever identification information
it contains to a reader. The reader converts the radio waves from the RFID tag into
digital information that software systems can use for processing [7]. Figure 84.2
shows the RFID system component. There are tag/transponder, antenna, reader and
application.
84 Integrating e-Learning with Radio Frequency Identification (RFID) … 899

Fig. 84.3 Growing number Google Scholar Science Direct Scopus

of radio frequency IEEE ISI
identification in learning 35.00
between years 2008 and 2012 30.00
25.00
20.00
15.00
10.00
5.00
0.00
2008 2009 2010 2011 2012

The emerging of the technology will revolutionize the development of learning

process among the LD children. Information and Technology (ICT) and RFID, are
among the examples that will instrument the learning revolution. We emphasize
towards the active learning, where it gives the new experience to the LD children,
to ‘‘actively’’ participate, simulate and interact with the tool and via-versa. The LD
children will interact directly with the tool using more intuitive techniques of
RFID that are attached to the learning items.
The term Learning Disability (LD) is used to refer to neurological disorders that
affect one or more of the basic psychological development involved in the pro-
cesses of speech, language, reading, writing, arithmetic, or other school subjects
[8, 9]. These children may not be the same and shows a different combination and
degree of difficulties. In difference to other disabilities, such as blindness, vision
and motor impairments, a learning disability does not become immediately
apparent in daily life [8]. If provided assistance related to children with learning
disabilities, they can sparkle and they can be successful.
According to The National Joint Committee on Learning Disabilities (NJCLD)
[10], the term ‘learning disability’ is defined as ‘‘A heterogeneous group of dis-
orders manifested by significant difficulties in the acquisition and use of listening,
speaking, reading, writing, reasoning or mathematical abilities. These disorders are
intrinsic to the individual and presumed to be due to Central System Dysfunction.’’
RFID has been used in variety of learning research. Result from the previous
similar research show that the RFID technology can help children learners to have
better learning experiences in terms of experiential learning, constructivist learn-
ing and more effective learning [16–18].
Figure 84.3 shows the number of publication related to RFID in learning,
starting from 2008 to 2012. The analysis is done based on the number publication
restricted in the Scopus, IEEE, Science Direct, ISI Web of Science, and Google
Scholar. Figure 84.3 shows a significant increase number of publications, due to
the features of RFID, it’s ideally suited for learning. Despite a decrease number of
publication in 2009, however it is increased again starting from 2010.
900 W.F.F. Yahya et al.

84.2.3 Learning Disability

In Malaysia, children with learning disabilities are categorised as special needs

children. The Malaysian Ministry of Education categorised special needs into three
categories, those who are visually handicapped, or partially or fully deaf, or suf-
fering from the disability to learn [11].
Traditional computer or web based learning environments offer the same
content and they do not consider the individual differences, preferences and
interests [12]. Modern technology offers great opportunities and continues to offer
novelties that assist students with disabilities to take part in education and life-long
learning. Emerging learning environments blended with technology offers a great
potential to promote and enhance students learning processes to live freely and
learn more easily. Many researches over the last 30 years have shown that tech-
nology can play an important role in specific disadvantaged groups such as the
blind, those with movement disabilities and LD [6, 9, 12].

84.2.4 Learning Style

Learning is the process of gaining knowledge and active process, so it may be

useful in life situations [2]. All people learn in different ways. Different learners
have different cognitive processes, learning style preferences and past experiences
that they apply when they are learning. An individual’s learning style will affect
the way in which information is processed and will effect on the learning effec-
tiveness and efficiency [19].
Many researches showed that knowing learning styles of individuals will assist
their learning process [13]. Learning style can be defined as individual preferences
of learning and differences in students’ learning and considered as one of the
factors influencing a learner’s achievement [13, 5]. Almost 71 different learning
styles models have been stated [5]. Learning styles that are most used in literature
are Felder–Silverman, Kolb, Dunn and Dunn, Honey and Mumford and Visual,
Auditory and Kinesthetic (VAK) learning style [5, 14]. The main purpose for
selecting VAK learning style for our system is the reason that this learning style is
suitable for structural characteristics of topics in creating the content of the system.
It can also apply e-Learning approaches that represent using media for all learning
modalities. For example, visually, there are present words and images. For audi-
tory, there are spoken words and sound explanation. Similarly, in kinesthetic,
interactive animation is presented. For many years, multi-sensory approaches have
been supported in order to help LD children [6].
84 Integrating e-Learning with Radio Frequency Identification (RFID) … 901

Fig. 84.4 Architecture of system

84.3 Proposed Approach

The purpose of our research is to combine an e-Learning with RFID technology in

order to model the learning style focused for LD children. Today’s technology
offers great chances to assist students with disabilities to live freely and learn more
easily. Developing the learning environments assisted by technology for these
students is of great importance in improving their learning processes.
In this study, an adaptive e-Learning was proposed based on learning style of a
learner using VAK learning style and RFID technology focused for LD children to
improve their learning performance. Based on the VAK learning style, different
contents will be prepared according to these three styles that are offered to the
student.
This system also connects the digital and physical world by providing a plat-
form for LD children to select tangible object and receive computer-based mul-
timedia instructions. By using RFID technology, the children only needs to
explore, select items, and simply move them over the RFID reader. A different
content which will be automatically launched is prepared according to these three
styles that are offered to the student to achieve greater impact on the learning
process.
General architecture of this system is described. This system will be designed as
an adaptive e-Learning environment where the content will be based on VAK
learning style and is supported with RFID technology. The architecture of this
system is shown in Fig. 84.4.
As a part of the research, scenarios appropriate for constructivist approach were
prepared for the teaching of these subjects. These scenarios will be designed in
902 W.F.F. Yahya et al.

three different ways according to the sub-areas of VAK learning style. While
transferring the activities into the digital environment according to sub-learning
styles, characteristics of each area were made prominent. For example, by using
RFID technology, the children only needs to explore, select items, and simply
move them over the RFID reader. A different content which will be automatically
launched is prepared according to this three styles that are offered which creates an
activity for auditory learning style, sound lectures and instructions. While making
an activity for visual learning style, more images, flow charts and animations were
used. Similarly, while making an activity for kinesthetic learning style, learners
were mainly presented with interactive animations and embedded with the sto-
rytelling concept.

84.4 Conclusions

This paper proposed a model for an adaptive e-Learning which is able to model the
learning style of learners using the VAK learning style with RFID technology.
Children learn the content that is appropriate to their own learning styles via this
system. It is possible to say that this system provides a fully individualized
environment to learners. In summary, the system offers what learners need. It
creates a promise to significantly improve the learning result. Indeed, it provides a
great opportunity improving the quality for learners to acquire knowledge faster
and flexible, by assist learners to study in ‘‘the best way’’.

Acknowledgments We would like to thank Senstech Sdn. Bhd. as contributing funds to conduct
this research. The authors also would like to express a deep gratitude to the anonymous reviewers
of this paper. Their useful comments have played a significant role in improving the quality of
this work.

References

1. Ghadirli, H.M., Rastgarpour, M.: A model for an intelligent and adaptive tutor based on web
by Jackson’s learning styles profiler and expert systems. In: Presented at the International
MultiConference of Engineers and Computer Scientists 2012, Hong Kong (2012)
2. Sun, P.-C., et al.: What drives a successful e-Learning? An empirical investigation of the
critical factors influencing learner satisfaction. Comput. Educ. 50, 1183–1202 (2008)
3. Akplotsyi, R., Mahdjoubi, L.: Effects of learning styles on engaging children in school
projects. In: presented at the ARCOM, Bristol, UK, 5–7 Sept 2011
4. García, P., et al.: Evaluating Bayesian networks’ precision for detecting students’ learning
styles. Comput. Educ. 49, 794–808 (2007)
5. Özyurt, Ö., et al.: Design and development of an innovative individualized adaptive and
intelligent e-learning system for teaching–learning of probability unit: Details of
UZWEBMAT. Expert Syst. Appl. 40 , 2914–2940 (2013)
6. Beacham, N.A., Alty, J.L.: An investigation into the effects that digital media can have on the
learning outcomes of individuals who have dyslexia. Comput. Educ. 47, 74–93 (2006)
84 Integrating e-Learning with Radio Frequency Identification (RFID) … 903

7. Hanson, J: An introduction to RFID development Available at: http://www.devx.com/

enterprise/Article/31108 (2006)
8. Savidis, A., et al.: Developing inclusive e-learning and e-entertainment to effectively
accommodate learning difficulties. Univ. Access Inf. Soc. 5, 401–419 (2007)
9. Adam, T., Tatnall, A.: Using ICT to improve the education of students with learning
disabilities. 281, 63–70 (2008)
10. Sulaiman, T., et al.: The level of cognitive ability among learning disabilities children in
Malacca Malaysia. Int. J. Psychol. Stud. 3(1), 69–77 (2011)
11. HJ, T., C, SK., P, Woo.: Student Learning Disability Experiences, Training and Services
Needs of Secondary School Teachers. Malays J. Psychiatary. 17 (2010)
12. Polat, E., et al.: Adaptive web-assisted learning system for students with specific learning
disabilities: a needs analysis study. In: Educational Sciences: Theory & Practice,
pp. 3243–3258 (2012)
13. Ozyurt, O., et al.: Uzwebmat: a framework for expert system based on personalized adaptive
and intelligent tutoring system for mathematics. In: IADIS International Conference e-
Learning, pp. 173–180 (2011)
14. Mahdjoubi, L., Akplotsyi, R.: The impact of sensory learning modalities on children’s
sensitivity to sensory cues in the perception of their school environment. J. Environ. Psychol.
32, 208–215 (2012)
15. Khan, B.H.: ELearning-Chapter 5: Layout 1. In: The Global e-Learning Framework
ed,STRIDE pp. 42–51 (2010)
16. Huang, K., et al: Breaking the sound barrier: designing an interactive tool for language
acquisition in preschool deaf children. In: Proceedings of the 7th International Conference on
Interaction Design and Children, IDC ‘08 (2008)
17. Parton, B.S., Hancock, R.: Vision 3D: digital discovery for the deaf. In: International
Symposium on Instructional Technology and Education of the Deaf, Rochester, New York,
June 2008
18. Wei, C.W., et al. A joyful classroom learning system with robot learning companion for
children to learn mathematics multiplication. Turkish Online Journalof Education. 11–23
(2011)
19. Bencheva, N.: Learning styles and e-learning face-to-face to the traditional learning. Vol.49
(2010)
Chapter 85
Palmprint Identification Using Invariant
Moments Algorithm Based on Wavelet
Transform

Inass Shahadha Hussein and M.J. Nordin

Abstract Because of the uniqueness of palmprints found on the palms of humans,

palmprint identification has been used in several applications. It is usually asso-
ciated with criminal identification, and has now become more popular in civilian
applications. Therefore, the aim of the proposed model is to improve personal
identification based on extracting shape feature using moments algorithm based on
wavelet transform and matching algorithm, which is proposed in this model. This
model has shown promising results without affecting rotation, translation and
scaling of objects, because it is associated with the use of a good description of
shape features. This system has been tested using databases from the Chinese
Academy of Sciences (CASIA), in Beijing. By using false rejection rate (FRR) and
false acceptance rate (FAR), we calculated the accuracy of identification. The
experiment shows 98 % identification rate in the CASIA database.

85.1 Introduction

A wide variety of systems require reliable personal authentication schemes to

either confirm or determine the identity of individuals requesting their services.
The purpose of such schemes is to ensure that the rendered services are accessed
by a legitimate user, and not anyone else. Traditionally, passwords (knowledge-
based security) and identification cards (token-based security) have been used to
restrict access to systems. However, security can be easily breached in these
systems when a password is divulged to an unauthorized user or a card is stolen by

I.S. Hussein (&)  M.J. Nordin

Faculty of Information Science and Technology, Universiti Kebangsaan Malaysia,
Bangi, Selangor, Malaysia
e-mail: [email protected]
M.J. Nordin
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 905

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_85
906 I.S. Hussein and M.J. Nordin

an impostor. Furthermore, simple passwords are easy to guess (by an impostor)

and difficult passwords may be hard to recall (by a legitimate user). The emergence
of biometrics has addressed the problems that plague traditional verification [1].
Among all biometric traits, palmprint has one of the highest levels of reliability
and has been extensively used by forensic experts in criminal investigations. A
palmprint is obtained from the central surface of the hand, between the wrist and
the fingers. Figure 85.1 shows the shape and details of a palmprint.
A simple biometric system has four important modules [2], which are the
following:
• The Sensor Module, which captures the biometric data of an individual. An
example is a palmprint sensor that captures palmprint impressions of a user.
• The Feature Extraction Module, in which the acquired data is processed to
extract feature values. For example, extract shape feature by using moments
algorithm.
• The Matching Module, in which the feature values are compared against those
in the template by generating a matching score. For example, in this module,
the number of matching minutiae between the query and the template can be
computed and treated as a matching score.
• The Decision-making Module, in which the user’s claimed identity is either
accepted or rejected based on the matching score generated in the matching
module.
The rest of paper organized as follows: Sect. 85.2 explains in details the
methodology that used in this paper. Section 85.3 introduces the results and dis-
cussion. Finally, Sect. 85.4 concludes the paper.

85.2 Methodology

This study explains and discusses the proposed method in performing palmprint
recognition. The first stage is the acquisition of hand images from the CASIA. The
second stage is pre-processing, which affects two main steps: including the bi-
narizing hand image and extracting the palmprint region from the whole hand.
This stage helps strip images from noise. The final stage includes feature extrac-
tion methods and matching techniques, such as invariant moments, used in com-
puter vision applications. In this stage, the shape descriptor approach is employed
in extracting global features of the palmprint, based on the wavelet transform, by
decomposing each palmprint image into four bands and deriving seven invariant
moments for each band. Finally, matching is carried out to perform identification.
This method can generate more accurate results in less time. Points of interest in
the identification of the palmprint are the features of an image, the change in the
scale and rotation of the image’s features.
85 Palmprint Identification Using Invariant Moments Algorithm … 907

Fig. 85.1 Shape and details

of a palmprint

85.2.1 Hand Image Acquisition (Data Set)

The databases were obtained from the CASIA. Images of the hand were obtained
using a flatbed scanner set at 256 grayscale. Grayscale images are mere repre-
sentations of the image using only one color, which is gray, to cover the entire
image. In a grayscale image, the density of the color component is used to rep-
resent the image. This component carries more information about the image. It
also reduces the size of the image dimensions, for example, instead of three bytes,
the image will be reduced gradually to a single byte (which is 0–255 in the decimal
system). The image, when converted to this type of color range, it becomes easier
to explain and promote. This property is also another feature of the grayscale mode
[3]. In this study, we used the CASIA database. The CASIA database contains 800
images of the hands of 50 subjects. All images measure 640 9 480 pixels. Each
user submitted eight pictures of their right hand and eight pictures of their left
hand. The images show 256 gray levels on the flatbed scanner. The scanner is link-
free, and therefore users are free to put their hands anywhere on the scanner as
long as their hands are placed face down. Figure 85.2 shows samples of the left
hand images. To ensure that the reproduction of your illustrations is of a rea-
sonable quality, we advise against the use of shading. The contrast should be as
pronounced as possible.

85.2.2 Image Pre-processing

Image processing is an area described by the need for intensified efforts to create
an experimental feasibility of the proposed solutions to a specific problem. Image
processing is a discipline in computer science. Each image has a value of more
908 I.S. Hussein and M.J. Nordin

Fig. 85.2 Samples of the left hand images

than tens of thousands of words and provides a number of meanings. The digital
images can be defined through the various definitions. For example, a digital image
can be formulated as a function of two dimensions, and (X, Y), where x and y are
spatial coordinates [4]. Values F (X, Y) represent the clarity or jamming of the
image at a certain point. All the digital images of different elements, each element
have a location and a value. These elements are referred to as image elements and
pixels. The image processing is the study of the use of a variety of algorithms that
were conducted on the pictures, where all of the input and output of the algorithms
is the picture. First, we binarized hand image from the original image, at this stage
a global threshold used to extract the image from rare hand. Figure 85.3 shows the
original image and binarized image.
After that, extraction of the region from the whole hand will be done [5].
Figure 85.4 shows a sample of the extracted images.

85.2.3 2D Wavelet Transform

After extracting the region from the whole hand, 2D filter banks will apply to these
regions, to get decomposed images into four bands [6], low–low pass filter, low–high
pass filter, high–low pass filter and high-high pass filter (LL, LH, HL, HH) (Fig. 85.5).

85.2.4 Invariant Moments Algorithm

Recognition of objects efficiently can extract feature points of objects (palmprint)

to create a powerful feature descriptor or the representation of objects. Hu [7] has
introduced a technique to extract the shape features form of images, which are
85 Palmprint Identification Using Invariant Moments Algorithm … 909

Fig. 85.3 Binarized hand images of the left and hands

Fig. 85.4 Samples of

extracted left palmprint of
one user from the CASIA
database

called invariant moments. These features are invariance to translation, rotation,

and change in scaling.
The advantage of the moments over other techniques is the implementation of
the previous descriptors is straight forward, and they also carry a ‘‘physical’’
interpretation of boundary shape. Hu was first to set out the mathematical foun-
dation for two-dimensional moment invariant and demonstrated their applications
910 I.S. Hussein and M.J. Nordin

Fig. 85.5 2D filter bank analysis

to shape recognition. Invariant moments were first applied to aircraft shapes and
have been proven to be fast and reliable. These values are invariant with respect to
translation, rotation and shape scaling.
Hu defines seven of these shape descriptor values computed from central
moments through order three that are independent to object translation, scale and
rotation.

U1 ¼ g20 þ g02 ð85:1Þ

U2 ¼ ðg20  g02Þ2 þ 4g211 ð85:2Þ

U3 ¼ ðg30  3g12Þ2 þ ð3g21  g03Þ2 ð85:3Þ

U4 ¼ ðg30 þ g12Þ2 þ ðg21 þ g03Þ2 ð85:4Þ

U5 ¼ ðg30  3g12Þ ðg30 þ g12Þ ½ðg30 þ g12Þ2

 3ðg21 þ g03Þ2 þ ð3g21g03Þðg21 þ g03Þ ð85:5Þ
½3ðg30 þ g12Þ2ðg21 þ g03Þ2

U6 ¼ ðg20g02Þ½ðg30þg12Þ2ðg21þg03Þ2
ð85:6Þ
þ4g11ðg30þg12Þðg21þg03Þ

U7 ¼ ð3g21  g03Þðg30þg12Þ½ðg30þg12Þ2
 3ðg21þg03Þ2þð3g12  g30Þ ðg21þg03Þ ð85:7Þ
½3ðg30þg12Þ2  ðg21þg03Þ2
85 Palmprint Identification Using Invariant Moments Algorithm … 911

85.2.5 Identification

To obtain the identification accuracy of our palmprint system, each of the palm-
print images was matched with all of the palmprint images in the database. A
matching is noted as a correct matching if two palmprint images are from the same
palm. If we suppose the query image is an X, and images in the database are N, we
must compare X with N.

85.3 Results and Discussion

The main objective of this study is to obtain higher authentication accuracy by

using a proposed algorithm after performing all previous stages. In this stage, 50
users from the CASIA database had features of their palmprints extracted using the
invariant moment algorithm. This stage is considered fundamental in building the
proposed system of recognition. To find out how accurate this algorithm is, we
need to determine the accuracy rate of the identification phase. For determining
accuracy rate, we have used two factors, which were FRR and FAR.

85.3.1 Results Based on Wavelet Transform

After applying wavelet transform (2D filter bank). The output is images that had
been decomposed into four bands (LL, LH, HL, HH). Figure 85.6 shows an
example of decomposed palmprint images.

85.3.2 Results Based on Moments Algorithm

After applying wavelet transform on palmprint images and decomposing these

images into four bands (the output from the previous stage), The invariant
moments algorithm was applied for deriving seven moments from each band.
Table 85.1 shows values of the seven moments of four bands for one palmprint
user (Table 85.2).

85.3.3 Accuracy Rate of Identification Phase

For the purpose of calculating the accuracy rate of the proposed method to perform
the identification phase by matching query palmprint image with each one stored
in the database based on FRR and FAR. To calculate the FRR factor, samples of
912 I.S. Hussein and M.J. Nordin

Fig. 85.6 Samples of

decomposed palmprint for
five users

Table 85.1 Values of the seven moments of four bands for one palmprint user
Moments LL LH HL HH
First moment 9.66809E-04 0.72658299392 0.49137043951 34.0659780176
Second moment 6.78459E-12 3.3876777E-03 4.8144644E-04 457.609538309
Third moment 2.08395E-13 8.92198E-03 6.57074E-03 14562.2738551
Fourth moment 1.14094E-13 9.5015245E-03 3.8902797E-03 1665.46619179
Fifth moment –3.1786E-20 7.7039764E-04 2.3161658E-04 –4608230.8966
Sixth moment 2.19293E-19 5.1328263E-04 –4.361702E-05 27447.6711358
Seventh moment –1.1909E-26 –6.772396E-05 –1.531511E-05 6526394.74011

Table 85.2 Final seven moments which were derived from Table 85.1 for one palmprint
Moment Value
First moment 8.82122456503965
Second moment 114.403351858058
Third moment 3640.57233696204
Forth moment 416.369895895987
Fifth moment –1152057.72390458
Sixth moment 6861.91790138718
Seventh moment 1631598.68500723
85 Palmprint Identification Using Invariant Moments Algorithm … 913

Table 85.3 Accuracy type

No. set FRR (%) FAR (%) Accuracy (%)
1 6.66 6.66 93.33
2 6.66 0 96.66
3 0 0 100
4 0 0 100
5 0 0 100
Rate 1.33 2.66 98

Fig. 85.7 Accuracy rate for Accuracy rate of identification

CASIA database of phase for CASIA database
identification phase 105
Accuracy (%)
100

95
accuracy
90
0 2 4 6
No.of palmprints’ set

five sets were taken, the size of each set was 15 palmprint templates. These
samples belong to the same samples stored in the database. To calculate the FAR,
samples of five sets were taken, the size of each set was 15 palmprint templates.
These samples were of out the stored database. Finally, accuracy rate was cal-
culated based on FRR and FAR as: Accuracy = (100 - (FRR + FAR)/2).
According to the results in Table 85.3, it was observable that the highest
accuracy was at set no. 3, 4 and 5, with a value of 100 %. The results had the
highest FRR, which was 6.66 %, which means, just one palmprint of the fifteen
valid users was rejected, and with the same high rate (6.66 %) of the FAR, means
that one of fifteen invalid palmprints was accepted. The accuracy rate is 98 %
(Fig. 85.7).

85.4 Conclusion

All methods used in this study are based on important steps in the processing and
recognizing of palmprints. The major important steps include preprocessing,
feature extraction, and matching. All of the aforementioned processes were carried
out in a variety of ways and algorithms, which are suitable for the structural
condition of the image itself. The results of the experiment show that the overall
goal of using invariant moments on the basis of the wavelet transform is
914 I.S. Hussein and M.J. Nordin

successfully implemented on palmprints. It was found that moments of extracting

palmprint to be invariant to rotation, translation and scaling. Transformation was
applied to decompose palmprint images, that made moments more stable imple-
mented to achieve its goal.

Acknowledgment The authors wish to thank Universiti Kebangasaan Malaysia (UKM) and
Ministry of Higher Education Malaysia for supporting this work by research grant ERGS/1/2012/
STG07/UKM/02/9.

References

1. Ross, A.: Information Fusion in Fingerprints Authentication. Michigan State University,

Michigan (2003)
2. Jain, A., Hong, L., Pankanti, S., Bolle, R.: An Identity Authentication System Using
Fingerprints Pattern Recognition and Image Processing Laboratory. Michigan State Univer-
sity, Michigan (1997)
3. Abdullah, A., Veltkamp, R.C., Wiering, M.A.: Fixed partitioning and salient points with
Mpeg-7 cluster correlograms for image categorization. Pattern Recogn. 43(3), 650–662 (2010)
4. González, R.C., Woods, R.E.: Digital Image Processing, 1st edn., pp. 703. Addison- Wesley,
Reading, Massachusetts (1993)
5. Pavlidis, T.: Algorithms for Graphics and Image Processing, p. 416. Computer Science Press,
Rockville, Maryland (1982)
6. Zainal, Z.A., Manaf, M., Shibghatullah, A.S., Jusoff, K., Ahmad, R., Ayop, Z., Anawar, S.,
Shaaban, A., Yusoff, M.: A New Hybrid Embedding Method in Iris Biometric System
Australian. J. Basic Appl. Sci. 7(3), 46–50 (2013)
7. Hu, D., Feng, G., Zhou, Z.: Two-dimensional locality preserving projection (2dlpp) with its
application to palmprint recognition. Pattern Recogn. 40(1), 339–342 (2007)
Chapter 86
Auto Mobile Ad Hoc Mechanism in Delay
Tolerant Network

Muhammad Affandy Azman, Sharifah Hafizah Syed Ariffin,

Norsheila Fisal, Mazlan Abbas, Mohd Husaini Mohd Fauzi
and Sharifah K. Syed-Yusof

Abstract Delay Tolerant Network (DTN) is known as the solution to an Internet

network where connectivity is an issue. There are existing project which uses
smartphone as a physical transport of data between these intermittent networks.
However, each smart phone needs to connect to each other via an infrastructure
which will result in lower successful transfer rate. An automated mechanism is
proposed for Ad Hoc connection between the smart-phones is to ensure connec-
tivity which leads to higher successful transfer rate. This report presents the
automated mechanism called Auto Mobile Ad Hoc Network in Delay Tolerant
Network that is able to provide better reliability for data transmitted through DTN.
This mechanism also allows application of the current Delay Tolerant Network to
be connected to other networks and other nodes automatically using Ad Hoc mode.

M.A. Azman
S.H.S. Ariffin (&)
N. Fisal
M.H.M. Fauzi
S.K. Syed-Yusof
UTM-MIMOS Center of Excellence, Universiti Teknologi Malaysia, UTM Skudai,
81310 Johor Bahru, Johor, Malaysia
e-mail: [email protected]
M.A. Azman
e-mail: [email protected]
N. Fisal
e-mail: [email protected]
M.H.M. Fauzi
e-mail: [email protected]
S.K. Syed-Yusof
e-mail: [email protected]
M. Abbas
MIMOS (M) Bhd, Technology Park, 57000 Kuala Lumpur, Malaysia
e-mail: [email protected]

Ó Springer International Publishing Switzerland 2015 915

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_86
916 M.A. Azman et al.

86.1 Introduction

Internet plays an important role to connect people regardless of distance. However,

not everyone has the opportunity to use the Internet, because of the challenging
issues due to delay and disruptions. One of the most challenging issues for a
certain geographical area to have communication infrastructure is the lack of
infrastructural and facilities. Delay Tolerant Network (DTN) architecture for
challenged network are categorized in [1] where DTN can be implemented in four
different challenged networks shown in Fig. 86.1. There are terrestrial mobile
networks where users are mobile and commuting from one network to another,
there is also exotic media networks where near-earth satellite communication that
is prone to latencies. There are two other challenged networks in DTN which are
the military Ad Hoc network where intentional jamming might be the cause of the
disconnection and network need to compete for bandwidth, and sensor/actuators
network where this networks are consist of millions of low powered nodes and
communication within this network often scheduled to conserve power.
There are projects which uses the DTN approaches such as in [2–4] to appre-
hend these challenges. One of the projects is Bytewalla: DTN on Android [5] that
uses mobile devices to transfer data between nodes. This project aims at con-
necting African rural villages using Android phones with delay-tolerant net-
working. The idea is that people carrying their mobile phones and travel from
villages to cities will carry data with them. Once they reach the city, they will
connect to any WiFi access point and upload the data. For the past years, Byte-
walla had evolved from version 1–5, each with new features such as security and
routing protocols. However, there are several issues in Bytewalla that can be
improved. In the current version of Bytewalla both nodes (i.e. transmitter and
receiver) need to be connected manually to an infrastructure or the same network
to allow data to be transferred. If both nodes are not connected in a same network,
the data will not be forwarded, and will be stored in the transmitter node until the
data expires, which affects the reliability of data transfer.
In the present report, we will discuss the implementation of an automatic
mechanism called Auto Mobile Ad Hoc function using Bytewalla. This mecha-
nism is to improve the probability of successful bundle (a protocol data unit of the
DTN bundle protocol) send and receive by enabling all nodes to connect to each
other automatically rather than to wait for an available Access Point (AP). Each
node will act dynamically according to the certain situation. This implementation
is still using the conventional way of transferring data by adding another layer in
the Internet protocol suite which is the Bundle Protocol [6] and will be imple-
mented on a smart-phone. This method is to enhance the existing Bytewalla
software and to increase reliability data transfer as well as reduce transmission
delay. Figure 86.1 shows the challenged network in DTN.
The report is organized as follows: Sect. 86.2 describes the protocols used in
DTN and the transmission routings. Section 86.3 presents the proposed concept of
86 Auto Mobile Ad Hoc Mechanism in Delay Tolerant Network 917

Fig. 86.1 Challenged

network in delay tolerant
network

Auto Mobile Ad Hoc Mechanism. In Sect. 86.4 development and implementation

in experimental DTN test bed is presented in detail. In Sect. 86.5, results and
discussion are elaborated. Last but not least, Sect. 86.6 concludes the report.

86.2 Data Protocols and Routings in DTN

The DTN is introduced to solve technical issues in networks that may lack of
continuous network connectivity. The DTN is designed to operate effectively over
extreme distances. An example of networks that are facing issue is the space-
network, such as Inter Planetary (IPN) Internet Project [7, 8]. The DTN is used to
tackle problems such as the intermittent connectivity, a long and variable delay,
asymmetric date rates, and high error rates by using the store and forward
switching [9, 10]. In order to implements this method, a device with a persistent
storage is needed to hold the message indefinitely. Example of the device is a
smart-phone. Figure 86.2 shows the store and forward method used.
It is possible for DTN to implement the store and forward method by overlaying
a new protocol layer which is the Bundle Protocol (BP), RFC5050 on top of the
Transport Layer. Basically, a bundle is a packet which contains a source-appli-
cation user’s data, control information and a bundle header. Each node is identified
by Endpoint Identifiers (EID), which can be treated as address. There are several
practical applications in DTN that have been done by other researchers. One of
them is DakNet, which was developed by the MIT media Lab researchers [11].
Their goal was to provide remote villagers with low-cost digital communication by
918 M.A. Azman et al.

Fig. 86.2 Store and forward method

equipping busses with a mobile access point that travels between village kiosk and
city collecting data.
Routing is the main aspect of DTN. It determines the delivery success rate and
the delay of the bundles. Currently there are two routing strategies in DTN. The
first strategy is flooding, where the message is replicated to several nodes in order
for destination EID to be received. Each node acts as a relay to store the bundle
until they are able to contact with another node. One of the routing protocols that
use flooding concept is Epidemic routing [12], where nodes replicate and transmit
messages to newly discovered contacts continuously. The problem with Epidemic
routing is that it may congest the network in clustered areas while wasting network
resource (bandwidth, storage and energy).
The second strategy is forwarding strategy which uses the best path to the
destination by making use of network topology and local or global knowledge to
find the best route path to deliver the message to the destination without replication
which can lead to less bandwidth and consumption, and faster. One of the routing
protocols that use forwarding strategy is the Probabilistic Routing Protocol using
History of Encounters and Transitivity (PRoPHET) [13]. This routing protocol
uses algorithm which maintain the set of probabilities for successful delivery for
known the destination. The downside for this protocol is that the actual value for
the probability is crucial for successful bundle transfer. The routing protocol used
in this project is the Epidemic routing. This is to ensure the successful transfer of
bundle because of the high transfer probability rate in Epidemic routing.

86.3 Auto Mobile Ad Hoc Network

With the current DTN mechanism, all transmission from one mobile device to
another need to be done manually even though there are potential node that can
take the data to be forwarded later. A mechanism is added to solve this connec-
tivity issue where the application will not connected to any network/node, if it is
not initiated manually. This proposed mechanism will scan and connect auto-
matically to another network/node.
The architecture in the experiment consists of three nodes (i.e. Android phones)
which had been rooted (i.e. ability to access to system files and modify). Fig-
ure 86.3 illustrates the overall experiment scenario architecture. Each node will
store and forward received bundle. The bundle will keep on forwarded until it has
reached the destination EID. Each of the nodes is connected with each other
through Ad Hoc network.
86 Auto Mobile Ad Hoc Mechanism in Delay Tolerant Network 919

Android A Android B

Android A

Android B Android C

Fig. 86.3 Experimental scenario of the auto mobile Ad Hoc network

Start

Connected to a network

Check
Status Not connected to any network

Create / Connect to Ad-Hoc

Connected to another
Network
network

Fig. 86.4 State diagram of the auto mobile Ad Hoc network

Figure 86.4 shows the state diagram for Auto Mobile Ad Hoc mechanism in the
DTN, which shows the flow for each node’s network connection. The node will
continuously check for its own network status, if the node is connected to any
network, the node will stay connected and resume the checking phase. If the node
is not connected to any network, the node will then scan and attempt to connect to
any nearby Ad Hoc node. If there are no Ad Hoc nodes nearby, the node will then
create its own Ad Hoc network.

86.4 Experimental DTN Test Bed

The experimental DTN test bed consist of DTN Servers and mobile node that have
been configured and embedded with auto mobile Ad Hoc function.
920 M.A. Azman et al.

86.4.1 DTN Server

The DTN 2 application uses DHCP server, DTN2 software on Ubuntu Platform. In
order to send a bundle, an EID and a payload (message) is needed. Each node
haves their own unique EID. For the smartphone, the EID is determined by the
MAC address. An EID can be treated as an address for each node. Referring to
Fig. 86.3, The bundle will be send to the mobile node that is in the same network
with the server (i.e. network A), which is Android A.

86.4.2 Mobile Node

For the mobile side, an application is called Bytewalla (Bytewalla is a DTN

application for Android1) is used. Each mobile node is installed with a modified
version of Bytewalla which has the Auto Mobile Ad Hoc functionality. Each
mobile node is also rooted, to enable the use of Ad Hoc in mobile node. Each node
has its own unique EID. For the Android phones, the EID is determined by the
MAC address. An EID can be treated as an address for each node. Referring to
Fig. 86.3, a bundle will be sent to Android C which is the EID, originated from
Android A. All of the nodes are not originally connected to each other. When
Android A is in range with a nearby node (in this case Android B), Android A and
Android B will initiate a connection automatically and the bundle will be send to
the Android B from Android A. The bundle will then store and forwarded to the
next node until it has reached the EID (Fig. 86.5).

Fig. 86.5 Test bed set up

1
Android-A linux based operating system for touch screen phones.
86 Auto Mobile Ad Hoc Mechanism in Delay Tolerant Network 921

86.5 Results and Discussions

We investigate the effects of message size and number of nodes to the overall
delay. The overall delay is calculated by calculating the time taken for a bundle to
arrive to an EID from source. The measurement of the overall delay for each
message size and number of nodes is carried out three times.
In the first experiment, we determined the delay based on the message size and
different approaches, Android A and B were used as the nodes. Android A sent the
bundle to Android B and both nodes are originally not in the same network. In this
experiment we compared the delay of data transmission between infrastructure
network and Auto Mobile Ad Hoc network.
Figure 86.6 shows the results of the first experimental setup for the delay of
transmitted data with variable data size. As the data size increased, the delay also
increased. Using the infrastructure setup, the minimum delay for 200 kB of
message size is about 14 s. The delay increased gradually based on the data size to
about 23 s for 1000 kB of message size. For the Ad Hoc setup, the minimum delay
is about 34 s for 200 kB of message size and increased gradually to about 42 s for
1000 kB of message size. The difference in delay between Ad Hoc and Infra-
structure is because of the time needed for an Android phone to create/establish an
Ad Hoc network which takes around 30–32 s.
In the second experiment, we determined the delay based on the number of
nodes and here, Android A, Android B and Android C were used. In this setup, all
of the nodes are originally not in the same network. Android A will be the source
sending the bundle to Android B or Android C which is the EID. In the second
experimental setup, we compared between 2 and 3 nodes of data transmission from
source to destination. In the 2-node setup, Android A and B are used. In the 3-node
setup, Android A, Android B, and Android C are used.
Figure 86.7 shows the results from the second experiment of the delay with
variable number of nodes. For 3 nodes, the minimum delay is about 65 s for
200 kB of message size which increases to about 83 s of delay for 1,000 kB of
message size. As we increased the number of nodes, the overall delay also
increased almost double. The delay increases from 2 to 3 nodes because all of the
nodes are not originally in the same network, thus each nodes creates their own Ad
Hoc which consume time.
Figure 86.8 shows the delay compared between the original Bytewalla and the
Auto Mobile Ad Hoc Network. As shown in Fig. 86.8, the Auto Mobile Ad Hoc,
the delays increased as we increase the message size to a maximum of about 42 s.
However, for the original Bytewalla, because of the nodes originally not in a same
network, the nodes will not be connected to any other node and the bundle will not
be send to the next node, hence, the delay is infinity.
922 M.A. Azman et al.

Fig. 86.6 Delay versus 2 Nodes 2 Nodes (Adhoc)

message size for (Infrastructure)
infrastructure and Ad Hoc 45
mode DTN 40

35

Delay (s)
30

25

20

15

10
200 400 600 800 1000
Message Size (kB)

Fig. 86.7 Delay versus the 3 Nodes 2 Nodes

number of mobile nodes the 90
message are transmitted in
80
DTN
70
Delay (s)

60

50

40

30
200 400 600 800 1000
Message Size (kB)

Fig. 86.8 Reliability of the 2 Nodes (Original)

Ad Hoc function in DTN
compared with the original
version of DTN
86 Auto Mobile Ad Hoc Mechanism in Delay Tolerant Network 923

86.6 Conclusion

This report has presented the implementation of Auto Mobile Ad Hoc function in
the existing DTN application using Bytewalla software embedded in mobile nodes.
The Auto mobile Ad Hoc function is to improve the probability of successful
bundle transfer to the destination by adding the ability to connect to each node
automatically. This network concept can be very fruitful for the current situation
since the bundle will keep on transferring to the next node instead of relayed only
on one node. From the results shown, the Auto Mobile Ad Hoc Network function
is able to connect automatically to other nodes without user prompt which
improved the probability of successful bundle transfer. However, there are delays
involves, which is the time needed for a node to create its own Ad Hoc network
and connects to other Ad Hoc network. The Auto Mobile Ad Hoc Network cer-
tainly has its drawback by having about 32 s of delay for the Ad Hoc network
creation. However, it will ensure higher reliability of data transfer through the
automated connection of nodes through Ad Hoc network.

Acknowledgments The authors would like to thank all those who contributed toward making
this research successful. The authors wish to express their gratitude to Ministry of Higher
Education (MOHE), Malaysia Research Management Center (RMC) for the sponsorships, Te-
lematic Research Group (TRG), Universiti Teknologi Malaysia and MIMOS (M) Bhd for the
financial support and advice of this project. (Vot number Q.J130000.7823.4L550).

References

1. Fall, K.: A Delay-Tolerant Network Architecture for Challenged Internets. ACM SIGCOMM
2003, Germany
2. Fall, K., Farrell, S.: DTN: An architectural retrospective. IEEE J. Sel. Areas Common 26(5),
826–828 (2008)
3. MacMahon, S.F.: Delay-and disruption-tolerant networking. IEEE Internet Comput. 13(6),
82–87 (2009)
4. Cerf, V., Hooke, A., Torgerson, L., Dust, R., Scott, K., Fall, K., Weiss, H.: Delay-tolerant
networking architecture. IETF RFC 4838 (2007)
5. Ntareme, H., Zennaro, M., Pehrson, B.: Delay tolerant network on smartphones: applications
for communication challenged areas. ExtremeComm11’, ACM International Conferences
Proceedings Series, doi10.1145/2414393.2414407
6. Scott, K., Burleigh, S.: Bundle Protocol Specification IETF RFC 5050 (2007)
7. Akyildiz, I.F., Akan, O.B., Chen, C., Fang, J., Su, W.: InterPlaNetary internet: state-of-the-art
and research challenges. Comput. Networks J. (Elsevier Science) 43(2), 75–112 (2003)
8. Burleigh, S., Hooke, A., Torgerson, L., Fall, K., Cerf, V., Durst, B., Scott, K.: Delay-tolerant
networking: an approach to Interplanetary Internet. IEEE Commun. Mag. 41(6), 128–136
(2003)
9. Chuah, M., Yang, P., Davidson, B.D.: Store-and Forward performance in DTN, IEEE VTC
(2006)
10. Macmahon, A., Farell, S.: Delay and disruption-tolerant network. Internet Computing IEEE
(2009)
924 M.A. Azman et al.

11. Pentland, A., Fletcher, R., Hasson, A.: Daknet: rethinking connectivity in developing
Nations. Computer 37(1), (2004) doi:10.1109/MC.2004.1260729
12. Jain, S., Fall, K., Patra, R.: Routing in a delay tolerant network. In: SIGCOMM ‘04
Proceedings of the 2004 Conference on Applications, Technologies, Architectures, and
Protocols for Computer Communications, vol. 34 (4), pp. 145–158
13. Vasilakos, A.V., Zhang, Y., Spyropoulus, T.: Delay Tolerant Network: Protocols and
Applications. CRC Press, ISBN 1439811083 (2012)
Chapter 87
An Exploratory Study on Blind Users’
Mental Model in Computer Accessibility

Manoranjitham Muniandy and Suziah Sulaiman

Abstract The growing needs in the area of HCI has led to many new researches
focusing on user-centric design, investigating the problems faced by a computer
user and ways to produce an effective yet efficient design. Often, designers need to
acquire the knowledge and the experience of the users to produce a good interface
design. This can be referred as a wise exploitation of user’s mental model. The
paper serves as a preliminary study which explores the mental model of blind users
as a contribution in improving the accessibility of computer applications. Based on
the blind user’s perception, imagination and interpretation, the study indicates that
touch sensation plays an important role in improving the representation of a
computer application to them. Associating touch sensation with information
familiar to the blind users enhances the learning process.

87.1 Introduction

The term ‘‘computer accessibility’’ in human-computer interaction spells out

clearly the accessibility of computer system to all people, regardless of disability
or severity of impairment [1]. The user interface of a computer should be acces-
sible by anyone, independent of their physical, perceptual-motor, social and cul-
tural capabilities [2]. In [3], it is stated that blind users usually use software known
as ‘‘screen readers’’ which works with all the programs installed in a computer to
read out the interface and details of an application to them. However, not all can be

M. Muniandy (&)
S. Sulaiman

Department of Computer and Information Sciences, Faculty of Science and Information
Technology, Universiti Teknologi PETRONAS, Bandar Seri Iskandar,
31750 Tronoh, Perak, Malaysia
e-mail: [email protected]; [email protected]
S. Sulaiman
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 925

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_87
926 M. Muniandy and S. Sulaiman

read by the screen readers. Braille devices able to produce the output on the
computer display however, some important information such as application names
and pop-ups are not displayed by the Braille device.
World Health Organization (WHO) recorded the estimated visually impaired
population as 285 million; with the blind holding the record of 39 million and the
rest occupied by the low vision category [4]. Eweek.com did addressed an
important question on why the blind community are being ignored when they are
able to create a big market? [5]. The challenges in life and acquiring a new skill
especially on IT field seem to be an issue to the blind. It is important to allow the
blind to further venture into the IT equipped world by providing an easy learning
platform for them. As a start, this paper presents a study that identifies the usability
problems based on users’ experience when interacting with a computer applica-
tion. The intention is to understand the blind users’ mental model. The study
further investigates the extent in which touch sensation could be used in accessing
the application.

87.2 Literature

Designers tend to address a user’s mental model as a valuable piece of information

due to the ability of these users to formulate the mental model using their expe-
rience and expectations. In order to design a user’s preferred system, one need to
utilize the user’s mental model as this information is found to be greatly con-
tributing during the development stage. Every designer considers a mental model
to be an absolutely necessary element as it helps them to understand a user’s view
on system content [6, 7]. Usually, when a user interacts with an existing or familiar
environment, they tend to use the previously developed mental model and if they
come across an unfamiliar or a new environment then they will create a new
mental model altogether [6].
Familiarity can be achieved when a user’s mental model is analyzed and some
of the important questions are answered such as (1) how a system should look
alike? Or (2) how a system supposedly should behave? Or (3) how a user expects a
system to respond? Mental model of a blind can be in any form, such as a diagram,
an image, a theory, a set of concepts, some guidelines or etc. [8]. This represen-
tation is said to be a great help in enhancing the understanding of usability of any
system [9, 10]. Usually, a system is build according to a designers’ view. In
another word, the developed system is based on a designer’s mental model. In this
study, mental model of blind users refers to the representation formed by the blind
when they use computer application.
According to [2], usability issues highlight the easy use of an interface which is
measured through the interaction between a user and the system. In [11], it was
addressed the need of assistive technology when a blind computer user accesses a
computer application. And that it is important for the designer to design a system
87 An Exploratory Study on Blind Users’ Mental Model in Computer Accessibility 927

or an application which is ‘‘software’’ friendly for the screen readers to be able to

read the presented content.
Often the blind users’ mental models are not taken into account and the
developed applications are tested without considering the fact that blind users have
the tendency to navigate through an application instead of listening carefully the
information that is being read [12]. A developed computer application must be
oriented to usability and accessibility for the users from any background. Referring
to Table 87.1, usability or also referred to as accessibility of an application is
defined in terms of five quality components [13]. Usually, an application is
designed by a designer having these five components in their mind as a benchmark
to ensure good applications with great accessibility level is produced for the users.
Current computer applications do not address the issues face by the blind. It can be
due to the fact that, the availability of screen reader is believed to perform its’
duty. And another reason can be due to the lack of awareness among the devel-
opers and also visually impaired are not even considered as a heavy computer user
but research has proven that a large number of blind users are now very inde-
pendent and learning new skills for survival.

87.3 Related Work

87.3.1 Mental Model and Blind Internet Users

In [14], the two-dimensional mental model created by the blind people when they
are using touch screen with audio feedback was discovered. These authors have
agreed that little work has been done on mental model for blind people. Mental
model of blind people created from surfing a two-dimensional web page is
investigated using two ways: firstly, is by using a touch screen display with audio
feedback and the other using a screen-reader only. Users were given 10 to 15 min
training on the each of this web page before the actual test was conducted. Five
users are required to surf the pages twice using screen reader for the first time and
using touch screen with audio feedback for the second time.
And another five did the same but using the touch screen with audio feedback
for the first time and screen reader for the second time. After that, all users are
given a set of foam blocks containing rough and smooth surface to construct the
diagrammatic representation of the pages. Rough surfaced blocks to represent the
headings while smooth surfaced blocks to represent the data. After the analysis of
these diagrammatic representation, it shows that previously (using screen reader)
the mental model of a blind user are in single column structure as quite a number
of the representation stacked up the headings foam blocks in one column or both
headings foam and data foam blocks in one column.
The same users had two dimensional mental model after using the touch screen
with audio feedback which shows that these assistive technology able to deliver
928 M. Muniandy and S. Sulaiman

Table 87.1 The five quality components [13]

Components Description
Learnability How easy is it for users to accomplish basic tasks the first time they encounter
the design?
Efficiency Once users have learned the design, how quickly can they perform tasks?
Memorability When users return to the design after a period of not using it, how easily can
they reestablish proficiency?
Errors How many errors do users make, how severe are these errors and how easily
can they recover from the errors?
Satisfaction How pleasant is it to use the design?

the content of a layout better than a screen reader. As these users, tend to stack up
the headings foam and data foam blocks in two-dimensional (table) format.
However, their research also shows that the theory is applicable better if the layout
is simple and not much different can be seen if the layout is complex. Therefore, it
can be concluded that the two-dimensional mental model of the blind people is
highly influenced by the complexity of a web page.
In [15], it was highlighted the issues faced by the blind society when navigating
web pages. They assess the web pages based on one important accessibility point
which is ‘‘Ease of navigation’’. They have also highlighted that ‘‘information
scent’’ plays an important role whereby related information are grouped or
arranged together and if a blind user skips their targeted topic, they will know it by
realizing the scent of the current topic which is different from the previous topic.
Though the research used a systematic evaluation process, the final outcome of this
research was merely an identification of the accessibility level among the e-
commerce sites in three different countries.

87.3.2 Difficulties Faced by the Blind Users Involving

the Sighted Individuals

A study by Fernando Alonso et al. focuses on the development of dual graphical

user interfaces for both visually impaired and sighted communities. They con-
centrated on how to come up with a design that will be able to satisfy both types of
users [16]. The main concern of this project was to deliver a set of guidelines that
can be used in the dual interface development in accordance with design for all
principle. Even though these authors did take into consideration for the blind users,
but the crossed checked guidelines had to drop out some issues that exist only in
the visually impaired environment since those issues does not co-exist in the
sighted user environment.
A research on the application of various stimuli to produce an effective pattern
which is believed to increase the effectiveness of represented information to the
user was conducted in [17]. The study suggested a design principle defining the
87 An Exploratory Study on Blind Users’ Mental Model in Computer Accessibility 929

stimulus pattern. It was a good study; however, the authors have failed to provide
the same experiment for all the users which are essential in deriving a conclusion.
To eliminate biasness during the evaluation, a change in the sequence of the tactile
pattern should be done instead of not giving certain patterns at all for some users.
Therefore, this shows that it is not right to conclude if the actual test subject is
differs for all the users.

87.3.3 Audio Metaphors for the Blind Users

A conversational metaphor was proposed in [18], in which the metaphor focuses

on the environment of 3D aural to allow the visually impaired for easy access of
information. The proposed hypermedial model contains direct graph nodes and
links which is used to map to certain speakers. Documents are represented by the
nodes and the semantic relationships between the documents are represented using
the links. A speaker is used to explain the detail of the nodes and links to the user.
Many samples of real people were used to generate the 3D voice of speakers using
an off-line way [18]. The authors have developed a metaphor for the visually
impaired, the proposed idea focused only in audio environment and did not address
the problems relating to computer accessibility or other usability issues related to
blind users.

87.3.4 General Improvement of the Surrounding Objects

for the Visually Impaired

A study was focused on improving the interactions between a blind and everyday
technological manmade object such as hand phones, software applications and etc.
In this article, interviews and observations were used to address the obstacles faced
by the blind in using the technological gadgets [19]. In order to improve the
gadgets, it is important to know the obstacles faced by the blind community while
using the surrounding objects. In another review, Gregory Petit et al. has generated
tactile graphics by translating the illustrations in the schoolbook to be accessed by
the visually impaired students. Two types of tactile hardware were used to gen-
erate the three different tactile rendering of an illustration [20]. A significant result
was shown in displaying the tactile illustration by applying a workable method-
ology; however, they did not propose any framework that can be used by others at
the end of the project.
930 M. Muniandy and S. Sulaiman

87.4 Motivation of the Study

Previous researches focuses on the problems faced by the blind in using a screen
reader thus not highlighting the inaccessibility of computer application to them [14,
18]. Some other findings relating to touch sensation were lacking of empirical study,
hence not supporting the contribution of touch sensation in learning computer
application [14, 17, 20]. Our initial investigation reveals that in order for a blind
user to select and to open an application such as Microsoft Word, the user need to
depend on the screen reader to read the file name or the name of the application.
This entire process takes plenty of time as user need to wait and listen to the entire
icon description to be read. Based on our findings, blind users possess rich tactile
sensation in dealing with their daily life. There is a need to conduct an exploratory
study on blind users to address their problems in accessing computer applications
and to evident the presence of the touch sensation as an important factor that
contributes effectively in improving the accessibility of computer applications.

87.5 The Study

The objectives of this study are twofold:

(i) To examine blind user’s experience on computer accessibility.
A thorough study will be conducted to identify the issues faced by the blind in
accessing computer applications.
(ii) To examine how touch feedback may assist in accessing computer
applications.
Touch sensation is known to work well with blind, however, in this study, it
will highlight to what extent touch can contribute in accessing computer
applications.

87.6 Study Methods

A group of blind users with the need to learn IT skills was selected as the study
object. As a good site to study, Malaysian Association of the Blinds (MAB) a
voluntary organization serving the blind was chosen. In here, some basic skills
such as mobility, Braille, English language and computer literacy are taught. The
researchers conducted their field work to study extensively the blind users’
experience and imagination during this entire rehabilitation course for one
semester (duration of five months). The following section discusses on the three
data collection methods used in this study.
87 An Exploratory Study on Blind Users’ Mental Model in Computer Accessibility 931

87.6.1 Personal Interviews and in-Depth Interviews

Several informal interviews and four in depth interviews were conducted with
seven blind users. One of them is the chairman, two of them are the instructors and
four of them are the students. The age group of the interviewees is above 18 and
their occupation varies tremendously. Interviews were chosen due to the nature of
this method which has a very high response rate and also the method encourages the
collection of true and correct responses addressing the nature of a problem [21]. The
section below describes some of the questions asked during the interview session:
• Can you easily open a Microsoft word application?
• How do you know that you are opening a word document instead of an excel
sheet?
• Have you ever made a mistake of opening other documents instead of the
intended document?
The questions somehow addressing the same issue which is on the users’
accessibility level when accessing an application during the learning process. The
main objective of the researcher for asking the similar questions is to ensure the
respondent’s response is consistent and correct directing to the main problem of
this research. Four in-depth interviews were also conducted to get the respondents
to further open up about their concern and dissatisfactions when dealing with
computer applications. In-depth interviews are useful when detailed information
about a person’s thoughts and imagination is required [21]. The section below
highlights two of the questions asked during the in-depth interview:
• In your imagination, how flexible do you want/prefer the existing computer
applications to be presented?
• In your opinion, what kind of senses seems appeals the most for a blind user?
These questions are designed to trigger the user to communicate their thoughts
and expectations. When a user describes their thoughts and expectations in detail,
it allows the researcher to identify the issues with the current application which
does not satisfy the user’s need.

87.6.2 Observation

Apart from interviews, observation was also done for the entire period of Braille
class and computer literacy class with the same set of participants from the
interview session except for the chairman. The observation was also video
recorded for multiple reviews. Observation was conducted due to the ability of this
technique recording the actual users’ behavior. Observation eliminates the ten-
dency of a respondent being bias or exaggerates their problems, captures the latest
happenings and only actual behavior is being captured [21]. The facial reactions
932 M. Muniandy and S. Sulaiman

and eagerness or lack of interest and irritations when a user encounters an issue
while working with computer applications is studied to identify the negative
emotions of a user and the factors which cause this negative emotion during their
learning process. By identifying a users’ emotion and factors which cause this
emotion, more accurate information is obtained by the researcher to support the
research questions of this research.

87.6.3 Hands-on Activities

Hands-on activities in the form of two simple games were conducted with four
participants to learn more about a blind person’s habit in applying their existing
experience on discovering knowledge. The experiment was performed with the
intention to explore and to identify the effect of touch sensation to the blind users.
Polystyrene blocks resembling the Braille cells are created and referring to
Table 87.2, blocks were used to represent some of the frequently used applica-
tions. The representations are new patterns and are not represented in Braille code.
The pattern of pop-up blocker and pattern of applications (apps) is a new creation
due to no suitable Braille code is available for the representation while the com-
bination of application (apps) code and alphabets referring to the first word of an
application is used for the complete representation. These inventions are used to
test the touch sensitivity level among the participants to explore the mental model
of the blind users to show that touch sensation able to contribute in improving the
accessibility of computer applications.

87.7 Results and Discussion

87.7.1 Results of Interviews

The information presented in Table 87.3, highlights the views of each key person in
terms of issues faced by the blinds in accessing computer applications. Referring to
the feedback, they have admitted that almost all of their students are unable to open a
simple application such as ms.word. This simple application needed repetitive
actions even with the aid of a screen reader such as listening and clicking repetitively.

87.7.2 Results of Observation

The below analysis was done based on the observation of these selected group of
participants and instructors. The video recorded observations were reviewed few
times in order to ensure that only real issues are captured for analysis purpose.
87 An Exploratory Study on Blind Users’ Mental Model in Computer Accessibility 933

Table 87.2 Patterns of the blocks to represent applications

Applications/description Block pattern Applications/description Block pattern
(1) Pop-up blocker (new) (4) Internet explorer (apps + i)
(2) Ms. Word (apps + w) (5) File (f + e)
(3) Ms. Excel (apps + e) (6) Folder (f + r)

Table 87.3 Issues and views of key person on the blinds

Role Issues and views
Chairman • The difficulty of the participants in following the entire program and some
abandoned half way due to the complex learning process
Instructors General issues faced by the students:
• Totally helpless when a pop up blocker pops out and inability to
understand the displayed graphics as the screen reader reads it out as
‘‘graphics’’. Expressed the need for additional assistive technology
Issues faced by the beginner students
• Inability to open a simple application such as Ms. Word and inability to
understand the words read by the screen reader due to the foreign accent
Participants/ • Inability to understand the word read by the screen reader, especially when
students a word spelled using local language (Malay language) is read
• Unable to figure out a pop-up blocker and inability to open some simple
applications. Express the need of some other additional technology

The second presented information in Table 87.4, points out the observed nature or
characteristics of the key persons during both interview session and observation
session.
The facial reaction and the body language of the participants were analyzed
carefully to derive the below emotions as stated in the Table 87.4. The observation
reveals that the instructor’s emotion is divided into both positive and negative
while the emotions of the students is mainly negative. This can be due to the
reason that the instructor is an advanced user while the students are just the
beginners. Based on Tables 87.3 and 87.4, the main factors contributing to the
‘‘inaccessible’’ elements for the blinds are identified.
The main factors contributing more to the ‘‘inaccessible’’ elements for the
blinds are arranged to the most frequently encountered to the less frequently
encountered by the users and also according to the most frustrating to the less
frustrating emotion felt by the users:
• Inability to open the right application.
• Blocks by the pop-up blocker.
• Having to listen repetitively the same information when missed out for the first
time.
• Inability to know the actual content of a graphic.
• Inability to understand the foreign accent.
934 M. Muniandy and S. Sulaiman

Table 87.4 Observed analysis of key person

Role Observed nature
Instructors • Appears calm and confident in moving around the applications and able to
confidently open some computer application and demo the class activities
• Seems a little impatient while waiting for the screen reader to read each
app and seems unhappy when a screen reader reads out a displayed graphic
as ‘‘graphic’’. Appears not happy when a pop up blocker blocks their input
Participants/ • Appears nervous and diffident when trying to open an app and appears
students frustrated when unable to open the right apps or when there is a pop blocker
block the screen. Appears blank when a screen reader reads out a displayed
graphic as ‘‘graphic’’
• Looks annoyed when unable to understand the word read by the screen
reader and having to listen again the same word in the attempt to understand

87.7.3 Results of Hands-on Activities

Two types of hands on activities were conducted with four players. Each player
was given about 50 blocks of polystyrene. For the first hands-on activity (arrange
the blocks activity), a list of applications names was read out and the players find
the blocks that represent the particular application and arranged it in front of them.
The second hands-on activity (Guess the blocks activity), blocks representing an
application are presented in front of the players and they feel the blocks and
guessed the application names. All the participants were given only 5–10 min to
learn the representation of blocks pattern for each application.
Referring to Table 87.5, three rounds of arranging the blocks activity were
conducted. At each round, five to six applications’ names were read out and the
players searched through the blocks given to them and arranged the correct rep-
resentation of the apps. Comparison between the arranged blocks by the players
and the actual blocks (as shown in Table 87.5) shows that the players are able to
arrange most of the blocks correctly. Referring to Table 87.6, player 1 and player 3
are presented with blocks of description (a) and player 2 and player 4 is presented
with blocks of description (b). Based on the observed result, all players are able to
correctly guess the presented blocks except for player 2 as player 2 is someone
who has become blind just recently and trying to adapt to blindness. Throughout
this activity, the players expressed their opinion by pointing out that having Braille
blocks to represent applications and pop-up blockers truly facilitates them in
learning computer skills.
Most of the previous research on assistive technologies is focused on the
problems faced by the blind in using a screen reader thus not highlighting the
inaccessibility of computer application to them [14, 18]. Some other findings
relating to touch sensation were lacking of empirical study, hence not supporting
the contribution of touch sensation in learning computer application [14, 17, 20].
87 An Exploratory Study on Blind Users’ Mental Model in Computer Accessibility 935

Table 87.5 Arrange the blocks activity

Throughout this activity, the players or better known as the blind participants
seems more confident and excited in using the blocks to represent computer
applications. They have also expressed their opinion by pointing out that having
braille blocks to represent applications and pop-up blockers truly facilitates them
in learning computer skills. Having only screen readers as their assistive tech-
nology in learning computer seems complex for the blinds. Combination of screen
reader and braille blocks will definitely help the blinds in learning the applications
faster and in an easier manner.
936 M. Muniandy and S. Sulaiman

Table 87.6 Guess the blocks activity

Desc/apps and icon names Player 1 (a) Player 2 (b) Player 3 (c) Player 4 (d)
(a) Internet explorer H H H H
(b) Pop-up blocker
(a) File H H H H
(b) Ms. Word
(a) Pop-up blocker H H H H
(b) File
(a) Ms. Word H H H H
(b) Internet explorer
(a) Folder H X H H
(b) Ms. Excel
(a) Ms. Excel H H H H
(b) Pop-up blocker
(a) Pop-up blocker H X H H
(b) Folder
(a) Internet explorer H H H H
(b) Ms. Word
(a) Ms. Word H X H H
(b) File
(a) File H X H H
(b) Internet explorer
H Able to guess correctly X Unable to guess correctly

87.8 Conclusion and Future Work

This exploratory research investigated the issues and problems faced by the blinds
society by conducting a series of interviews, recorded observations and hands on
activities. Challenges faced by the blinds in performing a simple task such as
opening an application in the computer has been highlighted. Views from various
key personnel such as chairman, instructors and blind participants from the blind
association has taken into consideration in concluding the problems of the blinds
which shows that the usual representation of computer accessibility is not suitable
for the blind IT users. And the analysis of the collected result has shown that the
touch sensation of a blind can be used in identification of an object. In this case,
braille blocks were used as the object. Thus, representation of a computer appli-
cation using a braille blocks improves the learning process of a blind person. The
hands on activity has also evident that touch sensation is an effective alert
mechanism for the blinds in navigating through computer applications.
Hence, this study discovers the mental model of a blind user as an important
contribution in improving the accessibility of computer applications. As a future
work, the formulation of a usable model representing the mental model of the blind
87 An Exploratory Study on Blind Users’ Mental Model in Computer Accessibility 937

users will be focused onto. This model will be established to overcome the
identified factors which contribute to the ‘‘inaccessible’’ element of an application
for the blind. It is foreseen that a developer may use the model as a system
development guide and as well as a system assessment guide in future.

References

1. http://en.wikipedia.org/wiki/Computer_accessibility—Accessed 30/09/2013/
2. Ferreira, S.B.L., Nunes, R.R., da Silveira, D.S.: Aligning usability requirements with the
accessibility guideline focusing on the Visually-Impaired. In: 4th International Conference on
Software Development for Enhancing Accessibility and Fighting Info-exclusion (DSAI
2012), Elsevier (2012)
3. Queiroz, M.A.:—Bengala legal—http://www.bengalalegal.com/—Access 2/4/2008
4. World Health Organization, Global data on visual impairments 2010, WHO/NMH/PBD/
12.01
5. Vaas, L.: Web blind spots. eWeek.com. Retrieved October 24, 2012, (2000) from http://
www.eweek.com/c/a/Web-Services-Web-20-and-SOA/Web-Blind-Spots/1/
6. Sulaiman, S., Hasbullah, H.: Siti Nur Syazana Mat Saei, Mental model of blind users to assist
designers in system development. In: ITSim (2010)
7. Potosnak, K.: Human factor: mental model: helping users understand software (1989)
8. Mental Models and Usability, Retrieved on December 20, 2011 from: http://www.lauradove.
info/reports/mental%20models.htm
9. Kurniawan, S.H., Sutcliffe, A.G.: Mental models of blind users in the windows environment.
In: Proceedings ICCHP, pp. 568–574 (2002)
10. Roth, S.P., Schmutz, P., Pauwels, S.L., Bargas-Avila, J.A., Opwis, K.: Mental models for
web objects: where do users expect to find the most frequent objects in online shops, news
portals, and company web pages?’’, Elsevier (2009)
11. Moloney, K.P., Jacko, J.A., Vidakovic, B., Sainfort, F., Leonard, V.K.E. Shi, B.: Leveraging
data complexity: pupillary behavior of older adults with visual impairment during HCI. In:
ACM transaction on computer-human interaction, pp. 376–402 (2006)
12. Takagi, H., Asakawa, C., Fukuda, K., Maeda, J.: Accessibility designer: visualizing usability
for the blinds. In: Proceedings of the ACM SIGACCESS Conference on Computers and
Accessibility, pp. 22–26, Atlanta (2004)
13. Nielsen, J.: Usability 101: Introduction to Usability by Nielsen Norman Group, http://www.
nngroup.com, 4th Jan 2012–Accessed 02/10/2013
14. Abidin, A.H.Z., Xie, H., Wai, Wong, K.W.: Blind users’ mental model of web page using
touch screen augmented with audio feedback. In: International Conference on Computer and
Information Science (ICCIS) (2012)
15. Takagi, H., Saito, S., Fukuda K., Asakawa, C.: Analysis of Navigability of web applications
for improving blind usability. In: ACM Transaction on Computer-Human Interaction, vol. 14
(2007)
16. Alonso, F., Fuertes, J.L., González, Á.L., Martínez, L.: A framework for blind user
interfacing. In: ICCHP pp. 1031–1038 (2006)
17. Nishino, H., Goto, R., Kagawa, T., Yoshida, K., Utsumiya, K., Hirooka, J., Osada, T.,
Nagatomo, N., Aoki, E.: A touch screen interface design with tactile feedback. In:
International Conference on Complex, Intelligent, and Software Intensive Systems, IEEE
(2011)
18. Lumbreras, M., Rossi, G.: A metaphor for the visually impaired: browsing information in a
3D auditory environment. In: CHI’95 (1995)
19. Shinohara, K.: Designing assistive technology for blind users. In: ASSETS (2006)
938 M. Muniandy and S. Sulaiman

20. Petit, G., Dufresne, A., Levesque, V., Hayward, V., Trudeau, N.: Refreshable tactile graphics
applied to schoolbook illustration for students with visual impairment. In: ASSETS’08 (2008)
21. Kumar, M., Talib, S.A., Ramayah, T.: Business Research Methods. Oxford University Press,
Oxford (2012)
Chapter 88
Case-Based Reasoning and Profiling
System for Learning Mathematics
(CBR-PROMATH)

Nur Azlina Mohamed Mokmin and Mona Masood

Abstract This paper discusses the architecture of a case-based reasoning profiling

system for learning mathematics (CBR-PROMATH). The adaptive system has the
ability to suggest suitable learning materials based on previous cases of learner
profiles and individual learning styles. The developed learning materials use a
learning tool which consists of so-called mastery, understanding, interpersonal and
self-expressive styles. Two sets of experiments were carried out to test the sys-
tem’s functionality. The first consisted of 10 sets of learners’ profile cases, stored
previously in the database. The second presented the system with 10 real new
cases. The system compared and calculated similarity values between the new and
stored cases. The learning material that was most similar was presented as a
solution for the new case. The experiment showed that the CBR algorithm was
successfully applied in the development of the CBR-PROMATH.

Keywords Case-based reasoning
Mathematics
Mathematics learning styles

88.1 Introduction

Mathematics is one of the basic pillars of scientific progress. Therefore, mathe-

matical literacy is crucial in building the foundation of a more complex skill [1].
Hodgen and Marks suggest that the ability to benefit from higher education and
then to be able to play a productive role in the workplace depends on the math-
ematical competence of the individual [2]. The lack of mathematical literacy will

N.A.M. Mokmin (&)
M. Masood

Center for Instructional Technology and Multimedia, Universiti Sains Malaysia,
11800 Pulau Pinang, Malaysia
e-mail: [email protected]
M. Masood
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 939

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_88
940 N.A.M. Mokmin and M. Masood

subsequently cause incorrect application or interpretation of mathematics

especially in the science and engineering fields [3, 4].
One reason why mathematic records poor achievement is because of the failure
to differentiate between the learning styles of students [5–8]. A student’s learning
style is the method s/he uses to understand a lesson [9]. Some students prefer
visual diagrams and pictures, whereas others prefer the hands-on approach. Edu-
cators must be able to identify the student’s preferred learning style to ensure the
effectiveness of the teaching and learning process [10].
Hundreds of applications have been created in the field of computer science to
assist students with their lessons. The advancement of technology in education has
led to the development of educational software that applies various learning style
models [11–13]. Alves believes that using case-based reasoning (CBR) can sup-
port educators to design appropriate learning activities to enhance students’
learning experience [14]. Therefore, this paper will first discuss the application of
CBR and related past research as well as the algorithm for developing an intel-
ligent profiling system that can suggest a learning material based on a specific
student profile. The architecture of the CBR profiling system for learning math-
ematics (CBR-PROMATH), students’ profile, and the learning engine are elabo-
rated. In addition, the results of an experiment on the system’s functionality are
also highlighted.

88.1.1 Case-Based Reasoning (CBR)

Artificial intelligence (AI) is the computer’s attempt at imitating human thoughts

and responses. The adoption of AI in education started at the end of the 1980s, in
the form of autonomous agents, intelligent tutoring systems, and educational
theories [14]. Case-based reasoning (CBR) is a problem-solving AI paradigm that
enables utilization of specific knowledge and experience [15, 16]. The field of
CBR is based on cognitive science. It focuses on how humans generate hypotheses
on new situations based on past experiences [17].
According to the CBR theory, an encountered problem (the new case) prompts
the individual to retrieve cases from memory, and reuse the old case, which then
suggests a solution. If the new solution is found to be effective, the knowledge is
then stored in memory for later use. Embedded within each case is a series of
indices which aids memory retrieval. If, however, the attempt to solve the problem
fails, the algorithm will find reasons for the failure and store the result to avoid
repeating similar mistakes [16, 18].
A case indicates a problem situation. Figure 88.1 show the CBR cycle. When a
new case is presented to the system, it will go through a CBR cycle starting from
establishment as a new case and ending when it has been solved by the system.
The problem-solving lifecycle in a CBR system consists of the following four
stages [19];
88 Case-Based Reasoning and Profiling System … 941

Fig. 88.1 The CBR cycle [18]

• Retrieve: retrieving similar cases from the existing knowledge base whose
problem is identified as similar to a given new case
• Reuse: reusing the solutions from the retrieved cases for the new case
• Revise: revising the solutions retrieved in an attempt to solve the new case
• Retain: retaining the new case once it has been solved.

88.1.2 Previous Studies

Many systems are developed for educational purposes; nonetheless, they com-
monly present the same materials without specifically considering individual
student differences. Conversely, many researchers in the education field have
developed systems that cater for differences in the learning process. These systems
have adopted the different learning styles identified.
Yang and Yen propose a framework for multi-agent CBR (MACBR) for per-
sonalized recommendation of e-learning resources [20]. This system gives per-
sonalized recommendation to the learner by taking account of the learner’s
characteristics. It works by noting students’ learning requirements, retrieving
related cases and submitting the case to the similarity calculator agents. The
system then calculates the degree of similarity between the target case and the
related cases and seeks a case similar to the student’s requirements by identifying
similar values.
942 N.A.M. Mokmin and M. Masood

An adaptive web-based educational system named Domus was developed in

[21] that applied CBR and fuzzy logic to adapt e-learning contents and contexts
according to student learning styles and individual needs. The Domus system
works by searching in the database for similar cases based on student learning
style, assuming that students with similar profiles have similar difficulties in the
learning process. Domus functions by providing adapted learning, recommending
the accomplishment of certain learning activities, collecting opinions about a
subject, suggesting Web resources, alerting the student to events, and supporting
agenda management.
In [22] a personalized learning system was constructed by means of a genetic
algorithm and a CBR approach. In this personalized e-learning system, CBR
worked by providing a summative assessment for each individual learner after
several units of instruction. The research considered the curriculum’s difficulty
level and the continuity of successive curricula in mastery learning.
Most adaptive applications apply a CBR algorithm developed for e-learning
content but very few applications are developed exclusively for mathematics
learning. An adaptive mathematics learning application must include mathematics
learning styles and mathematics achievements in the learning engine. Therefore,
the current study uses the mathematics learning style recommended by [23] in the
design and development process.

88.1.3 Similarity and Weighting in CBR

The learning engine uses outcomes to analyze similarities of a given problem in

order to re-weight case features. A similarity measure is a critical component in
any CBR system [24, 25]. Similarity measures were described in [26] as a means
to compare two cases. Local similarity measures are similar on the feature level
and global similarity measures are similar on the case or object level.

88.2 Architecture of Case-Based Reasoning

Profiling System for Learning Mathematics
(CBR-PROMATH)

CBR-PROMATH is a web-based system programmed in Hypertext Preprocessor

(PHP), a server-side scripting language designed not only for web development but
also as a general-purpose programming language. The related database manage-
ment system applied in this system is MySQL, which is based on Structured Query
Language (SQL). MySQL and PHP are both free for download and use because
they are open-sourced. The PHP and MySQL combination has become a popular
88 Case-Based Reasoning and Profiling System … 943

Fig. 88.2 The CBR-PROMATH architecture

choice for database-driven websites. In contrast, the learning materials are

developed from multimedia software.
The main characteristic of CBR-PROMATH is that it can give suggestions on
the appropriate learning materials based on learning styles and student profiles.
The system has three basic components; the student profile database, the learning
engine and the learning material database (Fig. 88.2). These three components
interact to adapt the different cases (problems) submitted by the user to the system.
These submitted data are transformed by the learning engine into variables that are
processed as information before the best learning material is suggested.

88.2.1 Student Profile

The student’s profile consists of information such as their identity card, the
Malaysian Certificate of Education or ‘‘Sijil Pelajaran Malaysis’’ (SPM) mathe-
matics result, learning styles and test results. The student’s profile is where
information on the learner is obtained and processed into variables by the system.
This information is then used by the learning engine to compute similarity values.
Figure 88.3 shows the graphical user interface (GUI) for the student’s profile.
Figure 88.4 shows the set of data that is stored in the system in the form of cases
for consultation by the learning engine.
944 N.A.M. Mokmin and M. Masood

Fig. 88.3 GUI of the student’s profile

Fig. 88.4 Set of stored cases

88.2.2 Learning Engine

The learning engine processes the information keyed in by the user, and converts
the information into variables for calculation by the system. When a new case is
presented to the system, the CBR algorithm calculates the local similarity value
and the global similarity value between each case for comparison with the new
case. The case with the highest similarity value is selected. The learning materials
suggested are used as a reference for the new case. Personalized learning materials
can then be recommended to a student with a similar learning profile.
88 Case-Based Reasoning and Profiling System … 945

88.2.3 Learning Materials

In [27] it was found that students who learn with adaptive learning materials have
significantly better learning achievement than those who learn with non-adaptive
materials. Providing students with learning materials and personalized recom-
mendations in terms of their preferred learning methods can make learning easier
for them [11]. Researchers involved in the development of personalized and
adaptive learning materials include learning styles as important factors in their
work [25, 28, 29]. It is important to diagnose the student’s learning style because
some students learn more effectively when taught according to their preferred
method. Information about different learning styles can help the system become
more sensitive towards students who use the system. As stated by [30], a student’s
learning style represents the type of learner they become.
In the development of the CBR-PROMATH, the mathematics learning style
that was suggested by [23] was used in the design of the learning materials. The
learning materials have four distinct styles of instruction:
• Mastery learning style (MLS) that emphasizes skill acquisition and retention of
critical mathematical terms.
• Understanding learning style (ULS) that builds student’s capacities to find
patterns and explain mathematical concepts.
• Self-expressive learning style (SLS) that capitalizes on students’ powers of
imagination and creativity
• Interpersonal learning Style (ILS) that invites students to find personal meaning
in mathematics by working together as part of a community of problem-
solvers.
The materials suggested by the system are from the learning engine. At the end of
every learning material presented, marks or grades for every assessment are stored in
the database. A low mark indicates the unsuitability of the learning materials sug-
gested, in which case the student is presented with the second learning material
which has the second-highest similarity value. If the result is a pass, the learning
material suggested will be used as a reference for the next case. Figures 88.5 and
88.6 show an example of the learning material that has been developed.

88.2.4 Experiments

In order to test the system’s functionality, a series of experiments was carried out.
In particular, it was set up to see whether the system is capable of carrying out
effective calculations when given a new case. Table 88.1 shows the set of new
cases submitted, learning materials suggested and the case referred to previously.
The experiments and calculation values given by the system suggest that CBR-
PROMATH successfully referred the cases with the highest similarity value to the
new case.
946 N.A.M. Mokmin and M. Masood

Fig. 88.5 Example of learning material 1

Fig. 88.6 Example of learning material 2

The suggested learning materials in Table 88.1 are the result of the similarity
value of each stored case in the database. In the table, Case Id Aida332 has the
highest similarity in terms of value to the New Case Id Nora12; therefore the
learning material for Aida332 is used as a learning material suggestion for New Id
Nora12.
88 Case-Based Reasoning and Profiling System … 947

Table 88.1 New cases, learning material suggestions and reference case
New case Learning Mathematics Test Reference to case Id Learning material
Id style result result in database suggestion
Nora12 ULS A C Aida332 Understanding
Akila225 ILS D D Ahmad Interpersonal
Danny67 SLS C B Asiyah22 Self-Expressive
Johnny89 MLS B A Ailing77 Mastery
Sitihajar34 ILS C D June88 Interpersonal
Kamal44 ILS B C Aida332 Understanding
Badri66 ILS D B Ahmad Interpersonal
Haikal33 SLS C B Asiyah22 Self-Expressive
Fiza78 ULS C A Ailing77 Mastery
Alia99 ILS C B Asiyah22 Self-Expressive

88.2.5 Conclusion and Future Work

This work shows that the design and development of CBR-PROMATH can sug-
gest suitable learning materials for the user based on the cases submitted to the
system. In case-based reasoning, a reasoning engine remembers a previous situ-
ation similar to the current one and uses that to solve the new problem [15]. The
learning materials that are developed are of value to math educators and
researchers. A useful suggestion for future research is the assessment of all
learning materials suggested by the system in terms of their effectiveness in the
teaching and learning process.

References

1. Kashefi, H. et al.: Creative problem solving in engineering mathematics through computer-

based tools. In: 2nd International Seminar on Quality and Affordable Education, pp. 207–211
(2013)
2. Hodgen, J., Marks, R.: The employment equation: why our young people need more maths for
today’s jobs. The Sutton Trust, London (2013) [Online]. Available: http://www.suttontrust.
com/news/publications/the-employment-equation-why-our-young-people-need-more-maths/
3. Haider, H., et al.: How we use what we learn in math: an integrative account of the
development of commutatively frontline learning research 1. Frontline Learn. Res. 2(1), 1–21
(2014)
4. Hong, K.S., et al.: Status of mathematics teaching and learning in Malaysia. Int. J. Math.
Educ. Sci. Technol. 40(1), 59–72 (2009)
5. Aral, A., Cataltepe, Z.: Learning styles for K-12 mathematics e-Learning. In: CSEDU 2012 -
4th International Conference on Computer Supported Education, pp. 317–322 (2012)
6. Middleton, K. et al.: Examining the Relationship between Learning Style Preferences and
Attitudes Toward Mathematics Among Students in Higher Education. Institute for Learning
Styles Journal, Howard University, vol. 1 (2013)
948 N.A.M. Mokmin and M. Masood

7. Havola, L.: Assessment and learning styles in engineering mathematics education. Licentiate
thesis, Aalto University, Espoo, Finland (2012)
8. Murat, G.: The effect of students’ learning styles on their academic success. Creative Educ.
4(10), 627–632 (2013)
9. Mohamad, B., Hashim, I.: Gaya Pengajaran Dan Pembelajaran. PTS Professional Publishing
Sdn. Bhd, Kuala Lumpur (2010)
10. Adnan, M., et al.: Learning style and mathematics achievement among high performance
school students. World Appl. Sci. J. 28(3), 392–399 (2013)
11. Kinshuk, S.G.: Providing adaptive courses in learning management systems with respect to
learning styles. In: Proceedings of the World Conference on E-Learning in Corporate,
Government, Healthcare, and Higher Education (e-Learn), pp. 2576–2583 (2007)
12. Mahnane, L., et al.: A model of adaptive e-learning hypermedia system based on thinking and
learning styles. Int. J. Multimedia Ubiquitous Eng. 8(3), 339–350 (2013)
13. Sarrafzadeh, A., et al.: How do you know that i don’t understand? a look at the future of
intelligent tutoring systems. Comput. Hum. Behav. 24(4), 1342–1363 (2008)
14. Alves, P.: Advances in Artificial Intelligence to Model Student-Centred VLEs, Advances in
Learning Processes, INTECH, pp. 113–124 (2010)
15. Kolodner, J.L.: An introduction to case-based reasoning. Artif. Intell. Rev. 6, 3–34 (1992)
16. Andrew, A.T., Keene, C.W.: Applying case-based reasoning theory to support problem-based
learning. J. Appl. Instr. Des. 3(2), 31–40 (2013)
17. Shiu, S.C., Pal, S.K.: Case-based reasoning: concepts, features and soft computing. Appl.
Intell. 21(3), 233–238 (2004)
18. Aamodt, A., Plaza, E.: Case-based reasoning: foundational issues, methodological variations,
and system approaches. AI Commun. 7(1), 39–59 (1994)
19. Swanson, R., Gordon, A.S.: Say anything: using textual case-based reasoning to enable open-
domain interactive storytelling. ACM Trans. Interact. Intell. Syst. (TiiS) 2(3), 16 (2012)
20. Yang, L., Yan, Z.: Personalized recommendation for learning resources based on case
reasoning agents. Electr. Control Eng. (ICECE) 3, 6689–6692 (2011)
21. Alves P. et al.: Case-based reasoning approach to adaptive web-based educational systems.
In: ICALT ‘08: Proceedings of the 2008 Eighth IEEE Int. Conference on Advanced Learning
Technologies, pp. 260–261 (2008)
22. Huang, M., et al.: Constructing a personalized e-learning system based on genetic algorithm
and case-based reasoning approach. Expert Syst. Appl. 33(3), 551–564 (2007)
23. Silver, H.F., et al.: Math Tools, Grade 3-12; 60 + Ways to Build Mathematical Practices,
Differentiate Instruction, and Increase Student Engagement, 2nd edn. Corwin, Thousand
Oaks (2012)
24. Long et al. J.: Adaptive similarity metrics in case-based reasoning. In:Proceedings of
Intelligent Systems and Control, Honolulu, Hawaii, USA (2004)
25. Soh et al. W.M.: Implicit detection of learning styles—the smalt way. In: Proceedings of the
9th International CDIO Conference, Massachusetts Institute of Technology and Harvard
University School of Engineering and Applied Sciences, Cambridge, USA (2013)
26. Bergmann, R., et al.: Representation in case-based reasoning. Knowl. Eng. Rev. 20(3),
209–213 (2005)
27. Hwang, G.J., et al.: A learning style perspective to investigate the necessity of developing
adaptive learning systems. Educ. Technol. Soc. 16(2), 188–197 (2013)
28. Anthony, P., et al.: Learning how to program in c using adaptive hypermedia system. Int.
J. Inf. Educ. Technol. 3(2), 151–155 (2013)
29. Sudhana, K.M., et al.: An architectural model for context aware adaptive delivery of learning
material. Int. J. Ad. Comput. Sci. Appl. 4(10), 80–87 (2013)
30. Mohamad, M.M., et al.: Disparity of learning styles and cognitive abilities in vocational
education. Int. J. Soc. Hum. Sci. Eng. 8(1), 6–9 (2014)
Chapter 89
What Is the Influence of Users’
Characteristics on Their Ability to Detect
Phishing Emails?

Ibrahim Alseadoon, M.F.I. Othman and Taizan Chan

Abstract Phishing emails cause significant losses for organisations and victims.
To fight back against phishing emails, victims’ detection behaviours must be
identified and improved. Then, the impact of victims’ characteristics on their
detection behaviours must be measured. Three methods, namely experiments,
surveys and semi-structured interviews, were applied in our research to gain a
richer understanding of victims’ behaviours concerning phishing emails. Several
user characteristics were measured using the deception detection model. The
results suggest that users’ characteristics either increase or decrease their suspi-
cions. Characteristics such as user extraversion, trust and submissiveness represent
variables that prevent victims from suspecting phishing emails. In contrast, email
experience increases victims’ suspicion of phishing emails. Furthermore, victims’
personality characteristics and a variable called the susceptibility variable play
important roles in increasing the tendency of victims to execute the actions
requested in phishing emails.

89.1 Introduction

Phishing emails are designed to imitate legitimate entities, causing users to trust in
their validity. Then, users are asked to perform an action included in the phishing
email, which is most likely connected to a fake website, in order to obtain victims’
sensitive information. A phishing attack has two aspects: the first is the way in
which the attacker reaches his or her victims; the second depends on the victims
themselves and whether they choose to respond to these emails [1–3]. Securing
users against phishing emails begins with the use of technology to detect such
emails before they reach users [4–6]. However, the perpetrators of phishing emails

I. Alseadoon (&)
M.F.I. Othman
T. Chan

Queensland University of Technology, Brisbane, QLD 4000, Australia
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 949

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_89
950 I. Alseadoon et al.

are always finding new ways to enable these emails to reach users [7, 8]. Undetected
phishing emails make users the last defence line of defence against these emails.
The number of phishing websites detected by the Anti-Phishing Working Group
in April of 2012 was 63,253, which is believed to be the highest number ever
recorded [9]. The financial losses incurred due to phishing attacks are consider-
able. These losses can vary from $61 million to approximately $3 billion per year
[10, 11]. Increased investigation and prevention of phishing attacks must occur to
minimise these losses.
The purpose of our research is to identify users’ detection behaviours,1 as well
as user characteristics that impact their detection behaviours. Several user char-
acteristics have been identified and measured using the deception detection model
proposed by Grazioli in 2004. If victims are found to have individual character-
istics that predisposed them to falling victim to deception, then knowing this will
aid in identifying other individuals who are at risk.
The paper is organised as follows: the next section describes relevant research
on phishing emails. Then, the theoretical framework section introduces the
deception detection model used and user characteristics measured using the model.
The methodology section details the steps taken to conduct our research. The
findings are described in the results section. Finally, the discussion and conclusion
sections conclude this paper.

89.2 Related Work

Education programs have been provided to increase users’ knowledge about

phishing emails. Such programs have increased the rate at which users become
detectors by 40 % [12]. Educating users on the design features of phishing emails
has helped reduce the rate at which users fall victim to phishing emails [13–16].
The main drawback of phishing email education is that some users who received
education programs still became victims [7, 8]. This suggested that we should
attempt to look for weaknesses that predisposed users to becoming victims.
Security tools have been provided to warn users about phishing attempts. In such
studies, the final decision concerning the validity of an email is made by the user,
but this is done with the help of security tools [17, 18]. The drawback of these
approaches is that they are limited to users’ discretion; one study revealed that users
often ignore these kinds of warnings and thus still fall victim to the deceit [19, 20].
To help prevent users from becoming victims, we must understand users’ detection
behaviours and which user characteristics influence their detection behaviours [21].
In other words, the main focus in phishing email studies has been improving users’
knowledge about phishing emails and developing new security tools. Unfortunately,
user characteristics and detection behaviours have not been well-investigated [21].

1
Users’ detection behaviours are the behaviours that users perform to detect phishing emails,
which include susceptibility, confirmation and response.
89 What Is the Influence of Users’ Characteristics … 951

89.3 Theoretical Framework

The theory of deception explains the process by which users detect deception [22,
23]. According to this theory, detection is a cognitive process that involves
examining various cues (e.g., words, tone and body language). Detectors look for
cues that lead them to conclude that the received message is a deceptive message.
Grazioli [24] applied this theory in a computer-based environment and found that
the detection process goes through four phases (see Fig. 89.1). Interestingly,
Grazioli found that detectors and victims both go through the four phases of the
deception detection model. However, only detectors were able to detect deception,
whereas victims failed to detect it [24]. In this research project, we aim to discover
the reasons victims failed to successfully detect deceptive messages.
We applied Grazioli’s model in order to understand users’ detection behaviours
when faced with phishing emails. From examining the Grazioli model, we were
able to determine users’ detection behaviours and divide them into three phases:
susceptibility, confirmation and response (see Fig. 89.1). Additionally, our research
is interested in finding the impact of users’ characteristics on their detection
behaviours (see Fig. 89.2), as well as the impact of the confirmation channel.

89.3.1 Variables Related to Users’ Detection Behaviours

These variables can be divided into three main variables, as explained in the
previous section: susceptibility, confirmation and response.

89.3.1.1 Response

Response is the last behavior, and it indicates that users have responded to the
phishing email. Ultimately, users were classified as either detectors or victims.
Victims were those who chose to perform the action in the phishing emails.
Detectors were those users who saw the phishing email and chose to ignore it.

89.3.1.2 Confirmation

Users who become doubtful about a phishing email generate hypotheses and
evaluate them. Hypothesis generation is a mental process that is not measured in
this research. On the other hand, hypothesis evaluation is a process that can be
captured by examining the ways in which users choose to act. For example, when
questioning the legitimacy of an email, some users contact organisations by phone,
while others email their friends. Confirmation was measured by asking users to
self-report the type of confirmation channels they used. Confirmation channels are
952 I. Alseadoon et al.

Fig. 89.1 Users’ detection, cognitive and behaviour processes

Fig. 89.2 Research model

classified based on their richness (face-to-face, telephone, email or self-investi-

gation). Face-to-face is considered to be a highly rich medium [25].
H1. Use of a highly rich confirmation channel decreases user response rates to
phishing emails.

89.3.1.3 Susceptibility

Susceptibility is the first phase in user detection behaviour. Users begin their
detection behaviour by suspecting the phishing email. Based on the level of sus-
ceptibility, users will decide upon their intended behaviour (see Fig. 89.1). To
measure users’ susceptibility, five phishing emails were presented to participants in
the form of a picture. Participants were asked to play the role of someone called John
and were told that John had received these five emails. Participants were asked to
rate the likelihood of responding to each of the five emails, with 1 being ‘definitely
89 What Is the Influence of Users’ Characteristics … 953

would ignore the email’ and 7 being ‘definitely would respond to the email’. This
variable was measured in the first survey, meaning that the participants had not been
informed about the phishing email experiment or that these five emails were
phishing emails. The aim was to capture users’ normal responses to emails and to
determine whether they were able to determine these emails to be phishing emails.
H2. A high level of susceptibility to phishing emails increases user response
rates to phishing emails.

89.3.2 Related Users’ Characteristics

Our research examines the impact of users’ characteristics on their detection

behaviours. Figure 89.2 illustrates the proposed model used in this research. The
diagram below presents related user characteristics.

89.3.2.1 Big Five Personality Dimensions

The Big Five personality dimensions divide a user’s personality into five main
dimensions: (1) extraversion, which describes a person who is more interactive
with others (2) agreeableness, which describes a person who is more kind and
warm to others (3) consciousness, which describes a person who is more deter-
mined to finish tasks (4) emotional stability, which describes a person who is more
likely to be calm and (5) openness, which describes a person who is more open to
new experiences [26]. These dimensions have been suggested to influence users’
susceptibility to phishing emails [27]. Our research investigates the impact of
extraversion (H3a, H4a), agreeableness (H3b, H4b), consciousness (H3c, H4c),
emotional stability (H3d, H4d) and openness (H3e, H4e) on users’ detection
behaviours. The measures used were developed by Gosling et al. [28], whose work
contained ten items.
H3. Certain big five personality dimensions impact user response to phishing
emails.
H4. Certain big five personality dimensions impact user susceptibility to
phishing emails.

89.3.2.2 Trust

Trust measures users’ tendency to trust others. One drawback of relying on victims
to make decisions regarding phishing emails is their trust that these emails are
legitimate. Our research aims to discover the impact of trust on users’ suscepti-
bility to phishing emails. Trust was measured using three items developed by
McKnight et al. [29].
H5. Trust increases users’ susceptibility to phishing emails.
954 I. Alseadoon et al.

89.3.2.3 Submissiveness

Wright et al. [30] performed a study on phishing emails, and based on interviews
they conducted, they suggested that submissiveness has an effect on users
becoming victims of phishing emails. The nature of phishing emails is such that
they cause users to believe that they are in a situation that demands their com-
pliance with the requests of the phishing emails. Wright et al. [30] indicate that
users with a high level of submissiveness are different than others. Submissiveness
is measured using 16 items developed by Allan and Gillbert [31].
H6. Submissiveness increases user susceptibility to phishing emails.

89.3.2.4 Email Experience and Email Richness

The theory of media richness supports the idea that users are more able to detect
deception carried out via media with a high degree of richness as compared to
media with a low degree of richness [25]. Richness is measured by the ability to
reproduce the information carried by the medium. Because email is considered to
have a low degree of richness, deception detection in emails tends to be difficult
for users. However, the theory of channel expansion suggests that the more
experience a user has with a medium, the more he or she will perceive the medium
to be rich [32]. Therefore, our research will measure the impact of email experi-
ence and richness on user ability to detect deception via email. The measures
include six items for email experience and four items for email richness developed
by Carlson and Zmud [32].
H7. Email experience decreases user susceptibility to phishing emails.
H8. Email richness decreases user susceptibility to phishing emails.

89.4 Methodology

A mixed methods approach was applied by using a combination of qualitative and

quantitative methods. The following subsections provide more details.

89.4.1 Subjects

The targeted participants were university undergraduate students. Aiming for this
age group was crucial because it has been reported in the literature that students in
this age group are more susceptible to phishing emails than those of older age
groups [33, 34]. A total of 187 students out of 430 contacted provided complete
answers to all the phases required by our research.
89 What Is the Influence of Users’ Characteristics … 955

89.4.2 The Survey

The survey was designed to collect information regarding participants’ charac-

teristics and detection behaviors, which required us to divide the survey into two
parts. The first survey was delivered before sending the phishing email to collect
information regarding the participants’ characteristics. The second survey was
delivered after sending the phishing email to collect information regarding par-
ticipants’ detection behaviours.

89.4.2.1 The First Survey

The first survey, which did not inform participants about the experiment, was
designed to collect information regarding participants’ characteristics, such as the
Big Five personality dimensions, trust, submissiveness, email experience and
email richness. In this survey, participants were informed that the survey was
designed to capture the impact of users’ characteristics on their normal day-to-day
email behaviours.

89.4.2.2 The Second Survey

This survey was designed to collect information regarding users’ detection

behaviors and recorded the type of confirmation channel used. The survey was sent
to participants after a sufficient amount of time had passed (i.e., around two weeks)
since their receipt of the phishing email. In this survey, the real goal of the
experiment was revealed to the participants (i.e., a phishing email study).

89.4.3 The Phishing Email

The aim of sending the phishing email was to classify participants into two groups:
detectors and victims. The phishing email was sent on behalf of the blog main-
tenance team. The body of the phishing email was designed as follows: the
phishing email informed participants about a problem that occurred in a blog,
which resulted in damaging the blog database that held students’ account infor-
mation. The phishing email emphasised that the students who had received this
email were those who had been affected by this recent problem and that they
needed to act quickly to resolve it. After students were given an explanation as to
why they had received the email, they were prompted with a solution that for this
purpose of this research, imitating a real phishing email design. Two solutions
were provided; students were either asked to reply to an email address with the
requested information or to click on a link embedded in the phishing email.
956 I. Alseadoon et al.

Both the reply email address and the link were hosted outside the university
domain (on a malicious domain).

89.4.4 Interviews

After collecting information about participants’ responses to the phishing email

and the two surveys, participants were sent an invitation to be interviewed about
their experiences with the phishing email. In these semi-structured interviews,
participants were asked questions regarding their responses when they received the
phishing email. The interviews helped to shed light on certain aspect of users’
detection behaviours that could not be identified in the previous surveys.

89.5 Results

Participants were randomly divided into two main groups: those who received
reply phishing emails and those receiving click phishing emails. The percentage of
victims was 26 % of the whole population. The results suggest that phishing
emails that ask about private information directly are less harmful than those that
ask about private information indirectly. Two-thirds (69.5 %) of the 187 partici-
pants were under 26 years of age, and 77.5 % (N = 145) were male. Some 57.2 %
(N = 107) spoke English as their first language, and 58.3 % (N = 109) identified
their nationality as Australian.
We applied regression analysis to measure the relationships between variables.
There were two dependent variables: susceptibility and response (see Fig. 89.2).
Linear regression was applied to measure the dependent variable, susceptibility,
and logistic regression was applied to measure the dependent variable, response.
The results show that R-square is equal to 0.643 for the linear regression (sus-
ceptibility) and that R-square is equal to 0.492 for the logistic regression
(response).

89.5.1 Susceptibility

The results of multiple regressions show that there are two positive, significant
relationships: one between submissiveness and susceptibility and another between
trust and susceptibility. There is also a negative, significant relationship between
perceived email richness and susceptibility. The results show that submissiveness,
trust and email richness explain 64 % of the variance in susceptibility (see
Table 89.1).
89 What Is the Influence of Users’ Characteristics … 957

Table 89.1 Summary of the hypotheses tested regarding susceptibility

Hypotheses Variables Significance Conclusion
Hypothesis 5 + Trust b = 0.475 P = 0.000 Supported
Hypothesis 6 + Submissiveness b = 0.190 P = 0.012 Supported
Hypothesis 8 - Email Richness b = -0.199 P = 0.007 Supported

Extraversion refers to a person who is more interactive. Due to the nature of

extraverted users, they are more susceptible to phishing emails. Those users with
high levels of trust are more likely to trust others. Phishing emails ask their victims
to trust them (i.e., phishing emails impersonate trustworthy entities). Submis-
siveness measures users’ obedience to others. Phishing emails directly ask their
victims to perform an action. High levels of submissiveness will increase the
chances of complying with the direct order received via phishing emails. Email
experience’s impact is opposite that of the previous three variables (extraversion,
trust and submissiveness). High email experience decreases user’s susceptibility to
phishing emails. Email experience may improve the baseline for distinguishing
between legitimate emails and illegitimate ones. Observing inconsistent cues is an
essential aspect of making users suspect phishing emails, which can be a result of
high levels of email experience. Finally, email richness has not been indicated to
have any significant impact on users’ susceptibility.

89.5.2 Response

The results of the linear regressions show that there are four positive, significant
relationships between the Big Five personality dimensions, susceptibility and
confirmation channel. There is also a negative, significant relationship between
confirmation channel and response. The results show that the identified variables
explain 49 % of the variance in responses (see Table 89.2).
As indicated in Table 89.2, susceptibility, openness, extraversion, agreeable-
ness and confirmation channel predicted the response as an outcome at the
p \ 0.05 level. Those with higher susceptibility and openness scores were two and
half times more likely than others (Exp (B)) to respond to the phishing emails.
Those with higher extraversion scores were two times more likely than others (Exp
(B)) to respond to the phishing emails. Those with higher agreeableness scores
were four times more likely than others (Exp (B)) to respond to the phishing
emails. Those with higher confirmation channel scores were one time more likely
than others (Exp (B)) to detect phishing emails.
Phishing emails ask their victims to interact with them and comply with their
requests, which may be the reason that extraversion and agreeableness signifi-
cantly impacting users’ responses to phishing emails. Openness describes users
who are more open to new experience and take more risks. Phishing emails are a
risk for users who comply with their requests. It can be suggested that openness
958 I. Alseadoon et al.

Table 89.2 Summary of hypotheses tested with responses

Hypothesis Variables Significance Conclusion
Hypothesis 3a + Extraversion B = 0.540 P = 0.002 Supported
Hypothesis 3b + Agreeableness B = 1.289 P = 0.000 Supported
Hypothesis 3e + Openness B = 0.859 P = 0.001 Supported
Hypothesis 2 + Susceptibility B = 0.914 P = 0.000 Supported
Hypothesis 1 Confirmation B = -0.432 P = 0.031 Supported

increases users’ tendency to take the risks included in phishing emails. High levels
of susceptibility will decrease users’ suspicion levels, which will increase users’
vulnerability. Finally, the type of confirmation channel did not have any significant
effect on users being victims.

89.6 Discussion

From the results above, it can be seen that users’ characteristics play an important
role in influencing users’ detection behaviours. The process of detection includes
various phases, and each phase is influenced by certain user characteristics. The
following paragraphs describe the impact of user characteristics on their detection
behaviours.
The first phase in the deception detection model is activation, which is mea-
sured by susceptibility. There are two variables responsible for increasing users’
susceptibility to phishing emails. These variables prevent users from suspecting
phishing emails. A lack of suspicion means that these victims are not able to tell
the difference between phishing emails and legitimate ones. In contrast, email
richness reduced users’ susceptibility to phishing emails and caused them to
engage in detection behaviour. Engaging in detection behaviour does not guar-
antee that users will be detectors; users must develop an explanation as to why
there are differences between what they observe and what they expected to
observe. This is called inconsistency, which is essential for detection [35].
Hypothesis generation and hypothesis evaluation are the next phases in the
deception detection model. Generating a weak hypothesis and performing weak
hypothesis evaluation inhibits users from detect phishing emails. In this research
project, one victim who responded to the phishing email reported that he suspected
the phishing email to be illegitimate. However, he became a victim because he
generated the following hypothesis: ‘‘If other students received this email, then it
is genuine.’’ After contacting one of his friends, who confirmed having received
the phishing email, he decided to respond to the phishing email simply because
other students had received the email. It can be seen that this victim had a com-
bination of weak hypothesis generation and weak hypothesis evaluation. Phishing
emails can easily fulfill this assumption because they target an enormous number
of users. In contrast, detectors have shown a high level of strong hypothesis
89 What Is the Influence of Users’ Characteristics … 959

generation. One detector forwarded the phishing email to the owner of the blog,
asking him whether or not the email was genuine. Confirming the legitimacy of a
suspected email with the legitimate entity that the phishing emails impersonate
will help in detection.
The final phase in the deception detection model is global assessment. In this
phase, users come to a decision regarding whether to respond to the phishing
email. In this research, four variables were found to increase victims’ tendency to
perform the action requested in the phishing emails. Most of these variables are
related to users’ personalities, which leads to the assumption that these victims are
risk takers. One participant acknowledged his knowledge that sending passwords
over emails was a low-security behaviour, but he still chose to send them. In the
interviews, our research found that some victims suspected the phishing email and
commenced the process of detection until they reached the final phase. However,
these victims still chose to respond to the phishing email. It can be seen that
personality characteristics influenced users’ decisions to perform risk-taking
behaviours. This means that these victims were aware of the risk involved in their
low-security behaviour but, due to their personality, chose to take the risk.

89.7 Limitations

Our research examines users who have a relationship with the impersonated entity.
It is assumed that these kinds of phishing email attacks are difficult to detect.
Emails received from impersonated entities that users have no relationship with
are easily rejected.
Our research tested two types of phishing emails. However, in the real world,
phishing emails come in various forms. Therefore, more experiments and time are
needed.
Because the number of participants in our research was limited, the results
cannot be generalized to the wider population of Internet users in Australia. To
generate broader results, similar experiments involving an enormous number of
participants must be conducted.

89.8 Conclusion

Our research aimed to identify user characteristics that cause users to become
victims of phishing emails, as well as their detection behaviour. Several variables
have been measured using the deception detection model to discover the impact of
user characteristics on the process of detecting phishing emails. The findings of the
research revealed that trust and submissiveness increase users’ susceptibility to
phishing emails (activation), which is the first phase in the deception detection
model. The personality dimensions of extraversion, agreeableness and openness, in
960 I. Alseadoon et al.

addition to susceptibility, are responsible for increasing users’ tendency to proceed

with the requested actions included in the phishing email, which is the last phase of
the deception detection model. Our research suggests that users’ ability to detect
phishing emails is complicated and should be investigated. Detectors are able to
conclude that they should ignore phishing emails on their own. In contrast, victims
need to be assisted with their decision making to increase their resistance to
phishing emails.
The overall suggestion produced by our research is that user characteristics play
an important role in improving users’ ability to detect phishing emails. Various
user characteristics impact different phases of the process of detecting phishing
emails. To increase user security against phishing emails, the user characteristics
that are important for each phase must be improved. Assisting users in detecting
phishing emails can be done by (1) educating users about phishing emails and their
techniques to decrease their susceptibility and (2) performing re-assistance to
reduce their risk-taking behaviours and thus decrease their tendency to respond to
phishing emails.

Acknowledgments Ibrahim Alseadoon would like to acknowledge the Ministry of Higher

Education, Saudi Arabia and the University of Ha’il for sponsoring his PhD studies at Queensland
University of Technology, Brisbane, Australia. Mohd Fairuz Iskandar Othman would like to
acknowledge the Ministry of Higher Education, Malaysia and Universiti Teknikal Malaysia
Melaka for sponsoring his PhD studies at Queensland University of Technology, Brisbane,
Australia.

References

1. Zhang, W., Luo, X., Burd, S.D., Seazzu, A.F.: How Could I Fall for That? Exploring Phishing
Victimization with the Heuristic-Systematic Model. In System Science (HICSS), 45th Hawaii
International Conference on, 2012, pp. 2374–2380 (2012)
2. Xun, D., Clark, J.A., Jacob, J.: Modelling User-phishing Interaction. In: Proceedings of
Human System Interactions May 25-27, 2008, Kraków, Poland (2008)
3. Wang, J., et al.: Phishing Susceptibility: An Investigation Into the Processing of a Targeted
Spear Phishing Email. Profess. Commun. IEEE Trans. 99, 1 (2012)
4. Fette, I., Sadeh, N., Tomasic, A.: Learning to detect phishing emails. In: Proceedings of the
16th International Conference on World Wide Web. pp. 649–656. ACM: Banff, Alberta,
Canada (2007)
5. Cook, D., Gurbani, V., Daniluk, M.: Phishwish A Stateless Phishing Filter Using Minimal
Rules. In: Financial Cryptography and Data Security. pp. 182–186. Springer, Heidelberg
(2009)
6. Bergholz, A. et al.: Detecting known and new salting tricks in unwanted emails. In: CEAS
2008: Proceedings of the Fifth Conference on Email and Anti-Spam. Citeseer (2008)
7. Wang, J., et al.: An exploration of the design features of phishing attacks. Inf. Assur. Secur.
Priv. Ser. 4, 29 (2009)
8. Sharma, K.: An anatomy of phishing messages as deceiving persuasion: a categorical content
and semantic network study. EDPACS 42(6), 1–19 (2010)
9. Anti Phishing Working Group Phishing Activity Trends Report, 2nd Quarter (2012)
89 What Is the Influence of Users’ Characteristics … 961

10. Pettey, C.: Gartner Says Number of Phishing Attacks on U.S. Consumers Increased 40
Percent in 2008 (2006) [cited 2009 2/9]; Available from: http://www.gartner.com/it/page.jsp?
id=936913
11. Herley, C., Florencio, D.: A profitless endeavor: phishing as tragedy of the commons. In:
Proceedings of the 2008 Workshop on New Security Paradigms. ACM: Lake Tahoe,
California, USA. pp. 59–70 (2008)
12. Sheng, S. et al.: Anti-Phishing phil: the design and evaluation of a game that teaches people
not to fall for phish. In: Proceedings of the 3rd Symposium on Usable privacy and Security.
ACM: Pittsburgh, Pennsylvania (2007)
13. Arachchilage, N.A.G., Love S., Scott M.: Designing a Mobile Game to Teach Conceptual
Knowledge of Avoiding ‘‘Phishing Attacks’’. Int. J. e-Learn. Secur. 2, 127–132 (2012)
14. Kumaraguru, P. et al.: School of phish: a real-world evaluation of anti-phishing training. In:
Proceedings of the 5th Symposium on Usable Privacy and Security. ACM: Mountain View,
California, pp. 1–12 (2009)
15. Bekkering, E., Hutchison, D., Werner, L.: A follow-up study of detecting phishing emails. In:
Proceedings of the Conference on Information Systems Applied Research. Washington DC
(2009)
16. Sven, D. et al.: Phishing IQ tests measure fear, not ability. In: Financial Cryptography and
Data Security. pp. 362–366. Springer, Heidelberg (2007)
17. Wu, M., Miller, R.C., Little, G.: Web wallet: preventing phishing attacks by revealing user
intentions. In: SOUPS ‘06: Proceedings of the Second Symposium on Usable Privacy and
Security. ACM, New York (2006)
18. Kirda, E., Kruegel, C.: Protecting users against phishing attacks. Comput. J. 49(5), 554
(2006)
19. Kim, Y.-G., et al.: Method for evaluating the security risk of a website against phishing
attacks. Intell. Secur. Inf. 5075, 21–31 (2008)
20. Wu, M., Miller, R.C., Garfinkel, S.L.: Do security toolbars actually prevent phishing attacks?
In: Proceedings of the SIGCHI Conference on Human Factors in Computing Systems.
pp. 601–610. ACM, Montreal, Quebec, Canada (2006)
21. Purkait, S.: Phishing counter measures and their effectiveness—literature review. Inf. Manag.
Comput. Secur. 20(5), 382–420 (2012)
22. Johnson, P.E., Grazioli, K.S.: Fraud detection: intentionality and deception in cognition. Acc.
Organ. Soc. 18(5), 467–488 (1993)
23. Johnson, P.E., et al.: Success and failure in expert reasoning. Organ. Behav. Hum. Decis.
Process. 53(2), 173–203 (1992)
24. Grazioli, S.: Where did they go wrong? an analysis of the failure of knowledgeable internet
consumers to detect deception over the internet. Group Decis. Negot. 13(2), 149–172 (2004)
25. Daft, R.L., Lengel, R.H.: Organizational information requirements, media richness and
structural design. Manage. Sci. 32(5), 554–571 (1986)
26. Costa, P.T., McCrae, R.R.: Four ways five factors are basic. Pers. Individ. Differ. 13(6),
653–665 (1992)
27. Parrish Jr, J.L., Bailey, J.L., Courtney, J.F.: A Personality Based Model for Determining
Susceptibility to Phishing Attacks. Little Rock: University of Arkansas (2009)
28. Gosling, S.D., Rentfrow, P.J., Swann, W.B.: A very brief measure of the big-five personality
domains. J. Res. Pers. 37(6), 504–528 (2003)
29. McKnight, H., Kacmar, C., Choudhury, V.: Whoops. Did i use the wrong concept to predict
e-commerce trust? modeling the risk-related effects of trust versus distrust concepts. In:
System Sciences, Proceedings of the 36th Annual Hawaii International Conference. IEEE,
Waikoloa, Hawaii, USA (2003)
30. Wright, R., Chakraborty, S., Basoglu, A., Marett, K.: Where Did They Go Right?
Understanding the Deception in Phishing Communications. Group Decision and Negotiation.
19(4), 391–416 (2009)
31. Allan, S., Gilbert, P.: Submissive behaviour and psychopathology. Br. J. Clin. Psychol. 36(4),
467–488 (1997)
962 I. Alseadoon et al.

32. Carlson, J.R., Zmud, R.W.: Channel expansion theory and the experiential nature of media
richness perceptions. Acad. Manag. J. 42(2), 153–170 (1999)
33. Sheng, S. et al.: Who falls for phish? A demographic analysis of phishing susceptibility and
effectiveness of interventions. In: Proceedings of the 28th International Conference on
Human Factors in Computing Systems. ACM, Atlanta, Georgia, USA (2010)
34. Kumaraguru, P. et al.: Getting users to pay attention to anti-phishing education: evaluation of
retention and transfer. In: Proceedings of the Anti-phishing Working Groups 2nd Annual
eCrime Researchers Summit, pp. 70-81. ACM, Pittsburgh, Pennsylvania (2007)
35. Buller, D.B., Burgoon, J.K.: Interpersonal deception theory. Commun. Theory 6(3), 203–242
(1996)
Chapter 90
Adaptive and Dynamic Service
Composition for Cloud-Based Mobile
Application

R. Kanesaraj Ramasamy, Fang-Fang Chua and Su-Cheng Haw

Abstract Adaptive and dynamic service composition are the main research
challenges in Service-oriented Architecture (SOA). We are looking into improving
the efficiency of dynamic composition and the methods on how to improve the
discovery of web services. Currently the approaches which are available do not
compose while checking for errors. In this paper, we will be composing web
services using Multi Agent Methods (MAS) and Petri Net for error checking to
improve the efficiency of the composition. We are implementing the composition
engine using Business Process Model (BPM) to compose web services while MAS
is used to locate web services. Proposed architecture is expected to produce a
dynamic web service composition engine and improve reusability for Cloud-Based
Mobile Application.

90.1 Introduction

Web services are becoming more important and have received great considerations
from the research community. Web services are reachable network interface to
application programs. They are generally created by using the standard Internet
technologies. Web services are also self-describing software applications that can
be advertised, located, and used across the Internet using a set of standards such as
SOAP, WSDL, and UDDI [1]. Web services composition allow application to
communicate on various platform.

R.K. Ramasamy (&)
F.-F. Chua
S.-C. Haw

Faculty of Computing and Informatics, Multimedia University, Cyberjaya, Malaysia
e-mail: [email protected]
F.-F. Chua
e-mail: [email protected]
S.-C. Haw
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 963

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_90
964 R.K. Ramasamy et al.

Service-oriented Computing paradigm (SOC), emphasis the web service com-

position, which exploits services as important basics for developing applications.
Web service helps to achieve interoperable Business to Business (B2B) interaction
offered by multiple business partners based on business processes [2]. This
interconnection of web services is called as web service composition. There are
two types of composition which are static and dynamic or automatic composition.
In this paper, we have elaborated on two different web service compositions and its
methods. The web service composition languages such as Business Process Exe-
cution Language for Web Services (BPEL4WS), Business Process Modeling
Language (BPML), Web Service Choreography Interface (WSCI), XML language
(XLANG), and Web Services Flow Language (WSFI) are used to specify process
models. The remainder of this paper is organized as follows. Section 90.2 intro-
duce service composition methods. Section 90.3 discusses the analysis of web
service composition methods. Section 90.4 describes about cloud service com-
position and Sect. 90.5 presents the proposed solution.

90.2 Service Composition Methods

90.2.1 Multi Agent Method

Intelligent behavior is intended as the ability of processing the input requirements

as introduced by the user, whereas the system is able to analyze the input and
identify the web services. It also carry out an automatic composition of web
services for corresponding business processes [3]. Agents are built with intelligent
behavior based on respective task to be completed. Agents are important to web
service composition because it being context-aware, autonomous and able to
interpret semantic with the help of ontological knowledge representation. Agents
interact with web services by requirement layer which is one of the three layers in
multi agent system [4]. Requirements are set to describe the interaction and roles
whereas contracts are used to describe the responsibilities of the roles in the
requirement layer which have three types of contract as follow:
• Permission shows the permit behavior of roles
• Obligation goals that the roles must accomplished
• Prohibitions describe the forbidden behavior of the roles
Agent layer is the second layer in MAS, which consists of role agent and
manage agent. Role agent will act according to the web service and manage agent
will monitor the agent and system behavior. Besides that, there are two approaches
that can be used to compose web services known as orchestration and
choreography.
Figure 90.1 shows orchestration approach. Orchestration can interact with both
internal and external web services and it occurs at the message level. Orchestration
90 Adaptive and Dynamic Service Composition … 965

Fig. 90.1 Orchestration

approach [4]

also represents control from one party’s perspective which is responsible for
coordinating the whole process of the system and does the composition while the
agent doesn’t need to know whom to contact. In the orchestration approach, agent
coordinator handles the composition. Orchestration is dynamic as the agent
coordinator take full responsibility of the new services instead of each service or
agent aware of it. Multi agent can self-update and self-compose the web services
and each agent knows its role hence it will be beneficial for us to use multi agent
method as it is autonomous and highly reliable. Figure 90.2 shows Choreography
approach. Choreography is collaborative and allows each invoked party to describe
its part in the interaction and track message sequence among multiple parties and
sources. Moreover, in this approach, agent is more independent compared to
orchestration approach where the entire agent is controlled by coordinator. All
service or agents in choreography approach have to be aware of business process,
exchange message, operation and timing of exchange message [4].

90.2.2 Complex Event Processing Method

Complex Event Processing (CEP) helps to control the events while multi events
can compose the complex event. According to David B. Robins, CEP is a part of
Event-Driven Architecture (EDA), which is an architecture dealing generally with
the production, detection, consumption of, and reaction to events [5]. CEP is using
detection of multiple event stream process and heterogeneous event. The events
occur when messages are sent and received Simple Object Access Protocol
(SOAP) messages and CEP adaptation needs the platform to consume incoming
messages, process them and send the results to its destination. Besides that, CEP
has the ability to recognize the complex events. This technology discovers a
relationship between events through the analysis and correlation of the multiple
events. CEP contains simple techniques and use new techniques such as patterns of
events including event relationships. A large part of CEP is pattern matching and it
consists filtering, which is to determine events of interest and extracting properties
from the event [5]. Table 90.1 describes the basic transformation patterns.
The CEP has higher level performances and capacity compared to other
approaches. It is an excellent platform which can carry many events and processes
966 R.K. Ramasamy et al.

Fig. 90.2 Choreography

approach [4]

Table 90.1 Five basic transformation patterns that can reconcile mis-matches
Patterns Description
Match-make It translates one message type to another
Solve the one-to-one transformation
Split The approach for the one-to-many patterns
It separates one message sent by the source into more than one to be received
separately
Merge The solution for the many-to-many patterns
operator It combines more than two messages into one message
Aggregate Solution for the one to one transformation
Disaggregate Solutions for the many to many transformations

(1000 to 100 k messages per second). The CEP plays its role as a great platform in
real time business implication however it has low efficiency in composing large
amount of services.

90.2.3 Petri Net Method

Petri net-based method is used for the data validation of web services composition.
It is developed with BPEL4WS. Petri net is used as validation tool in web service
composition to detect errors, misconceptions and failures. This method is focused
on selecting the best possible path for web services composition based on the user
request. In this method, the movement of token (black dots) from one place to
another caused by firing of transition. The firing represents an occurrence of the
event or an action taken. Petri net is a process modeling technique that has a
formal semantic that has been used to model and analyze several types of pro-
cesses including protocols and business process. Every node is either a place or a
transition and it is connected and directed. When a token occupied in place, it is
able to place at least one token that connected to the transition. The enabled
transition fire one token from every input place and depositing one token from
every output place [6].
90 Adaptive and Dynamic Service Composition … 967

The early forms of high level petri net are Coloured Petri Nets (CPNs). CPNs
are used as an underlying formalism and BPEL as inputs and define service
workflow net as a kind of CPNs-as a united formalism to describe services,
composition and mediator [7]. In [7], they separated the control flow with the
message exchange and provide a united formalism to describe services, compo-
sition, and mediator. Then High-level Petri net (HPN) was developed to overcome
the problems in web service, such as the use of complex structured data as tokens
and using algebraic expressions to annotate net elements. Figure 90.3 shows the
process flow of Petri net.
Basically petri net has three steps:
• Input places allowing messages to get into the process
• Output places, to go out from the process
• Internal places allowing for modeling of the process behavior.
Transitions represent the dynamic element (activity of the business process).
Based on the customer requests from the Petri net, once the services are identified
through Petri Net, the corresponding aspects developed are bonded to the join
points in the main code. It also can be used to evaluate the effectiveness of services
composition. The tokens describe the state of the net. Hereby an arc connects a
condition (place) with an event (transition) and vice versa [3].
If an arc goes from a condition to an event, it means that the condition is
necessary for the occurrence of the event. In this case, we have a pre-condition.
Analogously an arc from an event to a condition means, that the occurrence of the
event makes this condition true, and it is called a post-condition. In large number
of places, there are several places with same semantic meaning. Figure 90.4 shows
the normal process in web service that has two web services ‘‘BookInfo’’ Query
and Book Deal. Both the web services require the same input ‘‘BookName’’ from
user. Figure 90.4 is using different places for both the services even it requires the
same semantic. Petri net method will group in one place in which the semantic
have the same meaning with the tokens as shown in Fig. 90.5. If the transition is
not effective then there is no matching between the transitions and inputs or desire
outputs, the service can be deleted because it will not be used.
Petri net graph gives all possible combination of composition available with the
selected web services. By implementing this method in web service composition, it
improves the reliability of web service selection. Petri net also reduce service
selection scope and improve the composition efficiency.

90.2.4 SWORD Method

According to [8], SWORD is a set of tools for the composition of Web services
including ‘‘information-providing’’ services. SWORD used rules to represent a
service that able to produce specific output based on certain input given to the
service. To determine automatically whether a desired composite can be
968 R.K. Ramasamy et al.

Fig. 90.3 Petri net process

Input Input Input message
flow [3]
transition

Finish Receive

Petri net N = (P, T, F)

P-A set of place (ellipses),
T-A set of transition (box),
F- A flow relation (arcs/activity).

Fig. 90.4 Normal web BookInfo

BookName
service process [11]

Output
BookInfo
Query
Input

Quantity BookDeal list

Book Deal

Fig. 90.5 Using petri net BookName BookInfo

(merge same semantic) [11]

BookInfo Query

BookName

BookDeal list

Quantity

Book Deal
90 Adaptive and Dynamic Service Composition … 969

appreciated with the existing services, a rule—based expert system is established.

Additionally, SWORD can compose information by providing services such as
information, email and other services. It also generates a functional composition
plan given the functional requirements. SWORD uses Entity-Relation (ER) model
to specify the Web services. From the inputs and outputs of service, the rule based
expert systems defined whether which outputs can be obtained by the service
which indicates the inputs. Developer needs to specify the inputs and outputs of
the composite service in the world model and submit it to SWORD at the phase of
deployment. It is to create and deploy a new composite service. SWORD generates
a composition plan for the composite service. Moreover, the developer can view
the generated plan and the developer can also request for a persistent represen-
tation of the plan generated if the plan is appropriate [9].
The sequence of services is included in the representation of the plan that need
to be invoked in order to obtain the composite service outputs from its inputs.
According to [9], when actual requests for the composite service is received, the
services specified in the plan are executed, starting with the known inputs, in order
to compute the desired outputs. SWORD has the ability to self-compose when
given certain inputs and it is efficient in responding to its environment. It allows
developers to compose the existing web services efficiently to produce new web
services composition.

90.3 Analysis of Web Service Composition Method

MAS is autonomous because it has two approaches that can control its own agent.
For the first approach, the coordinator will take full responsibility of the services,
whereas the second approach, each agent acts independently. MAS can stay
operational and carry out the task immediately when the user gives an input.
Moreover, the agent will give the output based on what user request and robustness
in response to the environment. MAS will be highly complex if combined with
HPN. Table 90.2 shows the characteristic comparison between methods.
CEP which uses pattern such as aggregate. CEP is autonomous because it is
independent but low in efficiency when composing an input. It keep composing till
receive checkout messages that indicate the input has been sent. Sword and MAS
uses mediator however, Sword is rule based method. It composes based on the
rules that has been set. Petri net is neither autonomous nor independent because
there are no simultaneous images of the flow of parts, resources, information and
control through the system. This method offers the ability to visualize the system
flow, which makes it easy to understand and learn. Petri Net method efficiently
reduces service selection scope by finding the shortest path or critical path to
compose the service but it is highly complex because of its formula and distributed
choice of transition has to be executed [7]. Petri net unable to self-compose
because the flow of web service have different interface or mismatches in the
interface. Based on Table 90.2, the recommended method to be used to compose
970 R.K. Ramasamy et al.

Table 90.2 Comparison between methods

Characteristics/methods Multi agent (MAS) CEP/patterns Sword Petri net
Autonomous Yes Yes Yes No
Reliability Yes Yes Yes Yes
Efficiency Yes Low Yes Yes
Complexity Low High High Very High
Self-compose Yes No Yes No

web service is MAS. MAS is autonomy where it can self-update and highly
reliable. It is more efficient compared to other methods because of its robustness in
response to the dynamic environment. Moreover, MAS has lower complexity than
CEP, Sword and Petri net.

90.4 Cloud Service Composition

There are a lot of different definitions available for cloud. We accept the definition
of cloud, in which provided by the National Institute of Standards and Technology
(NIST) [10]. The NIST cloud computing definition: Cloud computing is a model
for enabling ubiquitous, convenient, on-demand network access to a shared pool
of configurable computing resources servers, storage, applications, and services)
that can be rapidly provisioned and released with minimal management effort or
service provider interaction. This cloud model is composed of five essential
characteristics, three service models, and four deployment models [10].
Cloud services work with all resources called as participants such as consumers,
brokers and service providers but participants may have incomplete knowledge of
all possible Cloud resources and even to other participants, due to constantly
changing Cloud environments and their distributed nature [4]. An efficient Cloud
service composition method should dynamically select the most appropriate ser-
vices to create the cheapest composition solution. As service providers charge a
fee for providing a service and this service fees may change over the time, based
on supply and demand. Referring to [4], to satisfy incoming client requirements,
there is a collaboration needed between brokers and service providers. Those
requirements can be received via Cloud resources which can be accessed by using
web service in an automated manner. However, constantly changing cloud com-
puting environments, which are deployed as self-contained components, are nor-
mally partial solutions that must be composed to provide a single virtualized
service to Cloud consumers. This composition of services should be carried out in
a dynamic and automated manner to satisfy consumer’s requirements. The idea of
adopting software agents for managing Cloud services was first introduced in [10].
Mell and Grance [10] are the earliest efforts in designing and constructing
90 Adaptive and Dynamic Service Composition … 971

Fig. 90.6 Architecture of web service composition method

Table 90.3 Explanation for stages

Data Description
flow
A User input from mobile phone to cloud (HTML 5 Application)
B User inputs transferred to business process model
C Business process model with embedded rules will decide which role agent to trigger
D Role agent will search for the respective web service to be used
E Selected web service will be sent to business process model to perform operation
F Upon completion of all the processes in business process model, the output will be
replied back to HTML 5 application which is placed in CLOUD
G The output received in HTML 5 application will be reflected on user’s mobile phone

negotiation agents for supporting the establishment of service-level agreements

among Cloud participants, this present research introduces a self-organizing agent-
based approach for dealing with Cloud service composition.

90.5 Proposed Solution

Based on the analysis from various sources as discussed above, we have proposed
a new architecture for a cloud-based mobile application which have the capability
to compose web services dynamically. Besides that, our focus is also on designing
the architecture that is adaptive for the provided environment. Figure 90.6 shows
the architecture of the proposed idea. Table 90.3 provides the explanation of data
flow based on Fig. 90.6.
972 R.K. Ramasamy et al.

Area 2
Area 3

Fig. 90.7 Area 2 and area 3 (adaptive and dynamic process)

Referring to Fig. 90.6, the proposed architecture is divided to 3 main areas. The
first area is the user input or end user input for the model. Second area is the
adaptable area finally the dynamic searching area.
Adaptability happens in area 2. We have adapted the orchestration approach as
used in [11] to improve the availability. HPN as proposed by [3], will be used to
reduce errors. We aim to reduce service selection scope and improve on the
composition efficiency. Business Process Execution Language (BPeL) will be used
to execute the designed Business Process Model. Our proposed solution will cover
exceptional behavior handling such as faults, events and compensation.
As shown in area 3 in Fig. 90.7, dynamic composition can be achieved by using
automated tools to analyze a user query, and select and assemble Web service
interfaces to fulfill the user demand. Choreography approach will be used to create
the dynamic environment. Agents will receive requests from BPM and it will
search the correct service and return it back to BPM. Agents will also update the
error log when the requested semantic was not found during the searching process.
The agent will refer to the error log before the searching is started to check on the
availability and the location of the services.

90.6 Conclusion

The proposed system will be tested using a real time mobile application such as
early warning system for natural disasters. Besides that, it will also support various
mobile platform such as Windows Phone, IOS and Android. As for future work,
90 Adaptive and Dynamic Service Composition … 973

we are planning for performance analysis using QoS attributes to test the efficiency
of adaptive service composition strategy. It is expected to provide an improved
search result and reduce composition time.

References

1. Liu, Y., Ngu A.H., Zeng L.Z.: QoS computation and policing in dynamic web service
selection. In: Proceedings of the 13th International World Wide Web Conference on
Alternate Track Papers & Posters, pp. 66-73. ACM (2004)
2. Papazoglou, M.P., Van Den Heuvel, W.: Service-oriented computing: state-of-the-art and
open research issues. IEEE Comput. 40(11), 1086 (2003)
3. García Coria, J.A., Castellanos-Garzón, J.A., Corchado, J.M.: Intelligent business processes
composition based on multi-agent systems. Expert Syst. Appl. 41, 1189–1205 (2014)
4. Maalal, S., Addou, M.: A new approach of designing multi-agent systems. Int. J. Adv.
Comput. Sci. Appl. 2, 148–149, 149–150 (2011)
5. Taher, Y., Parkin, M., Papazoglou, M.P., Van Den Heuvel, W.: Adaptation of Web Service
Interactions Using Complex Event Processing Patterns, pp. 601–609. Springer, Heidelberg
(2011)
6. Zhao, L., Li, B., Wu, J.: A multi-agent collaborative model for web services composition. In:
ISECS International Colloquium on Computing, Communication, Control, and Management.
CCCM 2009, vol. 1, pp. 145–148. IEEE (2009)
7. Bencomo, N., Whittle J., Sawyer, P., Finkelstein, A., Letier, E.: Requirements reflection:
requirements as runtime entities. In: Proceedings of the 32nd ACM/IEEE International
Conference on Software Engineering, Vol. 2, pp. 199–202. ACM (2010)
8. Chan, P.P., Lyu, M.R.: Dynamic web service composition: a new approach in building
reliable web service. In: 22nd International Conference on Advanced Information
Networking and Applications. AINA 2008, pp. 20–25. IEEE (2008)
9. Ponnekanti, S.R., Fox, A.: Sword: a developer toolkit for web service composition: In
Proceedings of the Eleventh International World Wide Web Conference, Honolulu, HI, vol.
45. (2002)
10. Mell, P., Grance, T.: The NIST Definition of Cloud Computing. Computer Security Division,
Information Technology Laboratory, National Institute of Standards and Technology, p. 50
(2009)
11. Yamaguchi, H., El-Fakih, K., Bochmann, G.V., Higashino, T.: A Petri net based method for
deriving distributed specification with optimal allocation of resources. In: Proceedings of
ASIC International Conference on Software Engineering Applied to Networking and Parallel/
Distributed Computing (SNPD’00), pp. 19–26 (2000)
Chapter 91
Web Service Composition Using Windows
Workflow for Cloud-Based Mobile
Application

R. Kanesaraj Ramasamy, Fang-Fang Chua and Su-Cheng Haw

Abstract There is a lot of support provided by the mobile OS giants in cloud

services. Besides providing their own services, they encourage developers to
upload services into their cloud environment. However, when more services are
being consumed in their cloud, more charges imposed to the users. Meanwhile,
there exists a large amount of services which are available in public web service
repositories. The cost of mobile development can be reduced by invoking and
composing the services in sequence and upload or use them as a single service. We
have used Windows Workflow for web service composition and the WSDL file
produced can be hosted in our cloud environment and to be used by all. In this
paper, we will be discussing on how to use Windows Workflow for web service
composition for cloud-based mobile application. With the proposed system we will
be able to save battery life time and increase the storage capacity.

91.1 Introduction

Web services are unique as it can be supported by various platforms although it

was developed in one particular platform. Web services which are available in
service repositories can be used in one single system, which is called as web
service composition. In other word, various web services can be composed in one
application [1]. In fact, web service interfaces are like remote procedure call and
the interaction protocols are manually written [2]. Application can be developed

R.K. Ramasamy (&)  F.-F. Chua  S.-C. Haw

Faculty of Computing and Informatics, Multimedia University, Cyberjaya, Malaysia
e-mail: [email protected]
F.-F. Chua
e-mail: [email protected]
S.-C. Haw
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 975

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_91
976 R.K. Ramasamy et al.

by invoking a couple of services using the composition method to reduce devel-

opment time. There are plenty of support being provided by various service pro-
viders such as Google and Microsoft since web services can be called remotely.
They have provided a platform with all the prebuilt services to be used for our
daily development whereby they also encourage the developer to add on more
services into their cloud environment. This indicates the widespread of cloud
environment in current mobile development.
Besides that, service composition can be executed by composing elementary or
composite services. When composing web services, the sequence of events that
occur in any business process will be implemented by few services. This will lead
to workflow management, where the logic is comprehended by composing
autonomous applications. These composite services also can be in turn recursively
composed with other services into higher level solutions, and so on [3]. Further-
more, this recursive composition feature allows developing new solution rapidly
based on the existing business services and becoming one of the most significant
features of SOA.
Deadlock, mismatching and availability of web service are the major issues to
be faced by developers in the development process of web service composition for
complex and distributed environment [4]. In order to solve the problems, they
divided the composition process into three parts: constraints before composition,
constraints in composition and constraints after composition. They used finite state
machine to model web services to discuss the constraints in web service compo-
sition. Before the composition occurs, the status of web service should be con-
firmed as active even though it is difficult to be determined.
The paper discusses about the development of web service composition for
Cloud-Based mobile application. We had tested the efficiency of stand-alone
mobile application with cloud-based mobile application. With this implementa-
tion, it is expected to have complex mobile application running on mobile devices
without compromising the resources of the mobile device.
The remainder of this paper is generally organized as follow: Sect. 91.2 will
introduce the related works which have been done by other researchers and ideas
that provide motivation for the proposed solution. Section 91.3 shows the pro-
posed solution on how it’s being implemented. Section 91.4, shows the results of
the implementation and discussion. In Sect. 91.5, analysis of the result will be
discussed to identify the problems which have been fixed with the proposed
method. Finally, a conclusion is drawn and future work is described.

91.2 Related Works

Web service composition provides an opportunity for enterprises to increase the

ability to adapt themselves to frequent changes in users’ requirements by inte-
grating existing services [5]. According to [6], they present a novel algorithm called
HRLPLA (Hierarchical Reinforcement Learning) for composing Web services.
91 Web Service Composition … 977

This algorithm deliberates functional and QoS properties simultaneously. However,

the plans generated by AI planning are not adaptive to running environment [6].
This lead to failure of composition plan when a component service stops func-
tioning. Additionally, it is not suitable for dynamic environment. To solve this
problem, they introduced MAXQ, a HRLPLA method where it provides efficient
service composition and can work with a large scale of services. Since it is using
hierarchical structure, information obtained in designed subtasks can be reused by
super tasks; therefore it accelerates the process of learning. Their approach is
proven through experimental result to achieve higher efficiency and effectiveness
than normal reinforced learning and AI planning.
Though every method provides different level of automation in service com-
position, they are not to prove that the higher automation is the best. This is
because the Web service environment is highly complex and it is not feasible to
generate everything in an automatic way [7]. Jinghai and Xiaomeng [7] presents a
technique to generate composite services from high level declarative description.
This method used composition rules to define whether two services can be com-
posed. This method contains 4 phases; first specification phase where Composite
Service Specification Language (CSSL) language used. Secondly, in matchmaking
phase, it accepts service requester’s specifications by generating compositions plan
using composition rules. Thirdly is the selection phase where service requester
select particular plan generated in selection phase based on quality of composition.
The final phase is the generation phase. In this phase, a detailed description of the
composite service is automatically generated and presented to the service requester
[7]. The composition rules deliberate the syntactic and semantic properties of Web
services and at the same time contribute the Web service’s attributes that could be
used in service composition. These rules can be used as a guideline for other Web
service methods also [7].
Another service composition approach is non-QoS based approaches. Jinghai
et al. [8] presents a mixed initiative framework for semantic web service discovery
and composition which allow user involvement over many key decisions by
suggesting and identifying the inconsistencies. In this approach, users decide the
number of functionalities to be supported. The quality and accuracy of the support
provided by their framework is intended to improve over time, as users learn to
develop richer and more accurate annotations. For this composition engine, it
combines Web Ontology Language (OWL) ontologies where it plans functionality
with Jess based on the GraphPlan algorithm. GraphPlan provides reach-ability
analysis to determine whether a given state can be reached from another state and
disjunctive refinement [8].
In [9], the authors utilize the semantic web service languages with the model
driven methodology to build composite web services. There are four phases and at
the end of first phase, abstract composite model is attained. It contains all the vital
information used for service discovery and selection. On the second phase, dis-
covery process depends on semantic descriptions where appropriate web services
are handled. While on third phase, a concrete composite model is attained to handle
mismatch between output of one service and input or other service. In the last phase,
978 R.K. Ramasamy et al.

different descriptions of the concrete composition model are used for the composed
service [9]. The methodology also deliberates a syntactic and semantic description
about the interfaces of service candidates at the same time processes QoS
requirements from the developer and offerings from the service providers [9]. The
benefit is this methodology provide good documentation of composition in the form
of graphical models. Executable compositions and the ability generate semantically
from a graphical model offers valuable advantage to the service developers, where
they don’t have to write a lot of low-level XML code.

91.3 Proposed Solution

Significantly, stand-alone mobile application responds faster compared to cloud

based application. This is because, a stand-alone mobile application only loads
internally meanwhile a cloud-based mobile application will load from cloud ser-
ver. We would like to test the performance of the applications towards Business
Process which are created by composing a few services. Figure 91.1 shows the
proposed architecture for our Cloud-Based mobile application. Based on Fig. 91.1,
the engine for the mobile application will be placed in cloud environment. We
consider Windows Workflow Foundation (WF) and HTML 5 pages as the engine.
WF is used to compose web services together to provide a complex mobile
application. Table 91.1 shows the description for proposed architecture for cloud-
based mobile application.
WF is a Microsoft technology that provides an API, an in-process workflow
engine, and a rehostable designer to implement long-running processes as work-
flows within .NET applications. The current version of WF was released as part of
the .NET Framework version 4 and is referred to as (WF4) [10]. The term
‘‘Workflow’’ here refers to the root activity that is executed by the host. In
addition, Workflows can use both out-of-box activities and custom activities [10].
For the deployment and testing, we are using Windows Phone (Nokia Lumia
920) and Android Emulator to test the efficiency in completing the composition
and getting the results. The test environment is being set up in a personal computer
which runs on Intel i7 2.66 GHz with 8 GB RAM.
We used BMI calculator as the domain for testing because it’s easy to under-
stand but it has more than one service to be composed. Proposed system uses three
different web services. Table 91.2 shows the description of the services used in
this project.
Proposed workflow will receive four values from the mobile application which
are two values with double data type and two values with the metric used in the
system for the measurements. In the workflow, the system will convert the width to
kilogram and height to centimetres using the service provided in workflow. Then
the converted value will be send to BMI calculator for BMI calculation and return
the BMI value to the user. (91.1) explains the formula being used for this system or
the calculation of BMI in this system. To ensure the values meeting the required
91 Web Service Composition … 979

Internet Information Services

B

HTML 5 Windows
Workflow
C
E

A F
Web Service
D Repository

Fig. 91.1 Proposed architecture for cloud-based mobile application

Table 91.1 Description for proposed architecture for cloud-based mobile application
Stage Description
A User input from mobile device will be sent to HTML 5 page
B HTML 5 will trigger the WSDL produced from WF
C WF will invoke web service related to functions in workflow
D The processed data in the related web service will be returned to WF
E The WSDL will be return the processed data to HTML 5 page
F Based on the processed data HTML 5 will produce the output to mobile device

Table 91.2 Description of services used

Service Description
BMI Calculates BMI but only receives weight in kilogram and height in
centimetres
Width converter Converts all the values received to desired width metric as required by user
Length Converts all the values received to desired length metric as required by user
converter

metrics for the calculation, we have included two separate services from a web
service repository to perform the conversion and provide the value to BMI cal-
culator service.
 
BMI ¼ Weight½kg= ðHeight ½cm=100Þ2 : ð91:1Þ

Figure 91.2 shows the flowchart of the Business Process (BP) where the
composition takes place. Every process in flowchart will be using a web service
from various web service repositories. In Fig. 91.3, we have shown the BP, which
980 R.K. Ramasamy et al.

Fig. 91.2 BMI BP flowchart

False True

False True

was developed on Windows Workflow based on Fig. 91.2. This BP will be

deployed in a form WSDL. We will have 3 different WSDL which have been
combined as one single WSDL which will call and perform the operation. This
will improve the reusability of the code as developers do not have to call all the
web services manually.
The composition is done using Windows Workflow 4.0 running on.NET
Framework 4.5. We have used Visual Studio 2013 RC to develop the Windows
Workflow project. The design that we have developed is able to handle the error or
exception catching. We have used the try catch blocks to develop the workflow so
that any error will be handled in workflow level and will not interrupt users operation
or generate any system error for the user. Furthermore, to improve the performance
in workflow, we have also included IF ELSE blocks, if user have selected the
required metrics, it doesn’t go through the conversion again. Figure 91.3 shows the
implemented workflow or the composition diagram for this system.
91 Web Service Composition … 981

Fig. 91.3 BMI BP in

windows workflow

91.4 Implementation and Testing Results

As shown in Fig. 91.4, a stand-alone Windows Phone Application was developed

and tested. Figures 91.5 and 91.6 shows the Cloud-Based Mobile Application
which was developed for this testing. Using this emulator and Nokia phone we
have collected the data of total response time to receive user input, invoke com-
posed WDSL and finally display the result to user’s screen in milliseconds as
shown in Table 91.3. As for the test data, we have fixed the same 10 sets as we
would like to test in all areas in this system such as converting the value, based on
the metric required by the service. Since we have IF ELSE blocks in our workflow
it is expected that it will provide a shorter response time compared to the flow that
requires metric conversion.
Results in Table 91.3, indicates that Cloud Based application respond faster
compared to a stand-alone application when there is WSDL to be executed from
another server. The set of data used are the same for the entire test. We have fixed
982 R.K. Ramasamy et al.

Fig. 91.4 WP stand-alone

application

Fig. 91.5 Cloud based

application in WP
91 Web Service Composition … 983

Fig. 91.6 Cloud based

application in android

Table 91.3 Test result

Weight Height Time (ms)
WPSA WPCBA ACBA
74 (kg) 1.74 (m) 813 91 808
57 (kg) 157 (cm) 678 571 560
59 (kg) 172 (cm) 645 568 568
71 (kg) 175 (cm) 283 565 438
65 (kg) 1.61 (m) 254 87 923
52 (kg) 166.5 (cm) 525 552 300
71 (kg) 156 (cm) 660 647 560
71000 (g) 172 (cm) 420 62 454
69 (kg) 173 (cm) 492 557 563
68 (kg) 1.65 (m) 748 442 142
Average 551.8 414.2 531.6
WPSA windows phone stand-alone application
WPCBA windows phone cloud-based application
ACBA android cloud based application

the measurement metrics for test data as the Business Process will convert the
metric if it is not the same as required by the BMI service.
Figure 91.7 was generated based on Table 91.3. We are able to identify the
pattern of total response time needed for every type of application. By average it
shows that Cloud Based Mobile Application takes shorter time compared to stand-
alone mobile application when there is a WSDL to be triggered or invoked.
984 R.K. Ramasamy et al.

Time (Miliseconds)
Fig. 91.7 Test result 1000
800
600
400
200
0
0 2 4 6 8 10 12
Test Cases
WPSA WPCBA ACBA

The results are also affected by the network load such as number of PC and the
operations in the network. The experiment was not conducted in an empty net-
work, as we were considering the network where usually mobile devices are used.
Most mobile devices are used in 3G network rather than office or home private
WIFI which is much slower compared High Speed Broadband.

91.5 Analysis and Discussion

Based on the result obtained, we are able to have a single process for all the
application in various platforms. This was a problem raised by [11] where pro-
grammers have to develop same process again for multiple platform. Besides that,
using service composition for mobile application will also help the developer to
build any web applications in future using the same Business Process which is
being created for mobile application.
Furthermore, as stated in [4], getting the status of the web service before a
composition happen is very important but it’s impossible to check the status of the
web service every time we would like to use it. It will increase the burden to the
server. For this experiment, we have implemented TRY CATCH blocks in our
Business Process. We will catch the error in workflow level. If the developer will
like to log the errors, they could request it from the return values in the workflow.
We are using the HRLPLA method in our service composition. It’s a hierar-
chical based composition where the data used by parent process will be assigned to
the child process. By using Cloud-Based Mobile application for service compo-
sition, it will increase the battery life time for the mobile device compared to
stand-alone mobile application. This is because all the processing power will be
handled by the cloud servers as compared to the stand-alone application which will
use up the mobile device resources. For storage purpose, stand-alone application
are not able to store a large amount of data compared to Cloud-Based Mobile
Application as it will be using the cloud server storage to place all the data. With
the proposed solution, mobile user can have more applications installed in the
mobile device and have longer battery life time even with complex applications
are loaded in the phone.
91 Web Service Composition … 985

91.6 Conclusion

Based on the experiment that we have completed, we have proved that Cloud-
Based Mobile application responds faster compared to a stand-alone application.
This have been considered as a very important test as mobile application that has
shorter loading time are the stand-alone applications, however when it comes to
processing from cloud servers, Cloud-Based mobile application have shorter
loading time. This application might not load as fast as stand-alone application for
the UI, but its responds faster when it comes to invoking web services from
different servers.
This project will be further improved to have more complex Business Process
where is be able to dynamically select web services based on the mobile appli-
cation requirement. This selection will be done in workflow level using software
agents in multiple levels. There are a couple issues in dynamic service composition
for mobile application such as semantic discovery for web services and avail-
abilities issues. We will be looking into all this issues in our future work and will
be tested using more complex case studies.

References

1. Samwell, J.: What is web service composition. http://stackoverflow.com/questions/1244425/

what-is-web-service-composition. Accessed 16 Oct 2013
2. Srivastava, B., Koehler, J.: Web service composition. Current solutions and open problems.
In: ICAPS 2003 vol. 1, p. 23 (2003)
3. Lublinsky, B.: Service composition. http://www.infoq.com/articles/lublinsky-soa-composition.
Accessed 25 Oct 2013]
4. Yan, H., Hui, W.: Constraints in web services composition, networking and mobile
computing. In: 4th International Conference Wireless Communications, pp. 1–4 (2008)
5. Roy, G., Michael, J.C.: Model-driven semantic web service composition. In: Asia-Pacific
Software Engineering Conference (APSEC’05), vol. 12, pp. 79–86 (2005)
6. Hongbing, W., Xiaohui, G.: An adaptive solution for web service composition. In: Services
(SERVICES-1), pp. 503–510 (2010)
7. Jinghai, R., Xiaomeng, S.: A survey of automated web service composition methods. In:
Proceedings of First International Workshop on Semantic Web Services and Web Process
Composition. vol. 1(2), p. 67 (2004)
8. Jinghai, R., Dimitar, D., Paul, H., Norman, S.: A mixed initiative approach to semantic web
service discovery and composition. In: SAPs Guided Procedures Framework. IEEE
International Conference on Web Services, pp. 401–410 (2006)
9. Groenm, R., Jaeger, M.C.: Model-driven semantic Web service composition. In: 12th Asia-
Pacific Software Engineering Conference (APSEC 2005), p. 8 (2005)
10. Pluralsight, M.: A developer’s introduction to windows workflow foundation (WF) in .NET
4. http://msdn.microsoft.com/en-us/library/ee342461.aspx. Accessed 17 Oct 2013
11. Anthony I.W.: Software Engineering Issues for Mobile Application Development. Genetic
Algorithms + Data Structures = Evolution Programs, 3rd edn, pp. 397–400. ACM, Berlin
(1996, 2010)
Chapter 92
An Effective Image Retrieval Method
Based on Fractal Dimension Using Kernel
Density Estimation

Zhang Qin, Huang Xiaoqing and Liu Wenbo

Abstract Fractal coding has been proved useful for image compression, and it is
also proved effective for image retrieval. In the paper, we present a statistical
method called variable bandwidth kernel density estimation to analyze fractal
coding parameters. Then retrieve images using the retrieval index constructed with
this method. Experimental results show that the proposed method with a variable
optimized bandwidth performs better than those with a fixed bandwidth and the
histogram method both in retrieval rate and retrieval speed. In this paper, the
Average Retrieval Rate (ARR) can reach 72.40 %, which is more than that
obtained by the existing methods.

Keywords Fractal coding  Kernel density estimation  Variable optimized


bandwidth Image retrieval

92.1 Introduction

Image retrieval has been an active research area for years, there are various kinds
of image retrieval technique mainly based on text, content and semantic. The
Content-Based Image Retrieval (CBIR) technique [1] is used to retrieve images
with image features directly, so we can find the most similar images from the
database through the comparison between the image features. Fractal coding
parameters can effectively represent essential features of images. Fractal coding, as
a new image compression technique, has been applied into image retrieval. Fractal
features provide geometric information of an image that is irrelevant to the shape
and size of an object in the image, therefore, fractal features are more robust than

Z. Qin (&)  H. Xiaoqing  L. Wenbo

College of Automation Engineering, Nanjing University of Aeronautics and Astronautics
(NUAA), Nanjing, China
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 987

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_92
988 Z. Qin et al.

color and texture features. Meanwhile, retrieving images in fractal domain can be
faster and more effective, especially for the compressed images.
Fractal image coding is a block-based scheme that exploits the self-similarity
hiding within an image. Fractal features generated by the block-based scheme are
quantitative measurements of self-similarity, therefore they can be used to con-
struct image features. Fractal image compression was originally developed by
Barnsley and Sloan [2]. Jacquin [3] implemented a block-based fractal compres-
sion, which is popularly known as fractal block coding. And fractal block coding
has been applied into image retrieval. Pi MH proposed to employ the histogram of
range block means and the 2D joint histogram of range block means and contrast
scaling parameters as an image index [4, 5], and this technique greatly improved
the retrieval rate. Some scholars proposed the histogram of collage error as an
image signature and combined fractal parameters with collage error to improve the
retrieval rate [6].
The features of an image can be acquired effectively with the statistical char-
acteristics of fractal coding parameters, and the performance in image retrieval has
already been confirmed. A statistical method called kernel density estimation is
proposed, which can estimate the density of samples more accurately. Compared
with the commonly used histogram method, the kernel density estimation can be
more accurate and smooth. Therefore, we apply this method into image retrieval.
Since the bandwidth of kernel function plays an important role in kernel density
estimation, we propose the method with a variable optimized bandwidth in con-
formity with data [7]. Experimental results show that this method has not only
higher retrieval rate but also less retrieval time than the existing methods.
The rest of the paper is organized as follows. The Sect. 92.2 introduces fractal
coding and collage error. Section 92.3 presents the proposed method. Section 92.4
presents the performance evaluation. The Sect. 92.5 presents conclusions and
future work.

92.2 Fractal Coding and Collage Error

92.2.1 Fractal Coding

In this paper, the orthogonalization fractal coding method is adopted [8]. An image
ðM  MÞ is first segmented into non-overlapping blocks of size B 9 B called
range blocks, recorded as R1 ; R2 ; . . .Rm . A domain block pool X is a set of domain
blocks of size 2B 9 2B, generated by dividing the ðM  MÞ image into overlapped
blocks. And the domain blocks are recorded as D1 ; D2 ; . . .Dn . In general, B ¼ 2t , t
is an integer. After the 4-neighborhood pixel average and compression transform,
the domain blocks are mapped into the images with size B 9 B. To improve the
quality of the images, eight kinds of isometric transform are applied into the
domain blocks. (In Jacquin’s scheme, rotation transformation of 0 , 90 , 180 ,
92 An Effective Image Retrieval Method … 989

270 , vertical midline, horizontal midline and diagonal reflection transformation of

45 , 135 are proposed)
According to the Partitioned Iterated Function System (PIFS), we can find out
the domain blocks matched with the range blocks using affine transformation
iterations.


R0 ¼ r U þ sqðD  dUÞ ð92:1Þ

For each range block R, orthogonalization fractal block coding is obtained by

minimizing the following equation
 2
EðR; DÞ ¼ R  ri U  sj qðD  dUÞ ð92:2Þ

The above minimization is performed over D 2 X by working with a set of pre-

quantized fractal parameters fri gIi¼1 and fsj gJj¼1 (I and J are the quantization levels
for ri and sj , respectively). Note that U is a matrix whose elements are all ones, s is
a contrast scaling parameter, q is the isometric transform, jj  jj is the 2-norm and r
and d are the average of range block and domain block respectively. ri is the
 
average of the ith range block. Since U; D  dU ¼ 0, we define Eq. (92.2) as
orthogonalization fractal block coding. Then range block R can be written as
ðr ; s; xD ; yD Þ ¼ arg min EðR; DÞ. Where ðxD ; yD Þ is the top-left corner coordinate of
D2X
the ‘best-matching’ domain block.
When all the domain blocks matched with the range blocks are found, the
fractal coding of the whole image is completed.

92.2.2 Collage Error

We define the collage error as follows:

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
_ EðR; DÞ
e ¼ EðRÞ ¼ min ð92:3Þ
D2X BB

Collage error is a quantitative measure of the similarity between range block

and ‘‘best-matching’’ domain block. It is relatively robust compared with other
fractal parameters which can be quite sensitive to changes in domain block pool. Pi
MH [9] has proved that the proposed indices not only reduce computational
complexities, but also enhance the retrieval rate, compared with the existing
fractal-based retrieval methods.
990 Z. Qin et al.

92.2.3 Program Code

[imagem imagen]=size(Image1);
Sr=4;Sd=8;
Rnum=(imagem/Sr)*(imagen/Sr);
Dnum=(imagem/Sd)*(imagen/Sd);
Image2=zeros(Dnum,Sr,Sr);
Image2=blkproc(Image1,[Sd/Sr,Sd/Sr],'mean(mean(x))');
RBlocks=zeros(Rnum,Sr,Sr);
DBlocks=zeros(Dnum,Sd,Sd);
DBlocksReduce=zeros(Dnum*8,Sr,Sr);
for i=1:imagem/Sr
for j=1:imagen/Sr
k=(i-1)*imagen/Sr+j;
RBlocks(k,:,:)=Image1((i-1)*Sr+1:i*Sr,(j-1)*Sr+1:j*Sr);
end
end
for i=1:imagem/Sd
for j=1:imagen/Sd
k=(i-1)*imagen/Sd+j;
m=Sr; n=Sr;
DBlocksReduce(k,:,:)=Image2((i-1)*Sr+1:i*Sr,(j-1)*Sr+1:j*Sr);
DBlocksReduce(k+Dnum,:,:)=DBlocksReduce(k,m:-1:1,:);
DBlocksReduce(k+2*Dnum,:,:)=DBlocksReduce(k,:,n:-1:1);
DBlocksReduce(k+3*Dnum,:,:)=DBlocksReduce(k,m:-1:1,n:-1:1);
DBlocksReduce(k+4*Dnum,:,:)=reshape(DBlocksReduce(k,:,:),Sr,Sr)';
A=reshape( DBlocksReduce(k+3*Dnum,:,:),Sr,Sr)';
DBlocksReduce(k+5*Dnum,:,:)=A(:,n:-1:1);
DBlocksReduce(k+6*Dnum,:,:)=
imrotate(reshape(DBlocksReduce(k,:,:),Sr,Sr),90);
DBlocksReduce(k+7*Dnum,:,:)=
imrotate(reshape(DBlocksReduce(k,:,:),Sr,Sr),270);
DBlocks(k,:,:)=Image1((i-1)*Sd+1:i*Sd,(j-1)*Sd+1:j*Sd);
end
end
RandDbest=zeros(Rnum,1)+256^3;
RandDbests=zeros(Rnum,1);
RandDbesto=zeros(Rnum,1);
RandDbestj=zeros(Rnum,1);
for i=1:Rnum
x=reshape(RBlocks(i,:,:),Sr*Sr,1);
meanx=mean(x);
for j=1:Dnum*8
y=reshape(DBlocksReduce(j,:,:),Sr*Sr,1);
meany=mean(y);
s=(x-meanx)'*(y-meany)/((y-meany)'*(y-meany));
o=(meanx-s*meany);
e=(x-s*y-o)'*(x-s*y-o);
if (RandDbest(i)>e)
RandDbest(i)=e;
RandDbests(i)=s;
RandDbesto(i)=o;
RandDbestj(i)=j;
end
end
end
92 An Effective Image Retrieval Method … 991

92.3 Proposed Method

Kernel Density Estimation (KDE), as a popular nonparametric density estimation,

is widely used in the field of pattern recognition, classification and image pro-
cessing. The histogram is the simplest non-parametric density estimation method
which is frequently used. It has been demonstrated that histograms of fractal
parameters capture statistical characteristic of texture images effectively [4]. Since
the histograms heavily depends on width of bins and end points of bins, different
bins and end points may result in different histogram distribution, meanwhile,
different distributions lead to different results of image retrieval, and thereby affect
the retrieval rate.
On the condition that the densities of data are unknown, the kernel density
estimation is used, and it has been applied into statistical characteristics of images
in massive literatures [10, 11] and has obtained considerable results. The prop-
erties of kernel density estimation are, as compared to histograms, smooth, no end
points and they depend on bandwidths heavily. This method is more simple and
effective than the histogram method, which reduces the computational complexity
of data in image retrieval.

92.3.1 Kernel Density Estimation of Fractal Parameters

Let x1 ; x2 ; . . .xn be an i.i.d (independent and identically distributed) sample drawn

from some distribution with an unknown density p(x). We are interested in esti-
mating the shape of this function p(x). Its kernel density estimation is

_ 1 Xn
x  xi
pn ðxÞ ¼ Kð Þ ð92:4Þ
nh i¼1 h

where KðÞ is the kernel—a symmetric but not necessarily positive function that
integrates to one—and h is a smoothing parameter called the bandwidth. Intui-
tively we want to choose h as small as possible, which will lead to instability.
However, we should balance the smoothness and stability of the estimation.
However, it’s the bandwidth not the function of the kernel that exhibits a strong
influence on the estimation results. A bandwidth h of the kernel may alter the
density estimation, and it can accordingly affect the goodness-of-fit of the density
function pn ðxÞ to the unknown underlying density pn ðxÞ. Generally the most
common optimal criterion used to select bandwidth is the Mean Integrated Squared
Error (MISE) [12]. This principle is applied to select a fixed optimal bandwidth.
The optimal bandwidth is the argument that minimizes the MISE.
992 Z. Qin et al.

Z
MISEð^pn ðxÞÞ ¼ E ½^pn ðxÞ  pn ðxÞ2 dx ð92:5Þ

The integrand of the MISE can be decomposed into three parts:

E^p2n  2pn E^
pn þ p2n . Then we will subtract p2n from the MISE since it is the
underlying density and does not depend on the choice of a kernel, thus the cost
function as a function of the bandwidth is defined as
Z b Z b Z b
0
MISE ¼ MISE  p2n ðxÞdx ¼ E^p2n dx 2 pn E^pn dx ð92:6Þ
a a a

Here (a, b) is an interval of interest, and the interval length is H. The minimum
of the cost function Eq. (92.6) is an estimate of the fixed optimal bandwidth, which
is denoted by h .
After obtaining the fixed bandwidth, we will introduce the proposed method to
obtain a variable bandwidth. First we define a formula:

1X N
y¼ dðx  xi Þ ð92:7Þ
n i¼1

where n is the number of estimated point. dðtÞ is the Dirac delta function. The
kernel density estimation is obtained by convoluting a Gaussian kernel k(s) to y.
Z
^pn ¼ yxs kðsÞds ð92:8Þ

The most commonly used kernel is Gaussian kernel function:

1 s2
kðsÞ ¼ pffiffiffiffiffiffi expð 2 Þ ð92:9Þ
2ph 2h

As we know, the fixed bandwidth is selected from an entire observation interval

(a, b), however, the estimation may be improved by using a kernel bandwidth
which is adaptively selected in conformity with data. Thus, the kernel density
estimation with the variable bandwidth hx is expressed as
Z
^pn ¼ yxs kh ðsÞds ð92:10Þ

Here, we provide a method for obtaining the variable bandwidth hx that min-
imizes the MISE by optimizing a local interval length among which the variable
bandwidth can be regarded as a fixed one. To conduct the local optimization, we
introduce the local MISE criterion as
92 An Effective Image Retrieval Method … 993

Z
localMISE ¼ E ½^pn ðxÞ  pn ðxÞ2 qH dx ð92:11Þ

The weight function qH localizes the integration in the interval H. According to

Eq. (92.6), we introduce the local cost function by subtracting the term irrelevant
for the choice of h as
Z
1X 2X
Cn ðh; HÞ ¼ localMISE  p2n qH dx ¼ 2
wh;H ðxi ; xj Þ 2 kh ðxi  xj ÞqxHi x
n i;j n i6¼j
ð92:12Þ

where
Z
wh;H ðxi ; xj Þ ¼ kh ðu  xi Þkh ðu  xj Þqux
H du ð92:13Þ

The optimal bandwidth h varies according to different interval length H. We

suggest selecting an interval length that scales with the optimal bandwidth as
c1 h , The parameter c is a smoothing parameter for the variable bandwidth, it
regulates the interval length for local optimization. With small c, the variable
bandwidth fluctuates slightly, while with large c, the variable bandwidth fluctuates
significantly.
In order to select a variable kernel bandwidth, firstly, compute the local cost
function wh;H ðxi ; xj Þ in Eq. (92.13) and find h that minimize the Eq. (92.13). Then
repeat the procedure above while changing H. Change H to find H  that satisfies
H ¼ c1 h . We could obtain the variable bandwidth by computing the cost
function
Z
^ n ðcÞ ¼
b
2X
C ^p2n dx  kh ðxi  xj Þ ð92:14Þ
a n2 i6¼j c

P
where ^pn ¼ i khc ðx  xi Þ. At last, we should repeat the procedure above to find c
that minimizes C ^ n ðcÞ, then apply it to obtain the variable bandwidth h . The
c
bandwidth is what we want to calculate kernel density estimation more precisely.
It has been proved that range block mean R,  contrast scaling parameter s and
collage error e [4, 6, 9] are effectively used to retrieve images. In this paper, we
apply these parameters into kernel density estimation directly. The optimal uni-
form quantization for R  is f0; 1; 2; . . .; 63g [13] and to ensure the convergence of
the decoding, the scaling factor s is restricted to the interval ðSmax ; Smax Þ, where
0\Smax \1, as for collage error e, it is real-valued, hence, before we calculate
kernel density estimation of collage errors, they are rounded into the closest
integer if collage errors are smaller than T  1, or are set as T  1 if they exceed
T  1 (T is a user-specified threshold). In this paper, we set T ¼ 20. Then we
994 Z. Qin et al.

Fig. 92.1 Examples of 128 9 128 images. a–d four similar images; e–h four different images

calculate kernel density estimation with the processed collage errors and fractal
coding parameters.
Figure 92.1 shows four similar and four different texture images from VisTex
texture database [14]. Kernel density estimation of range block mean R, contrast
scaling parameter s and collage error e corresponding to these images are plotted
in Fig. 92.2. In most cases, the curves are close for similar texture images, and
different for the dissimilar texture images.
The left column of Fig. 92.2 shows the kernel density estimation of range block
mean R, contrast scaling parameter s and collage error e respectively according to
the first four similar images (a–d). Obviously the curves are close for the similar
texture images.
The right column of Fig. 92.2 shows the kernel density estimation of range
block mean R,  contrast scaling parameter s and collage error e respectively
according to the other four different images(e–h). We can see that the curves are
different for the dissimilar texture images.

92.3.2 Similarity Measurement

We define that VQ ðÞ ¼ fu1 ; u2 ; . . .; uH g and VC ðÞ ¼ fv1 ; v2 ; . . .; vH g are the

features of the query and candidate images respectively. The vectors of range
block mean R, contrast scaling parameter s and collage error e of a query image are
expressed as fuR1 ; uR2 ; . . .; uRH g, fuS1 ; uS2 ; . . .; uSH g and fue1 ; ue2 ; . . .; ueH g
respectively. The same with vector VC for candidate images.
In addition, the subscript variable H represents the amount of range block mean
 contrast scaling parameter s or collage error e of an image. To measure the
R,
92 An Effective Image Retrieval Method … 995

Fig. 92.2 Comparison of the KDE of fractal coding parameters

similarity between two images, we can calculate the deviation between their
fractal coding parameters instead since these parameters can express images
compactly.
In this paper, we adopt the most commonly used L2 and KLD (Kullback-Leibler
divergence) [15] as the distance criterion to measure the similarity between the
query and the candidate images, the distance between the two images is calculated
as follows:
996 Z. Qin et al.

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u H
uX
dL2 ðQ; CÞ ¼ t ðub  vb Þ2 ð92:15Þ
b¼1

X
H
ub
KLDðQ; CÞ ¼ ub logð Þ ð92:16Þ
b¼1
vb

The distances between similar images are much smaller than those between
dissimilar images. Experiments show that the similarity measurement Eq. (92.7)
performs much better than Eq. (92.8). Thus we only discuss the retrieval rate using
L2 distance. The obtained distances are sorted in an ascending order.

92.4 Performance Evaluation

We have performed experiments on VisTex texture database. The set of VisTex is

the classical selection of 40 classes of texture images that are used by many
literatures for image retrieval [16]. The images are listed as follows: Bark0, Bark6,
Bark8, Bark9, Brick1, Brick4, Brick5, Buildings9, Fabric0, Fabric4, Fabric7,
Fabric9, Fabric11, Fabric14, Fabric15, Fabric17, Fabric18, Flowers5, Food0,
Food5, Food8, Grass1, Leaves8, Leaves10, Leaves11, Leaves12, Leaves16,
Metal0, Metal2, Misc2, Sand0, Stone1, Stone4, Terrain10, Tile1, Tile4, Tile7,
Water5, Wood1, and Wood2. These are real world 512 9 512 images from dif-
ferent natural scenes. Only gray-scale levels of the images are used. As for our
experiments each image is divided into sixteen non-overlapping 128 9 128 sub-
images, thus creating a test database of 640 texture images.
In the retrieval experiments, each sub-image in the database is used once as a
query image. For comparison purpose, retrieved images are the first 16 most
similar images for each query. The relevant images for each query consist of all
the sub-images from the same original texture image. All experiments are con-
ducted on a 2 GHz PC using Matlab7.8.0 as a programming tool.
The retrieval method using feature vectors of range block mean, contrast
scaling and collage error are respectively named as KM (kernel density estimations
of range block mean), KS (kernel density estimations of contrast scaling) and KE
(kernel density estimations of collage error). A large number of texture images in
database are used to do testing, from which we could conclude that the KE method
works much better than the KM and KS. Therefore, we apply the KE method into
images retrieval in this paper.
92 An Effective Image Retrieval Method … 997

Table 92.1 Retrieval rate of three methods (%)

Image HE FKE VKE
Bark0 16.80 21.09 52.34
Bark6 14.84 11.72 37.50
Bark8 22.27 20.31 47.65
Bark9 21.48 22.66 46.87
Brick1 22.27 39.06 64.84
Brick4 34.77 30.86 83.98
Brick5 21.48 24.22 76.56
Buildings9 21.88 26.17 75.39
Fabric0 46.88 41.80 76.56
Fabric4 36.80 41.48 75.00
Fabric7 33.69 38.91 75.78
Fabric9 58.20 59.53 74.21
Fabric11 38.28 42.58 76.56
Fabric14 35.94 40.63 74.60
Fabric15 52.34 46.88 76.56
Fabric17 30.47 38.67 74.60
Fabric18 25.00 46.88 75.39
Flowers5 57.03 58.59 73.82
Food0 67.19 63.44 81.17
Food5 65.40 28.52 73.43
Food8 46.48 51.17 76.17
Grass1 32.03 42.50 73.82
Leaves8 36.72 64.06 76.17
Leaves10 23.83 27.34 75.78
Leaves11 48.44 57.81 75.00
Leaves12 42.58 57.81 76.56
Leaves16 50.47 32.03 71.87
Metal0 38.67 54.53 81.17
Metal2 34.77 48.67 76.09
Misc2 39.45 46.25 77.73
Sand0 37.11 35.16 73.04
Stone1 26.56 33.44 78.12
Stone4 31.25 33.98 75.00
Terrain10 37.97 22.66 77.34
(continued)
998 Z. Qin et al.

Table 92.1 (continued)

Image HE FKE VKE
Tile1 72.27 63.28 85.82
Tile4 69.92 74.22 88.04
Tile7 63.28 71.41 86.60
Water5 25.00 22.27 70.31
Wood1 22.66 34.27 75.39
Wood2 38.75 38.13 87.10

Table 92.2 Average Retrieval Rate (ARR) compared with other literatures (%)
Method HE FKE literature [3] literature [4] Proposed method
GGD + MM rmm + CT (VKE)
ARR 35.42 39.88 67.27 69.52 72.40
(%)

92.4.1 Average Retrieval Rate and Retrieval Speed

We use 40 512 9 512 VisTex texture images. Each image is divided into 16
128 9 128 non-overlapping sub-images. Finally a test database of Z ¼ 640 texture
images is created. Each sub-image is encoded using full search. Let the number of
ideally retrieved images of one class be denoted by F (in this case F ¼ 16) and mz
be the number of correctly retrieved images of one class from the top 16 images at
the z-th test. The performance is measured in Average Retrieval Rate (ARR) that is
defined the same with literature [6], which is then calculated as
PZ
z¼1 mz
ARR ¼ ð92:17Þ
FZ

Experiments show that KE (kernel density estimation of collage error) method

works better than the others. We can see from Table 92.1 and Table 92.2 that our
method performs better than HE (histogram estimation) method, FKE (fixed
bandwidth kernel density estimation of collage error) method and other methods.
From Table 92.2 we can see that the ARR of the proposed method is 72.40 %,
which is more that the other listed methods. The runtime of the retrieval, which is
completely determined by the performance of the similarity measurement process,
is also a key index to indicate the performance. Compared with literature [16], our
runtime is largely reduced since the basic arithmetic operations are adopted in our
method, while the computationally expensive log; ex and xr operations with more
iterations are applied in literature [16], which leads to an increase in computation
time.
92 An Effective Image Retrieval Method … 999

92.5 Conclusion

In this paper, we apply orthogonalization fractal coding algorithm into image

retrieval, which has been verified that the decoding speed is higher than that of the
basic fractal coding. Meanwhile, we propose an image retrieval method based on
fractal coding parameter with a variable optimized bandwidth kernel density
estimation method. The kernel bandwidth can adjust according to the data dis-
tribution. Thus, the statistical characteristics of fractal coding parameters are
employed as retrieval indices. Experiments show the superiority in both retrieval
rate and retrieval speed when compared with the existing methods. In the future,
we will combine some other features of images with fractal parameters to improve
performance of image retrieval.

References

1. Chun, Y.D., Kim, N.C., Jang, I.H.: Content-based image retrieval using multi-resolution
color and texture features. IEEE Trans. Multimedia 10, 1073–1084 (2008)
2. Barnsley, M., Sloan, A.D.: A better way to compress images. Byte 13, 215–233 (1988)
3. Jacquin A.E.: Fractal image coding: a review. Proc. IEEE 81(10), 1451–1465 (1993)
4. Pi M, Mandal M, Basu A.: Image retrieval based on histogram of new fractal parameters. In:
Proceedings of the ICASSP’, vol. 3, pp. 585–588 (2003)
5. Pi M., Mandal M., Basu A.: Image retrieval based on histogram of fractal parameters. IEEE
Trans. Multimed. 7(4), 597–605 (2005)
6. Pi, M., Li, H.: Fractal indexing with the joint statistical properties and its application in
texture image retrieval [J]. IET Image Proc. 2(4), 218–230 (2008)
7. Shimazaki, H., Shinomoto, S.: Kernel bandwidth optimization in spike rate estimation.
J. Comput. Neurosci. 29, 171–182 (2010)
8. Øien, G.E., LepsØy, S.: A class of fractal image coders with fast decoder convergence. Sign.
Proces. 40(1), 105–117 (1994)
9. Pi, M., Tong, C.S., Basu, A.: Improving fractal codes based image retrieval using histogram
of collage errors. In: 2nd International Conference on Image and Video Retrieval, vol. 2728,
pp. 121–130 (2003)
10. Xiaoqing, H., Qin, Z., Wenbo, L.: A new method for image retrieval based on analyzing
fractal coding characters. J. Vis. Commun. Image Represent. 24(1), 42–47 (2013)
11. Zhi, L., Ran, S., Liquan, S.: Unsupervised salient object segmentation based on kernel density
estimation and two-phase graph cut. IEEE Trans. Multimedia 14(4), 1275–1289 (2012)
12. Deng Biao, Yu Chuanqiang, Li Tianshi, Su Wenbin, Pan Yangke.: Dual-bandwidth kernel
density estimation algorithm based on estimate points. Chinese Journal of Scientific
Instrument, vol.32(3), pp.615-620 (2011)
13. Tong, C.S., Pi, M.: Fast fractal image compression using adaptive search. IEEE Trans. Image
Process. 10, 1269–1277 (2001)
14. VisTex texture database. http://vismod.media.mit.edu/vismod/imagery/VisionTexture/
15. Do, M.N., Vetterli, M.: Wavelet-based texture retrieval using generalized Gaussian density
and Kullback-Leibler distance. IEEE Trans. Image Process. 11(2), 146–158 (2002)
16. Kwitt, R., Uhl, A.: Lightweight probabilistic texture retrieval. IEEE Trans. Image Process.
19(1), 241–253 (2010)
Chapter 93
Bio Terapi Solat: 3D Integration in Solat
Technique for Therapeutic Means

Arifah Fasha Rosmani, Noor Azura Zainuddin, Siti Zulaiha Ahmad

and Siti Zubaida Ramli

Abstract Solat has been known for its health benefits, however not many people
are aware of it. Solat involves stretching that is in harmony with the body needs for
health care. Therefore, this research focuses on solat guidance in three dimensional
(3D) ways. The purpose of this research is to show the complete solat techniques
as a therapy to the human body and health using 3D model. This application is
developed using 3D Autodesk Maya software that includes the human model,
movement (rigging) and other multimedia elements that can create interactivity in
this application. The application shows a guideline of learning a perfect Solat
Technique and the benefit of solat in the therapeutic means. Usability test is
performed to observe the usability and learnability aspects of the application.
Respondents were selected among adults to experience this application. It even-
tually guides the users to focus on solat therapy techniques and its benefits.

93.1 Introduction

There are two types of solat: solat fardhu and solat sunat, solat fardhu is the second
Rukun Islam that is compulsory to all Muslims. Solat is mandatory and it must be
performed five times every day. Solat (dhuhr, asr, maghrib, isha’, and fajr) is
important to each Muslim because through solat, Muslims can get closer to Allah
S.W.T and avoid doing bad things. In the Al-Quran it is also stated that solat is
mandatory to be performed as follows:

A.F. Rosmani (&)
S.Z. Ahmad
S.Z. Ramli

Faculty of Computer and Mathematical Sciences, Universiti Teknologi MARA, 02600 Arau,
Perlis, Malaysia
e-mail: [email protected]
N.A. Zainuddin
Academy of Contemporary Islamic Studies, Universiti Teknologi MARA, 02600 Arau,
Perlis, Malaysia

 Springer International Publishing Switzerland 2015 1001

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_93
1002 A.F. Rosmani et al.

Which means: Maintain with care the [obligatory] prayer and [in particular]
the middle solat and stand before Allah, devoutly obedient. (Al-Baqarah, 238)
The benefits from solat can make it an option for medical therapies for those
who perform it. Unfortunately, many people do not realize that solat can be a good
health treatment and they do not know that the solat technique can be a therapy to
relax their mind [1].
Solat has many techniques and since Islam is simple, it does not burden Muslims
in performing the solat. There are many techniques for solat in this world according
to different mazhabs and manners of people. However, people do not know which
one is a better technique for them to follow. As a Muslim, solat has to be carried out
in the manner of solat advocated by the prophet (Nabi Muhammad S.A.W) [2].
Most of the guides for learning the solat technique for solat therapy are
available in written form. The materials, such as books, make the readers bored
and they are not interested to learn solat. An effective way to make it more
interesting and engaging is by illustrating Solat Technique using 3D animation. 3D
model of animation is intended to connect users with the application. There are
many 3D softwares available in the market. However, 3D Autodesk Maya has
been chosen based on the extensive creative features such as 3D computer ani-
mation, character modeling, visual effects, rendering, and compositing process.
Therefore, this study intends to develop an application to show the solat
technique in 3D animation because the existing application uses limited materials
for learning solat such as books, video-based learning, and 2D models. As the 3D
is very popular and interesting in the real world environment, this project utilizes
the technology to develop 3D animation as an alternative to learn solat and
becomes a technique for therapy.

93.2 Research Background

This project provides an alternative way to learn the solat technique for therapeutic
means for all Muslims. It focuses on the 3D animation as a way of learning Solat
Technique. This project is suitable for all users but the main targets of users are
children from 7 years old to teenagers around 18 years old. The Prophet (Nabi
Muhammad S.A.W) said: Order your children to pray at the age of seven. And
beat them (lightly) if they do not do so by the age of ten. From the hadith it tells
that children should start practicing solat from the age of 7. Hence, the develop-
ment of this project enables children to interact and understand how to learn solat.
93 Bio Terapi Solat: 3D Integration in Solat Technique … 1003

Children can learn the correct way or the technique of solat as the correct way of
Solat Technique is good for health.
Besides that, this project can also be used by Muslims to introduce the benefit
from solat technique to the other Muslims who are not aware of it. The devel-
opment of this project is also to illustrate the right movement and the technique of
breathing during solat.

93.2.1 Solat

Solat in the Arabic language is a worshiping act with physical and body move-
ments as well as a silent Quranic recitation though mind and soul. This Muslim
way of worshiping also involves some quranic recitations and as well as suppli-
cations (du’a). These specific recitations and supplications must be verbalized
when the worshipper assumes certain positions and performs movements between
positions [3].
The Messenger of Allah S.W.T, Rasulullah S.A.W, says: There are five solat
that Allah obligated the slaves to perform. Whoever performs them properly
without belittling their obligation, Allah promised to admit him into Paradise.
Whoever leaves them out does not have a promise from Allah to have Paradise
without torture before. If He willed, He tortures him, and if He willed, He forgives
him. (Narrated by Ahmad in his Musnad.)
It is obligatory to perform each of these five solat in its due time. It is better to
perform each solat earlier in its due time. Solat is also physical as well as a
spiritual act which involves total obedience and submission to Allah S.W.T.

93.2.2 Solat Techniques

Each solat has continuous sequences of body movements consisting of bowing,

kneeling, prostrating, and sitting. When performing solat, most of the muscles and
joints are in action. All the movements when performing the solat fardhu 17
raka’at per day, is equal to 119 physical postures per day, 3570 per month, and
43,435 per year. The movements increase went performing the other optional
solat. Through the solat, unique postures are possible which can give significant
reactions to the human body.

93.2.3 Solat as a Therapy

Solat helps to remove all sources of tension due to the constant change of
movements. It is known that such changes lead to an important physiological
1004 A.F. Rosmani et al.

relaxation. Therefore, the Prophet commands that the Muslims, when afflicted with
a state of anger, should resort to prayers. It is proved that prayer has an immediate
effect on the nervous system as it calms agitation and maintains balance. They are
as well a successful treatment for insomnia resulting from nervous dysfunction [4].
In order to visualize the implementation of therapeutic benefit in Solat, 3D
model with animation should be integrated to increase the learnability and
awareness among Muslim.

93.2.4 Dimensional (3D) Model

Three Dimensional is one of a techniques to create or build interactive multimedia

and animation film or cartoon [5]. Multimedia is more interesting when 3D ani-
mation is used rather than 2D animation. According to Yongguang et al. [6], 3D
describes an image that provides the view in depth. When 3D images are made
interactive, the users will feel involved with the scene. Nowadays, 3D has already
developed in the film industry and cartoon production and now is growing in the
field of education, game, and also in advertisement.

93.3 Related Works

A few observations have been done on the previous available solat applications to
find the advantages and disadvantages of each application before developing our
own application. Figure 93.1 shows solatSim application which is a review of each
position in Zuhur obligatory Solat. It also contains male and female respondent’s
image illustration solat position. When the respondents do the technique of solat,
SolatSim also view the blood circulation flow and the muscle condition for each
position of solat [4]. In SolatSim it include five (5) main topics which are ‘‘Male’’,
‘‘Female’’, ‘‘Research’’, ‘‘Credit’’ and ‘‘Solat Info & Demonstration’’.
‘‘The Right Way to Pray’’ is another application that is related with this
research as shown in Fig. 93.2. It includes all the techniques of solat that can guide
users to learn solat and includes information about Islam. This application is
developed by Islamic finder at http://www.islamicfinder.org/prayer/index.html.
This application has provided a user friendly interface to users. It is also interactive
and users can control the page. It includes buttons such as ‘‘back’’, ‘‘repeat’’ and
‘‘next’’. It can be repeated by user if they did not understand on certain topics.
However, the disadvantage of this application is the model is not clear. The
developer used a blur model to describe each of the movements of solat. Users
might not see clearly the right position of solat. Moreover, it also does not include
the Arabic writing of the positions of solat.
93 Bio Terapi Solat: 3D Integration in Solat Technique … 1005

Fig. 93.1 SolatSim interface

Fig. 93.2 The right way to pray interface

93.4 Methodology

In this research, three phases were involved which are Designing, Developing, and
Testing. The particular activities of all phases are briefly explained as below.

93.4.1 Design Site Map

A site map provides the flow of content of all processes that are included in the
application. Figure 93.3 shows the site map of this project. The site map shows the
contents of the application. This application has 3 menus which are ‘‘Sunnah
Rasulullah S.A.W’’, ‘‘Teknik Solat’’ and ‘‘Glosari’’. ‘‘Teknik Solat’’ is the main
menu of this application. In the site map, each page contains a ‘‘KELUAR’’ button
to ease users to exit this application. When a user clicks the ‘‘KELUAR’’ button, a
message box will be appeared to ask users for confirmation to exit the application.
Besides that, this application includes maximizing and minimize button to ease its
usage.
1006 A.F. Rosmani et al.

Fig. 93.3 Site map of ‘‘Bio Terapi Solat’’ application

93.4.2 Design Storyboard

The storyboard is designed to show the flow of the application. Usually it is

sketched on a piece of paper before designing. It is easier to design the interface
and get an idea when it is sketched on paper. When the storyboard has been
completed, the sketch of the interface will be scanned using a scanner to make it in
a digital form.

93.4.3 Development

The storyboard made will be used as a guide to draw the button and the interface.
The button plays an important role to enable users to navigate in an application.
Each element in the application must match the interface. Color and appropriate
font is made to attract users to use this application. The home icon is used to
identify the ‘‘Laman utama’’ application ‘‘Bio Terapi Solat.’’ The book icon is
used to identify a button of ‘‘Sunnah Rasulullah S.A.W’’. ‘‘Solat Teknik’’ icon that
shows a picture of the bow position is used to show the button for ‘‘Solat Teknik’’.
While the icon ‘‘Glosari’’ is used to visualize the search to find a piece of infor-
mation. Figure 93.4 shows the button in the application using icon.
93 Bio Terapi Solat: 3D Integration in Solat Technique … 1007

Fig. 93.4 Button for ‘‘Bio Terapi Solat’’ application

Fig. 93.5 Interface of

‘‘Laman Utama’’

93.4.3.1 Main Menu

The home page is the first interface for this application. On the homepage, there is
information about this application. It also has 4 buttons including ‘‘Laman Utama’’
(Main Menu), ‘‘Sunnah Rasulullah S.A.W’’, ‘‘Teknik Solat’’ (Solat Technique)
and ‘‘Glosari’’ (Glossary). In addition, there are buttons to exit, minimize and
maximize the application. Figure 93.5 shows the interface of ‘‘Laman Utama’’.

93.4.3.2 Sunnah Rasulullah Page

On this page, there are six buttons of ‘‘Sunnah Rasulullah S.A.W’’, users can click
on each button to access all the available ‘‘Sunnah’’. This application comes with
‘‘Sunnah Rasulullah S.A.W’’ to provide additional information to the users. When
a user clicks the button, the information will be displayed on the ‘‘Sunnah’’.
Figure 93.6 shows an example interface for ‘‘Sunnah’’ information.

93.4.3.3 Terapi Solat Page

Figure 93.7 is the interface to see prayer techniques in 3D animation. Users can
select the desired prayer technique. There are nine techniques provided as
‘‘Takbir’’, ‘‘Qiam’’, ‘‘Tomaninah’’, ‘‘rukuk’’, ‘‘sujud’’, ‘‘duduk di antara dua
1008 A.F. Rosmani et al.

Fig. 93.6 ‘‘Sunnah Rasulullah’’ page

Fig. 93.7 Terapi solat interface

sujud’’, ‘‘Tahiyad Awal’’, ‘‘Tahiyad Akhir’’ and ‘‘Salam’’. For every single
movement in ‘‘Solat’’, there are many beneficial elements that can provide the
therapeutic means for Muslims that have been emphasized in the application. For
example as shown in Fig. 93.7, the benefits for solat movements have been
included in ‘‘Terapi’’ and ‘‘Teknik’’ buttons where the correct way of performing
solat and its benefits are highlighted.

93.4.3.4 Glossary Page

The glossary is a page that allows the user to access information contained in this
multimedia application. It contains information such as a dictionary and only
contains jargons such as medical terms that may not be understood by the users.
Users can enter a word in the textbox and if the requested information is in the
database then it will be displayed as shown in Fig. 93.8.

93.5 Usability Testing

After the completion of Development Phase, usability test is carried out to get the
feedback on flaws of the application. Generally, usability testing is the method
used to test prototypes and find out the ease of use level. The participants in
93 Bio Terapi Solat: 3D Integration in Solat Technique … 1009

Fig. 93.8 Glossary page

Usability Test are real users. From the testing, the developer reviews the attitude of
the users towards the project. Participants are required to complete the designed
tasks under simulation scenarios—usually in the laboratory. It can help to find out
the problems and modify the design in the early stages of product design [7].
The usability testing was done by the lecturers of UiTM, Perlis. The application
has been tested on the lecturers in Center of Islamic Teachings and Understanding
(CITU) and Computer Science Department. Besides that, participants of usability
testing are the students of UiTM, Perlis. This usability testing had been done by
determining the navigation, design, sound, graphic, animation, and interactivity
between the user and the application. Participants were asked to perform a ques-
tionnaire after exploring the application; they were also required to complete the
testing. After exploring applications, developer’s has distributed a set of ques-
tionnaires to the participants. They were given 5–10 min to complete the relevant
questionnaire.

93.5.1 Usability Test Result

From the observation, it is found that the participants were confused with the
Glossary interface. They did not know what terms that are provided in the Glos-
sary and it does not provide any direction or instruction to the participants. The
‘‘Terapi’’ and ‘‘Teknik’’ buttons also made the participant uncomfortable with
them. The buttons required the user to hold the mouse click to enable them to read
the information. However, these drawbacks have been refined and suitable
enhancement has been implemented.
Based on the usability test result as shown in Table 93.1, the user is satisfied
with the application. From the observation, the participants obtained information
regarding Solat as a therapy. They are able to use the application as their guidance
to perform solat. Additionally, the therapeutic benefits and the integration of 3D
model have enhanced their learnability.
1010 A.F. Rosmani et al.

Table 93.1 Usability result

Description Feedback
Task 1 Easy to click on the button Button Most evaluator satisfied with the use of a
Task 2 Understand the usage of button. The icon of a button makes it interesting to
each button provided the participant
Task 3 Easily navigates from one Navigation The evaluator finds the menu navigation
screen to another screen easily
Task 4 Interface suitable to the Interface The interface includes interactivity with
application the user
Task 4 Text clear and readable Text Most participants agree the font suitable to the
Task 5 Suitable text color application. It’s easy to read and suitable with the
background
Task 6 Easy to understand
Task 7 Sound could assist in the Sound Sound is redundant when user clicks the
learning process button (sound button in ‘‘solat teknik’’)
Task 8 Sound is clear
Task 9 Suitable with application Color Satisfied with the color
Task 10 3D model useful in learning 3D The participants agree 3D model could easily to
process be understood

93.6 Conclusion and Future Works

Bio Terapi Solat can be used as a guideline for Muslim or others for learning and
gaining knowledge about the benefits of solat. Many people did not realize solat
can be utilized as a therapy and is beneficial for us. Therefore, this application will
guide and show users how to perform the correct way of solat. The application also
provides information on techniques and highlights the therapeutic benefits that can
be achieved from the prayer technique that has been done.
In addition, the 3D model performs solat in 360 view. Each technique of solat,
including the therapy is displayed in this application. This application differs with
other applications because it includes the additional information such as therapy of
solat, Sunnah Rasulullah S.A.W and glossary.
As a conclusion, the objective of this project has been achieved. The suitable
multimedia element has been developed as interactive multimedia for guideline
and learning of solat as a therapy. The model integrates with a 3D animation
human model for making it interesting and more understandable. This application
can be used by Muslims as a guide for them on the right technique to perform
solat. The therapy is a good treatment to us if performed correctly during solat.
This application which provides information about the prayer therapy could guide
and educate users while using this application.
Furthermore, the evaluators recommend a 3D model over virtual reality modes.
Users can control the 3D model where they can select the different part or angle of
the model.
93 Bio Terapi Solat: 3D Integration in Solat Technique … 1011

References

1. Doufesh, H., Faisal, T., et al.: EEG spectral analysis on muslim prayers. Appl. Psychophysiol.
Biofeedback 37(1), 11–18 (2012)
2. Rahman, U.Z.A.: Formula Solat Sempurna. Sri Damansara, Kuala Lumpur, Telaga Biru Sdn.
Bhd (2008)
3. Salleh, N.A., Lim, K.S., et al.: AR modeling as EEG spectral analysis on prostration. In: 2009
International Conference for Technical Postgraduates (TECHPOS) (2009)
4. Aziz, N.A.A., Samsudin, S.: Computerized simulation development for blood circulation and
bodily movement during obligatory prayers (SolatSim). In: Second World Congress on
Software Engineering (WCSE) (2010)
5. Baran, I., Popovi, J., et al.: Automatic rigging and animation of 3D characters. ACM Trans.
Graph. 26(3), 72 (2007)
6. Yongguang, L., Mingquan, Z., et al.: Using depth image in 3D model retrieval system. Adv.
Mater. Res. (Comput. Mater. Sci.) 268–270, 981–987 (2011)
7. Wang, H., Yan, B.: A data-processing mechanism for scenario-based usability testing. In: 2011
IEEE 2nd International Conference on Computing, Control and Industrial Engineering (CCIE)
(2011)
Chapter 94
Enhanced Interactive Mathematical
Learning Courseware Using Mental
Arithmetic for Preschool Children

Siti Zulaiha Ahmad, Noor Asmaliyana Ahmad,

Arifah Fasha Rosmani, Umi Hanim Mazlan
and Mohammad Hafiz Ismail

Abstract Interactive Mathematical Learning Courseware 2.0 (iMLc2.0) is an

enhanced version of multimedia application which aims to expose the pre-school
children in mental arithmetic technique with; (i) larger range of numbers, (ii)
implementation of standard written method and, (iii) video support to visualize
both techniques. The application integrates all multimedia elements and mental
arithmetic techniques in interactive and supportive ways which are suitable for
pre-school children. It is divided into three modules, addition, subtraction, and
quizzes. The addition operation has been designed to apply the finger-brain
approach which is a part of mental arithmetic technique while, mental-imagery
approach is adapted for subtraction operation. Experimental test and user accep-
tance test have been conducted to evaluate the application. The implementation of
iMLC2.0 in enriched interactive multimedia environment can be used as additional
teaching and learning tool since it is supportive and attractive. In addition, it can
motivate the preschooler to be prepared for future mathematical learning.

94.1 Introduction

The process of learning mathematics, such as numbers and basic operations like
addition and subtraction, takes a long time to build for new learners, especially for
preschoolers. Some of them found it hard to memorize numbers and mathematical
concepts even though they already knew and understood them. Technology and
computer are main aspirations and they are now integrated in education curricu-
lums [1].

S.Z. Ahmad (&)
N.A. Ahmad
A.F. Rosmani
U.H. Mazlan
M.H. Ismail
Faculty of Computer and Mathematical Sciences,
Universiti Teknologi MARA Perlis, Perlis, Malaysia
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 1013

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_94
1014 S.Z. Ahmad et al.

With the evolution of computer technology, learning of mathematics for

pre-school children can be improved. Generally, childrens’ attention and interest
can be developed if the learning process is assisted by multimedia tools.
Furthermore, children nowadays are already exposed to computer at an early age.
Multimedia is an interactive computer-based environment with the combination
of texts, voices, pictures and animations [2]. Thus the teachers can teach more
consistently and effectively by using it as an alternative teaching tool. It is dif-
ferent from the traditional method, where only the exercise book is used as their
single source in learning mathematics due to lack of tools. This results a boring
environment and causing children to lose their focus.
Mental Arithmetic Technique can help children to build cognitive thinking
when they need to use mental representations and fingers in order to do basic
operation exercises. With the support of multimedia, their imagination of the
mathematical concepts can be at enhanced. This improves their performance and
allows them to be more active and excited due to the utilization of multimedia
combination elements during learning sessions.
Most children in traditional method environment use only exercise books as
their source of information. It is hard to develop their understanding and imagi-
nation using mental arithmetic if this method is used as a single approach in the
classroom. Without an alternative learning tool, children will easily get bored and
lose their interest. As a result, their focus on the mental representations of
mathematical concepts decreases. This may affect their mathematical skills per-
formance in the future, especially when they are in primary school. Therefore, the
purpose of this study is to improve the design of previous iMLc by introducing
mental arithmetic with standard written method using additional interactive ele-
ments for pre-school children.

94.2 Literature Review

94.2.1 Mental Arithmetic Technique

Mental arithmetic can be defined as the action of adding numbers together, mul-
tiplying them and other mathematical operations by using the brain, without the
involvement of writing or the use of calculators [3]. This technique is usually used
in mathematical operations such as simple addition, subtraction, multiplication,
and division.
Mental arithmetic is also one of the methods that use the movement of fingers for
counting numerical values [4, 5] as well as the working brain. A previous study
proved that children move their real fingers to start the counting process after they
build their imagination in the brain [6]. In addition, a study on introductory of mental
arithmetic using simple application has shown a significant improvement in basic
mathematic addition operation [7]. Therefore in this study, we would like to apply
mental arithmetic with a large range of numbers using standard written method.
94 Enhanced Interactive Mathematical Learning Courseware Using … 1015

94.2.2 Children’s Cognitive Thinking

Every child develops their learning preference along mental process simulta-
neously in day-to-day activities. Cognitive or mental processes are the processes of
recognizing, understanding, and learning something [3]. The evolution of tech-
nology become as a catalyst towards development of children’s cognitive thinking
and skill to enable them progresses well in learning process.
Furthermore, cognitive skills in children can be sharpened through interaction
and communication with computers [8]. Based on Piaget’s Theory, children
around five to eight years old are able to obtain the skills regarding objects, events,
people, and use the symbols to imagine and represent real life and some examples
are the symbols of words, numbers, and images. Therefore, children can establish
their cognitive thinking and build mental representations when learning sessions
are supported with multimedia elements such as sound, audio, video, images,
graphics, and animations.

94.2.3 Multimedia

Multimedia is defined as the interactive computer-base, which is included with the

texts, images, audio, video, and animations [2, 7, 9, 10] The use of multimedia is
able to attract the children’s interest and attention towards learning mathematics.

94.2.3.1 Elements in Multimedia

An interactive multimedia is composed of many elements that are important to

deliver attractive prototype to children’s learning. Animation is primarily used to
illustrate the ideas and concepts. At an early age, children usually like to learn
from the moving images because of their characters, which are attractive and
motivating [11]. Animation is essential to improve understanding and brings out
the interests of young learners [11, 12]. Next, video is an alternative element that
could provide better visualization on specific activity. In any instructional
courseware we need to match learning preferences with different types of multi-
media instruction in order to provide easy-manage learning environment [13].
Images or graphics are considered to be part of multimedia where creativity and
imagination are needed in a learning session. Basically, the use of graphics can
help to sharpen memory [11] and stimulate mental model representations [11, 14].
The audio or sound also assists the children to give optimum attention and
provides a way to learn mathematical concepts or terms from the oral speech. The
implementation of sound can improve learning and pronunciation skills after lis-
tening to the spoken explanation [14]. Besides that, text is one of the important
element in providing interaction and information as well as an effective way to
communicate [10].
1016 S.Z. Ahmad et al.

Fig. 94.1 Research model

94.3 Methodology

The study was conducted by applying research model as depicted in Fig. 94.1. The
activities have been subdivided into three main phases:- Analysis, Design &
Construction and Testing & Evaluation.

94.3.1 Analysis Phase

The first phase was to collect information on three areas of study:- preschoolers’
preferences, multimedia elements principles and mental arithmetic. Interview and
observation technique were applied in order to gain insight of the target users and
the classroom environment. Teachers from two pre-schools in Kedah were selected
in the interview session and preschoolers have been observed during mathematics
classes. The observation was focused on the implementation of mathematic class,
children behavior and their reaction towards learning process.

94.3.2 Design and Construction Phase

The second phase consists of two main activities: design and construction. Design
activity involved storyboard design process by sketching interface design concept
on a paper. It is based on the information gathered from the previous stage. The
following process was designing user interface that incorporated multimedia ele-
ments such as graphics, sounds, texts, animations and videos using Adobe Flash,
Adobe Photoshop and Adobe Illustrator software. Those elements were specifically
chosen to stimulate the children’s learning experience and to entertain them. Next,
the design was translated into multimedia application in a construction activity.
94 Enhanced Interactive Mathematical Learning Courseware Using … 1017

Scripting and integrating the designed user interface were the main tasks of con-
struction activity using Adobe Flash, Audacity and AVS Video Editor. iMLc2.0
was delivered as a multimedia application prototype.

94.3.3 Testing and Evaluation Phase

The final phase involved a series of testing and evaluation processes. The proto-
type was tested using usability testing and some refinements on the prototype were
done based on the users’ comments. Next, experimental testing was conducted to
measure the children’s performance which consist of Pre-Test and Post-Test
evaluation. In Pre-Test session, the children were tested with mathematic question
using normal learning process while during the Post-Test session, they had to
answer a different set of question by applying mental arithmetic learning process.
In order to assess courseware’s acceptability among the targeted audiences, an
acceptance test was conducted with kindergarten teachers. They were given a set
of acceptance test questionnaires consist of 13 Likert Scale Based questions. The
result was then analyzed to determine the acceptance rate of the application.

94.4 Courseware Architecture

The courseware consists of three main menus: (1) Number, (2) Operation, and (3)
Activity modules. The content has been enhanced in all aspects including the
design, the interaction approach and the range of number. Figure 94.2 shows the
architecture of Interactive Mathematical Courseware application.

Fig. 94.2 The architecture of

interactive mathematical
courseware application Applicatio Main
Sub Menu
n Menu
Number
Number Introduction

Addition
Interactive Operations
Mathematical Subtraction
Courseware
Matching
Activities Numbers

Quiz Addition

Quiz Subtraction
1018 S.Z. Ahmad et al.

Fig. 94.3 Number

introduction interface

94.4.1 Number

Number menu consists of Number Introduction sub menu as shown in Fig. 94.3.
The children can learn starting from number one until number nine. It is also
provided with the animation object and spelling that represents the numbers. The
interface design includes graphic that familiar to children environment in order to
attract their attention.

94.4.2 Operation

There are two sub menu involved in the operation menu of the application. They
are addition and subtraction operation. Figures 94.4a, b show the addition interface
which applied mental arithmetic technique, provided with the video for addition
process. In order to introduce larger range of number, we include an example as a
tutorial shown in Fig. 94.4c which is blended with standard written method. In
Fig. 94.5, the mental-imagery of subtraction shows the subtraction operation
interface. It implemented object animation with narration to visualize the sub-
straction process in three different range of number examples.

94.4.3 Activity

Activity menu consists of three different sub menus which are matching numbers,
addition quizzes and subtraction quizzes. The concepts that were applied in the
activity menu are drag and drop, perfect choice and writing concept. One of the
94 Enhanced Interactive Mathematical Learning Courseware Using … 1019

Fig. 94.4 a Addition operation interface with video assistance. b Addition operation interface.
c Addition operation interface in larger range of number to adapt mental arithmetic technique
using standard written method

Fig. 94.5 Subtraction

operation interface

activities is shown in Fig. 94.6, number matching activity, which uses drag and
drop concept. Figure 94.7 shows other activity to strengthen mental arithmetic
skill in order to solve addition operation.
1020 S.Z. Ahmad et al.

Fig. 94.6 Drag and drop

interface: number matching

Fig. 94.7 Drag and drop

interface: reinforcement of
mental arithmetic skill

94.5 Finding and Discussion

Experimental test and User Acceptance test were conducted during testing phase
of the application. Experimental test was done with preschoolers at two schools in
Kedah to measure their understanding level towards mental arithmetic technique
using iMLc2.0 by carrying out pre-test (conventional method) and post-test
(mental arithmetic technique). The level of understanding for both schools were
represented by the mean score of pre-test and post-test as shown in Fig. 94.8. It
shows that children are capable to achieve higher scores after being exposed with
the mental arithmetic technique. Besides that, as depicted in Fig. 94.9, the children
took more than 15 min to solve basic mathematic questions during pre-test session
by using conventional method such as stick bar counting or finger counting.
94 Enhanced Interactive Mathematical Learning Courseware Using … 1021

Fig. 94.8 Comparison of Fundamental test for the understanding

understanding level for pre- level
10
test and post-test
9.5

Mean Score
9 Pre-Test
8.5
Post-Test
8
7.5
7
SK Tanah SK Haji Abdul
Merah, Rahman
Pendang Tokai, Alor
Star
Schools

Fig. 94.9 Comparison of

time taken using two different
methods

However, in the post-test session, average time taken for the children to solve the
mathematical problem using mental arithmetic obviously improved as they only
took less than 10 min to complete all questions. Time improvement and higher
score in post-test session proved that participants can think faster and accurate as
they can simplify the calculation for the given questions.
User Acceptance Test was conducted with preschool teachers in order to
determine the confidence level of end users of the application and to ascertain
whether it can be accepted as a teaching tool in preschool. They were given time to
explore the application before answering a set of questionnaires that covers four
criteria. They are Perceived Ease Use (PEU), Perceived Usefulness (PU), Attitude
(ATT), and Intention to Use (BI). Mean score for all criteria were calculated and
plotted into a bar chart as shown in Fig. 94.10. The highest total mean score is BI,
which is 5.00 where all participants strongly agreed to adopt the application in
future as a teaching aid. The total mean score for PEU is 4.63 where most of the
participants agreed that the application was easy to use (user friendly) and
learnable. The total mean for PU is 4.71 which reveals that the application is useful
1022 S.Z. Ahmad et al.

Fig. 94.10 Bar chart of Criteria of User Acceptance

criteria for user acceptance 5.00

Total mean score

test 4.00 PEU

3.00 PU

2.00 ATT
1.00 BI
0.00
PEU PU ATT BI
Criteria

to attract children’s interest and attention. The total mean score for ATT is 4.67,
which means that users are more likely to use the application because it uses
multimedia elements.

94.6 Conclusion

The main contribution in this research is more towards enhancing the interaction
design and problem solving technique of the previous application [7] which are the
utilization of multimedia elements (video and text animation) and content
enrichment (larger range of number and standard written method). Based on this
study, it can be summarized that the iMLC2.0 application was successfully
accepted to be used in teaching and learning mathematics in preschool.
The inclusion of video element enable the children to increase understanding in
mental arithmetic technique as they can visualize it better compared to the utili-
zation of animations, text and sound alone. The result proved that learning
mathematics through multimedia application can increase the fun and interest
during the learning session. Yet, it could improve their mathematical skill in
addition and subtraction operations.
The application achieves its target to be a helpful tool in improving users’
performance and sharpening their cognitive thinking. In addition, this technique
could be implemented in different school levels: pre-school, primary school (lower
and higher level) and secondary school. However, the approach might be different
for each level.
In conclusion, this application achieves the objectives of this research, where an
Interactive Mathematical Learning Courseware was designed and developed by
adapting suitable multimedia elements such as text, graphics, animation, sound,
and video. The implementation of mental arithmetic technique and standard
written method were successfully introduced in the application. Our next research
paper will be focusing specifically on instructional design diversity and its
effectiveness in implementing mobile iMLc.
94 Enhanced Interactive Mathematical Learning Courseware Using … 1023

References

1. Ktoridou, D., Etekleous, N., Gregoriou, G.: Preschoolers Developing Mathematical

Understanding through Computer-Based Activities. In: The International Conference on
Computer as a Tool, pp. 787–790 (2005)
2. Weiss, I., Kramarski, B., Talis, S.: Effects of multimedia environments on kindergarten
children’s mathematical achievements and style of learning. Educ. Media Int. 43(1), 3–17
(2006)
3. Longman: Longman dictionary of contemporary english online (Pearson Education Limited).
http://www.ldoceonline.com/. Accessed October 2011
4. Wu, S.S., Meyer, M.L., Maeda, U., Salimpoor, V., Tomiyama, S., Geary, D.C.: Standardized
assessment of strategy use and working memory in early mental arithmetic performance.
Dev. Neuropsychol. 33(3), 365–393 (2008)
5. Klein, E., Moeller, K., Willmes, K., Nuerk, H., Domahs, F.: The influence of implicit hand-
based representations on mental arithmetic. Frontiers Psychol. Cogn. 2, 1–7 (2011)
6. Cheah, B.L., Ong, S.L.: Perbandingan Kebolehan Menyelesaikan Masalah Matematik Antara
Murid Yang Belajar Abakus-Aritmetik Mental Dengan Murid Yang Tidak Belajar Abakus-
Aritmetik Mental. Jurnal Pendidik dan Pendidikan. Universiti Sains Malaysia, vol. 21,
pp. 85–100 (2006)
7. Ahmad, S.Z., Rosmani, A.F., Ismail, M.H., Suraya, M.S.: An introductory of mental
arithmetic using interactive multimedia. Comput. Inform. Sci. 3(4), 72–79 (2010)
8. Gelderblom, H., Kotze, P.: Designing technology for young children: what we can learn from
theories of cognitive development. In: Proceeding of SAICSIT ’08 (South African Institute of
Computer Scientists and Information Technologists), pp. 66–75 (2008)
9. Shujuan, Z., Wei, S., Zheng, L., Qibo, H.: the principles of multimedia teaching design based
on cognitive load theory. In: 2010 2nd International Conference on Education Technology
and Computer (ICETC), pp. 110–112 (2010)
10. Segers, E., Verhoeven, L., Hulstijn, N.-H.: Cognitive processes in children’s multimedia text
learning. Appl. Cogn. Psychol. J. 22, 375–387 (2008)
11. Betrancourt, M., Chassot, A.: Making sense of animation: how do children explore
multimedia instruction? cognitive processes in children’s multimedia text learning. Appl.
Cogn. Psychol. 22, 375–387 (2008)
12. Fang, H., Li, X., Helge, H.: Research on Improving Learning Interests for Elementary
Students Based-on Mental Mode by Using Mathematics Animations. In: International
Conference on Application of Information and Communication Technologies, pp. 1–5 (2009)
13. Zaidel, M., Luo, X.: Effectiveness of multimedia elements in computer supported instruction:
analysis of personalization effects, student’s performance and cost. J. Coll. Teach. Learn.
7(2), 11–16 (2010)
14. Ali, B.B., Badioze H.Z.: Framework of adaptive multimedia mathematics courseware.
Proceeding of the 2nd IMT-GT Regional Conference on Mathematics, Statistics and
Applications (2006)
Chapter 95
Comparative Evaluation of Ensemble
Learning and Supervised Learning
in Android Malwares Using
Network-Based Analysis

Ali Feizollah, Nor Badrul Anuar, Rosli Salleh and Fairuz Amalina

Abstract With the prevalence of mobile devices, the security threats are growing
in number and seriousness. Among the mobile operating systems, Google’s
Android has been attacked more than others have. From April 2013 until June
2013, the number of malwares were doubled for the Android. In this paper, we
evaluate the mobile malwares detection using the ensemble learning and super-
vised learning. Furthermore, we compare the two learning approaches based on the
experimental results. We compared our experimental results with a similar work.
The network traffic generated by mobile malwares are analyzed. We use 600
malware samples from the MalGenome data sample to build the dataset. We use
two versions of random forest algorithm as our evaluating algorithm, ensemble
learning and supervised learning. The empirical results show that the ensemble
learning improves the detection of the Android malwares. The ensemble learning
achieved 99.6 % of true positive rate while the supervised learning attained
99.4 %.

Keywords Mobile malware 

Ensemble learning  Supervised learning 

Android Network-based analysis Malgenome

A. Feizollah (&)  N.B. Anuar  R. Salleh  F. Amalina

Security Research Group (SECReg), University of Malaya, Kuala Lumpur, Malaysia
e-mail: [email protected]
N.B. Anuar
e-mail: [email protected]
R. Salleh
e-mail: [email protected]
F. Amalina
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 1025

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_95
1026 A. Feizollah et al.

Fig. 95.1 Android Malware Growth [1]

95.1 Introduction

The heavy use of mobile devices is undeniable in today’s world. They have
facilitated life in many aspects. People are able to email, chat, facebook and do
mobile banking on the go from their mobile devices. However, there is a drawback
inherent in such conveniences. Since the mobile devices contain sensitive and
personal information, they are prone to attacks by hackers. Based on Symantec
report that was published in October 2013, in just 2 months, from April 2013 until
June 2013, the volume of Android malwares has doubled [1]. The Fig. 95.1 shows
the growth in the Android malwares between July 2011 and June 2013.
Among the mobile operating systems, the Google’s Android operating system
has been attacked more than other mobile operating systems since it is open source
[2]. F-secure corporation reported that Android operating system comprises of
79 % of malwares in 2012 [3]. It is estimated that the sales of Android based
mobile devices will grow by 12 % in 2014 compare to other operating systems [4].
Such growth in malwares and the sale of Android based mobile devices raise
concerns over the security of mobile devices.
In addition, the official Android market, known as Google Play, contains
malwares. The report from Symantec [1] shows the top ten application categories
with highest percentage of malwares. The Google Play is included in this report as
well. The Fig. 95.2 represents the application categories with percentage of
malwares.
We used MalGenome data sample as our dataset for this study. It is one of the
well-accepted data samples of Android malwares among the research community.
It is a collection of Android malwares collected by the University of North Car-
olina in the period of August 2010 and October 2011 [5]. They published the paper
95 Comparative Evaluation of Ensemble Learning and Supervised Learning … 1027

Fig. 95.2 Application Category with the percentage of malwares [1]

along with the data sample in 2012. It is a collection of 1260 malwares in 49

different malware families. In this paper, 600 malware samples were selected from
the MalGenome data sample.
The supervised learning is defined as methods applied on a labelled data set. It
entails two stages: training and testing. In the training stage, the algorithm tries to
find a pattern in data with the same label in order to group them. In the test stage,
the learned algorithm is used with the new set of data to examine how successful
the algorithm is.
The ensemble learning comprises of the individually trained algorithms. In the
testing stage, the prediction is the combination of trained algorithms. In other
words, the selected algorithms are trained separately but the decision on new data
is done together. There are two methods in ensemble learning: bagging and
boosting. In the bagging method, the individual algorithms are trained with the
random distribution of the training sets. The distribution of the training sets is
independent of the previous algorithm’s performance. The Fig. 95.3 shows a
sample of assigning training sets in the bagging algorithm [6].
The boosting method is similar to the bagging method, but the difference is that
the assignment of the training sets for each algorithm is based on the performance
of the previous algorithm. For instance, the Fig. 95.4 depicts a sample of boosting
method in which the set C is repeated in the second algorithm onward. The reason
is that the boosting method keeps track of the incorrectly classified data. When the
first algorithm incorrectly classifies the data in C, the boosting method repeats the
C set in the training set for second algorithm. The boosting method is proved to
1028 A. Feizollah et al.

Fig. 95.3 Training sets for

bagging method Original Training Set A,B,C,D,E

Training Set for 1st Algorithm B,C,A,E,D

Training Set for 2nd Algorithm D,C,E,B,A

Training Set for 3rd Algorithm E,D,A,D,B

Fig. 95.4 Training sets for

boosting method Original Training Set A,B,C,D,E

Training Set for 1st Algorithm B,C,A,E,D

Training Set for 2nd Algorithm D,C,E,C,A

Training Set for 3rd Algorithm C,D,C,D,C

produce better results than the bagging method [6]. In this paper, we have selected
the Adaboost algorithm, which uses the boosting method.
In analysis of the Android malwares, there are two methods: behavior-based
and network-based. The behavior-based analysis refers to the internal behavior of
the malwares in the mobile operating system such as system calls. Malwares
represent some specific behavior by which analyzing them makes it easier to detect
the malwares. The network-based analysis, on the other hand, focuses on the
malwares’ behavior in the network traffic. Every mobile application, including
malicious one, has to communicate through the network. The communication of
the malicious applications have some features that differentiate them from normal
network traffic [7].
Thus, there is a vital need in analyzing the malwares and developing effective
methods to confront the mobile malwares. This paper aims at evaluating the
ensemble learning and supervised learning in mobile malware detection using the
network-based analysis. It is imperative to study various types of algorithms in
order to determine the best one to confront the massive growth of the mobile
malwares.
This paper is organized as follows. Section 95.2 discusses some of the related
works done by other researchers. Section 95.3 presents backgrounds of the algo-
rithms used in this paper. The methodology and the flow of the proposed method is
described in the Sect. 95.4. The results of this work is presented in the Sect. 95.5
along with discussions. Section 95.6 discusses the future works that can be done.
95 Comparative Evaluation of Ensemble Learning and Supervised Learning … 1029

95.2 Related Works

95.2.1 Supervised Learning Versus Ensemble Learning

In a study [8], researchers applied ensemble learning to the generic algorithm and
feature clustering. They experimented with the ensemble learning. The results
showed an improvement in the detection.
Yang Li et al. [9] published a study in which they employed the ensemble
learning on the KDD CUP 99 dataset. The KDD dataset is the collection of the
DDoS attacks. They reported a significant increase in the detection rate via the
ensemble learning. Thus, the ensemble learning is proven to increase the effec-
tiveness of other algorithms. In this paper, we prove that the same effect of the
ensemble learning is applied to the Android malwares.
A study was published in 2008 [10] in which authors used the ensemble
learning and supervised learning in order to predict breast cancer survivability.
The results show that the ensemble learning was more successful than the
supervised learning.

95.2.2 Behavior-Based Versus Network-Based Analysis

As discussed in the introduction, there are two methods in analyzing the Android
malwares: behavior-based and network-based. The behavior-based analysis
involves analyzing the system calls inside the mobile operating system such as
read, write and open.
The Crowdroid [7] analyzed the system calls by collecting them as a log file and
processing them to discover anomalies. They tested the Crowdroid using self-
written malwares as well as real malwares. The results were as high as 100 %, but
the downside is that they used self-written malwares that are not as realistic as the
real world malwares.
However, the Android operating system is based on Linux kernel [11] and col-
lecting the system calls is a toil [7]. Whereas, in the network-based analysis, collecting
the network traffic is as simple as installing the tPacketCapture application [12].
Sue et al. [13] extracted nine features from the network traffic and used two
algorithms, decision tree and random forest, for the experiments. The features are
the average and standard deviation of the number of sent/received packets, the
average and standard deviation of the number of bytes sent/received and the
average TCP/IP session duration. They achieved 96.70 % of detection rate with
random forest.
Similarly, a study was done by Feizollah et al. [14] in which they analyzed the
network traffic of 100 malwares for malware detection. They chose three network
features namely, packet size, connection duration and number of GET/POST
parameters. The results were as high as 99.94 % of true positive rate. Thus, the
network-based approach was chosen for this paper.
1030 A. Feizollah et al.

95.3 Background of Algorithms

We have used adaboost algorithm along with the random forest for the ensemble
learning approach.

95.3.1 Adaboost

It is one the most successful and popular ensemble algorithms. It is a self-rated

algorithm meaning that it scores the reliability of its own prediction. It has a high
flexibility for combining with other methods such as supervised learning. It also
requires less knowledge about the data set in order to improve the accuracy of the
algorithm [15]. In this work, we employed the Adaboost.M1 [16] in our experiment.
As discussed in the introduction section, the boosting algorithm works by assigning
the weight on the training sets. Suppose training set ðx1 ; y1 Þ; . . .. . .; ðxn ; yn Þ where
each xi belongs to instance space x and each label yi is in the label set y. The
Adaboost.M1 assigns the weight on the training sets. Initially, the weights are
equal. Based on the performance of each algorithm, the weight of misclassified
training sets are increased in order to focus on the misclassified sets in the next
rounds of training [10].

95.3.2 Random Forest

Random forest algorithm is one of the most popular among the researchers. It has
been proven that it is a popular and powerful algorithm in the pattern matching and
machine learning [17]. The random forest algorithm constructs a collection of
decision trees, which use the classification and regression methods. They are sets
of rule-based methods to generate the decision trees. The trees are developed
independent of each other and they vote for the best class to form the random
forest. The error in the random forest algorithm depends on the robustness of
individual trees and the correlation between them [18].

95.3.3 Adaboost and Random Forest

The combination of Adaboost and random forest was utilized in this paper.
Adaboost is the boosting algorithm that is used to improve the performance of the
random forest algorithm. For instance, researchers used the Adaboost with the
random forest and the final results show improved performance [10].
95 Comparative Evaluation of Ensemble Learning and Supervised Learning … 1031

95.4 Methodology

The methodology of this study comprises of three phases: building the data set,
feature selection and extraction, and the supervised learning and ensemble
learning. Figure 95.5 shows the architecture of the methodology. The details of
each phase are explained in the following sections.

95.4.1 Building the Data Set

The data set consists of the selected features in the network traffic. The first step is
to collect the network traffic generated by the malwares as well as normal appli-
cations in order to construct the data set. The capturing process of the network
traffic is done by installing the tPacketCapture Pro [12] on a real device and
running the applications for 30 min. The generated network traffic is captured via
the application and the result is prepared in the pcap file format. We then trans-
ferred the file to the computer to extract selected features. We used 600 malware
samples from the MalGenome data sample in this paper ranging from the simplest
malware to the most sophisticated ones like AnserverBot.

95.4.2 Feature Selection and Extraction

Among the massive number of network features, we have selected four of the most
important ones. They are packet size, connection duration, frame length and the
source port.
The packet size is the size of each packet; the packet that leaks user’s data from
a device to the hacker has larger size than normal packets. Most of the time,

Supervised Learning and

Building the Dataset Feature Selection and Extraction Ensemble Learning

Internet

Supervised
Normal Applications Learning
Feature Selection
and Extraction
Malicious Applications Ensemble
Learning
Dataset of Refined
Collected Network Traffic Dataset

Fig. 95.5 Methodology architecture

1032 A. Feizollah et al.

Table 95.1 Results of Algorithm TPR (%) FPR (%)

the experiment
Adaboost and random forest 99.6 0.4
Random forest 99.4 0.6

Android malwares connect to the hacker to check for new commands. Thus, most
connections are in the form of simple handshakes, which is a plausible reason to
choose the connection duration for this study. The frame length is the length of
each frame in bytes. The source port is important in the detection of the malwares.
The selected features were extracted using the tshark [19] program, which is a
command line version of the Wireshark program.

95.4.3 Supervised Learning and Ensemble Learning

The prepared data set is fed to the supervised algorithm as well as ensemble
algorithm. We chose the random forest algorithm for the supervised phase and the
adaboost and random forest for the ensemble phase. The purpose is to compare the
results of two learning methods and to prove that the ensemble learning provides
better results and performance than the supervised learning in the Android mal-
wares detection.

95.5 Results and Discussions

The results of the experiment are expressed in terms of true positive rate (TPR)
and false positive rate (FPR). The random forest algorithm achieved the TPR of
99.4 % while the Adaboost and random forest algorithm resulted in 99.6 % of the
TPR. Table 95.1 shows the results of the study.
Receiver operating characteristic (ROC) curve is one of the common methods
of analyzing the results of an experiment. It indicates how the true positive rate
changes as the internal thresholds of the algorithm are varied to generate more or
fewer false alarm. It plots true positive rate against false positive probability. ROC
curves show the tradeoff between false positive rate and true positive rate that
means any increase in the true positive rate is accompanied with a decrease in the
false positive rate. The top left corner of the diagram is considered the optimum
solution. The Figs. 95.5 and 95.6 illustrate the ROC curve for our experiment
(Fig. 95.7).
As presented, the ROC curves are very similar. In order to analyze ROC curves,
the area under the curve (AUC) is examined. In this case, the AUC for Adaboost
and random forest is 0.999 and the AUC for the random forest is 0.998. As the
95 Comparative Evaluation of Ensemble Learning and Supervised Learning … 1033

Fig. 95.6 Adaboost and 1

random forest ROC 0.9

True Positive Rate

0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1
False Positive Rate

Fig. 95.7 Random forest 1

ROC 0.9
0.8
True Positive Rate

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1
False Positive Rate

AUC approaches to 1.00, the algorithm has better performance. Thus, the Ada-
boost and random forest has the better performance than the random forest.
In order to authenticate this work, a comparison is done with a similar work. In
2013, a study was conducted [20] on Android malwares for classification purpose.
The authors used ensemble learning approach with the random forest as the chosen
algorithm. They achieved 99 % of detection rate in their experiment. However, in
this work, we attained 99.6 % of detection rate. In addition, we conducted another
experiment with supervised learning to prove that the ensemble learning is more
effective than the supervised learning.
Overall, it is evident that the ensemble learning, adaboost and random forest,
performs better than the supervised learning, random forest, in the Android mal-
ware detection and the ensemble learning has improved the Android malware
detection over the supervised learning.

95.6 Conclusion and Future Works

In this paper, we analyzed the network traffic generated by the Android malwares
using two versions of random forest algorithm, which are supervised learning
approach and ensemble learning approach. The experimental results proved that
1034 A. Feizollah et al.

the ensemble learning improved the Android malware detection. In addition, we

compared our work with a similar work done by other researchers and we con-
cluded that our work achieved higher detection rate.
As a future work, the real time experiment on the mobile malwares can be
conducted. It is worth mentioning that experimenting with more data samples is
suggested for future works.

Acknowledgments This work was supported in part by the Ministry of Higher Education,
Malaysia, under Grant FRGS FP034-2012A and the Ministry of Science, Technology and
Innovation, under Grant eScienceFund 01-01-03-SF0914.

References

1. Symantec: http://www.symantec.com/connect/blogs/android-madware-and-malware-trends
2. Teufl, P., Ferk, M., Fitzek, A., Hein, D., Kraxberger, S., Orthacker, C.: Malware detection by
applying knowledge discovery processes to application metadata on the Android Market
(Google Play). Security and Communication Networks, (2013)
3. Techcrunch: http://techcrunch.com/2013/03/07/f-secure-android-accounted-for-79-of-all-mobile-
malware-in-2012-96-in-q4-alone/
4. Cnet-News: http://news.cnet.com/8301-1035_3-57614451-94/heads-up-apple-android-to-make-
big-gains-in-2014-idc-predicts/?part=rss&subj=news&tag=title
5. Yajin, Z., Xuxian, J.: Dissecting android malware: characterization and evolution. In: 2012
IEEE Symposium on Security and Privacy (SP), pp. 95–109. (2012)
6. Maclin, R., Opitz, D.: Popular ensemble methods: an empirical study. J. Artif. Intell. Res. 11,
169–198 (1999)
7. Burguera, I., Zurutuza, U., Nadjm-Tehrani, S.: Crowdroid: behavior-based malware detection
system for Android. In: 1st ACM Workshop on Security and Privacy in Smartphones and
Mobile Devices, pp. 15–26. ACM, 2046619 (2011)
8. Tao, H., Ma, X.-p., Qiao, M.-y.: Support vector machine selective ensemble learning on
feature clustering and genetic algorithm. In: Wang, X., Wang, F., Zhong, S. (eds.) Electrical,
Information Engineering and Mechatronics 2011, vol. 138, pp. 1617–1625. Springer, London
(2012)
9. Li, Yang, Jian Lin, Li, Song Jie, Yue, Wang, Z.: Research of intrusion detection based on
ensemble learning model. Appl. Mech. Mater. 336, 2376–2380 (2013)
10. Thongkam, J., Guandong, X., Yanchun, Z.: Ada boost algorithm with random forests for
predicting breast cancer survivability. In: IEEE International Joint Conference on Neural
Networks (IJCNN), pp. 3062–3069. (2008)
11. Shabtai, A., Fledel, Y., Kanonov, U., Elovici, Y., Dolev, S., Glezer, C.: Google android: a
comprehensive security assessment. Secur. Priv. IEEE 8, 35–44 (2010)
12. tPacketCapturePro: https://play.google.com/store/apps/details?id=jp.co.taosoftware.android.
packetcapturepro
13. Su, X., Chuah, M., Tan, G.: Smartphone dual defense protection framework: detecting
malicious applications in android markets. In: 2012 Eighth International Conference on
Mobile Ad hoc and Sensor Networks (MSN), pp. 153–160. (2012)
14. Feizollah, A., Anuar, N.B., Salleh, R., Amalina, F., Ma’arof, RuR, Shamshirband, S.: A study
of machine learning classifiers for anomaly-based mobile botnet detection. Malays.
J. Comput. Sci. 26, 251–265 (2013)
15. Schapire, R.E.: A brief introduction to boosting. In: the International Joint Conference on
Artificial Intelligence, pp. 1401–1405. (1999)
95 Comparative Evaluation of Ensemble Learning and Supervised Learning … 1035

16. Freund, Y., Schapire, R.E.: Experiments with a new boosting algorithm. In: Thirteenth
International Conference on Machine Learning, pp. 148–156. (1996)
17. Meinshausen, N.: Quantile regression forests. Mach. Lear. Res. 7, 983–999 (2006)
18. Breiman, L.: Random forests. Mach. Learn. 45, 5–32 (2001)
19. tshark: http://www.wireshark.org/docs/man-pages/tshark.html
20. Alam, M.S., Vuong, S.T.: Random forest classification for detecting android malware. In:
2013 IEEE and Internet of Things Green Computing and Communications (GreenCom),
IEEE International Conference on and IEEE Cyber, Physical and Social Computing (iThings/
CPSCom), pp. 663–669. (2013)
Chapter 96
Tailored MFCCs for Sound Environment
Classification in Hearing Aids

Roberto Gil-Pita, Beatriz López-Garrido and Manuel Rosa-Zurera

Abstract Hearing aids have to work at low clock rates in order to minimize the
power consumption and maximize battery life. The implementation of signal
processing techniques on hearing aids is strongly constrained by the small number
of instructions per second to implement the algorithms in the digital signal pro-
cessor the hearing aid is based on. In this respect, the objective of this paper is the
proposal of a set of approximations in order to optimize the implementation of
standard Mel Frequency Cepstral Coefficient based sound environment classifiers
in real hearing aids. After a theoretical analysis of these coefficients and a set of
experiments under different classification schemes, we demonstrate that the sup-
pression of the Discrete Cosine Transform from the feature extraction process is
suitable, since its use does not suppose an improvement in terms of error rate, and
it supposes a high computational load. Furthermore, the use of the most signifi-
cative bit instead of the logarithm also supposes a considerable reduction in the
computational load while obtaining comparable results in terms of error rate.

96.1 Introduction

A hearing aid capable of automatically classifying the acoustic environment that

surrounds his/her user, and selecting the amplification ‘‘program’’ that is best
adapted to such environment (‘‘self-adaptation’’) would improve the user’s

R. Gil-Pita (&)  M. Rosa-Zurera

Signal Theory and Communications Department, University of Alcalá, Madrid, Spain
e-mail: [email protected]
M. Rosa-Zurera
e-mail: [email protected]
B. López-Garrido
Servicio de Salud de Castilla La Mancha (SESCAM), Castilla La Mancha, Spain
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 1037

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_96
1038 R. Gil-Pita et al.

comfort [1]. The ‘‘manual’’ approach, in which the user has to identify the acoustic
surroundings, and to choose the adequate program, is very uncomfortable and
frequently exceeds the abilities of many hearing aid users [2]. Furthermore, sound
classification is also used in modern hearing aids as a support for the noise
reduction and source separation stages, like, for example, in voice activity
detection [3], in which the objective is to extract information from the sound in
order to improve the performance of these systems.
There is a number of interesting features that could potentially exhibit different
behavior for speech, music and noise and thus may help the system classify the
sound signal. One of the most typical features used for information extraction in
audio analysis are the Mel Frequency Cepstral Coefficients (MFCCs), that have
already been used for sound environment classification in hearing aids [4]. The
problem of implementing an MFCC based sound classifier in a hearing aid is that
DSP-based hearing aids have constraints in terms of computational capability and
memory. The hearing aid has to work at very low clock rates in order to minimize
the power consumption and thus maximize the battery life. Additionally, the
restrictions become stronger because a considerable part of the DSP computational
capabilities are already being used for running the algorithms aiming to com-
pensate the hearing losses.
Therefore, the design of any automatic sound classifier is strongly constrained
to the use of the remaining resources of the DSP: roughly speaking, the compu-
tational power does not use to exceed 5 MIPS. This restriction in number of
operations per second enforces us to put special emphasis on signal processing
techniques and algorithms tailored for properly classifying while using a reduced
number of operations.
This paper presents a tailored efficient implementation of MFCC based sound
environment classifier for digital hearing aids. The idea is to propose solutions for
the implementation of these systems, bearing with the finite-precision limitations
and the computational cost, with the aim of establishing a set of tradeoffs between
performance in terms of error rate and number of required assembler instructions
per second.

96.2 Description of the Sound Environment Classifier

Figure 96.1 shows the block diagram of the sound environment classifier, in
relationship with the main loss compensation algorithm of a hearing aid. As we
can appreciate, the sound environment classifier basically consists of a feature
extraction block, and the aforementioned classifier. Each of this block will be
studied in detail in this section.
96 Tailored MFCCs for Sound Environment Classification … 1039

Fig. 96.1 Block diagram of

the proposed system showing
the different stages of the
hearing aid

96.2.1 Feature Extraction

There is a number of interesting features that could potentially exhibit different

behavior for speech, music, and noise and thus may help the system classify the
sound signal. Classical approaches use tailored specific features such as the
spectral centroid, spectral flux, voice to white or short time energy, which are
computationally efficient and they obtain considerably good results in the problem
at hand [5]. In a different approach, MFCCs render very good results in terms of
error rate, but with a considerably increase of the required computational resources
[6]. In this paper we will focus on the last ones, with the aim of analyzing their
drawbacks in terms of computation complexity, allowing a posterior definition of a
computationally simplified version of the MFCCs.
Obtaining MFCC coefficients [7] has been regarded as one of the techniques of
parameterization most important used in speech processing. They provide a
compact representation of the spectral envelope, so that most of the energy is
concentrated in the first coefficients. Perceptual analysis emulates human ear non-
linear frequency response by creating a set of filters on non-linearly spaced fre-
quency bands. Mel cepstral analysis uses the Mel scale and a cepstral smoothing in
order to get the final smoothed spectrum.
The main stages of MFCC analysis are:
• Short Time Fourier Transform (STFT): MFCCs are evaluated using informa-
tion derived from the STFT. In order to overcome the non-stationary of speech,
it is necessary to analyze the signal in short time periods, in which it can be
considered almost stationary. So, time frames or segments are obtained
dividing the signal in frames, which usually have an overlap factor of 50 %.
Then, DFT is calculated to each windowed time frame, generating Xt[k], a
time-frequency decomposition.
• Mel scale non uniform filter bank: In order to determine the MFCCs, phase
information is discarded, since we work with the energy of STFT, |Xt[k]|2. This
signal |Xt[k]|2 is then multiplied by a triangular filter bank, using Eq. (96.1).
1040 R. Gil-Pita et al.

X
N=2
Emt ¼ jXt ½kj2 Hm ½k; 1  m  F; ð96:1Þ
k¼0

where Hm[k] are the triangular filter responses, whose area is unity. These
triangles are spaced according to the Mel frequency scale, matching F the
number of final MFCCs [8].
• Discrete Cosine Transform (DCT): Through the DCT, expressed in Eq. (96.2),
the logarithm of the spectral coefficients are re-transformed, so the mel fre-
quency spectral coefficients are converted to a cepstral domain.

X
F
MFCCmt ¼ logðEmt Þcosðmðk  1=2Þp=N Þ; m ¼ 1; . . .; F ð96:2Þ
k¼1

Once MFCCs are evaluated, features are determined from statistics of each
MFCC. Some of the most common used statistics are the mean and the standard
deviation, which have been successfully used in sound environment problems for
hearing aids [5]. So, in this paper we study the implementation of mean and
standard deviation of the MFCCs as features, since we have found that these
values obtain very good results with a considerably low computational complexity.

96.2.2 Classification Algorithms

In order to analyze the performance of the proposals under different classification

scenarios, four different classifiers are studied: the Least Squares Linear Classifier
(LSLC), the Least Squares Diagonal Quadratic Classifier (LSDQC), the Multilayer
Perceptron (MLP) and the Quadratic Multilayer Perceptron (QMLP).
Least Square Classifiers are a classifiers that render very good results with a
very fast learning process and low computational complexity after training and
therefore they have been selected for the experiments carried out in this paper. Let
us considerer a set of training patterns x = [x1, x2, …, xL]T (for instance, and as we
described above, the mean and/or standard deviation of the used features), where
each of these patterns is assigned to one of the C possible classes, c = 1,…, C. In a
linear or quadratic classifier, the decision rule can be obtaining looking for the
class that maximizes a combination from a set of C combinations, as shows
Eq. (96.3).

X
L L X
X L
yc ¼ vc0 þ vcn xn þ wcnp xn xp ; ð96:3Þ
n¼1 n¼1 p¼1
96 Tailored MFCCs for Sound Environment Classification … 1041

where vcn are the linear weighting values, and wcnp are the quadratic weighting
values. So, in the LSLC the terms wcnp is zero, giving a linear combination of the
input features.
Another particular case is the LSDQC which is referred to the use of only the
diagonal coefficients of the matrix of quadratic terms ðwcnp ¼ 0; 8n 6¼ pÞ. In the
least squares approach, the weights are adjusted in order to minimize the mean
square error, which leads to the solution of the equations of Wiener-Hopf [9]. So,
the use of the LSDQC allows obtaining more complex solutions with the drawback
of duplicating the effective features of the classifier.
Concerning MLPs, they are feed forward artificial neural network models that
have successfully been implemented in hearing aids as sound environment clas-
sifiers [5]. They consist of multiple layers of nodes in a directed graph, with each
layer fully connected to the next one. Each node is denominated neuron, and it
processes data applying a nonlinear function called activation function to a linear
combination of the inputs of the node [10]. MLPs are typically trained to minimize
the mean squared error of the outputs using back propagation algorithms. In this
paper two-layer MLPS have been trained using the Levenberg-Marquardt opti-
mization algorithm [11]. In the MLP, each neuron of the first layer divides the
input space in two by means of a hyperplane, and the second layer combines these
hyperplanes to generate more complex boundaries. So, the complexity of the
solutions implemented in an MLP can be controlled by the number of available
neurons.
At last, QMLPs, also known as second order neural networks [12], are a variant
of the MLPs inspired in the differences between the LSLC and the LSDQC. It
consists in an MLP whose inputs are doubled adding quadratic terms of the input
pattern. So, in this case each neuron in the first layer can divide the input space in
two by a quadratic function, highly increasing the capability of the classification
system.
It is important to highlight all the four classifiers described in this paper are not
influenced by shifts or scaling of the input patterns, since these kind of changes
will only suppose a corresponding change in the input weights and they will not
alter the performance of the classifier. This property will be exploded in the next
section in order to reduce the computational cost of the implemented solution.

96.3 Description of the Proposals

In this section we will focus on two specific points that highly increase the
computational complexity of the sound environment classifier, and we will pro-
pose possible solutions to overcome each problem.
In order to determine the MFCCs in a hearing aid, we start from the squared
values of the STFT. These values are usually calculated in the main algorithms of
the DSP, since they are used for the multi-band compression-expansion algorithm
1042 R. Gil-Pita et al.

used to adapt sounds to the hearing losses of the patient. In some cases and in order
to optimize the power consumption, specific coprocessors are implemented in
order to determine the values of |Xt[k]|2 [13]. So, the computational complexity of
the evaluation of the terms |Xt[k]|2 will not be considered in the present paper.
The computational cost associated to the evaluation of the Mel scale triangular
filters supposes a number of instructions that must be taken into account. Ideally,
the number of Multiplication/AC cumulation (MAC) instructions required can be
proportional to the product F(N/2 + 1), being F the number of filter-banks and
N the frame length (see Eq. 96.1). Fortunately, since most of the terms of
Hm[k] are zero, the number of operations is drastically reduced. For instance, in the
particular case at hand the proportion of non-zero terms of Hm[k] is around 7.3 %
of the total values.
The next group of operations required to determine the MFCCs consists in the
evaluation of the DCT and the logarithm at the end of the MFCC extraction. These
two parts of the feature extraction process usually consume a high proportion of
the available DSP power, and therefore they will be deeply studied in this section.

96.3.1 Proposal 1: Approximated Evaluation

of the Logarithm

The logarithms are implemented in a typical DSP architecture using 13 assembler

instructions. Taking into account that every time frame we must determine the
logarithm of F terms (see Eq. 96.2), this operation requires a considerable amount
of instructions in the problem at hand. In order to reduce the number of instruc-
tions, in this paper we propose the use of the Most Significative Bit (MSB) (the
position of the first non zero bit in the binary representation of the register) in order
to approximate the logarithm function. This operation only requires one assembler
instruction, and it can be seen as an approach of the logarithm function. So, we can
obtain a very fast way of approximating the log2 by means of the position of the
most significative bit in the accumulator using Eq. (96.4).

logðaÞ ¼ logð2Þ  log2 ðaÞ ’ logð2ÞðmsbðaÞ þ 0:5Þ; ð96:4Þ

where msb() represents the MSB operation.

The multiplication by the constant log(2) and the addition of 0.5 can be sup-
pressed, since it only supposes a scale and shift over the features of the classifier,
and its use will only change the values of the trained weights of the classifiers, but
it will not alter the decision rule. So, the proposed replacement of the logarithm
function in Eqs. (96.2) and (96.5) by the MSB function significantly reduces the
number of required assembler instructions.
96 Tailored MFCCs for Sound Environment Classification … 1043

96.3.2 Proposal 2: Removal of the DCT Block

The DCT is evaluated by means of a set of $F$ linear combinations of the F filter
banks using Eq. (96.2). This supposes F2 MAC operations for each time frame.
In the particular case of determining the mean value of the MFCCs as features
with a linear classifier, the use of the DCT can be suppressed, since the perfor-
mance of the classifier without the DCT is completely equivalent. The evaluation
of the mean value supposes a linear combination of the DCT, and the classifier
implements linear combinations of these linear combinations. Since the weights of
the linear classifier are determined in order to minimize the mean square error, the
use of the DCT supposes a change in the values of the weights vcn but it will not
alter the values of the linear combination of the classifiers (given by Eq. (96.3)
with wcnp ¼ 0; 8n 6¼ p), since a linear combination of a linear combination is
another linear combination. Therefore, the final error of this particular classifica-
tion scheme (mean of the MFCCs and linear classifier) is completely independent
of the use of the DCT in the feature extraction process.
So, the proposed tailored MFCCs removes the DCT block extraction, so that
Eq. (96.2) is replaced by Eq. (96.5).
0
MFCCmt ¼ logðEmt Þ ð96:5Þ

In the case of a more complex classifier, or in case of the use of the standard
deviation of the features, the error rate might vary with the suppression of the DCT
block, but the changes in terms of error rate could not compensate the drawback of
the added computational complexity.

96.4 Experimental Work

Prior to the description of the different experiments we have carried out, it is worth
having a look at the sound database we have used. It consists of a total of 2343 s of
audio, including both speech in quiet, speech in noise, speech in music, vocal
music, instrumental music and noise. The database was manually labelled,
obtaining a total of 781 s of speech in quiet, speech in music and speech in noise,
781 s of music, and 781 s of noise, so we will work in a three class problem
(C = 3, speech, music and noise). All audio files are monophonic, and were
sampled with a sampling frequency of Fs = 16 kHz and 16 bits per sample.
Speech and music files were provided by D. Ellis, and recorded by E. Scheirer and
M. Slaney [14]. This database has already been used in a number of different
works [15]. Speech was recorded by digitally sampling FM radio stations, using a
variety of stations, content styles and levels, and contains samples from both male
and female speakers. The sound files present different input levels, with a range of
30 dB between the lowest and the highest, which allows us to test the robustness of
1044 R. Gil-Pita et al.

the classification system against different sound input levels. Music includes
samples of jazz, pop, country, salsa, reggae, classical, various non-Western styles,
various sorts of rock, and new age music, both with and without vocals. Finally,
noise files include sounds from the following environments: aircraft, bus, cafe, car,
kindergarten, living room, nature, school, shop, sports, traffic, train, and train
station. These noise sources have been artificially mixed with those of speech files
(with varying degrees of reverberation) at different Signal to Noise Ratios (SNRs)
ranging from 0 to 10 dB. In a number of experiments, these values have been
found to be representative enough regarding the following perceptual criteria:
lower SNRs could be treated by the hearing aid as noise, and higher SNRs could be
considered as clean speech.
For training and testing, it is necessary for the database to be divided into two
different sets. 1414 s (H 60 %) for training, and 930 s (H 40 %) for testing. This
division has been randomly carried out, ensuring that the relative proportion of
files of each category is preserved for each set. The training set is used to deter-
mine the weights of the classifiers in the training process, and the test set is used to
assess the classifier’s quality after training. The test set has remained unaltered for
all the experiments described in this paper.
Each file was processed using the hearing aid simulator described in [16]
without feedback. The features were computed from the output of the Weighted
Overlap-Add (WOLA) filter-bank with N = 128 DFT points and analysis and
synthesis window lengths of 256 samples. So, the time/frequency decomposition is
performed with 65 frequency bands. Concerning the architecture, the simulator has
been configured for a 16-bits word-length Harvard Architecture with a MAC unit
that multiplies 16-bit registers and stores the result in a 40-bit accumulator.
In order to study the effects of the limited precision, the classifiers were con-
figured for taking a decision with time slots of 20 ms. The objective is to study the
effects of the limited computational capability in a classification scenario in which
a small time scale is required like, for example, in noise reduction or sound source
separation applications. The results we have illustrated below show the average
probability of classification error for the test set and the computational complexity
of the considered in percentage of computational load for a 5 MIPS standard
hearing aid DSP. The probability of classification error represents the average
number of time slots that are misclassified in the test set.
It is important to highlight that in a real classification system the classification
evidence can be accumulated across the time for achieving lower error rates. This
fact makes necessary a study of the tradeoff between the selected time scale, the
integration of decision for consecutive time slots, the performance of the final
system and the required computational complexity. This analysis is out of the
scope of this paper, since our aim is not to propose a particular classification
system, that must be tuned for the considered hearing aid application, but to
illustrate a set of tools and strategies that can be used for determining the way an
MFCC based classifiers can efficiently be implemented in real time for sound
environment classification tasks with limited computational capabilities.
96 Tailored MFCCs for Sound Environment Classification … 1045

With the aim of carrying out an experimental validation of the two proposals,
several experiments have been carried out under different scenarios. In these
experiments, the objective is to determine the effectiveness of the proposed
approximations over both the error rate and the computational cost. The param-
eters of this second group of experiments are:
• Features: We have carried out experiments with 25 MFCCs using the Proposal
1 (MSB approximation of the logarithmic function) and the Proposal 2
(removal of the DCT block in the MFCC estimation), described in the above
section. Furthermore, we have also carried out experiments combining both
proposals. For comparison purposes, we also run experiments using (1) a set of
four classic features (spectral centroid, spectral flux, voice to white and short
time energy) described in [5], (2) the 25 standard MFCCs, and the first 12
MFCCs, as it was described in [6].
• Statistics: In order to evaluate the importance of the selected statistics of the
features, two different choices have been considered and compared: first, the
use of the mean value of the different features along the 20 ms time slot has
been studied; second, both the mean and the standard deviation of the features
have been used as input vector for the corresponding classifiers.
• Classifiers: Four different classifiers have been evaluated for each feature and
statistic combination: the LSLC, LSDQC, a two layer MLP, and a two layer
QMLP. Both the MLP and the QMLP have been configured with 10 tan-
sigmoidal neurons in the first layer and three linear neurons in the output layer
(corresponding to the three classes considered in the experiments).
Table 96.1 shows the results obtained in the experiments. As we can see, the
use of the 4 classic features supposes the worst results in terms of error rate, but the
associated computational cost is quite low when compared to the use of 25
standard MFCCs. On the other hand, the use of the first 12 MFCCs supposes a
reduction in the DSP load, but in most of the cases with a consequent increment in
the average error rates.
Concerning the approximation of the logarithm by the MSB operator (Proposal 1),
we can see that it supposes a slight increment in the error rate in all the cases when
compared to the standard use of MFCCs (an average relative increase of 1.6 % in
error rate), but a considerable reduction in the computational cost (an average rel-
ative reduction of 24 % in the DSP load). This fact makes the proposed MSB
approximation suitable for most of the real time hearing aid scenarios.
As it was expected, we can check that the results achieved with the mean value
of 25 standard MFCCs and with the Proposal 2 for the LSLC are equal. As we
demonstrated in the last section, in that case the DCT block does not alter the
classifier performance and therefore it can be removed. But, furthermore, in the
case of removing the DCT (Proposal 2) with other classifier or in the case of also
using the standard deviation, the error rate is not only different but even lower than
in the case of the standard use of MFCCs (average relative reduction of 2.66 %).
Moreover, the removal of the DCT block supposes an average reduction of 50 %
 1046

Table 96.1 Test error rate and DSP load for the experiments described in the paper
Mean of the features Mean and standard deviation of the features
Features Classifier LSLC LSDQC MLP QMLP LSLC LSDQC MLP QMLP
(%) (%) (%) (%) (%) (%) (%) (%) (%)
Referenced 4 classic features [5] Error rate 37.23 35.83 29.01 28.43 36.40 34.52 28.34 27.01
DSP load 1.43 1.45 1.60 1.65 2.02 2.06 2.22 2.31
25 standard MFCCs Error rate 22.64 19.23 15.32 16.62 20.60 17.20 14.24 12.72
[8]
DSP load 5.55 5.68 5.87 6.17 6.20 6.45 6.70 7.30
First 12 MFCCs [6] Error rate 24.49 21.58 16.35 15.85 21.95 18.81 15.15 13.92
DSP load 3.89 3.95 4.12 4.26 4.50 4.62 4.81 5.10
Proposed Proposal 1: MSB- Error rate 22.63 19.26 15.57 14.36 20.67 17.36 14.26 13.30
MFCCs
DSP load 4.05 4.18 4.37 4.67 4.70 4.95 5.20 5.80
Proposal 2: NODCT Error rate 22.64 19.38 15.03 13.67 19.80 16.15 13.83 11.75
DSP load 2.43 2.55 2.75 3.05 3.08 3.33 3.57 4.17
Proposal 1 + 2 Error rate 22.63 19.40 15.34 13.99 19.96 16.27 13.22 12.35
DSP load 0.93 1.05 1.25 1.55 1.58 1.83 2.07 2.67
R. Gil-Pita et al.
96 Tailored MFCCs for Sound Environment Classification … 1047

of the DSP load, which is quite important in order to minimize power

consumption.
When both proposals are implemented (MSB approach and DCT removal) we
get the lowest computational costs (74 % lower than the use of standard MFCCs,
and 14 % lower than the use of the classic features), and the error rates are even
lower than those obtained by the standard MFCCs. This choice might be suitable
for those applications in which the computational cost is more constrained.
Concerning the comparison between classifiers, we can se that their use allow to
establish tradeoffs between computational cost and error rate. So, the best result in
terms of computational cost is obtained by the LSLC classifier with the mean value
of the MFCCs under Proposals 1 and 2 (22.63 % of error rate with a computational
load of 0.93 %), and in those cases in which we desire to get the best performance
in terms of error rate, the QMLP achieves the lowest error rates (for instance,
12.35 % with the mean and standard deviation of the same features with only a
2.67 % of the DSP load).

96.5 Conclusions

This paper has been motivated by the fact that the implementation of signal
processing techniques on hearing aids is strongly constrained by the small number
of instructions per second to implement the algorithms on the digital signal pro-
cessor the hearing aid is based on. In this respect, the objective of this paper has
been the proposal of a set of approximations in order to optimize the implemen-
tation of standard MFCC based sound environment classifiers in real hearing aids.
The performance of the proposed solutions show a balance between keeping error
classification probability within low values (in order to not disturb the user’s
comfort) and achieving this by using a small number of instructions per second.
The reason underlying these restrictions is that hearing aids have to work at low
clock rates in order to minimize the power consumption and maximize battery life.
The final, global conclusion is that the suppression of the DCT block from the
MFCC extraction process is suitable, since its use does not suppose an improve-
ment in terms of error rate, and it supposes a high computational load. Further-
more, the use of the MSB operator instead of the logarithm in the MFCC supposes
a considerable reduction in the computational load while obtaining comparable
results in terms of error rate.

Acknowledgments This work has been partially funded by the University of Alcalá (CCG2013/
EXP-074), the Spanish Ministry of Economy and Competitiveness (TEC2012-38142-C04-02)
and the Spanish Ministry of Defense (DN8644-ATREC).
1048 R. Gil-Pita et al.

References

1. Hamacher, V., Chalupper, J., Eggers, J., Fischer, E., Kornagel, U., Puder, H., Rass, U.: Signal
processing in high-end hearing aids: state of the art, challenges, and future trends. EURASIP
J. Appl. Sig. Process. 18, 2915–2929 (2005)
2. Büchler, M., Allegro, S., Launer, S., Dillier, N.: Sound classification in hearing aids inspired
by auditory scene analysis. EURASIP J. Appl. Sig. Process. 18, 2991–3002 (2005)
3. Marzinzik, M.: Noise reduction schemes for digital hearing aids and their use for hearing
impaired. PhD thesis, Carl von Ossietzky University Oldenburg (2000)
4. Dong, R., Hermann, D., Cornu, E., Chau, E.: Low-power implementation of an hmm-based
sound environment classification algorithm for hearing aid application. In: Proceedings of the
15th European Signal Processing Conference (EUSIPCO 2007), vol. 4 (2007)
5. Gil-Pita, R., Alexandre, E., Cuadra, L., Vicen, R., Rosa-Zurera, M.: Analysis of the effects of
finite precision in neural network-based sound classifiers for digital hearing aids. EURASIP
Journal on Advances in Signal Processing, vol. 2009. (2009)
6. Xiang, J.J., McKinney, M.F., Fitz, K., Zhang, T.: Evaluation of sound classification
algorithms for hearing aid applications. In: Acoustics Speech and Signal Processing
(ICASSP), 2010 IEEE International Conference on, IEEE, pp. 185–188. (2010)
7. Hunt, M., Lennig, M., Mermelstein, P.: Experiments in syllable-based recognition of
continuous speech. In: Acoustics, Speech, and Signal Processing, IEEE International
Conference on ICASSP’80, IEEE vol. 5, pp. 880–883. (1980)
8. Mohino-Herranz, I., Gil-Pita, R., Alonso-Diaz, S., Rosa-Zurera, M.: Synthetical enlargement
of mfcc based training sets for emotion recognition. Int. J. Comput. Sci. Inf. Technol. 6(1),
249–259 (2014)
9. Ye, J.: Least squares linear discriminant analysis. In: Proceedings of the 24th international
conference on Machine learning, ACM pp. 1087–1093. (2007)
10. Bishop, C.M., Nasrabadi, N.M.: Pattern recognition and machine learning, vol. 1. Springer,
New York (2006)
11. Marquardt, D.W.: An algorithm for least-squares estimation of nonlinear parameters. J. Soc.
Ind. Appl. Math. 11(2), 431–441 (1963)
12. Kosmatopoulos, E.B., Polycarpou, M.M., Christodoulou, M.A., Ioannou, P.A.: High-order
neural network structures for identification of dynamical systems. Neural Networks, IEEE
Trans. 6(2), 422–431 (1995)
13. Cuadra, L., Alexandre, E., Gil-Pita, R., Vicen-Bueno, R., Álvarez-Pérez, L.: Influence of
acoustic feedback on the learning strategies of neural network-based sound classifiers in
digital hearing aids. EURASIP Journal on Advances in Signal Processing, pp. 14. (2009)
14. Scheirer, E., Slaney, M.: Construction and evaluation of a robust multifeature speech/music
discriminator. In: ICASSP, pp. 1331–1334. (1997)
15. Thoshkahna, B., Sudha, V., Reemigration, K.: A speech-music discriminator using hiln
model based features. In: IEEE International Conference on Acoustics, Speech and Signal
Processing, pp. 425–428 (2006)
16. Vicen-Bueno, R., Gil-Pita, R., Utrilla-Manso, M., Álvarez-Pérez, L.: A hearing aid simulator
to test adaptive signal processing algorithms. In: IEEE Int. Symposium on Intelligent Signal
Processing (WISP), pp. 619–624. (2007)
Chapter 97
Metamodelling Architecture
for Modelling Domains with Different
Mathematical Structure

Vitaliy Mezhuyev

Abstract The new metamodelling approach for domain specific modelling is

proposed in the paper. The additional level of the metamodelling architecture is
introduced, which gives the possibility of metamodels development in the different
mathematical semantics. This allows to take into account the mathematical
structure of modelled domains, and to use the mathematical operations for
development of new effective methods for solving domain specific tasks. The
applicability of the approach for development of metamodels for modelling dif-
ferent domains is shown.

Keywords Domain specific modelling Metamodel   Metamodelling architec-

 
ture Mathematical structure Formal system

97.1 Introduction

The methodology of Domain Specific Modelling (DSM) becomes more and more
popular today, allowing to overcome the known issues of the ‘‘universal’’ mod-
elling approach [1]. The sense of DSM is development of Domain Specific Lan-
guages (DSLs), applicable for modelling properties of particular domains. A DSL
is built inside a so called metamodel, defining the concrete syntax of the language.
The abstract syntax of a DSL is defined in the frame of the meta-metamodel as e.g.
MOF [2], GOPPRR [3], MGA [4] etc.

V. Mezhuyev (&)
Faculty of Computer Systems and Software Engineering, University Malaysia Pahang,
Gambang, Malaysia
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 1049

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_97
1050 V. Mezhuyev

Emphasizing the power of the existing DSM approaches, they have a number of
issues, caused by the lack of generalisation and formalisation:
• the metamodel based DSLs are mostly descriptive, i.e. not expressive for the
definition of methods for solving domain specific tasks;
• the applicability of a DSL by the generation of software data and code is
limited;
• while the DSM approach is intended for using by domain experts, the oblig-
atory involvement of IT specialists for development of code generators is
needed;
• for code generation an additional external language should be used, which is
not linked with specifics of a modelled domain;
• the meta-metamodel, used for metamodels development, does not reflect the
mathematical structure of a considered domain and is hardcoded inside a DSM
tool.
Let’s consider the principles of the proposed approach to the metamodels
development, allowing to overcome the specified above issues:
• the formal definition of the object of modelling—the domain, as the set of
entities, linked by the forming mathematical structure and the domain specific
relationships;
• the definition of the meta-metamodel and the metamodel as the formal systems,
allowing to fix correspondingly the structural and domain specific properties;
• the mathematical structure of a domain is defined at the meta-metamodel level
and next is used as the carrier of domain specific properties;
• the additional level of the metamodelling architecture is introduced, which
allows to develop the meta-metamodels, having different mathematical
semantics.
While the existing metamodelling approaches use the predefined mathematical
formalisms (mostly, graphs) for structuring domain properties, here the develop-
ment of meta-metamodels in the different mathematical semantics is possible.
Additional level of the metamodelling architecture allows to express properties of
domains in terms of set theory and to reflect different mathematical structures
(algebraic, topological, differential, geometrical etc.). Corresponding mathemati-
cal operations are integrated in the metamodel and used for solving domain spe-
cific tasks. Generation of software data and code becomes the partial case of the
proposed metamodelling approach.
The paper is organized as follows. First the new metamodelling architecture is
discussed in comparison with existing approaches. Section 97.3 of the paper shows
applicability of the proposed approach for producing the graph based metamodels
for modelling software systems. Section 97.4 expands the practical applications
for requirements engineering, business process modelling and solving tasks of
multidimensional physical domains. The conclusion, plan of future research and
references list finalize the paper.
97 Metamodelling Architecture for Modelling Domains … 1051

97.2 Metamodelling Architectures

The methodology MOF (Meta Object Facility) [2] was used by the OMG (Object
Management Group) consortium for development of the Unified Modelling Lan-
guage (UML). MOF has the four levels of the metamodelling architecture. The top
level is the meta-metamodel (M3), defining the language for development of the
metamodels (having the level M2). The level M2 (here, UML) used for devel-
opment of the domain models of the level M1 (the UML-models). The last is the
level of data (M0), describing the concrete instances of M1. The MOF architecture
is based on the object-oriented methodology of software systems design.
The meta-metamodel GOPPRR (Graph-Object-Property-Port-Role-Relation-
ship) allows to produce metamodels inside the graph based notations, by means of
connection of objects by relationships, definition of domain properties (attributes)
and roles [3]. Each of the GOPPRR concepts a metatype is called. As MOF, the
metamodelling architecture of the GOPPRR in four levels can be shown (see the
Fig. 97.1).
The proposed approach also has the multiple-level metamodelling architecture,
but it semantics differs from the existing methodologies. All of the metamodels are
considered to be formal systems; they contain an alphabet of types, a grammar and
operations. We introduce the additional level of the metamodelling architecture—
the meta-meta-metamodel (M4), as a formal system, that is built on the basis of set
theory. M4 includes the meta-metatype ‘‘element of a set’’, set operations and
grammar rules, which (taken together) allow us to specify a set structure. This
approach allows us to consider a domain as a set of heterogeneous entities, having
domain specific properties and linked by different kinds of mathematical structures.
Formally, we define a domain as a set of entities D, linked by structural S and
domain specific P relationships:

D ¼ fd1 ; d2 . . .dN g; S; P  D  D ð97:1Þ

where N is a power of D. Each element of D can have attributes, which we

consider as unary relationships on D. 0-ary relationships are used to identify
elements of D. Binary and other relationships are used to fix mathematical
structure of D.
All of the levels of the proposed metamodelling architecture contain not only
descriptive elements, such as in MOF or GOPPRR, but also procedural part,
implemented with software functions.
Following our proposal, the architecture for development of the graph based
metamodel on the Fig. 97.2 is shown. Here a node and an edge of a graph serve as
the mathematical metatypes for development of domain specific metamodels types
(an attribute is the inherent part of a node and of an edge). The node and the edge
are produced from the meta-meta-metamodel as the having algebraic structure
subsets of the composing domain entities. Note, while GOPPRR [3] and MGA [4]
also use the graphs for structuring domain specific properties, this is a partial case
1052 V. Mezhuyev

M3 (Meta-metamodel) GOPPRR

M2 (Metamodel) Object Property Relationship

Power:
M1 (Model) Device Connector
Integer

M0 (Data) Electrical 100W

TV-Set Cable
socket

Fig. 97.1 The GOPPRR metamodelling architecture

of the proposed approach, where development and using the different mathematical
structures is possible.
The implementation of mathematical operations of the metamodels at all levels
of the proposed architecture, forms the Application Program Interface (API) of the
corresponding software tool. The API of M4 contains the methods for manipu-
lation with the elements of a set of composing domain entities. The API of M3 is
the operations with subsets (e.g., with a node and an edge of a graph, and in the
general case with any model objects of the considered domain). For M2, the API
contains the metamodel processing routines (here, the metatypes of the level M3
become domain-specific types, i.e., to the mathematical subsets the semantics of
the domain is assigned). M1 contains instances of the types and definitions of
domain-specific methods, implemented with the APIs of all the previous levels.
M0 is data values and processes in the computer memory (instances of the
methods, defined at the level M1).

97.3 Development of Graph-Based Metamodels

Let us consider the mathematical method for producing the graph-based meta-
metamodel in the context of proposed approach. Its alphabet includes the meta-
types node N and edge E of the graph Gr = (N, E); the grammar GGr is the set of
rules, defining the possibility {true, false} of the connection of nodes ni, nj by the
edge ek = (ni, nj), n e N, e e E


GGr ¼ ni ; nj jgk  ftrue; falseg; ni ; nj  N; i; j ¼ 1::M; k ¼ 1::K ð97:2Þ

where M is a power of N. The number of rules K depends on the properties of the

graph Gr (is it directed, are loops possible, etc.).
At the level of metamodel development, to the nodes and the edges of the meta-
metamodel the semantics of domain is assigned. For example, the node N can be
the metatype for definition of the types of software tasks and synchronization
97 Metamodelling Architecture for Modelling Domains … 1053

M4 (Meta-meta-
metamodel ) Element of a set
Set theory

M3 (Meta-
metamodel ) Node Edge
Graph theory

M2 (Metamodel ) Entity Interaction

Interaction Entities

M1 (Model) Device Connector

Electrical
M0 (Data) TV-Set Cable
socket

Fig. 97.2 The levels of the proposed metamodelling architecture

objects, and the edge E can be the metatype for definition of the types of channels
(communication protocols) between tasks and synchronization objects. This
metamodel will include the alphabet, containing typical for parallel programming
synchronisation objects (critical section, mutex, semaphore, resource, FIFO etc.)
and software tasks (driver, application etc.); the grammar rules, specifying the
valid interactions of software tasks via synchronisation objects, and operations,
used for definition of code generation functions.
Table 97.1 shows an example of the definition of the metamodel for modelling
the parallel concurrent software system inside the graph based meta-metamodel.
In this example, a Node and an Edge are the mathematical metatypes of graph
based meta-metamodel M3. Domain specific types are the nodes Task, Sync and
the edges PutData, GetData, which compose the alphabet of M2 metamodel and
are used to create instances at the M1 level. M2 also defines the grammar rules for
combining instances of the types by using predicates PutData(Task, Sync) and
GetData(Sync, Task). These grammar rules correspond to the edges of the graph-
based meta-metamodel and are used for development of code generation methods
(implemented by walking the graph based model M1). The M1 model of software
system includes instances of Task1, Task2 … TaskT and synchronization objects
Sync1, Sync2 … SyncS, linked by the channels of interaction PutData, GetData
(where T, S—are the number of tasks and the number of synchronization objects in
the model respectively).
For the interesting reader, to show the applicability of described graph based
metamodel, we can refer to the metamodel of interacting entities [5], which was
used for development of a real-time operation system [6] and for modelling dis-
tributed parallel real-time software [7].
1054 V. Mezhuyev

Table 97.1 Levels of metamodelling architecture for a software system modelling

Level Alphabet Grammar Operations/
Methods
M4 Math Elements d of the set D The rules of grammar, Create /delete
structures based on the relations element d,
d 2 D, fdg  D subset {d}
M3 Node n e Node and edge Connection of nodes by Add edge
e e Edge of graph edges ek (ni, nj), ni, nj e G0 = G + e
G = (Node, Edge), Node, ek e Edge, i, j = 1.. Delete edge
Node, Edge , D |Node|, i 6¼ j, k = 1..|Edge| G0 = G - e
Add node
G0 = G + n
Delete node
G0 = G - n
M2 Domain Edge PutData, GetData; PutData (Task, Sync); Add /delete a
specific Node Task, Sync; GetData (Sync, Task) type of task
properties Task /sync
object Sync,
create
communication
channel
PutData,
GetData
M1 Task Task1, Task2;
Sync Sync1;
PutData (Task1, Sync1);
GetData (Sync1, Task1);

The definition of the metamodel alphabet as the set of attributed types and the
domain model as the instances of the types, having the concrete values of attri-
butes, make possible the formal checking a model in its state space. Due to
including mathematical methods in the metamodel the checking properties of
behaviour of a real-time system (e.g. absence of deadlocks) was applied. The
graph based methods (e.g. Dijkstra’s algorithm) for development of the code
generation functions (e.g. routing table of a real time operation system) were used.

97.4 Other Applications of the Metamodelling Approach

Except development of graph-based metamodel for software systems design, the

applicability of the proposed approach was proven for the next domains:
• requirements engineering (RE), where conceptual metamodelling for systems
specification was used. The set of the typical for the RE concepts formed the
alphabet of the metamodel, which symbols were the types for instantiation—
definition of the concrete statements describing a system properties and
behaviour. The methods of the graph based meta-metamodel were used to
97 Metamodelling Architecture for Modelling Domains … 1055

check correspondence of the graph of architectural decomposition to the graph

of initial requirements, generate the document of systems specifications, made
the control of versions etc. The conceptual metamodel was further expanded by
the Finite State Machine formalism [5]. This allows us to build the domain
specific models of processes on the base of the ontology of a domain. To each
concept of ontology the state transition attribute was added. The process
grammar was the set of rules, defining the state transitions of conceptual model
of a system description. e.g. only after capturing requirements user can move to
the specification stage, next to the phase of architectural modelling etc. Such
the approach allows us to manage users activity to achieve the goal of a process
in a given time (up to deadline);
• development of the metamodel, based on the vector algebra and the logic of
syllogisms. Here vectors were used as the metatypes for producing the logical
types of the metamodel alphabet. In the practical implementation [8], the
alphabet of the metamodel on the base of the types of categorical syllogisms
was developed. Due to using vector algebra for the definition of the metamodel,
the operations on syllogisms as operations on vectors in linear vector space
were implemented. This allows us to develop the algorithm for automatic
geometrical theorem proving. The approach was used for development of the
logic for optical computers, where at physical level vectors were implemented
as laser beams;
• development of the metamodel for multidimensional physical domains [9]. The
alphabet of the meta-metamodel was defined as the set of the basic (corre-
sponding to the dimensions of the physical space) geometrical objects, i.e.
point, line, surface and 3D region. For metamodels development we set dis-
tributions of physical properties among the defined with the meta-metamodel
geometrical structures. Due to considering objects as the sets of geometrical
points in the physical space, the grammar of the metamodel in the terms of
Boolean operations on geometrical subsets was defined. This grammar limits
the possible compositions of the geometrical objects in the 3D space. The
mathematical methods of the metamodel correspond to the solutions of mul-
tidimensional tasks of the integral and deferential calculus. As the interesting
application of the metamodelling approach for physical domains the design of
metamaterials (artificial composites with specific optical properties) can be
mentioned [10].

97.5 Plan of Future Research

The plan of research is further exploring the properties of the metamodels,

allowing to fix different mathematical structures:
• the formal definition of the metamodels, the mathematical structure of its types,
grammars and operations at all levels of the metamodelling architecture;
1056 V. Mezhuyev

• learning the linguistic properties of the metamodels, incl. the possibility of

reduction of the grammars into the normal Chomsky form;
• definition of the method for metamodels composition, allowing to combine the
declarative and imperative constructs (alphabet, grammar and operations);
• exploring the textual and the visual forms of expression of metamodels and
development of the method for its combination;
• expansion of the approach on the other types of mathematical structures
(metric, geometrical, differential, topological, etc.).

97.6 Conclusion

The new approach for metamodels development is proposed. The metamodelling

architecture is decomposed into the layers, allowing to fix the structural and the
domain specific properties. This allows to take into account the mathematical
structure of considered domains. The additional set-based level of the metamod-
elling architecture is introduced, which allows to define the meta-metamodels in
the different mathematical semantics.

References

1. France, R.B., Ghosh, S., Dinh-Trong, T., Solberg, A.: Model-Driven development using
UML 2.0: promises and pitfalls. IEEE Comp. 39(2):59–66 (2006)
2. ISO/IEC 19502:2005, Information technology. Meta object facility.—ANSI, p. 292 (2007)
3. Kelly, S., Juha-Pekka, T.: Domain-Specific Modeling: Enabling Full Code Generation,
p. 427. Wiley-IEEE Computer Society Pr. (2008)
4. Nordstrom, DD.: Metamodeling—Rapid Design and Evolution of Domain-Specific Modeling
Environments/Dissertation for the Degree of Doctor of Philosophy in Electrical Engineering,
p. 170. Nashville, Tennessee (1999)
5. Mezhuyev, V., Sputh, B., Verhulst, E.: Interacting entities modelling methodology for robust
systems design. In: Second International Conference on Advances in System Testing and
Validation Lifecycle, pp. 75–80. CPS Publishing (2010)
6. Boute, R.T., Miguel, J., Faria, S., Sputh, B.H.C.: Vitaliy Mezhuyev Formal Development of a
Network-Centric RTOS/Eric Verhulst, pp. 227. Springer, Berlin (2011)
7. Mezhuyev, V.: domain specific modelling distributed parallel real time applications. Syst.
Inf. Process. 5(86), 98–103 (2010)
8. Mezhuyev, V.: Vector logic: Theoretical Principles and Practical Implementations,
pp. 91–97. The papers of Zaporizzia National University, Zaporizzia, ZNU (2006)
9. Mezhuyev, V., Lytvyn, O.: Metamodel for visual modelling multidimensional domains and
its practical applications. Control Syst. Mach. 4, 31–43 (2010)
10. Mezhuyev, V., Pérez-Rodríguez, F.: Visual Environment for Metamaterials Modelling. Some
Current Topics in Condensed Matter Physics, pp. 1–13. Universidad Autónoma del Estado de
Morelos (2010)
Chapter 98
Use Case Based Approach to Analyze
Software Change Impact and Its
Regression Test Effort Estimation

Avinash Gupta, Aprna Tripathi and Dharmendra Singh Kuswaha

Abstract Software needs to be maintained and changed to satisfy the new

requirement and existing faults. Without analyzed changes, change implemented
to software often cause unexpected ripple effects. To avoid this and diminish the
risk of performing undesirable changes, an impact analysis of the change is done.
Software Change Impact Analysis (SCIA) needs to be computed at every change
request for software systems, to access the impact information for several critical
software engineering tasks such as risk analysis, effort estimation, and regression
testing. The use of UML analysis/design models on large projects lead to a large
number of interdependent UML diagrams. Paper proposes a UML model based
approach strictly to use use-case diagram for impact analysis that is applicable in
early decision making and change planning. Later, by using the impact set we
estimate the regression test effort required for the effected change in the software.
The reduction in test effort observed ranges from 47 to 95 %, saving significant
software testing cost.

Keywords Impact analysis  Use-case  Test case  Regression testing  SCIA

98.1 Introduction

Requirement of users are increasing day by day. To fulfil user’s requirements and to
compete in market, software companies need to enhance their product’s perfor-
mance and functionality before let it being obsolete in market. So software change

A. Gupta  A. Tripathi (&)  D.S. Kuswaha

MNNIT Allahabad, Allahabad, India
e-mail: [email protected]
A. Gupta
e-mail: [email protected]
D.S. Kuswaha
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 1057

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_98
1058 A. Gupta et al.

Fig. 98.1 Software change impact analysis process model

Fig. 98.2 Impact analysis classification be Lehnart

management is essential activity to enrich software age. SCIA is an essential activity

before implementing change in software. It allows to measure the effort required
implementing a change and its ripple effects, impact analysis suggests those soft-
ware artefacts which may be changed, and helps to identify test cases which should
be re-executed to ensure that the change was implemented correctly [1].
SCIA has following steps as shown in Fig. 98.1.
Common techniques used to implement change impact analysis are based on
either traceability or dependency relationships between the software artefacts.
Traceability-based impact analysis techniques work on analyzing the relationships
between requirements and other development artefacts (such as design, imple-
mentation and test cases) to determine the scope of the anticipated changes,
dependency-based impact analysis techniques work on a more detailed level by
analyzing the relationships between the artefacts of the same development phase.
Lehnart et al. [2] proposes impact analysis technique’s classification based on
evaluation artifacts as shown in Fig. 98.2.

98.1.1 Model Based Software Change Impact Analysis

Model based analysis provides facility of SCIA for software models to keep their
quality and correctness at early stage. Models, such as UML diagrams, enable the
assessment of architectural changes on a more abstract level than source code.
98 Use Case Based Approach to Analyze Software … 1059

Fig. 98.3 Use-case diagram designed by visual paradigm tool

This enables SCIA in earlier stages of development and in MDD, which has
become more important in recent years. But dependent on the underlying mod-
elling language, even model based analysis provides effective impact analysis
results, for example when analyzing detailed UML class diagrams. Thus, SCIA is
an essential activity of software Maintenance. Proposed approach is concerned
towards the automation of SCIA in MDD environment that provides an automated
SCIA approach for use-case model scheme by using flow of events written in
Visual Paradigm [3] Modelling Tool.
This paper has been organized into the following sections. Section 98.2
describes the related work done in this field. Section 98.3 provides the proposed
approaches and implementation. Section 98.4 presents simulation and comparison
results. Finally, Sect. 98.5 includes the conclusion and future directions (Fig. 98.3).

98.2 State of the Art

There are various strategies for performing SCIA, These strategies are based on
some parameters which is considered during their requirements engineering pro-
cess the execution phase. These strategies can be broadly classified as following:
Automatable and Manual. These strategies require fewer infrastructures, but may
be harder in their impact estimation than the automatable ones.
1060 A. Gupta et al.

98.2.1 Traceability and Dependency Analysis

Antoniol et al. [4] and Corley et al. [5] proposes that traceability and dependency
analysis both involve examining relationships among entities in the software. They
differ in scope and detail level; traceability analysis is the analysis of relationships
among all types of models, while dependency analysis is the analysis of low-level
dependencies extracted from source code. By extracting dependencies from source
code, it is possible to obtain call graphs, control structures, data graphs and so on
[6]. Since source code is the most exact representation of the system [5], any
analysis based on it can very precisely predict the impact of a change. Dependency
analysis is also the most effective strategy for impact analysis available. The
drawback of using source code is that it is not available until late in the project [7],
which makes dependency analysis narrow in its field of application. The identi-
fication of the primary fundamental point of traceability is based on a predefined
search strategy and a multi-step selection process [8] When requirements trace-
ability exists down to the source, it can, however, be very efficient to use source
code dependencies in order to determine the impact of requirements changes. A
drawback is that very large systems have huge amounts of source code depen-
dencies, which make the dependency relationship difficult to use.

98.2.2 Program Slicing Technique

Slicing [9] attempts to understand dependencies using different independent slices

of the program. The whole program is sliced into a decomposition slice, which
contains the place of the change, and the rest of the program, a complement slice.
Slicing is based on data and control dependencies in the program. Changes made
to the decomposition slice around the variable that the slice is based on are
guaranteed not to affect the complement slice. Slicing technique limits the scope
for propagation of change and makes that scope explicit. Slicing technique is also
used for slicing of documents in order to account for ripple effects as a part of
impact analysis. Slicing techniques can be useful in requirements engineering to
isolate the impact of a requirements change to a specific part of the system. In
order to provide a starting point for the slicing technique, the direct impact of the
change must first be assessed.

98.2.3 Model Driven Development

Model-Driven Development (MDD) is a model-based software development

approach which aims at improving build time and quality of software artifacts by
focusing on UML models as abstract level artifacts rather than code [10].
98 Use Case Based Approach to Analyze Software … 1061

UML models can be defined at different levels of abstraction to represent various

aspects of the system. Dependencies between software life cycle objects are
becoming more complex as many software systems grow beyond a million of lines
of code. So, MDD is helpful in such type of systems. The potential benefits of
using models are significantly greater in software than in other engineering dis-
ciplines because of the potential for a seamless link between models and the
systems they represent.
In the context of MDD, models also used for many reasons such as, to handle
immediate errors, to add new functional requirements, to enhance some quality
aspects, or to adapt to a new technological or architectural environment. UML is a
standard for modelling software systems and it is extensively used in the area of
model-driven development [11]. For this reason, our work is restricted to UML
models. There are many UML Models e.g. class diagram, use case diagram,
sequence diagram, activity diagram etc. some author considered the hierarchies
between the models. Use case model is considered at a very abstract level UML
model, and this is very close to main functionality of the software system. Class
diagram and sequence diagram is considered the next level of UML models. So our
research is focused on use case diagram. A measure of distance between a changed
element and potentially impacted elements is proposed by Briand et al. [12] to
prioritize the results of impact analysis according to their likelihood of occurrence.

98.2.4 Informational Retrieval

SCIA approach in this thesis, is based on Information Retrieval discussed in [14],

used to derive the information contained in use-case flow events, with respect to
requested change. Information retrieval processes contains following steps: (1)
Building a corpus, (2) Natural-language processing (NLP), (3) Querying, and (4)
Estimating an impact set.

98.3 Proposed Approach

In the proposed approach, the use-case diagram with descriptions is used in SCIA.
A software system contains one or more use cases. Each use case has its main flow
of event and alternate flow of event, Details of core functionality of use cases in
defined in terms of flow events.
Dependency of use cases are utilized in the proposed approach. In the proposed
approach, flow of events of use cases is written by using the tool Visual Paradigm
Trial version software [3]. The proposed approach includes the steps shown in
Fig. 98.4.
Read SCRF: Users or Developer, who want some change in the developed
system, fill SCRF as per requirement. After getting the filled SCRF, Controller
1062 A. Gupta et al.

Fig. 98.4 Flow chart of

proposed approach

Configuration tool (CCT) will analyze the SCRF. The important field of the SCRF
i.e. the change requested field is stored in a new directory.
Parsing and Extraction: In this phase, change requested file is parsed. Parser
parses the stop words like is, are, am, this, that etc., from the stop word file. The
parsed keywords are stored in an output file. Impacted Use Cases: This phase is the
information retrieval phase. After parsing the SCRF, for all extracted keywords,
we search the flow of events directory and gives the impacted use case with
respected to each keyword, this process follows recursively run for all keywords.
For each flow of event file, respected use case name is stored.
Check Similar Use Cases: In this step similar use cases names are checked.
Delete Similar Use Cases: In previous step, it might be possible that, there may be
some redundant use cases. In this phase these redundant use cases are removed to
avoid the ambiguity. Final Impacted Use Cases: The final outcome of this step is
the name of impacted use cases.
98 Use Case Based Approach to Analyze Software … 1063

Table 98.1 Mapping of classes with use cases

Use-cases Mapped classes
Transaction Log, Logpanel, Receipt, Simreceiptprinter, Transaction, Inquiry, Balances,
Inquiry Message
Session Simcardreader, Simdisplay, Simcarderader, Card, Cardreader, Carddispencer,
Withdrawal Session, Simkeyboard, Accountinformation, Withdrawal, Money, Balance,
Deposit, Transfer
Transfer
Invalidpin Invalid, Authentication, Status, Checker

98.3.1 Implementation and Results

Proposed scheme is illustrated by the ten open source java based and some self
made projects.
The proposed approach has been implemented in three phases. To gather the
change request from the user or developer, SCRF is needed, thus in first phase a
SCRF is designed in java language. This SCRF form contains many fields like
priority, Type of change, Document affected etc.
In second phase, a NLP parser is written in java programming language. This
NLP parses the stop words (i.e. in computing, stop words are words which are
filtered out prior to, or after, processing of natural language data (text) [2]. There is
not one definite list of stop words which all tools used. Stop word list are updates
as per CCT decisions. More than 600 stop words are managed to parse. These stop
words will growing, based on decisions taken by the CCT. After parsing change
request we got some keywords as per desired change requested. To validate our
approach we have taken ten examples, these examples flow of events are written in
visual paradigm trial version software [3].
In third phase i.e. IR part, with the help of parsed keywords, dependent use case
flow of events are filtered with respect to change request. Extracted keywords of
change request are used and retrieve impacted use case names from the flow of
events of the system. For this, a C program is developed. Output of this program is
names of the impacted use cases with respected to the change request.

98.4 Comparison

Proposed approach is compared with Yin Li et al. [13] approach, which is based on
dependences between the system classes. For comparison, Li approach is applied
on the project that is used to evaluate proposed approach. Since, results achieved
through proposed approach are based on the use-case diagram, thus use-cases are
mapped with the classes. Table 98.2 maps the use-cases with classes of project
ATM. Table 98.1 filters the impacted classes with respect to the impacted use-
cases of project ATM for the change request ‘‘Ask for PAN validation if transfer or
1064 A. Gupta et al.

Table 98.2 Impacted use-cases identified through proposed approach and respected classes for
Project ‘‘ATM’’
Impacted use-cases Impacted classes
Transaction Log, Logpanel, Receipt, Simreceiptprinter, Transaction,
Withdrawal Accountinformation, withdrawal
Transfer Money, Balance, Deposit, transfer

Table 98.3 Comparison of proposed approach and Yin Li Existing approach impacted classes
for project ‘‘ATM’’
Proposed approach impacted classes Yin Li impacted classes
Log. Logpanel, Receipt Log. Logpanel, Receipt
Simreceiptprinter, Transaction Simreceiptprinter, Transaction
Accountinformation, Withdrawal, Money, Accountinformation, Withdrawal, Money,
Balance, Deposit, Transfer Balance, Deposit, Transfer

withdrawal is more than 3,000’’. Table 98.2 illustrates the impacted classes found
by proposed approach and Yin Li approach. After analysing the Table 98.3, it is
found that the results from proposed approach are equivalent to Li’s approach.
From this comparison, it is concluded that proposed approach is more efficient in
terms of abstraction.
The next section elaborates a case study to illustrate the reduction in test cases.

98.4.1 Estimation of Regression Testing Effort

from the Impact Set

When a change is made to software, the total effort made is the sum of modifi-
cation effort and testing effort. The testing effort can be reduced if we can identify
those test cases from the test suite that may be impacted due to the modifications
made. Other test cases need not be run again and again. Once we have the impact
set of use cases and hence the classes, when a change is made in any use case,
effort required for regression testing and total effort reduction in testing is com-
puted. Effort Required in Regression Testing (E0 ) = (Total number of test cases of
impacted classes /Total number of test cases in Test suite) * E

E  E0
% Reduction in Test Effort ¼  100
E

where E is the effort required to run the existing test suite. In the next section we
will analyze the area of the class diagram impacted because of the change class
and its impact on regression test effort (Table 98.4).
98 Use Case Based Approach to Analyze Software … 1065

Table 98.4 Classes and their impact area along with percentage reduction in effort required
during regression testing
Classes Number of use cases/ Number of test cases after % reduction of
classes impacted reduction test cases
Arithmetic 6 99 47
Equality 5 86 54
Bitwise 4 86 54
Additive 4 60 68
Multiplicative 3 47 75
Prefix 3 34 82
Shift 3 34 82
Unary 3 34 82
Compute 2 8 95

98.5 Simulation and Results

We demonstrate the proposed methodology work using a small application

‘‘Computer Operations Tutorial (COT)’’ developed in JAVA that takes a mathe-
matical expression as input, parses it and displays the solution as the output. The
exit condition for the application is to give the input ‘‘exit’’.
For COT, we have following classes whose names are stored in an array A:
A[ ] = {‘‘Additive’’, ‘‘Arithmetic’’, ‘‘Bitwise’’, ‘‘Compute’’, ‘‘Main’’, ‘‘Multi-
plicative’’, ‘‘Prefix’’, ‘‘Shift’’, ‘‘Unary’’, ‘‘Equality’’}.
Now suppose we make a change to the Equality class for implementing the
requested change, then the impact set (S) thus produced would be: Equality,
Bitwise, Compute, Multiplicative and Main. This can be verified by applying
Breadth First Search algorithm to find all connected components with the id of
Equality class as source, on matrix DM shown in Fig. 98.5. The zero value of
DM[i][j] represents the independency between class i and class j while the one
shows that the class j depends on the class i.
The requested change can impact the system in two ways: directly or indirectly,
as has been depicted in the dependency matrix shown above. When the equality
class is considered as the element of change set, the directly impacted class will be
bitwise, while compute, multiplicative and main will be indirectly impacted
classes. This can be easily observed from Fig. 98.5. The brown colored class (row
labelled E in matrix) is the initial class to which the change is requested. The green
colored class (row labelled B in matrix) represent the directly impacted classes.
Blue colored classes (row labelled M and N in matrix) are the indirectly impacted
classes, and the black colored classes are those that are unaffected by the change
that has been requested.
For a change requested in Equality classes, in the existing test suite there are
190 test cases, out of which we require to re-run only 86 test cases, thus required
1066 A. Gupta et al.

Fig. 98.5 Dependency

matrix for equality class (E)
of COT

regression testing effort for the change will be 0.46E where E is the effort required
to run the existing test suite and the percent of average reduction in testing effort
will be 54 % of the existing system testing effort.
As the number of impacted classes rises for a class, more the effort is required
for regression testing.

98.6 Conclusion and Future Work

Proposed approach reduces the effort that is unnecessary required in source code
based SCIA. With the help of proposed approach, impacted use cases are found in
an early stage. SCIA is important activity of SCM process. Software system doc-
umentation and coding is increasing in massive rate, it is very difficult to impact
analysis of whole system manually, for small systems manual SCIA might be
useful, but for Large System, SCIA is a tedious job. The approach used in this paper
finds the regression test effort as well as test suite required for regression testing
based on the impact set that is the sub set of the existing test suite of the system.
Obtained results have coarse granularity over the existing approaches. Benefits of
Proposed approach, at the use case level, save time and effort required in later stages
of change management of the system. Thus the proposed approach allows efficient
SCIA of the systems and reduces effort with a significant improvement over the
traditional methods. In the proposed approach we are using use case diagram for
impact analysis. To refine the SCIA, this approach may be applied in other UML
models like in class diagram, sequence diagram, activity diagram. To find the
98 Use Case Based Approach to Analyze Software … 1067

impact use case, search based information retrieval approach is used. By using
advance approaches like artificial intelligence etc., in information retrieval tech-
niques, SCIA will be more efficient and improved. The reduction in test effort
observed ranges from 47 to 95 %, saving significant software testing cost.

References

1. Arnold, R., Bohner, S.: Impact analysis-towards a framework for comparison. In Software
Maintenance,1993. CSM-93, Proceedings., Conference on, pp. 292–301, (1993)
2. Lehnert, S.: A taxonomy for software change impact analysis. In Proceedings of the 12th
International Workshop on Principles of Software Evolution and the 7th annual ERCIM
Workshop on Software Evolution, IWPSE-EVOL’11, ACM, pp. 41–50, New York, NY,
USA, (2011)
3. Visual paradigm: http://www.visual-paradigm.com/. [Online; accessed 03-June- 2013]
4. Antoniol, G., Canfora, G., Casazza, G., De Lucia, A., Merlo, E.: Recovering traceability links
between code and documentation. Soft. Eng. IEEE Trans. 28(10), 970–983 (2002)
5. Corley, C.S., Kraft, N.A., Etzkorn, L.H., Lukins, S.K.: Recovering traceability links between
source code and fixed bugs via patch analysis. In Proceedings of the 6th International
Workshop on Traceability in Emerging Forms of Software Engineering, pp. 31–37. ACM,
(2011)
6. Bohner, S.A.: A graph traceability approach for software change impact analysis. Ph.D.
Thesis, Fairfax, VA, USA, UMI Order No. GAX95-42995 (1995)
7. Khalil, A., Dingel, J.: Supporting the evolution of UML models in model driven software
developmeny: A Survey, Technical Report, School of Computing. Queens University,
Canada (2013)
8. Breivold, H.P., Crnkovic, I., Larsson, M.: A systematic review of software architecture
evolution research. Inf. Softw. Technol. 54(1), 16–40 (2012)
9. Tip, F.: A Survey of Program Slicing Techniques. Technical Report. CWI, Amsterdam, The
Netherlands (1994)
10. Göknil, A., Kurtev, I., van den Berg, K.G.: Change impact analysis based on formalization of
trace relations for requirements. In: ECMDA Traceability Workshop (ECMDA-TW),
pp. 59–75. Berlin, Germany (2008)
11. Fowler, M.: UML Distilled: A Brief Guide to the Standard Object Modeling Language, 3rd
edn. Addison-Wesley Longman Publishing Co., Inc, Boston (2003)
12. Briand, L.C., Labiche, Y., O’Sullivan, L.: Impact analysis and change management of uml
models. In Proceedings of the International Conference on Software Maintenance, ICSM’03,
IEEE Computer Society, pp. 256–,Washington, DC, USA, (2003)
13. Li, Y., Li, J., Yang, Y., Li, M.: Requirement-centric traceability for change impact analysis: a
case study. In: Wang, Q., Pfahl, D., Raffo, D. (eds.) Making Globally Distributed Software
Development a Success Story. Lecture Notes in Computer Science, vol. 5007, pp. 100–111.
Springer, Berlin Heidelberg (2008)
14. Marcus, Sergeyev, A., Rajlich, V., Maletic, J.I.: An information retrieval approach to concept
location in source code. In: Reverse Engineering, 2004. Proceedings. 11th Working
Conference on, pp. 214–223. IEEE, (2004)
Chapter 99
A Review of Image Segmentation
Methodologies in Medical Image

Lay Khoon Lee, Siau Chuin Liew and Weng Jie Thong

Abstract A precise segmentation of medical image is an important stage in

contouring throughout radiotherapy preparation. Medical images are mostly used
as radiographic techniques in diagnosis, clinical studies and treatment planning.
This review paper defines the limitation and strength of each methods currently
existing for the segmentation of medical images.

Keywords Advantages
Image segmentation
Medical image

99.1 Introduction

In medical fields nowadays, medical imaging is a crucial component in a many

applications. Such applications take place throughout the clinical track of events;
not only within diagnostic settings, but prominently in the area of preparation,
carrying out and evaluation before surgical operations.
Generally, image segmentation is the procedure of separating an image into
several parts. Instead of considering the whole data presented in an image all at
once, it is better to focus on a certain region-based semantic object in image
segmentation. Image segmentation has been widely implemented in medical
imaging to separate homogeneous area. Studied proposed by Liew et al. in 2012
shows that region of interest (ROI) segmentation plays a crucial role in multilevel
authentication [1]. Thus the goal of image segmentation is to find the regions that

L.K. Lee (&)
S.C. Liew
W.J. Thong

Faculty of Computer Systems and Software Engineering, Universiti Malaysia Pahang,
Lebuhraya Tun Razak, 26300 Gambang Kuantan, Pahang Darul Makmur, Malaysia
e-mail: [email protected]
S.C. Liew
e-mail: [email protected]
W.J. Thong
e-mail: [email protected]

Ó Springer International Publishing Switzerland 2015 1069

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_99
1070 L.K. Lee et al.

represent meaningful parts of objects for easier analyzation purpose .In this review
paper, author aims to gather and analyze methods used in image segmentation. So,
in general this paper will summarize suitable image segmentation methods to be
used for each types of medical images scan.

99.2 Method

In this section, several techniques that are being widely used on medical image
segmentation had been briefly described by the author. Segmentations are divided
mainly in four different techniques, which are thresholding-based, region-based,
edge-based, and clustering-based. Additionally there are also other methods for
image segmentations. The Fig. 99.1 below illustrates the types of image seg-
mentation available.

99.2.1 Thresholding-Based

99.2.1.1 Gray-Level Thresholding

The work of Beveridge [2] and his friends offered a decent example of a procedure
that integrates gray level thresholding which is a technique under the thresholding-
based image segmentation. In their paper, an input image can either be a grayscale
or colored image. The image is then divided into sectors of fixed size and fixed
location. An intensity histogram is calculated for each sector (and on colour images,
for each colour channel), and used to produce a local segmentation. For every
sector, information from its neighbors is used to detect clusters for which there may
not be enough local support due to the artificially induced partition of the image.

99.2.1.2 Ostu’s Method

In 1979, Ostu’s [3] research shows that thresholding method is based on a very simple
idea, which is to figure out the threshold that minimizes the weighted within-class
variance. In the experiment, Ostu’s method resulted in an advance in threshold-based
technique by converting a gray scale image into a binary image by automatically.

99.2.1.3 Gaussian Mixture Method

Gaussian mixture approach for image segmentation is a method to estimate the number
of components with their means and covariance sequentially without requiring any
initialization as proposed by Nicole and Alexander in 2012 [4]. The experiment
99 A Review of Image Segmentation Methodologies in Medical Image 1071

Fig. 99.1 Methodologies of

image segmentation

procedure starts from a single mixture component covering the whole data set and
sequentially splits incrementally during the expectation maximization steps. This
Gaussian mixture method successfully shows its effectiveness after many experiments.

99.2.2 Region-Based

99.2.2.1 Region Growing

In 2001, Cootes and Taylor proved that region growing method can be accom-
plished by adding an offset and by scaling to transform all gray values to zero-
mean and unit-variance. As an alternative to work with intensities, Cootes and
Taylor suggested a method which is storing the gradients’ direction and strength,
the latter is mapped non-linearly, this method proved to be better than the nor-
malized intensity. In 2002, Boasch et.al described a non-linear procedure in his
paper. The latter method proved especially useful in images with a strongly non-
Gaussian distribution as encountered, e.g. in ultrasound images. Studies continued
by Scott et al. in 2003, used gradient orientation, corner and edge strength for the
detection of vertebrae in dual energy X-ray images [5–7].
1072 L.K. Lee et al.

99.2.2.2 Region Splitting and Merging

Region splitting and merging technique is well known in region-based approach.

This technique is a combination of region splitting and merging. Manousakes et al.
in 1998 had [8] applied splitting and merging technique in trying to overcome the
difficulties occurred when using homogeneity measures. This experiment suc-
cessfully verified that the split-and-merge methods can be implemented in 2D and
3D MRI with a directionally uniform performance. Improvements involving
simulated annealing and boundary elimination can also be applied effectively in
3D or 2D MRI. The execution process is less time consuming.

99.2.3 Edge-Based

99.2.3.1 Edge Detection

In general edge detection methods are the process of identifying and locating sharp
discontinuities in an image. Edge detection is important for the object recognition
of human organs in medical images. In the years of 2006, Y.Q Zhang et al. [9]
introduced basic mathematical morphological theory and operations, the novel
mathematical morphological edge detection is proposed to detect the edge of lungs
CT image with salt -and -pepper noise. The experimental result shows that the
method proposed is more efficient for both medical image de-noising and edge
detection than the existing edge detection method.

99.2.3.2 Prewitt Edge Detection

According to Prewitt and J.M, 1970 [10], the Prewitt edge detector is an appro-
priate way to estimate the magnitude and orientation of an edge. Although dif-
ferential gradient edge detection needs a rather time consuming calculation to
estimate the orientation from the magnitudes in the x and y-directions, the compass
edge detection obtains the orientation directly from the kernel with the maximum
response. The Prewitt operator is limited to eight possible orientations; however
experiments show that most direct orientation estimations are not that accurate.
This gradient based edge detector is estimated in the 3 9 3 neighbourhood for
eight directions. All the eight convolution masks are calculated. One convolution
mask is then selected, namely that with the largest module. In this experiment,
Prewitt detector is able to clearly define the edges.
99 A Review of Image Segmentation Methodologies in Medical Image 1073

99.2.3.3 Laplacian of Gaussian

Laplacian of Gaussian was firstly introduced by Marr and Hildreth in 1980 who
combined Gaussian filtering with the Laplacian technique. This algorithm is not
often used in machine vision. Researchers that continued this method were Berzins
in 1984, Shah, Sood and Jain in 1986, Huertas and Medioni in 1986 [11, 12].

99.2.3.4 Watershed

Watershed is an image segmentation method introduced by S. Beucher and F. Meyer

in 1990 which separates overlapping objects [13]. They proposed using mathe-
matical morphology in image segmentation. The objective of the paper is to avoid
the problem of oversegmentation. This method involved two tools which are
watershed transform and the homotopy modification. The number of experiments
and the positive results were attached in the author’s published paper.

99.2.4 Clustering-Based

99.2.4.1 Fuzzy C-Mean Clustering Method

The fuzzy set theory was introduced by Zadeh, and successfully applied in image
segmentation. The fuzzy c-means algorithm was proposed by Bezdek in 1981
based on the fuzzy theory, it is the most widely studied and used algorithm in
image segmentation for its simplicity and the ability to obtain more information
from images. The studies was extended by many researchers, one of the latest
researchers’ studies was done by Krinidis and Chatzis in 2010. They proposed a
fuzzy local information c-means (FLICM) to overcome the problem of setting
parameters in the FCM-based methods. This algorithm uses both spatial and gray-
level local information, and is fully free of parameter adjustment (except for the
number of clusters) [14–16].

99.2.4.2 K-Means Clustering

According to Kaus et al. in 2004, [17] K-means Clustering Method is a process to

classify a given data set through a certain numbers of K-clusters. After clustering
all features into n classes with the k-means algorithm, each landmark is assigned to
the cluster containing the largest number of training features from that point. In the
research, the features sampled at a certain landmark only need to be compared with
the assigned cluster. Different appearance models for different boundary segments
are thus supported automatically.
1074 L.K. Lee et al.

99.2.4.3 Hierarchical Clustering

Hierarchical clustering is one of the edge based method developed by D. Cordes

et al. in 2002. The author had done a research by using hierarchical clustering to
measure connectivity in functional MRI. This method is able to detect similarities
of low-frequency fluctuations, and the results indicated that the patterns of func-
tional connectivity can be obtained with hierarchical clustering that resembles
known neuronal connections [18].

99.2.4.4 Mean-Shift

In the year of 2002, Comaniciu and Meer described a segmentation method based
on the mean-shift algorithm. The mean-shift algorithm by Cheng in 1995 is
designed to locate points of locally-maximal density in feature space. Feature
vectors containing gray-scale or colour information as well as pixel coordinates are
computed for each pixel [19, 20].

99.2.5 Other Method

99.2.5.1 Level-Set Method

Level-set method were presented by Osher and Sethian in 1988 then extended by
Malladi et al. in 1995. This paper features an implied shape representation and can
be employed with regional or edge-based features. It is slightly different with the
research by Leventon et al. whereby they extended the original energy formulation
by an additional term which deforms the contour towards a previously learned
shape model. A frequent criticism is that the signed distance maps which the shape
model is based on, do not form a linear space, which can lead to invalid shapes if
training samples vary too much. In 2006, Pohl et al. presented a method of
embedding the signed distance maps into the linear Log Odds space, which could
solve the modeling problems. To keep this review at a reasonable length, they had
to ignore level-set theory and techniques: The conceptual differences between the
implicit representation and the discrete models they intend to focus on would have
required a special treatment for all following sections [21–23].

99.2.5.2 Artificial Neural Network

In 2003, Pal and Pal [24] had done a research on ANNs method used for seg-
mentation. As what Pal and Pal have predicted, ANN becomes widely applied in
image processing. Lately the research of this method has been continued by Indira,
S.U., and Ramesh in the years of 2011 [25]. In this paper, an unsupervised method
99 A Review of Image Segmentation Methodologies in Medical Image 1075

Kohonen’s Self-Organizing Maps (SOM) form ANN and (Genetic Algorithm) GA

were combined to identify the main featured present in the image. SOM combined
with GA and some of the variants of SOM like the Variable Structure SOM
(VSSOM), Parameterless SOM (PLSOM) are compared and their performance is
evaluated. A new unsupervised, nonparametric method is developed by combining
the advantages of VSSOM and PLSOM. The experiments performed on the
satellite image shows that the modified PLSOM is efficient and the time taken for
the segmentation is less when compared to the other methods.

99.3 Modality

The imaging modalities employed can be divided into two global categories:
anatomical and functional. This paper only covered anatomical modalities. Images
are presented in 2D as well as in 3D domain. In the 2D domain each element is
called pixel, while in 3D domain it is called voxel [26]. Medical imaging is
performed in various modalities and in this paper the author will discuss images
such as CT, MRI, X-Ray and ultrasound.
CT (Computed tomography) is the image of sectional planes (tomography)
which are harder to interpret. CT can visualize small density difference, e.g. grey
matter, white matter and CSF. CT can detect and diagnose disease that cannot be
seen with X-ray. However it is more expensive than X-ray, lower resolution and
lower ionizing radiation [27].
Magnetic resonance imaging (MRI) is a sophisticated medical imaging tech-
nique that uses magnetic fields and radiofrequency to visualize the body’s internal
structures. Magnetic field gradients cause signals from different parts of the body
to have different frequencies. Signals collected with multiple gradients are pro-
cessed by computer to produce an image [27]. MR imaging of the body is per-
formed to get the structural details of brain, liver, chest, abdomen and pelvis which
helps in diagnosis or treatment.
In ultrasound imaging, computerized images are produced by sound waves
reflected by organs and other interior body parts in real time. It can be used for
interventional procedures. Ultrasound has no known harmful effects (at levels used
in clinical imaging). Ultrasound equipment is inexpensive, but many anatomical
regions cannot be visualized with ultrasound, for instance, head [28].
X-ray is the oldest non-invasive imaging of internal structures. It is rapid, short
exposure time and cheaper than other modalities. However X-ray is unable to
distinguish between soft tissues in head and abdomen. Real time X-ray imaging is
possible and used during interventional procedures. X-Ray contains ionizing
radiation which is a cause of cancer [29].
1076 L.K. Lee et al.

99.4 Discussion

Table 99.1 below summarizes the advantages and limitation of each methodology.

99.5 Analysis

In this section, Table 99.2 is presented as recommendations of methodologies for

CT scan, ultrasound, MRI and X-ray. Analysis is done according to the advantages
and disadvantaged of each methods has mentioned above.
Segmentation of CT scan usually involves three main image related problems;
noise that can alter the intensity of a pixel such that its classification becomes
uncertain, intensity inhomogeneity where the intensity level of single tissue class
varies gradually over the extent of the image, and image that have finite pixel size
and are subject to partial volume averaging where individual pixel volumes con-
tain a mixture of tissue classes so that the intensity pixel in the image may not be
consistent with any one class. For the segmentation of bone via CT scan, it is
recommended to use thresholding-method and region-based, or a combination of
region growing and watershed methodologies. By using these two methods, image
renderings are often used to provide detailed visualization of skeletal structure.
The main goal in brain MR segmentation is to segment gray matter, white
matter and cerebrospinal fluid. Segmentation is also used to find out the regions
corresponding to lesions tumors, cyst, edema, and other pathologies. Thus the
recommended method is clustering-based method. This is because this method is
based on the intensity of information, and a major concern is the presence of
intensity inhomogeneity.
For 3D MRI, it is slightly different with the 2D MRI, it is recommended to use
ANN or level-set method, due to complexity of dimensionality.
Thresholding-based method has been performed mostly in ultrasound for
extracting a variety of structures. The thresholding of intensity and texture sta-
tistics is used to segment ovarian cysts which will result in a good output, since is
simple to use.
Edge-based and watershed method is recommended for X-ray segmentation.
The edge-based methods works well for noise-free image, but the performance
degrades with noisy images or when fake and weak edges are present in the X-ray
imaging. Better result could be obtained when the edge-based method is combined
with watershed method.
99 A Review of Image Segmentation Methodologies in Medical Image 1077

Table 99.1 Advantages and limitation of methodologies

Methodologies Advantages Disadvantages
Thresholding- Gray level Easy to implement and Noisy and blurred edges.
based thresholding efficient. No need prior Lack of sensitivity and
-based information or calculation sharpness, hard to define,
complex for
multidimensional
Otsu’s Functions well and stably. False maxima may occur
method General: no specific during optimization function.
histogram shape assumed. The method tends to
Able to extend to multilevel artificially enlarge small
thresholding classes to obtain ‘better
separation’; small classes
might be merged and missed
Gaussian Relatively general histogram Many histograms are not
mixture model. When model is Gaussian. Intensities are
approach invalid, minimizes finite and non-negative.
classification error Difficult to detect close and
probability. Applicable to flat modes. Requires
small-size classes significant simplification of
model form extension to
multithresholding
Region-based Region Perform better in noise Seed point must be specified.
growing image. Easy to compute Costly. Images may be
undersegmentated or
oversegmentated
Region The image can be split May produce blocky
merging and progressively according to segments
splitting demanded resolution. Can
split the image by mean or
variance of segment pixel
value. The merging criteria
can be different to the
splitting criteria
Edge-based Edge Enclose large areas. Only applicable to simple
detection Applicable to images with background. closed contours
uneven illumination are not guaranteed
Prewitt edge Simplicity and able to detect Sensitivity to noise and
detection edges and their orientations inaccurate
Laplacian of Able to find the correct Malfunctioning at corners,
Gaussian places of edges. Can be used curves and where the gray
to test wider area around the level intensity function
pixel varies. Not finding the
orientation of edge because
of using the Laplcian filter
Watershed Perform better in noise Seed point must be specified
image. Fast speed and output and easily over-segmentated.
reliable Only applied on gradient and
time consuming
(continued)
1078 L.K. Lee et al.

Table 99.1 (continued)

Methodologies Advantages Disadvantages
Clustering- Fuzzy C- Simple and easy to Undefined optimal solution.
based mean understand Sensitive initialization
clustering process. Not compatible with
noisy data
Hierarchical Simple, output reliable. The Involves in detailed level, the
clustering process and relationships can fatal problem is computation
check the dendrogram. Only time. For the reason that
need to compute the hierarchical clustering
distances between each involves in detailed level, the
pattern, instead of fatal problem is the
calculating the centroid of computation time
clusters
K-mean Fast number of clustering is User has to select the number
clustering fixed. K-means algorithm is of desired output clusters
easy to implement. It is faster before starting to classify
than the hierarchical data. The result is sensitive
clustering to the selection of the initial
random centroids. Cannot
show the clustering details
Mean shift An extremely versatile tool The kernel bandwidth is the
for feature space analysis. only factor that can control
Suitable for arbitrary feature the output. The computation
spaces time is quite long
Other method Level set Versatile, robust, accurate, Require considerable thought
method and efficient in order to construct
appropriate velocities for
advancing the level set
function
Neural Applicable to a variety of Lack of profound theoretical
network problems. Easy to implement basis for ANNs. Problem in
based choosing the best
architecture. Black-box
problem

Table 99.2 Recommended Modalities Recommended method

methodologies
CT Scan Thresholding and region-based
Region-growing and watershed
3D MRI ANN
Level set method
MRI Clustering-based
Ultrasound Thresholding-based
X-ray Edge-based and watershed
99 A Review of Image Segmentation Methodologies in Medical Image 1079

99.6 Conclusion

In conclusion, this paper surveys on the existing methods of image segmentation

and recommends image segmentation methods to be applied on CT scan, 3D MRI,
MRI, ultrasound and X-Ray. The scope only includes digital watermarking in
medical image. Each image segmentation method has its own limitations. Hence
this paper can be used as a reference. However, there are certain limitations for
image segmentation in digital watermarking for medical images. The accuracy of
the segmentation remains the most concerned issue in determining critical cases
such as detection of tumor via medical imaging. Recommendation for future works
includes improving the accuracy and the speed of image segmentation for the
digital watermarking in medical imaging.

References

1. Liew, S.C., Liew, S.W., Zain, J.M.: Tamper localization and lossless recovery watermarking
scheme with ROI segmentation and multilevel authentication. J. Digit. Imaging 26(2),
316–325 (Springer) (2012). DOI: 10.1007/s10278-012-9484-4
2. Beveridge, J.R., Griffith, J., Kohler, R.R., Hanson, A.R., Rise-man, E.M.: Segmenting images
using localized histograms and region merging. Int. J Comput. Vision 2(3), 311–347 (1989)
3. Ostu, N.: A threshold selection method from gray-level histogram. IEEE Trans. Syst. Man
Cybern. SMC-8, 62–66 (1978)
4. Nicola, Greggio, et al.: Fast estimation of Gaussian mixture models for image segmentation.
Mach. Vis. Appl. 23(4), 773–789 (2012)
5. Cootes, T.F., Taylor, C.J.: On representing edge structure for model matching In:Proceedings
of the IEEE CVPR, vol. 1, (2001)
6. Bosch, J., Mitchell, S., Lelieveldt, B., Nijland, F., Kamp, O., Sonka, M., Reiber, J: Automatic
segmentation of echocardiographic sequences by active appearance motion models. IEEE
Trans. Med. Imaging (Cootes, T.F, Taylor, ging) 21(11), 1374–1383 (2002)
7. Scott, I.M., Cootes, T.F., Taylor, C.J.: Improving appearance model matching using local
image structure. In: Proceedings of the IPMI.LNCS, vol. 2732, Springer, Heidelberg (2003)
8. Manousakkas, I.N., Undrill, P.E., Cameron, G.G., Redpath, T.W.: Department of Biomedical
Physics and Bioengineering, University of Aberdeen, Foresterhill, Aberdeen, AB25 2ZD,
Scotland, United Kingdom Received December 30, (1996)
9. Yu-qian, Z., Wei-hua, G., Zhen-cheng, C., Jing-tian, T., Ling-Yun, L.: Medical images edge
detection based on mathematical morphology. In: Engineering in Medicine and Biology
Society, 2005. IEEE-EMBS 2005. 27th Annual International Conference of the IEEE,
pp. 6492–6495. (2006)
10. Prewitt, J.M.: Object Enhancement and Extraction, vol. 75. Academic Press, New York
(1970)
11. Albovik : Handbook of Image and Video Processing. Academic Press, New York (2000)
12. Haralick, R.M., Shapiro, L.G.: Computer and Robot Vision. vol. 1, Addition-Wesley
Publishing Company Inc., Boston (1992)
13. Beucher, S., Meyer, F.: The morphological approach to segmentation: The watershed
transform. In: Dougherty, E.R. (ed.) Mathematical Morphology in Image Processing, vol. 12,
pp. 433–481. Marcel Dekker, New York (1993)
14. Zadeh, L.: Fuzzy sets. Inform. Control 8, 338–353 (1965)
1080 L.K. Lee et al.

15. Bezdek, J.C.: Pattern Recognition with Fuzzy Objective Function Algorithms. Kluwer
Academic Publishers, Norwell (1981)
16. Krinidis, S., Chatzis, V.: A robust fuzzy local information c-means clustering algorithm.
IEEE Trans. Image Process. 19, 1328–1337 (2010)
17. Kaus, M.R., von Berg, J., Weese, J., Niessen, W., Pekar, V.: Automated segmentation of the
left ventricle in cardiac MRI. Med. Image Anal. 8(3), 245–254 (2004)
18. Cordes, D., Haughton, V., Carew, J.D., Arfanakis, K., Maravilla, K.: Hierarchical clustering
to measure connectivity in fMRI resting-state data. Magn. Reson. Imaging 20(4), 305–317
(2002)
19. Comaniciu, D., Meer, P.: Mean shift: a robust approach toward feature space analysis. IEEE
Trans. Pattern Anal. Mach. Intell. 24(5), 603–619 (2002)
20. Cheng, Y.: Mean shift, mode seeking and clustering. IEEE Trans. Pattern Anal. Mach. Intell.
17(8), 790–799 (1995)
21. Malladi, R., Sethian, J., Vemuri, B.: Shape modeling with front propagation: a level set
approach. IEEE Trans. Pattern Anal. Mach. Intell. 17(2), 158–174 (1995)
22. Leventon, M.E., Grimson, W.E.L., Faugeras, O.: Statistical shape influence in geodesic
active contours. In: Proceedings of the IEEE CVPR, vol. 1, (2000)
23. Pohl, K.M., Fisher, J., Shenton, M., McCarley, R.W., Grimson, W.E.L., Kikinis, R.,
Wells,W.M.: Logarithm odds maps for shape representation. In: Proceedings of the MICCAI.
LNCS, vol. 4191, Springer, Heidelberg (2006)
24. Pal, N.R., Pal, S.K.: A review on image segmentation techniques. Pattern Recogn. 26(9),
1277–1294 (1993)
25. Indira, S.U., Ramesh, A.C.: Image segmentation using artificial neural network and genetic
algorithm: a comparative analysis. In: Process Automation, Control and Computing (PACC),
2011 International Conference on IEEE, pp. 1–6. (2011)
26. Subach, B.R., et al.: Reliability and accuracy of fine-cut computed tomography scans to
determine the status of anterior interbody fusions with metallic cages
27. McRobbie, D.W.: MRI from Picture to Proton. Cambridge University Press, Cambridge
(2007)
28. Semelka, R.C., Armao, D.M., Elias, J., Huda, W.: Imaging strategies to reduce the risk of
radiation in CT studies, including selective substitution with MRI. J. Magn. Reson. Imaging
25(5), 900–909 (2007)
29. Spiegel, P.K.: The first clinical X-ray made in America—100 years. Am. J. Roentgenol.
(Leesburg, VA: American Roentgen Ray Society) 164(1), 241–242 (1995)
Chapter 100
The Utilization of Template Matching
Method for License Plate Recognition:
A Case Study in Malaysia

Norazira A. Jalil, A.S.H. Basari, Sazilah Salam, Nuzulha Khilwani

Ibrahim and Mohd Adili Norasikin

Abstract Automatic License plate detection and recognition system is special

form of optical character recognition and has been an active research domain in
image processing field. However, the accuracy is varied due to different styles of
number plates endorsed. Besides, the characters on Malaysia license plate are in one
or two lines. Thus, the proposed license plate recognition (LPR) technique of this
research is able to achieve the best recognition performance based on Malaysia
license plate vehicle registration number characters. This paper presents a study of
applying the template matching method for character image recognition. The
database of characters and license plate image has been created by collecting
images from various type of car. The initial pre-processing involves image
enhancement, binarization, filtering and segmenting of license plate. There are 100
license plates that contain 693 characters have been tested, and the result shown that
92.78 % of all characters is correctly recognized. Thus, template matching can be
classified as one of the promising algorithm for recognizing Malaysia license plate.

Keywords Malaysia license plate  Image processing  Template matching

N.A. Jalil (&)  A.S.H. Basari  S. Salam  N.K. Ibrahim  M.A. Norasikin
Center for Advanced Computing Technology (C-ACT), Faculty of Information and
Communication Technology, Universiti Teknikal Malaysia Melaka, Durian Tunggal,
Melaka, Malaysia
e-mail: [email protected]
A.S.H. Basari
e-mail: [email protected]
S. Salam
e-mail: [email protected]
N.K. Ibrahim
e-mail: [email protected]
M.A. Norasikin
e-mail: [email protected]

 Springer International Publishing Switzerland 2015 1081

H.A. Sulaiman et al. (eds.), Advanced Computer and Communication
Engineering Technology, Lecture Notes in Electrical Engineering 315,
DOI 10.1007/978-3-319-07674-4_100
1082 N.A. Jalil et al.

100.1 Introduction

The advanced of computer application processed more than textual data solving
everyday problems. Image processing techniques has been applied in many real-
life applications with great societal value. It has been applied in numerous domain
applications such as medical, security, engineering, forensic and biometric pur-
poses. The ability to process image or picture and translate it into something
meaningful has made the image processing fields become an active research area
until todays. A picture is worth a thousand words. Thus, with the great demands in
intelligent transportation system, the technology of image processing has adopted
for vehicle and traffic surveillance [1], managing vehicle parking system and
highway electronic toll collection. The fundamental issues in real-time license
plate recognition are the accuracy and the recognition speed in different envi-
ronmental conditions.
A license plate is the unique identification of a vehicle. License plate recognition
system is a complex image processing application which recognizes the characters
on a car plate based on the given conditions and situation. The license plate rec-
ognition system is installed in many places with multiple purposes and even polices
are using this application to detect speeding vehicles and monitoring surveillance
from distance away. These systems are beneficial because it can automate car park
management [2], improve the security of car park operator and the users as well.
Moreover, this system can eliminate the usage of swipe cards and parking tickets
that lead to green technologies [3], improve traffic flow during peak hours, detect
speeding cars on highways, and detect cars which run over red traffic lights. License
plate recognition system has been intensively studied worldwide; however, in
Malaysia cases, the implementation of these systems are less accurate due to the
different styles of car plates applied. The main reason is the developments of image
processing applications are still inadequate where it is unable to reach the 100 %
accuracy in recognition. Due to that, it is recommended that research is still con-
ducted for this application because of the importance of car plate recognition.
The Road Transport Department of Malaysia has endorsed a specification for
car plates that are allowed for use includes the font and size of characters that must
be followed by car owners. In Malaysia, the standard number plate has black color
for the background and white color for the font. All number plate is in the form of
vertical and horizontal with some specified font and size. Figure 100.1 below
shows samples of common and special Malaysian car plates. These various fonts
and sizes of characters will lead to problems during recognition phase.
One of the factors that contribute to the failure in achieving 100 % accuracy in
recognition was unable to recognize similar pattern characters. The common
confused characters such as in the case of recognizing character ‘B’ and ‘8’ or ‘3’,
‘0’ and ‘D’, ‘E’ and ‘F’, ‘G’ and ‘6’ or ‘Q’, ‘A’ as ‘4’, ‘6’ with ‘S’ and ‘7’ with ‘T’
or ‘1’. However, to avoid confusion between the letters and numbers, certain letters
are not used in the Malaysian registration system. The alphabets ‘I’ and ‘O’ are
omitted from the alphabetical sequences due to their similarities with the numbers
100 The Utilization of Template Matching Method for License Plate … 1083

Fig. 100.1 Vehicle number plate specification approved by Jabatan Pengangkutan Jalan
Malaysia (JPJ)

‘1’ and ‘0’. The alphabet ‘Z’ is omitted and reserved for use on Malaysian military
vehicles. Besides, there can be no leading zeroes in the number sequence.
Therefore, the experiments are conducted by using template matching to rec-
ognize the characters of license plate. Both vertical and horizontal Malaysia
license plate is applied as testing images.
The rest of the paper is organized as follows: the entire recognition processes
are briefly described in Sect. 100.2. Testing and results were discussed in
Sect. 100.3 and finally concluding remarks were given in Sect. 100.4.

100.2 Materials and Methods

Based on the previous studies [4, 5], the overview of image processing technique
in license plate recognition system includes 5 stages: (a) pre-processing, (b) fil-
tering, (c) feature extraction, (d) segmentation and (e) character recognition. The
final output of the sample experiment is to recognize the alphanumeric characters
on the license plate.

100.2.1 Preprocessing

Digital image preprocessing is an initial step to image processing improving the

data image quality for more suitable for visual perception or computational pro-
cessing. Preprocessing is an important step of applying a number of procedures for
smoothing and enhancing the image [6], removing background noise, normalizing
the intensity of individual image particles, image blurring and remove image
1084 N.A. Jalil et al.

reflections. This phase is used for making a digital image usable by subsequent
algorithm in order to improve their readability for the next phases. Preprocessing
for car license plate number uses three common sub processes [1], which are
geometric operation, gray scaling process and binarization process.

100.2.1.1 Geometric Operation

Geometric operation is a process to locate the car license plate. The purpose of this
operation is to localize the car plate for faster character identification over a small
region. The input image is first cropped manually to minimize the processing time.
Only the lower part of the input image that contains the required information will
be processed [1, 5, 7].

100.2.1.2 Grayscaling Process

Grayscaling is a process to produce shades of gray image from a multi-colour

image. Grayscale is usually the preferred format for image processing. Often, the
grayscale intensity is stored as an 8-bit integer giving 256 possible different shades
of gray from black to white. Basically, grayscale images are entirely sufficient for
many tasks and so there is no need to use more complicated and harder-to-process
colour images.
The method for the conversion of RGB to grayscale, as follows [8, 9]:

Intensity value ¼ 0:299  R þ 0:587  G þ 0:114  B ð100:1Þ

100.2.1.3 Binarization Process

Binarization is a process of converting grayscale image into black and white image
or ‘‘0’’ and ‘‘1’’. In digital image processing, binarization process enable of
removing unwanted information, detect the information of target area and increase
the processing speed [10]. Previously, the grayscale image consists of different
level of gray values; from 0 to 255. To improve the quality and extract some
information from the image, the image needs to be process a few times and thus
make the binary image more useful. Gray threshold value of an image is required
in the binarization process as it is important to determine whether the pixels that
having gray values will be converted to black or white.
The purpose of binarization is to identify the extent of objects and to con-
centrate on the shape analysis, where the shape of a region is more significant than
the intensity of pixels [6]. To avoid from data loss during the binarization process,
as both desired (characters) and undesired image (noise) will appeared; a variable
thresholding technique [11] is applied. This technique helps in determines the local
optimal threshold value for each image pixel in order to avoid problem originating
from non-uniform illumination [12].
100 The Utilization of Template Matching Method for License Plate … 1085

100.2.2 Filtering

In this stage, filtering process is used for blurring and for noise reduction while
preserving the sharpness of the image. Blurring is used in filtering process to
remove and clean-up the small noise and boundaries. In order to reduce the
background noise, softening is applied in low pass filter and thus making the image
blur.
A low-pass filter [13, 14] is also known as blurring or smoothing filter. The
method is used by calculating the average of a pixel and all of its eight immediate
neighbours. The result will replace the original value of the pixel. The process is
repeated for every pixel in the image.

100.2.3 Feature Extraction

In order to obtain an accurate location of the license plate, localization and

extraction of needed area from the vehicle image is necessary. This phase will give
a greatly affects for recognition rate and overall speed of the whole process. After
removing most of the unwanted object in an image, the next step is to search for
regions with high values which are most likely contain a license plate [15]. This is
done by constructing a horizontal and vertical projection profile of the edge image.
Due to the presence of some characters and text, the region containing the number
plate has a higher concentration of edges than the surrounding regions [16]. By
combining both horizontal and vertical projection [17], the location of the number
plate within the region of interest can be detected. The number of white pixels
presence in each row and column is counted. The vertical boundaries of the
number plate manifest as peaks in the graph of the vertical projection. The hori-
zontal boundaries of the number plate manifest as troughs in the graph of the
horizontal projection indicate the regions has strong edged and can be used for
horizontal position of license plate.

100.2.4 Image Segmentation

Segmentation is an important and crucial [15, 18] stage in License Plate Recog-
nition system that influences the accuracy of the whole system. The goal of
character segmentation is to find and segment the isolated characters on the plates,
without losing features of the characters. The correct segmentation process will
ease the process of character recognition. There are many factors that make the
character segmentation task difficult, such as image noise, plate frame, rivet, and
rotation and illumination variance. Thus, preprocessing is very important for
ensuring the good performance of character segmentation.
1086 N.A. Jalil et al.

Width = 15 pixels

Height = 24 pixels

Fig. 100.2 Size of segmented character

For Connected Component Analysis [19], the threshold of the plate image is
calculated and search for the connected components in the image, each connected
component will be assigned a special label in order to distinguish between different
connected components. Then, the character is resized to the standard height and
width in order to be used for recognition system as illustrated in Fig. 100.2.

100.2.5 Character Recognition

Character recognition is the most important task in recognizing the plate number
[6]. The recognition of characters has been a problem that has received much
attention in the fields of image processing, pattern recognition and artificial
intelligence. It is because there is a lot of possibility that the character produced
from the normalization step differ from the database. The same characters may
differ in sizes, shape and style [20] that could result in recognition of false char-
acter, and affect the accuracy and complexity of the whole system. In Malaysian
car plate, there are two groups of character, which is alphabet and numeric. It is
important for the system to differentiate the character correctly as sometimes the
system may confuse due to the similarities in the form of shape.

100.2.5.1 Template Matching

Template matching [1, 21, 22] is one of the most and common and easy classi-
fication methods for recognizing the segmented characters [23]. This technique is
applied by finding the small part of image that match with the template. The size of
each character image is normalized according to the template stored in the data-
base. Each character is compared to its corresponding pixel in the template and the
highest coefficient result is identified as the character of the input [24]. This
method has shown high accuracy but requires efficient searching method and needs
a large storage to save all the numbers and character templates [25].
However, due to processing time, the small size of template is commonly used
and thus may lead to inaccurate detection. Besides, slight deviations in shape, size,
100 The Utilization of Template Matching Method for License Plate … 1087

(a) (b)

(c) (d)

(e) (f)

(g)

Fig. 100.3 Proposed technique, a Input image, b Cropped input image, c Binary image,
d Filtered number plate, e Extracted number plate, f Segmented character, g Output from
template matching

and orientation, would prevent template matches from reading even the limited
0–9 numbers and A–Z English alphabets.
The recognition process using template matching is grouped into two main sub-
modules which are template creation and template matching. Template database
contain the characters that may be present in a license plate are from A to Y and
the digits that may also be present in a license plate are from 0 to 9. There are four
set of 33 alphanumeric characters (23 alphabets and 10 digits) is used for template
creation. The alphabets ‘I’, ‘O’ and ‘Z’ are omitted and reserved due to their
similarities with few digits and for Malaysian military vehicles purposes only. The
number of templates can be increased in order to increase the accuracy of the
license plate recognition system. Fitting approach is necessary for template
matching. For matching the characters with the database, input images must be
equal-sized with the database characters. The matching is done on a pixel by pixel
basis. In order to achieve optimum recognition rate, the template characters fit to
15 9 24 pixels is prepared. The segmented number plate region must be at the
same size of the template, although the real number plate region is bigger or
smaller than the template size [1]. The extracted characters segmented from plate
and the characters on database are now equal-sized [26].
The next step is template matching. Template matching is an effective algo-
rithm for recognition of characters. The character image is compared with the ones
1088 N.A. Jalil et al.

in the database and the highest correlation value is selected. Correlation is an

effective technique for image recognition [27]. This method measure the degree of
similarity corresponding pixel values of two images, template and input image. An
image with the highest correlation value means strong relation between input
image and templates that will produces the best match.
This type of comparison does not involve much computation. It is restricted to
the number of samples to be compared and the number of templates available in
the database, proportionate to which the duration of time taken to compare and
throw up a result would vary. Another limitation of the method is that when exact
matches are not found the probability of misinterpreted increases. In this method
the computational time increases as the number of sample increases [28].
Figure 100.3 shows the detail process flow.

100.3 Results and Discussion

In this experiment, 693 characters from 100 license plate images were employed
and the size of each segmented images has been resized to 15 9 24 pixels. For
these images, all of them were taken by digital camera from various scenes and
under different lighting conditions of the real world, varied distances from the
vehicle and weather. The distance between the camera and the vehicle varied from
3 to 7 m. The success rate has been achieved where the accuracy of license plate
recognition using template matching yield up to 92.78 %.
From 693 characters tested, 50 have been wrongly recognized. Out of 33
characters, there are 11 characters that are wrongly recognized which are B, G , M,
Q, U, 0, 3, 5, 6, 8 and 9. Out of 50 errors, number 8 and 9 has the highest
misrecognized characters where system tends to recognize as 6 and 0 respectively.
Other common misrecognized characters are M and 0. The failure of this section is
caused by some defectives on input image that lead to failure. Based on the result,
the recognition accuracy and efficiency of the system can be increases with more
number of training samples.
The highest degree of similarity between each character is 1. Thus, based on the
experiment, even though the degree of similarity closes to 1, yet there are errors
detected. The highest degree of misrecognizes character is 0.8157 while the lowest
is 0.4005.

100.4 Conclusion

In this paper, a simple yet efficient method of detecting and recognizing characters
image in Malaysia license plate has been presented. The proposed method utilizes
template matching to perform the recognition. Firstly, pre-processing and filtering
is conducted to remove the unwanted images. Next, extraction is used to locate the
100 The Utilization of Template Matching Method for License Plate … 1089

position of license plate characters, and then segmentation is done to separate the
plate characters individually. Finally, template matching with the use of correla-
tion is applied for recognition of plate characters. The performance of template
matching with Malaysia license plate has been tested and evaluated. The result has
shown that the developed algorithm works well and has gained the high accuracy
value up to 92.78 %.
As future work, it will be interesting to test the method with more number of
English alphabet and numeric samples. Another suggestion is to improve the
performance of template matching by hybridizing with back-propagation neural
network.

Acknowledgments This work is part of Master by research in Information and Communication

Technology supported by MyBrain Scheme of Universiti Teknikal Malaysia Melaka and the
Malaysian Technical University Network—Centre of Excellence (MTUN CoE) funding grant,
MTUN/2012/UTeM-FTMK/11 M00019.

References

1. Zakaria, M.F., Suandi, S.A.: Malaysian car number plate detection system based on template
matching and colour information. Int. J. Comput. Sci. Eng. 02(04), 1159–1164 (2010)
2. Ibrahim, N.K., Kasmuri, E., Norazira, A.: License plate recognition (LPR): a review with
experiments for Malaysia case study. Int. J. Soft Comput. Softw. Eng. 3(3), (2013)
3. Tare, S.R.: Review paper on carpooling using android operating system-a step towards green
environment. Int. J. Adv. Res. Comput. Sci. Softw. Eng. 3(4), 54–57 (2013)
4. Dandu, B.R., Chopra, A.: Vehicular number plate recognition using edge detection and
characteristic analysis of national number plates. Int. J. Comput. Eng. Res. 2(3), 795–799
(2012)
5. Teo, L.A.K.T.K., Wong, F.: Smearing algorithm for vehicle parking management system. In:
Proceedings of the 2nd Seminar on Engineering and Information Technololgy, pp. 331–337.
(2009)
6. Indira, B., Shalini, M., Murthy, M.V.R., Shaik, M.S.: ‘‘Classification and recognition of
printed hindi characters using artificial neural networks. Int. J. Image Graph. Signal Process.
4(6), 15–21 (2012)
7. Mousa, A.: Canny edge-detection based vehicle plate recognition. Int. J. Signal Process.
Image Process. Pattern Recognit. 5(3), 1–8 (2012)
8. Sedighi, A., Vafadust, M.: A new and robust method for character segmentation and
recognition in license plate images. Expert Syst. Appl. 38, 13497–13504 (2011)
9. Dashtban, M.H.: A novel approach for vehicle license plate localization and recognition. Int.
J. Comput. Appl. 26(11), 22–30 (2011)
10. Jin, L., Xian, H., Bie, J., Sun, Y., Hou, H., Niu, Q.: License plate recognition algorithm for
passenger cars in Chinese residential areas. Sensors (Basel) 12(6), 8355–8370 (2012)
11. Mukherjee, A.: Enhancement of image resolution by binarization. Int. J. Comput. Appl.
10(10), 15–19 (2010)
12. Sivanandan, S., Saiyyad, Y.: Automatic vehicle identification using license plate recognition
for Indian vehicles. Int. J. Comput. Appl. 2012, 23–28 (2012)
13. Goyal, R., Kaur, A.: A Review of optimal binarization techniques on documents with
damaged background. Int. J. Comput. Sci. Technol. 4333, 237–239 (2011)
1090 N.A. Jalil et al.

14. Sun, G., Sun, X., Han, X.: A new method for edge detection based on the criterion of
separability. J. Multimed. 6(1), 66–73 (2011)
15. Gaur, S.B.C.: Comparison of edge detection techniques for segmenting car license plates. Int.
J. Comput. Appl. Electron. Inf. Commun. Eng. 5, 8–12 (2011)
16. Li, P., Connan, J.: Numberplate detection using double segmentation. In: Proceedings of the
2010 Annual Research Conference of the South African Institute for Computer Scientists and
Information Technologists—SAICSIT’10, pp. 386–389. (2010)
17. Zheng, L., He, X., Samali, B., Yang, L.T.: An algorithm for accuracy enhancement of license
plate recognition. J. Comput. Syst. Sci. 79(2), 245–255 (2013)
18. Akter, S.: Automatic license plate recognition (ALPR) for Bangladeshi vehicles. Glob.
J. Comput. Sci. Technol. 11(21), (2011)
19. Yilmaz, K.: A smart hybrid license plate recognition system based on image processing using
neural network and image correlation. In: 2011 International Symposium on Innovations in
Intelligent System and Applications (INISTA), pp. 148–153. (2011)
20. Prasad, K., Nigam, D.C., Lakhotiya, A., Umre, D., Durg, B.I.T.: Character recognition using
matlab’s neural network toolbox. Int. J. u- e- Serv. Sci. Technol. 6(1), 13–20 (2013)
21. Adebayo Daramola, T.F.S., Adetiba, E., Adoghe, A.U., Badejo, J.A., Samuel, I.A.:
Automatic vehicle identification system using license plate. Int. J. Eng. Sci. Technol. 3(2),
1712–1719 (2011)
22. Kranthi, S., Pranathi, K., Srisaila, A.: Automatic number plate recognition. Int. J. Adv.
Technol. 2(3), 408–422 (2011)
23. Reshma, P.: Noise removal and blob identification approach for number plate recognition.
Int. J. Comput. Appl. 47(8), 13–17 (2012)
24. Mahalakshmi, T., Muthaiah, R., Swaminathan, P., Nadu, T.: Review article : an overview of
template matching technique in image processing. Res. J. Appl. Sci. Eng. Technol. 4(24),
5469–5473 (2012)
25. Khalil, M.I.: Car plate recognition using the template matching method. Int. J. Comput.
Theory Eng. 2(5), 3–7 (2010)
26. Ibrahim, N.K., Kasmuri, E., Jalil, N.A.: A review on license plate recognition with
experiments for Malaysia case study. Middle-East J. Sci. Res. 14(3), 409–422 (2013)
27. Gilly, D., Raimond, D.K.: License plate recognition- a template matching method. Int. J. Eng.
Res. Appl. 3(2), 1240–1245 (2013)
28. Suvarna, M.: Diagnosis of burn images using template matching, k-nearest neighbor and
artificial neural network. Int. J. Image Process. 7, 191–202 (2013)