(navigation image)
Home American Libraries | Canadian Libraries | Universal Library | Community Texts | Project Gutenberg | Children's Library | Biodiversity Heritage Library | Additional Collections
Search: Advanced Search
Anonymous User (login or join us)
Upload
See other formats

Full text of "Journal of Computer Science and Information Security, January 2010"

UCSIS Vol. 7, No. 1, January 2010 
ISSN 1947-5500 



International Journal of 
Computer Science 
& Information Security 



© IJCSIS PUBLICATION 2010 



IJCSIS Editorial 

Message from Managing Editor 

IJCSIS Volume 7 No. 1 January 2010 issue presents high-quality e-publications 
(with acceptance rate of ~ 28%) that deals will all aspects of computer science. 
Paper selection undergoes journal-style peer review process. The editorial board 
and technical review committee contain some of the most renowned specialists 
in their areas of expertise. In this way, both the high standard of the published 
papers as well as a broad spectrum of publications is guaranteed. 

In this issue, you will find high-quality contributions from researchers in the 
diverse area of computer science, networking, information retrieval, emerging 
technologies and information security. With a continuing open-access policy, we 
are pleased to disseminate this collection of remarkable computer science 
research works with our readers. 

Special thanks to our reviewers and technical sponsors for their valuable service. 



Available at http:/ / sites.qooqle.com/ site/ iicsis/ 

IJCSIS Vol. 7, No. 1, 
January 2010 Edition 
ISSN 1947-5500 
© IJCSIS 2010, USA. 

Indexed by (among others): 

— Saentffktom mms 

Google schoEar — rit**<:o»r* OwllUb 



"^■^MiTti *n&irm tor scifr>ct 



IJCSIS EDITORIAL BOARD 



Dr. Gregorio Martinez Perez 

Associate Professor - Professor Titular de Universidad, University of Murcia 
(UMU), Spain 

Dr. M. Emre Celebi, 

Assistant Professor, Department of Computer Science, Louisiana State University 
in Shreveport, USA 

Dr. Yong Li 

School of Electronic and Information Engineering, Beijing J iaotong University, 
P. R. China 

Prof. Hamid Reza Naji 

Department of Computer Enigneering, Shahid Beheshti University, Tehran, Iran 

Dr. Sanjayjasola 

Professor and Dean, School of Information and Communication Technology, 
Gautam Buddha University 

Dr Riktesh Srivastava 

Assistant Professor, Information Systems, Skyline University College, University 
City of Sharjah, Sharjah, PO 1797, UAE 

Dr. Siddhivinayak Kulkarni 

University of Ballarat, Ballarat, Victoria, Australia 

Professor (Dr) Mokhtar Beldjehem 

Sainte-Anne University, Halifax, NS, Canada 

Dr. Alex Pappachen J ames, (Research Fellow) 

Queensland Micro-nanotechnology center, Griffith University, Australia 



TABLE OF CONTENTS 

1. Paper 06011051: Scenario Based Worm Trace Pattern Identification Technique (pp. 001-009) 

Siti Rahayu S., Robiah Y., Shahrin S., Mohd Zaki M., Irda R., Faizal M. A., 

Faculty of Information and Communication Technology, Univeristi Teknikal Malaysia Melaka, Durian 

Tung gal, Melaka, Malaysia 

2. Paper 07011068: Avoiding Black Hole and Cooperative Black Hole Attacks in Wireless Ad hoc 
Networks (pp. 010-016) 

Abderrahmane Baadache, Laboratory of Industrial Technology and Information, University of A. Mira, 
Targua Ouzemour, 06000, Bejaia, Algeria. 

Ali Belmehdi, Laboratory of Industrial Technology and Information, University of A. Mira, Targua 
Ouzemour, 06000, Bejaia, Algeria. 

3. Paper 06011053: Design of Current Controller for Two Quadrant DC Motor Drive by Using 
Model Order Reduction Technique (pp. 017-024) 

K. Ramesh, EEE Department, Velalar College ofEngg. & Tech., Erode - 638012, India 
K. Ayyar, EEE Department, Velalar College ofEngg. & Tech., Erode - 638012, India 
Dr. A. Nirmalkumar, EEE Department, BIT, Sathyamangalam, India 
Dr. G. Gurusamy, EEE Department, BIT, Sathyamangalam, India 

4. Paper 09011080: Wireless Congestion Control Protocol For Multihop Ad Hoc Networks (pp. 025- 
031) 

Mahendra kumar. S, Department of Electronics and Communication Engineering, Velalar College of 
Engineering and Technology, Tamil Nadu, India. 

Senthil Prakash. K, Department of Electronics and Communication Engineering, Velalar College of 
Engineering and Technology, Tamil Nadu, India. 

5. Paper 06011058: Saturation Throughput Analysis of IEEE 802.11b Wireless Local Area Networks 
under High Interference Considering Capture Effects (pp. 032-039) 

Ponnusamy Kumar, Department of Electronics and Communication Engineering, K. S. Rangasamy 
College of Technology, Tiruchengode, Namakkal, Tamilnadu, India. 

A.Krishnan, Department of Electronics and Communication Engineering, K.S. Rangasamy College of 
Technology, Tiruchengode, Namakkal, Tamilnadu, India. 

6. Paper 06011057: Performance Evaluation of Unicast and Broadcast Mobile Ad-hoc Network 
Routing Protocols (pp. 040-046) 

Sumon Kumar Debnath, Dept. of Computer Science and Telecommunication Engineering, Noakhali 

Science and Technology University, Bangladesh 

Foez Ahmed, Dept. of Networks and Communication Engineering, College of Computer Science, King 

Khalid University, Kingdom of Saudi Arabia 

Nayeema Islam, Department of Information & Communication Engineering, Rajshahi University, 

Rajshahi- 6205, Bangladesh 

7. Paper 07011069: Deriving Relationship Between Semantic Models - An Approach for cCSP (pp. 
047-054) 

Shamim H. Ripon, Department of Computing Science, University of Glasgow, UK 

Michael Butler, School of Electronics and Computer Science, University of Southampton, UK 

8. Paper 14120912: The Importance Analysis of Use Case Map with Markov Chains (pp. 055- 062) 

Y aping Feng, Dept. of Computer engineering, Kumoh National Institute of Technology, Korea 
Lee-Sub Lee, Dept. of Computer engineering, Kumoh National Institute of Technology, Korea 

9. Paper 30120931: Convergence of Corporate and Information Security (pp. 063-068) 

Syed (Shawon) M. Rahman, PhD, Assistant Professor of Computer Science 



University of Hawaii-Hilo, HI, USA and Adjunct Faculty, Capella University, MN, USA Email: 

Shannon E. Donahue, CISM,CISSP, Ph. D. Student, Capella University, 225 South 6th Street, 9th Floor 

Minneapolis, MN 55402, USA 

10. Paper 01011035: Image Retrieval Techniques based on Image Features: A State of Art approach 
for CBIR (pp. 069-076) 

Mr. Kondekar V. H., Department of Electronics & Telecommunication Engineering,Walchand Institute of 
Technology, Solapur, Solapur University. 

Mr. Kolkure V. S., Bharat Ratna Indira Gandhi Collage of Engineering, Solapur, Solapur University. 
Prof. Kore S.N., Walchand College of Engineering, Sangali. Shivaji University. 

11. Paper 01011037: AHB Compatible DDR SDRAM Controller IP Core for ARM BASED SOC (pp. 

077-085) 

Dr. R. Shashikumar, C. N. Vijay Kumar, M. Nagendrakumar, ECE dept, SJCIT, Chikkaballapur, 

Karnataka, India 

C S. Hemanthkumar, JVIT, Bidadi 

12. Paper 01011038: High Throughput of WiMAX MIMO-OFDM Including Adaptive Modulation 
and Coding (pp. 086-091) 

Had] Zerrouki, Mohamed Feham 

Laboratoire de Systemes de Technologies de I'Information et de Communication (STIC), University Abou 

Baker Belkaid, Tlemcen, Algeria. 

13. Paper 02011039: Performance Modeling and Evaluation of Traffic management for Mobile 
Networks by SINR Prediction (pp. 092-094) 

K. K. Guatam, Department of Computer Science & Engineering, Roorkee Engineering & Management 

Technology Institute, Shamli (247774) India 

Anurag Rai, Department of Information Technology, College of Engineering Roorkee, Roorkee (247667) 

India 

14. Paper 03011041: Thai Rhetorical Structure Analysis (pp. 095-105) 

Somnuk Sinthupoun, Department of Computer Science, Maejo University, Chiangmai, Thailand 50290 
Ohm Sornil, Department of Computer Science, National Institute of Development Administration, Bangkok, 
Thailand 10240y 

15. Paper 04011042: Mobility Impact on Performance of Mobile Grids (pp. 106-111) 

A. S. Nandeppanavar, Department of ISE, Basaveshwar Engineering College, Bagalkot-587102, India 
M. N. Birje, Department of ISE, Basaveshwar Engineering College, Bagalkot-587102, India 
S. S. Manvi, Department ofE&CE, Reva Institute of Technology & Management, Bangalore, India 
Shridhar, Department of E&CE, Basaveshwar Engineering College, Bagalkot-587102, India 

16. Paper 04011043: Analysis of Birth weight using Singular Value Decomposition (pp. 112-115) 

D. Nagarajan, Department of Mathematics, Salalah College of Technology, Salalah, Sultanate of Oman. 

P. Sunitha, Department of Mathematics, S. T. Hindu College, Nagercoil, Kanyakumari, Tamil Nadu, India 

V. Nagarajan, Department of Mathematics, S. T. Hindu College, Nagercoil, Kanyakumari, Tamil Nadu, 

India. 

V. Seethalekshmi, Department of Mathematics, James College Technology, Nagercoil, Tamil Nadu, India 

17. Paper 04011044: A Simple Method of Designing Dual loop Controller for Cold Rolling Mill 
(pp. 116-120) 

S. Umamaheswari, Dept. ofEIE, Mahendra Engg, College, Namakkal(Dt), India. 
V. Palanisamy, Principal/ Info Institute of Technology, Coimbatore, India. 
M. Chidambaram, Director/ NIT, Trichy, India. 

18. Paper 05011046: Detection of Microcalcification in Mammograms Using Wavelet Transform and 
Fuzzy Shell Clustering (pp. 121-125) 



T. Balakumaran, Department of Electronics and communication Engineering, Velalar College of 

Engineering and Technology, Erode, TamilNadu, India. 

Dr. ILA. Vennila, Department of Electrical and Electronics Engineering, PSG College of Technology, 

Coimbatore, TamilNadu, India. 

C Gowri Shankar, Department of Electrical and Electronics Engineering, Velalar College of Engineering 

and Technology, Erode, TamilNadu, India. 

19. Paper 06011048: The Fast Haar Wavelet Transform for Signal & Image Processing (pp. 126-130) 

V. Ashok, Department ofBME, Velalar College ofEngg.&Tech., Erode, India - 638012 

T. Balakumaran, Department ofECE, Velalar College ofEngg.&Tech, Erode, India - 638012 

C Gowrishankar, Department ofEEE, Velalar College ofEngg.&Tech, Erode, India - 638012 

Dr. ILA.Vennila, Department of ECE, PSG College of Technology, Coimbatore, TamilNadu, India 

Dr. A. Nirmal kumar, Department of EEE, Bannari Amman Institute of Technology, Sathyamangalam, 

TamilNadu, India 

20. Paper 06011054: A Survivability Strategy in Route Optimization Mobile Network by Memetic 
Algorithm (pp. 131-134) 

K. K. Guatam, Department of Computer Science & Engineering, Roorkee Engineering & Management 

Technology Institute, Shamli (247774) India 

Anurag Rai, Department of Information Technology, College of Engineering Roorkee, Roorkee (247667) 

India 

21. Paper 06011055: Analysis of Large-Scale Propagation Models for Mobile Communications in 
Urban Area (pp. 135-139) 

M. A. Alim, M. M. Rahman, M. M. Hossain, A. Al-Nahid 

Electronics and Communication Engineering Discipline, Khulna University, Khulna 9208, Bangladesh. 

22. Paper 06011056: Performance Evaluation of TCP over Mobile Ad-hoc Networks (pp. 140-146) 

Foez Ahmed, College of Computer Science, King Khalid University, Abha, Kingdom of Saudi Arabia 

Sateesh Kumar Pradhan, College of Computer Science, King Khalid University, Abha, Kingdom of Saudi 

Arabia 

Nayeema Islam, Department of Information & Communication Engineering, Rajshahi University, 

Rajshahi- 6205, Bangladesh 

Sumon Kumar Debnath, Dept. of Computer Science and Telecommunication Engineering, Noakhali 

Science and Technology University, Bangladesh 

23. Paper 07011060: Vision Based Game Development Using Human Computer Interaction (pp. 147- 
153) 

Ms. S. Sumathi, Bharath University, Chennaijndia 

Dr. S. K. Srivatsa, St. Joseph College of Engineering, Chennaijndia 

Dr. M. Uma Maheswari, Bharath University, Chennaijndia 

24. Paper 07011061: Path Traversal Penalty in File Systems (pp. 154-159) 

M.I. Lali, F. Ahsan and A.F \M Ishaq 

Dept. of Computing Science, CUT, Islamabad 

25. Paper 07011066: Using Statistical Moment Invariants and Entropy in Image Retrieval (pp. 160- 
164) 

Ismail I. Amr, College of Computers and Informatics, Misr International University, Cairo, Egypt 
Mohamed Amin , Passent El-Kafrawy , and Amr M. Sauber, Faculty of science Department of Math and 
Computer Science, Menoufia University, Shebin-ElKom, Egypt 

26. Paper 07011067: Genetic Algorithm Based Optimization of Clustering in Ad-Hoc Networks (pp. 
165-169) 

Bhaskar Nandi, Subhabrata Barman, Soumen Paul, Haldia Institute of Technology, W.B, India. 



27. Paper 07011070: Multi-Product Inventory Optimization using Uniform Crossover Genetic 
Algorithm (pp. 170-179) 

S. Narmadha, Assistant Professor, Department of Computer Science and Engineering, Park College of 

Engineering and Tekhnology, Coimbatore - 641659, Tamilnadu, India 

Dr. V. Selladurai, Professor and Head, Department of Mechanical Engineering, Coimbatore Institute of 

Technology, Coimbatore - 641014, Tamilnadu, India 

G. Sathish, Research Scholar, Department of Computer Science and Engineering, Anna University - 

Coimbatore, Tamilnadu, India 

28. Paper 07011071: Efficient Inventory Optimization of Multi Product, Multiple Suppliers with 
Lead Time using PSO (pp. 180-189) 

S. Narmadha, Assistant Professor, Department of Computer Science and Engineering, Park College of 

Engineering and Tekhnology, Coimbatore - 641659, Tamilnadu, India 

Dr. V. Selladurai, Professor and Head, Department of Mechanical Engineering, Coimbatore Institute of 

Technology, Coimbatore - 641014, Tamilnadu, India 

G. Sathish, Research Scholar, Department of Computer Science and Engineering, Anna University - 

Coimbatore, Tamilnadu, India 

29. Paper 07011072: Test Case Generation using Mutation Operators and Fault Classification (pp. 
190-195) 

Mrs. R. Jeevarathinam, Department of Computer Science, SNR Sons College, Coimbatore, Tamilnadu, 

India. 

Dr. Antony Selvadoss Thanamani, Associate Professor and Head, Department of Computer Science, NGM 

College, Pollachi, Tamilnadu, India. 

30. Paper 14120913: An Energy Efficient and Reliable Congestion Control Protocol For Multicasting 
In Mobile Adhoc Networks (pp. 196-201) 

G Sasi Bhushana Rao, Senior Professor, ECE Department, Andhra University, Visakhapatnam 

M. RajanBabu, Associate Professor, ECE Department, Lendi Institute of Engineering and Technology, 

Jonnada, Vizianagaram, AndhraPradesh, India 

31. Paper 07110911: An Intelligent System For Effective Forest Fire Detection Using Spatial Data (pp. 

202-208) 

K. Angayarkkani, Senior lecturer, D.G Vaishnav College, Arumbakkam, Chennai, India. 

N. Radhakrishnan, Geocare Research Foundation, #23/30, First main Road, Pammal, Chennai, India 

32. Paper 07120905: Performance Analysis and Optimization of Lumped Parameters of Electrostatic 
Actuators for Optical MEMS Switches (pp. 209-215) 

D. Mohana Geetha, Department of Electronics and Communication Engineering, Kumaraguru college of 
Technology, Coimbatore641006, India. 

M. Madheswaran, Center for Advanced Research, Department of Electronics and Communication 
Engineering, Muthayammal Engineering College, Rasipuram 637408 , India. 

33. Paper 08011075: Modeling of Human Criminal Behavior using Probabilistic Networks (pp. 216- 
219) 

Ramesh Kumar Gopala Pillai, Research Scholar, R. V. Center for Cognitive Technologies, Bangalore, India 
Dr. Ramakanth Kumar .P, Professor, R.V. Center for Cognitive Technologies, Bangalore, India 

34. Paper 08120906: Reaching the Unreached - A Role of ICT in Sustainable Rural Development (pp. 
220-224) 

Mr.Nayak S.K., Head, Dept. of Computer Science, Bahirji Smarak Mahavidyalaya, Basmathnagar, 
Dist.Hingoli. (MS), India 

Dr. S. B. Thorat, Director, Institute of Technology and Mgmt. Nanded, Dist.Nanded. (MS), India 
Dr.Kalyankar N.V., Principal, Yeshwant Mahavidyalaya, Nanded Nanded (MS), India 

35. Paper 08120909: A proof Procedure for Testing Membership in Regular Expressions (pp.225-227) 



Keehang Kwon and Hong Pyo Ha, Dong- A University Department of Computer Engineering 

Busan, Republic of Korea 

Jiseung Kim, Kyung-IL University Department of Industrial Engineering, Daegu, Republic of Korea 

36. Paper 09011078: Impact of Random Loss on TCP Performance in Mobile Ad-hoc Networks 
(IEEE 802.11): A Simulation-Based Analysis (pp. 228-233) 

Shamimul Qamar, Department of Computer Science, Colloge of Artsv & Science, King Saud University, 

Riyadh, Kingdom of Saudi Arabia 

Kumar Manoj, IIT Roorkee, Saharanpur campus, Saharanpur, India 

37. Paper 16120916: Automatic diagnosis of retinal diseases from color retinal images (pp. 234-238) 

D. Jayanthi, PG Scholar, Dept of IT, Sri Venkateswara college of Engineering 
N Devi, Senior Lecturer, Dept of IT, Sri Venkateswara college of Engineering 
S. SwarnaParvathi, Senior Lecturer, Dept of IT, Sri Venkateswara college of Engineering 

38. Paper 21120921: Changing Neighbors k-Secure Sum Protocol for Secure Multi-Party 
Computation (pp. 239-243) 

Rashid Sheikh, & Beerendra Kumar, SSSIST, Sehore, INDIA 

Durgesh Kumar Mishra, Acropolis Institute of Technology and Research Indore, INDIA 

39. Paper 23120926: A Probabilistic Model For Sequence Analysis (pp. 244-247) 

Amrita Priyam, Dept. of Computer Science and Engineering, Birla Institute of Technology, Ranchi, India. 
B. M. Karan, Dept. of Electronics and Electrical Engineering, Birla Institute of Technology, Ranchi, India 
G. Sahoo, Dept. of Information Technology, Birla Institute of Technology, Ranchi, India 

40. Paper 24120927: Dual Watermarking Scheme with Encryption (pp. 248-253) 

R. Dhanalakshmi, PG Scholar, Dept of CSE, Sri Venkateswara college, Of Engineering, India 

K. Thaiyalnayaki, Assistant Professor, Dept of IT, Sri Venkateswara college, of Engineering, India 

41. Paper 28011029: Effort minimization in UI development by reusing existing DGML based UI 
design for qualitative software development (pp. 254-261) 

P. K. Suri, Professor, Department of Computer Sc. & Application, Kurukshetra University, Kurukshetra, 

Haryana, India 

Gurdev Singh, Department of Computer Sc. & Application, Kurukshetra University, Kurukshetra, Haryana, 

India 

42. Paper 22120924: Medical Image Compression using Wavelet Decomposition for Prediction 
Method (pp. 262-265) 

S. M. Ramesh, Senior Lecturer, Dept. ofECE, Bannari Amman Institute of Technology, Erode, India 
Dr. A. Shanmugam, Professor, Dept. ofECE, Bannari Amman Institute of Technology, Erode, India 

43. Paper 31120934: High Performance Hybrid Two Layer Router Architecture for FPGAs Using 
Network-On- Chip (pp. 266-272) 

P. Ezhumalai, Department of Computer Science and Engineering, Sri Venkateswara College of 

Engineering, Pennalur, Sriperumbudure-602105, Chennai, Tamilnadu, India 

S. Manojkumar, Department of Computer Science and Engineering, Sri Venkateswara College of 

Engineering, Pennalur, Sriperumbudure-602105, Chennai, Tamilnadu, India 

Dr. C Arun, Department of Electronics & Communication Engineering, Rajalakshmi Engineering College 

Thandalam, Chennai - 602 105, Tamilnadu, India 

Dr. P. Sakthivel, Department of Electronics & Communication Engineering, College of Engineering, 

Guindy Anna University, Chennai, Tamilnadu 

Dr. D. Sridharan, Department of Electronics & Communication Engineering, College of Engineering, 

Guindy Anna University, Chennai, Tamilnadu 

44. Paper 13050924: New System for Secure Cover File of Hidden Data in the Image Page within 
Executable File Using Statistical Steganography Techniques (pp. 273-279) 



Md. Rafiqul Islam, A. W. Naji, A. A. Zaidan and B. B. Zaidan 

Department of Electrical and Computer Engineering, Faculty of Engineering, International Islamic 

University Malaysia (HUM), P.O. Box 10, 50728 Kuala Lumpur, Malaysia 

45. Paper 07011064: An innovative platform to improve the performance of exact-string-matching 
algorithms (pp. 280-283) 

Mosleh M. Abu-Alhaji, M. Halaiyqahi, Muhannad A. Abu-Hashemi, Adnan A. Hnaif, O. Abouabdallai , and 

Ahmed M. Manasrah. 

i : National Advanced IPv6 Center of Excellence, i : Computer Science 

University Sains Malaysia, Penang Malaysia 



46. Paper 18011012: A MAC Layer Based Defense Architecture for Reduction-of-Quality (RoQ) 
Attacks in Wireless LAN (pp. 284-291) 

Jatinder Singh, Director, Universal Institute of Engg. & Tech. Lalru-CHD (India) 

Dr. Savita Gupta, Prof. Deptt. Of Computer Engg., UIET, Punjab University, CHD, India 

Dr. Lakhwinder Kaur, Reader, UCOE, Punjabi University, Patiala, India 

47. Paper 06011052: Application of k-Means Clustering algorithm for prediction of Students' 
Academic Performance (pp. 292-295) 

Oyelade, O. J, Department of Computer and Information Sciences, College of Science and Technology, 

Covenant University, Ota, Nigeria. 

Oladipupo, O. O, Department of Computer and Information Sciences, College of Science and Technology, 

Covenant University, Ota, Nigeria. 

Obagbuwa, I. C, Department of Computer Science, Lagos State University, Lagos, Nigeria. 

48: Paper : Secured Cryptographic Key Generation From Multimodal Biometrics: Feature Level 
Fusion of Fingerprint and Iris (pp. 296-305) 

A.Jagadeesan, Research scholar/ Senior Lecturer /EIE, Bannari Amman Institute of Technology, 
Sathyamangalam-638 401, Tamil Nadu, India 

Dr. K.Duraiswamy, Dean/ Academic, K.S.Rangasamy College of Technology, Tiruchengode - 637 209, 
Tamil Nadu, India 

49. Paper : Evolutionary Computation Algorithms for Cryptanalysis: A Study (pp. 306-310) 

Poonam Garg, Information Technology and Management Dept., Institute of Management Technology, 
Ghaziabad - India 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 

SCENARIO BASED WORM TRACE PATTERN 
IDENTIFICATION TECHNIQUE 

Siti Rahayu S., Robiah Y., Shahrin S., Mohd Zaki M., Irda R., Faizal M. A. 

Faculty of Information and Communication Technology 

Univeristi Teknikal Malaysia Melaka, 

Durian Tunggal, Melaka, 

Malaysia 



Abstract — The number of malware variants is growing 
tremendously and the study of malware attacks on the Internet 
is still a demanding research domain. In this research, various 
logs from different OSI layer are explore to identify the traces 
leave on the attacker and victim logs, and the attack worm trace 
pattern are establish in order to reveal true attacker or victim. 
For the purpose of this paper, it will only concentrate on 
cybercrime that caused by malware network intrusion and used 
the traditional worm namely blaster worm variants. This 
research creates the concept of trace pattern by fusing the 
attacker's and victim's perspective. Therefore, the objective of 
this paper is to propose on attacker's, victim's and multi-step 
(attacker/victim) 's trace patterns by combining both 
perspectives. These three proposed worm trace patterns can be 
extended into research areas in alert correlation and computer 
forensic investigation. 

Keywords — trace pattern, attack pattern, log 

I. Introduction 

Nowadays the numbers of cases of internet threat causes 
by malware have been tremendously increased. Malware that 
consist of Trojan, virus and worm had threatened the internet 
user and causes billion of losses to the internet users around 
the world. The exponential growth of malware variants as 
reported by AV-test.org [1] is quite alarming. This can be 
seen from the higher growth rate of the malware collection 
shown in the graph of the new malware samples over the last 
several years as depicted in Fig.l. 



Total Number of Unique Samples in AV-Test.org's Malware Collection 




■Actual balance 

: crec5i: 



f ///f /ff/f ////// f/f/////f//// 



Fig. 1 The continuous growth of the number of malware [1] 

In the report, the numbers of malware is not cumulative 
where they represent only the new variants in the time frame 
specified without including the previously identified ones. 



This does not mean that there are only unique pieces of 
malware as there are also many variants of the same pieces 
of malware. Variants are often created to defeat the security 
tools, for instance a worm can mutate to a different variants, 
sometimes in only one hour [2]. Thus make it difficult for 
security tool to detect the threat. 

Due to this reason the study of attacks on the internet has 
becoming an important research domain. The study on 
internet attack is very crucial, especially in developing an 
effective security tool to defend the internet user from the 
attack threat. Collecting data related to Internet threats has 
become a relatively common task for security researchers 
and from the data the researcher can investigate the attack 
pattern in order to find the root cause and effect of an attack. 
Attack pattern can also be used as a guide to the investigator 
for collecting and tracing the evidence in forensic field [3]. 
According to [4], the anatomy of attack consists of attacker 
and victim perspective. 

To address this challenge, this paper propose on 
attacker's, victim's and multi-step (attacker/victim) 's trace 
patterns by collaborate both attacker and victim perspectives. 
This research explore the various logs from different OSI 
layer to identify the traces leave on the attacker and victim 
logs, and establish the attack trace pattern in order to reveal 
true attacker or victim. The research only focuses on 
cybercrime that caused by malware network intrusion 
specifically traditional worm namely blaster worm variants. 

II. RELATED WORK 

A. Blaster Worm 

Generally malware consists of virus, worm and trojan 
horse [5]. For the purpose of this research, the researchers 
focuses on one types of worm that described by [6], which is 
traditional worm specifically blaster worm. 

Blaster uses a public exploit for the RPC DCOM 
vulnerability in order to obtain a system-level command shell 
on its victims [7]. The worm start searching for IP addresses 
at a time for hosts with listening TCP port 135 open by 
sending a connection attempt to each one simultaneously. 
Once found, it tries to figure out the windows version and 
then sends the RPC DCOM exploit that binds a system level 
shell to TCP port 4444. Once the target is successfully 
compromised, the worm transmits the msblast.exe executable 
via TFTP on UDP port 69 to infect the host where the 
payloads used in the public DCOM exploit, as well as the 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) 

TFTP functionality, are both encapsulated within msblast.exe. 
Once the executable has been transferred, or after 20 seconds 
have elapsed, the TFTP server is shut down and the worm 
then issues further commands to the victim to execute 
msblast.exe [8]. Assuming the executable was downloaded 
successfully, the propagation cycle then begins again from 
the newly infected host, while the infecting instance of the 
worm continues iterating through IP addresses. 

B. Trace Pattern 

Trace is described as a process of finding or discovering 
the origin or cause of certain scenario and pattern is defined 
as a regular way in which certain scenario happened [9]. 
Trace pattern is essential in assisting the investigators tracing 
the evidence found at crime scenes. In the computer crime 
perspective, it can be found in any digital devices. These 
traces consist in a variety of data records their activities such 
as login and logout to the system, visit of pages, documents 
accesses, items created and affiliation to groups. Traces data 
are typically stored in log files and normally takes on several 
selected attributes such as port, action, protocol, source IP 
address and destination IP address. 

The trace data can be used to identify a victim or attacker 
by analyzing the attack pattern which is represented in the 
form of trace pattern can helps determine how a crime is 
being committed. Attack pattern is type of pattern that is 
specified from attacker perspective. The pattern describes 
how an attack is performed, enumerates the security patterns 
that can be applied to defeat the attack, and describes how to 
trace the attack once it has occurred [10]. 

An attack pattern provides a systematic description of the 
attack goals and attack strategies for defending against and 
tracing the attack. Hence, attack patterns can guide forensic 
investigators in searching the evidence and the patterns can 
serve as a structured method for obtaining and representing 
relevant network forensic information. This also helps the 
forensic investigator at the data collection phase that requires 
the investigator to determine and identifying all the 
components to be collected, deciding the priority of the data, 
finding the location of the components and collecting data 
from each of the component during the investigation process 
[11]. 



International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
software developers use attack pattern to learn about how 
their software may be attacked. Armed with knowledge 
about potential attacks, developers can take steps to mitigate 
the impact of these attacks. Similarly, network administrator 
or network engineers use attack pattern to study on how the 
potential attacker attack their network in order to detect and 
block any vulnerabilities in their network. 

The study from [1 2] [1 3] [14] discussed the concept of 
attack patterns as a mechanism to capture and communicate 
at the attacker's perspective that shows the common methods 
for exploiting software, system or network while [10] and 
[11] discussed on the attack pattern on how the attack is 
performed, the attack goals, how to defences against the 
attack and how to trace once it has occurred. 

All of the researchers are only focusing on the attacker's 
perspective while victim's perspective is omitted. Therefore, 
this research proposed the trace patterns by focusing on the 
attacker's, victim's and attacker/victim's (multi-step) 
perspectives to provide clear view on how the attacker 
performed the attack and what is the impact caused by the 
attack. 

Multi-step perspective proposed in this research is 
motivated based on the study by [15] in order to reveal the 
true attacker or victim. A multi-step attack is a sequence of 
attack steps that an attacker has performed, where each step 
of the attack is dependent on the successful completion of the 
previous step. These attack steps are the scan followed by the 
break-in to the host and the tool-installation, and finally an 
internal scan originating from the compromised host [16]. 

In the next section, researchers present the experimental 
approach used in this research to gather and analyse logs for 
designing the proposed worm trace pattern. 

III. EXPERIMENTAL APPROACH 

This experimental approach used four phases: Network 
Environment Setup, Attack Activation, Trace Pattern Log 
Collection and Trace Pattern Log Analysis. Further details on 
these phases are explained as follows and depicted in Fig. 2. 



Various descriptions provided by several researchers to 
describe the term attack pattern. According to [12], they use 
the term attack pattern to describe the steps in a generic 
attack. Meanwhile, [13] describe the term attack pattern as 
the attack steps, attack goal, pre-conditions and post- 
conditions of an attack. [14] describe an attack pattern as the 
approach used by attackers to generate an exploit against 
software in which it is a general framework for carrying out a 
particular type of attack, such as a method for exploiting a 
buffer overflow or an interposition attack that leverages 
certain kinds of architectural weaknesses. 

Although there are different descriptions provided by 
several researchers, they have the same idea and concept that 
the attack pattern is very important to provide a way to 
protect them from any potential attack. For example, 




Attack A ft rvat ion 




Trace Pattern Lag 
Atiar/sEi 



Fig. 2 Trace Pattern Experimental Approach Framework 

A. Network Environment Setup 

The network environment setup use in this experiment is 
following the network simulation setup done by the MIT 
Lincoln Lab [17] and it only consists of Centos and 
Windows XP operating systems to suit our experiment's 
environment. The network design is shown in Fig. 3. 

This network design consists of two switches, one router, 
three servers for Intrusion Detection System (IDS Arowana 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



and IDS Sepat) and Network Time Protocol (NTP) run on 
Centos 4.0, seven victims and one attacker run on Windows 
XP. 




(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 

A. Attack Scenario 

The attack scenario as depicted in Fig. 4 is derived based 
on the log analysis in the experimental approach framework 
in section III. 



192.168.11.20 



192,158.10,4 
192.169.12.3 192.169.13.16 Victim 

Victim Victim Rpglan 



Fig. 3 Network Setup Environment for Blaster Trace Pattern 

In this experiment, host 192.168.2.10 (Tarmizi) is 
Attacker, host 192.168.4.20 (Sahib) and 192.168.11.20 
(Yusof) are the selected Victims for this research. The log 
files that are expected to be analyzed are personal firewall 
log, security log, system log and application log. Each log 
are generated by host level device and one log files by 
network level device (alert log by IDS). Wireshark were 
installed in each host and tcpdump script is activated in IDS 
to capture the whole network traffic. The Wireshark and 
tcpdump script were used to verify the traffic between 
particular host and other device. 

B. Attack Activation 

The attacker machine, Tarmizi is selected to launch the 
Blaster variant attack. This experiment runs for one hour 
without any human interruption in order to obtain the 
attacker and victim logs. 

C. Trace Pattern Log Collection 

Each machine generated personal firewall log, security 
log, application log, system log and wireshark log. The IDS 
machine generates alert log and tcpdump log. The trace 
pattern logs are collected at each victim and attacker machine. 
The wireshark and tcpdump files are used to verify the traffic 
between particular host and other device. 

D. Trace Pattern Log Analysis 

In this trace pattern log analysis process the researchers 
analyze the logs generated by the attacker and victim 
machine. The objective of the trace pattern log analysis is to 
identify the victim and attacker trace pattern by observing the 
traces leave on the selected logs. The output from the 
analysis is used to construct worm trace pattern as proposed 
in Section V. 

IV. Analysis and findings 

The researchers have collected all logs generated during 
the experiment and the attack scenario is identified. Based 
on the scenario, various logs from attacker's host, victim's 
host, and network are analyzed. 




Fig. 4 Blaster Attack Scenario 

Based on the analysis, the worm attack is activated in 
Tarmizi and successfully exploited all hosts that mark with 
1 35, 4444 and 69 except for host Yusof and Selamat. These 
hosts was mark with 135 and 4444 and indicated that the 
attacker is already open the backdoor but unable to transfer 
the malicious codes through port 69. Then, one of the 
infected hosts that are Sahib has organized attack on host 
Yusof. 

B. Trace Pattern Analysis 

The purpose of trace pattern analysis is to construct the 
victim's, attacker's and multi-step (victim/attacker) 's trace 
pattern by observing the traces leave on the selected logs. 
The various logs involve in this analysis are host logs: 
personal firewall log, security log, system log, application 
log and network logs: alert log by IDS. Based on the attack 
scenario described previously, the logs are further analyzed 
to extract the trace pattern generated by the attack from the 
attacker's and victim's machine by tracing the fingerprints 
emerged in the logs. 

In this analysis, the researchers have selected Sahib and 
Yusof as victims; and Tarmizi as attacker as shown in Fig. 3. 
The tracing tasks involve the log from devices at host and 
network level and initially started at victim's logs followed 
by attacker's logs: personal firewall log, security log, system 
log and application log. The network logs are used to 
complement the finding from the host level tracing tasks. 
The details of the analysis for victim, attacker and multi-step 
attack (attacker/victim) trace pattern are further elaborated in 
the next sub-section. 

1) Victim's Traces 

The victim's data traces are extracted from the logs at the 
victim's host and network log. The summary of the data 
traces are depicted in Fig. 5 and the evidences are found in 
personal firewall log, security log, system log and 
application log and the details of the traces are discussed. 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 



Level 



Kvidiwe'lxig Category Truce Pattern 



Extracted Data 



Conceptual Diagram 



Personal 

Firewall 
Log 



Scan and 

Exploit 



Action, Protocol and Dcstiiiu 
tion Port: 

OPEN-INBOUND 135 TCP 
QFHHNBOUND 4444 TCP 
OPEN69UDP 

Source IP Address and Desti 
nation IP Address: 
Refer to Tablet 



20O9-O9-O714 41 09 OPEH-DfflOCM) ICP 192.:6S2 1C 192.1684 2C 3993 13?-- 
im-09-01 1441 120PEPHMBOrjHDTCP 1K.I68.2.1Q 192.163.4,2040024444-. 
2005-05-07 14.41:12 OPEN UDP 152,168.4.20 152.168.2.103027 69 



PERSONAL 

FIREWALL 

LOG 



1:l<OPFN--r\TSOr\T>TCT 
*M4 OPEN-INBOUND TCP 
69 OPEN UDP 
Source IP Address 
Destination IP Address 



Security 



Impact / 
Effect 



Most 



Event ID: 
592 

User ID: 

System 

Image File Name (IFN): 

• %WINDIR%\System32 
\tftp.exe 

• %WM)IR%\System32 
\msblast.exe 



Mfflt2m 14:41:12 Security Success Audit Detailed Tracking 592 NTALTfflORlTVSYSTEM 

SAHlb "A ntw process has been created; 
New Process ID: 1016 

Image File Name: C: \\VLNDOWS^itm32Vtfitp.exe 
Creator Process ID: 1226 User Name: SAHBt Domain: WORKGROUP 
Logon ID: (0xC,053E7) 

09A)7/2009 14:41 25 Security Success Audit Detailed Tricking 592 NT AUTHOPJTtfSYSTEM SA- 
HIB "A ntw proce a has totm creattd: 
New Process ID: 1752 

Image File Name: C:\VVINDOWStystem32^sMast.e:a 
Creator Process ID: 122S User Name: SAHIBS Domain: WORKGROUP 
Logon ID: (Ox0>0x3E7) 



SECURITY LOG 



t . User System 

/ W ; %WINDER%\ SystOlflZ 

f \trtp.exe 

A \ J 

\ ID: 552 

\ User. System 

1 nrtJ:%wncr,-'w ?-x\c.m" 



M:%™©IR%\ System^ 
\msblastexe 



System Log 



Event ID: 
7031 

1074 

Image File Name (IFN): 
RPC sendee terminated unex- 
pectedly 
Windows Restart 



09/07/2009 14:41:29Sen.ice Control Manager Error None7031N/A SAHIB 
The Remote Procedure Call (RPC) service terminated unexpectedly. 

It has done this 1 tinned). The following corrective action will be taken in 60000 millisec- 
onds: Reboot the machine. 

09/07/2009 14:41 :29USER32 Information None 1074 
NT AUTHORITYASYSTEM SAHIB The process winlogoaexe has initiated the restart 
of SAHIB for the following reason: No title for this reason could he found Minor Reason: 
Oxff 

Shutdown Type: reboot Comment: Windows must now restart because the Remote Pro- 
cedure Call (RPC) service terminated unexpectedly 



SYSTEM 
LOG 



TO: 7031 
RPC serriee terminated unexpectedly 



A s 

* \ ID: 1074 

\ Windows Restart 



Activity 



TFTP Get 



l**| [1:1444:3] TFTP Get [**] 

[CJassificatJon: Potentially Bad Traffic] [Priority: 2] 



DS ALERT 
LOG 



Network 



Alert IDS 

Log 



Alarm 



Source IP; 
SRC IP: Victim 



Destination D?: 
Dest. IP: Attacker 



Destination Port: 
69 



09^7-14:4 1:03.846332 192.1*8.420:3027 -> 192.168.2.10:69 
UDP TTL1 28 TOSfixO ID337 IpLen:2C DgmL«i:48 Len: 20 



IDS ALERT 
LOG 



SRCIP: Victim 

DesrIP: Attacker 

Dest Pent 65 



Fig. 5 Summary of Blaster Traces on Victim's Log 



In Personal Firewall log, ports 135 TCP, 4444 TCP and 
69 UDP are considered as part of this victim's trace pattern 
due to the information gain from [7], [18] and [19]. These 
ports are known as vulnerable ports that can be used by 
malicious codes to exploit the system-level command shell 
on its victims. These ports also provide an inter-process 
communication mechanism that allows programs running on 
one host to execute code on remote hosts [20]. 

The source IP address of OPEN-INBOUND TCP 
indicates the remote host and the source IP address of OPEN 
UDP indicates the local host. While the destination IP 
address of OPEN-INBOUND TCP indicates the local host 
and the destination IP address of OPEN UDP indicates the 
remote host. The summarized of the IP address dependency 
is depicted in TABLE 1. 

Table 1: IP Address Dependency 



Action 


Source IP 
Address 


Destination IP 
Address 


OPEN-INBOUND 


Remote Host 


Local Host 


OPEN 


Local Host 


Remote Host 



The patterns of the communication between Source IP 
address and Destination IP address indicate that the local 
host has permitted the TFTP service and the incoming traffic 
from the remote host. All communication activity is done by 
vulnerable ports open exploitation. 

In Security log, the traces data from the security log 
shows that there is a new process created by system proves 
by the existence of event id 592. It has initiated the TFTP 
service used to download and upload the blaster worm code 
and execute the remote blaster worm code (msblast.exe). 
This activity is logged in Image File Name as %WINDIR%\ 
System32\tftp.exe and %WINDIR%\Sy stem32\msblast.exe. 

System log shows the traces data of the Blaster-infected 
machine stops its TFTP daemon after a transmission or after 
20 seconds of TFTP inactivity by showing the new process 
created on event id 7031 and 1074 that indicates the RPC 
service terminated unexpectedly and windows restart 
respectively. 

The alert IDS log shows that there is an activity called 
TFTP Get is occurred on port 69 UDP where the source IP 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



address from alert IDS log is similar as the destination IP 
address in the victim's firewall log and vice versa. These 
traces identify that there is a pattern exists on how the blaster 
worm initiates the client to download the worm code. The 
Source IP address is the victim and the Destination IP 
address is the attacker. 

Action OPEN-INBOUND, OPEN-INBOUND and OPEN 
are known as a complete sequence of communication of 
blaster worm to gain access and upload the malicious codes 
to be exploited [21] where the OPEN action shows that an 
outbound session was opened to a remote host and the 
OPEN-INBOUND action shows that an inbound session was 
opened to the local host 

2) Attacker's Traces 

The attacker's data traces are extracted from the logs at 
the attacker's host and network log. The summary of the 
data traces on the attacker's and network logs are illustrated 
in Fig. 6 and the evidences are found in personal firewall 
log, security log, system log and application log and the 
details of the traces of the attacker's logs are discussed as 
following. 

The traces data leaved in attacker's Personal Firewall 
Log shown the vulnerable ports that are used by the attacker 
to exploit the system-level command shell on its victims. The 
pattern of the traces data are OPEN TCP 135, OPEN TCP 
4444 and OPEN-INBOUND UDP 69 as referred to [7], [18] 
and [19]. 

The source IP address of OPEN TCP indicates the local 
host and the destination IP address action indicates the 
remote host. While, the source IP address of OPEN- 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 
INBOUND UDP indicates the remote host and the 
destination IP address indicates the local host. 



The patterns of source IP address and destination IP 
address indicate that the local host is opened an outbound 
session to the remote host which allow the local host transmit 
the payload (worm codes) to the remote host by exploiting 
the vulnerable ports open. The traces data from the security 
log shows that there is a new process created (Event ID: 592) 
that shows the blaster worm is activated based on the trace 
shows on the image file name. 

In the alert IDS log, (Portscan) TCP Portsweep presents 
the pattern of scanning activity which shows the behavior of 
traditional worm attack in general and blaster worm attack in 
specific [22]. Therefore, this trace discovers that the owner 
of the source IP address is a potential attacker who launched 
the worm. 

3) Multi-step (Attacker/Victim) Trace Pattern 

Based on the extracted data from the logs at the victim's 
host and network log, the multi-step (Attacker/Victim) 's 
traces data is identified. The summary of the data traces on 
the multi-step and network logs are represented in Fig. 7 and 
the evidences are found in personal firewall log, security log, 
system log and application log and the details of the traces of 
the multi-step's logs are discussed. 

There are two different patterns existing in personal 
firewall log that discover attacker and victim traces as shown 
in TABLE 2. 



Level 


Evidence/ 
Log 


Category 


Trace Pattern 


Extracted Data 


Conceptual Diagram 


Host 


Personal 
Firewall 
Log 


Scan and 
Exploit 


Action, Protocol and 
Destination Fort: 

OPEX Hi TCP 
OPEN +444 TCP 
OPEN-INBOUND 6? 
UDP 

Source IP Address md 
Destination IP Ad- 
dress: 

Refer to Table 1 


2009-09-07 14:41:09 OPEN TCP 192,168.2,10 192,168.4.203993 135 

2009-09-07 14:41:11 OPEN TCP 192.168.2.10 192.168.4.204002 4444 

2009-09-07 14:41:11 OPEN-EVBOOT) UDP 192.168.4.20 192.168.2.10 3027 69---- 














r 133 OPEN TCP 
4444 OPEN TCP 
©OPEN-INBOUND UDP 

Source IP Address 
Destination IP Address 




PERSONAL 

FIREWALL 

LOG 
















Security 
Log 


Symptom 


Event ID: 

592 

Image Filename 
(IEK): 

~\blasterA.exe 


09/07/2009 14:40:01 Security Success Audit Detailed Tracking 592 Kamal TARMIZI 

"A new process has been created: 
New Process ID: 1408 

Image File Name: C -'.Documents and Settings'KamaT.Desktop\blaster A.ese 
Creator Process ID: 1492 
User Name: Kamal 
Domain: TARMIZI 
Logon ID: (0x0 : 0x2273F) 














r 






SECURITY 
LOG 




ID: 592 

Uisr: Kamal 

IFN: ~Wa*rAjSB 












Network 


Alert IDS 
Log 


Activity 


Message: 

(Portscan) TCP 

Portsweep 


["] [122:3:0] (portscan) TCP Portsweep ["] 

[Priority: 3] 














IDS ALERT 
LOG 




f * 

(Portscan) TCP 

Portsweep 
















Alarm 


Source IP: 

SRCJP: Attacker 


09/07-14:44:18.729996 192.168.2.10 -> 192.168.11.1 

PROT0255 TTL:0 TOS:0x0 TD:3307 IpLen20 DgmLen:166 














IDS ALERT 
LOG 




5RCJP: Attacker 

















Fig. 6 Summary of Blaster Traces on Attacker's Log 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 






I TJtee J'jllvr u 



L \W-n-, lul l>.i^ 



t'outeplual Diagram 



ELsptail 



OPEK-PBO'US'D. OfEN- 



OFfcS. UJt\ OJ'iN L^aOl'MJ 



359**P-0^ UiJSHJOPE.n.IXBOL'MJ TCP tM.lil 3.t(J JP3.I41.4 3D J*?J 11? - -■ 
?C4SM*=fr" MM 142 OFOMNDOL'ND TCP t« tliS lift I9H«S.4 2&4flo: J4+t" 
3fl». H-0- lit 1 1 3 OPF \ 1 IIP S K 1« i 3C- 1« L 41 : 1 3 H ?'**----.-- - 

jphw o- [4«.;iopi> Kttfl ld! J.JOiilitt.i 1.20 4? jiuf ■ 

305?.»-g: U-iJ^OFIMCPjaiHi*.^ L93.16S.L] 204Tl"4*«-. .^^ . 
200? 39 Q ■' L Ml .1 J OFi > IMHJ LSD I UV If J. itlM :Q i?Z 1M.4 JO >P 1 ! frM 



PERSON.*! 

rlKiWALL 

LOO 



1 5i UPtK LNHULMJ rtV 
WJ OPES ISBOUND 7C? 
»0P£NO3F 
fcura E? Addmi 
Dm limine j 1 Aii/r si 



TCT.TTP. LTJP 



PjcfrMoTiUi I 



l^nfTN'TO* 

iUJOPF%"TfT> 

M UffFNlMlUl^n LDP 

burr IK Wm 

•ViItjIi.t. IP i di-a 



Iinpim 
Effet 



YirLiin 
lwrlD: 

Ahjrkrr 

Si95::jif-i»j 
image rue Nam* ur»: 

* i„WCST)m!. SynemJ] 
[ftp. fat 

-i swpcdib.*., statmftz 



» 01 3JD? 144143 StodLv SoKLrn Aril CflBlxdTiKrkng 5?; XT AUTHOR 

[TYSYSriM SAirJTJ, 'AmprHnihihticreinJ 
Nr* Promts* □> J0L4 

[nuui kib S'ih C:U IMMH1S iiTlrm.1: th, ftp 

Cnitef ?[«(« D i:JS UietKidk : SAKED* Donuai WORKGROUP 
L*BsnM>, (OriLlhJi?] 

W a ' DM J H .« .21 SkwtK Slkmu, Audi Bribed Tiiclng JK M AVI HUE 

[TYSVSTFU SAHIB -AuipwiiliHkrrirmnJ 

Nr»Pt[K«iID IVJ3 

[m»(f Kir- r'.imr tl U IMH)USinrHLlZBitliii.r» 

C:ms« Ft«*h ff> t33 1 Vj« nW SAJlJflJ Do-num WORKGROUP 

f»DtJ»flJ Ji:l2:445Kunt}.-Succ«iiAii4l Dllhhil Till UltMII 11111 1HI 1*1 

L?l~Ldt(ii: K»!>K:il5 l<IK< SAH;il "Am- prniiibihfirmlii 

Stu PtKtHlD 213 

[QU E t UrNimr fi \\ [MJOtt S ii*HmK aitluLtir 

C rriiflf PrarH-i HI t"4E Uirf N«nr- Shihfm nnman 5AHSH 

UpmiD.CIUQ.Oi&itt) 




-on 
IBM 






<i Window. RrMjM 



09 7 2Q09 J 4.4 1 .2-9 Service C?i«vJ Ntnjifl 5C T Eii^i Nvnc 7(01 N A 

SAHIB 

Tfco RtnaTt PntftdureC alt LRPCMerviee terminated DiwjpemdK. It biu 

cLitir rhr. I lunHs) Thr fnillou. m it iiirrrvlivp ju-licvn will he [nkrn in fiOOOO mill: 

stcoods Eteboot the nmckiiit 

1SQT2W9 14:41:2? USE R32 laforastiaA None H» T * 

NT ALTIIORITV SYSTEM SAIDB Tucixkch wuIordd .cm Jus JnitwHtd [be 
tf sun of SAHIB for die foHowing rtasco: No title tor this reason could be found 
Mioftr EtfA^n QxPf 

Shutdo'-i a Typ« icboot Cu-ramirnl ^ l n f|(i " i ^lll^l dii« rrslarl bvcamv tbv Rv 
multPrwrdart? CiJ (RPC} »nkt tvrmiBa trd jnti^tirtf dh 



[ 4TSTOILOG T**^ 



Virhm 



.\ltiiag«: 

(TJrjltican r TCP PrjltiTAfrp 



["IP 1444 }]TttT9mr*\ 

[tlasiifi eaten PotcDlialLr Eiad Traffic J [Prior itj": 2 ] 



["] [122:3:0] (porf«an) TCPPorcm-ttpl"] 

Priority 11 



i \A 

1 V 



Atnl IDS 
Log 



EHC IF Uurlr, 



09 Of -14:4 1 03 5+6iS2 L 91. 168.4.10 :302 7 - • ]!12.I68.MU:6V 
LT3P TTL 1 25 TOS^sO ID.JJT TpLen :aDgmLea:-4S Leu 20 



0907,14.4? 38 3S4313 LK. 1614.20 -> I92.16E.11.1 

PROTO: 255 "ITL TOS: OkO ID: 1395 SpLcn 20 DgmLcn: 264 



TCMAURTLM 



< 



SP.C_T? Vtraa 
D«i rj.AaiiloM 
Di^l ; J .:- M 1 



S.E4 - JP AhkLk 



Fig. 7: Blaster Traces on Multi-step (Victim/Attacker) Log 



Table 2: Traces Data on Personal Firewall Log 



Victim 


Attacker 


Action, Protocol and 


Action, Protocol and 


Destination port: 


Destination port: 


OPEN-INBOUND TCP 135, 


OPEN TCP 135, OPEN TCP 


OPEN-INBOUND TCP 4444 and 


4444and OPEN -INBOUND 


OPEN UDP 69 - These ports are 


UDP 69 - These ports are 


known as vulnerable ports that 


known as vulnerable ports that 


provide hole and opportunities to 


can be used by the attacker to 


the attacker to exploit in order to 


exploit the system-level 


gain access to the system. 


command shell on its victims. 


Source IP Address and 


Source IP Address and 


Destination IP Address: 


Destination IP Address: 


OPEN-INBOUND TCP - the 


OPEN TCP - the source IP 


source IP address indicates the 


address indicates the local host 


remote host and the destination IP 


and the destination IP address 


address indicates the local host. 


indicates the remote host. 


OPEN UDP - the source IP address 


OPEN-INBOUND UDP - the 


indicates the local host and the 


source IP address indicates the 


destination IP address indicates the 


remote host and the destination 


remote host. 


IP address indicates the local 




host. 



From the victim perspective (OPEN-INBOUND TCP 135, 
OPEN-INBOUND TCP 4444 and OPEN UDP 69), the 
patterns shows that the source IP address and destination IP 
address indicate that the local host is permitted the TFTP 
service and the incoming traffic from the remote host. 

While, from the attacker perspective (OPEN TCP 135, 
OPEN TCP 4444 and OPEN -INBOUND UDP 69), the 
patterns of source IP address and destination IP address 
indicate that the local host is opened an outbound session to 
the remote host which allow the local host transmit the 
payload (worm codes) to the remote host. 

All communication activity is done by vulnerable ports 
open exploitation. Therefore, the traces data found are 
significant to the multi-step attack (victim/attacker) where 
this host was infected (act as victim) and as long as the 
computer was infected with the worm code (msblast), it (act 
as attacker) continued to generate traffic to attempt to infect 
other vulnerable computers [23]. 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



The traces data in TABLE 3 are taken from the security 
log in Fig. 9 and it shows that there is a new process created 
by system which initiates the TFTP service. This service is 
used to receive and sent the blaster worm code and execute 
the blaster worm code imsblast.exe) remotely. 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 
Based on the analysis, the researchers have identified the 
significant attributes from the victim, attacker and multi-step 
traces data. These findings are further use to construct the 
proposed worm trace pattern. 



V. PROPOSED WORM TRACE PATTERN 



Table 3: Traces Data on Security Log 



Victim 


Attacker 


Event ID: 592 


Event ID: 592 


User ID: System 


User ID: S-l-5-21-725345543- 
1547161642-839522115-1003 


Image File Name: 

%WINDIR%\System32\ 

tftp.exe, 

% WINDIR % \Sy stem32\msbla 

st.exe 


Image File Name 

%WINDIR%\System32\msblast.exe 



The pattern in TABLE 3 has indicates that this host is a 
victim of blaster worm attack. Another traces data found is a 
new process created (Event ID: 592) by unidentified user (S- 
1-5-21-725345543-1547161642-839522115-1003) which 
executed the msblast as shown on the image file name. It 
identify that this host was infected previously and 
automatically attempt to transfer the worm code by 
generating traffic to exploit other vulnerable computers. 

System log shows the traces data of the Blaster-infected 
machine stops by showing the new process created on event 
id 7031 and 1074 that indicates the RPC service terminated 
unexpectedly and windows restart respectively. This pattern 
is significant with the victim pattern in which if the host is 
infected by blaster worm, the RPC service is terminated 
unexpectedly by exploiting the RPC DCOM and force the 
windows restart. 

As depicted in TABLE 4, the alert IDS log shows that 
there are traces found on TFTP Get and (Portscan) TCP 
Ports weep activities for victim and attacker respectively. 

The TFTP Get activity on port 69 trace indicates that 
there is a pattern exists on how the blaster worm initiates the 
client to download the worm code. The source IP address is 
the victim and the destination IP address is the attacker. On 
the other hand, the (Portscan) TCP Portsweep trace proves 
that there is scanning activity on the vulnerable open port. 
This trace indicates that the source IP address is the attacker 
who activated the worm. Both traces found in TABLE 4 are 
significant to the pattern that found in victim and attacker as 
depicted in Fig. 8 and Fig. 9 respectively. 

Table 4: Traces Data on Alert IDS Log 



This research proposed the worm trace pattern based on 
victim, attacker and multi-step point of view. The following 
section describes the details. 

A. Victim 's Trace Pattern 

Victim's trace pattern is useful for forensic in order to 
provide clear view on how the victim attacked by the 
potential attacker. According to the analysis and findings 
from Fig. 5, the overall Blaster victim's trace pattern is 
summarized in Fig. 8. 



Victim 


Attacker 


Message: TFTP Get 


Message: (Portscan) TCP 
Portsweep 


Source IP Address and Destination IP 
Address: 

Source IP address indicates the victim and 
the Destination IP address indicates the 
attacker 


Source IP Address: 

Source IP address indicates 
the attacker 


Destination Port: 69 





135 OPEN-MOOD TCP 
#4QPEN-INB01NDICP 
i5S OPENOT 
Source IP Address 
Destination IP Address 



TEN °cWIM)ffi° 5 SysiemJlipeve 



M2 

UN: %WM% System32tatkst.ese 



ID: 7031 

RPC service laminae - .- -1 



ID: 1C"- 
.:: :"-::■ 




IDS ALERT 

LOG 





SRCJP: Victim 
Deal Attacker 
DdPttt® 






LT3S ALERT 
LOG 




ALARM 















Fig. 8 Proposed Blaster Victim's Trace Pattern 

In Fig. 8, the traces data indicated the blaster worm 
pattern at the victim's host used port 135 TCP to permit the 
scanning and transmitting RPC DCOM exploit codes from 
remote host which launch the windows shell to initiate worm 
code download used port 4444 TCP. 

Then it launched the TFTP client service using port 69 for 
downloading the worm code. The traces on TFTP Get on 
port 69 UDP also found in the network log that supports all 
the traces found on the host log. 

B. Attacker's Trace Pattern 

Attacker's trace pattern provides a systematic description 
of the attack goals and attack strategies for defending against 
and tracing the attack. This pattern is useful to guide forensic 
investigators in searching the evidence and provide a 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



structured method for obtaining and representing 
network forensic information. 




Fig. 9 Proposed Blaster Attacker's Trace Pattern 

The overall Blaster attacker's trace pattern depicted in Fig. 
9 indicate the blaster worm pattern at the attacker's host used 
port 1 35 TCP to allow the local host scan and transmit RPC 
DCOM exploit codes to the remote host which launch the 
windows shell to initiate worm code download used port 
4444 TCP. Then it launched the TFTP client service using 
port 69 to permit the client (remote host) download the worm 
code from the local host. The activity of TCP Portsweep 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 
relevant trace also existed in the network log that supports all the 
traces found on the host log. 

C. Multi-step (Attacker/Victim) Trace Pattern 

Multi-step' s trace pattern is used as a guide for forensic 
investigators to reveal and prove the true attacker or victim. 
This trace pattern is a combination of victim's and 
attacker' s trace pattern in which the traces data is extracted 
from a log for the same host. 



The traces data on multi-step at the host's logs from 
victim/attacker perspective illustrated in Fig. 10 indicate that 
the blaster worm used port 135 TCP to permit the scanning 
activity and it is supported by the traces found in network 
logs that show (Portscan) TCP Portsweep activities. 

The worm then transmit RPC DCOM exploit codes from 
remote host which launch the windows shell to initiate 
downloading the worm code using port 4444 TCP and it 
launched the TFTP client service on port 69. This worm 
activity is shown by the traces found in network logs that 
confirm the existence of TFTP Get on port 69 UDP activities. 
Once the host is infected (act as victim), it's (act as attacker) 
then generate traffic; attempt to infect other vulnerable hosts. 

The source IP address from host log indicates that the 
remote host is the victim and the destination IP address 
which is the local host is the attacker. Hence, multi-step 
(victim/attacker) trace pattern could identify the true victim 
or attacker. 



Victim 

13 5 OPEN-INBOUND TCP 

44++ OPEN-INBOUND TCP 

69 OPEN UDP 

Source IP Address 

Destination IP Address 



135 OPEN TCP 

4444 OPEN TCP 

69 OPEN-INBOUND UDP 

Source IP Address 




1— [ 



ID 592 

User: Svstem 
IFN: %WTNDIR%. System32 tftp.e: 



ID: 592 

Usee System 
IFN :%WINDIR%\ System3 2\msblast.e: 



Attacker 
ID: 592 

User: S-* 

3X: "g'wixdifj'g SvstemS: 1 




SECURITY LOG 






SYSTEM LOG 



[Portscan) TCP Portsweep 



D>c 



SCAN £: EXPLOITING 




IMPACT EFFECT 




BIS ALERT LOG 



D-C 



ACTIVITY 



Victim 

SRC_IP: Victim 
Dest_D?: Attacker 



SRC_IP: Attacker 



P* 



IDS ALERT LOG 



v-c 




NETWORK LOG 




http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 



Fig. 10 Proposed Multi-step 

VI. Conclusions and future works 

Trace pattern of an attack in an attack scenario is 
constructed by analyzing various logs from heterogeneous 
devices in victim, attacker and victim/attacker (multi-step) 
perspectives. These trace patterns offer a systematic 
description of the impact of the attack, the attack goals and 
attack steps along with strategies for defending against and 
tracing the attack. For example, personal firewall logs 
provide information on how the attacker entered the network 
and how the exploits were performed; meanwhile event 
logging such as security log, system log and application log 
enables network administrators to collect important 
information such as date, time and result of each action 
during the setup and execution of an attack. Therefore, the 
propose victim, attacker and multi-step (victim/attacker) 
trace patterns in this paper can be extended into research 
areas in alert correlation and computer forensic investigation. 



(Victim/Attacker) Trace Pattern 
[18] McAfee. (2003). Virus Profile: W32/Lovsan.worm.a [Electronic 

Version]. Retrieved 23 July 2009 from 

http://home.mcafee.com/Vi ruslnfo/VirusProfile.aspx?key=100547 . 
[19] Symantec. (2003). W32.Blaster.Worm [Electronic Version]. 

Retrieved 23 July 2009 from 

http://www.symantec.com/security response/writeup.isp?docid=20 

03-081113-0229-99 . 
[20] Sachs, M. H. (2003). SANS Top-20 Entry: W5 Windows Remote 

Access Services [Electronic Version]. Retrieved 10 December 2009. 
[21] Dubendorfer, T., Arno Wagner, Theus Hossmann, & Plattner, B. 

(2005, July 7-8). Flow-Level Traffic Analysis of the Blaster and 

SobigWorm Outbreaks in an Internet Backbone. Paper presented at 

the Detection of Intrusions and Malware & Vulnerability Assessment, 

IEEE, Vienna, Austria. 
[22] Cliff, C. Z., Don, T., & Weibo, G. (2006). On the Performance of 

Internet Worm Scanning Strategies. ACM Performance and 

Evaluation Journal, 63(7), 700-723. 
[23] Braverman, M. (2005). Win32/Blaster: A Case Study from 

Microsoft's Perspective. On VIRUS BULLETIN CONFERENCE 

OCTOBER 2005. 



References 

[I] Seltzer, L. (2009). The Growth of Malware: Up, Up, and Away 
[Electronic Version]. Retrieved 24 Nov 2009 from 
http://blogs.pcmag.com/securitywatch/2009/07/the_growth_of_malw 
ar e_up_up_an . php . 

[2] Gadelrab, M., Abou El Kalam, A., & Deswarte, Y. (2008). Execution 

Patterns in Automatic Malware and Human-Centric Attacks. Paper 

presented at the Proceeding of Seventh IEEE International 

Symposium on Network Computing and Applications. 
[3] Olivier, T., & Marc, D. (2008). A framework for attack patterns' 

discovery in honeynet data. Journal of Digital Investigation, 5, 128- 

139. 
[4] McHugh, J. (2001). Intrusion and intrusion detection. International 

Journal of Information Security, 1, 14-35. 
[5] Karresand, M. (2003). A proposed taxonomy of software weapons 

(No. FOI-R-0840-SE): FOI-Swedish Defence Research Agency. 
[6] Lazarevic, A., Kumar, V., & Srivastava, J. (2005). Managing Cyber 

Threats (Pg 19-78). On Massive Computing: Springer US. 
[7] Microsoft. (2003). Virus alert about the Blaster worm and its variants 

[Electronic Version]. Retrieved 23 July 2009 from 

http://support.microsoft.com/kb/826955 . 
[8] Bailey, M., Cooke, E., Jahanian, F., Watson, D., & Nazario, J. (2005). 

The Blaster Worm: Then and Now. IEEE Computer Society. 
[9] Hornby, A. S. (2005). Oxford Advanced Learner's Dictionary (7 ed.). 

New York: Oxford University Press 
[10] Fernandez, E., Pelaez, J., & Larrondo-Petrie, M. (2007). Attack 

Patterns: A New Forensic and Design Tool. IFIP International 

Federation for Information Processing, 242, 345-357. 

[II] Kent, K., Chevalier, S., Grance, T., & Dang., H. (2006). Guide to 
Integrating Forensic Techniques into Incident Response, NIST 
Special Publication 800-86. 

[12] Hoglund, G., & McGraw, G. (2004). Exploiting Software: How to 

Break Code. Boston, Massachusetts: Addison- Wesley/Pearson. 
[13] Moore, A., Ellison, R., & Linger, R. (2001). Attack modeling for 

information security and survability, Technical Note CMU/SEI-2001- 

TN-001. Pittsburgh, Pennsylvania: Software Engineering Institute, 

Carnegie Mellon University 
[14] Sean, B., & Amit, S. (2006). Introduction to Attack Patterns 

[Electronic Version]. Retrieved 19 Nov 2009. 
[15] Liu, Z., Wang, C, & Chen, S. (2008). Correlating Multi-Step Attack 

and Constructing Attack Scenarios Based on Attack Pattern Modeling. 

IEEE Computer Society, 214-219. 
[16] Valuer, F., Vigna, G., Kruegel, C, & A. Kemerrer, R. (2004). A 

Comprehensive Approach to Intrusion Detection Alert Correlation 

IEEE Transaction on dependable and Secure Computing, 7(3). 
[17] Lincoln Lab, M. (1999). 1999 DARPA Intrusion Detection 

Evaluation Plan [Electronic Version]. 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 



Avoiding Black hole and Cooperative Black hole 
Attacks in Wireless Ad hoc Networks 



Abderrahmane Baadache 

Laboratory of Industrial Technology and Information, 

University of A. Mira, Targua Ouzemour, 06000, 

Bejaia, Algeria. 



Ali Belmehdi 

Laboratory of Industrial Technology and Information, 

University of A. Mira, Targua Ouzemour, 06000, 

Bejaia, Algeria. 



Abstract — In wireless ad hoc networks, the absence of any 
control on packets forwarding, make these networks vulnerable 
by various deny of service attacks (DoS). A node, in wireless ad 
hoc network, counts always on intermediate nodes to send these 
packets to a given destination node. An intermediate node, which 
takes part in packets forwarding, may behave maliciously and 
drop packets which goes through it, instead of forwarding them 
to the following node. Such behavior is called black hole attack. 
In this paper, after having specified the black hole attack, a 
secure mechanism, which consists in checking the good 
forwarding of packets by an intermediate node, was proposed. 
The proposed solution avoids the black hole and the cooperative 
black hole attacks. Evaluation metrics were considered in 
simulation to show the effectiveness of the suggested solution. 

Keywords- wireless ad hoc network, routing protocol, security, 
black hole, cooperative black hole. 



I. Introduction 

A wireless ad hoc network is a collection of nodes 
connected by wireless links. Although it offers the advantage of 
being easy to deploy, the wireless ad hoc network paradigm, 
characterized by the absence of any control at the routing 
operation or data forwarding, introduced truths problems of 
security making thus less powerful the operation of such 
network. 

Early conceived routing protocols suppose saint the 
environment in which the network is deployed [1]. However, 
that it is the contrary in practice, and it proves to be difficult the 
guarantee of the security within a network, where the medium 
of communication is open, a central authority of certification 
misses, what facilitates the interception, the modification or 
even the manufacture of packets so that then to inject them into 
the network, in order to disturb the correct operation or to make 
entirely non operational the network. Guarantee the security is 
not limited to ensure, individually, the following services: 
Participants authentification, data integrity and confidentiality, 
non repudiation, access control to the communication medium 
and anonymity. But also, how to put individual solutions 
together to produce a solution whose range will hold in account 
the beforehand listed considerations and requirements. The 



literature contains security solutions which are protocols whose 
originating objective was security, and other protocols 
conceived at the beginning without security like primary 
objective, and they were secured in improved versions because 
attacks of which they were victims. Obviously, security 
remains always an arising problem and the remedy is far from 
being found. 

Various attacks against wireless ad hoc networks can be 
conducted [1]. They are qualified passive ones, if they are 
limited to the listening of the network traffic to take note, or 
active if the traffic is modified by the intruder. Security attacks 
can be internal when the malicious node belongs to the 
network, or external if not. Deny of service attacks are easy to 
carry out, and difficult to detect. Their principle is the violation 
and the non respect of the network protocol specification and 
their finality is the disturbance of the correct network 
operation. The no relaying of the traffic (of control or data) by 
an intermediate node constitutes a behavioral deviation, whose 
consequence is the violation of the objective for which the 
network is deployed. Such malicious behavior is called the 
black hole attack. In this paper and after having specified how a 
black hole attack is conducted, a solution consisting in 
checking the good forwarding of the traffic by an intermediate 
node, was proposed. The solution is based on the well-known 
principle which is the Merkle tree [2], [4]. 

The rest of the paper is organized as follows: Section 2 
summarizes the related work, follow-up by a necessary 
background in section 3, then the black hole attack 
specification is presented in section 4 and detailed description 
of the suggested solution is the subject of section 5. Simulation 
and results are analyzed and discussed in section 6. The paper 
is achieved by a conclusion and our future work. 

II. RELATED WORK 

They did not have satisfied solutions to solve the black hole 
attack problem, what led researchers to be addressed to this 
attack to find remedies for it. In the literature, the black hole 
attack solutions can be solutions which are interested in the 
black hole attack acting in an individual manner or those which 
are interested in the cooperative black hole attack or general 



10 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



security mechanisms being interested 
addition to the black hole attack. 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. I, 2010 
in others attacks in with the little strong assumption that a neighborhood of any 
node has more trusted than malicious nodes. 



In [4], Hongsong et al. proposes an intrusion detection 
model to combat the black hole attack in AODV [5] routing 
protocol. In this model, a security agent, established by a 
hardware thread in network processor uses parallel 
multithreading architecture, try to detect two cases of figure of 
attack. Those exploiting AODV control messages RREQ 
(Route REQuest) and RREP (Route REPly). The agent 
monitors the RREQ-RREP messages at real-time and if any 
detection rule is violated, the black hole attack is detected and 
the malicious node is isolated and recorded to a black list. This 
solution requires a special material for its implementation. It is 
dedicated to AODV protocol and it considers only control 
messages, however that black hole attack can target data 
messages. Considering always AODV, authors of [6] try to 
detecting abnormality occurs during the black hole attack by 
defining a normal state from dynamic training data that is 
updated at regular time intervals. To express the state of the 
network, the following features are used: Number of sent out 
RREQ messages, Number of received RREP messages and the 
change of the sequence number value used by AODV to 
determine the route freshness degree. Through the simulation, 
this method shows significant effectiveness however a more 
processing overhead is needed for its implementation and 
consequently it can suffer from scalability problems. In [7], the 
authors are always interested in AODV and propose a solution 
in which the receiving node of RREP message compares the 
sequence number value with a dynamic updated threshold. If 
the sequence number value is found to be higher than the 
threshold value, the node is suspected to be malicious and it 
adds the node to the black list. Here still, except message 
RREP is controlled. It is necessary to hold in account also data 
packets because a black hole node can behave normally in the 
route establishment phase and maliciously in the data 
transmission phase. In more, the threshold considered can miss 
exactitude what brings back to false alarms. 

A cooperative black hole attack is when several malicious 
nodes work together as a group. To identify multiple black hole 
nodes acting in cooperation, Fu's team, in [8], [9], proposes 
slightly modified AODV protocol by introducing Data Routing 
Information (DRI) table, that contains information on routing 
data packet from/through the node and cross checking process 
that determines the reliable nodes to discover secure paths from 
source to destination. In [10], authors propose an enhancement 
of the basic AODV routing protocol to combat the cooperative 
black hole attack. They use a structure, which they call fidelity 
table, wherein every participating node will be assigned a 
fidelity level that acts as a measure of reliability of that node. In 
case the level of any node drops to 0, it is considered to be a 
black hole node and is eliminated. In their approach, they 
assume that nodes are already authenticated which is a little 
strong assumption. Agrawal et al. [11] proposes a complete 
protocol to detect a chain of cooperating malicious nodes in an 
ad hoc network. The proposed protocol is based on sending 
equal and small sized blocks of data, and monitoring the traffic 
flow at the neighborhoods of both source and destination, then 
gathering results of monitoring by a trusted backbone network, 



Other black hole attack solutions were summarized in [1] 
and dedicated solutions for securing particular routing 
protocols against all possible attacks, not only against the black 
hole, were proposed. SAR [12] and SEAD [13] are examples of 
such secure routing protocols. The watchdog and pathrater [14] 
mechanism is also proposed for mitigating misbehavior. The 
watchdog for identifying misbehaving nodes and the pathrater 
helps routing protocols avoid these nodes. This mechanism is 
functional only with nodes equipped by interfaces that support 
promiscuous mode operation. 

III. Background 

Our solution is based on the principle of Merkle tree [2], [3] 
and uses the AODV [5] routing protocol like case of study, 
thus the knowledge of Merkle tree principle and AODV 
functioning is necessary. 

A. Merkle tree 

A Merkle tree is a binary tree in which, each leaf carries a 
given value and the value of an interior node (including the 
root) is a one-way hash function of the node's children values. 
Figure 1 illustrates an example of a Merkle tree in which: 

• h denotes a one-way hash function. For example, the 
function SHA-1 [15]. 

• 1 1 is the concatenation operator. 

• Values of leaves 1,2,4, respectively, are: h(a), h(b), 
h(c). 

• The value of the interior node 3 is: h(h(a)Hh(b)) 
which is the hashing result of the concatenation of 
values of children 1 and 2. Idem for the node 5 
whose value is h(h(h(a)Hh(b))Hh(c)) and children 
are: 3 and 4. 

h(h(a||b)||h(c)) 




h( a ||b) 



h(a) h(b) 

Figure 1. Example of a Merkle tree 



B. 



AODV routing protocol 

AODV (Ad hoc On-Demand Distance Vector) [5] is a 
reactive routing protocol composed of two modules: 

• Route discovery module: To send data to a given 
destination D, the source node S consults its routing 
table. If it finds a valid entry (a route) towards this 
destination D, it uses it immediately, else it launches a 
route discovery procedure (see Figure 2.), witch 
consists in broadcasting, by the source node S, a route 
request (RREQ) message (containing amongst other 
information: destination's address, destination's 
sequence number) towards neighbors. When RREQ is 



11 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



received by an intermediate node, this last consults its 
routing table to find a fresh route (the route is fresh if 
the sequence number of this route is larger than that of 
RREQ) towards the requested destination in RREQ. If 
such a route is found, a route reply (RREP) message is 
sent through the pre-established reverse route 
(established when RREQ pass through intermediate 
nodes) towards the source S. If the intermediate node 
does not find a fresh route, it updated its routing table 
and sends RREQ to these neighbors. This process is 
reiterated until RREQ reaches the destination node D. 
The destination node D sends RREP to S by using the 
pre-established reverse route. It should be noted that 
the source S can receive several RREP, it will choose 
that whose destination's sequence number is larger, if 
destination's sequence numbers of several RREP are 
equal, that of which the smallest hope counter will be 
selected. 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 
the manner with which the malicious node fits in the data route 
differs. Since our case of study is the AODV protocol, we will 
see how a malicious node can make a success of its attack in 
AODV. 




>@t---zd3>'' 



Figure 2. Route discovery process of AODV 

• Route maintenance module: AODV uses Hello 
messages to maintain the connectivity between nodes. 
Each node periodically sends a Hello message to these 
neighbors and awaits Hello messages on behalf of 
these neighbors. If Hello messages are exchanged in 
the two directions, a symmetrical link between nodes is 
always maintained if no link interrupt occurs. The 
broken link can be repaired locally by the node 
upstream, else a route error (RERR) message is sent to 
the source S (see Figure 3.). This last can launch again, 
if necessary, the route discovery procedure. It should 
be noted that the link interrupt is the consequence of 
the mobility or the breakdown of nodes. 




>®-- ~<g>' 

Figure 3 . Route maintenance process of AODV 

IV. Black hole Att a c k specification 

In a black hole attack, the malicious node refuses to 
forward data packets to the following node in the route 
connecting a given source and destination. To conduct its 
attack, the malicious node must initially belong to the data 
route, then it pass to the action which is the data dropping. 
According to the specification of the target routing protocol, 



Two kinds of black hole attack can be distinguished: 

• Internal black hole attack: The malicious node is an 
internal node which does not seek to fit in an active 
route between a given source and destination, and if the 
chance would have it, this malicious becomes element 
of an active data route, it will be able to conduct its 
attack as the transmission of the data starts. This attack 
is internal because the malicious node belongs already 
to the data route. Here, there is no violation of AODV 
specification and the malicious node has anything to 
make to carry out with success its attack. 

• External black hole attack: The malicious node is an 
external node which seeks to fit in an active route. For 
that, it violates the routing protocol specification and 
executes the process schematized in Figure 4. and 
summarized in the following points: 

- The malicious node detects the existence of an 
active route and takes note of the destination 
address. 

- The malicious node prepares a route replay packet 
(RREP) in which: the destination address field is set 
to the spoofed destination address, the sequence 
number is set to a greatest value and the counter 
hope is set to a smallest value. 

- The malicious node sends this route reply RREP to 
the nearest intermediate node belonging to the real 
active route (not necessarily to the data source node 
himself). 

- The route reply RREP received by the intermediate 
node will be relayed through the preestablished 
inverse route towards the data source node. 

- The source node updates its routing table by the 
new information received in the route reply. 

- The source uses the new route to sending data. 

- The malicious node starts to drop the data in the 
route to which it belongs. 




Xb>- --®' 



Figure 4 . Black hole attack specification 

The malicious node M detects the existence of active route 
(S,A,E,D) between the source S and destination D. The 
malicious node M sends to the node A a route reply message 
(RREP) containing the spoofed destination address, a relatively 
large sequence number and a small hope count. The node A 



12 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



forwards this route reply message to S. This last updates its 
routing table and considers the new route (S,A,M). The node S 
uses this route to send data and while arriving at M, these data 
will be quite simply dropped. Nodes source and destination 
will not be able any more to communicate in the presence of 
the black hole attack. 

V. Our solution 

The table I contains the notations used to describe our 
solution. 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 
cannot conduct a cooperative black hole attack because they do 
not know the secret S D held by D. 



TABLE I. 



Notations 



Notation 


Significance 


idi 


Identity of node i. 


Si 


Secret generated by node i. 


h 


One-way hash function. 


II 


Concatenation operator. 



In Figure 5, we consider a piece of network made up of 3 
nodes A, B and C. On this last, a Merkle tree is juxtaposed. We 
point out that our goal is to check that B conveys well, towards 
C, the traffic sent by A. 

5=h(n||k}© 

ji=h(a||R) o 



a=h(id,\||S, v ) 




P=h(idH||S,i) X=h(id c ||Sc) 

Figure 5 . Basic principle of the solution, single black hole case 

Each node / holds the value h(idiHSi), \e{A,B,C}. The 
transmitter node A has, in more, the value 5 (value of the root 
of the Merkle tree). So that A checks that B forwards well the 
traffic to C, the node C sent the value X (value held by C) to B 
and S, in turn, sent to A, (3 (value held by B) and ^.When both 
P and X are received by A, the node A recalculates 5 from a 
(value owned by A), (3 and X, then compares the result with the 
value of 5 already held, if equality, the node B routed out the 
traffic to C, otherwise, B is a black hole node. Obviously, to 
leading a black hole attack, the node B must generate the value 
X which is impossible, because it does not know the C's secret 
S, 

Nodes B and C can cooperate to conduct black hole attack, 
this is easy if C communicates to B its secret S c . To prevent 
such a scenario, the idea of the solution can be generalized as it 
is shown in Figure 6. 

When (3, X and co are received by A, the node A recalculates 
x\f from a, P, X and co, then compares the result with the value 
\\i of already held, if equality, the route (A,B,C,D) is secured, 
otherwise, the route contains a black hole nodes. B and C 



V=li{6| | to) © 



5=h(n||X)o 




a=h(id A ||SA)X^ 



H(idn||S n ) 



P=h(idii||S B ) fc=h(idc||Sc) 



Figure 6. Basic principle of the solution, cooperative black hole case 

It is important to note that the secret S is generated by node 
/ independently of other nodes and this to prevent another node 
to replay the role of the node /. 

Once detected, the black hole node can be put in a black list 
to avoid it in all future communication and, if necessary, 
another discovery route operation can be started again. For the 
implementation of our solution, an initialization process is 
necessary. It enables to nodes to generate each one its secret 
and communicate the root value of the Merkle tree to the 
transmitter node (communicate the value 5 to the node A in 
Figure 5). 

The gray hole attack is a variant of the black hole attack in 
which a malicious node, selectively, destroys packets of the 
traffic that passes through it. Our solution can easily be adapted 
to thwart such attacks. It is remarkable that this solution can be 
integrated in any routing protocol, at the time when its 
operation is not related to any specificity of a particular routing 
protocol. Obviously, the specification of the black hole attack 
is different from a routing protocol to another. 

VI. Simulation and analysis 

In the network of the Figure 7, we use a source data node, a 
destination data node and one or more black hole nodes chosen 
randomly among nodes of the network. OPNET Modeler 
version 11.5 [16] is used as a simulator. The table II contains 
the simulation parameters. 



T AB LE II. SIMULATION PARAMETRES 


Simulation parameter 


Value 


Nodes number 


10 


Network size 


1km* 1km 


Simulation duration (sec) 


600 


Packet Inter- Arrival Time (sec) 


exponential 1) 


Packet size (bits) 


exponential(1024) 


Transmit Power (watt) 


0.0001 


Routing protocol 


AODV 


Hash function 


SHA-1 



13 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 




(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 
single black hole node and the traffic received by the 
destination node when the solution is used. The traffic received 
by the destination node is null under the effect of the attack 
without using the solution, this is justified by the destruction of 
packets per black hole node. Under the effect of the solution, le 
black hole node is eliminated and the destination node receives 
normally the traffic. 

Figure 9. shows the simulation results in the presence of 
two malicious nodes cooperating together to conduct a black 
hole attack (cooperative black hole). 



Figure 7. Network used in simulation 

In this section, we show the effect of the black hole attack 
on the network functioning in first place, and how our solution 
can neutralize such an attack in second. This, in the presence of 
one or more black hole attacks (cooperative black hole). For 
this, we consider the following metrics: 

• Traffic sent by the node source (packet/sec): indicate 
the number of packets/second sent by the source node. 

• Traffic received by the destination node (packet/sec): 
indicate the number of packets/second received by the 
destination node. 



■ Object Sovce node 

■ average (Traffic 5ertf (packets/sec)) 

■ Object: Destination node 

■ average (Traffic Received (packet&/*ec))-wthoiJt attack 

■ Object : Destination node 

■ average (Traffic Received (packets/sec))-with» attack- without solution 
Object: Destination node 

average (Traffic Received (packfi£s/sec))-with attack and sobtton 




Figure 8 . Traffic Sent & Traffic Received - Single Black hole 

Figure. 8 shows the traffic sent by the source node, the 
traffic received by the destination node without attack, the 
traffic received by the destination node in the presence of a 



I Object: Source node 

I average (Traffic Sent <packets/sec)) 

I Object: Destination node 

I average (Traffic Received (packfits/sec))-vwthoot attack 

I Object: Destination node 

I average (Traffic Received (packets/sec)>w(th attack-^nithoLit solution 

Object; Destination node 

average (Traffic Received (packets/sec))-wfth attack and solution 




Figure 9. Traffic Sent & Traffic Received - Cooperative Black hole 

Our solution detects well the cooperative black hole attack 
but with a little time lag because the detection of the 
cooperative black hole attack requires more calculation than in 
the case of a simple black hole, then an alternative route is 
found to convey the traffic to the destination node. 

To evaluate the performance of our solution, we consider 
the following evaluation metrics: 

• End to end delay (sec): indicate the delay, in second, 
for sending a bit of the source node to the destination 
node. 

• Network load (bits/sec): indicate the traffic quantity, in 
bits/sec, in the entire network. 

Figure 10. and Figure 11. show, respectively, the end to end 
delay and the network load, in cases : without black hole 
attack, with black hole attack and solution, and cooperative 
black hole attack and solution. 

A very small time lag due to tests and calculations carried 
out by nodes, after which, the delay is stabilized and become 
almost identical in all scenarios, which shows that our solution 
does not influence the degradation of the network performance. 

A slight increase in the network load because of messages 
exchanged between nodes to communicate various hash values. 
More messages in the case of cooperative black hole, 
comparatively with the case of single black hole, which 
justified graphs pace in Figure 1 1 . 



14 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



0,30 



0,25 



0,20 



0,15 



0,10 



0,05 



I Without Uacl<hcfe attack 
I Single blackhole attack with solution 
I Cooperative blackhole attack with solution 
Delay (sees) 




mAtAm/k ^A/UfojfYW 



200 



~1~ 
400 



"1 
600 
time (sec) 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 
Moreover, Gray hole, a variant of the black hole, can easily be 
detectable by our solution. Simulation results show well its 
effectiveness in the detection of such attacks and the slight 
influence of this solution on the network performance. It was 
announced that our solution is general and it is not dedicated to 
any routing protocol. 

We think that the network density, nodes mobility and the 
number of black hole nodes are determining factors in our 
solution performance, in term of end to end delay and network 
load. In a future work, we will study the influence of these 
factors and will find adequate mechanisms which will make the 
solution more powerful. 

Acknowledgment 

Colleagues, H. Moumen and S. Mustapha are thanked for 
the second reading of this paper and for pertinent criticisms and 
constructive remarks. 



Figure 10. End to end delay 



I Without btackhole attack 

I Single Mackhole attack with solution 

I Cooperative blackhote attack with solution 

average (Network Load (bits/fee)) 




Figure 1 1 . Network load 

In the issue of these results, the network performance 
remains acceptable, if it is only a negligible variation in the end 
to end delay and a slight increase in the network load. 

VII. CONCLUSION ANDFUTURWORK 

Ad hoc network security is a serious problem by which 
researchers were concerned. Several security solutions are 
suggested but perfect security is far from beings obvious. We 
focused in this paper to the black hole attack, which refuses to 
convey the traffic and drop it. After the black hole attack 
specification in an example of routing protocol (AODV), we 
proposed a solution, based on the principle of Merkle tree, for 
avoiding the black hole and the cooperative black hole attacks. 



References 

[I] B. Kannhavong, H. Nakayama, Y. Nemoto, N. Kato, and A. Jamalipour, 
"A survey of routing attacks in mobile ad hoc networks," in IEEE 
Wireless Communications, Oct. 2007, pp. 85-91. 

[2] A. J. Menezes, P. V. Oorschot, and S. A. Vanstone, "Handbook of 
Applied Cryptography", CRC Press, 1996. 

[3] J. L. Munoz, J. Forne, O. Espazara, and M. Soriano, "Certificate 
revocation system implementation based on the merkle hash tree," 
International Journal of Information Security, vol. 2, no. 2, pp. 110-124, 
Jan. 2004. 

[4] C. Hongsong, J. Zhenzhou, and H. Mingzeng, "A novel security agent 
scheme for aodv routing protocol based on thread state transition," Asian 
Journal of Information Technology, vol. 5, no. 1, pp. 54-60, 2006. 

[5] C. Perkins, E. B. Royer, and S. Das, "Ad hoc on-demand distance vector 
(aodv) routing," RFC: 3561, Nokia Research Center, 2003. 

[6] S. Kurosawa, H. Nakayama, N. Kato, A. Jamalipour, and Y. Nemoto, 
"Detecting blackhole attack on aodv-based mobile ad hoc networks by 
dynamic learning method," International Journal of Network Security, 
vol. 5, no. 3, pp. 338-346, Nov. 2007. 

[7] P. N. Raj and P. B. Swadas, "DPRAODV: A dyanamic learning system 
against blackhole attack in aodv based manet," IJCSI International 
Journal of Computer Science Issues, vol. 2, pp. 54-59, 2009. 

[8] H. Weerasinghe and H. Fu, "Preventing cooperative black hole attacks 
in mobile ad hoc networks: Simulation implementation and evaluation," 
International Journal of Software Engineering and Its Applications, vol. 
2, no. 3, pp. 39-54, Jul. 2008. 

[9] S. Ramaswamy, H. Fu, M. Sreekantaradhya, J. Dixon, and K. Nygard, 
"Prevention of cooperative black hole attack in wireless ad hoc 
networks," in International Conference on Wireless Networks 
(ICWN'03), Las Vegas, Nevada, USA, 2003. 

[10] L. Tamilselvan and V. Sankaranarayanan, "Prevention of co-operative 
black hole attack in manet," Journal of Networks, vol. 3, no. 5, pp. 13- 
20, May 2008. 

[II] P. Agrawal, R. K. Ghosh, and S. K. Das, "Cooperative black and gray 
hole attacks in mobile ad hoc networks," in Proceedings of the 2 nd 
International Conference on Ubiquitous Information Management and 
Communication (ICUIMC'08). New York, NY, USA: ACM, 2008, pp. 
310-314. 

[12] S. Yi, P. Naldurg, and R. Kravets, "Security-aware ad hoc routing for 
wireless networks," in Proceedings of the 2nd ACM international 
symposium on Mobile ad hoc networking & computing (MobiHoc'01). 
New York, NY, USA: ACM, 2001, pp. 299-302. 

[13] Y. Hu, B. J. David, and P. Adrian, "Sead: secure efficient distance 
vector routing for mobile wireless ad hoc networks," Journal of Ad Hoc 
Networks, vol. 1, no. 1, pp. 175-192, Jul. 2003. 



15 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. I, 2010 
[14] S. Marti, T. J. Giuli, K. Lai, and M. Baker, "Mitigating routing Oued Smar, Algiers, Algeria in 2001 and the magister degree, option 

misbehavior in mobile ad hoc networks," in Proceedings of the 6 th networks and distributed systems from the doctoral school of networking and 

annual international conference on Mobile Computing and networking distributed systems, department of computer science, A. Mira University, 

(MobiCom'00). New York, NY, USA: ACM, 2000, pp. 255-265. Bejaia, Algeria in 2005. Currently, he is preparing his PhD under the direction 

[15] D. Eastlake and P. Jones, "Us secure hash algorithm 1 (shal)," RFC: of the professor A. Belmehdi. His research topic is wireless networking, in 

3174, Network Working Group, Sep. 2001. particular, security in ad hoc, sensor, mesh and vehicular networks. 

[16] Opnet website. [Online]. Available: http://www.opnet.com/ Ali Belmehdi is a professor at the University of A. Mira, Bejaia, Algeria. 

Besides his teaching activities, he is the header of the Laboratory of Industrial 

AUTHORS PROFILE Technology and Information. His career is rich in scientific publications and 

. , , . t» i , , • j .i • i • • supervising students in PhD or magister degree. His research topic includes 

Abderrahmane Baadache obtained the engineer degree in computer science, . ° , Al . _*■ i i • , . i • 

. ° . .7 c • mm the automatic and the computer science, particularly, wireless networking, 

option computing systems, Irom the national institute ol computing (INI), 



1 6 http://sites.google.com/site/ijcsis/ 

ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 



Design of Current Controller for Two Quadrant 
DC Motor Drive by Using Model Order Reduction 

Technique 



K.Ramesh 
EEE Department, 
Velalar College of Engg. & Tech., 
Erode -638012, India 



K.Ayyar 
EEE Department, 
Velalar College of Engg. & Tech., 
Erode -638012, India 



Dr.A.Nirmalkumar 
EEE Department, 
BIT, Sathyamangalam, 
India 



Dr.G.Gurusamy 
EEE Department, 
BIT, Sathyamangalam 
India 



Abstract -In this paper, design of current controller for a two 
quadrant DC motor drive was proposed with the help of model 
order reduction technique. The calculation of current controller 
gain with some approximations in the conventional design 
process is replaced by proposed model order reduction method. 
The model order reduction technique proposed in this paper 
gives the better controller gain value for the DC motor drive. 
The proposed model order reduction method is a mixed method, 
where the numerator polynomial of reduced order model is 
obtained by using stability equation method and the denominator 
polynomial is obtained by using some approximation technique 
preceded in this paper. The designed controller's responses were 
simulated with the help of MATLAB to show the validity of the 
proposed method. 

Keywords- Current controller, Model order reduction, Integral 
Square Error. 

I. INTRODUCTION 

DC motors used in many applications such as steel 
rolling mills, electric trains and robotic manipulators require 
current and speed controllers to perform tasks. Major 
problems in applying a conventional control algorithm in a 
controller design are the effects of nonlinearity in a DC motor. 
The nonlinear characteristics such as friction and saturation 
could degrade the performance of conventional controllers. 
Many advanced model-based control methods such as variable 
structure control and model reference adaptive have been 
developed to reduce these effects. However, the performance 
of these methods depends on the accuracy of system models 
and parameters. In this paper current controller of two 
quadrant DC motor drive is considered. The linear operation 
of DC motor drive was taken in to account in the design stage 
of current controller. In conventional design methods, some of 
simplification processes are considered to design the controller 
parameter values but where as in this proposed method, the 
model order reduction technique was introduced for the 
controller parameter design values. 

Both in systems and control engineering and in 
numerical analysis, a wealth of model order reduction 
techniques have been developed. Balanced truncation, Krylov 
subspace methods, proper orthogonal decomposition and other 



SVD-based methods are just a few classes of methods that 
have been developed. 

The computation of equivalent linear system models of 
large linear dynamic systems is a topic of considerable 
practical interest. This interest is motivated by the reduced 
complexity obtained by reducing the large linear sub-network 
in a linear (or nonlinear) network. Ideally, linear analysis on 
these sub-networks is performed by first computing a state 
space model or equivalent transfer function form, followed by 
the application suitable analysis method. However, the 
applicability of this method is limited since typical dynamic 
systems are represented by very large scale matrices that 
require specialized large-scale eigen analysis programs and 
computer resources. To avoid this practical limitation, model- 
order reduction methods are widely used in the solution of 
such systems. The basic idea behind model-order reduction is 
to replace the original system equations with a much smaller 
state-space or transfer function dimension. In particular, the 
identified reduced order model frequency characteristics must 
approximate those of the full order model. 

In the analysis of many systems for which the physical 
laws are well known, one is frequently confronted with 
problems arising from the high dimensions of descriptive state 
model, the famous curse of dimensionality. The reduction of 
such high order systems (also termed as large scale systems) 
into low order models is one of the important problems in 
control and system theory system and is considered important 
in analysis, synthesis and simulation of practical systems. The 
exact analysis of high order systems is both tedious and costly. 

To overcome the stability problem Hutton & Friedland 
[1] and Appiah [2] gave different methods, called stability 
based reduction methods which make use of some stability 
criterion. Other approaches in this direction include the 
methods such as Shamash [3] and Gutman, Mannerfelt & 
Molandor [4] which do not make use of any stability criterion 
but always lead to the stable reduced order models for stable 
systems. Bosley and Lees [5] and others have proposed a 
method of reduction based on the fitting of the time moments 
of the system and its reduced model but these methods have a 
serious disadvantage that the reduced order model may be 
unstable even though the original high order system is stable. 
Some combined methods are also given for example Shamash 



17 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 



[6], Chen, Chang and Han [7] and Wan [8] in which the 
denominator of the reduced order model is derived by some 
stability criterion method while the numerator of the reduced 
model is obtained by some other methods. [9]. 

In this paper, a new model order reduction method is 
proposed and its helps in finding the current controller gain 
value. Simulation results were shows the validity of the 
proposed method. The proposed model order reduction 
method is a mixed method, where the numerator polynomial 
of reduced order model is obtained by using the stability 
equation method and numerator polynomial is obtained by the 
method proposed in the paper [10]. 

II.DC MOTOR DRIVE 

The control schematic of a two-quadrant converter- 
controlled separately-excited DC motor drive is shown in 
figure 1 . The motor drive shown is a speed controller system. 
The thyristor bridge converter gets its ac supply through a 
three phase transformer and fast acting ac contactors. The dc 
output is fed to the armature of the dc motor. The field is 
separately excited, and the filed supply can be kept constant or 
regulated, depending on the need for the field weakening 
mode of operation. The DC motor has a tachogenerator whose 
output is utilized for closing the speed loop. The motor is 
driving a load considered to be factional for this treatment. 
The output of the tachogenerator is filtered to remove the 
ripples to provide the signal, co mr . The speed command co r * is 
compared to the speed signal to produce a speed error signal. 
This signal is processed through a proportional-plus-integral 
(PI) controller to determine the torque command. The torque 
command is limited, to keep within the safe current limits and 
the current command is obtained by proper scaling. The 
armature current command i a * is compared to the actual 
armature command i a to have a zero current error. The PI 
controller produces the equivalent control signal V c when an 
error signal is occurred. The control signal accordingly 
modifies the triggering angle a to be sent to the converter for 
implementation. 

The current control loop of DC motor drive is shown in 
the figure 2. The DC machine contains an inner loop due to 
the induced emf. It is not physically seen; it is magnetically 
coupled. The inner current loop will cross this back-emf loop, 
creating a complexity in the development of the model. The 
inner current loop assures a fast current response and also 
limits the current to a safer level. The inner current loop makes 
the converter a linear current amplifier. The outer speed loop 
ensures that the actual speed is always equal to the 
commanded speed and that any transient is overcome with in 
the shortest feasible time without exceeding the motor and 
converter capability. 

The operation of the closed loop speed controlled drive 
is explained from one or two particular instances of speed 
command. A speed from zero to rated value is recommended, 
and the motor is assumed to be at standstill. This will generate 
a large speed error and a torque command and in turn an 
armature current command. The armature current error will 



generate the triggering angle to supply a a preset maximum 
DC voltage across the motor terminals. The inner current loop 
will maintains the level permitted by its commanded value, 
producing a corresponding torque. As motor starts running, the 
torque and current are maintained at their maximum level, thus 
accelerating the motor rapidly. 



Hira^Mi'Jlsmif 



'«- 1 ^JFm^f- 







Fig. 1 . Speed-controlled two-quadrant DC motor drive 

When the motor attains the commanded value, the torque 
command will settle down to a value equal to the sum of the 
load torque and other losses to keep the motor in steady state. 
The DC machine contains an inner loop due to the induced 
emf. It is not physically seen; it is magnetically coupled. The 
inner current loop cross this back-emf loop, creating a 
complexity in the development of the model and is shown in 
the fig.2. 

Tliree-phase 
ac supply Xi 

r 




Fig.2. DC motor and current-control loop 

The design of the gain constants of the current 
controllers is of paramount importance in meeting the dynamic 
specifications of the motor drive. 

III.DESIGN OF CONTROLLERS 

The overall closed-loop system of DC motor drive is 
shown in fig.3. The design of control loops starts from the 
innermost (fastest) loop to the outer (slowest) loop. The reason 



18 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 



to proceed from the inner to the outer loop in the design 
process is that the gain and time constants of only one 
controller at a time are solved, instead of solving for the gain 
and time constants of all controllers simultaneously. 



*&£bF 



Ooiiholla 




This is a fourth order system, and simplification is 
necessary to synthesize a controller without resorting to a 

computer. Noting that T m is on the order of a second and in 

the vicinity of the gain crossover frequency, we see that the 
following approximation is valid. 



(l + sT m ) = sT m 
This reduces the loop gain function to 



(2) 



GH i (s) = 



(l + ^)(l + 57 2 )(l + ^ r ) 



(3) 



Fig. 3. Block diagram of the DC motor drive 

In some applications such as traction application, the 
motor drive need not be speed controlled but may be torque 
controlled. In that case, the current loop is essential and exists 
regardless of whether the speed control loop is going to be 
closed. The performance of the outer loop is dependent on the 
inner loop; therefore, the tuning of the inner loop has to 
precede the design and tuning of the outer loop. 

IV.CURRENT CONTROLLER 

The current control loop is shown in figure 4. 



K c (l+iT c ) 



*T< 



V, 



l+sT r 



V a 



K r 



l+sT ir 



(1+sTO (l+sT 2 ) 



H ( 



Fig.4. Current-control loop 
The loop gain transfer function is, 

,w T c sQ+sT,)<\+ S T 2 )(l+sT r ) 

Where, 
T b T 2 = Electrical time constant of the motor, 
B t = Total friction co-efficient in N-M/ (rad/sec) 
J = Moment of inertia, kg-m2 
R a = DC machine armature resistance 
L a = DC machine armature inductance 
K c = Current controller gain 
T c = Current controller time constant 
K r = Converter gain 
T r = Converter delay time 
co m = DC motor speed. 



(i) 



Where, 



K = K \ K C K r H C T m 



(4) 



The time constants in the denominator are seen to have 
the relationship 



T r <T 2 <T x 



(5) 



The equation (3) can be reduced to second order, to 
facilitate a simple controller synthesis, by judiciously selecting 



T C =T 2 
Then the loop function is 
GH t (s) = 



K 



(i + .r 1 )(i + ^ r ) 



(6) 



(7) 



The conventional design methods uses the 
approximations used in the equations from (2)-(7).We can't 
assure that the approximations were made are good in the 
design aspect. So the proposed Model order reduction 
technique is applied to get the equivalent reduced order model 
of order two. The validity of the proposed method is verified 
through the time and frequency domain analysis. 

The characteristic equation or denominator of the 
transfer function between the armature current and its 
command is 



(l+sTi)(l+sT r )+K=0 
This equation is expressed in standard form as 



(8) 



T x T r 



s + s 



r T x +T r ^ 



+ - 



K + \ 
TT 

1 \ 1 r J 



= 



(9) 



19 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 



by, 



The general second order characteristic equation is given 



s 2 +2%6) n s + (o n = 



(10) 



On comparing the equations (9) and (10) with the 
damping ratio % as 0.707 for satisfactory operation of system, 

we get the value for the term K. From that value, the current 
controller gain K c can be easily estimated. 



Step 2: The numerator polynomial of reduce order system is 
obtained by using the method proposed in [10]. 

Consider the given system transfer function given in (1 1) 
G(s)= a + a Y s + a 2 s 2 + + <V m 



b + b Y s + b 2 s + + b n _ l s n + b n s n 

(l + A l s + A 2 s 2 + + A m 5 m ) ;n>m 

(1 + B x s + B 2 s 2 + + B n _ x s n ~ l + B n s n ) 



(16) 



V.MODEL ORDER REDUCTION METHOD 
Let the nth order system is given by the transfer function 



Os). 



Yd/ 






a^+c^s+c^s + ±a m s 

j D(s) b +b 1 s+b 2 s 2 + i-b^s"^ +b n s" 



(11) 



and the corresponding r (r<n) order reduced order is of the 
form 



Where, K= 



<3n 



A i= — ;i = 0,1,2,3. ..m and 



Bi= -;j = 0,l,2,3...n 

J J ? -J 7 7 7 



Let the transfer function of the approximating low-order 
system be, 



Gr(s)= 



N(s) 



d +d x s+d 2 s 2 + ^rd r _ x s r l 



D(s) e +e x s+e 2 s + +e r _ x s r +e r s r 



(12) 



(l + CiS + C 2 s 2 + +Cs q ) 

Gr(s) = K l — q -— 

(l + D lS + D 2 s 2 + +D p _ l s p ~ 1 +D p s p ) 

where n>p>q 



(17) 



The proposed model order reduction method consists of three 
steps 



The coefficients of D b D 2 , .. were obtained from the equation 
(15). 



Step 1 : The denominator polynomial of reduced order model 
is obtained by using the stability equation method [11]. 



The following relation should be satisfied as closely as 
possible 



For stable G(s) the eve and odd parts of D(s) 
may be factored as the following stability equations. 



\ G U C °)\ 2 =1 for 0<co<oo 

\G r (Jv)\ 2 



(18) 



D e (s) = e o yi 



\ + ^ 



-t j 



("-!)/ 



D (s) = e l H 

Where, 



i=i 



1 + 



2 A 



pfj 



(13) 
(14) 



2 ^ 2^2^ 2 ^ 

z x <p x <z 2 <p 2 < . 



After discarding the factors with larger magnitude of zi 
and pi, the stability equations are reduced, which are 
combined to give a reduced polynomial of order r, as: 



D r (s)=b n 



■2/ j\ 

1+- 

V Z i J 



(r -y 2 

i=i 






= 5V 



(15) 



J 



j=o 



G(s)_(l+A^+V+ ±AJ n )$+D,s+D 2 s 2 + ±D p _/~ l +D/) 

G r (s)~ (\+B l s+B 2 s 2 + ±B n _/- 1 +B/)([+C [ s+C 2 s 2 + .+C/) 

= (1 + m x s + m 2 s 2 + + m u s u ) 

(1 + l Y s + l 2 s 2 + + / v _ 1 ^ v " 1 + l v s v ) 



Where, u=m+p and v=n+q 

\G(jco)\ 2 _ G(s)G(-s) 



\G r (jo)\ 2 G r (s)G r (-s) 



(19) 



(20) 



The equation produces the even powers of s' and can be 
written as, 



1 + L 2 s 2 + L 4 s 4 + . 



+ L lu s I 



\G(jm)\ 2 

\G r (j(u)\ 2 l + M 2 s 2 +M 4 s 4 + +M 2v s 2v 



(21) 



20 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



\G(M\ 2 


_ u (L 2 -M 2 )s 2 +(L i -M i )s 4 +.. 


.....+ (L lu -M i y u 


\G r (J0))\ 1 


\ + M 2 s 2 +M/+.... 


...+ M lv s lv 



To satisfy the above equation, 
L 2 = M 2 

L 4 = M 4 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 

(22) ±10V. The tachogenerator has the transfer function 
G ( s ) = QQ65 • The speed reference voltage has a maximum 

" 1 + 0.0025 

of 10V. The maximum current permitted in the motor is 20A. 
Converter transfer function: 

1.35V 1.35x230 



L 2u = M 2v ;ifu=v (23) 

If u<v, then the error generated by the lower order model is, 

(24) 



I i \GUa>)\* 1 



\G r {Ja)\ 2 
From the above equations, we can obtain the conditions as 



^-=Z(-l)''2m,.m 2x _,. + (-ir 

i=0 

forx= 1,2, 3 u andm =l 



m v 



(25) 



and 



j-i 



M 2y =X(-l)'2/ / / 2 ,_ / +(-!)'/, 



(26) 



fory= 1,2, 3 vandl =l 

From the equation (11), the reduced order model 
numerator coefficients can be obtained. Finally the reduced 
order model is in the form of 



Gr(s) = K 



(l + C lS + C 2 s 2 + + C q s q ) 

(l + D lS + D 2 s 2 + +D p _ x s p - 1 +D p s p ) 



(27) 



Step 3: The coefficients of the reduced order denominator 
polynomial is adjusted further to get the better approximation 
with original system. The coefficient of V term in the 
denominator is increased by n% and the same was reduced 
from the coefficient of the term ( s \ The value of n is choosen 
by trial and error method. Normally the value of n is ranging 
between 1 tol5. 

VI.DESIGN EXAMPLE 

The motor parameters and ratings of a speed controlled 
DC motor drive maintaining the field flux constant are as 
follows 220V, 8.3A, 1470 rpm, R a =4Q, J=0.0607kg-m 2 , 
L a =0.072H, B t =0.0869N-m/rad/sec, K b =1.26V/rad/sec. The 
converter is supplied from 230V, 3-PhaseAC at 50Hz. The 
converter is linear, and its maximum control input voltage is 



K = 



31.05 



V 10 

cm 

V dc (max) = 310.5 V 

The rated DC voltage required is 220V, which 
corresponds to a control voltage of 7.09V. The transfer 
function of the converter is, 



G r (s) = 



31.05 



(1 + 0.001385) 



(28) 



Current transducer gain: The maximum safe control voltage 
is 7.09V, and this has to correspond to the maximum current 
error: 

Imax — 2,0 A. 



7.09 
H r = = 0.355V/ A 



Motor transfer function: 



K, = 



1 K 2 b +* a B t 



= 0.0449 



Tl=0.1077 sec and T2=0.0208 sec 



T m = — = 0.7 sec 

m B 



The subsystem transfer functions are 

I M = K (1 + ^J = 0.0449(1 + 0.7.) (29) 

V a (s) ''(l + sT.Xl + sTz) (1 + 0.0208^(1 + 0.10775) 

co m (s) = K b IB t = 14.5 (30) 

'«« ( 1 + ^ m) (1 + 0.75) 

Design of current controller: 

T 2 = 0.0208 sec; Tc=0.03 sec 

By applying the proposed model order reduction 
technique, with the value of ^=0.707, the value of K is 
obtained as, 

K= 357.192 



21 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. I, 2010 



K c = 



KT C 



K Y H c K r T m 



- = 35.719 



By using trial and error method the value of Kc can be 
adjusted in to the value of 3 . 1 to obtain the better response for 
the current controller. The validity of this proposed method is 
evaluated by plotting the frequency response of the closed 
loop current to its command. This is shown in fig. 5. 




Fig.5 Frequency responses of actual and reduced order 
systems 

From this figure, it is evident that the proposed method is 
quite valid in the frequency range of interest. The same in time 
domain analysis is shown in fig. 6. 





Step Response 
i i i i i i i i i 




1 

0.8 


r 




I 06 


I 




0.4 












— Actual system response 




0.2 


" 


— Proposed approximation method 






■ 






1 234567391 






Time (sec) 



The time responses are important to verify the design of 
the controller and it is shown in fig. 7 for different values of 
current controller gain, K c . 










Step Response 




4.5 












' 


' 


i i i i 






4 












3.5 












3 
% 2.5 
1 2 


L 










w~ 








1.5 
1 












0.5 






Current controller response with Kc=35.719 






















0.05 


0.1 


0.15 0.2 0.25 0.3 


0.35 








Time (sec) 





Fig.6 Time responses of actual and reduced order systems 




22 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. I, 2010 




Fig.7 (a) Current controller response with K c =50 (b) Current 

controller response with K c =35.719 (c) Current controller 

response with K c =10 (d) Current controller response with 

K c =3.1 

The value of K c can be optimized by trail and error 
method. The integral square error (ISE) was calculated in 
between the actual and approximated system as 0.0204 and 
this shows the validity of the proposed method. 



VII.CONCLUSION 

The design of the gain constant of current controller is 
of paramount importance in meeting the dynamic 
specifications of the motor drive. This paper helps to obtain 
the same with the help of model order reduction technique. 
The validity of the proposed method was illustrated through 
the design example. The proposed model order reduction 
method is mathematically simple and produces the stable 
reduced order system if the given system is stable. The 
optimal value of current controller gain to suppress the 
oscillations at the output was obtained by trail and error 
method but it may be selected optimally by applying the 
neural network concepts or genetic algorithm. 

REFERENCES 

[1] M.F. Hutton and B. Friedland, "Routh approximations for reducing 

order linear time invariant systems", IEEE Trans, on Auto. Control, 

Vol.20, pp. 329-337, 1975. 
[2] R.K. Appiah, "Linear model reduction using Hurwitz polynomial 

approximation", Int. J. Control, Vol.28, No.3, pp. 477-488, 1978. 
[3] Y. Shamash, "Truncation method of reduction: a viable alternative", 

Electronics Letters, Vol.17, pp. 97-99, 1981. 
[4] P.O. Gutman, C.F. Mannerfelt and P. Molander, "Contributions to the 

model reduction problem", IEEE Trans, on Auto. Control, Vol.27, 

pp.454-455, 1982. 
[5] M.J. Bosley and F.P. Lees, "A survey of simple transfer-function 

derivations from high order state variable models", Automatica, Vol.8, 

pp. 765-775, 1972. 




[6] Y. Shamash, "Model reduction using the Routh stability criterion and the 

Pade-approximation technique" Int. J. Control, Vol.21, pp. 475-484, 

1975. 
[7] T.C. Chen, C.Y. Chang and K.W. Han, "Reduction of transfer functions 

by the stability equation method", Journal of Franklin Institute, 308, pp. 

389-404, 1979. 
[8] Bai-Wu Wan, "Linear model reduction using Mihailov criterion and 

Pade approximation technique", Int. J. Control, Vol.33, pp. 1073-1089, 

1981. 
[9] T.C. Chen, C.Y. Chang and K. W. Han, "Model Reduction using the 

stability-equation method and the Pade approximation method", Journal 

of Franklin Institute, Vol.30, No.9, pp. 473-490, 1980. 
[10] K. Ramesh, A. Nirmalkumar and G. Gurusamy, "Design of Digital IIR 

filters with the Advantages of Model Order Reduction Technique", 

International Journal of Electronics, Communications and Computer 

Engineering, Vol.1, No.2,pp 117-122, 2009 



AUTHORS PROFILE 

Ramesh.K received B.E. degree in Electrical 
and Electronics Engineering from 
K.S.R.College of Technology, Tamilnadu, 
India in 2002 and M.E. in Applied 
Electronics from Kongu Engg. College, Anna 
University, Chennai in 2005 and also 
received the MBA in System from PRIDE, 
Salem (Tamilnadu), India in 2005. Since then, he is working 
as a Assistant Professor in Velalar College of Engineering and 
Technology (Tamilnadu), India. Presently he is a Part time 
(external) Research Scholar in the Department of Information 
and Communication at Anna University, Chennai (India). His 
fields of interests include Model order reduction, Controller 
design and Optimization Techniques. 

Ayyar.K received B.E. Degree in Electrical 
and Electronics Engineering from Ranipettai 
Engineering College, Tamilnadu, India in 
2003 and M.E. in Power Electronics and 
Drives from Government College of 
Engineering, Salem, Anna University, 
Chennai in 2007. Since then, he is working 
as a Lecturer in Velalar College of Engineering and 
Technology (Tamilnadu), India. Presently he is a Part time 
(Internal) Research Scholar in the Department of Electrical 
and Electronics Engineering at Anna University, Coimbatore 
(India). His fields of interests include Power Drives and 
control, Controller design and System Optimization. 

Dr.Nirmalkumar.A, received the 

B.Sc.(Engg.) degree from NSS College of 
Engineering, Palakkad in 1972, M.Sc.(Engg.) 
degree from Kerala University in 1975 and 
completed his Ph.D. degree from PSG Tech 
in 1992. Currently, he is working as a 
Professor and Head of the Department of 
Electrical and Electronics Engineering in Bannari Amman 
Institute of Technology, Sathyamangalam, Tamilnadu, India. 





23 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 



His fields of Interest are Power quality, Power drives and 
control and System optimization. 

Dr.G.Gurusamy obtained his Pre-University 
education at St Johns College, Palayamkottai, 
and Thirunelveli district. He then joined PSG 
College of Technology, Coimbatore, in the 
year 1962 to pursue his Engineering course. 
He was graduated in Electrical Engineering in 
1967. He latter obtained M.E (Applied 
Electronics) in 1972 and Ph.D in Control 
Systems in 1983. He is currently working as a Dean of 
Department of Electrical and Electronics Engineering in 
Bannari Amman Institute of Technology, Sathyamangalam. 
His fields of Interest are Advanced control, Digital control, 
Optimization and Biomedical electronics. 




24 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 



WIRELESS CONGESTION CONTROL 
PROTOCOL FOR MULTIHOP AD HOC 

NETWORKS 



Mahendra kumar.S 
Department of Electronics and Communication 

Engineering 

Velalar College of Engineering and Technology 

Tamil Nadu, India. 



Senthil Prakash.K 
Department of Electronics and Communication 

Engineering 
Velalar College of Engineering and Technology 

Tamil Nadu, India. 



Abstract — The traditional TCP congestion control mechanism 
encounters a number of new problems and suffers a poor 
performance when the IEEE 802.11 MAC protocol is used in 
multihop ad hoc networks. Many of the problems result from 
medium contention at the MAC layer. In this paper, I first 
illustrate that severe medium contention and congestion are 
intimately coupled, and TCP's congestion control algorithm 
becomes too coarse in its granularity, causing throughput 
instability and excessively long delay. Further, we illustrate 
TCP's severe unfairness problem due to the medium contention 
and the tradeoff between aggregate throughput and fairness. 
Then, based on the novel use of channel busyness ratio, a more 
accurate metric to characterize the network utilization and 
congestion status, 

I propose a new wireless congestion control protocol 
(WCCP) to efficiently and fairly support the transport service in 
multihop ad hoc networks. In this protocol, each forwarding node 
along a traffic flow exercises the inter-node and intra-node fair 
resource allocation and determines the MAC layer feedback 
accordingly. The end-to-end feedback, which is ultimately 
determined by the bottleneck node along the flow, is carried back 
to the source to control its sending rate. Extensive simulations 
show that WCCP significantly outperforms traditional TCP in 
terms of channel utilization, delay, and fairness, and eliminates 
the starvation problem. 

Keywords-component; Medium access control; Congestion 
control; Fairness; Multihop ad hoc networks; WCCP; 



I. Introduction 

Wireless ad hoc networks have found many applications in 
battlefield, disaster rescue and conventions, where fixed 
communications infrastructures are not available and quick 
network configurations are needed. To provide reliable 
transport services over and hence fully exploit the potential of 
ad hoc networks, efficient congestion control is of paramount 
importance. Unfortunately, the traditional TCP congestion 
control mechanism performs very poorly. As shown in recent 
studies [4]. 



TCP congestion control has an implicit assumption, i.e., 
any packet loss is due to network congestion. However, this 
assumption is no longer valid in the ad hoc networks as packet 
losses may well be due to channel errors, medium contention, 
and route failures. 

In this paper, I mainly focus on the problems arising from 
medium contention when the IEEE 802.11 MAC protocol is 
used in the multi-hop ad hoc networks. I show that a rate based 
congestion control protocol is more appropriate than its 
window based counterpart in multihop ad hoc networks. We 
illustrate the close coupling between congestion and medium 
contention, which explains the instability of TCP. We also find 
that the optimum congestion window size of TCP may be less 
than one even in a very simple topology, say chain topology, in 
order to maximize the end-to-end throughput and minimize the 
end-to-end delay. Thus TCP tends to overshoot the network 
capacity and its granularity of sending rate adjustment is too 
coarse because the minimum increase in window size is the 
size of one packet upon each TCP acknowledgment or during 
each round trip time. Then we further show that medium 
contention also results in severe unfairness and starvation 
problems for TCP flows. Therefore, we conclude that 
congestion control, fairness and medium contention are all 
closely coupled in multihop ad hoc networks. 

I propose a new wireless congestion control protocol 
(WCCP) based on the channel busyness ratio. In this protocol, 
each forwarding node determines the inter-node and intra-node 
fair channel resource allocation and allocates the resource to 
the passing flows by monitoring and possibly overwriting the 
feedback field of the data packets according to its measured 
channel busyness ratio. The feedback is then carried back to the 
source by the destination, which copies it from the data packet 
to its corresponding acknowledgment. Finally, the source 
adjusts the sending rate accordingly. Clearly, the sending rate 
of each flow is determined by the channel utilization status at 
the bottleneck node. In this way, WCCP is able to approach the 
max-min fairness in certain scenarios. We compare WCCP 
with TCP through extensive simulations in Section IV. We 
observe that WCCP significantly outperforms TCP in terms of 
channel utilization, delay, and fairness. Especially, it solves the 
starvation problem suffered by TCP. 



25 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



A. MAC layer 

The Media Access Control data communication protocol 
sub-layer, also known as the Medium Access Control, is a 
sublayer of the Data Link Layer. It provides addressing and 
channel access control mechanisms that make it possible for 
several terminals or network nodes to communicate within a 
multipoint network [3] [5], typically a local area network or 
metropolitan area network . The hardware that implements the 
MAC is referred to as a Medium Access Controller. The MAC 
sub-layer acts as an interface between the Logical Link Control 
sub layer and the network's physical layer. The MAC layer 
emulates a full-duplex logical communication channel in a 
multipoint network. This channel may provide unicast, 
multicast or broadcast communication service. 

B. AODV 

Ad hoc On-Demand Distance Vector Routing is a routing 
protocol for mobile ad hoc networks and other wireless ad-hoc 
networks [2] [4]. AODV is capable of both unicast and multicast 
routing. It is a reactive routing protocol, meaning that it 
establishes a route to a destination only on demand. In contrast, 
the most common routing protocols of the Internet are 
proactive, meaning they find routing paths independently of the 
usage of the paths. AODV is, as the name indicates, a distance- 
vector routing protocol. AODV avoids the counting-to-infinity 
problem of other distance-vector protocols by using sequence 
numbers on route updates, a technique pioneered by DSDV. 

II. LITERATURE SURVEY 

Good and bad network performance is largely dependent on 
the effective implementation of network protocols. TCP, easily 
the most widely used protocol in the transport layer on the 
Internet (e.g. HTTP, TELNET, and SMTP), plays an integral 
role in determining overall network performance. Amazingly, 
TCP has changed very little since its initial design in the early 
1980's. A few "tweaks" and "knobs" have been added, but for 
the most part, the protocol has withstood the test of time. 
However, there are still a number of performance problems on 
the Internet and fine tuning TCP software continues to be an 
area of work for a number of people. The design of TCP was 
heavily influenced by the end-to end argument. Method of 
handling congestion and network overload. Are the key 
components of the end-to-end argument. 

The four congestion control algorithms used in TCP are, 
Slow Start, Congestion Avoidance, Fast Retransmit and Fast 
Recovery. Slow Start, a requirement for TCP software 
implementations is a mechanism used by the sender to control 
the transmission rate [3], otherwise known as sender-based 
flow control. This is accomplished through the return rate of 
acknowledgements from the receiver. In other words, the rate 
of acknowledgements returned by the receiver determines the 
rate at which the sender can transmit data. When a TCP 
connection first begins, the Slow Start algorithm initializes a 
congestion window to one segment, which is the maximum 
segment size (MSS) initialized by the receiver during the 
connection establishment phase. When acknowledgements are 
returned by the receiver, the congestion window increases by 
one segment for each acknowledgement returned. Thus, the 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 
sender can transmit the minimum of the congestion window 
and the advertised window of the receiver, which is simply 
called the transmission window. 



In the Congestion Avoidance [3] [9] algorithm a 
retransmission timer expiring or the reception of duplicate 
ACKs can implicitly signal the sender that a network 
congestion situation is occurring. The sender immediately sets 
its transmission window to one half of the current window size 
(the minimum of the congestion window and the receiver's 
advertised window size), but to at least two segments. If 
congestion was indicated by a timeout, the congestion window 
is reset to one segment, which automatically puts the sender 
into Slow Start mode. If congestion was indicated by duplicate 
ACKs, the Fast Retransmit and Fast Recovery algorithms [3] 
are invoked. As data is received during Congestion Avoidance, 
the congestion window is increased. However, Slow Start is 
only used up to the halfway point where congestion originally 
occurred. This halfway point was recorded earlier as the new 
transmission window. After this halfway point, the congestion 
window is increased by one segment for all segments in the 
transmission window that are acknowledged. This mechanism 
will force the sender to more slowly grow its transmission rate, 
as it will approach the point where congestion had previously 
been detected. 

The Media Access Control data communication protocol 
sub-layer, also known as the Medium Access Control, is a sub 
layer of the data link layer[3]. It provides addressing and 
channel access control mechanisms that make it possible for 
several terminals or network nodes to communicate within a 
multipoint network, typically a local area network or 
metropolitan area network. The MAC sub-layer acts as an 
interface between the Logical Link Control sub layer and the 
network's physical layer. The MAC layer emulates a full- 
duplex logical communication channel in a multipoint network. 
This channel may provide unicast, multicast or broadcast 
communication service. 

The channel access control mechanisms provided by the 
MAC layer are also known as a multiple access protocol. The 
most widespread multiple access protocol is the contention 
based CSMA/CD protocol used in Ethernet networks. This 
mechanism is only utilized within a network collision domain, 
for example an Ethernet bus network or a hub network. An 
Ethernet network may be divided into several collision 
domains, interconnected by bridges and switches [3] [5]. A 
multiple access protocol is not required in a switched full- 
duplex network, such as today's switched Ethernet networks, 
but is often available in the equipment for compatibility 
reasons. 

Several works have pointed out that greedy TCP can result 
in severe congestion in ad hoc networks and hence 
performance degradation. RED is one of the schemes proposed 
to overcome this problem [2]. Random early detection, also 
known as random early discard or random early drop is an 
active queue management algorithm. It is also a congestion 
avoidance algorithm [6]. 

In the traditional tail drop algorithm, a router or other 
network component buffers as many packets as it can, and 



26 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



simply drops the ones it cannot buffer. If buffers are constantly 
full, the network is congested. Tail drop distributes buffer 
space unfairly among traffic flows. Tail drop can also lead to 
TCP global synchronization as all TCP connections "hold 
back" simultaneously, and then step forward simultaneously. 
Networks become under-utilized and flooded by turns. RED 
addresses these issues. It monitors the average queue size and 
drops (or marks when used in conjunction with ECN) packets 
based on statistical probabilities. If the buffer is almost empty, 
all incoming packets are accepted. As the queue grows, the 
probability for dropping an incoming packet grows too. When 
the buffer is full, the probability has reached 1 and all incoming 
packets are dropped. RED is considered more fair than tail drop 
[2]. 

The more a host transmits, the more likely it is that its 
packets are dropped. Early detection helps avoid global 
synchronization. RED makes Quality of Service differentiation 
impossible. Weighted RED [5] and RED In/Out provide early 
detection with some QoS considerations. The Adaptive RED 
algorithm infers whether to make RED more or less aggressive 
based on the observation of the average queue length. If the 
average queue length oscillates around min threshold then 
Early Detection is too aggressive. On the other hand if the 
average queue length oscillates around max threshold then 
Early Detection is being too conservative. The algorithm 
changes the probability according to how aggressive it senses it 
has been discarding traffic. 

RED has some disadvantages too, Unfairness is caused as 
nodes drop packets unaware of other nodes and Queue at any 
single node cannot reflect the network congestion state to 
overcome these problems Neighborhood Random Early 
Detection is proposed. It is based on the principle that Queue 
size of a neighborhood reflects the degree of local network 
congestion. 

III. SYSTEM DESIGN - WIRELESS CONGESTION 
CONTROL PROTOCOL (WCCP) 

TCP's congestion control suffers from a coarse granularity 
when applied to the multihop ad hoc environment. To 
overcome this problem a rate based wireless congestion control 
protocol is defined. There are two components in WCCP. One 
is at the transport layer. It replaces the window adjusting 
algorithm of TCP with a rate control algorithm to regulate the 
sending rate. The other is between the networking layer and the 
MAC layer. It monitors and possibly modifies the feedback 
field in TCP data packets when it passes the outgoing packets 
from the networking layer to the MAC layer and the incoming 
packets in the reverse direction. 

A. System Architecture 

Based on the estimate of the available bandwidth, the inter- 
node and intra-node resource allocation schemes are proposed 
to determine the available channel resource for each node and 
for each flow passing through that node and accordingly 
modify the MAC layer feedback. Then an end to- end rate 
control scheme is proposed to carry the feedback from the 
bottleneck node to the source node which accordingly adjust 
the sending rate to make full and fair utilization of the channel 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
resource at the bottleneck node without causing severe medium 
contention and packet collision. 



Inter and intra 

node Resource 

Allocation 



Data 



i 



Inter and intra node 
Resource 
Allocation 













Sender 
Node 


-► 

<- 


Intermediate 
Node 


ack 






End to end 

rate Control 

Scheme 





Bottleneck 
Node 



Receiver 
Node 



Figure 1. WCCP Mechanism 

B. Measurement of Channel Busyness Ratio 

The channel busyness ratio rb is an easily measured metric 
at the location of each node under the current architecture of 
the IEEE 802.11 standard. Notice that the IEEE 802.11 is a 
CSMA-based MAC protocol, working on the physical and 
virtual carrier sensing mechanisms. There is already a function 
to determine whether the channel is busy or not, i.e., the 
channel is determined busy when the measuring node is 
transmitting, receiving, or its network allocation vector 
indicates the channel is busy, and is idle otherwise. The 
channel busyness ratio can be simply calculated as the ratio of 
the total lengths of busy periods to the total time during a time 
interval, which is the average control interval. 

C. Bandwidth Calculation 

In the rate-based congestion control algorithm, to calculate 
the ideal sending rate, the source is in direct need of a timely 
and easily measured metric which should satisfy two 
requirements. First, since MAC contention is tightly coupled 
with congestion, a candidate of congestion signal should reflect 
the condition of MAC contention and collision. Second, in 
order to fully utilize the shared channel without causing severe 
congestion and packet collision, the candidate should indicate 
the available bandwidth. Channel busyness ratio rb, meets these 
two requirements, the channel utilization cu indicates the ratio 
of the channel occupancy time for successful transmissions to 
the total time, the normalized throughput s indicates the 
achievable data rate for payload divided by the channel data 
rate and is proportional to cu and the collision probability p 
indicates the probability that each transmission encounters a 
collision, the threshold thb should be chosen such that 

rb ~ cu(rb ^ thb) 
Since the performance of rb is not sensitive to n, we can fix 
n and observe the effect of the payload size. After choosing 
thb, according to Equation (3), we can estimate the available 
bandwidth of each node, denoted by BWa, as follows 



27 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



BWa = BW (thb - rb) data/Ts, (rb<thb) 



Where BW is the transmission rate in bits/s for the 
DATA packets, data is the average payload size in unit of 
channel occupancy time, and Ts is the average time of a 
successful transmission at the MAC layer. 

D. Inter Node Resource Allocation 

According to equation above equation, each node could 
calculate the total available bandwidth for its neighborhood 
based on the measured channel busyness ratio in a period called 
average control interval, denoted by Tc. To determine the 
available bandwidth for each node, WCCP accommodates the 
channel resource AS for each node proportionally to its current 
traffic load S in Tc. 

AS = (thb - rb)/rb * s (2) 

Because both the incoming traffic and outgoing traffic of 
each node consume the shared channel resource, S should 
include the total traffic (in bytes), i.e., the sum of the total 
incoming and outgoing traffic.There is two cases when we 
compare the observed rb with thb, i.e., rb <thb and rb ^ thb. 
When rb < thb , AS is positive, meaning we should increase the 
traffic Since the available bandwidth is proportional to thb - rb 
according to equation (1), we may increase S by such an 
amount that after the increase AS, S is proportional to thb, 
which is the optimal channel utilization. Actually, equation (2) 
achieves our desired increase as it can be easily seen that 

(AS + S)/thb=s/rb (3) 

Therefore, rb will approach thb after one average control 
interval Tc when all the nodes in the neighborhood increase the 
total traffic rate according to equation (2). When rb ^ thb, AS is 
negative, meaning we decrease the traffic. In this case, 
however, the linear relationship between the available 
bandwidth and rb no longer exists, and the collision probability 
increases dramatically as the total traffic rate increases. In 
addition, when the node enters saturation, both collision 
probability and rb amount to their maximum values and do not 
change as the traffic increases, although the total throughput 
decreases. It thus appears that ideally, WCCP needs to 
aggressively decrease the total traffic rate. However, since it is 
difficult to derive a simple relationship between the traffic rate 
and rb when rb > thb, WCCP uses the same linear function as 
for the case rb < thb.. Indeed, this brings two advantages. First, 
as the increase and decrease use the same law, it is simple to 
implement at each node. Second, opting out of aggressive 
decrease helps achieve smaller oscillation in channel 
utilization. 

E. Intra Node Resource Allocation 

After calculating AS, the change in the total traffic or the 
aggregate feedback at each node, WCCP needs to apportion it 
to individual flows traversing that node in order to achieve both 
efficiency and fairness. WCCP relies on an Additive-Increase 
Multiplicative- Decrease policy to converge to efficiency and 
fairness: If AS > 0, all flows increase the same amount of 
throughput. And if AS < 0, each flow decreases the throughput 
proportionally to its current throughput. 



(IJCSIS) International lournal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 

(1) Before determining the feedback when AS >= 0, WCCP 

needs to estimate the number of flows passing through the 
considered node. Again, since the channel is shared by both 
incoming and outgoing traffic, 



J L 



Transmitted \ 

and ACKed Si as , 



TCP's Congestion 
.Window . 

I I— I I 



Queue of packets 
without a perm it 



t 

Sla SI +W s -l 



Packet arriving queie 



Maximum size is the 
receiver's advertised 
window size 



S last = oldest unacknowledged packet 
Siast+Ws-I = highest numbered packet 
I ha I can be trans mi I led 
Wg=reeeiver s advertised window 



yj Queue of packets 
wi lh a perm it 



Permit 
queue 




Arriving permits at a 
rate of one per l/r p sec 

Figure 2. Rate control mechanism 

The number of flows J used by WCCP should be different 
from the real number of flows. For those flows that either 
originate or terminate at the node, the node counts each as one 
flow, whereas for those flows only passing through the node, 
the node counts each as two flows, i.e., one in and one out. This 
is because that flows passing through the node occupy twice 
channel resource of that for flows originating or terminating at 
the node Let rpk denote the packet sending rate (pkt/s) of the 
flow which the kth observed packet during the period Tc at 
node i belongs to. Since rpkTc is equal to the number of 
packets which are observed from the corresponding flow at 
node I during Tc, we can convert the summation over each 
flow to the summation over each packet by the fact that, for all 
the packets of each flow, the summation of 1 /rpkTc is equal to 
l.Thus J can be calculated at node i as 



J=l/rph*Tc fork->ltoK 



(4) 



Where K is the total number of packets seen by node i in 
Tc. Thus each node only needs to do the summation for each 
received and transmitted packet. Therefore, if AS >= 0, the 
increasing amount of traffic rate for each flow Cp, and per 
packet feedback pfk will be 



Cp=AS/Tcj 
Pfk=Cn*Rpk/Rpk*Tc 



(5) 
(6) 



where Cn is a constant and WCCP aims to make full use of 
the channel resource while not introducing severe medium 
contention, i.e., rb should be as close to thb as possible but 
never exceed thb too much. Therefore, when the aggregate 
feedback of previously passing packets is equal to AS, the node 
sets local feedback value as zero until the next control interval 
starts. With this mechanism in place, the channel busyness ratio 
rb should be around thb at the bottleneck nodes and be smaller 
at other nodes. 



28 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



F. End-To-End Rate-Based Congestion Control Scheme 

A leaky bucket (permit queue) is attached to the transport 
layer to control the sending rate of a WCCP sender. The permit 
arrival rate rp of the leaky bucket is dynamically adjusted 
according to the explicit feedback fb carried in the returned 
ACK whenever a new ACK arrives (henceforth, ACKs refer to 
the transport layer acknowledgments). Namely, 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 
parameters of a simulation without having to recompile any 
source code. 



Rp=rp+fb 



(7) 



To enable this feedback mechanism, each WCCP packet 
carries a congestion header including three fields, i.e., rp, Tc, 
and fb, which is used to communicate a flow's state to the 
intermediate nodes and the feedback from the intermediate 
nodes to the source. The field rp is the sender's current permit 
arrival rate, and the field Tc is the sender's currently used 
control interval. They are filled in by the sender and never 
modified in transit. The last field, fb, is initiated by the sender 
and all the intermediate nodes along the path may modify it to 
directly control the packet sending rate of the source. 

IV. SYSTEM IMPLEMENTATION 

A. Tool Description 

The network simulator ns-2 is an object-oriented, discrete 
event-driven network simulator developed at the OC Berkley 
and ISC ISI as part of the VINT project [VIN03]. It is a very 
useful tool for conducting network simulations involving local 
and wide area networks. In the recent years its functionality has 
grown to include wireless and ad hoc networks as well. 

The ns-2 network simulator has gained an enormous 
popularity among participants of the research community, 
mainly because of its simplicity and modularity. The network 
simulation allows simulation scripts, also called simulation 
scenarios, to be easy written in a script-like programming 
language TCL. Emulation refers to the ability to introduce the 
simulator into a live network. Special objects within the 
simulator are capable of introducing live traffic into the 
simulator and injecting traffic from the simulator into the live 
network. The interface between the simulator and live network 
is provided by a collection of objects including Tap Agents and 
Network Objects. Tap agents embed live network data into 
simulated packets and vice-versa. Network objects are installed 
in tap agents and provide an entry point for the sending and 
receipt of live data. Both objects are described in the following 
sections. When using the emulation mode, a special version of 
the system scheduler is used: the Real Time Scheduler. This 
scheduler uses the same underlying structure as the standard 
calendar-queue based scheduler, but ties the execution of 
events to real time 

More complex functionality relies on C++ code that either 
comes with ns-2 or is supplied by the user. The utilization of 
the two programming languages increases the flexibility of ns- 
2. C++ is mainly used for event handling and per-packet 
processing tasks for which TCL would become too slow. TCL 
is most commonly used for simpler routing protocols, general 
ns-2 code and simulation scenario scripts. The usage of TCL 
for simulation scenario scripts allows the user to change 



Simulations in ns-2 can be logged to trace files, which 
include detailed information about packets in the simulation 
and allows for post-run processing with some analysis tool. It is 
also possible to let ns-2 generate a special trace file that can be 
used by NAM (Network Animator), a visualization tool that is 
part of the simulator distribution. 

B. Resource Requirements 

• Hardware requirements 



Processor Type 




Pentium-IV,512MB 
RAM 


Hard Disk 




20 GB 


• Software requirements 


Operating 
System 




Linux 


Programming 
Language 




C++ 


Tool 




Network Simulator 

2 



V. RESULTS AND DISCUSSION 

A. NAM Output 

The nam output shows the 9-node chain topology. There 
are three cases considered to compare the performance of TCP 
and Wireless congestion Control Protocol (WCCP). In case 1 
there is only one flow between nodeO to node 8 and in case 2 
there are three flows one flow is between nodeO to node4 and 
the second flow is between nodeO to node8 and third flow is 
between node4 to node 8. 



flaw 1 , 



(a) Scenario 1 



-o- 



flow 



(b) Scenario 2 




^^ 



<n 



low 2- 



L flow2, 3,4,5, 

to 



(c) Scenario 3 



^ 



flaw 6 

\ 



-o- 



-o- 



-o- 



-o- 



-o- 



^7low 1 



5 o 

S 9 



Figure 3. 9-Node Chain Topology with Different Traffic Distribution 



29 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 
Figure 3 shows various flows for the nine node topology. &^-^^-y^ 

Scenario 1 has only one flow from node 1 to node9 and 
Scenario 2 has 3 flows and Scenario 3 has 6flows 



& 



AppK jflotit Rkh S^nlMi ff 



■ 



■" QQ "a 1 ** •fa* may ». aaw if 



!*. 9-i i 



uinn ■■*<»*" 




' ' ' ' ' ' 



a **» «<»*- 



Figure 4. Nam Output 

Figure 4 shows the Nam Output for the nine node topology 
as shown in the figure 3 it shows the Nam output for 3 flows 

B. Graphs 
1) Scenariol 



U A«»c*ra"i h« r! S-Hirni (V ' i> Q " "" '** **•' •»* - 1 " J,J • 1 -" H 


OMal5SiSIj| 


Tlmil/onTT 


ra 




















HUVWSQ 




| 






























































































■1 




















' 
























































r 


T 


' 




I 


n 






, 1 








. 














1 








t J 






I 


TT 


- 






u 










1 -ft 






1 


fl 






T 


■ - 








l .; 








, 


1 




L» 


Mi 


, 


, 


| 


, :."■:, JW 


• 


. 





* f 


nfl 


L 




* , 4 " J " w 


■ : U^ 


1 V 

roocw k r 


• ■ 


llvki 


:*s 




t ' " 






i * 


ifjrj 


















^ 












OPdOi idOWO 20 ( 


"•' •'"•"" -i...i:iii ' • ■ '•«- 


30 WlO-Hld .T1DHX! BODOCQ SDOfHH] lOd 


moo 






Figure 5. Round Trip Time of TCP vs WCCP 

Figure 5. Shows the comparison of round trip time of TCP 
vs WCCP for scenariol 



2) Scenario 2 
The graph shows the comparison of throughput for TCP 
and WCCP. Throughput of WCCP is higher than TCP. 

Figure 6 shows the throughput of TCP for various flows. 
Here there are three flows. Throughput of the second flow 
which is the main flow is much smaller than the other flows. 




.Z m ." ..Jir- -(.r,M .h. 



Figure 6. TCP Throughput 



6**^' "-- • *»«■"# £ i. #0 ».*'«* =-««»WJl4 




FlowsVsThroughpul 


TtaM*nl 








































































































































































































CS 


rim ioow fm» jwoq ?hmq 3 mm 35cm 


n ■ '<KM*l*tfVJM.Sy«i a #«i-* ■4rB-L.rt-. / (»(*<,W|**H> Q.qt^* EH 



Figure 7. WCCP Throughput 



Figure 7 shows the throughput of WCCP for various flows. 
Here there are three flows. Throughput for the second flow is 
very small for TCP but it is overcome by Wireless Congestion 
Control Protocol. Here it overcomes the problem by allocating 
the resources properly at each node and by providing the 
required feedback to the source node. 

3) Scenario 3 
The graph shows the comparison of throughput for TCP 
and WCCP. Throughput of WCCP is higher than TCP. 

Figure 8 shows the throughput of TCP for various flows. 
Here there are six flows. Four flows from nodel to node3 and 
one from node7 to node9 and the main flow is from nodel to 
node9 and due to many flows the throughput for main flow is 
zero 



30 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



Alc«*f»baf« "»£« i*wl»m ,g - *> O ft -" "** iJi hn» in. JO 11 || 




















FloweVeThrouQhput 


Scen3) 










































































































































































































































































































































































































«« "** 


H ■ rm>,» 




i*" B fliwM *J»**n*i« 


. 


rtlHU 


..... 


a *™* a to 



Figure 8. TCP Throughput 




Figure 9. WCCP Throughput 

Figure 9 shows the throughput of TCP for various flows. 
Here there are six flows. Four flows from nodel to node3 and 
one from node7 to node9 and the main flow is from nodel to 
node9 and due to many flows the throughput for main flow is 
zero in TCP but it is overcome by WCCP. 

VI. CONCLUSIONS AND FUTURE ENHANCEMENT 

Congestion control is critical to reliable transport 
service in wireless multihop ad hoc networks. Unfortunately, 
traditional TCP suffers severe performance degradation and 
unfairness. Realizing that the main reason is the poor 
interaction between traditional TCP and the MAC layer, we 
propose a systematic solution named Wireless Congestion 
Control Protocol (WCCP) to address this problem in both 
layers. WCCP uses channel busyness ratio to allocate the 
shared resource and accordingly adjusts the sender's rate so 
that the channel capacity can be fully utilized and fairness is 
improved. We evaluate WCCP in comparison with TCP in 
various scenarios. The results show that our scheme 
outperforms traditional TCP in terms of channel utilization, 
end-to-end delay, and fairness, and solves the starvation 
problem of TCP flows. 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. I, 2010 

References 

[l] [1] X. Chen, H. Zhai, J. Wang, and Y. Fang, "TCP 
performance over mobile ad hoc networks," Canadian J. 
Electric. Comput. Engin., vol. 29, no. 1/2, pp. 129-134, 
Jan./April 2004. 

[2] H. Zhai, X. Chen, and Y. Fang, "Rate-based transport 
control for mobile ad hoc networks," in Proc. IEEE 

WCNC, vol. 8, no. 11, pp. 12-19, march 2005. 

[3] S. Xu and T. Safadawi, "Does the IEEE 802.11 MAC 
protocol work well in multihop wireless ad hoc networks? 
IEEE Commun. Mag., vol. 39, no. 6, pp. 130-137, June 
2001. 

[4] Z. Fu, X. Meng, and S. Lu, "How bad TCP can perform in 
mobile ad-hoc networks," in Proc. IEEE Symposium on 
Computers and Communications, vol. 6, no. 2, pp. 219- 
228 2002. 

[5] H. Zhai, J. Wang, X. Chen, and Y. Fang, "Medium access 
control in mobile ad hoc networks: challenges and 
solutions," Wireless Commun. Mobile Comput., vol. 6, 
no. 2, pp. 151-170, March 2006. 

[6] Z. Fu, P.Zerfos, H. Luo, S. Lu, L. Zhang, and M. Gerla, 
"The impact of multihop wireless channel on TCP 
throughput and loss," in Proc. IEEE INFOCOM.,vol 3, no 
6 pp. 1744-1753, April 2003. 

[7] K. Chen, Y. Xue, and K. Nahrstedt, "On setting TCP's 
congestion window limit in mobile ad hoc networks," in 
Proc. IEEE ICC, vol 6, no 11, pp.153-167, May 2005. 

[8] T. D. Dyer and R. V. Boppana, "A comparison of TCP 

Eerformance over three routing protocols for mobile ad 
oc networks," in Proc. ACM Mobihoc ,vol 6, no 2, 
pp.662-678, June 2004. 

[9] J. P. Monks, P. Sinha, and V. Bharghavan, "Limitations 
of TCP-ELFN for ad hoc networks," in Proc. MOMUC, 
vol 8, no 1 1, pp.47-67, March 2002. 

[10] X. Yang and N. Vaidya, "On physical carrier sensing in 
wireless ad hoc networks," in Proc. IEEE INFOCOM , 
vol 11, no 5, pp.603-615, May 2005. 

AUTHORS PROFILE 





Mahendra Kumar. S received the Bachelors 
degree in Electronics and Communication 
Engineering from from Anna University, Chennai in 
2006. and the Master degree in Communication 
System from Kumaraguru College of Technology, 
Coimbatore in 2009. He is currently working in 
Velalar College of Engineering and Technology as 
Lecturer of Electronics and Communication 
Department since 2009. His fields of interests 
include Ad Hoc & Sensor Network, He has published two papers in National 
Conferences and one International Conference in Ad Hoc Networks. He is a 
life member of ISTE. 

Senthil Prakash.K received the Bachelors degree 
in Electronics and Communication Engineering from 
Bharathidasan University, Tiruchirappalli in 2003. 
and the Master degree in Communication System 
from Kumaraguru College of Technology, 
Coimbatore in 2009. He is currently working in 
Velalar College of Engineering and Technology as 
Lecturer of Electronics and Communication 
Department since 2009. His area of interest includes 
Mobile communication and Ad Hoc Network, He has 

published two papers in National Conferences in Ad Hoc Networks. He is a 

life member of ISTE. 




31 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) Inter nationalJ our nal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 



Saturation Throughput Analysis of IEEE 802.1 lb 

Wireless Local Area Networks under High 

Interference considering Capture Effects 



Ponnusamy Kumar 

Department of Electronics and Communication 

Engineering 

K.S.Rangasamy College of Technology 

Tiruchengode, Namakkal, Tamilnadu, India 



A.Krishnan 

Department of Electronics and Communication 

Engineering 

K.S.Rangasamy College of Technology 

Tiruchengode, Namakkal, Tamilnadu, India 



Abstract — Distributed contention based Medium Access Control 
(MAC) protocols are the fundamental components for IEEE 
802.11 based Wireless Local Area Networks (WLANs). 
Contention windows (CW) change dynamically to adapt to the 
current contention level: Upon each packet collision, a station 
doubles its CW to reduce further collision of packets. IEEE 
802.11 Distributed Coordination Function (DCF) suffers from a 
common problem in erroneous channel. They cannot distinguish 
noise lost packets from collision lost packets. In both situations a 
station does not receive its ACK and doubles the CW to reduce 
further packet collisions. This increases backoff overhead 
unnecessarily in addition to the noise lost packets, reduces the 
throughput significantly. Furthermore, the aggregate throughput 
of a practical WLAN strongly depends on the channel conditions. 
In real radio environment, the received signal power at the access 
point from a station is subjected to deterministic path loss, 
shadowing and fast multipath fading. In this paper, we propose a 
new saturation throughput analysis for IEEE 802.11 DCF 
considering erroneous channel and capture effects. To alleviate 
the low performance of IEEE 802.11 DCF, we introduce a 
mechanism that greatly outperforms under noisy environment 
with low network traffic and compare their performances to the 
existing standards. We extend the multidimensional Markov 
chain model initially proposed by Bianchi[3] to characterize the 
behavior of DCF in order to account both real channel conditions 
and capture effects, especially in a high interference radio 
environment. 

Keywords-throughput; IEEE802.il; MAC; DCF; capture 

I. Introduction 

The use of IEEE 802.11 wireless local area networks 
(WLANs) has been spreading quickly. One of the channel 
access mechanisms in IEEE 802.1 1 is Distributed Coordination 
Function (DCF). DCF is based on Carrier Sense Multiple 
Access with Collision Avoidance(CSMA/CA) algorithm. In 
this mechanism, a station waits for a quiet period in wireless 
media, and then begins to transmit data while detecting 
collisions. The time lapse between successive carrier senses, 
when channel is occupied, is given by a back-off counter which 
has an initial random value within a predetermined range. The 
standard also defines an optional access method, PCF, which is 
for time bounded traffic. Since the DCF mechanism has been 



widely adopted in wireless networks, we focus our analysis 
only on this mechanism. 

Many research efforts have been done to study the IEEE 
802. 1 1 DCF performance, by both of analysis and simulation. 
Most of them assume the ideal channel condition, which means 
that the packet corruptions are only due to collisions[3-5]. A 
detailed analysis for the multi-access behavior in the 802.11 
DCF is presented in [6], where the packet sending probability 
p, which depends on different contention window size, is 
computed by approximating the 802.11 DCF under saturated 
traffic as a p-persistent CSMA protocol. This approximation is 
very useful for the analysis of the DCF and several papers 
[7-10] have adopted this approach and analyzed saturated DCF 
performance. Bianchi [3] suggests a Markov model to represent 
the exponential backoff process, and Wu et al. [8] use the same 
model and take the packet retry limit into account. 

In [11] , based on the IEEE 802.11 DCF, a novel scheme 
named DCF is proposed to improve the performance of 
Wireless Local Area Network (WLAN) in fading channel. 
Impact of bursty error rates on the performance of wireless 
local area network is studied in [12]. The throughput and delay 
were analyzed in ideal and error-prone channels. 

In [13], the authors proposed a fast collision resolution 
(FCR) algorithm. In this algorithm, when a station detects a 
busy period, it exponentially increases its contention window 
and generates a new backoff counter. In case that a station 
detects a number of consecutive idle slots, it exponentially 
reduces the backoff counter. In [14], the authors proposed a 
new backoff algorithm to measure the saturation throughput 
under several conditions and several set of parameters which 
are adjusted dynamically according to the network conditions. 

In [15], the authors presented an extension of Bianchi' s 
model to a non saturated environment. They modified the 
multi-dimensional Markovian state transition model by 
including state, characterizing the system when there are no 
packets to be transmitted in the buffer of a station. These states 
are called post backoff states and denote a kind of virtual 
backoff counter initiated prior to packet arrival. In [16], the 
authors propose a new scheme for IEEE 802.11 DCF to 
alleviate the low performance of the high date rate stations for 



32 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



asynchronous networks. They introduce an adaptive 
mechanism to adjust the packet size according to the data rate, 
in which the stations occupy the channel equal amount of time. 

In real radio environment, the signal power at access point 
from a given station will be subjected to deterministic path 
loss, shadowing and fast multi-path fading. Due to this, when 
more than one station simultaneously transmit to access point, 
the channel is successfully captured by a station whose signal 
power level is stronger than the other stations and thus 
increases the actual throughput. This phenomenon is called 
capture effect. In [17], the authors presented a Markov model 
to analyze the throughput of IEEE802.il considering 
transmission errors and capture effects over Rayleigh fading 
channels in saturated network conditions. Their model is very 
accurate when the contention level of a network is high. In 
[18], we have presented a novel scheme for DCF under non 
saturated traffic condition. In [19], we extend [3] and presented 
the non saturation throughput analysis for heterogeneous 
traffic. Hadzi-velkov and Spasenovski [20] have investigated 
the impact of capture effect on IEEE 802.11 basic service set 
under the influence of Rayleigh fading and near/far effect. 

Liaw et al. [22] introduced an idle state, not present in 
Bianchi's model [3], accounting for the case in which the 
station buffer is empty after a successful completion of a packet 
transmission. The probability that there is at least a packet in 
the buffer after a successful transmission is assumed to be 
constant and independent of the access delay of the transmitted 
packet. In [23], we presented the performance study for 
multihop network in string topology. 

In this paper, we present an analytical model to study the 
saturation behavior of the IEEE802.il DCF considering 
erroneous channel and capture effects. We differentiate channel 
induced errors from packet collision in order to optimize the 
performance of CSMA/CA under the saturated network 
condition. 

The rest of the paper is organized as follows: Section II 
describes our model for basic access mechanism under 
saturated traffic condition. Performance of the proposal scheme 
is analyzed in section III. Finally, section VI concludes this 
paper. 

II. Performance Analysis For 802. 1 1 DCF In 
Saturated Trafic Condition 

In this section, we present a discrete time bi-dimensional 
Markov model for evaluating the saturation throughput of the 
DCF under non ideal channel conditions considering capture 
effects. Saturated traffic condition means that all users always 
have a packet available for transmission. Throughput under 
saturated traffic situation is the upper limit of the throughput 
achieved by the system, and it represents the maximum load 
the system can carry in the stable condition. 

Let process s(t) be the stochastic process representing the 
backoff stage of a given station at the given time t. A second 
process b(t) is defined, representing backoff time counter of the 
station. Backoff time counter is decremented at the start of 
every idle backoff slot. The backoff counter is an integer value 
uniformly chosen from [0,Wi-l] where Wi denotes the 



(IJCSIS) Inter nationalJ our nal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 
contention window at the i th backoff stage. The backoff stage 
T is incremented by one for each failed transmission attempt 
up to the maximum value m, while the contention window is 
doubled for each backoff stage up to the maximum value 
W =2 m W • 

vv max ** vv min- 



Letting W^ = W , we can summarize the W as, 



W ; 



\2 l W 09 0<i<m\ 
[2 m W , i>m 



(1) 



The main aim in this section is to modify the MAC protocol 
in order to enhance the performance of MAC protocol in the 
event of channel induced errors. The assumption that all frame 
losses are due to collisions between WLAN devices is 
generally not true in a noisy wireless environment. However, 
unsuccessful reception of the data frame can also be caused by 
channel noise or other interference. 

In case of unsuccessful transmission the basic BEB 
mechanism will double the contention window size by 
considering channel errors as a packet collision. This process 
will unnecessarily increase the backoff overhead and intern 
increases channel idle slots. In order to alleviate this problem 
we propose a new mechanism that takes advantage of a new 
capability to differentiate the losses, and thereby sharpen the 
accuracy of the contention resolution process. When the frame 
is corrupted by the channel induced noise, we maintain the 
same contention size instead of doubling it. 

A. Loss differentiation method for basic access mechanism 

Basic access mechanism is the default access method in 
DCF and employs a two-way handshaking procedure. The loss 
differentiation for basic access is not straightforward because it 
provides only ACK feedback from receivers. The loss 
differentiation method for WLAN has been proposed in [24]. 
The following describes a loss differentiation method for basic 
access which requires minimum modifications to the legacy 
standard to provide additional feedback. The data frame can be 
functionally partitioned into two parts: header and body. The 
MAC header contains information such as frame type, source 
address and destination address, and comes before the MAC 
body, which contains the data payload. 

In a WLAN with multiple stations sharing a common 
channel, a collision occurs when two or more stations starts 
transmission in the same time slot, which will likely corrupt the 
whole frame (header plus body) at the receiver end. On the 
other hand, a frame transmission that is not affected by 
collision with transmission from another WLAN station may 
still be corrupted by noise and interference. However, under the 
condition that the signal-to-noise-plus-interference ratio 
(SINR) is reasonable to maintain a connection between the 
sending and receiving stations, the receiver is likely able to 
acquire the whole data frame and decode it, as the physical 
header is transmitted at the base data rate for robustness (e.g., 
in 802.1 lb, the 192 -bit physical header is always transmitted at 
1Mbps). In this case, the noise or interference may result in a 
few bit-errors that cause a Frame Check Sequence (FCS) error 
in the decoded data frame, which is then discarded by the 
receiver station. As the MAC header (18-30 bytes) is 
typically much shorter than the MAC body (e.g., a typical 



33 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) Inter nationalJ our nal of Computer Science and Information Security, 



Internet Protocol datagram is several hundreds to a couple of 
thousands bytes long), when FCS fails, it is much more likely 
caused by bit errors in the body than the header. 

If the MAC header is correctly received but the body is 
corrupted, the receiver can observe the MAC header content to 
learn the identity of the sender and to verify that it is the 
intended receiver. To verify the correctness of the MAC 
header, a short Header Error Check (HEC) field can be added 
at the end of the header as shown in Fig. 1 in order to provide 
error checking over the header, while the FCS at the end of the 
frame provides error checking over the entire MAC frame. 
Note that the use of HEC in the header is not a new concept as 
it has been adopted in many other communication systems, 
such as asynchronous transfer mode and Bluetooth, all of 
which includes a 1-byte HEC or header check sequence (HCS) 
field in their header. With the HEC, when a data frame is 
received and FCS fails, the HEC can be verified to see if the 
header is free of error, and if so, proper feedback can be 
returned to the sender identified by the MAC header. 

As discussed above, FCS failure but correct HEC in a frame 
reception is a good indication that the frame has been corrupted 
by transmission errors rather than a collision. Because in the 
basic access mechanism, only ACK frames are available to 
provide positive feedback, a new control frame NAK needs to 
be introduced to inform the sender that the data frame 
transmission has failed and the failure is due to transmission 
errors; i.e., the data frame has suffered a transmission loss. On 
the other hand, if the sender receives neither a NAK nor an 
ACK after sending a data frame, it is a good indication that the 
frame transmission has suffered a collision loss. The NAK 
frame can be implemented with exactly the same structure as 
the ACK frame except for a one-bit difference in the frame 
type field in the header, and is sent at the same data rate as an 
ACK frame. The transmission of a NAK does not consume 
more bandwidth or collide with other frames because it is 
transmitted SIFS after the data frame transmission and occupies 
the time that would have been used by the transmission of an 
ACK. 

The HEC field is a necessary modification to the standard 
because without it, when the FCS fails, the receiver would not 
be able to determine if the header is in error and would not be 
able to trust the sender address in the header for returning the 
NAK. The HEC field (which can be 1 or 2 bytes) costs an extra 
overhead. But it can be calculated that the overhead due to the 
extra field to the total transmission time is much less than 1%. 
Therefore the overhead is negligible. Comparing the two loss 
differentiation methods, RTS/CTS access is useful when the 
data frame size or the number of stations is very large or there 
are hidden terminals. However, as it consumes extra time for 
RTS/CTS exchange, RTS/CTS access is less efficient than 
basic access in other cases. 



Frame control/ Duration/ Ad dresses 


HEC 


Frame Body 


FCS 



Vol. 7, No. 1, 2010 
The loss differentiation method for RTS/CTS access does 

not involve any modification to the standard. The loss 

differentiation method for basic access, however, needs two 

minor modifications to the current standard: the HEC field and 

the NAK frame, which are both easy to implement. 

B. Analitical modeling of new backoff algorithm 

Based on the above consideration, let us discuss the 
Markovian model shown in Fig.2, assuming saturation network 
conditions. We assume that each station has m+1 stages of 
backoff process. The value of the backoff counter is uniformly 
chosen in the range (0,Wi-l), where Wi= 2'Wmin and depend on 
the station's backoff stage /. A station in (i,0) state will transit 
into (i+l,k) state in the event of collision without capture 
effect. On the other hand, the model transits from (i,0) to (0,k) 
state if frame is successfully captured. From state (i,0), the 
station re-enters the same backoff stage (i,k) in case of 
unsuccessful transmission due to transmission errors. 

The main approximation in our model is that, at each 
transmission attempt, each packet collides with constant and 
independent probability P co i regardless of previously suffered 
attempts and transmission errors occur with probability P e due 
to the erroneous channel. We also assume that the channel is 
captured by a station with the probability P cap in the event of 
collision. Based on the above assumptions we can derive the 
transition probabilities: 

P{i,k/i,k + \} = \, ke[0,Wi-2], ie[0,m] 

P{0,k/i,0} = ((l-P col )(l-P e ) + P col P cap )/W Qt ke[0,W t -ll ie[0,m] 

P{i,k/i,o} = (i-P col )Pjw , ke[o,w t -i\, fe[o,m] 
P{i,k/i - 1,0} = (l - P cap )P col /w it k e [o,w t - l], ie [i,m] 

P{m, k/m, 0} = ((1 - P cap )P col + (1 - P col )P e )/W m> k 6 [0, W m - 1] (2) 

The first equation represents that, at the beginning 
of each time slot, the backoff time is decremented. The second 
equation states that, the initialization of backoff window after 
successful transmission for a new packet. The third equation 
accounts that, the maintenance of backoff window in the same 
stage, if channel error is detected. The fourth and fifth 
equations represent that, the rescheduling of backoff stage after 
unsuccessful transmission. 

Let the stationary distribution of the chain be 
b i j=ftn t _^P{s(t) = iMt) = k}ae(0,m),ke(0,W i _ 1 ) • To obtain 
the closed form solution we first consider the following 
relations: 

ho = h-lA P col(X ~ Pcap)} + hfiiPed ~ Pool)} 



"coM "cap) 



Figure 1 . Frame format for Basic access mechanism 



*coA*~ "cap) 



i-a-^,)**. 



1 b Q 0<i <m 



(3) 



and, 



34 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



a-^rftf - - - 



((i-^xi-p e )+^^+(i-^)p fi )/r 




(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 

m Wi-1 




: 



\1 ^eejtJ^eel ' vv m 

Figure 2. Markov chain model for the backoff procedure of a station 

Ko =K-^-Pca p )P C ot +b m , W-Pca P ) +b mfi 0-P„ l )P t ( 4 ) 



Ui- 






l-(l-^/K 



«/ 






(5) 



from which we obtain the following relation, 



K n -- 



P (1-P ) 

1 coA X l cap) 

i-a-p„ / )p. 






A closed- form solution to the Markov chain owing to the chain 
regularities, for each k e (l,Wi-l), shown as: 

m 

io.-p a ,,yx-p.)+p a „p c JZ b w +h^-Poi)P> t=o 



b s „ = 



W, 



Po,(\ - PJb,_ l0 + (1 - P ml )Pjb lfi , l<i<m 
iPotO- - ^))(Vi.o +b m ,o) + (1 " P„,)Pb m ,o, i = m 



(7) 



By means of relations (3), (5) and remembering 



I*« 



1- 



"coV^- "cap) 



= K 



i-a- p c jp e/ 

we rewrite the relation (7) as: 

b, k = ^-^- b, i e (0, m), k e (0, Wj - 1) 

VV; 



(8) 



Thus, by relations (3), (5) and (8), all the values of bik are 
expressed as a function of b ,o- Considering normalization 
conditions, and making use of the above equations we 
obtain the following relation: 



^0,0 

2 



fm-l 



w n 



Ew 



V*=o 






i 



i-^ 



from which, we obtain, 



6 o,o=- 



2(l-P t )(l-2P t ) 



W (l-P t )(l-(2PX)+W (2P t ) m (l-2P t ) + (l-2P t ) 
where we assume, 

*coA^~ *cap) 



(9) 



P, =■ 



Assuming error free channel with no capture effects, i.e., P e 
P cap = 0, then (9 ) can be rewritten as, 

u _ 2(l-2P col )(l-P col ) 



(W + W-2P col ) + Wa-(2P col ) m )P col 



(10) 



which is similar to b ,o found in Bianchi's model [3] under 
saturated load conditions. 

Now we can express the probability T that a station transmits 
in a randomly chosen slot time when the backoff time is zero 
as, 



m ft 

El ^0,0 



(11) 



^ ) By substituting (9) in (1 1), we obtain the following relation. 



2(1 -2P t ) 



w (i-^)(i-(2^r)+w (2^ni-2^)+(i-2^) 



(12) 



Note that, when m=0, that is no exponential backoff is 
considered, and assuming P ca p = P e = 0, the probability T results 
to be independent of collision probability under saturated 
traffic condition 



(13) 



W +l 



which is the result found in [3] for constant backoff window. 

However, in general, the probability x depends on the 
conditional collision probability P co i, capture probability P cap 
and probability of packet loss P e . In our model we assume 
basic access method to compute the conditional collision 
probability P co i. To determine the value P co i it is sufficient to 
note that the probability that a transmitted packet encounters a 
collision if in a given time slot, at least one of the remaining 
(n-1) stations transmits another packet simultaneously. The 
conditional collision probability also depends on the capture 
probability because capture effect is the sub event of collision, 
i.e. without collision there is no capture effect. Therefore the 
probability P co i can be expressed as, 



Pcol=l-(l-^) "^ 



(14) 



35 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) Inter nationalJ our nal of Computer Science and Information Security, 



Our proposed model considers deterministic power loss and 
multipath fast fading of transmitted signals into account. We 
also assume that there is no direct path between transmitter and 
receiver within Basic Service Set(BSS), which means the 
envelop of received signal is Rayleigh faded. To compute the 
capture probability, we use the model proposed by Hadzi- 
velkov and Spasenovski[10]. In Rayleigh fading channel, the 
transmitted instantaneous power is exponentially distributed 
according to 



/(/?) = — exp(-^), 

Po Po 



p>0 



(15) 



where p represent the local mean power of the transmitted 
frame at the receiver and is determined by 



Po 



-An 



•Pt 



where riis the mutual distance from transmitter to receiver, x is 
the path loss exponent, A.ri" x is the deterministic path loss and 
p t is the transmitted signal power. The path loss exponent for 
indoor channels in picocells is typically taken as 4. During 
simultaneous transmission of multiple stations, a receiver 
captures a frame if the power of detected frames sufficiently 
exceeds the joint power of 'n' interfering contenders 



11 



k=\ 



by a certain threshold factor for the duration of a certain time 
period. Thus capture probability is the probability of signal to 
interference ratio 



7 



Pint 



(16) 



exceeding the product z g(Sf) where z is known as the capture 
ratio and g(Sf) is processing gain of the correlation receiver. 
The processing gain introduces a reduction of interference 
power by a factor g(Sj), which is inversely proportional to 
spreading factor Sf. The conditional capture probability P cap can 
be expressed over i interfering frames as, 



Pca P ^ g(S f )\ i) =prob( r > z g(S f )/i) 

=n+z g(s f )r i 



(17) 



For Direct Sequence Spread Spectrum(DSSS) using 11 chip 
spreading factor (Sf =1 1), 



g(S f ) = 



3S f 



Now the frame capture probability can be expressed as, 

Pca P {z Q ,n) = Y j R i P cap {z g(S f )\i) 



(18) 



where Ri is the probability of 'i' interfering frames being 
generated in the generic time slot, according to 



R ; 



v 



n 
i + l 



Vol. 7, No. 1, 2010 



.1+1/1 _\n-i-i (\9) 



'(!-*■)" 



Next step is the computation of the saturation system 
throughput, defined as the fraction of time the channel is used 
successfully to transmit the bits. Let P tr be the probability that 
there is at least one transmission in the considered time slot, 
with n stations contending for the channel, and each transmits 
with probability x, 



P tr =l-(l-r)" 



(20) 



The probability P s that a transmission on the channel is 
successful is given by the probability that exactly one station 
transmit on the channel or probability that two or more stations 
transmit simultaneously where one station captures the channel 
due to capture effects, 



p = nT(l-T)"-' + P cap 
l-(l-r)" 
Now we can express throughput as, 

E[ pay load transmitted in a time slot] 



(21) 



S=- 



E[lengthof a time slot] 



P tr P s (l-P e )E[PL] 



(\-P lr )a + P tr (\-P s )T c +P lr P s P e T e +P tr P s {\-P e )T s 



(22) 



where, T c is the average time that the channel sensed busy due 
to collision, T s is the average time that the channel sensed busy 
due to successful transmission, T e is the average time that the 
channel is occupied with error affected data frame and o is the 
empty time slot. For the basic access method we can express 
the above terms as, 

T c =H + E[PL] + ACK tuneout 

T s = H + E[PL] + SIFS + ACK + DIFS + 2x d 

T e =H + E[PL] + NAK 

Here, H - Physical header + MAC header 

E[PL] - Average payload length and 

id - propagation delay 

III. Performance Evaluations 

In what follows, we shall present the results for the data rate 
of 11Mbps. In the results presented below we assume the 
following values for the contention window: W m j n =32, m=5 
and W max =1024. The network parameters of 802.1 lb are given 
in Table I. We have also examined 802.1 lb with other possible 
parameter values. We use the method given in the IEEE 
standards [2] to calculate the bit error rates (BERs) and frame 
error rates (FERs) in a WLAN. This method has also been used 
in [25]- [27] to study WLAN performance. It is briefly 
described as follows. 



36 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



TABLE I. 



NETWORK PARAMETERS 



MAC header 


24 bytes 


PHY header 


16 bytes 


Payload size 


1024 bytes 


ACK 


14 bytes 


NAK 


14 bytes 


Basic rate 


1Mbps 


Data rate 


11Mbps 


Td 


1 JUS 


Slot time 


20 jus 


SIFS 


10 jus 


DIFS 


50 jus 


ACK timeout 


300 jus 



(IJCSIS) Inter nationalJ our nal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 

SER 2Mbps = Q^S.SxSINR) (29) 

When the BERs have been determined, the FERs of the data 
and control frames are derived from the BERs and the frame 
lengths. 

Now we can drive the frame error rate, combining BER values 
for both header and payload as: 



A. Bit error rate (BER) model for 802.11b 

First, the symbol error rate (SER) is calculated based on the 
signal-to-noise-plus-interference ratio (SINR) at the receiver. It 
is assumed that the interference and noise affect the desired 
signal in a manner equivalent to additive white Gaussian noise 
(AWGN). Given the number of bits per symbol, the SER is 
then converted into an effective BER. IEEE 802.11b uses 
DBPSK modulation for basic data rate at 1Mbps and 
complementary code keying (CCK) modulation to achieve its 
higher data rates (5.5 Mbps and 11 Mbps). The SER in CCK 
[2] has been determined as: 



SER = ^G(72xSCVRx/? c xD c ) 



(23) 



where, R c is the code rate, D c is the codeword distance, and, 2 
is over all codewords. For 1 1 Mbps data rate, the SER is given 
by 

SER 11Mbps = 24 X Q(^4XSINR) + 16 X Q(y/6xSINR) 



+ 174(V8 X SINR) +16 xQ(y/l0xSINR) 
+ 24 X Q(Vl2 XSINR) + Q(^4XSINR) 

As each symbol encodes 8 bits in 1 1 Mbps, the BER is 



BER 



11Mbps ^8 i 



x SER X lMbps = — — x SER X lMbps 



The SER for 5.5 Mbps is calculated as, 

SER 5SMbps = 14 X Q(^SxSINR) + Q(^16xSINR) 



And 
BER 



5.5Mbps 



2 4 -l 



xSER, 



1 \Mbps 



= — xSER, 

15 



1 \Mbps 



(24) 



(25) 



(26) 



(27) 



The SER in DBPSK modulation scheme has been determined 
as: 



SER 



1Mbps 



: q(vh XSINR) 



(28) 



For 1Mbps mode, because each symbol encodes a single bit, 
the BER is the same as SER. In case of 2Mbps, the BER is 
calculated as, 



P=\-{\-BER lMb J PHY + {\-BER 



\Mbps' 



M 1Mbps 



(MAC+DATA) 



(30) 



where, PHY is the length of the physical header, MAC is the 
length of the MAC header and DATA is the length of the 
packet payload. 

B. Numerical results and discussions 

The behavior of the transmission probability 'x' is depicted 
in Fig. 3 for basic access method as a function of SINR. The 
curves have been drawn for the capture threshold 6dB, number 
of contenting stations 10 and payload size 1024 bytes. The 
results shows that for increasing the channel quality, the 
transmission probability 'x' increases and reaches the steady 
state, above which the channel is assumed as ideal. The 
transmission probabilities of our proposed model and model 
[17] are clearly highlighted in the above results. The Bianchi's 
transmission probability is depicted as horizontal lines due to 
independence of the Bianchi's model on both capture effects 
and channel errors. 

Fig.4 shows the behavior of the saturation throughput as a 
function of the number of the contending stations for basic 
access mechanism. The curves have been drawn for the capture 
threshold 6dB, SINR 7dB and payload size 1024 bytes. The 
curves clearly show that, when the network load is moderate 
our proposed algorithm performs well. On the other hand, 
throughput can be higher than the model[17] for a low number 
of contending stations in the considered scenario. When the 
number of contending stations increases then the achievable 
throughput will approach the saturation throughput obtained in 
model[17]. 



0.045 
0.04 



^ 0.035 

I 003 

£ 0.025 

s 

I 0.02 

| 0.015 

0.01 

0.005 





- 














/ j* - " ~ 


. . & - - 


- - -& - - - 


— a — 


— 4 


Bianchi[3] 




/ Jf* 








~ 


Proposed ^ 
model / 




Model[17] 








- 


■ J- 


J 










; 


^--TV--^' 




i i i 











5.5 



6.5 



7 7.5 3 
SINR[dB] 



8.5 



9.5 



10 



Figure 3. Transmission probability as a function of SINR for basic access 
mechanism. Curves have been obtained for the capture threshold 6dB, payload 
size 1024 bytes and number of contending stations 10. 



37 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) Inter nationalJ our nal of Computer Science and Information Security, 



0.5 



0.4 



g 0.3 



-3 0.2 



0.1 



Ql 



1 ¥.' 

1 


;*-""' 


1 








* * ^ i 


_. .- - «- - " 

1 


_--:*--- 


1 




- 


T-l ,,, , , 




)( Fi op os eel Model 
---*--- Model[ 17] 


i i i 


" 



4 5 6 

Contending Stations 



10 



Figure 4. Saturation throughput for basic access mechanism as a function of 
the number of contending stations for capture threshold 6dB and SINR 7dB 
while pay load size 1024 Bytes. 

In order to assess throughput performance as a function 
capture threshold, Fig.5 shows throughput performance as a 
function SINR, for three different values of the capture 
threshold. Depending upon the channel quality as exemplified 
by the SINR on the abscissa in the figure, it could be preferable 
to operate at low capture threshold in order to gain higher 
throughput performance. The throughput predicted by Bianchi 
assuming SINR = go and capture threshold = go, is depicted as a 
horizontal line along with the proposed model for comparison 
purpose. For decreasing capture threshold, the system 
throughput increases above the Bianchi' s maximum achievable 
throughput performance. This is essentially due to the fact that, 
the capture effect tends to reduce the collision probability 
experienced by the contending stations which attempt 
simultaneous transmission. 

Fig. 6 shows the behavior of system throughput for basic 
access method as a function SINR, for two different payload 
sizes and for 5 transmitting stations. The upper curves are 
plotted for the payload size of 1 024bytes and the bottom curves 
are plotted for the payload size of 128 bytes. In both curves, 
saturation throughput is depicted for two different values of 
capture threshold. 



0.8 




- X — X 



— * — Capture Threshold = 1 dB 
- -*— ■ Capture Threshold = lOdB 
---■*--- Capture Threshold = 30 dB 
— * — Bianchi[3] 



7 7.5 
SINR[dB] 



8.5 



9.5 



10 



Figure 5. Saturation throughput for basic access mechanism as a function of 
SINR for capture thresholds ldB, lOdB and 30dB, while payload size 1024 
bytes and number of contending stations 5. 



Vnl 7 Nn / 7010 



0.8 

0.7 

^ 0-6 

*f 0.5 

o 

^ 0.4 

o 

| 0.3 

OS 

™ 0.2 

D.1 





Payload Size = 1024Bytes 




Payload Size = 12SBytes 



Capture Threshold = o"dB 
Capture Threshold = 30dB 



5.5 



6.5 



7.5 8 
SINR[dB] 



8.5 



9.5 



10 



Figure 6. Saturation throughput for basic access mechanism as a function of 
SINR for payload sizes 1024 and 128 bytes, number of contending stations 5 
and capture thresholds 6dB and 30dB. 



Upon comparing the curves, it is easily seen that the system 
throughput performance is poor for low values of payload size. 
On the other hand, when the capture threshold is high, collision 
probability increases, that tend to reduce the throughput 
performance. 

Fig. 7 shows the behavior of saturation throughput for basic 
access method as a function SINR for two different capture 
thresholds. It can be easily noticed that, when channel errors 
are more, the achievable throughput is high due to proper 
rescheduling of contention window. On the other hand, for 
increasing capture threshold, throughput tends to reduce, as 
expected, in the presence of capture. 

IV. Conclusion 

In this paper we have proposed a new MAC protocol for 
IEEE802.il Distributed Coordination Function taking into 
account of both erroneous channel and capture effects. This 
avoids unnecessary idle slots by differentiating noise lost 
packets from collision lost packets, increasing throughput 
considerably. It performs as well as IEEE 802.11 in noisy 
environment considering low traffic conditions. Using the 

0.7 



O.o" 
* 0.5 
I 0.4 
■S 0-3 
3 0.2 

0.1 



1 


' 


' 


1 


1 


- 




A/j^ Mod 


3l[17] 


- 


Propt 


used model -~~^J)/ 






- 


L_ 




i i 


1 1 


1 



5.5 



6.5 



7 7.5 
SINR[dB] 



9.5 



10 



Figure 7. Saturation throughput for basic access mechanism as a function of 
SINR for capture thresholds 6dB and 24dB while payload size 1024 Bytes and 
number of contending stations 2. 



38 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 



proposed model we have evaluated the throughput performance 
of IEEE802.il DCF for basic access method. Based on this 
model we derive a novel and generalized expression for the 
station's transmission probability, which is more realistic, such 
as non ideal channel conditions. To the best of our knowledge, 
this paper is the first to show the undesirable behavior of the 
standard backoff procedure when transmission losses occur, to 
develop a practical solution to this problem, and to give a 
theoretical performance analysis under homogeneous link 
conditions. 

References 

[I] IEEE Standard for Wireless LAN Medium Access Control (MAC) and 
Physical Layer (PHY) Specification, P802.ll, 1999. 

[2] IEEE LAN/MAN Standards Committee. Part 15.2: Coexistence of 
wireless personal area networks with other wireless devices operating in 
unlicensed frequency bands, August 2003. 

[3] G.Bianchi, " Performance Analysis of the IEEE 802.11 Distributed 
Coordination Function", IEEE Journal on Selected Area in 

Communications, vol.18, no.3, pp. 535-547, March 2000. 

[4] F. Cali, M. Conti, E. Gregori, IEEE 802.11 protocol: design and 
performance evaluation of an adaptive backoff mechanism, IEEE J. 
Select. Area Commun. 18 (2000) 1774-1786. 

[5] H. Wu, Y. Peng, K. Long, S. Cheng, J. Ma, Performance of reliable 
transport protocol over IEEE 802.11 wireless LAN: analysis and 
enhancement, in: INFOCOM 2002 - 21th Annual Joint Conference of 
the IEEE Computer and Communications Societies, vol. 2, 2002, pp. 
599-607. 

[6] F. Cali, M. Conti, E. Gregori, "Dynamic tuning of the IEEE 802.11 
protocol to achieve a theoretical throughput limit", IEEE/ACM Trans. 
Netw. Vol.8, no.6, pp.785-799, December 2000. 

[7] Y.C. Tay, K.C. Chua, A capacity analysis for the IEEE 802.11 MAC 
protocol, Proceedings of Wireless Networks 7 (2001) 159-171. 

[8] H.-T. Wu, Y.P., K. Long, S.-D. Cheng, J. Ma, Performance of reliable 
transport protocol over IEEE 802.11 wireless LAN: analysis and 
enhancement, in: Proceedings of IEEE INFOCOM 2002, New York, 
NY, June, 2002. 

[9] Y. Xiao, Saturation performance metrics of the IEEE 802.11 MAC, in: 
Proceedings of the IEEE Vehicular Technology Conference (IEEE VTC 
2003 Fall), Orlando, FL, USA, October 6-9, 2003. 

[10] Z.-H. Velkov, B. Spasenovski, Saturation throughput — delay analysis of 
IEEE 802.11 DCF in fading channel, in: Proceedings of the IEEE 
ICC_03, Anchorage, AK, USA, May, 2003 . 

[II] Xiaohui Xu, Xiaokang Lin, "Throughput Enhancement of the IEEE 
802.11 DCF in fading channel", In proc. IEEE international conference 
on wireless and optical communications networks, Augest 2006, 
Bangalore, India. 

[12] J. Yin, X. Wang, and D. P. Agrawal, "Impact of Bursty Error Rates on 
the Performance of Wireless Local Area Network (WLAN)", Ad Hoc 
Networks, vol. 4. no.5, pp. 651-668, 2006. 

[13] Y.Kwon, Y.Fang, and H.Latchman, "Design of MAC protocols with fast 
collision for wireless local area networks," IEEE Trans. Wireless 
commun., vol. 3, no. 3, pp. 793-807, May 2004. 

[14] Hadi Minooei, Hassan Nojumi "Performance evaluation of a new 
backoff method for IEEE 802.11", Elsevier journal on Computer 
Communication, vol. 30, issue. 18 pp. 3698-3704, December 2007. 

[15] David Malone, Ken Duffy, and Doug Leith, "Modeling the 802.11 
Distributed Coordination Function in Nonsaturated Heterogeneous 

Conditions", IEEE/ ACM Trans. Networking, vol.15, no.l, pp 159-172, 
February 2007. 

[16] Ergen and P. Varaiya, Formulation of Distributed Coordination Function 
of IEEE802.il for Asynchronous Networks: Mixed Data Rate and 



Vol. 7, No. 1, 2010 
Packet Size, IEEE Transaction on Vehicular Tech, Vol.57, no.l, pp.436- 

447, January 2008. 
[17] F. Daneshgaran, M. Laddomada, F. Mesiti, and M. Mondin, "Saturation 
Throughput Analysis of IEEE 802.11 in Presence of Ideal Transmission 
Channel and Capture Effects", IEEE Trans. Commun., vol. 56, no. 7, 
pp. 1178-1188, July 2008. 

[18] P.Kumar, A.Krishnan, K.Poongodi and T.D.Senthilkumar, "Performance 
Analysis of the IEEE 802.11 Distributed Coordination Function in High 
Interference Wireless Local Area Networks considering Capture 
Effects", In Proc. of IEEE International Advanced Computing 
Conference(IACC), Vol.1, pp. 234-456, March 2009. 

[19] T.D. Senthilkumar, A. Krishnan, P. Kumar, Nonsaturation Throughput 
Analysis of IEEE 802.11 Distributed Coordination Function. In Proc. 
of IEEE International Conference on Control Communication and 
Automation (Indie on' 08), IIT Kanpur, India, pp. 154-158, 2008. 

[20] Z. Hadzi-Velkov and B. Spasenovski, "Capture effect in IEEE 802.11 
basic service area under influence of Rayleigh fading and near/far 
effect", In Proc. of 13th IEEE International Symposium on Personal, 
Indoor and Mobile Radio Communications, Vol.1, pp.172 - 176, Sept. 
2002. 

[21] J. G. Proakis, Digital Communications, New York: Mc-Graw Hill, 2001 
(4 th ed). 

[22] Y. S. Liaw, A. Dadej, and A. Jayasuriya, "Performance analysis of IEEE 
802.11 DCF under limited load," in Proceedings of Asia-Pacific 
Conference on Communications, pp. 759-763, Perth, Western Australia, 
October 2005. 

[23] T.D. Senthilkumar, A. Krishnan, P. Kumar, New Approach for 
Throughput Analysis of IEEE 802.11 in AdHoc Networks. In Proc. of 
IEEE International Conference on Control Communication and 
Automation (Indicon'08) , IIT Kanpur, India, pp. 148-153, 2008. 

[24] Q. Pang, S.C. Liew, and V. Leung, "Design of an effective loss- 
distinguishable MAC protocol for 802.11 WLAN," IEEE 
Communications letters, vol. 9, no. 9, pp.781-783, Sep. 2005. 

[25] P. Chatzimisios, A.C. Boucouvalas, and V. Vitsas, "Performance 
analysis of IEEE 802.11 DCF in presence of transmission errors," IEEE 
ICC 2004, pp. 3854-3858, Jun. 2004. 

[26] J. Yeo and A. Agrawala, "Packet error model for the IEEE 802.1 1 MAC 
protocol," IEEE PIMRC 2003, pp.1722-1726, Jan. 2003. 

[27] J. Yin, X. Wang, and D.P. Agrawal, "Optimal packet size in error-prone 
channel for IEEE 802.11 distributed coordination function," IEEE 
WCNC2004, pp. 1654-1659, Mar. 2004. 



AUTHORS PROFILE 



Ponnusamy Kumar received his B.E degree from Madras 
University, Chennai, India, in 1998, and the M.Tech degree 
from Sastra University, Tanjavore, India, in 2002. Since 
2002 he is working as a Assistant Professor in 
K.S.Rangasamy College of Technology, Tamilnadu, India. 
He is currently pursuing his Ph.D degree under Anna 
University, Chennai, India. His research is in the general 
area of wireless communication with emphasis on adaptive 
protocols for packet radio networks, and mobile wireless communication 
systems and networks. Mr.P.Kumar is a member in IETE and ISTE. 

j0tt^ A. Krishnan received his Ph.D degree from IIT Kanpur, 

^k Kanpur, India. He is currently a professor with 

K.S.Rangasamy College of Technology, Tiruchengode, 

Tamilnadu, India. He has published over 150 papers in 

national and international journals and conferences. His 

^^A » :^^^ research interests are in the area of communication 

networks, transportation, and hybrid systems. Dr. A. 

Krishnan is a senior member in IEEE, and member in IETE and ISTE. 




39 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. I, 2010 



Performance Evaluation of Unicast and Broadcast 
Mobile Ad-hoc Networks Routing Protocols 

Sumon Kumar Debnath 1 , Foez Ahmed 2 , and Nayeema Islam 3 

x Dept. of Computer Science and Telecommunication Engineering, Noakhali Science and Technology University, Bangladesh 

2 Dept. of Computer Networks Engineering, College of Computer Science, King Khalid University, Kingdom of Saudi Arabia 

department of Information & Communication Engineering, Rajshahi University, Rajshahi- 6205, Bangladesh 



Abstract — Efficient routing mechanism is a challenging issue for 
group oriented computing in Mobile Ad Hoc Networks 
(MANETs). The ability of MANETs to support adequate Quality 
of Service (QoS) for group communication is limited by the 
ability of the underlying ad-hoc routing protocols to provide 
consistent behavior despite the dynamic properties of mobile 
computing devices. In MANET QoS requirements can be 
quantified in terms of Packet Delivery Ratio (PDR), Data 
Latency, Packet Loss Probability, Routing Overhead, Medium 
Access Control (MAC) Overhead and Data Throughput etc. This 
paper presents an in-depth study of one-to-many and many-to- 
many communications in MANETs and provides a comparative 
performance evaluation of unicast and broadcast routing 
protocols. Dynamic Source Routing protocol (DSR) is used as 
unicast protocol and BCAST is used to represent broadcast 
protocol. The performance differentials are analyzed using ns-2 
network simulator varying multicast group size (number of data 
senders and data receivers). Both protocols are simulated with 
identical traffic loads and mobility models. Simulation result 
shows that BCAST performs better than DSR in most cases. 



Keywords-MANETs, DSR, BCAST, 
Random way point model 



Unicast, Broadcast, 



I. 



Introduction 



A mobile ad hoc network is a self-organizing network 
comprising wireless mobile nodes that move around arbitrarily 
and can able to communicate among themselves using 
wireless radios, without the aid of any preexisting 
infrastructure [1]. Each participating mobile node can act as 
sender, receiver and even as a router at the same time and able 
to build, operate and maintain these networks [2]. 

Due to limited radio coverage of these wireless devices 
efficient support of group oriented communication is 
extremely critical in most MANET applications. In MANET 
group communications issues differ from those in wired 
environments for the following reasons: The wireless medium 
has variable and unpredictable characteristics. The signal 
strength and propagation fluctuate with respect to time and 
environment resulting disconnection of the network at any 
time even during the data transmission period [3]. The strength 



of the received signal depends on the power of the transmitted 
signal, the antenna gain at the sender and receiver, the distance 
between two mobile nodes, the obstacles between them, and 
the number of different propagation paths the signals travel 
due to reflection. Further node mobility also creates a 
continuously changing communication topology in which 
existing routing paths break and new ones form dynamically. 
Since MANETs have limited channel bandwidth availability 
and low battery power, their algorithms and protocols must 
conserve both bandwidth and energy [3]. Wireless devices 
usually use computing components such as processors, 
memory, and I/O devices, which have low capacity and 
limited processing power. Thus their communication protocols 
should have lightweight computational and information 
storage capability fulfilling some key features like robustness, 
simplicity and energy conserving. For this reason, several 
prominent unicast, multicast and broadcast protocols deployed 
in static wired networks that can not perform well in ad-hoc 
networks [3, 4]. 

In-group oriented communication system, routing protocols 
can be classified into three main categories [5, 6, 7] based on 
the number of senders and receivers in MANETs. Unicast 
communication is the point-to-point transmission with one 
sender and one receiver. While unicasting is a simple 
mechanism for one-to-one communication, for one-to-many or 
many-to-many dissemination it brings the network to its knees 
due to huge bandwidth demands [8]. This can also introduce 
significant traffic overhead, sender and router processing, 
power consumption, high packet latency and poor throughput 
in the network. To minimize these overhead for one-to-many 
or many-to-many communication multicast and broadcast Ad- 
Hoc routing protocols play an important role. Multicast 
communications are both one-to-many and many-to-many 
traffic pattern [9] i.e. to transmit a single message to a select 
group of recipients where as in broadcast routing 
communications is one-to-all traffic pattern. It is a basic mode 
of operation in wireless medium that provides important 
control and route establishment functionality for a number of 
unicast and multicast protocols. When designing broadcast 



40 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



protocols for ad hoc networks, developers seek to reduce the 
overhead such as collision and retransmission or redundant 
retransmission, while reaching all the network's nodes. In 
practice, the scope of the broadcast is limited to a broadcast 
domain. Broadcasting is largely confined to local area network 
(LAN) technologies, most notably Ethernet and Token Ring, 
where the performance impact of broadcasting is not as large 
as it would be in a wide area network. Because Broadcasting is 
used to carry huge amount of traffic and requires more 
bandwidth, neither X.25 nor frame relay supply a broadcast 
capability, nor Internet explicitly support broadcasting at the 
global level [10]. 

This paper compares two Ad Hoc routing protocols: 
unicast reactive DSR and BCAST protocol over group 
oriented communication system. Performance comparisons are 
based on Shadowing path loss model and Random way point 
mobility model. The simulation of two routing protocols 
focuses on their differences in their dynamic behaviors that 
can lead to performance differences. 



II. UNICAST AND BROADCAST ROUTING 
PROTOCOLS 

A. Dynamic Source Routing(DSR) Protocol 

DSR [11, 12] is an on-demand unicast reactive source- 
routed routing protocol. This means the source node determines 
the complete sequence of route information between source 
and destination and explicitly lists each hop of the path in the 
packet's header. Route is determined dynamically without any 
prior configuration necessary. The intermediate nodes do not 
require huge memory resources because they do not need to 
maintain consistent global routing information in order to route 
the packets that they forward. The basic operation of DSR 
contains two phases: route discovery and route maintenance. 
Route Discovery mechanism is used to find a source route to 
destination only when source attempts to send a packet to 
destination and does not already knows a route. To reduce the 
cost of Route Discovery, each node maintains a Route Cache of 
source routes it has learned or overheard. Route Maintenance is 
the mechanism used to detect if the network topology has 
changed such that it can no longer use its route to the 
destination because some of the nodes listed on the route have 
moved out of range of each other. 

B. Broadcast Routing ( BCAST) Protocol 

BCAST is an optimized scalable broadcast routing 
protocol [13]. It keeps track of one-hop and two-hop neighbor 
knowledge information that are exchanged by periodic 
"Hello" messages. Each "Hello" message contains the node's 
IP address and list of known neighbors. When a node receives 
a "Hello" packet from all its neighbors, it has two-hop 
topology information i.e. only packets that would reach 
additional neighbors are re-broadcast. For example if a node, 
B receives a broadcast packet from another node A, it knows 
all neighbors of A. If B has neighbors not covered by A, it 
issues the broadcast packet with a random delay. During this 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 
delay, if B receives another copy of this broadcast from C, it 
can check whether its own broadcast will still reach new 
neighbors. If this is no longer the case, it will drop the packet. 
Otherwise, the process continues until #'s timer goes off and 
B itself rebroadcasts the packet. 



The determination of Random delay time is very critical. 
To solve this problem a dynamic strategy is suggested in 
literature [13]. Each node searches its neighbor table for the 
maximum neighbor degree of any neighbor node, say MAX. If 
its own node degree is N, it calculates the random delay as 
MAX/N. Every node also buffers the most recent X packets. X 
can be any arbitrary integer number. To keep the memory 
requirement at each node low; set X to a small number. This 
mechanism improves the packet delivery ratio in BCAST. 

When a node receives a packet with sequence number N 
from source node A, it checks whether it also received packet 
AM from the same source. If not, it issues a one-hop broadcast 
to the neighbors, asking for retransmission of this packet by 
sending Negative Acknowledgement, NACK(N-1, A) message. 
Each neighbor, upon receiving the NACK packet, will check its 
local buffer and if they have this packet buffered, will schedule 
a retransmission. To reduce collisions, the NACKs and the 
packet retransmissions are jittered randomly by few 
milliseconds. 

III. Simulation model 

This section describes the simulation tools and parameters 
chosen to simulate the routing protocols 

A. Simulation Environment 

Network Simulator NS-2 [14, 15] is chosen to compare 
DSR and BCAST routing protocols. NS-2 is discrete event 
packet-level simulators with CMU's Monarch group's 
mobility extensions. It includes implementations of models of 
signal strength, radio propagation, wireless medium contention, 
capture effect, and node mobility. A simulation model with 
MAC and physical models are used to study the interlayer 
interaction and their performance. An unslotted carrier sense 
multiple access (CSMA) technique with collision avoidance 
(CSMA/CA) is used to transmit the data packets. In this 
experiment, the Distributed Coordination Function (DCF) of 
IEEE 802.11 for wireless LAN is used as MAC layer. The 
simulated radio interface model is the Lucent WaveLAN. 
WaveLAN is modeled as shared-media radio with channel 
capacity of 2Mbits/sec and transmission range of 250m and 
the carrier sensing range is 471.5m. All protocols use an 
interface queue (IFQ) of 50 packets. The IFQ is a FIFO 
priority queue where routing packets gets higher priority than 
data packets. All MAC and Network layer operations of the 
wireless network interfaces are logged in trace files. 



B. Radio Propagation Model 

The shadowing path loss model [15] is used in this 
simulation study. It attempts more realistic situation than free 
space and two-ray path loss models. It takes into account multi- 



41 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



path propagation effects. Both free space and two-ray models 
predict the mean received signal strength as a deterministic 
function of distance and consequently represent 
communication radius as an ideal circle. But in realistic 
environment, when the fading effects are considered it can be 
seen that, the received power at a certain distance is a random 
variable. Hence shadowing model is widely used in real 
environment. 

The available parameters that are used in our simulation code 
are shown in table I. 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 
the mobility behavior of nodes. If the node movement is small 
and the Pause Time is long, the topology of Ad Hoc network 
becomes relatively stable. On the other hand, if the node 
moves fast and the pause time is small; the topology is 
expected to be highly dynamic. 



IV. Performances Metrices 

The performance of DSR and BCAST protocols are compared 
using the following important Quality of Service (QoS) 
metrics 



TABLE I. 



PARAMETERS USED IN SIMULATION 



Parameters 


Value 


Comment 


Transmission Range 


250m 


Fixed (Considered) 


Frequency 


914 x 10 6 Hz 


Fixed (Considered) 


Path Loss Exponent 


2.0 


Fixed (Considered) 


Standard Deviation 


4.0 


Fixed (Considered) 


Reference Distance 


1.0m 


Fixed (Considered) 


CPThreshold 


10.0 Watt 


Fixed (Considered) 


RXThreshold 


6.76252 x 10 " 10 


Calculated 


CSThrehold 


2.88759x10" 


Calculated 
(RXThreshold*0.0427) 


Power (Pt) 


0.28183815 Watt 


Fixed (Considered) 


System Loss 


1.0 


Fixed (Considered) 



C. Traffic and Mobility Model 

In this simulation Continuous bit rate (CBR) traffic sources 
are used. The source-destination pairs are spread randomly 
over the network. Only 512- byte data packets are used. The 
number of source destination pairs and the packet- sending rate 
in each pair is varied to change the offered load in the 
network. 

A mobility model accurately represents the movement of 
mobile nodes in MANET. Random waypoint mobility model 
[16] (RWM) is used in this simulation study. The model 
includes networks with 50 mobile nodes placed on a site with 
dimensions 1500x300 meters. Each packet starts its journey 
from a random location to a random destination with a 
randomly chosen speed (uniformly distributed between 0-20 
m/s). Once the destination is reached, another random 
destination is targeted after a pause time and then repeats the 
process. The pause time, which affects the relative speeds of 
the mobiles, is also varied. Five randomly generated scenarios 
are run for each parameter combination, and each point in the 
graphs is the average results of these five scenarios. Identical 
mobility and traffic scenarios are also used across protocols to 
gather fair results. In RWM model, Pause Time and Max 
Speed of a mobile are the two key parameters that determine 



Packet Delivery Ratio (PDR): The ratio of the number of 
packets received by the CBR sinks at the final destination to 
those generated by the CBR sources. 

Packet Latency: This includes all possible delays caused by 
buffering during route discovery, queuing delay at the 
interface queue, retransmission delays at the MAC, 
propagation and transfer times [17]. The lower the packet 
latency the better the application performance as the average 
end-to-end delay is small. 

Normalized routing Load (NRL): The ratio of the number of 
routing packets sent to the number of data packets received. 
Each hop-wise transmission of these packets is counted as one 
transmission [18]. 

Normalized MAC Load (NML): The number of routing, 
Address Resolution Protocol (ARP), and control packets (e.g., 
RTS, CTS and ACK) transmitted by the MAC layer, including 
IP/MAC headers for each delivered data packet [18]. It 
considers both routing overhead and the MAC control 
overhead. This metric also accounts for transmission at each 
hop. 

Throughput: The ratio of the total data received by the end 
user and the connection time [19]. A higher throughput 
directly impacts the user's perception of the QoS. 

V. Simulation results and discussions 

The results of this simulation study are separately considered 
into two sections. 

• Varying Number of data Senders: 

• Varying Number of data Receivers 

A. Effect of number of senders on QoS metrics 

To investigate the effect of number of senders on the 
performance of DSR and BCAST, the data send rate and 
number of data receivers is kept constant at 2 packets/sec. and 
20 respectively. The numbers of data senders are increased 



42 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



from 1 to 10 and several QoS metrics are measured and plotted 
into logarithmic graphs. For the fairness of protocols 
comparison and network performance, each ad hoc routing 
protocol is run over the same set of scenarios. Table II. Shows 
Simulation parameters for the different Senders Scenarios 



TABLE II. 



Simulation parameters for the different Senders 
Scenarios 



Parameter 


Value 


Number of senders (variable) 


1, 2, 5, 7 and 10 


Number of receivers (keep constant) 


20 


Pause Time 


Om/s 


Max. Speed 


20m/s 


Antenna Range 


250 m 


CBR Rate 


2 packets/sec. 


Simulation Time 


200 s 



The PDR and data packet latency simulation results as a 
function of number of Senders are given in fig 1. 



PDR versus Multicast Group 
Members (Number of Senders) 



-BCAST 



-DSR 



150 n 






uu < 


> ♦ 


♦ — 


♦-= 


-♦ 


50 


^ — _ 











^ ▲ 



2 5 7 

Number of Senders 

(a) 



10 



Latency versus Multicast Group 
Members (Number of Senders) 




(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 
This is due to the fact that with relatively many senders (two 
or more) the traffic sources are more spread throughout the 
MANET and hence the overall performances of on-demand 
unicast DSR deteriorate rapidly. For example in DSR with five 
senders and 30 receivers, each sender needs to generate and 
maintain 30 simultaneous unicast connections to connect 30 
multicast group members. Hence overall 150 unicast 
connections are required. Each data source also requires 
generating 30 data packets. This provides an inefficient use of 
wireless medium and causes congestion in the network. Since 
for many senders the PDR value drops drastically, the 
performance of DSR protocol is not very attractive for many- 
to-many application and it also increases the packet 
transmission cost. BCAST shows higher packet delivery ratio 
for most scenarios. Fig 1(a) also shows that for five senders, 
the PDR of BCAST and DSR are 99.45% and 47.82% 
respectively. This is because that BCAST has less redundancy 
and dynamically selects only a subset of nodes to re-broadcast 
a packet. It keeps 2-hop neighbor topology information and 
each node also buffers most recent few packets. A NACK 
based retransmission scheme of BCAST protocol further 
increase PDR. 

From fig 1(b) it is observed that the average packet latency 
increases with increasing number of senders. BCAST protocol 
performs better than DSR in this case. For example for seven 
senders, the packet latency of BCAST and DSR are 14.43 ms 
and 3.86 (LIS respectively. This is due to the fact that DSR 

maintains unicast connections. As the number of sender's 
increases, it requires to generate more packets in order to 
reach the group members, more routing packets causes delay 
in the interface queue before reaching the intended destination. 
This causes more packet delay with increasing the number of 
senders. Since BCAST maintains broadcast connections and 
keep two hops topology information, the average packet delay 
is significantly lower than DSR. Lower packet latency is the 
desirable property for real-time applications because these 
applications can tolerate loss but very sensitive to delay. 
Hence BCAST is more effective for real time applications. 

The NRL and NML simulation results as a function of number 
of Senders are given in fig 2. 



12 5 7 

Numbr of Senders 

(b) 

Figure 1 . PDR (a) and Latency (b) results as a function of Number of 
Senders. 

From fig 1(a) it is observed that with increasing number of 
senders, the PDR of BCAST protocol is higher and relatively 
consistent as compare to DSR routing protocol. A relatively 
high PDR is a desirable property for the scalability of Ad Hoc 
routing protocols. Unicast DSR shows lower PDR than 
BCAST for one-to-many and many-to-many communication. 



Normalized Routing Load 
versus Number of Senders 




2 5 7 

Number of Senders 

(a) 



10 



43 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 




Normalized MAC Load versus 
Number of Senders 
» BCAST A DSR 



2 5 7 

Number of Senders 

(b) 



10 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 
of overhead is directly related to the throughput growth. From 
fig 3 it can also be observed that with increasing the number of 
senders, the data throughput increases and the throughput of 
DSR does not increases too drastically and provides poor 
performance. For few senders (one to five) DSR throughput 
increases but lower than BCAST protocols. With more senders 
the DSR throughput highly degrades. Since DSR packet drop 
probability increases with increasing the number of senders, it 
affects the DSR throughput. It can be observed that as the 
number of senders increase, the throughput of BCAST 
increases and the maximum throughput are achieved for 
BCAST protocol. For example at five senders, fig 3 mentions 
that the data throughputs of BCAST and DSR protocols are 
17.54 kbps and 3.91 kbps respectively. 



Figure 2. .NRL (a) and NML (b) results as a function of Number of Senders 



From fig 2(a) it can be observed that the NRL of DSR is 
higher than BCAST and hence provides poor performance. 
This is because that with increasing the number of senders 
DSR requires more routing packets to maintain unicast 
connections among group members. For example for five 
senders, NRL of DSR and BCAST are 44.23 packets and 
0.735 packets respectively. From fig 2(b), it is also observed 
that in DSR normalized MAC load is also extremely high than 
BCAST protocols. In this case almost all MAC transmissions 
are unicast, a high fraction these transmitted packets are MAC 
layer control packets (RTS, CTS and ACK). Due to unicast 
nature, as the number of senders increase DSR requires to 
generate more MAC control packets. BCAST gives better 
performance in this case. Since in BCAST all MAC 
transmissions are multicast, it generates only fraction of MAC 
layer control packets than DSR. This effect results lower 
transmission collision and offers high packet delivery 
guarantee. For example for 10 senders the NML of BCAST 
and DSR protocols are 0.41 and 511.03 packets respectively. 

The data throughput results as a function of number of Senders 
are given in fig 3. 



B. Effect of number of receivers on QoS metrics 

In this case, to evaluate the performance of DSR and 
BCAST, the data send rate and number of data senders are kept 
constant at 2packets/sec. and 05 respectively. The numbers of 
multicast receivers are varied and QoS metrics are measured. 

The simulation parameters considered for performance 
evaluation are provided in table III. 



TABLE III. 



Simulation parameters for the different receivers 
Scenarios 



Parameter 


Value 


Number of senders (keep constant) 


05 


Number of receivers (variable) 


10,20, 30, 40 and 50 


Pause Time 


Om/s 


Max. Speed 


20m/s 


Antenna Range 


250 m 


CBR Rate 


2 packets/sec. 


Simulation Time 


200 s 



Data Throughput versus 

Number of Senders 

— ♦— BCAST— A— DSR 




2 5 7 

Number of Senders 



10 



Figure 3. .Data throughput results as a function of Number of Senders. 



The PDR, average packet latency and throughput results as a 
function of number of receives is shown in fig 4. 



PDR versus Number of Receivers 




10 20 30 40 50 

Number of Receivers 

(a) 



This throughput result is consistent with the control overhead 
(NRL and NML) shown in fig 3. This is because that the decay 



44 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



Latency versus Number of 
Receivers 



100000 n 

10000 
^ 1000 

B ioo 

10 

1 



^ 




-DSR 
























^ 








♦ ♦ 


♦ 


» 



10 20 30 40 

Number of Receivers 



50 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 

NML versus Number of 
Receivers 

♦ BCAST-A— DSR 



(b) 



Throughput versus Number of 
Receivers 
'g 1000 -n » BCAST A DSR 

B 100 



10 

1 

0.1 



10 20 30 40 

Number of Receivers 



50 



(C) 

Figure 4. .PDR (a), Latency (b) and Data throughput (c) results as a function 
of Number of Receivers. 

From fig 4(a) and fig 4(b), it is observed that, DSR gives 
similar PDR and Latency results as in the case of number of 
senders. BCAST protocol also shows better performance and 
these performances are independent of the number of multicast 
receivers. From fig 4(c), it is observed that, the throughput 
performance of BCAST is independent of the number of 
multicast receivers. BCAST protocol shows better 
performance in this case. In DSR when the number of 
multicast receivers below 30, DSR throughput performance 
increases with increasing the number of receivers. After this 
the throughput deteriorates again. 

The NRL and NML simulation results as a function of number 
of multicast receivers are given in fig 5. 




10 20 30 40 

Number of Receivers 



50 



(b). 

Figure 5. .NRL (a) and NML (b) results as a function of versus Number of 
Receivers 

From fig 5(a) and fig 5(b), it is observed that the normalized 
routing and MAC load of DSR and BCAST protocols are also 
similar as in the case of number of senders. 



vi. Conclusions 

There are number of alternatives when delivering data 
from one or more senders to a group of receivers such as 
setting up dedicated unicast connections from each sender to 
each receiver, employing a unicast, multicast protocol and 
broadcasting packet to every node. This paper compares the 
performance of BCAST and DSR routing protocols over group 
communication in MANETs. BCAST is an optimized 
neighbor knowledge based broadcast protocol provides robust 
performance with less delay time (keeps two hop neighbor 
information and minimizes network congestion) and less 
traffic overhead (partial source route) in terms of PDR, 
latency, normalized routing and MAC load, packet drop 
probability and data throughput. On-demand unicast DSR 
protocol suffers more from scalability issue as the number of 
data senders and data receivers increase. For scenarios with N 
senders and M receivers, NxM unicast connections have to be 
discovered and maintained by the underlying unicast routing 
algorithm. This introduces substantial overhead and causes 
high network load. Unicast DSR is also more sensitive to data 
send rate. The simulation result shows that the broadcast 
protocol BCAST work very well in most scenarios and are 
more robust even with high traffic environments. 



NRL versus Number of Receivers 




10 20 30 40 50 

Number of Receivers 

(a) 



References 

[1] G. Holland and N. Vaidya, (2002), "Analysis of TCP Performance over 
Mobile Ad Hoc Networks," 5 th Annual Intl. Conf. Mobile Comp. And 
Net, pp. 275-288, 2002 [Academic Publishers. Manufactured in The 
Netherlands]. 

[2] E. Baburaj and V. Vasudevan (2008), "An Intelligent Multicast Ad-hoc 
On demand Distance Vector Protocol for MANETs" Journal of 
Networks, Vol. 3, No. 6, June 2008. 

[3] P. Mohapatra, C. Gui, and J. Li, (2004), "Group communications in 
mobile ad hoc networks", University of California, Davis, Computer 
Volume 37, Issue 2, Pages 52 - 59, Feb 2004 [published by the IEEE 
Computer Society]. 

[4] G. H. Lynn, (2003), "ROMR: Robust Multicast Routing in Mobile Ad- 
hoc Networks", Ph.D. Thesis, University of Pittsburgh, November 25, 
2003. 



45 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



[5] L. M. Feeney, (1999), "A Taxonomy for Routing Protocols in Mobile 
Ad Hoc Networks", Swedish Institute of Computer Science, sept 29, 
1999. 

[6] X. Hong, K. Xu and M. Gerla, (2002), "Scalable Routing Protocols for 
Mobile Ad Hoc Networks", University of California, Los Angeles CA 
90095, IEEE network, 2002. 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 
Assistant Proctor in that university. His research interests 
include Mobile Ad-hoc networks, Vehicular Ad-hoc networks, 
Sensor Network, MIMO, OFDM, and MCCDMA. 



[7] P. Kuosmanen, F. D. Forces and N. Academy, (2002), 
Ad Hoc Routing Protocols", 2002. 



'Classification of 



[8] 
[9] 



[10] 



[11] 



[12] 



[13] 



A. S. Tanenbaum, (2003), "Computer Networks", Fourth Edition, 2003. 

C. E. Perkins, E. Royer, S. R. Das, and M. K. Marina, (2001), 
"Performance Comparison of Two On-demand Routing Protocols for Ad 
hoc Networks", IEEE Personal Communications Magazine special issue 
on Ad hoc Networking, pp. 16-28, February 2001. 

B. A. Forouzan, (2000), "TCP/IP Protocol Suite", Third Edition, Tata 
McGraw-Hill Edition, 2000. 

D. B. Johnson and D. A. Maltz, (2004), "The Dynamic Source Routing 
Protocol for Mobile Ad Hoc Networks (DSR)", IETF MANET working 
group Internet draft, 19 July 2004. 

X. Li, (2006), "Multipath Routing and QoS Provisioning in Mobile Ad 
hoc Networks", Ph.D. Thesis, Department of Electronic Engineering 
Queen Mary, University of London, April, 2006. 

T. Kunz, (2003), "Reliable Multicasting in MANETs", Contract Report, 
Defence R&D Canada - Ottawa Communications Research Centre, July 
2003. 

[14] "The Network Simulator ns-2", http://www.isi.edu/nsnam/ns/. 

[15] K. Fall, and K. Varadhan, (2008), "The ns Manual (formerly ns Notes 
and Documentation)", The VINT Project, UC Berkeley, LBL, USC/ISI, 
and Xerox PARC, September 2008. 

[16] J. Broch, D. A. Maltz, D. B. Johnson, Y-C. Hu and J. Jetcheva, (1998), 
"A Performance Comparison of Multihop Wireless Ad Hoc Network 
Routing Protocols", Proc. IEEE/ACM Mobicom Conference, pp. 85-97, 
zoctober 1998. 

[17] M. K. Marina and S. R. Das, (2003), "Ad hoc on-demand multipath 
distance vector routing", Technical Report, Computer Science 
Department, Stony Brook University, April 2003 

[18] G. Jayakumar and G. Ganapathy, (2007), "Performance Comparison of 
Mobile Ad-hoc Network Routing Protocol", IJCSNS International 
Journal of Computer Science and Network Security, VOL.7 No. 11, 
November 2007. 

[19] L. Wang, L., Y. Shu, M. Dong, L. Zhang and O. Yang, (2001), 
"Adaptive Multipath Source Routing in Ad Hoc Networks", IEEE 
International Conference on Communications, ICC 2001. 




Foez Ahmed joined as a Lecturer 
in the Department of Electronics 
and Communication Engineering, 
Northern University Bangladesh 
(NUB), Dhaka, Bangladesh in 
the year of 2008. At present he is 
on leave from Northern 
University and working as a 
Lecturer with the Dept of 
Networks and Communication 
Engineering, College of 

Computer Science, King Khalid 
University, Kingdom of Saudi Arabia. He did his under 
graduation and post graduation in Information and 
Communication Engineering in 2007 and 2009 respectively 
from Rajshahi University, Bangladesh. He has received 
various Awards and Scholarships for the under graduation and 
post graduation results. His research interests include Mobile 
Ad-hoc Networks and Routings, Cognitive Radio Networks, 
Cooperative Communications, Sensor Networks and Sparse 
Signal Processing in Wireless Communication. 



Nayeema Islam was born 1976 in Bangladesh. She 
received her M.S. in Telecommunication Engineering from 
Asian Institute of Technology, Thailand in 2004. Also she 
received her M.Sc. and B.Sc. in Computer Science and 
Technology from Rajshahi University, Bangladesh in 1998 
and 1997 respectively. She is presently working as Assistant 
Professor in the department of Information and 
Communication Engineering, University of Rajshahi since 
2004. Her fields of interest include Telecommunications, 
computer networking, mobile ad-hoc networks and QoS 
routing. 



Authors Profile 




Sumon Kumar Debnath 

obtained his M.Sc degree in 
Information & Communication 
Engineering from Rajshahi 
University, Bangladesh during 
the year 2009. He has joined 
Noakhali Science & Technology 
University, Bangladesh and 
working as a lecturer in the dept. 
of Computer Science and 
Telecommunication Engineering. 
He is also working as an 



46 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



Deriving Relationship Between Semantic Models 

An Approach for cCSP 

Shamim H. Ripon 1 , Michael Butler 2 

1 Department of Computing Science, University of Glasgow, UK 

2 School of Electronics and Computer Science, University of Southampton, UK 



Abstract — Formal semantics offers a complete and rigorous 
definition of a language. It is important to define different 
semantic models for a language and different models serve 
different purposes. Building equivalence between different se- 
mantic models of a language strengthen its formal foundation. 
This paper shows the derivation of denotational semantics from 
operational semantics of the language cCSP. The aim is to show 
the correspondence between operational and trace semantics. We 
extract traces from operational rules and use induction over 
traces to show the correspondence between the two semantics 
of cCSP. 

Index Terms — Compensating CSP, semantic relationship, trace 
semantics, operational semantics. 

I. Introduction 

A formal semantics offers a complete, and rigorous defi- 
nition of a language. Operational and denotational semantics 
are two well-known methods of assigning meaning to pro- 
gramming languages and both semantics are necessary for a 
complete description of the language. Denotational semantics 
associates an element of a semantic domain to each expression 
in the language and the semantic is compositional. Traces are 
one of the ways to define denotational semantics. A trace 
gives the global picture of the behaviour. The common way 
of defining operational semantics is to provide state transition 
systems for the language, where the transition system models 
the computation steps of expressions in the language and 
allows the formal analysis of the language. 

Compensating CSP (cSCP) [1] is a language defined to 
model long running business transactions within the frame- 
work of Hoare's CSP [2] process algebra. Business trans- 
actions need to deal with faults that can arise at any stage 
of the transactions. Compensation is defined in [3] as an 
action taken to recover from error in business transactions 
or cope with a change of plan. cCSP provides constructs for 
orchestration of compensations to model business transactions. 
With the introduction of the language, both traces Q and 
operational [4] semantics have been defined. Both semantics 
have valuable non-overlapping application and we want to use 
them both. The key question is "How they are related?". 

This paper draws the correspondence of two different se- 
mantic representation of a language which strengthen the 
formal foundation of the language. In particular, the aim is 
to accomplish the unification between operational and deno- 
tational approach of cCSP. The unification is based on the 



approach where we use the transition rules from operational 
semantics to derive the traces and then show that these derived 
traces correspond to the original traces by using induction over 
the derived traces. Completing the derivation means that any 
of the presentations can be accepted as a primary definition of 
the meaning of the language and each of the definitions can 
even safely and consistently be used at different times and for 
different purposes. 

The reset of the paper is organised as follows. A brief 
overview of cCSP along with an example is given in SectionllTl 
The trace and the operational semantics of cCSP are outlined 
in Section Jill We describe the how we define and prove 
a relationship between the semantic models in Section [V] 
We define theorems and supporting lemmas to establish the 
relationship for both standard and compensable processes. We 
outline some lessons from the experiment and then summarise 
some related work in Section [V] and Section Ivj respectively. 
We draw our conclusion in Section IVIII 



II. Compensating CSP 

The introduction of the cCSP language was inspired by 
two ideas: transaction processing features, and process algebra. 
Like standard CSP, processes in cCSP are modelled in terms of 
the atomic events they can engage in. The language provides 
operators that support sequencing, choice, parallel composition 
of processes. In order to support failed transaction, compen- 
sation operators are introduced. The processes are categorised 
into standard, and compensable processes. A standard process 
does not have any compensation, but compensation is part of 
a compensable process that is used to compensate a failed 
transaction. We use notations, such as, P, g, .. to identify 
standard processes, and PP, gg, .. to identify compensable 
processes. A subset of the original cCSP is considered in this 
paper, which includes most of the operators, is summarised in 
Fig.ffl 

The basic unit of the standard processes is an atomic event 
(A). The other operators are the sequential (P ; g), and the 
parallel composition (P || g), the choice operator (P □ g), the 
interrupt handler (P > g), the empty process SKIP, raising an 
interrupt THROW, and yielding to an interrupt YIELD. A pro- 
cess that is ready to terminate is also willing to yield to an in- 
terrupt. In a parallel composition, throwing an interrupt by one 
process synchronises with yielding in another process. Yield 



47 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



Standard Processes: 


Compensable Processes: 


P, Q ::= A (atomic event) 


PP 


QQ 


:= P -i- Q (compensation pair) 


| P ; Q (sequential composition) 




\PP]QQ 


\ P □ Q (choice) 






\PPU QQ 


| P || Q (parallel composition) 






1 PP II QQ 


SKIP (normal termination) 






| SKIPP 


THROW (throw an interrupt) 






THROW W 


YIELD (yield to an interrupt) 






YIELDD 


\ P \> Q (interrupt handler) 








[PP] (transaction block) 








Fig. 1. 


cCSP syntax 



execution, with the courier possibly having been booked and 
possibly some of the items having being packed. In case of 
failure, the semantics of the transaction block will ensure that 
the appropriate compensation activities will be invoked for 
those activities that already did take place. 



points are inserted in a process through YIELD. For example, 
(P ; YIELD ; Q) is willing to yield to an interrupt in between 
the execution of P, and Q. The basic way of constructing a 
compensable process is through a compensation pair (P -=- g), 
which is constructed from two standard processes, where P 
is called the forward behaviour that executes during normal 
execution, and Q is called the associated compensation that 
is designed to compensate the effect of P when needed. The 
sequential composition of compensable processes is defined 
in such a way that the compensations of the completed tasks 
will be accumulated in reverse to the order of their original 
composition, whereas compensations from the compensable 
parallel processes will be placed in parallel. In this paper, 
we define only the asynchronous composition of processes, 
where processes interleave with each other during normal 
execution, and synchronise during termination. By enclosing 
a compensable process PP inside a transaction block [PP], 
we get a complete transaction and the transaction block itself 
is a standard process. Successful completion of PP represents 
successful completion of the block. But, when the forward 
behaviour of PP throws an interrupt, the compensations are 
executed inside the block, and the interrupt is not observable 
from outside of the block. SKIPP, THROWW, and YIELDD 
are the compensable counterpart of the corresponding standard 
processes and they are defined as follows: 

SKIPP = SKIP + SKIP, 

YIELDD = YIELD -=- SKIP 

THROWW = THROW -=- SKIP 

To illustrate the use of cCSP, we present an example of 
a transaction for processing customer orders in a warehouse 
in Fig El The first step in the transaction is a compensation 
pair. The primary action of this pair is to accept the order and 
deduct the order quantity from the inventory database. The 
compensation action simply adds the order quantity back to 
the total in the inventory database. After an order is received 
from a customer, the order is packed for shipment, and a 
courier is booked to deliver the goods to the customer. The 
PackOrder process packs each of the items in the order in 
parallel. Each Packltem activity can be compensated by a 
corresponding Unpackltem. Simultaneously with the packing 
of the order, a credit check is performed on the customer. 
The credit check is performed in parallel because it normally 
succeeds, and in this normal case the company does not wish 
to delay the order unnecessarily. In the case that a credit check 
fails, an interrupt is thrown causing the transaction to stop its 



OrderTransaction 

ProcessOrder 

FulfillOrder 



[ ProcessOrder ] 

(AcceptOrder -i- RestockOrder) ; FulfillOrder 
BookCourier -=- CancelCourier \\ 
PackOrder || 

CreditCheck ; (Ok ; SKIPP 

UNotOk • THROWW) 
PackOrder = \\i E Items • (Packltem(i) + Unpackltem (i)) 

Fig. 2. Warehouse order processing 

III. Semantic Models 

This section briefly outlines the trace and the operational 
semantics of cCSP. 

A. Trace Semantics 

A trace of a process records the history of behaviour up to 
some point. We show the operators on traces which are then 
lifted to operators on set of traces. Traces considered for cCSP 
are non-empty sets. 

The trace of a standard process is of the form s{uj) where 
s e S* (E is alphabet of normal events) and uo e ft 
(Q = {/ ? ! 3 ?}), which means all traces end with any of the 
events in Q, which is called a terminal event. The terminal 
events represent the termination of a process. Successful 
termination is shown by a /. Termination by either throwing 
or yielding an interrupt is shown by ! or ? respectively. In 
sequential composition (p ; q), the concatenated observable 
traces p and q, only when p terminates successfully, (ends 
with /), otherwise the trace is only p. The traces of two 
parallel processes are p(w)\\q(uj') which corresponds to the set 
(p 1 1 1 q), the possible interleaving of traces of both processes 
and followed by cj&u/, the synchronisation of ou and u/. The 
trace semantics of standard processes are shown in Fig. [3j 

Compensable processes are comprised of forward and com- 
pensation behaviour. The traces of compensable processes are 
of pair of traces of the form (s(oj),s'(uj')), where s(ou) is the 
forward behaviour and s'(u') is the compensation behaviour. 
In sequential composition, the forward traces correspond to 
the original forward behaviour and followed by the traces of 
the compensation. Traces of parallel composition are defined 
as the interleaving of forward traced and then follows the 
interleaving of compensation. The traces of a compensation 
pair are the traces of both of the processes of the pair when 
the forward process (P) terminate with a (/), otherwise the 
traces of the pair are the traces of the forward process followed 
by only a (/). The traces of a transaction block are only the 
traces of compensable processes inside the block when the 
process terminates with a (/), otherwise when the forward 
process inside the block terminates with a (!) the traces of 



48 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



Atomic Action: For A G U T(A) = {(A, /)} 
Basic Processes: 



T(SKIP) = {(/)}, T(THROW) 
Sequential Composition: 
p(/) ; q = p.q, and p(uj) ; q = p(w), 
T(P;Q) = {p;q\pePAqeQ} 
Parallel Composition: 



{(!)}, T(YIELD) = {(?>,{/>} 



where u ^ / 



p(a;)||g(a/) = {r(cj&c/) | r G (p ||| g)} where a/ 

r(P||0) = {r \re(p\\q)A P ePAqe Q} ^S? 

Interrupt Handler: 

p(\) \> q = p.q and p(cj) > q = p(cj) where cj 7^ ! 

T(P D> Q) = {p > g | p G P A q G Q} 

Choice: T(PDQ) = T(P) U T(Q) 

Transaction Block: 

[p(\),p'] = p.p' and \p(S),p']=p(S) 

T{[PP]) = {[p,p']\{p,p')ePP} 



! ! ! ? ? / 
! ?/?// 



! ! ! ? ? / 



Fig. 3. Trace semantics of standard processes 



the block are the traces of the forward process followed by 
the traces of the compensation. Fig. [4] outlines the traces of 
compensable processes. 



Basic Processes: 

T(SKIPP) = T(SKIP + SKIP) = {((?>, (/)), ((/>, (/))} 
T(THROWW) = T(THROWW -=- SKIP) = {((?), (/)), ((!), (/))} 
T(YIELDD) = T(YIELD -=- SliT/P) = {((?), (/))} 
Compensation Pair: 

p(S) -T- q = (p{/), q) and p(cj) -=- g = (p{uj), (/)) where u ^ / 
T(P + Q) = {((?>, (/))} U{p-g|p£PA ge g} 
Sequential Composition: 

(p(/),P , );(ffi9 , ) = (w,ff / ;p') 

(p(u),p') ; (g,g') = (p(u),p f ) where a; 7^ / 
T(PP ; 00) = {pp ;qq\ppePPAqqe QQ} 
Parallel Composition: 

(p,p')ll(*, *') = {(r, r') I r G (p||g) Ar'e {p'\W)} 
T(PP\\QQ) = {rr \ rr G (pp\\qq) A pp G PP A qq G 0<2} 
Choice: T(PPUPQ) = T(PP) U T(Q0) 



Fig. 4. Trace semantics of compensable processes 

The following healthiness conditions declare that processes 
consist of some terminating or interrupting behaviour which 
ensures that the traces of processes are non-empty: 

• /?(/) G T(P) or p( ! ) G r(P), for some p 

• (p(S),p') G r(PP) or (p( ! ),/>') G T^PP), for some p,p' 

5. Operational Semantics 

By using labelled transition systems 0, the operational 
semantics specifies the relation between states of a program. 
Two types of transitions are define to present the transition 
relation of process terms: normal and terminal. A normal 
transition is defined by a normal event {a G £) and a terminal 
transition is defined by a terminal event (cj G fi) . 



For a standard process, a normal transition makes the 
transition of a process term from one state to its another state 
(P to P'). The terminal transition, on the other hand terminates 
a standard process to a null process (0): 



P' 







In sequential composition (P ; g), the process g can start 
only when the process P terminates successfully (with /). If 
P terminates with ! or ? the process Q will not start. In par- 
allel composition each process can evolve independently and 
processes synchronise only on terminal events. The transition 
rules for standard processes are outlined in Fig. 



Atomic Action: A — > 


&R7P (A G E) 


Basic Processes: 




SKIP -A 0, THROW - 


-U 0, yplpd — ■> 0, ytpld -A 


Sequential Composition: 


P -% P' 


p^oaqAq' p^o 


^/) 


(P;Q)^ (P> ; Q) 


(P ; Q) -A Q> (P;Q)^0 


Parallel Composition: 




P -% P' Q - 


A Q' ?A0AQ^>0 _ u ^ 


!!!??/ 

! ?/ ?// 


P\\Q-^P'\\Q P\\Q- 




->P||Q' P||g c ^0 w&w ' 


! ! ! ? ? / 


P -A P' 

Choice: 

p □ Q _^> pi 


PUQ -^> Q> 


Interrupt handler: 




P -^>P' 


P — ^ A Q -^ Q' P^>0 


^0 


P > Q -1+P' > Q 


P \> Q ^ Q' P t> Q -^»0 


Transaction Block: 




PP -^ PP' PP 


-Ap PP ^ P A P ^ P' 


[PP] -^ [PP'] [PP] 


A [PP] ^ i 57 



Fig. 5. Operational semantics of standard processes 

For compensable processes, the normal transitions are same 
as standard processes. However, the terminal events terminate 
the forward behaviour of compensable processes, additionally, 
the compensation are stored for future reference. 

PP -1+PP', PP ^P (P is the compensation) 

In sequential composition (PP ; QQ), when PP terminates, 
its compensation (P) is stored and QQ starts to execute. In 
this scenario, we get an auxiliary construct ((gg,P)) where 
the processes have no particular operational relation between 
them. After termination of the process QQ, its compensation 
(2) is accumulated in front of P i.e., (Q ; P). In the parallel 
composition, the main difference with the standard processes 
is that after termination of the forward behaviour the com- 
pensations are accumulated in parallel. The transition rules of 
compensable processes are summarised in Fig. [6J 

A non-terminal event changes the state of the process inside 
the block. Successful completion of the forward process inside 
the block means completion of the whole block, but throwing 
a interrupt by the compensable process inside the block results 
the compensation to run. In compensation pair, after successful 
completion of the forward behaviour the compensation will be 
stored for future use, however, unsuccessful termination, i.e, 
terminates by ! or ? results an empty compensation (Fig. O. 



49 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



Choice: 








PP -±> PP' QQ -^ QQ' 


PP — 


> P QQ^U Q 


PPUQQ -^ PP' PP C 


QQ -% QQ' 


PPnQQ 


-^P PPnQQ^U Q 


Sequential Composition: 






PP -^ PP' 


PP^PAQQ^Q 


PP ^ P 
PP ; QQ^P 


PP ■ QQ^ PP' ■ QQ 


PP ; QQ — 


+ Q;P 


PP ^ PAQQ^ QQ' 
PP; QQ^(QQ',P) 


QQ — 


>QQ' 


QQ^Q 


(QQ,P) — 


>(QQ',P) 


(QQ,P)-^Q;P 


Parallel Composition: 








PP -^ PP' 


QQ — 


>QQ' 


PP -^P AQQ^Q 

PP || QQ ^' P || Q 


PP || QQ -A PP' || QQ 


PP || QQ — 


> PP || QQ' 


Compensation Pair: 








P -^ P' P 


Ao 


P^O 


a , _/ /\ 


P-rQ^P'-rQ P ^ 


Q^Q P 


^gA skip 



Compensable processes have both forward and compensa- 
tion behaviour. A compensable process is defined as a pair 
of traces. Hence, it is required to extract traces from both 
forward and compensation behaviour. The forward behaviour 
of a compensable process PP is defined as follows: 

PP -A R (t ends with u) 

where t is the trace of the forward behaviour. R is the attached 
compensation. The behaviour of compensation is similar to 
standard processes and by reusing that we get the following 
definition: 



pp^o 



3RPP 



R A R 







Fig. 6. Operational semantics of compensable processes 



IV. Relating Semantic Models 

In this section we describe the steps to derive a relationship 
between the two semantic models of cCSP. We follow a 
systematic approach to derive the relationship where traces 
are first extracted from the transition rules and prove that the 
extracted traces correspond to the original trace definition. The 
steps of deriving the semantic relation are shown in Fig. [TJ 



where t' is the trace of the compensation. For a compensable 
process PP, the derived traces DT(PP) is defined as follows: 



0/) eDT(PP) 



pp^o 





























Operational 
Semantics 








Trace 
Semantics 




tra 


ces e 


Ktr 


acted from operational rules 












Derived 
Traces 










I 












induction over traces 

i 


' 










Correspondence between 

derived trace and original trace 

for each term of cCSP 










I 

structural induction over terms 

i 










Establish correspondence 














bel 


tween semantic models 









By using the definition of derived traces and the original 
traces we state the following theorem to define the relationship 
between the semantic models, 

Theorem 1: For any standard process term P, where P^O 

DT(P) = T(P) 

For any compensable process terms PP, where PP ^ and 
does not contain the term (PP, P) , 



DT(PP) 



T(PP) 



Traces are extracted for each term of the language, and its 
correspondence is shown with the corresponding traces in the 
trace semantics. Assume P and Q are standard process terms, 
then for all the operators, we prove that 



t G DT(P Q) 



t£T(P® Q) 



(2) 



For each such operator 0, the proof is performed by induction 
over traces. In the proof we assume that, DT(P) = T(P) and 
DT(Q) = T(Q). 

We follow similar style for compensable processes. Assum- 
ing DT(PP) = T{PP) and DT(QQ) = T(QQ) we show that, 



(f,0 GDT(PP®QQ) 



(t,t')eT(PP®QQ) (3) 



Fig. 7. Steps to derive relationship between semantic models 

The operational semantics leads to lifted transition relations 
labelled by sequences of events. This is defined recursively. 
For a standard process P, 



P^Q 



P^Q 

3P f .p-l+p* A pf 



Q 



The derived traces of a standard process P is defined as DT(P). 
Let t G DT(P), then we get the following definition, 



In the following sections we outline the proof steps showing 
the correspondence in © and ^ for both standard and 
compensable process terms. 

A. Standard Processes 

Sequential Composition: By using Q the relationship be- 
tween the semantic models is derived by showing that, 

t G DT(P ; Q) = teT(P;Q) 

From ([T]) we get the derived traces of the sequential compo- 
sition, 



t G DT(P) 







(1) 



t G DT(P ; Q) 



i?\Q) 



o 



50 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



We also expand the definition of trace semantics as follows: 

te T{P\Q) 

= 3p,q-t=(p;q) A p e T(P) A q e T(Q) 

= 3/7, q • t = (p ; q) Ape DT(P) A q e DT(Q) 

= 3p,q-t=(p\q) A P -A A Q -U 

Finally, from the above definitions of traces, the following 
lemma is formulated for the sequential composition of stan- 
dard processes: 

Lemma 1: 

(P; Q) -^O = 3p : q-t=(p ;q)AP^0AQ^0 

The lemma is proved by applying induction over the trace t, 
where t = (uj) is considered as the base case, and t = (a)t is 
considered as the inductive case. To support the proof of the 
lemma, two equations are derived from the transition rules. 
These derived equations are based on the event by which the 
transition rules are defined: 



Inductive step: t = (a)t 

3P- (P;Q) 



p;e^o 



R A R 



(P;Q) 


-^Ao = P^Ao A g^AO 




(p\Q) 


-^R = 3P'-P^P'AP = 

v pAoa q^r 


= (?' ; G) 


Proof: 
Basic step: 


t=(u>) 




OP; 2)^0 = (P;g)^Ao 

"From transition rules sequential composition" 






= pAo a e^Ao 


(4) 




V PAO A w^/ 


(5) 


From Q 


P^O a e^Ao 




= 


3p,q- p= (/) A q= (uj) A (uj) = 


--(p>q) 


A 


pAoAe^o 




= 


3p,q> (uj) = {p ; q) A p = (/) 




A 


pAo a e^o 




From 







A 



A 



pAo a w^/ 

3p,q-p= (uj) A uj ^ / A (uj) = (p \ q) 

pAo a e^o 

3/?,#. (uj) = (p ;q) A p^ (/) 

pAo a e^o 

Therefore, for t = (uj) from @ and ([5]) 

3p,g- (uj) = (p \q) A p = (/) 

apAo a e^o 

V 3/7,</» = (p;</) A^(/) 

apAo A e^o 

= 3M-(w) = (p;«)APA0Afi-^0 



"From operational rules" 

3P' • P -^ P' A (P' ; Q) -A (6) 

VBg'-PAo A gAe' AQ'AO (7) 
From © 

3P'- P^P' A {P' ; g) ^0 
= "Inductive hypothesis" 

3P' • P ^P' A 3p',q- t=(p' ; tf) 

A P'AOAgAo 

= "Combining existential quantifications" 

= "Using trace rule (a)/ = (a)(p f ; g) = {(a)p f ) ; g" 

3//,#- (a)t=((a)p f ;q) A P^Ko A Q^-+0 
= 3p,q-p= (a)p f A (a)t = (p \ q) 
A PAO A g^O 
From © 

3 g' • P ^A A g ^ g' A Q f -A 

= pAo a e^io 

= 3/?,g • p = (/) A q = (a)t A (a)t = (p ; q) 
A PAO A g^O 
= 3p,q- (a)t =(p\q) A p = (/> 
A PAO A g^O 
Therefore for f = (a)t, from @vfl 

3p,q-p = (a)p' A (a)t = (p ; q) APAO A g^0 
V 3p, q ■ p = (/) A (a)r =(p;^) APAOAfiAo 
= "Combining existential quantifications" 

3/?,</ • (p = (/> V p = (a)p') A (a)/ = (p ; </) 
A PAO A g^O 
= 3/?,0- (<z)f =(/?;^)APA0A6A0 

■ 
This completes the proof of the lemma. We follow the same 
approach to prove other lemmas in the rest of the paper. 
Parallel Composition: The parallel composition of two 
processes is defined to be the interleaving of their observ- 
able events followed by the synchronisation of their terminal 
events. For example, considering asynchronous actions, A \\ B 
can execute A followed by B or B followed by A. For traces 
p and q we write p \ \ \ q to denote the set of interleaving of p 
and q and it follows the following definition: 

ep\\\q = p = () A q=() 
(a)t £ p \\\ q = 3// • p = (a)p f A t E p' \\\ q 
V 3q' • q = (a)q' A t e p \\\ q f 



51 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



By following similar steps as sequential composition, we 
define the following lemma for parallel composition: 
Lemma 2: 



(P II 2)^0 



3/?, q ■ te (p || q) A P -^ A Q -U 



We derive two supporting equation from the transition rules 
of parallel composition: 

P || Q -^ R = P -^P f A R = P' ||e 

V Q^Q' A R = P\\ Q' 

P\\Q^0 = P^0A2^0AwG uj1Luj2 

Proof: The proof of the base case is trivial and omitted 
from the presentation. The inductive case is described here: 



(p\\Q)^o 

3R-(P\\Q)^R A R — 







= "Using the operational rules" 
3P' P-^P 1 A (/>' || Q) -^0 

V 3 Qf ■ Q -^ Qf A (P || g') -A 

= "Inductive hypothesis" 

3P'- P^P' A 3p',0- f € (p' |U) 

A P'AOAO^O 



A 



P -A A g' — 







"Combining existential quantifications" 

{a)p: OAfiAo 



A0 A g^O 



= 3p',q. te(p' \\q) A P 

V 3/7,^- fe(p||tf') A P 
= 3p,q-p = {a)p' A t £ (p' \\ q) 

apAo a e^o 

V 3p,q-q = (a)q' A f G (p || #') 

apAo a g^o 

= "Combining" 

3p,q- (p = (a)p f A t e (p f \\ q) V q = (a)q' 
A te (p || #')) AP^OAg^O 
= "By the definition the interleaving of traces" 

3/7, 4 • (a)t e || q) A P -^ A Q -U 



P. Compensable Processes 

Sequential Composition: For compensable processes PP 
and gg, let (f , *') e DP(PP ; QQ) and according to trace 
derivation rule we get 



(f,0 G DJ(PP ; gg) = 3R(PP ;QQ)^RAR 







The following lemma is stated to define the relationship for 
the lifted forward behaviour of sequential composition of 
compensable processes: 



Lemma 3: 

(PP;QQ)-^R = 3P,Q,p,q-t = (p ; q) 
A PP^P AQQ^Q 

A R = COND(last(p) = /, (Q ; P),P) 
Where, COND(true,el,e2) = el 

COND(false,el,e2) = e2 

COND expression is used to state that when process PP 
terminates successfully (terminate by /), compensation from 
both PP and QQ are accumulated in reverse order, otherwise 
only compensation from PP is stored. The following equations 
are derived from the transition rules to support the proof of 
the above lemma. 



(PP ; QQ) -^ RR 



(PP \QQ)^R 



V 
A 

V 



PP -^ PP' A RR = (PP f ; QQ) 
PP^P A QQ^ QQ' 
R=(QQ',P) 

PP -A P A QQ -^ Q A R = (Q ;P) 

PP^>PAcj^/AP = P 



In the inductive case of the lemma we get the following 
intermediate step involving the auxiliary construct (QQ,P). 



PP.QQ^R 



V 
A 



3RR • PP \QQ ^ RR /\ RR —^ R 
3PP' • PP -^ PP' A PP f \QQ-^R 
3P,QQ' PP^P A QQ^ QQ' 
(QQr',P)-L+R (8) 



To deal with this we need another lemma which will support 
the removal of auxiliary construct in ([8]). This lemma considers 
the situation where the forward behaviour of the first process 
of sequential composition is terminated with / and its com- 
pensation is stored and the second process of the composition 
has started. Here to mention that t in ^ above is a complete 
trace. 

Lemma 4: 

(QQ,P)^R = 3Q- QQ^Q AR=(Q\P) 

The lemma is proved by induction over traces. By using this 
lemma, we prove Lemma [5] by following the similar approach 
of applying induction over traces. 

Parallel Composition:Let (t,t') e DT(PP || QQ) By using 
the trace derivation rule we get, 

(t,t')eDT(PP\\ QQ) = 3R • (PP || QQ) -U R A R -A 

We then define the following lemma to establish the seman- 
tic correspondence for parallel composition of compensable 
processes: 
Lemma 5: 
(PP\\QQ)^R =3P,Q,p,q.t£(p\\q) 

A PP ^P A QQ^P A R = P\\Q 
The lemma is proved by using induction over traces similar 
to other lemmas. 



52 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



Compensation Pair: A compensation pair (P -=- Q) consists 
of two standard processes: a standard process (P) and its 
compensation (Q). The semantics of compensation pair is 
defined in such a way that the behaviour of the compensation 
Q is augmented only with successfully completed forward 
behaviour of P, otherwise, the compensation is empty. For 
a compensation pair, we show that 



{t,t')eDT{P + Q) 



(t,i*)eT(p + Q) 



To prove the semantic correspondence between the semantics 
model, we state the following lemma: 
Lemma 6: 



(P-g)^O = 3/>,tf-(f,O = (p-H2)AP-^0Ae 







The lemma is proved by induction as previous lemmas. To 
support the inductive proof, the following two equations are 
derived from the transitions rules shown earlier, 



(P-rQ)^R 



(P + Q) -^RR 



P^O A R = Q 

P^UO A uj^ S A R = SKIP 

P^P f A RR = P' + Q 



Unlike the lemmas defined earlier for compensable processes, 
Lemma [6] includes the traces of both forward and compensa- 
tion behaviour. The following trace rules for the compensation 
pair are used in the proof of the lamma: 



when/? =p'(/){t,t') 
when/? = p'{(jS) Auj ^ S(t, t') 



(p'(S)+q) = (p,q) 
(p»- 9 ) = (p,(/» 



Transaction Block: Transaction block is a standard process. 
We let t G DT([PP]) and by following the trace derivation rule 
we get 



t£DT([PP}) 



\PP] 







The semantic correspondence is then derived by proving the 
following lemma: 

Lemma 7: 
[PP] -U = 3p,p' • t = [/?,//] APP^O 
The operational semantics provide us the following equations 
to support the proof of the above lemma. 



\PP] 



\PP] 



R 



V 



PP 
PP 
PP 
PP 



PP' A R 
P A P — 
P A P — 
P A P — 



:[PP'] 

P' A R 






P' 



The block operator runs the compensation of a terminating 
forward behaviour and discards the compensation of success- 
fully completed forward behaviour. It removes the traces of 
an yielding forward behaviour. 

We left two operators from the correspondence proof pre- 
sented here. First one is the choice operator (PBQ). The trace 
of choice is the union of their traces and the operational rules 
shows that either process (P or Q) can evolve independently. 
Correspondence proof of this operator is trivial. Another 



operator that was left is interrupt handler (P > Q). It is 
quite similar to standard sequential composition except that 
the flow of control from the first to the second process is 
caused by a throw (!) rather than a / and hence, showing its 
correspondence proof would be repetitive. 

V. Lessons Learned 

We have adopted a systematic approach to show the corre- 
spondence between the two semantic models of cCSP. Traces 
are derived from the operational rules and then applying 
induction over the traces we showed the correspondence. Due 
to the way of defining operational rules the trace derivation 
was done easily. We used labelled transition system to define 
the operational rules. In [6] operational rules are defined for a 
similar language as ours but same symbol is used to define the 
labels of different transition rules. However, we used special 
symbols for different kinds of transitions. Transition between 
states are caused by two kinds of events: normal and terminal 
and we used these events as labels in our transition rules. 
The advantage of this approach of defining labels is that these 
labels are the traces of the transition and we can then derive 
these traces from the transition rules. 

The trace operators play a significant role in defining the 
lemmas as well as in the correspondence proofs. The operators 
are used both at the trace levels and at the process levels. All 
the lemmas defined in this chapter have a common pattern 
applicable to both standard and compensable processes. For 
example, for standard processes P and g, and their traces p 
and q, the lemmas for all the operators are defined as follows: 



(P<g> G) — >0 = 3p,q-t=(p< 
(for parallel operator use t G (p C 



) q) instead of t = (p ® q)) 



Similar definitions are also given for the forward behaviour 
of compensable processes. The use of operators at both trace 
and process levels allow us to apply appropriate rules for 
the operators (rules for terminal and observable events from 
operational and trace semantics). 

The correspondence was proved by using structural induc- 
tion. First, the induction was applied on process terms of 
the language and then on the derived traces. The lower level 
induction which is on traces support the induction on upper 
level which is on process terms 

VI. Related Work 

The semantic correspondence presented here is based on the 
technique of applying structural induction. A similar approach 
is also applied by S. Schneider [7], where an equivalence 
relation was established between the operational and denota- 
tional semantics of timed CSP [8][9]. Operational rules are 
defined for timed CSP and then timed traces and refusals 
are extracted from the transition rules of a program, and it 
is shown that the pertinent information corresponds to the 
semantics obtained from the denotational semantic function. 
By applying structural induction over the terms of timed CSP, 
it was proved that the behaviour of the transition system is 
identical to those provided by the denotational semantics. 



53 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



A similar problem was also investigated in [10], where a 
metric structure was employed to relate the operational and 
denotational models of a given language. In order to relate 
the semantic models it was proved that the two models co- 
incide. The denotational models were extended and structural 
induction was applied over the terms of the language to relate 
the semantic models. 

Other than using induction, Hoare and He [ 1 1 ] presented the 
idea of unifying different programming paradigms and showed 
how to derive operational semantics from its denotational 
presentation of a sequential language. They derive algebraic 
laws from the denotational definition and then derive the 
operational semantics from the algebraic laws. Similar to our 
work, Huibiao et al. Q2] derived denotational semantics from 
operational semantics for a subset of Verilog [13]. However the 
derivation was done in a different way than our method where 
the authors defined transitional condition and phase semantics 
from the operational semantics. The denotational semantics 
are derived from the sequential composition of the phase 
semantics. The authors also derived operational semantics 
from denotational semantics lfT4) . 

Unlike our approach, the unification between the two se- 
mantics was shown in [fT5l by extending the operational 
semantics to incorporate the denotational properties. The 
equivalence was shown for a language having simple models 
without any support for concurrency. Similar problem was also 
investigated in [16] for a simple sequential language, which 
support recursion and synchronisation in the form of inter- 
leaving. The relation between operational and denotational 
semantics is obtained via an intermediate semantics. 

VII. Concluding Remarks 

It is of great importance to have the description of both 
operational and denotational semantics. Having both of the 
semantics we need to establish a relationship between these 
two. Demonstrating the relationship between these two seman- 
tics of the same language ensures the consistency of the whole 
semantic description of the language. 

The main contribution of this paper is to show the corre- 
spondence between the operational semantics and the trace 
semantics of a subset of cCSP language. The correspondence 
is shown by deriving the traces from the operational rules and 
then applying the induction over the derived traces. Two level 
of induction is applied. In one level induction is applied over 
the operational rules and in the next level induction is applied 
over the derived traces. 

The correspondence shown here are completely done by 
hand which is error prone and there are strong possibilities to 
miss some of the important parts during the proof. As part of 
the future work our goal is to use an automated/mechanized 
prover which will help us to use the similar approach that we 
followed here i.e, mathematical induction, and at the same time 
prove the theorems automatically. Among several tools we are 
currently using PVS (Prototype Verification System) [ FTl for 
our purpose. The specification language of PVS is based on 
classical, typed, high order logic and contains the constructs 



intended to ease the natural development of specification. The 
PVS proof checker is interactive and provides powerful basic 
commands and a mechanism for building re-usable strategies 
based on these. 

The parallel operator of cCSP does not support synchroniza- 
tion on normal events. Synchronization of events is significant 
for the development of a language. Currently we are working 
on adding synchronization to cCSP. Adding synchronization 
and then using mechanized theorem prover for showing the 
correspondence will strengthen the formal foundation of the 
language. 

References 

[1] M. Butler, T. Hoare, and C. Ferreira, "A trace semactics for long-running 
transaction," in Proceedings of 25 Years ofCSP, ser. LNCS, A. Abdallah, 
C. Jones, and J. Sanders, Eds., vol. 3525. London: Springer- Verlag, 
2004. 

[2] C. Hoare, Communicating Sequential Process. Prentice Hall, 1985. 

[3] J. Gray and A. Renter, Transaction Processing : Concepts and Tech- 
niques. Morgan Kaufmann Publishers, 1993. 

[4] M. Butler and S. Ripon, "Executable semantics for compensating CSP," 
in WS-FM 2005, ser. LNCS, M. Bravetti, L. Kloul, and G. Zavattaro, 
Eds., vol. 3670. Versailles, France: Springer- Verlag, September 1-3 
2005, pp. 243-256. 

[5] G. D. Plotkin, "A structural approach to operational semantics." Aarhus 
University, Computer Science Department, Tech. Rep. DAIMI FN- 19, 
September 1981. 

[6] R. Bruni, H. Melgratti, and U. Montanari, "Theoretical foundations for 
compensations in flow composition languages," in POPL, 12-14 January 
2005, pp. 209-220. 

[7] S. Schneider, "An operational semantics for timed CSP," Journal of 
Information and computing, vol. 116, no. 2, pp. 193-213, 1995. 

[8] G. M. Reed and A. W. Roscoe, "A timed model for communicating 
sequential processes," Theoretical Computer Science, vol. 58, no. 1-3, 
pp. 249-261, June 1988. 

[9] S. Schneider, J. Davies, D. M. Jackson, G. M. Reed, J. N. Reed, and 
A. W. Roscoe, "Timed CSP: Theory and practice," in REX Workshop, 
ser. LNCS, vol. 600, 1991, pp. 640-675. 
[10] F. van Breugel, "An introduction to metric semantics: operational and 
denotational models for programming and specification languages," 
Theoretical Computer Science, vol. 258, no. 1-2, pp. 1-98, May 2001. 
[11] C. Hoare and H. Jifeng, Unifying Theories of Programming. Prentice 

Hall International Series in Computer Science, 1998. 
[12] H. Zhu, J. P. Bowen, and J. He, "From operational semantics to 
denotational semantics for Verilog," in CHARME 2001, ser. LNCS, 
T. Margaria and T. F. Melham, Eds., vol. 2144, 2001, pp. 449-466. 
[13] M. Gordon, "The semantic challenge of Verilog HDL," in Proceedings 
of the 10th Annual IEEE Symposium on Logic in Computer Science 
(LICS '95: ). IEEE Computer Society, June 1995, pp. 136-145. 
[14] H. Zhu, J. P. Bowen, and J. He, "Deriving operational semantics 
from denotational semantics for Verilog," in 8th Asia-Pacific Software 
Engineering Conference (APSEC 2001). IEEE Computer Society, 4-7 
Dec 2001, pp. 177 - 184. 
[15] S. F. Smith, "From operational to denotational semantics," in Proceed- 
ings of the 7th International Conference on Mathematical Foundations 
of Programming Semantics, ser. LNCS, vol. 598, 1992, pp. 54-76. 
[16] J.-J. C. Meyer and E. Vink, On Relating Denotational and Operational 
Semantics for Programming Languages with Recursion and Concur- 
rency, ser. Open Problems in Topology. Elsevier, 1990, ch. 24, pp. 
387-406. 
[17] S. Owre, J. Rushby, and N. Shankar, "PVS: A Prototype Verification 
System," in 11th International Conference on Automated Deduction 
(CADE), ser. Lecture Notes in Artificial Intelligence, D. Kapur, Ed., 
vol. 607. Springer- Verlag, June 1992, pp. 748-752. 



54 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 



The Importance Analysis of Use Case Map 
with Markov Chains 



Yaping Feng 

School of Computer engineering, Kumoh National Institute 

of Technology, Korea 



Lee-Sub Lee 

School of Computer engineering, Kumoh National Institute 

of Technology, Korea 



Abstract — UCMs (Use Case Maps) model describes functional 
requirements and high-level designs with causal paths 
superimposed on a structure of components. It could provide 
useful resources for software acceptance testing. However until 
now statistical testing technologies for large scale software is not 
considered yet in UCMs model. Thus if one applies UCMs model 
to a large scale software using traditional coverage-based 
exhaustive tasting, then it requires too much costs for the quality 
assurance. Therefore this paper proposes an importance analysis 
of UCMs model with Markov chains. With this approach not only 
highly frequently used usage scenarios but also important objects 
such as components, responsibilities, stubs and plug-ins can also 
be identified from UCMs specifications. Therefore careful 
analysis, design, implementation and efficient testing could be 
possible with the importance of scenarios and objects during the 
full software life cycle. Consequently product reliability can be 
obtained with low costs. This paper includes an importance 
analysis method that identifies important scenarios and objects 
and a case study to illustrate the applicability of the proposed 
approach. 

Keywords- Use Case Maps; Markov chain; Usability Testing. 



I. Introduction 

UCMs (Use Case Maps) [1, 2, 3] is a set of semi-formal 
notations for describing scenarios of a system. This notation is 
being standardized as a part of the URN (User Requirement 
Notation) that is the most recent addition to ITU-T's 
(International Telecommunication Union-Telecommunication) 
family of languages. UCMs model provides a scenario-based 
model for a system. And it has been successfully applied to 
wide range of systems, including telecommunication systems 
[4], distributed systems [1] and etc. Furthermore, Daniel 
Amyot applied UCMs model for customer-oriented acceptance 
tests for Web applications [5]. But existing research works 
have not considered statistical testing method yet. Traditional 
coverage-based exhaustive testing might be impractical for the 
large scale software products because of the size as well as the 
generally uneven distribution of problems and usage 
frequencies in different areas and product components. 

The general solution of testing for large software products 
uses product reliability goals as a stopping criterion. The use of 
the criterion requires testing to be performed under an 
environment that resembles actual usage by target customers so 

This paper was supported by Research Fund, Kumoh National Institute of 
Technology 



that realistic reliability assessment can be obtained, resulting in 
the so-called SUT (Statistical Usage Testing) [3]. 

A prerequisite to such statistical testing and reliability 
analysis strategies is collection of usage information and 
construction of corresponding usage models. UCMs describe 
functional requirements and high-level designs with causal 
scenarios superimposed on a structure of components. 
Furthermore individual scenarios from UCMs can be generated 
automatically from related tools [6]. Hence, it is 
straightforward to collect usage information from UCMs to 
construct usage model. 

This paper proposes an importance analysis of UCMs 
model with Markov chains which applies statistical technique 
to UCMs model. If one simply applies traditional statistical 
techniques to usecase model, only highly frequently used usage 
scenarios can be identified and tested more thoroughly than 
less frequently used ones. But the approach of this paper is 
based on UCMs model which includes more objects like 
components, responsibilities, stubs and plug-ins. So this paper 
has proposed mechanisms and object model of UCMs which 
can calculate the importance of these objects. Thus efficient 
testing can be given to both highly frequently used usage 
scenarios and important objects to assure and maximize the 
product reliability from a customer's perspective with low cost. 
Since the combination of UCMs model and statistical 
technique can identify much more important information, it 
will be very beneficial for quality assurance engineering. 

This paper is organized as follows. Related works are 
discussed in section 2. Section 3 describes the background 
concepts of UCM scenario models and Markov chain usage 
models. An importance analysis of UCM with Markov chains 
is presented in section 4 with a case study. Conclusions and 
future works are discussed in section 5. 



II. Related Works 

Markov chain usage model is a probabilistic model based 
on FSM (Finite State Machine) model. It can be generated from 
textual use cases [7, 8] and then the most likely usage scenarios 
are identified as test cases. Since these works are based on the 
normal use cases, with these traditional approaches important 
components and events can not be identified. 

Kallepalli and Tian developed UMMs (Unified Markov 
Models) to support statistical testing, performance evaluation, 



55 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



and reliability analysis [9]. They applied the model to web 
application testing and proposed an automated way of 
constructing UMMs model from web server logs [9, 10]. With 
this approach, one web page or one information file serves as a 
state, so components such as Java components and events 
cannot be included in the scenarios. Due to the complexity of 
the model level in the web applications, identifying important 
components and events did not be covered. 



Daniel Amyot applied UCM model to customer-oriented 
acceptance tests for Web applications [5]. But complete 
coverage of acceptance tests in this approach is impractical for 
the large projects. It is necessary to identify highly frequently 
used usage scenarios and components in order to achieve 
efficient tests to assure and maximize the product reliability. 



III. Background 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
number is checked against the list (chk). If the call is denied, a 

relevant message is prepared for the originating party (md). 

This successful request causes the called agent to verify (vrfy) 

whether the called party is idle or busy (terminating plug-in in 

stub ST). If he is idle, then there will be some status update 

(upd) and a ring signal will be activated on UserT's side (ring), 

concurrently, preparation of a ring-back targeted to the 

originating party (mrb). Otherwise, a message stating that 

UserT is not available will be prepared (mb) and sent back to 

UserO (msg). For a detailed description of UCMs notation, 

readers should refer to [2] and [12]. 



Compared to the traditional use case diagrams, UCMs 
contain more object information such as components, 
responsibilities, stubs and plug-ins. Thus this paper applies 
statistical technique to UCMs model which can analyze not 
only scenarios but also these additional objects to assure and 
maximize the product reliability from a customer's perspective. 



A. Use Case Map 

UCM is part of user requirements notation standards 
proposed to the ITU-T for describing functional requirements 
as causal scenarios. UCM allows developers to model dynamic 
behavior of systems where scenarios and structures may 
change at run-time. For these reasons, UCM has been widely 
used in a range of systems [3]. 

UCM notation consists of three formal elements that form 
the basis for all maps: paths represent scenarios; components 
including system and non-system entities perform 
responsibilities; and responsibilities such as actions, events and 
so on are linked to paths and may be contained within 
components to indicate that the component executes that event. 
UCM scenarios begin with start points which represent 
triggering events and/or pre-conditions which are required for 
the commencement of the scenario. Scenario paths are 
followed to endpoints that represent terminating events or post 
conditions of the scenario execution. The UCM notation also 
provides a hierarchical abstraction mechanism in the form of 
stubs (diamonds) and plug-ins (sub-maps). Each hierarchy of 
maps has a root map that contains stubs where lower-level 
maps can be plugged in. 

UCM are primarily visual, but a formal textual 
representation also exists. Based on the XML (extended 
Markup Language) 1.0 standard, this representation allows for 
tools to generate UCMs or use them for further processing and 
analysis [11]. 

We mentioned Figure 1 to demonstrate the fundamental 
understanding of UCMS and it will be used as an example in 
the rest of the paper. The figure shows a simple UCMs model 
where UserO attempts to establish a telephone call with another 
UserT through some network of agents. Each user has an agent 
responsible for managing subscribed telephony features. UserO 
first sends a connection request (req) to the network through his 
agent. UserO' s agent has an originating dynamic stub SO 
which has two plug-ins, default originating and OCS. UserO 
can subscribe to one of services. The OCS plug-in shows an 
object OCSlist that represents a list of screened numbers that 
the originating user (UserO) is forbidden to contact. The called 



B. Markov chain usage models 

Markov chains is the simplification based on the so called 
memoryless property, or the Markovian property, which states 
that the state transitions from a given state depend only on the 
current state, but not the history or how we reached that 
particular state [13]. 

As shown in Figure 2, a usage chain consists of states, The 
Markov chain is completely defined when transition 
probabilities are established that represent the best estimate of 
real usage. For each state the summation of output arcs or 
probabilities should be 1. Transition probabilities can be 
obtained from various sources containing information about the 
actual usage counts and relative frequencies. Several methods 
can be employed to extract this information, including 
subjective evaluation based on expert opinions, survey of target 
customers, and measurement of actual usage logs. 



default ml default out] 
• > 1 



Default originating plug- in 
Oscr:0 Mwrtag 



ocs in I 



v±fl 



OCSlist 



[aLlo«d] 
[i allowed.] 



OCS plug-in 




terminating phig-in 



Figure 1 . A Simple telephone system UCMs model 



56 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vnl 7 Nn 1 7010 



An operational sequence consisting of visits to multiple 
states can be constructed by following the state transitions. The 
likelihood for a particular sequence to happen can also be 
easily calculated by the product of its individual state transition 
probabilities. Therefore, Markov chains can be used to ensure 
performance and reliability based on usage scenarios and 
frequencies by target customers. Furthermore, Kallepalli and 
Tian developed UMMs to support statistical testing, 
performance evaluation, and reliability analysis. UMMs 
possess a hierarchical structure formed by a collection of 
Markov chains. It captures information about execution flow 
(control flow), transaction processing (workload creation, 
handling, and termination), and associated probabilistic usage 
information. 

As introduced in section 3.1, UCMs model also has a 
hierarchical structure, a root map that contains stubs where 
lower-level maps can be plugged in. Thus it is straightforward 
to convert UCMs model to UMMs. 

For example as shown in Figure 3, the top level Markov 
chain, depicted by the upper part of the diagram, represents 
high-level operational units (states), associated connections 
(transitions), and usage probabilities. Various sub operations 
may be associated with an individual state and can be modeled 
by more detailed models or sub models, as shown in the down 
Markov chains. 



IV. An Importance Analysis of UCM with Markov 
Chains 

This paper proposes an importance analysis of UCMs 
model with Markov chains which can identify highly 
frequently used usage scenarios and generate lists of important 
objects such as components, responsibilities, stubs and plug-ins 
in UCMs specifications. This section will introduce overall 
phases and detail procedure with an example. 

The main procedure of this approach is illustrated in Figure 
4. 



P(E,B)=1 




P(C,D)=0.99 
Figure 2. A Markov usage chain example 




Figure 3. UMMs example 
Probabilities 



(i) 



(iii) 



UCM model 



(ii) 



Usage model 



(iv) 



Scenario definitions 



Individual scenario 
usase models 




Highly frequently 
used usage scenarios 



Important 
components 



Important 
responsibilities 



Important 

stubs 



Important 
plug-ins 



Figure 4. Overall procedure of the proposed approach 

This approach includes five steps as follows: 

1) Construct a UCMs model during requirement phase: A 
UCMs model should be constructed during requirement phase 
(i); UCMNav or jUCMNav tool can be used to draw UCM 
notations [3J. 

2) Define all the possible scenarios from the UCMs 
model: Since it is easier to identify transition probability on 
scenario basis; the next step is generating all of the possible 
scenarios from the UCMs model The generation of scenarios 
is for Transition probabilities should be prepared for each 
scenarios path 

3) Convert the UCMs model to an usage model with 
probabilities: After assigning probabilities, the UCMs model 
is converted to a usage model based on some rules that will be 
discussed later 

4) Generate each individual scenario usage model. 

5) An importance analysis of the objects of each object 
types of the UCMs: With the individual scenario usage models, 
an importance analysis should be followed to identify highly 
frequently used usage scenarios and generate lists of objects 

(components, responsibilities, stubs and plug-ins) ranked by 
their relative importance factor in the early phases of software 
life cycle. 



57 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



TABLE I. Scenario definitions for simple telephone system 



Scenario name 


Pre-condition 


Post- 
condition 


NormalldleCall 


Start point = req 
SO = default inl 
ST = in2 
[idle] = true 


null 


NormalBusyCall 


Start point = req 
SO = default inl 
ST = in2 
[idle] = false 


null 


OCSDeniedCall 


Start point = req 
SO = ocs inl 
[idle] = false 


null 


OCSAllowedldleCall 


Start point = req 
SO = ocs inl 
ST = in2 
[allowed] = true 
[idle] = true 


null 


OCSAllowedBusyCall 


Start point = req 
SO = ocs inl 
ST = in2 
[allowed] = true 
[idle] = false 


null 



TABLE II. The correspondences between UCMs model and usage 

MODEL 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
C. Identifying Scenario definitions for UCMs model 

For extracting individual scenarios, all possible scenarios 
should be defined. For each scenario, one needs to provide a 
scenario name, pre-condition including the list of start points to 
be triggered and initial values of global Boolean variables, and 
(optionally) a post-condition used to assert the validity of a 
scenario once the traversal has completed. 

As shown in Figure 1, a simple telephone system is 
modeled with jUCMNav editor. The following five scenarios 
are identified from the UCMs model as shown in Table 1 : 

1) NormalldleCall: UserO subscribes default originating 
plug-in and gives a call to UserTwho is idle. 

2) NormalBusyCall: UserO subscribes default originating 
plug-in and gives a call to UserTwho is busy. 

3) OCSDeniedCall: UserO subscribes originating call 
screening plug-in and gives a call to UserT, but this number is 
in the list of screened numbers that the originating UserO is 
forbidden to contact. 

4) OCSAllowedldleCall: UserO subscribes originating 
call screening plug-in and gives a call to UserT whose number 
is allowed and UserT is idle. 

5) OCSAllowedBusyCall: UserO subscribes originating 
call screening plug-in and gives a call to UserT whose number 
is allowed and UserT is busy. 



UCMs model 


Usage Model 


Start/end points 


States 


Responsibilities 


States 


Scenario path 


Transition path 


AND/OR forks and joins 


AND/OR forks and joins 



AND relation 



OR relation 



-a ^>-k 



UCMs model Usage model 

Figure 5. "AND/OR" relation translation 



B. An UCMs model construction 

UCMs graphical models describe functional requirements 
and high-level designs with causally linked responsibilities, 
superimposed on structures of components. With the help of 
editor UCMNav or jUCMNav, it is easy to construct UCMs 
model. A detailed introduction of how to use these editors can 
be found in [3]. 

For an example, a UCMs model of a simple telephone 
system is illustrated in Figure 1. It is designed with jUCMNav 
editor. 



D. Converting UCMs model to usage model with 
probabilities 

The general approach of usage model construction includes 
two steps. First, construct a basic model with basic states 
(nodes) and state transitions (links) identified from product 
specification. Second, complete the usage model by assigning 
transition probabilities. 

There are some quite close correspondences between some 
of the scenario entities in UCMs model and usage model 
elements. Table 2 illustrates these correspondences. In the Table 
2, UCM start/end points or responsibilities can represent states 
in usage model. Causal relationships path between 
responsibilities in UCMs model can serve as transition path 
between states in usage model. AND/OR forks and joins 
relationships serve as AND/OR forks and joins in usage model. 
The AND/OR relation is shown in Figure 5. 

"Stub" elements in UCM root map can be converted to 
states in top-level models of UMMs. "Plug-in" elements in 
UCM sub-map can be converted to sub-level models of UMMs 
as shown in Figure 6. The big round with "S" means a state. 
The "PI" and "P2" identify the pathl and path2. If the "Stub" 
is a dynamic stub, an additional "or" relation by a small round 
notation is added. 

The second step of the construction of a usage model is 
assigning transition probabilities. Several methods such as 
expert opinions, survey of target customers, and measurement 
of actual usage logs can be employed to extract these 
probabilities [10]. Basic UCMs model is augmented with 



58 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 



assigning probabilities to start points, including plug-ins' start 
points, and branches on OR-forks. 

As shown in Figure 1 , a simple telephone system has been 
modeled by UCM. This UCMs model should be converted to 
usage model that is illustrated in Figure 7. This usage model is 
an UMM, the top-level UMM represents the root map in the 
UCMs model, and there are 2 sub-level UMMs which represent 
plug-ins of two stubs in top-level model. The numbers assigned 
to transitions represent transition probabilities. 

E. Individual scenario usage model generation 

As shown in Table 1, five scenarios are defined for the 
example simple telephone system. Thus based on the whole 
usage model, individual scenario usage models can be 
generated. Figure 8 shows NormalldleCall scenario usage 
model. Other four scenarios usage models can also be 
generated with the same way. 




-<*£ ^ 



-®<fl 



Sub-level UMM 




Sub-level UMM 



UMMs usage model 



Figure 6. "stub" and "plug-in" elements of UCMs in UMMs 



Top-level UMM 



Low-level UMM _-™ J _ + 
Stub (SO) 



U-:c- ■ 




/oen 


:0 




Ageni T 




I.I;,;.; | 






reg 




i.i 


StubfSO'J 




Si 


StutKST) 






ring 




I 




.-•■ 




Cut! 






.= !■:•; 














out; 








ouM 














m$a 

































out! 



^ 



outl 
1 

out! 



out! 



Low-level UMM 
Stub (ST) 



m2 

t 




Figure 7. Usage model for the simple telephone system 



Vnl 7 Nn / 7010 



Top-level UMM 



Lew-level UMM 

Stub (SO) 



Low-level UMM 
Stub (ST) 



usei o 




Agent:0 




AaeniT 




User:T 






f&a 




ir:1 


StubCSO) 




in2 


StiiDCST) 




... 


■!■"■ 




1 




- 




M..II 






„,;-. 




















DUI4 














mss 































0.5.. C'-M vuit i-n „ Lc-,-ii.iia.ii 0ii: j 



jpd 




out3 










^p^j-^--^ 



Figure 8. NormalldleCall scenario usage model 

F. An Importance Analysis of UCMs 

In this section, an importance analysis methodology of 
UCMs model will be introduced. Firstly we will provide 
several definition of the model. Secondly with an example the 
insight of the equations for importance will be illustrated. 
Finally the equations are presented. 

The definition of importance of an object is given as 
Definition 1. According to the definition, if an object is used 
more frequently by users, the importance of the object is higher. 

Definition 1: Importance of an object is usage probability 
of the object from a customer's perspective. 

As shown in the background of UCMs, there are several 
kinds of objects existed in UCMs model such as scenarios, 
components, responsibilities (start/end points), stubs and plug- 
ins. According to the description of UCMs, a component can 
contain responsibilities, stubs, plug-ins, and other components. 
And also a plug-in can contain responsibilities, stubs and 
components. We can easily notice that is has a composite 
pattern. Therefore all the objects in the UCMs model are 
classified as three types: scenarios, primitive objects and 
containers. The definitions of these objects are given as 
follows: 

Definition 2: A scenario is a path which consists of 
transitions from start point to end point across the related 
responsibilities which may bind to the components. 

Definition 3: A primitive object is an object which is not 
includes any other objects. 

Definition 4: A container is an object that consists of other 
objects such as primitive objects or containers. 

The importance of a scenario can be calculated as product 
of related individual transition importance or probability. For a 
primitive object, the importance is summation of product of 
each related scenario's importance and the number of 
appearances of primitive object in this scenario. The 
importance of a container can be calculated as summation of 
the importance of its contained objects. 

Figure 9 shows an example of an object model to illustrate 
the insight of the calculation of the importance of each type of 
the object. This object model is actually a lattice model. In this 
example, four scenarios (S b S 2 , S 3 and S 4 ) are defined, and 



59 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 



there are six primitive objects and four containers. Scenario Si 
has three transitions T b T 3 and T 5 . Because the transition 
probabilities should be prepared initially, the importance of 
scenarios SI can be calculated as I (Si) = I (Ti) * I (T 3 ) * I (T 5 ). 
The primitive object R01 appears in scenario Si once and in 
scenario S 2 also once. So the importance of primitive object 
R01 can be calculated as I (R01) = I (SO * 1 + I (S 2 ) * 1. For 
the container C03, it contains two child objects R01 and R03. 
Thus the importance of the container C03 is calculated as I 
(C03) = I (R01) + I (R03). By this way, the importance of all 
the objects can be calculated. 



COI CBg 



ST 




co: 


T; ■'■ 

T 4 


L 


EFI 


HI 
mj 






4 containers 



,Q™) 6 primitive objects 



4 scenarios 



T 4 J 6 transitions 



Figure 9. An example of object model for importance analysis of UCM 



After generalizing the above example we can define formal 
equations of calculating the importance of all the objects in 
UCMs model as follows. The Eq.l is used for the importance 
of a scenario. The Eq.2 is for the importance of a primitive 
object and the Eq.3 is used for a container. 



TeS 



/OS) =n 7(7*0 



(1) 



Where: 

S represents a scenario 
I(S) is importance of a scenario S 
Tj is a transition included in scenario S 
I(Tj) is importance of a transition T t in this scenario which 
is initially defined 



Si^S(C) 



I(PO)= ^(l(Si)xN(PO 9 S0) 



(2) 



Where: 



PO represents a primitive object 

I(PO) is importance of a primitive object PO 

Sj represents a scenario 

I(Sj) is importance of scenario Sj 

S(C) is a set of scenarios which includes the primitive 

object PO 
N(PO, Sj) is the number of the primitive object PO 

appeared in scenario S t 



Oi^O 



Vol 7, No. 1, 2010 
TABLE III. Scenarios importance for the simple telephone system 



Scenarios 


Importance 


NormalldleCall 


0.48 


NormalBusyCall 


0.12 


OCSDeniedCall 


0.12 


OCSAllowedldleCall 


0.224 


OCSAllowedBusyCall 


0.056 



TABLE IV. Important scenarios with overall importance 

THRESHOLD 0.2 



Scenarios 


Importance 


NormalldleCall 


0.48 


OCSAllowedldleCall 


0.224 



TABLE V. 



Table 5. Important scenarios with alternative 
importance threshold 0.3 



Scenarios 


Importance 


NormalldleCall 


0.6*0.8=0.48 


OCSDeniedCall 


0.4*0.3=0.12 


OCSAllowedldleCall 


0.4*0.7*0.8=0.224 



Where: 

C represents a container 

1(C) is importance of a container C 

O(C) is a set of child objects of container C. 

Oj represents a contained object in the container C. The 

contained objects may be child primitive objects or 

containers 
I(Oj) is importance of object O f 

G. Case study of the importance analysis of UCM: a simple 
telephone system 

In this sub-section we will show an example according to 
the importance analysis method explained in 4.5. For the 
simple telephone system, five scenarios are identified from the 
UCMs model (Table 1). Thus by Eq. (1), the importance of 
these five scenarios is calculated and the result is shown in 
Table 3. 

After calculating importance of scenarios, to identify the 
highly frequently used usage scenarios, some thresholds are 
used. In practical applications, thresholds can be adjusted to 
control the numbers of test cases to be generated and executed. 
Several thresholds have been initially proposed in [14] and 
used in developing UMMs. Two kinds of thresholds are used in 
this paper: 

1) Overall importance threshold for complete end-to-end 
operations to ensure that commonly used complete operation 
sequences by target customers are covered and adequately 
tested. 

2) Alternative importance threshold to ensure commonly 
used operation pairs, their interconnections and interfaces are 
covered and adequately tested. 



/(C) 



I coo 



(3) 



60 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



TABLE VI. 

SYSTEM 



Importance of all responsibilities in simple telephone 



Responsibilities 


Importance 


req 


1 


msg 


1 


vrfy 


0.88 


out4 


0.88 


in2 


0.88 


upd 


0.704 


ring 


0.704 


out3 


0.704 


mrb 


0.704 


default outl 


0.6 


default inl 


0.6 


ocs inl 


0.4 


chk 


0.4 


ocs outl 


0.28 


mb 


0.176 


out2 


0.12 


md 


0.12 




Five scenarios 

Figure 10. Composition Object model of UCMs model for simple telephone 

system 



Percent of each responsibility 




El req (10*) 

■ nsg (LOU) 
Q vrfy (9*) 
Oout4 (9») 

■ in*: (9tt) 
D upd (7%j 

■ rati* iW 
OoutS (7%) 

■ mrb (7%) 

■ default out! <M) 
D default iaX (6*) 
Oocs inl (4%) 

■ chk (4%) 

■ ocs outl (3%) 

■ mb (2%) 

■out 2 tin) 

■ ffid (1%) 



Figure 1 1 . Percent of each responsibility importance 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
So if overall importance threshold is given as 0.2, then the 

highly frequently used usage scenarios can be identified as 

shown in Table 4. If alternative importance threshold is given 

as 0.3, Table 5 shows the highly frequently used usage 

scenarios. 

For calculating the importance of components, 
responsibilities, stubs and plug-ins in UCMs model of simple 
telephone system, firstly a composition object model as shown 
in Figure 10 is given. This object model clearly describes the 
architecture of relations among different objects in UCMs 
model. Based on this object model, importance of each object 
can be easily calculated. 

Responsibilities in this object model are all primitive 
objects, so with Eq. (2), the importance of all the 
responsibilities can be calculated as shown in Table 6. 

Plug-ins in this object model are all containers, importance 
of each plug-in can be calculated as summation of its children's 
importance by Eq. (3). The result is shown as follows: 

• Importance of plug-in Default ORIGINATING =1.2 

• Importance of plug-in TERMINATING = 4.928 

• Importance of plug-in OCS = 1.32 



Importance of each stub and component also can be 
calculated with Eq. (3). The result is shown as follows: 

• Importance of stub SO = 2.52 

• Importance of stub ST = 4.928 

• Importance of component AgentT = 4.928 

• Importance of component AgentO = 4.928 

• Importance of component UserO = 2.52 

• Importance of component UserT = 0.704 

After getting all objects importance, a percent of 
importance for each object type can be calculated to show 
which objects are more important clearly. Percent of 
responsibility importance is shown in Figure 1 1 . 

The calculation of the importance is very simple and 
straightforward. As presented above the importance of each 
scenario can be easily calculated by Eq. (1) using transition 
probabilities, and after getting the importance of scenarios, the 
importance of primitive objects can be calculated with Eq. (2) 
based on the object model. For containers, the importance can 
be calculated as simple summation of its children's importance 
byEq.(3). 

In the early phases of the software life cycle (requirement 
analysis phase) the importance of all scenarios, components, 
responsibilities, stubs and plug-ins, and the highly frequently 
used usage scenarios and important objects can be identified 
from a customer's perspective. Since the importance of each 
object type can be identified at the early phase where UCM is 
defined, the quality engineering efforts can be applied from 



61 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



3 UCM > 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 




Acceptance Testing f 



eKmmary Design] f Svstan Testina 1* 

system design) J ^ \ J 



: Lnportant scenarios and objects ! J" Detailed Design I 1 

(.program design) I [_ 



Integration testing 




Figure 12. UCM in V model 

first phase to the last phase. That means this approach can 
support in the full software life cycle for quality assurance 
activities. For instance as shown in Figure 12, since the 
importance of each objects is identified at the Requirement 
Analysis phase, the important scenarios can be heavily tested 
on the Acceptance test, the important containers are can be 
took cared at the system testing and integration testing and the 
important primitive object can be focused at unit testing. 

V. Conclusion 

This paper proposes an importance analysis of UCM with 
Markov chains which applies statistical technique to UCMs 
model. Like other traditional statistical techniques, highly 
frequently used usage scenarios can be identified. 

Furthermore, since the approach of this paper is based on 
UCMs model, it can identify more important object types such 
as components, responsibilities, stubs and plug-ins. Therefore 
this paper also proposes an object model of UCM which can 
calculate the importance of these objects. Thus the efficient 
testing can be possible for both highly frequently used usage 
scenarios and important object types to assure and maximize 
the product reliability from a customer's perspective. Since the 
combination of UCMs model and statistical technique can 
identify much more important information, it is very beneficial 
for quality assurance engineering. 

The importance of each object is identified at the 
Requirement Analysis phase, which means the approach make 
possible the quality driven development during the full 
software life cycle. The method of the importance calculation is 
very clear, easy and straightforward that it is very practical to 
the real environment. To show these characters we include a 
simple telephone example with importance calculations. 

In this approach, it requires manual operation to generate 
usage model and object models. Future work will extend the 
UCM Navigator (UCMNav or jUCMNav) editor to support 
automatic generation. Future work will also include the 
research of an algorithm to generate all possible scenarios 
automatically. 



References 



[i] 



R.J.A. Buhr, "Use Case Maps as Architectural Entities for Complex 
Systems", IEEE Transactions on Software Engineering, Special Issue on 
Scenario Management. December 1998, Vol. 24, No. 12, pp. 1131-1155. 

[2] "Use Case Maps Web Page and UCM User Group", 
http://www.UseCaseMaps.org, 1999. 

[3] J.A. Whittaker, and M.G. Thomason, "A Markov chain model for 
statistical software testing", IEEE Trans on Software Engineering, Oct. 
1994,Vol. 20, No. 10, pp. 812-824. 

[4] Amyot, D., Specification and Validation of Telecommunications System 
wtih Use Case Maps and Lotos, in School of Technology and 
Engineering. 2001, University of Ottawa: Ottawa, d 

[5] D.Amyot, J.F. Roy, and M.Weiss, "UCM-Driven Testing of Web 
Applications", 12th SDL Forum, June 2005, pp. 247-264. 

[6] D. Amyot, X.Y. He, Y. He and D.Y. Cho, "Generating scenarios from 
use case map specifications", Third International Conference On Quality 
Software (QSIC), Dallas, USA, Nov. 2003, pp. 108-115. 

[7] M. HUBNER, I. PHILIPPOW, and M. RIEBISCH, "Statistical Usage 
Testing Based on UML", Proceedings of the 13th Annual IEEE 
International Symposium and Workshop on Engineering of Computer 
Based Systems, 2006. 

[8] B. Regnell, P. Runeson, and C. Wohlin, "Towards Integration of Use 
Case Modelling and Usage -Based Testing", Journal of Systems and 
Software, 2000, pp. 50:117-130. 

[9] C. Kallepalli, and J. Tian, "Measuring and Modeling Usage and 
Reliability for Statistical Web Testing", IEEE Transactions on Software 
Engineering, Nov. 2001, vol. 27, no. 11, pp. 1023-1036. 

[10] C. Kallepalli, and J. Tian, "Usage Measurement for Statistical Web 
Testing and Reliability Analysis", Proceeding of Seventh International 
Software Metrics Symposium, London, IEEE Computer Society Press, 
2001, pp. 148-158. 

[11] D. Amyot, and A. Miga, Use Case Maps Linear Form in XML, version 
0.13, May 1999. 

[12] R.J.A. Buhr, and R.S. Casselman, "Use Case Maps for Object-Oriented 
Systems", Prentice-Hall, USA, 1995. 

[13] S. Karlin, and H. M. Taylor, A First Course in Stochastic Processes, 2nd 
Ed. Academic Press, New York, 1975. 

[14] J. D. Musa. Software Reliability Engineering. McGraw-Hill, New York, 
1998. 



AUTHORS PROFILE 

Yaping Feng received his B.S. degree in softtware 
engineering from East China Normal University, China in 
2005 and received his M.S degree in computer 
engineering from Kumoh National Institute of 
Technology, Korea in 2007. He has worked as a developer 
in POSData China from 2008. His current research 
interests include Software Engineering and Mobile 
Computing. 

Lee-Sub Lee received his B.S., M.S. degrees in 
mathematics and computer science from Sogang 
University, Seoul, Korea, in 1988 and 1990, respectively. 
He ha a Ph.D degree in computer science & engineering 
from Korea University, Seoul, Korea in 2004. He is 
assistant professor of Kumoh National Institute of 
Technology since 2004. He had worked as a senior 
manager of the IT R&D Center, Samsung SDS, Ltd from 
1990 in SungNam, Korea. His current research interests 
include Software Engineering and Mobile Computing. 





62 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 



Convergence of Corporate and 
Information Security 



Syed (Shawon) M. Rahman, PhD 

Assistant Professor of Computer Science 

University of Hawaii-Hilo, HI, USA and 

Adjunct Faculty, Capella University, MN, USA 



Shannon E. Donahue CISM,CISSP 

Ph. D. Student, Capella University, 

225 South 6th Street, 9th Floor 

Minneapolis, MN 55402, USA 



Abstract — As physical and information security boundaries have 
become increasingly blurry many organizations are experiencing 
challenges with how to effectively and efficiently manage security 
within the corporate. There is no current standard or best 
practice offered by the security community regarding 
convergence; however many organizations such as the Alliance 
for Enterprise Security Risk Management (AESRM) offer some 
excellent suggestions for integrating a converged security 
program. This paper reports on how organizations have 
traditionally managed asset protection, why that is changing and 
how to establish convergence to optimize security's value to the 
business within an enterprise. 



Keywords-component; convergence; 
management; corporate; threats 



security; 



risk 



I. 



Introduction 



Throughout history organizations have had security and 
loss prevention departments to protect their physical assets. In 
the last 10 to 20 years however there has been a major shift in 
what is considered an asset. Information and intangible assets 
have increased significantly. In fact, one could make a very 
strong argument that information has become an enterprise's 
most important asset. In the not so distant past, organizations 
considered their most valuable assets to be physical assets. 
With the growth of the internet and increased methods for 
communication which has given most organizations the ability 
to do business globally, along with the expansion of data 
warehousing and electronic storage, information has fast 
become the most critical element to the success of an 
organization [1]. 

Information provides organizations with data on their 
customers, finances, inventories, suppliers, partners and 
competitors. Metadata (data formed by combining groups of 
information) provides organizations with vital information that 
organizations use for decision making on a daily basis. Even 
the organizations who do still mainly rely on physical resources 
need information to forecast and communicate with vendors 
and suppliers. 

Before intangible assets became the largest value to 
organizations throughout the world, most companies counted 
their physical assets as their primary asset. Organizations had 
loss prevention units and security departments that safeguarded 



the company's assets with cameras, physical access control 
measures, and security guards. 

These corporate security departments were made up largely 
of former law enforcement officers that reported to legal, 
compliance or risk management divisions. The main charges 
of this department were intrusion protection and investigations 
in the workplace. As the digital age came to life, organizations 
saw more and more information being stored on servers, in 
databases, and in files. 

Organizations began to realize the value that intangible 
assets provide and subsequently created departments 
responsible for securing that information. IT security 
departments and information security departments were tasked 
with managing the risk that surrounded information. 

These information security departments were created after 
organizations saw threats to their information in the form of 
hackers, malware, unavailability, and data theft. Additionally, 
both regional and international laws have surfaced which 
require dedicated information security managers to be 
responsible for a formal information security program which is 
responsible for data protection. Many of the people responsible 
for information security were moved into the role due in part to 
their technical backgrounds which were very useful when 
considering technical controls and designs of a logical 
perimeter. However, due to their technical backgrounds most 
of these information security professionals had no experience 
in traditional corporate or physical security. 

Initially having two separate groups responsible for the 
security of different types of assets was not a problem. 
However, some security functions have begun to converge on 
their own and this creates unique challenges for organizations 
with stove piped security functions. 

As technology evolved corporate security departments 
began to use some advanced tools and technology such as 
traditional closed circuit cameras running over IP networks. 
The cameras were the responsibility of the corporate security 
team, but IT had control of the IP network. The same problem 
happened with access cards. The corporate security department 
traditionally had control over physical access control but with 
databases housing all of the data, IT and information security 
again were clearly involved. Vendors have recognized this 



63 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



trend and have began to offer converged solutions for risks 
such as access control [12]. 



It occurred to many that having the corporate and 
information security departments work together may be able 
help to reduce an organization's overall risk profile while 
streamlining some redundant security processes. However, 
convergence did not look so appealing to everyone. While the 
unintentional convergence of systems was happening, the two 
departments who were initially affected were not working 
together to improve the risk. 

When discussing information and corporate security, 
convergence is defined as "a trend affecting global enterprises 
that involves the identification of risks and interdependencies 
between business functions and processes within the enterprise 
and the development of managed business process solutions to 
address those risks and interdependencies" [1]. This definition 
speaks to the need for organizations to look at convergence and 
begin to disassemble the organizational silos in order to 
encourage collaboration to manage risk holistically. Fostering 
an environment rich in collaboration will help the organization 
lower its overall risk and reach its business objectives. 

According to a 2007 article in security magazine [15], "the 
silo approach to managing enterprise risks is inadequate 
because it leaves too many gaps and provides no reliable way 
to evaluate an enterprise's risk position" [15]. Due in no small 
part to these gaps, many organizations have tried to begin to 
explore converging the functions. In reality, convergence 
between information and corporate security is still very 
immature. In most organizations, even if the departments both 
report to a single Chief Security Officer (CSO) or Chief Risk 
Officer (CRO) the departments are still as different as apples 
and oranges acting in their traditional silos and worse than that 
oftentimes shutting one another out. 

II. Reasons for convergence 

One reason for convergence is that threats continually 
increase and become threats to both corporate and information 
security safeguards. If an organization's corporate and 
information security departments do not work together they 
may miss out on valuable information that could be beneficial 
to both areas regarding particular risks and threats. 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 
Additionally, working in silos may be detrimental as 
downstream risks that are an emerged result of other risks may 
not be considered. 



United States Computer Emergency Readiness Team (US- 
CERT) [2] interacts with federal agencies, industry, the 
research community, state and local governments, and others to 
collect reasoned and actionable cyber security information and 
to identify emerging cyber security threats. Based on the 
information reported, US -CERT was able to identify the 
following cyber security trends (figure 2 and 3) for fiscal year 
2009 first quarter [2]. 

Figure 1 displays the overall distribution of cyber security 
incidents and events across the six major categories. The 
percentage of Category 5 (Scans, Probes, or Attempted 
Access) reports decreased for the second consecutive quarter. 
This was a 2.9% decrease in CAT 5 incidents compared to the 
previous quarter. The percentage of Malicious Code incidents 
increased by 3.3%. Figure 2 is a breakdown of the top five 
incidents and events versus all others. Phishing remained the 
most prevalent incident type, accounting for 70% of all 
incidents reported. This was a slight percentage decrease of 
1.8% from the previous quarter, on the other hand, The 
sophistication of attack tools has gone up, while the level of 
skill required to use those tools has gone down(refer to figure 
3). At this stage, the attacker takes advantage of his or her 
ability to steal confidential and proprietary data and sells it for 
profit or uses it for military intelligence [3]. 

Another strong argument for convergence is the blurring of 
boundaries between corporate and information security. For 
example, if a corporate security department is in charge of 
corporate access control to restricted areas via card readers, but 
IT owns the systems who responds when there is a major 
breach? Likewise, if there is a disaster and information security 
and corporate security are not aligned with their plans whose 
plan does the organization follow ? 

Convergence offers the organization the opportunity to 
restructure systems. Currently, systems in physical and 
information security are oftentimes segregated from one 
another and are not aware of what the other systems are doing. 
Once an organization decides to move forward with 
convergence, systems can be combined which saves the 




■ Unauthorised Access 4,2% 

■ Malicious Code 6.9% 

■ Improper Usage 5.5% 

■ Scansj Probes & Attempted Access 73.4% 
Under Investigation / Other 10.0% 
Total: 100.0% 




I Phishing 70.0% 

I Malware 6.6% 

I Policy Violation 5.3% 

I Non Cyber 4.8% 
I Suspicious Network Activity 4.3% 

Others 8.9% 

Total: 100.0% 




Figure 1 : Incidents and Events by 
Category[2] 



Figure 2: Top Five Incidents vs. All Others 
[2] 



Figure 3. Attack tool trends [3] 



64 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



organization money on upkeep and maintenance of 
infrastructure, not having to purchase any new hardware and 
also lowers bandwidth uses on the company network. 

III. Benefits of convergence 

Joining the information and corporate security departments 
in some way will help to dispel some of the device 
management confusion. Whether it is through just working 
together to have information security and corporate security 
handle a joint presentation to promote awareness training, or 
actually joining the two separate functions to report to one 
CRO or one CSO benefits can be derived from some sort of 
convergence. If organizations choose to organizationally link 
the two functions there will be some immediate cost savings as 
a Chief managing both areas should have an understanding of 
which team members should be managing particular areas and 
can set goals that are achievable by both departments. This 
helps to enable the departments to work together towards a 
common goal and hopefully will reduce the 
amount of overall risk to the enterprise. 

Gaps between corporate and information 
security have caused problems in the past 
and convergence including collaboration and 
training can help to minimize these gaps. 
For example, if the information security 
department has all controls applied 
appropriately and theft occurs as a result of a 
thief posing as an employee the breach has 
still occurred. In fact, a good example of 
information and corporate security not being 
on the same page occurred recently at the 
Sumitomo Mitsui Bank in London, England. 
Criminals who posed as janitors within the 
bank had installed devices on computer 
keyboards that allowed them to obtain login 
information. The criminals tried 

unsuccessfully to steal £220 million. 
Information security controls had been applied but there was a 
physical security breach that could have been devastating to the 
organization [6]. 

While convergence could surely have helped in the above 
mentioned situation, many organizations are slow to adopt a 
converged approach. CSO's and CISO's have spoken up and 
want the business community to understand the benefits of 
convergence. It appears that business leaders are becoming 
more accepting of the idea of convergence. According to a 
2005 PricewaterhouseCoopers and CIO magazine survey of 
8,200 IT and security executives in 63 countries of, 53% of 
organizations have some level of integration between their 
corporate and IT security divisions [8]. That's up from just 29% 
in 2003. The projections seem to be growing as well. The 
Alliance for Enterprise Security Risk Management expected 
2005 global spending on convergence activities expected to 
reach $1.1 billion dollars and significantly move upward after 
that. 

Some of the major benefits that organizations are seeing as 
a result of converged information and corporate security are 



(IJCSIS) International lournal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
significant. Many organizations are saving millions of dollars 
by streamlining these functions. 



Other benefits exist as well. It is hard to put a dollar 
amount on how safe people feel in the office, or how not being 
the latest company to lose millions of customer's data is 
effecting the organization. But through these types if 
situations, security is not only keeping its assets safe but is 
preventing unnecessary funds from going out the door as well. 

According to a computerweekly.com report executives are 
seeing the benefits of convergence to organizational risk, 
"According to the results of a global survey conducted by the 
Economist Intelligence Unit (EIU) for AT&T, the majority of 
executives (52%) believe that having a converged network 
gives their companies a better defense against IT security 
breaches" [11]. 

Table 1 : Global security convergence spending forecast [4]. 





2004 


2005 


2006 


2007 


2008 


Large-scale convergence projects 
in NA and Europe 


19 


68 


175 


382 


856 


Physical/ logical access control 
projects in NA and Europe 


50 


150 


413 


903 


1,656 


Other projects performed jointly 
by Hand physical security 
departments in NA and Europe 


13 


45 


118 


246 


406 


Public sector: border control 
convergence systems, law 
enforcement projects in NA 
and Europe 


410 


820 


1,899 


4,202 


8,003 


Small projects (data center 
security, communications security, 
etc) in NA and Europe 


14 


40 


108 


229 


369 


Total 


506 


1,123 


2,713 


5,962 


11,289 



Cost control and productivity may also be improved as a 
result of convergence. By having systems joined, an 
organization can eliminate steps that are redundant. This helps 
to improve processes, eliminate human errors, and increase 
productivity which ultimately generates revenue for the 
organization. 

Organizations have learned quite a bit about business 
continuity and disaster recovery in the last few years. 
September 11, Hurricane Katrina, and devastating natural 
disasters around the world have brought an increased focus on 
business continuity (BCP) and disaster recovery planning 
(DRP). BCP and DRP are another area where convergence 
plays an important role. We believe information technology 
security and corporate security must work together to ensure 
that they are meeting one another's business needs when it 
comes to project their assets from inside and outside threats and 
recover data and resources from any attacks. 

The collaboration will also streamline efforts so that 
everyone can understand what to do in cases of emergencies. 
Collaborating on these efforts will be instrumental in restoring 



65 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



services in a timely manner. Restoration of these services is an 
extremely critical issue and is an indirect revenue generator 
since the quicker that services are restored the faster money is 
being made. 

While the benefits of convergence seem very obvious, there 
are quite a few challenges that will create some issues for the 
person who is charged with convergence. 

IV. Challenges of convergence 

Security convergence is not without its challenges. The 
information security department and corporate security 
departments have long operated in parallel. They have not 
shared and have not wanted to and, in a lot of situations have 
through of the other as an adversary. In order to have 
convergence be effective, collaboration between these two 
areas must happen. It is not enough to have them reporting 
through the same channel, they must both have common goals 
and have mutual discussions on how to meet those goals. In 
order to get to this level of cooperation there are huge cultural 
challenges that will need attention. 

Information security professionals and corporate security 
professionals do not usually have the same background. Many 
information security professionals have technical backgrounds 
while many of the corporate security professionals have law 
enforcement backgrounds. According to Steve Hunt, President 
of 4A international LLC a security consulting firm in Chicago, 
these differences can lead to a gap in how the two departments 
evaluate security technologies and controls [8]. 

Salaries are another issue. Corporate security professionals 
are not earning comparable salaries generated by information 
security professionals. Information security professionals often 
collect six figure salaries early in their career whereas the 
traditional corporate security professional might be making 
only half of that closer to the end of their career. Parity 
adjustments are not practical because there are ways to justify 
the salary disparity such as technical skills, higher levels of 
education, professional certifications, and a better job market. 
However, when an organization combines groups and has them 
working in tandem, if half of them are earning 50% of what the 
others are earning there will be some bitter feelings which may 
result in a hostile environment where people will not work 
together. 

One additional challenge is the training gap. Corporate 
security professionals often have not been trained on 
information security or technology. They are not aware of 
what to watch for when performing their own duties. Often 
they are unhappy with the convergence plan because they are 
concerned that they will not retain their positions. A cross 
training plan that would help the corporate security people 
understand better what social engineering is and how to spot it, 
what phishing is and how to spot it and some other technical 
training would go a long way towards team building and 
improved skill sets. Additionally, information security 
professionals likely have very little skill in surveillance, 
disaster response, investigations and loss prevention. Both sets 
of individuals need to have some basic training so they are at 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
least aware of what the other is doing and how it complements 
their own security efforts. 



Organizational culture is the most difficult thing to 
influence so any help from senior management is going to have 
added benefit. It is important that senior management really 
demonstrate support for the security program so that employees 
understand that it is a critical piece of the business that is 
everyone's responsibility in the same way that customer service 
and quality management are. 

Table 2: Summary of Benefits and Challenges of Convergence 



Benefits and challenges of Convergence 


Benefits 


Challenges 


Cost Savings 


Culture 


More holistic view of risk 


Salary Differences 


Reduction of risk profile 


Training requirements 


Streamline process 


Lack of Collaboration 



V. Beginning a converged program 

Once the organization decides that they want to converge 
their security program there are many things to consider. 
Organizational structure is one, budgets are another and the 
overall risk profile is a third. 



A. Organizational Structure 

Organizationally there are many routes that can be taken to 
align the physical and information security organizations. Each 
of the options has positives and negatives and may work in one 
organization and not at all in another. Security managers need 
to know the organization that they work in, consider their 
budgeting options as well as objectives and select a solution 
that will meet the needs of their business. 

The first method is to combine the physical and information 
security departments into one security organization reporting to 
a CSO or CRO. This group would have responsibility for all 
things security from the guard at the front desk to encryption 
protocols for sensitive data. 

While the model of combining all the players does force 
some level of integration, and if successful can be extremely 
effective in business process integration and fraud detection it 
is not without drawbacks. Going right to a fully immersed 
security program will increase the likelihood that the security 
staff will be upset. This unhappiness can result in a lack of 
cooperation and hostile work environments and may even 
cause damage to the risk profile because staff is so worried 
about their work situation that they aren't paying attention to 
the organizational risks. 

If this is the route chosen the security manager will need to 
spend plenty of time working on opening communication 
channels and fostering collaboration. Cross training will be an 
immediate need if expectations are for people to learn each 



66 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



other job functions and additional funding for corporate 
training may be necessary depending on the gap in skills. 

The next option is to keep the physical and logical 
departments as separate departments with their own budgets 
and reporting lines but ultimately have them report to the same 
executive, most likely a CSO. The CSO would receive 
information from both security areas and be able to make 
decisions based on the information provided to them by the 
business unit director. The groups would most likely not work 
joint projects or go through any cross training, but would still 
benefit from updated processes based on organizational goals. 
The departments would still need to collaborate to ensure all 
risks are addressed. While this level of integration cannot be 
considered complete immersion, security does not have to be a 
completely combined area to reap some of the benefits of 
convergence. This type of management can still be extremely 
effective at driving down costs and eliminating redundant work 
and systems. 

The third option which is becoming more popular these 
days is to keep the functions completely separate and assist in 
process management and collaboration through bringing 
security issues to a risk council staffed with business and 
security management that could make decisions regarding 
security. This approach would be helpful in minimizing culture 
issues and would probably be well received by members of 
both security teams. 

Whichever organizational structure choice is taken it is 
important to consider the culture of the organization. Culture is 
an often overlooked area that can be critical to the success of 
implementing convergence and should not be underestimated. 

Many things can be done to smooth the progress of 
convergence. Combining processes where possible, gaining 
senior management support for the initiative and beginning to 
look at organizational risks instead of just risks to individual 
departments will help to incorporate convergence smoothly. 

Many frameworks exist to help an organization lower their 
risk profile. It is important to remember that the reason for 
discussing convergence of information security and physical 
security is to show how security professionals can help to 
improve the overall risk profile in your organization. 

Some corporate risk management frameworks that are used 
internationally are the Committee of Sponsoring 
Organizations' (COSO) Enterprise Risk Management (ERM) 
framework, the Operationally Critical threats, Asset and 
Vulnerability Evaluation model (OCTAVE) and one more 
accepted framework for managing enterprise risk is the Risk 
and Insurance Management Society (RIMS) Risk Management 
Model (RMM). All of these models focus on managing 
operational risk and reducing the overall risk profile. 

We have found the AESRM recommendation useful while 
beginning a converged program [4]: 

♦ Establish a governance framework for managing 
security risks. 

♦ Define security requirements early in the planning 
stage. 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 
♦ Understand the technology better. 



♦ Analyze and understand security-related cost-benefit 
trade-offs. 

♦ Develop a unified set of meaningful standards. 

♦ Deploy special network security controls. 

♦ Implement effective authorization, accountability and 
auditability controls. 

♦ Include critical systems in organization continuity 
plans. 

♦ Protect information important for investigations. 

♦ Increase auditing and logging. 

♦ Require tailored training and awareness programs. 

♦ Pressure vendors to play a more active role in security. 

♦ Expand audit coverage of systems and devices. 

VI. Conclusion 

Organizational risks and threats are changing every day. 
As new technologies expand into the organization the risk 
profile continues to expand. Many organizations are in a 
constant state of growth. The organizational risk profile needs 
to constantly be reviewed and updated. 

Since the beginning of information security departments, 
corporate and information security have operated in their own 
silos with separate management teams, different risks, different 
processes, and different budgets. In the last decade, 
organizations have seen dramatic changes in their risk profile. 
With technical advances, traditional corporate security 
functions have evolved to include the use of networks, 
databases, and file servers. Compliance requirements have 
dictated how sensitive information needs to be transmitted and 
so traditional corporate security departments and information 
security departments find some of their functions that used to 
have clear management delineations becoming blurry. 

Although convergence has been a buzz word for a long 
time, companies have just recently started to notice the benefits 
of converging traditional siloed security functions to allow for 
an enterprise wide view of risk. Some of the benefits that can 
be derived from converging information and corporate security 
areas including streamlining processes, saving money on 
infrastructure, increasing productivity, positioning the 
organization better to have an broader view of organizational 
threats and risks and meeting compliance requirements. 

Many organizations have attempted to move towards a 
converged security area and have realized that there are many 
hurdles to be jumped on the way to a converged organization. 
One of these challenges is getting the security staff in both 
areas to work together. Organizations can work to encourage 
collaboration by ensuring that employee concerns are 
addressed and people feel confident that communication will 
be well received. Gaining support from senior management 
and combining processes to minimize confusion are two more 



67 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



techniques that will prove useful during the implementation of 
a converged security structure. 

While the challenges may be difficult to work through, the 
benefits that the organization will realize as a result of taking a 
wider view of risk will be immense. As risks can change at any 
moment it is absolutely critical to have as many well trained 
professionals as possible working together to improve the 
productivity and sustainability of an organization with a 
holistic approach to security management through convergence 
in enterprise security risk management 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. I, 2010 
[12] L. Stepanek, "Convergence here & how. SDM: Security Distributing & 
Marketing," 2007 . Business Source Complete Database, 37(2), 58-60 

[13] J. Watson, Physical and IT security must go together, May 
2005; http ://www. computing, co.uk/computing/news/207 1 725/integrat 
ion-way-ensure-security . Web retrieve on December 22, 2009 

[14] Willoughby, M. (2003, May 28). "Bridging the gap: Information 
security meets physical security." 

http ://www. computerworld. com/securitytopics/security/story/0, 1 080 L8 1 
589.00.html Web retrieve on December 22, 2009 

[15] B. Zalud, (2007) "Neo-security: Inclusive Enterprise Risk Management. 
Security: For Buyers of Products, Systems & Services" Business Source 
Complete Database 44(10), 12-20. 



References 



[I] Alliance for Enterprise Security Risk Management "Convergence of 
physical and information security in the context of risk management", 
2007; http './/www. aesrm.org/aesrm_convergence_in_ERM_pdf , Web 
retrieve on December 22, 2009 

[2] US-CERT, "Cyber Security Trends, Metrics, and Security Indicators", 
June 16, 2009. Volume 4, Issue 1. http://www.us- 
cert.gov/press_room/trendsanalysisQ109.pdf, Web retrieve on December 
22, 2009 

[3] Liu, Simon and Cheng, Bruce; "Cyberattacks: Why, What, Who, and 
How", IT Pro, IEEE Computer Society, May/June 2009. 

[4] Alliance for Enterprise Security Risk Management "Convergence of 
Enterprise Security Organizations", 2005; 

http://www.asisonline.org/newsroom/alliance.pdf , Web retrieve on 
December 22, 2009 

[5] Alliance for Enterprise Security Risk Management "Convergent security 
risks in physical security systems and IT infrastructures". ISACA 
Information Security Management Conference 2006. 

[6] CNN. "Police Foil $423m Cyber Robbery", Mar.2005; 
http://www.cnn.com/2005/WORLD/europe/03/17/bank.robbery/index.ht 
ml, Web retrieve on December 22, 2009 

[7] J. Burris, (2008). "Mission Impossible?", Community Banker, 17(6), 28- 
29. 

[8] T. Hoffman, "Security convergence", 2006; 

http ://www. computerworld. com/action/article. do?command=viewArticl 
eBasic&articleId= 108571 &pageNumber= 1 , Web retrieve on December 
22, 2009 

[9] C. Huang, R. Behara, O. Hu, 2008; "Managing Risk Propagation in 
Extended Enterprise Networks"; IT Professionals Magazine, 10(4), 14- 
19. 

[10] D. Radcliffe, "Companies move to combine, physical, IT security 
efforts" 2007; 

http ://www. landfield.com/isn/mailarchive/200 l/Feb/0 153 .html ; Web 
retrieve on December 22, 2009 

[II] A. Saavas, (2007, March 22). "Converged networks deliver the best 
security, say execs", 2007; 
http ://www. computerweekly . com/Articles/2007/03/22/2226 1 4/converge 
d-networks-deliver-the-best-security-say-execs.htm , Web retrieve on 
December 22, 2009 



AUTHORS PROFILE 

Syed (Shawon) M. Rahman 

Syed (Shawon) Rahman is an Assistant Professor in the Department of 
Computer Science & Engineering at the University of Hawaii-Hilo (UHH), 
Hawaii, USA and an Adjunct Faculty in the Capella University, Minneapolis, 
USA. He earned his Ph.D. in Software Engineering and MS in Computer 
Science degree from North Dakota State University. Dr. Rahman's research 
interests include Information Assurance & Security, Data Visualization, Data 
Modeling, Web Accessibility, and Software Testing & Quality Assurance. His 
research focus also includes Software Engineering Education, Search Engine 
Optimization, and Lightweight Software Development Methodologies such as 
extreme Programming and Test-driven Development. 

Before joining at the UHH in Fall 2009, Dr. Rahman taught the last three 
years in the Dept. of Computer Science and Software Engineering at the 
University of Wisconsin-Platteville as an Assistant Professor. Dr. Rahman is 
always interested in applying Emerging Technologies and Tools in classrooms 
to improve students' learning experience and performance. He has published 
enormous number referred articles and presented his works around the world. 
He is an active member of many professional organizations including ACM, 
ASEE, ASQ, IEEE, and UPE. 



Shannon E. Donahue 

Shannon is a Certified Information Security Manager (CISM) and Certified 
Information Systems Security Professional (CISSP) with more than 10 years of 
experience in information technology and security. She was the Information 
Security Officer at William Rainey Harper College where she was responsible 
for developing and managing the information security program. She was 
responsible for securing data on all systems as well as incident management, 
policy development, and security awareness. 

As a network support analyst and security architect for AT&T, Shannon 
provided tier 3 network support and helped to develop the virus response plan 
for managed service customers. She also was a lead member responsible for 
incident response and disaster recovery plans. In her current position with 
ISACA, Shannon is responsible for managing the security program and for 
serving the needs of the security profession through research projects and 
publications. Shannon has a masters degree in Management & Systems from 
New York University and is currently working on a PhD in Information 
Security at Capella University. 



68 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



Image Retrieval Techniques based on Image Features: 
A State of Art approach for CBIR 

x Mr. Kondekar V. H., 2 Mr. Kolkure V. S., 3 Prof.Kore S.N. 

department of Electronics & Telecommunication Engineering, 
(^Walchand Institute of Technology, Solapur, 

2 Bharat Ratna Indira Gandhi Collage of Engineering, Solapur) Solapur University. 

3 
Walchand College of Engineering, Sangali. Shivaji University. 



Abstract-The purpose of this Paper is to describe our 
research on different feature extraction and matching 
techniques in designing a Content Based Image 
Retrieval (CBIR) system. Due to the enormous 
increase in image database sizes, as well as its vast 
deployment in various applications, the need for CBIR 
development arose. Firstly, this paper outlines a 
description of the primitive feature extraction 
techniques like: texture, colour, and shape. Once these 
features are extracted and used as the basis for a 
similarity check between images, the various matching 
techniques are discussed. Furthermore, the results of 
its performance are illustrated by a detailed example. 

Keyword - CBIR, Feature Vector, Distance metrics, Similarity 
check, similarity matrix, Histogram, Wavelet Transform, variance, 
standard deviation. 

I. Introduction 

As processors become increasingly powerful, and 
memories become increasingly cheaper, the deployment 
of large image databases for a variety of applications 
have now become realisable. Databases of art works, 
satellite and medical imagery have been attracting more 
and more users in various professional fields — for 
example, geography, medicine, architecture, advertising, 
design, fashion, and publishing. Effectively and 
efficiently accessing desired images from large and 
varied image databases is now a necessity 

CBIR or Content Based Image Retrieval is the 
retrieval of images based on visual features such as 
colour, texture and shape. Reasons for its development 
are that in many large image databases, traditional 
methods of image indexing have proven to be 
insufficient, laborious, and extremely time consuming. 
These old methods of image indexing, ranging from 
storing an image in the database and associating it with a 
keyword or number, to associating it with a categorized 
description, have become obsolete. This is not CBIR. In 
CBIR, each image that is stored in the database has its 
features extracted and compared to the features of the 
query image. It involves two steps: 

• Feature Extraction: The first step in the process 
is extracting image features to a distinguishable 
extent. 



• Matching: The second step involves matching 
these features to yield a result that is visually 
similar. 

Examples of CBIR applications are: 

• Security Check: Finger print or retina scanning 
for access privileges. 

• Intellectual Property: Trademark image 
registration, where a new candidate mark is 
compared with existing marks to ensure no risk 
of confusing property ownership. 

• Medical Diagnosis: Using CBIR in a medical 
database of medical images to aid diagnosis by 
identifying similar past cases. 

• Crime prevention: Automatic face recognition 
systems, used by police forces. 




Uuery Image 



Image 
Database 



Fig. 1 CBIR (Content Based Image retrieval ): Block Diagram. 

Problem Statement 

The problem involves entering an image as a 
query into a software application that is designed to 
employ CBIR techniques in extracting visual properties, 
and matching them. This is done to retrieve images in 
the database that are visually similar to the query image. 

II. Feature Extraction Techniques. 

2.1. Colour 
2.1.1 Definition 

One of the most important features that make 
possible the recognition of images by humans is colour. 
Colour is a property that depends on the reflection of 
light to the eye and the processing of that information in 
the brain. We use colour everyday to tell the difference 
between objects, places, and the time of day . Usually 
colours are defined in three dimensional colour spaces. 



69 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



These could either be RGB (Red, Green, and Blue), 
HSV (Hue, Saturation, and Value) or HSB (Hue, 
Saturation, and Brightness). The last two are dependent 
on the human perception of hue, saturation, and 
brightness. 

Most image formats such as JPEG, BMP, GIF, 
use the RGB colour space to store information. The 
RGB colour space is defined as a unit cube with red, 
green, and blue axes. Thus, a vector with three co- 
ordinates represents the colour in this space. When all 
three coordinates are set to zero the colour perceived is 
black. When all three coordinates are set to 1 the colour 
perceived is white. The other colour spaces operate in a 
similar fashion but with a different perception. 

2.1.2 Methods of Representation 

The main method of representing colour 
information of images in CBIR systems is through 
colour histograms. A colour histogram is a type of bar 
graph, where each bar represents a particular colour of 
the colour space being used. In MatLab for example you 
can get a colour histogram of an image in the RGB or 
HSV colour space. The bars in a colour histogram are 
referred to as bins and they represent the x-axis. The 
number of bins depends on the number of colours there 
are in an image. The y-axis denotes the number of pixels 
there are in each bin. In other words how many pixels in 
an image are of a particular colour. 

An example of a colour histogram in the HSV 
colour space can be seen with the image shown. To view 
a histogram numerically one has to look at the colour 
map or the numeric representation of each bin. As one 
can see from the colour map each row represents the 
colour of a bin. The row is composed of the three 
coordinates of the colour space. The first coordinate 
represents hue, the second saturation, and the third, 
value, thereby giving HSV. The percentages of each of 
these coordinates are what make up the colour of a bin. 
Also one can see the corresponding pixel numbers for 
each bin, which are denoted by the blue lines in the 
histogram. 

Quantization in terms of colour histograms 
refers to the process of reducing the number of bins by 
taking colours that are very similar to each other and 
putting them in the same bin. By default the maximum 
number of bins one can obtain using the histogram 
function in MatLab is 256. For the purpose of saving 
time when trying to compare colour histograms, one can 
quantize the number of bins. Obviously quantization 
reduces the information regarding the content of images 



but as was mentioned this is the trade-off when one 
wants to reduce processing time. 

There are two types of colour histograms, Global 
colour histograms (GCHs) and Local colour histograms 
(LCHs). A GCH represents one whole image with a 
single colour histogram. An LCH divides an image into 
fixed blocks and takes the colour histogram of each of 
those blocks. LCHs contain more information about an 
image but are computationally expensive when 
comparing images. "The GCH is the traditional method 
for colour based image retrieval. However, it does not 
include information concerning the colour distribution of 
the regions" of an image. Thus when comparing GCHs 
one might not always get a proper result in terms of 
similarity of images. 



Insert Tools Desktop 



, fc «t et 3"? ® « n IS 



3000 
2000 
1000 





Fig. 2: Histogram of two Image of same class. 




Edit View Insert Tools Desktop Window Help 



- ^ X. 



, I k | @i ei r? ® v- □ id | 





Fig. 3: Histogram of two Images of two different classes. 



desktop Window Help 



I- II □ Ileal 



. | t* | <~k <^k <r> i§> : I We J I □ ED I i r— l 




Fig. 4: Histogram of two Image of same class. 



70 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



2. 2 TEXTURE 

Texture is an intuitive concept that describes 
properties like smoothness, coarseness, and regularity of 
a region. Texture is an important element to human 
vision, it provides cues to scene depth and surface 
orientation. In the next sections, Intensity-based texture 
will be described which has been the topic of 
investigation for many years and has proven useful. For 
example, the black and white television proves the 
usability of Intensity-based texture: people are able to 
see 3D in a 2D black and white screen. So, it seems 
important to look at Intensity-based textures before 
looking at colourful textures because the techniques 
used by Intensity-based textures can probably be 
expanded to colour-texture. 

2.2.1 Texture Definition 

Texture is that innate property of all surfaces 
that describes visual patterns, each having properties of 
homogeneity. It contains important information about 
the structural arrangement of the surface, such as; clouds, 
leaves, bricks, fabric, etc. It also describes the 
relationship of the surface to the surrounding 
environment. In short, it is a feature that describes the 
distinctive physical composition of a surface. 
Texture may be defined as a local arrangement of image 
irradiances projected from a surface patch of 
perceptually homogeneous irradiances. Texture regions 
give different interpretations at different distances and at 
different degrees of visual attention. At a standard 
distance with normal attention, it gives the notion of 
macro-regularity that is characteristic of the particular 
texture. When viewed closely and attentively, 
homogeneous regions and edges, sometimes constituting 
texels, are noticeable. 




(a) Clouds 



(b) Bricks 

Figure 5: Examples of Texture 




(c) Rocks 



2.2. 3Texture properties include: 



Moment 


Expression. 


Description 


Mean 


L-l 
i=0 


To measure 
the average 
intensity 


Standard 
deviation 


1 


To measure 
the average 
contrast 


a = ^jU 2 (Z) =^a z 


Smoothness 




To measure 
the relative 
smoothness of 
the intensity in 
a region. 


Third 
moment 


L-l 

M3 =^( z i- m ) 3 P( z i) 
i-0 


To measure 
the skewness 
of a histogram 


Uniformity 


L-l 
1=0 


To measure 
the uniformity 


Entropy 


L-l 

e = -Y J P(Z i )hg 2 P(Z i ) 
i=0 


To measure 

the 

randomness 



• Coarseness 

• Contrast 

• Directionality 



Table: 1 statistical parameters 

• Line-likeness 

• Regularity 

• Roughness 

Texture is one of the most important defining 
features of an image. It is characterized by the spatial 
distribution of grey levels in a neighbourhood. In order 
to capture the spatial dependence of gray-level values, 
which contribute to the perception of texture, a two- 
dimensional dependence texture analysis matrix is taken 
into consideration. 

This two-dimensional matrix is obtained by decoding 
the image file; jpeg, bmp, etc. 

2.2.4 Texture Features: 

Methods of Representation 

There are three principal approaches used to 
describe texture; statistical, structural and spectral. 



71 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



• Statistical techniques characterize textures 
using the statistical properties of the grey 
levels of the points/pixels comprising a 
surface image. Typically, these properties 
are computed using: the grey level co- 
occurrence matrix of the surface, or the 
wavelet transformation of the surface. 

• Structural techniques characterize textures 
as being composed of simple primitive 
structures called "texels" (or texture 
elements). These are arranged regularly on a 
surface according to some surface 
arrangement rules. 

• Spectral techniques are based on properties 
of the Fourier spectrum and describe 
global .periodicity of the grey levels of a 
surface by Identifying high-energy peaks in 
the Fourier spectrum. 

For optimum classification purposes, what concern are 
the statistical techniques of characterization. This is 
because it is these techniques that result in computing 
texture properties. The most popular statistical 
representations of texture are: 

• Co-occurrence Matrix 

• Tamura Texture 

• Wavelet Transform. 

2.3. SHAPE 

2.3. 1 Definition of Shape: 

Shape may be defined as the characteristic surface 
configuration of an object; an outline or contour. It 
permits an object to be distinguished from its 
surroundings by its outline. Shape representations can be 
generally divided into two categories: 

Boundary-based, and Region-based. 



1 '•••■ 



.!*•, 



<)■:■■•■•;::• 

I * • 



Figure 6: Boundary -based & Region-based 

Boundary-based shape representation only uses 
the outer boundary of the shape. This is done by 
describing the considered region using its external 
characteristics; i.e., the pixels along the object boundary. 
Region-based shape representation uses the entire shape 
region by describing the considered region using its 



internal characteristics; i.e., the pixels contained in that 
region. 

2.3.2 Methods of Representation 

For representing shape features mathematically, we have: 

Boundary-based: 

Polygonal Models, boundary partitioning 

Fourier Descriptors: 

Splines, higher order constructs 

Curvature Models 

Region-based: 

Super quadrics 

Fourier Descriptors 

Implicit Polynomials 

Blum's skeletons 

The most successful representations for shape 
categories are Fourier Descriptor and Moment Invariants: 
The main idea of Fourier Descriptor is to use the Fourier 
transformed boundary as the shape feature. 
The main idea of Moment invariants is to use region- 
based moments, which are invariant to transformations 
as the shape feature. 

iilImage Feature Matching 

3.1 Similarity Distance measures 

CBIR employs low level image features such as 
color, shape or texture to achieve objective and 
automatic indexing, in contrast to subjective and manual 
indexing in traditional image indexing. For contend 
based image retrieval, the image feature extracted is 
usually an N-dimensional feature vector which can be 
regarded as a point in a N-dimensional space. Once 
images are indexed into the database using the extracted 
feature vectors, the retrieval of images is essentially the 
determination of similarity between the features of query 
image and the features of target images in database, 
which is essentially the determination of distance 
between the feature vectors representing the images. The 
desirable distance measure should reflect human 
perception. That is to say, perceptually similar images 
should have smaller distance between them and 
perceptually different images should have larger 
distance between them. Therefore, for a given shape 
feature, the higher the retrieval accuracy, the better the 
distance measure. Various distance measures have been 
exploited in image retrieval, are discussed below. 



72 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



3.2 SIMILARITY MEASUREMENTS 

A similarity measurement is normally defined as 
a metric distance. In this section different similarity 
measurements are described in details. 

Minkowski-form distance metrics 

The Minkowski metric between two point's p = (xl, yl) 

and q = (x2, y2) is defined as: 

d k (P,Q) = (*i-yi\ K +\x 2 -y 2 \ K ^ (1) 

The histogram distance 

The histogram distance, calculated per bin m, between a 
query image q and a target image t is denoted as: 

Af-l 
D t (q,t)= V J h q [m] - h t [m] I 

m=0 (2) 

Where M is the total number of bins, hq is the 

normalized query histogram, and ht is the normalized 

target histogram. We recognize Di(q; t) as the 
Minkowski form metric with k=l. 

Histogram the Euclidean distance 

The Euclidean distance is a Minkowski form with k=2: 



M-\ 



D e (q,t)-. 



^M-\ 2 

(hJm]-h t [m]) 



(3) 



The distances (i.e., calculated Minkowski-form distance 
measures) only take account for the correspondence 
between each histogram bin. and do not make use of 
information across bins. This issue has been recognized 
in histogram matching. As a result, quadratic distance is 
proposed to take similarity across dimensions into 
account. It has been reported to provide more desirable 
result than only matching between similar bins of the 
color histograms. However, since the histogram 
quadratic distance computes the cross similarity between 
colours, it is computationally expensive. 

Histogram quadratic distance 

The quadratic-form distance between two feature vectors 
q and t is given by: 

D e (q,t) = (h q -h t ) T A(h q -h i ) 



(4) 



Where A = [aij ] is a similarity matrix, aij denotes the 
similarity between elements with indexes i and j. Please 
note, that h q and ht are denoted as vectors. 

In order to determine the intersection similarity 
(S) we adapt Equation to give: 



s q4 =Y}-%^-h^ 



ra=0 



(5) 



Minkowski-form distance metrics compare only the 
same bins between color histograms and are defined as: 



d{Q,I) = Y^ i JH Q [i\-H I m\ r 



(6) 



Where Q and I are two images ,N , is the number of bins 
in the color histogram (for each image we reduce the 
colours to N , in the RGB color space, so each color 
histogram has N bins, HQ[i] is the value of bin i in 
color histogram HQ , which represents the image Q , and 
HI[i] is the value of bin i in color histogram HI , which 
represents the image I. 

When r=l, the Minkoski-form distance metric 
becomes LI. When r=2 , the Minkoski-form distance 
metric becomes the Euclidean distance. In fact, this 
Euclidean distance can be treated as the spatial distance 
in a multi-dimensional space. 



m. 



y*g@^ 


— — 


-h. 



fal (h) 

Figure 7 : (a) Minkowski-form distance metrics compare only 
similar bins between histograms. 

(b) Quadratic -form distance metrics compare multiple bins 
between histograms using similarity matrix A = [aij ]. 

Quadratic-form distance metrics 

The quadratic distance, also called cross distance, is 
used in the QBIC-system. This method considers the 
cross-correlation between histogram bins based on the 
perceptual similarity of the colours represented by the 
bins. The set of correlation values is represented in a 
similarity matrix. 

Quadratic-form distance metric compares not only the 
same bins but multiple bins between color histograms 
and is defined as: 



d(QJ) = (H Q -H I ) t A(H Q -H I ) 



(7) 



Where Q and I are two images HQ is the color 
histogram of image Q, HI is the color histogram of 
image I, IAI=[ai,j] is a NXN 54 , matrix, N is the 
number of bins in the color histograms, and aij denotes 
the similarity between colours i and j.The similarity 
matrix is obtained through a complex algorithm: 



73 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



\-vf% C0 <h}-Si™w 



+ (c sini 



(hjs^hP 2 



S 

where (hq, sq, vq) and (hi si , vi) represent hue, 
saturation, and value components for two colours 
indexed by two histogram bins. Quadratic-form distance 
metrics overcome a shortcoming of the Minkoski-form 
distance metrics in that the latter assumes that bins in 
color histograms are totally unrelated, while the former 
does not. 

Chebyshev Distance 

This distance calculation metric is named after Pafnuty 
Lvovich Chebyshev, it is also known as chessboard 
distance, the equation is defined as 
d = max(l xl - xl I, I yl - yl I) /ox 

Bray Curtis distance 



dtj 



ljx ik - 



k=\ 



Lu Xtk Xjk 



(9) 



Manhattan Distance 

Manhattan distance is also known as Taxicab distance. 
This is because it comes from the fact that it represents 
the shortest distance a car will drive in a city laid out in 
square blocks. For example, in the plane, the Manhattan 
distance between the point PI with coordinates (xl, yl) 
and the point P2 at (x2, y2) is 



l x l -x 2 ul y l -y 2 l 

Hamming Distance 
HD = ±-±X j (XOR)Y j 



N 



(10) 



(11) 



two images although in some cases, different users can 
give different opinions. Two evaluation measures were 
used here to evaluate the effectiveness of the image 
search engine system. 

The first measure is Recall. It is a measure of 

the ability of a system to present all relevant items. The 

equation for calculating recall is given below: 

number of relevant items retrieved 



Recall = 



number of relevant items in collection 

(12) 

The second measure is Precision. It is a measure of the 
ability of a system to present only relevant items. The 
equation for calculating precision is given below, 
number of relevant items retrieved 



Precision 



total number of items retrieved 



(13) 



A retrieval score 



A retrieval score can be computed according to 
the following evaluation criterion: for each query, the 
system returns the 'x' closest images to the query, 
including the query image itself (as the distance from the 
query image to itself is zero). The number of 
mismatches can be computed as the number of images 
returned that belong to a class different than that of the 
query image, in addition to the number of images that 
belong to the query image class, but that have not been 
returned by the system. The retrieval score for one class 
can be then computed as 

f mismatches \ 



Re trivalScore = 100 x 



1- 



% 



(14) 



IV System performance Evaluation. 

Precision and Recall: 

Testing the effectiveness of the image search engine is 
about testing how well can the search engine retrieve 
similar images to the query image and how well the 
system prevents the returned results that are not relevant 
to the source at all in the user point of view. The big 
question here is how we know that which image is 
relevant. Determining whether or not two images are 
similar is purely up to the user's perception. Human 
perceptions can easily recognise the similarity between 



V Conclusions 

CBIR at present is still topic of research interest. 
Different features are used for retrieval of images such 
as Image colour quadratic distance for image histogram, 
Image Euclidian distance for image wavelet transform; 
image Hamming Distance. And corresponding retrieval 
Recall and Precision parameters are calculated for each 
feature. For as to increase retrieval efficiency 
combination of these features should be used instead of 
using a single feature for image retrieval. 

The retrieval efficiency and timing performance 
can be further increased if the image collection is trained 



74 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



(pre-processed) and grouped using supervised learning 
such as classification or unsupervised learning such as 
clustering. With that, the image with high similarities in 
the feature space will be group together and result a 
smaller search space. This will greatly enhance the 
search time and precision. 

Future work 



Classification and Clustering: 

The retrieval efficiency and timing performance 
can be further increased if the image collection is trained 
(pre-processed) and grouped using supervised learning 
such as classification or unsupervised learning such as 
clustering. With that, the image with high similarities in 
the feature space will be group together and result a 
smaller search space. This will greatly enhance the 
search time and precision. 



Acknowledgements 

I would like to take this opportunity to thank my 
Guide, Prof. S. N. Kore ,Head of electronics 
Engineering Department Walchand College of 
Engineering, Sangli for his valuable guidance, constant 
inspiration, advice, and encouragement throughout the 
course of this work. I am also thankful to him for his 
timely and helpful advices to understand the nature of 
topic and careful reviews on the subject. 

Also, I would like to our Principal Dr. S. A. 
Halkude and Head Dr. S. K. Dixit for their continuous 
encouragement. 

References 

1. Lenina Birgale, Manesh Kokare and Dharmpal 
Doye, "Colour and Texture Features for Content 
Based Image Retrieval", Proceedings of the 
International Conference on Computer Graphics, 
Imaging and Visualisation (CGIV'06) 0-7695- 
2606-3/06 $20.00 © 2006IEEE. 

2. ZHEN-HUA ZHANG, YONG QUAN, WEN- 
HUI LI, WU GUO, "A NEW CONTENT- 
BASED IMAGE RETRIEVAL", Proceedings of 
the Fifth International Conference on Machine 
Learning and Cybernetics, Dalian, 13-16 August 
2006. 

3. P.S.Hiremath, S. Shivashankar, and Jagadeesh 
Pujari, "WAVELET BASED FEATURES FOR 
COLOR TEXTURE CLASSIFICATION WITH 
APPLICATION TO CBIR", IJCSNS 



International Journal of Computer Science and 
Network Security, VOL.6 No.9A, September 
2006. 

4. P.S.Hiremath, S. Shivashankar, " WAVELET 
BASED FEATURES FOR TEXTURE 
CLASSIFICATION", GVIP Journal, Volume 6, 
Issue 3, December, 2006 

5. Subrahmanyam Murala, Anil Balaji Gonde, R. P. 
Maheshwari," Color and Texture Features for 
Image Indexing and Retrieval", 2009 IEEE 
International Advance Computing Conference 
(IACC 2009) Patiala, India, 6-7 March 2009. 

6. L. Kotoulas and I. Andreadis, "Colour 
histogram content-based image retrieval and 
hardware implementation", IEE Proc. -Circuits 
Devices Syst., Vol. 150, No. 5, October 2003. 

7. T F. Ang , 2 Y K. Cheong, 3 L Y. Por, 4 K K. 
Phang, "Colour-Based Image Retrieval Using 
Global Colour Histogram and Harbin", 
Multimedia Cyberscape Journal, Volume 5, 
Number 1, Year 2007. 

8. Igor Marino vi and Igor Ftirstner Manufaktura 
d.o.o., Subotica, Serbia "Content-based Image 
Retrieval ", l-4244-2407-8/08/$20.00 ©2008 
IEEE. 

9. Dengsheng Zhang and Guojun Lu 
"EVALUATION " OF SIMILARITY 
MEASUREMENT FOR IMAGE 
RETRIEVAL", l-4240-2407-8/08/$25.00 
©2004 IEEE 

10. Greg Pass Ramin Zabih*, "Histogram 
Refinement for Content-Based Image Retrieval", 
0-8186-7620-5/96 $5.00 1996 IEEE. 

11. Gonzalez, R.C., Woods, R.E, "Digital Image 
Processing" 2nd Ed., Prentice Hall 

12. Robi Polikar, "The Engineers Ultimate Guide 
To Wavelet Analysis", [Online Document], 
Available: 

http://engineering.rowan.edu/~polikarAVAVEL 
ETS/WTtutorial.html 

13. VisTex, "color image database", at 
http://www.white.media.mit.edu/vismod/imager 
y/ vision 

14. Texture, MIT media Lab. 

15. MATLAB Manual 



75 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 




Mr. Kondekar Vipul H. received 
Master of Engineering in 

Electronics Engineering with 
specialization in Computer 

engineering from Walchand College 
of Engineering, Sangali Maharastra- 
India He is working as Lecturer for 
Last 6 Years in Electronics and 
Telecommunication Engineering 
Department at Walchand institute of Technology, 
Solapur University, Solapur, maharastra-India. His 
current research interest areas are Image Processing and 
Microcontroller based system design. He has authored 
and co-authored more than 10 technical papers 
published in various prestigious national/international 
journals and referred conference,symposium, workshop 
proceedings. 




Mr. Kolkure Vijaykumar S. has 
completed Master of Engineering 
(ME) in Electronics Engineering 
with specialization in Computer 
engineering from Walchand College 
of Engineering; Sangli Maharastra- 
India .He has 3 years' experience of 
working in IT industry. Presently he 
is working as Lecturer for Last 1 Years in Electronics 
and Telecommunication Engineering Department at 
Bharat Ratna Indira Gandhi College of Engineering, 
Solapur University, Solapur, Maharastra-India. His 
current research interest areas are Image Processing and 
Embedded systems. He has more than 8 technical 
papers published in various prestigious 
national/international journals and referred 
conference/symposium/workshop proceedings. 



76 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



AHB Compatible DDR SDRAM Controller IP Core 

for ARM BASED SOC 

^r.R.Shashikumar, 2 C.N. Vijay Kumar, 3 M.Nagendrakumar, 4 C.S.Hemanthkumar 

Professor Asst.prof Asst.Prof Sr.Lecturer 

ECE dept, SJCIT, Chikkaballapur, Karnataka, India JVIT, Bidadi 



Abstract — DDR SDRAM is similar in function to the regular 
SDRAM but doubles the bandwidth of the memory by 
transferring data on both edges of the clock cycles.DDR 
SDRAM most commonly used in various embedded application 
like networking, image/video processing, Laptops ete. Now a 
day's many applications needs more and more cheap and fast 
memory. Especially in the field of signal processing, requires 
significant amount of memory. The most used type of dynamic 
memory for that purpose is DDR SDRAM. For FPGA design 
the IC manufacturers are providing commercial memory 
controller IP cores working only on their products. Main 
disadvantage is the lack of memory access optimization for 
random memory access patterns. The 'data path' part of those 
controllers can be used free of charge. This work propose an 
architecture of a DDR SDRAM controller, which takes 
advantage of those available and well tested data paths and can 
be used for any FPGA device or ASIC design.[5]. In most of the 
SOC design, DDR SDRAM is commonly used. ARM processor 
is widely used in SOC's; so that we focused to implement AHB 
compatible DDR SDRAM controller suitable for ARM based 
SOC design. 

Keywords-AHB; DDR SDRAM; Verilog;IP core; 



1 INTRODUCTION 

The DDR SDRAM is a high-speed CMOS, dynamic 
random-access memory. It is internally configured as a quad 
bank DRAM. The DDR SDRAM uses double data rate 
architecture to achieve high-speed operation. The double data 
rate architecture is essentially 2nprefetch architecture with an 
interface designed to transfer two data words per clock cycle 
at the I/O pins. A single read or write access for the DDR 
SDRAM effectively consists of a single 2n-b\t wide, one- 
clock-cycle data transfer at the internal DRAM core and two 
corresponding /7-bit wide, one-half clock-cycle data transfers 
at the I/O pins. A bidirectional data strobe (DQS) is 
transmitted externally, along with data, for use in data 
capture at the receiver. DQS is a strobe transmitted by the 
DDR SDRAM during READs and by the memory controller 
during WRITEs. DQS is edge-aligned with data for READs 
and center-aligned with data for WRITEs. 

Read and write accesses to the DDR SDRAM are 
burst oriented; accesses start at a selected location and 
continue for a programmed number of locations in a 
programmed sequence. Accesses begin with the registration 
of an ACTIVE command, which is then followed by a READ 
or WRITE command. The address bits registered coincident 



with the ACTIVE command are used to select the bank and 
row to be accessed. The address bits registered coincident 
with the READ or WRITE command are used to select the 
bank and the starting column location for the burst access. 

The DDR SDRAM provides for programmable 
READ or WRITE burst lengths of 2, 4, or 8 locations. An 
auto precharge function may be enabled to provide a self 
timed row precharge that is initiated at the end of the burst 
access This model has implemented in RTL by Verilog. The 
focus of this work is to implement behavioral model of DDR 
SDRAM and also implemented on the Xilinx Spartan series 
FPGA.The Top level model is as shown in Fig.l. The core 
contains mainly two parts, AHB Slave and DDR SDRAM 
controller. 



TOP MODULE 



A 






H 






B 






S 
L 


A N 

M V 


Controller 
Module 


A 






V 






E 







Signal 
Module 



n 



Data path 
Module 



<^^> 



DDR 
SDRAM 



Figure . 1 Top Module 

2. FUNCTIONAL BLOCK DIAGRAM 

The functional block diagram of the DDR controller is shown 
in Fig.2. It consists of four modules, AHB Salve, the main 
control module, the signal generation module and the data 
path module. The AHB slave normally connected to the AHB 
bus Arbiter in the design. AHB master sends data to the AHB 
slave based on that protocol. Burst mode read and writes and 
split transaction read and writes transactions supported. AHB 
slave complies with the processor interface protocol of ARM 
processor. The main control module has two state machines 
and a refresh counter, which generates proper istate and 
estate outputs according to the system interface control 
signals. The signal generation module generates the address 
and command signals required for DDR based on istate and 
estate. The data path module performs the data latching and 
dispatching of the data between the Processor and DDR. 



77 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



The DDR SDRAM provides for programmable 
READ or WRITE burst lengths of 2, 4, or 8 locations. An 
AUTO PRECHARGE function may be enabled to provide a 
self-timed row precharge that is initiated at the end of the 
burst access. As with standard SDR SDRAMs, the pipelined, 
multibank architecture of DDR SDRAMs allows for 
concurrent operation, thereby providing high effective 
bandwidth by hiding row precharge and activation time. An 
auto refresh mode is provided, along with a power-saving 
power-down mode. 



sys.CLK ■ 



sysjneset 

ipJNtJm ■* 



E*5_f_wn - 



sys_ref_ack 
sys_cyc_ena ■*■[■ 



:■:■- 



[ffi^ ■> 



-Iiyncil 
Generation 






•^ ddf_add 



Figure .2 Functional Block Diagram 

3. FUNCTIONAL DESCRIPTION 
3.1. Initialization 

DDR SDRAMs must be powered up and initialized 
in a predefined manner. Operational procedures other than 
those specified may result in undefined operation. 
DDR SDRAM requires a 200 |is delay prior to applying an 
executable command [7]. Once the 200|is delay has been 
satisfied, a DESELECT or NOP command should be applied. 
Following the NOP command, a PRECHARGE ALL 
command should be applied. Next a LOAD MODE 
REGISTER command should be issued. A PRECHARGE 
ALL command should then be applied, placing the device in 
the all banks idle state. Once in the idle state, two AUTO 
REFRESH cycles must be performed (tRFC must be 
satisfied.) Additionally, a LOAD MODE REGISTER 
command for the mode register is issued [16]. 

3 .2 Register Definition 

The mode register is used to define the specific 
mode of operation of the DDR SDRAM. This definition 
includes the selection of a burst length, a burst type, a CAS 



latency and an operating mode. The mode register is 
programmed via the MODE REGISTER SET command 
(with BAO = and BA1 = 0) and will retain the stored 
information until it is programmed again or the device loses 
power (except for bit A8, which is self-clearing). 
Reprogramming the mode register will not alter the contents 
of the memory, provided it is performed correctly. The mode 
register must be loaded (reloaded) when all banks are idle 
and no bursts are in progress, and the controller must wait the 
specified time before initiating the subsequent operation. 
Violating either of these requirements will result in 
unspecified operation. Mode register bits A0-A2 specify the 
burst length, A3 specifies the type of burst (sequential or 
interleaved), A4-A6 specifies the CAS latency, and A7-A12 
specifies the operating mode. Fig. 3 shows the description of 
mode register. 

W.1 BAJ0A12 A11 A10 A3 AB 47 AS A5 U A3 AZ Al AD Address Ele 

T .VM* VM * T .M. T . T .V 

/3yii/li»vfrn^e /H/7 /6/B Si /a 7z/i/q/ Mode p^^ \u;. 

| Q- | J | Ope rj ting Mnde | CAS Latency , ET Bust Length _ .' 



MM «nd Ml ] PA1 md BJIflQ 
murt be "0. 0T -Lo select the 
ba^e mode regrter fre. -tire- 
extended rriodc- re-girter|. 



M1^ 
G 



Mi Ml Mil 


9lt£ LengtTi 


n n n 


Reserved 


O O 1 


2 


O 1 O 


i ; 


O 1 1 


& 


i ■:■ ■:■ 


Reserved 


1 O 1 


Reso-ved 


1 1 


Reserved 


1 1 1 


Reserved 



Billet 1'fPCr 



vrqu:-nciJ 



Interleaved 



M6 M5 U4 



Cfl.-.-LjU-n<,y 



Qyaraltnfl Mode- 



N or ma I Cptntion 



Normal Operjtion/He^et DLL 



.in :ii---- x.xt: T'.ttyc-d 



Figure.3 Mode register definition 

3.3 DDR Commands 

Table 1 presents the commands issued by the 
controller. These commands are passed to the memory using 
the following control signals[7]: 

• Row Address Select (RAS) 

• Column Address Select (CAS) 

• Write Enable (WE) 

• Clock Enable (CKE) (always held High after device 
configuration) 

• Chip Select (CS) (always held Low during device 
operation) 



78 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



Signal No. 


Function 


RAS 


CAS 


WE 


1 


Load Mode Register 


L 


L 


L 


2 


Auto Refresh 


L 


L 


H 


3 


Precharged 


L 


H 


L 


4 


Select Bank Activate Row 


L 


H 


H 


5 


Write Command 


H 


L 


L 


6 


Read Command 


H 


L 


H 




Signal No. 


Function 


RAS 


GAS 


WE 


7 


No Operation (NOP) 


H 


H 


H 



Table 1 DDR SDRAM Commands 
3.3.1 Command Functions 

The fallowing commands are used in DDR SDRAM 
controller core. 

• Mode Register : 

The Mode register is used to define the specific 
mode of DDR SDRAM operation, including the selection of 
burst length, burst type, CAS latency, and operating mode. 

• Auto Refresh : 

The REFRESH command instructs the controller to 
perform an AUTO REFRESH command to the SDRAM. The 
controller will acknowledge the REFRESH command with 
ACK. DDR SDRAM is somewhat similar to regular 
SDRAM. Both will break the RAM into smaller chunks for 
simultaneous, synchronized request-and-reply access. In 
addition, both types of memory can be packaged in DIMM 
modules. However, DDR SDRAM will perform the 
alternating request-and-reply rhythm on both the rise and fall 
of the clock cycle. This method effectively doubles the 
bandwidth available and increases the speed the system can 
access data in memory 

• Precharge: 

The PRECHARGE command is used to deactivate 
the open row in a particular bank. The bank is available for 
subsequent row activation for a specified time (tRP) after the 
PRECHARGE command is issued. Input A10 determines 
whether one or all banks are precharged. 

• ACTIVE Command: 

The ACTIVE command activates a row in a bank, 
allowing any READ or WRITE commands to be issued to a 
bank in the memory array. After a row has been opened, 
READ or WRITE commands can be issued to that row, 
subject to the tRCD specification. When the controller 
detects an incoming address that refers to a row in a bank 
other than the currently opened row, the controller issues an 
address conflict signal. A PRECHARGE command is also 
issued by the controller to deactivate the open row. The 
controller also issues another ACTIVE command to the new 
row. 

• READ Command: 

The READ command is used to initiate a burst read 
access to an active row. The value on BAO and BA1 selects 
the bank address. The address inputs provided on A0 - Ai 
select the starting column location. After the read burst is 



over, the row is still available for subsequent access until it is 
precharged. 

• WRITE Command: 

The WRITE command is used to initiate a burst 
access to an active row. The value on BAO and BA1 selects 
the bank address, while the value on address inputs A0 - Ai 
selects the starting column location in the active row. The 
value of Write Latency is equal to one clock cycle. 

4 .DDR SDRAM Controller Block 

The controller block mainly consists of two FSMs 

• Initial FSM 

• Command FSM 

It generates the required commands for initializing 
the DDR SDRAM. The block diagram for the DDR SDRAM 
controller is as shown in Fig.4. 



Sys Dly 200us ■ 

Sys C Ik ■ 
Sys Reset ■ 

Sys_Adsn ■ 

Sys R Wn ■ 

Sys Ref Req . 



Initial FSM 



Command FSM 



— ► SysInitDone 
-+ c State 



-+ Sys Ref Ack 



-► Sys Cyc End 



Figure 4 Block Diagram of Controller Block 

4.1 DDR SDRAM Initial FSM 

Before normal memory accesses can be performed, 
DDR needs to be initialized by a sequence of commands. The 
INITFSM state machine handles this initialization. Fig. 5 
shows the state diagram of the INITFSM state machine. 
During reset, the INITFSM is forced to the ilDLE state. 
After reset, the sys_dly_200US signal will be sampled to 
determine if the 200 |is power/clock stabilization delay is 
completed. After the power/clock stabilization is complete, 
the DDR initialization sequence will begin and the 
INITFSM will switch from ilDLE to i_NOP state and in 
the next clock to IPRE. 

The initialization starts with the PRECHARGE ALL 
command. Next a LOAD MODE REGISTER command will 
be applied for the extended mode register to enable the DLL 
inside DDR, followed by another LOAD MODE REGISTER 
command to the mode register to reset the DLL. Then a 
PRECHAGE command will be applied to make all banks in 
the device to idle state. Then two, AUTO REFRESH 
commands, and then the LOAD MODE REGISTER 
command to configure DDR to a specific mode of operation. 
After issuing the LOAD MODE REGISTER command and 
the tMRD timing delay is satisfied, INITFSM goes to 
iready state and remains there for the normal memory 



79 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



access cycles unless reset is asserted. Also, signal 
sysinitdone is set to high to indicate the DDR initialization 
is completed. The i_PRE, i_ARl, i_AR2, iEMRS and 
i_MRS states are used for issuing DDR commands. The 
LOAD MODE REGISTER command configures the DDR by 
loading data into the mode register through the address bus. 
The data present on the address bus (ddradd) during the 
LOAD MODE REGISTER command is loaded to the mode 
register. The mode register contents specify the burst length, 
burst type, CAS latency, etc. A PRECHARGE/AUTO 
PRECHARGE command moves all banks to idle state. As 
long as all banks of the DDR are in idle state, mode register 
can be reloaded with different value thereby changing the 
mode of operation. However, in most applications the mode 
register value will not be changed after initialization. This 
design assumes the mode register stays the same after 

initialization. 



4.1.2 Different states of Initial FSM 

a) Idle: 

When reset is applied the initial fsm is forced to 
IDLE state irrespective of which state it is actually in when 
system is in idle it remains idle without performing any 
operations. 

b) No Operation: 

The NO OPERATION (NOP) command is used to 
instruct the selected DDR SDRAM to perform a NOP (CS# 
is LOW with RAS#, CAS#, and WE# are HIGH). This 
prevents unwanted commands from being registered during 
idle or wait states. Operations already in progress are not 
affected. 



4.1.1 Initial FSM State Diagram: 




Figure .5 Initial FSM State Diagram 

As mentioned above, certain timing delays (tRP, 
tRFC, tMRD) need to be satisfied before another non-NOP 
command can be issued. These SDRAM delays vary from 
speed grade to speed grade and sometimes from vendor to 
vendor. To accommodate this without sacrificing 
performance, the designer needs to modify the HDL code for 
the specific delays and clock period (tCK). According to 
these timing values, the number of clocks the state machine 
will stay at itRP, itRFCl, i_tRFC2, itMRD states will be 
determined after the code is synthesized. In cases where tCK 
is larger than the timing delay, the state machine doesn't 
need to switch to the timing delay states and can go directly 
to the command states[15]. The dashed lines in Fig 5 show 
the possible state switching paths. 



c) Precharge: 

The PRECHARGE command is used to deactivate 
the open row in a particular bank or the open row in all banks 
as shown in Fig. 6. The value on the BAO, BA1 inputs selects 
the bank, and the A10 input selects whether a single bank is 
precharged or whether all banks are precharged. 




t-=-:-z*n*i- " Dont Cdf*" 



Figure .6 Precharge Commands. 



d) Auto Refresh: 

AUTO REFRESH is used during normal operation 
of the DDR SDRAM and is analogous to CAS#-before- 
RAS# (CBR) refresh in DRAMs. This command is 
nonpersistent, so it must be issued each time a refresh is 
required. All banks must be idle before an AUTO REFRESH 
command is issued. 

e) Load Mode register(LMR): 

The mode registers are loaded via inputs AO-An. 
The LOAD MODE REGISTER command can only be issued 
when all banks are idle, and a subsequent executable 
command cannot be issued until tMRD is met. 

f) Read/Write Cycle: 

The Fig. 5 shows the state diagram of CMDFSM 
which handles the read, write and refresh of the SDRAM. 



80 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



The CMDFSM state machine is initialized to 
cidle during reset. After reset, CMDFSM stays in cidle as 
long as sysINITDONE is low which indicates the SDRAM 
initialization sequence is not yet completed. Once the 
initialization is done, sysADSn and sysREFREQ will be 
sampled at the rising edge of every clock cycle. A logic high 
sampled on sysREFREQ will start a SDRAM refresh 
cycle. This is described in the following section. If logic low 
is sampled on both sysREFREQ and sysADSn, a system 
read cycle or system write cycle will begin. These system 
cycles are made up of a sequence of SDRAM commands. 
Initialization: 

Prior to normal operation, DDR SDRAMs must be 
powered up and initialized in a predefined manner. 
Operational procedures, other than those specified, may 
result in undefined operation. 

4.2 DDR SDRAM COMMAND FSM 

The fig. 7 shows the state diagram of CMDFSM, 
which handles read, write and refresh of the DDR. The 
CMDFSM state machine is initialized to cidle during reset. 
After reset, CMDFSM stays in cidle as long as 
sysinitdone is low which indicates th e DDR initialization 
sequence is not yet completed. From this state, a 
READA/WRITEA/REFRESH cycle starts depending upon 
sysadsn/ rd_wr_req_during_ refreq signals as shown in the 
state diagram. All rows are in the "closed" status after the 
DDR initialization. The rows need to be "opened" before 
they can be accessed. However, only one row in the same 
bank can be opened at a time. Since there are four banks, 
there can be at most four rows opened at the same time. If a 
row in one bank is currently opened, it needs to be closed 
before another row in the same bank can be opened. ACTIVE 
command is used to open the rows and PRECHARGE is used 
to close the rows. When issuing the commands for opening or 
closing the rows, both row address and bank address need to 
be provided. 




In this design, the ACTIVE command will be issued for each 
read or write access to open the row. After a tRCD delay is 
satisfied, READA or WRITEA commands will be issued 
with a high ddr_add[10] to enable the AUTO REFRESH for 
closing the row after access. Therefore, the clocks required 
for read/write cycle are fixed and the access can be random 
over the full address range. Read or write is determined by 
the sysrwn status sampled at the rising edge of the clock 
before the tRCD delay is satisfied. If logic high is sampled, 
the state machine switches to cREADA. If a logic low is 
sampled, the state machine switches to cWRITEA. 

For read cycles, the state machine switches from, cREADA 
to ccl for CAS latency, then switches to crate for 
transferring data from DDR to processor. The burst length 
determines the number of clocks the state machine stays in 
crdata state. After the data is transferred, it switches back to 
cidle. 

For write cycles, the state machine switches from 
cWRITEA to cwdata for transferring data from bus master 
to DDR, then switches to ctDAL. After the clock rising 
edge of the last data in the burst sequence, no commands 
other than NOP can be issued to DDR before tDAL is 
satisfied. 

4.2.1 Different states of Command FSM 

a) Refresh Cycle: 

DDR memory needs a periodic refresh to hold the 
data. This periodic refresh is done using AUTO REFRESH 
command. All banks must be idle before an AUTO 
REFRESH command is issued. In this design all banks will 
be in idle state, as every read/write operation uses auto pre 
charge. 

b) Active: 

The ACTIVE command is used to open (or activate) 
a row in a particular bank for a subsequent access, like a read 
or a write, as shown in Fig. 8. The value on the BAO, BA1 
inputs selects the bank, and the address provided on inputs 
AO-An selects the row. 



■ZMJ= 



^^^ 



^gaa»^ 



gsssssr 



.aaaay 



^53555 



JS4444*: 



^•^^^ 



"%£££££ 




Figure .7 Command FSM State Diagram 



EttA E»|--TT 



Figure. 8 Activating a Specific Row in a Specific Bank. 



81 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



c) Read: 

The READ command is used to initiate a burst read 
access to an active row, as shown in Fig. 8. The value on the 
BAO, BA1 inputs selects the bank, and the address provided 
on inputs AO-Az (where A/ is the most significant column 
address bit for a given density and configuration) selects the 
starting column location. 



n-j i km 






■*.¥> 
MUX BAT 

EJjJEJj norn CUE 
EN AP - an±44 auto pr«<|-ur^e^ EM'S AP - clublc- Duto pc«-:l~k] 



Figure .9 Read Command 

d) Write: 

The WRITE command is used to initiate a burst 
write access to an active row as shown in Fig. 10. The value 
on the BAO, BA1 inputs selects the bank, and the address 
provided on inputs AO-A/ (where A/ is the most significant 
column address bit for a given density and configuration) 
selects the starting column location. 



* 



Mi 



^mrrua 
Mote: EH AP - andDltii ajto ppocfiargjB- aid QSAP - dLubla juto pracharga. 

Figure .10 Write Command 

Similar to the FP and EDO DRAM, row address and column 
address are required to pinpoint the memory cell location of 
the SDRAM access. Since SDRAM is composed of four 
banks, bank address needs to be provided as well. 

The SDRAM can be considered as a four by N array 
of rows. All rows are in the "closed" status after the SDRAM 
initialization. The rows need to be "opened" before they can 
be accessed. However, only one row in the same bank can be 
opened at a time. Since there are four banks, there can be at 
most four rows opened at the same time. If a row in one bank 
is currently opened, it must be closed before another row in 



the same bank can be opened. ACTIVE command is used to 
open the rows and PRECHARGE (or the AUTO 
PRECHARGE hidden in the WRITE and READ commands, 
as used in this design) is used to close the rows. When 
issuing the commands for opening or closing the rows, both 
row address and bank address need to be provided. 

For sequential access applications and those with 
page memory management, the proper address assignments 
and the use of the SDRAM pipeline feature deliver the 
highest performance SDRAM controller. However, this type 
of controller design is highly associated with the bus master 
cycle specification and will not fit the general applications. 
Therefore, this SDRAM controller design does not 
implement these custom features to achieve the highest 
performance through these techniques. 

In this design, the ACTIVE command will be issued 
for each read or write access to open the row. After a tRCD 
.delay is satisfied, READ or WRITE commands will be issued 
with a high sdr_A[10] to enable the AUTO REFRESH for 
closing the row after access. So, the clocks required for 
read/write cycle are fixed and the access can be random over 
the full address range. Read or write is determined by the 
sysRWn status sampled at the rising edge of the clock 
before tRCD delay is satisfied. If a logic high is sampled, the 
state machine switches to cREADA. If a logic low is 
sampled, the state machine switches to cWRITEA.For read 
cycles, the state machine switches from cREADA to ccl 
for CAS latency, then switches to crdata for transferring 
data from SDRAM to bus master. The number of clocks the 
state machine stays in crdata state is determined by the burst 
length. After the data is transferred, it switches back to 
cidle.For write cycles, the state machine switches from 
cWRITEA to cwdata for transferring data from bus master 
to SDRAM, then switches to ctDAL. Similar to read, the 
number of clocks the state machine stays in cwdata state is 
determined by the burst length. The time delay tDAL is the 
sum of WRITE recovery time tWR and the AUTO 
PRECHARGE timing delay tRP. After the clock rising edge 
of the last data in the burst sequence, no commands other 
than NOP can be issued to SDRAM before tDAL is 
satisfied.As mentioned in the INITFSM section above, the 
dash lines indicates possible state switching paths when tCK 
period is larger than timing delay spec. 

e) Refresh cycle: 

Similar to the other DRAMs, memory refresh is 
required. A SDRAM refresh request is generated by 
activating sdrREFREQ signal of the controller. The 
sdrREFACK signal will acknowledge the recognition of 
sdrREFREQ and will be active throughout the whole 
refresh cycle. The sdrREFREQ signal must be maintained 
until the sdrREFACK goes active in order to be recognized 
as a refresh cycle. Note that no system read/write access 
cycles are allowed when sdrREFACK is active. All system 
interface cycles will be ignored during this period. The 



82 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



sdrREFREQ signal assertion needs to be removed upon 
receipt of sdrREFACK acknowledge, otherwise another 
refresh cycle will again be performed. 

Upon receipt of sdrREFREQ assertion, the state 
machine CMDFSM enters the c_AR state to issue an AUTO 
REFRESH command to the SDRAM. After tRFC time delay 
is satisfied, CMDFSM returns to cidle. 

5. DATA PATH 

The fig. 11 shows the data path module with inputs 
and outputs as shown in the figure 



Ok. 
Clk2x . 

Reset ■ 
iState . 
estate - 



Data 

Path 

Module 



4 ► Sysdata 64 



4 ► ddid(Ll6 



-► data valid 



Figure . 1 1 .Data Path Module. 

The data flow design between the SDRAM and the 
system interface. The module in this reference design 
interfaces between the SDRAM with 16-bit bidirectional data 
bus and the bus master with 64 -bit bidirectional data bus. The 
user should be able to modify this module to customize to fit 
his/her system bus requirements. The data path module 
performs the data latching and dispatching of the data 
between the processor and DDR. 



pICLK 




n 


IB 


ra 


n 


IS 


T9 


IT 


IE 


ia 


T1D 


— ( 




-^" 


S^A 


TZL 




l*\ 




Y///S///////////A- 


M i 




^ 














:■,■: .-.::!=- 


!/////; 


f////. 












^R Wn 


//)/, 


\S /////////////// ///////, 






































i^qME) 






















/ 


\ 


\ 


'jft D 'ALID 




































rp_CVZ EYD 




L 


/ 




\ 




























■■•■.:■■ 




lomiXR. 


1^ 


>t 




1 








K 




X=Jo™=J[ C 
































HJC.DCE 

:■]■ ::;,■.'.■.' 


l «- 


JL^ 


K*n« 


J_^ 


l^E 


X ™ 


X - 


| ** 


| ** 


| » 


K - 


K »> 


Xj«p 


X™ 


sv z-: 




























































mi 4H| 






\ - 


~r~ 


T^= 


1 
















1[ *. 


mrjpq 






\ - 


~r~ 


a-fidfcU 


























i~ 


T^ 


f 
















1[ -~ 



























Y~\ Dontcara 



Figure .12 Read Cycle Timing Diagram 




■^^E2z2izi 



!/)i//Y//y/y/k//V/}//W>> . 



?;, v.. :: 









HZHC 



n {= 



x± 



i/\/)//k zz. 



U////// ZJU 



TIElJEE. 



xn 



£Z 



^ 



Y/7// J 



]CE 



■ 

X 



Z2 Dm " lci 



Figure. 13 Write Cycle Timing Diagram. 

The signal generation module generates the address and 
command signals required for DDR based on istate and estate 



6. TIMING DIAGRAMS 



6. IMPLEMENTATION AND RESULT 



The fig. 12 and Fig. 13 are the read cycle and write 
cycle timing diagrams of the reference design with the two 
CAS latency cycles and the burst length of four. In the 
example shown in the figures, the read cycle takes 10 clocks 
and the write cycle takes 9 clocks. The state variable cState 
of CMDFSM is also shown in these figures. Note that the 
ACTIVE, READ, WRITE commands are asserted one clock 
after the cACTIVE, cREADA, cWRITEA states 
respectively. The values of the region filed with slashes in the 
system interface input signals of these figures are "don't 
care." For example, signal sysRWn needs to be valid only 
at the clock before CMDFSM switches to the cREADA or 
cWRITEA states. Depending on the values of tRCD and 
tCK, this means the signal sysRWn needs to be valid at 
state c ACTIVE or the last clock of state c tRCD. 



DUT mainly contains two parts, AHB slave and 
DDR SDRAM controller. The core is developed by using 
verilog [2]. In this work we have designed a High speed 
DDR SDRAM Controller with 64-bit data transfer which 
synchronizes the transfer of data between DDR RAM and 
External AHB compatible devices. This can be used in ARM 
based SOC design. The advantages of this controller 
compared to SDR SDRAM is that it synchronizes the data 
transfer, and the data transfer is twice as fast as previous, the 
production cost is also very low. This core is verified by 
using testbench and several testcases, which cover most of 
the functionality of the core. The simulations of specified 
functions were conducted by the Modelsim tool [3]. The 
fig. 14, Fig. 15 and Fig 16 shows the waveforms initial, 
command and data path of the core. Fig. 17 and 18 shows the 
RTL schematic of top module and controller respectively. 



83 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



jtfQ£ £fefto:il*r^ |^#I tiSn^iMiirfc^ttfc fcjfltsrl 









Figure. 14 Waveforms of initial FSM 




Figure. 15 Waveforms of Command FSM 



b 6t Yen lisw font To* Whfcn 



j*b«: »*«^::'*i5:^j*ail«!|t. ,, j!riKri'ii» ! a , T | iii l Esifi[;. Mit*!'* vg-s^ts.-i 1 * 




Figure .16 Waveforms Of Data path 

For verification of core, two BFM's are used, AHB 
BFM at the input side and DDR memory model BFM at the 
output side. Core has been tested for main functionality of 
DDR SDRAM controller. AHB slave tested for both split and 
burst transfers. Our implementation of the design was 



analyzed by using ISE FPGA tool from Xilinx [1]. Designs 
are mapped on to Spartan FPGA. The whole system has been 
placed and routed in to the XC3S500e-5-PQ208 FPGA chip. 
The numbers of slices utilized by the design are 1340 out of 
4656, 28% of the utilization. 




Figure .17.RTL Schematic of Top Module 




Figure .18.RTL Schematic of Controller 

HDL Synthesis report of the core as shown below 
Macro Statistics 

# Adders/Subtractors : 7 



# 3 -bit adder 

#3 2 -bit sub tractor 

# Counters 

# 32-bit down counter 



84 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



# Registers 

# 1 -bit register 

# 16-bit register 
#3 -bit register 

# 32-bit register 
#4-bit register 

# Latches 

# 64-bit latch 

Device utilization summary: 
Selected Device : 3s400tql44-5 



18 
8 

2 
1 
4 

2 
2 
1 



[10] 128Mb DDR SDRAM, Device Specification, Hynix, April 2006 

[11] Altdq & Altdqs Megafunction, User Guide, Altera, March 2005 

[12] PLLs in Stratix II & Stratix II GX Devices, April 2006 

[13] How to Use DDR SDRAM, User's Manual, Document No. E0234E40, 

ALPIDA, September 2005 

[14] Initialization Sequence for DDR SDRAM, Technical Note, TN-46-08, 

Micron. 

[15] DDR SDRAM Controller, Reference Design RD1020, Lattice, 

Semiconductor Corporation, April 2004. 

[16] DDR SDRAM Controller Using Virtex-4 FPGA Devices, Oliver 

Despaux, Application Note, March 27, 2006. 

[19] ALTERA, DDR SDRAM Controller White Paper, Verl.l, 2002, 8. 

[20] Guo Li, Zhang Ying, Li Ning, and Guo Yang, "The Feature of DDR 

SDRAM and the Implementation of DDR SDRAM Controllers via VHDL", 

The Journal of China Universities of Posts and Telecommunications, 2002, 

vol.9, no. 1, pp. 61-65. Proceedings 



482 out of 3584 
468 out of 7168 
8944 out of 7168 

165 

165 out of 97 

71 

3 out of 8 



14% 

6% 

12% 



629% (*) 



37% 



Number of Slices 

Number of Slice Flip Flops 

Number of 4 input LUTs 

Number of IO s 

Number of bonded IOBs 

IOB Flip Flops 

Number of GCLKs 

Timing Summary: 

Speed Grade: -5 

Minimum period: 9.802ns (Maximum Frequency: 

102.021MHz) 

Minimum input arrival time before clock: 5.954ns Maximum 

output required time after clock: 10.704ns 

Maximum combinational path delay: 8.451ns 

DDR2 SDRAM is the second generation of DDR 
SDRAM. DDR2 SDRAM improves on DDR SDRAM by 
using differential signaling and lower voltages to support 
significant performance advantages over DDR SDRAM. 
DDR SDRAM standards are still being developed and 
improved [20]. 

REFERENCES 

[1]. http://www.xilinx.com 

[2]. Samir Palnitkar, Pearson 2 nd edition "Verilog HDL, A Guide to Digital 

Design and Synthesis. 

[3]. http://www.micron.com 

[4]. Bandwidth, Area Efficient and Target Device Independent DDR 

SDRAM Controller by T. Mladenov, F. Mujahid, E. Jung, and D. Har , 

Proceedings Of World Academy Of Science, Engineering And Technology 

Volume 18 December 2006 Issn 1307-6884 

[5]. Application of DDR Controller for High-speed Data Acquisition Board 

by Zude Zhou, Songlin Cheng, and Quan Liu School of Information 

Engineering, Wuhan University of Technology, Proceedings of the First 

International Conference on Innovative Computing, Information and Control 

(ICICIC'06)IEEE 

[6] An All-Digital Delay-Locked Loop for DDR SDRAM Controller 

Applications Ching-Che Chung, Pao-Lung Chen, and Chen-Yi Lee ,Dept. of 

Electronics Engineering / National Chiao Tung University OO ©2006 

IEEEz[7] A Novel Design of DDR-based Data Acquisition Storage Module 

in a Digitizer by Jie Guo, Yibing Shi, Zhigang Wang School of Automation 

Engineering University of Electronic Science and Technology of China May 

2, 2009 at 02:42 from IEEE Xplore. 

[8] Double Data Rate (DDR) SDRAM Specification, JEDEC STANDARD, 

JESD79E, May 2005 

[9] The Love/Hate Relationship with DDR SDRAM Controllers, Graham 

Allan, MOSAID 



AUTHORS PROFILE 

Til Dr. R. Shashikumar is presently working as a Professor 

_ in E & C dept, SJCIT, Chikballapur, 
Karnataka, India. He is having 10 years of 
teaching and 6 years of Industry experience. 
His areas of interest includes ASIC, FPGA, 
HJP'S Network Security. 

f 



[2] Prof.C.N.Vijayakumar me ,miste,mie,miete is presently working 

as a HOD and Assistant Professor in the 
department of Telecommunication engg , 
SJCIT, Chikballapur, Karnataka, India. He is 
having 15 years of teaching experience. His 
areas of interest are Power Electronics, Low 
Power VLSI, ASIC and Control System. 



[3]Mr. M.N.NagendraKumar m .e, miste is working as a 
As st. Professor in the Dept of E & C, SJCIT, Chikballapur, 
Karnataka, India. He is having 13 years of teaching experience. 
His areas of interest are Power electronics, Control System and 
VLSI. 

[4]Mr. C.S.Hemanthkumar M .Tech, miste is working as a Sr.Lecturer in 
the Dept of E&C, JVIT, Bangalore, India. His areas of interest are 
VLSI, Embedded, Signal Processing. 



85 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) InternationalJournal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 



High Throughput of WiMAX MIMO-OFDM 
Including Adaptive Modulation and Coding 



Hadj Zerrouki 1 , Mohamed Feham 2 

Laboratoire de Systemes de Technologies de reformation 

et de Communication (STIC) 

University Abou Baker Belkaid, Tlemcen, Algeria. 



Abstract— WiMAX technology is based on the IEEE 802.16 
specification of which IEEE 802.16-2004 and 802.16e amendment 
are Physical (PHY) layer specifications. IEEE 802.16-2004 
currently supports several multiple-antenna options including 
Space-Time Codes (STC), Multiple-Input Multiple- Output 
(MIMO) antenna systems and Adaptive Antenna Systems (AAS). 
The most recent WiMAX standard (802. 16e) supports broadband 
applications to mobile terminals and laptops. Using Adaptive 
Modulation and Coding (AMC) we analyze the performance of 
OFDM physical layer in WiMAX based on the simulation 
results of Bit-Error-Rate (BER), and data throughput. The 
performance analysis of OFDM-PHY is done. In this paper, an 
extension to the basic SISO mode, a number of 2x2 MIMO 
extensions are analysed under different combinations of digital 
modulation (QPSK, 16-QAM and 64-QAM) and Convolutional 
Code (CC) with 1/2 , 23 and 3/4 rated codes. The intent of this 
paper is to provide an idea of the benefits of multiple antenna 
systems over single antenna systems in WiMAX type 
deployments. 

Keywords-WiMAX; MIMO; OFDM; AMC; Space Time Block 
Codes; Spatial Multiplexing 

I. Introduction 

The first WiMAX systems were based on the IEEE 802.16- 
2004 standard [1]. This targeted fixed broadband wireless 
applications via the installation of Customer Premises 
Equipment (CPE). In December, 2005 the IEEE completed the 
802.16e-2005 [2] amendment, which added new features to 
support mobile applications. The resulting standard is 
commonly referred to as mobile WiMAX. 

An ever crowded radio spectrum implies that future 
demands must be met using more data throughput wireless 
technologies. Since system bandwidth is limited and user 
demand continues to grow, spectral efficiency is vital. One way 
to improve link capacity, and potentially increase spectral 
efficiency, is the application of MIMO. It is well reported in 
the literature that MIMO physical (PHY) layer techniques have 
the potential to significantly increase bandwidth efficiency in a 
rich scattering environment [3]. Orthogonal Frequency 
Division Multiplexing (OFDM) is a well-established technique 
for achieving low-cost broadband wireless connectivity, and 
has been chosen as the air interface for a range of new 
standards, including IEEE802.16d/e. The ideas of MIMO and 
OFDM have been combined by a number of authors to form a 



new class of MIMO-OFDM system [4] [5]. This approach 
represents a promising candidate for WiMAX applications. 

WiMAX standard supports a full-range of smart antenna 
techniques, including spatial transmit diversity and spatial 
multiplexing (SM). Spatial transmit diversity is achieved by 
applying Alamouti's Space-Time coding. SM can also be 
employed to increase the error-free peak throughput. Higher 
order modulation schemes with SM increase the link 
throughput, but require high SNR to achieve low Packet Error 
Rates (PER). Space-Time Block Coding (STBC) provides 
strong diversity gain, but cannot increase the link throughput 
without the use of Adaptive Modulation and Coding (AMC), 
and therefore AMC has become a standard approach in recently 
developed wireless standards, including WiMAX. The idea 
behind AMC is to dynamically adapt the modulation and 
coding scheme to the channel conditions so as to achieve the 
highest spectral efficiency at all times. Adaptive modulation 
changes the coding scheme and/or modulation method 
depending on channel-state information - choosing it in such a 
way that it squeezes the most out of what the channel can 
transmit. 

This paper investigates the performance of the WiMAX 
standard when MIMO techniques are applied. Bit Error Rate 
(BER) and throughput results are presented for a MIMO with 
OFDM system that uses the coding and modulation schemes 
defined in the WiMAX standard IEEE 802.16-2004. Results 
are compared with basic SISO operation. 

II. The WiMAX PHY Description 

The IEEE 802.16 standard was firstly designed to address 
communications with direct visibility in the frequency band 
from 10 to 66 GHz. Due to the fact that non-line-of-sight 
transmissions are difficult when communicating at high 
frequencies, the amendment 802.16a was specified for working 
in a lower frequency band, between 2 and 1 1 GHz. The IEEE 
802. 16d specification is a variation of the fixed standard (IEEE 
802.16a) with the main advantage of optimizing the power 
consumption of the mobile devices. The last revision of this 
specification is better known as IEEE 802.16-2004 [1]. 

On the other hand, the IEEE 802. 16e standard is an 
amendment to the 802.16-2004 base specification with the aim 
of targeting the mobile market by adding portability. 



86 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



WiMAX standard-based products are designed to work not 
only with IEEE 802.16-2004 but also with the IEEE 802. 16e 
specification. While the 802.16-2004 is primarily intended for 
stationary transmission, the 802. 16e is oriented to both 
stationary and mobile deployments. 

A. PHY Layer Overview 

WiMAX is not truly new; rather, it is unique because it was 
designed from the ground up to deliver maximum throughput 
to maximum distance while offering 99.999 percent reliability. 
To achieve this, the designers (IEEE 802.16 Working Group D) 
relied on proven technologies for the PHY including 
orthogonal frequency division multiplexing (OFDM), time 
division duplex (TDD), frequency division duplex (FDD), 
Quadrature Phase Shift Keying (QPSK), and Quadrature 
Amplitude Modulation (QAM), to name only a few. WiMAX 
has a scalable physical-layer architecture that allows for the 
data rate to scale easily with available channel bandwidth. This 
scalability is supported in the OFDMA 1 mode, where the FFT 
(fast Fourier transform) size may be scaled based on the 
available channel bandwidth. For example, a WiMAX system 
may use 128, 512, or 1024 FFTs based on whether the channel 
bandwidth is 1.25MHz, 5MHz, or 10MHz respectively. This 
scaling may be done dynamically to support user roaming 
across different networks that may have different bandwidth 
allocations. 

B. OFDM Parameters in WiMAX 

As mentioned previously, the fixed and mobile versions of 
WiMAX have slightly different implementations of the OFDM 
physical layer. Fixed WiMAX, which is based on IEEE 
802.16-2004, uses a 256 FFT-based OFDM physical layer. 
Mobile WiMAX, which is based on the IEEE 802.16e-2005 2 
standard, uses a scalable OFDMA-based physical layer. In the 
case of mobile WiMAX, the FFT sizes can vary from 128 bits 
to 2048 bits. 

Table I shows the OFDM-related parameters for both the 
OFDM-PHY and the OFDMA-PHY. The parameters are 
shown here for only a limited set of profiles that are likely to be 
deployed and do not constitute an exhaustive set of possible 
values. 

C. WiMAX OFDMA-PHY 

In Mobile WiMAX, the FFT size is scalable from 128 to 
2048. Here, when the available bandwidth increases, the FFT 
size is also increased such that the subcarrier spacing is always 
10.94 kHz. This keeps the OFDM symbol duration, which is 
the basic resource unit, fixed and therefore makes scaling have 
minimal impact on higher layers. A scalable design also keeps 
the costs low. The subcarrier spacing of 10.94 kHz was chosen 
as a good balance between satisfying the delay spread and 
Doppler spread requirements for operating in mixed fixed and 
mobile environments. This subcarrier spacing can support 
delay-spread values up to 20 jus and vehicular mobility up to 
125 km/h when operating in 3.5GHz. A subcarrier spacing of 
10.94 kHz implies that 128, 512, 1024, and 2048 FFT are used 
when the channel bandwidth is 1.25MHz, 5MHz, 10MHz, and 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 
20MHz, respectively. It should, however, be noted that mobile 
WiMAX may also include additional bandwidth profiles. 



TABLE I. 



OFDM Parameters Used in WiMAX 



Parameter 


Fixed 
WiMAX 
OFDM 


Mobile WiMAX 
Scalable 
OFDMA a 


FFT size 


256 


128 


512 


1024 


2048 


Number of used data subcarriers 


192 


72 


360 


720 


1440 


Number of pilot subcarriers 


8 


12 


60 


120 


240 


Number of null/guardband subcarriers 


56 


44 


92 


184 


360 


Cyclic prefix or guard time 


1/4, 1/8, 1/16, 1/32 


Channel bandwidth (MHz) 


3.5 


1.25 | 5 | 10 | 20 


Subcarrier frequency spacing (kHz) 


15.625 


10.94 


Useful symbol time (^is) 


64 


91.4 


Guard time assuming 12.5% (^s) 


8 


11.4 


OFDM symbol duration ((is) 


72 


102.9 


Number of OFDM symbols in 5 ms frame 


69 


48.0 



a. Boldfaced values correspond to the OFDMA parameters used in our evaluation of 

the Fixed WiMAX standard. 

III. System Model Description 

Since increase the link throughput systems is the main goal 
of this work. Figure 1 depicts the block scheme of a typical 
structure of a system combining Space Time Coding (STC) 
with MIMO OFDM enabled WiMAX simulator used in this 
paper. The Block diagram represents the whole system model 
or the signal chain at base band. The block system is divided 
into 2 main sections namely the transmitter and the receiver. 



Data bits 



1. 



Randomization 



Convolutional 
Coding 



Interleaving 



r. 



BER & TP 
calculation 

:zrz 



Derandomization 

~1 



Convolutional 
Decoding 



Deinterleaving 



Modulation scheme 
(QPSK, M-QAM) 



"1L 



Symbol 
Demodulation 



Space time 
encoder 



Sub-Channel 
Mapping 



inn 



Sub-Channel 
Mapping 



IFFT + CP 



uni 



Space time 
Decoder 

_± 4 



Sub-Channel 
Demapping 



IFFT + CP 



Ly 



i 



TfTTT 



Sub-Channel 
Demapping 



FFT - CP 



^ 



nm 



FFT - CP 



Ly u 



Transmitter 



Receiver 



Figure 1 . System block diagram for WiMAX MIMO OFDM 
simulator used in this paper. 



Orthogonal Frequency Division Multiplexing Access. 
2 Although the scalable OFDMA scheme is referred to as mobile WiMAX, it 
can be used in fixed, nomadic, and mobile applications. 



87 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) InternationalJournal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 



A. Transmitter 

The data is generated from a random source, consists of a 
series of ones and zeros. Since the transmission is done block 
wise, when forward error correction (FEC) is used, the size of 
the data generated depends on the block size used, modulation 
scheme used to map the bits to symbols (QPSK, M-QAM), and 
whether FEC is used or not [1]. The generated data is passed on 
to the next stage, either to the FEC block or directly to the 
symbol mapping if FEC is not used. 

1) Channel coding: There are various combinations of 
modulations and code rates available in the OFDMA burst. 
Channel coding includes the randomization of data, forward 
error correction (FEC) encoding, interleaving, and modulation. 
In some cases, transmitted data may also be repeated on an 
adjacent subcarrier. 

2) Randomization: Randomization of the data sequence is 
typically implemented to avoid the peak-to-average power 
ratio (PAPR) increasing beyond that of Gaussian noise, thus 
putting a boundary on the nonlinear distortion created in the 
transmitter's power amplifiers. It can also help minimize 
peaks in the spectral response. 

3) Forward error correction (FEC): In our case, the error 
correcting codes are used, the data generated is randomized so 
as to avoid long run of zeros or ones, the result is ease in 
carrier recovery at the receiver. The randomized data is 
encoded using tail biting convolutional codes (CC) whose 
constraint length is 7 and the native code rate is V 2 (puncturing 
of codes is provided in the standard to produce higher code 
rates). Finally interleaving is done by two stage permutation, 
first to avoid mapping of adjacent coded bits on adjacent 
subcarriers and the second permutation insures that adjacent 
coded bits are mapped alternately onto less or more significant 
bits of the constellation, thus avoiding long runs of lowly 
reliable bits. 

4) Modulation: There are three modulation types available 
for modulating the data onto the subcarriers : QPSK, 16QAM, 
and 64QAM used with gray coding in the constellation map. 
In the Uplink, the transmit power is automatically adjusted 
when the modulation coding sequence (MCS) changes to 
maintain the required nominal carrier-to-noise ratio at the BS 
receiver. 64QAM is not mandatory for the Uplink. 

5) Space Time Encoder (MIMO encoder): The Space Time 
Encoder stage converts one single input data stream into 
multiple output data streams. How the output streams are 
formatted depends on the type of MIMO method employed. 
Different symbols are simultaneously transmitted over these 
antennas to reduce noise interference. The receiver after 
receiving the signal retrieves the bits using Maximum 
Likelihood decoding algorithm and passes the data to the 
guard band removal block. 

6) IFFT and cyclic prefixand: An'iV point inverse discrete 
fourier transform (IDFT) of 'X^'is defined as: 



1 7V-1 , >» 

*(«) = -£*(*>" for V =0,1,.... ^-7. 



Nto 



(1) 



From the equation we can infer that this is equivalent to 
generation of OFDM symbol. An efficient way of 
implementing IDFT is by inverse fast Fourier transform 
(IFFT). Hence IFFT is used in generation of OFDM symbol. 
The addition of cyclic prefix is done on the time domain 
symbol obtained after IFFT. The IFFT size (W value) is 
considered as 512 in simulations. This data is fed to the 
channel which represents 'Rayleigh fading channel model' 
and also implements multipath as shown in block diagram. 

B. Receiver 

The first thing done at receiver (in simulation) is removal 
of cyclic prefix, thus eliminating the inter symbol interference 
(ISI). Data is then passed through the serial to parallel 
converter of size 512 and then fed to the FFT for frequency 
domain transformation. The signal was distorted by the 
channel, to reconstruct the original signal we need information 
as to how the channel acted on the transmitted signal so that 
we can mitigate its effect. This is called equalization. In an 
OFDM system, this is done by channel estimation and 
interpolation, and reverse process (including deinterleaving 
and decoding) is executed to obtain the original data bits. As 
the deinterleaving process only changes the order of received 
data, the error probability is intact. 

C. Rayleigh Fading Channel 

Rayleigh Fading is one kind of statistical model which 
propagates the environment of radio signal. According to 
Rayleigh distribution magnitude of a signal which has passed 
though the communication channel and varies randomly. 
Rayleigh Fading works as a reasonable model when many 
objects in environment which scatter radio signal before 
arriving of receiver. When there is no propagation dominant 
during line of sight between transmitter and receiver on that 
time Rayleigh Fading is most applicable. On the other hand 
Rician Fading is more applicable than Rayleigh Fading when 
there is dominant line of sight. During our simulation we used 
Rayleigh Fading when we simulate the performance of BER 
and throughput Vs Signal to Noise Ratio. 

IV. Enhancement with MIMO link adaptation 

It is well-understood that spectral efficiency is the key to 
good system design. Without loss of generality, the normalized 
effective system efficiency can be written as: 



£■ system data throughput 

^ sys total radio resouces allocated 



(2) 



Two clear approaches emerge to improve the effective 
efficiency. Firstly the system data throughput can be improved 
by using methods such as high level AMC and MIMO. 
Secondly, improvements can be made to reduce the amount of 
radio resource required in the system. This section focuses on 
the system performance enhancements with MIMO Link 
Adaptation (LA) in combination with OFDMA. Using a 



88 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



statistical approach, it is possible to demonstrate the potential 
benefits of a relay enhanced mobile WiMAX deployment. 



A. Throughputs and coverages for WiMAX system 

It is well-known that MIMO promises transmission 
efficiency enhancement which achieves one aspect of 
efficiency enhancement. IEEE 802. 16e standard supports a full 
range of smart antenna technologies. Together with modulation 
and coding, the link throughput for each user can be calculated 
from Packet Error Rate (PER) by: 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
In our case, by using six (6) of the common WiMAX 
burst profiles; it is possible to achieve a large range of spectral 
efficiencies. This allows the throughput to increase as the 
signal-to-interference-plus-noise ratio (SINR) increases 
following the trend promised by Shannon's formula. In this 
case, the lowest offered data rate is QPSK and rate 1/2 
Convolutional codes; the highest data-rate burst profile is with 
64 QAM and rate 3/4 Convolutional codes. The achieved 
throughput normalized by the bandwidth is defined as: 



r _ N D N b R FEC R STC 



: (l - PER) 



(3) 



Where, T s , N D , N b , R FEC and R ST c denote the OFDMA 
symbol duration, the number of assigned data subcarriers, the 
number of bits per subcarrier, FEC coding rate, and space-time 
coding rate for the user. Equation (3) implies that, a 
combination of MIMO, AMC and flexible sub-channelization 
is required to maximize the link performance. 

B. MIMO scenarios description 

1) Space-Time Block Coding (STBC): Our Fixed WIMAX 
simulator implements the Alamouti scheme [7] on the 
Downlink to provide transmit and receive diversity. This 
scheme uses a transmission matrix [si, -s 2 ; s 2 , Si ], where Si 
and s 2 represents two consecutive OFDMA symbols. 

2) Spatial Multiplexing (SM): WiMAX system supports 
SM to increase the peak error free data rate [8]. The idea 
behind spatial multiplexing is that multiple independent 
streams can be transmitted in parallel over multiple antennas 
and can be separated at the receiver using multiple receive 
chains through appropriate signal processing. This can be done 
as long as the multipath channels as seen by the various 
antennas are sufficiently decorrelated, as would be the case in 
a scattering-rich environment. Spatial multiplexing provides 
data rate and capacity gains proportional to the number of 
antennas used, a 2x2 SM system can double the peak data rate. 
This comes at the expense of sacrificing diversity gain, and 
hence a much higher SNR is required. 

3) Adaptive Modulation and Coding: WiMAX systems 
use adaptive modulation and coding in order to take advantage 
of fluctuations in the channel. The basic idea is quite simple: 
Transmit as high a data rate as possible when the channel is 
good, and transmit at a lower rate when the channel is poor, in 
order to avoid excessive dropped packets. Lower data rates are 
achieved by using a small constellation, such as QPSK, and 
low-rate error-correcting codes, such as rate convolutional or 
turbo codes. The higher data rates are achieved with large 
constellations, such as 64 QAM, and less robust error 
correcting codes; for example, rate convolutional, turbo, or 
LDPC codes. In all, 52 configurations of modulation order and 
coding types and rates are possible, although most 
implementations of WiMAX offer only a fraction of these[9]. 



T = (1 - BLER) r log 2 (M) bps I Hz 



(4) 



where BLER is the block error rate, r < 1 is the coding rate, and 
M is the number of points in the constellation. For example, 64 
QAM with rate 3/4 codes achieves a maximum throughput, 
when BLER— >0; QPSK with rate 1/2 codes achieves a best- 
case throughput. 

V. Simulation results 

In this section SISO and MIMO BER and Throughput results 
are presented using the Fixed WiMAX simulator. The 
Simulation model was implemented in Matlab® 7. The PHY 
parameters used in simulation are given in Table I, with 
boldfaced values correspond to the OFDMA-PHY parameters. 
A carrier frequency of 2GHz is considered. For Spatial 
Multiplexing, an MMSE receiver is used to remove the inter- 
stream interference on a per sub-carrier basis. The link 
throughput is calculated from the PER as given by equation 
(3) in section IV. On the downlink, we consider that no 
sharing of OFDMA symbol. In this case of single-user MIMO, 
the multiple streams are intended for the same receiver; we 
have considered transmission formats with only a single 
stream for a single user as a basic OFDM in fixed WiMAX. 

Figure 2 and 3 displays the performance of SISO and 2x2 
MIMO STBC systems. For comparison purposes, the SNR 
versus BER graph is plotted for different modulation schemas 
and Convolutional Coding (CC) rates are showed. 




10 15 

SNR (dB) 



Figure 2. BER performance of SISO WiMAX system. 



89 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



It is evident from the graph that the QPSK modulation 
scheme with 1/2 CC rate is better suited for OFDM 
transmission in terms of BER performance, and it can be see 
that the BER performance MIMO system with 2x2 STBC 
yields a gain of about 3 dB over the corresponding SISO 
system at a BER of 10" 3 . 




(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
directly exploits the MIMO channel capacity to improve the 
data throughput by simultaneously transmit different signals 
on different transmit antennas, at the same carrier frequency. 
The main disadvantage is that no redundancy is added and, 
thus, it might suffer from poor link reliability. To overcome 
this problem additional channel coding can introduced. This, 
however, reduces its data rate advantage. As expected, the SM 
2x2 modes doubles the peak error-free throughput of every 
link-speed. However, at low SNR values the throughput of SM 
is less than STBC. 



10 15 

SNR (dB) 



Figure 3. BER performance of 2x2 STBC WiMAX system. 

Figure 4 and 5 present the throughput versus SNR graphs 
for the SISO and MIMO 2x2 STBC scenarios respectively. We 
observe that STBC offers a significant performance gain of 
3dB to 4dB, the exact value depend on the selected link-speed. 
As we see here, STBC does not improve the data throughput, 
however at a given SNR STBC can provide a significant 
increase in throughput when combined with suitable link 
adaptation, since higher throughput modes can be used at much 
lower values of SNR. STBC increases the robustness of the 
WiMAX system by coding over the different transmitter 
branches and over space and temporal dimension. In this way, 
the spectral efficiency of MIMO is exploited by adding extra 
redundancy to improve the performance. 




Figure 4. Throughput of SISO WiMAX system. 

Figure 6 illustrates the simulated MIMO 2x2 SM system 
throughputs versus SNR. The main advantage of SM is that it 



-QPSK, 1/2 
- QPSK, 3/4 
-16QAM, 1/2 
-16QAM, 3/4 

64QAM, 1/2 
-64QAM, 2/3 

64QAM, 3/4 _ 




Figure 5 . Throughput of 2x2 STBC WiMAX system. 




Figure 6. Throughput of 2x2 SM WiMAX system. 

Figure 7 shows the throughput envelope versus SNR for all 
the investigated fixed WiMAX scenarios: SISO, 2x2 STBC, 
and 2x2 SM. This envelope assumes the use of adaptive 
modulation and coding (AMC) to maximize the expected 
throughput. Obviously, both MIMO schemes outperform the 
SISO scenario. However, for a very spatially correlated 
channel, the SM method can be worse than SISO. In this case 
STBC performance would tend to that of SISO. The STBC 
produces the best performance at low to medium values of 
SNR, due to its robustness in poor channel conditions. On the 



90 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



other hand, at high SNR the increased error-free data rate 
makes SM the best choice. 

WiMAX system supports Adaptive MIMO Switching 
(AMS) to select the best MIMO scheme. Figure 7 clearly 
shows that for the channel conditions analysed here, the 
switching point between STBC and SM is 20dB. This value 
will increase with increasing spatial correlation. 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
the benefit of MIMO when applied to WiMAX. Throughput 
results were presented for diverse scenarios. At lower values of 
SNR STBC is preferred. However, at high SNR Adaptive 
MIMO Switching should be used to switch to SM. Give that 
SM 2x2 doubles the error-free throughput, at high SNR this 
scheme leads to the highest throughput. In practice, the 
viability of SM (and the value of the SNR switching threshold) 
depends on the level of spatial correlation. 




Figure 7. Adaptive Modulation and Coding (AMC) and switching point 
between STBC 2x2 and SM 2x2. 



VI. Conclusion 

In this paper, we considered the calculation of data 
throughput of the downlink of OFDMA-based IEEE802.16 
WiMAX systems; this paper has presented a detailed study of 



References 

[1] IEEE Std 802.16TM-2004, "Part 16: Air interface for fixed broadband 
wireless access systems," Oct 2004. 

[2] IEEE Std 802.16Etm-2005, "Part 16: Air interface for fixed and mobile 
broadband wireless access systems," Feb. 2006. 

[3] A. Paulraj R. Nabar and D. Gore, Introduction to space-time wireless 
communications. Cambridge university press, 2003. 

[4] H. Sampath, S. TAlwar, J. Tellado, V. Erceg and A. Paulraj, "A form- 
generation MIMO-OFDM broadband wireless system: design 
performance, and field trial results," IEEE Communications Magazine, 
Vol. 40, No. 9, pp 143-149, Sep. 2002. 

[5] A. Zelst and T. Schenk, "Implementation of a MIMO OFDM-based 
wireless LAN system," IEEE Trans. Axoustics, Speech and Signal 
Processing, Vol. 52, No. 2, pp 483-494, Feb. 2004. 

[6] C. Tarhini, T. Chahed, On capacity of OFDMA-based IEEE802.16 
WiMAX including Adaptive Modulation and Coding (AMC) and 
intercell interference, LANMAN'2007, Princeton NJ, June 2007. 

[7] M. Alamouti, "A simple transmit diversity technique for wireless 
communications," IEEE JSAC, Vol. 16, No. 8, Oct. 1998. 

[8] G. J. Foschini, "Layered Space-Time Architecture for Wireless 
Communication in a Fading Environment when Using Multi-element 
Antennas," Bell Labs Tech. J. pp. 41-59, Autumn 1996. 

[9] J. G. Andrews, A. Ghosh, R. Muhamed, "Fundamentals of WiMAX: 
Understanding Broadband Wireless Networking, " Prentice Hall PTR, 
Feb, 2007. 



91 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



Performance Modeling and evaluation of Traffic 
management for Mobile Networks by SINR 

Prediction 



K.K.Guatam 

Department of Computer Science & Engineering 

Roorkee Engineering & Management Technology Institute 

Shamli(247774) India 



Anurag Rai 

Department of Information Technology 

College of Engineering Roorkee 

Roorkee (247667) India 



Abstract — Over the recent years a considerable amount of effort 
has been devoted towards the performance evaluation and 
prediction of Mobile Networks. Performance modeling and 
evaluation of mobile networks are very important in view of their 
ever expending usage and the multiplicity of their component 
parts together with the complexity of their functioning. The 
present paper addresses current issues in traffic management 
and congestion control by (singal-to-interference-plus-noise ratio) 
SINR prediction congestion control, routing and optimization of 
cellular mobile networks. 

Keywords- Mobile Networks, Modeling, call admission control, 
QoS (Quality of Service) SINR. 



I. 



INTRODUCTION 



Over the recent years a considerable amount of effort has 
been devoted towards the performance, evaluation of wireless 
mobile networks (WMN). A considerable amount of research 
has been used to characterize user and calling behavior and 
their performance impact on wireless mobile networks. At 
present the mobility in most mobile networks is confined to 
the end users only. 

With the development of mobile compression, the CAC 
schemes are generally adopted by setting thresholds for hand 
off calls and new call differently given the traffic condition 
and it is the maximum number of users that can be 
supported. [1,2] In realistic systems, information about the 
traffic management quality cannot; be instantaneous but is 
outdated to some degree. Firstly, the SINR estimation in the 
receiver takes some time and secondly, the user has to wait for 
the next channel allocation to report his SINR to the base 
station. The system may need to block incoming users if all of 
the entire band width has been used up to provide the highest 
QoS to existing users. However, if these existing users can be 
degraded to a lower but acceptable QoS level, it is possible to 
reduce the blocking probability without degrading the QoS of 
existing users. A graceful degradation mechanism is proposed 
in (1). Thus a system could free some bandwidth allocation for 
new users. In this paper, we address current issues in traffic 
management of cellular mobile networks by the use of SINR 
prediction, the SINR is calculated for mobile user equipments 



in every transmission time interval for the traffic 
management of mobile networks. In traffic management and 
congestion control, courcoubetis and series device new 
procedures and tools for the analysis of network traffic 
measurement. 

II. MODEL DESCRIPTION 

We consider uplink communication in a wireless mobile 
network. As an accepted call does not always send data 

frames, then for best traffic, we consider the activity factor 

£ as the probability that a call is active. We represent QoS 
requirement of traffic by required transmission rate. The 
required transmission rate can be obtained by setting the target 
level. 

Often these intra-and inter-traffic interferences of calls 
can be large so that the target bit error rate of traffic 
interferences(BERIT) cannot be achieved temporarily, which 
is called outage. [3] The outage probability needs to approach 
zero (as close as possible) and can be different for each class. 
Here we assume for traffic management the allowed outage 
probability is the same. 

III. Outage Probability For Traffic 

In a mobile network, a traffic management that supports a 
single class of calls, the outage probability is given by [2,4]. 

Pout = Pr {N a + M a > (3/2) G(x 1 - (Yb/N0 n ) +1 } (1) 

When N a , M a , G, X, Y b , and NO represent the number of active 
calls in the current call. Similarly, in a network that support L- 
Class of calls, we obtain 

L 

Pr { ^ (Ybi / Ybj) Ci (Ni a +Mi a ) > Aj}=Pr 



P J out 



£ #I(Ni a +Mi a ) > 77 j} (2) 



When i, j represent traffic call classes (TCC), Ci is the 
number of orthogonal codes needs for a TCC, T. By the 



92 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



Gaussian random variable from the limit theorem and we can 
write control the outage probability of a TCC 'j', as 
P j out =Q(77J- XI dX ) (3) 



as <U we can show that 



= a ; 



dP„ 



>0 



w 



hereQ ( E, ) = 1/ ^~ In J e" x2/2 



dx 



And represent the total traffic receive single power (TRSP) i.e. 

L 

Y, #I(Ni a +Mi a ) 

_ L 

Therefore I = (1+fl) ^ 01 N a i, (4) 

i=\ 
L 

And d 2 A = Y 2 l(dUf2 N'i) 

i=\ 

Where N a i and d I indicate the mean and variance of 
N a I. 

According to the assumption of TCC equal outage probability 
for each class, r/ 1 = r/ j for all I and j, therefore TCC received 

single power meets the following relation. 

01 10] = CiXi (3G - 2CJXJ) /cjXj(3G+ 2CiXj) (6) 

This indicates that the power allocation refers to the target 
of TCC outage probability Call Admission Control (CAC): 

IV. SYSTEM MODEL 
The Communication system under consideration can be 

L 

defined as r[k] = ^ h [1]& [k-1] + z [k] (7) 

i=0 

Where r [k] received call sequence h [1] unknown channel 
for traffic with memory L\ z[k] is an independent and 
identically distributed Gaussian notice sequence. [5,6] 

Then traffic management symbol sequence s [k] is drawn 
from M-ary alphabet, A with equal probability, the vector 
version of (1) can be written as 



r[k] 



r[l] 
r[0] 



S[K-L] 



S[K] 



5[1-L] ... 5[1] 
S[-L] ... S[0] 



\WY 


+ 


~z[k] 


hm 




z[l] 


[h[0]_ 




_z[0]_ 



Where Sk is toeplitz data matrix. 

V. Call Admission Control for Traffic 

Management 

I call Admission Control for traffic Management, 
(CACTM) the outage Probability is very small, defined 



d?j dN t l dN t 

where OC i is the active factor for (CACTM) a class I call. It is 

clear that the average rate for mobile network (ARRMN) and 
outage probability increase with the number of users. 

VI. Numerical Result 

We now compare the performance of the two CACs 
through numerical analysis. [7] The system bandwidth is 
2.50MHz and each code can carry information bits at the rate 
of 19.2(kbps) so that the processing gain is 256. Two types of 
calls are considered to manifest the effect of traffic parameters 
on performance. Class 1 and 2 calls are voice traffic and we 
set their transmission rates after channel coding at 
19.29(kbps). They have different Mobile Network Average 
Revenue Rate (MNARR) for the traffic management 
requirement of less than 10" 4 and 10 6 , respectively, and their 
activity factors are set at 1.0. The coefficient for intercall 
interference modeling are chosen as fl = 0.114 and f2 = 
0.44(12). 



VII. 



Framework 



Angle -SINR Table :- 

In order to make the directional routing effective for call 
admission control system, a node should know how to set its 
transmission direction effectively to transmit a packet to its 
neighbors. So each node periodically collects its neighborhood 
information and forms an Angle- SINR Table (AST). Sinu s 
m(t) (Signal - to - Interference and Noise Ratio) is a number 
associated with each link 1 u n, m, and is a measurable indicator 
of the strength of radio connection from node n to node m at an 
angle u with respect to n and as perceived by m at any point of 
time t for call admission control. AST of node n specifies the 
strength of radio connection of its neighbors with respect to n 
at a particular direction for call admission control. Angle - 
SINR Table for node n time t is shown below (Table I) where, 
we assume that nodes I, j and k are the neighbors of n. 
[9,10,11,12] 

TABLE I. ANGLE - SINR TABLE (AST) FOR NODE n 



Azimuth 

Angle 

(degree) 


SINR value as perceived by neighbors of rooters n at 
different angle w.r.t rooters n 


i 


j 


K 





SINR°n,i (t) 


SINR°n,j (t) 


SINR°n,j (t) 


30 


SINR 30 n,i (t) 


SINR 30 n,j (t) 


SINR 30 n,j (t) 


60 


SINR 60 n,i (t) 


SINR 60 n,j (t) 


SINR 60 n,j (t) 










330 


SINR 330 n,i (t) 


SINR 330 n,j (t) 


SINR 330 n,j (t) 


360 


SINR 360 n,i (t) 


SINR 360 n,j (t) 


SINR 360 n,j (t) 



In order to form ANGLE - SINR TABLE (AST), each 
node periodically sends a directional request in the form of a 
directional broadcast for the call admission control, 
sequentially in all direction. In this work, it has been done at 30 
degree interval, covering the entire 360 degree space 



93 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



sequentially. A node is i in the neighborhood of n will wait 
until it receives all the request packets generated by n in all 
direction in that occasion. In other words, node I accumulates 
the entire column of the AST of n for node I, I accumulates the 
entire column of the AST of n for rooters i. Here, rooters i, 
after receiving the first request from n, has to wait a pre- 
specified amount of time to make sure that the directional 
broadcasts by n in all direction are over. Rooters I sends this 
information from all the neighbors of n, the Angle- SINR Table 
of n would be complete. [13] 

Conclusion 

In this paper, we consider Call Admission Control for 
Traffic Management (CACTM) in Mobile Networks. Through 
the mathematical analysis and also present outage probability 
and a system model's for CAC. We also present an example 
for Call Admission Control for Traffic Management 
(CACTM) and present a frame work for the set up-the call 
admission control. 

Acknowledgment 

The author would like to thank Dr. H.N. Dutta, Director, 
REMTech for his moral support in carrying out the work. 

References 



[1] J. Zhang , J. W. Mark, and S.Xuemin, "An adaptive handoff priority 
scheme for wireless MC-CDMA cellular networks supporting 
multimedia application ," in Proc. IEEE Globecom, Nov/Dec. 2004, pp. 
3088-3092. 

[2] Z. Liu and M. E. Zarki, "SIR - based call admission control for 
DSCDMA cellular system," IEEE J. Select. Areas Commun, vol . 12, 
May 1994, pp.638-644, 

[3] R. j. Boucherie and Nico M. Van Dijk, "On a queueing network model 
cellular mobile telecommunications networks, "Operations Research, vol 
48, no, 2000, pp . 38—49, 

[4] X.Chao and W. Li, "Performance analysis of a cellular network with 
multiple classes of calls," IEEE Trans. Coomun, vol. 53, no. 9, pp. 1542- 
1550,2005. 



[5] C. T. Chou and K.G. Shin, " Analysis of adaptive bandwidth allocation 
in wireless networks with multilevel degradable quality of service," 
IEEE Trans. Mobile Compute, vol. 3 no. 1, 2004, pp. 5-17, 

[6] Maruf Mohammad , William Tranter " Blind Acquisition of short Burst 
with Per - Survivor Processing" IEEE TRANSCATION ON 
WIRELESS COMMUNICATION, vol. 6. No. 2. February 2007. 

[7] Wei Li , and Xiulichao " Call admission Control for an adaptive 
Heterogeneous Multimedia Mobile Network" IEEE TRANSCATION 
ON WIRELESS COMMUNICATION, vol. 6. no. 2. February 2007. 
page no. 515-525. 

[8] Jin-Cho Choi, Yound-June Choi, and Saewoong Bank "power - Based 
Admission Control for Multi Class Calls in QoS -Sensitive CDMA 
NETWORKS" IEEE TRANSCATIONS ON Wireless communication, 
vol, 6 no. 2 February 2007 page no. 469-472. 

[9] Markus Jordan Gerd Ascheid and Heinrich Meyr , "Performance 
Evaluation of Opportunistic Beamforming with SINR Prediction for 
HSDPA", NEW Jersy 1993. 

[10] M. Kawai, M.Nozaki and K.Gyoda, "A wireless Ad Hoc Community 
Network with Recongfigurable Topology Architecture", Proc.of the 
GLOBECOMM'98.1998. 

[11] T. Ohira and K.Gyoda , "Electronically Steerable Passive Array Radiator 
(ESPAR),"Antennas for Law -cost Adaptive Beam Forming", IEEE 
International Conference on Phased Array System , dana Point , CA 
May 2000. 

[12] S. Mandyopadhyay, K. Gyoda K. Hasuike,S.Horisawa,Y.Kado And 
s.Tawara, "An Adaptive MAC Protocol for wireless AD Hoc 
Community Network (WACNet) Using Electronically Steerable Passive 
Array Radiator Anteena ", submitted to GLIBECOM 2001. 

[13] Fan Bai, N Sadagopan and A Helmy , " IMPROTANT : a Framework to 
systematically analyze the impact of mobility on performance of 
Routing Protocol for AdHoc Neworks", IEEE 2003 



AUTHORS PROFILE 

Authors Profile .. K K Gautam is the Dean in the Roorkee Engineering 
& Management Technology Institute, Shamli-247 774, India. 

Anurag Rai is the Head of the Information Technology Department in 
College of Engineering Roorkee,Roorkee-247 667, India. 



94 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 



THAI RHETORICAL STRUCTURE ANALYSIS 



Somnuk Sinthupoun 

Department of Computer Science 

Maejo University 

Chiangmai, Thailand 50290 



Ohm Sornil 

Department of Computer Science 

National Institute of Development Administration 

Bangkok, Thailand 10240y 



Abstract — A rhetorical structure tree (RS tree) is a 
representation of discourse relations among elementary discourse 
units (EDUs). A RS tree is very useful to many text processing 
tasks employing relationships among EDUs such as text 
understanding, summarization, and question-answering. Thai 
language with its unique linguistic characteristics requires a 
unique RS tree construction technique. This paper proposes an 
approach for Thai RS tree construction which consists of three 
major steps: EDU segmentation, Thai RS tree construction, and 
discourse relation (DR) identification. Two hidden markov 
models derived from grammatical rules are used to segment 
EDUs, a clustering technique with its similarity measure derived 
from Thai semantic rules is used to construct a Thai RS tree, and 
a decision tree whose features extracted from the rules is used to 
determine the DR between EDUs. The proposed technique is 
evaluated using three Thai corpora. The results show the Thai RS 
tree construction and the DR identification effectiveness of 
94.90% and 82.81%, respectively. 



Keywords- Thai Language, Element Discourse 
Rhetorical Structure Tree, Discourse Relation. 

I. Introduction 



Unit, 



A RS tree is a tree-like representation of discourse relations 
among elementary discourse units (EDUs) which can be 
defined as follows: RS tree = (Status, DR, Promotion, Left, 
Right) where Status is either nucleus or satellite EDU (nucleus 
expresses what is more essential to writer's purpose than 
satellite); DR is a Discourse Relation (DR); Promotion is a 
subset of EDUs; and Left, Right can be either NULL or 
recursively defined objects of type RS tree [4, 6]. 

Some researchers consider an EDU to be a clause or clause- 
like [6] excerpt, while others consider them to be a sentence 
[10] in discourse parsing. A number of techniques are 
proposed to determine EDU boundaries for English such as 
using discourse cues [5, 8, 9], punctuation marks [6, 9], and 
syntactic information [6, 10, 12]. 

There are many DRs. Some have a single nucleus such as 
elaboration and condition, while others have multiple 
nucleuses such as contrast [20]. A number of techniques for 
determining the DR between EDUs are proposed, such as 
using verb semantics [13] to build verb based events which 
represent EDUs, using cue phrase/discourse marker (e.g., 



"because", "however") [5], and using machine learning 
techniques [6]. 

For Thai RS tree construction, Sukvaree, et.al. [16] 
purposes a technique to construct a tree by using a spanning 
tree which make decision by discourse marker and focus of 
EDU, into two phases which consist of local and global EDU 
spanning tree. Using spanning tree, RS tree construction used 
the right adjacent rule to add right hand side to attach left hand 
side EDU. 

For Thai DR recognition, Sukvaree, et.al. [16] purposes a 
technique to recognize a DR by using DR marker tag to 
recognize DR. If DR marker tag has many semantic DRs, the 
maximum probability value in chain of DR marker is used to 
identify the DR. There are four relations: cause-result, 
constion, contrast and elaboration. Wattanamethanont, et.al. 
[11] purposes a technique to recognize a DR by using Naive 
bayes classifier. The feature of the machine learning are DR 
marker, key phrase and word co-occurrence. There are three 
relation: elaboration, logical and sequence. 

This article proposes a new approach for Thai RS Tree 
construction which consists of two major steps: EDU 
segmentation and Thai RS tree construction. Two hidden 
markov models taking into account syntactic properties of 
Thai language are employed to segment EDUs, and a 
clustering technique with its similarity measure derived from 
linguistic properties of Thai language is employed to construct 
a Thai RS tree. Once the tree is created, DR between EDUs is 
then determined by a decision tree. 

II. Issues in Thai RS tree 

Thai language possesses unique characteristics syntactically 
and semantically. This makes techniques proposed for other 
languages not directly applicable to Thai language. A number 
of important issues are discussed in this section. 

A. Issues in Thai EDU Extraction 

No Explicit EDU boundary Thai language has no 
punctuation masks (comma, semi-colon and blank) and no 
special symbols (full stop) to identify the start and the end of 
EDUs. Unlike English contains specific symbols (e.g. '.', 
comma, semi-colon and blank) to specify the start and the end 
of EDUs, they can be used to separate a text into EDUs. 



95 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



Therefore, EDU identification becomes a problem for Thai RS 
tree construction. 

EDU1 EDU2 EDU3 

A A _ A fc 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
Omission Problem The absence of Subject, Object or 
Preposition which is modifier nucleus of VP especially in 
anaphoric EDU of Thai language are often occur. 



Thai : [wiw 2 . . . w m w m+1 w m+2 . . .w n w n+1 w n+2 . . . w ] 
English : [w { w 2 ... w m ],[w m + 1 w m+2 ... wj;[w n+1 w n+2 ... w ]. 
where w } is a word in text. 
Omission Problem Given two EDUs, an absence of 
subject, object or conjunction in anaphoric EDU may happen. 
Such as anaphoric EDU omits the subject that refers back to 
the object of antecedent EDU. Accordingly, EDU 
segmentation is ambiguous. 



Thai 



There are 
three 
Possible : 



"memzuBouMuaflfQ mnsnTvelu'lfl" (A friend's going 

to borrow this book. Because she hasn't been 

able to buy it.) 

1) [S(m0 t u)V( ! O^ 6 U0Su)O( t H t U^^0)]EDUl 

[Because S(O) V(hw91sj"1^)] EDU 2 

2) [S(m0 t u)V(D^ 6 U9Sjj)O( t H t lI^^0)] ED u J 

[Because(0)S(0)V(Tn4lu'¥)] EDU2 

3) [S(m9 t u)V( ! o^ 6 U0Su)O(O)] EDU i 
[Because(O)S(TiyM0)V(iiw0 , lu"lli)] ED u 2 



B. Issues in Rhetorical Structure Tree 
Construction 

After EDUs are extracted correctly, relationships among 
EDUs need to be determined, and a number of issues need to 
be considered. 

Adjacent Marker Problem Given three EDUs and two 
markers indicating different relationship, as shown in Ex. 1, 
there are two possible for the RS Tree. First, relationship 
between EDU1 and EDU2 is determined by using discourse 
marker "uei" (but), next that between (EDU1, EDU2) and 
EDU3 is determined. On the other hand, the relationship 
between EDU2 and EDU3 is determined first by using 
discourse marker 'm (if), next that between (EDU2, EDU3) 
and EDU1 is determined. 

Ex. 1. EDU1: fttfllflfifliftltfuanmimntT (A court has order to 
separate marriage property.) 
EDU2: uflmfifmffftmiijffafli'i'u (But if a wife or a 

husband disagree,) 
EDU3: fnnfls^EjnmnmiuEjnlfl (a court will have order to 
cancel separation.) 



EDUl 



EDU1 


EDU2 


EDU3 



EDU2 




EDU3 



a) A RS Tree (BUT) b) A RS Tree (IF) 

Fig. 1. Adjacent marker problem 



Ex. 2. EDUl: fnfllflfiriiftliiimnmimntT (A court has order to 
separate marriage property.) 
EDU2: 01 rf^nmnrrmmn 02 1a (A court will cancel 
to separate marriage property.) 

In Ex. 2, EDU2 omits subject 'Via" (court) and object 
'mmwitt' (marriage property). Therefore, word co-occurrence 
only is not enough to solve relationship between EDUl and 
EDU2. This research use Absence rules to solve relationship 
between EDUl and EDU2. 

Implicit Marker Problem The absences of discourse 
marker in Thai language are often occurred. In Ex. 3, 'W 
(but) is a discourse marker which is omitted but relationship 
between EDUl and EDU2 still have relationship similar to 
that EDUl and EDU2 have the discourse marker. 

Ex. 3. EDUl: fnalmlfli^liiuofimimjifY (A court has order to 
separate marriage property.) 
EDU2: 01 n?a™f0tnSflViiriii (But a wife or a husband 
disagrees to separate marriage property.) 

Therefore, only use cue phrase is not enough to solve 
relationship between EDUs such as EDU2. 

C. Issues in DR identification 

After RS trees are constructed correctly, DRs among 
EDUs need to be identified, and a number of issues need to be 
considered. 

Adjacent Marker Problem Given three EDUs and two 
markers indicating different relationship, as shown in Ex. 1, 
there are two possible for the DR. First, relationship between 
EDUl and EDU2 is "contrast" relation which is determined by 
using discourse marker "m" (but), next that between (EDUl, 
EDU2) and EDU3 is determined. On the other hand, the 
relationship between EDU2 and EDU3 is "condition" relation 
which is determined first by using discourse marker 'm (if), 
next that between (EDU2, EDU3) and EDUl is determined. 

Implicit DR Marker Problem The absences of DR 
marker in Thai language are often occurred. In Ex. 3, 'W 
(but) is a "contrast" relation marker which is omitted but 
relationship between EDUl and EDU2 still have "contrast" 
relation similar to that EDUl and EDU2 have the DR marker. 
Therefore, only use DR marker is not enough to solve DR 
between EDUs such as EDU2. 

DR Marker Ambiguity Problem one DR marker can 
have many semantics DR such as "ifk" (when) can infer 

"condition" or "cause-result" relation, "uri" (but) can infer 
"contrast" or "elaboration" relation in Ex. 4. 



Identify applicable sponsor/s here, (sponsors) 



96 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 



Ex. 4. EDU1: fnalflSfliftltfuanmimjitT (A court has order to 

separate marriage property.) 
EDU2: iiwfiiaTHfetnSflflfliu (But contrastre i a tion a wife or a 

husband disagree,) 
EDU1: fnalflSfliftltfuafimmsjitr (A court has order to 

separate marriage property.) 

EDU2: UM^f1?EJ1imstnSmi4^QU(ButOnly e i a boration relation a 

wife and a husband agree,) 

III. Structures of Th a i EDUs 

A Thai EDU consists of infrastructure and adjunct 
constituents. There are twelve possible arrangements of Thai 
EDU [17], as shown in Table 1. The structure of an EDU "A 
teacher usually doesn't drink alcohol" is shown in Fig. 2. 



EDU 

I 



FinajfimipnmljJUjJivian 
(A teacher usually doesn't drink alcohol) 



JZ 



Adjunc 



Infrastructure of sentence 



Modal 
Adjunc 



Subject 



ziz 



Noun 
phrase 

I 



Head 



Transitive verb 

l — 



usually 



teacher 



Verb phrase 
I 



Pre- 

Nucleus 
I 



doesn't 



=i_ 



Nucleu 



Object 



Noun 



Head 



drink 



alcohol 



Fig. 2. The structure of the EDU "A teacher usually doesn't drink alcohol". 
Table 1 : The twelve patterns of Thai EDUs. 



EDU 


Examples 


Rules 


Vi 


vb (I'm hungry.) 


NPs-V/-NPs 


S-Vi 


du-nn (It's rain.) 


Vi-S 


ilu1<nu-?nfli(Are you pain?) 


Vt-O 


■hi- ih (I'm hungry.) 


NPo-NPs-W- 
NPo 


S-Vt-O 


?fl-vu-ifln (The car hit the boy.) 


O-S-Vt 


(I've already seen this 
photograph.) 


Vtt-O-I 


» « ill » » VI V 

m lu la m-oi-mi h 

(I haven't given the patient 

the medicine.) 


NPs-Vff-NPo- 

NPi 


S-Vtt-O-I 


W-l'w-firifmA-'W'u 

(Who gave you the sweet?) 


O-S-Vtt-I 


fniiuMJ-lfisf^-^fmifniJ- f\m 
(Who would dare to ask you 
the secret?) 


NPo-NPs-Vtt- 

NPi 


I-S-Vtt-O 


■Hiji-ih-flsl'H-ijnmi 

(Niece, I am going to give 

you this house.) 


NPi-NPs-Vft- 

NPo 


N 


ili (Auntie) 


NP N -NP N 


N-N 


Tjihnm-W (Whose pen is this?) 



IV. 



Vol. 7, No. 1, 2010 

EDU Segmentation 



This section describes the EDU segmentation process 
proposed in this research. To reduce the segmentation 
ambiguity caused form modifiers, omissions of words and 
discourse markers, the first noun phrases and verb phrases 
which can be constituents of EDUs are determined, based on 
the possible structures. These phrases are then used to identify 
boundaries of EDUs. 

A noun phrase (NP) is a noun or a pronoun and its 
expansion which may function as one of the four Thai EDU 
constituents, namely the subject (S), the object (O), the 
indirect object and the Nomen (N). The structure of a noun 
phrase consists of five constituents which are: head (H), 
intransitive modifier (Mi), adjunctive modifier (Ma), 
quantifier (Q), and determinative (D). 

A verb phrase (VP) is a verb and its expansion which may 
function as one of the three Thai EDU constituents, namely 
intransitive verb (Vi), transitive verb (Vt) and double 
transitive verb (Vtt). The structure of a verb phrase consists of 
four constituents which are: nucleus (Nuc), pre-nuclear 
auxiliary (Auxl), post-nuclear auxiliary (Aux2), and modifier 
(M). 

An Arrangement of NP and VP constituent [17] is shown in 
Table 2. There are twenty five possible arrangements of noun 
phrase and ten possible arrangements of verb phrases. 

Table 2: The twenty five patterns of NP and ten patterns of VP. 



Noun Phrase 


Noun Phrase 


Verb Phrase 


H-Ma 


H 


Nuc 


H-Mi-Ma 


H-Mi 


Nuc-Aux2 


H-Q-Ma 


H-Q 


Nuc-M 


H-Ma-Q 


H-D 


Nuc-Aux2-M 


H-D-Ma 


H-Mi-Q 


Nuc-M-Aux2 


H-Mi-Q-Ma 


H-Q-Mi 


Auxl -Nuc 


H-Q-Mi-Ma 


H-Mi-D 


Auxl -Nuc- Aux2 


H-Mi-D-Ma 


H-Q-D 


Auxl -Nuc-M 


H-Q-D-Ma 


H-D-Q 


Aux 1 -Nuc- Aux2-M 


H-D-Q-Ma 


H-Mi-Q-D 


Aux 1 -Nuc-M- Aux2 


H-Mi-Q-D-Ma 


H-Mi-D-Q 




H-Mi-D-Q-Ma 


H-Q-Mi-D 




H-Q-Mi-D-Ma 







A. Phrase Identification 

To perform the phrase identification, word segmentation 
and part of speech (POS) tagging are performed using 
SWATH [15] which extracts words and classifies them into 44 
types such as common noun (NCMN), active verb (VACT), 
personal pronoun (PPRS), definite determiner (DDAC), unit 
classifier (CNIT) and negate (NEG). A hidden markov model 
(HMM) [14] employs these POS tag categories to determine 
phrases. The model assumes that at time step t the system is in 
a state PC (t) which has a probability of emitting a particular 
visible state POS tag (t), the transition probabilities a 4 j among 



97 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



hidden states and b jk for the probability of the emission of a 
visible state: 



a l(p =n(nxjT+i)inx l (T)). 



P 9K = TtCrayMinX^T)). 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
Given a sequence of visible state tag T , the decoding is to find 
the most probable sequence of hidden states. This research 
uses Virterbi [14] to calculates p(tag T ) of visible 
POS states recursively because each term p(tag(t)IPC(t)) 
p(PC(t)IPC(t-l)) involves only tag(t), PC(t) and PC(t-l) by 
defining follows: 



(1) 



(2) 



where PC (t) is Phrase Constituent at time step t, tag(t) is POS 
tag at time step t. 

The transition probability for the hidden states where a 
particular sequence PC T = {PC (1), PC (1), ..., PC(T) } of T 
hidden states can be written as: 



(0, t = and) ^ initial state 
1, t = and j = initial state 
arg max t St-^Oa^bj^ otherwise, 



(6) 



where b jkt represents the transition probability b jk selected by 
the visible state emitted at time t. Thus the only nonzero 
contribution to the arg is for index k which matches the visible 
state tag(t). 



p(PC T ) = Y\p(PC(t)\PC(t-l)) 



(3) 



The hidden state and the corresponding visible state where the 

model generated the particular sequence of T visible POS tag 

state tag T can write as: 

T 



p(tag T \PC T ) = Y\p(tag(t)\PC(t) 



(4) 



The probability that model produces a sequence tag of visible 
Pos tag states is 



p(tag T ) = 

T 

argmaxV\p(tag(t)\PC(t)MPC(t)\PC(t - 1)) 



(5) 



The expression, p(PC(t)IPC(t-l)) is the probability of PC(t) 
given the previous PC(t-l), and p(tag(t)IPC(t)) is the 
probability of POS tag(t) given the Phrase Constituent(t). 

This research use Baum- Welch [14] learning to determine 
model parameters, the transition probabilities a^ and b jk , from 
an ensemble of training samples. 





Fig. 3. A phrase model. 



Output Start ^ H <^ Auxl <^ Nuc 



Fig 4. The result of Viterbi tagging on the phrase model. 

Figure 3 shows a phrase model of string '^oimnjolomiMo 
muu m?is0i'toj ,l l# < w002 ^tfu03^0oAvm^oiiru" (A friend's going to 
borrow this book. Because she (0^ hasn't been able to buy it 
(0 2 ). Therefore she (0 3 ) must borrow it from me.) which POS 
tags of string is "mmi (A Friend-NCMN) u^o (is going to- 
XVMM) Su (borrow-VACT) tnMo (book- NCMN) m'u 
(numerative-CNIT) if (this-DDAC) mns (Because-CONJ) no 
(she(Oi)- PPRS) lii (hasn't been-NEG) 1* (able to-XVMM) 4 
(buy-VACT) flu (it(0 2 )) ^(Therefore: CONJ) no (she(0 3 ): 
PPRS) Mm (must: XVMM) mj (borrow: VACT) *nMo(book: 
NCMM) $u (me: PPRS)". The hidden state of a phrase model 
consists of H(NCMN- book (2/4), friend (1/4); PPRS- 
me(l/4)), D (CNIT- numerative (1/2); DDAC- this(l/2)), 
Discourse-marker (CONJ- because(l/2), therefore (1/2) ), 
Auxl (XVMM- is going to(l/4), must(l/4), able to(l/4); NEG- 
hasn't been (1/4)) and Nuc (VACT- borrow (2/3), buy (1/3)). 
The hidden state of Thai EDU model consists of S (H- friend 
(1)), O (H- book (2/4); D- numerative, this (2/4)) I (H- me 
(1)), Discourse-marker (marker- because, therefore(l)) and Vt 
(Auxl- must, is going to, able to, hasn't been (4/7); Nuc- 
borrow, buy (3/7)). 



98 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



B. EDU Boundary Determination 

After we determine NP and VP, another HMM on EDU 
constituents (shown in Fig. 5.) is then created to determine the 
starting and the ending of boundaries of EDUs. This model 
can handle the subject and object absence problems, discussed 
earlier. 

Fig. 5 shows an example of the EDU segmentation model 
of an EDU ''wQu-^njQ-mj-tmMQ-m'ij-u'' (A friend's going to 
borrow this book.) The EDU segmentation model can be 
expressed as: 



p{tag T ) = 



arg max T^P(tag(t)\EDUC(t))p(EDUC(t)\EDUC(t - 1)) 



(7) 



where EDUC(t) is EDU Constituent at time step t, tag(t) is 
Phrase tag at time step t. 

The expression, p(EDUC(t)IEDUC(t-l)) is the probability 
of EDUC(t) is EDU constituent at time t given the previous 
EDUC(t-l) and p(tag(t)IEDUC(t)) is the probability of Phrase 
tag(t) given the EDU Constituent (t). 




Fig. 5. A HMM ofThaiEDU. 



BsnjQ 



Start 


1 K 





S 


1 


1[1/6*1] 





| 





I 


1 


\ 


Uarke 





1 1 [2/6*0] 


Vt 


o N 


V 1 [3/6*0] 


End 










Output 



< 



<7 Vt ^Vt^O ^O^lO^l 



Fig. 6. The result from Viterbi taged on EDU segmentation model. 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
C. EDU Constituent Function 
Determination 

After we determine EDU constituent by using EDU 
segmentation model, rule base in Table 1 and Viterbi 
algorithm are then used to determine the grouping of function 
in EDUs. Evaluation of EDU constituent function 
determination is process which insert EDU constituent 
function tag for using in RS tree step. For example a string 
"m9i4-D^ s u9-^3j- c H^^9-m , u-y" (A friend's going to borrow this 
book.), the result from Viterbi tagged on the EDU 
segmentation model is shown as: S, Vt, Vt, O, O, O 

Rules in Table 1 are used to group constituents into 
appropriate functions.. For example, the rule based "NP -NP S - 
Vt-NPo" is used with label "S, Vt, Vt, O, O, O" because this 
label consists of Vt. Therefore, the result from Evaluation of 
EDU constituent function determination is shown as: NP S - (V, 
V) t - (NP, NP, NP) 

V. Thai RS Tree Construction 

In this section, we describe our technique to construct an 
RS tree from a corpus based on rules derived from Thai 
linguistic characteristics. The rules are classified into three 
types which are Absence, Repetition and Addition [1, 3, 17, 
18, 19]. Given a pair of EDUs, an author may write by using 
any or all of the three rules. The model calculates an EDU- 
EDU similarity matrices based on these sets of rules. Finally, a 
hierarchical clustering employs these matrices to build a 
rhetorical structure tree. 



A. Linguistic Rules for EDU Relations 

Absence rules In Thai language, it has been observed that 
frequently in writing some constituents of EDUs may be 
absent while the meanings are still the same. In the example 
below, the NP (object) "nmu" (dessert) is absence from the 
anaphoric EDU according to rule, i.e., rule O 

(O, O). 
Cataphoric EDU (VtO) : eennsjsvhmujl'mj (Would you like to 

make dessert?) 
Anaphoric EDU (Vt) : eenfmvh (Yes, I like to make.) 

Repetition rules It has been observed that frequently the 
anaphoric EDU is related to its cataphoric EDU by repetition 
of NP (subject, object), preposition phrase (PP) where it 
functions as a modifier of a nucleus or a verb phrase (VP). In 
the following example, two EDUs relate by a repetition of NP 
(object) "irm" (House), i.e., rule a (O, O). 

Cataphoric EDU (VttOI) : mimsnnejiJTum (I'm going to sell 

him a house.) 
Anaphoric EDU (VtO) : u^iiuinim^'ttm (Which house are 

you going to sell?) 

Addition rules It has been observed that frequently the 
anaphoric EDU is related to its cataphoric EDU by addition of 
discourse marker, and possibly accompanied by Absence and 



99 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 



Repetition rules. In the example below Discourse Marker 
'W w (because) is added in front of the anaphoric EDU, i.e., 
rule J\ (Marker, Before). 

Cataphoric EDU (VttOI) : ^umimzunnm (I like to borrow 

films.) 
Anaphoric EDU (VtO) : irmswiqfo'hflfli (Because have not 

been able to buy it.) 

Table 3 lists Repetition, Absence, and Addition rules, a (S, 
S) means that subject of the anaphoric EDU is repeated in the 
cataphoric EDU. O (S, S) means that subject is absence from 
the anaphoric EDU which subject appeared in the cataphoric 
EDU. Jl (Marker, Before) means that Discourse Marker is 
added in front of the anaphoric EDU. 

B . ED U Sim ila rity Sco ring 

A similarity score between two EDUs is calculated from 
contents of the EDUs, based on the three types of rules. A 
feature vector consists of Subject, Absence of Subject, Object, 
Absence of Object, Preposition, Absence of Preposition, 
Nucleus, Modifier Nucleus, Head, Absence of Head, Modifier 
Head, Absence of Modifier Head, Marker Before, and Marker 
After elements, each corresponding to a rule in the rule sets. 
The value of an element is dependent upon the type of rule, as 
follows: 

Features based on absence rules: 

A feature is generated from each absence rule with a value 
of: 
// (CI, C2) is true then 

\01 
value = 1 



021 



Total number of sentences 
Else, value = 



(8) 



where Ol is the order of cataphoric EDU, 02 is the order of 
anaphoric EDU, CI is constituent of cataphoric EDU, C2 is 
constituent of anaphoric EDU. 

Given the following example: 



Vol. 7, No. 1, 2010 
Table 3: Repetition, Absence, and Addition rules 



Repetition ( .H) 


Absence ( O) 


Addition ( J\) 


a (S, S) 


<J> (S, S) 


J\ (Marker, After) 


a (0, S) 


O (0, S) 


J\ (Marker, Before) 


a (S, 0) 


<J> (S, 0) 


J\ (Key Phrase, After) 


a (0, 0) 


O (0, 0) 


J\ (Key Phrase, Before) 


a ((S, S), (S, 0)) 


O ((S, S), (S, 0)) 




a ((S, 0), (S, 0)) 


O ((S, 0), (S, 0)) 




a ((0, S), (S, 0)) 


O ((0, S), (S, 0)) 




a ((0, 0), (S, 0)) 


((0, 0), (S, 0)) 




a (S, Prep) 


(Only H, H) 




a (0, Prep) 


((H, M), H) 




a (Prep, S) 


((H, M), M) 




a (Prep, 0) 


((H, M), (H, M)) 




a ((S, Prep), (S, 0)) 


(S, Prep) 




a ((0, Prep), (S, 0)) 


(0, Prep) 




a ((Prep, S), (S, 0)) 


(Prep, S) 




a ((Prep, 0), (S, 0)) 


(Prep, 0) 




a ((S, S), (S, Prep)) 


((S, Prep), (S, 0)) 




a ((S, 0), (S, Prep)) 


((0, Prep), (S, 0)) 




a ((S, Prep), (S, Prep)) 


((Prep, S), (S, 0)) 




a ((0, S), (S, Prep)) 


((Prep, 0), (S, 0)) 




a ((0, 0), (S, Prep)) 


0((S,S),(S,Prep)) 




a ((0, Prep), (S, Prep)) 


((S, 0), (S, Prep)) 




a ((Prep, S), (S, Prep)) 


0((S, Prep), (S, Prep)) 




a ((Prep, 0), (S, Prep)) 


((0, S), (S, Prep)) 




a ((Prep, Prep), (S, Prep)) 


((0, Prep), (S, Prep)) 




a ((S, S), (0, Prep)) 


((Prep, S), (S, Prep)) 




a ((S, 0), (0, Prep)) 


((S, S), (0, Prep)) 




a ((S, Prep), (0, Prep)) 


((0, S), (0, Prep)) 




a ((0, S), (0, Prep)) 


((S, Prep), (0, Prep)) 




a ((0, 0), (0, Prep)) 


((S, 0), (0, Prep)) 




a ((0, Prep), (0, Prep)) 


((0, 0), (0, Prep)) 




a ((Prep, S), (0, Prep)) 


((0, Prep), (0, Prep)) 




a((Prep, Prep), (0, Prep)) 


0((S, 0, S), (S,0, Prep)) 




a((S, 0, Prep), (S, 0, 
Prep)) 


0((S, S, 0), (S,0, Prep)) 




a (Only H, Only H) 






a (H, M) 




a (Only M, Only Nuc) 




a (Only M, Only M) 




a ((Nuc, M), (Nuc, M)) 





In the example, the properties of EDU 1 and EDU2 match 
with the rule "The anaphoric EDU2 is related to its cataphoric 
EDU1 by absence of subject", i.e., O (S, S), with the absence 
of subject 'Vmrm" (Villager). 



[1-21 

Value EDUl,Subject — 1 o 



— Value EDU2 Absence of Subject 



(9) 



EDU1: <mirm (/Subject) ibsneu (/Nucleus) ^trmfmulu 

fiieufifi (/Object) (The villager performs the 

family-industry.) 
EDU2: ims (/Before) (/Subject, Absence of subject) mmnu 

(/Nucleus) ttuijmjQ^™ (/Object) (and saves property 

of the nation.) 
EDU3: Qpmi'Hfmul'ufnQimfi (/Subject) Imilu (/Nucleus) tfuTji 

nja^ii (/Object) (Therefore, the family- industry is a 
property of the nation.) 



Features based on repetition rules: 

A feature is generated from each repetition rule with a 
value of: 



// r (CI, C2) is true then 

\01-02\ 



value = 1 



Total number of sentences 

Total of repeating words 



(10) 



Total of words in sentences 



Else, values = 



where Ol is the order of cataphoric EDU, 02 is the order of 



100 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



anaphoric EDU, CI is constituent of cataphoric EDU, C2 is 
constituent of anaphoric EDU. 

In the example, the properties of EDU1 and EDU3 match 
with the rule "The anaphoric EDU is related to its cataphoric 
EDU by repetition of subject", i.e., a (O, S), with the absence 
of object "qflflrmfniulufiioufifa" (Family-Industry). 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
Addition rule: 



Value EDJJ i object — va l ue EDU3, Subject 



= d-^)*(H) 



en) 



Features based on addition rules: 

A feature is generated from each addition rule with a value 
of: 

If /J (M, L) is true then 

values = 1 
Else, value = 

Where M is Marker of EDU, L is link. 



(12) 



In the example, the properties of EDU1 and EDU2 match 
with the rule "The anaphoric EDU is related to its cataphoric 
EDU by addition of discourse marker "mz" (AND) ", i.e., fl 
(Marker, Before), link EDU2 with in front of EDU2 (EDU1). 
If addition rule is true, value will be assigned 1. But if addition 
is false, value will be assigned 0. 



Similarity Calculation 

Similarity between two EDUs (cataphoric and anaphoric) 
can be calculated as: 



ZijuiXapizy/ S k {MR t j) = 



max k (MR t j) -S k (MR t j) 
max k (MRij) -min k {MRij) 



(14) 



where S k (k \in {1, 2, 3}) is the normalized matrix whose 
values in matrix range from one (for the inter-EDU ranked 
best) down to zero (for the inter-EDU ranked worst), Matrix 
rule (MR) is a matrix from absence rule class, repetition rule 
class, and addition rule class, respectively; and Max's and 
Min's are taken over all occurring ranks for each rule [7]. The 
input score are normalized into range [0, 1] before 
computation takes standard combination algorithm [7]. Sj can 
be calculated separately for each type of rule, as follows: 

Absence and repetition rules: 

MP l(p = \Magnitude^ t ^ v oric * M agnitude^y P oric \ (15) 

where i and j are the order of EDU and li-jl < MD 

These rules consist of two parts (Cataphoric and 
Anaphoric). If two parts of absence and repetition rules are 
true, then absence and repetition rules are true. But if one part 
of absence and repetition rules is false, then absence and 
repetition rules are false. 



MP l(p = \Magnitude c E l t ^ horic + Magnitude j 



Anaphoric , 
EDUj I 



(16) 



Where i and j are the order of EDU and li-jl < MD 



This rule consists of two parts (Cataphoric and Anaphoric). 
If one part of addition rule are true, then addition rule are true. 

The similarity above is used with a disjunctive hypothesis, 
where MD is the maximum distance of appropriate disjunctive 
hypothesis. Essentially, the disjunctive hypothesis enumerates 
relations of EDUj over number of the Cartesian product {i, i + 
1, ..., i + MD + 1} X { j,j + 1, ...,j +MD+ 1}, i.e., all the 
pairs of EDUs that separated by an imaginary line drawn 
between EDUj and EDUj. The maximum distance 4 is the 
appropriate of disjunctive hypothesis. 

VI. Thai RS Tree Construction 

The methods used in this part are a hierarchical clustering. 
Because results from hierarchical clustering can be represent 
relation between EDUs in binary tree. 

Each sample (an EDU in this case) begins in a cluster of its 
own and while there is more than one cluster left. The two 
closest clusters are combined into a new cluster and the 
distance between the newly formed cluster and each other 
cluster is calculated. The method of hierarchical clustering is 
shown in Table 4 and the result from hierarchical clustering is 
shown in Fig. 7. 

Table 4. The hierarchical clustering method. 



Clustering Method 


Distance Between Clusters A and B 


Single Linkage 


The smallest distance between a sample 
in cluster A and a sample in cluster B 


Un weighted Arithmetic 
Average 


The average distance between a sample 
in cluster A and a sample in cluster B 


Neighbour Joining 


A sample in cluster A and a sample in 
cluster B are the nearest. Therefore, 
define them as neighbours. 


Weighted Arithmetic 
Average 


The weighted average distance between 
a sample in cluster A and a sample in 
cluster B. 


Minimum Variance 


The increase in the mean squared 
deviation that would occur if clusters 
A and B were fused 



VII. Thai DR Recognition 

In this section, we describe our technique to recognize Thai 
DRs from a corpus based on rules derived from Thai linguistic 
characteristics. The rules are classified into three types which 
are Absence, Repetition and Addition [1, 3, 17, 18, 19]. A 
Decision tree (C5.0) employs these features to recognize a 
DR. 

A. Linguistic Rules for DR Recognition 

A feature score of DRs is calculated from contents of the 
EDUs, based on the three types of rules. A feature score 
consists of Cataphoric score (Subject, Object, Preposition, 
Nucleus, Marker Before and Marker After) and Anaphoric 
score (Subject, Absence of Subject, Object, Absence of 



101 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



Object, Preposition, Absence of Preposition, Nucleus, 
Modifier Nucleus, Head, Absence of Head, Modifier Head, 
Absence of Modifier Head, Marker Before, and Marker After) 
elements, each corresponding to a rule in the rule sets. The 
value of an element is dependent upon the type of rule, as 
follows: 

Features based on absence rules: 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
In the example, the properties of EDU1 and EDU3 match 
with the rule "The anaphoric EDU is related to its cataphoric 
EDU by repetition of subject", i.e., a (O, S), with the absence 
of object "qwfrmfiiiuluflieijflfj" (Family-Industry). 



A feature is generated from each absence rule with a value 



of: 



value } 



H,x = 1 



EDU1, Obj, 



Value EDU2t Subject - 1 



(20) 



Features based on addition rules: 



If 0(C1,C2) is true then 



A feature is generated from each addition rule with a value 



Value, 
Value, 



ELSE, 



Cataphoric ' 
Anaphoric = 



Value c a 
Value A „ 



1 

"Omit" =2 



-1 



(17) 



where CI is constituent of cataphoric EDU, C2 is constituent 
of anaphoric EDU. 

Given the following example: 

EDU1: <mirm (/Subject) ibsneu (/Nucleus) ^trmfmulu 

fiiQiifrh (/Object) (The villager performs the 

family-industry.) 
EDU2: uns; (/Before) (/Subject, Absence of subject) mmnu 

(/Nucleus) truijniiB^'jfiw (/Object) (and saves property 

of the nation.) 
EDU3: qflfTTHfimjliiflTBijflfj (/Subject) t*n\u (/Nucleus) ttjjijfl 

njQ^mi (/Object) (Therefore, the family-industry is a 

property of the nation.) 

In the example, the properties of EDU1 and EDU2 match 
with the rule "The anaphoric EDU2 is related to its cataphoric 
EDU1 by absence of subject", i.e., O (S, S), with the absence 
of subject 'Vmjhu" (Villager). 



ValUeEDUl, Subject — 1 



Value EDU2i Absence of Subject = "Omit"=2 



(18) 



Features based on repetition rules: 

A feature is generated from each repetition rule with a value 
of: 

If h (C1,C2) is true then 

Value Cataphoric = 1 



Else, 
Value c a 



(19) 



= 1 



vaiue^ na p nor j C — -i 



where CI is constituent of cataphoric EDU, C2 is constituent 
of anaphoric EDU. 



of: 

Ifff (M, L) is true then 

valuesi = M 
Else, value i = -1 

where M is Marker of EDU, L is link. 



(21) 



In the example, the properties of EDU1 and EDU2 match 
with the rule "The anaphoric EDU is related to its cataphoric 
EDU by addition of discourse marker "ims" (AND) ", i.e., fl 
(Marker, Before), link EDU2 with in front of EDU2 (EDU1). 
If addition rule is true, value will be assigned 1. But if addition 
is false, value will be assigned 0. 



value EDU1> 



Before 



= "AND" = value 



EDU2, Before 



(22) 




andEDU2 Anap horic by feature of EDUlcmph. 
iftcrl and EDU2 AnaDh 



Fig.7.RlisDRofEDUl C ataphoric 

consists of values {S, O, Prep, Nuc, Marker bei 
consists of values { S, O, Prep, Nuc, Marker before or after I Repetition 0S, 00, 0Prep 
I Absence Marker be f re or after Udditkm }, respectively and R2 is DR of Rlcataphoric and 
EDU3 Ana phoric by Rl consists of values {S, O, Prep, Nuc, Marker be fore or after 
}edui+edu2 and EDU3 Anap horic consists of { S, O, Prep, Nuc, Marker before or after I 

Repetition 0S, 00, 0Prep I Absence Marker before or after Uddition } , respectively. 

VIII. Experimental Evaluation 

A. Evaluation of Thai EDU 
Segmentation 

In order to evaluate the effectiveness of the EDU 
segmentation process, a consensus of five linguists familiar 
with rhetorical structures of Thai texts is used to manually 



102 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



segment EDUs of Thai family law which consists of 10,568 
EDUs in total. 

The EDU segmentation model is trained with 8,000 random 
EDUs, and the rest are used to measure performance. 

| The correct FRS tree 



-G> 




(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
EDUs). The Absence data contains EDUs mostly following 
absence rules while the Repetition data contains those mostly 
following repetition rules. The analysts created training and 
testing data sets by manually building EDU constituents and 
inserting appropriate tags into each text. 

The results of the Viterbi algorithm are labels of EDU 
constituents, then rules based on NP and VP are applied to 
group EDU constituents and insert function tags. 

Table 5 : Performance of determining EDU constituent functions 



| Minimum Variance | 




| Un Weight ed Arithmetic Average | 



10' - 9' - 



r^ 



EDU1 


4' ~T~^\ EDU * 


I EDU3 


EDU4 

^-^EDUS 


EDU6 
r— <B>(EDU7 


EDUS 



Fig. 8. Results from two hierarchical clustering algorithms 

The training proceeds till the estimated transition 
probability changes no more than a predetermined value of 
0.02 or the accuracy achieves 98%. 

The performances of both phrase and EDU segmentations 
are evaluated using recall (Eq. 23) and precision (Eq. 24) 
measures, which are widely used to measure the performance. 
Recall = 

# correctly s Thai (Phrases; EDUs) identified by using model 

# Thai (Phrases; EDUs) identified by analysts \-^3) 



Precision = 

# correctly s Thai (Phrases; EDUs) identified by using model 
Total number of Thai (Phrases; EDUs) identified by using HMM 



(24) 



Rules 


Absence data 


Repetition data 


Family law data 


NPs-Vi-NPs 


NP S (100%) 


NP S (100%) 


NP S (100%) 


NPo -NP S -Vt- 
NP 


NPs & NPo 
(100%) 


NPs &NP 
(100%) 


NP S &NP 
(100%) 


NPS-Vtt- 
NPO-NPI 


NPs &NP &NPi 
(100%) 


NPs &NP0&NP1 
(100%) 


NPs &NP0&NP1 
(100%) 


NPo -NP S - 
Vtt-NPi 

NPi-NP s -Vtt- 
NP 


NP S (100%), 

NPo&NP, 

(91.37%) 


NP S (100%), 
NPO&NP, 

(79.59%) 


NP S (100%), 

NPo&NP, 

(90.21%) 


N-N 


NP N (100%) 


NP N (100%) 


NP N (100%) 



Table 5 shows the results of grouping EDU constituents 
(Subject (S), Object (O), Indirect Object (I) and Nomen (N)) 
by using rules based on NPs assuming the position of verb 
phrases (Vi, Vt and Vtt) are known. In general, all rules except 
NP -NP S -Vtt-NPi and NP r NP s -Vtt-NP perform well. 

To resolve ambiguities with these two rules, a probability 
table of words in positions of NP r and NP coming after 
position of Vtt (P(Vttl NP T , NP )) is used. The results of 
evaluating EDU constituents by using rules based on NP 
together with the probability table yield higher performance of 
NP and NP T in absence data (92.24%), Repetition data 
(85.78%) and Family law (93.71%). 

C. Evaluation of Thai RS Tree 
Construction 

In order to evaluate the effectiveness of the Thai RS Tree 
construction process, linguists manually built the rhetorical 
structure trees of three texts which are Absence, Repetition 
and Family Law data sets, with a total of 568 EDUs. The 
performances of the rhetorical structure were evaluated by 
using recall and precision measures, i.e., (Eq. 25) and (Eq. 26) 
[6], respectively. 



The results show that the proposed method achieves the 
recall values of 84.8% and 85.3%; and the precision values of 
93.5% and 94.2% for phrase and EDU segmentations, 
respectively. 

B. Evaluation of EDU Constituent 
Function Determination 

In order to evaluate the effectiveness of grouping EDU 

constituents and determining their functions in an EDU using 

the proposed HMM model on NP and VP, the experiments 

employ three texts (Absence data with 84 EDUs, Repetition 

data with 117 EDUs and a subset of the Family law with 367 



Recall 

#correctlys internal nodes identified by RST (25) 

^internal nodes identified by analysts 
Precision 

Ucorrectlys internal nodes identified by RST (26) 

~ Total number of internal nodes identified by RST 

In the Absence and Repetition data sets, though relations 
between EDUs follow mostly Absence rules and Repetition 
rules, respectively, many EDUs also follow other kinds of 
rules. For example, 

Anaphoric EDU (SvtO) : l^ulikwni (S) ^flVufmn (Vt) mnmu 
(a O) (A Postman will select 



103 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 



letters) 
Cataphoric EDU ((S)VtO) : uns (A) (O S) nsfutf<i (Vt) mthjiu 
(a O) (And will receive and 
send letters) 

Recall and precision are calculated with respect to the 
ability of an algorithm to construct an RS tree structure similar 
to that created by the linguists. 

Table 6 shows example calculations of recall and precision 
of Thai RS Trees on a text created by the Minimum Variance 
and Unweighted Arithmetic Average algorithms in Fig. 8. 

Table 7 shows the results of evaluating Thai RS Tree 
construction on the three data sets. The performance on the 
Family law which combines many kinds of rules in its content 
is 94.90% recall and 95.21% precision. The results also show 
that Unweighted Arithmetic Average clustering algorithm 
gives the best performance for Thai RS Tree construction. 

A. Evaluation of Thai DR Recognition 

In order to evaluate the effectiveness of the Thai DR 
recognition process, linguists manually built the DRs of 
Family Law data sets from RS tree, with a total of 1,248 
EDUs. The DR model is trained with 828 random EDUs, and 
the rest are used to measure performance. The performances of 
the DR recognition were evaluated by using analysis module 
of Clementine version 12.0. 

Table 8 shows the results of evaluating ten Thai DRs 
recognition on the data sets. The performance on the Family 
law which combines many kinds of rules in its content, found 
marker and not found marker is 82.50%, 85.09% and 81.28%, 
respectively. 

Table 9 shows the results of comparison found marker and 
not found marker to recognize ten Thai DRs on the data sets. 
The performance on the Family law which found and not found 
marker in its content is 99.40% and 100%, respectively. 

IX. Conclusions 

Thai rhetorical structure tree (RST) construction is an 
important task for many textual analysis applications such as 
automatic text summarization and question-answering. This 
article proposes a novel two-step technique to construct Thai 
RS tree combining machine learning techniques with linguistic 
properties of the language. 

First, phrases are determined and then are used to segment 
elementary discourse units (EDUs). The phrase segmentation 
model is a hidden markov model constructed from the possible 
arrangements of Thai phrases based on part-of- speech of 
words, and the EDU segmentation model is another hidden 
markov model constructed from the possible arrangements of 
Thai EDUs. 



Vol 7, No. 1, 2010 

Table 6: Computing the performance of rhetorical structure tree 
(P=Precision; R=Recall). 



The correct 
RS tree 


Minimum Variance 


Unweighted Arithmetic 
Average 




3' 


3' 




4' 


4' 




r 


1' 




9' 


6' 




2' 


2' 




5' 


5' 




6' 






7' 






8' 








7' 






8' 






9' 






10' 




P=9/9 


P=6/10 




R=9/9 


R=6/9 



As a side effect, functions of the EDU constituents can be 
determined by the EDU segmentation model together with 
linguistic rules grouping related constituents into a large unit. 
Experiments show the EDU segmentation effectiveness of 
85.30% and 94.2% in recall and precision, respectively. 

A hierarchical clustering algorithm with EDU similarity 
derived from semantic rules of the language is proposed to 
construct an RS tree. The technique is experimentally 
evaluated and the effectiveness is 94.90% and 95.21% in 
recall and precision, respectively. 

Table 7: The results from RS tree experiment. 



Data 


N. 


Clustering Method 


Recall 


Precision 


Absence 


84 


Neighbour Joining 


87.23 


89.13 






Single Linkage 


82.97 


84.78 






Un weighted Arithmetic 
Average 


87.23 


89.13 






Minimum Variance 


89.40 


91.30 






Weighted Arithmetic 
Average 


87.23 


89.13 


Repetition 


117 


Neighbour Joining 


89.70 


91.04 






Single Linkage 


83.82 


85.07 






Unweighted Arithmetic 
Average 


89.70 


91.04 






Minimum Variance 


77.94 


79.10 






Weighted Arithmetic 
Average 


89.70 


91.04 


Family- 


367 


Neighbour Joining 


85.98 


86.26 


Law 




Single Linkage 


64.01 


64.21 






Unweighted Arithmetic 
Average 


94.90 


95.21 






Minimum Variance 


63.37 


63.57 






Weighted Arithmetic 
Average 


90.44 


90.73 



104 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



Table 8: The results from DR recognition experiment. 



Data 


N. 


Balance node (boost) 


Correct 


Family law 


367 


1244 


82.81% 


Not found 
Marker 




715 


81.28% 


Found marker 




529 


85.09% 



Table 9: Comparison DR of found and not found marker. 



DR 


Correct 


Correct 




Not found Marker 


found Marker 


flfifieiismjj (consent) 


93.10% 


98.10% 


pTtbetn (example) 


52.40% 


54.00% 


mwmzQt (characteristic) 


69.40% 


99.30% 


flfilpmjj (summary) 


96.10% 


- 


wfliAii (condition) 


59.60% 


85.30% 


Jfirufn (option) 


97.70% 


99.40% 


nan (time) 


62.50% 


90.50% 


iwwwa (reason) 


90.80% 


91.20% 


ii^niw (explanation) 


100.00% 


- 


iipinpi"u (contrast) 


92.00% 


98.90% 



A decision tree (C5.0) algorithm with DR features derived 
from semantic rules of the language is proposed to recognize a 
DR of EDUs in RS tree. The technique is experimentally 
evaluated by setting boosting 10 number of trials, pruning 
severity 75, maximum record per child branch 2 and winnow 
attributes and the effectiveness is 82.81%. 

References 



[1] D. Chamnirokasant, "Clauses in the Thai Language". Unpublished 
master's thesis, Chulalongkorn University, Thailand, 1969. 

[2] D. K. Harman, editor. "The second Text Retrieval conference (TREC- 
2)", Gaithersburg, MD, USA, March 1994. U.S. Government Painting 
Office, Washington D.C, 1994. 

[3] D. Mahatdhanasin, "A study of sentence groups in Thai essays". 
Unpublished master's thesis, Chulalongkorn University, Thailand, 1980. 

[4] D. Marcu, "Build Up Rhetorical Structure Trees". In American 
Association for Artificial Intelligence, 1996. 

[5] D. Marcu, "A decision-based approach to rhetorical parsing". In the 37th 
Annual Meeting of the Association for Computational Linguistics, ACL, 
Maryland, pp. 365-372 , 1999. 

[6] D. Marcu, "The theory and Practice of Discourse Parsing and 
Summarization". The MIT Press, Cambridge, MA, 2000. 

[7] E. A. Fox, and J. A. Shaw, "Combination of multiple search". In 
Harman [2], pages 243-249, 1997. 

[8] J. Charoensuk, and A. Kawtrakul, "Thai Elementary Discourse Unit 
Segmentation by Discourse Segmentation Cues and Syntactic 
Information". In the Sixth Symposium on Natural Language Processing 
2005 (SNLP 2005), Chiang Rai, Thailand, December 13-15, 2005. 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
[9] L. Alonso, and I. Castellon, Towards a delimitation of discursive 
segment for Natural Language Processing applications, International 
Workshop on Semantics, Pragmatics and Rhetorics, San Sebastian 22-24 
November, 2001. 

[10] L. Polanyi, "A formal model of the structure of discourse". Journal of 
Pragmatics, 12,601-638, 1988. 

[11] M. Wattanamethanont, T. Sukvaree, and A. Kultrakul, "Discourse 
relation recognition by using Naive Bayesian classifier", The 9 th 
National Computer Science and Engineering Conference (NCSEC 
2005), University of the Thai Chamber of Commerce (UTCC), Bangkok, 
Thailand, October 27-28, 2005. 

[12] R. Soricut, and D. Marcu, "Sentence Level Discourse Parsing using 
Syntactic and Lexical Information". In Proceedings of the 2003 Human 
Language Technology and North American Association for 
Computational Linguistics Conference (HLT/NAACL), May 27-June 1, 
Edmonton, Canada, 2003. 

[13] R. Subba, , B. Di Eugenio, S. N. K. "Learning FOL rules based on rich 
verb semantic representations to automatically label rhetorical relations 
(EACL)", In Workshop on learning Structured Information in Natural 
Language Applications , 2006. 

[14] S. Levinson, R. Rabiner, and M. Sondhi, "An introduction to the 
application of the theory of probabilistic function of a Markov proceeds 
to automatic speech recognition". Bell System Technical Journal, 
62:1035-1074, 1983. 

[15] T. Charoenporn, V. Sornlertlamvanich, H. Isahara, "Building A Large 
Thai Text Corpus— Part-Of-Speech Tagged Corpus: ORCHID—". In 
proceedings of the Natural Language Processing Pacific Rim 
Symposium, 1976. 

[16] T. Sukvaree, J. Charoensuk, M. Wattanamethanont, and A. Kultrakul, 
"RST based Text Summarization with Ontology Driven in Agriculture 
Domain". Department of Computer Engineering, Kasetsart University, 
Bangkok, Thailand, 2004. 

[17] V. Panupong, Inter- Sentence Relations in Modern Conversational Thai. 
The Siam Society, Bangkok, 1970. 

[18] W. Aroonmanakun, "Referent Resolution for Zero Pronouns in Thai". 
Southeast Asian Linguistic Studies in Honour of Vichin Panupong. 
(Abramson, Arthur S., ed.). pp. 11-24. Chulalongkorn University Press, 
Bangkok. ISBN 974-636-995-4 , 1997. 

[19] W. Aroonmanakun, "Zero Pronoun Resolution in Thai: A Centering 
Approach". In Burnham, Denis, Interdisciplinary Approaches to 
Language Processing: The International Conference on Human and 
Machine Processing on Human and Machine Processing of Language 
and Speech. NECTEC: Bangkok, 127-147, 2000. 

[20] W. C. Mann, and S. A. Thompson, "Rhetorical structure theory. Toward 
a functional theory of text organization". Text, 8(3): 243-281, 1988. 

AUTHORS PROFILE 

Somnuk Sinthupoun is a teacher at the Department of Computer 
Science, Maejo University, Thailand. He holds a M.S. in Computer Science 
from National Institute of Development Administration (NIDA.), and a B. in 
Computer Science from Maejo University. His main research interests include 
artificial intelligence, information retrieval, data mining, and related areas. 

Ohm Sornil is an Assistant Professor at the Department of 
Computer Science, National Institute of Development 
Administration, Thailand. He holds a Ph.D. in Computer 
Science from Virginia Polytechnic Institute and State 
University (Virginia Tech), an M.S. in Computer Science from 
Syracuse University, an M.B.A. in Finance from Mahidol 
University, and a B.Eng. in Electrical Engineering from 
Kasetsart University. His main research interests include 
computer and network security, artificial intelligence, 
information retrieval, data mining, and related areas. 



105 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 



Mobility Impact on Performance of Mobile Grids 

A. S. Nandeppanavar 1 , M. N. Birje 2 , S. S. Manvi 3 , Shridhar 4 
l ' department of ISE, Basaveshwar Engineering College, Bagalkot-587102, India 
department of E&CE, Reva Institute of Technology & Management, Bangalore, India 
department of E&CE, Basaveshwar Engineering College, Bagalkot-587102, India 



Abstract — Wireless mobile grids are one of the emerging 
grid types, which help to pool the resources of several willing 
and cooperative mobile devices to resolve a computationally 
intensive task. The mobile grids exhibit stronger challenges like 
mobility management of devices, providing transparent access 
to grid resources, task management and handling of limited 
resources so that resources are shared efficiently. Task 
execution on these devices should not be affected by their 
mobility. 

The proposed work presents performance evaluation of 
wireless mobile grid using normal walk mobility model. The 
normal walk model represents daily motion of users and the 
direction of motion is mostly symmetric in a real life 
environment; thus it is effective in location updating of a 
mobile station and in turn helps task distribution among these 
available mobile stations. Some of the performance parameters 
such as Task Execution Time, task failure rate, communication 
overhead on Brokering Server and Monitoring Cost are 
discussed. 

Keywords- Mobile Grids; Normal Walk Mobility Model; 
Mobility management; Collaborative Problem Solving; Task 
Management 



I. 



Introduction 



Mobile grid computing is one of the emerging grid 
types, with two underlying technologies: Mobile Computing 
and Grid Computing. Mobile grids focus on resource 
sharing among wireless mobile devices for distributed 
applications and are characterized by relatively high 
mobility and limited resources. Common resources shared 
are: computing power, data storage/network file system, 
communication and bandwidth, application software. These 
mobile grids exhibit stronger challenges like mobility 
management of devices, providing transparent access to grid 
resources, and handling of limited resources so that 
resources are shared efficiently. Therefore it is necessary to 
develop an environment model which can represent the 
issues mentioned above. 

The work given in [1] focuses on wireless mobile grid 
for collaborative problem solving considering mobility 
related issues such as effect of mobility on performance of 
grid and network instability due to mobility. This mobile 
grid framework allows mobile devices to work 
collaboratively on computationally expensive tasks. Such a 
task is decomposed into smaller tasks and distributed across 
the other mobile devices willing to share their computational 
power with others. 



Mobility models are important parameters for location 
updating of Mobile Stations (MS). We consider Normal 
Walk Mobility Model [2] to represent mobility pattern of 
users and to decide their location. The work also focuses on 
finding the direction of movement of MS and to predict 
instantaneously the Base Station Controller (BSC) with 
which handover occurs. Performance is analyzed based on 
parameters like task execution time, task failure rate, 
communication overhead on Brokering Server (BS) and 
monitoring cost. 

The paper is organized as follows: Section II presents 
the related works on wireless mobile grid architectures, 
mobility models, mobility management and handovers in 
wireless mobile networks. Proposed work is described in 
section III. Section IV discusses about the results. And 
section V concludes the work. 

II. Related works 

Ian Foster et al have defined grid [3] as "flexible, secure, 
coordinated resource sharing among dynamic collection of 
individuals, institutions, and resources what we refer to as 
virtual organizations". They have highlighted the need for 
grid technology in virtual organization. 

T. Phan et al [4] have presented the challenge of 
integrating mobile devices with computational grid. The 
integration is provided through the use of an Interlocutor, 
which acts as proxy for cluster of Minions. There is no 
selection strategy to replace an interlocutor which has 
moved to another cell. 

Kurkovsky et al [1], [5], and [6] have proposed an agent 
based approach to the design of wireless grid architecture to 
solve computationally expensive tasks. This architecture 
enables mobile devices within a wireless cell to form 
computational grid. It has several limitations, one of which 
being the inadequate consideration for the mobility of the 
mobile agents. Tasks are indiscriminately aborted by 
Subordinates and/or Initiators whenever these mobile agents 
move to neighboring cells. 

P. Mudali et al [7] have proposed an extension to the 
architecture by Kurkovsky et al in [1]. They have proposed 
a multi-cell wireless computational grid, which is based on 
location area concept in GSM cellular networks. The 
proposed wireless computational gird is capable of greater 
device mobility tolerance than proposed in [1]. But still 



106 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



there is need to introduce mobility management schemes in 
proposed architecture. 



Ian F. Akyildiz [8] et al have proposed a simplified 
random walk model for hexagonal cell configuration where, 
probability states gives performance of the model. Further 
Guoliang Xue [9] and Md. Imdadul Islam and A.B.M. 
Siddique Hossain [10] have proposed improved models by 
reducing number of probability states to improve the 
performance. 

Chiu-Ching Tuan et al have proposed a novel normal 
walk model for PCS networks with mesh cell configuration 
in [11] and compact normal walk model for hexagonal cell 
configuration in [2]. They represent the daily mobility 
behavior of an individual mobile station that moves from a 
cell to another, in PCS networks. 

Jingyuan Zhang has described various location 
management schemes and mobility modeling method used 
for cellular networks in [12], and also provided the 
comparison of these methods. 

M. N. Birje [13] et al have proposed a prediction based 
handover model for multiclass traffic in wireless mobile 
networks by using software agents, considering two cases: 
local handoff (between BSC's connected to same mobile 
switching center (MSC)), and global handoff (between 
BSC's connected to different MSC). 

III. Proposed Work 

In this section we describe the architecture used in 
proposed work, mobility modeling, and location updating 
mechanisms. 

A. Architecture 

Figure 1 shows considered architecture. It includes two 
Virtual Organizations spanning over different Actual 
Organizations, which are monitored and controlled by a 
BSC. Its components are described briefly as follows: 

Virtual Organization (VO): is the one which spans multiple 
actual organizations and transcends greater amount of 
geographical, organizational, and other type's related to 
intellectual property rights and national laws. 

Actual Organization (AO): represents a single organization 
in a single place. 

Base Station Controller (BSC): provides service to number 
of AOs through BTS. It has two servers BS and BSMS. 

Base Transceiver Station (BTS): it supports the 
communication between BSC and mobile stations within 
AO. The BTS is fixed and is able to communicate with 
mobile stations using its radio transceiver. 

Mobile Station (MS): is nothing but the mobile node which 
the user is holding. MS within an AO can play two roles: 
Initiator and Subordinator. Any device which is ready to 
take part in problem solving in grid environment is called as 
Subordinator. Initiator is the one which initiates or requests 
a task to be solved. Any subordinate may become an 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
Initiator of distributed task if its user requests a large and/or 
computationally intensive task to be solved. In this case 
initiator is responsible for submitting such a distributed task 
to Brokering Server. 




Figure 1. Architecture 

Brokering Server (BS): it has the responsibility of task 
distribution and coordination in solving computationally 
intensive tasks. It knows the initial residing status of all 
devices in VO. Task distribution is based on available 
resources. BS has two data stores as below: 

Resource Repository (RR): keeps track of available 
resources. Details of currently available resources of all 
mobile stations are stored in Resource Repository. 

Job Allocation Database (JADB): keeps track of job 
distribution during task execution. Information about all 
subtasks allotted to different mobile stations and also results 
after performing operation are stored in this data base. 

Base Station Monitoring Server (BSMS): It supports 
communication among mobile stations in wireless grid. It 
also keeps track of all the mobile stations available in the 
wireless grid. Each node entering or leaving the wireless 
grid should inform Base Station Monitoring Server. It 
maintains information about the mobile station's current 
location, which in turn helps to predict handoff in advance. 



SUBORDINATE 



Brokering Server 




MOBILITY LOCATION 

MANAGEMENT 



RR-Resoiu ce Repositoi y 
JADB - Job Allocation Database 



Figure 2. Interaction Diagram 

Figure 2 explains the interaction among different 
components of the architecture. (1) Initiator sends the job to 



107 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



be solved to the BS; BS looks for eligible nodes in RR, so 
that sub jobs can be distributed to them. (2) Details of sub 
job and nodes to which jobs are to be distributed are stored 
in JADB. (3) BS distributes sub jobs to eligible subordinates 
in the grid. Subordinates solve the sub jobs and return their 
partial results to the BS, which is stored in JADB (4) the 
initiator collects the partial results from the BS. BSMS 
supports the communication among all the nodes and 
servers. It is also responsible for location and mobility 
management of the mobile node. 

B. Mobility modeling and location updating 

Proposed architecture considers Normal Walk mobility 
model described in [11] for modeling movements of mobile 
stations within Actual Organizations. Normal Walk mobility 
model is a multi-scale, straight-oriented, mobility model. It 
represents the daily mobility patterns of a mobile station and 
the direction of motion is mostly symmetric in a real life 
environment. The mobile station moves in one step. Each 
move is based on previous move and is obtained by rotating 
previous move by an angle of 6 in anticlockwise. This angle 
is called as moving angle, and helps to determine the next 
relative direction in which an MS (mobile station) moves 
across a cell in single step. 

The drift angle 6 in this model is a continuous random 
variable and its value helps to decide angle of movement of 
mobile station. The probability distribution of 6 is assumed 
to approach normality rather than randomization with two 
parameters: 

Mean (ju) with zero-degree. 

Standard Deviation (o) in the interval [5°, 90°] 

Varying o can redistribute the probabilities associated 
with 6 and in turn helps to make the movement patterns 
more realistic to represent the user mobility. Thus any 
movement yielded from this model is function of o, and is 
called a normal walk. The normal distribution of 6 is 
represented as: 

e~N(0°,o 2 ) (1) 

And probability density function of 6 is defined by: 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
The direction k is not absolute for a mesh cell, but is 
relative to the inlet that an MS is currently visiting. The 
range of each direction k is confined to lie between two 
fixed angles. The confining angles are calculated as: 



fiP> 



_±_ 



2 V <^ 



Incj 



-270<6k270 (2) 



This drift angle 6 also helps to determine one of the six 
relative directions in which an MS handoffs from a 
hexagonal cell (in figure 3) to another in next step. The six 
directions are indexed by 'k' and given below: 

Making U-turn or turning back (B, k=0). 

Turning right (R, k=l). 

Moving front-right (Fr, k=2) 

Moving front or forward (F, k=3). 

Moving front-left (Fl, k=4) 

Turning left (L, k=5). 



angF = tan 



'* ^ 



angFl = tan 



-1 



y 4Rij 



\R'J 



angL = tan (go) 



(3) 



where Ri and Ro represent inner and outer radii of 
hexagonal cell respectively. 

Thus angF = 16.1 , angFl = 49.1 and angL = 90 . 







90 " Uft --KJ.I' 


From-kfi 






)sO\ 




K / / ■--''' JT 


-V% 


Back I 
■-1 




\\ 


H J 


-i 


^*~ '^S 








^^ — ^ 








^ _ / / 




\^\ ^ 


^l y 






_ 9{r {R,k=Y) \-49.r 


(Fr, k=2) 



16.1" 



front 



16.1" 

-■.:■■ 



Figure 3. Hexagonal Cell 

To find the probability of each of the six directions at 
next step, 6 is standardized into Z (a continuous random 
variable), to find cumulative probabilities. Let Z= 6/o such 
that Z has a standard normal distribution, Z ~ N (0, 1). The 
probability distribution function (pdf) and cumulative 
distribution function (cdf) are respectively given by 

<p(z) = -== • e 



(4) 



<fi(z) = \ (p(co)dco 



Considering the equations (3), (4) the six probabilities 
associated with each direction k, with given o are given by: 

angF 



108 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



angL angFl 
I = (/)( ) - 0( ) 



angFl angF 



angL 



fr=fl 
r = l 



(5) 



Movement of mobile node follows normal walk model, 
where BS knows the initial location of the device. Location 
updating of the device during movement follows the 
algorithm below: 

1. Find the new location of the MS based on its current 
location and moving angle 

2. If the MS moves out of VO, but within same AO, 
then information update is done to reflect its new 
location 

3. If the MS moves out of VO, and also from local 
AO, then handover occurs 

C. Job Distribution and Execution 

BS of the proposed architecture is responsible for job 
handling, so that sub jobs of a given large job are distributed 
and partial results are collected back. BS has a data base 
called as JADB to keep track of job distribution among 
mobile devices within a wireless grid. It includes sub jobs 
distributed, identity of node to which jobs allotted and 
partial results. If a mobile device within wireless grid 
initiates a new job, the source code needed to run the 
corresponding job and all relevant parameters are submitted 
to the BS. BS generates sub jobs and assigns the sub jobs to 
available eligible subordinates, and waits to collect back the 
partial results. 

Steps used to distribute and solve a computationally 
expensive job in the wireless grid environment can be 
described as below: 

1. The user of a mobile device initiates a 
computationally intensive job 

2. The Initiator creates sub jobs and transmits them to 
Brokering Service 

3. The Brokering Service stores the received sub jobs 
in the JADB 

4. The Brokering Service uses heuristics to find a 
subordinate in the RR for each sub job 

5. The Brokering Service transmits sub jobs to the 
Subordinates 

6. Each chosen Subordinate receives a sub job 



(IJCSIS) InternationalJournal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 

7. Subordinates execute the sub jobs. During job 
execution 

a) if the subordinate moves out of VO, but 
within same AO, then the task assigned to 
that subordinate is continued to execute on 
it and the location of the subordinate is 
updated 

b) if the subordinate moves out of VO, and 
also from local AO, then task assigned to 
it is terminated. BS then redistributes it to 
another eligible subordinate and updates 
its JADB. 

8. The Brokering Service receives partial results from 
subordinates that have finished their sub jobs and 
store them in the JADB 

9. The initiator may chose to accept partial results or 
wait until all results are received from Brokering 
Service 



The efficiency of the proposed work is measured using 
the average time taken for a task to be initiated, distributed, 
solved by the subordinates and returned back to the initiator. 
This average task execution time depends on different 
parameters like population of wireless grid and device 
mobility. Population of wireless grid specifies number of 
mobile nodes present in the wireless grid. Device mobility 
refers to the probability of a new device joining a wireless 
grid or a current device leaving the wireless grid per unit of 
time. 

Some of the performance parameters evaluated are as 
follows: 

Task execution time: It is the time taken for a task to be 
distributed by initiator, solved by the subordinates, and 
returned back to the initiator. It depends on different 
parameters like mobility factor (percentage of mobile nodes 
joining or leaving the wireless grid) of nodes within grid 
environment and grid population. To study the effect of grid 
population on execution time, the simulator is run 
considering different grid population. And to study the 
effect of mobility on execution time, the simulator is run 
under two scenarios: with and without mobility of nodes. In 
without mobility, nodes are considered to be at fixed 
location and tasks are distributed among them to find 
execution time. Where as in mobility, number of nodes is 
assumed fixed and mobility factor is varied. 

Location Monitoring overhead: Location monitoring is 
keeping track of nodes status. BSMS has the responsibility 
of keeping track of nodes. It updates the database whenever 
a node enters or leaves the grid. Thus number of updates 
helps to determine the location monitoring overhead. The 
wireless grid simulator is run with different mobility factors 
and number of updates for each run is tabulated. 

Communication overhead: location updating overhead 
leads to communication overhead because, location 
updating increases number of communications among BS 



109 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



and BSMS. This in turn leads to increase in 
utilization. Bandwidth utilization for different mobility 
factors is recorded to study its effect on communication. 

IV. Simulation and Results 

The proposed model has been simulated for various 
wireless grid scenarios. It considers m number of VOs 
controlled by a single BSC, with c number of AOs in each 
VO. The number of nodes in a VO varies between nl to n2. 
Each VO is allocated v Mbps bandwidth. Mobility factor in 
any VO is represented by mf 

The inputs considered for simulation are: m = 2, c = 2 
for both VOs, nl = 30 and n2 = 90. v = 40. mf = varies 
between 10% to 40%. Results are depicted using graphs as 
below: 

Figure 4 shows the execution time calculated 
considering different grid size (grid population), where 
mobile nodes are stationary. The graph depicts that increase 
in grid size reduces the execution time i.e. increasing grid 
size in turn helps to provide more resources for executing 
the tasks and thus can be executed faster with more 
resources. 

Figure 5 shows effect of mobility on execution time 
considering different mobility factors. The graph depicts 
that execution time gradually increases with an increase in 
mobility factor, because higher mobility factor leads to 
higher rate of task abortion and reallocation of task. 

Figure 6 shows the effect of mobility factor on number 
of updates. The graph depicts that location update rate is 
more as mobility is more. Increasing mobility factor lead to 
increase in number of updates i.e. location update cost. 

Figure 7 shows the effect of mobility factor on 
bandwidth utilization. The graph depicts that bandwidth 
utilization rate is more as mobility is more. 



(IJCSIS) InternationalJournal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 
bandwidth 



Exeartionjime Vs Grid .population 




50 70 90 

Grid _p opu I at ion 



Figure 4. Execution time Vs. Grid_population 



400 n 
= 350 ■ 

s 

8 300- 

1 250 ■ 
= 200 ■ 

1= 150 - 

i m ' 

u 50- 

B n 

LU ■ 


ExecutionTime Vs. Mobility 


Factor 


-^^ 


— ■ — & ec ution_Tirne (Nodes=1 0) 
— a— &ecution_Time (Nodes=6) 


^~^ 




10% 20% 30% 40% 
Mobility Jador 





Figure 5. Execution time Vs. Mobilityfactor 



Location update \fe. Mobility^ factor 



10% 20% 30% 40% 

M obi lity _f actor 



Figure 6. No. of updates Vs. Mobilityfactor 





Bandwidth utilization Vs. Mobil ity factor 


18 -i 




16- 




I 14- 
| 12 . 

-I 8 ' 

l 6 

I 4- 

CD 










—■—Bandwidth utilization 
(Mbps) 










2- 




0- 












10% 20% 30% 40% 




Mol)ility_factor 



Figure 7. Bandwidthutilization Vs. Mobilityfactor 

V. Conclusion 

The proposed work presents implementation of normal 
walk mobility model in a wireless mobile grid environment 
for a Virtual Organization with two actual organizations. 
Normal walk mobility model helps to decide the next 
relative direction in which mobile station moves and in 
which it handoffs form one AO to another. This model in 
turn helps BSMS to keep track of number of mobile stations 
available in AOs; so that Brokering Server can distribute 
tasks among available nodes for tasks request sent to 
Brokering Server. 



110 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



The proposed work considers Normal walk model 
because, it represents daily motion of users. The directions 
of motion are mostly symmetric in a real life environment, 
thus it is effective in location updating of a MS. The work 
also shows the effect of grid population and mobility factor 
on execution time and monitoring cost. It can be concluded 
that, high rate of resource availability decreases task 
execution time and higher mobility factor increases 
execution time. Finally increase in mobility factor increases 
monitoring overhead. 

References 

[I] Stan Kurkovsky, Bhagyavati, Arris Ray, "A Collaborative Problem- 
Solving Framework for Mobile Devices", ACMSE April 2-3, 2004 
Huntseville, Alabama, USA. 

[2] Chiu-Ching Tuan and Chen-Chau Yang, " A Compact Normal Walk 
Model for PCS Networks" Mobile Computing and Communications 
Review, Volume 7, Number 4, 2004. 

[3] Ian Foster, Carl Kesselman, and Steven Tuecke, "The Anatomy of the 
Grid: Enabling Scalable Virtual Organizations", May 2001. 

[4] T. Phan, L. Huang, C. Dulan, "Challenge: Integrating Mobile 
Wireless Devices into the Computational Grid", in Proceedings of the 
8 th International Conference on Mobile Computing and Networking, 
Atlanta, GA, September 2002. 

[5] Stan Kurkovsky, Bhagyavati, "Modeling a Computational Grid of 
Mobile Devices as a Multi-Agent System", in Proceedings of the 
2003 International Conference on Artificial Intelligence, Las Vegas, 
NV, June 2003. 

[6] Stan Kurkovsky, Bhagyavati, Arris Ray, "Modeling a Grid-Based 
Problem Solving Environment for Mobile Devices", in Journal of 
Digital Information Management Vol.2 No.2, June 2004. 

[7] P. Mudali, M.O. Adigun, and J.O. Emuoyibofarhe, "Minimizing the 
Negative Effects of Device Mobility in Cell-based Ad-hoc Wireless 
Computational Grids". 

[8] I. F. Akyildiz, Y.-B. Lin, W.-R. Lai, and R. -J.Chen, "A new random 
walk model for PCS networks", IEEE J. Selected Areas Commun. 18, 
7, pp. 1254-1260, July 2000. 

[9] Guoliabg Xue, "An improved Random Walk Model for PCS 
Networks", IEEE Trans. Commun. 50, 8, pp. 1224-1226 August 
2002. 

[10] Md. Imdadul Islam and A.B.M. Siddique Hossain, "A proposed 
random walk model for mobile cellular network", December 2004. 

[II] Chiu-Ching Tuan and Chen-Chau Yang, "A novel normal walk model 
for PCS networks with mesh cell configuration", 2005. 

[12] Jingyuan Zhang, "Location Management in Cellular Networks", 
August 2004. 

[13] M. N. Birje, S. S. Manvi, M. S. Kakkasageri, S. V. Saboji, "Prediction 
Based Handover for Multiclass Traffic in Wireless Mobile Networks: 
An Agent Based Approach" ICICS 2007. 

[14] Ashish Agarwal, Douglas O. Norman, Amar Gupta, "Wireless Girds: 
Approaches, Architectures, and Technical Challenges", MIT Sloan 
School of Management, January 2004. 

[15] WITS A (World Information Technology and Services Alliance), 
Background Paper on Grid Computing, September 2004. 



(IJCSIS) InternationalJournal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 
[16] Buyurman Baykal, and Azgar B. Akan, "A Qos-Aware Handoff using 
RSVP in next generation Wireless Networks", Eighth IEEE 
Symposium on Computers and Communications, 2003. 

[17] S. V. Saboji, S. S. Manvi, M. S. Kakkasageri, "Prediction based 
handover for multicast traffic in mobile networks", Proc. International 
Conference on Wireless Networks (ICWN-2007), The 2007 World 
Congress in Computer Science, Computer Engineering, Applied 
Computing, Las Vegas, Nevada, USA, pp. 293-298, June 25-28, 
2007. 



[18] Brijesh K R. Gupta, Mohan Lai, S. C. Sharma, "A Hybrid Scheme for 
Handover in ATM Based Personal Communication, vol. 3, no. 1, 
2006. 

[19] Cellular Networks, 

http//www.bridgeportnetworks. com/technology /wpimshandover.pdf 

[20] Jidong Wang and Lichun Bao, "Mobile Context Handoff in 
Distributed IEEE 802.11 Systems" IEEE Wireless Communications 
and Networking Conference (WCNC), pp. 225-230, 2004. 

AUTHORS PROFILE 

Anupama S. Nandeppanavar received B.E. in ISE from BLDEA's 
college of Engineering, Bijapur in 2005, and now she is PG student in 
Basaveshwar Engineering College, Bagalkot. Currently she is working as 
Lecturer Basaveshwar Engineering College, Bagalkot. Her interested area 
is computer networking. 

Mahantesh N. Birje received B.E. in CSE from UBDT college of 
Engineering, Davangere in 1997, M.Tech. in CSE from Basaveshwar 
Engineering College, Bagalkot in 2005, and currently he is the research 
scholar at Visvesvaraya Techological University, Belgaum. He is working 
as Asst. Professor in the department of Information Science and 
Engineering, Basaveshwar Engineering College, Bagalkot Karnataka, 
INDIA. His area of interest include Grid computing, Multimedia 
Communications, and Agent technology. He has published 4 refereed 
journal papers, and 7 international conference papers. 

Sunilkumar S. Manvi received M.E. degree in Electronics from the 
University of Visweshwariah College of Engineering, Bangalore, Ph.D 
degree in Electrical Communication Engineering, Indian Institute of 
Science, Bangalore, India. He is currently working as a Professor and Head 
of Department of Electronics & Communication Engineering, REVA 
Institute of Technology and Management, Bangalore, India. He is involved 
in research of Agent based applications in Multimedia Communications, 
Grid computing, Ad-hoc networks, E-commerce and Mobile computing. He 
has published 3 books, 3 book chapters, 25 refereed journal papers, and 
about 75 refereed conference papers. He has given many invited lectures 
and has conducted several workshops/seminars/conferences. 

Shridhar received B.E. in E & CE from Gulbarga University in 1988, 
M.Tech. in digital electronics & advanced communications from KREC, 
Surathkal in 2000. He is currently working as Asst. Professor in the 
department of Electronics and Communications Engineering, Basaveshwar 
Engineering College, Bagalkot. His area of interest include signal 
processing and control systems. 



111 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



Analysis of Birth weight using Singular Value 

Decomposition 



D.Nagarajan 

Department of Mathematics, 

Salalah College of Technology, 

Sal alah, Sultanate of Oman. 



V.Nagarajan 

Department of Mathematics, 

S.T.Hindu College,Nagercoil, 

Kanyakumari, Tamil Nadu,India. 



P.Sunitha 

Department of Mathematics, 

S.T.Hindu College,Nagercoil, 

Kanyakumari,Tamil Nadu,India 



V.Seethalekshmi 
Department of Mathematics, 
James College Technology, 
Nagercoil,TamilNadu,India. 



Abstract — The researchers have drawn much attention about 
the birth weight of newborn babies in the last three decades. 
The birth weight is one of the vital roles in the baby's health. 
So many researchers such as [2],[1] and [4] analyzed the birth 
weight of babies. The aim of this paper is to analyze the birth 
weight and some other birth weight related variable, using 
singular value decomposition and multiple linear regression. 

Keywords- Birth weight; Haemoglobin concentration; 
Maternal Weight; Maternal height; Singular value 
decomposition. 



proportion of morbidity and mortality in childhood 
whereas child born with adequate birth weight are reported to 
do well even under adverse environment, researchers often 
use birth weight as a measure of morbidity risk. 



II. 



DATABASE 



The data for the present study were collected from 
Chennai some private hospital during January 2008 to 
December 2008. The data for the selected variables were 
collected from the hospital case records. 



I. INTRODUCTION 

For a successful reproduction a good health throughout 
childhood, adolescence adult life and pregnancy is necessary. 
Special care has to be taken during pregnancy to get a 
healthy baby. Birth weight is an important determination of 
child health. It is a well known fact that the birth weight in 
influenced by the factors such as gestational period, maternal 
height, maternal weight, age, parity, haemoglobin 
concentration, rate of intrauterine growth, nutrition and many 
other socio-economic factors. Low birth weight neonate is 
defined as any neonate weighing less than 2500 grams at 
birth. It raises grave health risks for children which is a 
public health problem in most developing countries. Low 
birth weight stems primarily from poor maternal health and 
nutrition. Three factors have most impact poor maternal 
nutritional status before conception, short stature due mostly 
to under nutrition and infections during child hood and poor 
nutrition during pregnancy .Less than 17 years and greater 
than 34 years of age are at increased risk of low birth weight 
delivery. It can arise as a result of a baby being born too 
earlier (<37 weeks, also known as premature birth) or being 
born too small for gestational age (small as a result of 
intrauterine growth restriction). These types of babies have 
worst prognosis. Since low birth weight children are 
responsible for a very significant 



III. DESCRIPTION OF MODEL 

The singular value decomposition closely associated to 
the companion theory of diagonaling a symmetric matrix. 
Hark back that if A is a symmetric real n x n matrix there is 
an orthogonal matrix V and a diagonal D such that 



A=VDV X 



(1) 



Here the columns of V are latent vectors for A and 
diagonal entries of D are eigen values of A for Singular 
Value Decomposition begin with m x n real matrix. There 
are orthogonal matrices U and V and a diagonal matrix S, 
such that 



A = USV T . 



(2) 



Here U is m x m and V is n x n, so that S is rectangular 
with the same dimensions as A .The matrix S can be 
formatted to be non negative and in order of decreasing 
order. The columns of U and V are called left and right 
singular vectors for A. [3]. Singular value decomposition is 
used in Latent Semantic Indexing (LSI) to determine the rank 
of the maternal variables and birth weight. Before scoring 
the maternal variables with Latent Semantic Indexing we 
need to onstruct a matrix with the maternal variables 
available as "A" . 



112 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



TABLE I : MATERNAL VARIABLES BY BIRTH WEIGHT 



Baby 
Weight 


Maternal 

height 
(145-162) 


Maternal 
Weight 
(50-70) 


Age 
(21- 

35) 


Blood 
Pressure 
(120/80) 


Haemo 
globin 
(9-14) 


<2000 


15 


10 


8 


15 


9 


2000 


78 


55 


58 


35 


60 


2400 


120 


105 


133 


102 


80 


2800 


206 


240 


230 


180 


150 


3200 


120 


135 


142 


60 


53 


3600 


8 


21 


15 


25 


19 


>3600 


3 


7 


4 


10 


3 



The "Baby weight, Y" is taken as the dependent variable 
and other maternal variables are treated as independent 
variables XI, X2, X3, X4, X5, where 



XI 
X2 
X3 
X4 
X5 



Maternal Height 
Maternal Weight 
Age of mothers 
Blood pressure 
Haemoglobin concentration 



Baby weight 

According to Singular Value Decomposition theory, an 
arbitrarily real rectangular matrix of order m x n can be 
decomposed into three matrices such that 



A —11 S V 

mxn mxl ^Ixn nxn 



(3) 



where U and V are orthogonal matrices and S is a 
singular matrix with eigen values as its diagonal entries 
,which are arranged in non - increasing order. The following 
analysis using MATLAB. 



15 10 8 15 9 

78 55 58 35 60 

120 105 133 102 80 

206 240 230 180 150 

120 135 142 60 53 

8 21 15 25 19 

3 7 4 10 3 



-0.0424 0.1362 

-0.2177 0.1121 

-0.4183 0.2287 

-0.7774 0.2715 

-0.4088 -0.8714 

-0.0637 0.2748 

-0.0201 0.0897 



-0.0520 0.0642 

-0.8923 -0.3417 

-0.1740 0.8457 

0.3157 -0.3517 

0.0160 0.0182 

0.2295 -0.1896 

0.1348 0.0114 



0.7905 0.5108 -0.294^ 

-0.0643 0.0783 0.1293 

-0.1576 0.0405 0.0284 

0.1051 -0.2749 -0.0906 

-0.0222 0.2636 0.0547 

-0.5022 0.7609 0.0261 

0.2869 0.0853 0.9400 





585.9583 

















49.1200 

















34.3363 










s = 










21.2226 




















6.6886 








































0.4737 


-0.1618 


-0.6429 


0.1375 


0.5632 " 


0.5133 


- 0.2822 


0.4454 


-0.6588 


0.1564 


0.5230 


-0.3830 


0.1032 


0.5093 


-0.5565 


0.3705 


0.6792 


0.4333 


0.3596 


0.2903 


0.3182 


0.5349 


-0.4358 


-0.3979 


-0.5142 



From S matrix shows that its consists of five non zero 
singular values, confirming that A is a rank 5 matrix. 

A. Dimensionality Reduction: 

Computing U K ,S K ,V K from U,S,V using MATLAB. 

Let us take the economic dimension K =2 ,that is rank 2 
approximation that means the first 2 columns of U and V and 
the first two rows and columns of S. 



B. Decomposed matrices 

Dimensionality reduction has been done my truncating 
the three matrices obtained form Singular Value 
Decomposition 



585.95 
49.12 







-0.0424 0.1362 






-0.2177 0.1121 






-0.4183 0.2287 


u 


K 


-0.7774 0.2715 
-0.4088 -0.8714 
-0.0637 0.2748 
-0.0201 0.0897_ 




"- 0.4737 -0.1618" 




-0.5133 -0.2822 


v K = 


-0.5230 -0.3830 




-0.3705 -0.6792 




_-( 


).3182 -0.5349 



0.0017 

0.02035 



It has been done by choosing the rank to be 2 ie, K = 2 is 
applied .Using 



x { =x 1 T u K sk- 1 , 



(4) 



and 



Y = Y T U k Sk -1 , 



(5) 



113 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



we get the new coordinate of vectors in this reduced 
space the new set of coordinate vectors are given below 

Y = [-8.9121 7.4489] 



\ =[-0.4709 


-0.1620] 


\ =[-0.5107 


-0.2826] 


3 =[-0.5205 


-0.3836] 


4 =[-0.3681 


0.6808] 


5 =[-0.3164 


0.5361] 



Vector determination using cosine similarity values we 
rank results in decreasing order. Using 



Sim(q,d) = 



q.d 

\q\ \d\ 



(6) 



Hence, 



Sim(Y,X l ) = 0.516924 
Sim(Y,X 2 ) = 0.360846 
Sim(Y,X 3 ) = 0.237192 
Sim(Y,X 4 ) = 0.929062 
Sim(Y,X 5 ) = 0.94228 

From the calculation 5 4 12 3 

.That means the value may be interpreted as the proportion of 
variability in Y that is explained by 

X X X X X 

1 ' 2 ' 3 ' 4 ' 5 .It reveals that birth weight is closely 

related to Haemoglobin absorption. Haemoglobin is one of 
the vital roles in the birth weight of babies. 

A comparison study is done between Singular Value 
Decomposition and multiple linear regression model to 
analyze the relationship between birth weight and maternal 
variable using the linear regression model of the form "Baby 
weight, Y" is taken as the dependent variable and the other 
variables are treated as independent variables. 

TABLE II : REGRESSION COEFFICIENT 



Predicator 


Coefficient 


t 


VIF 


Constant 


3342.8 


5.68 




*. 


-96.82 


-1.09 


283.827 


x 2 


9.05 


0.22 


75.032 


Xj, 


68.39 


0.78 


309.698 


x 4 


-29.46 


-0.58 


59.953 


x 5 


45.02 


0.56 


106.884 



From the above TABLE II it is observed that the value of 
R 2 is 0.738. It includes the maternal weight, height, age of 
mothers, blood pressure and Haemoglobin. It seems to 
73.8% of the variation in baby weight is explained by the 



fitted model. The remaining 16.2 of variation can be 
explained by the factors other than these variables like 
socioeconomic factors. 

Analyzing has been done for each maternal variable with 
the birth weight as shown below. 

TABLE III: R 2 VALUES OF ALL INDEPENDENT VARIABLE 



Dependent 
variable 


Independent 
variable 


R 2 


Y 


x, 


7.3 


Y 


x 2 


1.5 


Y 


x 3 


2.7 


Y 


x 4 


3.0 


Y 


x 5 


8.1 



From the above Table III shows that the relation between 
Y and XI is 7.3%, Y and X2 = 1.5, Y and X3 = 2.7, Y and 
X4 = 3.0YandX5 = 8.1 

Hence also hemoglobin concentration is very close to 
birth weight .Hemoglobin is one of the vital role in the birth 
weight of babies. 

TABLE IV: COMPARISON STUDY 



Variable related to 
Y 


Cosine Similarity 


Multiple regression 

R 2 


x, 


0.5169 


7.3 


x 2 


0.3608 


1.5 


x, 


0.2371 


2.7 


x 4 


0.9298 


3 


x 5 


0.94228 


8.1 



Above table reveals that birth weight is very close to its 
haemoglobin concentration 

IV. CONCLUSION 

From this study, it is observed that, the birth weight 
mainly depends on the Haemoglobin concentration in both 
Sigular Value Decomposition analysis and multiple linear 
regression analysis. Hence the mother with high 
Haemoglobin concentration can avoid low birth weight. So 
the pregnant women should intake additional nutritional food 
to increase the Haemoglobin concentration and to avoid the 
health risk problems among the neonates. Some following 
reasons are the case of lowbirthweight, that is the mother has 
not obtained the appropriate nutrition, early marriage, late 
pregnancy at around 35 years, mother below 40 Kilograms, 
mother has anemia problems, small placenta, chronic 
placenta insufficiency can also lead to low birth weight , lack 
of Oxygen also leads low birth weight and due to mothers 
hypertension and malnutrition. Some action taken before 
birth to avoid low birth weight, which is regular checkup, 
appropriate nutrition intake, checks the haemoglobin status, 
take iron. The case should be identified and corrected in the 



114 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



mother inability to gain weight should be deleted early to 
avoid problems. The following actions are taken after birth. 
In such circumstance the child needs extra care to maintain 
the temperature using like incubator, overhead warmer and 
kangaroo mother care. 

REFERRENCE: 

[1 ] Arbuckle T.E and Sherman G J,Comparison of the risk factors for pre 
terms delivery and intrauterine growth retardation ,Peachtree Perinat, 
Epidemol,( 1989) Vol 3 ,ppl 15-129. 

[2] Bhatia B.D and Tyagi N.K, Birth weight relationship with other 
Foelat Anthrometric parameter ,Indian pediatrics(1984),vol 21, 

pp833-838 

[3] I.J Good, Some applications of the singular decomposition of matrix, 
Technometrics (1969). Vol 1 l,no 4,pp 823-831. 

[4] Philip Steer u Kondaveeli c Bary-Kinsella, Relation between maternal 
haemoglobin concentration and birth weight in different ethic groups. 
Biomedical journal. (1995).310(l),pp 489-491. 



1 1 5 http://sites.google.com/site/ijcsis/ 

ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 



A Simple method of designing dual loop 
controller for cold rolling mill 



S.Umamaheswari, Dept. of EIE, 
Mahendra Engg, College, 
Namakkal(Dt), INDIA 



V.Palanisamy M.Chidambaram 

Principal/ Info Institute of Technology, Director/ NIT, 

Coimbatore, INDIA Trichy, INDIA 



Abstract-The mathematical model (Interval Plant) of the web guide 
in rolling mill is controlled using PID controller. The given 
interval plant is approximated to first order plus time delay with 
integrator (FOPTDI) system. The dual loop control (DLC) method 
proposed by Jacob and Chidambaram for design PID controllers 
is extended for FOPTD+I systems. The performance of the 
closed loop system is evaluated for both the original and the 
approximated model. The controllers are also tuned using Internal 
Model Control (IMC) and the performance is compared by 
simulation. 

Index Terms -Interval plant, Double loop control , Direct synthesis 
method, IMC(Internal Model Control) 



I. 



Introduction 



There are many intermediate web guides in cold rolling mills 
process such as CRM (cold rolling mill), CGL(Continuous 
galvanizing line) and so on. The main functions of the web 
guide are to adjust the center line of the strip to the center line 
of the steel process, so they are called centre position control 
(CPL). Rapid process speed cause large deviation between the 
center position of the strip and the process line. So the 
difference between the center position of the strip and the 
process line should be compensated. In general, the centre 
position control (CPC) of the web is obtained by hydraulic 
driver and electrical controller. The model of the web guide 
system is obtained from Sang Min Kim et.al( 2004)[1]. 

For the purpose of designing controllers, the dynamics of many 
processes can be described by first-order plus time delay with 
an integrator (FOPTDI) model. 



Methods for tuning PID controllers for such models are based 
on stability analysis such as Internal model controller (Rivera 
et.al.(1986)[5]. The dual loop method is extended for designing 
PID controller for FOPTD+I systems. The inner loop is of PD 
controller and the outer loop is of PID controller designed using 
direct synthesis method proposed by shesagiri Roa[ 6]. PID 
controllers are also designed by IMC method. The performance 
of the controller is compared by simulation. 



II. INTERVAL PLANT 

The model of the web guide system by using geometrical 
relations of the guide ignoring the mass and stiffness of the web 
is given by Sang Min Kim et.al.(2004). 



G(s) = 



a 3 s 3 +a 2 s 2 +a x s l +a 
s 5 +b 4 s 4 +b 3 s 3 +b 2 s 2 +b x s l + b 



Where the coefficients are of interval in nature, which is given 

by, 

a 3 = [4.5 24] ; a 2 = [5 29] ; a x = [0.9 5] ; a, =[0.05 3] 

b 5 = [0.05 0.3] ; & 4 = [ 4 9 ] ; 6 3 =[4 9]; b 2 =[0J2]\ 

^ = [0.03 0.08]; b =[0 0] 

This is represented as model 1, model2 and model3 by 
considering only the minimum, maximum and average values 
of the interval. The model, model2 and model3 of the plant are 
given by, 



Model 1 = 



Model2 = 



0.05 + 0.9^+5^+4.5^ 
0.03s l +OJs 2 +4s 3 +4s*+s s 



0.3 + 5^+ 29/+ 24s 3 
0.0Ss 1 +2s 2 +9s 3 +9s 4 + s 5 



Model3 = 



0.175 + 2.95^+17^+ 14.25s 3 
0.055s 1 +L35? + 6.5s 3 + 6.5s 4 + s 5 



These models have been approximated to FOPTD+I system. 
Approximation is done by taking 1/s from the actual model 
separately, and applying step input, the time constant T is 
calculated. Letting s to zero, the magnitude k is calculated. 



116 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 



Hence the FOPTD is obtained in which the 1/s term is again 
combined to get FOPTD+I model. 
The approximated system is given by, 



RModell 



RModel2 



RModel3 



1.667 



s(0.5584s + l) 

3.75 _ 

s(\.256s + \) 6 

3.1818 
s{\. 0659 s + 1)' 



-0.3526^ 



-0.793s 



-0.613s 



In the present paper dual loop system in which the inner loop is 
of PD controller and the outer loop is of PID for FOPTD+I 
system is designed. The performance of the control system is 
compared with that of the IMC method. 



III. DUAL LOOP PID 

The general transfer function of the process to be 

it 



controlled is given by, 



-Ls 



s(rs + l) 



(1) 



w here k is the process gain T is the time constant and i is 



the process lag. 



Load 



3fa 



^PIDLgHA(^) 



-©-> 



Kp e -Ls 



sn+TAi 



Figure 1. Block diagram of the dual loop control 



ses 



The inner loop is of PI controller and the outer loop is of PID 
controller designed using direct synthesis method proposed by 
sheshagiri roa for unstable system. In this paper it has been 
extended for stable system. The closed loop Transfer function 

-Ls 



of the inner loop is given by, 
is approximated to 



K k p e ~ 



s + kk p e~ 



, the delay 




Kk p (l-(L/ 2 )s) 



s(l + (y 2 )s) + k c k p (l-(L/ 2 )s) 



k M 1+ (%) s Y 



i l+ ( L A) s ) +k M l -( L A) s ) 



*^( 1+ (%)4 



1/2 



l- Kk o L /, 



'\ 



s + k c k p 



Closed loop transfer function of the inner loop is given by, 
k c k p (l + (%)s)e-» 



T / 2 

y 2 s +s 



kk.L 



s + k c k p 



^~2 



FID 



L/2 4*+( 1 +kckiJL./2^ s+kckp 



Figure 2. Block diagram of the simplified dual PID 
after simplification the controller is obtained as, 

1 {as + V) 
G r =k(l + — + r d s) 

where 

k =■ 



7i 



k(3A z +1.50A + O.50tj 1 -tj 2 ) 

T = ^ 

L d 

a = 0.50,(3 



t(3A 2 +\.56X + 0.56ri l -ri 2 ) 
in which and 

(0.59 -r) A 3 +(3r 2 -\50r)X 2 + Wt 2 X + 0.59 V 



J h 



t(0.56 + t) 



117 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. I, 2010 



Tuning Parameter Selection 



The tuning parameter X should be selected in such a way that 
the resulting controller gains should be positive for positive 
values of k. Hence, as per the method given by Seshagirirao et. 
al. [6] to get positive values of controller gain(kc), the 
constraint to be followed is, 

77 <3A 2 +1.50/1-%+ 0.50ft 

Also, X should be selected in such a way that the resulting 
controller gives good robust control performances. The initial 
value of the tuning parameter can be taken as equal to half of 
the time delay of the process to get good control performances. 
If not then, the tuning parameter can be increased from this 
value till good nominal and robust control performances are 
achieved. For suitable value of X and f5 , the controller 
designed on DS method gives good control performances. 
However for high value of /? , the phase lag imposed by the 

term (/fa + 1) in the controller is more and thus the designed 
controller with this value of ft is not able to give robust control 

performances which results in low gain and phase margins of 
the open loop system than the required values (gain margin 
should be >1.7 and phase margin should be >35° for robust 
control of a process [6]. 

Based on many simulation studies, it is observed that taking 
'0.1 P ' instead of /3 gives good compromise between nominal 
performances and robust control performances. Thus, in the 
present work, the value of j3 obtained is modified as '0.1 j3 ' 
for simulation studies. 

The performances of the closed loop system are evaluated by 
giving a unit step in the set point. Fig. 1-3 shows the comparison 
of Rmodel 1 , 2 & 3 with the controller tuned using double loop 
MRC method. 

5. IMC METHOD 

The IMC based PID controller settings for FOPTDI system is 
given by, (Dale, et. al. (2004) and Rivera et.al. (1986)[2&5] 
respectively. 

k c k = (2r c + T + 0)/(r c +0) 2 

t, = (2r c + x + 6) 
t d =(2t c +O)t/(2t c + t + 0) 

The choice of design parameter T c is a key decision in both the 
direct synthesis (DS) and IMC design methods. In general 
increasing T c produces a more conservative controller because 

k c decreases while T l increases. Several IMC guidelines for 

T have been published for FOPTD with an integrator system. 



As per Rivera et al., (1986), tJ6 > 0.8 and T c > 

0.1 T ,k p exp(-T d s)/s(TS + 1) . By simulation it was found 

that T c =4 as per Rivera et. al., gives best result for all three 
models. 

6. SIMULATION RESULTS 



Let us consider the Rmodel2 plant with k =,3.75 T = 1.256 



and T d = 0.793. The PID controller settings by the dual loop 
method are k c =0.452 r. = 0.96 and T D =0366 The PID 

settings by IMC method are k c =0.12 and T. =0.012 T D = 

0.1319. Figure 1, 2 and 3 show the comparison of the servo 
response of all three models. The robustness of the closed loop 
is evaluated by using the maximum controller settings on the 
other two models of the system. Figures 4, 5 and 6 show the 
regulatory response of the systems. Comparison of ISE and IAE 
values of the Rmodel 1, Rmode2 and RmodeB for both the 
servo problem and the regulator problems using the controller 
settings for the IMC and the dual loop method are shown in 
table I. 




Fig.l Servo response of Rmodell 
Solid -> DLC ; dash-dot -> IMC 




Fig. 2 Regulator response of Rmodell ; 
Legend as in Fig.l 



118 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. I, 2010 




Fig. 3 Servo response of Rmodel 2. 
Legend as in fig. 1 




Fig. 4 Regulator response of Rmodel 2 
Legend as in fig. 1 





Fig. 6 Regulator response of Rmodel 3 
Legend as in fig. 1 

The servo and regulatory response of the dual loop control 
method gives a better response than that of IMC method. Table 
I and II gives the comparison of ISE and IAE values for both 
regulator and servo response. The performance of the controller 
designed by the dual loop method gives the best. Here the 
controller setting for the Rmodel2 is chosen and robustness is 
evaluated for other models. 



TABLE I 
ISE, IAE VALUE COMPARISON FOR REGULATOR RESPONSE 



Fig. 5 Servo response of Rmodel 3 
Legend as in fig. 1 



Models 


Method 


Controller Settings 


REGULATOR 
Response 






Kc 


Kc 
1*1 


Kc* 


ISE 


IAE 


Model 

1 




0.45^ 
0.12 


- 0.96 
0.012 


0.366 
0.131 








DUAL 
PID 


3.79 


498.6 




IMC 


4.92 


509.3 


Model 
2 










DUAL 
PID 


4.99 


487.6 




IMC 


6.78 


500.9 


Model 
3 










DUAL 
PID 


4.98 


501.2 




IMC 


5.03 


504.1 



119 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 



TABLE II 
ISE, IAE VALUE COMPARISON FOR SERVO RESPONSE 



Models 


Method 


Controller Settings 


Servo 
Response 






Kc 


Kc 


K C *T D 


ISE 


IAE 


Model 1 




0.45: 

0.12 


I 0.96 
0.012 


0.366 
0.131 








DUAL 
PID 


0.116 


9.62 




IMC 


0.416 


14.9 


Model 2 










DUAL 
PID 


0.166 


0.07 




IMC 


0.33 


120.92 


Model 3 










DUAL 
PID 


0.14 


0.17 




IMC 


0.34 


19.81 



Also the same controller parameters are checked for their 
corresponding original model. The response shows that the dual 
loop method seems to be better compared to the other two 
models. The ISE and IAE values for regulator response are 
Reported in Table I. The dual loop method has got least ISE 
and IAE values. 

7. CONCLUSION 

The mathematical model of the web guide system is tuned 
using IMC and dual loop control method. The controller 
designed for FOPTD +Integrator system works well for even 
the corresponding original model. Also the robustness of the 



system is verified by simulation. The controller has been 
designed using dual loop control and IMC. Among this, 
robustness of the dual loop method is seems to be better when 
compared to IMC method. Referring to the Table I and II, the 
ISE and IAE values are comparatively less for dual loop 
method than that of the IMC method. 

References 

[1] Sang Min Kim, Byoung joon Ahn, et.al., "The modeling 
and control of web guide process in cold rolling mill", IEEE 
Industrial Electronics Society, Korea, Nov2-6,(2004) 

[2] E.Dale, Seborg, Jhomas F.Edgar, "Duncan A.Mellichamp, 
"PID controller design, tuning and troubleshooting", 2 nd ed., 
pp. 304-309, John Wiley & sons(Asia) Pte.Ltd., (2004) 

[3] M.Chidambaram and JacobE.F. "Control of unstable 
delay plus first order systems", Computer& Chemical 
Engineering, Vol.20, pp.579 -584, (1996) 

[4] M.Chidambaram and JacobE.F. "Robust Control of an 

unstable bioreactor", process control & Quali., Vol.8, pp.167 
- 175, (1996) 

[5]Rivera,D.E.,Morari,M.&Skogestad.S,"IMC-PID controller 
design. Industrial Engineering and Chemical Process, Design 
and Development", 25, 252 (1986). 

[6] Seshagirirao, A., V.S.R Rao, and M.Chidambaram, "Direct 
synthesis - based controller design for integrating processes 
with time delay," Journal of the Franklin Institute, 2008. 



120 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. I, 2010 



Detection of Microcalcification in 
Mammograms Using Wavelet Transform and Fuzzy 

Shell Clustering 



T.Balakumaran 

Department of Electronics and 

Communication Engineering 

Velalar College of Engineering and 

Technology, Erode, TamilNadu, 

India. 



Dr.ILA.Vennila 
Department of Electrical and 

Electronics Engineering 

PSG College of Technology, 

Coimbatore, TamilNadu, India 



C.Gowri Shankar 

Department of Electrical and 

Electronics Engineering 

Velalar College of Engineering and 

Technology, Erode, TamilNadu, 

India. 



Abstract — Microcalcifications in mammogram have been mainly 
targeted as a reliable earliest sign of breast cancer and their early 
detection is vital to improve its prognosis. Since their size is very 
small and may be easily overlooked by the examining radiologist, 
computer-based detection output can assist the radiologist to 
improve the diagnostic accuracy. In this paper, we have proposed 
an algorithm for detecting microcalcification in mammogram. 
The proposed microcalcification detection algorithm involves 
mammogram quality enhancement using multirresolution 
analysis based on the dyadic wavelet transform and 
microcalcification detection by fuzzy shell clustering. It may be 
possible to detect nodular components such as microcalcification 
accurately by introducing shape information. The effectiveness of 
the proposed algorithm for microcalcification detection is 
confirmed by experimental results. 

Keywords: Computer-aided diagnosis, dyadic wavelet 
transform, skewness and kurtosis, Fuzzy shell clustering. 



I. 



Introduction 



Breast cancer is the second leading cause of cancer deaths 
for women and is found as one in eight women in the United 
States. It is a disease in which cells in the tissues of the breast 
become abnormal and divide without order or control. These 
abnormal cells form too much tissue and become a tumor. 
According to WHO report, nearly two million women are 
diagnosed with breast cancer every year worldwide. The 
disease can be treated if discovered so early enough. The 
effective detection of breast cancer in earlier stage increases 
the survival rate. The appropriate method for early detection of 
pre cancerous symptoms is screening mammography, which 
has to be conducted as a regular test for women. 

Calcification clusters are an early sign of breast Cancer. 
Microcalcifications are quite very small bits of calcium 
deposits present inside the soft breast tissue. It shows up in 
clusters or in patterns (like circles or lines) associated with 
extra cell activity in breast region. 



Microcalcifications appear in mammogram image as small 
localized granular points with high brightness. It is not easy to 
detect by naked eye because of its miniaturized dimension. 
However about 10%-40% of Microcalcification clusters are 
missed by radiologists due to its small size and nonpalpable 
[1], [2]. To avoid these problems, a New CAD (computer 
Aided diagnosis) system has to be developed to improve the 
diagnostic rate. By incorporating the expert knowledge of 
radiologists, the CAD system can be made to provide a clear 
insight about the disease and saves the society from breast 
cancer. 

Many researchers have proposed the algorithms for 
Microcalcification detection based on discrete wavelet 
transform, which is a powerful tool for analyzing the image 
hierarchically on the basis of scale. Some researchers have 
developed a CAD system using fuzzy clustering, artificial 
neural network and genetic algorithm. R. N. Strickland et al. 
[3], H.Yoshida et al. [4] used a wavelet transform to detect 
microcalcification and the fuzzy logic was tried by N.Pandey 
et al. [5]. Anne Strauss et al. [6] presented an identification 
scheme based on watershed Processing. Valverde et al. [7] 
used a deformable-based model for Microcalcification 
detection. Some of these studies detect approximately 70% to 
80% of correct calcification. Objective of CAD system is to 
reduce the false positives and consistency of radiologists in 
image interpretation [8]. Naturally Microcalcifications are 
nodular in structure, other tissue such as mammary ducts 
blood vessels are linear in structure [9]. The Fuzzy shell 
clustering algorithm (FSC) is best in identifying circular 
objects present in an image [10]. In the proposed method, 
wavelet has been combined with fuzzy shell clustering (FSC) 
algorithm in order to mark the Microcalcification region 

The rest of this paper is organized as follows. Section II 
presents microcalcification enhancement by using dyadic 
wavelet transform; the detection part using Fuzzy shell 
clustering in Section III, Results obtained on execution of 
algorithm are presented in section IV and Conclusion as last 
section. 



121 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



II. MlCROCALCIFICATION ENHANCEMENT 



The fundamental operation needed to assist 
microcalcification detection in mammogram is image contrast 
enhancement. In many image-processing applications, the 
grayscale histogram equalization (GHE) is one of the simplest 
and effective techniques for contrast enhancement. Histogram 
equalization improves contrast of the microcalcification in 
mammogram image by reassigning the intensity values of 
pixels based on the image histogram. But Histogram 
equalization technique doesn't enhance the microcalcification 
region because it modifies the intensity of the image with 
some unpleasant visual artifacts. 

Microcalcifications appear as subtle and bright spots, 
whose size varies from 0.3mm to 1mm in the mammogram 
image. It is not easy to enhance the microcalcification regions 
since surrounding dense breast tissue makes the abnormality 
areas almost invisible. Microcalcifications are high frequency 
in nature. So it can be extracted by using high pass filtering. 
But conventional enhancement technique like unsharp 
masking, homomorphic filters and high boost filtering tends to 
change the characteristics of microcalcification. To overcome 
these limitations microcalcification regions can be enhanced 
by dyadic wavelet transform without modifying characteristics 
of microcalcification. 

Wavelet analysis permits the decomposition of image at 
different levels of resolution. In Fig. 1, the filter bank structure 
of the two-dimensional wavelet transform is shown from level 
j to level j+1, which generates four sub-images at level j+1. Sj 
be original image, the approximation sub-image Sj +1 is 
obtained by applying the vertical low-pass filter followed by 
horizontal low-pass filter to Sj. The sub-image Sj + 1 L H is 
obtained by applying the vertical low-pass filter followed by 
the horizontal high-pass filter. The sub-image Sj + 1 H L is 
obtained by applying the vertical high-pass filter followed by 
horizontal lowpass filter. Finally, the response Sj+ 1 H H is 
obtained by applying the vertical and horizontal high-pass 
filters successively [9]. The downsampling by a factor 2 is 
introduced after each level of filtering. The same procedure is 
repeated for each level of approximation coefficients till Sj +n is 
achieved. 



Ai 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 

The digitized mammogram incorporated with a size of 
1024 x 1024 pixel was taken from Digital Database for 
screening mammogram(DDSM). Mammogram image was 
decomposed upto 10 levels by applying dyadic wavelet 
transform with a decimation factor 2. The original 1024x1024 
grayscale digital mammogram image was decomposed to 10 
levels by applying Daubechies4 wavelet transform. Finally the 
lowest approximation image Sio is of single pixel width. Since 
microcalcification appears as high frequency behavior in 
mammogram, the enhancement is achieved by setting the 
value of Sio as zero. 

The detail coefficients are enhanced by as per the 
expression number (1) 









r-~ 


H L (Z) 


— 


I' 


1 — ► 


H L (Z) 


— 


1' 


—- 














— 


H H (Z) 


— 


I' 














' 


Hl(Z) 


— 


I' 


— 


H H (Z) 


— 


V 


— 














— 


Hh(Z) 


— ' 


1' 



■ °i+i 



C HL 

' °j+1 



• C HH 
°j+1 



Sj D (x,y)= Sj D (x,y) 



if | Sj u (x,y) | < Tj 
g*Sj U (x,y) if |Sj D (x,y)| > Tj 



(1) 



where, x and y are spatial coordinates, D represents all 
horizontal, vertical and diagonal subbands. Tj be a non 
negative threshold obtained by taking standard deviation of 
respective subimage. The best visual quality of 
microcalcification is obtained while the gain(g) is set as 1.2. 
The reconstruction of weighted higher frequency subbands 
provides better visibility of microcalcification region than the 
other breast regions. 



III. Microcalcification Detection 

The CAD system was developed to assist the radiologist in 
detecting the suspicious areas on the basis of certain feature 
extracted from the images. After enhancing the 
microcalcification regions, next step in the journey of 
climbing the aim is the extraction of microcalcification 
through Soft computing tools. Microcalcification appears in 
mammogram as nodular points with higher brightness, 
localized or broadly diffused along the breast tissue, whereas 
normal tissues such as blood vessels are linear in structure. So 
detecting the nodular structure in image is a key in detecting 
the microcalcification. 

A. Region of Interest (ROI) Identification 

The first stage of microcalcification detection is ROI 
identification. The enhanced mammogram image is 
decomposed by undecimated wavelet transform (filter bank 
implementation without downsampling). The resulting 
horizontal detailed image or vertical detailed image is used to 
identify the region encircling the microcalcification clusters. 
Third and fourth order statistical parameters, skewness and 
kurtosis [11], are used to find the regions of microcalcification 
clusters. 

An estimate of the skewness is given by 



S k : 



i=\ 

(Af-l)cr 3 



(2) 



Fig. 1 Two dimensional dyadic wavelet structure 



and the statistical parameter kurtosis holds the expression 



122 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



K„ 



£(*<-m) 4 

i = l 

(# -l)o- 4 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 
8 is minimum threshold value, which is used to minimize 
the number of iteration. 



(3) 



where x, is the input data over N observations, m is the 
ensemble average of xj and a with its standard deviation. The 
third and fourth order statistical estimates were calculated for 
every overlapping 32x32 square regions of horizontal 
bandpass image or vertical bandpass image. The area having 
skewness value greater than 0.2 and kurtosis value greater than 
4 is marked as a region of interest (ROI). 

B. Fuzzy Shell clustering 

Extraction of features is the key process in the 
development of CAD system. Fuzzy shell clustering (FCS) 
algorithm is used to perceive nodular structure from the ROI. 
It extracts the clusters with spherical symmetry of a given data 

set X={xi,x 2 , x n } with the prototype cluster centers 

V={vi,v 2 , v c }. Here c denotes the number of clusters 

formed during processing. Each data point x k has a degree of 
membership u ik to the i th cluster. The data point x k is assigned 
to a cluster if the below given weighted objective function is 
minimum. 



j=XI<^ 2 +2>,X(i-^r 



k=l i=\ 



i=\ 



k=\ 



C. Nodular extraction 

After identifying ROI, the next step is to detect edges of 
the enhanced image. Edges are detected by applying first order 
derivatives. First order derivative is implemented by gradient 
operator (G x &G y ) . The edges are detected by computing the 
gradient of each pixel in the enhanced image in x direction and 
y direction. The gradient of f(x,y) is 



V/=Gl+G v 



(8) 



(4) 



By applying first order derivative, Enhanced edges were 
detected. These enhanced edge pixels are connected in sets to 
form group. The Groups containing less than 5 pixels are 
discarded. Fuzzy Shell clustering algorithm was applied to 
only edge point, which belongs to ROI. FCS was applied 
several times with c= 1 and W, = 9. As a result, different 
circles with different radius were obtained. Final goal is to 
extract only circular (nodular) region, whose radius is equal to 
microcalcification radius. 

After obtaining different circular regions, 
microcalcification detection is performed to distinguish 
between valid microcalcification region and invalid one. The 
two validity measurement parameters are Cluster density and 
Relative Shell Thickness [12]. These two parameters are used 
to identify microcalcification regions. 



Where D ik is distance from data point x k to cluster center vj 
and Wj be the 3-db width of the i th cluster. The weighting 
exponent me(l,oo) is the fuzziness controlling parameter with 
m as a real number greater than 1, it was chosen as 2 for better 
segmentation. Microcalcification appears as nodular (circular) 
in structure, the FCS algorithm identifies nodules described by 
their center vj and radius rj. The geometric distance function of 
clustering is 



Dik= ** ~ v i\\~ r i\ 



(5) 



The necessary condition on membership leads to following 
equation used to update u ik for the minimization of objective 
function 



U ik 



1 



D 






2 
m -1 



(6) 



The following equation is used to speed up the algorithm 



u j+l -u J 



(7) 



C 



Z "a 



2nr : 



(9) 



Numerator denotes sum of membership function on most 

\u jk I > yC 
characteristic points ( / z ) and denominator denotes 

area of the circular region. 

Relative shell Thickness is defined by 



RST , 



_k=i 

k = l 



(10) 



Where r^ is radius of circular region. Correct 
microcalcification was detected according to the rule 

If(C di >1.15&&RSTj<0.2) 

Microcalcification nodular 
End 

The above thresholds were found by different experiments. 
Microcalcification was successfully detected by our proposed 
method. 



123 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 



IV. Results 



To test the proposed method, experiments were performed 
on the set of mammogram image with different size and 
features which were obtained from DDSM database. In 
DDSM database, the size of each pixel is 0.0435mm and gray 
level depth is 12 bits. 

Fig. 2(a) shows a low contrast mammogram image of size 
1024x1024. Fig. 2(b) shows enhanced image where one can 
clearly visible microcalcification than the original image. 
Fig. 2(c) shows Regions of Interest (ROI) was calculated by 
high order correlation parameters such as Skewness and 
kurtosis. Fig. 2(d) shows edge map of enhanced image by 
applying first order derivatives. Fig. 2(e) shows edge map after 
removal of small unwanted groups and Fig. 2(f) shows the 
microcalcification region after applying FCS. Here FCS was 
applied several times to each edge point which is belonging to 
ROI. 




" : . 




(e) (f) 

Fig. 2 Steps of microcalcification detection: (a) Original Mammogram 

Image (b) Enhanced image by wavelet transform (c) Region of Interest 

(ROI)identified by using Skewness & kurtosis (d) Edge map of enhanced 

image (e) Edge map after removal of unwanted groups (f) Microcalcification 

Detection by FCS algorithm 






96 
95 
94 
93 
92 

. 






FROC Curve 












































^ 









0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 

FP number/image 



Fig. 3 FROC Curve of Microcalcification Detection 

Free-response receiver operating characteristic (FROC) 
curve is used to evaluate the performance of microcalcification 
detection [13]. An FROC plot is achieved by calculating of the 
true-positive detection ratio (TP) with average number of false 
positives (FPs) per image. True positive detection ratio here 
refers that how many true abnormalities (microcalcifications) 
are correctly detected by computerized scheme and false 
positive per image refers to how many true abnormalities are 
missed. We have tested 112 images for FROC analysis. Our 
proposed method was achieved 95.23% TP ratio with 0.9 FP 
number/image. 

V. Conclusion 

In this paper, an algorithm for image enhancement based 
on wavelet transform and microcalcification detection using a 
Fuzzy shell clustering were proposed. Original Mammogram 
image is decomposed by applying dyadic wavelet transform. 
Enhancement is performed by setting lowest frequency 
subband of decomposition to zero. Then image is 
reconstructed from the weighted detailed subbands, the 
visibility of the resultant's image is greatly improved. 
Experimental results were confirmed that microcalcification 
can be accurately detected by fuzzy shell clustering algorithm. 

We proposed an algorithm for Microcalcification detection 
by introducing shape information. FCS algorithm can detect 
microcalcification if it's nodular in structure. We have 
testedll2 images from our database. In these images, 95% of 
microcalcifications were detected correctly and 5% of 
microcalcifications were failed to detect because of they are 
not nodular in structure. Thus the research is still being 
performed to find out the better way to detect 
microcalcification without nodular structure. 



References 

[1] R.G.Bird,T.W.Wallace,B.C.Yankaskas, "Analysis of cancers 

missed at screening mammography," Radiology, vol. 184, pp. 613- 
617,1992 

[2] H. Burhenne, L. Burhenne, F. Goldberg, T. Hislop, A.J.Worth, 

P.M.Rebbeck, and L.Kan, "Interval breast cancers in the screening 
mammography program of British Columbia: Analysis and 
classification," Am. J Roentgenol., vol. 162, pp. 1067- 

1071,1994 

[3] R.N. Strickland, H.L. Hahn, "Wavelet transforms for detecting 

microcalcifications in mammograms" IEEE Trans. Med. Imag., 
vol. 15, no. 2, pp. 218- 229, April 1996 



124 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 



[4] H.Yoshida,K Doi, and R. M. Nishikawa, "Automated detection of 

clustered microcalcifications," Proc. SPIE (Digital Mammograms 
Using wavelet Transform Tech., Med.Imag. 1994: Image Process.), 
vol. 2167, pp. 868-886, February 1994 

[5] N. Pandey, Z. Salcic, and J. Sivaswamy, "Fuzzy logic based 

microcalcification detection", Proceedings of the IEEE for Signal 
Processing Workshop, pp. 662- 671,2000 

[6] Anne Strauss, Abdeljalil Sebbar, Serge DCsarnaud et al.. 

"Automatic detection and segmentation of microcalcifications on 
digitized mammograms", In Proceedings of the Annual 
International Conference of the IEEE,14(5),pp.l938-1939, 1992 

[7] F. Valverde, N. Guil, J. Munoz, Q. Li, M. Aoyama and K. Doi, "A 

Deformable Model for Image Segmentation in Noisy Medical 
Images", International Conference on Image Processing, vol.3, pp. 
82-85, Oct. 2001. 

[8] K. Doi, H. MacMahon, S. Katsuragawa, R. M.Nishikawa, and Y. 

Jiang, "Computer-aided diagnosis in radiology: Potential and 
pitfall," Eur. JRadiol.,vol. 31, pp. 97-109,1999 

[9] Nakayama R,Uchiyama Y,Yamamoto K,Watanabe R, Namba, K, 

"Computer-aided diagnosis scheme using a filter bank for 
detection of microcalcification clusters in mammograms", IEEE 
Transactions on Biomedical Engineering, Vol 53.No.2,p.273-283, 
February 2006 

[10] R. N. Dave , "Fuzzy shell clustering and application to circle 

detection in digital images. International Journal on General 
Systems, 16:343-355,1990 

[11] M.N. Gurcan, Y. Yardimci, A.E. Cetin and R.Ansari , "Automated 

Detection andEnhancement of Microcalcification on 

DigitalMammograms using Wavelet Transform Techniques", 
Dept. of Radiology, Univ.of Chicago, 1997 

[12] M.Barni, A.Mecocci and G.Perugini, "Application of possibilistic 

shell-clustering to the detection of craters in real-world 
imagery",Proceedings of the IEEE for Geoscience and Remote 
Sensing Symposium, vol.1, pp. 168-170,2000 

[13] C.E.Metz,"Some practical issues of experimental design and data 

analysis in radiological ROC studies," Invest. Radiol., vol. 24, no. 
3, pp. 234-245, Mar. 1989 

[14] Hiroyuki Yoshida,"Diagnostic Scheme Using wavelet Transform 

for Detecting Clustered Microcalcifications in Digital 
Mammograms", Acad Radiol, Vol. 3,1996 

[15] S.A. Hojjatoleslami and J. Kittler, "Automatic detection of 

calcification in mammograms", The IEEE Fifth International 
Conference on Image Processing and Its Applications, pp. 139-143. 
July 1995 

[16] Heng-Da Cheng,* Senior Member, IEEE, Yui Man Lui, and Rita I. 

Freimanis "A Novel Approach to Microcalcification Detection 
Using Fuzzy Logic Technique" ;IEEE transactions on medical 
imaging, vol. 17, no. 3, June 1998 

[17] F. Laine, J. Fan, andW. Yang, "Wavelets for contrast enhancement 

of digital mammography," IEEE Eng. Med. Biol. Mag., vol. 14, 
no. 5, pp.536-550, Oct. 1995. 

[18] S. Yu and L. Guan, "A CAD system for the automatic detection of 

clustered microcalcifications in digitized mammogram film," IEEE 
Trans. Med. Imaging, vol.19, no.2, pp.1 15-126, 1998. 






Dr.ILA.Vennila received the B.E Degree in 
Electronics and Communication Engineering from 
Madras University, Chennai in 1985 and ME 
Degree in Communication System from Anna 
University, Chennai in 1989. She obtained Ph.D. 
Degree in Digital Signal Processing from PSG 
Tech, Coimbatore in 2006. Currently she is 

f | working as Assistant Professor in EEE 

Department, PSG Tech and her experience started from 1989; she 
published about 35 Research Articles in National, International 
Conferences National and International journals. Her area of interests 
includes Digital Signal Processing, Medical Image processing, Genetic 
Algorithm and fuzzy logic. 



Mr.C.Gowri Shankar received the B.E. 

Electrical and Electronics Engineering from 

Periyar University in 2003 and M.E Applied 

Electronics from Anna University, Chennai in 

2005. Since 2006, he has been a Ph.D. 

candidate in the same university. His research 

Wk interests are Multirate Signal Processing, 

Computer Vision, Medical Image Processing, 

and Pattern Recognition. Currently, he is working in Dept of Electrical 

and Electronics Engineering, Velalar College of Engineering and 

Technology, Erode. 




T.Balakumaran received the Bachelors degree in 
Electronics and Communication Engineering 
from Bharathiyar University, Coimbatore in 
2003 and the Master degree in Applied 
Electronics from Anna University, Chennai in 
2005. Since then, he is working as a Lecturer in 
Velalar College of Engineering and Technology 
(Tamilnadu), India. Presently he is a Part time 
(external) Research Scholar in the Department of 
Electrical Engineering at Anna University, 
Coimbatore (India). His fields of interests 
include Image Processing, Medical Electronics and Neural Networks. 




125 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 



The Fast Haar Wavelet Transform for Signal & Image 

Processing 



V.Ashok 
Department of BME, 
Velalar College of Engg.&Tech. 
Erode, India -638012. 



T.Balakumaran 

Department of ECE 

Velalar College of Engg.&Tech 

Erode, India -638012 



C.Gowrishankar 
Department of EEE 
Velalar College of Engg.&Tech 
Erode, India -638012 



Dr.ILA.Vennila 
Department of ECE, 
PSG College of Technology, 
Coimbatore, 
TamilNadu, India 



Dr.A.Nirmal kumar 

Department of EEE, 

Bannari Amman Institute of Technology, 

Sathyamangalam, 

TamilNadu, India 



Abstract- A method for the design of Fast Haar wavelet for 
signal processing & image processing has been proposed. In 
the proposed work, the analysis bank and synthesis bank of 
Haar wavelet is modified by using polyphase structure. 
Finally, the Fast Haar wavelet was designed and it satisfies 
alias free and perfect reconstruction condition. 
Computational time and computational complexity is 
reduced in Fast Haar wavelet transform. 

Keywords- computational complexity, Haar wavelet, 
perfect reconstruction, polyphase components, Quardrature 
mirror filter. 



I. 



Introduction 



The wavelet transform has emerged as a cutting edge 
technology, within the field of signal & image analysis. 
Wavelets are a mathematical tool for hierarchically 
decomposing functions. Though routed in approximation 
theory, signal processing, and physics, wavelets have also 
recently been applied to many problems in computer 
graphics including image editing and compression, 
automatic level-of-detailed controlled for editing and 
rendering curves and surfaces, surface reconstruction from 
contours and fast methods for solving simulation 
problems in 3D modeling, global illumination and 
animation [1]. 

Wavelet theory was developed as a consequence in the 
field of study the multi-resolution analysis. Wavelet 
theory can determine the nature and relationship of the 
frequency and time by analysis at various scales with 
good resolutions. 

Time-Frequency approaches were obtained with the 
help of Short Time Fourier Transform (STFT). For the 
better time (or) frequency resolution (but not both) can be 
determined by individual preference (or) convenience 
rather than by necessity of the intrinsic nature of the 
signal, the wavelet analysis gives the better resolution [2] . 



According to the applications, the biomedical 
researchers have large number of wavelet functions from 
which to select the one that most closely fits to the 
specific application. Wavelet theory has been successfully 
applied to a number of biomedical problems [3-5]. Many 
applications such as image compression, signal & image 
analysis are dependent on power availability. In this 
paper, a method for design of Haar wavelet for low power 
application is proposed. The main idea of this proposed 
method is the decimated wavelet coefficients are not 
computed. This makes the conservation of power and 
reduces the computation complexity. The Haar wavelet 
which makes the low power design is simple and fast. The 
proposed design approach introduces more savings of 
power. 

This paper organised as follows. In Section II, the 
existing Haar wavelet is introduced. In section III presents 
Haar wavelet analysis bank reduction. In section IV 
presents Haar wavelet synthesis bank reduction. In section 
V presents Haar wavelet and Fast Haar wavelet 
experimental results are shown as graphical output 
representation to the signal and image processing and we 
conclude this paper with section VI. 



II. 



HAAR WAVELET STRUCTURE 



d(n) 



1 — f 


bo(n) 


► 


I 2 


► 


t 2 


► 


ao(n) 


i 


— ► 


bi(n) 


► 


I' 


► 


t 2 


# 


■ ai(n) 





m 



Fig. 1 Two channel wavelet structure 



126 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 



The wavelet transform can be implemented by a two 
channel perfect reconstruction (PR) filter bank [6] . A filter 
bank is a set of filters, which are connected by sampling 
operators. Fig.l shows an example of a two-channel filter 
bank applied by one dimensional signal. d(n) is an input 
signal and d R (n) is reconstructed signal. In the analysis 
bank, b (n) is a analysis low pass filter and bi(n) is a 
analysis high pass filter. However in practice, the 
responses overlap, and decimation of the sub-band 
signals, which are results in aliasing. The fundamental 
theory of the QMF bank states that the aliasing in the 
output signal d R (n) can be completely canceled by the 
proper choice of the synthesis bank [7]. In the synthesis 
bank, a (n) is the reconstruction low pass filter(LPF) and 
ai(n) is the reconstruction high pass filter (HPF). Low 

pass analysis coefficients of Haar Wavelet is ^v™ "Y^l. 
High pass analysis coefficients of Haar Wavelet is 

y*. i.M.-. Low pass synthesis coefficients of Haar 

Wavelet is "V. v2t. High pass synthesis coefficients of 

[** -X] 

Haar Wavelet is *i S y2P. 



D 1 (Z) = |[Z)(Z 1/2 )5 1 (Z 1/2 )] 



(4) 



III. 



HAAR WAVELET ANALYSIS BANK REDUCTION 





* 


B (Z) 


» 


I 2 








w 

D(Z) 














1 


Bi(Z) 


* 


i 2 









* Do(Z) 



* Di(Z) 



Fig. 2 Analysis bank of wavelet structure 

Fig. 2 shows analysis bank of wavelet structure. d(n) is 
an input signal, d (n) is an low pass output of d(n) and 
di (n) is high pass output of input signal. 



For simplicity write in Z domain 

D (Z) = |[D(Z 1/2 )B (Z 1/2 ) + D(-Z 1/2 )B (-Z 1/2 )] 

D 1 (Z) = ^[D(^ /2 )B 1 (Z 1/2 )+D(-^ /2 )B 1 (-Z 1/2 )] 



(1) 

(2) 



At Perfect Reconstruction condition, No Aliasing 
Components presents 



From Quadrature Mirror Filter by [7], analysis filters 
are chosen as follows 



B (Z) = B(Z)<^b(n) 

B l (Z) = B(-Z)^(-l) n b(n) 



(5) 
(6) 



Transfer function B(Z) of an LTI system can 
decomposed into its polyphase components [9] . 



B(Z) can be decomposed into 

B (Z)=Y J :: o l Z^B A Z M ) 

In Haar Wavelet M=2 

So Low pass filter & High pass filter is 

B (Z) = B 00 (Z 2 ) + Z - l B 0l (Z 2 ) 

B l (Z) = B m (Z 1 )- Z - , B 0l (Z 1 ) 

Sub B (Z), B, (Z) in Eq (3) & (4) 
D (Z) = \[D(Z U2 )(B m (Z) + z -" 2 B m (Z))] 



D {Z)=\[I\Z V2 ){BJ^^z m IXZ V2 )B m {Z)] 



In Haar wavelet B o(Z) = B i(Z) 

D Q (Z) = B QQ (Z)[\D(Z m ) + \Z- m D(Z m )] 



Like 



D l (Z) = D(Z m )B 00 (Z)-±Z- y2 D(Z m )B 0l (Z) 



D l (Z) = B 00 (Z)[±D(Z m )-±Z- m D(Z m )] 
Combining Eq (11) & (12) 



D(Z) 



B W (Z) 






I 2 


+ 










7; y 








— ► 


i 2 





Fig. 3 Modified analysis bank structure 



(7) 



(8) 
(9) 



(10) 



(11) 



(12) 



Dod) 



D t <Z> 



D„(Z) = ^[D(Z 1,2 )£ (Z 1 ' 2 )] 



(3) 



127 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 



B 00 (Z)^ — ►boo(n). In haar wavelet b 00 (n) = -=. 




Fig. 4 Fast Haar wavelet analysis bank 

Shifting the down sampler to the input bring reduction 
in the computational complexity of factor 2 along with it. 
Fig. 4 shows Fast Haar wavelet analysis structure 
compared to original Haar wavelet structure, Number of 
arithmetic calculations are reduced in Fast Haar wavelet 
structure. But using above method computational 
complexity [10] reduced in less than quarter of original 
computational complexity. 



IV. HAAR WAVELET SYNTHESIS BANK REDUCTION 



D:ilj 



^7 



A 5 (Z) 




D=CJ 



[-.'.::■ 



'— t 



A,(Z) 



Refer to Eq (7) 

A(Z) is decomposed into 

A(Z) = ^Z-"AA(Z M ) 

In Haar Wavelet M=2 

A (Z) = A 00 (Z 2 )+ Z - l A m (Z 2 ) 
A 1 (Z) = -A 00 (Z 2 ) + z" 1 A 01 (Z 2 ) 



(16) 

(17) 
(18) 



SubEq. 17 & 18 in (13) 

D^(Z)=D (Z 2 )[4^ 

D R (Z) = A 00 (Z)[D (Z 2 )-D l (Z 2 )] + z-\ l (Z 2 )[D (Z 2 ) + D l (Z 2 )^ 

(19) 

Up sampler at the input of the synthesis filter bank will 
moved to output. So Eq.(19) can be drawn by 




Di(Z) 7 



Fig. 6 Modified synthesis bank structure 

In Haar wavelet A o(Z)= A i(Z)= B 00 (Z) 
In Haar wavelet b o(n) = a o(n)= "7» f 
Draw in time domain 



Fig. 5 Synthesis bank of wavelet structure 

Fig. 5 shows synthesis bank of wavelet structure. d (n) 
is low pass input signal, di (n) is high pass input signal and 
d R (n) is reconstructed signal 



For simplicity write in Z domain 

D R (Z) = A (Z)D (Z 2 ) + A l (Z)D l (Z 2 ) 



(13) 



From Quadrature Mirror Filter by [8] at perfect 
reconstruction, filters are chosen as follows 



A (Z) = 2B(Z) <-> 2b (n) 



A X {Z) = -A(-Z) = -2B(-Z) <-> 2(-l) n+l b(n) 



(14) 



(15) 



UQ — J |2 I — ^Q — >Q+ a R (n; 




\ifi 



Fig. 7 Fast Haar wavelet synthesis bank 

Combining Fig. 4 & Fig. 7, Fast Haar Wavelet 
Structure is obtained. Compared to Fig. 2, Number of 
Mathematical calculations are reduced in Fast Haar 
Wavelet Structure is shown in Fig. 8. 



128 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 




Detail data 



Fig. 8 Fast Haar wavelet structure 
V. EXPERIMENTAL RESULTS 

The results of applying, for one subject, which the 
signal is taken from laser based noninvasive Doppler 
indigenous developed equipment, the novel Fast Haar 
wavelet with approximation data are shown in Fig. 9 
shows that difference between original haar wavelet and 
Fast haar wavelet are matched well. The Error rate 
between existing and proposed Fast Haar wavelet at - 
90dB are shown in Fig. 11. 



60.5 
60 

59.5 
59 



57.5 
57 

56.5 
56 



Approximation data 



Original Haar wavelet 
- Fast Haar wavelet 



mu\kw^^^^H 







it f t 



10 20 



40 50 60 
Time Period 



80 90 



100 



Fig. 9 Results of approximation data compared to existing and Proposed 
Fast Haar wavelet Transform. 

Similarly from the same novel Fast Haar wavelet with 
detail data are shown in Fig. 10 shows that difference 
between original Haar wavelet and Fast Haar wavelet are 
matched well. The Error rate between existing and 
proposed Fast Haar wavelet at -160dB to -220dB are 
shown inFig.ll. 




Fig. 10 Results of detail data compared to existing and Proposed Fast 
Haar wavelet Transform. 

Error Signal 



Approximation Error 
Detail Error 




10 20 30 40 50 60 70 80 90 100 

Time Period 

Fig. 1 1 Results of Error rate compared to existing and Proposed Fast 
Haar wavelet Transform 

We have checked our proposed method in image 
processing also. Lowpass output was obtained by applying 
original Haar wavelet and proposed Fast Haar wavelet. 
Fig. 12(a) shows Lena image, Fig. 12(b) shows lowpass 
image of lena by applying original Haar wavelet transform 
and Fig. 12(c) shows lowpass image by applying Fast Haar 
wavelet transform. Fig. 12(d) shows difference between 
Fig. 12(b) & Fig.l2(c) From the Fig. 12(d), it is clearly 
visible difference value for all coefficients are less. 



129 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 




(c) (d) 

Fig. 12 Comparison of Fast haar wavelet with original Haar wavelet 
a) Lena image (b) Lowpass of Lena image by original Haar wavelet 

(c) Lowpass of Lena image by Fast Haar wavelet 
(d) Difference between lowpass output by original Haar wavelet & 
Fast Haar wavelet 

VI. CONCLUSION 

This work presents a novel Fast Haar wavelet 
estimator, for application to biosignals such as 
noninvasive doppler signals and medical images. . In this 
paper, signals and images are decomposed and 
reconstructed by Haar wavelet transform without 
convoution. The proposed method allows for the dynamic 
reduction of power and computational complexity than 
the conventional method. The error rate between the 
conventional and the proposed method was reduced in the 
signal and image procesing. 

REFERENCES 

[1] Eric J.stollnitz, Tony D.Derose and david H.Salesin, Wavelets for 
computer graphics - theory and applications book, Morgan 
kaufmann publishers, Inc. San Francisco California 

[2] O. Rioul and M. Vetterli , 'Wavelets and Signal Processing', IEEE 
Signal Processing Mag, pp 14-18, Oct 1991. 

[3] Fig.liola and E. Serrano, 'Analysis of physiological time series 
using wavelet transforms,' IEEE Eng. Med. Biol, pp 74 - 80, 
May/June 1997. 

[4] Aldroubi and M.A. Unser, Eds , Wavelets in Medicine and Biology. 

[5] P. Salembier, 'Morphological multiscale segmentation for image 
coding,' Signal Process, vol. 38, pp. 359-386, 1994. 

[6] P.P. Vaidyanathan , 'Multirate Systems and Filter Banks,' 
Englewood Cliffs, NJ: Prentice-Hall, 1993. 

[7] P.P. Vaidyanathan; Quadrature Mirror filter banks, M band 
extensions and Perfect Reconstruction Techniques. IEEE ASSP 
Magazine, Vol 4, pp 4-20, July 1987. 

[8] J.H.Rothweiler: polyphase, Quadrature filters - A new subband 
coding technique. IEEE ICASSP'83, pp.1280-1283, 1983 

[9] R.E.crochiere,L.R.Rabiner: Multirate Digital signal processing. 
Englewood Cliffs :prientice hall, 19 83. 

[10] MJ.T Smith, T.P.Barnwell III: A Procedure for designing exact 
Reconstruction filter banks for tree structured Sub-band Coders. 
Proc IEEE ICASSP'84, pp.27. 1.1 -27. 1.4, March 1984. 





AUTHORS PROFILE 

Mr.V.Ashok received the Bachelors degree in 
Electronics And Communication Engineering 
from Bharathiyar University, Coimbatore in 
2002 and the Master degree in Process 
Control And Instrumentation Engineering 
form Annamalai University, Chidambaram in 
2005. Since then, he is working as a Lecturer 
in Velalar College of Engineering and 
Technology (Tamilnadu), India. Presently he is a Part time 
(external) Research Scholar in the Department of Electrical 
Engineering at Anna University, Chennai (India). His fields of 
interests include Medical Electronics, Process control and 
Instrumentation and Neural Networks. 



Mr.T.Balakumaran received the Bachelors 
degree in Electronics and Communication 
Engineering from Bharathiyar University, 
Coimbatore in 2003 and the Master degree 
in Applied Electronics from Anna 
University, Chennai in 2005. Since then, he 
is working as a Lecturer in Velalar College 
of Engineering and Technology 
(Tamilnadu), India. Presently he is a Part time (external) Research 
Scholar in the Department of Electrical Engineering at Anna 
University, Coimbatore (India). His fields of interests include 
Image Processing, Medical Electronics and Neural Networks. 

Mr.C.Gowri Shankar received the B.E 
Electrical and Electronics Engineering from 
Periyar University in 2003 and M.E Applied 
electronics from Anna University, Chennai 
in 2005. Since 2006, he has been a Ph.D. 
candidate in the same university. His 
research interests are Multirate Signal 
Processing, Computer Vision, Medical 
Image Processing, and Pattern Recognition. Currently, he is 
working in Dept of Electrical and Electronics Engineering, Velalar 
College of Engineering and Technology, Erode. 

Dr.ILA.Vennila received the B.E Degree in 
Electronics and Communication 

Engineering from Madras University, 
Chennai in 1985 and ME Degree in 
Communication System from Anna 
university, Chennai in 1989. She obtained 
Ph. D. Degree in Digital Signal Processing 
from PSG Tech, Coimbatore in 2006. 
Currently she is working as Assistant Professor in EEE 
Department, PSG Tech and her experience started from 1989; she 
published about 35 Research Articles in National, International 
Conferences National and International journals. Her area of 
interests includes Digital Signal Processing, Medical Image 
processing, Genetic Algorithm and fuzzy logic 





^^^^^ Dr.A.Nirmalkumar. A, received the 

JM B.Sc.(Engg.) degree from NSS College of 

Engineering, Palakkad in 1972, 
M.Sc.(Engg.) degree from Kerala 
University in 1975 and completed his Ph.D. 
degree from PSG Tech in 1992. Currently, 
he is working as a Professor and Head of 
the Department of Electrical and 
Electronics Engineering in Bannari Amman 
Insititute of Technology, Sathyamangalam, Tamilnadu, India. His 
fields of Interest are Power quality, Power drives and control and 
System optimization. 




130 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



A Survivability Strategy in route optimization 
Mobile Network by memetic algorithm 



K .K. Gautam 

Department of Computer Science & Engineering 

Roorkee Engineering & Management Technology Institute, 

Shamli (247774) India 



Anurag Rai 

Department of Information Technology 

College of Engineering Roorkee 

Roorkee (247667) India 



Abstract — The capability to provide network service even 
under a significant network system element disruption is the 
backbone for the survival of route optimize of mobile network 
Technology in today's world. Keeping this view in mind, the 
present paper highlights a new method based on memetic 
algorithm . 

Keywords- Survivability, Mobile Network, memetic algorithm., 
PAN,. 

I. Introduction 

Network survivability is considered to cope, it with 
increasing demand for reliable network system. Network 
survivability for the route optimization is an essential aspect of 
reliable communication service. Survivability consists not 
only of robustness against failure occurring due to natural 
faults. In mobile networks infrastructure element such as base 
station (BS), base station Controller (BSC), wired links, and 
mobile switch centre (MSC), are employed to provide and 
maintain essential services, hence the operation interruption of 
a network component affects overall or partial network 
services. Wireless access network have unique characteristics 
to support mobile users, which can result in different 
survivability and security aspect [3]. Therefore wireless 
survivability strategies must be designed to improve the 
service available rate of the network component system (4), 
(3). 

Mobile user authentication is necessary when a mobile user 
wants to request service provided by the service providers 
survivable (SPS) in the visited domains. In this paper, we 
present a survivability strategy in mobile networks method by 
the use of memetic algorithm. 

A network could be as simple as a forum held in a city 
between people, where people use the opportunity to 
communicate with each other, they use a network by the use of 
memetic algorithm has the potential for setup the survivability. 
Fundamental to distribute mechanics is the effect of 
measurement on a state. If some properly of a general state is 
measured, it collapses to an eigenstate of the property and 
cannot be 'rebuilt' into the original state. Information can be 
encoded into a general optimize set up. Mobile networks can 
have very complex form of hierarchy e.g. Mobile networks in 
a mobile network visiting mobile nodes (VMNS) in mobile 



networks and so on. This situation is called as a nested mobile 
network. 

Many important problems arising in science, industry and 
commerce, mobile networks fall very neatly into the read- 
made category of optimization problem. 

II. Survivability 

Traditional security research is primarily focused on the 
detection and prevention of intrusion and attacks rather than 
on continued correct operation while under attack. Fault 
tolerance is usually concerned with redundancy that is 
required to detect and correct up to a given number of 
naturally occurring faults. Nature is not malicious and 
conventional failure model make significant assumptions, in 
particular, assuming faults to be independent and random. The 
presence of intelligent adversarial attacks can protocol 
vulnerability often become more important considerations in 
the presence of an adversary [1,2] 

There are a number of definitions of survivability. The 
one we use here is from the Software Engineering Institute, 
which emphasizes timeliness, survivability under attack and 
failure, and that detection of attack is a vital capability. 

Survivability is the capability of a system to fulfill its 
mission in a timely manner, even in the presence of attacks or 
failures. Survivability goes into the demon of security and 
fault. Tolerance of focus on delivery of essential service even 
when system is entered or experiences failures, and rapid 
recovery of full service, when conditions improve. Unlike 
traditional security measures that require central control and 
administrative, survivability addresses highly distributed 
unbounded network environment that lacks central control and 
unified security policies. Mobile networks can have very 
complex form of hierarchy e.g. Mobile networks in a mobile 
network visiting mobile nodes (VMNS) in mobile networks 
and so on. 



131 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



III. The Three Rs: Resistance, Rrcognition, And 
Recovery 



The focus of survivability is on delivery of essential 
services and preservation of essential assets. Essential service 
and asserts are those system capabilities that are critical o 
fulfilling mission objectives. Survivability depends on three 
key capabilities: resistance, recognition, and recovery. 
Resistance is the capability to detect attacks as they occur and 
to evaluate the extent of damage and compromise. Recovery, a 
hallmark of survivability is the capability to maintain essential 
service and asserts during attacks, limit the extent of damage 
and restore full service following attack. 

We further extend this definition to require that 
survivability system be able to quickly incorporate lesion 
learned from failure, evolve, and adapt to emerging threats. We 
call this survivability feature refinement. 

We can classify survivable mobile wireless networking 
requirement into four categories based on (2): (i) resistance 
requirement; (ii) recognition requirement; (iii) recovery 
requirements; and (iv) refinement requirement. We can also 
describe a requirement definition process (3). This includes the 
definition of system and survivability requirement, legitimate 
and intruder usage requirement, development requirement, 
operation requirement, and evolution requirement. Essential 
service must be identified and specified for the penetration, 
exploration, and exploitation phases of the attack. 

The approach has guided this work and is recommended for 
more detailed requirement analyses for future mobile wireless 
network. 

Ultimately, there are two distinct aspects of survivability 
that apply at all networking layers. 

IV. Information Access Requirement 

Does the user have access to the information or service 
required to complete the task in the presence of failure or 
attack? For eg. it is possible to replicate service or information 
and provide them locally when the network gets partitioned. 
However End-to-end communication should not be mandated 
in these cases. 



V. End-To-End Communication 

On the other hand there are interactive application , inter- 
personal communication such as voice calls, or dynamically 
generated information such as current sensor data, which 
require true end-to-end connectivity. Do existing sessions 
survive? Can the user create new session to reach the intended 
communication end- point even in the presence of failures and 
attacks? This requires that the communication end-point 
themselves survive and that the communication end-points 



themselves survive and that the adversary must not be able to 
permanently partition the network. Furthermore, the adversary 
must not be able to permanently disable access to required 
services such as authentication, naming, resource discovery, or 
routing. 

VI. Mobile Network Survivability 

Existing work on survivability in the context of cellular 
telephone networks concentrates primarily on infrastructure 
survivability (for e.g. see the outage index metrics (80) and 
does not consider adversarial attacks. However, they offer 
some insight on quantifying survivability and the role of 
network management tools. 

Networks are vulnerable during upgrades, especially those 
involving software. Furthermore, rapid evolution leads to 
learning - cure problems as well as - over - concentration 
leads or service into single points of failure. This problem is 
exacerbated by deficits in network management tools to 
operate and maintain increasingly complex system. 
Architectural improvement applicable to mobile include the 
use of redundant networks 



VII. 



Base Station 



In more environment, a cell that is geographical region unit 
is covered by the radio frequency of a base station. Each call is 
controlled by a BS which has a fixed connection to a BSC (or 
RNC). In mobile network infrastructure element such as base 
station controller (BSC), wired links and mobile switch centre 
(MSC) are employed to provide and maintain essential 
service. Hence the operation interruption of a network 
component affects overall or partial network services. 

A radiation antenna is classified as omni directional and 
directional with an ommni directional antenna, a single 
frequency spreads out in all directions of 360 coverage. A cell 
is directional antenna with each different set channel. 



VIII. 



System State Of Base Station 



The BS system, including antenna parts cannot provide 
partial or whole service function for coverage cell when single 
or more fatal failures occur in the BS system. In this paper, we 
consider that system failures are caused by key distribution 
method. For example, by interrupt sequence mishandling, 
overall system operation falls into failure state because of 
unanticipated handled interruption to a component of the 
system. 

IX. Personal Area Network 

A mobile network can have a hierarchical structure e.g. 
mobile network within another mobile network. This situation 
is referred to as nested mobile network. A personal area 
network (PAN) may travel a vehicle, which also contains a 
mobile network of larger on scale, fig 1 illustrate a simple 



132 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



larger scale. MR-1, MR-2 are attached to their own home 
links. A wireless personal area network moves as a single unit 
with one or more mobile routers that connect it to global 
internet. 

X. Multi Objective Optimization (MOO) 



This fig is also defined a routing inefficiency for the traffic 
management and designed an imported rout optimization 
schemes for traffic management of mobile networks. Thes 
concept of traffic management for the network mobility was 
introduce the signaling over heads of a number of hosts 
moving as a group as MRs. 



An unaccompanied multi objective optimization problem 
is an example of route optimization for mobile network. 
Because mobile moves as a single unit with one or more 
mobile routers that connect it to the global internet[9,ll] We 
defined this problem as 
"Minimize" z = f(x) 

Where f(x) = (fl(x), £2(x) m(x)) 

Subject to xGx 

Z 2 minimize PF {0} 




f : x- 



->z 



fig-1 



Fig-1 an example of multi objective optimization problem 
with mobile search space (MSS) x, as vector fitness function f 
that maps solution in x to objective vector made up of two 

component (mobile routers) 'costs' z x and the z 2 

minimized. 

Here if we define 

a=mobile router - 1 

b=mobile router-2 



XL Memetic Algorithm Approach 

The impressive record of Memetic Algorithms producing 
high quality solution in combinalarial optimization and in real 
- world application (e.g. see page 220[5]) is sometimes cited 
as a testament of their inherent effectiveness of robustness as 
black box search. However, since the advent of the no free 
lunch theorems (6,7,8) we know that MA;S like any other 
search algorithm, are only reality good to the extent to which 
they can be "aligned" to the specific features of a route 
optimization problems in mobile networks. None the less, 
MAs, like therefore bears evolutionary algorithms (EAs), do 
have unassailable advantage over other more traditional search 
techniques: that is their flexibility. This flexibility has 
important advantage, as has to solve mobile route optimization 
problems: one is to choose some traditional techniques. And 
them simplify or otherwise other the problems. 

As in any other optimization scenario as route 
optimization problems, we should know that the out set what 
is a desirable outcomes; the Memetic Algorithm frame work 
proposed above required. The operators and procedures be 
selected based on their current success abs. When a mobile 
network moves from one place to another it change its point 
attachment to the internet, which also makes changes to its 
reach ability and to the internet topology. [9, 10,1 1] 

XII. Performance Measures In Mas For Moo 

If one is developing or using an algorithm for optimization 
it almost goes without saying that there should be some way to 
measures its performance. In MOO the situation is the same 
regarding the time aspect of performance assessment but the 
quality aspect is clearly more difficult, the extensive array of 
existing met heuristic, issues and methods reviewed in the 
section above gives a richer basis from which to design 
new MAs than do the existing MAs for MOO themselves. In a 
typical cellular network, the area of coverage is after 
geographically divided into hexagonal cells. The call is the 
basic unit of a cellular system. 

In recent years, Muscat and Krasnogor have provided a 
guiding manifesto for putting the "Memetic" back in Memetic 
algorithm (9,10) advocating. 



And 

l=access router -1 

2=access router-2 



Candidate MA framework for MOO 

• MN: = initialize(MN) 

• A: = Nondom (MN) 



133 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



Algorithm Candidate MA framework for MOO 

MN: = Initialize (MN) 
MN: = Nondom (MN) 
while stopcriterion not satisfied do 
while stagnation criterion not satisfied do 
SAMN:=SelectFrom(PUA,sel_sc hed(succ(SEL))) 
SAMN:=Vary(SAMN,var_sched(succ(VAR))) 
SAMN":=LocalSearch(SAMN",Is_sched(succ(LS))) 
MN:=Replace(PUC",rep_sched(succ(RED))) 
A:=Reduce(Nandom(AUSAMN),red_sched 
(succ(RED))) 
end while 

MN:=Randomlmmigrants(P,imm_sched(succ(IMM))) 
return (A) 

Here we represent an Algorithm, we put forward a simple 
framework that could serve as a guide for making a more 
Memetic MA for MOO. In linel, MN (Mobile Networks) of 
solution is initialized. As usual this procedure may be simply 
random or it may employ some heurists (s). Line 2 sets the 
archive A to the no dominated solution from MN. Thereafter, 
the main loop of the MA begins line 4 sets up an inner loop in 
which a stagnation criterion is checked. This should be based 
on some memeplex which monitors progress in diversity, 
proximity, and /or some other criteria. Line5-9 gives a very 
high level description of the update of the MN and archive. 
Five different 'schedulers' are employed, basically 
corresponding to mating selection, reproduction, lifetime 
learning, survival selection and update of the archive, 
respectively. Each scheduler chooses from a memeplex of 
operators, based on estimates of the current success of those 
operators. E.g. in line 5, Select from is the operation of mating 
selection, the domain of which is the union of the MN and 
archive, and co-domain is a Small Area Mobile Networks 
(SAMN), the selection is controlled by the scheduler, 
selsched, which use a success measures to choose one 
operators for the set SEL, of currently available operators for 
selection. Notice that MN and A are potentially of variable 
size, in this scheme. In line 11, the MN is updated using some 
immigration policy to rerelease it from stagnation, the archives 
of no dominated solution are returned in line 13. 

The frame work proposed is rather broad and actually 
instantiating it requires us to consider how we should resolve 
many choices, including those considered in the following 
sections, at the very least? 



Conclusion 
In this paper, we have proposed a scheme for mobile 
service use of BS system and memetic alogithm. The 
survavility of Route optimization scheme in nested mobile 
network modifying the process of Memetic Algorithm . And 
hence the basic support protocol for survivability of Route 
optimization scheme for mobile network needs to be 
extended with an appropriative route optimization scheme, we 
proposed scheme can achieve the mobile route optimization 
environment, it may get a survivability scheme. 

Acknowledgment 
The author would like to thank Dr. H.N. Dutta Director, 
REMTech for his moral support in carrying out the work 



References 

[I] J.Kabara , P. Krishna Murthy, and D. Tipper, "Information auuurance 

in wireless network". In proc. IEEE workshop on DIREN'02. 
[2] U. Varshney A.P Snow and A.D. Malloy, "Designing Survivable 

wireless and mobile network" in proc. IEEE WCNC'99, neworeleans, 

LA, Sep. 1994, pp.30-34. 
[3] D. Tipper, S. Rammaswamy, and T. Dahiberg,"pcs network 

survivability" in proc. IEEE WCNC, new or leans LA, sep 1999, 

invited paper. 
[4] D. Samfat, R. molva, N. Asokan, "Untraceability in mobile Network" 

Proceeding of Mobi COM' 95, Berkely, November 1995. 
[5] Zhang Bin, Wujing-Xing Proc. Of the feb.2003 ICCNMC'03 IEEE. 
[6] Sangjoon Park, Jiyoung Song, Byunaggi Kim, IEEE Trans of Veh. 

Tech. Vol. 55 pp 328-339. 
[7] Ashotosh Dutta , James Burns , K. Daniel Wong, Ravi jain, Ken 

Young Telcordia Technologies, 445 South Street, Morristown, NJ 

07960 pp 1-6 
[8] Sangjoon Park , Jiyoung song, and Byunggi Kim, Member, IEEE "A 

Survivability Strategy in Mobile Networks. IEEE TRANSACTION 

ON TECHNOLOGY, VOL. 55,NO.328-340. 
[9] M.A. Abido, A new multiobjective evolutionary algorithm for 

environmental /economic power dispatch. In Power Engineering 

Society Summer Meeting, Vol. 2, page 1263-1268, IEEE, 2001. 
[10] Pragya Lamsal, "Network Mobility", Research Seminar on Hot 

Topics in Internet Protocols, page 1-2. 

[II] Joshua Knowles and David Corne, "Memetic algorithms for 
multiobjective optimization: issues, methods and prospect, page 1-40. 



AUTHORS PROFILE 

Authors Profile .. K K Gautam is the Dean in the Roorkee Engineering & 
Management Technology Institute, Shamli-247 774, India. 

Anurag Rai is the head of the Information Technology Department in 
College of Engineering Roorkee, Roorkee- 247 667, India. 



134 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 



Analysis of Large-Scale Propagation Models for 
Mobile Communications in Urban Area 



M. A. Alim , M. M. Rahman, M. M. Hossain, A. Al-Nahid 

Electronics and Communication Engineering Discipline, 

Khulna University, 

Khulna 9208, Bangladesh. 

* Corresponding author. 



Abstract — Channel properties influence the development of 
wireless communication systems. Unlike wired channels that are 
stationary and predictable, radio channels are extremely random 
and don't offer easy analysis. A Radio Propagation Model 
(RPM), also known as the Radio Wave Propagation Model 
(RWPM), is an empirical mathematical formulation for the 
characterization of radio wave propagation as a function of 
frequency. In mobile radio systems, path loss models are 
necessary for proper planning, interference estimations, 
frequency assignments and cell parameters which are the basic 
for network planning process as well as Location Based Services 
(LBS) techniques. Propagation models that predict the mean 
signal strength for an arbitrary transmitter-receiver (T-R) 
separation distance which is useful in estimating the radio 
coverage area of a transmitter are called large-scale propagation 
models, since they characterize signal strength over large T-R 
separation distances. In this paper, the large-scale propagation 
performance of Okumura, Hata, and Lee models has been 
compared varying Mobile Station (MS) antenna height, 
Transmitter-Receiver (T-R) distance and Base Station (BS) 
antenna height, considering the system to operate at 900 MHz. 
Through the MATLAB simulation it is turned out that the 
Okumura model shows the better performance than that of the 
other large scale propagation models. 

Keywords- Path Loss; Okumura model; Hata model; Lee 
model; 

I. Introduction 

In mobile radio systems, path loss models are necessary for 
proper planning, interference estimations, frequency 
assignments, and cell parameters which are basic for network 
planning process as well as LBS techniques that are not based 
on GPS system [3]. A Radio Propagation Model (RPM), also 
known as the Radio Wave Propagation Model (RWPM) or the 
Radio Frequency Propagation Model (RFPM), is an empirical 
mathematical formulation for the characterization of radio 
wave propagation as a function of frequency, distance and 
other conditions. Most radio propagation models are derived 
using a combination of analytical and empirical methods. In 
general, most cellular radio systems operate in urban areas 
where there is no direct line-of-sight path between the 
transmitter and receiver and where the presence of high rise 
buildings causes severe diffraction loss. 



Propagation models that predict the mean signal strength 
for an arbitrary transmitter-receiver seperation distance are 
useful in estimating the radio coverage area of a transmitter and 
are called large-scale propagation model. On the other hand, 
propagation models that characterize the rapid fluctuations of 
the received signal strength over very short travel distances or 
short time durations are called small scale or fading models [1]. 

In this paper, the wideband propagation performance of 
Okumura, Hata, and Lee models has been compared varying 
Mobile Station (MS) antenna height, propagation distance, and 
Base Station (BS) antenna height considering the system to 
operate at 900 MHz. Through the MATLAB simulation it 
turned out that the Lee model outperforms the other large scale 
propagation models. 



II. 



Literature Review 



Path loss characteristics of a channel are commonly 
important in wireless communications and signal propagation. 
Path loss may occur due to many effects, such as free-space 
loss, refraction, diffraction, reflection, aperture-medium 
coupling loss and absorption. Path loss is also influenced by 
terrain contours, environment (urban or rural, vegetation and 
foliage), propagation medium (dry or moist air), the distance 
between the transmitter and the receiver, and the height of 
antennas [4]. 

Path loss normally includes propagation losses caused by 

• The natural expansion of the radio wave front in free 
space (which usually takes the shape of an ever- 
increasing sphere) 

• Absorption losses (sometimes called penetration 
losses) 

• When the signal passes through media not transparent 
to electromagnetic waves, diffraction losses when part 
of the radiowave front is obstructed by an opaque 
obstacle and 

• Losses caused by other phenomena. 

The signal radiated by a transmitter may also travel along 
many and different paths to a receiver simultaneously; this 
effect is called multipath. Multipath can either increase or 
decrease received signal strength, depending on whether the 



135 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



individual multipath wavefronts interfere constructively or 
destructively. 

In wireless communications, path loss can be represented 
by the path loss exponent, whose value is normally in the range 
of 2 to 4 (where 2 is for propagation in free space, 4 is for 
relatively lossy environments. In some environments, such as 
buildings, stadiums and other indoor environments, the path 
loss exponent can reach values in the range of 4 to 6. On the 
other hand, a tunnel may act as a waveguide, resulting in a path 
loss exponent less than 2 [4]. 



The free-space path loss is denoted by L (d) , which is 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
A. Okumura Model 

The Okumura model for urban areas is a Radio propagation 
model that was built using the data collected in the city of 
Tokyo, Japan. This model is applicable for frequencies in the 
range of 150 MHz to 1920 MHz and distances of 1 km to 100 
km. It can be used for the base stations antenna heights ranging 
from 30m to 100m. 

To determine path loss using Okumara's model, the free 
space path loss between the points of interest is first determined 



L P (d) = -20 log 



10 



4jzd 



'(dB) 

where, c = velocity of light, fc = carrier frequency and d : 
distance between transmitter and receiver [2]. 

For log-distance path loss with shadowing the path loss is 

denoted by L p (d) , which is 



L p (d)cc 
or equivalently, 



d 



V^o J 



, d >d 



L p (d) = L p (d ) + I0nlog 



r d^ 



10 



V^o J 



(dB), d > d 



where, n = path loss component, dO = the close-tin 
reference distance (typically 1 km for macrocells, 100m for 
microcells), d = distance between transmitter and receiver [2]. 

III. Materials And Methods 

Calculation of the path loss is usually called prediction. 
Exact prediction is possible only for simpler cases, such as the 
above-mentioned free space propagation or the flat-earth 
model. For practical cases the path loss is calculated using a 
variety of approximations. 

Statistical methods (also called stochastic or empirical) are 
based on fitting curves with analytical expressions that recreate 
a set of measured data. Among the most commonly used such 
methods are Okumura Model, Hata Model, and Lee's Model. 

In the cities the density of people is high. So the more 
accurate loss prediction model will be a good help for the BTS 
mapping for optimum network design. Among the 

Radio Propagation Models (RPM) city models are to be 
analysed in this paper to find the best fitting city model. The 
well known propagation models for urban areas are: 

i) Okumura Model 

ii) Hata Model 

iii) Lee's Model 



KAf^d), 



is added to it along with 
the type of terrain. The 



and then the value of 
correction factors according to 
expression of the model [1]: 

L^ (dB) = L F+ A mu (f,d)-G(h te )-G(h re )-G AREA (1) 

where, L 50 = The 50 th percentile (i.e. median) value of 
propagation path loss., L F = The Free Space propagation Loss 
in dB, A mu = Median attenuation.relative to free space in dB, 
G(h te ) = The base station antenna height gain factor, G(h re ) = 
Mobile antenna height gain factor and G A rea = The gain due 
to the type of environment. 

f K A 



G(\) = 201og 



10 



G(h re ) = \0log 



200^ 



G(h re ) = 20 log 



10 
V 

f K 



10 



V 



I00m)h te )30m 



/z < 3m 



I0m)h re )3m 




Figure 1. 



30 40 50 BO 70 80 90 100 
BTS Antsnna Height (m) 



Propagation Path Loss due to the change in BTS antenna height for 
Okumura model. 



136 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



£ 






o 

s 

S 2B 

01 

s 
a 

g. 22 


V i ii 

\ j Cflrrier FreqU ency = 9ffl MHi 
\ MS dislancp from BTS = 10 km 
\^k Sl^antEnnp hEigFil ^STm 
i \. ■ i i 

i ■■ -p ■!■■■■ t ■ JV? ■■ »■■■■■-■ -"P"- ■-■■■■ -^"-m ■■-■»■ — 

1 ^^« ■ 1 

1 "■- 1 

I ^x__ I I 

, , ,1 ^-l 1 

i rV. 

i i ^""""^-^i 

i i i ^^-t_^ 




9 
a 


2 4 6 E 10 
ME Ajutenna Heighl (m) 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol, 7, No. J, 2010 

a(h re ) = 3.2(log 10 1 1 .75h„ f - 4.97dB 
for f c > 300MH Z 



Figure 2. 



Propagation Path Loss due to the change in MS antenna height for 
Okumura model. 




Figure 3. Propagation Path Loss due to T-R separation for Okumura model. 

B. Hata Model 

Hata Model is based on the Okumara's model where some 
correction factors are included. It works in the frequencies 
range from 150 MHz to 1500 MHz.. The standard formula for 
median path loss in urban areas is given by [1]. 



L 50 (urban)(dB) = 69.55 + 26. 1 6 log 10 f c -13.82 log 10 /zj 
-a(/O + (44.9-6.551og 10 /*Jlog 10 d 



.(2) 



where, fc =The frequency from 150 MHz to 1500 MHz, 
hte = The effective base station antenna height (30m to 200m), 
hre = The effective mobile antenna height (lm to 10m), d = 
The trnasmitter-receiver (T-R) distance in km and a(hre) = The 
correction factor for effective mobile antenna hieght which is a 
function of the size of coverage area. For a large city it is given 
by, 

a(h re ) = %.2%\og w \Mh re f-\.\dB 
for f c <300MH z 




Figure 4. Propagation Path Loss due to the change in BTS antenna height for 
Hata model. 




n 


140 


T3 










130 


Q 




—1 




_Z 

*3 


120 


□_ 




c 


110 


« 




— 




!□. 


100 


a 




0_ 





90 



---- 


■ III 


Ca[riEr Frequency = 90p MHz 

M? J?l s A*™ i$uy\ B JSj = J0 _km _ 
\~ BTS anlcnn a| height = 50 m 

\. ■ i i 




■ ^V ' 1 




^%J 1 




1 ^~*J 

1 1 ^ V ^"->—. 



4 B 

MS Antenna Height (m) 



10 



Figure 5. Propagation Path Loss due to the change in MS antenna height for 
Hata model. 









„ 130 

£ 120 

a 

1 no 

e 

a ioo 

m 

EL. 

£ 90 


Sl ■» ■ T ■ fTiTiiri 

■..■ m m * m k\ ■ _ _ _ _ 1 




/ Carrier F>e que ncy = EDO MHz 
/ l BTS antenna height = 50 m 

i ■ i J ■ ■ ■ ■ * f ■ ■ \«.CJ ■ w ■ ■ i, ■ r ■ ■■ ■ .r J^i ■ a.r ■ ■ ■ ■ r u m m ■ ■■ 

/ MS antenna height 4 3 m 
j p 1 1 

Jr ■ 4 1 


f r ; — — r - 




5 10 15 2 

MS Distance from BTS (km) 






Figure 6. Propagation Path Loss due to T-R separation for Hata model. 



137 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 



C. Lee Model 

Lee's path loss model is based on empirical data chosen 
so as to model a flat terrain. Large errors arise when the model 
is applied to a non-terrain. However, Lee's model has been 
known to be more of a "North American model" than that of 
Hata. 



The propagation loss calculated as: 



L(dBm) = 124 + 30.5 log, 



V"oy 



+ 10Hog 1( 






-a n 



..(3) 



where, d is in km, f and fc is in MHz, k = 2 for fc < 450 
MHz and in suburban/open area and 3 for fc > 450 MHz and in 
urban area, dO = 1.6 km. 



are: 



In this analysis the parameter values taken for calculations 

Carrier Frequency fc = 900 MHz 

Nominal (Calibration) Distance do = 1.6 km 

Base Mobile Station Antenna hb = 30.48 m 

Mobile Station Antenna Height hm = 3 m 

Base Station Transmit Power Pb = 10 W 

Base Station Antenna Gain Gb = 6 dB 

Mobile Station Antenna Gain Gm = dB with respect to 
isotropic antenna. 

/ is the transmitted frequency, d is the Transmitter- 
Receiver distance and a0 is a correction factor to account for 
BS and MS antenna heights, transmit powers and antenna gains 
that differ from the nominal values. As such, when the 
prevailing conditions differ from the nominal ones, then a0 is 
given by: 

a = 10 log lQ (a x a 2 a 3 a 4 a 5 ) 

where: 

al = (new BTS antenna height (m) / 30.48 m)2 

a2 = (new MS antenna height (m) / 3 m)v ;for MS antenna 
height < 3, v =1 and for MS antenna height >3, v =2; 

a3 = (new transmitter power / 10 W)2, in this paper the 
value of a3 is taken 1 . 

a4 = new BS antenna gain correction factor = (Gb / 4), 
Here a4 is considered 6 dB. 

a5 = frequency correction factor = (///c)-n ; for 2<n<3 and 
f and /c is in MHz. /c is taken 900 MHz. 




Figure 7. Propagation Path Loss due to the change in BTS antenna height for 
Lee model. 







m HO 

£ 136 

-j 

S 136 

L. 

"■H 134 

_- 
*? 

o. 

£ 132 


\ \ _i L I- 


\ | Cams* Frgq ujancy = 9QD bAHz 

_\ MS .disiartcE ! jfram BTS i= 1 D km 

V\ BTfe ensenn ^heighf = ^0 m 


! i !^~^-^_ 




2 4 6 8 10 
MS Antenna Height (m) 



Figure 8. 



Propagation Path Loss due to the change in MS antenna height for 
Lee model. 



Q 

^ 130 

1 

| 12G 

m 

£110 






^ — 


! 


......... 

if 


H ^__ I _r^ -- 1 

^*r J *^ 1 

-i*^ m 1 
. ■" 1 


^ 


Carrieti Frequ-enc y \= "900 MHz 
BTS ahtenna height = 90 nn 
L ~ "MS aiienria "Fieigrl] ^" 3" rn 






: ; 






S 10 15 2 

MS Distance from BTS (km) 


° 



Figure 9. Propagation Path Loss due to the T-R separation for Lee Model 

IV. Result 

From Fig. 10, it is seen that the propagation path loss 
decreases due to the increase in BTS antenna height for all the 
models. For Hata model the loss is maximum, for Lee model 
the loss is medium and for Okumura model the loss is 
minimum. From Fig. 11, it is seen that the propagation path 
loss increases with the decrease in MS antenna height for all 
the models. For Lee model the loss is maximum, for Hata 



138 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



model the loss is medium and for Okumura model the loss is 
minimum. 



£- 



14JQ 
120 


■■- 




Hata Model 
Lee Model 




100 








■ 


s: 








- 


en 










40 






Okumura Model 


" 




„ 








5Q BO 70 BQ 

BTS Artlenrta Height (rn) 



90 



ICO 



Figure 10. Comparison of Propagation Path Loss due to the change in BTS 
antenna height. 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
From Fig. 12 it is seen that for Lee model the propagation 
path loss is highest due to the increase in MS distance from 
BTS than the other two models and Okumura model has the 
lowest path loss. From the analyses it is seen that overall 
Okumura model shows the better performance than that of the 
other two models. 

V. Conclusion 

In this paper, three widely known large scale 
propagation models are studied and analyzed. The analyses and 
simulation was done to find out the path loss by varying the 
BTS antenna height, MS antenna height, and the T-R 
separation. Okumura model was seen to represent low power 
loss levels in the curves. The result of this analysis will help the 
network designers to choose the proper model in the field 
applications. Further up-gradation in this result can be possible 
for the higher range of carrier frequency. 





140 






to 
to 


120 


C3 




<— 


1DD 






03 
□_ 


SO 


1= 




CD 


60 


CT3 






40 






e 




D_ 





Lee Model 
Hata Modef 

Okumura Model 



2 4 6 B li 

MS Antenna Height (m) 

Figure 11. Comparison of Propagation Path Loss due to the change in MS 
antenna height. 



References 

[1] Theodore. S.R., 2006, "Wireless Communications Principles and 
Practice", Prentice-Hall, India. 

[2] J. W. Mark, Weihua Zhuang, 2005, "Wireless Communications and 
Networking", Prentice-Hall, India. 

[3] F. D. Alotaibi and A. A. Ali, April 2008, "Tuning of lee path loss model 
based on recent RF measurements in 400 MHz conducted in Riyadh city, 
Saudi Arabia," The Arabian Journal for Science and Engineering, Vol 
33, no IB, pp. 145-152. 

[4] http:// en.wikipedia.org/wiki/Radio_Propagation_model, June, 2008. 




5 10 15 20 

MS distance from BTS (km) 

Figure 12. Comparison of Propagation Path Loss due to T-R separation in 
BTS antenna height. 



139 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 



Performance Evaluation of TCP over Mobile Ad-hoc 

Networks 

Foez ahmed 1 , Sateesh Kumar Pradhan 1 , Nayeema Islam 2 , and Sumon Kumar Debnath 3 

*Dept. of Computer Networks Engineering, College of Computer Science, King Khalid University, Kingdom of Saudi Arabia 

department of Information & Communication Engineering, Rajshahi University, Rajshahi- 6205, Bangladesh 
3 Dept. of Computer Science and Telecommunication Engineering, Noakhali Science and Technology University, Bangladesh 



Abstract — With the proliferation of mobile computing devices, 
the demand for continuous network connectivity regardless of 
physical location has spurred interest in the use of mobile ad 
hoc networks. Since Transmission Control Protocol (TCP) is 
the standard network protocol for communication in the 
internet, any wireless network with Internet service need to be 
compatible with TCP. TCP is tuned to perform well in 
traditional wired networks, where packet losses occur mostly 
because of congestion. However, TCP connections in Ad-hoc 
mobile networks are plagued by problems such as high bit 
error rates, frequent route changes, multi-path routing and 
temporary network partitions. The throughput of TCP over 
such connection is not satisfactory, because TCP misinterprets 
the packet loss or delay as congestion and invokes congestion 
control and avoidance algorithm. In this research, the 
performance of TCP in Ad-hoc mobile network with high Bit 
Error rate (BER) and mobility is studied and investigated. 
Simulation model is implemented and experiments are 
performed using the Network Simulatior-2 (NS-2). 

Keywords- Ad Hoc Network, TCP, High Bit Error Rate, 
Route re-computation. 

I. Introduction 

An ad hoc network is a collection of mobile nodes 
forming a temporary wireless network without the aid of 
any established infrastructure or centralized administration 
[1]. Since each node in the network is potentially mobile, 
the topology of an ad-hoc network can be highly dynamic. 
Ad hoc networks are very useful in emergency search- and- 
rescue operations, meetings or conventions in which persons 
wish to quickly share information, and data acquisition 
operations in inhospitable terrain. 

In ad hoc networks all nodes are mobile and can be 
connected dynamically in an arbitrary manner. All nodes of 
these networks behave as routers and take part in discovery 
and maintenance of routes to other nodes in the network. 
This ad-hoc routing protocols can be divided into two 
categories: Table-driven routing protocol and On-Demand 
routing protocols. In table driven routing protocols, 



consistent and up-to-date routing information of all nodes is 
maintained at each node. In On-Demand routing protocols, 
the routes are created when required. When a source wants 
to send data to a destination, it invokes the route discovery 
mechanisms to find the path to the destination. In recent 
years, a variety of new routing protocols targeted 
specifically at this environment have been developed. There 
are four multi-hop wireless ad hoc network routing 
protocols that cover a range of design choices: Destination- 
Sequenced Distance- Vector (DSDV) [2], Dynamic Source 
Routing (DSR) [1], On-Demand Distance Vector Routing 
(AODV) [3] and Temporally Ordered Routing Algorithm 
(TORA) [4]. While DSDV is a table-driven routing 
protocol, TORA, DSR, AODV, fall under the On-demand 
routing protocols category. 

As the standard network protocol on the Internet, the use 
of the TCP/IP over the ad-hoc network is a certainty. It is 
because that the use of TCP/IP supports a large number of 
popular applications and allows seamless integration with 
the internet. In addition, TCP/IP provides network-level 
consistency for multi-hop networks that consist of hosts 
using a variety of physical-layer media. However, TCP 
assumes a relatively reliable underlying network where most 
packet losses are due to congestion, which conflicts strongly 
with characteristics of wireless links in ad-hoc mobile 
network. In mobile ad hoc environment losses are more 
often caused by high Bit Error Rate (BER), route re- 
computation, temporary network partition, and multi-path 
routing [5]. If the losses are not due to congestion, then TCP 
unnecessarily reduces throughput leading to poor 
performance. 

In this paper, the effect of high bit error rate and route re- 
computation on the performance of TCP in mobile ad hoc 
network is analyzed. All four mentioned routing protocols 
are used throughout the experimentations. The TCP variants 
used in these simulations are TCP Tahoe [6], Reno [7], New 
Reno [8] and Sack [9]. 



140 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 



The rest of this paper is organized as follows. In section 
II the impact of multi-hop wireless networks on TCP 
performance is discussed briefly. The network topologies 
together with simulation parameters are presented in section 
III. Simulation results are also discussed in the same 
section. Finally, the recommendations for future work in 
this area and concluding remarks are provided in section IV. 

II. The Problems with TCPin Ad Hoc Networks 

This section describes the effects of High Bit Error Rate 
and Route re-computation on the performance of TCP and 
the possible reasons behind these effects. 

A. Effect of High Bit Error Rate (BER) 

Bit errors cause packets to get corrupted resulting in lost 
TCP data segments or acknowledgements. When 
acknowledgements do not arrive at the TCP sender within a 
given short amount of time (the retransmit timeout or RTO), 
the sender retransmits the segment, exponentially backs off 
its retransmit timer for the next retransmission, and decreases 
its congestion window. Repeated errors result in a small 
congestion window at the sender and causes low throughput. 

B. Effect of Route Re-computations 

When an old route is no longer available, the routing 
protocol at the sender attempts to find a new route to the 
destination. It is possible that discovering a new route may 
take long. As a result, the TCP sender times out, retransmits 
a packet, and invokes congestion control mechanisms. This 
is clearly an undesirable behavior because the TCP 
connection will be very inefficient. If we imagine a network 
in which route computations are done frequently (due to high 
node mobility), the TCP connection will never get an 
opportunity to transmit at the maximum negotiated rate (i.e., 
the congestion window will always be significantly smaller 
than the advertised window size from the receiver). 



include a set of mobile ad hoc routing protocols as well as an 
802.1 1 MAC layer and a radio propagation model. 

C. Effect of High Bit Error Rate 

To investigate the effect of high BER on the TCP 
performance two different types of topologies are 
considered: Grid topology and Chain topology. Both 
topologies are considered to be static. 

C.l. Experimental setup for Chain Topology and Result 
analysis 

A 7 -node chain or string topology is shown in fig. 1. In a 
chain topology, TCP packets travel along a chain of 
intermediate nodes toward the destination. 

0OGKIKZKZKZ) 

Figure 1 . Chain Topology 

Simulations are performed to investigate the performance 
of the TCP flow whose source and destination are placed at 
both ends of the chain topology. A single wireless channel is 
shared for transmissions, and only receivers within the 
transmission range of the sender can receive the packets. 
The two adjacent nodes are about 200m apart. In this 
simulation the data rate of the wireless channel is the two-ray 
ground model with transmission range 250m, carrier sensing 
range 550m, and interference range 550m. The network 
model consists of 4, 7 and 10 nodes respectively in a 
2200*500 meter flat, rectangular area. At first simulations 
are performed without introducing any error on the links. 
Throughputs (in Mbps) of various TCP variants which are 
TCP Tahoe, TCP Reno, TCP New Reno and Sack over 
mobile ad hoc networks are identified. For performance 
comparison all four routing protocols i.e., DSDV, AODV, 
DSR, TORA are considered. These results for 4-node, 7- 
node and 10 node networks are shown in table I. 



III. Experimental setup, simulation results and 
discussions 

A. Measurement parameter 

The performance metrics of the interest in this study is 
throughput, which is the measure of how soon an end user is 
able to receive data. It is determined as the ratio of the total 
data received by the end user and the connection time. A 
higher throughput will directly impact the user's perception 
of the quality of service. 

B. Simulation Tools 

The results in this study are based on simulations using 
the network simulator (NS2) from Lawrence Berkeley 
National Laboratory (LBNL) , with extensions from the 
MONARCH project at Carnegie Mellon [10]. The extensions 



TABLET 



Throughput Of TCP Versions for Chain topology 





4-Node Chain 
Topology 


7-Node Chain 
Topology 


10-Node Chain 
Topology 








1 


I 






1 


I 






1 


I 


Tahoe 


oo 
CN 


CO 

o 

vq 


CO 

o 

vq 


CN 

o 
vq 


oo 

'nI- 

CN 


CO 

o 

vq 


CO 

o 

^q 


CN 

o 
vq 


oo 

CN 


CO 

o 
vq 


CO 

o 

vq 


CN 

o 

vq 


Reno 


oo 

CN 


CO 

o 


CO 

o 

vq 


(N 

O 

*q 


OO 
'nI- 

CN 


CO 

o 

vq 


CO 

o 

^q 


CN 

o 


oo 

CN 


CO 

o 

vq 


CO 

o 

^q 


CN 

O 

vq 


New 
Reno 


CO 
CO 


OO 

vq 


in 

oo 

vq 


m 

oo 

vq 


OO 
CO 


oo 

vq 


oo 

vq 


in 

oo 

vq 


CO 
CO 


oo 

vq 


oo 

vq 


in 

oo 
vq 


Sack 


oo 
(N 


CO 

o 


CO 

o 


CN 

O 


oo 
CN 


CO 

o 


CO 

o 


CN 

O 


oo 

CN 


CO 

o 


CO 

o 


CN 

O 



141 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 



From the table I, it is observed that maximum throughput 
is achieved for New Reno protocol for all four ad hoc routing 
protocols. However if routing protocols are compared, 
AODV shows the best performance. From these results it can 
be concluded that New Reno- AODV combination shows the 
best performance for all the sizes of chain topologies. Hence 
this combination is used to identify the effect of high BER. 
Simulations are performed to investigate the performance of 
TCP for a variety of wireless packet error rates ranging from 
0.000 to 0.017. Each simulation is performed for 200 
seconds. 

The observed throughputs for all the networks are plotted 
in fig. 2. From figure it can be observed that TCP throughput 
highly degrades with increasing packet error rate that proves 
its poor performance. Since TCP's congestion control 
algorithm is responsible for TCP throughput it leads to poor 
performance in High error rate environment. This figure also 
compares the impact of varying network sizes in presence of 
Bit Error. It can be observed that as the network size 
increased, throughput tends to decrease earlier with 
increasing PER. For example when the packet error rate is 
0.015 the throughput for 4 node chain topology is 0.0296559 
Mbps, the throughput of 7 node chain topology is 0.001024 
Mbps and finally the throughput of 10 node chain topology 
is 0.00Mbps. Hence, not only the increasing bit error rates 
effects the network throughput but also the increasing 
network size have significant effect on TCP performance. 



-•-4 nodes (AODV) 
-■- 7 nodes (AODV) 
-±- 10 nodes (AODV) 




0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 

Packet Error Rate (PER) 



Figure 2. Throughput of TCP New Reno over ad hoc network for PER 
ranging from 0.0 to 0.016 for various sizes of Chain Topologies. 




Figure 3. 16-node Grid Topology 

The performance of the TCP flow is investigated whose 
source and destination is placed at two opposite corners of 
the grid (Bottom-left and Upper-Top of the grid). For the 16- 
node network the distance between two adjacent nodes is 
100 meter. For other two networks the nodes are 50 meters 
apart. The data rate of the wireless channel is the two-ray 
ground model with transmission range 250m, carrier sensing 
range 550m, and interference range 550m. The size of all 
networks is 500m x 500m. . Again simulations are 
performed without introducing any error on the links. 
Throughputs (in Mbps) of all TCP variants all four 
mentioned routing protocols are calculated. These results are 
shown in table II. 



TABLE II. 



Throughput of TCP versions for Grid topology 





16-Node Grid 
Topology 


25-Node Grid 
Topology 


49-Node Grid 
Topology 




^ 

3 
Q 






I 


^ 

3 
Q 




c* 

&0 

Q 


I 








I 


Taho 

e 


m 

On 
On 


On 
(N 

O 

vq 


(N 

o 

vq 


OO 

co 

On 

in 


VO 
On 
On 
CM 


On 
CN 

O 

vq 


CN 

O 

vq 


l> 

CO 
On 

in 


o 
o 

vo 
in 


On 
CM 

O 

vq 


m 

On 
O 


On 
CM 

On 

in 


Reno 


oo 

On 


On 
(N 

O 

vq 


(N 

O 

vq 


m 

On 

in 


VO 
On 
CM 


On 
CN 

O 

vq 


<* 

CM 

O 

vq 


On 

in 


VO 
VO 

vo 
in 


On 
CM 

O 

vq 


m 

On 
O 


vo 

CO 
On 

in 


New 
Reno 


CM 

On 


On 

m 

oo 

VO 


co 
m 

oo 
vq 


On 
VO 
O 

vq 


co 

CO 
CO 


On 

m 

oo 

vq 


m 

oo 

vq 


oo 

VO 

l> 

vq 


o 

CO 

vq 


On 

m 

oo 
vq 


VO 

oo 
o 


o 
vo 
o 
vq 


Sack 


o 
oo 

On 


On 
(N 

O 

VO 


(N 

O 

VO 


oo 
co 

On 

in 


o 
o 

CO 


On 
CN 

o 

VO 


CM 

O 

VO 


o 

CO 
On 

in 


o 

vo 
vo 
in 


On 
CM 

O 

vo 


m 

On 
O 
O 


On 
CN 

On 

in 



C.2. Experimental setup for Grid Topology and Result 
analysis 

Three different sizes of Grid topologies are used for 
simulation. The numbers of nodes of these grid topologies 
are 16, 25, and 49 respectively. A 16-node grid topology is 
shown in fig. 3. 



For 16-node grid topology New Reno achieves maximum 
throughput for all the ad hoc routing protocols. AODV 
routing protocol gives the best performance when New Reno 
is chosen as the transport layer protocol. Same result is 
obtained for 25 node grid topology. In case of 49 node grid 
topology New Reno shows the best performance. However 



142 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 



in this case DSR is the dominant routing protocol since it 
helps to achieve maximum throughput. 

Next, errors are generated for all the networks. The 
packet error rate is varied from 0.000 to 0.017. The 
calculated throughputs are plotted in fig. 4. From figure 4 
can be observed that increasing error decreases the TCP 
throughput for all network sizes. For 25 -node network, 
throughput decreases rapidly than 16-node network with 
increasing PER. For 49 -node network throughput decreases 
slowly than other two. Again, the reduction of throughput of 
TCP is due to mistaking wireless error as an indication of 
congestion. 



16 nodes (AODV) 
25 nodes (AODV) 
49 nodes (DSR) 




0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 
Packet Error Rate (PER) 

Figure 3. Throughput of TCP New Reno over ad-hoc network for PER 
ranging from 0.000 to 0.016 for various sizes of grid topologies 



D. Effect of Route Re-computation on the Performance of 
TCP 

The effects of route re-computation are investigated for 
AODV and TORA routing protocols. 

D.l. Effect of Route Re-computation for AODV Routing 
Protocol 

To experiment the effect of route re-computation on the 
performance of TCP the following network topology (fig. 5) 
is used. 





node4 


o 




node5 






nodel 




node3 






nodeO 








o 


node2 



There are 6 nodes in this network. The scenario area is 
600 m x 600 m and transmission Range of each node is 250 
m. All nodes communicate with identical wireless radios, 
which have a bandwidth of 2Mbps. It is assumed that all 
links are error free. New Reno is used as the TCP standard. 
The simulation is run for a period of 200 seconds. The initial 
positions of the nodes are, nodeO: (25, 25), nodel: (156,143), 
node2: (396,143), node3: (303,242), node4: (160,331), 
node5: (399,329). 

During simulation nodeO starts sending packets at 5 
seconds to node5 through the route 

node0^nodel^>node3^node5 (figure 5). At time 10 
seconds node3 starts moving towards the coordinate position 
(599,599) at the speed of 30m/s. At the same time node2 also 
start moving towards the coordination position (303,242) at 
the same speed as the node3. Finally a new network topology 
is formed as shown in fig. 6. 




Figure 4. Network Topology before Mobility (AODV Routing Protocol) 



Figure 5 . Changed Network Topology after Mobility (AODV Routing 
Protocol) 



Due to mobility node3 moves out of the transmission 
range and within few seconds the transmission link breaks. 
Now the old route is no longer available and hence the 
routing protocol at the sender tries to find a new route to the 
destination. From the simulation scenario it can be observed 
that if all other nodes are fixed it is not possible for the 
sender to find out the new route due the unavailability of the 
wireless transmission range. To establish a connection a 
node must move towards the transmission range, which we 
generate through the movement of node2. At time nearly 40 
seconds node2 reaches the transmission range. At the same 
time a new route is established and nodeO again starts 
sending packets through the new route node0^> 
nodel^node2^node5. It is seen that discovering a new 
route takes about 35 seconds. Meanwhile the TCP sender has 
timed out and has invoked congestion control mechanism, 
which results sever degradation of TCP throughput. Fig. 7 
shows the performances of this network for the entire 
simulation time. From the graph it can be observed that 
during time interval 10 to 40 seconds no packet is delivered 



143 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. I, 2010 



from the source to the destination. Hence the throughput 
remains zero for that time period. After 40 seconds the 
connection resumes due to the new route and throughput 
begins to increase. 



throughput (Tvlbps) 

I. SOQO - 

1.7000 - 

1 (MMM) - 

1.5000 - 

1 4CM1C) - 

1 JUM.H) - 

1 20<M) - 

1 Kill* I - 

1.0000 - 

0.9000 - 

0.8000 - 

(I 7IMM) - 

0.6000 - 

0.5000 - 

0.4000 - 

0.3000 - 

0.2000 - 

0.1000 - 

(I I Kill* I - 



« i i, k h k „ I 



ft 1 » i i, k B k 



■:i onou 



Figure 6. Throughput of TCP over Mobile Ad-hoc network for the entire 

Simulation time of 200 seconds in presence of Dynamic Network Topology 

(AODV Routing Protocol). 



D.2. Effect of Route Re -computation for TORA Routing 
Protocol 

To investigate the effect of route re-computation on the 
performance of TCP in case of TORA network protocol the 
same topology is used as AODV (Figure 5). The network 
parameters are also same. In this case the sender nodeO has 
established the route through nodel and node2 to send 
packets to node5. The route is shown in fig. 8 





node4 ^-^ 


node5 


nodeO 


node3 

o 


node2 



Figure 7 . Network Topology before Mobility (TORA Routing Protocol) 

At time 10 seconds node2 starts moving towards the 
coordinate position (599,599) at the speed of 30m/s. When 
this node moves out of the wireless transmission range the 



connection breaks and packets start dropping. TCP assumes 
these packet drops as the indication of congestion and 
invokes congestion control algorithm. Meanwhile nodeO tries 
to find out a new route and at near about 20 seconds it 
established a new route through the nodes 1 and node3 and 
again resumes sending packets. The new route is shown in 
fig. 9. 



node4 fl j 


\^J node2 


nodel J*~^ 
nodeO /~-\ (T y. 


node5 


( U^^ node3 





Figure ! 



Changed Network Topology after Mobility (TORA Routing 
Protocol) 



The TCP throughput for the entire simulation time is 
shown in the graph of figure 10. From the graph it is clear 
that between the time intervals 10 to 20 seconds no packet is 
successfully delivered to the destination and hence the 
throughput reduces zero. After 20 seconds when the 
connection is re-established throughput starts increasing and 
remains almost same for the remaining time. 



Tlirougliput (Zvlbps) 
1 7OO0 - 
1.6000 - 
1.5000 - 
1.4000 - 
1 J5OO0 - 
1 2OO0 - 
1.1000 - 
1.0000 - 
0.9000 - 
0.8000 - 
0.7000 - 
0.6000 - 
0.5000 - 
0.4000 - 
JWHIO _ 
0.2000 - 

i> mini - 

li OOOO - 



Figure 9 . Throughput of TCP over Mobile Ad-hoc network for the entire 

Simulation time of 200 seconds in presence of Dynamics Network 

Topology (TORA Routing Protocol) 



200.0000 
time (sec) 



144 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. I, 2010 



IV. Conclusions And Recommendations 

In this paper, TCP performance is evaluated in a multi- 
hop wireless network environment. The impact of high bit 
error rate of wireless links on the performance of TCP is 
investigated through simulation for both the chain and grid 
topologies. TCP cannot differentiate packet lost due to 
congestion and packet lost due to bit error. For both cases, 
the TCP sender assumes that networks congestion has 
occurred and invokes congestion control algorithm. From 
simulations it is observed that this algorithm results in 
significant reduction in throughput and unacceptable inter- 
reactive delays for active connections, thus severely 
degrading performance. Due to dynamic topology of ad hoc 
network the route between a sender and receiver changes 
rapidly. From simulation it is seen that the route change 
results in link disconnections, which reduces TCP 
throughput. Also sometimes it takes long to find a new route 
to resume data transfer. Meanwhile TCP sender times out 
and invokes congestion control mechanisms. From 
experiments it is also observed that due to difference of 
routing protocols, same topology uses different routes to 
transmit data. 

Further investigation can be carried out to study the 
effects of network partitions and multi-path routing on the 
TCP performance. The impact that the link-layer has on TCP 
performance, such as aggregate delay caused by local 
retransmissions over multiple wireless hops can be 
experimented through simulation. Experiments can be 
carried out to evaluate the performance in a congested ad hoc 
network. 

References 



[1] Johnson, D. and Maltz, D. (1996). "Dynamic source routing in ad hoc 
wireless networks," in Mobile Computing (ed. T. Imielinski and H. 
Korth), Kluwer Academic Publishers, Dordrecht, The Netherlands. 

[2] Perkins, C, E., and Bhagwat, P., (1994), "Highly dynamic 
Destination-Sequenced Distance-Vector routing (DSDV) for mobile 
computers," Proc. of the DIGCOMM '94 Conference on 
Communications Architectures, Protocols and Applications, pages 
234-244, August-1994. 

[3] Perkins, C, Elizabeth, M., R., and Samir, R. D., (1999), "Ad hoc on 
demand distance vector (AODV) routing," Internet draft, IETF, 
January 1999. 

[4] Park, V., and Corson, S., "Temporally Ordered Routing Algorithm 
(TORA)," Version 1 Internet Draft, draft-ietf-manet-tora-spec- 03.txt, 
work in progress, June 2001 

[5] Dong, X., Lee, D., and Wang, j., "Improving TCP Performance over 
Ad-Hoc Mobile Networks," Course Project Repost for EE228A, 
November 2000. 

[6] Jacobson, V., "Congestion avoidance and control," Proceedings of the 
ACM Symposium on Communications Architectures and Protocols 
Vol. 18, No. 4, pp. 314-329, Stanford, CA, USA, August 16-18, 1988 

[7] Jacobson, V., "Modified TCP Congestion Avoidance Algorithm,' 
end2end-interest mailing list, April 30, 1990. <URL: 
ftp://ftp.isi.edu/end2end/end2end-interest-1990.mail.> 




[8] Hoe, J. C, "Improving the Start-behavior of a Congestion Control 
Scheme for TCP," Annual conference of the Association for 
Computing Machinery's Special Interest Group on Data 
Communication (ACM SIGCOMM '96), pp.270-280, California, 
USA, August 26-27, 1996. 

[9] Mathis, M., Mahdavi, J., Floyd, S. and Romanow, A., "TCP Selective 
acknowledgement options," IEFT, RFC 2018 (Status Proposed 
Standard), 1996. <URL: www.rfc-editor.org/rfc/rfc2018.txt> 

[10] Fall., K., and Varadhan., K.,"NS-Documentation, 

http://www.isi.edu/nsnam/ns/ns- documentation.html," 

December 2006. 



Authors Profile 



Foez Ahmed joined as a 
Lecturer in the Dept. of 
Electronics and Communication 
Engineering, Northern 

University Bangladesh (NUB), 
Dhaka, Bangladesh in the 2008. 
At present he is on leave from 
Northern University and 
working as a Lecturer with the 
Dept of Networks and 
Communication Engineering, 
College of Computer Science, 
King Khalid University, Kingdom of Saudi Arabia. He did 
his under graduation and post graduation in Information and 
Communication Engineering in 2007 and 2009 respectively 
from Rajshahi University, Bangladesh. He has received 
various Awards and Scholarships for the under graduation 
and post graduation results. His research interests include 
Mobile Ad-hoc Networks and Routings, Cognitive Radio 
Networks, Cooperative Communications, Sensor Networks 
and Sparse Signal Processing in Wireless Communication. 



Sateesh Kumar Pradhan 

obtained his PhD degree in 
Computer Science from 

Berhampur University, India 
during the year 1999. He joined 
Berhampur University, as 
Assistant Professor in the 1987 
and promoted to Associate 
Professor in 1999. He was Head of 
the Department of Computer 
Science, Utkal University, India during 2001-2003. He was 
the Secretary and Vice-President Orissa IT Society from 
2003-2005 and 2005-2007 respectively and was the 
organizing Chair of the International Conference on 
Information Technology - 2005 (ICIT - 2005). At present he 
is on leave from Utkal University and working with the 




145 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. I, 2010 



Department of Computer Engineering, King Khalid 
University, KSA. His research interests include Neuron 
based Algorithms, Computer Architecture, Ad-hoc Network 
and Computer Forensic. 



Nayeema Islam was born 1976 in Bangladesh. She received 
her M.S. in Telecommunication Engineering from Asian 
Institute of Technology, Thailand in 2004. Also she 
received her M.Sc. and B.Sc. in Computer Science and 
Technology from Rajshahi University, Bangladesh in 1998 
and 1997 respectively. She is presently working as Assistant 
Professor in the department of Information and 
Communication Engineering, University of Rajshahi since 
2004. Her fields of interest include Telecommunications, 
computer networking, mobile ad-hoc networks and QoS 
routing. 



Sumon Kumar Debnath 

obtained his M.Sc degree in 
Information and 

Communication Engineering 
from Rajshahi University, 
Bangladesh during the year 
2009. He has joined Noakhali 
Science and Technology 
University, Bangladesh and 
working as a lecturer in the 
department of Computer 
Science and Telecommunication 
Engineering. He is also working as a Assistant Proctor in 
that university. His research interests include Mobile Ad- 
hoc network, Sensor Network, MIMO, OFDM, and 
MCCDMA. 




146 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 



Vision Based Game Development Using Human 

Computer Interaction 



Ms.S.Sumathi 
Bharath University 
Chennaijndia 



Dr.S.K.Srivatsa 
St. Joseph College of Engineering 
Chennaijndia 



Dr.M.Uma Maheswari 
Bharath University 
Chennaijndia 



Abstract— A Human Computer Interface (HCI) 
System for playing games is designed here for 
more natural communication with the machines. 
The system presented here is a vision-based 
system for detection of long voluntary eye blinks 
and interpretation of blink patterns for 
communication between man and machine. This 
system replaces the mouse with the human face 
as a new way to interact with the computer. 
Facial features (nose tip and eyes) are detected 
and tracked in real-time to use their actions as 
mouse events. The coordinates and movement of 
the nose tip in the live video feed are translated 
to become the coordinates and movement of the 
mouse pointer on the application. The left/right 
eye blinks fire left/right mouse click events. The 
system works with inexpensive USB cameras and 
runs at a frame rate of 30 frames per second. 

Keywords: Human Computer Interface (HCI), 
SSR Filter, Hough transform 

i. Introduction 

One of the promising fields in artificial 
intelligence is HCI. Human-Computer Interface 
(HCI) can be described as the point of 
communication between the human and a computer. 
HCI aims to use human features to interact with the 
computer. The system tracks the computer user's 
movements with a video camera and translates them 
into the movements of the mouse pointer on the 
screen. The tip of the user's nose can be tracked and 
captured with a webcam and monitor its movements 
in order to translate it to some events that 
communicate with the computer. In our system, the 
nose tip is the pointing device, because of the 



location and shape of the nose, as it is located in the 
middle of the face it is more comfortable to use it as 
the feature that moves the mouse pointer and 
defines its coordinates, and it is located on the axis 
that the face rotates about, so it basically does not 
change its distinctive convex shape which makes it 
easier to track as the face moves. We use the human 
feature Eyes to simulate mouse clicks, so the user 
can fire their events as he blinks. While different 
devices were used in HCI (e.g. infrared cameras, 
sensors, microphones) we used a webcam that 
affords a moderate resolution and frame rate as the 
capturing device in order to make the ability of 
using the program affordable for all individuals. 

We present an algorithm that distinguishes true 
eye blinks from involuntary ones, detects and tracks 
the desired facial features precisely, and fast enough 
to be applied in real-time. In our work we are trying 
to implement this in playing video games. Getting 
the player physically involved in the game provides 
a more immersive experience and a feeling of 
taking direct part rather than just playing as an 
external beholder. Already motion sensors have 
been implemented to recognize physical activity, we 
can even use finger gestures in playing games but 
we increase human computer interaction by using 
our eyes in playing games. 

We propose an accurate algorithm that 
distinguishes true eye blinks from involuntary ones, 
detects and measures their duration, and fast enough 
to be applied in real-time to control a non - 
intrusive interface for computer users in playing 
games. This system can be used in playing games 
like first shooter game, a role playing game and an 
action game. When we compare eye-based and 
mouse-based control it is found that using an eye 



147 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



tracker can increase the immersion and leads to a 
stronger feeling of being part of the game. We are 
tracking the nose movements and eye blinks for 
playing the games, the nose movements are 
interfaced with the mouse movements and the eye 
blinks are interfaced with the mouse clicks. The 
eye-movement or eye blink controlled human- 
computer interface systems are very useful for 
persons who cannot speak or use hands to 
communicate. There are no lighting requirements or 
offline templates needed for the proper functioning 
of the system. It works with inexpensive USB 
cameras and runs at a frame rate of 30 frames per 
second. 

The automatic initialization phase is triggered 
by the analysis of the involuntary blinking of the 
current computer user, which creates an online 
template of the eye to be used for tracking. This 
phase occurs each time the current correlation score 
of the tracked eye falls below a defined threshold in 
order to allow the system to recover and regain its 
accuracy in detecting the blinks. This system can be 
utilized by users for applications that require mouse 
clicks as input for e.g. games. The main 
contribution of this paper is to provide a 
reimplementation of this system that is able to run 
in real time at 30 frames per second on readily 
available and affordable webcams in playing video 
games. 

ii. Related Work 

With the growth of attention about computer 
vision, the interest in HCI has increased 
proportionally. Different human features and 
monitoring devices were used to achieve HCI, but 
during our research we were only into works that 
involved the use of facial features and webcams. 

The current evolution of computer technologies 
has enhanced various applications in human- 
computer interface. Face and gesture recognition is 
a part of this field, which can be applied in various 
applications such as in robotic, security system, 
drivers monitor, and video coding system. 

We noticed a large diversity of the facial features 
that were selected, the way they were detected and 
tracked, and the functionality that they presented for 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 

the HCI. Researchers chose different facial features: 
eye pupils, eyebrows, nose tip, lips, eye lids' 
corners, mouth corners for each of which they 
provided an explanation to choose that particular 
one. 



Different detection techniques were applied where 
the goal was to achieve more accurate results with 
less processing time. To control the mouse pointer 
various points were tracked ranging from the middle 
distance between the eyes, the middle distance 
between the eyebrows, to the nose tip. To simulate 
mouse clicks; eye blinks, mouth opening/closing, 
and sometimes eyebrow movement were used. Each 
HCI method that we read about had some 
drawbacks, some methods used expensive 
equipments, some were not fast enough to achieve 
real-time execution, and others were not robust and 
precise enough to replace the mouse. We tried to 
profit from the experience that other researchers 
gained in the HCI field and added our own ideas to 
produce an application that is fast, robust, and 
useable. 

Eye movement events detected in EOG signals 
such as saccades, fixations and blinks have been 
used to control robots or a wearable system for 
medical care givers. Patmore et al. described a 
system that provides a pointing device for people 
with physical disabilities. All of these systems use 
basic eye movements or eye-gaze direction but they 
do not implement movement sequences which 
provide a more versatile input modality for gaming 
applications. 

III. SYSTEM OVERVIEW 

This system design can be broken down into 
three modules, (1) Facial Features tracking (2) 
Integrating Nose tip movements with the mouse 
cursor (3) Replacing the eye blinks with the mouse 
click events. 

Face Detection 

In this module, we propose a real-time face 
detection algorithm using Six-Segmented 
Rectangular (SSR) filter, distance information, and 
template matching technique. Since human face is a 
dynamic object and has a high degree of variability, 



148 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. I, 2010 



we propose the method combine both feature-based 
and image-based approach to detect the point 
between the eyes by using Six-Segmented 
Rectangular filter (SSR filter). The proposed SSR 
filter, which is the rectangle divided into 6 
segments, operates by using the concept of bright- 
dark relation around Between-the-Eyes area. 
Between-the-Eyes is selected as face representative 
in our detection because its characteristic is 
common to most people and is easily seen for a 
wide range of face orientation. 

Firstly, we scan a certain size of rectangle divided 
into six segments throughout the face image. Then 
their bright-dark relations are tested if its center can 
be a candidate of Between-the-Eyes. Next, the 
distance information obtained from stereo camera 
and template matching is applied to detect the true 
Between-the-Eyes among candidates. Between-the- 
Eyes has dark part (eyes and eyebrows) on both 
sides, and has comparably bright part on upper side 
(forehead), and lower side (nose and cheekbone). 
This characteristic is stable for any facial 
expression. 

We use an intermediate representation of image 
called integral image to calculate sums of pixel 
values in each segment of SSR filter. Firstly, SSR 
filter is scanned on the image and the average gray 
level of each segment is calculated from integral 
image. Then, the bright-dark relations between each 
segment are tested to see whether its center can be a 
candidate point for Between-the- Eyes. Next, the 
stereo camera is used to find the distance 
information and the suitable Between-the- Eyes 
template size.. Finally the true Between-the-Eyes 
can be detected. 

1) Integral Image 

The SSR filter is computed by using intermediate 
representation for image called integral image. For 
the original image i(x, y), the integral image is 
defined as, 



ii(x,y) = Y d ^i(x,y) 



x'<x y'^y 



(l) 



The integral image can be computed in one pass 
over the original image by the following pair of 
recurrences. 



s(x ,y ) = s(x , y - 1) + i(x ,y) (2) 
ii(x ,y ) = ii(x -l,y) + s(x ,y ) (3) 

Where s(x, y) is the cumulative row sum, 

s(x , -1) = 0, and ii(-l, y) = 0. 

Using the integral image, the sum of pixels within 
rectangle D (r s ) can be computed at high speed with 
four array references as shown in Fig.l. 

s r = (ii (x ,y ) + ii(x - W, y - L)) - (ii (x - W, 
y)+ ii(x,y-L)) (4) 



A 


B 


L 

V 


C 


D 


W 


M 



Figure 1. Integral Image 

2) SSR filter 

At the beginning, a rectangle is scanned throughout 
the input image. This rectangle is segmented into 
six segments as shown in Fig. 2 (a). 



EL 


E2 


E3 


B4 


B5 


B5 




00 

Figure 2. SSR Filter 



We denote the total sum of pixel value of each 
segment (Bl B6) as 1 6 b b S S. The proposed SSR 
filter is used to detect the Between-the-Eyes based 
on two characteristics of face geometry. 



149 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(1) The nose area ( n S ) is brighter than the right 
and left eye area ( er S and el S , respectively) as 
shown in Fig. 2 (b), where 

S n = Sb2 + Sb5 
S e r = Sbl + Sb4 
S e l = Sb3 + Sb6 



Then, 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 

Then right and left half of p m n is re-defined again 
separately as p r ij (i=0,...,15, j = 3, ...., 15) and p ij 
(i=0,...,15, j = 3, ...., 15), respectively, each has been 
converted to have average value of 128 and 
standard deviation of 64. 

Then the left mismatching value (Di) and the right 
mismatching value (D r ) are calculated by using the 
following equation. 



S n > S e l 



(5) 
(6) 



(2) The eye area (both eyes and eyebrows) ( e S ) is 
relatively darker than the cheekbone area (including 
nose) ( c S ) as shown in Fig. 2 (c), where 

S e = Sbl + Sb2 + Sb3 
S c = Sb4 + Sb5 + Sb6 

Then, 



o e <C o c 



(7) 



When expression (5), (6), and (7) are all satisfied, 
the center of the rectangle can be a candidate for 
Between-the-Eyes. 




Figure 3. Between-the-Eyes candidates from SSR 
filter 

In Fig. 3 (b), the Between-the-Eyes candidate area 
is displayed as the white areas and the non- 
candidate area is displayed as the black part. By 
performing labeling process on Fig. 3 (b), the result 
of using SSR filter to detect Between-the-Eyes 
candidates is shown. 

Because the SSR filter extracts not only the true 
Between-the- Eyes but also some false candidates, 
so we use the average Between-the-Eyes template 
matching technique to solve this problem. 

To evaluate the candidates, we define the Between 
the- Eyes pattern as p mn (m=0,...,31, n = 0, ...., 15) . 



Di-Y 



Dr = Z 



(Pij l -tij l ) 
W 

(Pij r -tij r f 
Vif 



(8) 



(9) 



The processing flow of Real-Time face detection 
system is shown in Fig. 4. 



SSR-filter 



Candidate point 

i 

Stereo camera to find distance informati 



* 



on 



Between the Eyes template 

i 

verage Between-the-Eyes templates matching 



Eyes detection 



Between -the - Eyes point detection 



J 



Figure 4. Processing Flow of Real-Time Face 
Detection 

After extracting the templates, we pass them to the 
support vector machine in order to classify them. 
Positive classification results mean true faces, while 



150 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



negative ones mean false faces. Since the program 
will be used by one person at a time, we need to 
pick one of the positive results as the final detected 
face. To achieve that, we pick the highest positive 
result, but before doing so, we will multiply each 
positive result by the area of the cluster that its 
template represents. 



Finding the Nose Tip 

After locating the eyes, the final step is to find the 
nose tip. From figure 5 we can see that the blue line 
defines a perfect square of the pupils and outside 
corners of the mouth; the nose tip should fall inside 
this square, so this square becomes our region of 
interest in finding the nose tip. 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 

In vertical intensity profiles we add horizontally to 
each column the values of the columns that 
precedes it in the ROI the same as in the horizontal 
profile, the values accumulate faster at the nose tip 
position so the maximum value gives us the 'y' 
coordinate of the nose tip. From both, the horizontal 
and vertical profiles we were able to locate the nose 
tip position. 




Figure 5 The square that forms the ROI 

The nose tip has a convex shape so it collects more 
light than other features in the ROI because it is 
closer to the light source. In horizontal intensity 
profiles we add vertically to each line the values of 
the lines that precedes it in the ROI, so since that 
the nose bridge is brighter than the surrounding 
features the values should accumulate faster at the 
bridge location; in other words the maximum value 
of the horizontal profile gives us the 'x' coordinate 
of the nose tip. 





Last 
column 



Figure 6:The horizontal profile of ROI 



Figure 7: Horizontal Sums of the columns 
After locating the nose bridge we need to find the 
nose tip on that bridge. Since each NBP represents 
the brightest S2 sector on the line it belongs to, and 
that S2 sector contains the accumulated vertical sum 
of the intensities in that sector from the first line to 
the line it belongs to, we will be using this 
information to locate the nose tip. Nose trills are 
dark areas, and the portion that they add to the 
accumulated sum in the horizontal profile is smaller 
than the contribution of other areas; in other words 
each NBP will add with its S2 sector a certain 
amount to the accumulated sum in the horizontal 
profile, but the NBP at the nose trills location will 
add a smaller amount, we will notice a local minima 
at the nose trills location, by locating this local 
minima we take the NBP that corresponds to it as 
the nose trills location, and the next step is to look 
for the nose tip above the nose trills. Since the nose 
tip is brighter than other features it will donate with 
its S2 sector to the accumulated sum more than 



151 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



other NBPs, which means a local maxima in the 
first derivate, so the location of the nose tip is the 
location of the NBP that corresponds to the local 
maxima that is above the local minima in the first 
derivate. Tracking the nose tip will be achieved by 
template matching inside the ROI. 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 

transform gives several lines so we approximate 
them to a final line which is the eye brow. 




Figure 8:The final result of the face and eye 
detection process 

Hough Transform 

The Hough transform is a technique which can be 
used to isolate features of a particular shape within 
an image. In our proposal we use this to find our 
eye brows. The classical Hough transform is most 
commonly used for the detection of regular curves 
such as lines, circles, ellipses, etc. 

The Hough transform can be used to identify the 
parameter of a curve which best fits a set of given 
edge points. This edge description is commonly 
obtained from a feature detecting operator such as 
the Roberts Cross, Sobel or Canny edge detector 
and may be noisy, i.e. it may contain multiple edge 
fragments corresponding to a single whole feature. 
Furthermore, as the output of an edge detector 
defines only where features are in an image, the 
work of the Hough transform is to determine both 
what the features are (i.e. to detect the feature(s) for 
which it has a parametric (or other) description) and 
how many of them exist in the image. To find the 
eyebrow line from the set of thresholding points we 
apply the Hough transform. Sometimes Hough 



Motion Detection 

To detect motion in a certain region we subtract the 
pixels in that region from the same pixels of the 
previous frame, and at a given location (x,y); if the 
absolute value of the subtraction was larger than a 
certain threshold, we consider a motion at that pixel. 

Blink Detection 

We apply blink detection in the eye's ROI before 
finding the eye's new exact location. The blink 
detection process is run only if the eye is not 
moving, because when a person uses the mouse and 
wants to click, he moves the pointer to the desired 
location, stops, and then clicks, so basically the 
same for using the face, the user moves the pointer 
with the tip of the nose, stops, then blinks. To detect 
a blink we apply motion detection in the eye's ROI; 
if the number of motion pixels in the ROI is larger 
than a certain threshold we consider that a blink was 
detected, because if the eye is still, and we are 
detecting a motion in the eye's ROI, that means that 
the eyelid is moving which means a blink. In order 
to avoid multiple blinks detection while they are a 
single blink (because motion pixels will appear 
while the eye is closing and reopening), the user can 
set the blink's length, so all blinks which are 
detected in the period of the first detected blink are 
omitted. 

IV Conclusion 

The proposed system is the best system for the 
users to play games interactively. The automatic 
initialization phase is greatly simplified in this 
system, with no loss of accuracy in locating the 
user's eyes and choosing a suitable open eye 
template. Another improvement in this system is, it 
is compatible with inexpensive USB cameras, as 
opposed to the high- resolution cameras. The 
experiments indicate that the system performs 
equally well in extreme lighting conditions. The 
accuracy percentages in all the cases were 
approximately the same as those that were retrieved 
in normal lighting conditions. 



152 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 



Another important consideration is the placement 
and orientation of camera. The experiments showed 
that placing the camera below the user's head 
resulted in desirable functioning of the system. 
However, if the camera is placed too high above the 
user's head, in such a way that it is aiming down at 
the user at a significant angle, the blink detection is 
no longer as accurate. This is caused by the very 
small amount of variation in correlation scores as 
the user blinks, since nearly all that is visible to the 
camera is the eyelid of the user. Thus, when 
positioning the camera, it is beneficial to the 
detection accuracy to maximize the degree of 
variation between the open and closed eye images 
of the user.Higher frame rates and finer camera 
resolutions could lead to more robust eye detection 
that is less restrictive on the user, while increased 
processing power could be used to enhance the 
tracking algorithm to more accurately follow the 
user's eye and recover more gracefully when it is 
lost. 

REFERENCES 

1.M. Bartlett, G. Littlewort, B. Braathen, T. Sejnowski, 
and J. Movellan. A prototype for automatic recognition 
of Spontaneous facial actions. In S. Becker, S. Thrun, 
and K. Obermayer, editors, Advances in Neural 

Information Processing Systems, volume 15, pages 1271— 
1278, 2003. 

2. T. N. Bhaskar, F. T. Keat, S. Ranganath, and Y.V. 
Venkatesh.Blink detection and eye tracking for eye 
localization. In Proc. TENCON 2003, volume 2, pages 821- 
824, 2003. 

3. A. Haro, M. Flickner, and I. Essa. Detecting and 
tracking eyes by using their physiological properties, 
dynamics, and appearance. In IEEE International 
Conference on Computer Vision and Pattern Recognition, 
volume 1, pages 163-168,2000. 

4.Q. Ji, H. Wechsler, A. Duchowski, and M. Flickner. 
Editorial: special issue: eye detection and tracking. 
Comput.Vis.Image Underst, 98(1): 1-3, 2005. 
5.D. G. Lowe. Distinctive image features from scale-invariant 
keypoints. Int. J. of Computer Vision, 60(2):91-110, 2004. 
6. M. Valstar, M. Pantic, Z. Ambadar, and J. Cohn. 
Spontaneous vs. posed facial behavior: Automatic analysis of 
brow actions. In Int. Conf. on Multimodal Interfaces, pages 
162-170, 2006. 





AUTHORS PROFILE 

She has completed her Bachelor of 
engineering in Computer Science 
from Raja Rajeswari engineering 
College,Anna University and 

Master of Engineering from P. S.N. A 
Engineering college ,Anna 

University. She is currently doing her Ph.D in 
Bharath University.Her research interest includes 
image processing,computer vision and database 
management systems. 

Dr.S.K.Srivatsa was born at 
Bangalore on 21 st July 1945.He 
received his Bachelor of Electronics 
and Telecommunication 

Engineering degree (Honors) from 
Jadavpur University(securing first 
rank and two medals). Master degree in Electrical 
Commuication Engineering(With distinction)from 
Indian Institute of Science and Ph.D also from 
Indian Institute of Scienc,Bangalore.In July 2005,he 
retired as professor of Electronics engineering from 
Anna University.He has taught twenty-two different 
Courses at P.G level during the last 33 years.He has 
functioned as a Member of the Board of studies in 
some Educational Institutions.His name is included 
in the Computer Society of India database of 
Resource professionals.He has received about a 
dozen awards.He has produced 24 Ph.Ds.He is the 
author of well over 400 publications. 

Dr.M.Uma maheswari received her 
bachelor of science in Computer 
Science from Bharathidasan 
university in 1995, Master of 
Computer Applications in 
Computer Sience from 

Bharathidasan University 

inl998,M.Phil in Computer Science from Alagappa 
University,Karaikudi in 2005,Master of Technology 
in Computer Science from Mahatma Gandhi Kasi 
Vidyapeeth university in 2005 and Ph.D in 
Computer Science from Magadh Universty,Bodh 
Gay a in 2007. She has 10 years of teahing 
experience and guided 150 M.C.A projects,23 
M.Tech projects and 6 Ph.D research works. 




153 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) INTERNATIONAL JOURNAL OF COMPUTER SCIENCE AND INFORMATION SECURITY, VOL. 7, NO. 1, 2010 



Path Traversal Penalty in File Systems 

M.L Lali *\ F. Ahsan * 2 , A.F.M. Ishaq * #3 

* Department of Computer Science, COMSATS Institute of Information Technology 

Islamabad, Pakistan 

1 mikramullah@comsats . edu .pk 

2 fahsan@comsats . edu .pk 

# Present address 

SZABIST, Dubai International Academic City, Dubai, UAE 

3 faiz@szabist.ac.ae 



Abstract — File systems are used to manage data in the form of 
files and directories. These directories are hierarchical in nature. 
Access to the stored data is achieved by traversing through the 
path from root level to the respective directory containing the 
required file. This complex nature of data storage mechanism has 
significant effects on the performance of file systems in terms of 
accessibility. For considering new optimizations for file system 
design, it is important to study existing ones. Therefore, we 
designed a benchmark application to measure the penalty over 
path traversal in different file systems. Here, we present our 
results for the impact of directory depth in Windows FAT32, 
NTFS, Linux EXT-2 and Solaris UFS files systems. Overall, It is 
found that there is a considerable performance degradation as 
we go deeper along the directory levels in all these file systems. 

Index Terms — File System Benchmark, File Server, File Sys- 
tems, File Access Efficiency, Directory Depth. 



I. Introduction 

Computers are used for accessing and retrieving information 
in the form of data. The data is stored on storage media and 
managed by file systems which have become an indispensable 
part of modern operating systems. Consistency and efficiency 
of file systems affects the reliability and performance of most 
of the running applications [1]. Thus, a prime concern of the 
researchers is to develop efficient and reliable file systems. 
To achieve the objective of efficiency and reliability, there 
has been always a need to explore different new possibilities 
in the area. These new possibilities can only be found by 
thorough examination of the existing file systems. The promi- 
nent Benchmarks to find the effect of different parameters 
over efficiency of file systems are shown in [2]. File system 
efficiency is greatly dependent on the data layout which 
depends on the structure of the file system [3]. Furthermore, 
the paper says that in a specific data layout, file access time 
depends on the directory depth at which it is located. File 
access efficiency can be measured and compared for different 
levels of directory depth by benchmarking applications. 

Most of the existing file system benchmarks measure the 
file read/write performance without considering the effect 
of directory layout. It has been found that the performance 
of the file system operations is heavily dependent on the 
hardware architecture and corresponding parameters like bus 
speed, bus width, memory size, protocol, etc. However if 
these parameters are held constant then the directory depth, 
where a file is located, becomes a major parameter influencing 



the performance of a file system. With the explosive growth 
of data to be stored, we need better metadata management 
techniques to improve the accessibility of actual contents. 

The present day popular file systems create and maintain 
directory files in the same manner as the data files are kept. 
Such approach allows the directory files to be placed randomly 
over the disk and receive non-contiguous space, resulting in 
increased overhead while resolving a path. However, existing 
file systems are flexible enough to accommodate different 
customized layouts of file storage by different types of users. 
With the help of optimization techniques used by the operat- 
ing system, the implementation of storage structures is kept 
transparent to the user. In terms of workstations, file access 
is usually a step-by- step procedure which is very much in 
the context of usage, allowing optimizations to take place. On 
the other hand in servers, request for a file from the storage 
medium is serviced which is generally out-of-context. Such a 
request needs to traverse the whole path in stages where no 
optimization technique like caching is effective, resulting in 
slow response. 

The study presented in this paper analyzes the effect of 
directory depth over the file system efficiency. We analyzed 
the file system performance for various directory depths while 
pertaining to the file system design. We developed a bench- 
mark application that emphasize on the parameters required 
for the file accessing behaviors of the file systems. In [4], 
Tanenbaum et al. argue that most of the prominent file system 
architectures are hierarchical in nature. In our studies, we 
analyzed the overhead involved in data read operations due 
to hierarchical directory layout in most common file systems. 

A. Related work 

The File layout and file system performance is studied in 
[3]. They found that there is a sifnificant performance degra- 
dation due to fragmented files on the storage media. The file 
size distribution on UNIX is studied by A.S. Tanenbaum and 
others in [4]. The file system space utilization is presented by 
[5], [6]. The File system usage in Windows NT 4.0 is presented 
in [7]. The authors present their results about parameters 
like file life time, data distribution, file access patterns, file 
opening and closing characteristics etc. Furthermore, there 
are many different benchmarking applications for file systems 
as available in [8], [9], [10], [11]. The performance impact 



154 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) INTERNATIONAL JOURNAL OF COMPUTER SCIENCE AND INFORMATION SECURITY, VOL. 7, NO. 1, 2010 



of stripe size on the network attached storage systems is 
presented in [12]. Metadata management for large scale file 
systems is presented in [13]. The metadata indexing and 
search in petascale data storage systems is given in [14]. In 
this manuscript, we contribute by presenting our results for 
the panelty due to hierarchical nature of file systems found 
through our benchmark application. This studies shares the 
main objective of improving the file system performance by 
studying the existing file systems with other related work given 
above. 

B. Overview 

In this paper we present our findings about the effect of 
directory depth over performance of FAT-32, NTFS, UFS and 
EXT-2 file systems. In section 2, we give a short description of 
our Benchmark application, section 3 presents the environment 
and settings during data collection and in section 4, we 
describe our results with discussions. Section 5 presents the 
conclusion and future illusions. 

II. Benchmark description and data collection 

The following are the major components of our benchmark- 
ing process. 

• Workload Generator 

• Supervisor 

• Client Application 

A. Workload Generator 

The workload is created on a logical partition of the storage 
medium on the server. The Workload Generator program 
creates files and directories on the storage media, which are 
used for collection of results. Since, in describing our results, 
different levels in the directory hierarchy will be periodically 
referred, therefore to avoid any confusion we will refer the root 
level as level-0, a directory on the root as level- 1 , subdirectory 
within a directory on the root as level-2, and so on; as shown 
in figure 1. In our experimental set up, the data was generated 
to 15 levels down the hierarchy. The workload taken into 
consideration consisted of 32,000 files (32,000 is upper limit of 
number of files in FAT32) and a similar number of directories 
at the root level of the logical partition of storage media 
i.e. level-0. For levels down the hierarchy, each sub-directory 
contains one file and one directory, which further has a file and 
a directory and so on. Thus, each level, including the root level 
contains 32,000 files and a similar number of directories. This 
workload can be increased to any number of levels, depending 
upon the partition size of the hard disk and cluster size being 
used by the operating system. 

B. Supervisor 

The supervisor is used for supervision of the client com- 
ponents. It issues a few supervisory commands to clients as 
described below. 

• Clients need some initializing parameters which are sent 
by the supervisor. In case of multiple clients, supervisor 



Level-0 



Root 
Directory 





I 


Level-1 


Directory 1 






Level-2 


Sub-Directory 1 




1 


Level -3 


Sub-Directory 2 




■ 

i 


Level-n+1 


Sub-Directory 'n' 



Directory 2 



Directory 



Sub-Directory 1 



Z 



Sub-Directory 1 



S j h -Directory 2 



_ 



Sub-Directory 2 



Sub-Directory 'n' 



Sub-Directory n" 



Fig. 1 . Sample Data Layout Created via Workload Generator 



is capable of sending the same parameters with varying 
values, dynamically. 

• The set of parameters include the number of requests to 
be sent from each client, the directory level for accessing 
files and a unique seed to generate a different random 
path for the files at the same directory level. 

• When a client sends a registration request, the supervisor 
adds it in the List of Clients and sends back the above 
parameters on the basis of which client will access 
random files located on the server. 

• Supervisor multicasts a go command when the start 
button is pressed. Thus, all the clients start their corre- 
sponding data read operation from files located at the 
Server, simultaneously. 

• After completion of the requests, all clients will send their 
results to the supervisor. Once all results are collected, the 
supervisor makes a log file when the Save Results button 
is pressed. 

C. Client Application 

The clients send requests to the file server to read from 
random paths and pursue the following steps: 

• When a client is initialized, it notifies the Supervisor for 
its existence. 

• In response, the supervisor sends back set of parameters, 
on the basis of which client application sets the values of 
different parameters required for benchmarking. 

• Client waits until it receives a go command from the 
Supervisor application 

• It starts sending a specified number of requests to the file 
server on the go command. 

• The Client reads the first byte of each request from a 
unique file at specified depth. 

• After completing the requests, client sends results to the 
supervisor. 

The communication between the Supervisor and the Clients 
contains a sequence of parameters. This sequence is initiated 
by the clients request to register itself at the Supervisor. In 
response, the Supervisor generates a random number based 
on the client-ID and the time at which request is received and 
sends it to the client with other essential parameters. Thus, 



155 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) INTERNATIONAL JOURNAL OF COMPUTER SCIENCE AND INFORMATION SECURITY, VOL. 7, NO. 1, 2010 



in response to its registration request, the client gets a wait 
signal and a random number. The clients use the random 
number to generate a list of files from a configuration file 
which contains paths to all the files on the server with respect 
to the particular directory level. On completion of the read 
requests, each client sends the elapsed time for the request to 
the supervisor. In reply, the supervisor again sends a random 
number as a seed for the next requests, on the basis of which 
the client regenerates a new list of files to access and waits 
for the next go command. 



III. Experimental environment and 

CONFIGURATIONS 

The studies presented in [5], [6], show the usage patterns of 
data in different working environments. For our observations, 
we generated data according to the findings in these papers. 
The data was quite fragmented on the hard drive to minimize 
the automatic optimizations. Additionally, traditional measures 
of performance optimizations for the working of a file system, 
like caching, were tried to be reduced by disabling them to 
acquire the actual overheads at different directory depths. 

We read only first byte of each file to make the file access 
time constant for each directory level. Furthermore, it kept 
seek time for a file to be constant. Therefore, as a result the 
variable time showed the latency due to the path traversal in 
the file systems. The cluster size for the file systems was kept 
to its default level. 

The systems, we used for our experiments had the same 
hardware configurations for Windows and Linux based vol- 
umes and were connected through a 100-bit Ethernet LAN. 
The hard disk drives were of 40GB capacity, out of which 
a 20GB partition was used for data generated by the appli- 
cation as in section II-A. Ten clients of the same hardware 
configuration and operating system were set up. All clients 
initially registered to a single supervisor as in section II-B on 
the network. Each client sent 1000 requests to the server by 
randomly selecting from the configuration file, i.e. a total of 
10,000 requests were sent at each directory level on the server. 
Zero depth of directory shows that the file being requested by 
the client was physically present on the root level of logical 
partition. The File server was bombarded with a chain of read 
requests via certain client-nodes over the network. This chain 
of requests was repeated for different directory depths. We 
performed our experiments separately for all the file systems 
i.e. FAT32, NTFS and EXT. During our experiments, we used 
a dedicated network, thus there was not any other network 
traffic. All the file servers had data generated from the same 
configuration file which means that all of them had same 
number of files and directories at same directory level with 
respect to their root level directory. 

In our client settings, we had 10 clients with similar 
hardware configurations. Prior to start the requests for data 
from the file server, clients (as described in II-C) got registered 
with a monitor for reporting errors or final results otherwise. 
We copied all the file paths to the clients in a file. 



TABLE I 
Penalty in FAT32 File System Levels 



Directory Level 


Avg. Time (seconds) 


Penalty Ratio w.r.t. Level-0 





121 


1 


1 


350 


2.89 


2 


405 


3.35 


3 


442 


3.65 


4 


464 


3.83 


5 


486 


4.02 


6 


505 


4.17 


7 


524 


4.33 


8 


557 


4.6 


9 


598 


4.94 


10 


632 


5.22 


11 


658 


5.44 


12 


691 


5.71 


13 


709 


5.86 


14 


736 


6.08 


15 


757 


6.26 



A. Windows based Volumes 

We completed our experiments for the FAT32 and NTFS 
file systems in the same environment. The cluster block size 
for both the file systems in use was kept to their default values 
(which are generally used), i.e. 4 KB for NTFS and 16 KB 
for FAT 32 on a logical partition of 20 GB. The size of the 
data generated on the disk drive formatted with FAT- 3 2 file 
system is 19.6GB and the logical drive with NTFS format 
is populated with 7GB of data. To avoid any performance 
increasing mechanism, caching client requests for shared data 
parameter was disabled. 

B. Linux and Solaris based volmues 

We used a dedicated sun system as file server with UFS. 
Similarly, we had a system with 3GHz processor, 1GB mem- 
ory and 40GB disk drive as a file server with EXT- 2 file 
system. Clients accessed the file servers through "SAMBA" 
utility for file sharing. 

It should be noted that for our observations for UFS, 
we used Sun Solaris systems which had different hardware 
configurations but generated data was of similiar nature as of 
others. Therefore, we donot compare the results. 

IV. Results and discussion 

We collected results for all FAT-32, NTFS and EXT-2 file 
systems in almost similar environment. The environment for 
UFS was different as it uses Sun Solaris systems but it is not a 
matter as we did not intend the comparision of the performance 
between different file systems. Our objective was to explore 
the performance degradation along the directory depth. 

Preliminary, experiments were performed before choosing 
the final values of the three main variable parameters: number 
of clients, number of requests, and directory depth. An increas- 
ing number of clients increase the network traffic, resulting in 
more collisions. The time to retransmit the request affects the 
measurements. Same effect is observed in terms of increase in 
the number of requests. We minimized the chances of caching 
on the server by using a moderate number of requests. We set 
directory depth at fifteen levels, starting from level which 



156 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) INTERNATIONAL JOURNAL OF COMPUTER SCIENCE AND INFORMATION SECURITY, VOL. 7, NO. 1, 2010 



800 
700 



^ 600 
S 500 



p 300 

200 
100 




Fig. 2. FAT32 10 Clients and 1000 Requests/Client 



is the root directory as observed in [5], [6]. Figure 2 shows 
the results for each directory level, plotted against time taken 
for a thousand requests from each client for the FAT32 file 
system. An average of five readings for the time taken at each 
directory level from 10 clients is plotted against each directory 
level. There is an abrupt increase in file access time when 
going from root level to the first level of depth. After this, 
there is a linear increase found in the observations on FAT- 3 2 
file system. 

Table I shows that the penalty for each directory other than 
the root directory, i.e. level 0, increases significantly. For an 
operation on a file located in the first subdirectory or level 1, 
it takes about three times as long and for the level- 10 it is 
more than five times as long as the time taken at the root 
directory. The difference of penalty ratios of two adjacent 
levels is somewhat consistent; except that of level and 1. 
Study of the FAT-32 file system architecture shows that it 
implements a linked based allocation scheme and every path 
traversal starts from the root directory which is treated as a 
reference point [14]. For this purpose the root directory is 
cached at system startup. Therefore, for any file located on 
the volume the file system will start from the root directory 
and traverse the path step by step before the file is located. 
The same experiment with similar parameters was carried out 



1,100 



950 



300 



6o0 



300 



350 



1 



3 4 



3 7 8 9 10 11 12 13 14 15 
Dir Level 



Fig. 3. NTFS 10 Clients and 1000 Requests/Client 

on the NTFS -based volume. Corresponding results of NTFS 
based experiments are plotted in figure 3. The curve in the 
graph shows the effect of directory depth on the file system 



performance in NTFS. It is also evident that the increase in 
time up to fourth level in the NTFS file system is less than the 
increase at higher directory levels. Our results show that from 
level 5 and onwards the increase in penalty with respect to the 
previous level is more than 10%. However, by level 10 file 
access penalty is doubled than that of the root directory and, 
for further directory depths, it continues to increase constantly. 
This less significant increase at the initial levels is due to the 
structure of NTFS file systems. The structure of NTFS shows 
that it uses master file tables (MFT) for managing data. The 
study of the structure of the NTFS file system conducted in 
[15] reveals that up to some directory depth data is stored 
directly in MFT which decreases the access time for initial 
levels of directory depth. 

Table II shows that for the first four levels the files were 
stored in MFT, but for higher levels a hierarchical directory 
structure was chosen for storage of files. It is also evident 
from the table that the difference of penalty ratios for any two 
adjacent levels is less till level 4, but increases there after. In 

TABLE II 
Penalty in NTFS File System Levels 



Directory Level 


Avg. Time (seconds) 


Penalty Ratio w.r.t. Level-0 





362 


1 


1 


378 


1.04 


2 


397 


1.1 


3 


411 


1.14 


4 


433 


1.2 


5 


475 


1.31 


6 


532 


1.47 


7 


583 


1.61 


8 


634 


1.75 


9 


683 


1.89 


10 


726 


2.01 


11 


786 


2.17 


12 


836 


2.31 


13 


880 


2.43 


14 


928 


2.56 


15 


972 


2.69 



Figures 4, 5, we show the results observed for EXT-2 and UFS 
file systems. It is seen that graphs are more linear in case of 
EXT-2 and UFS file systems. We see that there is a steady 
increase in penalty along the increase in directory depth. The 
performance degradation is obvious. Tables III and IV display 



3000 



2500 
IK 

2 2000 
o 

| 1500 



1000 



500 



12 3 4 



5 6 7 8 9 
Dir Level 



10 11 12 13 14 15 



Fig. 4. EXT- 10 Clients- 1000 Requests/Client 



157 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) INTERNATIONAL JOURNAL OF COMPUTER SCIENCE AND INFORMATION SECURITY, VOL. 7, NO. 1, 2010 



TABLE III 
Penalty in EXT-2 File System Levels 



TABLE IV 

Penalty in UFS File System Levels 



Directory Level 


Avg. Time (seconds) 


Penalty Ratio w.r.t. Level-0 





345 


1.00 


1 


360 


1.04 


2 


522 


1.51 


3 


685 


1.98 


4 


837 


2.42 


5 


1000 


2.90 


6 


1168 


3.38 


7 


1320 


3.82 


8 


1492 


4.32 


9 


1654 


4.79 


10 


1821 


5.27 


11 


1981 


5.74 


12 


2142 


6.21 


13 


2299 


6.66 


14 


2486 


7.20 


15 


2657 


7.70 



Directory Level 


Avg. Time (seconds) 


Penalty Ratio w.r.t. Level-0 





117 


1.00 


1 


148 


1.26 


2 


185 


1.58 


3 


220 


1.87 


4 


255 


2.18 


5 


290 


2.48 


6 


325 


2.77 


7 


364 


3.11 


8 


393 


3.36 


9 


434 


3.70 


10 


468 


3.99 


11 


499 


4.26 


12 


532 


4.54 


13 


573 


4.89 


14 


609 


5.20 


15 


644 


5.50 



700 

600 

c 500 
o 

I 400 

1 300 

200 

100 



~^ 



12 3 4 5 



) 7 8 9 10 11 12 13 14 15 
Dir Level 



Fig. 5. UFS-10 Clients- 1000 Requests/Client 



the raw results for further calculations. We can calculate the 
penalty factor F from these values. 

The purpose of this study was to explore the time taken 
in accessing files located at different levels of the directory 
hierarchy. We did not investigate about the loopholes present 
in the existing file systems. Furthermore, we did compare 
different file systems in terms of using better optimizing 
techniques. The objective of the conducted experiment was 
purely to analyze relative performance of different file systems 
in terms of speed of file accessing at different levels. For 
this purpose, known optimizing techniques like caching were 
explicitly disabled during the data collection process. Thus, the 
only factor investigated was the operating expense of directory 
traversal at different levels. 

V. Conclusion and future work 

This paper gives a brief overview of the effect of directory 
depth on file system efficiency. Our results showed that 
there is a significant increase in access time with increasing 
directory depth. We noted that the performance of the file 
system was mainly affected by the directory depth where 
many clients access files from different directories, which is a 
fundamental requirement for a file server. Benchmark results 
showed that a linear but significant increase was found in 
terms of time as we accessed a file at deeper directory level. 



The results shown in table II indicate that the difference in 
percent increase of time is approximately 70 sec, amongst 
levels 5, 10 and 15 on NTFS volumes. Similarly, results in 
the FAT-32 file system, shown in table-2, indicate that an 
average difference in percent increase of file access time 
is approximately 112 sec between levels 5, 10 and 15. 
Similiarly, a considerable performance degration is observed 
in EXT-2 and UFS file systems as show in tables III, IV. 

In this manuscript, we have presented our observations for 
different file systems. The larger objective of our work is to 
develop a more efficient and flexible design of storage media. 
The idea is to create a data server based on factual data 
patterns. The benchmarking results presented here reveal that 
there is a strong need to consider new techniques for data 
management on storage media. We propose a new file system 
where metadata should be completely separated from the data 
on the disk drives. This will decrease this perfomance overhead 
due to dispersed directories over the disk drives. 

VI. Acknowledgments 

Authors would like to acknowledge support provided by the 
Higher Education Commission Islamabad, Pakistan through 
the Indigenous Ph.D. Fellowship Program for conducting this 
research. 

References 

[1] Oracle, "Linux file system performance comparison for oltp with ext2, 
ext3, raw, and ocfs on direct- attached disks using oracle 9i release 2," 
January 2004. 

[2] F. Ahsan, M. I. Lali, I. Ahmad, A. F. M. Ishaq, and S. Mohsin, "Explor- 
ing the effect of directory depth on file access for FAT and NTFS file 
systems," in ISTASC'08: Proceedings of the 8th conference on Systems 
theory and scientific computation. Stevens Point, Wisconsin, USA: 
World Scientific and Engineering Academy and Society (WSEAS), 
2008, pp. 130-135. 

[3] K. A. Smith and M. Selzter, "File layout and file system performance," 
Harvard Computer Science, USA, Technical Report TR-35-94, 1994. 

[4] A. S. Tanenbaum, J. N. Herder, and H. Bos, "File size distribution on 
unix systems: then and now," Operating Systems Review, vol. 40, no. 1, 
pp. 100-104, 2006. 



158 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) INTERNATIONAL JOURNAL OF COMPUTER SCIENCE AND INFORMATION SECURITY, VOL. 7, NO. 1, 2010 



[5] M. I. Ullah, F. Ahsan, and A. F. M. Ishaq, "Study of File System 
Space Utilization Patterns in MS-Windows Volumes," Proceedings of 
International Conference on ICT in Education and Development, pp. 
157-164, December 16- 18th, 2004. 
[6] M. I. Ullah, F. Ahsan, I. Ahmad, and A. F. M. Ishaq, "Analysis of File 
System Space Utilization Patterns in UNIX Based Volumes," Proceed- 
ings of The IEEE International Conference on Emerging Technologies, 
(ICET 2005), Septemebr 17-18, 2005. 
[7] Vogels and Werner, "File system usage in windows nt 4.0," in SOSP '99: 
Proceedings of the seventeenth ACM symposium on Operating systems 
principles. New York, NY, USA: ACM, 1999, pp. 93-109. 
[8] T. Bray, "Bonnie file system benchmark," [Online], Available: 
http://www.textuality .com / bonnie/. (Last visited 01/10/2009), Novem- 
ber 1990. 
[9] F. John, "Whos best? how good are they? how do we get that good?" 
[Online], Available: http://management.about.com/cs/benchmarking/a/- 
Benchmarking.htm (Last visited 11/10/2009), November. 

[10] J. B. A. Park, "Iostone: A synthetic file system benchmark," pp. 45-52, 
June 1990. 

[11] Iozone, "Iozone filesystem benchmark," [Online], Available: 
http://www.iozone.org/, July, 2005 bonnie/. (Last visited 07/10/2009), 
July 2005. 

[12] Y Deng and F. Wang, "Exploring the performance impact of stripe size 
on network attached storage systems," Journal of Systems Architecture, 
vol. 54, no. 8, pp. 787-796, 2008. 

[13] S. Weil, S. A. Brandt, E. L. Miller, and K. Pollack, "Intelligent metadata 
management for a petabyte-scale file system," May 2004. 

[14] A. Leung, M. Shao, T. Bisson, S. Pasupathy, and E. L. Miller, "High- 
performance metadata indexing and search in petascale data storage 
systems," Jul. 2008. 

[15] D. Mikhailov, "FAT and NTFS performance," [Online], 
Available:http://www.digit-life.com/articles/ntfs/index3.html (Last 

visited 15/09/2009), June 2007. 




Dr. Ishaq earned a doctorate in Physics from Mc- 
Master University, Hamilton, Canada in 1972. He 
switched over to Computer Systems in the late sev- 
enties. He has 43 years of professional experience in 
university teaching, research, training, academic ad- 
ministration, technical consulting and management. 
His last full time employment was with CUT, Islam- 
abad, as a Professor and Dean. He moved to Dubai, 
UAE, in 2007, where he works as a Consultant and 
is associated with SZABIST in Dubai International 
Academic City. 



Author's Profile 




Mr. M.I. Lali received his master in software 
engineering degree from COMSATS Institute of 
Information Technology (CUT), Islamabad, Pakistan 
in 2002. Later, after spending a few years in industry, 
he joined CUT for PhD program. Currently, he is 
pursuing his studies for PhD in the area of Formal- 
ism in File Systems and Software Design. He has 
worked at University fo Groningen, Netherlands for 
some time for his research. 




Mr. F. Ahsan received his Bachelors in Computer 
Science degree from FAST, National University, 
Karachi, and masters degree from SZABIST, Is- 
lamabad, Pakistan. Currently, he is pursuing his 
PhD studies at COMSATS Institute of Information 
Technology, Islamabad. His research is focused on 
distributed systems and computer networks, sup- 
ported by HEC. 



159 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



Using Statistical Moment Invariants and 
Entropy in Image Retrieval 

Ismail I. Amr*, Mohamed Amin^, Passent El-Kafrawy^, and Amr M. Sauber^ 

* College of Computers and Informatics 
Misr International University, Cairo, Egypt 
^ Faculty of science Department of Math and Computer Science 
Menoufia University, Shebin-ElKom, Egypt 



Abstract 

Although content-based image retrieval (CBIR) is 
not a new subject, it keeps attracting more and 
more attention, as the amount of images grow 
tremendously due to internet, inexpensive hardware 
and automation of image acquisition. One of the 
applications of CBIR is fetching images from a 
database. This paper presents a new method for 
automatic image retrieval using moment invariants 
and image entropy, our technique could be used 
to find semi or perfect matches based on query- 
by-example manner, experimental results 
demonstrate that the purposed technique is scalable 
and efficient. 

Keywords 

Moment invariants, content-based image 
retrieval, image entropy. 



analyze the actual contents of the image. The term 
content' in this context might refer to color, shape, 
texture, or any other information that can be derived 
from the image itself. Without the ability to examine 
image content, retrieval must rely on metadata such as 
captions or keywords, which may be laborious or 
expensive to produce. 

What is desired is a similarity matching, independent 
of translation, rotation, and scale, between a given 
template (example) and images in the database. 
Consider the situation where a user wishes to retrieve 
all images containing cars, people, etc. in a large visual 
library. Being able to form queries in terms of sketches, 
structural descriptions, color, or texture, known as 
query-by-example (QBE), offers more flexibility over 
simple alphanumeric descriptions. This paper is 
organized as follows: section 2 provides the necessary 
background for CBIR. Section 3 defines the image 
segmentation technique using the Moments and 
Entropy. Section 4 explains the pro- posed model with 
the experiments conducted. Finally, section 5 
concludes the paper and lists some further work. 



1 Introduction 

In many areas of commerce, government, academia, 
and hospitals, large collections of digital images are 
being created. Many of these collections are the 
product of digitizing existing collections of analogue 
photographs, diagrams, drawings, paintings, and prints. 
Usually, however, technologies related to archiving, 
retrieving, and editing images/video based on their 
content are still in their infancy, the only way of 
searching these collections was by keyword 
indexing, or simply by browsing. Digital image 
databases however, open the way to content-based 
searching. "Content-based" means that the search will 



1.1 Background 

Among the approaches used in developing early 
image database management systems (IDMS) are 
textual encoding [1], logical records [2], and 
relational databases [3]. The descriptions, employed 
to convey the content of the image, were mostly 
alphanumeric. Furthermore, these were obtained 
manually or by utilizing simple image processing 
operations designed for the application at hand. 
Later generations of IDMS have been designed in an 
object-oriented environment [4], where image 
interpretation routines form the backbone of the 
system. However, queries still remain limited to a set 



160 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



of predetermined features that can be handled by the 
system. The reader is referred to [5] for a survey of 
IDMS. 



Most recent systems reported in the literature for 
searching, organizing, and retrieving images based 
on their content include IBMs Query-by-Image- 
Content (QBIC) [6], MITs photo-book [7], the 
Trademark and Art Museum applications from ETL 
[8], Xenomania from the University of Michigan 
[9], and Multimedia/VOD test bed applications from 
the Columbia University [10]. IBMs QBIC is a system 
that translates visual information into numerical 
descriptors, which are then stored in a database. It 
can index and retrieve images based on average 
color, histogram color, texture, shape, and sketches. 
MITs photobook describes three content-based tools, 
utilizing principal component analysis, finite element 
modes, and the Wold transform to match 
appearances, shapes, and textures, respectively, from 
a database to a prototype at run time. Xenomania is 
a system for face image retrieval, which is based on 
QBE. Its embedded routines allow for segmentation 
and evaluation of objects based on domain 
knowledge, yielding feature values that can be 
utilized for similarity measures and image retrieval. 
The database management system of the Columbia 
University proposes integrated feature maps based 
on texture, color, and shape information for image 
indexing and query in transform domain. 
Similarity-based searching in medical image databases 
has been addressed in [11]. A variety of shape 
representation and matching techniques are currently 
available, which are invariant to size, position, 
and/or orientation. They may be grouped as: (1) 
methods based on local features such as points, 
angles, line segments, curvature, etc. [12]; (2) template 
matching methods [13]; (3) transform coefficient based 
methods, including Fourier descriptors [14] or 
generalized Hough transform [15]; (4) methods using 3 
modal and finite element analysis [16]; (5) methods 
based on geometric features, such as local and 
differential invariants [17]; and (6) methods using B- 
Splines or snakes for contour representation [18]. 
Comprehensive surveys of these methods can be found 
in [19]. 



local diffusive segmentation method [20] is used. There 
exist a wide variety of ways to achieve segmentation; 
however, it is not the subject of this paper. All 
contiguous pixels, which share a given point-based 
characteristic of the object or are surrounded by those 
that do, are considered as object pixels and those 
outside the included region, are considered as 
background. The result is a group of contiguous pixels, 
which collectively represent the object. The boundary 
pixels of the object are then extracted from the 
segmented object pixels by a simple iterative trace, 
around the outside of the object that continues until the 
starting point is reached. This trace produces a second 
group of pixels collectively representing the objects 
exterior contour [20]. 

B. Entropy 
Entropy is a scalar value representing a statistical 
measure of randomness that can be used to characterize 
the texture of the input image. Entropy is defined as 

S=-f>>g 2 (/>) 



The value of entropy is also an invariant that is neither 
affected by rotation nor scaling. 

C. Moments 
Region moment representations interpret a normalized 

gray level image function as a probability density of a 

2D random variable. Properties of this random variable 

can be described using statistical characteristics - 

moments. A moment of order (p+q) is dependent on 

scaling, translation, rotation, and even on gray level 

transformations and is given by 



m 



pq 



= | jx p y q g(x,y)dxdy 



The central moment 



Mpq= J \(* -x c )"(y -y c ) q g(x,y)dxdy 

-oo -oo 

Let f(x,y) = (x-x c ) p (y -y c ) q g(x,y) 



III. Background 

A. Image Segmentation 
The shape representation method described here 
assumes that the object has been fully segmented from 
the original image, such that all pixels representing the 
objects shape have been identified as distinct from 
those pertaining to the rest of the image. In this paper, a 



We use composite trapezoidal rule for 
evaluation the double integral 



161 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



M N 



hk 



I T = J jf(x,y)dxdy 

-GO 

{/o,0 +/o,M + 2 (/o,l +/0.2 + +/o,M-l)} 

+2EV«.o + fi,M +2(/,, 1 +/,, 2 +...+/ itM _ 1 )} 

\J N ,0 ~^J N ,M + ^V iV,l ~^J N ,2 ~^""^~J N ,M~l )) 

Where M and N image size, x € , y c are the 

co-ordinate of the region's centriod , h =k =1 
and 



y c 



m 



01 



m 



> *c =" 



m 



00 



m 



00 



2 The Proposed Technique 

Our proposed method consists of two parts: region 
selection and shape matching. In the first part, the 
image is partitioned into disjoint, connected regions. 
The second part, the shape of each potential object is 
tested to determine whether it matches one from a set 
of given template. To this effect, we use image 
Moments and Entropy. Although many techniques 
[21]-[23] use moment invariants to overcome rotation 
and scaling, we extend this technique by using both 
moments and entropy to support two level filtering 
which is more appropriate in an image database case. 
In this paper we focus on part two and part one may be 
implemented as shown in section III-A. 



The normalized un- scaled central moments 



M P 



* Pq= '(Moo) A 
Where X = T ^- + \ 



A less general form of invariance is given by 
seven rotation, translation, and scale invariant 
moment characteristics [31]. 

A = ^20 + $02 

A - (^20 ~~ 3$i 2 ) + (3i9 21 - 19 03 ) 

^4 = (^30 _l ^12) + (^21 _l H)3) 



<*5 = 



'30 "12 > 

\2 



21 "03 > 

2" 



(i9 30 -3i9 12 )(i9 30 +i9 12 ) (t9 30 +i9 12 ) -3(i9 21 + l 9 03 ) 
+(3t9 21 -t9 03 )(t9 21 +t9 03 ) 3(t9 30 -t9 21 ) -(^ 21 + l 9 03 ) 



a = 



(^20 "^02) 



(i9 30 +i9 12 ) -(# 21 +i9 03 ) 



h = 



_+4 l 9 11 ( l 9 30 + l 9 12 )( l 9 21+l 9 03 ) 
(3x9 21 -i9 03 )(i9 30 +i9 12 ) (i9 30 +i9 12 ) -3(i9 21 +i9 03 ) 

-(i9 30 -3i9 12 )(i9 21 +i9 03 ) 3(i9 30 +i9 21 ) -(i9 21 +i9 03 ) 



A. Implementation 

Images, as figure 1, are stored in the database after its 
subdivided into distinct sub-images as stated 
previously. For each sub-image stored we compute the 
entropy and moments, that is indexed accordingly. 
Each time a template image is requested to find its 
match in the database, the following steps are 
implemented. First we compute the entropy and 
moments for the image that is being searched for. 
Second, the computed entropy is used to filter the 
images in the database. At last, the computed moments 
are used to determine and fetch the most matching 
images in the filtered subset. 

B. Experimental Results 

The experiments run on an image database that 
contains one hundred images. Figure 1 shows all the 
images, where each of these sub-images are stored 
independently in the database. We picked some images 
in the database scaled, rotated, or scaled and rotated 
and used as testing templates. Figure 2 and 3 show 
multiple experimental results that demonstrate the 
performance of our technique, each experiment 
contains a template (example) and a result set 
containing the matches. All the results are promising 
except when the template is scaled larger than double 
the size of the original image (i.e. image in the 
database). 



162 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



••00 






Figure 1 Images in database 
Experiments 



Template 


Result 


□ 


□ 




s 




Conclusions 

This paper presents a method for automatic CBIR 
based on query-by-example by region-based shape 
matching. The proposed method consists of two parts: 
region selection and shape matching. Each image is 
segmented to a set of sub images by using a local 
diffusive segmentation method. Consequently, the 
image entropy is computed and used to narrow the 
search space. Later, the moment invariants of the image 
are matched, which are independent to translation, 
scale, rotation and contrast, to every sub image and to 
the template. Retrieval is performed by returning those 
sub images whose invariant moments are most similar 
to the ones of a given query image. 
In the future we would like to find more image 
characteristics that can be used to query images in the 
database. Our focus now is to explore other 
characteristics that can be used for highly scaled 
images. Moreover, we will research colored images 
and develop new techniques to handle color. 
Acknowledgment 

The author would like to thank the valuable comments 
and encouragement received from colleagues in 
Menoufia University. 



REFERENCES 

[1] S. K. Chang, "Pictorial information systems," IEEE Computer, 

vol. 14, no. 11, 1981. 

[2] W. I. Grosky, "Toward a data model for integrated pictorial 

databases," Computer Vision Graphics Image Process, vol. 25, no. 

3, pp. 371-382, 1984. 

[3] N. S. Chang and K. S. Fu, "Picture query languages for pictorial 

database systems," IEEE Computer, vol. 14, no. 11, pp. 23-33, 

1981. 

[4] A. C. A. Pizano, A. Klinger, "Specification of spatial integrity 

constraints in pictorial databases," IEEE Computer, vol. 22, no. 12, 

pp. 59-70, 1989. 

[5] W. I. Grosky and R. Mehrotra, "Image database management," 

Advances in Computers, vol. 35, no. 5, pp. 237-291, 1992. 

[6] W. N. J. A. Q. H. B. D. M. G. J. H. D. L. D. P. D. S. P. Y. M. 

Flickner, H. Sawhney, "Query by image and video content: The 

qbic system," Computer, vol. 28, no. 9, pp. 23-32, 1995. 



163 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



[7] S. S. A. Pentland, R. W. Picard, "Photobook: Content-based 

manipulation of image databases," in Proc. SPIE Storage and 

Retrieval for Image and Video Databases II, W. N. R. C. Jain, Ed., 

vol. 2, no. 185, SPIE. Bellingham, Wash., 1994, pp. 34-47. 

[8] H. S. T. Kato, T. Kurita, "Intelligent visual interactionwith 

image database systems-toward multimedia personal interface," 

Journal of Information Processing, vol. 14, no. 2, pp. 134-143, 

1991. 

[9] R. J. J. R. Bach, S. Paul, "A visual information management 

system for the interactive retrieval of faces," IEEE Trans. 

Knowledge Data Engineering, vol. 5, pp. 619-628, 1993. 

[10] J. R. Smith and S. Chang, "Visualseek: A fully automated 

content-based image query system," in ACM Multimedia 

Conference, Boston, MA, 1996. 

[11] C. F. E. G. M. Petrakis, "Similarity searching in medical image 

databases," University of Maryland, Tech. Rep. UMIACS-TR-94- 

134, 1994. 

[12] F. Stein and G. Medioni, "Structural indexing: Efficient 2-d 

object recognition," IEEE Trans. Pattern Anal. Mach. Intell., vol. 

14, pp. 1198-1204,1992. 

[13] W. J. R. D. P. Huttenlocher, G. A. Klanderman, "Comparing 

images sing the hausdorff distance," IEEE Trans. Pattern Anal. 

Mach. Intell., vol. 15, pp. 850-863, 1993. 

[14] P. Kumar and E. Foufoula-Georgiou, "Fourier domain shape 

analysis methods: A brief review and an illustrative application to 

rainfall area evolution," Water Resources Res., vol. 26, pp. 2219- 

2227, 1990. 

[15] S. C. Jeng and W. H. Tsai, "Scale and orientation invariant 

generalized hough transform — a new approach," Pattern Recognit., 

vol. 24, no. 11, pp. 1037-1051, 1991. 

[16] S. Sclaroff and A. P. Pentland, "Modal matching for 

correspondence and recognition," IEEE Trans. Pattern Anal. Mach. 

Intell., vol. 17, pp. 545-561, 1995. 

[17] E. Rivlin and I. Weiss, "Local invariants for recognition," 

IEEE Trans. Pattern Anal. Mach. Intell., vol. 17, pp. 226-238, 

1995. 

[18] Z. Y. F. S. Cohen, Z. Huang, "Invariant matching and 

identification of curves using b-splines curve representation," IEEE 

Trans.Image Process., vol. 4, pp. 1-10, 1995. 

[19] W. E. L. Grimson, Object recognition by computer: The role of 

geometric constraints. Cambridge, MA: MIT Press, 1990. 

[20] T. Bernier, "An adaptable recognition system for biological 

and other irregular objects," Ph.D. dissertation, McGill University, 

2001. 

[21] J. Flusser, "On the independence of rotation moment 

invariants," Pattern Recognition, vol. 33, no. 9, pp. 1405-1410, 

2000. 

[22] S. M. S. Rodtook, "Numerical experiments on the accuracy of 

rotation moments invariants," Image and Vision Computing, vol. 

23, no. 6, pp. 577-586, June 2005. 

[23] H. L. Dong Xu, "Geometric moment invariants," Pattern 

Recognition, vol. 41, no. 1, pp. 240-249, January 2008. 



164 http://sites.google.com/site/ijcsis/ 

ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



Genetic Algorithm Based Optimization of 
Clustering in Ad-Hoc Networks 



Bhaskar Nandi 



#2 



Subhabrata Barman 



Haldia Institute of Technology, W.B, India. 

#2 



#3 



Soumen Paul 



Abstract: 

In this paper, we have to concentrate on implementation of 
Weighted Clustering Algorithm with the help of Genetic 
Algorithm (GA).Here we have developed new algorithm for the 
implementation of GA-based approach with the help of 
Weighted Clustering Algorithm (WCA) [4]. Cluster-Head 
chosen is a important thing for clustering in ad-hoc networks. 
So, we have shown the optimization technique for the 
minimization of Cluster-Heads(CH) based on some parameter 
such as degree-difference , Battery power (P v ), degree of 
mobility, and sum of the distances of a node in ad-hoc networks. 
Cluster-Heads selection of ad-hoc networks is an important 
thing for clustering. Here, we have discussed the performance 
comparison between deterministic approach and GA-based 
approach. In this performance comparison, we have seen that 
GA does not always give the good result compare to 
deterministic WCA algorithm. Here we have seen connectivity 
(connectivity can be measured by the probability that a node is 
reachable to any other node.) is better than the deterministic 
WCA algorithm [4]. 

Keywords- Adhoc Networks, GA, Cluster Head (CH), WCA. 

I. INTRODUCTION 
A wireless ad-hoc network consists of nodes that move freely 
and communicate with each other using wireless links. Ad- 
hoc networks do not use specialized routers for path 
discovery and traffic routing. One way to support efficient 
communication between nodes is to develop wireless 
backbone architecture; this means that certain nodes must be 
selected to form the backbone. Over time, the backbone must 
change to reflect the changes in the network topology as 
nodes move around. The algorithm that selects the members 
of the backbone should naturally be fast, but also should 
require as little communication between nodes as possible, 
since mobile nodes are often powered by batteries. One way 
to solve this problem is to group the nodes into clusters, 
where one node in each cluster functions as cluster head, 
responsible for routing. A clusterhead does the resource 
allocation to all the nodes belonging to its cluster. Due to the 
dynamic nature of the mobile nodes, their association and 
dissociation to and from clusters perturb the stability of the 
network and thus reconfiguration of cluster heads is 
unavoidable. Thus, it is desirable to have a minimum number 
of clusterheads that can serve the network nodes scattered 
evenly in the area. An optimal selection of the clusterheads is 
an NP-hard problem . Therefore, various heuristics have been 
designed for this problem . we apply genetic algorithms (GA) 
as an optimization technique to improve the performance of 



165 



clusterhead election procedure. In particular, GAs are defined 
as search algorithms that use the mechanics of natural 
selection and genetics such as reproduction, gene crossover, 
mutation as their problem- solving method. The goal is to he 
able to find out a better solution in the form of new 
generations that have received advantages and survival- 
enhancing traits from the previous Generations. We have to 
target artificial-life simulation is created where survival of the 
fittest logic is applied for the string structures that are the 
living organism equivalent in real world. Even though the 
representation is structured, there is a randomization in data 
exchange to simulate the evaluation of real life forms. As 
each generation brings up a new set of strings by different 
combination of bits of pieces of the previous generation, the 
results are not guaranteed to come up with a generation that 
has a better fitness value hut by performing different genetic 
operations, the probability of achieving the desired results is 
increased. 

II. CLUSTERING IN ADHOC NETWORKS 

The weight-based distributed clustering algorithm that takes 
into consideration that the number of nodes that a cluster head 
can handle the ideal degree, transmission power, mobility and 
battery power of a mobile node. We try to keep the number of 
nodes in a cluster around a pre-defined threshold to facilitate 
the optimal operation of the medium access control (MAC) 
protocol. Our cluster head election procedure is n periodic as 
in earlier research, but adapts based on the dynamism of 
threshold value of nodes. This on-demand execution of WCA 
aims to maintain the stability of the network, thus lowering the 
computation and communication cost associated with it. 
A cluster head may not be able handle a large number of 
nodes due to resource limitations even if these nodes are its 
neighbors and lie well within its transmission range. Thus, the 
load handling capacity of the cluster head puts an upper bound 
on the node-degree. In other words, simply covering the area 
with the minimum number of cluster heads will put more 
burden on the cluster heads. At the same time, more cluster 
heads will lead to a computationally expensive system. This 
may result in good throughput, but the data packets have to go 
through multiple hops resulting in high latency. In summary, 
choosing an optimal number of cluster heads which will yield 
high throughput but incur as low latency as possible, is still an 
important problem. As the search for better heuristics for this 
problem continues, we propose the use of a combined weight 
metric, that takes into account several system parameters like 

http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



the ideal node-degree, transmission power, mobility and the 
battery power of the nodes. We could have a fully distributed 
system where all the nodes share the same responsibility and 
act as cluster heads. However, more cluster heads result in 
extra number of hops for a packet when it gets routed from the 
source to the destination, since the packet has to go via larger 
number of cluster heads. Thus this solution leads to higher 
latency, more power consumption and more information 
processing per node. On the other hand, to maximize the 
resource utilization, we can choose to have the minimum 
number of cluster heads to cover the whole geographical area 
over which the nodes are distributed. The whole area can be 
split up into zones, the size of which can be determined by the 
transmission range of the nodes. This can put a lower bound 
on the number of cluster heads required. Ideally, to reach this 
lower bound, a uniform distribution of the nodes is necessary 
over the entire area. Also, the total number of nodes per unit 
area should be restricted so that the cluster head in a zone can 
handle all the nodes therein. However, the zone based 
clustering is not a viable solution due to the following reasons. 
The cluster heads would typically be centrally located in the 
zone, and if they move, new cluster heads have to be selected. 
It might so happen that none of the other nodes in that zone 
are centrally located. Therefore, to find a new node which can 
act as a cluster head with the other nodes within its 
transmission range might be difficult. Another problem arises 
due to non-uniform distribution of the nodes over the whole 
area. If a certain zone becomes densely populated then the 
cluster head might not be able to handle all the traffic 
generated by the nodes because there is an inherent limitation 
on the number of nodes a cluster head can handle. We propose 
to select the minimum number of cluster heads which can 
support all the nodes in the system satisfying the above 
constraints. 



Step2: Compute the degree- difference 

for every node v . Here 5 is ideal nodWrmmlper^f a cluster 

except the cluster head. 

Step3: For every node, compute the sum of the distances, D v 
,with all its neighbors, as 

D v = IWM} 

v eN(v) 

Step 4. Compute the running average of the speed for every 
node till current timer . This gives a measure of mobility and 
is denoted by Mv , as 

Mv = jifc-^-Wrr7/-i) 2 
t=\ 

Where [Xt>Yt ) anc ^ \Xt-\>Yt-\) are the coordinates of 
the node v at time t and t- 1 respectively. 

Step 5. Compute the cumulative time, P v during which a node 
v acts as a cluster head. P v implies how much battery power 
has been consumed which is assumed more for a cluster head 
than an ordinary node. 

Step 6. Calculate the combined weight Wv for each node v, 

W v = WjAv + w 2 Dv + w 3 Mv + w 4 Pv 

wl, w2, w3, w4 are the weighing factors for the 
corresponding system parameters and 
wl+ w2+ w3+ w4=l. 

Step 7. Choose that node with the smallest Wv as the cluster 
head. All the neighbors of the chosen cluster head are no 
longer allowed to participate in the election procedure. 



III. CLUSTER HEAD ELECTION PROCEDURE 

The network formed by the nodes and the links can be 
represented by an undirected graph G=(V,E) where V 
represents the set of nodes vi and E represents the set of links 
e t . Dominant set S is subset of V(G).such that 

Union of N(V)=V(G) 
Here N(V) is the neighborhood of node v , defined as 



d v = \ N ( v )\= ^{dist(v, v )<tx rc 



where tx range is the transmission range of v . 

Clustering Algorithm use a combined weight metric to search 

dominant set, the combined weight is composed by cluster 

head degree, battery power, mobility, distance. The Cluster 

head election procedure consists of eight steps as described 

below: 

Step 1. Find the neighbors of each node v which defines its 
degree dv as 



d v = \ N M\= T,{dist(v, v )<tx rc 



Step 8. Repeat steps 2 — 7 for the remaining nodes not yet 
selected as a cluster head or assigned to a cluster. 

IV. PROPOSED WORK 

Factors that influence the implementing the GA 
A brief discussion of four factors is given below: 

1. degree-difference: A v = \d v ~ S | for every node v . Here 
5 is ideal node number of a cluster except the cluster head. 

2. Battery power (P v ): Obviously, the higher the battery 
power, the higher the probability that the node will become 
CH. 

3. Degree of mobility: The mobility of the node has great 
impact on the network lifetime. The topology of the network 
will be change very frequently due to the high mobility of 
nodes, which leads to reselection of CHs rapidly. 

4. sum of the distances, D v with all its neighbors, as 

D v = Z W (v,v')} 

V E N(v) 



166 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



Optimization Approach For Cluster Head Selection Using 
GA: 

Algorithm: 

Alg. Clustering_GA(int chromosome [][] ) 

{ 

Take dataset(chromosome matrix) according to the node's 
neighbourhood at time t; 
while(not end of all chromosome in chromosome matrix) 

{ 

Take the first row(chromosome) from chromosome 
matrix; 

Generate the Gene matrix using the parameter Av, D v , M v 
P v from the first chromosome row; 
while(convergence criteria is not met ) 

{ 

Calculate the Wv , value for each Gene (For i=l to 4) 
{ W vi =W!Av +w 2 D v +w 3 M v +w 4 P v 
W v , = W v + W vi 
If(i==4) 

{ j=i; 

bO]=W v 

j++; 
} 
} 

Maximum and Minimum value is taken from b array; 

Minimum value of b array position row is replaced 

Maximum value of b array position row; 

Getting a new Gene matrix ; 

Take two parent from Gene matrix; 

Mod_Gene[][]=Crossover(Gene); 

Mutation(Mod_Gene[] []); 
}/End For/ 
}/End While/ 
One of the CH is choosen from the chromosome; 
Take another chromosome; 
}/End main while/ 

A set of CH will be choosen among the data set; 
The duplicate node in the set will be deleted to get the 
desired result; 
} /End of alg./ 

III. METHODOLOGY 

Our goal is to search best nodes among hundreds of nodes, 

so that they can act as CHs. 

Conventional search methods are not robust, while the GA is 

a search procedure that uses random choice as a tool to guide 

a highly exploitative search through a coding of a parameter 

space. According to Goldberg the GA has 4 major 

characteristics: 

1. GAs with a coding of the parameter set, not the parameters 
themselves. 

2. GAs search from a population of points, not a single point. 

3. GAs use payoff (objective function) information, not 
derivatives or other auxiliary knowledge. 

4. GAs use probabilistic transition rules, not deterministic 
rules. 



In many optimization methods, we move carefully from a 
single point in the decision space to the next using 
some transition rule to determine the next point. This point- 
to-point method is dangerous because it is a perfect 
prescription for locating false peaks in multi modal (many 
peaked) search spaces. By contrast, GA works from a rich 
database of points simultaneously (a population of strings), 
climbing many peaks in parallel; thus, the probability of 
finding a false peak is reduced. A GA starts with a 
population of strings and thereafter generates successive 
populations of strings. A simple GA consists of three 
operators: 

1 . Reproduction 

2. Crossover 

3. Mutation 

The chromosome of the GA contains all the building blocks 
to a solution of the problem at hand in a form(fig-l) that is 
suitable for the genetic operators and the fitness function. 
Each individual node is represented by a 4 number called 
"gene'. These four parameter which define the feature of the 
node and are represented as follows: 

Node ID — ► XI X2 X3 X4 

XI: degree-difference 
X2: Battery power (Pv), 
X3: its degree of mobility, and 
X4: sum of the distances 

Let's take an example. To start off, select an initial 
chromosome of total population are neighbours of particular 
node ID . Here, we select a population of size equal to the no 
of nodes . Then we have to operate on each chromosome 
using the 4 parameter for each neighbor nodes of particular 
node ID. Corresponding node ID has a cluster haead that 
sould be determined by some fitness value. This value can be 
evaluated from a fitness function, 

fix) =fix\; x2; x3; x4)= Wi*v +W 2 *P V +W 3 *M V +W 4 *D V . 

case of Ad-hoc the fitness function depends upon the four 
factors, discussed in above. And minimum of f(x) should be 
selected as cluster head. A generation of the GA begins with 
reproduction. We select the mating pool of the next 
generation by spinning the weighted roulette wheel four 
times. From this, the best string get more copies, the average 
stay even, and the worst die off. Above procedure should be 
applied for each of the chromosome. 



Chromo- 
some 



Wv Node Id Nodes neigliioiir list 



-* 6 



-* 8.. 



population 



Size 



W^ value All Node Id 

ForallNodeld 



Fignre{4.1)( Array Representation of Nodes) 



IV. GRAPHICAL ANALYSIS 



167 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



Here, we have shown the comparison between deterministic 
approach and GA-based approach of weighted clustering 
algorithm. And we see that sometime genetic algorithm based 
approach is better than the deterministic approach which is 
shown in figure( 6. 5). and sometime show both approach 
produces the same number of clusterheads as well as cluster. 
Sometime deterministic gives the lower number of cluster 
than the number of cluster in GA-based approach. In 
figure(6.5) green color curve represents the deterministic 
approach of clustering and yellow color curve represents the 
GA-based approach .How average number of cluster are 
changing with respect to the varying transmission range with 
fixed displacement equal to 5 

In figure (6.6) shows the comparison 
of deterministic and GA-based approach between average 
number of cluster and varying displacement, and we see that 
GA-based approach always provides the better result than the 
deterministic approach. 

In figure(6.7) shows the comparison of 
deterministic and GA-based approach between Connectivity 
and Transmission range .Here connectivity can be measured 
by the probability that a node is reachable to any other node. 
For a single component graph ,any node is reachable to the 
any other node and the connectivity is 1 .If the network does 
not not result in single component graph, then we can say that 
all the other node in the largest component can communicate 
with each other and the connectivity can be ratio of the 
cardinality of the largest component to the cardinality of the 
graph. From figure(6.7) we have shown the transmission 
range of the cluster head can be large enough to yield the 
connected network. If we compare the deterministic approach 
and GA-based approach ,there we have shown GA gives the 
better connectivity than the deterministic approach. A well 
connected graph can be obtained at the cost of a higher 
transmission range. If we see the graph of transmission range 
versus average number of cluster heads. There we can see the 
cluster head will be minimum by incrementing the 
transmission range .But in GA-based approach gives the 
better result than deterministic approach. So that in respect 
of connectivity ,GA-based approach gives the better result. 



Maximum Displacement=5 



25 n 

|j 20 

B 15 



2 10 



5 



-*—'-__*- — 



10 20 30 40 

Transmission Range 



50 



Green Curve = Deterministic Approach 
Yellow Curve = GA-based Approach 

Figure(6.5)(Comparison Between Deterministic and Soft Computing 
Approach with Fixed Displacement) 



Tx_Range=30 



1 7 - 5 ^^m^ 

S 7 # ♦ ♦ 




3456789 10 11 
Maximum Displacement 



Green Curve = Deterministic Approach. 
Yellow Curve = GA-based Approach 

Figure(6.6)Comparison Between Deterministic and Soft Computing 
Approach With Fixed Transmission Range) 



0.5 
0.4 



0.1 





*~*r ♦ 



10 20 30 40 50 60 70 

Trnsmission Range 



Yellow Color Curve= GA-Based Approach 
Red Curve = Deterrministic Approach 
Figure(6.7) Connectivity Vs Transmission Range 



V. CONCLUSION 

From the graphical analysis, we have done comparison 
analysis between deterministic WCA and GA-based WCA 
and there we have seen that, we can not get always optimistic 
result in genetic algorithm because genetic algorithm is a 
randomized searching technique. We have seen when 
transmission range increases then average number of clusters 
decreases (Figure(6.5)),so that connectivity of network 
should be better to compare with the deterministic WCA. 

VI. REFERENCES 

[1] D.E. Goldberg, "Genetic Algorithms in Search, Optimization, and 
Machine Learning", Addison- Wesley, 1953. 

[2] L. Davis, "Applying Adaptive Algorithms to Epistatic Domains", 
Proceedings of International Joint Conference on Artificial Intelligence, 
1985. 

[3] Jian Zhang, Bin Wang, Fei Zhang.School of Computer, WuHan 
University, WuHan 430072, China. " A Distributed Approach of WCA in 
Ad-hoc Network" 

[4] D. Turgut, B. Turgut, R. Elmasri, & T.V. Le,Optimizing clustering 
algorithm in mobile ad hoc networks using simulated annealing, Proc. IEEE 
Wireless Communication and Networking Conference . 
[5] D.E. Goldberg, "International Conference on Genetic Algorithms", 
Proceedings of the Fourth International Conference on Genetic Algorithms, 
San Diego, July 13-16, 1991. 



168 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



VII. AUTHORS' PROFILE 



Bhaskar Nandi is a lecturer with the department of 
Computer Science and Engineering, Seacom Engineering 
College,Howrah,Kolkata,West Bengal, India. He has a 
teaching experience of about two years, and 1 year of 
research experience, ,more than two years of industry 
experience. His research interest are in soft computing, Ad- 
hoc Networking, Information Security and Data Mining. He 
has publication in different national journal and conferences. 
Presently he is working Data Mining and Network Security. 



Subhabrata Barman is a Senior lecturer with the department 
of Computer Science and Engineering, Haldia Institute of 
Technology, Haldia, West Bengal, India. He has a teaching 
experience of more than 6 years and a research experience of 
more than 2 years. His research interests are in the field of 
Mobile Networking and Computing, Computational 
Intelligence, Image Processing, Speech and Signal 
Processing. He has several publications in several national 
and international conferences and journals. Currently he is 
working in the area QoS issues and Energy Management in 
Wireless Adhoc and Sensor Networks. 



Soumen Paul is an Assistant Professor with department of 
the Information Technology, Haldia Institute of Technology, 
Haldia, West Bengal, India. He has a teaching experience of 
more than 8 years, industry experience of 11 years and a 
research experience of more than 2 years. His research 
interests are in the field of Control Engineering, Soft 
Computing and Mobile Networking. He has publications in 
several national and international conferences and journals. 
His doctoral work is in the area of Deadbeat realization of 
linear, non-linear, time invariant control systems of nth order. 



1 69 http://sites.google.com/site/ijcsis/ 

ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 



Multi-Product Inventory Optimization using 
Uniform Crossover Genetic Algorithm 



S.Narmadha 

Assistant Professor 

Department of Computer Science and Engineering 

Park College of Engineering and Tekhnology 

Coimbatore - 641659, Tamilnadu, India 



Dr.V.Selladurai 

Professor and Head 
Department of Mechanical Engineering 

Coimbatore Institute of Technology 
Coimbatore - 641014, Tamilnadu, India 



G.Sathish 

Research Scholar 
Department of Computer Science and Engineering 
Anna University - Coimbatore, Tamilnadu, India 



Abstract - Inventory management is considered to be an 
important field in Supply Chain Management because the cost of 
inventories in a supply chain accounts for about 30 % of the value 
of the product. The service provided to the customer eventually 
gets enhanced once the efficient and effective management of 
inventory is carried out all through the supply chain. The precise 
estimation of optimal inventory is essential since shortage of 
inventory yields to lost sales, while excess of inventory may result 
in pointless storage costs. Thus the determination of the 
inventory to be held at various levels in a supply chain becomes 
inevitable so as to ensure minimal cost for the supply chain. The 
minimization of the total supply chain cost can only be achieved 
when optimization of the base stock level is carried out at each 
member of the supply chain. This paper deals with the problem 
of determination of base-stock levels in a ten member serial 
supply chain with multiple products produced by factories using 
Uniform Crossover Genetic Algorithms. The complexity of the 
problem increases when more distribution centers and agents 
and multiple products were involved. These considerations 
leading to very complex inventory management process has been 
resolved in this work. 

Keywords: Supply Chain Management, Inventory 
Optimization, Base Stock, Uniform Crossover, Genetic 
Algorithm (GA), Supply Chain Cost 

I. INTRODUCTION 

Supply Chain Management (SCM) is an efficient 
management of the complete end to end process, starting from 
the design of the product or service to the time when it has 
been sold, consumed and finally gotten rid of by the 
consumer. This complete process includes product design, 
procurement, planning and forecasting, production, 
distribution, fulfillment and after sales supports. A company's 
competitiveness in the global economy can be increased only 
with the aid of effective SCM. This involves complex 
strategic, tactical and operational decisions that often require 
an in-depth understanding of industry- specific issues, which 
ranges from network design to production sourcing and from 
production planning and inventory management to scheduling 
[1]. 

The inventory management problem is one of maintaining 
an adequate supply of some item to meet an expected pattern 



of demand, while striking a reasonable balance between the 
cost of holding the items in inventory and the penalty (loss of 
sales and goodwill, say) of running out. The item may be a 
commodity sold by a store; it may be spare machine parts in a 
factory; it may be railway wagons; it may be cash in the bank 
to meet the customers' demand. It is indeed surprising to find 
that a very wide variety of seemingly different problems can 
be mathematically formulated as an inventory-control 
problem. There are, of course, several different models of 
inventory systems. There are three types of expenses 
associated with inventory systems. The relative importance of 
these will depend on the specific system. They are: (i) 
administrative cost of placing an order, called reorder cost or 
set cost; (ii) cost of maintaining an inventory, called inventory 
holding cost a carrying cost, which includes storage charge, 
interest, insurance, etc., a (iii) shortage cost is a loss of profit, 
goodwill, etc., when run out of stock. All the above should be 
optimized for efficient supply chain management. 

A. Inventory Control in Supply Chain Management 

It has been stated by several people that the focus point of 
supply chain management is inventories and inventory 
control. To transfer their focus from scheming logistical costs 
to investigate supply chains [2] few food manufacturers and 
grocers formed Efficient Consumer Response in the year 
1992. The major competitive factor for companies focused on 
value creation for end consumers is the customer service. In 
general, firms hold inventory for two major reasons, to lessen 
costs and to improve customer service. The inspiration for 
each varies as firms stabilize the problem of having too much 
inventory (which can direct to high costs) versus having very 
small inventory (which can direct to lost sales) [3]. 

Supply chain management leads to cost savings, mainly in 
the course of lessening in inventory. Inventory costs have got 
reduced by about 60% from 1982, whereas transportation 
costs have fallen by 20% [4]. These cost savings have led 
many people to follow inventory-reduction strategies in the 
supply chain. To deal with inventory, firms make use of one 
of three common approaches. First of all, the majority of 
retailers make use of an inventory control approach, 
monitoring inventory levels by item. The second thing is, 
manufacturers are typically more concerned with production 
scheduling and use flow management to deal with inventories. 



170 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



Third, numerous firms (for the majority part those handling 
raw materials or in extractive industries) do not keenly deal 
with inventory [5]. 



The inventory management is influenced by the nature of 
demand, depending on whether demand is derived or 
independent. Independent demand comes up from demand for 
an end product. End products are found all through the supply 
chain. By definition, a self-governing demand is uncertain, 
meaning that extra units or safety stock must be accepted to 
guard against stock outs. While managing uncertainty, the 
objective should be to minimize the inventory levels and also 
meet customer expectation. Supply chain coordination can 
reduce the ambiguity of intermediate product demand, in that 
way reducing inventory costs [3, 6]. 

Since Ford Harris' renowned Economic Order Quantity 
(EOQ) model was first proposed in 1913, the inventory 
control has been rewarded immense awareness for a long time 
because of its significance in the cost control. To lessen the 
total expected inventory costs per unit time while satisfying 
the customer demand on time [7] is one of the major 
objectives. Inventory control for large-scale supply chains is 
well recognized [8-10] as an essential problem with several 
applications together with manufacturing systems, logistics 
systems, communication networks, and transportation systems 
[11]. It is essential to locate the apt mechanism for 
coordinating the inventory processes that are controlled by 
independent partners, in order to find out the right ordering 
quantity and inventory level amid partners in the chain. For 
example, the manufacturer make use of the periodic review 
and lot sizing policy to manage its inventory and the retailer 
employs the periodic review with target stock level to control 
its inventory and more [12]. 

B. Inventory Optimization in Supply Chain Management 

The effective management of the supply chain has become 
unavoidable these days due to high expectation in customer 
service levels [13]. The supply chain cost was immensely 
influenced by the overload or shortage of inventories. Thus 
inventory optimization has transpired into one of the most 
important topics as far as supply chain management is 
considered [14-16]. 

To exploit economies of scale and order in large lots, the 
important issues in supply chain is to optimize the inventory 
level by considering various costs in maintaining a high 
service level towards the customer. Since, the cost of capital 
tied up in inventory is more, the inventory decision in the 
supply chain should be coordinated without disturbing the 
service level. The coordination of inventory decision within 
an entity is viable, but not between the entities. So the 
integration of the entities to centralize the inventory control is 
needed. 

Inventory Optimization [10] application organizes the 
latest techniques and technologies, thereby assisting the 
improved inventory visibility, the enhancement of inventory 
control and its management across an extended supply 
network. Some of the design objectives of inventory 
optimization are to optimize inventory strategies, thereby 
enhancing customer service, reducing lead times and costs and 



(IJCSIS) International lournal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
meeting market demand [14-16]. The design and management 
of the storage policies and procedures for raw materials, work- 
in-process inventories, and typically, final products are 
illustrated by the inventory control. The costs and lead times 
can be reduced and the responsiveness to the changing 
customer demands can be significantly improved and 
subsequently inventory can be optimized by the effective 
handling of the supply chain [17]. 



There are several reasons for manufacturers' increasing 
focus on optimizing inventory by applying the latest tools and 
techniques for inventory control. Traditionally, competitive 
pressure has always driven manufacturers to seek enhanced 
capabilities to reduce inventory levels; to enhance service 
levels and supply availability; and to establish the right 
product inventory mix and level in each geography and 
channel. A key driver of the renewed focus on inventory lies 
in the recognition that traditional techniques are failing to 
reign in inventories in the wake of increased supply chain 
complexity. This complexity is characterized by increased 
uncertainty. Demand is more volatile and therefore less 
predictable. This is true not only for aggregate demand but for 
forecasting splits and volumes across channels and markets. 
Traditionally three strategies have been employed by 
manufacturers to address uncertainty; a) increase inventory 
levels to hedge against uncertainty; b) develop supply chain 
flexibility to be more responsive to uncertainty; c) improve 
forecast accuracy so that less uncertainty propagates to the 
manufacturing floor. Inventory optimization techniques and 
technologies map to the flexibility and accuracy strategies. 
[18]. 

Inventory Optimization characterizes the supply network 
uncertainty present in a variety of specific steps or links in 
manufacturing and distribution processes. Advanced 
mathematical models are then solved to identify optimal 
inventory policies, stocking locations, or quantities. 
The uncertainty addressed by 10 include: demand uncertainty, 
cycle time variability and replenishment lead time 
variability. [18] Efficient management of the supply chain, i.e. 
the reduction of the costs and lead times and vastly enhanced 
responsiveness to the changing customer demands lead to an 
optimized inventory. 

II. Related Works- Re view 

A. Review of Base-Stock based Inventory Control Models 

The inventory control problem for a single class assembly 
network which operates under a modified echelon base-stock 
policy was studied by Vecchio and Paschalidis [19]. 
An approach to find close-to-optimal echelon stock levels that 
minimize inventory costs while guaranteeing stockout 
probabilities to stay below some predefined levels was 
developed by them. They reduced the safety stock selection to 
a deterministic nonlinear optimization problem on the basis of 
the large deviations techniques. In addition, they analyzed as 
to how a supplier can interact with a buyer to reach a mutually 
beneficial mode of operations, using their inventory control 
approach. Their interaction takes the form of a supply contract 
by which explicit QoS guarantees is enforced. 
The applications in a distributed fashion with neither the 



171 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



supplier nor the buyer revealing their corresponding cost 
structures applying the joint optimization algorithm was 
proposed by the authors. 

A model of supply chain consisting of n production 
facilities in tandem and producing a single product class was 
considered by Ioannis CH. Paschalidis et al. [20]. The finished 
goods inventory maintained in front of the most downstream 
facility is used to meet the external demand while backlogging 
of unsatisfied demand was performed. The facility at stage 1 
produced if inventory has fallen below a certain level w A and 
idles otherwise on the basis of a base-stock production policy 
adopted at each stage of the supply chain. In order to 
minimize expected inventory costs at all stages subject to 
maintaining the stock out probability at stage 1 below a 
prescribed level, they necessitated the optimization of the 
hedging vector W= (wi,...,w n ). They made assumptions on 
demand and production processes that included auto 
correlated stochastic processes, which were relatively general 
modeling. They have combined analytical (Large derivations) 
and sample path based (perturbation analysis) techniques to 
solve the stochastic optimization problem. The existence of a 
natural synergy between those two approaches has been 
demonstrated. 

An attempt was made to optimize the inventory (i.e. base- 
stock) levels of a single product at different members in a 
serial supply chain with the objective of minimizing the Total 
Supply Chain Cost (TSCC), by Sudhir Ryan Daniel and 
Chandrasekharan Rajendran [21] and P. Radhakrishnan et al. 
[22]. The performance measure considered, which is a good 
representation of the system-wide total cost is the TSCC. 
In order to optimize the base -stock levels, a genetic algorithm 
(GA) has been proposed. To analyze the performance of the 
supply chain (operating with deterministic and stochastic 
replenishment lead times) for the base- stock levels that are 
generated by the proposed GA and other solution procedures 
considered in this study, different supply chain settings are 
simulated. They demonstrated that their proposed GA required 
significantly less computing effort to perform very well in 
terms of yielding solutions that are not significantly different 
from the optimal solutions (obtained through complete 
enumeration of solution space). 

A beneficial industry case applying Genetic Algorithms 
(GA) has been proposed by K.Wang and Y.Wang [23]. 
The case has made use of GAs for the optimization of the total 
cost of multiple sourcing supply chain system. The system has 
been exemplified by a multiple sourcing model with stochastic 
demand. A mathematical model has been implemented to 
portray the stochastic inventory with the many to many 
demand and transportation parameters as well as price 
uncertainty factors A genetic algorithm which has been 
approved by Lo [24] deals with the production-inventory 
problem with backlog in the real situations, with time-varied 
demand and imperfect production due to the defects in 
production disruption with exponential distribution. Besides 
optimizing the number of production cycles to generate a (R, 
Q) inventory policy, an aggregative production plan can also 
be produced to minimize the total inventory cost on the basis 
of reproduction interval searching in a given time horizon. 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 
The inventory levels across supply chain members were 
obtained with the aid of a search routine. 



B. Review of Optimization based Inventory Control Models 

Sukran Kadipasaoglu et al. [1] provided a study on the 
market characteristics and competitive priorities, 
manufacturing environment, logistics and distribution 
activities, and supply chain planning and control activities for 
polymer manufacturers. They have described polymer 
distribution network optimization, production/distribution 
planning, production scheduling, demand management, 
available-to-promise, and inventory planning activities 
pertaining to supply chain planning and control. Besides, they 
illustrated the applications existing in a commercial DSS that 
support these activities. They have as well described about 
diverse issues that continue to confront supply chain managers 
in polymer manufacturing. It encompasses forecasting for the 
huge number of product-customer combinations, 
identification of safety stock requirements, administering 
production schedule changes, business process management 
throughout DSS implementation and data mapping for 
decision support. Their research contributes to the supply 
chain literature by proffering a suitable context for 
investigating supply chain-related issues. Through discussion 
and characterization of the polymer supply chain, they 
recognized the specific issues of concern to potential 
researchers and to supply chain professionals. 

The effect of product variety on supply-chain 
performance, which is measured in terms of expected lead 
time and expected cost at the retailers, was analyzed by Ulrich 
W. Thonemann and James R. Bradley [25]. They took a 
supply chain with a single manufacturer and multiple retailers 
into account. If setup times are significant, the effect of 
product variety on cost where the cost increases proportionally 
to the square root of product variety is substantially greater 
than that suggested by the risk-pooling literature for perfectly 
flexible manufacturing processes. An illustration that 
underestimates the cost of product variety, leads companies to 
offer product variety that is greater than optimal was made as 
well. In conclusion, they illustrated that by reducing the setup 
time, the unit manufacturing time, the number of retailers, or 
the demand rate the supply-chain performance can be 
managed. The fact that complex mathematical approaches are 
often not applied in practice was recognized by the authors. 
Nevertheless, practitioners who used the simple models to 
estimate the effect of their decisions often appreciated these 
models. 

The inventory and supply chain managers are mainly 
concerned holding of the excess stock levels and hence the 
increase in the holding cost. Meanwhile, there is possibility 
for the shortage of products. For the shortage of each product 
there will be a shortage cost. Holding excess stock levels as 
well as the occurrence of shortage for products lead to the 
increase in the supply chain cost. The factory may 
manufacture any number of products, each supply chain 
member may consume a few or all the products and each 
product is manufactured using a number of raw materials 
sourced from many suppliers. All these factors pose additional 



172 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) 

challenge in extracting the exact product and the stock levels 
that influence the supply chain cost heavily. 

Many well-known algorithmic advances in optimization 
have been made, but it turns out that most have not had the 
expected impact on the decisions for designing and optimizing 
supply chain related problems. For example, some 
optimization techniques are of little use because they are not 
well suited to solve complex real logistics problems in the 
short time needed to make decisions. Also some techniques 
are highly problem-dependent and need high expertise. This 
adds difficulties in the implementations of the decision 
support systems which contradicts the tendency to fast 
implementation in a rapidly changing world. 10 techniques 
need to determine a globally optimal placement of inventory, 
considering its cost at each stage in the supply chain and all 
the service level targets and replenishment lead times that 
constraint each inventory location about the estimation of the 
exact amount of inventory at each point in the supply chain 
free of excesses and shortages although the total supply chain 
cost is minimized. Owing to the fact that shortage of inventory 
yields to lost sales, whereas excess of inventory may result in 
pointless storage costs, the precise estimation of optimal 
inventory is indispensable [26]. In other words, there is a cost 
involved in manufacturing any product in the factory as well 
as in holding any product in the distribution center and agent 
shop. More the products manufactured or held, higher will be 
the holding cost. Along with this, low lead time results in 

III. Objectives 

The supply chain cost can be minimized by maintaining 
optimal stock levels in each supply chain member. There is a 
necessity of determining the inventory to be held at different 
stages in a supply chain that will minimize the total supply 
chain cost i.e., minimizing holding and shortage cost. The 
inventory control for more number of products along with 
different levels of supply chain is a complex task. The 
approach aims to make use of the meta heuristic algorithms 
like Genetic algorithm for the prediction of the optimal stock 
levels to be maintained, so as to minimize the total supply 
chain inventory cost, comprising holding and shortage costs at 
all members of the supply chain. The genetic algorithm is 
proposed that considers all these factors that are mentioned 
hitherto such that the analysis paves the way for minimizing 
the supply chain cost by maintaining optimal stock levels in 
each supply chain member. 

A. Genetic Algorithm 

Genetic algorithm is a randomized search methodology 
having its roots in the natural selection process. Initially the 
neighborhood search operators (crossover and mutation) are 
applied to the preliminary set of solutions to acquire 
generation of new solutions. Solutions are chosen randomly 
from the existing set of solutions where the selection 
probability and the solution's objective function value are 
proportional to each other and eventually the aforesaid 
operators are applied on the chosen solutions. Genetic 
algorithms have aided in the successful implementation of 
solutions for a wide variety of combinatorial problems. 



International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
The robustness of the Genetic algorithms as search 
techniques have been theoretically and empirically proved 
[27]. The artificial individual is the basic element of a GA. An 
artificial individual consists of a chromosome and a fitness 
value, similar to a natural individual. The individual's 
likelihood for survival and mating is determined by the fitness 
function [28]. In accordance with the Darwin's principle, 
individuals superior to their competitors, are more likely to 
promote their genes to the next generations. In accordance 
with this concept, in Genetic Algorithms, a set of encoded 
parameters are mapped into a potential solution, named 
chromosome, to the optimization problem [29]. 
The population of candidate solutions is obtained through the 
process of selection, recombination, and mutation performed 
in an iterative manner. [30]. 

Chromosomes refer to the random population of encoded 
candidate solutions with which the Genetic algorithms initiate 
with. [27]. Then the set (called a population) of possible 
solutions (called chromosomes) are generated [31]. A function 
assigns a degree of fitness to each chromosome in every 
generation in order to use the best individual during the 
evolutionary process [32]. In accordance to the objective, the 
fitness function evaluates the individuals [30]. 
Each chromosome is evaluated using a fitness function and a 
fitness value is assigned. Then, three different operators- 
selection, crossover and mutation- are applied to update the 
population. A generation refers to an iteration of these three 
operators [33]. The promising areas of the search space are 
focused in the selection step. The selection process typically 
keeps solutions with high fitness values in the population and 
rejects individuals of low quality [30]. Hence, this provides a 
means for the chromosomes with better fitness to form the 
mating pool (MP) [31]. After the process of Selection, the 
Crossover is performed. 



B. Uniform Crossover 

In the crossover operation, two new children are formed 
by exchanging the genetic information between two parent 
chromosomes. Multipoint crossover defines crossover points 
as places between loci where an individual can be split. 
Uniform crossover generalizes this scheme to make every 
locus a potential crossover point. A crossover mask, the same 
length as the individual structure is created at random and the 
parity of the bits in the mask indicate which parent will supply 
the offspring with which bits. This method is identical to 
discrete recombination. 

Consider the following two individuals with 1 1 binary 
variables each: 

Individual 1 01110011010 
Individual 2 10 10 110 10 1 

For each variable the parent who contributes its variable to 
the offspring is chosen randomly with equal probability. Here, 
the offspring 1 is produced by taking the bit from parent 1 if 
the corresponding mask bit is 1 or the bit from parent 2 if the 
corresponding mask bit is 0. Offspring 2 is created using the 
inverse of the mask, usually. 

Sample 1 01100011010 

Sample 2 10011100101 



173 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



After crossover the new individuals are created: 
offspring 1 1110 1111111 
offspring 2 001 10000000 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 
products that would be supplied to the distribution centers. 
From the distribution center, the stocks will be moved to the 
corresponding agents. 



Uniform crossover has been claimed to reduce the bias 
associated with the length of the binary representation used 
and the particular coding for a given parameter set. This helps 
to overcome the bias in single-point crossover towards short 
substrings without requiring precise understanding of the 
significance of the individual bits in the individual's 
representation. How uniform crossover may be parameterized 
by applying a probability to the swapping of bits was 
demonstrated by William M. Spears et al.[34]. 

This extra parameter can be used to control the amount of 
disruption during recombination without introducing a bias 
towards the length of the representation used. 
The chromosome cloning takes place when a pair of 
chromosomes does not crossover, thus creating off springs 
that are exact copies of each parent [28]. 

The ultimate step in each generation is the mutation of 
individuals through the alteration of parts of their genes [26]. 
Mutation alters a minute portion of a chromosome and thus 
institutes variability into the population of the subsequent 
generation [27]. Mutation, a rarity in nature, denotes the 
alteration in the gene and assists us in avoiding loss of genetic 
diversity [26]. Its chief intent is to ensure that the search 
algorithm is not bound on a local optimum [28]. 

IV. Inventory Optimization Analysis using Uniform 
Crossover Genetic Algorithm 

The proposed method uses the Genetic Algorithm with 
Uniform Crossover to study the stock level that needs 
essential inventory control. This is the pre-requisite 
information that will make any kind of inventory control 
effective. In practice, the supply chain is of length n, means 
having n number of members in supply chain such as factory, 
distribution centers, suppliers, retailers and so on. 
The exemplary supply chain taken for the implementation of 
the proposed method is shown in Fig. 1. 

Factors? i 



Distribution Center 1 



Distribution Center 2 



Distribution Center 3 



Ag0iil J 



Agent 2 



AjLTcnt 3 



AgciiL 4 Agcnl 5 



Agent (> 



Fig. 1 Three Stage - 10 Member Supply Chain 



Fig. 1 exhibits that a factory is the parent of the chain and 
it is having three distribution centers Distribution Center 1, 
Distribution Center 2 and Distribution Center 3. Each 
distribution center further comprises of several agents but as 
stated in the example case, each Distribution center is having 
two agents. So, in aggregate there are six agents, Agent 1 and 
Agent 2 for Distribution Center 1, Agent 3 and Agent 4 for 
Distribution Center 2 and Agent 5 and Agent 6 for 
Distribution Center 3. The factory manufactures different 



To make the inventory control effective, the most primary 
objective is to predict where, why and how much of the 
control is required which is made through the proposed 
GAUX methodology. The proposed methodology is aimed at 
determining the specific product that needs to be concentrated 
on and the amount of stock levels of the product to be 
maintained by the different members of the supply chain. The 
methodology also analyses whether the stock level of the 
particular product needs to be in abundance, in order to avoid 
shortage of the product or needs to be held minimal in order to 
minimize the holding cost. 



Proposed 

Generic 

Alf^irilhin 





Selsr.ion af 
Valid Contents 






1 








Generation of 

K>:nu:;.m 

Individuals 








i 






I>eLe:[ 1 1 i i LZ.LJtui <if 






1 






CafuulaLifin nf 
rii.-iejisFu.-ictkiri 






|_ 




F 




- decfifcTI 




1 


t Cr^ri^Lii^ 




1.. ■'iitovin 
( h--fiflanver 


■OpuiaL: kts 




1 




Mutation 








r 











Fig. 2 Genetic Algorithm steps for the proposed 
inventory management analysis 

The methodology as shown in Fig. 2 would analyze the 
past records very effectively and thus facilitate efficient 
inventory management with the aid of Genetic Algorithm. 
The analysis is initiated by the selection of valid records. 
The validation of records is done over the records of past 
periods. The stock levels at the different supply chain 
members are held in the dataset for different products, namely 
PI, P2, P3, P4, P5, P6, P7, etc. Seven products have been 
considered for the analysis, hence the stock levels for the 
seven products at each member of the chain throughout the 
past period are considered as data set as shown in the Table 1 . 
For the valid record set selection, records having nil values are 
neglected and the records having positive or negative values 
are selected for the analysis. This can be done by means of 
clustering algorithms, extraction algorithms or by any of the 
data mining functions. Hence the extraction function results in 
data sets having either positive or negative values. The record 
set having positive values represents excess stock levels and 
the negative values represent shortage level of a particular 
product at a particular member of the supply chain. Then the 



174 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



data set is subjected to Genetic Algorithm and the various 
steps performed in the genetic algorithm are discussed below. 

A. Generation of Individuals 

The randomly generated initial chromosome is created by 
having the stock levels within the lower limit and the upper 
limit for all the contributors of the supply chain, factory and 
the distribution centers. As known, chromosome is constituted 
by genes which defines the length of the chromosomes. 
The stock level of each member of the chromosome is referred 
as gene of the chromosome. Hence for n length supply chain, 
the chromosome length is alsow . Since a 10 member supply 
chain is used for illustration, the length of the chromosome n 
is 10, i.e. 10 genes. And the chromosome representation is 
pictured in Fig. 3. Each gene of the chromosome is 
representing the amount of stock that is in excess or in 
shortage at the respective members of the supply chain. 



(IJCSIS) International lournal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 
inventory values; and log [1- (N rep / N t )] will ensure minimum 
value corresponding to the maximum probability; So, the 
fitness function is structured to retain the minimum value 
corresponding to the various chromosomes being evaluated 
iteration after iteration and this in turn ensures that the fitness 
function evolution is towards optimization. 



P3 7000 -200 -600 -500 -150 



| ?2 | 5000 j 400 | -800 500 _445 | 315 | -820 | 405 -150 100 | 

Fig. 3 Random individual generated for the genetic operation 

These kinds of chromosomes are generated for the genetic 
operation. Initially, only two chromosomes will be generated 
and from the next generation a single random chromosome 
value will be generated. The chromosomes thus generated is 
then applied to find its number of occurrences in the database 
content by using a Select count ( ) function. The function will 
give the number of occurrences/ repetitions of the particular 
amount of stock level for the ten members N rep that are going 
to be used further in the fitness function. 

B. Evaluation of Fitness function 

A specific kind of objective function that enumerates the 
optimality of a solution in a genetic algorithm in order to rank 
certain chromosome against all the other chromosomes is 
known as Fitness function. Optimal chromosomes, or at least 
chromosomes which are near optimal, are permitted to breed 
and merge their datasets through one of the several techniques 
available in order to produce a new generation that will be 
better than the ones considered so far. 



The fitness function is given by: 



f(k)-- 



f 
logl- 

V 



N 



rep 



N, 



ik =1,2,3- 



(1) 



t J 



where, 

N re is the number of repetitions of records of similar 

stock levels that occurs throughout the period; 

N t is the total number of records of inventory values 

obtained after clustering; 

m is the total number of chromosomes for which the 

fitness function is calculated. 

In the fitness function, the ratio (N rep / N t ) plays the role of 
finding the probability of occurrence of a particular record of 



C. Selection 

The selection operation is the initial genetic operation 
which is responsible for the selection of the fittest 
chromosome for further genetic operations. The fitness 
function is carried out for each chromosome and the 
chromosomes are sorted on the basis of the result of the 
fitness function and ranked. The chromosome generating 
value as minimum as possible will be selected by the fitness 
function and will be subjected further to the genetic 
operations, crossover and mutation. 

D. Uniform Crossover 

Among the numerous crossover operators in practice, a 
uniform crossover is chosen in this proposed method for its 
advantages over the other forms. Uniform crossover is global 
and less biased when compared to that of standard and one 
point crossover. Uniform crossover does not select a set of 
crossover points. It simply considers each bit position of the 
two parents, and swaps the two bits with a probability of 50%. 
With large search spaces, a GA using uniform crossover 
outperforms a GA using one point crossover, which in turn 
outperforms a GA using two point crossover [35-36]. From 
the matting pool, two chromosomes are subjected for the 
uniform crossover. The chromosomes initially selected and 
after undergoing uniform crossover operation performed in 
this analysis is pictured in Fig. 4. As soon as the crossover 
operation is completed, the genes of the two chromosomes 
present get interchanged. 

Before Crossover 



?} 7000 | -200 | -600 | -500 | 150 | -JpOTftOO -400 | 700 | -600 I 



| P2 | 5000 I 400 | -800 500 445 | 3 ] 5 | -320 | 405 -150 100 
After Crossover 

P3 | 7000 | -150 -820 -400 ! 315 | -350 -750 | 405 500 | 600 | 

I P2 5000 450 I 700 I -600 350 I -200 I 800 I 500 I -100 [50l 



Fig. 4 Chromosomes after uniform crossover operation 

E. Mutation 

The crossover operation is succeeded by the final stage of 
genetic operation known as Mutation. In the mutation, a new 
chromosome is obtained. This chromosome is totally new 
from the parent chromosome. The concept behind this is the 
child chromosome thus obtained will be fitter than the parent 
chromosome. The performance of mutation operation is 
shown in Fig. 5. 



175 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



TJefore Mutation 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 

V . Experimental Re s u l t s 



Mp 



Mp 2l 



N%, 



Mp, 



P2 | 5000 | jjfj | 700 -600 ' 350 | -200 SOP ; 5()(i -1()(J | 15 



After Mutation 



Evf.p.. 



Vpi 



Mpa, 



Mp, 



1 



P2 1000 -GOD 700 4.^0 3:10 500 £00 -200 4.00 1.50 



Fig. 5 Chromosome subjected to mutation operation 

Four mutation points are chosen as shown in Fig. 5. 
The mutation is done on the particular gene present at the 
Mutation points. This pointing of gene is done randomly. 
Hence, the four mutation points may point any of the ten 
genes. 

The process explained so far will be repeated along with 
the new chromosome obtained from the previous process. 
In other words, at the end of each of the iteration, a best 
chromosome will be obtained. This will be included with the 
newly generated random chromosome for the next iteration. 
When the number of iterations is increased then the obtained 
solution moves very closer to the accurate solution. More the 
number of iterations the more accurate the optimal solution 
will be. Eventually with the help of the Genetic algorithm, the 
best stock level to be maintained in the members of the supply 
chain could be predicted from the past records, so that the loss 
due to the holding of excess stock level and shortage level can 
be reduced leading to an optimal inventory solution. 



The approach suggested for the optimization of inventory 
level and thereby an efficient supply chain management has 
been implemented in the platform of Lab VIEW 2009. 
The database consists of the records of stock levels held by 
each member of the supply chain for every period. For 
implementation, seven different products in circulation with 
the ten member supply chain network have been considered. 
A sample set of data from a large database used in the 
implementation is given in Table 1 . 

Table 1. A Sample Dataset Constituted by the Product Identification 
along with its Stock Levels in Each Member of the Supply Chain 



PI 


Fl 


DC1 


DC2 


DC3 


Al 


A2 


A3 


A4 


A5 


A6 


7 


-371 


-736 


-299 


634 


448 


756 


340 


-736 


-778 


863 


5 


-407 


379 


-981 


-864 


-391 


999 


-196 


307 


-171 


-529 


2 


-146 


-604 


443 


746 


-561 


-734 


445 


424 


-891 


-824 


4 


-962 


-524 


-685 


-254 


205 


446 


-469 


108 


346 


840 


3 


-834 


266 


969 


965 


735 


244 


-752 


133 


-554 


-939 


3 


-449 


-282 


577 


-926 


-414 


-200 


-743 


850 


196 


851 


4 


540 


-830 


-835 


882 


-379 


768 


-635 


-112 


539 


107 


3 


-778 


-313 


629 


-690 


824 


-927 


850 


307 


-171 


-529 


2 


351 


293 


328 


-732 


357 


-566 


685 


424 


-891 


-824 


1 


500 


108 


490 


-345 


-236 


108 


-931 


-260 


-144 


162 


5 


844 


-728 


286 


740 


686 


-421 


424 


-792 


-927 


-879 


4 


-321 


902 


-450 


-260 


-144 


162 


238 


307 


-171 


-529 


3 


775 


-394 


-520 


-792 


-927 


-879 


-507 


424 


-891 


-824 


4 


794 


932 


-584 


307 


-171 


-529 


-503 


108 


346 


840 


2 


-122 


-686 


-620 


424 


-891 


-824 


941 


133 


-554 


-939 


6 


235 


464 


401 


108 


346 


840 


-934 


464 


401 


108 


5 


218 


-848 


836 


133 


-554 


-939 


-834 


-848 


836 


133 


4 


489 


409 


148 


850 


196 


851 


-495 


-144 


162 


238 


3 


-422 


638 


676 


-112 


539 


107 


-440 


-927 


-879 


-507 


5 


893 


520 


-423 


-736 


-778 


863 


-335 


676 


-112 


539 



[fa T:t Vir.:- : r -i*rh ^-fit- It.-.- V.'nHrv^ 1 1* r — 


\± H k J ,.^ h „..u, k,i . |:_.|^.|^ii^i m^ 




























































Inventory Optimization - Using GA 


















































































n li:i mP. -i 


a 




-j.:.i- 






























L" Mi. <.■ U ■. Lipliii L:jlivii-..'d.LLd..\ 
'f-.rU^d.rr.- 


f 

-JJ1-- 















































































































































































/ 


/' 




































































V ~\ 






















































/ 




































" 






















































/ 

/ 

/ 








































v:i 




..„i-,H,- 




























































U H 5J >1 HU Lii .A \.<y iJi 
































































































































































































u-,r ■■„.-, ■„„.■ v 






























inMai. 




i 


hlHJtlJ.111111 


ffi~|7nnmrani 


llfl 
































■-1j. l L. -:... 1 ^uLj^KulJ 




i.i:«jo:ii Di-ii 


J > iiflimrant.il 


|lfi 






Generate V\<A 
































p |... s 




ti (\ (\ ii n:n 


■■ iiJLmjiiimjtii 








































MHJClWHJimjlIL 


| , i immi r \cr. 


































■,1.1-jhP Irr Miitor* Rw 7 










































l T I7 ? 1 


i 


Iclution 




































■-l.n:t: -■:< 1 MutriKo:,: 


'I h || IHH1 


-non || uo tw || em ?m 


■ KB 


-Mil || IflO ?>5 










































































































R„™,V,I„h 


f.n.i.ili:,-, 


































i} \-- I 






QJWijVJJ'Ji 


•» 


























































■ .. 1 


































■ 



Fig. 6 Inventory Optimization Tool created in Lab VIEW 2009 



176 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



In the sample database tabulated in Table 1, the first field 
comprises of the product Identification (PI) and the other 
fields are related with the stock levels that were held by the 
respective ten members of the supply chain network. For 
example, the first attribute and first field of the database is '7' 
which refers the Product I.D. '7'. The corresponding fields of 
the same attribute denote the stock levels of the product I.D. 
'7' in the respective members of the supply chain. Similarly, 
different sets of stock levels for different products are held by 
the database. 

As per the proposed analysis based on Uniform Crossover 
GA, two random initial chromosomes are generated as shown 
in Fig. 3. These initial chromosomes are subjected for the 
genetic operators, Uniform Crossover and Mutation. 
The resultant chromosome thus obtained after the application 
of crossover and mutation is shown in Fig. 5. 

Fig. 6 shows the main window of the tool created for 
inventory optimization using genetic algorithm in LabVIEW 
2009. The tool created is based on the uniform crossover 
genetic algorithm. 

The window displays the fitness function used, uniform 
crossover sample bit pattern randomly generated with 
probability of 0.5, the points at which mutation should occur, 
the end iteration count upon which the fitness function 
stabilizes, no. of supply chain members and population size. 

The best chromosome obtained after the required iterations 
and the fitness value and plot generated for the iteration value 
are also displayed. 

The organization can decide about the quantum of 
iterations for running the simulation to arrive at the optimal 
solution. As long as minimization of the fitness function is 
still possible, then the iteration continues till such a time that 
no improvement in the fitness function value is noticeable. 
After a certain number of iterations, if the fitness function 
value is not improving from the previous iterations, then this 
is an indication that the fitness function value is stabilizing 
and the algorithm has converged towards optimal solution. 
This inference is useful for deciding the number of iterations 
for running the GA simulation as well as this may be used as 
the stopping criteria for the algorithm. For greater accuracy, 
the number of iterations should be sufficiently increased and 
run on the most frequently updated large database of past 
records. 

As for our iteration value of '200', the resultant 
chromosome moved towards the best chromosome after each 
iterative execution. Hence at the end of the execution of 200 th 
iteration, best chromosome obtained is shown in Fig. 7. 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 
sufficient to reduce the loss either due to the holding of excess 
stocks or due to the shortage of stocks. By focusing on the 
excess/shortage inventory levels and initiating appropriate 
steps to eliminate the same at each member of the chain, it is 
possible to optimize the inventory levels in the upcoming 
period and thus minimize the supply chain cost. 



6 


1800 


-300 


800 


350 


650 


700 


-800 


-200 


100 


225 



Fig. 7 The final best chromosome obtained after 200 th iteration 

VI. Discussion of Results 

The final chromosome obtained from the GA based 
analysis shown in the Fig. 7 is the inventory level that has the 
potential to cause maximum increase of supply chain cost. 
It is inferred that controlling this resultant chromosome is 



The organization should take necessary steps to decrease 
the production of product 6 in the factory by 1800 units to 
make up for the predicted excess; increase the inventory level 
of product 1 by 300 units in distribution center 1 to make up 
for the predicted shortage, reduce inventory level of product 1 
by 800 units and 350 units in distribution centers 2 and 3 
respectively to make up for the predicted excess. 

Agent 1 should decrease the inventory level of product 6 
by 650 units. Agent 2 should decrease the inventory level of 
product 6 by 700 units. Agent 3 and Agent 4 should increase 
the inventory level of product 6 by 800 units and 200 units 
respectively to make up for the predicted excess / shortage. 
The inventory level of product 6 should be decreased by 100 
units and 225 units by Agent 5 and Agent 6 respectively. 
Thus by following the predicted stock levels, the 
excess/shortage inventory levels can be avoided in the 
upcoming period and thus the increase of supply chain cost 
can also be avoided. The analysis extracts an inventory level 
that made a remarkable contribution towards the increase of 
supply chain cost, and in turn enabled to predict the future 
optimal inventory levels to be maintained in all the supply 
chain members with the aid of these levels. Therefore it is 
possible to minimize the supply chain cost by maintaining the 
optimal stock levels that was predicted from the inventory 
analysis, and thus making the inventory management more 
effective and efficient. 

VII. Conclusion 

Inventory management is an important component of 
supply chain management. An innovative and efficient 
methodology that uses Genetic Algorithms with Uniform 
Crossover to precisely determine the most probable excess 
stock level and shortage level required for inventory 
optimization in the supply chain such that the total supply 
chain cost is minimal is proposed using LabVIEW 2009. 

The optimized stock level at all members of the supply 
chain is obtained by following the proposed genetic algorithm. 
Thus the proposed work gives a better prediction of stock 
levels amid diverse stock levels at all members of the supply 
chain. The complexity of increasing the number of products 
through the supply chain has been resolved by the proposed 
approach. Products due to which the members of the supply 
chain incur extra holding or shortage cost are also determined. 
More specifically, the inventory is optimized in the whole 
supply chain regardless of the number of products and the 
number of members in the supply chain. The proposed 
approach of inventory management has achieved the 
objectives which are the minimization of total supply chain 
cost and the determination of the products due to which the 
supplier endured either additional holding cost or shortage 
cost. 



177 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) 



References 



[1]. Sukran Kadipasaoglu, Jennifer Captain and Mark James, "Polymer 
Supply Chain Management", International Journal on Logistics Systems 
and Management, Vol. 4, No. 2, pp. 233-253, 2008. 

[2]. R. King and P. Phumpiu, "Reengineering the food supply chain: The 
ECR initiative in the grocery industry", American Journal of 
Agricultural Economics, Vol. 78, pp. 1181-1186, 1996. 

[3]. Frank Dooley, "Logistics, Inventory Control, and Supply Chain 
Management", CHOICES: The magazine of food, farm and resource 
Issues, Vol. 20, No. 4, 4 th Quarter 2005. 

[4]. R. Wilson, 15 th Annual State of Logistics Report. Council of Supply 
Chain Management Professionals, 2004. Available online: 
http://www.cscmp.org/. 

[5]. R. Ballou, "Business logistics/supply chain management", 5 th Ed. Upper 
Saddle River, NJ: Prentice Hall, 2004. 

[6]. M. Fisher, "What is the right supply chain for your product?", Harvard 
Business Review, Mar/ Apr., pp. 105-116, 1997. 

[7]. Guangyu Xiong and Hannu Koivisto, "Research on Fuzzy Inventory 
Control under Supply Chain Management Environment", Lecture Notes 
in Computer Science, Vol. 2658, pp. 673, 2003. 

[8]. M.C. Bonney, "Trends in Inventory Management", International 
Journal on Production Economics, Vol. 35, No. 1-3, pp. 107-114, 1994. 

[9]. M. Muller, "Essentials of Inventory Management", NewYork: Amer. 
Manage. Assoc, 2002. 

[10]. S. Nahmias, "Production and Operations Analysis", New York: 
McGraw-Hill/Irwin, 2004. 

[11]. Krishnamurthy, Khorrami and Schoenwald, "Decentralized Inventory 
Control for Large-Scale Reverse Supply Chains: A Computationally 
Tractable Approach", IEEE Transactions on Systems, Man, and 
Cybernetics, Vol. 38, No. 4, pp. 551-561, July 2008. 

[12]. Kanit Prasertwattana, Yoshiaki Shimizu and Navee Chiadamrong, 
"Evolutional Optimization on Material Ordering and Inventory Control 
of Supply Chain through Incentive Scheme", Journal of Advanced 
Mechanical Design Systems, and Manufacturing, Vol. 1, No. 4, pp. 562- 
573,2007. 

[13]. Mileff, Peter, Nehez, Karoly, "A new inventory control method for 
supply chain management", 12 th International Conference on Machine 
Design and Production, 2006. 

[14]. "Optimization Engine for Inventory Control", Golden Embryo 
Technologies Pvt. Ltd., Maharastra, India, 2004. 

[15]. Jinmei Liu, Hui Gao, Jun Wang, "Air material inventory optimization 
model based on genetic algorithm", Proceedings of the 3 rd World 
Congress on Intelligent Control and Automation, Vol. 3, pp. 1903 - 
1904,2000. 

[16]. CM. Adams, "Inventory optimization techniques, system vs. item level 
inventory analysis", 2004 Annual Symposium RAMS - Reliability and 
Maintainability, pp. 55 - 60, 26-29, Jan, 2004. 

[17]. P. Pongcharoen, A. Khadwilard and A. Klakankhai, "Multi-matrix real- 
coded Genetic Algorithm for minimizing total costs in logistics chain 
network", World Academy of Science, Engineering and Technology, 
Vol. 26, pp. 458-463, 14-16, December, 2007. 

[18]. Greg Scheuffele and Anupam Kulshreshtha, "Inventory Optimization A 
Necessity Turning to Urgency", SETLabs Briefings, Vol. 5, No. 3, 
2007. 

[19]. Vecchio and Paschalidis, "Enforcing service-level constraints in supply 
chains with assembly operations", Proceedings of IEEE Conference on 
Decision and Control, Vol. 5, pp. 5490- 5495, December 2003. 

[20]. Ioannis CH. Paschalidis, Yong Liu, Christos G. Cassandras and Christos 
Panayiotu, "Inventory control for supply chains with service level 
constraints: A synergy between Large Deviations and Perturbation 
analysis", Annals of Operations Research, Vol. 126, pp. 231-258, 2004. 

[21]. J. Sudhir Ryan Daniel and Chandrasekharan Rajendran, "A simulation- 
based genetic algorithm for inventory optimization in a serial supply 
chain", International Transactions in Operational Research, Vol. 12, 
pp. 101-127,2005. 

[22]. P. Radhakrishnan, V.M. Prasad and M.R. Gopalan, "Inventory 
optimization in supply chain management using genetic algorithm", 
International Journal of Computer Science and Network Security, 
Vol. 9, No. 1, January 2009, pp. 1-8. 



International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 

[23]. K. Wang and Y. Wang, "Applying Genetic Algorithms to Optimize the 
Cost of Multiple Sourcing Supply Chain Systems - An Industry Case 
Study", Studies on Computational Intelligence, Vol. 92, pp. 355-372, 
2008. 

[24]. Chih-Yao Lo, "Advance of Dynamic Production-Inventory Strategy for 
Multiple Policies Using Genetic Algorithm", Information Technology 
Journal, Vol. 7, pp. 647-653, 2008. 

[25]. Ulrich W. Thonemann and James R. Bradley, "The effect of product 
variety on supply-chain performance", European Journal of Operational 
Research, Vol. 143 No.3, pp. 548-69, 2002. 

[26]. S. Buffett and N. Scott, "An Algorithm for Procurement in Supply 
Chain Management", AAMAS-04 Workshop on Trading Agent Design 
and Analysis, New York, 2004. 

[27]. S. Behzadi, Ali A. Alesheikh and E. Poorazizi, "Developing a Genetic 
Algorithm to solve Shortest Path Problem on a Raster Data Model", 
Journal on Applied Sciences, Vol. 8, No. 18, pp. 3289-3293, 2008. 

[28]. Aphirak Khadwilard and Pupong Pongcharoen, "Application of Genetic 
Algorithm for Trajectory Planning of Two Degrees of Freedom Robot 
Arm With Two Dimensions", Thammasat International Journal on 
Science and Technology, Vol. 12, No. 2, April- June 2007. 

[29]. M.A. Sharbafi, M. Shakiba Herfeh, Caro Lucas and A. Mohammadi 
Nejad, "An Innovative Fuzzy Decision Making Based Genetic 
Algorithm", World Academy of Science, Engineering and Technology, 
Vol. 19, pp. 172-175, May 2006. 

[30]. Thomas Butter, Franz Rothlauf, Jorn Grahl, Hildenbrand Jens Arndt, 
"Developing Genetic Algorithms and Mixed Integer Linear Programs 
for Finding Optimal Strategies for a Student's Sports Activity", Working 
Papers in Information Systems, University of Mannheim, 2006. 

[31]. S.A. Qureshi, S.M. Mirza and M. Arif, "Fitness Function Evaluation for 
Image Reconstruction using Binary Genetic Algorithm for Parallel Ray 
Transmission Tomography", International Conference on Emerging 
Technologies, 2006. ICET '06. 13-14, Nov. 2006, pp. 196-201. 

[32]. Saifuddin Md. Tareeq, Rubayat Parveen, Liton Jude Rozario and Md. 
Al-Amin Bhuiyan, "Robust Face detection using Genetic Algorithm", 
Journal on Information Technology, Vol. 6, No. 1, pp. 142-147, 2007. 

[33]. M. Soryani and N. Rafat, "Application of Genetic Algorithms to Feature 
Subset Selection in a Farsi OCR", World Academy of Science, 
Engineering and Technology, Vol. 18, pp. 113-116.. 

[34]. William M. Spears and K.A. De Jong, "On the Virtues of Uniform 
Crossover", 4 th International Conference on Genetic Algorithms, 
La Jolla, California, July 1 99 1 . 

[35]. Syswerda, Gilbert, "Uniform Crossover in Genetic Algorithms", Proc. 
3 rd International Conference on Genetic Algorithms, Morgan Kaufman 
Publishing, 1989. 

[36]. Riccardo Poli and W.B. Langdon, "On the Search Properties of 
Different Crossover Operators in Genetic Programming", Proceedings 
of Genetic Programming'98, Madison, Wisconsin, 1998. 

ABOUT AUTHORS 

Mrs. S. Narmadha is working as Assistant Professor 
in the Department of Computer Science and 
Engineering in Park College of Engineering and 
Technology, Coimbatore. She obtained her Bachelor's 
degree in Computer Science and Engineering from 
Tamilnadu College of Engineering, Coimbatore under 
Bharathiar University and Master's degree in 
Mechatronics from Vellore Institute of Technology, 
Vellore. She is currently pursuing Ph.D. under Anna 
University, Chennai. She has 8 years of Teaching 
Experience and 2 years of Industrial experience. She has published 14 papers 
in International Conferences, 2 papers in International journals and a book on 
'Open Source Systems'. She is life member of ISTE. Her field of interest 
includes Supply Chain Management, Automation, Database Management 
Systems, Virtual Instrumentation, Soft Computing Techniques and Image 
Processing. 




178 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 



Dr. V. Selladurai is the Professor and Head, Department 

of Mechanical Engineering and Principal, Coimbatore 

Institute of Technology, Coimbatore, India. He holds a 

Bachelor's degree in Production Engineering, a Master's 

degree in Industrial Engineering specialisation and a 

Ph.D. degree in Mechanical Engineering. He has two 

years of industrial experience and 22 years of teaching 

experience. He has published over 90 papers in the 

proceedings of the leading National and International 

Conferences. He has published over 35 papers in international journals and 22 

papers in national journals. His areas of interest include Operation Research, 

Artificial Intelligence, Optimization Techniques, Non- Traditional 

Optimization Techniques, Production Engineering, Industrial and 

Manufacturing Systems, Industrial Dynamics, System Simulation, 

CAD/CAM, FMS, CIM, Quality Engineering and Team Engineering. 



Mr. G. Sathish is a full time research scholar under 
Anna University, Coimbatore. He holds a Bachelor's 
degree and a Master's degree in Computer Science and 
Engineering. He has 8 years of industrial experience. 
He has published 10 papers in the proceedings of the 
leading International Conferences. His field of interest 
includes Supply Chain Management, Optimization 
Techniques, Data Mining, Knowledge Discovery, 
Automation, Soft Computing Techniques and Image 
Processing. 




1 79 http://sites.google.com/site/ijcsis/ 

ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 



Efficient Inventory Optimization of Multi Product, 
Multiple Suppliers with Lead Time using PSO 



S.Narmadha 

Assistant Professor 

Department of Computer Science and Engineering 

Park College of Engineering and Tekhnology 

Coimbatore - 641659, Tamilnadu, India 



Dr.V.Selladurai 

Professor and Head 
Department of Mechanical Engineering 

Coimbatore Institute of Technology 
Coimbatore - 641014, Tamilnadu, India 



G.Sathish 

Research Scholar 
Department of Computer Science and Engineering 
Anna University - Coimbatore, Tamilnadu, India 



Abstract - With information revolution, increased globalization and 
competition, supply chain has become longer and more complicated 
than ever before. These developments bring supply chain 
management to the forefront of the management's attention. 
Inventories are very important in a supply chain. The total investment 
in inventories is enormous, and the management of inventory is 
crucial to avoid shortages or delivery delays for the customers and 
serious drain on a company's financial resources. The supply chain 
cost increases because of the influence of lead times for supplying the 
stocks as well as the raw materials. Practically, the lead times will not 
be same through out all the periods. Maintaining abundant stocks in 
order to avoid the impact of high lead time increases the holding cost. 
Similarly, maintaining fewer stocks because of ballpark lead time 
may lead to shortage of stocks. This also happens in the case of lead 
time involved in supplying raw materials. A better optimization 
methodology that utilizes the Particle Swarm Optimization algorithm, 
one of the best optimization algorithms, is proposed to overcome the 
impasse in maintaining the optimal stock levels in each member of 
the supply chain. Taking into account the stock levels thus obtained 
from the proposed methodology, an appropriate stock levels to be 
maintained in the approaching periods that will minimize the supply 
chain inventory cost can be arrived at. 

Keywords: Supply Chain Management, Inventory Optimization, 
Base Stock, Multiple Suppliers, Lead Time, Particle Swarm 
Optimization (PSO), Supply Chain Cost 



I. 



INTRODUCTION 



Inventory takes many forms, ranging from raw materials to 
finished goods. While holding large amounts of inventory 
enables a company to be responsive to fluctuations in customer 
demand, the associated costs can be excessive. In order to 
operate in a lean environment at maximum efficiency levels, 
companies must minimize all unnecessary expenses, including 
those associated with production and storage of inventories. 

Inventory control is typically a key aspect of almost every 
manufacturing and/or distribution operation business. The 
ultimate success of these businesses is often dependent on its 
ability to provide customers with the right goods, at the right 
place, at the right time. The right goods are those that the 
customer wants; the right place is your "available" inventory, 



not the supplier's warehouse, and in today's economy the right 
time is immediately. 

Failure to have the right goods in the right place at the 
right time often leads to lost sales and profits and, even worse, 
to lost customers. Today's reality is that there is very little 
differentiation between commodity products of the same type, 
and customers will, more often than not, choose to return to 
businesses that meet all three conditions, even choosing 
relatively unknown brands over known brands. 

The role of inventory management is to coordinate the 
actions of all business segments, particularly sales, marketing 
and production, so that the appropriate level of stock is 
maintained to satisfy customers' demands. The goal of 
inventory management is to balance supply and demand as 
closely as possible in order to keep customers satisfied and 
drive profits. 

Inventory management is a fundamental requisite to supply 
chain optimization. The processes and controls of effective 
inventory management are critical to any successful business. 
Since it is rarely the case that any business has the luxury of 
unlimited capital, inventory management involves important 
decisions about what to buy or produce, how much to buy or 
produce and when to buy or produce within the capital limits. 
These are "value decisions." Excessive inventory investments 
can tie up capital that may be put to better use within other 
areas of the business. On the other hand, insufficient inventory 
investment can lead to inventory shortages and a failure to 
satisfy customer demand. A balance must be struck and 
maintained. 

The aim of inventory management is to reduce inventory 
holdings to the lowest point without negatively impacting 
availability or customer service levels. This can be done while 
still maximizing the business' ability to exploit economies of 
scale to positively impact profitability. 

Inventory optimization takes inventory management to the 
next level, enabling businesses to further reduce inventory 
levels while improving customer service levels and 
maximizing capital investments. 



180 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



Inventory management is an ongoing process that relies on 
inputs from forecasts and product pricing, and should be 
executable within the cost structure of the business under an 
overall plan. Inventory control involves three inventory forms 
of the flow cycle: 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 
treated by easy expansion of the method. Hence, the 
mixed integer nonlinear optimization problems with both 
continuous and discrete variables can be treated with this 
method. 



• Basic Stock - The exact quantity of an item required to 
satisfy a demand forecast. 

• Seasonal Stock - A quantity buildup in anticipation of 
predictable increases in demand that occur at certain times 
in the year. 

• Safety Stock - A quantity in addition to basic inventory 
that serves as a buffer against uncertainty. 

The challenge is to weigh the balance in favor of basic 
stock so that the business holds as little safety stock as 
possible and provides 'just the right amount' of seasonal stock. 
However, the predictability of demand has a direct impact on 
how much safety stock a business must hold. When demand is 
unpredictable, higher levels of safety stock must be 
maintained. Therefore, the search for the optimal inventory 
levels to achieve a lean manufacturing environment becomes a 
key objective. 

A. Benefits of Inventory Optimization 

The primary function of an Inventory Optimization 
solution is to allow companies to effectively fulfill demand 
and identify how to gain additional profits from their 
inventories. Improved efficiencies through effective resource 
management and optimization lead to an increase in service 
level, improved performance against customer request dates 
and improved return on equity. These gains are derived in 
three ways: a) System Benefits b) Value-Added Benefits and 
c) Strategic Benefits. 

B. Particle Swarm Optimization 

In 1995, Kennedy and Eberhartin, inspired by the 
choreography of a bird flock, first proposed the Particle 
Swarm Optimization (PSO). In comparison with the 
evolutionary algorithm, PSO, relatively recently devised 
population-based stochastic global optimization algorithm, has 
many similarities and the robust performance of the proposed 
method over a variety of difficult optimization problems has 
been proved [1]. In accordance with PSO, either the best local 
or the best global individual affects the behavior of each 
individual in order to help it fly through a hyperspace [2]. 
Simulation of simplified social models has been employed to 
develop Particle Swarm Optimization techniques. The 
following are the features of the method [3]: 

• The researches on swarms such as fish schooling and bird 
flocking are the basis of the method. 

• The computation time is short and it requires little 
memory as it is based on a simple concept. 

• Nonlinear optimization problems with continuous 
variables were the initial focus of this method. 
Nevertheless, problems with discrete variables can be 



In addition to PSO, several evolutionary paradigms exist 
which include Genetic algorithms (GA), Genetic programming 
(GP), Evolutionary strategies (ES) and Evolutionary 
programming (EP). Biological evolution is simulated by these 
approaches which are based on population [4]. Genetic 
algorithm and PSO are two widely used types of evolutionary 
computation techniques among the various types of 
Evolutionary Computing paradigms [5]. 

PSO and evolutionary computation techniques such as 
Genetic Algorithms (GA) have many similarities between 
them. A population of random solutions is used to initialize 
the system which updates generations to search for optima. 
Nevertheless, PSO does not have evolution operators such as 
crossover and mutation that are available in GA. 

In PSO, the potential solutions, called particles follow the 
current optimum particles to fly through the problem space. 
Every particle represents a candidate solution to the 
optimization problem. The best position visited by the particle 
and the position of the best particle in the particle's 
neighborhood influences its position. 

Particles would retain part of their previous state using 
their memory. The particles would still remember the best 
positions they ever had even as there are no restrictions for 
particles to know the positions of other particles in the 
multidimensional spaces. An initial random velocity and two 
randomly weighted influences: individuality (the tendency to 
return to the particle's best previous position), and sociality 
(the tendency to move towards the neighborhood's best 
previous position) form each particle's movement [6]. 

When the neighborhood of a particle is the entire swarm, 
the global best particle refers to the best position in the 
neighborhood and in this case, gbest PSO refers the resulting 
algorithm. Generally, lbest PSO refers the algorithm in cases 
when smaller neighborhoods are used [5]. A fitness function 
that is to be optimized evaluates the fitness values of all the 
particles [6]. 

PSO uses individual and group experiences to search the 
optimal solutions. Nevertheless, previous solutions may not 
provide the solution of the optimization problem. The optimal 
solution is changed by adjusting certain parameters and 
putting random variables. The ability of the particles to 
remember the best position that they have seen is an advantage 
of PSO [6]. 



II. 



Related Review 



A fresh Genetic Algorithm (GA) approach for the 
Integrated Inventory Distribution Problem (IIDP) has been 
projected by Abdelmaguid et al. [7]. They have developed a 
genetic representation and have utilized a randomized version 
of a formerly developed construction heuristic in order to 
produce the initial random population. 



181 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



Pongcharoen et al.[8] have put forth an optimization tool 
that works on basis of a Multi-matrix Real-coded Generic 
Algorithm (MRGA) and aids in reduction of total costs 
associated with in supply chain logistics. They have 
incorporated procedures that ensure feasible solutions such as 
the chromosome initialization procedure, crossover and 
mutation operations. They have evaluated the algorithm with 
the aid of three sizes of benchmarking dataset of logistic chain 
network that are conventionally faced by most global 
manufacturing companies. 

A technique to utilize in supply-chain management that 
supports the decision-making process for purchases of direct 
goods has been projected by Buffett et al.[9]. RFQs have been 
constructed on basis of the projections for future prices and 
demand and the quotes that optimize the level of inventory 
each day besides minimizing the cost have been accepted. The 
problem was represented as a Markov Decision Process 
(MDP) that allows for the calculation of the utility of actions 
to be based on the utilities of substantial future states. The 
optimal quote requests and accepts at each state in the MDP 
were determined with the aid of Dynamic programming. A 
supply chain management agent comprising of predictive, 
optimizing and adaptive components called the TacTex-06 has 
been put forth by Pardoe et al. [10]. TacTex-06 functions by 
making predictions regarding the future of the economy, such 
as the prices that will be proffered by component suppliers and 
the degree of customer demand and then strategizing its future 
actions so as to ensure maximum profit. 

Beamon et al.[ll] have presented a study on evaluations of 
the performance measures employed in supply chain models 
and have also displayed a framework for the beneficial 
selection of performance measurement systems for 
manufacturing supply chains Three kinds of performance 
measures have been recognized as mandatory constituents in 
any supply chain performance measurement system. New 
flexibility measures have also been created for the supply 
chains. The accomplishment of beam-ACO in supply-chain 
management has been proposed by Caldeira et al.[12]. Beam- 
ACO has been used to optimize the supplying and logistic 
agents of a supply chain. A standard ACO algorithm has aided 
in the optimization of the distributed system. The application 
of Beam-ACO has enhanced the local and global results of the 
supply chain. 

A beneficial industry case applying Genetic Algorithms 
(GA) has been proposed by Wang et al.[13]. The case has 
made use of GAs for the optimization of the total cost of a 
multiple sourcing supply chain system. The system has been 
exemplified by a multiple sourcing model with stochastic 
demand. A mathematical model has been implemented to 
portray the stochastic inventory with the many to many 
demand and transportation parameters as well as price 
uncertainty factors. A genetic algorithm which has been 
approved by Lo [14] deals with the production-inventory 
problem with backlog in the real situations, with time-varied 
demand and imperfect production due to the defects in 
production disruption with exponential distribution. Besides 
optimizing the number of production cycles to generate a (R, 
Q) inventory policy, an aggregative production plan can also 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
be produced to minimize the total inventory cost on the basis 
of reproduction interval searching in a given time horizon. 



B arias et al.[15] have developed a System Dynamics 
simulation model of a typical retail supply chain. The intent of 
their simulation exercise was to build up inventory policies 
that enhance the retailer's revenue and reduce costs at the same 
instant. Besides, the research was also intended towards 
studying the implications of different diversification strategies. 
A supply chain model functioning under periodic review base- 
stock inventory system to assist the manufacturing managers 
at HP to administer material in their supply chains has been 
introduced by Lee et al.[16]. The inventory levels across 
supply chain members were obtained with the aid of a search 
routine. 

The inventory and supply chain managers are mainly 
concerned holding of the excess stock levels and hence the 
increase in the holding cost. Meanwhile, there is possibility for 
the shortage of products. For the shortage of each product 
there will be a shortage cost. Holding excess stock levels as 
well as the occurrence of shortage for products lead to the 
increase in the supply chain cost. The factory may 
manufacture any number of products, each supply chain 
member may consume a few or all the products and each 
product is manufactured using a number of raw materials 
sourced from many suppliers. All these factors pose additional 
holding of the excess stock levels and hence the increase in the 
holding cost. Meanwhile, there is possibility for the shortage 
of products. For the shortage of each product there will be a 
shortage cost. Holding excess stock levels as well as the 
occurrence of shortage for products lead to the increase in the 
supply chain cost. All these factors pose additional challenge 
in extracting the exact product and the stock levels that 
influence the supply chain cost heavily. 

Many well-known algorithmic advances in optimization 
have been made, but it turns out that most have not had the 
expected impact on the decisions for designing and optimizing 
supply chain related problems. Some optimization techniques 
are of little use because they are not well suited to solve 
complex real logistics problems in the short time needed to 
make decisions and also some techniques are highly problem- 
dependent which need high expertise. This adds difficulties in 
the implementations of the decision support systems which 
contradicts the tendency to fast implementation in a rapidly 
changing world. 10 techniques need to determine a globally 
optimal placement of inventory, considering its cost at each 
stage in the supply chain and all the service level targets and 
replenishment lead times that constraint each inventory 
location. 

III. Objectives 

The supply chain cost increases because of the influence of 
lead times for supplying the stocks as well as the raw 
materials. Practically, the lead times will not be same 
throughout all the periods. Maintaining abundant stocks in 
order to avoid the impact of high lead time increases the 
holding cost. Similarly, maintaining fewer stocks because of 
ballpark lead time may lead to shortage of stocks. This also 



182 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 



happens in the case of lead time involved in supplying raw 
materials. A better optimization methodology would consider 
all these above mentioned factors in the prediction of the 
optimal stock levels to be maintained such that the total supply 
chain cost can be minimized. Here, an optimization 
methodology that utilizes the Particle Swarm Optimization 
(PSO) algorithm, one of the best optimization algorithms, is 
proposed to overcome the impasse in maintaining the optimal 
stock levels in each member of the supply chain. Taking into 
account the stock levels thus obtained from the proposed 
methodology, an appropriate stock levels to be maintained in 
the approaching periods that will minimize the supply chain 
inventory cost can be arrived at. 

Supply chain model is broadly divided into four stages in 
which the optimization is going to be performed. The supply 
chain model is illustrated in the Fig. 1. 



A. Inventory Optimization of Multiproduct, Multiple 
Suppliers with Lead Time 

Effective supply chain strategies must take into account the 
interactions at various levels in the supply chain. It is 
challenging to design and operate a supply chain so that total 
system wide costs are minimized and system wide service 
levels are maintained. 

Uncertainty is inherent in every supply chain. Supply 
chains need to be designed to eliminate as much uncertainty as 
possible to deal effectively with the uncertainty that remains. 
Demand is not the only source of uncertainty. Delivery lead 
times, manufacturing yields, transportation times and 
component/raw material availability can also have significant 
impact on supply chain. Inventory and back order levels 
fluctuate considerably across the supply chain, even when 
customer demand for specific products does not vary greatly. 
The two desired objectives of improved service and inventory 
levels seem to be not achieved at the same time since 
traditional inventory theory tells us that to increase service 
level, the firm must increase inventory and therefore cost. But 
recent developments in information and communications 
technologies, have led to the innovative approaches that allow 
the firm to improve both the objectives simultaneously. 



Raw material 
source ] 



Raw materia] 
source 2 



Factory 



Raw material 
source m 



Stage 1 



Stage 2 



Distribution 
Center ] 



^h 



Agent ] 



-L-* 



Agent 2 



Ayjni ,V.., 



Distribution 
Center 2 



^ 



Agent ] 



Distribution 
Center i 



stage 3 




A sent ; 



l 



Agent A d 



End 

Level 

customer 



•n 



Agent ] 



-u* 



Agent 2 



I s * A°e 



ent A d . 




Stage 4 



Fig. 1 Four Stage Supply Chain Model 



183 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 



In this present research, a prediction analysis that considers all 
these factors that are mentioned above, such that the analysis 
paves the way for minimizing the supply chain cost has been 
proposed. The supply chain cost can be minimized by 
maintaining optimal stock levels in each supply chain 
member. Such optimal stock levels can be predicted only by 
analyzing the past records. This minimization of supply chain 
cost is effective only if the optimal level is predicted with the 
knowledge of the lead times of the stocks. Hence a 
methodology is developed that analyze the past records and 
predict the emerging excess/shortage of stock levels that are to 
be considered to identify the optimal stock levels which will 
be maintained by each of the supply chain member. 

Particle Swarm Optimization algorithms, one of the 
optimization algorithms in Evolutionary computation is used 
for analysis purpose. The stock levels that are obtained from 
the analysis are the stock levels that contribute more to the 
increase of total supply chain cost. These stock levels are used 
for the prediction of the optimal stock levels that need to be 
maintained in each supply chain member. 

B. PSO Model for Prediction Analysis 

The methodology proposed here will minimize the total 
supply chain cost by predicting optimal stock levels not only 
by considering the past records with respect to the stock 
levels, but also the lead time of the products to reach each 
supply chain member from its previous stage as well as the 
lead time taken in supplying the raw materials to the factory. 
Usually, shortage for a particular stock at a particular member, 
excess stock levels at a particular member, time required to 
transport stock from one supply chain member to another i.e. 
lead time of a stock at a member, time taken to supply raw 
materials to the factory to manufacture certain products i.e. 
lead time of raw materials in factory are some of the key 
factors that play vital role in deciding the supply chain cost. A 
better optimization methodology should consider all these 
factors. In the proposed methodology all the above mentioned 
key factors in predicting the optimal stock levels are 
considered. Also, different priorities are assigned to those 
above factors. As per the priority given, the corresponding 
factors will influence the prediction of optimal stock levels. 
Hence as per the desired requirement, the optimal stock level 
will be maintained by setting or changing the priority levels in 
the optimization procedure. 

The optimization is going to be performed in the supply 
chain model as illustrated in the Fig. 1 . 

The members participating in the supply chain model are 
raw material sources (r 1 ,r 2 ,r 3 , •••,r m } , a factory /, i 

distribution centers D ={d v d 2 , d 3 , •••,d i } and the agents 

A { Aj , A^ , A d ,• • • , A d } , A d is the number of agents for 

the distribution center d i . Hence, the total number of agents in 
the supply chain model can be arrived using formula : 



where N A is the total number of agents used in the supply 
chain model. 

The factory is manufacturing k number of products. The 
database holds the information about the stock levels of each 
product in each of the supply chain member, lead time of 
products at each supply chain member and lead time of raw 

material. For / members from factory to end-level- Agents, 
there are /— 1 lead times for a particular product and these 
times are collected from the past records. Similarly, the lead 
time for raw materials from r m to / is also taken from the 

earlier period and thus the database is constituted. Each and 
every dataset recorded in the database is indexed by a 
Transportation Identification (TID). For p periods, the TID 

will be IT ,"f ,"f ,• • • ,T } . This TID will be used as an index in 

mining the lead time information. 

Now, the particle Swarm Optimization (PSO) is utilized to 
predict the emerging excess/shortage of stock levels which are 
vital information for optimal stock levels to be maintained in 
the future to minimize the supply chain cost. The procedures 
involved in determining the optimal stock levels are illustrated 
in Fig. 2. 

As the particle swarm optimization (PSO) is more suitable 
for finding the solution for the optimization problem, PSO is 
utilized in finding the optimal stock levels to be maintained in 
each member of the supply chain. The flow of procedures is 
discussed below. 

The individuals of the population including searching 

points, velocities, Pfr est and g^ est are initialized randomly 

but within the lower and upper bounds of the stock levels for 
all supply chain members, which have to be specified earlier. 
Hence the generated searching point individual is 



4P k S 1 S 2 S 3 SJ,[ =1,2,3- 



(2) 



where,P LB <P k <P UB ,S LB < S, <S UI 



P LB , Py B and S L B , S UB are the lower and upper bound values 
of the number of products and stock levels respectively. 

The generated population is having the size of N p i.e. the 
number of individuals. Since the total number of members that 

are maintaining the stock levels from / is / , the dimension d 

of each individual is given by 



d =1 + 1 
and hence the equation (2). 



(3) 



N A=IX 



(1) 



Similarly, the initial velocity for the individual will be 

v i v 2 v 3 v 1+1 ],v mm <v 1+1 <v max (4) 

where V •„ and v mov are the minimum and maximum limit 

II11I1 IllaX 

for velocities respectively. 



184 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



Star: 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 

f P(occ) A 



Generation of searching points, 
velocities, p^t and g^* 



Calculation of 
Evaluation value 



Take down pteat which is having 
a best evaluation value to gbest 



Calculate new velocities 
and new particle 




Fig. 2 Particle Swarm Optimization Steps in Optimizing 
the Stock Levels 

Then each individual is queried into the database for 
obtaining the details regarding the TID and frequency of the 
individual. This will bring T , q ep andP ( occ ) , number of 
periods of occurrence of that particular individual. Here, q 
need not to be consecutive integers. This obtained TID is 
queried into the database having the lead time of a particular 
product to a particular supply chain member. The product I.D. 
obtained from the individual is queried into the database 
having the lead time each raw material for the particular 
product. After all these queries, the lead time of stocks 
obtained is as follows 



T= [t 

s L a. 



VJ 



- q,l q,2 q,l-l 

And the lead time for raw materials is obtained as 



(5) 



T=ft t 2 t r ] (6) 

where r is the number of raw materials required for a certain 
product. 



f (a) = w, 



1- 



V 



T(jperiod^ 



+ log(w 2 .l 



itock+Ws'U 



=1,2,3- 



(7) 



where T {periods) is the total number of periods of records 
in database. 

In equation (7), W x , W 2 and w 3 are the weightings of the 

factors, stock levels, lead time of stocks and lead time of raw 
materials in optimization, respectively and they are determined 
as 

(8.1) 

(8.2) 

(8.3) 
J R 1 +R 2 +R 3 

R { ,R 2 and R 3 are the priority levels of influence of stock 

levels, lead time of stocks and lead time of raw materials in 
optimization respectively. Increasing the priority level of a 
factor increases the influence of the corresponding factor in 

the evaluation function. Hence this R { , R 2 and 7? 3 decides the 

amount of influence of the factors. The lead time of the stocks 

t stock * s determined as follows 



w, 


R, 


R, +R 2 +R 3 


w 2 


R 2 


R, +R 2 +R 3 




R 3 



/-l 



1 stock 



=zz« 

i=\ q 



qj 



(9) 



And the lead time required to fill the raw materials is given 



as 



-s«. 



(10) 



i=\ 



Equation (9) and (10) is substituted in the equation (7) 
gives an evaluation value for each individual. 

For every individual, a comparison is made between its 
evaluation value and its p best .The g best indicates the most 

excellent evaluation value among the p best . This g best is 

nothing but an index that points the best individual we have 
generated so far. 

Subsequently the adjustment of the velocity of each 
particle a is as follows: 



Then for each individual the evaluation function is 
calculated. 

Determination of Evaluation function where, 

The evaluation function is determined for each randomly 
generated individual. The function is given by 



Ja,b) =w*v cnt (a) +c 1 *r 1 *[p best (a,b) ■ 
+c 2 *r 2 *[g best (b)-I cnt (a,b)] 

a= U-— ,N p 
=4,2, ,d 



Icn t (a,b)] 



(11) 



185 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



Here v cnt (a) represents current velocity of the particle, 
V new (a,b) represents new velocity of a particular parameter 
of a particle, r x and r 2 are arbitrary numbers in the interval 
[0,1] , c x and c 2 are acceleration constants (often chosen as 

2.0), W is the inertia weight that is given as 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
different products are utilized and these products are in 
circulation in the seven member supply chain network 
considered. The sample database which consists of the past 
records is shown in Table 1 . 



- x iter 



iter 



(12) 



where w rriQY and W m : n are the maximum and minimum inertia 

IIldA 111111 

weight factors respectively that are chosen randomly in the 
interval [0,1] 

iter max is the maximum number of iterations 

iter is the current number of iteration 

Such newly obtained particle should not exceed the limits. 
This would be checked and corrected before proceeding 
further as follows, 

Ifv new (a,b) >v max (b),then^ iew (a,b) = v mas (b) 



ifv (a,b)<v (b),thenf (a,b) 



,(b) 



Then as per the newly obtained velocity, the parameters of 
each particle is changed as follows 



Xa,b) =€ cn (a,b) +v new (a,b) 



(13) 



Then the parameter of each particle is also verified whether 
it is beyond the lower bound and upper bound limits. If the 
parameter is lower than the corresponding lower bound limit 
then replace the new parameter by the lower bound value. If 
the parameter is higher than the corresponding upper bound 
value, then replace the new parameter by the upper bound 
value. For instance, 

If/> <P LB ,thenP k =P LB 
Similarly, if P k > P v B , then P k = P v B 

This is to be done for the other parameters also. 

This process will be repeated again and again until the 
maximum number of iterations is reached. Once the maximum 
number of iterations is attained the process is terminated. The 
latest g best pointing the individual is the best individual which 

is having the stock levels that are to be considered and these 
stock levels are utilized in taking the necessary steps for 
maintaining the optimal stock levels at each of the supply 
chain members. 



IV. EXPERIMENTAL RESULTS 

The approach suggested for the optimization of inventory 
level and thereby efficient supply chain management has been 
implemented in the MATLAB 7.4. The database consists of 
the records of stock levels held by each member of the supply 
chain for every period. For implementation purpose, five 



In the database tabulated in Table 1, the second field 
comprises of the product Identification (PI) and the other 
fields are related with the stock levels that were held by the 
respective seven members of the supply chain network. For 
example, the second attribute and second field of the database 
is '3' which refers the Product I.D. '3'. The corresponding 
fields of the same attribute denote the stock levels of the 
product I.D. '3' in the respective members of the supply chain. 
Similarly, different sets of stock levels are held by the 
database. 







Table 1: Sample data from database 
of different stock levels 




TI 


PI 


Fl 


F2 


F3 


F4 


F5 


F6 


F7 


1 


3 


632 


424 


247 


-298 


-115 


365 


961 


2 


5 


-415 


488 


-912 


979 


-492 


-922 


205 


3 


2 


369 


-686 


-468 


-807 


183 


-386 


-228 


4 


2 


459 


289 


-522 


-316 


130 


-854 


468 


5 


3 


-663 


944 


856 


451 


-763 


657 


484 


6 


4 


-768 


-937 


-768 


242 


369 


-890 


289 


7 


3 


-890 


619 


-629 


-844 


791 


285 


596 


8 


3 


193 


-263 


-474 


325 


-409 


-216 


-738 


9 


2 


578 


-890 


675 


-321 


-411 


239 


-916 


10 


1 


-192 


-421 


593 


394 


-141 


955 


-456 


11 


5 


494 


-317 


600 


363 


698 


927 


621 


12 


3 


-838 


-355 


776 


682 


-813 


-350 


-513 


13 


5 


673 


-593 


628 


637 


643 


622 


-244 


14 


1 


-974 


-311 


-319 


-189 


-449 


663 


520 


15 


3 


-725 


797 


184 


406 


-888 


-575 


-144 


16 


4 


811 


-569 


-473 


615 


467 


-748 


192 


17 


1 


-102 


-703 


-859 


983 


206 


-803 


-445 


18 


2 


426 


689 


-735 


-465 


680 


-913 


147 


19 


3 


655 


626 


-158 


-485 


-622 


-928 


515 


20 


5 


-175 


662 


-847 


819 


239 


-902 


372 



Table 2: Sample data from database which is 



1 


lavin 


g leac 


I time 


stor 


stock 


s 


TI 


TI 


T2 


T3 


T4 


T5 


T6 


1 


28 


27 


19 


9 


19 


19 


2 


35 


33 


16 


4 


24 


15 


3 


38 


38 


20 


8 


10 


18 


4 


25 


25 


9 


21 


22 


13 


5 


45 


40 


15 


4 


16 


11 


6 


36 


43 


7 


13 


21 


3 


7 


31 


27 


20 


5 


22 


4 


8 


45 


28 


5 


8 


10 


6 


9 


41 


45 


6 


22 


8 


20 


10 


34 


26 


5 


21 


12 


4 


11 


45 


38 


1 


2 


11 


16 


12 


40 


38 


9 


18 


22 


23 


13 


38 


40 


11 


17 


7 


10 


14 


47 


47 


22 


8 


23 


23 


15 


35 


27 


16 


12 


22 


4 



186 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



Sample data from database which is having lead times for 
stocks is given in Table 2; and sample data set taken from a 
database which is having raw material lead time for different 
products is given in Table 3. Initial random individuals for 
running the PSO algorithm is given in Table 4; and initial 
random velocities corresponding to each particle of the 
individual required for the PSO algorithm is given in Table 5. 

Table 3: Raw material lead time 
fo r different produ cts 



PI 


RM 


T 


1 


1 


20 


1 


2 


3 


1 


3 


8 


2 


1 


10 


2 


2 


3 


2 


3 


9 


2 


4 


23 


2 


5 


7 


3 


1 


24 


3 


2 


23 


3 


3 


20 


3 


4 


22 


4 


1 


3 


4 


2 


22 


4 


3 


19 


4 


4 


18 


5 


1 


23 


5 


2 


16 


5 


3 


23 


5 


4 


21 



Table 4: Initial Random Individuals 



PI 


Fl 


F2 


F3 


F4 


F5 


F6 


F7 


3 


855 


61 


215 


863 


24 


75 


-757 


5 


854 


-154 


145 


-241 


-215 


415 


845 



Table 5: Initial Random Velocities corresponding 
to each particle of the individual 


PI 


Fl 


F2 


F3 


F4 


F5 


F6 


F7 


-0.1298 


0.0376 


-0.3439 


0.3567 


0.0982 


-0.0560 


-0.1765 


-0.0409 


-0.4997 


0.0863 


0.3573 


-0.0113 


0.0524 


0.2177 


0.6550 


0.0342 



V. Discussion Of Results 

An iteration involving all these processes was carried out 
so as to obtain the best individual. Here for instance, the 
iteration value of '100' is chosen and so hundred numbers of 
iterative steps will be performed. The best individual obtained 
as a result is 



3 -602 -280 -821 398 382 -764 -125 


and its database format is depicted in the table 6. 
Table 6: Database format of Final Best Individual 


PI 


Fl 


F2 


F3 


F4 


F5 


F6 


F7 


3 


-602 


-280 


-821 


398 


382 


-764 


-125 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
As long as minimization of the fitness function is still 
possible, then the iteration continues till such a time that no 
improvement in the evaluation function value is noticeable. 
After a certain number of iterations, if the evaluation function 
value is not improving from the previous iterations, then this is 
an indication that the evaluation function value is stabilizing 
and the algorithm has converged towards optimal solution. 
This inference is useful for deciding the number of iterations 
for running the PSO simulation as well as this may be used as 
the stopping criteria for the algorithm. 

For greater accuracy, the number of iterations should be 
sufficiently increased and run on the most frequently updated 
large database of past records. In our experimentation, the 
iteration value of 100 is chosen and its value is 3.8220 and the 
values for weighting factors wl, w2 and w3 obtained are 
0.6250; 0.3125 and 0.0625 respectively. 

The final individual obtained from the PSO based analysis 
shown in the Table 6 is the inventory level that has potential to 
cause maximum increase of supply chain cost. It is inferred 
that controlling this resultant individual is sufficient to reduce 
the loss either due to the holding of excess stocks or due to the 
shortage of stocks. By focusing on the excess/shortage 
inventory levels and initiating appropriate steps to eliminate 
the same at each member of the chain, the organization can 
optimize the inventory levels in the upcoming period and thus 
minimize the supply chain cost. 

That is, the organization should focus on the most potential 
product among multi products that could cause maximum 
inventory cost and in this specific case it is product 3. 
The organization should take necessary steps to increase the 
production of product 3 in the factory by 602 units to make up 
for the predicted shortage; increase the inventory level of 
product 3 by 280 units in distribution centre 1 to make up for 
the predicted shortage; increase the inventory level of product 
3 by 821 units in distribution centre 2 to make up for the 
predicted shortage; decrease inventory level of product 3 by 
398 units in agent 1 to make up for the predicted excess; 
decrease inventory level of product 3 by 382 units in agent 2 
to make up for the predicted excess; increase the inventory 
level of product 3 by 764 units in agent 3 to make up for the 
predicted shortage; increase the inventory level of product 3 
by 125 units in agent 4 to make up for the predicted shortage. 

Thus by following the predicted stock levels, the increase 
in the supply chain cost due to excess/shortage inventory 
levels can be avoided. The analysis provided an inventory 
level that had maximum potential to cause a remarkable 
contribution towards the increase of supply chain cost. The 
organization can predict the future optimal inventory levels in 
all the supply chain members with the aid of these levels. 
Therefore it is concluded that it is possible to minimize the 
supply chain cost by maintaining the optimal stock levels in 
the upcoming period among various partners in the supply 
chain that was predicted from the inventory analysis, making 
the inventory management further effective and efficient. 



187 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



VI. Conclusion 



Inventory management is an important component of 
supply chain management. As the lead time plays vital role in 
the increase of supply chain cost, the complexity of predicting 
the optimal stock levels increases. The novel and proficient 
approach based on PSO algorithm, one of the best 
optimization algorithms, is proposed to overcome the impasse 
in maintaining the optimal stock levels in each member of the 
supply chain. The proposed methodology reduced the total 
supply chain cost as it undoubtedly established the most 
probable surplus stock level and shortage level along with the 
consideration of lead time in supplying the stocks as well as 
raw materials that are required for inventory optimization. 

The organizations can make use of the proposed 
techniques in this present research for inventory optimization 
by capturing the database in the desired format to suit their 
respective supply chain structure, replacing the simulated data 
used in this research with the real data of the organization. 
For greater accuracy, the number of iterations should be 
sufficiently increased and run on the most frequently updated 
large database of past records. Also the organization can 
mention the maximum possible lower limit and upper limit for 
the shortage and excess inventory levels respectively, within 
which the inventory is expected to fluctuate among the various 
members of the supply chain to make the convergence faster 
towards optimal solution. Further if the organization can 
evaluate and quantify the cost involved for each shortage as 
well as excess of inventory at each member of the supply 
chain, then the exact savings due to inventory optimization in 
the supply chain can be calculated for the organization. 

VII. Future scope 

Organization can adopt decomposition technique and use 
the proposed techniques for inventory optimization of 
high/medium/low value independent products. In this case, the 
organization should apply the technique for the high value 
items separately, medium value products separately and low 
value products separately for inventory optimization in the 
supply chain. 

References 

[1]. H. Lu, 2003, "Dynamic Population Strategy Assisted Particle Swarm 
Optimization in Multiobjective Evolutionary Algorithm design", IEEE 
Neural Network Society, IEEE NNS Student Research Grants 2002, 
Final reports. 

[2]. Hirotaka Yoshida Kenichi Kawata and Youshikazu Fukuyama Yosuke 
Nakanishi, 1999, "A Particle Swarm Optimization for Reactive Power 
and voltage control considering voltage stability", Proceedings of IEEE 
International Conference on Intelligent System Applications to Power 
Systems, pp. 117- 121, April 4- 8. 

[3]. Mahamed G.H. Omran, Andries P Engelbrecht, and Ayed Salman, 2005, 
"Dynamic Clustering using Particle Swarm Optimization with 
Application in Unsupervised Image Classification", Proceedings of 
World Academy of Science, Engineering and Technology, Vol. 9, 
pp. 199-204 

[4]. S. A. Hamdan, 2008, "Hybrid Particle Swarm Optimiser using multi- 
neighborhood topologies", INFOCOMP - Journal of Computer Science, 
Vol.7, No. l,pp.36-43. 

[5]. Ling-Feng Hsieh, Chao-Jung Huang and Chien-Lin Huang, 2007, 
"Applying Particle Swarm Optimization to Schedule Order Picking 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 

Routes in a Distribution Center", Asian Journal on Management and 

Humanity Sciences, Vol. 1, No. 4, pp. 558- 576. 
[6]. Alberto Moraglio, Cecilia Di Chio, Julian Togelius and Riccardo Poli, 

2008, "Geometric Particle Swarm Optimization", Journal on Artificial 

Evolution and Applications, Article ID 143624. 
[7]. T.F. Abdelmaguid and M.M. Dessouky, 2006, "A genetic algorithm 

approach to the integrated inventory-distribution problem", International 

Journal of Production Research., 44: 4445-4464. 
[8]. P. Pongcharoen, A. Khadwilard and A. Klakankhai, 2007, "Multi-matrix 

real-coded Genetic Algorithm for minimizing total costs in logistics 

chain network", Proceedings of World Academy of Science, Engineering 

and Technology, Vol. 26, pp. 458-463, 14-16, December. 
[9]. S. Buffett and N. Scott, 2004, "An Algorithm for Procurement in Supply 

Chain Management", AAMAS-04 Workshop on Trading Agent Design 

and Analysis, New York. 
[10]. D. Pardoe and P. Stone, 2007, An Autonomous Agent for Supply Chain 

Management. In: Handbooks in Information Systems Series: Business 

Computing, Adomavicius, G. and A. Gupta (Eds.). Elsevier. Amsterdam. 

http://www.cs.utexas.edu/~pstone/Papers/bib2html/b2hd-TacTex- 

Book07.html 
[11]. B.M. Beamon, 1998, "Supply chain design and analysis: Models and 

methods", International Journal on Production Economics, 55: 281-294. 
[12]. J.L. Caldeira, R.C. Azevedo, C.A. Silva and J.M.C. Sousa, 2007, 

"Supply-chain management using ACO and beam-ACO algorithms", 

Proceedings of the IEEE International Fuzzy Systems Conference, July 

23-26, London, pp. 1-6. 
[13]. K. Wang and Y. Wang, 2008, "Applying Genetic Algorithms to 

Optimize the Cost of Multiple Sourcing Supply Chain Systems - An 

Industry Case Study", Studies on Computational Intelligence, Vol. 92, 

pp. 355-372. 
[14]. Chih-Yao Lo, 2008, "Advance of Dynamic Production-Inventory 

Strategy for Multiple Policies using Genetic Algorithm", Information 

Technology Journal, Vol. 7, pp. 647-653. 
[15]. Y. Barlas and A. Aksogan, 1996, "Product Diversification and Quick 

Response Order Strategies in Supply Chain Management", 14 th 

International Conference of the System Dynamics Society 1996 

Cambridge, Massachusetts, USA, pp. 51-56. 
[16]. H.L. Lee and C. Billington, 1995. The evolution of supply-chain- 
management models and practice at Hewlett-Packard. Interface, 25: 42- 

63. 
[17]. K.E. Parsopoulos and M.N. Vrahatis, 2005 "Recent approaches to global 

Optimization problems through Particle Swarm Optimization", Natural 

Computing, 1 : 235-306. 
[18]. http://swarmintelligence.org 



ABOUT AUTHORS 

Mrs. S. Narmadha is working as Assistant Professor in 
the Department of Computer Science and Engineering 
in Park College of Engineering and Technology, 
Coimbatore. She obtained her Bachelor's degree in 
Computer Science and Engineering from Tamilnadu 
College of Engineering, Coimbatore under Bharathiar 
University and Master's degree in Mechatronics from 
Vellore Institute of Technology, Vellore. She is 
currently pursuing Ph.D. under Anna University, 
Chennai. She has 8 years of Teaching Experience and 2 
years of Industrial experience. She has published 14 papers in International 
Conferences, 2 papers in International journals and a book on 'Open Source 
Systems'. She is life member of ISTE. Her field of interest includes Supply 
Chain Management, Automation, Database Management Systems, Virtual 
Instrumentation, Soft Computing Techniques and Image Processing. 




188 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 



Dr. V. Selladurai is the Professor and Head, Department 

of Mechanical Engineering and Principal, Coimbatore 

Institute of Technology, Coimbatore, India. He holds a 

Bachelor's degree in Production Engineering, a Master's 

degree in Industrial Engineering specialisation and a 

^^ Ph.D. degree in Mechanical Engineering. He has two 

years of industrial experience and 22 years of teaching 

experience. He has published over 90 papers in the 

proceedings of the leading National and International 

Conferences. He has published over 35 papers in international journals and 22 

papers in national journals. His areas of interest include Operation Research, 

Artificial Intelligence, Optimization Techniques, Non-Traditional 

Optimization Techniques, Production Engineering, Industrial and 

Manufacturing Systems, Industrial Dynamics, System Simulation, 

CAD/CAM, EMS, CIM, Quality Engineering and Team Engineering. 




Mr. G. Sathish is a full time research scholar under Anna 
University, Coimbatore. He holds a Bachelor's degree 
and a Master's degree in Computer Science and 
Engineering. He has 8 years of industrial experience. He 
has published 10 papers in the proceedings of the leading 
International Conferences. His field of interest includes 
Supply Chain Management, Optimization Techniques, 
Data Mining, Knowledge Discovery, Automation, Soft 
Computing Techniques and Image Processing. 



189 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



Test Case Generation using Mutation Operators 

and Fault Classification 

Mrs. R. Jeevarathinam #1 , Dr. Antony Selvadoss Thanamani 2 

# Department of Computer Science 

SNR Sons College, Coimbatore, Tamilnadu, India. 

i 

Associate Professor and Head 

Department of Computer Science 

NGM College, Pollachi, Tamilnadu, India. 



Abstract — Software testing is the important phase of 
software development process. But, this phase can be easily 
missed by software developers because of their limited time to 
complete the project. Since, software developers finish their 
software nearer to the delivery time; they don't get enough time 
to test their program by creating effective test cases. . One of the 
major difficulties in software testing is the generation of test 
cases that satisfy the given adequacy criterion Moreover, 
creating manual test cases is a tedious work for software 
developers in the final rush hours. A new approach which 
generates test cases can help the software developers to create 
test cases from software specifications in early stage of software 
development (before coding) and as well as from program 
execution traces from after software development (after coding). 
Heuristic techniques can be applied for creating quality test cases. 
Mutation testing is a technique for testing software units that has 
great potential for improving the quality of testing, and to assure 
the high reliability of software. In this paper, a mutation testing 
based test cases generation technique has been proposed to 
generate test cases from program execution trace, so that the test 
cases can be generated after coding. The paper details about the 
mutation testing implementation to generate test cases. The 
proposed algorithm has been demonstrated for an example. 

I. Introduction 

Software testing is a standard method of assuring 
software quality. Software testing is an important activity to 
assure the quality of software. Unfortunately, software testing 
is very labor intensive and very expensive. It can take about 
50 percents of total cost in software developing process [1]. 
The software testers may need to spend a longer time using 
many test cases if the test data used are not of high quality. 
Therefore, a performance of executing test case is an 
important issue to reduce the testing time. Software testing is 
usually the first part of software development stages, which 
software developers decide to omit when there is a limited 
time to deliver the software. In other word, developers may 
not have enough time after they finished their coding to create 
test cases to test their code. Generating test cases can resolve 
these problems. This not only helps developers to test their 
program when they finish coding but also controls the 
developers to program the software as defined in the software 
specification [2]. In this case the software specifications are 



the main sources for generating test cases as these documents 
describe the software system to be developed in detail. 
One of the most difficult and expensive parts of applying 
these techniques has been the actual generation of test data- 
which has traditionally been done by hand 

The general aim of the research reflected in this 
paper is to formalize, and mechanize where possible, routine 
aspects of testing. Such formalization has two benefits. First, 
it makes it easier to analyse a given test set to ensure that it 
satisfies a specified coverage criterion. Second, it frees the test 
engineer to concentrate on less formalization, and often more 
interesting tests. Developers have responded to this need in 
many ways, including improving the process, increasing the 
attention on early development activities, and using formal 
methods to describe requirements, specifications, and designs. 
Although all of these improvements help create software that 
is of higher quality and higher reliability, the software still 
needs to be tested, and the more stringent needs for the 
product also means that the testing method must be more 
effective at finding problems in the software. Project and test 
managers are more than ever in a position where they need 
solid information for how to apply scarce resources. Applying 
structured, precisely defined testing techniques allows 
development resources to be used more wisely. 

Specification-based testing refers to creating test 
inputs from the software specifications. Specification-based 
testing allows tests to be created earlier in the development 
process, and be ready for execution before the program is 
finished. Additionally, when the tests are generated, the test 
engineer will often find inconsistencies and ambiguities in the 
specifications, which allows problems to be found and 
eliminated early. Specifications can be used as a basis for 
output checking, which significantly reduces one of the major 
costs of testing. Another advantage is that the essential part of 
the test data can be independent of any particular 
implementation of the specifications. Specification-based 
testing is also important for conformance testing, where 
access to the code is not provided, but specifications for the 
product are. 

There is an increasing need for effective testing of 
software for growing applications, such as web applications 



190 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



and e-commerce require software that exhibits more reliability 
than most traditional application areas. Without software that 
functions reliably, businesses that operate on the web will lose 
money, sales, and customers. In recent years, the phrase 
"fault-based testing" has been applied to techniques that 
choose test data that attempt to show the presence (or absence) 
of specific faults during unit testing. Techniques have been 
developed that determine whether test data can detect specific 
faults (i.e., mutation analysis [3]), and the theoretical 
properties of fault-based testing have been explored [5, 4]. 

The mutation testing is a fault based testing strategy 
that measures the quality of testing by examining whether the 
test set, test input data used in testing can reveal certain type 
of faults. Mutation testing helps testers create test data by 
interacting with them to strengthen the quality of the test data. 
Faults are introduced into programs by creating many versions 
of the software, each containing one fault. Test cases are used 
to execute these faulty programs with the goal of causing each 
faulty program to produce incorrect output (fail). Hence the 
term mutation; faulty programs are mutants of the original, 
and a mutant is killed when it fails. When this happens, the 
mutant is considered dead and no longer needs to remain in 
the testing process because the faults represented by that 
mutant have been detected. 

II. RELATED WORK 

If testers want to test functional requirements, they 
may use black-box testing technique. Black-box testing [6] 
does not need knowledge of how software is programmed. 
Test oracles are specified by software design or software 
specifications. Testers inject test data to execute program, then 
compare actual result with the specified test oracle. By 
contrast, white-box testing needs knowledge of how software 
is programmed. In white-box testing, paths or statements 
which has been executed are test oracle. These are called 
coverage criteria. There are three main types of coverage 
criteria: statement, coverage, branch coverage, and path 
coverage. Statement coverage reports whether each statement 
is encountered by the test suite or not. Branch coverage 
reports whether every branch structure (if - else clause or 
while clause) has been executed for true and false condition in 
each branch. Finally, path coverage reports whether all 
possible paths in function has been tested. 

In Object-oriented context, the structure of software 
is more complicated than the structural one. Conventional test 
approaches may not be enough for testing. The combination of 
those two traditional approaches is called Gray-box testing [7]. 
In Gray-box testing, test data generates based on the high 
level design which specifies the expected structure and 
behaviour of system. Gray-box testing investigates the 
coverage criteria of white-box method and finds all possible 
coverage paths. In addition, the generated test case should be 
satisfied with functional requirement as in the black-box 
testing criteria. 

Many automated test case generation techniques 
produce test cases based on Gray-box method. Not only does 



Gray-box testing concern functional requirement as black box 
testing, but also concerns on behaviours of system. Clarke [8] 
proposed an empirical study which compared efforts between 
automate test generation and manual test generation. In his 
report test data was generated from extended finite state 
model (EFSM). The research shows that the automate test data 
generation could reduce an effort from manual test data 
generation for more than 88 percents. Xu and Yang [9] 
proposed test data generation framework called JMLAutoTest 
framework. JMLAutoTest framework generates test data from 
Java Modelling Language (JML) [[10] [11]]. JML is a 
notation for specifying behaviour and interface in Java class 
and method. Since JML is a formal specification, developers 
should spend efforts to understand JML before writing 
specification. Because UML diagrams are now widely used 
for software development [12], generating test data from 
UML diagrams should help developer to reduce a great 
number of efforts. 

Wang, et.al [13] proposed test data generation from 
activity diagram. They extracted a test scenario from activity 
diagram. The test scenario is a sequence of possible paths in 
activity diagram. From these paths, the executing sequence of 
program has been generated in order to cover all possible 
paths. However, activity diagram describes flows of system, 
not the behaviour of the system. Due to performance of 
generating test data and a concern of size of test data set, 
heuristic techniques are applied for test data generation. 
GADGET [14] and TGEN [15] use genetic algorithm to 
improve quality of generating test data. GADGET generates 
test data from a control flow graph generated from source 
code. A fitness function is defined for each condition node in 
control flow graph. An empirical study showed that test data 
generated by GADGET covers more than 93 percents of 
source code, while random testing achieves around 55 
percents. TGEN transforms a control flow graph to a control 
dependency graph (CDG). Each part of CDG represents the 
smallest set of predicate to traverse every node in control flow 
graph. Both GADGET and TGEN generate test data using 
white box method; therefore, test data can be generated only 
after software is finished. Using Genetic algorithm to generate 
test data from software model is proposed in [16]. JML is a 
model for generating test data. Fitness function is calculated 
by coverage of paths and post condition defined by JML. 

Because of the large number of mutant programs that 
must be generated and run, early designers of mutation 
analysis systems considered individually creating, compiling, 
linking, and running each mutant more difficult, and slower, 
than using an interpretive system [[17] [18]] . It was 
considered likely that the cost of compiling large numbers of 
mutants would be prohibitive. Of the interpreter-based 
systems that have been developed, Mothra is the most recent 
and comprehensive [[19] [21]]. In these conventional, 
interpreter-based mutation analysis systems, the source code is 
translated into an internal form suitable for interpretive 
execution and mutation. For each mutant, a mutant generator 
program produces a "patch" that, when applied to the internal 
form, creates the desired alternate program. The translated 



191 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



program plus the collection of patches represents a program 
neighbourhood. To run a mutant against a test case, the 
interpreter dynamically applies the appropriate patch and 
interpretively executes the resulting alternate internal form 
program. A number of attempts to overcome the performance 
problem have been made. Some approaches attempt to limit 
the number of mutants that must be run. In selective mutation 
[26], only a subset of the possible mutagens is used, resulting 
in fewer mutants being created. Preliminary results suggest 
that selective mutation may provide almost the same test 
coverage as non-selective mutation under certain conditions. 
Running only a sample of the mutants [27] has also been 
suggested. 

In extreme cases, however, it is necessary to run 
almost all the mutants. In other approaches, the use of non- 
standard computer architectures has been explored. 
Unfortunately, full utilization of these high performance 
computers requires an awareness of their special requirements 
as well as adaptation of software. Work has been done to 
adapt mutation systems to vector processors [23], to SIMD 
[22] and hypercube (MIMD) machines [[24] [25]]. However, 
it is the very fact that these architectures are non-standard that 
limits the appeal of these approaches. Not only are they not 
available in most development environments, but testing 
software designed for one operational environment (machine, 
operating system, compiler, etc.) on another is fraught with 
risks. 

The approaches above do not squarely address the 
primary factor that causes conventional systems to be slow: 
interpretative execution. As noted previously, the overhead of 
compiling many mutant programs outweighs the benefit of 
increased execution speed. Compiler-integrated [20] program 
mutation seeks to avoid excessive compilation overhead and 
yet retain the benefit of compiled speed execution. In this 
method, the program under test is compiled by a special 
compiler. As the compilation process proceeds, the effects of 
mutations are noted and code patches that represent these 
mutations are prepared. Execution of a particular mutant 
requires only that the appropriate code patch be applied prior 
to execution. Patching is inexpensive and the mutant executes 
at compiled- speeds. Unfortunately, crafting the needed special 
compiler is an expensive undertaking. Modifying an existing 
compiler reduces this burden somewhat, but the task is still 
technically demanding. Moreover, for each new computer and 
operating system environment, this task must be repeated. 

III. FAULT CLASSIFICATION 

A test case that distinguishes the program from its 
mutant is considered to be effective at finding faults in the 
program. The effectiveness of mutation testing, like other 
fault-based approaches, depends heavily on the types of faults 
that the mutation system is designed to represent. Since 
mutation testing uses mutation operators to implement faults, 
the quality of the mutation operators is crucial to the 
effectiveness of mutation testing. Although mutation testing 
has a rich history, most mutation operators have been 
developed for procedural programs. 00 languages contain 



new features such as encapsulation, inheritance, and 
polymorphism. These features introduce the potential for new 
faults. Therefore, existing mutation operators for procedural 
programming languages are not sufficient for programs 
written in 00 languages and new 00- specific language 
operators are needed. 

The effectiveness of mutation testing depends 
heavily on the types of faults that may be represented. In these 
new kinds of faults, some of which are not modelled by 
traditional mutation operators. Which are insufficient to test 
these 00 language features, particularly at the class testing 
level. This paper introduces a new set of class mutation 
operators for the 00 languages. These operators are based on 
specific 00 faults and can be used to detect faults involving 
inheritance, polymorphism, and dynamic binding, thus are 
useful for inter-class testing. The faults modelled by these 
operators are not general; they can be application- specific or 
programmer- specific. Therefore, to execute mutation testing 
with these operators, they should be selected based on the 
characteristic of the program to be tested. The previous 
attempts suffered from not having a general fault model. The 
previous 00 mutation operators do not handle several fault 
types and did not handle all 00 features. Faults can be 
classified as occurring at the intra-method level, inter-method 
level, intra-class level, and inter-class level. 

Intra-method level faults occur when the 
functionality of a method is implemented incorrectly. A 
method in a class corresponds to the unit of the conventional 
program testing. Inter-method and intra-class level faults are 
made at the interactions between pairs of methods of a single 
class or between pairs of methods that are not part of a class 
construct in non-00 languages. Because methods are getting 
smaller and interactions among methods are increasingly 
encoding the design complexity. Inter-class level faults 
include faults that occur due to the object-oriented specific 
features such as encapsulation, inheritance, polymorphism, 
and dynamic binding. 

IV. MUTATION OPERATORS 

There are three kinds of mutation operators available 
namely statement level operators, method level operators and 
class level operators. 

Statement level mutation operators involve the 
creation of a set of mutant programs of the program being 
tested. Each mutant differs from the original program by one 
mutation. A mutation is a single syntactic change that is made 
to a program statement. 

Operand Replacement Operators (ORO) - Replacing a single 
operand with another operand or constant. 
Expression Modification Operators (EMO) - Replacing an 
operator or inserting a new operator. 

Statement Modification Operators (SMO) - Replacing or 
deleting a statement or part of the statement. 

Method level mutation operators are used in unit and 
integration level testing and can be classified into two levels: 
(1) intra-method, (2) inter-method. This classification follows 



192 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



definitions by Harrold and Rothermel [25] [27] and Gallagher 
and Offutt [26] [23]. 

Intra-method level faults occur when the 
functionality of a method is implemented incorrectly. Testing 
within classes corresponds to unit testing in conventional 
programs. So far, researchers have assumed that traditional 
mutation operators for procedural programs will suffice for 
this level (with minor modifications to adapt to new 
languages). 

Inter-method level faults are made on the 
connections between pairs of methods of a single class. 
Testing at this level is equivalent to integration testing of 
procedures in procedural language programs. Interface 
mutation which evaluates how well the interactions between 
various units have been tested, is applicable to this level. 

Class level mutation operators can be classified into 
two levels: (1) intra-method, (2) inter-method. 

Intra-class testing is when tests are constructed for a 
single class, with the purpose of testing the class as a whole. 
Intra-class testing is a specialization of the traditional unit and 
module testing. It tests the interactions of public methods of 
the class when they are called in various sequences. Tests are 
usually sequences of calls to methods within the class, and 
include thorough tests of public interfaces to the class. 

Inter-class testing is when more than one class is 
tested in combination to look for faults in how they are 
integrated. Inter-class testing specializes the traditional 
integration testing and seldom used subsystem testing, where 
most faults related to polymorphism, inheritance, and access 
are found. 

Based on the fault classification, Ma et al. [28] 
developed a comprehensive set of class mutation operators for 
Java. There are 24 mutation operators explained below. Each 
mutation operator is related to one of the following six 
language feature groups. The first four groups are based on 
language features that are common to all object oriented 
languages. The fifth group includes language features that are 
Java- specific, and the last group of mutation operators are 
based on common object oriented programming mistakes. As 
is usual with mutation operators, they are only applied in 
situations where the mutated program will still compile. 

A. Information Hiding 

Access control is one of the common sources of mistakes 
among object oriented programmers. The semantics of the 
various access levels are often poorly understood, and access 
for variables and methods is not always considered during 
design. Poor access definitions do not always cause faults 
initially, but can lead to faulty behaviour when the class is 
integrated with other classes, modified, or inherited from. The 
Access Control mutation operator, Access modifier change 
(AMC) has been developed for this category. 

B. Inheritance 

Although inheritance is a powerful and useful abstraction 
mechanism, incorrect use can lead to a number of faults. 
Seven mutation operators have been defined to test the various 
aspects of using inheritance, covering variable hiding, method 



overriding, the use of super, and definition of constructors and 

are listed below. 

IHD-Hiding variable deletion IHI-Hiding variable 

insertion 

IOD-Overriding method deletion IOP- Overriding method 

calling position change 

IOR- Overriding methods rename ISK-Super keyword 

deletion 

IPC-Explicit call of a parent's constructor deletion 

C. Polymorphism 

Polymorphism and dynamic binding allow object 
references to take on different types in different executions 
and at different times in the same execution. That is, object 
references may refer to objects whose actual types differ from 
their declared types. In most languages (including Java and 
C++), the actual type can be any type that is a subclass of the 
declared type. Polymorphism allows the behaviour of an 
object reference to differ depending on the actual type. Four 
operators have been developed for this category. 
PNC- new method call with child class type 
PMD- Instance variable declaration with parent class type 
PPD -Parameter variable declaration with child class type 
PRV- Reference assignment with other comparable type 

D. Overloading 

Method overloading allows two or more methods of the 
same class or type family to have the same name as long as 
they have different argument signatures. Just as with method 
overriding (polymorphism), it is important for testers to 
ensure that a method invocation invokes the correct method 
with appropriate parameters. Four mutation operators have 
been defined to test various aspects of method overloading. 
OMR- Overloading method contents change 
OMD- Overloading method deletion 
OAO- Argument order change 
OAN- Argument number change 

E. Java Specific Features 

Because mutation testing is language dependent, mutation 
operators need to reflect language- specific features. Java has a 
few object-oriented language features that do not occur in all 
object oriented languages and four operators have been 
defined to ensure correct use of these features. They cover use 
of this, static, default constructors and initialization. 
JTD- this keyword deletion 
JSC- static modifier change 
JID- Member variable initialization deletion 
JDC -Java- supported default constructor creation 

F. Common Programming Mistakes 

This category attempts to capture typical mistakes that 
programmers make when writing object oriented software. 
These are related to use of references and using methods to 
access instance variables. Four operators have been developed 
for this category. 

EOA- Reference assignment and content assignment 
replacement 



193 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



EOC- Reference comparison and content comparison 

replacement 

EAM- Accessor method change 

EMM- Modifier method change 

V. RELATIONSHIP BETWEEN FAULTS AND 
OPERATORS 

The following table relates the fault types and our 
mutation operators. All faults are covered, and some required 
multiple mutation operators. Conversely, some of the 
mutation operators cover more than one fault 
Faults Class Mutation Operators 



State visibility anomaly 
State definition inconsistency 
(due to state variable hiding) 
State definition anomaly 



IOP 



IHD, IHI 



(due to overriding) 




IOD 


Indirect inconsistent state definitior 


L 


IOD 


Anomalous construction behaviour 




IOR, IPC, P> 


Incomplete construction 




JID, JDC 


Inconsistent type use 


PID, PNC, PPD, PRV 


Overloading methods misuse 


OMD 


, OAO, OAN 


Access modifier misuse 




AMC 


Static modifier misuse 




JSC 


Incorrect overloading- 






methods implementation 




OMR 


Super keyword misuse 




ISK 


This keyword misuse 




JTD 


Faults from common programming 






mistakes 


EOA, 


EOC, EAM, EN 



VI. EFFECTIVENESS OF MUTATION OPERATORS 

TCAS/Siemens has an internal state which is large relative 
to the number of inputs and outputs. TCAS, aircraft collision 
avoidance, is a part of a set of C programs that came 
originally from Siemens Corporate Research and was 
subsequently modified by Rothermel and Harrold [26]. These 
programs are used in research on program testing, so they 
come with extensive test suites and sets of faulty versions. 
There are 12 input variables specifying parameters of own 
aircraft and another aircraft and one output variable, alt_sep, a 
resolution advisory to maintain safe altitude separation 
between the two aircrafts. The program computes intermediate 
values and prints alt_sep to the standard output. The program 
has minimal documentation, and we wrote a formal 
specification for it. The following table gives the results of 
various mutation operators in terms of number of mutants 
generated, number of traces produced and the percentage of 
fault coverage. 

There are several issues that need to be considered to 
evaluate the usefulness and effectiveness of the 00 class level 
mutation operators. First is the issue of equivalent mutants. 
Equivalent mutants do not affect the semantics of the program; 
therefore they are useless for mutation testing. Second, some 
operators can generate mutants that are easily killed. Although 
mutants make simple syntactic changes to the program, their 
semantic impacts can vary greatly. The impacts of 00 



operators on the semantics can vary from affecting the method 
to the semantics of the entire class. For example, the IOD 
operator swaps overriding methods with its parent's. The 
effect of IHD, on the other hand, extends over the whole class 
because it handles instance variable, which determine the 
class state. It is possible that some of these mutation operators 
will create mutants that are too easily killed. Finally, the 
mutation operators need to be evaluated in terms of their 
effectiveness of detecting faults in 00 programs. The AMC 
and JSC operators produced a lot of equivalent mutants, and 
the PNC, PMD, PPD and IHI operators produced equivalent 
mutants when overriding was present. 

TABLE I 

Result of mutation operators and theiir fault coverage 



Operator 


No.of 
Mutants 


No.of 
Traces 


Coverage 


AMC 


202 


24 


96.6% 


IOD 


72 


21 


87.9% 


ISK 


130 


21 


93.1% 


IHD 


116 


14 


62.9% 


PNC 


74 


18 


94.2% 


PPD 


72 


21 


90.7% 


PMD 


144 


29 


83.7% 


JTD 


12 


4 


52.4% 


JSC 


83 


17 


85.2% 


IHI 


97 


22 


76.4% 



VII. 



Conclusion 



This paper presents a comprehensive set of mutation 
operators to test for faults in the use of object-oriented 
features. These mutation operators are based on an exhaustive 
list of 00 faults, which gives them a firm theoretical basis. As 
a result, they correct several problems. These mutation 
operators are designed with an emphasis on the integration 
aspects of Java to support interclass level testing, and will 
help testers find faults with the use of language features such 
as access control, inheritance, polymorphism and overloading. 
Thus, this provides a way to improve the reliability of 00 
software. 

References 

[1] Myers, G., The Art of Software Testing. 2 ed. 2004: John Wiley & Son. 

Inc. 234 

[2] Beck, K., Test-Driven Development by Example. 2003: Addison- Wesley. 

220. 

[3] R. A. DeMillo, R. J. Lipton, and F. G. Sayward. Hints on test data 

selection: Help for the practicing programmer. IEEE Computer, 11(4):34{41, 

April 1978. 

[4] L. J. Morell. A Theory of Error-Based Testing. PhD thesis, University of 

Maryland, College Park MD, 1984. Technical Report TR-1395. Rel-work- 

mutation testing 



194 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



[5] T. A. Budd and D. Angluin. Two notions of correctness and their relation 
to testing. Acta Informatica, 18(1):31{45, November 1982. 
[6] Beizer, B., Black-box testing : techniques for functional testing of 
software and systems. 1995: John Wiley & son Inc. 294. 
[7] Hung, N.Q., Testing Application on the Web. 2003: John Wiley & Sons. 
[8] Clark, J.M. Automated Test Generation from a Behavioral Model. In the 
11th International Software Quality Week (QW98). 1998. 
[9] Xu, G. and Z. Yang. JMLAutoTest: A Novel Automated Testing 
Framework Based on JML and JUnit. in Lecture Notes in Computer Science. 
2004. 

[10] Burdy, L., et al. An overview of JML tools and applications, in Eighth 
International Workshop on Formal Methods for Industrial Critical Systems 
(FMICS '03), ser. Electronic Notes in Theoretical Computer Science. 2003. 
Elsevier. 

[11] Leavens, G.T., et al., JML Reference Manual. 2005. 
[12] Lange, C.F.J., M.R.V. Chaudron, and J. Muskens, In practice: UML 
software architecture and design description. Software, IEEE, 2006. 23(2): p. 
40-46. 

[13] Wang, L., et al. Generating test cases from UML activity diagram based 
on Gray-box method, in Software Engineering Conference, 2004. 11th Asia- 
Pacific 2004. 

[14] Michael, C, G. McGraw, and M.A. Schatz, Generating software test data 
by evolution. Software Engineering, IEEE Transactions on, 2001. 27(12): p. 
1085-1110. 

[15] Pargas, R., M. Harrold, and R. Peck, Test-data generation using genetic 
algorithms. Software Testing, Verification and Reliability, 1999. 9(4): p. 263- 
282. 

[16] Cheon, Y., M.Y. Kim, and A. Perumandla. A Complete Automation of 
Unit Testing for Java Programs, in Proceedings of the 2005 International 
Conference on Software Engineering Research and Practice (SERP '05). 2005. 
Las Vegas, Nevada, USA,. 

[17] Timothy A. Budd. Private correspondence, February 24 1992 
[18] Timothy A. Budd, Richard J. Lipton, Frederick G. Sayward, and Richard 
A. DeMillo. The Design of a Prototype Mutation System for Program Testing. 
In Proceedings of the National Computer Conference, pages 623-627,. 
[19] Richard A. DeMillo, Dany S. Guindi, Kim N. King, W. Michael 
McCracken, and A. Jefferson Offutt. An Extended Overview of the Mothra 
Software Testing Environment. In Proceedings of the Second Workshop on 
Software Testing, Veri_cation, and Analysis, pages 142-151, Ban_, Alberta, 
Canada, July 19{21 1988. IEEE Computer Society Press. 
[20] Richard A. DeMillo, Edward W. Krauser, and Aditya P. Mathur. 
Compiler-Integrated Program Mutation. In Proceedings of the Fifteenth 
Annual International Computer Software and Applications Conference 
(COMPSAC), pages 351(356, Tokyo, Japan, September 11{13 1991. IEEE 
Computer Society Press 

[21] Kim N. King and A. Jefferson Offutt. A Fortran Language System for 
Mutation-based Software Testing. Software-Practice and Experience, 
21(7):685{718, July 1991. 

[22] Edward W. Krauser, Aditya P. Mathur, and Vernon J. Rego. High 
Performance Software Testing on SIMD Machines. IEEE Transactions on 
Software Engineering, SE-17(5):403- 423, May 1991. 

[23] Aditya P. Mathur and Edward W. Krauser. Mutant Unification for 
Improved Vectorization. Technical Report SERC-TR-14-P, Software 



Engineering Research Center, Purdue University, West Lafayette, IN, April 

25 1988. 

[24] ByoungJu Choi and Aditya P. Mathur. High Performance Mutation 

Testing. The Journal of Systems and Software, 20(2):135{152, February 1993. 

[25] A. Jefferson Offutt, Roy P. Pargas, Scott V. Fichter, and Prashant K. 

Khambekar. Mutation Testing of Software Using a MIMD Computer. In 

Proceedings of the 1992 International Conference on Parallel Processing, 

pages II{257{266, St. Charles, IL, August 17{21 1992. 

[26] A. Jefferson Offutt, Gregg Rothermel, and Christian Zapf. An 

Experimental Evaluation of Selective Mutation. In Proceedings of the 

Fifteenth International Conference on Software Engineering, Baltimore, MD, 

May 17{21 1993. IEEE Computer Society Press. 

[27] Mehmet Spahinoglu and Eugene H. Spafford. A Sequential Statistical 

Procedure in Mutation-Based Testing. In Proceedings of the 28th Annual 

Spring Reliability Seminar, pages 127{148, Boston, MA, April 19 1990. 

Central New England Council of IEEE. 

[28] Y. S. Ma, Y. R. Kwon, and J. Offutt. Inter-class mutation operators for 

Java. In IEEE Computer Society Press, editor, 13th International Symposium 

on Software Reliability Engineering, pages 352-363, Annapolis MD, 

November 2002. 

AUTHORS PROFILE 



Mrs R. Jeevarathinam graduated with MCA in 
2001 from Bharathiar University, India and 
completed M.Phil from Bharathidasan University, 
India during 2003-04. Her areas of Interest include 
Software Engineering & Data Mining. She has 
about 8 years of teaching experience. Currently 
she is working as a Sr. Lecturer in CS department 
at SNR Sons College, Coimbatore, India and also 
pursuing PhD of Mother Teresa Women University, India. She has published 
a number of papers in various national & international journals & conferences. 
She is an active IEEE student member 



Dr.Antony Selvadoss Thanamani is presently 
working as Reader in the Dept of Computer 
Science, NGM College, India. He has published 
more than twenty papers in national/journals and 
more than ten books. His areas of interest includes 
E-Learning, Software Engineering, Data Mining, 
Networking and etc. He has about 20 years of 
teaching experience. He is guiding many research scholars and has published 
many papers in national and international conference and in many 
international journals 




195 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



An Energy Efficient and Reliable Congestion Control 
Protocol For Multicasting In Mobile Adhoc Networks 



Dr.G.Sasi Bhushana Rao 

Senior Professor 

Department of Electronics and Communication Engineering 

Andhra University 

Visakhapatnam 



M.RajanBabu 

Associate Professor 

Department of Electronics and Communication Engineering 

Lendi Institute of Engineering and Technology 

Jonnada, Vizianagaram, AndhraPradesh, India 



Abstract — This paper presents an energy efficient and reliable 
congestion control protocol for multicasting in mobile adhoc 
networks (MANETs). Our proposed scheme overcomes the 
disadvantages of existing multicast congestion control protocols 
which depend on individual receivers to detect congestion and 
adjust their receiving rates. In the first phase of our protocol, we 
build a multicast tree routed at the source, by including the nodes 
with higher residual energy towards the receivers. In the second 
phase, we propose an admission control scheme in which a 
multicast flow is admitted or rejected depending upon on the 
output queue size. In the third phase, we propose a scheme which 
adjusts the multicast traffic rate at each bottleneck of a multicast 
tree. Because of the on-the-spot information collection and rate 
control, this scheme has very limited control traffic overhead and 
delay. Moreover, the proposed scheme does not impose any 
significant changes on the queuing, scheduling or forwarding 
policies of existing networks. Simulation results shows that our 
proposed protocol has better delivery ratio and throughput with 
less delay and energy consumption when compared with existing 
protocol. 

Keywords-Congestion Control; Mobile Adhoc Networks; 
Multicasting; admission control; multicast tree. 

I. Introduction 

A mobile ad-hoc network (MANET) is composed of 
mobile nodes without any infrastructure. Mobile nodes self- 
organize to form a network over radio links. The goal of 
MANETs is to extend mobility into the realm of autonomous, 
mobile and wireless domains, where a set of nodes form the 
network routing infrastructure in an ad-hoc fashion. The 
majority of applications of MANETs are in areas where rapid 
deployment and dynamic reconfiguration are necessary and 
wired network is not available. These include military 
battlefields, emergency search, rescue sites, classrooms and 
conventions, where participants share information dynamically 
using their mobile devices. These applications lend themselves 
well to multicast operations [1]. 

Multicasting is aimed to deliver data to a set of selected 
receivers. There is no restriction on the location or number of 
members in a host group. Multicast can be classified into one 
to many or many to many communication applications. The 
important member identifications and functions are: group 
member, sources, destination, forwarding nodes, non-group 



member. The group membership is dynamic means that hosts 
may join and leave groups at any time Multicast packets are 
delivered to each member of a multicast group with the same 
best-efforts reliability and performance as unicast packets to 
members. Multicast groups may be of arbitrary size, may 
change membership dynamically, and may have either a 
global or local scope. The senders do not need to know 
membership groups, and needs not to be a member of that 
group. [2]. In addition, within a wireless medium, it is crucial 
to reduce the transmission overhead and power consumption. 
Multicasting can improve the efficiency of the wireless link 
when sending multiple copies of messages by exploiting the 
inherent broadcast property of wireless transmission. Hence, 
reliable multicast routing plays a significant role in MANETs 
[1]. 

Multicasting can be used to improve the efficiency of the 
wireless link when sending multiple copies of messages to 
exploit the inherent broadcast nature of wireless transmission. 
So multicast plays an important role in MANETs Unlike 
typical wired multicast routing protocols, multicast routing for 
MANETs must address a diverse range of issues due to the 
characteristics of MANETs, such as low bandwidth, mobility 
and low power. MANETs deliver lower bandwidth than wired 
networks; therefore, the information collection during the 
formation of a routing table is expensive [1]. 

A. Multicast Issues in MANET 

Scalability: A multicast routing protocol is scalable with 
respect to some constraints posed by MANETs. 

Multicast service support: The multicast protocol defines 
conditions for joining/leaving groups, multicast participants 
should be able to join or leave groups at will. On the other 
hand, service providers can be convinced to support multicast 
protocols. 

Traffic control: Both source and core-based approaches 
concentrate traffic on a single node. In stateless multicast 
group membership is controlled by the source, which leads to 
the vulnerability of multicast protocols for MANETs. Still 
need to be investigated is how to efficiently distribute traffic 
from a central node to other member nodes for MANETs. 



196 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



QoS: QoS defines a guarantee given by the network to 
satisfy a set of predetermined service performance constraints 
for the user in terms of end-to-end delay, jitter, and available 
bandwidth. Therefore, multicast routing protocols must be 
feasible for all kinds of constrained multicast applications to 
run well in a MANET. However, it is a significant technical 
challenge to define a comprehensive framework for QoS 
support, due to dynamic topology, distributed management 
and multi-hop connections for MANET s. 

Multiple sources: Most of the existing multicast routing 
protocols in ad-hoc networks are designed for single source 
multicasting. However, a multicast group may contain 
multiple sources due to different kinds of services or 
applications simultaneously provided by the networks. Each 
single source multicast routing protocol induces a lot of 
overhead and thus wastes tremendous network resources in a 
multi-source multicast environment. 

The QAMNet [9] depends on the traffic pattern hence it is 
difficult to accurately estimate the threshold rate. The 
protocols which support QoS for multicasting introduce 
network state and additional signaling. Such additional 
signaling packets for reservation protocol must be avoided as 
this adds to network congestion, especially in high mobility 
scenarios. 

The QMR and E-QMR protocols calculate approximately 
the available bandwidth based on the channel status. This 
results in some problem. Each node can listen to the channel to 
determine the channel status and computes the idle duration 
only for a period of time [10]. A lantern- tree topology is used 
to provide QoS multicast routing. Need for a centralized MAC 
scheme in ad hoc mobile networks with dynamic wireless 
environments is its main disadvantage [12]. 

B. Proposed Solution 

In this paper, we propose to design an energy efficient and 
reliable congestion control (EERCCP) protocol for 
multicasting with the following phases. 

In its first phase, it builds a multicast tree routed at the 
source, by including the nodes with higher residual energy 
towards the receivers. Most of the existing schemes depend 
on individual receivers to detect congestion and adjust their 
receiving rates which are much disadvantageous. In the second 
phase, we propose an admission control scheme in which a 
multicast flow is admitted or rejected depending upon on the 
output queue size. 

In the third phase, we propose a scheme which adjusts the 
multicast traffic rate at each bottleneck of a multicast tree. 

II. Related Work 

Hua Chen, Baolin Sun [3] introduces an Entropy-based 
Fuzzy controllers QoS Routing algorithm in MANET 
(EFQRM). The key idea of EFQRM algorithm is to construct 
the new metric-entropy and fuzzy controllers with the help of 
entropy metric to reduce the number of route reconstruction so 
as to provide QoS guarantee in the ad hoc network. 

Tolga Numanoglu and Wendi Heinzelman [4] propose a 
mesh networking inspired approach that adapts the amount of 



redundancy according to the current link conditions. They 
simultaneously reduce unnecessary energy dissipation. 

D. Agrawal, T. Bheemarjuna Reddy, and C. Siva Ram 
Murthy [5] propose a Robust Demand-driven Video Multicast 
Routing (RDVMR) protocol. Their protocol uses a novel path 
based Steiner tree heuristic to reduce the number of forwarders 
in each tree. They construct multiple trees in parallel with 
reduced number of common nodes among them. Moreover, 
unlike other on-demand multicast protocols, RDVMR 
specifically attempts to reduce the periodic (non on-demand) 
control traffic. 

Guojun Wang, Jiannong Cao, Lifan Zhang, Keith C. C. 
Chan [6] proposes a logical Hypercube-based Virtual 
Dynamic Backbone (HVDB) model for QoS -aware multicast 
communications. In this model, high fault tolerance and small 
diameter of hyper cubes are the basis for high availability, and 
regularity and symmetry of hyper cubes contribute to good 
load balancing. 

Vida Lashkari B. O, Mehdi Dehghan [7] proposes an 
efficient algorithm named is proposed to improve the route 
discovery mechanism in MAODV for QoS multicast routes. 
QoS-MAODV especially can establish a multicast tree with 
the minimum required bandwidth support and decrease the 
end-to-end delay between each destination and the source 
node. It can establish QoS routes with the reserved bandwidth 
on per chosen flow. To perform accurate resource reservation, 
they have developed a method for estimating the consumed 
bandwidth in multicast trees by extending the methods 
proposed for unicast routing. 

Zeyad M. Alfawaer, GuiWei Hua, and Noraziah Ahmed 
[8] introduced MANHSI (Multicast for Ad hoe Network with 
hybrid Swarm Intelligence) protocol, which relies on a swarm 
intelligence based optimization technique to learn and 
discover efficient multicast connectivity. The proposed 
protocol instances that it can quickly and efficiently establish 
initial multicast connectivity and/or improved the resulting 
connectivity via different optimization techniques. 

Harald Tebbe and Andreas J. Kassler [9] present QAMNet, 
an approach to improve the Quality of Service (QoS) for 
multicast communication in MANETs. They extend existing 
approaches of mesh based multicasting by introducing traffic 
prioritization, distributed resource probing and admission 
control mechanisms, adaptive rate control of non-real-time 
traffic based on Medium Access Control (MAC) layer 
feedback so as to maintain low delay and required throughput 
for real-time multicast flows. 

Mohammed Saghir, Tat-Chee Wan, Rahmat Budiarto [10] 
has extended QMR to make it more effective than the previous 
work. They propose a cross-layer framework to support QoS 
multicasting. They have enhanced the IEEE 802.11 MAC 
layer to estimate the available bandwidth at each node. 

Ravindra Vaishampayan, J.J. Garcia-Luna-Aceves [11] 
proposed a protocol for unified multicasting through 
announcements (PUMA) in ad-hoc networks, which 
establishes and maintains a shared mesh for each multicast 
group, without requiring a unicast routing protocol or the pre 
assignment of cores to groups. PUMA achieves a high data 



197 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



delivery ratio with very limited control overhead, which is 
almost constant for a wide range of network conditions. 

N. Ben Ali, A. Belghith, J. Moulierac, M. Molnar [13] has 
proposed a new algorithm coined mQMA. This deals with two 
main problems of traditional IP multicast, which are multicast 
forwarding state scalability and multi-constrained QoS 
routing. The algorithm mQMA is a QoS multicast aggregation 
algorithm which handles multiple additive QoS constraints. It 
builds few trees and maintains few forwarding states for the 
groups. The multicast tree aggregation technique, allows 
several groups to share the same delivery tree. The mQMA 
algorithm builds trees satisfying multiple additive QoS 
constraints. 

III. Energy Efficient And Reliable Congestion 
Control Protocol 

A. Energy Efficient Tree Construction 

In our energy efficient and reliable congestion control 
protocol we build a multicast tree routed at the source towards 
the receivers. The distance i.e. the geographical location of the 
nodes is assumed. Their residual energy is measured. The 
nodes are sorted based on its location from the source and 
arranged in a sequence order. A threshold value Q is set and 

the nodes which are less than Q(n < Q) are unicast from the 

source and the nodes which are greater than Q(n > Q) are 

multicast. In case of multicasting the node which has the 
minimum energy per corresponding receiver is set as the relay 
node. The relay node then forwards the packets from the 
source to the corresponding receivers. 

1 ) Calculating Residual Energy of a Node: Consider a 
network with multicast groups G\,G2,-,G X . Each group 
{G/} consists of N nodes. Every node in the MANET 
calculates its remaining energy periodically. The nodes may 
operate in either transmission or reception mode. Let 
{Ei,E2,---,E n }zre the residual energies of the nodes 
measured by the following method. 

The power consumed for transmitting a packet is given by 
(1) 



Consumed energy = TP*t 



(i) 



Where TP is the transmitting power and t is transmission 
time. 

The power consumed for receiving a packet is given by (2) 

Consumed energy = RP * t (2) 

Where RP is the reception power and t is the reception time. 
The value t can be calculated as 



t = D s ID r 



D s is Data size and D r is Data rate 



(3) 



Hence, the residual energy ( E ) of each node can be 
calculated using (1) or (2) and (3) 

E = Current energy - Consumed energy 



2) Algorithm: 

1 . Consider a group G j = { A/} , A/^ , • • • , A/3 } 

2. Measure the distance d of each node from source S 

d(S,Nj) where i = 1,2, •••,/! 

3. Sort the nodes N( in ascending order of d . 

4. Create the partitions XI and X2 of the nodes N( such 

that 

Xl = {N h -,N Q } 
X2 = {N Q + h ..-,N n } 
Where Q is the distance threshold. 

5. Source unicast the packets to XI 

6. In X2 find a relay node N r which has max (E() 

7. Then S unicast the packets to N r which in turn 

multicast the packets to the rest of the nodes in X2 . 




N 



° ,OV VO n. 



On. 



Figure 1 . Energy efficient tree construction 

Source S unicast the packets to nodes N\ , A^ , A/3 , A/4 
and A/5 is the relay node. A/5 multicast the packets to the rest 
of the nodes Afg , • • • , N\ \ . 

B. Multicast Admission Control 

Most of the existing schemes depend on individual 
receivers to detect congestion and adjust their receiving rates 
which are much disadvantageous. We propose a scheme which 
adjusts the multicast traffic rate at each bottleneck of a 
multicast tree. Each node estimates its current traffic load and 
arrival rate. Based on its traffic load, it estimates the receiving 
rate. If the receiving rate is less than the arrival rate, it 
adaptively adjusts its receiving rate. 



198 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



In order to adjust the total number of multicast flows 
which traverse a bottleneck, the following procedure is used. 
In our proposed scheme, based on the link's output queue 
state, multicast flows at a bottleneck can be blocked or 
released. Let the number of packets in the queue is N . Let 
QTl and QTl (QTl < QTl) are two thresholds for the 

queue size. Then the flow is released or blocked based on the 
following conditions. 

If N < QTl , then the multicast flow is released. 

If N > QTl , then the multicast flow is blocked. 

In most of the existing schemes, in order to detect 
congestion and for adjusting the receiving rate they depend on 
the individual receivers. In our proposed scheme multicast 
traffic rate is adjusted at each bottleneck of a multicast tree. 
Whenever congestion happens or about to, then the multicast 
sessions which traverse the branch are blocked. Thus the 
packets are stopped from entering the branch. The blocked 
flows are released to traverse the branch when the branch is 
lightly utilized. 

C. Multicast Traffic Rate Adjustment 

When the available bandwidth is less than the required 
bandwidth or the queue size is less than a minimum threshold 
value, it indicates the possibility of congestion or packet loss. 
The behaviour of the multicast session is expressed as 



R(t + l) = {R(t)-g 
R(t) + g 
R(t) 



If R(t) > B 
If R(t) < B 
otherwise} 



Here R(t) denotes the instantaneous rate of the multicast 
session at time t . B is the bottleneck bandwidth. 

When R(t) > B then the network is congested and the 
multicast session decreases its rate by a step g . 

If R(t) < B then the network is not congested and the 
multicast session increases its rate by a step g . 

The proposed scheme overcomes most of the 
disadvantages of existing schemes: 

1. Link errors cannot cause the proposed scheme to 
wrongly block a layer, because instead of the loss information 
at receivers, the queue state at a bottleneck is used as the 
metric to adjust the multicast traffic rate at the bottleneck. 

2. Link access delay caused by competition in MANETs 
cannot hinder the rate adjustment in this scheme, because, it 
blocks multicast layers right at each bottleneck of a multicast 
tree instead of depending on receivers to request pruning to 
drop layers. 

3. Because of the on-the-spot information collection and 
rate control this scheme has very limited control traffic 
overhead. 

Moreover, the proposed scheme does not impose any 
significant changes on the queuing, scheduling or forwarding 
policies of existing networks. 



IV. Simulation Results 

A. Simulation Model and Parameters 

We use NS2 to simulate our proposed protocol. In our 
simulation, the channel capacity of mobile hosts is set to the 
same value: 2 Mbps. We use the distributed coordination 
function (DCF) of IEEE 802.11 for wireless LANs as the 
MAC layer protocol. It has the functionality to notify the 
network layer about link breakage. 

In our simulation, 50 mobile nodes move in a 1000 meter x 
1000 meter region for 50 seconds simulation time. We assume 
each node moves independently with the same average speed. 
All nodes have the same transmission range of 250 meters. In 
our simulation, the minimal speed is 5 m/s and maximal speed 
is 5 m/s. The simulated traffic is Constant Bit Rate (CBR). 

Our simulation settings and parameters are summarized in 
table I 

Table I. Simulation Parameters 



No. of Nodes 


50 


Area Size 


1000 X 1000 


Mac 


802.11 


Radio Range 


250m 


Simulation Time 


50 sec 


Traffic Source 


CBR 


Packet Size 


250,500,... 1000 


Mobility Model 


Random Way Point 


Speed 


5 m/s 


Receivers 


5,10,. ..25 


Pause time 


5s 


Transmit Power 


0.660 w 


Receiving Power 


0.395 w 


Idle Power 


0.335 w 


Initial Energy 


3.1 J 



B. Performance Metrics 

We compare our EERCCP protocol with the multicast 
AODV [14] protocol. We evaluate mainly the performance 
according to the following metrics. 

Average end-to-end delay: The end-to-end-delay is 
averaged over all surviving data packets from the sources to 
the destinations. 

Average Packet Delivery Ratio: It is the ratio of the No. 
of packets received successfully and the total no. of packets 
sent. 

Average Energy Consumption: The average energy 
consumed by the nodes in receiving and sending the packets 
are measured. 

Throughput: It is the number of packets received by all 
the nodes in the network. 

C. Results 

1) Based On Receivers: In this experiment, we vary the 
group size or the number of receivers per group as 5,10 25. 



199 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



Receivers Vs Delay 




-♦— AODV 
-m— EERCCP 



10 15 20 25 
Receivers 

Figure 2. Receivers Vs Delay 

Receivers Vs DelRatio 




-♦—AODV 
-m— EERCCP 



10 15 20 25 
Receivers 



When the number of receivers is increased, 

Figure 2 shows the end-to-end delay occurred for both 
AODV and EERCCP. As we can see from the figure, the 
delay is less for EERCCP, when compared to AODV. 

Figure 3 shows the delivery ratio for both AODV and 
EERCCP. As we can see from the figure, the delivery ratio is 
high for EERCCP, when compared to AODV. 

Figure 4 shows the energy consumption for both the cases. 
As we can see from the figure, the energy consumption is less 
for EERCCP, when compared to AODV. 

Figure 3 shows the throughput occurred for both the cases. 
As we can see from the figure, the throughput is high for 
EERCCP, when compared to AODV. 

2) Based on Psize: In this experiment, we vary the packet 
size as 250,500 1000. 

Psize Vs Delay 




Figure 3. Receivers Vs Delivery Ratio 



250 500 750 1000 
Psize 



-♦—AODV 
-■ — EERCCP 



Receivers Vs Energy 



Figure 6. Psize Vs Delay 




3000 



-♦—AODV 
-m— EERCCP 



5 10 15 20 25 
Receivers 

Figure 4. Receivers Vs Energy 



Receivers Vs Throughput 




-♦—AODV 
-m— EERCCP 



10 15 20 25 
Receivers 



Figure 5. Receivers Vs Throughput 





Psize Vs DelRatio 










0.4 
| 0.3 
2 0.2 

0.1 


. 






■~ 








— ♦- 


-AODV 








^ ^^ 






1=1=0000 


♦^^•^^^ 
















o J 
















250 


500 750 1 000 
Psize 





Figure 7. Psize Vs DelRatio 



Psize Vs Energy 




-♦—AODV 
-m— EERCCP 



Figure 8. Psize Vs Energy 



200 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



Psize Vs Throughput 



tL\J\J\J 

ts 1 Ron 


V 


Q. 

.C 

o> 1000 


^V_ 


2 


^=t=* 


- 





-♦—AODV 
-■— EERCCP 



250 500 750 1000 
Psize 

Figure 9. Psize Vs Throughput 



When the Psize is increased, 

Figure 6 shows the end-to-end delay occurred for both 
AODV and EERCCP. As we can see from the figure, the 
delay is less for EERCCP, when compared to AODV. 

Figure 7 shows the delivery ratio for both AODV and 
EERCCP. As we can see from the figure, the delivery ratio is 
high for EERCCP, when compared to AODV. 

Figure 8 shows the energy consumption for both the cases. 
As we can see from the figure, the energy consumption is less 
for EERCCP, when compared to AODV. 

Figure 9 shows the throughput occurred for both the cases. 
As we can see from the figure, the throughput is high for 
EERCCP, when compared to AODV. 

V. Conclusion 

In this paper, we have proposed an energy efficient and 
reliable congestion control protocol for multicasting in mobile 
adhoc networks. Our proposed protocol overcomes the 
disadvantages of existing multicast congestion control 
protocols which depend on individual receivers to detect 
congestion and adjust their receiving rates. In the first phase 
of our protocol, we have built a multicast tree routed at the 
source, by including the nodes with higher residual energy 
towards the receivers. In the second phase, we have proposed 
an admission control scheme in which a multicast flow is 
admitted or rejected depending upon on the output queue size. 
In the third phase, we have proposed a scheme which adjusts 
the multicast traffic rate at each bottleneck of a multicast tree. 
Because of the on-the-spot information collection and rate 
control, this scheme has very limited control traffic overhead 
and delay. Moreover, the proposed scheme does not impose 
any significant changes on the queuing, scheduling or 
forwarding policies of existing networks. Simulation results 
have shown that our proposed protocol has better delivery 
ratio and throughput with less delay and energy consumption 
when compared with existing protocol. 

References 

[1] Luo Junhai, Xue Liu, Ye Danxia," Research on multicast routing 
protocols for mobile ad-hoc networks", Elsevier, 2007 

[2] Abdussalam Nuri Baryun, and Khalid Al-Begain," A Design Approach 
for MANET Multicast Protocols", ISBN, 2008 



[3] Hua Chen, Baolin Sun," An Entropy-Based Fuzzy Controllers QoS 
Routing Algorithm in MANET",IEEE,2009 

[4] Tolga Numanoglu and Wendi Heinzelman," Improving QoS in 
Multicasting Through Adaptive Redundancy", University of Rochester 
Center for Electronic Imaging Systems 

[5] D. Agrawal, T. Bheemarjuna Reddy, and C. Siva Ram Murthy," Robust 
Demand-Driven Video Multicast over Ad hoc Wireless 
Networks",IEEE,2006 

[6] Guojun Wang, Jiannong Cao, Lifan Zhang, Keith C. C. Chan," A Novel 
QoS Multicast Model in Mobile Ad Hoc Networks", IEEE International 
Parallel and Distributed Processing Symposium,2005 

[7] Vida Lashkari B. O., Mehdi Dehghan" QoS-aware Multicast Ad hoc On- 
Demand Distance Vector Routing", WCE 2007 

[8] Zeyad M. Alfawaer, GuiWei Hua, and Noraziah Ahmed," A Novel 
Multicast Routing Protocol for Mobile Ad Hoc Networks", ISSN, 2007 

[9] Harald Tebbe and Andreas J. Kassler ," QAMNet: Providing Quality of 
Service to Ad-hoc Multicast Enabled Networks", Wireless Pervasive 
Computing, 2006 1st International Symposium on, IEEE,2006 

[10] Mohammed Saghir, Tat-Chee Wan, Rahmat Budiarto," QoS Multicast 
Routing Based on Bandwidth Estimation in Mobile Ad Hoc Networks", 
ICCCE2006 

[11] Ravindra Vaishampayan, JJ. Garcia-Luna-Aceves," Efficient and 
Robust Multicast Routing in Mobile Ad Hoc Networks", IEEE, 2004 

[12] Y. Chen and Y. Ko, "A Lantern-Tree Based QoS on Demand Multicast 
Protocol for A wireless Ad hoc Networks," IEICE Trans. 
Communications, vol. E87-B, 2004 

[13] 13 N. Ben Ali, A. Belghith, J. Moulierac, M. Molnar," QoS multicast 
aggregation under multiple additive constraints", Elsevier, 2008 

[14] Elizabeth M. Royer, Charles E. Perkins," Multicast Operation of the Ad- 
hoc On-Demand Distance Vector Routing Protocol",ACM,1999 



Dr.G.Sasi Bhushana Rao received his BE from 
GITAM, Visakhapatnam. ME, PhD, MBA (HRD & 
Marketing) from Osmania University. He has 22 years 
of research and development, administrative 
experience as asst. General Manager (CNS) in Airport 
Authority of India, Ministry of Civil Aviation and 
ISRO, Govt, of India. Presently he is working as 
Senior Professor in E C E Dept. at Andhra University 
Visakhapatnam. His areas of interest are GPS, Signal 
Processing and Mobile Communications. He has more than 100 in various 
International and National Journals and conferences (including IEEE, IEE) 
presently ongoing project works under his guidance are DRDO (N S T L), 
CSIR, UGC, AICTE and WIPRO research projects in the department of 
electronics and communication engineering Andhra University 
Visakhapatnam. He is senior member in IEEE. 



M. RajanBabu received B.Tech from Bapatla 
Engineering College-Bapatla and M.Tech from J N T 
U College of Engineering-Kakinada in the 
Department of Electronics And Communication 
Engineering, He is pursuing his PhD under the 
guidance of Dr. G. Sasi Bhushana Rao on wireless 
networks from Andhra University He is having 12 
I / years of teaching experience currently he is working 

as Associate Prof in Lendi Institute of Engineering and technology Jonnada 

Vizianagaram (dt) A.P -India 





201 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



An Intelligent System For Effective Forest Fire 
Detection Using Spatial Data 



K.Angayarkkani 

Senior lecturer 

Department of ComputerApplications 

D.G. Vaishnav College, Arumbakkam, Chennai 



Dr.N.Radhakrishnan 

Geocare Research Foundation 

#23/30, First main Road, 

Pammal, Chennai - 600 075, India 



Abstract — The explosive growth of spatial data and extensive 
utilization of spatial databases emphasize the necessity for the 
automated discovery of spatial knowledge. In modern times, 
spatial data mining has emerged as an area of voluminous 
research. Forest fires are a chief environmental concern, causing 
economical and ecological damage while endangering human 
lives across the world. The fast or early detection of forest fires is 
a vital element for controlling such phenomenon. The application 
of remote sensing is at present a significant method for forest 
fires monitoring, particularly in vast and remote areas. Different 
methods have been presented by researchers for forest fire 
detection. The motivation behind this research is to obtain 
beneficial information from images in the forest spatial data and 
use the same in the determination of regions at the risk of fires by 
utilizing Image Processing and Artificial Intelligence techniques. 
This paper presents an intelligent system to detect the presence of 
forest fires in the forest spatial data using Artificial Neural 
Networks. The digital images in the forest spatial data are 
converted from RGB to XYZ color space and then segmented by 
employing anisotropic diffusion to identify the fire regions. 
Subsequently, Radial Basis Function Neural Network is 
employed in the design of the intelligent system, which is trained 
with the color space values of the segmented fire regions. 
Extensive experimental assessments on publicly available spatial 
data illustrated the efficiency of the proposed system in 
effectively detecting forest fires. 

Keywords- Data Mining, Remote Sensing, Spatial data, Forest 
Fire Detection, Color Space, Segmentation, Anisotropic diffusion, 
Radial Basis Function Neural Network (RBFNN). 

I. Introduction 

The rapid progress in scientific data collection has led to 
enormous and ever-increasing quantity of data making it 
unfeasible to be manually interpreted. Therefore, the 
development of novel techniques and tools in assist for 
humans, aiding in the transformation of data into useful 
knowledge, has been the heart of the comparatively new and 
interdisciplinary research area called the "Knowledge 
Discovery in Databases (KDD)" [3]. Data mining is the vital 
step in KDD, which facilitates the discovery of buried but 
valuable knowledge from enormous databases. Data Mining is 
formally defined as "The non-trivial extraction of inherent, 
new, and potentially valuable information from databases" 
[5]. Data mining combines machine learning, pattern 
recognition, statistics, databases, and visualization techniques 



into a single unit so as to enhance efficient information 
extraction from large databases [6]. Data mining techniques 
profit a number of fields like marketing, manufacturing, 
process control, fraud detection and network management. 
Other than this, a huge variety of data sets like market basket 
data, web data, DNA data, text data, and spatial data [7] have 
benefited as well. 

The progress in scientific data collection has resulted in 
huge and continuously rising amount of spatial data [1]. Thus 
the need, for automated discovery of spatial knowledge from 
massive amount of spatial data, arises. The process of 
identifying previously hidden but valuable information from 
vast spatial databases is known as spatial data mining. It is 
comparatively tedious to extract patterns of value and interest 
from the spatial databases owing to the complexity of spatial 
data types, spatial relationships, and spatial autocorrelation 
than that of the conventional numeric and categorical data [2]. 
Spatial data mining technologies facilitate the comprehension 
of spatial data, discovery of relationships among spatial and 
non-spatial variables, determination of the spatial distribution 
patterns of a specific phenomenon further supporting the 
envisagement of the pattern trends. The elemental parts of 
spatial data mining are spatial statistics and data mining. 
Spatial data mining techniques involves visual interpretation 
and analysis, spatial and attribute query and selection, 
characterization, generalization and classification, detection of 
spatial and non spatial association rules, clustering analysis and 
spatial regression in addition to a wide variety of other fields 
[8]. 

Spatial data mining and relational data mining vary from 
one another because of the fact that in the former the attributes 
of the neighbors of some object of interest need to be taken into 
consideration as well, since they have a prominent influence on 
the object [9]. Some distinguishing characteristics of spatial 
data that forbid the usage of regular data mining algorithms 
include: (i) rich data types (e.g., extended spatial objects) (ii) 
inherent spatial relationships between the variables, (iii) other 
factors that influence the observations and (iv) spatial 
autocorrelation among the characteristics [2]. The extraction of 
patterns of interest and rules from the spatial data sets like the 
remotely sensed imagery and related ground data significantly 
benefits the application areas like precision agriculture, 
community planning, resource discovery and more[10]. Spatial 



202 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



data mining is comprehensively employed in change detection, 
modeling deforestation, disaster analysis, forest fire detection 
and other related fields. Our research focuses on the detection 
of forest fires from the spatial data analogous to forest regions. 

A. Forest Fires 

For a long time, fires have been a source of trouble. Fires 
have notable influence over the ecological and economic 
utilities of the forest, being a prime constituent in a great 
number of forest ecosystems [11]. Past has witnessed multiple 
instances of forest and wild land fires. Fires play a remarkable 
role in determining landscape structure, pattern and eventually 
the species composition of ecosystems. The integral part of the 
ecological role of the forest fires [13] is formed by the 
controlling factors like the plant community development, soil 
nutrient availability and biological diversity. Fires are 
considered as a significant environmental issue because they 
cause prominent economical and ecological damage despite 
endangering the human lives [12]. Due to the forest fires, 
several hundred million hectares (ha) of forest and other 
vegetation are destroyed every year [14]. 

Occasionally, forest fires have forced the evacuation of 
susceptible communities in addition to heavy damages 
amounting to millions of dollars. As per the forest Survey of 
India 19.27% or 63.3 million ha of the Indian land has been 
classified as forest area, of which 38 million ha alone are 
hoarded with resources in great quantity (crown density above 
40%).Thus the country's forests face a huge threat. 
Degradation caused by forest fires [15] jeopardizes the Indian 
forests. Fires caused huge damage in the year 2007 affecting 
huge territories in addition to the prominent number of human 
casualties [16]. Forest fires remains to be a potential threat to 
ecological systems, infrastructure and human lives. The 
practical and effective option to minimize the damage caused 
by the forest fire is to detect the fires at their early stages and 
reacting fast to prevent the spread of the fire. Hereafter, hefty 
efforts have been taken to ease the early detection of forest 
fires, usually being carried out with the help of human 
surveillance. Forest fire, Drought, Flood and many other 
phenomena especially the ones with large spatial extent are 
some of the spatial phenomena which contain predictable 
spatial patterns that are evident through remote sensing 
Images/products. 

B. Our Contributions 

In our earlier work [45], we have presented an efficient 
forest fire detection system using Fuzzy logic. The primary 
intention of this research is to extract valuable information 
from spatial data and employ them for locating the regions 
vulnerable to forest fire with the aid of Image Processing and 
Artificial Intelligence techniques. This paper presents an 
intelligent system that is capable of detecting forest fires. The 
presented intelligent system utilizes the images in the spatial 
data that corresponds to forest regions, obtained from remote 
sensing. The Radial Basis Function Neural Network is 
employed in the design of the presented intelligent system. The 
images in the forest spatial data with the presence of fires are 
utilized in training the neural network. Initially, the digital 
images in the forest spatial data are converted from RGB to 

XYZ color space. Then, the segmentation of the image in 



XYZ color space is carried out with the aid of the renowned 

anisotropic diffusion approach. The X-YL color space values 
of the regions with fires, resulting from the segmentation, are 
fed as input to training the radial basis function neural network. 

For a given XYZ co lor space value of a pixel, the trained 
radial basis function neural network will identify whether that 
pixel corresponds to fire region or not. The presented 
intelligent system effectively detects forest fire, which is very 
well illustrated by the experimental evaluation on the publicly 
available spatial data. 

The rest of the paper is organized as follows: Section II 
presents a brief review of some recent researches existing in the 
literature related to forest fire detection. A concise description 
of the concepts utilized in the presented intelligent system is 
given in Section III. The proposed intelligent system for 
effective forest fire detection is presented in Section IV. The 
experimental results are given in Section V. The conclusions 
are summed up in Section VI. 

Ii. Review Of Related Researches 

The proposed research has been motivated by several 
earlier researches in the literature related to forest fire detection 
using spatial data and artificial intelligence techniques. A 
concise description of some of the recent researches is given in 
this section. 

Armando et al. [17] have studied on the automatic 
recognition of smoke signatures in lidar signals attained from 
very small-scale experimental forest fires using neural-network 
algorithms. A scheme of multi- sensorial integrated systems for 
early detection of forest fires has been presented by Ollero et 
al. [18]. The system presented by the authors uses infrared 
images, visual images, and data from sensors, maps and 
models. To facilitate the minimization of perception errors and 
the improvement in reliability of the detection process, it is 
necessary for the integration of sensors, territory knowledge 
and expertise, according to their study. 

An improved fire detection algorithm which provides 
increased sensitivity to smaller, cooler fires as well as a 
significantly lower false alarm rate has been presented by Louis 
Giglio et al. [19]. The Theoretical simulation and high- 
resolution Advanced Space borne Thermal Emission and 
Reflection Radiometer (ASTER) scenes are employed to 
establish the performance of their algorithm. Seng Chuan Tay 
et al. [20] have presented an approach to reduce the false 
alarms in the hotspots of forest fire regions which uses 
geographical coordinates of hot spots in forest fire regions for 
detection of likely fire points. The authors employ clustering 
and Hough transformation to determine regular patterns in the 
derived hotspots and classify them as false alarms on the 
assumption that fires generally do not spread in regular patterns 
such as straight lines. In this work demonstrate the application 
of spatial data mining for the reduction of false alarm from the 
set of hot spots is derived from NO A A images. 

A graph based forest fire detection algorithm based on 
spatial outlier detection methods has been presented by Young 
Gi Byun et al. [21]. By using the spatial statistics the authors 
have achieved spatial variation in their algorithm. This 



203 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



algorithm illustrates higher user and producer accuracies, when 
compared with the MODIS fire product provided by the NASA 
MODIS science team. The ordinary scatter plot algorithm was 
proved to be inefficient by the authors because it is insensitive 
to small fires, while Moran's scatter plot was also weak 
because of the numerical criterion's absence for spatial 
variation which requires a more and less high commission 
error. 

An approach to predict forest fires in Slovenia using 
different data mining techniques has been presented by Daniela 
Stojanova et al. [22]. The authors have employed the predictive 
models based on the data from a GIS (Geographical 
Information System) and the weather prediction model - Aladin 
and MODIS satellite data. The work examined three different 
datasets: one for the Kras region, one for Primorska region and 
one for continental Slovenia. The researchers demonstrated that 
Bagging and boosting of decision trees offers the best results in 
terms of accuracy for all three datasets. Yasar Guneri Sahin 
[23] has proposed a mobile biological sensor system for prior 
detection of forest fires which utilizes animals as mobile 
biological sensors. This system is based on the existing animals 
tracking systems used for the zoological studies. The work 
illustrates that the combination of these fields may lead to 
instantaneous development of animal tracking as well as forest 
fire detection. A number of serious forest fires were detected 
by the system in the earliest, which reduced their effect and 
therefore contributes to the reduction of the speed of global 
warming. 

A fully automated method of forest fire detection from TIR 
satellite images on the basis of random field theory has been 
presented by Florent Lafarge et al. [24]. The results of the 
system rely only on the confidence coefficient. The obtained 
values for the both detection rate and false alarm rate were 
convincing. The estimation of fire propagation direction 
presents interesting information associated to the evolution of 
the fires. In Movaghati et al. [25], the capability of agents to be 
applied in processing of remote sensing imagery has been 
studied. An agent based approach for forest fire detection has 
been presented in this paper. The tests used in MODIS version 
4 contextual fire detection algorithms were used by the agents 
to determine agent behavioral responses. The performance of 
their algorithm was compared against that of MODIS version 4 
contextual fire detection algorithm and ground-based 
measurements. The results portray a good agreement between 
the algorithms and field data. 

In our earlier work [45], we have presented an efficient 
system to detect forest fires using spatial data collected from 
forest. Image Processing and Artificial Intelligence techniques 
were utilized in the design of the presented system. Anisotropic 
diffusion and fuzzy logic are employed for segmentation and 
fire detection processes respectively. The images are converted 
to YCbCr color space and segmentation is performed. The Cr 
value of YCbCr color space of fire pixels is utilized in the 
formation of fuzzy sets and fuzzy rules are derived from the 
formed fuzzy sets. The publicly available spatial data has been 
employed in the evaluation process. The fuzzy rules derived 
using the presented system, have successfully detected the 
forest fires in the spatial data. 



In. Description Of Concepts Utilized In The 
Presented Intelligent System 

The concepts utilized in the presented intelligent system for 
effective forest fire detection such as color space, anisotropic 
diffusion segmentation and artificial neural networks are 
detailed in this section. 

A. Color Space Conversion 

A color space is defined as a means by which the 
specification, creation and visualization of colors is performed. 
A computer screen produces colors based on the varied 
combinations of red, green and blue phosphor emission 
required to form a color. Typically color is represented by three 
coordinates or parameters [26]. The location of the color in the 
color space is exemplified by these parameters. Color space 
conversion is defined as the transformation and description of a 
color from one source to another. Normally, color space 
conversion is performed while converting an image that is 
represented in one color space to another color space, with the 
objective of making the translated image appear as similar as 
possible to the original. The commonly used color spaces are 

RGB, CIE XYZ 9 cm YUV, CIE L*a*b*, YCbCr and HSV. In 
the proposed intelligent system, the images in RGB color space 
are converted to XYZ color space. 

1 ) Cie Xyz Color Space: CIE XYZ color space [27] is one 
of the first mathematically defined color spaces created by the 
International Commission on Illumination in 1931. Any color 
can be generated as a mixture of three other colors or 
"Tristimuli" and commonly RGB for CRT based systems (TV, 
computer) or XYZ (fundamental measurements). The XYZ 
color space is defined such that all visible colors can be 
represented using only positive values, and, the Y value is 
luminance. As a result, the colors of the XYZ primaries 
themselves are invisible [28]. The chromaticity diagram is 
extremely non-linear, in that a vector of unit magnitude 
denoting the difference between two chromaticities is not 
uniformly visible. A 3x3 matrix transform is used to transform 
the R G B values in a particular set of primaries to and from 
CIE XYZ . These transformations involve tristimulus values 
which are a set of three linear-light components that conform 
to the CIE color-matching functions. CIE XYZ i s a special set 
of tristimulus values. The equations to convert RGB into 
XYZ color space are as follows: 



0.412453 0.357580 0.180423 
0.212671 0.715160 0.72169 
0.019334 0.119193 0.950227 



(1) 



B. Anisotropic Diffusion Segmentation 

Image segmentation is defined as a low-level image 
processing task meant for partitioning an image into identical 
regions [29]. The segmentation results can be used to identify 
the regions of interest and objects in the scene that is very 
beneficial to the subsequent image analysis. Because of the fact 
that the inherent multi-features not only possess non linear 
relation independently but also encompass inter-feature 



204 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



dependency between R, G, and B, color image segmentation is 
more monotonous when compared to the grey image 
segmentation. In our system, we have employed an anisotropic 
diffusion approach for the segmentation of images in the spatial 

data. The segmentation is carried out on the XYZ color space 
converted image. 

Literature offers numerous models of linear and nonlinear 
diffusion for achieving image smoothing and segmentation. 
Nonlinear anisotropic diffusion has been one of the commonly 
used approaches by researchers [30], [31] in their works. The 
anisotropic diffusion enhances the response of edge detection 
algorithms by a series of operations namely: smoothing the 
image interiors to emphasize boundaries for segmentation, 
eliminating the spurious detail and eradicating noise from 
images efficiently [34]. The relaxation processes that 
implement anisotropic diffusion tends to leave out the low 
frequency objects that are complex to be dispersed without 
over-processing the image. 

Anisotropic diffusion in image processing discretizes the 
family of continuous partial differential equations, which 
incorporate both the physical processes of diffusion and the 
Laplacian. Provided that there are no sinks or sources that exist 
[32], the following equation formulates the abovementioned 
process (for any dimension): 



~\ 

— u (x/) = div (c (x/) Vu (x/)) 
dt 



(2) 



c(x 9 t) 



Diffusion strength is controlled by^ vyv '* y . Vector x 
represents the spatial coordinate(s).The ordering parameter is 

the variable^. The function ^ ' ^ is considered as image 
I(x,t) 



intensity 



[33]. 



C. Artificial Neural Networks (ANN) 

Artificial Neural Networks are a branch of the artificial 
intelligence, developed to reproduce human reason and 
intelligence. ANN possesses the abilities to recognize patterns, 
manage data and learn like the brain [35]. The weights and the 
input-output function (transfer function) that is specified for the 
units are used to characterize the behavior of an ANN [37] . The 
most significant pros in using artificial neural networks are 
solving the very complex problems of conventional 
technologies, not formulating an algorithmic solution or using 
the very complex solution [35]. In the presented intelligent 
system, Radial Basis Function Neural Network is employed 
and is detailed below. 

1) Radial Basis Function Neural Network (RBFNN): In the 
late 80' s, a variant of artificial neural network emerged by the 
name, Radial Basis Functions. Nevertheless, their roots are 
well-established in much older pattern recognition techniques 
for instance potential functions, clustering, functional 
approximation, spline interpolation and mixture models [39]. 
Radial Basis Function Neural Network (RBFNN) is based on 
supervised learning. RBF networks were autonomously 
proposed by many researchers [40], [41], [42], [43], [44] and 



are a popular variant to the MultiLayer Perceptron MLP. RBF 
networks are also excellent at modeling non-linear data and 
can be trained in one stage rather than using an iterative 
process as in MLP and also learn the given application 
speedily. The RBF network has a feed forward structure 
consisting of a single hidden layer of J locally tuned units, 
which are fully interconnected to an output layer of L linear 
units. All hidden units concurrently receive the n - 
dimensional real valued input vector X (Figure. 1). The 
prime difference from that of MLP is the absence of hidden- 
layer weights. The hidden-unit outputs are not computed using 
the weighted- sum mechanism/sigmoid activation; rather each 
hidden-unit output Z / is obtained by closeness of the input 
X to an n -dimensional parameter vector jUj associated with 
f hidden unit [4]. The response characteristics of the 



the 

■ th 



hidden unit (j = 1,2, ,7) is assumed as, 



K 



f\\ \\\ 

X- Mj 



(3) 



Where K is a strictly positive radially symmetric function 

Li 

(kernel) with a unique maximum at its 'centre' J and which 

drops off rapidly to zero away from the centre. The parameter 

G 

J is the width of the receptive field in the input space from 

/ z. 

unit- 7 . This implies that J has an appreciable value only 

|x-//.| a 

when the distance " " is smaller than the width J . 

Given an input vector X , the output of the RBF network is the 

L -dimensional activity vector * , whose ^ component 
( / = 1 > 2 >—> Misgiven by [36], 

Y l (X) = f j w lj Z j (X) 

j=1 (4) 

For / = 1 , mapping of (3) is similar to a polynomial 
threshold gate. However, in the RBF network, a choice is made 
to use radially symmetric kernels as 'hidden units'. 




Figure 1. Radial Basis Function Neural Network (RBFNN) 



205 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



RBF networks are best suited for approximating continuous 

or piecewise continuous real-valued mapping J 
where n is sufficiently small. These approximation problems 
include classification problems as a special case. From (3) and 
(4), the RBF network can be viewed as approximating a desired 

function ** ^ ' by superposition of non-orthogonal, bell- 
shaped basis functions. The degree of accuracy of these RBF 
networks can be controlled by three parameters: the number of 
basis functions used, their location and their width [38]. 

Iv. Intelligent System For Effective Forest Fire 
Detection 

The proposed intelligent system for effective detection of 
forest fires is presented in this section. The spatial data 
collected from forest regions are utilized by the presented 
intelligent system. With the aid of the images in the spatial 
data, forest fire detection is performed. The Radial Basis 
Function Neural Network is employed in the design of the 
presented intelligent system. The images in the forest spatial 
data with the presence of fires are employed in training the 
radial basis function neural network. Initially, the images with 

the presence of fires are converted from RGB to XYZ color 
space. The color space conversion from RGB to XYZ is carried 
out with the help of (1). Figure 2 shows the image in RGB 
color space and its corresponding XYZ color space converted 
image. 




(a) (b) 

Figure 2. a) Image in RGB color space, b) XYZ color space converted image 

Afterwards, the XYZ color space converted image is 
segmented using anisotropic diffusion segmentation, which 
locates the regions of fire. The result of anisotropic diffusion 
segmentation is depicted in Figure 3. 



•%«i^ '" 






t ^ *"***• 2S!% 



Figure 3. Anisotropic Diffusion Segmented image 

The regions of fire obtained using the segmentation is 
utilized in training the radial basis function neural network. The 

radial basis function neural network is trained with the XYZ 
color space values of the pixels that belong to fire regions. 
With the help of the trained neural network, we can effectively 



detect the presence of forest fires in an image. The presence of 
forest fire in an image is detected using the following steps. 

Initially the image is converted from RGB to XYZ color 
space. Then, the color space converted image is segmented 
using anisotropic diffusion segmentation. Subsequently, the 

XYZ color space values of pixels in the segmented regions 
are fed as input to the trained neural network for detecting the 
presence of fires. The designed intelligent system will aid the 
people in surveillance to detect forest fires and to take 
appropriate actions. 

V. Experimental Results 

This section presents the results obtained from the 
experimentation on the presented intelligent system. The 
proposed intelligent system is implemented in MATLAB 
(Matlab 7.4). The publicly available forest spatial data with the 
presence of fires are employed in training the radial basis 
function neural network. Consequently, forest spatial data with 
and without the presence of fires are fed as input to the 
proposed system for evaluation. The presence of fires is 
detected effectively by the presented intelligent system with the 
aid of the trained neural network. The intermediate results of 
the presented system are depicted in Figure 4. From the results 
we can conclude that the presented intelligent system can be 
used for effectively detecting forest fires in the spatial data 
using artificial intelligence techniques. 




Figure. 4 Intermediate results of the presented intelligent system a) Input 

Image, b) XYZ color space converted image, c) Output of Anisotropic 

Diffusion Segmentation, d) Fire detected Image 

Vi. Conclusion 

Forest fires cause noteworthy environmental demolition 
while menacing human lives. In the last two decades, a 
significant effort was made to develop automatic detection 
tools that could aid the Fire Management Systems (FMS). The 
three chief trends used for the detection of forest fires are: the 
use of satellite data, infrared/smoke scanners and local sensors 
(e.g. meteorological). In this paper, we have presented an 
intelligent system for effective forest fire detection using 
spatial data. The proposed system made use of image 
processing and artificial intelligence techniques. The images in 
the spatial data, obtained from remote sensing, have been 



206 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



utilized by the presented system for the detection of forest fires. 
The color space conversion, anisotropic diffusion segmentation 
and Radial Basis Function Neural Networks have been 
employed in the presented intelligent system. The experimental 
results have demonstrated the effectiveness of the proposed 
intelligent system in detecting forest fires using spatial data. 

References 

[I] Deren Li, Song Xia, Haigang Sui, Xiaodong Zhang, "Change Detection 
Based On Spatial Data Mining", Wuhan University, white paper, pages- 
8,2007. 

[2] Shekhar, S., Zhang, P., Huang, Y. and Vatsavai, R.R., Trends in spatial 
data mining. In: Kargupta, H., Joshi, A. (Eds.), Data Mining: Next 
Generation Challenges and Future Directions, AAAI/MIT Press, pp. 

357-380, 2003. 

[3] Martin Ester, Hans-Peter Kriegel, Jorg Sander, "Algorithms and 
Applications for Spatial Data Mining", Published in Geographic Data 
Mining and Knowledge Discovery, Research Monographs in GIS, 
Taylor and Francis, 2001. 

[4] Park, J. and Sandberg, I. W., "Universal approximation using radial 
basis function networks", Neural Computation, vol.3, pp. 246-257, 
1991. 

[5] W. J. Frawley, P. G. Shapiro, and C. J. Matheus, "Knowledge discovery 
in databases - an overview," Ai Magazine, vol. 13, pp. 57-70, 1992. 

[6] Tadesse, T., J.F. Brown, and M.J. Hayes. 2005. A new approach for 
predicting drought-related vegetation stress: Integrating satellite, 
climate, and biophysical data over the U.S. central plains. ISPRS Journal 
of Photogrammetry and Remote Sensing 59(4):244-253. 

[7] K. Julish, Data Mining for Intrusion Detection, a Critical Review, in 
Applications of Data Mining in Computer Security, D. Barbara and S. 
Jajodia (Eds.), Kluwer Academic Publisher, 2002. 

[8] Hong Tang, and Simon McDonald, "Integrating GIS and spatial data 
mining technique for target marketing of university courses", ISPRS 
Commission IV, Symposium 2002, Ottawa Canada, July 9-12 2002. 

[9] Imam Mukhlash and Benhard Sitohang, "Spatial Data Preprocessing for 
Mining Spatial Association Rule with Conventional Association Mining 
Algorithms," Proceedings of the International Conference on Electrical 
Engineering and Informatics Institute Teknologi Bandung, Indonesia 
June 17-19, 2007. 

[10] Q.Ding, W. Perrizo, Q.Ding, "On Mining Satellite and Other Remotely 
Sensed Images," DMKD-2001, pp. 33-40, Santa Barbara, CA, 2001. 

[II] Gonzalez, J.R., Palahi, M., Trasobares, A., Pukkala, T., "A fire 
probability model for forest stands in Catalonia (north-east Spain)," 
Annals of Forest Science, 63: 169-176, 2006. 

[12] Cortez, P. and A. Morais, "A data mining approach to predict forest fires 
using meteorological data.", New Trends in Artificial Intelligence, 
Proceedings of the 13th EPIA - Portuguese Conference on Artificial 
Intelligence, pp:5 12-523, 2007. 

[13] P.S. Roy , "Forest Fire and Degradation Assessment Using Satellite 
Remote Sensing and geographic Information System", Proceedings of a 
Training Workshop Satellite Remote Sensing and GIS Applications in 
Agricultural Meteorology, pages 361-400, 2003. 

[14] Y. Rauste, "Forest Fire Detection with Satellites for Forest Fire Control," 
Proc. XVIII Congress of ISPRS, Int'l Soc. for Photogrammetry and 
Remote Sensing, Vol.31, No. B7, pp. 584-588, 1996. 

[15] Bahuguna, V.K. & Singh, S., "The forest fire situation in India", Int. 
Forest Fire News, no. 26, pp. 23-27, 2001. 

[16] European-Commission. "Forest Fires in Europe 2007", Technical report, 
Report No-8, 2008. 

[17] Armando M. Fernandes, Andrei B. Utkin, Alexander V. Lavrov, Rui M. 
Vilar, "Neural Network Based Recognition of Smoke Signatures from 
Lidar Signals", Neural Processing Letters, Vol. 19, No. 3, pp. 175-189, 
June 2004. 

[18] Ollero, J.R. Martinez-De Dios and B.C. Arrue, "Integrated systems for 
early forest-fire detection", III International Conference on Forest Fire 
Research 14th Conference on Fire and Forest Meteorology VOL II, pp 
1977-1988, Luso, 16/20 November 1998 



[19] Louis Giglio, Jacques Descloitres, Christopher O. Justice, Yoram J. 
Kaufman, "An Enhanced Contextual Fire Detection Algorithm for 
MODIS", Remote Sensing of Environment, vol. 87, pp. 273-282, 2003. 

[20] Seng Chuan TAY, Wynne HSU, Kim Hwa LIM, "Spatial Data Mining: 
Clustering of Hot Spots and Pattern Recognition", Proceedings. 2003 
IEEE International Geoscience and Remote Sensing Symposium, 
Volume: 6, pp: 3685- 3687, 21-25 July 2003. 

[21] Young Gi Byun, Yong Huh, Kiyun Yu, Yong H Kim, "Evaluation of 
Graph-based Analysis for Forest Fire Detections", Proceedings of world 
academy of science, engineering and technology, volume. 10, December 
2005, ISSN 1307-6884. 

[22] Daniela Stojanova, Pance Panov, Andrej Kobler, Saso Dzeroski, 
Katerina Taskova, "Learning to predict forest fires with different Data 
Mining techniques", Conference on Data Mining and Data Warehouses 
(SiKDD 2006),Ljubljana, Slovenia, pp. 255-258, October 9, 2006. 

[23] Yasar Guneri Sahin, "Animals as Mobile Biological Sensors for Forest 
Fire Detection", Sensors, vol. 7, pp. 3084-3099, 2007. 

[24] Florent Lafarge, Xavier Descombes, Josiane Zerubia, "Forest fire 
detection based on Gaussian field analysis", In Proc. European Signal 
Processing Conference (EUSIPCO), Poznan, Poland, September 2007. 

[25] S. Movaghati, F. Samadzadegan, A. Azizi, "An agent-based algorithm 
for forest fire detection", The International Archives of The 
Photogrammetry, Remote Sensing and Spatial Information Sciences, 
ISPRS Congress Beijing 2008, Volume. XXXVII, Part. B7, 
Commission: VII/4, Advanced classification techniques, 2008. ISSN 
1682-1750. 

[26] Adrian Ford and Alan Roberts, "Color Space Conversions", Technical 
report, August 11, 1998. 

[27] CIE, Commission internationale de l'Eclairage proceedings, 1931. 
Cambridge University Press, Cambridge, 1932. 

[28] Smith, Thomas; Guild, John, "The CLE. colorimetric standards and 
their use", Transactions of the Optical Society, vol. 33, no. 3, pp. 73- 
134, 1931-32. 

[29] R. Duda and P. Hart, "Pattern Classification and Scene Analysis", Bayes 
Decision Theory. John Wiley & Sons, pp. 10-13, 1973. 

[30] P. Perona and J. Malik, "Scale space and edge detection using 
anisotropic diffusion", IEEE Transactions on Pattern Analysis and 
Machine Intelligence, Volume 12, Issue 7, July 1990, pp.629 - 639. 

[31] L. Alvarez, P.L. Lions, and J.M. Morel, "Image Selective Smoothing and 
Edge Detection by Nonlinear Diffusion, II," SIAM Journal of Numerical 
Analysis, vol. 29, no. 3, pp. 845-866, 1992. 

[32] Gerig, R. Kikinis, O. Kubler and F. A. Jolesz, "Nonlinear anisotropic 
filtering of MRI data", IEEE Transactions on Medical Imaging , vol. 11, 
Issue 2 , pp. 221-232, June 1992. 

[33] Sankar Seramani, Zhou Jiayin, Chan Kap Luk, N. Malmurugan and A. 
Nagappan, "Denoising of MR Images using Non Linear Anisotropic 
Diffusion Filtering as a Preprocessing Step", International Journal of 
BioSciences and Technology, Volume 1, Issue 1,2008, pp. 17-21. 

[34] G. Sapiro, "Cromaticity Diffusion", Proc. of ICIP 2000, Vancouver, BC, 
Canada, Sept. 2000, vol. II, pp. 784-787. 

[35] P. A. Maiellaro, R. Cozzolongo, P. Marino, "Artificial Neural Networks 
for the Prediction of Response to Interferon Plus Ribavirin Treatment in 
Patients with Chronic Hepatitis C", Current Pharmaceutical Design, vol. 
10, issue. 17, pp. 2101-2109(9), July 2004. 

[36] P. Venkatesan and S. Anitha, "Application of a radial basis function 
neural network for diagnosis of diabetes mellitus", Current Science, vol. 
91, no. 9, pp. 1195- 1199, 10 November 2006. 

[37] J.G. Wolff, "Medical Diagnosis as Pattern Recognition in a Framework 
of Information Compression by Multiple Alignment, Unification and 
Search", Decision Support Systems, Volume 42, Issue 2, pp: 608 - 625, 
November 2006. 

[38] Park, J. and Sandberg, I. W., "Approximation and radial basis function 
networks", Neural Computation, vol. 5, pp. 305-316, 1993. 

[39] Tou, J.T., Gonzalez, R. C, "Pattern Recognition Principles", Reading, 
MA: Addison - Wesley, 1974. 

[40] Broomhead, D. S. and Lowe, D., "Multivariate functional interpolation 
and adaptive networks", Complex Systems, vol. 2, pp. 321-355, 1988. 



207 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 
Vol. 7, No. 1,2010 



[41] Niranjan, M. A., Robinson, A. J. and Fallside, F., "Pattern recognition 
with potential functions in the context of neural networks", Proceedings 
Sixth Scandinavian Conference on Image Analysis, Oulu, Finland, vol. 
1, pp. 96-103, 1989. 

[42] Moody, J. and Darken, C. J., "Fast learning in networks of locally tuned 
processing units", Neural Computation, Vol. 1, issue 2, pp. 281-294, 
1989. 

[43] Hanson, S. J. and Burr, D. J., "Minkowski-r back propagation: learning 
in connectionist models with non-Euclidean error signals", Neural 
Information Processing Systems, American Institute of Physics, New 
York, pp. 348-357, 1988. 

[44] Poggio, T. and Girosi, F., "Regularization algorithms for learning that 
are equivalent to multilayer networks", Science, vol. 247, pp. 978-982, 
1990. 

[45] K. Angayarkkani and N. Radhakrishnan, "Efficient Forest Fire Detection 
System: A Spatial Data Mining and Image Processing Based Approach", 
International Journal of Computer Science and Network Security 
(IJCSNS), Vol.9 No.3, pp. 100 -107, March 2009. 



The author is a postgraduate in Computer Science followed 
by Master of Philosophy in Computer Science. The author 
has thirteen years of teaching experience in various fields of 
computer science. She has enrolled in Mother Teresa 
Women's University Kodaikanal for her Ph.D. doctoral 
degree. The author is currently doing research work on 
spatial data mining and image processing based techniques. 



fThe co-author is a post-graduate in Applied Geology (1985) 
followed by M.Tech degree in Remote Sensing (1990) and 
have completed the Ph.D doctoral degree in 1999 on 
spatial techniques - Remote sensing, GPS and GIS - 
watershed environment. The co-author, to his credit, has 
Eighteen years of research and field experience in spatial 
data and geoinformatics - Remote sensing, GIS and GPS - 
applications. To his credit, he has published twelve research 
papers in refereed journals - national and international - and international 
conferences and two papers are under peer review. He has also had the distinct 
honor of acting as Chairperson for a session on "Ecosystem and Bio-diversity" 
in an International conference held at Tsukuba University, Ibaraki, Japan, 
apart from participating many national level seminars, workshops and training 
programs. He has also been involved in consultancy service to UNESCO, 
New Delhi, and developed Computer Based Learning tutorial on Geology 
using VB as front end tool with various graphic utilities (A/V Support) under 
ICT program. He has been appointed as Lesson writer for M.Sc 
Geoinformatics covering Satellite Remote sensing and GIS by Annamalai 
University and has been acting as Resource person for Academic Institutions. 




208 http://sites.google.com/site/ijcsis/ 

ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 



Performance Analysis and Optimization of Lumped 
Parameters of Electrostatic Actuators for Optical 

MEMS Switches 



D.Mohana Geetha 
Department of Electronics and Communication 
Engineering 
Kumaraguru college of Technology 
Coimbatore641006, India. 



M.Madheswaran 

Center for Advanced Research 

Department of Electronics and Communication 

Engineering 

Muthayammal Engineering College, 

Rasipuram 637408 , India. 



Abstract - This Paper deals design and simulation of electro 
statically actuated clamped-clamped beam and cantilever beam 
using finite element analysis method (FEM). A detailed study and 
performance analysis for various bias voltages is provided in this 
paper. The pull in voltages for different dimensions of the beams and 
the natural or the dominant modes and the corresponding eigen 
values have been studied for different bias voltages. The displacement 
of the beam is also studied for various dimensions of the beam. The 
results were obtained for the length and the width of both clamped - 
clamped beam and cantilever beam through extensive simulations. 
The results obtained shows that pull in voltages varies from 2348V to 
772V and the natural frequencies vary from 102.92 KHz to 916.35 
KHz. 

Keywords- electrostatic, optical MEMS, clamp ed,cantilever, pull-in 
voltage. 



I. I. INTRODUCTION 

The development in the field of MEMS and its application in optical 
domain have increased the attention of research in the recent 
past.MEMS are found suitable for optical applications because these 
devices can be matched to optical wavelengths, and manufactured in 
high volume and high density arrays in the semiconductor 
manufacturing processes [1]. The inherent advantages of MEMS have 
started replacing the optical transmitting switches and tunable filters 
with MEMS actuators [2]. The most common process in all MEMS 
devices is actuation which affect the mechanical motion, forces, and 
work by a device or system on its surroundings in response to the 
application of a bias voltage or current. The most common types of 
actuators are electrostatic, thermal, magnetic, piezoelectric, shape 
memory alloys, and hydraulics. MEMS based devices are highly used 
for several applications like biomedical sensors, miniature biomedical 
instruments ,cardiac management systems, neuro stimulation ,engine 
and propulsion control, automotive safety, braking and suspension 
systems, telecommunication optical fiber components and switches, 
data storage systems, electromechanical signal processing and also for 
military applications. The drive mechanism of these devices includes a 
constant voltage source (voltage drive) or constant current source 
(current drive) to enable electrostatic actuation or capacitive sensing. 
Sazzadur Chowdhury, M. Ahmadi, W. C. Miller have 



demonstrated that the electrostatic MEMS devices can be implemented 
with less complexity[3]. Optical MEMS devices is an electro statically 
actuated micromechanical mirror [1]. 

Jin Cheng, Jiang Zhe, Xingtao Wu, K.R. Farmer, 
V.Modi, Lu Frechette,2004 [4] carried out the pull in analysis for 
cantilever beam and fixed- fixed beam and have calculated the travel 
ranges of free deformable actuators to 47.2% and 42% respectively. 
Ofir Bochobza -Degani, Eran Socher,Yael Nemirovsky [5] had 
modeled a general electrostatic actuator with a charge distribution in the 
dielectric coating and results were obtained for pull in voltages. Joseph 
I.Seeger and Bernhard E.boser [6] demonstrated and showed that the 
structure can move beyond the well known pull in limit at resonance 
Sazzadur Chowdhury, M. Ahmadi, W. C. Miller provided the 
comparative study of closed form methods for calculating the pull in 
voltage of electro statically actuated fixed- fixed beam actuators [3 ]. 

The electrostatic force associated with the constant voltage 
drive mode becomes nonlinear and gives rise to the well-known 
phenomenon of 'pull-in'. The pull-in phenomenon causes an electro 
statically actuated beam to collapse on the ground plane if the drive 
voltage exceeds certain limit depending on the device geometry. 
Accurate determination of the pull-in voltage is critical in the design 
process to determine the sensitivity, frequency response, instability, 
distortion, and the dynamic range of the device [3]. Since the 
determination of accurate pull in voltage and natural frequency is 
essential, extensive simulation is carried out by varying the dimensions 
of clamped -clamped beam and cantilever beam. 



II. MODELING OF ELECTROSTATIC ACTUATORS 

Electrostatic actuators have fast response and low power 
consumption. Application and release of forces take virtually the same 
time which is not the case of thermo actuation because of fast heating 
and slow cooling. Electro statically driven actuators are less sensitive to 
environmental conditions than others. They are two metal structures 
separated by an air gap. A bias voltage is applied between the metal 
structures, which results in a separation of charges between them. This 
produces an electrostatic force that can be used to decrease the gap 
between the plates as shown in the Fig 1. 



209 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 





Figure. l.Schmatic structure of parallel plate capacitor 



The two plates have an overlapping area of A and a spacing 
of d. The dielectric constant or relative electrical permittivity of 
the media between the two plates is denoted 8 r .The permittivity 
of the media is e = e r 8 ,where 8 is the permittivity of the 
vacuum[7]. 

The value of the capacitance C, between the two parallel 
plates is given by 



C = 



Q 
v 



(i) 



where Q is the amount of stored charge and V is the 
electrostatic potential. 



The electric energy stored in the capacitor is given by 
2 2C 



(2) 



According to Gauss's law, the magnitude of the primary 
electric field E is given by 



sA 



(3) 



The magnitude of the voltage is the electric field times the 
distance between two plates'd'. 

The capacitance of the parallel plate capacitor is 



c = Q= Q 



Q 



V E.d Q_ 
sA 



d 



eA 

d 



(4) 



dU 

dx 


1 

~2 


dC 
dx 



V 



(5) 



where x is the dimensional variable. 

If the plate moves, the gap between the plates changes and 
the magnitude of the force can be given as 



F = 



eu 



dx 



lsA jr2 1CV 

— -y 2 = 

2d 2 2 d 



(6) 



with normal dimension changed from x to d. 



The suspended plate is attracted towards the bottom 
plate due to the resultant electrostatic force. The suspended plate 
move towards the bottom plate until an equilibrium exits 
between them. The suspended plate makes contact with the 
bottom plate when there is maximum electrostatic force. 
The capacitance of the device over a range of motion can be 
used to characterize the electromechanical response of the 
device. 

An electrostatic actuator can be modeled as a variable 
capacitor suspended by elastic springs. An important design 
aspect for electrostatic actuators is to determine the amount of 
static displacement under a certain biased voltage as shown Fig 
2. The upper beam is supported by a mechanical spring with a 
force constant K m .Gravitational force can be neglected because 
the mass of the beams are very small. 



Movable 



F mechanical 




Figure.2.A coupled electromechanical model 



The capacitor can be used as an actuator to generate force or 
displacement .As a differential voltage is applied between the 
two parallel plates, an electrostatic attraction force is developed. 
The magnitude of the forces equals the gradient of the stored 
electric energy V s with respect to the dimensional variable .The 
magnitude of the force is 



When a voltage is applied, an electrostatic force F dectric is 
developed. The magnitude of F electric is given by 



electric 



IcV 2 

2 d 



(7) 



210 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



This force will tend to decrease the gap and gives displacement 
and mechanical restoring force. Under static equilibrium, the 
mechanical restoring force has an equal magnitude but opposite 
direction as the electrostatic force. 

The electrostatic force modifies the spring constant and the 
spring will be softened. The spatial gradient of the electric force 
is defined as an electrical spring constant and given by 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
From equation (8) and (12) 



K = 



dF 

electric 


= 


- 


U 2 J 


dd 



d 2 



(8) 



The electrostatic force at equilibrium when the beam is 
applied with a bias voltage is given by 



1 



electric 



= -sAV 2 (x +x) 2 =- 



:(x)V 2 






(9) 



The magnitude of the mechanical restoring force is given by 



F =-K x 

m echanical m 



(10) 



Equating the magnitudes of F r 

rearranging the terms, the displacement can be calculated as 



mechanical and F electric at X and 



-x- 



mechanical 



electric 



:(x)V 2 



K„ 



K m 2(xJ)K m 



(11) 



At a particular bias voltage, mechanical restoring force and 
the electrostatic force balance each other. The magnitude of 
electric force constant equals the mechanical force constant. 
The effective force constant of the spring is zero. The bias 
voltage invokes such a condition is called the pull in voltage 



'V p \ If the bias voltage is increased beyond V p 



the 



equilibrium position disappears. The electrostatic force 
continues to rise while the mechanical force increases linearly 
only. The two beams are pulled against each other and they 
make contact. This condition is called pull in or snaps in. 
At the pull in condition the magnitudes of electrical force and 
mechanical force and can be equated as 



V 2 =- 



2K m x(x + x ) 2 __ 2K m x(x + x f 



sA 



(12) 



The value of x is negative when the spacing between two 
electrodes decreases. The gradients of these two forces at the 
intersection point are equal and given by 



K\ = \K„ 



(13) 



K =■ 



cV 2 



■ = -2K m x(x + x ) 



(x + x o y 



The solution of x can be obtained as x = - 



(14) 



(15) 



From the equation (15), it may be concluded that the relative 
displacement of the parallel plate from its initial position is 
exactly one third of the original spacing at the critical pull-in 
voltage. 

The voltage at the pull in can be estimated from (12) and 
equation (15). 



The pull in voltage is given by 



2x n K„ 



v p=- 



1.5c 



(16) 



III. COMPUTATIONAL TECHNIQUE 

The clamped-clamped beam and cantilever beam actuators 
can be simulated and the performance in the electrical domain 
can be analyzed using Finite element method. The flow graph of 
the simulation is shown in Fig 3. The structure of the clamped - 
clamped and cantilever beam actuators are shown in Fig 4.The 
pull in voltage and the natural frequency of the proposed 
actuators can be simulated using ANSYS. The region between 
the two plates can be considered as the active region and the 
capacitance can be estimated by dividing the beam into meshes. 
Reduced order modeling of the coupled electrostatic system can 
be used to replace the electrostatic mesh for improving the 
execution time. 

The eigen value and the eigen vectors can be calculated 
using 

[K]{0 1 }=A i [M]{O i } (18) 

where [K] = structure stiffness matrix 
{Oi} = eigenvector 
A* = eigen value 
[M] = structure mass matrix 



211 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 



START 



INPUT Parameters 

Length, breadth, 

thickness of the beam 



Define the elements for 
electrostatic domain 



Mesh the Beam 



Extract the capacitance values 



Get the saved file as 
input file 



Define the capacitances, 
master nodes 



Apply element loads 



Extract the displacements of the 
beam due to element loads 



Save the output 
results file as (2) 




Apply the sweep 

voltage and run the 

sweep voltage 




No 



Save the result as 
file (3) 



Get the saved file 
(3) as input file 



Get the saved file as 
input file 



Extract the capacitance values 




Save the results 
file 



Stop 



Figuer.3.Flow chart for Pull in Yottaze and extracting capacitance .... . ........ 

2T2 nffp://sites. google.com/site/ijcsis/ 

ISSN 1947-5500 



The structure of the stiffness and mass matrix can be reduced 
by considering the set of degrees of freedoms (DOFs).Then 
using Guyan reduction technique the relationship can be 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
has been calculated for different beam width of clamped- 
clamped beam actuators and shown in Fig .6 



rewritten as 



[*]{£,} = A, M jj 
Where [k\ = reduced stiffness matrix (known) 



(19) 



M 



<f>i( = eigenvector (unknown) 
Xi = eigen value (unknown) 
= reduced mass matrix (known) 




" l - n-» 



i 



Figure 4(a).Clamped-Clamped beam 

" — — ., 1 

■ ■■ !_ ^-^ 1 




Figure 4(b) Cantilever beam 

Then the actual eigen value can be extracted using HBI 
(Householder-Bisection-Inverse iteration). Using FEM and the 
ANSYS , the pull in voltage and the natural frequency can be 
obtained to estimate the performance of both the actuators to 
find their capacity to use them in optical MEMS switches. 

V. RESULTS AND DISCUSSION 



The electrostatic actuators with clamped-clamped beam and the 
cantilever beam structures has been simulated using ANSYS 
and the parameters have been estimated using FEM 
analysis.The simulated output of the clamped-clamped beam 
and cantilever electrostatic actuators is shown in Fig 5(a) and 
5(b)respectively. The pull in voltage for various beam length 



It is seen that the pull in voltage decreases with beam length as 
well as beam width. However it is clear from the results that the 
variation in pull in voltage is much significant for the lower 
beam width. The pull voltage for cantilever beam with various 
beam width and beam length is shown in Fig. 7. It is seen from 
the Fig that the pull in voltage for the beam length 90 jam and 2 
|um beam width is 450V.It is found that at the same beam width 
and the beam length the pull voltage was estimated as 1 lOOv in 
clamped-clamped beam actuators. The natural frequency for 
both clamped -clamped beam and cantilever beam actuators are 
shown in Table. I 



It is seen from the table that the natural frequency is inversely 
proportional to the beam length in both the cases. The variation 
of natural frequency with the length and width of the cantilever 
beam is shown in the Fig .8 



The displacement of the beam for length L=85 |um for both 
clamped -clamped and cantilever beam in Fig.9.It is seen from 
the Fig that the displacement is much significant in cantilever 
beam for the beam with less width. 




Figure 5(a).Output of displaced clamped-clamped 
beam. 



213 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 




Figure 5(b).Output of displaced cantilever beam. 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 



TABLE I. 



NATURAL FREQUENCIES FOR CLAMPED AND CANTILEVER BEAMS 



2500 - 

<2 2000 - 
o 

> 

« 1500 - 

<D 
O) 
CO 

1 1000 - 
°- 500 - 

o - 








a 




- - B - - 1 5 urn width of 

Beam 

- ^>^ - 1 urn width of 

Beam 


Beam 

— *■ - 2 urn width of 
Beam 


Beam 


100 95 90 85 80 75 
length in um 







Beam 

Width in 

um 


Beam 

Length in 

urn 


Natural Frequencies (Hz) 


Clamped- 
clamped beam 


Cantilever beam 


15 


100 


0.10292E+08 


0.48965E+07 


10 


100 


0.10289E+08 


0.48960E+07 


5 


100 


0.10285E+08 


0.48957E+07 


2 


100 


0.10284E+08 


0.17325E+07 


1 


100 


0.10284E+08 


0.17325E+07 


15 


95 


0.10294E+08 


0.54232E+07 


10 


95 


0.11337E+08 


0.54226E+07 


5 


95 


0.11333E+08 


0.54222E+07 


2 


95 


0.11332E+08 


0.19193E+07 


1 


95 


0.11332E+08 


0.19193E+07 


15 


90 


0.12566E+08 


0.60395E+07 


10 


90 


0.12562E+08 


0.60389E+07 


5 


90 


0.12558E+08 


0.60384E+07 


2 


90 


0.12556E+08 


0.21381E+07 


1 


90 


0.12556E+08 


0.21381E+07 


15 


85 


0.1401 1E+08 


0.67670E+07 


10 


85 


0.14006E+08 


0.67662E+07 


5 


85 


0.14001E+08 


0.67657E+07 


2 


85 


0.14000E+08 


0.23965E+07 


1 


85 


0.13999E+08 


0.23965E+07 


15 


80 


0.15730E+08 


0.76340E+07 


10 


80 


0.15725E+08 


0.76330E+07 


5 


80 


0.15719E+08 


0.76324E+07 


2 


80 


0.15717E+08 


0.27046E+07 


1 


80 


0.15716E+08 


0.27046E+07 


15 


75 


0.17797E+08 


0.86784E+07 


10 


75 


0.17791E+08 


0.86773E+07 


5 


75 


0.91649E+07 


0.86765E+07 


2 


75 


0.17781E+08 


0.30762E+07 


1 


75 


0.17781E+08 


0.30762E+07 



Figure. 6 Pull in voltages of clamped-clamped 



800 

700 

600 - 
, 500 

400 - 

300 

200 - 
' 100 







— ■ — 1 5 |jm width of 
Beam 

— *. _ I0|jm width of 

Beam 

— x- - 5 um width of 

Beam 

— -o- - 2 |im width of 

Beam 

— n — 1 |jm width of 

Beam 



85 80 75 
Length in |im 



3.6 T 
3.4 
i.2 
3 
.8 
:.6 
!.4 

2.2 - 
2 




♦ cant ilever beam 



-clamped-clamped 
beam 



Figure.7 Pull in voltage of cantilever beam. 



Figure.9Variation of displacement of the beams 
for length L=85 um. 



214 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



VI. 



CONCLUSION 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
AUTHORS PROFILE 



The pull in voltages calculations was made using finite 
element method. ANSYS with Reduced order modeling was 
used to extract results quickly. Fringing field effects are also 
considered to yield accurate results. Even with the strongly 
coupled lumped transducers, convergence issues were 
experienced when applied to the difficult hysteric pull-in and 
release analyses. The cause of the problem can be attributed to 
the negative total system stiffness matrix and can be resolved 
using the augmented stiffness method. 



[1] 



[2] 



[3] 



[6] 



REFERENCES 



Jinghong Chen, Member, IEEE, Wendelin Weingartner, 
Alexi Azarov, and Randy C. Giles "Tilt- Angle 
Stabilization of Electro statically Actuated 
Micromechanical Mirrors Beyond the Pull-in Point" 
Journal of Microelectromechanical Systems, Vol. 13, No. 
6, December 2004. 

Marcel W. Pruessner, Student Member, IEEE, Kuldeep 
Amarnath, Madhumita Datta, Member, IEEE, Daniel P. 
Kelly, S. Kanakaraju, Ping-Tong Ho, and Reza Ghodssi, 
Member, IEEE " InP-Based Optical Waveguide MEMS 
Switches With Evanescent Coupling Mechanism" Journal 
of Microelectromechanical Systems, Vol. 14, No. 5, 
October 2005. 

Sazzadur Chowdhury, M. Ahmadi, W. C. Miller "A 
Comparison of Pull-in Voltage Calculation Methods for 
MEMS-Based Electrostatic Actuator Design" 1 st 
International Conference on Sensing Technology, 
November 21-23, 2005 Palmer ston North, New Zealand. 

[4] Jin Cheng, Jiang Zhe, Xingtao Wu, K.R.Farmer, V.Modi, 
Lu Frechette" Analytical and FEM Simulation Pull- in 
study on Deformable Electrostatic Micro Actuators" 
Nanotech 2002 vol.1 Technical proceedings of the 2002 
International conference on Modeling and simulation of 
Microsystems. 

[5] Ofir Bochobza -Degani, Eran Socher,Yael Nemirovsky" 
On the effect of residual charges on the pull- in parameters 
of electrostatic actuators "Sensors and 

Actuators,2002,Elsevier press. 

Joseph I.Seeger and Bernhard E.boser"Parallel Plate driven 
oscillations and resonant pull in"solid state sensor, actuator 
and Microsystems workshop, 2002. 

[7]. Chang Liu"foundations of MEMS"Illinois ECE 
series,Pearson International edition 2006. 




Dr. M. Madheswaran has obtained his Ph.D. degree in 
Electronics Engineering from Institute of Technology, 
Banaras Hindu University, Varanasi in 1999 and M.E 
degree in Microwave Engineering from Birla Institute 
of Technology, Ranchi, India. He has started his 
teaching profession in the year 1991 to serve his parent 
Institution Mohd. Sathak Engineering College, 
Kilakarai where he obtained his Bachelor Degree in ECE. He has served KSR 
college of Technology from 1999 to 2001 and PSNA College of Engineering 
and Technology, Dindigul from 2001 to 2006. He has been awarded Young 
Scientist Fellowship by the Tamil Nadu State Council for Science and 
Technology and Senior Research Fellowship by Council for Scientific and 
Industrial Research, New Delhi in the year 1994 and 1996 respectively. His 
research project entitled "Analysis and simulation of OEIC receivers for tera 
optical networks" has been funded by the SERC Division, Department of 
Science and Technology, Ministry of Information Technology under the Fast 
track proposal for Young Scientist in 2004. He has published 120 research 
papers in International and National Journals as well as conferences. He has 
been the IEEE student branch counselor at Mohamed Sathak Engineering 
College, Kilakarai during 1993-1998 and PSNA College of Engineering and 
Technology, Dindigul during 2003-2006. He has been awarded Best Citizen of 
India award in the year 2005 and his name is included in the Marquis Who's 
Who in Science and Engineering, 2006-2007 which distinguishes him as one of 
the leading professionals in the world. His field of interest includes 
semiconductor devices, microwave electronics, optoelectronics and signal 
processing. He is a member of IEEE, SPIE, IETE, ISTE, VLSI Society of India 
and Institution of Engineers (India). 

Mrs.D.Mohanageetha obtained her B.E degree from 
Bharathiar University in 1996, M.E Degree in 
Communication systems from Madurai Kamaraj 
University, Madurai in 2000. She has started her 
teaching profession in the year 1996 at V.L.B 
Janakiammal College of Engineering and 
Technology, Coimbatore. At present ,she is an 
Assistant Professor in Department of Electronics 
and Communication, Kumaraguru college of Technology, Coimbatore, Tamil 
nadujndia . She has published 10 research papers in International and national 
conferences. She is a part time Ph.D research scalar in Anna University 
Chennai. Her areas of interest are Optical networking,, MEMS,EMI/EMC and 
Image processing. She is a life member of ISTE and member of IEEE. 




215 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) InternationalJournal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 



Modeling of Human Criminal Behavior using 

Probabilistic Networks 



Ramesh Kumar Gopala Pillai 

Research Scholar 

R.V. Center for Cognitive Technologies 

Bangalore, India 



Dr. Ramakanth Kumar .P 
Professor 
R.V. Center for Cognitive Technologies 
Bangalore, India 



Abstract — Currently, criminal's profile (CP) is obtained from 
investigator's or forensic psychologist's interpretation, linking 
crime scene characteristics and an offender's behavior to his or her 
characteristics and psychological profile. This paper seeks an 
efficient and systematic discovery of non-obvious and valuable 
patterns between variables from a large database of solved cases via 
a probabilistic network (PN) modeling approach. The PN structure 
can be used to extract behavioral patterns and to gain insight into 
what factors influence these behaviors. Thus, when a new case is 
being investigated and the profile variables are unknown because 
the offender has yet to be identified, the observed crime scene 
variables are used to infer the unknown variables based on their 
connections in the structure and the corresponding numerical 
(probabilistic) weights. The objective is to produce a more 
systematic and empirical approach to profiling, and to use the 
resulting PN model as a decision tool. 



Keywords-component; Modeling, criminal profiling, criminal 
behavior, probabilistic network, Bayes Rule 



I. 



Introduction 



Modeling human criminal behavior is challenging due to 
many variables involved and the high degree of uncertainty 
surrounding a criminal act and the corresponding 
investigation. Probabilistic graphs are suitable modeling 
techniques because they are inherently distributed and 
stochastic. In this paper, the system variables comprising the 
PN are offender behaviors and crime scene evidence, which 
are initialized by experts through their professional experience 
or expert knowledge. 

The mathematical relationships naturally embedded in a set of 
crimes [3, 4, 8] are learned through training from a database 
containing solved criminal cases. The PN model is to be 
applied when only the crime scene evidence is known to 
obtain a useable offender profile to aid law enforcement in the 
investigations. A criminal profile is predicted with a certain 
quantitative confidence. 

The PN approach presented here seeks to build on the ideas of 
behavior correlations in order to obtain a usable criminal 
profile when only crime scene evidence is known from the 
investigation. 



This paper proposes a systematic approach for deriving a 
multidisciplinary behavioral model of criminal behavior. The 
proposed offender behavioral model is a mathematical 
representation of a system comprised of an offender's actions 
and decisions at a crime scene and the offender's personal 
characteristics. 

The influence of the offender traits and characteristics on the 
resulting crime scene behaviors is captured by a probabilistic 
graph or PN that maps cause-and-effect relationships between 
events, and lends itself to inductive logic for reasoning under 
uncertainty [1]. The use of PNs for CP may allow 
investigators to take into consideration various aspects of the 
crime and discover behavioral patterns that might otherwise 
remain hidden in the data. The various aspects of a crime 
include a victimology assessment (victim's characteristics, 
e.g., background characteristics, age, gender, and education), 
crime scene analysis (evidence from the crime scene, e.g., 
time and place where the crime occurred), and a medical 
report (autopsy report, e.g., type of non-deadly and deadly 
lesions and signs of self defense). 

The PN approach to criminal profiling is demonstrated by 
learning from a series of crime scene and offender behaviors. 
The learning techniques employed in this modeling research 
are evaluated on a set of validation cases not used for training 
by defining a prediction accuracy based on the most likely 
value of the output variables (offender profile) and its 
corresponding confidence level. 

II. APPROACH 

To start with, a graphical model of offender behavior is 
learned from a database of solved cases. The resulting CP 
model obtained through training is then tested by comparing 
its predictions to the actual offenders' profiles. 

Let the database sample space = D, 

Let D consist of 'd' solved cases {Ci, ...,Cd}, 

where Ci is an instantiation of X, which is randomly 
partitioned into two independent datasets such as a training set 



216 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



T and a validation set V, such that D = T[/V. The variables in 
X are partitioned as follows: the inputs are the crime scene 
(CS) variables X 1 (evidence) for X 1 = (X\ , ...,x\ ), and the 
outputs are the offender (OFF) variables comprising the 
criminal profile X°, for X° = (X° l , ...,X° m ), where (X 1 ,X°) s 
X. 



The PN model is learned from T, as explained later, and it is 
tested by performing inference to predict the offender 
variables (OFF) in the validation cases V. An offender profile 
is estimated based on crime scene evidence, with a prediction 
being the most likely value of a particular offender variable. 
During the testing phase, the predicted value of X°i , denoted 
by x P i a where a=l or 2 for a binary variable, is compared to the 
observed state x° ijb obtained from the validation set V, where 
b=l or 2. An example of an offender variable is "gender", with 
states "male" and "female". The overall performance of the 
PN model is evaluated by comparing the true (observed) states 
x°i b to the predicted output variable values x p ia in the 
validation cases. This process tests the generalization 
properties of the model by evaluating its efficiency over V. 

III. VARIABLES CONSIDERED 

The relevant categories of variables that have emerged from 
the criminal profiling research as selected by investigators, 
criminologists, and forensic psychologists are described as 
follows: 

• Crime Scene Analysis (CSA): CSA variables are systematic 
observations made at the crime scene by the investigator. 
Examples of CSA variable pertain to where the body was 
found (e.g., neighborhood, location, environment 
characteristics), how the victim was found (e.g., the body was 
well-hidden, partially hidden, or intentionally placed for 
discovery), and the correlation between where the crime took 
place and where the body was found (e.g., the body was 
transported after the murder). 

• Victimology Analysis (VA): VA variables consist of the 
background characteristics of the victim independent of the 
crime. For example, VA variables include the age, sex, race, 
education level, and occupation of the victim. 

• Forensic Analysis (FA): FA variables rely on the medical 
examiner's report that deals with the autopsy. Examples of this 
are time of death, cause of death, type of non-lethal wounding, 
wound localization, and type of weapon that administered the 
wounds. 

The set of CP variables used in this paper were defined in 

previous research [4, 5, 7, 8]. The selection criteria for 

variable selection [6] are: 

(i) Behaviors are clearly observable and not easily 

misinterpreted 

(ii) Behaviors are reflected in the crime scene, e.g., type of 

wounding, and 

(iii) Behaviors indicate how the offender acted toward and 

interacted with the victim 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 
e.g., victim was bound/gagged, or tortured. Some crime scene 
(CS) variables describing the observable crime scene and 
offender (OFF) variables describing the actual offender were 
selected based on the above criteria. Examples of the CS 
variables are multiple wounding to one area, drugging the 
victim, and sexual assault. Examples of the offender variables 
include prior offenses, relationship to the victim, prior arrests, 
etc. The variables all have binary values representing whether 
the event was present or absent. 



IV. TRAINING THE MODEL 

The basic schematic of the training software, including the 
validation process, is shown in Figure 1, where P h is the 
proposed PN and P opt is the trained (or optimized) PN. The 
software is intended to aid law enforcement in the 
investigation of violent crimes. Because the cases are unsolved 
and only the crime scene inputs are known, the criminal 
profiling software consists of a trained PN model that has been 
previously trained and validated with D. Also, the model has 
the potential to be updated by means of an incremental 
training algorithm when additional cases are solved by the 
police. Thus, p tramed consistently reflects the model of an 
evolving criminal profile over time. 




T+ 



v 



x° 



Model 

Evaluation 



Error 



... ; 



V" 



Criminal Profiling Software 
Tigure 1: CP model Training and validation ^ofhvare 



V. DATA BASE OF SOLVED CASES 

A set of single offender/single victim homicides was collected 
by psychologists from solved homicide files of the British 
police [4, 6]. In order to examine the aggressive behavioral 
patterns of a particularly violent offense, the criteria for case 
selection is: single offender/single victim homicide cases; a 
mixture of domestic (where the victim and offender were 
known to each other (e.g., family member, spouse, co-worker) 
and stranger (the offender is unknown to the victim, thus they 
had no previous links to each other) cases; offenders are adults 
at least 17 years of age, as defined by the court system. 
Excluded from the sample were cases when the cause of death 
was not aggressive or extremely intentional. Homicides by 
reckless driving are not included due to the lack of 
interpersonal interaction between the offender and victim. 



VI. SAMPLING 

A simulation set is built to produce an artificial CP database to 
study the PN learning and inference capabilities. This included 
a more extensive list of crime scene, offender characteristics 



217 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) InternationalJournal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 



and multiple- valued variables. A PN is used to simulate a set 
of cases where the crime scene and offender variables can be 
chosen by the user. An initial structure thus relating the 
variables and the corresponding initial probabilistic parameters 
O are declared based on the prior knowledge, through 
experience, or by sampled statistics. Cases are simulated by 
feed forward sampling, where variables are sampled one at a 
time in order from top-level variables (variables without 
parents), to the mid-level variables (variables with both 
parents and children), ending with the bottom-level variables 
(children variables with parents only). For each variable, the 
discrete conditional prior probabilities in vector form are given 
as: 

[P(X U \ 7Ci) t (X U \ 7Ci), , (X itri \ 7Ci)J 

where ri is the maximum state for X { and 7ri disappears if X { is a 

top-level variable. A value v,- is drawn from a uniform 

continuous distribution between '0' and 'V and the 

conditional prior probability vector as a vector of ranges 

becomes 

[P(x u \ 7c t ),P(x u \ n t )+ P(x h2 \ ttO, , 



/ j (xjj\ 7Ti)J which refers to 



Xi= 



f x u if 0<v<P(x u |ti 1 ) 

X 1;2 if P(X U |71 1 )<V< / .PCXijlTCj) 



< 



Xiri if / j P(x u \k, 



,)<V<1 



(1) 




P(X 2 IX,) 



P(X 3 |Xi) 



Figure 1\ Three-nodal model 

To simulate a set of cases for the system represented by the 
three-nodal model illustrated in Figure 2, the variables are 
ordered as (X b X 2 , X 3 ), where Xi is the parent of X 2 and X 3 , 
and Xi = (x u , x 1;2 ), X2 = (x 2J , x 2?2 ) and X3 = (x 3;1 , X3 52 ). 
Starting with X h it has three possible states with the prior 
probabilities P(x u ) = 0.2, P(x 12 ) = 0.5, and P(x 13 ) = 0.3, 
which becomes a vector [0.2, 0.7, 1] referring to: 



Xi= i 



x u if 0<vi<0.2 

x u if 0.2<vi<0.9 

x u if 0.9<vi<l 



(2) 



If Vi = 0.11 which makes X x = x 1;lj and the Conditional 
Probability Table (CPT) for X2 is listed in Table 1. 



x 2 


PfeilXi) 


FtaJXi) 


Xi = i i.i 


0.2 


0.8 


Xi=I 1.2 


0.9 


0.1 



Table 1 Conditional probability Tab It 
Then the conditional prior probability vector of ranges for a 
newly generated v 2 becomes 



X 2 



x 2 ,iif 0<v 2 <0.2 
x 22 if 0.2<v 2 < 1 



(3) 



X 3 is sampled following the same procedure as X x and X 2 . 
This is repeated until the desired number of cases as specified 
by the user is reached. The Matlab function utilized for the 
sampling exercise is sample bnet in the Bayes Net Toolbox 
[2]. 

VII. PN PREDICTIONS AND ACCURACY 

When a PN model of offender behavior on the crime scene is 
learned from solved cases, it is implemented on a set of solved 
validation cases in order to test the trained model's 
performance. Performance is tested through probabilistic 
inference. Inference is the process of updating the probability 
distribution of a set of possible outcomes based upon the 
relationships represented by the PN model and the 
observations of one or more variables. With the updated 
probabilities, a prediction can be made from the most likely 
value of each inferred variable. Thus, in order to test the 
trained model, only the crime scene evidence is inserted into 
the model, with the predicted offender profile being compared 
to the actual offender characteristics. Because this is a 
probabilistic model, a certain confidence accompanies the 
offender variable predictions. 

VIII. CONCLUSIONS 

This paper presents an approach for deriving a network model 
of criminal profiling that draws on knowledge-based systems 
and on fields of criminology and offender profiling. 
Implementing probabilistic networks makes it possible to 
represent multidimensional interdependencies between all 
relevant variables that have been identified in previous 
research as playing a role in determining or reflecting the 
behavior of offenders at the crime scene. Hence, a valid 



218 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 
network model can be used to predict unknown variables characteristics from crime scene behavior. Scandinavian 
composing an offender profile based on the variables observed Journal of Psychology, 44: 107-1 18, 2003. 
from the crime scene. 



References 

[1] R. Cowell. Introduction to inference for Bayesian 

networks. In M.I. Jordan, editor, Learning in Graphical 

Models, pages 9-26. 1998 

[2] K. Murphy. How To Use Bayes Net Toolbox. Refer 

http://www.ai.mit.edu/ murphyk/Software/BNT/bnt.html, 

2004. 

[3] R.K. Ressler, A. Burgess, and J.E. Douglas. Sexual 

Homicide: Patterns and motives. New York: Lexington Books, 

1988. 

[4] C.G. Salfati. Profiling homicide: A multidimensional 

approach. Homicide Studies,4:265-293, 2000. 

[5] C.G. Salfati. Greek homocide, a behavioral examination of 

offender crime-scene actions. Homicides Studies, 5(4):335- 

362, November 2001. 

[6] C.G. Salfati. Offender interaction with victims in 

homicide: A multidimensional analysis of crime scene 

behaviors. Journal of Interpersonal Violence, 18(5):490- 512, 

2003. 

[7] C.G. Salfati and F. Dupont. Canadian homicide: An 

investigation of crime scene actions. Homicide Studies, In 

Press for 2005. 

[8] P. Santtila, H. Hakkanen, D. Canter, and T. Elfgren. 

Classifying homicide offenders and predicting their 




under Kuvempu 
papers in reputed 



AUTHORS PROFILE 



Mr. Ramesh Kumar Gopala Pillai is a 

post graduate in Information Technology 
from Kuvempu University. He also 
possesses a MBA degree in Banking and 
Finance. Currently he is pursuing his 
research in the area 'Modeling of Human 
Criminal Behavior and its Implications' 
University. He has published a couple of 
International conferences and journals. 



jprflfc Dr. Ramakanth Kumar P. is a 

doctorate from Mangalore University. 

He has successfully guided research 

scholars to their PhD degree. He has 

executed several research projects as 

Principal Investigator for Government 

and Research establishments like 

DRDO, ISRO etc. He has several 

publications in reputed refereed International Journals and 

Conferences. Currently he is working as a Professor at R V 

Center for Cognitive Technologies, Bangalore. 




219 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 



Reaching the unreached 

A Role of ICT in sustainable Rural development. 



Mr.Nayak S.K. 

Head, Dept. of Computer Science 

Bahirji Smarak Mahavidyalaya, 

Basmathnagar, Dist.Hingoli. (MS), 

India 



Dr.S.B.Thorat 

Director, 

Institute of Technology and Mgmt. 

Nanded, Dist.Nanded. (MS), 

India 



Dr.Kalyankar N. V. 

Principal 

Yeshwant Mahavidyalaya, Nanded 

Nanded (MS) 

India 



Abstract — 
We have seen in last few decades that the progress of 
information technology with leaps and bounds, which have 
completely changed the way of life in the developed 
nations. While internet has changed the established working 
practice and opened new vistas and provided a platform to 
connect, this gives the opportunity for collaborative work space 
that goes beyond the global boundary. 

ICT promises a fundamental change in all aspects of our 
lives, including knowledge dissemination, social interaction, 
economic and business practices, political engagement, media, 
education, health, leisure and entertainment. 

In India ICT applications such as Warana, Dristee, Sari, 
Sks, E-Chaupal, Cybermohalla, Bhoomi, E-Mitra, Deesha, 
Star, Setu, Friends, E-Seva, Lokmitra, E-Post, Gramdoot, 
Dyandoot, Tarahaat, Dhan, Akshaya, Honeybee, Praja are in 
functioning for rural development. 

ICT offers an opportunity to introduce new activities, new 
services and applications into rural areas or to enhance existing 
services. With more than 70% of the Indian population living 
in rural areas and earns its live hood by agriculture and allied 
means of income. 

ICTs can play a significant role in combating rural 
and urban poverty and fostering sustainable development 
through creating information rich societies and supporting 
livelihoods. If ICTs are appropriately deployed and realize the 
differential needs of urban and rural people, they can become 
powerful tools of economic, social and political 
empowerment 

This paper introduces the application of ICT for rural 
development. The paper aims at improving the delivery of 
information to rural masses such as: technology information, 
marketing information, and information advice. This paper 
focuses digital divide and poverty eradication, good 
governance and the significance of internet for rural 
development. The paper concludes that ICTs offer the 
developing country, the opportunity to look ahead several 
stages of rural development by the use of internet. Effective 
use of ICT can demolish geographical boundaries and can 
bring rural communities closer to global economic systems 
and be of meaningful help to the underprivileged. 



Key words-ICT,RD, digital divide, ,e-governance,procurement, 
po verty alleviation, empowement. 



I. 



Introduction 



A. ICT Defined 



ICT can be interpreted broadly as "technologies that 
facilitate communication and the processing and transmission 
of information by electronic means." 

B. ICT and Development 

According to Nelson Mandela, "Eliminating the 
distinction between information rich and information poor 
countries is critical to eliminating other inequalities between 
North and South and to improving the quality of life of all 
humanity." (Mandela, 1998). 

C. Rural India 

"Just as the whole universe is contained in the Self, so is 
India contained in the villages"... This has been said by 
Mahatma Gandhi, the father of our nation and the visionary 
architect of India's rural development programme. The villages 
epitomize the soul of India. Rural India reflects the very 
essence of Indian culture and tradition. 



II. 



Internet 



Internet is another great achievement of ICT. Internet is one 
of the technologies available for global resources and 
information haring. Internet has become a common resource of 
the whole mankind, inquiring and sharing information become 
easier than ever. With Internet, geographical distance and state 
borders are eliminated. 

The Internet can be a powerful democratizing force, 
offering greater economic, political and social participation to 
communities that have traditionally been undeserved and 
helping developing nations meet pressing needs. It needs to 
ensure that everyone has a chance to share in the benefits of the 
Digital Age, information technology. 

Internet can be a tool for rural development. 



220 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



TABLE I. 


World-wide Internet Users Penetration 


Country 


Users 


Population 


Percentage 


(I 


n Millions) 


(In Millions) 


(As on 07/12/09) 


India 


81.00 


1156.89 


7.00 % 


USA 


227.71 


307.21 


74.10% 


Japan 


95.97 


127.07 


75.52 % 


China 


360.00 


1338.61 


26.30 % 


UK 


46.68 


61.11 


76.40 % 


Germany 


54.22 


82.32 


65.90 % 


Canada 


25.08 


33.48 


9.80 % 


France 


43.10 


62.15 


69.30 % 


Australia 


17.03 


21.26 


61.00% 


Russia 


45.25 


140.04 


32.31% 


Indonesia 


30.00 


240.27 


12.50 % 



www.internetworldstats.com/stats.htm 



Above table depicts that there is need to promote internet 
uses in developing country like India. 

III. Impact of ICT on Society 

History has seen a move from the agricultural society, 
through the industrialized society and to the information 
society. Society is better informed as a result of developments 
in ICT. Now a day ICT is used in daily practices. For ex. 
training, education and entertainment these help society to a 
great extent. There is impact of increased availability of 
information using ICT on a variety of public services. 

A. Digital Divide 

ICTs threaten to expand the already wide socio-economic 
gap between urban and rural populations in developing 
countries. Simultaneously offering opportunities to reduce it. 
ICT implementation can be successfully adapted for the 
development of challenges faced by rural communities. Such 
implementations can be evaluated so that the often unexpected 
and desirable results that emerge can be revealed and 
accounted for. 

B. E-Governance and Empowerment 

The poverty can be adequately addressed by effective use 
of e-governance and ICT application in environmental 
management. Improved governance by using ICT can have 
direct impact in reducing poverty and improving the 
environment. 

ICT can contribute fostering empowerment and 
participation by making government processes more efficient 
and transparent by encouraging communication and 
information sharing among rural and marginalized people. 

C. ICT and Agriculture 

The vast majority of poor people lives in rural areas and 
derives their livelihoods directly or indirectly from agriculture. 
ICTs can deliver useful information to farmers about 
agriculture like crop care and animal husbandry, fertilizer and 
feedstock inputs, pest control, seed sourcing and market prices. 



(IJCSIS) International lournal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
The ICT application in agriculture sector is still very 
limited in India, and full range of potential benefits that such 
technology can provide us yet to be realized. 

D. ICT and Poverty Alleviation 

ICT applications in developing countries are often part of 
an overall strategy for economic growth. The role of ICT in 
poverty reduction is not limited to reducing income poverty, 
but also includes non-economic dimensions - in particular 
empowerment. 

E. ICT and Health 

Health care is one of the most promising areas for poverty 
alleviation. ICTs are being used in India to facilitate remote 
consultation, diagnosis and treatment. 

Delivering health care with ICTs enables health care 
professionals and institutions to address the critical medical 
needs of rural communities, especially those in remote 
locations and those that lack qualified medical personnel and 
services. 



F. ICT for Education 

Moreover, appropriate use of ICTs in the classroom fosters 
critical, integrative and contextual teaching and learning; 
develops information literacy (the ability to locate, evaluate 
and use information). Thus, it improves the overall efficiency 
of the delivery of education in schools and educational 
management institutions at the national, state/provincial and 
community level. 

The use of ICTs in education aims to improve the quality of 
teaching and learning as well as democratize the access to 
education. 

G. ICT for Economic Development 

Information and Communication Technology has a vital 
role in connecting the rural community to outside world for 
exchange of information, a basic necessity for economic 
development. Effective use of ICT can demolish geographical 
boundaries and can bring rural communities closer to global 
economic systems and be of meaningful help to the 
underprivileged. 

H. Employment Opportunities 

Poor people in rural localities have lack of opportunities for 
employment because they often do not have access to 
information about them. One use of ICTs is to provide on-line 
services for job placement through electronic labor exchanges 
in public employment service or other placement agencies. 

IV. CHALLENGES AND ISSUES 

The barriers that currently limit the development of rural 
areas include Distance barriers: to access to administrative and 
governmental structures, Economic barriers: to access to 
wider business and labor markets, Social barriers: to 
information, education facilities, health and social services, 



221 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 
Information barriers : many rural areas and their amenities are remote locations and those that lack qualified medical 

undiscovered, unknown for the outer world. personnel and services. 



For successful implementation of E- rural development it is 
necessary to define ICT requirements and issues for rural areas 
like vision, policy, awareness, technology, Infrastructure, 
services, applications. It is extremely important to focus on all 
aspects of rural development. For introduction and 
implementation of E-rural policy it is necessary to establish 
cooperation of different sectors of government and on national 
level to introduce coherent E-rural policy, establish formal 
platform, supporting exchange information and knowledge 
which will join researchers, developers, regional and local 
government and which could coordinate research and 
implementation activity and which can also support E-rural 
policy 

Information-and-ICT initiatives are political. The 
effectiveness and potential of ICTD initiatives can be inhibited 
or circumscribed by national and/or local power relations. 
Political awareness and analysis is an important aspect of 
ICT4D planning at all levels. 

Harnessing ICTs for human development requires 
awareness-raising and constituency-building across all levels of 
society. 

The challenge for governments is to ensure the 
convergence of their initiatives and those taken up by various 
donors, multilaterals, NGOs and other organizations and to 
address the digital divide. 

V. Discussion / Conclusions 

A. Agriculture 

• The farmers are also not able to know about the prices 
prevailing in other markets, as the Market Committees 
are able to disseminate information mostly in respect of 
their own markets. 

• The availability of prompt and reliable market 
information about what is happening in the market, 
what quantities are arriving and what prices are quoted 
for different commodities considerably improves the 
decision making capability of the farmers and 
strengthens their bargaining power. 

B. Education 

• With introduction of ICT based education at school 
level our children and youngsters will grow as 
"Computer kids". Their exposure will get increased 
due to which the knowledge level will get definitely 
improved. 

• It enhances the quality of education. 

C. Health Services 

• Delivering health care with ICTs enables health care 
professionals and institutions to address the critical 
medical needs of rural communities, especially those in 



• General knowledge and awareness regarding health 
can be promoted among the people. 

D. Commerce and Trade 

• Still the benefits are not percolating down to the 
farmers, as they are unable to plan their strategies for 
sale of their produce at remunerative prices, in the 
absence of correct and timely market information and 
advice about arrivals, prices, market trend, etc. 

• In terms of market opportunities emerging agricultural 
technologies are increasingly information intensive 
and the rural poor must now cope with increasingly 
sophisticated input and output markets. 

E. Local Governance and Community life 

• ICT can empower rural communities (particularly 
marginalized group) and give them "a voice" which 
permits them to contribute to the development process. 

• With ICT government can serve the public better. 
Computers and the Internet make it possible for 
government to contact and transact with lots of people 
at the same time. 

VI. Suggestions / Recommendations 

• ICT should be harnessed for the benefit of ordinary 
farmers. 

• ICT policy is the most important factor on the 
introduction of information communication technology 
to the rural mass with a view to empower them by 
providing all types of information and communication 
with regard to their day to day life. 

• Awareness building on ICT usage should be conducted 
to the rural mass through the institutions such as gram 
panchayat, government office, schools, ICT Centers etc. 

• Government has to encourage the software developers 
to develop software packages in their respective mother 
tongue. 

• While the dedicated and qualified English teachers are 
appointed to the schools to enhance the English 
knowledge. Computer knowledge has to be given in 
their mother tongue with the support of English upto the 
students are able to understand and handle the 
computers. 

• Established cyber cafes in rural areas to facilitate the 
rural people who cannot purchase computers by 
themselves. To encourage the private sector to establish 
cyber cafe to provide ICT services in the rural areas. 

• It is necessary to open dialogue and professional 
discussions to create awareness on ICT in the rural 
areas. 



222 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. I, 2010 



To support and facilitate to develop locally relevant 
contents for the Internet. 

Significant attention needs to pay for the development 
of business models for content creation in the rural 
computing context. 

VII. Strategies 
Adopt one village one computer scheme. 

Start Community learning and Information Centers 
(CLIC) centers. 

Establish Market Information Centers in remote areas. 

Establish Tele Centers in remote areas. 

Encourage the use of computers and Internet in rural 
areas. 

Establish Information Technology (IT) parks in remote 
areas. 

Government has to reduce taxation of ICT-related 
components, products and services. 

Establish partnerships with NGOs engaged in awareness 
and innovative for ICT4RD. 

Explore the use of Free and Open Source Software 
(FOSS). 

Explore the use of local language software's. 

Promote the benefits of ICTs to private sector and 
academic institutions, and encourage computerization. 

Begin basic ICT skills workshops for all rural students 
at tertiary level. 

Encourage ICT awareness programmes, especially 
among primary and secondary school students in rural 
areas. 

Promote ICT-related courses at university/college level 
and expand the base of supportive certificate and 
diploma level at college level. 

Connect schools, universities and research centers to 
national and international distance education facilities, 
national and international databases, libraries, research 
laboratories and computing facilities. 

Encourage corporations to appreciate ICT competent 
staff and conduct/sponsor ICT training for staff 
members/professionals. 

Encourage assessment and promotion of civil servants 
to include ICT competency. 

Encourage ICT4D research and development and 
partnership with the private sector and international 
educational/research centers. 



VIII. The Road Ahead 

Rural Development forms an important agenda of the 
Government. However, the uptake of e-governance in the Rural 
Development sector has been relatively slow. The main reasons 
for this are poor ICT infrastructure in rural areas, poor ICT 
awareness among agency officials working in rural areas and 
local language issues. Efforts are, however, on to extend 
infrastructure up to village level. Already, many states have 
gone ahead to provide connectivity up to block level. This has 
helped in taking the e-governance efforts further closer to the 
people. 

The important requirement of establishing infrastructure in 
rural areas is now being taken up as a high-agenda project after 
the President of India envisioned the idea of providing urban 
amenities in rural areas (PURA). PURA (Provision of urban 
amenities in Rural Areas) has been conceived as a scheme 
under MoRD and envisages to achieve its objective by bridging 
the various kinds of divide that exists between rural and urban 
areas by providing four major kinds of connectivity to rural 
areas: physical (road, power), electronic (telecommunication, 
internet), knowledge and market. With the provision of such 
connectivity, it is hoped that the benefits of e-governance in the 
Rural Development sector would reach its true beneficiaries. 

Crucial success factors to realize this dream are strong 
political & administrative will, Government Process Reform, 
capacity building of provider (government functionaries) and 
consumer (rural citizens), utilization of ICTs as a medium to 
share information and deliver services that are demand-driven 
and people-centric. 

Acknowledgment (Heading 5) 

We are thankful to Hon. Ashok Chavan (Chief Minister, 
Maharashtra) India, Society members of Shri. Sharada Bhawan 
Education Society, Nanded. Also thankful to Shri. Jaiprakash 
Dandegaonkar (Ex- State Minister, Maharashtra), Society 
members of Bahiri Smarak Vidyalya Education Society, Wapti 
for encouraging our work and giving us support. 

Also thankful to our family members and our students. 
References 

[I] Annual report, 2002-2003, Ministry of Rural Development Government 
of India. 

[2] Data at Nasscom www.nasscom.org 

[3] E-Seva, an information brochure of Department of Information 
Technology and communications 200 1 . 

[4] http://ruralinformatics.nic.in 

[5] Governance for sustainable human development UNDP 
[6] What is good governance UN-ESCAP 
[7] Keith Yeomans ICTs in India, September 2001 . 

[8] Roger Harris, "ICT for poverty Alleviation Framework", December 
2002 

[9] 'Information and communications Technology for Development.' A 
source book for Parliamentarians. 

[10] Rural Informatics in India - An approach paper. 

[II] Information for Development (i4d) Vol. II No. 12 December 2004. 
[12] www.i4donline.net 

[13] http : //natfm. ac . in/j ournal/rural . doc 



223 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. I, 2010 
[14] Dr Yusaf Samiullah & Mr. Srinivasa Rao, "Role of ICTs in Urban and 
Rural Poverty Reduction". 

[15] Journal of the Eighth National Conference on e-Governance 3-5 
February, 2005. 

[16] Bastavee Barooah _October 2003, Significance of Internet Access for 
Rural India. 



AUTHORS PROFILE 





Dr. N.V.Kalyankar 

Principal 

Yeshwant Mahavidyalaya, Nanded (Maharashtra) 

Completed M.Sc. (Physics) from Dr.B.A.M.U, Aurangabad. In 
1980 he joined as a lecturer in department of physics at Yeshwant 
Mahavidyalaya, Nanded. In 1984 he completed his DHE. He 
completed his Ph.D. from Dr.B.A.M.U, Aurangabad in 1995. From 
2003 he is working as a Principal to till date in Yeshwant 
Mahavidyalaya, Nanded. He is also research guide for Physics and 
Computer Science in S.R.T.M.U, Nanded. He is also worked on 
various bodies in S.R.T.M.U, Nanded. He also published research 
papers in various international / national journals. He is peer team 
member of NAAC (National Assessment and Accreditation 
Council, India). He published a book entitled "DBMS concepts and 
programming in FoxPro". He also got "Best Principal" award from 
S.R.T.M.U, Nanded in 2009. He is life member of Indian National 
Congress, Kolkata (India). He is also honored with "Fellowship of 
Linnean Society of London (F.L.S.)" on 1 1 November 2009. 



S.K.Nayak 

M.Sc. (Computer Science), D.B.M, B.Ed. 

He completed M.Sc. (Computer Science) from S.R.T.M.U, 
Nanded. In 2000 he joined as lecturer in Computer Science at 
Bahirji Smarak Mahavidyalaya, Basmathnagar. From 2002 he is 
acting as a Head of Computer Science department. He is doing 
Ph.D. He attended many national and international conferences, 
workshops and seminars. He is having 2 international publications. 
His interested areas are ICT, Rural development, Bioinformatics. 




Dr.S.B.THorat 

M.E. (Computer Science & Engg.) 

M.Sc. (ECN), AMIE, LM-ISTE, Ph.D. (Comp.Sci. & Engg.) 

He is having 24 years teaching experience. From 2001 he is 
working as a Director, at ITM. He is Dean of faculty of Computer 
studies at Swami Ramanand Teerth Marathwada University, 
Nanded (Maharashtra). Recently he is completed his Ph.D. He 
attended many national and International conferences. He is having 
7 international publications. His interested area are AI, Neural 
network, Data mining, Fuzzy systems, Image processing. 



224 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) InternationalJournal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 



A proof Procedure for Testing 
Membership in Regular Expressions 



Keehang Kwon and Hong Pyo Ha 
Dong-A University Department of Computer Engineering 
Busan, Republic of Korea 



Jiseung Kim 
Kyung-IL University Department of Industrial Engineering 
Daegu, Republic of Korea 



Abstract 

We propose an algorithm that tests membership for 
regular expressions and show that the algorithm is 
correct. This algorithm is written in the style of a sequent 
proof system. The advantage of this algorithm over 
traditional ones is that the complex conversion process 
from regular expressions to finite automata is not needed. 
As a consequence, our algorithm is simple and extends 
easily to various extensions to regular expressions such 
as timed regular expressions or regular languages with 
the intersection. 

Key words: regular expressions, proof theory, linear 
logic, algorithm. 

1. Introduction 

Since its introduction, regular expressions [2,5] have 
gained much interest for applications such as text search 
or compiler components. One important question is, 
given a string w and a regular expression r, to decide 
whether w is in the set denoted by r. Testing membership 
in a regular expression has traditionally concentrated on 
converting a regular expression to finite automata. Such 
a conversion technique is unsatisfactory for at least two 
reasons. First, the conversion process itself requires a lot 
of extra overloads. Second, the conversion technique has 
only a limited number of applications and does not 
extend well to various — even simple — extensions 
(regular expressions with time [1], regular expressions 
with intersection, etc) to regular expressions. 
Little has been studied about the algorithms for testing 
membership for regular expressions themselves. This 
paper introduces such an algorithm. It is simple, easy 
to understand, nondeterministic and has some 
resemblance to the proof theory of intuitionistic linear 
logic [3] . In addition, it is a simple matter to observe that 



this algorithm extends well to other extensions to regular 

expressions. 

In this paper we present our algorithm, show some 

examples of its working, and discuss further 

improvements. The remainder of this paper is 

structured as follows. We describe our algorithm in the 

next section. 

In Section 3, we present some examples. Section 4 

concludes the paper with some considerations for further 

improvements. 

2. The language 

The regular expression is described by r-formulas given 
by the syntax rules below: 

r :: = \ s \ a \ r - r \ r + r \ r* 

In the rules above, an alphabet a represents the set {a}. 
8 represents {e}. represents the empty set. r • s 
represents the concatenation of two sets r and s. r + s 
represents the union of r and s. The Kleene closure of r - 
r* - indicates there are any number of r. 
We often write rr in place of r -r. The degree d{r) of a 
regular expression r is defined as follows: d(0)=O, 
d(a)=l, d(r s) = d(r + s) = d(r) + d{s) +1 ,d(r*) = d(r) +1. 
Further, <i(ri,...,r n ) = d(ri)+...+d(r n ) where r x ,...,r n is a 
sequence of regular expressions. 

The question of whether a string is a member of a 
regular expression is quite interesting. Such an algorithm 
needs to cope with the following: (1) the associativity of 
the operators, e.g., abc ^ a • (b • c), abc ^ (a -b) -c, and 
(2) the multiplicity of the Kleene closure, e.g., aaa ^ 
a*- a*, in an elegant fashion. We will present an 
algorithm for this task in the style of a proof system. 



225 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



Let w be a regular expression and r\ ,...,r n be a list of 
regular expression. Then a sequent of the form ri,...,r w h 
w - the notion that is an element of the concatenations 
of 7*i ? . . .,r n - is defined constructively by two axioms and 
eight inference rules. This is shown below. 

Algorithm for Testing Membership 



— ■ — jlasoml 
a\-a 

p • 3fr £>. \- w 

z±\-w 



p>&\-w 



€L 



p* 7 p* 7 £±\-w 

p 7 z±\-w 
o+S^I-iw 



CL 



+ A 



\-€ 



Asdom2 



^jhWj 






R 



p 7 A h W 



I*Z 



i?i 






+ A 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
empty) string. By the hypothesis, r 2 ..., r n \- w' has a proof. 
Putting the proofs for a ^a and r 2 ....r n \-w' using a • R 
rule, we obtain a proof satisfying the theorem. 

If r\ is r+s , then it must be the case that w ^ 
r,r 2 . . .r n or w ^ s,r 2 ....,r n . Consider the former case. 
By the hypothesis, r,r 2 ,.. .r n h- w has a proof. 
Putting this proof using a +ii rule, we obtain a proof 
satisfying the theorem. The same argument can be 
supplied for the latter case. 

The arguments for rs follow a similar pattern. 
For the case that r\ is of the form r*, we have to consider 
the three cases depending on whether r is never used, 
used once, or used more than once. Consider the case 
where r is never used. Then w must be an element of 
r 2 ,...,r n . By the hypothesis, r 2 ,...,r n ^w has a proof. 
Putting this proof using a WL rule, we obtain a proof 
satisfying the theorem. This is shown below. 



-d. 



r,7v-,7v 



-\L. 



WL 



In the above rule, a is an alphabet, A denote a list of 
regular expressions and p, i// denote a single regular 
expression. An inference rule can be read as follows: if 
all the sequents above are true, then the sequent below is 
true. A sequent A ,_ w has a proof if A ,_ w can be 
obtained from the axioms by applying the inference rules. 
In dealing with p* construct, the proof system can either 
discard it, use p once, or use p at least twice. 

Let us refer to the above collection of axioms and 
inference rules as DS. The following theorem shows the 
sound and completeness of the proof system DS. 



If r is used once, then there must be a string z such that w 
= zw',z ^ r, and w' ^ (r 2 ,...,r n ). By the hypothesis, both 
r \-z and r 2 ,. . ., r n i- w' have proofs. Putting the proofs for 
r\-z and r 2 ,...,r n i- w' using the rules, we obtain a proof 
satisfying the theorem. 



This is shown below. 



r— *2 



Tv--- 7 r T — *w 



r>r 3l — >r a — «?w 



R 






>r B 



!-±j. 



-D£ 



The arguments for the remaining case follow a 
similar pattern. The additional observation is that the CL 
rule is needed for this case. 



Theorem 2.1: Let r x ... ; r n be a list of regular expressions 
and let w be a string. Then, w is an element of r\ . . . . r n if 
and only if r b . . . , r n h- w has a proof in AS . 

Proof. The reverse direction is straightforward. In the 
forward direction, we prove the theorem by an induction 
on the degree of the sequence n..., r n , w. If the degree is 
1 or 2, then it must be of the form s ^ s , s ^,or a ^ a 
where a is an alphabet. It is easy to see that the theorem is 
true. 

If the degree is greater than 2, we consider the 
cases for the structure of r L 

If ri is a, then it must be the case that w is of the 

form aw' and w' ^ r 2v .. 5 r n where w' is a (possibly 



3. Examples 

This section describes the use of our algorithm. 
An example is provided by the following proof of aa ^ 

a*. 

It is interesting to note that the CL rule is used to 
control the multiplicity of a*. 



o— *a 



-DL 



a— *g 



•J?£ 



■oa 



/? 



►aa 



a: 



226 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



Another example of our algorithm is provided by the 
following proof of the sequent ca^(b+c) -a 



h+c- 



-+Z* 



a— *q 



b + c 7 a- 



■ca 



H 



(b + c) 



*co 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol 7, No. 1, 2010 
[2] S.C. Kleene, Introduction to Metamathematics, North- 
Holland, Amsterdam, 1964. 

[3] J.Y. Girard, "Linear logic", Theoretical Computer Science, 
vol.50,pp.l-102, 1987. 

[4] J.Hodas and D. Miller, " Logic programming in a fragment 
of intuitionistic linear logic", Journal of Information and 
Computation, 1992. Invited to a special issue of submission to 
the 1991 LICS conference. 

[5] J.E.Hopcroft, R.Motwani, and J.D. Ullman, Automata Theory, 
Languages and Computation, Ad. 



A computation process typically searches for a proof 
from the bottom-up in a sequent calculus for reasons of 
efficiency. Thus, given a conclusion sequent, it attempts 
to find its proof from bottom-up. 



4. Conclusion 



We have described an algorithm for testing membership 
in regular expressions. The advantage of this algorithm is 
that it does not require the complex conversion process to 
finite automata. As a consequence, it extends easily to 
various extensions to regular expressions. For example, 
our algorithm extends easily to the one that deals with 
algebraic laws, i.e., regular expressions with variables [5]. 
Two regular expressions with variables are equivalent if 
whatever expressions we substitute for the variables, the 
results are equivalent. For example, V L V 
M{ L+M=M+L ). 

Regarding the performance of our algorithm, 
non-determinism is present in several places of this 
algorithm. In particular, there is a choice concerning 
which way the text is split in the -R rule. Hodas and 
Miller [4] dealt with this rule by using IO-model in which 
each goal is associated with its input resource and output 
resource. The idea used here is to delay this choice of 
splitting as much as possible. This observation leads to a 
more viable implementation. Our ultimate interest is in 
a procedure for carrying out computations of the kind 
described above. It is hoped that these techniques may 
lead to better algorithms. 

5. Acknowledgements 

This paper was supported by Dong-A University 
Research Fund. 

References 

[1] E. Asarin and P. Caspi, and O. Maler, "A Kleene theorem 
for timed automata", Logic in Computer Science, 1997, 
pp.160-171. 



227 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. 1, 2010 



Impact of Random Loss on TCP Performance in 

Mobile Ad-hoc Networks (IEEE 802.11): A 

Simulation-Based Analysis 



Shamimul Qamar 

Department of Computer Sceience 
CAS, King Saud University 
Riyadh, Saudi Arabia 



Abstract-Initially TCP was designed with the notion in mind 
that wired networks are generally reliable and any segment loss 
in a transmission is due to congestion in the network rather than 
an unreliable medium (The assumptions is that the packet loss 
caused by damage is much less than 1%) . This notion doesn't 
hold in wireless parts of the network. Wireless links are highly 
unreliable and they lose segments all the time due to a number of 
factors. Very few papers are available which uses TCP for 
MANET. In this paper, an attempt have been made to justify the 
use of TCP variants (Tahoe and Reno) for loss of packet due to 
random noise introduces in the MANET. For the present analysis 
the simulation has been carried out for TCP variants (Tahoe and 
Reno) by introduces 0%, 10%, 20% and 30% noise.The 
comparison of TCP variants is made by running simulation for 
0%, 10%, 20% and 30% of data packet loss due to noise in the 
transmission link and the effect of throughput and congestion 
window has been examined. During the simulation we have 
observed that throughput has been decreased when a drop of 
multiple segments happens, further we have observed in the case 
of TCP variant (Reno) throughput is better at 1% (Figure 5) 
which implies a network with short burst of error and low BER, 
causing only one segment to be lost. When multiple segments are 
lost due to error prone nature of link, Tahoe perform better than 
Reno (Figure 13), that gives a significant saving of time (64.28%) 
in comparison with Reno (Table 4). Several simulations have 
been run with ns-2 simulator in order to acquire a better 
understanding of these TCP variants and the way they perform 
their function. We conclude with a discussion of whether these 
TCP versions can be used in Mobile Ad hoc Network? 

Index Term- TCP, Tahoe, Reno, MANET, ns-2, Random noise 

1. Introduction 

TCP is the widely used for Internet traffic. In the wired 
network losses are mostly due to congestion. In practice, 
losses may also be caused from noisy links. This is especially 
true in case of radio links, e.g. in cellular network, Mobile Ad- 
hoc network or in satellite links. A link may become in fact 
completely disconnected for some period of time or it may 
suffer from occasional interferences (due to shadowing, fading 
etc.) that cause packets to contain errors and then to be 
dropped resulting low throughput. Transport protocols should 



Kumar Manoj 

ECED,DPT 
I.I.T Roorkee 
India 



be independent of the technology of the underlying 
network layers. It should not care whether IP is running 
over fiber or radio. But in reality, it does matter, since 
most TCP implementations that have been designed for 
wired networks, don't cope very well on wireless 
networks. Different version of TCP in a local network 
with lossy link are compared and analyzed in [7]. Several 
scheme have been reported in the literature, to alleviate 
the effect of losses on TCP performances over wireless 
network (or networks with lossy link) [8]. Very few 
papers are available which uses TCP for MANET This 
paper deals with for Tahoe and Reno TCP variants and its 
variation of error by introducing different level of noise . 
The ns-2 network simulator used to understand the 
behaviors and characteristics of TCP variants. Wireless 
links are highly unreliable and they lose segments all the 
time due to a number of factors. These include fading, 
interference, higher bit error rate and mobility related 
processes such as handovers. 

In this paper, we will make a comparison between Tahoe 
and Reno TCP variants in a noisy environment. Our 
simulation show that Tahoe copes well in a high segment 
loss environment. Two algorithms, Slow Start and 
Congestion Avoidance, modify the performance of TCP's 
sliding window to solve some problems relating to 
congestion in the network [4]. The idea of congestion 
control is for the sender to determine the capacity that is 
available on the network. The sender, in standard TCP 
implementation, keeps two state variables for congestion 
control: a slow- start/congestion window, cwnd, and a 
threshold size, ssthresh. These variables are used to 
switch between the slow start and congestion avoidance 
algorithms. Slow start provides a way to control initial 
data flow at the beginning of a communication session 
and during an error recovery. This is based on received 
acknowledgements. Congestion avoidance increases the 
cwnd additively, so that it grows by one segment for each 
round trip time. These two algorithms are implemented 
together and work as if they were one. The combined 



228 



http://sites.google.com/site/ijcsis/ 
ISSN 1947-5500 



(IJCSIS) International Journal of Computer Science and Information Security, 

Vol. 7, No. I, 2010 



effect of these two algorithms on the congestion window is