Full text of "Journal of Computer Science IJCSIS September 2010"

See other formats

IJCSIS Vol. 8 No. 6, September 2010
ISSN 1947-5500

International Journal of
Computer Science
& Information Security

Editorial
Message from Managing Editor

IJCSIS is an open access publishing venue for research in general computer science
and information security.

Target Audience: IT academics, university IT faculties; industry IT departments;
government departments; the mobile industry and computing industry.

Coverage includes: security infrastructures, network security: Internet security,
content protection, cryptography, steganography and formal methods in information
security; computer science, computer applications, multimedia systems, software,
information systems, intelligent systems, web services, data mining, wireless
communication, networking and technologies, innovation technology and management.

The average paper acceptance rate for IJCSIS issues is kept at 25-30% with an aim to
provide selective research work of quality in the areas of computer science and
engineering. Thanks for your contributions in September 2010 issue and we are
grateful to the experienced team of reviewers for providing valuable comments.

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

IJCSIS Vol. 8, No. 6, September 2010 Edition
ISSN 1947-5500 © IJCSIS, USA.

Abstracts Indexed by (among others):

Gooole scholar = r\it*^e*r* 5>UUUa

*- ' \— m I I ^J vJ \j ul !■■■■■■ ^anrti * no in* r&r cei* nr*

b&rcMinch engine r&r science

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

Dr. T.C. Manjunath,

ATRIA Institute of Tech, India.

TABLE OF CONTENTS

1. Paper 23081018: Improvement Dynamic Source Routing Protocol by Localization for Ad hoc
Networks (pp. 1-6)

Mehdi Khazaei

Kermanshah University of Technology, Information Technology Engineering Group, Kermanshah, Iran

2. Paper 31081053: Steganalysis of Reversible Vertical Horizontal Data Hiding Technique (pp. 7-12)

Thorn Ho Thi Huong, Faculty of Information Technology, Haiphong Private University, Haiphong,

Vietnam

Canh Ho Van, Dept. of Professional Technique, Ministry of Public Security, Hanoi, Vietnam

Tien Trinh Nhat, College of Technology, Vietnam National University, Hanoi, Vietnam

3. Paper 21081012: Off-line Handwritten Signature Recognition Using Wavelet Neural Network (pp.
13-21)

Mayada Torek, Computer Science Department, Faculty of Computers and Information Sciences, Mansoura,

Egypt

Taher Hamza, Computer Science Department, Faculty of Computers and Information Sciences, Mansoura,

Egypt

Elsayed Radwan, Computer Science Department, Faculty of Computers and Information Sciences,
Mansoura, Egypt

4. Paper 21081013: A Black-Box Test Case Generation Method (pp. 22-31)

Nicha Kosindrdecha , Autonomous System Research Laboratory, Faculty of Science and Technology,
Assumption University, Bangkok, Thailand

Jirapun Daengdej, Autonomous System Research Laboratory, Faculty of Science and Technology,
Assumption University Bangkok, Thailand

5. Paper 21081014: White-Box Test Reduction Using Case-Based Maintenance (pp. 32-40)

Siripong Roongruangsuwan, Autonomous System Research Laboratory, Faculty of Science and Technology,
Assumption University, Bangkok, Thailand

Jirapun Daengdej, Autonomous System Research Laboratory, Faculty of Science and Technology,
Assumption University, Bangkok, Thailand

6. Paper 23081021: Nat Traversal for Video Streaming Applications (pp. 41-46)

Omar A. Ibraheem #1 , Omer Amer Abouabdalla * 2 , Sureswaran Ramadass n

# National Advanced IPv6 center of Excellence (NAV6), Universiti Sains Malaysia (USM), Pulau penang,

Malaysia

7. Paper 25081026: The Integration of GPS Navigator Device with Vehicles Tracking System for
Rental Cars Firms (pp. 47-51)

Omarah O. Alharaki, KICT, International Islamic University, Kuala Lumpur, Malaysia
Fahad S. Alaieri, KICT, International Islamic University, Kuala Lumpur, Malaysia
Akram M. Zeki, KICT, International Islamic University, Kuala Lumpur, Malaysia

8. Paper 29081036: Process Framework in Global extreme Programming (pp. 52-59)

Ridi Ferdiana, Lukito Edi Nugroho, Paulus Insap Santoso

Department of Electrical Engineering and Information Technology, Gadjah Mada University (UGM)

Yogyakarta, Indonesia

Ahmad Ashari, Department of Computer Science and Electronics, Gadjah Mada University (UGM),

Yogyakarta, Indonesia

9. Paper 29081038: A Hybrid PSO-SVM Approach for Haplotype Tagging SNP Selection Problem
(pp. 60-65)

Min-Hui Lin, Department of Computer Science and Information Engineering, Dahan Institute of
Technology, Sincheng, Hualien County 971, Taiwan, Republic of China

Chun-Liang Leu, Department of Information Technology, Ching Kuo Institute of Management and Health,
Keelung 336, Taiwan, Republic of China

10. Paper 07110912: PAPR Reduction Technique for LTE SC-FDMA Systems Using Root-Raised
Cosine Filter (pp. 66-71)

Md. Masud Rana, Jinsang Kim and Won-Kyung Cho

Deptartment of Electronics and Radio Engineering, Kyung Hee University

1 Seocheon, Kihung, Yongin, Gyeonggi, 449-701, Republic of Korea

11. Paper 22081017: Survey of Routing Protocols and Channel Assignment protocols in Wireless
Mesh Networks (pp. 72-77)

Vivek M Rathod, Suhas J Manangi, Satish E, Saumya Hegde
National Institute of Technology Karnataka - Surathkal

12. Paper 29071044: An Approach For Designing Distributed Real Time Database (pp. 78-87)

Dr. Dhuha Basheer Abdullah, Computer Sciences Dept./Computers Sciences and Mathematics College
/Mosul University, Mosul- Iraq

Ammar Thaher Yaseen, Computer Sciences Dept./Computers Sciences and Mathematics College /Mosul
University, Mosul- Iraq

13. Paper 31081051: Person Identification System using Static-dynamic signatures fusion (pp. 88-92)

Dr. S.A Daramola 1 and Prof.T.S Ibiyemi 2

1 Department of Electrical and Information Engineering, Covenant University Ota Ogun State, Nigeria.

2 Department of Electrical Engineering, University ofllorin, Ilorin, Kwara-State, Nigeria

14. Paper 31081061: Short term flood forecasting using RBF static neural network modeling a
comparative study (pp. 93-98)

Rahul P. Deshmukh, Indian Institute of Technology, Bombay, Powai, Mumbai, India

A. A. Ghatol, Former Vice-Chancellor, Dr. Babasaheb Ambedkar Technological University, Lonere,

Raigad, India

15. Paper 31031083: Analysis of impact of Symmetric Encryption Algorithms in Data Security Model
of Grid Networks (pp. 99-106)

N. Thenmozhi, Department of Computer Science, N.K.R. Govt. Arts College for Women, Namakkal-637
001, India.

M. Madheswaran, Department of Electronics and Communication Engg., Muthayammal Engineering
College, Rasipuram-637 408, India.

16. Paper 09071005: Low Power and Area Consumption Custom Networks-On-Chip Architectures
Using RST Algorithms (pp. 107-115)

1 P.Ezhumali 2 Dr.CArun

1 Professor, Dept of Computer Science Engineering

2Asst. Professor, Dept of Electronics and Communication

Ralalakshmi Engineering College, Thandalam-602 105, Chennai, India

17. Paper 16081005: Prediction of Epileptic form Activity in Brain Electroencephalogram Waves
using Support vector machine (pp. 116-121)

Pavithra Devi S T , M.Phil Research Scholar, PSGR Krishnammal College for Women , Coimbatore
Tamilnadu, India

Vijaya M S, Assistant Professor and Head, GRG School of Applied Computer Technology, PSGR
Krishnammal College for Women, Coimbatore Tamilnadu, India

18. Paper 16081006: Deployment of Intelligent Agents in Cognitive networks (pp. 122-127)

Huda Fatima, Dept. of CS, Jazan University, Jazan, K.S.A

Dr. Sateesh Kumar Pradhan, Dept. of Comp. Engineering, King Khalid University, Abha, K.S.A

MohiuddinAli Khan, Dept. of Comp. Networks, Jazan University, Jazan, K.S.A

Dr. G.N.Dash, Dept. of Comp. Science, Sambalpur University, Orissa, India

19. Paper 21081011: A Performance Study on AES Algorithms (pp. 128-133)

B.D. C.N. Prasad \ P.E.S.N.krishna Prasad \ P Sita Rama Murty 3 andKMadhavi 4

1. Dept. of Computer Applications, PVP Siddardha Institute of Technology, Vijayawada,

2. Dept. ofCSIT, Sri Prakash College of Engineering, Tuni,

3. Dept. ofCSIT, Sri Prakash College of Engineering, Tuni,

4. Dept. ofCSE, Dadi Institute of Technology, Anakapalli,

20. Paper 24091018: Hybrid Fingerprint Image compression and Decompression Technique (pp. 134-
138)

Dr.R.Seshadri, ,B.Tech„M.E,Ph.D, Director, S.V.U. Computer Center S.V. University, Tirupati

Yaswanth Kumar. Avulapti , M.C.A, M.Tech, (PhD), Research Scholar, Dept of Computer Science, S. V.

University, Tirupati

Dr.M.Usha Rani M.C.A, PhD, Associate Professor, Dept. of Computer Science,, SPMW, Tirupati

21. Paper 25081025: Punctured Self -Concatenated Trellis Codes with Iterative Decoding (pp. 139-144)

Labib Francis Gergis, Misr Academy, Mansoura City, Egypt

22. Paper 26081028: Application of Fuzzy Composition Relation For DNA Sequence Classification
(pp. 145-148)

Amrita Priyam, Dept. of Computer Science and Engineering, Birla Institute of Technology, Ranchi, India.

B. M. Karan , G. Sahoo +

+ Dept. of Electrical and Electronics Engineering, + + Dept. of Information Technology

Birla Institute of Technology, Ranchi, India

23. Paper 26081029: Data Security in Mobile Ad Hoc Networks using Genetic Based Biometrics (pp.
149-153)

B. Shanthini, Research Scholar , CSE Department , Anna University , Chennai, India

5. Swamynathan, Assistant Professor, CSE Department , Anna University , Chennai, India

24. Paper 28081031: Effective Multi-Stage Clustering for Inter- and Intra-Cluster Homogeneity (pp.
154-160)

Sunita M. Karad f , Assistant Professor of Computer Engineering, MIT, Pune, India
V.M.Wadhai ff , Professor and Dean of Research, MITSOT, MAE, Pune, India
M.U.Khar at fff , Principle ofPankaj Laddhad IT, Yelgaon, Buldhana, India
Prasad S.Halgaonkar ffff , Faculty of Computer Engineering, MITCOE, Pune, India
Dipti D. Patil fffff , Assistant Professor of Computer Engineering, MITCOE, Pune, India

25. Paper 28081033: A Pilot Based RLS Channel Estimation for LTE SC-FDMA in High Doppler
Spread (pp. 161-166)

M. M. Rana

Department of Electronics and Communication Engineering, Khulna University of Engineering and

Technology, Khunla, Bangladesh

26. Paper 28081034: Priority Based Congestion Control for Multimedia Traffic In 3G Networks (pp.
167-173)

ProfV.S Rathore 1, Neetu Sharma 2, Amit Sharma 3, Durgesh Kumar Mishra 4

123 Department of Computer Engineering, Rajasthan, India

12 Rajasthan College of Engineering for women, Rajasthan, India

3 Shri Balagi College of Engineering & Technology, Rajasthan, India

4 Acropolis Institute of Technology and Research, Indore, MP, India

27. Paper 31081050: Adaptive Sub-block ARQ techniques for wireless networks (pp. 174-178)

A. N. Kemkar, Member, ISTE and Dr. T. R. Sontakke, Member, ISTE

28. Paper 30071074: Trigon-based Authentication Service Creation with Globus Middleware (pp.
179-185)

Ruckmani V , Ramakrishna Engineering College, Coimbatore, India
Anitha Kumari K , PSG College of Technology, Coimbatore, India
Sudha Sadasivam G , PSG College of Technology, Coimbatore, India
Dhaarini M P , PSG College of Technology, Coimbatore, India

29. Paper 30081040: Performance Evaluation of Speaker Identification for Partial Coefficients of
Transformed Full, Block and Row Mean of Speech Spectrogram using DCT, WALSH and HAAR
(pp. 186-198)

Dr. H. B. Kekre, Senior Professor, MPSTME, SVKM's NMIMS University, Mumbai, 400-056, India

Dr. Tanuja K. Sarode, Assistant Professor, Thadomal Shahani Engg. College, Bandra (W), Mumbai, 400-

050, India

Shachi J. Natu, Lecturer, Thadomal Shahani Engg. College, Bandra (W), Mumbai, 400-050, India

Prachi J. Natu, Assistant Professor, GVAIET, Shelu, Karjat 410201, India

30. Paper 30081041: A Research Proposal for Mitigating DoS Attacks in IP-based Networks (pp. 199-
201)

Sakharam Lokhande, Assistant Professor, School of Computational Science, Swami Ramanand Teerth

Marathwada University, Nanded, MS, India, 431606.

Parag Bhalchandra **, Assistant Professor, School of Computational Science, Swami Ramanand Teerth

Marathwada University, Nanded, MS, India, 431606.

Nilesh Deshmukh, Assistant Professor , School of Computational Science, Swami Ramanand Teerth

Marathwada University, Nanded, MS, India, 431606.

Dr. Santosh Khamitkar, Assistant Professor , School of Computational Science, Swami Ramanand Teerth

Marathwada University, Nanded, MS, India, 431606.

Santosh Phulari, Assistant Professor, School of Computational Science, Swami Ramanand Teerth

Marathwada University, Nanded, MS, India, 431606.

Ravindra Rathod, Assistant Professor, School of Computational Science, Swami Ramanand Teerth

Marathwada University, Nanded, MS, India, 431606

31. Paper 30081042: An Efficient and Minimum Cost Topology Construction for Rural Wireless
Mesh Networks (pp. 202-209)

Prof. V. Anuratha, H.O.D - PG. Comp. Science, Sree Saraswathi Thyagaraja college, Pollachi, Tamil

Nadu, India

Dr. P. Sivaprakasam , Associate Professor, Sri Vasavi college of Arts & Science, Erode, Tamil Nadu,

India

32. Paper 31071081: Reinforcement Learning by Comparing Immediate Reward (pp. 210-214)

Punit Pandey, Department Of Computer Science and Engineering, Jaypee University Of Engineering And

Technology

Dr. Shishir Kumar, Department Of Computer Science and Engineering, Jaypee University Of Engineering

And Technology

Deepshikha Pandey, Department Of Computer Science and Engineering, Jaypee University Of Engineering

And Technology

33. Paper 31081043: Information realization with statistical predictive inferences and coding form
(pp. 215-220)

D. Mukherjee, Sir Padampat Singhania University, Udaipur-313601,Rajasthan,India

P.Chakrabarti* , AKhanna , V.Gupta

Sir Padampat Singhania University, Udaipur-313601,Rajasthan,India

34. Paper 31081045: Scaling Apriori for Association Rule Mining using Mutual Information Based
Entropy (pp. 221-227)

S. Prakash, Research Scholar, Sasurie College of Engineering, Vijayamangalam,Erode(DT), Tamilnadu,

India.

Dr. R. M. S. Parvathi M.E.(CSE),Ph.D., Principal, Sengunthar College ofEngg.for Women, Tiruchengode.

Tamilnadu, India.

35. Paper 31081046: Clustering of High- Volume Data Streams In Network Traffic (pp. 229-233)

M. Vijayakumar, Research Scholar, Sasurie College of Engineering, Vijayamangalam, Erode(Dt) ,
Tamilnadu, India.

Dr. R.M.S. Parvathi M.E.(CSE),Ph.D., Principal, Sengunthar College of Engg.for Women, Tiruchengode.
Tamilnadu, India.

36. Paper 31081049: A.R.Q. techniques using Sub-block retransmission for wireless networks (pp.
234-237)

A. N. Kemkar, Member, ISTE and Dr. T. R. Sontakke, Member, ISTE

37. Paper 22081015: Performance Analysis of Delay in Optical Packet Switching Using Various
Traffic Patterns (pp. 238-244)

AKavitha , IT dept, Chettinad College of Engineering & Technology, Karur, Tamilnadu, India

V.Rajamni, Indra Ganesan College of Engineering, Trichy, Tamilnadu, India

P. Anandhakumar, IT Dept, Madras Institute of Technology, Chennai, Tamilnadu, India

38. Paper 29081039: A Feedback Design for Rotation Invariant Feature Extraction in
Implementation with Iris Credentials (pp. 245-254)

M. Sankari, Department of Computer Applications, Nehru Institute of Engineering and Technology,

Coimbatore, India.

R Bremananth, School ofEEE, Information Engg. (Div.), Nanyang Technological University, Singapore -

639798.

39. Paper 31081047: Empirical Mode Decomposition Analysis of Heart Rate Variability (pp. 255-258)

C. Santhi, M.E., Assistant Professor, Electronics and Communication Engineering, Government College of
Technology, Coimbatore-641 013

N. Kumaravel, Ph.D, Professor, Head of the Department, Electronics and Communication Engineering,
Anna University, Chennai-600 025.

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Improvement Dynamic Source Routing Protocol by
Localization for Ad hoc Networks

Mehdi Khazaei

Kermanshah University of Technology

Information Technology Engineering Group

Kermanshah, Iran .

Abstract-Ad hoc networks are temporary networks with a
dynamic topology which don't have any established
infrastructure or centralized administration. Consequently, in
recent years many researchers have focused on these networks.
These networks need efficient routing protocols in terms of
Quality of Services (QOS) metrics. Ad hoc networks suffer from
frequent and rapid topology changes that cause many challenges
in their routing. Most of the routing protocols like this proposed
protocol try to find a route between source and destination
nodes and when any route is expired, a new route would be
formed. Rapid route reconstruction may cause the network
inefficiency. Therefore, we have to decrease this processes. The
proposed protocol as DSR routing protocol build one routes
between source and destination but create backup routes during
the route reply process, route maintenance process and use local
recovery process in order to improve the data transfer and
attended to QOS. The protocol performance is demonstrated by
using the simulation results obtain from the global mobile
simulation software (Glomosim). The experimental results show
that this protocol can decrease the packet loss ratio and increase
data transfer rather than DSR that, it is useful for the
applications that need a high level of reliability.

Keywords; Protocol, Routing, Local Recovery, Mobile Ad-hoc
Networks

I. Introduction

Routing in ad hoc networks is a very challenging issue due
to nodes mobility, dynamic topology, frequent link breakage,
limitation of nodes (memory, battery, bandwidth, and
processing power), and limited transmission range of the node
and lack of central point like base stations or servers. On the
other hand, there are a lot of performance metrics and quality
services which should be satisfied in an ad hoc network like
end -to-end data throughput, average end-to-end data delay,
packet loss ratio, Normalized Routing Load, Packet Delivery
Ratio, and route optimality. Each protocol can satisfy some of
these metrics and has some drawbacks in terms of other
metrics. Furthermore, due to the nature of ad hoc networks
(distributed and cooperated routing), even for a fixed metric,
each protocol can show a different performance with different
networks features like number of mobile nodes, mobility of
nodes, pause time and.... So by comparing between different
ad hoc routing protocols we can extract very important
information about the performance of these protocols in the
Different situations. In the other hands, the nodes mobility and
the probability of links failure may cause the fault tolerance

Issues more important for routing problem in ad hoc network
therefore, each routing protocol should be fault tolerant in
probable route failures [1].

Routing protocols in conventional wired networks are
usually based upon either distance vector or link state routing
algorithms as a DSDV [2], CGSR [2] and FSR [2]. These
algorithms require periodic routing advertisements to be
broadcast by each router. These conventional routing
algorithms are clearly not efficient for type of dynamic
changes which may occur in an ad-hoc network [2, 3]. A new
class of on-demand routing protocols e.g., DSR [4, 5], TORA
[2], AODV [6, 7]) for mobile ad hoc networks has been
developed with the goal of minimizing the routing overhead.
These Protocols reactively discover and maintain only the
needed routes, in contrast to proactive protocols (e.g., DSDV
[2]) which maintain all routes regardless of their usage. The
key characteristic of an on-demand protocol is the source-
initiated route discovery procedure. Whenever a traffic source
needs a route, it initiates a Route discovery process by sending
a route request for the destination (typically via a network-
wide flood) and Waits for a route reply. Each route discovery
flood is Associated with significant latency and overhead. This
is particularly true for large networks. Therefore, for on-
demand routing to be effective, it is desirable to keep the route
discovery frequency low [8].

Single route routing allows the establishment of one route
between a source and single destination node. Because of node
mobility, the route may be broken frequently; therefore,
having replacement route in cache memory to transmit data
will improve the fault tolerance and higher aggregate
bandwidth in these networks. Beside of this, by repairing the
broken routes locally, the number of route rediscovery
processes can be decreased. This paper improves the fault
tolerance and increase reliability by obtain replacement routes
in RREP 1 and RRER 2 processes and local recovery process
together. This optimization is done on the DSR protocol.

The rest of this paper is organized as follows. In section II
the DSR protocol is explained. Section III deals with the
related works and Section IV describe the proposed protocol
mechanism in detail. Performance evaluation by simulation is
presented in section V and concluding remarks are made in
section VI.

1 Route reply
2 Route error

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

II. Dynamic Source Routing Protocol (DSR)

DSR is an on-demand routing protocol for ad hoc
networks. Like any source routing protocol, in DSR the source
includes the full route in the packets' header. The intermediate
nodes use this to forward packets towards the destination and
maintain a route cache containing routes to other nodes. In
following subsections DSR operation are briefly described [4].

A. Route discovery

If the source does not have a route to the destination in its
route cache, it broadcasts a RREQ 3 message specifying the
destination node for which the route is requested. The RREQ
message includes a route record which specifies the sequence
of nodes traversed by the message. When an intermediate
node receives a RREQ, it checks to see if it is already in the
Route record. If it is, it drops the message. This is done to
prevent routing loops. If the intermediate node had received
the RREQ before, then it also drops the message. The
intermediate node forwards the RREQ to the next hop
according to the route specified in the header. When the
destination receives the RREQ, it sends back a route reply
message. If the destination has a route to the source in its route
cache, then it can send a RREP message a long this route.
Otherwise, the RREP message can be sent along the reverse
route back to the source. Intermediate nodes may also use their
route cache to reply To RREQs. If an intermediate node has a
route to the destination in its cache, then it can append the
route to the route record in the RREQ, and send an RREP back
to the source containing this route. This can help limit
flooding of The RREQ. However, if the cached route is out-
of-date it can result in the source receiving stale routes [4].

B. Route maintenance

When a node detects a broken link while trying to forward
a packet to the next hop, it sends a RERR message back to the
source containing the link in error. When an RERR message is
received, all routes containing the link in error are deleted at
that node [4].

III. Related Works

Ad hoc routing protocols such as ADOV, DSR, DSDV and
OLSR have been investigated on the ad hoc networks in the
past few years. The investigations of the performance of these
protocols on the ad hoc networks have produced many useful
results. However, we have seen very limited findings of how
these Ad-hoc routing protocols perform on wireless ad hoc
networks. Nonetheless, we can see many attempts at
developing routing protocols for ad hoc networks under the
different deployment of ad hoc networks [8, 9 and 10]. In
following are brought some of these attempts.

SMR is an on demand routing protocol that uses
maximally disjoint routes to transmit data packets. Unlike
DSR, intermediate nodes do not allow to send RREP packets
back to the source instead, only destination nodes reply to the
RREQ packets and selects maximally-disjoint routes [11].

MP-DSR is a multi-route QOS aware extension to DSR. It
focuses on a QOS metric, end-to-end reliability. End-to-end

Route Request

Reliability is defined as the probability of sending data
successfully from the source to the destination node within a
time window. MP-DSR selects a set of routes that satisfy a
specific end-to-end reliability Requirement [12].

MSR is attempts to minimize the end-to-end delay for
sending a data from source to destination by using multi-route
routing and intelligent traffic allocation [13].

CHAMP is multi-route protocol that uses round-robin
traffic allocation to keep routes fresh. It also employs
cooperative packet caching to improve fault tolerance and
takes advantage of temporal locality in routing, where a
dropped packet is a recently sent packet [9].

The local recovery techniques have been used in some
routing protocols for route maintenance processes. This
technique aims to reduce the frequency of RREQ floods
triggered by nodes that are located in the broken routes [14].

SLR is one of these routing protocols. It modifies DSR,
using a new route recovery mechanism called bypass routing.
Bypass routing utilizes both route caches and local error
recovery techniques during failures to reduce the control
overhead [15].

LRR is also another routing protocol that uses local
recovery techniques. In this protocol the information of next-
to-next (NN) node is stored at each intermediate node along
the route. After detecting a broken link by an upstream node, it
sends out the non-propagating requests to find another node
which is in contact with itself and the NN node on the route;
therefore the routes can be repaired locally in the shortest
possible time [16].

MRFT protocol improves fault tolerance in DSR and SMR
protocols. To achieve the goal of decreasing the packet loss
ratio and increasing fault- tolerance, MRFT uses both multi-
route and local recovery techniques together [17].

IV. The Proposed Protocol

This paper proposes IM-DSR 4 protocol to improve fault
tolerance and QOS in DSR protocol. To achieve the goal of
decreasing the packet loss ratio and increasing fault-tolerance,
IM-DSR uses local recovery techniques and alternate route
during route reply and route maintenance that reliability in the
network would be increased. IM-DSR modifying the route
discovery, route reply and route maintenance processes in
DSR. The IM-DSR protocol is including route discovery,
route reply, route maintenance and local recovery processes
that discussed in the following subsection.

A. Route Discovery Process

IM-DSR is an on demand routing protocol that builds
single route using request/reply cycles. When the source node
needs to send data to the destination but no route information
is known, it floods RREQ packets over the entire network.
When an intermediate node receives a RREQ that is not a
duplicate, it appends its ID to the packet and rebroadcasts it. In
IM-DSR all of the duplicate RREQs that are received by
intermediate nodes are dropped. In IM-DSR, intermediate

4 Improvement-DSR

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Nodes are allowed to send RREPs back to the source even
when they have route information to the destination in their
route caches.

B. Route Reply Process

When receiving the first RREQ, the destination sends a
RREP back to the source. After that, the destination node
consumes other RREQs. The Route-Number of the RREP is
one.

After receiving RREP packet by the intermediate nodes, if
it has not route with same length to destination node, they
store the routes in their route caches. The Route-Number of
this routes are zero and used in route maintenance process for
improving the break routes also sending data if there is not
main route.

Look at the Fig. 1, suppose that node (H) sends the RREP
to the source node (A), a route is found and sent to node (A)
by RREP is A-C-D-G-F-H. Now, suppose that the RREP is
received by node (C) which is middle node. Node (C) saves
the routes to destination (H) which is C-D-G-F-H, additionally
node (C) save C-D, C-D-G and C-D-G-F routes in the route
caches.

When the source node receive the RREP, it will store the
route and use that for transmit data.

C. Route Maintenance and Local Recovery Processes

During a transmission session, a problem such as node
mobility, or low battery power might be raised, which can lead
to break an existing route and lose route connectivity. This
may force a route rediscovery process by flooding RREQs
over the network. To avoid this phenomenon, IM-DSR uses
following mechanism that one of them is local recovery
techniques. Using local recovery techniques is very useful
despite they consume the limited power of each nodes.

Suppose that a node finds a broken link, while sending a
packet. At first, it seeks the route cache and deletes all routes
include the broken link, and then according to kind of the
packet one of the following items is done:

• If transitional packet would be a RREQ, the node would
not send RRER to the source node.

• If transitional packet would be a RREP, send RRER to the
node which makes the RREP.

^ G

Information In Cache Memory
N etwork Connections

Figure 1 . Routes Structure in an Ad-hoc Network

• If transitional packet is a RRER, it would examine how
many times the packet would be saved. If it was the first
time, the meaning is, the packet would be saved by
examining a route cache and finding alternate route. The
RRER is sent to destination through that route then the
RRER is made and it will report the broken link to the
source of RRER. If it were not first time or if alternate
route were not in the route cache of node, the RRER
would be deleted and only a RRER would be sent to the
source node. Therefore RRER is saved only for one time
by the IM-DSR protocol.

• If transitional packet would be data, it would examine
how many times the packet would be saved. If this time
were less than three, the data packet would be sent by
examining their route cache and alternate route then it
will send a RRER to the source node. If these times were
more than three or if alternate route was not in the route
cache of node the data packet would be deleted and only a
RRER would be sent to the source node. Therefore data
packet is saved for three times by the IM-DSR protocol.

If very data packet passed the same route towards
destination node and they faced the broken link (while
sending), the node which recognized an error, for every data
packet send a RRER to the source node. In order to avoid this
item every node before sending RRER to source node,
examine this is a first RRER or not. If it was not send, a new
RRER send to source node.

Every node which recognizes the broken link and makes
the RRER, examined the route cache in order to find alternate
route and put it in the RRER, which the node that received the
RRER, replaces the route in the RRER with the previous
invalid route in the route cache.

Fig. 2 shows this matter. The source node (A) sends data to
the destination node (H) through A-C-D-H. When node (D)
sends the data packet, it will find the failure in node (H). By
examine the route cache, it chooses the alternate route D-G-F-
H, hence, the data packet is sent to destination node (H)
through this route, then RRER is made and it is sent to the
node (A). This packet includes the alternate route D-G-F-H.
The node (A) receives the RRER and deletes A-C-D-H from
route cache and replaces A-C-D-G-F-H.

Every middle and source node which receives the RRER,
examine those route in route cache which includes the broken
link and should be deleted from cache and if packet included
alternate route, exploited that and saved in route cache with
number two. If in buffer, data packet waiting to send toward
alternate route destination, it will send through that route.
Such as the Fig. 2, while passing the RRER, node (C) adds C-
D-G-F-H to the route cache.

If a node who detected a broken link cannot find any
alternate route in its route cache, so it drops the data packet
and sends a RERR without any repaired route to the source.
After that, because of performing local recovery process by
the node that detects the broken link, the source node does not
trigger the rediscovery process immediately. After detecting
the broken link, node sends a RERR to the source and starts
the local recovery process simultaneously. To repair the route

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

J?
B

Information In Cache Memory
N etwork C onne ctions

Figure 2. Ad-hoc Network Structure, when occur break link

Locally, node triggers a local query to its neighbors. The
neighbors reply if they have any valid route to the destination.
When node receives the RRP 5 , it repairs the primary route and
then sends a RRP back to the source, like shown in the Fig. 3.

In Fig. 3 when node (D) finds the failure in node (G), First
examined the route cache for replacement route, if not found
then begun local recovery request which node (F) reply with
F-G-H as repaired route. When node (D) receives this route
update its memory cache and sends data from this route to
destination and send RRP to the source node that here is node
(A). When node (A) receives the RRP deletes A-C-D-G-H
from route cache and replaces A-C-D-F-G-H. Every middle
node which receives RRP, examine if included repaired route,
then exploited that and saved in route cache with number two.
In Fig. 3 while passing RRP, node (C) adds C-D-F-G-H to the
route cache more node (C) adds C-D, C-D-F and C-D-F-G to
route cache.

If the source node has not primary route to send data, it
will use the repaired route. If no data packet containing
repaired route was reported to the source node for a certain
amount of time, then it sends a new RREQ to the destination
node. After receiving the new RREQ by the destination node,
it performs the route rediscovery process that was described in
subsection A.

In the IM-DSR protocol, route with Route-Number one is
main route. Route with Route-number zero are obtain by route
reply process and route with Route-Number two are those
which are obtain in route maintenance process and known as
repaired route.

When the source node wants to send data to a destination,
it tries to use the primary, which has the highest priority (the
routes with route-Number one) at first, if not exist, it will use
obtain route (the route with Route-Number zero) otherwise
use repaired route (the route with Route-Number two) to send
data.

V. Performance Evaluation

A. Simulation Environment

In order to demonstrate the effectiveness of IM-DSR
protocol, we evaluate our proposed protocol and compare its
performance to the DSR (uni-route). We have implemented
IM-DSR protocol using the Global Mobile Simulation library

^ G

Information In Cache Memory
N etwork C onne ctions
■ Rep aire d Route by Lo c al Re c o very Process

^ ~~ — ~~ ~ Send Repaired Route packet

I igure 3. Ad-hoc Network Structure, when Occur Local Recovery

(GLOMOSIM) [18]. The Simulation environment consists of
50 numbers of nodes in a rectangular region of size 1500
meters by 1500 meters.

The nodes are randomly placed in the region and each of
them has a radio propagation range of 250 meters. 200
constant bit rate (CBR) flows are deployed for data
transmission. Simulation time is 300 seconds. The random
waypoint model is chosen as the node mobility model. All
data packets are 512 bytes. Band width is 2 Mbps and
simulation done for 0, 1, 3, 5 and 10 second as stop time.
Minimum and maximum speed for nodes are m/s and 30
m/s. IEEE 802.1 1 selected for MAC layer protocol.

B. Simulation Results

Fig. 4 shows packet delivery ratio (PDR) in every two
protocol. It is defined as ratio of the number of data packets
delivered to the destinations generated by sources. In DSR
protocol if sending node of data has not any routes to send
data, it would start route request process by RREQ. In every
request finds a route and if it was invalid, the route request
process begins again. The IM-DSR protocol against DSR
protocol obtains routes in route reply and Maintenance process
and used them for sending data. This would lead to increase
the packet delivery ratio in IM-DSR protocol.

Fig. 5 shows the number of RREQ (NRQ) in every two
protocols. This number is sum of the RREQ in request
process. Fig. 5 shows that this number in DSR protocol is
greater than IM-DSR protocol because in every route
discovery find only one route and if this route would be
invalid begin route discovery again. But IM-DSR finds routes
in route reply, maintenance and local recovery processes,
hence reduces this number.

' Route Repaired Packet

Stop Tirne

Figure 4. Packet Delivery Ratio (PDR)

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

3:C'
250

a 200

150
100

3
Stop Time

a 5

-DSR
- IM-DSR

Stop Time

Figure 5. Number of RREQ (NRQ)

Fig. 6 shows sum of the hop-count (SHC) in the network.
This number is summation of the all routes hop count that
lower value leads to less delay in network. This number is less
in IM-DSR protocol because of less route request process and
found optimal routes.

Fig. 7 shows the number of broken links (NBL) in the
network. Problem such as node mobility, low battery power or
congestion might be raised, which can lead to break an
existing route and lose route connectivity. Number of broken
link in IM-DSR protocol is less than DSR protocol because of
used alternate routes that making in route request, reply and
maintenance processes and used local recovery process.

Fig. 8 shows an average of delay (AVD) in two protocols.
Because of sending data from route with long hop-count and
data packets waiting more time in buffer, delay in DSR
protocol is more than IM-DSR protocol. The IM-DSR
protocol which, between the source and destination, selected
optimal routes and saved multi-route node decreases the delay
in comparisons with DSR protocol.

800 i

700

600

500

ffi 400 -

-■-IM-DSR

300

200 ■

100 -

Stop Tune

;

Figure 6. Sum of Hop-Count (SHC)

4000

3500

3000

2500
I
j 2000

15 00

1000

500

-DSR
-IM-DSR

Figure 8. Average of Delay (AVD)

VI. Conclusion

In this paper, we proposed a new routing protocol called
IM-DSR to provide higher QOS in ad hoc networks. This
protocol is an extension of DSR to increase the reliability by
modifying the route discovery and route maintenance
processes in DSR also added the local recovery techniques to
DSR. The simulation results show that IM-DSR is very
effective in decreasing the packet loss and also increasing the
fault tolerance ad hoc networks. In all of the cases, IM-DSR
has the higher packet delivery ratio than the DSR protocol
while improving the overhead of route maintenance,
maintaining acceptable overhead. Therefore the proposed
routing protocol is very useful for the applications that need a
high level of reliability.

References

[1] V. Nazari Talooki, J. Rodriguez, "Quality of Service for Flat Routing
Protocols in Mobile Ad hoc Networks", Mobimedia'09, London, UK,
September, 2009.

[2] E. M. Royer, "A Review of current Routing Protocols for Ad Hoc Mobile
Wireless Networks", IEEE Personal Communications, vol. 6, no. 2, pp.
46-55, 1999.

'The handbook of ad hoc wireless networks", CRC press,

Figure 7. Number of Broken Link (NBL)

[3] M. IILyas,
2003.

[4] D. Johnson, Et. al, "The Dynamic Source Routing Protocol for Mobile Ad
Hoc Network" 1 1 I m I i I lull www.ietf.org/internetdrafts/ draft-ietf
manet-dsr-10.txt (July 2004).

[5] D. B. Johnson, D. A. Maltz, "Dynamic Source Routing in Ad Hoc
Wireless Networks", Mobile Computing, Edited by Imielinski, Korth,
Chapter 5, pp. 153-181, Kluwer Academic Publishers, 1996.

[6] C. Perkins, E. Royer, "Ad Hoc On-Demand Distance Vector (AODV)
Routing" 1 1 1 1 1 1 i l 1 1 1 1 1 www.ietf.org/internet-drafts/draft-ietfmanet-
aodv-13.txt (Feb 2003).

[7] C. Perkins, E. Royer, "Ad-hoc On-demand Distance Vector Routing", 2nd
IEEE Workshop on Mobile Computing Systems and Applications, New
Orleans, LA, United States, pp. 90-100, Feb. 1999.

[8] Q. Jiang, D. Manivannan, "Routing protocols for Sensor networks", IEEE,
pages 93-98, 2004.

[9]- S. Mueller, R.P. Tsang and D. Ghosal, "Multipath Routing in Mobile Ad
hoc Networks Issues and Challenges", Proceedings of CA

USA,27,USA,2007.

[10] K. Wu, J. Harms, "On-Demand Multipath Routing for Mobile Ad Hoc
Networks", 4th European Personal Mobile Communications Conference
(EPMCC 01), Vienna, Austria, Feb. 2001.

[11] S. Ju-Lee, M. Gerla, "Split Multipath Routing with Maximally Disjoint
Paths in Ad Hoc Networks", IEEE International Conference on
Communications, vol. 10, pp. 3201-3205, Helsinki, Jun. 2001.

[12] R. Leung, E. Poon, A. C. Chan, B.Li, "MP-DSR: A QoS-Aware Multi
path Dynamic Source Routing Protocol for Wireless Ad-Hoc Networks",

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

26th Annual IEEE International Conference on Local Computer
Networks (LCN2001), pp. 132-141, Tampa, FL, United States, Nov.
2001.

[13] X. Li, L. Cuthbert, "On-Demand Node-Disjoint Multipath Routing in
Wireless Ad Hoc Networks", 29th Annual IEEE International
Conference on Local Computer Networks (LCN2004), Tampa, FL,
United States, pp. 419-420, Nov. 2004.

[14] E. Esmaeili, P. Akhlaghi, M. Dehghan and M. Fathi, "A New Multi-Path
Routing Algorithm with Local Recovery Capability in Mobile Ad hoc
Networks", Fifth International Symposium, Patras, Greece, July, 2006.

[15] C. Sengul, R. Kravets, "Bypass Routing: an On-Demand Local Recovery
Protocol for Ad Hoc Networks", Ad Hoc Networks, In press (Available
online 2005).

[16] R. Duggirala, R. Gupta, Q. Zeng, D. P. Agrawal, "Performance
Enhancements of Ad Hoc Networks with Localized Route Repair",
IEEE Transactions on Computers, vol. 52, no. 7, pp. 854-861, 2003.

[17] M. khazaei, R. berangi "A multi-path routing protocol with fault
tolerance in mobile ad hoc networks ", Proceedings of IEEE
international CSI, 14th, tehran, iran, Oct, 2009.

[18] UCLA Parallel Computing Laboratory and wireless Adaptive Mobility
Laboratory, GloMoSim: A Scalable Simulation Environment for
Wireless and Wired Network.

Mehdi khazaei received the bachelor's and master's
degrees from Iran University of Science and
Technology (IUST), in 2004 and 2007, respectively.
Currently He is lectureship in Kermanshah University
of Technology (KUT) and his researches focused on
wireless networks, especially ad hoc networks.

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

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

Vol 8, No. 6, September 2010

Steganalysis of Reversible Vertical Horizontal Data

Hiding Technique

Thorn Ho Thi Huong,

Faculty of Information Technology,

Haiphong Private University,

Haiphong, Vietnam

Canh Ho Van

Dept. of Professional Technique,

Ministry of Public Security,

Hanoi, Vietnam

Tien Trinh Nhat

College of Technology,

Vietnam National University,

Hanoi, Vietnam

Abstract — This paper proposes a steganalysis scheme for
detecting the reversible vertical horizontal (RVH) data hiding [1],
The RVH scheme was introduced in the IJCSIS International
Journal Vol. 7, No. 3, March 2010. In the RVH data hiding, the
message bits are embedded into cover-image by two embedding
phases: the horizontal embedding procedure HEm and the
vertical embedding procedure VEm. The pixel pairs belonging to
the horizontally embeddable and vertically embeddable pixel pair
domain are transformed to mark message bits. Through analysis,
we detect out that, the two histograms of LSB scanning
horizontally and vertically vary from a stego-image to the cover
image. Based on this observation, we design a specific steganlytic
method for attacking the RVH steganography. Experimental
results show the detection accuracies of the steganography with
various embedding rates are acceptable. The proposed technique
can be applied in detecting the misuse of steganographic
technology in malicious activities.

Keywords-Steganography, steganalysis, watermarking, cover image,
stego image, payload, reversible data hiding.

Introduction

Steganography is [1, 3, 4, 5] is the art and science of
concealed communication. The basic concept is to hide the
very existence of the secret message. Digital object such as a
text, image, video, or audio segment can be used as the cover
data. To obtain acceptable hiding payload and keep fidelity of
the stego-image, the LSB replacement techniques [2,4 or other
references] are popular and widely studied in the literature.
These methods usually hide more data in image areas with
higher spatial variations. Reversible steganography [1,3-5] is
one of the interesting branches of steganographic technology in
which the original cover image can be reconstructed without
any loss.

Steganalysis is the counterpart of steganography, the goal
of the steganalysis is to detect the hidden message,
equivalently, to discriminate the stego object from the non-
stego-object. The steganalysis techniques proposed in the
literature can be classified into two categories: the universal
steganalysis which is designed to detect the hidden message
embedded with various data embedding algorithms such as a
technique proposed in [6] is used to attack the LSB
steganography, and the specific steganalysis which is designed

to attack a specific steganography technique such as a
steganalytic method was presented in [2] for detecting stego-
images using the method proposed in [3].

In this paper, we proposed a steganalytic scheme to detect
the RVH watermarking scheme introduced in brief in the
abstract. Our experimental results show the feasibility of the
proposed method. It is useful in detecting malicious activities
on stego-images and also suggests a design consideration for
future development of steganographic techniques. The rest of
this paper is organized as follows. In the next section, we
present again the RVH scheme in brief. Section III describes
the proposed steganalytic method. Experimental results are
given in section IV, and conclusions are made finally in
Section V.

II. Review of the RVH data hiding scheme

In the steganographic method proposed in [1] used the
multiple embedding strategies to improve the image quality
and the embedding capacity. Basically, this method embeds
each message bit b of the secret bit stream into each grayscale
cover pixel pair of a grayscale cover image in raster scan order.
This scheme includes two main stages, namely, the horizontal
embedding procedure HEm and the vertical embedding
procedure VEm. For the HEm procedure, the input image is
horizontally scanned in raster scan order (i.e., from left to right
and top to bottom) to gather two neighboring pixels x and y
into a cover pixel pair (x, y). If y is an odd value, then the cover
pixel pair (x, y) is defined as a horizontally embeddable pixel
pair. Otherwise, the cover pixel pair (x, y) is defined as a
horizontally non-embeddable pixel pair. For the VEm
procedure, the input image is vertically scanned in raster scan
order to group two neighboring pixels u and v into a pixel pair
(u, v). If v is an even value, then the pixel pair (u, v) is defined
as a vertically embeddable pixel pair. Otherwise, the pixel pair
(u, v) is defined as a vertically non-embeddable pixel.

The secret bit sequence S is divided into two subsequence
SI and S2. The bit stream Bl is created by concatenating the
secret subsequence SI and the auxiliary data bit stream Al
(i.e., B1=S1||A1). Similarly, the bit stream B2= S2||A2. The
generation of Al and A2 will be described latter. The overview
of the RVH embedding process is shown in Fig. 1.

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

Fig. 1. Embedding phase of RVH steganographic system [1]

Firstly, the bit sequence Bl is horizontally embedded into
O by using the HEm procedure to obtain the output image T
sized H x W pixels. Secondly, the compressed location map
CM1 whose length is LCI (will be described later), is
embedded in to T by using the least significant bit (LSB)
replacement technique to obtain the output image U with size
of H x W pixels. Thirdly, the bit sequence B2 is vertically
embedded into U by using the VEm procedure to get the output
image V with size of H x W pixels. Fourthly, the compressed
location map CM2 whose length is LC2 is embedded into V by
using the LSB replacement technique to get the final stego
image with size of H x W pixels.

Each bit b in stream Bl is horizontally embedded into each
horizontally embeddable pixel pair (x, y) at a time by using the
horizontal embedding rule HR defined below until the whole
bit stream Bl is completely marked into O to obtain the output
image T.

Each bit b in B2 is vertically embedded into each vertically
embeddable pixel pair (u, v) at a time by using the vertical
embedding rule VR defines below until the entire bit sequence
B2 is concealed into U to get the output image V.

The horizontal embedding rule HR: For each pair (x, y), we
apply the following embedding rules:

• HR1: If the tobeembedded bit b=l, then the stego
pixel pair is unchanged by (xO, yO) = (x, y).

• HR2: If the tobeembedded bit b=0, then the stego
pixel pair is changed by (xO, yO) = (x, y-1).

The vertical embedding rule VR: For each pair (u,v), we
apply the following embedding rules:

• VR1 : If the tobeembedded bit b=0, then the stego
pixel pair is unchanged by (uO, yO) = (u,v).

• VR2: If the tobeembedded bit b=l, then the stego
pixel pair is changed by (uO, yO) = (u, v+1).

It is noted that the rule HR and VR don't cause the
underflow and overflow problem. That is the changed pixel
pairs are assured to fall in the allowable range [0,255].

The auxiliary data bit sequence Al is actually the LSBs of
the first LCI (LCI is the length of the compressed location
map CM1 ended with the unique end of map indicator EOM1)
pixels in the image T and generated as follows. Initially, Bl is

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

Vol 8, No. 6, September 2010
equal to SI (i.e., B1=S1). During the execution of the HEm
procedure, for the first LCI pixels in O, when each pixel has
been processed for embedding, its LSB is taken as an auxiliary
data bit of Al and appended to the end of Bl. That is, Bl is
gradually grown until the LCI auxiliary data bits in Al are
concatenated into Bl. Finally, the tobeembedded bit stream
is B1=S1||A1, which is completely embedded into O.

Similar to the generation of Al, the auxiliary data stream
A2 is actually the LSBs of the first LC2 (LC2 is the length of
the compressed location map CM2 ended with the unique end
of map indicator EOM2) pixels in the image V and generated
as follows. B2 initially equals the secret bit sequence S2.
During the execution of the procedure VEm, for the first LC2
pixels in the image U, when each pixel has been processed for
embedding, its LSB is taken as an auxiliary data bit of A2 and
append to the end of B2 until the LC2 auxiliary data bits in A2
are concatenated into B2. Finally, the information bit sequence
is B2=S2||A2, which is fully marked into the image U.

For the purposes of extracting Bl and recovering O, a
location map HL sized H x (W/2) is needed to record the
positions of the horizontally embeddable pixel pair (x, y) in O.
The location map HL is a one-bit bitmap. All the entries of HL
are initialized to 0. If cover pixel pair (x, y) is the horizontally
embeddable pixel pair, then the corresponding entry of HL is
set to be 1. Next, the location map HL is losslessly compressed
by using the JBIG2 codec (Howard et al, 1998 [8]) or an
arithmetic coding toolkit (Carpenter, 2002 [7]) to obtain the
compressed location map CM1 whose length is LCI. The
compressed location map CM1 is embedded into the image T
by using the LSB replacement technique as mentioned above.
Similarly, for the purposes of extracting B2 and recovering the
image U, we also need a location map VL sized (H/2) x W to
mark the position of the vertically embeddable pixel pairs (u, v)
in U. Then, VL is also losslessly compressed by using the
JBIG2 codec or an arithmetic coding toolkit to obtain the
compressed location map CM2 whose length is LC2. Next, the
map CM2 is concealed into the image V by using the LSB
steganography as mentioned above.

The final output of the embedding phase is the final stego
image X with size of H x W pixels.

III. The Proposed Steganalytic Scheme for the RVH
Steganography

After embedding a large message sequence M (its ratio is
about 90% of maximum embeddable capacity of image) into
the original image Baboon sized 512x512 pixels (show Fig. 2)
using the RVH scheme to obtain the stego-image Baboon, we
calculate histogram of the two images (cover Baboon image
and stego Baboon image), resulted in Fig. 3. It's very hard to
detect any difference between the two images.

However, when we separably calculate two histograms on
all pixel odd columns and all pixel even rows of the cover
Baboon image, shown in Fig. 4. Similarly, calculate two
histograms on all pixel odd columns and all pixel even rows of
the stego Baboon image, resulted in Fig. 5. It's easy to
difference between pair histogram in Fig.4 (a) and Fig. 5 (a), in
Fig. 4 (b) and Fig. 5 (b). The informality appears in Fig 5 (a)

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

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

Vol 8, No. 6, September 2010
and (b) due to embedding process of RVH scheme following
description in detail below. 20

Fig. 2. The Baboon image sized 512x512 pixels

(a)

Fig. 3. Histogram of the tested two images: (a) the cover Baboon
(b) the stego Baboon image

(b)
image,

=>(a)

Fig. 4. Histogram of the cover Baboon images: (a) histogram on all
columns, (b) histogram on all pixel even columns

300 (b)

pixel odd

(a)

300 ( b )

Fig. 5. Histogram of the stego Baboon images: (a) histogram on all pixel odd
columns, (b) histogram on all pixel even columns

According to the horizontal embedding procedure HEm,
from an input image O, the pixels of the image O are
horizontally grouped into pixel pairs (x, y), these pairs are
partitioned into two sets, one set is El and other set is El, the
set El contains pixels pair which are horizontally embeddable
pixel pairs, while the set El consists of those pixel pairs which
are horizontally non-embeddable pixel pairs.

Now, we examine the migration of LSB histogram of the
image O and the image T obtained after embedding secret bit
Bl. Without loss of generality, let (x, y) and (x,y) be the
corresponding pixel pairs in the image O and the stego-image
T, respectively. In the horizontal embedding procedure HEm,
pixel pairs (x, y) E El, i.e. the LSB of pixel y be bit 1, are
selected to embed message bits. Here, We don't examine
change of LSB histogram of pixels x on pixel even-columns
because they are still remained value after embedding message
bits. In the image T, the LSB of y is changed to either or 1,
and each of them appears in the same probability. It is
obviously that the probability of bit and bit 1 are 0.5 and 0.5,
respectively. For pixel pairs (x, y) e El, i.e. the LSB of pixel y
be bit 0, after embedding secret bits, y is unchanged. So the
probability of bit and 1 are 1 and 0, respectively.

Next, the compressed location map CM1 (CM1 is a binary
stream, whose length is LCI) are marked into the image T by
the LSB replacement technique to obtained image U. That
changes a part of probability of LSB of bit 1 and bit on all
pixel even-columns in the image T. Assume that the bits are
randomly distributed, so the probability of bit and bit 1 are

Pmapl(O) = Pmapl(l)-

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

Base on above discussions, the probabilities of bit and bit
1 of all pixels on even-column in the image U can be
calculated. Assume the probability of pixel pairs belonging to
El and the probability of pixel pairs belonging to El be P E1
and Pet, respectively. After marking the location map CM1,
P E i and Pet are changed to P' E1 and P'et • Let P R . H is the
embedding ratio defined by dividing the number pairs actually
used to hide data by the total number of pairs the image O. The
probability of bit b={0,l} of LSB of the image U can be
calculated using the following equation

p rM _ [Pr-h x (0-5 x P' E1 + P'ei) + P^h x 0.5 if b =
n S B_HW - [ Prh x (q 5 x p , Ei) +P _ x0i5 ifb = l ( }

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

Vol 8, No. 6, September 2010
used to embed message, i.e. the embedding ration of P R _
H =0.45 X 0.9=0.405. From (1), we have P L sb-h(0)
=0.405x(0.5x0.45+0.55) + 0.595x0.5 = 0.611375 and P LSB -
H (l)=0.405 X (0.5 X 0.45)+0.595 x 0.5 = 0.388625. Next,
calculated the probability of bit and the probability of bit 1 of
the output image X. We know that the probability of E2 equals
to the probability of the LSB of bit of all pixels on even -
rows, i. e. P E2 = P LSB (0)/2 + P LSB h(0)/2 = (0.5+0.61 1375)/2 =
0.5556875 and ?^2 =0.4443125. After covering a part of the
LSB of image V by the location map CM2 with a probability
0.05 (assumed), the two probability P E2 and ?^2 change to
P' E2 =0.5056875 and P'e2= 0.4943 125, respectively.

For the vertical embedding procedure VEm, vertically scan
the output image U in raster scan order to group pixel pairs (u,
v), we classify the pairs into two sets, one is E2 and other is
E2, the set E2 contains all pixel pairs which are vertically
embeddable pixel pairs, the set E2 consisting pixel pairs are
vertically non-embeddable pixel pairs. Let (u, v) and (u, v) be
pixel pairs of the image U (before using the procedure Vem)
and the stego-image V (after embedding secret message using
the procedure Vem). In the procedure VEm, only pixel pairs (u,
v) E E2, i.e. the LSB of v is bit 0, are embedded message bits.
After embedding message bits, the LSB of the pixels v
(obtained from E2) is either or 1. So, the probabilities of bit
and bit 1 of (u, v) are equals to 0.5. For pixel pairs (u, v) e E2,
i.e. the LSB of pixel v be bit 1, after embedding secret bits, v is
unchanged. So the probability of bit and 1 are and 1,
respectively.

Next, the compressed location map CM2 are marked into the
image V by the LSB replacement technique to obtained image
X. That changes a part of probability of LSB of bit 1 and bit
on all pixel even-rows in the image V. Assume the bits of the
map CM2 are randomly distributed, so the probability of bit
and bit 1 are P map2 (0) = P map 2(l)-

From above discussions, the probabilities of bit and bit 1
in the LSB of image X after using the vertical embedding
procedure VEm can be calculated. Assume the probability of
pixel pairs belonging to E2 and the probability of pixel pairs
belonging to E2 be P E2 and P^l, respectively. After marking
the location map CM2, P E2 and Pe2 are changed to P' E2 and
P'e2- Let P R _ V be the embedding ratio defined by dividing the
number pairs actually used to hide data by the total number of
pairs the image V. The probability of bit b={0,l} of LSB of
the image V can be calculated using the following equation

Ab) = \ P p R -

P R . V x (0.5 x P' E2 ) + P R - =V x 0.5

ifb =

x (0.5 x P' E2 + p' m ) + P-^y x 0.5 ifb = l

(2)

For a natural image, assume that the LSB is randomly
distributed, then the expected probability of bit and the
probability bit 1 of all pixel on pixel even-columns are the
same, i.e. P LS b(0) = P L sb(1)=0.5. So probability P E i=0.5,
Pet = 0.5. After covering a part of the LSB of image T by the
location map CM1 with a probability 0.05 (assumed), the two
probability P E1 and Pet change to P' E i=0.45 and P'et=0.55,
respectively. Consider the Baboon stego-image from the
Baboon cover image, the probability of the embeddable pairs
(i.e. those pixel pairs belonging to the procedure HEm) of an
input image T is P' E1 , and 90 % of the embeddable pairs are

The embedding ratio of 90 % of the embeddable pairs are
used to embed message, i.e. the embedding ration of P R _
v=0.5056875 x0.9=0.45511875. So probability of LSB of bit
and bit 1 of the output image X from (2) we obtain P LSB -
v(0)=0.4551 1875 X (0.5 X 0.5056875) + 0.54488125 X 0.5
^0.3875, P LSB _v(l)^0.61248.

Now, we check again the probability of LSB of bit 1 and the
probability of LSB of bit of all pixels on pixel even-columns,

PLSB even column(0)=P LS B H (0)/2 + P LSB V (0)/2 =(0.61 1375+

0.3875)/2=0.4994375,~ P LSB even coiumnO) = (Plsbh(I) +
Plsb_v(0))/2=(0.388625+0.61248)/2=0.5005525. We found out
that the probability of bit P L sv_even coiumn(O) and probability of
bit 1 P LSB even coiumn(l) are the same, that is after completing the
vertical procedure VEm, it make the value of these
probabilities be balanced. However, the probability of LSB of
bit and bit 1 of all pixels on pixel odd-columns don't equal
based on the following calculating:

PLSB_odd_column(0) = (PLSB_org_odd_column(0)/2+P LSB _v(0)/2)=(0 .5/2 +
0.3875/2)=0.44375, PLSB_odd_column(l)KPLSB_org_odd_column(iy2 +

Plsb_v(1)/2)=(0.5/2+0.612"48/2)=0.55624. Where

PLSB_or g _odd_coiumn(0) and P L sB_or g _odd_coiumn( 1 ) be the probabilities
of the LSB of bit and bit 1 of all pixels on the pixel odd-
column of the image X. A half of them isn't changed during
process of the RVH scheme, so P L sB_or g _odd_coiumn(0)/2 and
PLSB_or g _odd_coiumn(l)/2 equal to 0.5/2 and 0.5/2, respectively. We
can see obvious difference of the occurrences of bit 1 and bit
in the LSB on all pixel oddcolumns and all pixel
evencolumn of the stego-image of the RVH scheme with
respect to a standard natural image. Based on the problem, the
following rule is given to discriminate a stego-image of the
RVH steganography from a nature image.

W(X)

(true,if\P LSB (0)-
\ false, otherwise

Plsb(X)\>T

(3)

From equation (3), an image is detected be stego-image
marked by the RVH scheme if one of the measured values
Plsb(O) - Plsb(1)| on all pixel odd columns (or pixel even
columns) or pixel even rows (or pixel odd rows) is greater than
threshold T (0 < T < 1). The threshold T is used to control the
decision boundary of nature images and stego images, its value
depends on specific applications.

IV. EXPERIMENTAL RESULT

To show the reliability of the proposed method, we take
500 image from USC-SIPI Image Database [9] and content
based image retrieval (CBIR) image database [10] and convert
them into 8-bit grayscale images. The images are used to test

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

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

Vol 8, No. 6, September 2010

proposed classification. The hidden messages used in our test
are made by the pseudo random number generator. We embed
different amount of message using the RVH scheme, and
measure the migration of LSB histogram in the stego images.
Five embedding ratios 0%, 25%, 50%, 75% and 100% are used
in the test, and the obtained |Plsb(0)-Plsb(1)| values on all pixel
even-rows of the stego-images are depicted in Fig. 6-10,
respectively. We also measure the accuracy of the proposed
method in detecting the RVH scheme in different embedding
ratio and likelihood threshold value T, shown in table 1.

0.2

§ 0.15

|Plsb(0) -
o

[(

h i

0.05

; ii

gb-.

50 100 150 200

250 300

350 400 450 500

Number of images

Fig

6. The distribution of |P L sb(0) - Plsb(1)| value of the 500 cover images on
all pixel even-rows

0.35 c

0.3

0.25

If 0.2

- T

«f °- 15

t * t t

\\w

-^HhH

0.1

0.05

4^^

^ ^£#^

I A. ^jfa^J" ^ Jast^^

r i

v -.-■-■

50 100 150 200 250 300

350

400 450 5(

Number of images

Fig.

7. The distribution of |P L sb(0) - Plsb(I)] value of the 500 stego images on
all pixel even-rows with embedding ratio 25%

S' 0.2 H

50 100 150 200 250 300 350 400 450 500
Number of images

Fig. 8. The distribution of |Plsb(0) - Plsb(1)| value of the 500 stego images on
all pixel even-rows with embedding ratio 50%

200 250 300
Number of images

350 400 450

Fig. 9. The distribution of |Plsb(0) - Plsb(1)| value of the 500 stego images on
all pixel even-rows with embedding ratio 75%

0.5

0.45

0.4

0.35

I 1

1 03

^ 0.25-

-Q

J2 0.2

Q-

0.15

0.1

0.05

150 200 250 300
Number of images

Fig. 10. The distribution of |P L sb(0) - Plsb(1)| value of the 500 stego images on
all pixel even-rows with embedding ratio 100%

TABLE I. The detection accuracy of the proposed method with

VARIOUS EMBEDDING RATIOS AND THRESHOLD VALUES

\. Embedding
^^atio (%)

ThresholdT\

100

0.01

Cover

70.6 %

0.6%

Stego

29.4 %

98%

99.4 %

100%

0.02

Cover

80.6 %

5.8 %

0.6 %

Stego

19.4 %

94.2 %

99.4 %

100%

0.03

Cover

84.4 %

7.4 %

2.2 %

Stego

15.6%

92.6 %

97.8 %

100%

0.04

Cover

86.8 %

11%

2.4 %

Stego

13.2%

89%

97.6 %

100%

0.05

Cover

89%

13.8 %

3.6%

0.2 %

Stego

11%

86.2%

96.4 %

99.8%

100%

From Fig. 6 , we can see that most of the value of |P L sb(0) -
Plsb(1)| approach zero for natural images, while the higher
value of |P L sb(0) - Plsb(1)| is obtained with embedding ratio
25%, 50%, 75% and 100% shown from Fig. 7 to Fig. 10. From
table 1, we see that when the likelihood threshold value T is set

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

0.035, we can obtain an acceptable result in detecting stego
images used the RVH steganography.

V. Conclusion

The paper presents a method to break the RVH
steganography based on the observation of the distribution of
and 1 bits of the LSBs on pixel odd-columns (pixel even-
columns) or pixel even-rows (pixel odd-rows) of the RVH
stego-images. The experimental results are shown that the
proposed method can detect stego - images reliably with
embedding ratio being greater 25%. On the other hand, we
show a problem of security of the RVH scheme in the data
embedding process.

Acknowledgment

Our special thanks to Haiphong Private University (HPU)
for their financial support to our research and College of
Technology, Vietnam National University, Hanoi for their
support to good working environment. We would like to
extend our thanks to my guide, our friends and family
members without whose inspiration and support our efforts
would not have come to success.

References

[1] P. Mohan Kumar, K. L. Shunmuganathan, A reversible high embedding
capacity data hiding technique for hiding secret data in images,
International Journal of Computer Science and Information Security,
Vol.7, No. 3, March 2010, pp. 109-115.

[2] Yeh-Shun Chen, Ran-Zan Wang, Yeuan-Kuen Lee, Shih-Yu Huang,
Steganalysis of reversible contrast mapping water marking, Proceedings
of the world congress on Engineering 2008 Vol I, WCE2008, July 2-4,
2008, London, U.K., pp. 555-557.

[3] D. Coltuc and J. M. Chassery, " Very fast watermarking by reversible
contrast mapping," IEEE Signal Processing Lett., vol. 14, no. 4, pp.
255- 258, Apr. 2007.

[4] J. Tian, " Reversible Data embedding using a difference expansion,"
IEEE Trans. Circuits Syst. Video technol., vol. 13, no. 8, pp. 890- 896,
Aug. 2003.

[5] Z. Ni, Y. Q. Shi, N. Ansari, and W. Su, " Reversible Data Hiding,"IEEE
Trans. Circuits Syst. Video technol., vol. 16, no. 3, pp. 354- 362, Mar.
2006.

[6] J. Fridrich, M. Goljan, and R. Du, " Reliable detection of LSB
steganography in color and grayscale images,"Proceedings of the ACM
International Multimedia Conference and Exhibition, pp. 27- 30, 2001.

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

Vol. 8, No. 6, September 2010
[7] Carpenter, B., 2002. Compression via Arithmetic Coding
http://www.colloquial.com/ArithmeticCoding/

[8] P.G. Howard, F. Kossentini, B. Martins, S. Forchhammer, W. J.
Rucklidge, 1998. The emerging JBIG2 standard. IEEE Transactions on
Circuits and Systems for Video Technology 8 (7), pp. 838-848.

[9] USC-SIPI Image Database,

"http://sipi.usc.edu/services/database/Database.htm"

[10] CBIR Image Database, University of Washington,
http ://www. cs .Washington, edu/research/imagedatabase/groundtruth/. .

AUTHORS PROFILE

Ho Thi Huong Thorn received the B.S. degree of
Information Technology department from Haiphong
Private University and the M.S. degree in Information
Systems from College of Technology, Vietnam National
University in Vietnam, in 2001 and 2005, respectively.
She has started her career as Lecturer in Department of Information
Technology in Haiphong Private University, Vietnam and served for 9 years.
Currently, she is pursuing Doctor of Information Systems from College of
Technology, Vietnam National University, Hanoi, Vietnam. Her research
interests includes Image processing, Information Security, Information Hiding.

Ho Van Canh received the B.S. degree in Mathematics
from Hanoi City University in Vietnam in 1973, the Dr.
Sci. degree in Faculty of statistology from KOMENSKY
University in Czechoslovakia in 1987. Currently, he has
been working as a cryptologist in Dept. of Professional
Technique, Ministry of Public Security, Vietnam. His

research interests include cryptography, information security, information

hiding.

Trinh Nhat Tien received the B.S degree from University
of Prague in Czechoslovakia in 1974, and the Dr. degree
from University of Prague, Czechoslovakia and University
of Hanoi, Vietnam in 1984. He has started as Lecturer in
Department of Information Technology of College of
Technology, Vietnam National University, Hanoi, Vietnam since 1974. His
research interests include algorithm, complexity of algorithm, information
security.

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

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

Vol. 8, No. 6, 2010

Off-line Handwritten Signature Recognition
Using Wavelet Neural Network

Mayada Tarek

Computer Science Department,

Faculty of Computers and Information

Sciences,

Mansoura, Egypt

Taher Hamza

Computer Science Department,

Faculty of Computers and Information

Sciences,

Mansoura, Egypt

Elsayed Radwan

Computer Science Department,

Faculty of Computers and Information

Sciences,

Mansoura, Egypt

Abstract^Automatic signature verification is a well-
established and an active area for research with numerous
applications such as bank check verification, ATM access,
etc. Most off-Line signature verification systems depend
on pixels intensity in feature extraction process which is
sensitive to noise and any scale or rotation process on
signature image. This paper proposes an off-line
handwritten signature recognition system using Discrete
Wavelet Transform as feature extraction technique to
extract wavelet energy values from signature image
without any dependency of image pixels intensity. Since
Discrete Wavelet Transform suffers from down-sample
process, Wavelet Neural Network is used as a classifier to
solve this problem. A comparative study will be illustrated
between the proposed combination system and pervious
off-line handwritten signature recognition systems.
Conclusions will be appeared and future work is proposed.

Keywords -Discrete Wavelet Transform (DWT); Wavelet
Energy; Wavelet Neural Network (WNN); Off-line
Handwritten Signature.

I. INTRODUCTION

In the field of personal identification, two types of
biometrics means can be considered; first, physiological
biometrics, which involves data derived from the direct
measurement of some part of the human body; for-
example fingerprint-, face-, palm print-, retina-based
verification. Second, behavioural biometrics, which
involves data derived from an action taken by a person,
or indirectly measures characteristics of the human
body; for-example: speech-, keystroke dynamics and
signature-based verification [1].

In the last few decades, researchers have made great
efforts on off-line signature verification [1] for-
example; using the statistics of high grey-level pixels to
identify pseudo-dynamic characteristics of signatures;
developing technique based on global and grid features

1 Corresponding Author
Mail: mayaatarek @ yahoo . com
Tel : 020108631688

in conjunction with a simple Euclidean distance
classifier; proposing a system for off-line signature
verification consists of four subsystems based on
geometric features, moment representations, envelope
characteristics and wavelet features; applying wavelet
on signature verification [2,3,4,5].

Although these methods achieved a good results, they
still suffer from the exchangeability of signature
rotation and the distinguish-ability of person signature
size. Most of these feature extraction methods depend
on signature shape or pixels intensity in specific region
of signature. However, pixels' intensity are sensitive to
noise and also the signature shape may vary according
to translation, rotation and scale variations of signature
image [6].

Two types of feature can be extracted from signature
image; first, global features which are extracted from
the whole signature, including block codes [7]; second,
local features which are calculated to describe the
geometrical and topological characteristics of local
segments [8]. Because of the absence of dynamic
information in offline verification system, global
features extraction are most appropriate [9]. One of the
most appropriate global features extraction techniques is
wavelet transform, since it extracts time-frequency
wavelet coefficients from the signature image [8].
Wavelet Transform is especially suitable for processing
an off-line signature image where most details could be
hardly represented by functions, but could be matched
by the various versions of the mother wavelet with
various translations and dilations [10]. Also, wavelet
transform is invariant to translation, rotation and scale
of the image. Because of the advantage of wavelet
transform, this paper uses it in feature extraction stage.

Since one of problems that face wavelet is the huge size
of its coefficients, statistical model can be introduced to
represent them. This paper uses wavelet energy as
statistical model to represent all wavelet coefficients in
efficient way. Another problem is down-sample process
which can lose some important extracted feature from
signature image[ll]. This paper proposes a Wavelet
Neural Network (WNN) technique for off-line signature
recognition to overcome the disadvantages of Discrete
Wavelet Transform (DWT) down- sample process.

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

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

Vol. 8, No. 6, 2010

WNN takes full advantages of the partial-resolution
characteristic of the wavelet transform and the nonlinear
mapping behaviour of Artificial Neural Networks
(ANN) [15].

This paper proposes a combination model between
DWT and WNN techniques for off-line handwritten
signature recognition system. DWT technique will
analysis signature image to extract wavelet detail
coefficients. To reduce the huge number of these
coefficients with the same accuracy, a statistical model
is represented by wavelet energy. Because of the
problem of down sample, WNN technique will be used
as a suitable classifier technique to overcome this
problem. Also, a modified back-propagation technique
is used in learning WNN. A testing stage examines the
unseen signature. Moreover, a comparative study will
be illustrated between the proposed combination system
and pervious off-line handwritten signature recognition
systems. Conclusions will be appeared and future work
is suggested.

The rest of this paper organized as; in Section 2,
Handwritten signature, wavelet transform (WT),
Wavelet Neural Network (WNN) are mentioned.
Methodology and applications using a combination
between DWT and WNN techniques is described in
Section 3. Section4, consists of the result of the
proposed combination system and a comparative study
between three strategies (signature image pixels
intensity value as input to ANN , signature wavelet
energy values as input to ANN and signature wavelet
energy values as input to WNN). Finally section 5
concludes the paper.

II. PRELIMINARIES

A. Handwritten Signature

Handwritten signatures are widely accepted as a means
of document authentication, authorization and personal
verification. For legality most documents like bank
cheques, travel passports and academic certificates need
to have authorized handwritten signatures. In modern
society where fraud is rampant, there is the need for an
automatic Handwritten Signature Verification system
(HSV) [6]. Dependency on automation is due to the
difficulty faced in visual assessment for different types
and different sizes of signatures. Simple, cursive,
graphical and not a connected curve pattern are some of
the different types of signatures and machines are far
superior when it comes to processing speed and
management of large data sets with consistency [12].

Automatic HSV systems are classified into two types:
offline HSV and online HSV: static or off-line system
and dynamic or on-line system .Static off-line system
gain data after writing process has been completed .In
this case the signature is represented as a grey level

image. Dynamic systems use on-line acquisition devices
that generate electronic signals representative of the
signature during the writing process [1].

It is well known that no two genuine signatures of a
person are precisely the same and some signature
experts note that if two signatures written on paper were
same, then they could be considered as forgery by
tracing .Unfortunately, off-line signature verification is
a difficult discrimination problem because of dynamic
information regarding the signing velocity, pressure and
stroke order are not available also an off-line
handwritten signature is depend for instance on , the
angle at which people sign may be different due to
seating position or due to support taken by hand on the
writing surface and all this information can't be extract
from static image [12].

B. Wavelet Transform :

Wavelet Transform (WT) [13] is become a powerful
alternative analysis tool to Fourier methods in many
signal processing applications. The main advantages of
wavelets is that they have a varying window size, being
wide for slow frequencies and narrow for the fast ones,
thus leading to an optimal time -frequency resolution in
all the frequency ranges. Furthermore, owing to the fact
that windows are adapted to the transients of each scale,
wavelets lack the requirement of stationary. There are
two types of Wavelet Transform; Continous Wavelet
Transform(CWT), Discrete Wavelet Transform (DWT).

The Continuous Wavelet Transform [14] of a 1-D signal
x(t) is defined as in equation (1):

^(a,b)(t)= ^^(t) V{±) dt

(1)

Where t//(t) is the mother wavelet or the basis function
which, in a form analogous to sins and cosines in
Fourier analysis. All the wavelet functions used in the
transformation are derived from the mother wavelet
through translation (shifting) b and scaling (dilation or
compression) a.

The Discrete Wavelet Transform [14], which is based
on sub -band coding is found to yield a fast computation
of wavelet transform. It is easy to implement and
reduces the computation time and resources required.

In CWT, the signals are analyzed using a set of basis
functions which relate to each other by simple scaling
and translation. In the case of DWT, a time-scale
representation of the digital signal is obtained using
digital filtering techniques. The signal to be analyzed is
passed through filters with different cut off frequencies
at different scales [14].

In DWT, the extension to 2-D is usually performed by
using a product of 1-D filters. The transform is
computed by applying a filter bank as shown in
Figure 1. L and H to denote the 1-D low pass and high

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

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

Vol. 8, No. 6, 2010

pass filter, respectively. The rows and columns of image
are processed separately and down sampled by a factor
of 2 in each direction which may cause losing important
feature. Resulting in one low pass image LL and three
detail images HL, LH, and HH. Figure 2a shows the
one-level decomposition of Figure 7 in the spatial
domain. The LH channel contains image information of
low horizontal frequency and high vertical frequency,
the HL channel contains high horizontal frequency and
low vertical frequency, and the HH channel contains
high horizontal and high vertical frequencies. Three-
level frequency decomposition is shown in Figure 2b.
Note that in multi-scale wavelet decomposition only the
LL sub-band is successively decomposed [13].

rows

Image .

-©-

columns

-*©- HH

r+ H

rj->@= HL

L -©

H_}-^> LH

^^2> LL
Figure 1 : A one-level wavelet analysis filter bank.

(a) L-level Decomposition. (b) 3 -level Decomposition.

Figure 2 : Wavelet frequency decomposition.

C. Wavelet Neural Network :

neuron parameters .The output of WNN is therefore a
linear combination of several multidimensional
wavelets [15].

Figure 3 : The structure of the Wavelet Neural Network

In this WNN model, the hidden neurons have wavelet
activation functions y/ and have two parameter a t ,b t
which represent dilation and translation parameter of
wavelet function and V is the weight connecting the
input layer and hidden layer and U is the weight
connecting the hidden layer and output layer.

Let X n ={ x { },i=l, ,L and n=l, N be the WNN

input to no. n sample ; T ={ j k },k=l, ,S represents

the output of WNN ; D={ d k },k=l, ,S represents the

expected output ; Vy represents the connection weight
between no. i node (input layer) and . j node (hidden
layer) ; U jk represents the connection weight between
no. j node (hidden layer) and k node (output layer) .
Where N is the number of Sample ; S is the number of
output node ; L is the number of input node ; M is the
number of hidden layer.

III. WAVELET NEURAL NETWORK FOR
OFF-LINE HANDWRITTEN SIGNATURE
RECOGNITION

WNN is a combination technique between neural
network and wavelet decomposition .The advantages of
the WNN are a high-speed learning and a good
convergence to the global minimum [15]. The reason for
the application of WNN in case of such a problem as
classification is that the feature extraction and
representation properties of the wavelet transform are
merged into the structure of the ANN to further extend
the ability to approximate complicated patterns [16].

The WNN can be considered an expanded perceptron
[17]. The WNN is designed as a three-layer structure
with an input layer, a wavelet layer, and an output layer.
The topological structure of the WNN is illustrated in
Figure 3.

In WNN, both the position and dilation of the wavelets
as well as the weights are optimized. The basic neuron
of a WNN is a multidimensional wavelet in which the
dilation and translation coefficients are considered as

According to the fact that there aren't two genuine
signatures of one person are precisely the same, many
efforts have been done in order to comprehend the
delicate nuances of person signatures [12]. Especially
off-line signature recognition needs more effort because
of the absence of dynamic information that can't be
extracted from static image [12]. Also, the problems of
translation, rotation and scale variation of signature
image are still found when dealing with signature image
pixels' intensity [6].

This paper presents an implementation for off-line
handwritten signature recognition system using DWT
technique in feature extraction phase and WNN in
classification phase to overcome all the above problems
with off-line handwritten signature recognition system.
DWT technique depends on analyzing all signature
shapes (continuous case) instead of analyzing the pixels
intensity or segmentation part of signature (discrete
case). Because of the problem of down-sample caused

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

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

Vol. 8, No. 6, 2010

by DWT technique, WNN technique will be used in
classification stage to overcome this problem.

The proposed Off-line Handwritten Signature
recognition system as depicted in Figure 4 involves
four stages:

♦ Scan and removing noise stage.

♦ Feature extraction stage.

♦ Classification stage.

♦ Test stage.

First stage, Scan and removing noise stage, each off-
line handwritten signature is scanned due to creating
signature image. Because of the scanning process,
removing noise from signature image is an important
task. In this paper, the median filter [18] is used to
remove noise for two reasons. First, it preserves the
structural shape of the signature without removing small
strokes. Second, the absence of dealing with median
filter in wavelet transform technique, which work to
analysis image with low/high-pass filters corresponding
to its wavelet function.

The median filter is a nonlinear digital filtering
technique which is often used to remove noise. Noise
reduction is a typical pre-processing step that improves
the results. The median filter considers each pixel in the
image in turn and looks at its nearby neighbours to
decide whether or not the pixel intensity value is
representative of its surroundings. The median filter
replaces the pixel with the median of its neighbouring
pixel intensity values. The median is calculated by first
sorting all the pixel intensity values from the
surrounding neighbourhood into numerical order and
then replacing the pixel being considered with the
middle pixel intensity value [19].

Second stage, Feature extraction stage is the most
important component for designing the intelligent
system based on pattern recognition. The pattern space
is usually of high dimensionality. The objective of the
feature extraction is to characterize the object by
reducing the dimensionality of the measurement space
(i.e., the original waveform). The best classifier will
perform poorly if the features are not chosen well [20].

According to the fact that there aren't two genuine
signatures of one person are precisely the same, the
differences in the same person signature may exist in
details. Because of the details of an image will access
by high pass filter, DWT is used to access high pass
information of person's signature images. This
information is fused to obtain pattern of each person's
signatures that contains all details information of his/her
signatures [21]. Details information extracted by DWT
technique must be extracted using suitable wavelet
function to off-line handwritten signature recognition
application. According to the previous work in off-line
handwritten signature recognition have apply
Daubechies 4, 12 and 20 wavelets functions as depicted
in Figure 5 [5] as a mother wavelet function, which can
preserve maximum details of the original image, reflect
outline of the image objectively and decrease the FRR.

Figure 5: Daubechies 4, 12 and 20 wavelets functions

After DWT is applied on the image, wavelet
coefficients from the approximation sub-band is discard
and interested in wavelet coefficients from the details
sub-bands of all the decomposition levels . This entire
coefficient is very large to be used as feature extraction
model from an image. These wavelet coefficients can be
represented as statistical features such as mean, median,
standard deviation, energy and entropy [22]. In this
paper, wavelet energy values for details wavelet sub-
band is the reduced vector that contain the main
information that represent person signature from the
huge wavelet decomposition values.

While off-line handwritten signature image is sensitive
to translation, rotation and scale changes; the same
images with different scale or rotational may have
different wavelet coefficients. The main reason is that
the efficient implementation of 2D -DWT requires
applying a filter bank along the rows and columns of an
image [23].

}

Feature

extraction using

Wavelet

Transform

Wavelet energy
coefficient

Neuron of 1 person

Neuron of 2 person

Neuron of 9 person

Figure 4: Proposed off-line Handwritten signature Recognition System

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

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

Vol. 8, No. 6, 2010

Due to the separability of the filters, the separable 2D-
DWT is strongly oriented in the horizontal and vertical
directions. This makes it hardly possible to extract
translation, rotation and scale invariant features from
the wavelet coefficients. Wavelet energy can keep the
main characteristic of these wavelet coefficients and
make the same images with different translation,
rotation and scale having the same wavelet energy
values[23]. Wavelet energy values can be computed
after analysis signature image to it's wavelet sub-image
coefficient at three level analysis (LLx, HLx, LHx,
HHx). The percentages of energy of these high
frequency sub-images at the &-level wavelet
decomposition is defined in equation (2,3,4)[24]:

The back-propagation algorithm seems to be superior in
this handwritten signature verification environment
[25]. In a back-propagation neural network[26], the
learning algorithm has two phases. First, a training input
pattern is presented to WNN input layer. The WNN
propagates the input pattern from layer to layer until the
output pattern is generated by the output layer. If this
pattern is different from the desired output, an error is
calculated and then propagated backward through the
WNN from the output layer to the input layer. The
weights and both the position and dilation of the
wavelets layer are modified as the error is propagated.
The modified back-propagation training algorithm in
WNN [27] as shown in Figure 6.

ehlw :

100 * Y*(HL decomposition vector at level K) 2
^(decomposition vector) 2

(2)

100 * 2(L// decomposition vector at level K) 2
ELH<& = — — t^ — (3)

EHH

^(decomposition vector) 2
100 * £(//// decompositionvector at level K) 2

^(decomposition vector) 2

(4)

Third stage, Classification stage, after we get the
suitable wavelet energy values that represent signature
image, we take this values as input to WNN and train
this network with a modified Back-propagation (BP)
training algorithm to get efficient off-line signature
recognition. Using WNN for two reasons; first,
traditional ANN has many trade-off because of complex
computations, huge iterations and learning algorithms
are responsible for slowing down the recognition rate
using ANN; second ,recover losing important
information from signature image in DWT technique
because of down-sample process as depicted in
Figure 1 .

In this work, The input layer represents wavelet energy
values feature vector to neural network. The output
layer represents the ability to recognize the human
signature. The middle layer determined the ability to
learn the person signature recognition. Because of the
ability of Morlet function to deal with big input domain
[28] and represents its wave form in equation, Morlet
function will be the suitable wavelet activation
functions y/ in WNN to recognize offline handwritten
signature application. Morlet function equation and it's
derivation in equation (5,6) [27]:

ip(x) = cos(1.75x) exp I

(5)

Then.

dib(x) ( x 2 ,

-j^- = ~[xcos(1.7Sx) + 1.75 sin(1.75x)] exp ( - — ) (6)

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

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

Vol. 8, No. 6, 2010

Input : wavelet energy values extracted from signature image
Output : Class of recognized Signature

> Stepl: Initialize weights and offsets.

Set all weights and node offsets to small random values.
Initialize position and dilation parameter for each wavelet neuron in wavelet layer.
To choose centre point (p)between interval [z!,z 2 ] (input domain),then

t>!=p , ai=0.5(z r z 2 )
Interval [z { ,z 2 ] is divided into two parts by point p
In each sub-interval, we recursively repeat the same procedure which will

initialize b 2 , a 2 and b 3 , a 3 and so on, until all the wavelet are initialize.

> Step2: Present input and desired outputs

Present a continuous valued input vector Xj , X 2 X L

and specify the desired output Dj ,D 2 , . . . .D s .

If the net is used as a classifier then all desired outputs are typically set to zero except for that corresponding
to the class the input is from. That desired output isl. The input could be new on each trial or samples from a
training set could be presented cyclically until stabilize.

> Step 3: Calculate Actual Output

't*=YjJv*

7=1

T^VjjXr-bj

> Step 4: Calculate Error function

n=lk=l

> Step 5: Propagate error to weights and position and dilation parameter

\si =1 Vtjxr-b,

dE lv

J n=1

Where

T = YUv M xf z b l

N S

dE lvv drb(T)

J n=lk=l

Yi^VyXT-b,

dE
db,

(7)

e = 1nYL^- d ^ 2 (8)

(9)

l J n =l k=\ J

(11)

r«H w - D?) ^V(-^-J (i2)

J n=lk=l x J/

> Step 6: Update weights and position and dilation parameter

where :a is learning rate

s is momentum factor

> Step 7: Repeat by going to step 2

Figure 6. B ack-propagation training algorithm in W avelet Neural Network

1 8 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

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

Vol. 8, No. 6, 2010

Feature

extraction using

Wavelet

Transform

Wavelet energy
coefficient

Second Person Signature

Neuron of 1 person

1 Neuron of 2 person

Neuronof9 person

Figure 7: Testing stage of off-line Handwritten signature Recognition System

Finally, Test stage, after learning WNN, we can
examine the ability of WNN to verify the signature of
any person as shown in Figure 7. In this stage our goal
is to input signature image and recognize the person
signature. After scanning and removing noise from
person signature image, wavelet coefficients produce
after analysis image with DWT technique and then
compute wavelet energy value from wavelet detail
coefficients, finally, this wavelet energy values are
taken as input to test WNN classifier to find result in
output layer with only 1 value in only one neuron.
Number of neurons in the output layer represents the
number of person that system recognize.

IV. RESLUT :

This section summarizes the results of using DWT
technique (wavelet energy values) as feature extraction
technique and WNN as classifier to off-line handwritten
signature recognition system. This paper uses nine
person handwritten signatures as show in Figure 8,
each person has twenty image of his handwritten
signature ,ten for train stage and ten for test stage .

GspbrYWn

4) Liu

HUM**

Figure 8: Sample Signature images

In feature extraction stage, wavelet detail coefficients
are extracted from signature image using (db4 or db 12
or db20) wavelet function. To determine the suitable
wavelet function to our database, WNN is used as a
classifier to evaluate the suitable one. Wavelet detail
coefficients (at one level analysis) of signature image
according to one wavelet function is taken as trained
data to WNN.

Three WNN will be found to compare the recognition
rate between tree wavelet function. Modified BP
training algorithm as in Figure 6 is used to train WNN.
Finally, testing WNN with trained signature . Figure 9
shows the recognition rate to (db4,dbl2,db20) wavelet
detail coefficients using WNN as mention above. As a
result from Figure 9, Db20 is recognizing to be the
suitable wavelet function which have high recognition
rate in our database to offline handwritten application.

Figure 9: Offline handwritten signature recognition rate using
(db4,dt>12,db20) wavelet detail coefficients

After determine the suitable extracting wavelet
function, wavelet energy from each signature image is
computed using equation 2,3,4 with db20 as wavelet
function at three level analysis. Nine wavelet energy
coefficients are represented each signature image.

Table 1: WNN architecture and training parameters

The number of layers

The number of neuron on the
layers

Input:9
Hidden: 18
Output: 9

The initial weights and biases

Random

Wavelet Activation functions

Morlet function

Learning rule

Back-Propagation

MSE

0.0001

Learning rate

0.1

Momentum factor

0.009

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

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

Vol. 8, No. 6, 2010

In classification stage, WNN is used with parameters
shown in Tablel. These parameters are selected for
WNN structure after several different experiments. In
these experiments, the WNN is employed with different
parameters such as the number of hidden layers, the size
of the hidden layers, value of the moment constant and
learning rate, and type of the activation functions.
Wavelet energy values for each signature image are the
input features to WNN input layer. Each neuron in
WNN output layer represent a person.

In the test stage , Appling the test wavelet energy
values of the test signature to trained WNN. Evaluating
the proposed off-line handwritten signature recognition
system by recognition rate to each person as shown in
Figure 10.

Figure 10: Proposed off-line handwritten signature recognition
system result

All system evaluation is made by two concept False
Acceptance Rate(FAR) which indicates how many
forgeries were incorrectly classified as genuine
signatures ,and False Rejected Rate(FRR) which
indicates how many genuine signatures were incorrectly
rejected by the system. To the training signatures FAR
and FRR is 0.01% and to the testing signatures FAR
and FRR is 0.07% .

To evaluate our proposed system a comparative study

between three off-line handwritten signature systems is

made:

1 -signature image pixels intensity value as input to

ANN(ANN)

2- signature wavelet energy values as input to ANN
(WE+ANN)

3- Our proposed system signature wavelet energy
values as input to WNN (WE+WNN).

Figure 11 represent the recognition rate to each
training person data and Figure 12 represent the
recognition rate to each testing person data. Figure 11
and Figure 12 concluded that our proposed system has
the highest recognition rate.

lUZ/o

inn% -

. QR% -

"■--.llxf

>&</

r— ^

\y7

j 96%-

I 94% -

1 Q7°/n -

--■ANN

-■-WE+ANN

90% -

.i. U/F+U/MM

Wt+WNIN

SR%

PI P2 P3 P4 P5 P6 P7 P8 P9

Person

Figure 11: Comparative Study between signature image pixels

intensity value as input to ANN (ANN)and signature wavelet energy

values as input to ANN (WE+ANN)and signature wavelet energy

values as input to WNN (WE+WNN)with training data.

102%
100%
« 98%
j 9696
I 94%
S 92%

Sv S*'?\\. <

\ ,-■""''' /

(

-■-WE+ANN

PI P2 P3 P4 P5 P6 P7 P8 P9

Figure 12: Comparative Study between signature image pixels

intensity value as input to ANN (ANN)and signature wavelet energy

values as input to ANN (WE + ANN)and signature wavelet energy

values as input to WNN (WE + WNN) testing data.

V. CONCLUSIONS AND FUTURE WORK

Handwritten signature recognition plays an important
role in our daily life especially in any bank and any
ATM system. Off -Line Handwritten Signature
recognition is a difficult task than On-line one because
of absence of dynamic information in off -Line signature
image such as angle of written style of written and so
on. This paper proposed an off-Line handwritten
recognition system with Four stages. First stage is
scanning signature image and removing noise using
median filter. Second stage, extract feature from each
signature image using DWT technique with the
advantage of multi-scale and with respect the
translation, rotation and scale variations of signature
image. Computing wavelet energy values from DWT
details sub-bands coefficient to all person signature
images using the suitable wavelet function to our
database. Daubechies 20 (db20) is recognize as a
suitable wavelet function with three levels analysis after
a comparative study with other wavelet function. Third
stage, taking the wavelet energy values as input to
WNN with Morlet function as activation function in
hidden layer. Finally, testing trained WNN with
seen/unseen signature to evaluate our proposed system

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

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

Vol. 8, No. 6, 2010

recognition rate. A comparative study between three
off-line handwritten signature systems is made
( signature image pixels intensity as input to ANN ,
signature wavelet energy values as input to ANN and
signature wavelet energy values as input to WNN). The
conclusion will found that our proposed system
(wavelet energy values as input to WNN) has high
recognition rate.

[12] K R Radhika, M K Venkatesha and G N Sekhar," Pattern
Recognition Techniques in Off-line hand written signature
verification - A Survey", PROCEEDINGS OF WORLD
ACADEMY OF SCIENCE, ENGINEERING AND
TECHNOLOGY ,vol. 36, ISSN 2070-3740 , 2008.

[13] Engin Avci , Abdulkadir Sengur, Davut Hanbay, " An optimum
feature extraction method for texture classification", Expert
Systems with Applications: An International Journal, Published
by Elsevier Ltd, Volume 36 , Issue 3,2009,p.p 6036-6043.

To improve our system recognition rate, each person
signature should have its own wavelet function in
feature extraction stage. Genetic algorithm will be used
as a searching strategy in the future work to found the
suitable wavelet function to each person signature.

ACKNOWLEDGEMENTS :

The authors would like to thank .Prof. Albert Swart for
making his signature database available to us.

The first author would like to thank Sarah El.metwally
and Eslam Foud for their encouragements.

REFERENCES:

[1] S. Impedovo, G. Pirlo," Verification of Handwritten Signatures: an
Overview", 14th International Conference on Image Analysis and
Processing , 2007, pp. 191-196 .

[2] M. Ammar, Y. Yoshida, T. Fulumura, " A New Effective
Approach for Off-line Verification of Signatures by Using
Pressure Features",. Proceedings of the 8th International
Conference on Pattern Recognition, 1986, p.p 566-569.

[3] Y. Qi , B.R. Hunt, "Signature Verification Using Global and Grid
Features," Pattern Recognition, vol.27, Issue. 12, 1994, p.p 1621-
1629.

[4] V.E. Ramesh, M.N. Murty, "Off-line Signature Verification Using
Genetically Optimized Weighted Features", Pattern Recognition,
vol.32, Issue.2, 1999, p.p 217-233.

[5] P.S. Deng, H.-Y. M. Liao, C.W. Ho, and H.-R. Tyan,: "Wavelet-
Based Off-line Handwritten Signature Verification", Computer
Vision and Image Understanding, vol.76, no. 3, 1999, p.p 173 -
190.

r61 http://dspace.mak.ac.ug/bitstream/123456789/599/3/karanja-
evanson-mwangi-cit-masters-report.pdf , 17-8-2010

[7] M. Kalera, S. Srihari, and A. Xu. "Offline signature verification
and identification using distance statistics", International Journal
of Pattern Recognition and Artificial Intelligence, vol. 18, no.7,
2004, pp. 1339-1360.

[8] V. Nalwa." Automatic on-line signature verification", Lecture
Notes In Computer Science, Proceedings of the Third Asian
Conference on Computer Vision,1998, p.p 10-15 .

r91 http://research.microsoft.com/pubs/69437/handwritingregistration
cvpr07.pdf , 17-8-2010. ~~ "

[10]Sing-Tze Bow, "Pattern recognition and
preprocessing" ,Marcel Dekker,Inc, chapter 15,2002.

image

[14] http://www.dtic.upf.edu/~xserra/cursos/TDP/referencies/Park-
DWT.pdf , 17-8-2010.

[15]S.Sitharama Lyengar,E.C.Cho,Vir V.Phoh /'Foundations of
Wavelet Networks and Application",

chapman&Hall/CRC Press LLC , chapter 4, 2002.

[16] Xian-Bin Wen, Hua Zhang, and Fa-Yu Wang," A Wavelet
Neural Network for SAR Image Segmentation", Sensors ,
Vol.9,No.9,2009,p.p 7509-7515 .

[17] Zhang Q. and Benveniste A,"Wavelet networks"
On Neural Networks ,Vol.3, ,1992,p.p 889-898.

IEEE Trans.

[11] G.Y. Chen, T.D. Bui, A. Krzyzak," Contour-based handwritten
numeral recognition using multi-wavelets and neural networks",
Pattern Recognition , Vol.36 ,2003,p.p 1597 - 1604.

[18] ri Gross and Longin Jan Latecki,"Digital geometric methods in
document image analysis",Pattern Recognition, Vol. 32,No.3,
1999,pp. 407.

[19] http://homepages.inf.ed.ac.uk/rbf/HIPR2/median.htm , 24-6-2010.

[20] Avci, E., Turkoglu, I., & Poyraz, M., "Intelligent target
recognition based on wavelet packet neural network", Experts
Systems with Applications: An International Journal, vol 29, Issue
1 ,2005 ,p.p 175-182.

[21] Samaneh Ghandali, Mohsen Ebrahimi Moghaddam, "Off-Line
Persian Signature Identification and Verification Based on Image
Registration and Fusion", JOURNAL OF MULTIMEDIA, VOL.

4, NO. 3, 2009.

[22] A. Wahi, E. Thirumurugan "Recognition of Objects by
Supervised Neural Network using Wavelet Features", First
International Conference on Emerging Trends in Engineering and
Technology,2008.

[23] Chi-Man Pun and Moon-Chuen Lee ,"Log-Polar Wavelet Energy
Signatures for Rotation and Scale Invariant Texture
Classification", IEEE TRANSACTIONS ON PATTERN
ANALYSIS AND MACHINE INTELLIGENCE, VOL. 25, NO.

5, 2003

[241 http://www.mathworks.com/access/helpdesk/help/toolbox/wavele
tAyenergy2.html , 17-8-2010.

[25] Alan McCabe, Jarrod Trevathan and Wayne Read," Neural
Network-based Handwritten Signature Verification", JOURNAL
OF COMPUTERS, VOL. 3, NO. 8, 2008.

[26] Insung Jung, and Gi-Nam Wang," Pattern Classification of Back-
Propagation Algorithm Using Exclusive Connecting Network",
World Academy of Science, Engineering and Technology, VOL.
36,2007.

[27] Ming Meng , Wei Sun," Short-term Load Forecasting Based on
Rough Set and Wavelet Neural Network ", International
Conference on Computational Intelligence and Security,2008.

[28] Mohd Fazril, Zaki Ahmad, Hj. Kamaruzaman, "The Performance
of Two Mothers Wavelets in Function Approximation", Joural of
Mathematical Research, Vol. 1 ,No.2,2009.

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

A Black-Box Test Case Generation Method

Nicha Kosindrdecha

Autonomous System Research Laboratory

Faculty of Science and Technology, Assumption University

Bangkok, Thailand

Jirapun Daengdej

Autonomous System Research Laboratory

Faculty of Science and Technology, Assumption University

Bangkok, Thailand

Abstract — Test case generation techniques have been researched
over a long period of time. Unfortunately, while many
researchers have found methods of minimizing test cases, there
are still a number of important related issues that need to be
researched. The primarily outstanding research issue is a large
single test suite containing a huge number of test cases. Our study
shows that this can lead to other two problems: unable to identify
suitable test cases for execution and those test cases are lack of
ability to cover domain specific requirement. Therefore, we
proposed an additional requirement prioritization process during
a test case generation process and an automated method to
generate multiple test suites while minimizing a number of test
cases from UML Use Case diagram 2.0. Our evaluation result
shows that the proposed method is the most recommendation
method to minimize size of test cases while maximizing ability to
cover critical domain specific requirements.

Keywords-component; Test generation, testing and quality, test
case generation, test generation technique and generate tests

I. Introduction

Software testing is known as a key critical phase in the
software development life cycle, which account for a large part
of the development effort. A way of reducing testing effort,
while ensuring its effectiveness, is to generate a minimize
number of test cases automatically from artifacts used in the
early phases of software development. Many test case
generation techniques have been proposed [2], [4], [10], [11],
[12], [15], [21], [22], [42], [47], [50], mainly random, path-
oriented, goal-oriented and model-based approaches. Random
techniques determine a set of test cases based on assumptions
concerning fault distribution. Path-oriented techniques
generally use control flow graph to identify paths to be covered
and generate the appropriate test cases for those paths. Goal-
oriented techniques identify test cases covering a selected goal
such as a statement or branch, irrespective of the path taken.
There are many researchers and practitioners who have been
working in generating a set of test cases based on the
specifications. Modeling languages are used to get the
specification and generate test cases. Since Unified Modeling
Language (UML) 2.0 is the most widely used language, many
researchers are using UML diagrams such as UML Use Case
diagram, UML Activity diagram and UML Statechart diagram
to generate test cases and this has led to model-based test case
generation techniques. The study shows that model-based test
generation methods (or also known as black-box test

generation) are widely-used for generating test cases in the
commercial industry.

Moreover, the study [2], [4], [10], [11], [12], [15], [21],
[22] shows that the primary research issue is that existing
black-box test case generation methods generate a huge single
test suite with a number of possible tests. The number of
possible black-box tests for any non-trivial software application
is extremely large. Consequently, it is unable to identify
suitable test cases for execution.

Also, the study shows that the secondary research issue is
that the existing black-box test case generation methods ignore
critical domain specific requirements [5] during a test case
generation process. These requirements are one of the most
important requirements that should be addressed during test
activities.

Therefore, we propose a new black-box test case
generation, with requirement prioritization approach, from
requirements captured as use cases, 2.0, [23], [24], [33]. A use
case is the specification of interconnected sequences of actions
that a system can perform, interacting with actors of the
system. Use cases have become one of the favorite approaches
for requirements capture. Our automated black-box approach
aims to generate a minimize number of suitable test cases while
reserving critical domain specific requirements. Additionally,
we introduce an automated test generation method derived
from UML Use Case diagram, 2.0. Our approach is developed
to automatically generate many test suites based on notions
announced in the latest version of UML.

The rest of the paper is organized as follow. Section 2
discusses an overview of test case generation techniques.
Section 3 describes motivated research issues. Section 4
introduces a new test generation process with requirement
prioritization step. Also, section 4 proposes a new black-box
test generation method. Section 5 describes an experiment,
measurement metrics and results. Section 6 provides the
conclusion and research directions in the test case generation
field. The last section represents all source references used in
this paper.

II. LITERATURE REVIEW

The literature review is structured into two sections. The
first section gives an overview of previous studies. The second
section provides the related works

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

A. An Overview of Recent Researches

Model-based techniques are popular and most researchers
have proposed several techniques. One of the reasons why
those model-based techniques are popular is that wrong
interpretations of complex software from non-formal
specification can result in incorrect implementations leading to
testing them for conformance to its specification standard [43].
A major advantage of model-based V&V is that it can be easily
automated, saving time and resources. Other advantages are
shifting the testing activities to an earlier part of the software
development process and generating test cases that are
independent of any particular implementation of the design [7].

The model-based techniques are method to generate test
cases from model diagrams like UML Use Case diagram [23],
[24], [33], UML Sequence diagram [7] and UML State diagram
[5], [43], [22], [2], [21], [15], [32], [4]. There are many
researchers who investigated in generating test cases from
those diagrams. The following paragraphs show examples of
model-based test generation techniques that have been
proposed for a long time.

Heumann [23] presented how using use cases, derived from
UML Use Case diagram 1.0, to generate test cases can help
launch the testing process early in the development lifecycle
and also help with testing methodology. In a software
development project, use cases define system software
requirements. Use case development begins early on, so real
use cases for key product functionality are available in early
iterations. According to the Rational Unified Process (RUP), a
use case is used to describe fully a sequence of actions
performed by a system to provide an observable result of value
to a person or another system using the product under
development." Use cases tell the customer what to expect, the
developer what to code, the technical writer what to document,
and the tester what to test. He proposed three-step process to
generate test cases from a fully detailed use case: (a) for each
use case, generate a full set of use-case scenarios (b) for each
scenario, identify at least one test case and the conditions that
will make it execute and (c) for each test case, identify the data
values with which to test.

Ryser [24] raised the practical problems in software testing
as follows: (1) Lack in planning/time and cost pressure, (2)
Lacking test documentation, (3) Lacking tool support, (4)
Formal language/specific testing languages required, (5)
Lacking measures, measurements and data to quantify testing
and evaluate test quality and (6) Insufficient test quality. They
proposed their approach to resolve the above problems. Their
approach is to derive test case from scenario / UML Use Case
diagram 1.0 and state diagram 1.0. In his work, the generation
of test cases is done in three processes: (a) preliminary test case
definition and test preparation during scenario creation (b) test
case generation from Statechart and from dependency charts
and (c) test set refinement by application dependent strategies.

Whereas all requirements are mandatory, some are more
critical than others. For example, failure to implement certain
requirements may have grave business ramifications that would
make the system a failure, while others although contractually
binding would have far less serious business consequences if
they were not implemented or not implemented correctly (b)
Help programs through negotiation and consensus building to
eliminate unnecessary potential "requirements" (i.e., goals,
desires, and "nice-to-haves" that do not merit the mandatory
nature of true requirements) and (c) schedule the
implementation of requirements (i.e., help determine what
capabilities are implemented in what increment). Additionally,
these researches in 1980-2008 [8], [27], [28], [29], [30], [38]
reveal that there are many requirement prioritization methods
such as Binary Search Tree (BST), 100-point method and
Analytic Hierarchy Process (AHP)

III. RESEARCH PROBLEM

This section discusses the details of research issues related
to test case generation techniques and research problems,
which are motivated this study. Every test case generation
technique has weak and strong points, as addressed in the
literature survey. In general, referring to the literature review,
the following lists major outstanding research challenges.

The first research problem is that existing test case
generation methods are lack of ability to identify domain
specific requirements. The study [5] shows that domain
specific requirements are some of the most critical
requirements required to be captured for implementation and
testing, such as constraints requirements and database specific
requirements. Existing approaches ignore an ability to address
domain specific requirements. Consequently, software testing
engineers may ignore the critical functionality related to the
critical domain specific requirements. Thus, this paper
introduces an approach to priority those specific requirements
and generates an effective test case.

The second problem is that existing black-box test case
generation techniques aim to generate a large single test suite
with all possible test cases which maximize cover for each
scenario. Basically, they generate a huge number of test cases
which are impossible to execute given limited time and
resources. As a result, those unexecuted test cases are useless
and it is unable to identify suitable test cases for execution.

IV. PROPOSED METHOD

A. Test Case Generation Process

This section presents a new high-level process to generate a
set of test cases introduced by using the above comprehensive
literature review and previous works [43].

B. Related Works

This section provides the related works used in this paper,
prioritize requirement methods. Donald Firesmith [16]
addressed the purpose of requirement prioritization as follows:
(a) Determine the relative necessity of the requirements.

Autonomous System Research Laboratory, Faculty of Science and
Technology, Assumption University.

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Existing Test Case Generation Process Proposed Test Case Generation Process

Requirement

1.1
Requirement

Prioritization

< j

Design

Test Case V
Generation f

Design

2.1

Test Case

Generation

▼

3 - I

Development

Developme^B

-J

Testing

5.
Maintenance

Figure 1 . A Proposed Process to Generate Test Cases

From the above figure, there are two test case generation
processes: existing and proposed process. The left-hand side
shows an existing process to generate test cases directly from
diagrams. Meanwhile, the right-hand side proposes to add an
additional requirement prioritization process before generating
test cases. The requirement prioritization process aims to be
able to effectively handle with a large number of requirements.
The objective of this process is to prioritize and organize
requirements in an appropriate way in order to effectively
design and prepare test cases [16], [25], [37]. There are two
sub-processes: (a) classify requirements and (b) prioritize
requirements.

Our study [51], [52], [53], [54] shows that a marketing
perspective concentrates on two factors: customer's needs and
customer satisfaction. We apply that perspective to the
requirement prioritization and propose the following:

Customer Satisfaction

Performance

Delight
(Nice to Have)

From the above figure, the horizontal axis presents a
customer's need while the vertical axis represents a customer
satisfaction. There are four groups of requirements based on
those two factors: delight, attractive, indifferent and basic.
First, the delight requirement is known as 'nice-to-have'
requirement. If this requirement is well fulfilled, it will increase
the customer satisfaction. Otherwise, it will not decrease the
satisfaction. Second, the attractive requirement is called as
'surprise' or 'know your customer' requirement. This
requirement can directly increase the customer satisfaction if it
is fulfilled. Marketers and sales [53] believe that if we can
deliver this kind of requirement, it will impress customers and
significantly improve the customer satisfaction. Third, the
indifferent requirement is a requirement that customer does not
concentrate and it will not impress customers at all. In the
competitive industry, this requirement may be fulfilled, but
there are no any impacts to the customer satisfaction. Last, the
basic requirement is a mandatory requirement that customers
basically expect. Therefore, if this requirement is well
delivered, it will not increase the customer satisfaction.

Furthermore, our study reveals that the requirement can be
simply divided into two types: functional and non-functional
requirement. Our study also presents that functional
requirements can be categorized into two groups: domain
specific requirement [5] (or known as constraints requirement)
and behavior requirement. The following shows the
requirement classification used in this paper:

Constraints/

Business

Domain Specific/

/ 1

Functional
Requirement

Rules >^

System \
Behavior

Constraints Req.

Behavioral Req.

Non-Functional

Requirement

{ )

Dissatisfk \
Figure 2. Classify Requirement on Marketing's Perspective

Figure 3. Classify Requirement on Software Engineer

From the above figure, functional requirement is a
requirement that customers directly are able to provide. The
non-functional requirement is a requirement that is given
indirectly. The domain specific or constraints requirement is a
requirement relative to any constraints and business rules in the
software development. Meanwhile, the behavior requirement is
a requirement that describes a behavior of system. Once the
requirement is classified based on previous two perspectives,
the next process is to prioritize requirements based on return on
investment (or ROI) [51], [52], [53]. From business
perspective, ROI is the most important factor to assess the
important of each requirement. The following presents a
ranking tree by combining those two perspectives.

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

ornain
Specific

Rank#l

Behavioral
Reg-

Rank #2

Rank #3

Domain

Specific

R e q, —

Rank hase<
on ROI

Behavioral

Rank based

on ROI

Rank based
on ROI

Domain

Specific

R e q.

Rank based
on ROI

Behavioral _[ Rank based
Req. on ROI

Rank based
on ROI

UiJHU i n ,

Specific Rank #10

R e q . J j

Behavioral
Reg.

Rank #11

Rank #12

Figure 4. Requirement Prioritization Tree

From the above figure, we give the highest priority for all
'basic requirements due to the fact that they must be
implemented even they do not increase the customer
satisfaction. We rank the lowest priority for all 'indifferent'
requirements, because customers do not concentrate on.
Additionally, we prioritize both of all 'delight' and 'attractive'
requirement based on ROI. In this paper, we propose to use a
cost-value approach to weight and prioritize requirements. This
paper proposes to use the following formula:

P(Req) = (Cost * CP)

(1)

Where:

• P is a prioritization value.

• Req is a requirement required to be prioritized.

• Cost is a total estimated cost of coding and testing for
each requirement.

• CP is an user-defined customer priority value. This value
is in the range between 1 and 10. 10 is the highest priority
and 1 is the lowest priority. This value aims to allow
customers to identify how important of each requirement
is from their perspective.

To compute the above cost for coding and testing, this
paper proposes to apply the following formula:

Cost = (EffCode*CostCode)+(EffTest*CostTest) (2)

Where

• Cost is a total estimated cost.

• EffCode is an estimated effort of coding for each
requirement. The unit is man-hours.

• CostCode is a cost of coding that is charged to customers.
This paper applies the cost- value approach to identify the
cost of coding for each requirement group (e.g. "Must-
Have", "Should-Have", "Could-Have" and "Wish"). The
unit is US dollar.

• EjfTest is an estimated effort of testing for each
requirement. The unit is man-hours.

• CostTest is a cost of testing that is charged to customers.
The approach to identify this value is similar to
CostCode' 's approach. The unit is US dollar.

In this paper, we assumed the following in order to
calculate CostCode and CostTest. Also, this paper assumes that
a standard cost for both activities is $100 per man-hours.

• A value is 1.5 of ("Must-Have", "Should-Have") - this
means that "Must-Have" requirements have one and half
times cost value than "Should-Have" requirements.

• A value is 3 of ("Must-Have", "Could-Have") - this
means that "Must-Have" requirements have three times
cost value than "Could-Have" requirements.

• A value is 2 of ("Should-Have", "Could-Have") - this
means that "Should-Have" requirements have two times
cost value than "Could-Have" requirements.

• A value is approximately 3 of ("Could-Have", "Wish") -
this means that "Could-Have" requirements have three
times cost value than "Wish" requirements.

Therefore, the procedure of requirement prioritization
process can be shortly described below:

1 . Provide estimated efforts of coding and testing for each
requirement.

2. Assign cost value for each requirement group based on
the previous requirement classification (e.g. "Must-
Have", "Should-Have", "Could-Have" and "Wish").

3. Calculate a total estimated cost for coding and testing,
by using the formula (2).

4. Define a customer priority for each requirement.

5. Compute a priority value for each requirement by
using the formula (1).

6. Prioritize requirements based on the higher priority
value.

Once the requirements are prioritized, the next proposed
step is to generate test scenario and prepare test case.

B. Test Case Generation Technique

This section presents an automated test scenario generation
derived from UML Use Case diagram 2.0. The big different
between UML Use Case diagram 1.0 and 2.0 is a package
notion that can group each use case into each package.

The following shows an example of UML Use Case
diagram 2.0.

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Customer

^Jnquiry

-d(^Withdraw

Release 1

Release 2

S*\t\\bx Bank

./"^Own Acct.
^^X^TYansfor

NjX^TG Airlirm
^v^Support

Release 3

Figure 5. An Example of UML Use Case Diagram 2.0

From the above figure, the new notion in UML Use Case
diagram 2.0 is a package that is used for grouping each
function. There are three packages or releases. Each release
contains different functional requirement. The first release
contains two functions: inquiry and withdraw. The second
release is composed of: transfer own account and transfer to
other banks. The last release has only one function to support
Thai (TG) airline tickets.

Our approach aims to generate three test suites to cover the
above three packages while existing test case generation
techniques do not concentrate on. The first test suite is
developed for: inquiry and withdraw functions. The second test
suite is used for transferring own banks and other banks. The
last suite aims to a TG airline ticket support.

The approach is built based on Heumann's algorithm [23].
The limitation of our approach is to ensure that all use cases are
fully dressed. The fully dressed use case is a use case with the
comprehensive of information, as follows: use case name, use
case number, purpose, summary, pre-condition, post-condition,
actors, stakeholders, basic events, alternative events, business
rules, notes, version, author and date.

The proposed method contains four steps, as follows: (a)
extract use case diagram (b) generate test scenario (c) prepare
test data and (d) prepare other test elements. These steps can be
shortly described as follows:

1. The first step is to extract the following information
from fully dressed use cases: (a) use case number (b)
purpose (c) summary (d) pre-condition (e) post-
condition (f) basic event and (g) alternative events.
This information is called use case scenario in this
paper. The example fully dressed use cases of ATM
withdraw functionality can be found as follows:

Case

Event

ve Events

Name

Rules

UC-

Withdra

To allow

1 . Insert

1. Select

(a)

001

bank's

Card

Inquiry

Input

customer

2. Input

2. Select

amount

s to

A/C Type

withdraw

3. Select

3. Check

Outstan

money

Withdraw

Balance

ding

from

4. Select

Balanc

ATM

A/C Type

machines

5. Input

(b) Fee

anywher

Balance

charge

e in

6. Get

if using

Thailand

Money
7. Get Card

differen
tATM
machin

uc-

Transfer

To allow

1 . Insert

1. Select

Amoun

002

users to

Card

Inquiry

t<=

transfer

2. Input

2. Select

50,000

money to

A/C Type

baht

other

3. Select

3. Check

banks in

Transfer

Balance

Thailand

4. Select

from all

bank

ATM

5. Select

machines

"To"
account

6. Select
A/C Type

7. Input
Amount

8. Get

Receipt

9. Get Card

The above use cases can be extracted into the following use
case scenarios:

TABLE II.

Extracted Use Case Scenarios

Scenario Id

Summary

Basic Scenario

Scenario-001

To allow bank's

1 . Insert Card

customers to

2. Input PIN

withdraw money from

3. Select Withdraw

ATM machines

4. Select A/C Type

anywhere in Thailand.

5. Input Balance

6. Get Money

7. Get Card

TABLE I.

Example Fully Dressed Use Case

Use

Summar

Basic

Alternati

Busine

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Scenario-002

To allow bank's

customers to

withdraw money from 3

ATM machines
anywhere in Thailand

Scenario-003

Scenario-004

1 . Insert Card

2. Input PIN
Select Inquiry

4. Select A/C Type

5. Check Balance

6. Select Withdraw

7. Select A/C Type

8. Input Balance

9. Get Money

10. Get Card

To allow users to
transfer money to

other banks in

Thailand from all

ATM machines

To allow users to
transfer money to

other banks in

Thailand from all

ATM machines

1 . Insert Card

2. Input PIN

3. Select Transfer

4. Select bank

5. Select "To"
account

6. Select A/C Type

7. Input Amount

8. Get Receipt

9. Get Card

1 . Insert Card

2. Input PIN

3. Select Inquiry

4. Select A/C Type

5. Check Balance

6. Select Transfer

7. Select bank
Select "To"

account

9. Select A/C Type

10. Input Amount

11. Get Receipt

12. Get Card

TS-002

To allow bank's

1 . Insert Card

customers to

2. Input PIN

withdraw money from 3. Select Inquiry

ATM machines

4. Select A/C Type

anywhere in Thailand.

5. Check Balance

6. Select Withdraw

7. Select A/C Type

8. Input Balance

9. Get Money

10. Get Card

TS-003

To allow users to

1 . Insert Card

transfer money to

2. Input PIN

other banks in

3. Select Transfer

Thailand from all

4. Select bank

ATM machines

5. Select "To"
account

6. Select A/C Type

7. Input Amount

8. Get Receipt

9. Get Card

TS-004

To allow users to

1 . Insert Card

transfer money to

2. Input PIN

other banks in

3. Select Inquiry

Thailand from all

4. Select A/C Type

ATM machines

5. Check Balance

6. Select Transfer

7. Select bank

8. Select "To"
account

9. Select A/C Type

10. Input Amount

11. Get Receipt

12. Get Card

2. The second step is to automatically generate test
scenarios from the previous use case scenarios [23].
From the above table, we automatically generate the
following test scenarios:

3 . The next step is to prepare test data. This step allows to
manually prepare an input data for each scenario.

4. The last step is to prepare other test elements, such as
expected output, actual output and pass / fail status

TABLE III.

Generated Test Scenarios

Test Scenario

Summary

Basic Scenario

TS-001

To allow bank's

1 . Insert Card

customers to

2. Input PIN

withdraw money from

3. Select Withdraw

ATM machines

4. Select A/C Type

anywhere in Thailand.

5. Input Balance

6. Get Money

7. Get Card

V. EVALUATION

The section describes the experiments design, measurement
metrics and results.

A. Experiments Design

1. Prepare Experiment Data. Before evaluating the
proposed methods and other methods, preparing
experiment data is required. In this step, 50
requirements and 50 use case scenarios are randomly
generated.

2. Generate Test Scenario and Test Case. A comparative
evaluation method has been made among the proposed
test scenario algorithm, Heumann's technique Jim [23],
Ryser's method [24], Nilawar's algorithm [33] and the
proposed method presented in the previous section. It

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

is included a prioritization requirement algorithm prior
to generate a set of test scenarios and test cases.

3. Evaluate Results. In this step, the comparative
generation methods are executed by using 50
requirements and 50 use case scenarios. These methods
are also executed for 10 times in order to find out the
average percentage of critical domain requirement
coverage, a size of test cases and total generation time.
In total, there are 500 requirements and 500 use case
scenarios executed in this experiment.

The following tables present how to randomly generate data
for requirements and use case scenarios respectively.

basic 7, basic 2 , . . . , basic n

TABLE IV.

Generate Random Requirements

Attribute

Approach

Requirement
ID

Randomly generated from the following
combination: Req + Sequence Number.

For example, Reql, Req2, Req3, ...,
ReqN.

Description

Randomly generated from the following

combination: Des + Sequence Number

same as Requirement ID.

For example, Desl, Des2, Des3, ...,
DesN.

Type of
Requirement

Randomly selected from the following
values: Functional AND Non-Functional.

MoSCoW
Criteria

Randomly selected from the following

values: Must Have (M), Should Have (S),

Could Have (C) and Won't Have (W)

Is it a critical
requirement

(Y/N)?

Randomly selected from the following
values: True (Y) and False (N)

TABLE V.

Generate Random Use Case Scenario

Attribute

Approach

Use case ID

Randomly generated from the

following combination: uCase +

Sequence Number. For example,

uCase h uCase 2 , ..., uCase n .

Purpose

Randomly generated from the

following combination: Pur +

Sequence Number same as Use case

ID. For example, Pur h Pur 2 , . . . ,

Pur n .

Pre-condition

Randomly generated from the

following combination: pCon +

Sequence Number same as Use case

ID. For example, pCon h pCon 2 , . . . ,

pCon n .

Basic Scenario

Randomly generated from the
following combination: uCase +
Sequence Number. For example,

B. Measurement Metrics

The section lists the measurement metrics used in the
experiment. This paper proposes to use three metrics, which
are: (a) size of test cases (b) total time and (c) percentage of
critical domain requirement coverage. The following describe
the measurement in details.

1. A Number of Test Cases: This is the total number of
generated test cases, expressed as a percentage, as follows:

% Size = (# Size I # of Total Size)*\W (3)

Where:

• % Size is a percentage of the number of test cases.

• # of Size is a number of test cases.

• # of Total Size is the maximum number of test cases in the
experiment, which is assigned 1,000.

2. A Domain Specific Requirement Coverage: This is an
indicator to identify the number of requirements covered in
the system, particularly critical requirements, and critical
domain requirements [5]. Due to the fact that one of the
goals of software testing is to verify and validate
requirements covered by the system, this metric is a must.
Therefore, a high percentage of critical requirement
coverage is desirable.

It can be calculated using the following formula:

% CRC = (# of Critical I # ofTotat)*100 (4)

Where:

• % CRC is the percentage of critical requirement coverage.

• # of Critical is the number of critical requirements
covered.

• # of Total is the total number of requirements.

3. Total Time: This is the total number of times the
generation methods are run in the experiment. This metric
is related to the time used during the testing development
phase (e.g. design test scenario and produce test case).
Therefore, less time is desirable.

It can be calculated using the following formula:

Total = PTime + CTime + RTime (5)

Where:

• Total is the total amount of times consumed by running
generation methods.

• PTime is the total amount of time consumed by
preparation before generating test cases.

• CTime is the time to compile source code / binary code in
order to execute the program.

• RTime is the total time to run the program under this
experiment.

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

C. Results and Discussion

This section discusses an evaluation result of the above
experiment. This section presents a graph that compares the
above proposed method to other three existing test case
generation techniques, based on the following measurements:
(a) size of test cases (b) critical domain coverage and (c) total
time. Those three techniques are: (a) Heumman's method (b)
Ryser's work and (c) Nilawar's approach. There are two
dimensions in the following graph: (a) horizontal and (b)
vertical axis. The horizontal represents three measurements
whereas the vertical axis represents the percentage value.

Compare Percentage of Size of Test Cases. Critical Domain Coverage and Total Time among Four
Test Case Generation Techniques

« ©0 00%

Critical Domain Coverage
Measurements

Figure 6. An Evaluation Result of Test Generation Methods

The above graph shows that the above proposed method
generates the smallest set of test cases. It is calculated as
80.80% where as the other techniques is computed over 97%.
Those techniques generated a bigger set of test cases, than a set
generated by the proposed method. The literature review
reveals that the smaller set of test cases is desirable. Also, the
graph shows that the proposed method consumes the least total
time during a generation process, comparing to other
techniques. It used only 30.20%, which is slightly less than
others. Finally, the graph presents that the proposed method is
the best techniques to coverage critical domains. Its percentage
is much greater than other techniques' percentage, over 30%.

From the above figure, this study determines and ranks the
above comparative methods into five ranking: 5 -Excellent, 4-
Very good, 3-Good, 2-Normal and 1-Poor. This study uses a
maximum and minimum value to find an interval value for
ranking those methods.

For a number of test cases, the maximum and minimum
percentage is 98% and 80.80%. The different between
maximum and minimum value is 17.2%. An interval value is
equal to a result of dividing the different values by 5. As a
result, the interval value is approximately 3.4. Thus, it can be
determined as follows: 5-Excellent (since 80.80% to 84.2%), 4-
Very good (between 84.2% and 87.6%), 3-Good (between
87.6% and 91%), 2-Normal (between 91% and 94.4%) and 1-
Poor (from 94.4% to 97.8%).

For an ability to cover critical domain specific requirement,
the maximum and minimum percentage is 53.20% and 19%.
The different value is 34.2%. The interval value is 6.84.
Therefore, it can be determined as follows: 5 -Excellent (since
46.36% to 53.2%), 4-Very good (between 39.52% and
46.36%), 3-Good (between 32.68% and 39.52%), 2-Normal
(between 25.84% and 32.68%) and 1-Poor (from 19% to
25.84%).

For a total time, the maximum and minimum percentage is
31.82% and 30.20%. The different between maximum and
minimum value is 1.62%. An interval value is equal to a result
of dividing the different values by 5. As a result, the interval
value is 0.324. Thus, it can be determined as follows: 5-
Excellent (since 30.2% to 30.524%), 4-Very good (between
30.524% and 30.848%), 3-Good (between 30.848% and
31.172%), 2-Normal (between 31.172% and 31.496%) and 1-
Poor (from 31.496% to 31.82%).

Therefore, the experiment result of those comparative
methods can be shown below:

TABLE VI.

A Comparison of Test Case Reduction Methods

Algorithm

Number of
Test Cases

Cover
Critical
Domain
Specific Req.

Total
Time

Heumann's
Method

Ryser's Method

Nilawar's
Method

Our Proposed
Method

In the conclusion, the proposed method is the best to
generate the smallest size of test cases with the maximum of
critical domain coverage and the least time consumed in the
generation process.

VI. CONCLUSION

In this paper, we introduced a new test case generation
method and process, with an additional requirement
prioritization process. The approach inserts an additional
process to ensure that all domain specific requirements are
captured during the test case generation. Also, the approach is
developed to minimize a number of test cases in order to be
able to select suitable test cases for execution. Additionally, we
proposed an automated approach to generate test cases from
fully described UML use cases, version 2.0. Our generated
method can generate many test suites derived from UML Use
Case diagram, 2.0. Existing test case generation methods
generate only a large single test suite that contains a lot of
numbers of test cases.

Furthermore, we conducted an evaluation experiment with
a random requirements and fully described use cases. Our

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

evaluation result reveals that the proposed method is the most
recommendation automated test case generation methods for
maximizing critical domain requirement coverage. Also, the
result present that the proposed method is one of best methods
to minimize a number of test cases.

The future research, we plan to enhance an ability to
prioritize requirements and conduct a large experiment for a
large system development

References

[I] Ahl, V. "An Experimental Comparison of Five Prioritization
Methods" Master's Thesis, School of Engineering, Blekinge
Institute of Technology, Ronneby, Sweden, 2005.

[2] Alessandra Cavarra, Charles Crichton, Jim Davies, Alan
Hartman, Thierry Jeron and Laurent Mounier. "Using UML for
Automatic Test Generation." Oxford University Computing
Laboratory, Tools and Algorithms for the Construction and
Analysis of Systems, TACAS'2000, 2000.

[3] Amaral. "A.S.M.S. Test case generation of systems specified in
Statecharts." M.S. thesis - Laboratory of Computing and
Applied Mathematics, INPE, Brazil, 2006.

[4] Annelises A. Andrews, Jeff Offutt and Roger T. Alexander.
"Testing Web Applications. Software and Systems Modeling.",
2004.

[5] Avik Sinha, Ph.D and Dr. Carol S. Smidts. "Domain Specific
Test Case Generation Using Higher Ordered Typed Languages
fro Specification." Ph. D. Dissertation, 2005.

[6] A. Bertolino. "Software Testing Research and Practice." 10th
International Workshop on Abstract State Machines
(ASM'2003), Taormina, Italy, 2003.

[7] A.Z. Javed, PA. Strooper and G.N. Watson "Automated
Generation of Test Cases Using Model-Driven Architecture."
Second International Workshop on Automation of Software Test
(AST'07), 2007.

[8] Beck, K. & Andres, C. "Extreme Programming Explained:
Embrace Change", 2nd ed. Boston, MA: Addison-Wesley, 2004.

[9] Boehm, B. & Ross, R. "Theory- W Software Project
Management: Principles and Examples." IEEE Transactions on
Software Engineering 15, 4: 902-916, 1989.

[10] B.M. Subraya, S.V. Subrahmanya "Object driven performance
testing in Web applications." in: Proceedings of the First Asia-
Pacific Conference on Quality Software (APAQS'00), pp. 17-26,
Hong Kong, China, 2000.

[II] Chien-Hung Liu, David C. Kung, Pei Hsia and Chih-Tung Hsu
"Object-Based Data Flow Testing of Web Applications."
Proceedings of the First Asia-Pacific Conference on Quality
Software (APAQS'00), pp. 7-16, Hong Kong, China, 2000.

[12] C.H. Liu, D.C. Kung, P. Hsia, C.T. Hsu "Structural testing of
Web applications." in: Proceedings of 11th International
Symposium on Software Reliability Engineering (ISSRE 2000),
pp. 84-96, 2000.

[13] Davis, A. "The Art of Requirements Triage." IEEE Computer
36, 3 p: 42-49, 2003.

[14] Davis, A. "Just Enough Requirements Management: Where
Software Development Meets Marketing." New York: Dorset
House (ISBN 0-932633-64-1), 2005.

[15] David C. Kung, Chien-Hung Liu and Pei Hsia "An Object-
Oriented Web Test Model for Testing Web Applications." In
Proceedings of the First Asia-Pacific Conference on Quality
Software (APAQS'00), page 1 1 1, Los Alamitos, CA, 2000.

[16] Donald Firesmith "Prioritizing Requirements. Journal of Object
Technology", Vol.3, No8, 2004.

[17] D. Harel "On visual formalisms." Communications of the ACM,
vol. 31, no. 5, pp. 514-530, 1988.

[18] D. Harel. "Statecharts: A Visual Formulation for Complex
System." Sci.Comput. Program. 8(3):232-274, 1987.

[19] Flippo Ricca and Paolo Tonella "Analysis and Testing of Web
Applications." Proc. of the 23rd International Conference on
Software Engineering, Toronto, Ontario, Canada, pp.25-34,
2001.

[20] Harel, D. "Statecharts: a visual formalism for complex system."
Science of Computer Programming, v. 8, p. 231-274, 1987.

[21] Hassan Reza, Kirk Ogaard and Amarnath Malge "A Model
Based Testing Technique to Test Web Applications Using
Statecharts." Fifth International Conference on Information
Technology, 2008.

[22] Ibrahim K. El-Far and James A. Whittaker "Model-based
Software Testing", 2000.

[23] Jim Heumann "Generating Test Cases From Use Cases."
Rational Software, 2001.

[24] Johannes Ryser and Martin Glinz "SCENT: A Method
Employing Scenarios to Systematically Derive Test Cases for
System Test", 2000.

[25] Karl E. Wiegers "First Things First: Prioritizing Requirements."
Published in Software Development, 1999.

[26] Karlsson, J. "Software Requirements Prioritizing." Proceedings
of the Second International Conference on Requirements
Engineering (ICRE'96). Colorado Springs, CO, April 15-18,
1996. Los Alamitos, CA: IEEE Computer Society, p 110-116,
1996.

[27] Karlsson, J. "Towards a Strategy for Software Requirements
Selection." Licentiate. Thesis 513, Linkoping University, 1995.

[28] Karlsson, J. & Ryan, K. "A Cost-Value Approach for
Prioritizing Requirements." IEEE Software September/October,
p67-75, 1997.

[29] Leffingwell, D. & Widrig, D. "Managing Software
Requirements: A Use Case Approach", 2nd ed. Boston, MA:
Addison-Wesley, 2003.

[30] Leslie M. Tierstein "Managing a Designer / 2000 Project."
NYOUG Fall'97 Conference, 1997.

[31] L. Brim, I. Cerna, P. Varekova, and B. Zimmerova
"Component-interaction automata as a verification oriented
component-based system specification." In: Proceedings
(SAVCBS'05), pp. 31-38, Lisbon, Portugal, 2005.

[32] Mahnaz Shams, Diwakar Krishnamurthy and Behrouz Far "A
Model-Based Approach for Testing the Performance of Web
Applications." Proceedings of the Third International Workshop
on Software Quality Assurance (SOQUA'06), 2006.

[33] Manish Nilawar and Dr. Sergiu Dascalu "A UML-Based
Approach for Testing Web Applications." Master of Science
with major in Computer Science, University of Nevada, Reno,
2003.

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

[34] Moisiadis, F. "Prioritising Scenario Evolution." International
Conference on Requirements Engineering (ICRE 2000), 2000.

[35] Moisiadis, F. "A Requirements Prioritisation Tool." 6th
Australian Workshop on Requirements Engineering (AWRE
2001 ). Sydney, Australia, 2001 .

[36] M. Prasanna S.N. Sivanandam R.Venkatesan R.Sundarrajan "A
Survey on Automatic Test Case Generation." Academic Open
Internet Journal, 2005.

[37] Nancy R. Mead "Requirements Prioritization Introduction."
Software Engineering Institute, Carnegie Mellon University,
2008.

[38] Park, J.; Port, D.; & Boehm B. "Supporting Distributed
Collaborative Prioritization for Win-Win Requirements Capture
and Negotiation." Proceedings of the International Third World
Multi-conference on Systemics, Cybernetics and Informatics
(SCI'99) Vol. 2. 578-584, Orlando, FL, July 31-August 4, 1999.
Orlando, FL: International Institute of Informatics and Systemic
(HIS), 1999.

[39] Rajib "Software Test Metric." QCON, 2006.

[40] Robert Nilsson, Jeff Offutt and Jonas Mellin "Test Case
Generation for Mutation-based Testing of Timeliness.", 2006.

[41] Saaty, T. L. "The Analytic Hierarchy Process." New York, NY:
McGraw-Hill, 1980.

[42] Shengbo Chen, Huaikou Miao, Zhongsheng Qian "Automatic
Generating Test Cases for Testing Web Applications."
International Conference on Computational Intelligence and
Security Workshops, 2007.

[43] Valdivino Santiago, Ana Silvia Martins do Amaral, N.L.
Vijaykumar, Maria de Fatima, Mattiello-Francisco, Eliane
Martins and Odnei Cuesta Lopes "A Practical Approach for
Automated Test Case Generation using Statecharts", 2006.

[44] Vijaykumar, N. L.; Carvalho, S. V.; Abdurahiman, V. "On
proposing Statecharts to specify performance models."

International Transactions in Operational Research, 9, 321-336,
2002.

[45] Wiegers, K. "E. Software Requirements", 2nd ed. Redmond,
WA: Microsoft Press, 2003 .

[46] Xiaoping Jia, Hongming Liu and Lizhang Qin "Formal
Structured Specification for Web Application Testing". Proc. of
the 2003 Midwest Software Engineering Conference
(MSEC'03). Chicago, IL, USA. pp.88-97, 2003.

[47] Yang, J.T., Huang, J.L., Wang, F.J. and Chu, W.C.
"Constructing an object-oriented architecture for Web
application testing." Journal of Information Science and
Engineering 18, 59-84, 2002.

[48] Ye Wu and Jeff Offutt "Modeling and Testing Web-based
Applications", 2002.

[49] Ye Wu, Jeff Offutt and Xiaochen "Modeling and Testing of
Dynamic Aspects of Web Applications, Submitted for
publication." Technical Report ISE-TR-04-01,

www.ise.gmu.edu/techreps/, 2004.

[50] Zhu, H., Hall, P., May, J. "Software Unit Test Coverage and
Adequacy." ACM Comp. Survey 29(4), pp 366-427, 1997.

[51] Kano Noriaki, Nobuhiku Seraku, Fumio Takahashi,Shinichi
Tsuji. "Attractive Quality and Must-Be Quality." Journal of the
Japanese Society for Quality Control. 14(2), pp 39-48, 1984.

[52] Cadotte, Ernest R., Turgeon, Normand "Dissatisfiers and
Satisfiers: Suggestions from Consumer Complaints and
Compliments." Journal of Consumer Satisfaction,
Dissatisfactions and Complaining Behavior. 1, pp 74-79, 1988.

[53] Brandt, D. Randall "How service marketers can identify value-
enhancing service elements." Journal of Services Marketing.
2(3), pp 35-41, 1988.

[54] Herzberg, Frederick, Mausner, B., Snyderman, B.B. "The
motivation to work." New York: Wiley, 2 nd edition, 1959.

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

White-Box Test Reduction Using Case-Based

Maintenance

Siripong Roongruangsuwan

Autonomous System Research Laboratory

Faculty of Science and Technology, Assumption University

Bangkok, Thailand

Jirapun Daengdej

Autonomous System Research Laboratory

Faculty of Science and Technology, Assumption University

Bangkok, Thailand

Abstract — Software testing has been proven that it takes around
50-70% of the costs associated with the large development of
commercial software systems. Many reduction techniques have
been proposed to reduce costs. Unfortunately, the cost is usually
over budget and those techniques are failed to reasonably control
costs. The primarily outstanding research issues, motivated this
study, are a large number of redundancy test cases and a
decrease of ability to detect faults. To resolve these issues, this
paper proposes new deletion algorithms to minimize a number of
white-box test cases, while maximizing an ability to reveal faults,
by using a concept of case-based maintenance. Our evaluations
have shown that the proposed techniques can significantly reduce
a number of unnecessary test cases while preserving the
capability of fault detection.

Keywords-component; Test reduction, test case reduction,
deletion method, case based maintenance and test case deletion

Introduction

Software Testing is an empirical investigation conducted to
provide stakeholders with information about the quality of the
product or service under test [13], with respect to the context in
which it is intended to operate. Software Testing also provides
an objective, independent view of the software to allow the
business to appreciate and understand the risks of
implementation of the software. The software testing
techniques include the process of executing a program or
application with the intent of finding software bugs. It can also
be stated as the process of validating and verifying that
software meets the business and technical requirements that
guided its design and development, so that it works as
expected. Software Testing can be implemented at any time in
the development process; however, the most test effort is
employed after the requirements have been defined and coding
process has been completed.

Many researchers [6], [7], [8], [9], [10], [19], [24], [25],
[26], [28], [30], [36], [37], [39] have proven that these test case
reduction methods can reserve the ability to reveal faults.
However, there are many outstanding research issues in this
area. The motivated research issues are: a large number of test
cases particularly redundancy test cases and a decrease of
ability to detect faults. The study shows that test case reduction
methods have been researched over a long period of time, such
as test case prioritization, random approach and coverage-based

test case reduction techniques. Also, the study reveals that
coverage-based approaches are wildly used and researched.
Therefore, we concentrate on an approach to reduce test cases
based on the coverage factor. Many coverage factors have been
proposed over a long period of time. Unfortunately, existing
factors and test case reduction methods ignore the complexity
and impact of test cases. Thus, we propose to reduce a number
of test cases by considering both of test case complexity and
impact.

Our study [5] shows that one of effective approaches that
significantly reduce a number of redundancy test cases is to
apply the concept of artificial intelligent. There are many
artificial intelligent concepts, such as neutral network, fuzzy
logic, learning algorithms and case-based reasoning (CBR).
CBR is one of the most popular and actively researched areas
in the past. The researches [4], [5], [26], [35] show that CBR
has identical problems as same as software testing topic, such
as a huge number of redundancy cases and a decrease of
system's ability to resolve problems.

Fundamentally, there are four steps in the CBR system,
which are: retrieve, reuse, revise and retain. These steps can
lead to a serious problem of uncontrollably growing cases in
the system. However, the study shows that there are many
proposed techniques in order to control a number of cases in
the CBR system, such as add algorithms, deletion algorithms
and maintenance approaches. CBR have been investigated by
CBR researchers in order to ensure that only small amounts of
efficient cases are stored in the case base.

The previous work [27] shows that deletion algorithms are
the most popular and effective approaches to maintain a size of
the CBR system. There are many researchers have proposed
several deletion algorithms [4], [20], [35], such as random
method, utility approach and footprint algorithm. These
algorithms aim to: (a) remove all redundancy or unnecessary
cases (b) minimize a number of cases and (c) maintain the
ability of solving problems. Nevertheless, each technique has
strength and weakness. Some methods are suitable for
removing cases. Some methods are perfectly suitable for
reducing time. Some may be used for reserving the problem
solving capability. Eventually, the previous work [27]
discovered several effective methods (e.g. confidential case
filtering method, coverage value algorithm and confidential
coverage approach) to remove those cases, minimize size of

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

CBR and reduce amount of time, while preserving the ability of
CBR system's problem solving skill. Therefore, this paper
applies those effective deletion techniques to resolve the
problems of software testing.

In the light of software testing, the proposed techniques
focus on how to maintain the test case while maintaining the
capability of fault detection. It is appear that test cases in this
paper are treated as cases in the CBR system. Also, there is an
assumption that a given set of test cases are generated by a
path-oriented test case generation technique. The path-oriented
technique is widely used for a white-box testing that derives
test cases from available source code.

Section 2 discusses an overview of test case reduction
techniques and approach to maintain CBR. Section 3 provides
a definition of terminologies used in this paper. Section 4 lists
the outstanding research issues motivated this study. Section 5
proposes deletion algorithms using the concept of CBR.
Section 6 describes an evaluation method and discusses an
evaluation result. The last section represents all source
references used in this paper.

II. Literature Review

This section describes an overview of test case reduction
techniques and the concept of case based maintenance. The
following describes those two areas in details.

A. Test Case Reduction Techniques

This section discusses and organizes test case reduction (or
TCR) techniques researched in 1995-2006. This study shows
that there are many researchers who proposed a method to
reduce unnecessary test cases (also known as redundancy test
cases), like Offutt [2], Rothermel [8], McMaster [25] and
Sampth [31]. These techniques aim to remove and minimize a
size of test cases while maintaining the ability to detect faults.
The literature review [6], [7], [8], [9], [10], [11], [19], [24],
[25], [36], [37], [39] shows that there are two types of
reduction techniques, which are: (a) pre-process and (b) post-
process. First, the pre-process is a process that immediately
reduces a size of test cases after generating. Typically, it is
occurred before regression testing phase. Second, the post-
process is a process that maintains and removes unnecessary
test cases, after running the first regression testing activities.
Although these techniques can reduce the size of test cases, but
the ability to reveal faults seems slightly to be dropped.
However, Jefferson Offutt [5] and Rothermel [6], [7], [8], [9],
[10], [19], [20], [21], [32] has proven that these test case
reduction techniques have many benefits, particularly during
the regression testing phase, and most of reduction techniques
can maintain an acceptable rate of fault detection. The
advantages of these techniques are: (a) to spend less time in
executing test cases, particularly during the regression testing
phase (b) to significantly reduce time and cost of manually
comparing test results and (c) to effectively manage the test
data associated with test cases. This study proposes a new "2C"
classification of test case reduction techniques, classified based
on their characteristics, as follows: (a) coverage-based
techniques and (b) concept analysis-based techniques.

Autonomous System Research Laboratory, Faculty of Science and
Technology, Assumption University.

B. Case-Based Maintenance (CBM)

Due to the CBR's life cycle [16], the case base size grows
rapidly. That is caused a serious problem directly, for instance,
duplicate data, inconsistency data, incorrect data, and an
expense of searching for an appropriate case in a large case
base size. CBR can be classified as one of the Artificial
Intelligence algorithms. CBR solves new problem by retrieving
the similar case from the existing case base and then adapts the
existing case according to the target problem. Over the time,
CBR is growing. When the uncontrollable case-based growth is
occurred, the performance of CBR is decreasing. Therefore, the
maintenance process is required in order to preserve or improve
the performance of the system. The process of maintaining
CBR is called CBM. David C. Wilson [8] presented the overall
concepts of CBR and case based maintenance. This paper
focused on the case based maintenance (CBM) approach in
term of the framework. In other words, this paper described the
type of data collection and how the case based maintenance
works. There were so many policies for CBM, for example,
addition, deletion, and retain.

"CBM was defined as the process of refining a CBR system's
case-base to improve the system 's performance. It implements
policies for revising the organization or contents
(representation, domain content, accounting information, or
implementation) of the case-base in order to facilitate future
reasoning for a particular set of performance objectives. "

These studies [2], [3], [4], [16], [17], [27], [35] reveal that
several deletion algorithms have been proposed. For example, a
random approach (RD), utility deletion algorithm (UD),
footprint deletion algorithm (FD), footprint utility deletion
algorithm (FUD) and iterative case filtering algorithm (ICF).

RD is the simplest approach, which removes the case
randomly. UD deletes the case that has minimum utility value.
Footprint algorithm uses the competence model and removes
the auxiliary case from the system. FUD is a hybrid approach
between Utility algorithm and Footprint algorithm, and is
concerned with the competence model and the utility value.
Finally, ICF focuses on the case, which the reachability set is
greater than the coverage set [16], [27].

III. DEFINITION

This section describes a definition of CBR terminologies
used in the software testing area.

TABLE I.

Definitions of CBR for Software Testing

Element

CBR

Software Testing

Coverage
set

Coverage Set is the
set of target problems,
which it can be used
to solve successfully

[4].

Coverage set means a
set of stages /
elements, which they
can be used to test
successfully and
reveal faults.

Reachability
set

Reachability Set is the
set of case bases that
can be used to solve

Reachability set means
a set of test cases that
can be used to reveal

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

the target problem [4].

faults.

Competence
set

Competence is the
range of the target
problem that can be
solved successfully
[4].

Competence is the
range of ability to
reveal faults that can
be used to test
successfully.

Auxiliary
set

Auxiliary Case is a
case that does not
have a direct effect on
the competence of a
system when it is
deleted [4].

Auxiliary case is a test
case that does not have
a direct effect on the
competence of a
system when it is
removed.

Pivot set

Pivotal Case is the
case that does have a
direct effect on the
competence of a
system if it is deleted
[1], [29].

Pivotal case is a test
case that does have a
direct effect on the
ability to reveal faults
if it is deleted.

IV. RESEARCH PROBLEM

This section discusses the details of research issues
motivated this study. The literature review reveals that [13],
[23], [24], [25], [31], [38] those research issues are: (a) a large
number of redundancy test cases and (b) a decrease of the
ability to reveal faults. These research issues can be elaborated
in details as follows:

First, the literature review shows that redundancy test cases
are test cases tested by multiple test cases. Many test cases that
are designed to test the same things (e.g. same functions, same
line of code or same requirements) are duplicated. Those
duplicated tests are typically occurred during testing activities,
particularly during regression testing activities [13], [23], [24],
[25], [31], [38]. Those duplicated tests can be eventually
removed in order to minimize time and cost to execute tests.

The following shows an example of redundancy test cases.

Description

input

Sequence

Expected
Result

Actual
Result

Status
Pass / Fail

06-1

Withdraw Honey
torn ATM

1. Insert ATM Card

2. Insert PIN

3. Select -Wthwatf
1 Select MC Type
5. Identify Amount
S. Ciek-OK-

7. Receive Money
'. Receive Carrl

Money is

pluintfl

Baldnceis T^
ialeulated.

ATM Card iJ-^
cturncd. ~

)

hquiryS
Mthdraw Money
broiTi ATM

PIN. Amount

1. Insert ATM Card
2 Insert PIN

». SeleetWCType
5. Click "OK"
> Select -Wth*aw
7. Select AiCType
El Identify Amount

9. Ckck-OK"

1 0. Receive Money

Cuirenlbalancel
s displayed. /U

Monty is /

eturned (

Balance is \f

sacculated. \J

ATM Card is

eturned.

Redu
"kit

umber of

Cases

Figure 1 . An Example of Control Flow Graph

From the above figure, there are two test cases, with the
duplicated sequence and expected result, designed to test a
"withdraw" feature in ATM machine. The sequence of the first
test case is: (a) insert ATM card (b) insert PIN (c) select
"withdraw" (d) select account type (e) identify an amount (f)
click "OK" button (g) receive money and (h) receive card. The
sequence of the second test case is similar to the first one.
However, the additional step in the second case is to inquiry a
balance amount before withdrawing the money. Therefore, it is
appear that the first test case is a part of the second test case.
We call the first test case as a redundancy test case.

The study shows that there are many proposed methods to
delete those duplicated test cases such as McMaster's work [24]
[25], Jeffs method [13] and Khan's approach [23]. Also, the
study shows that one of the most interesting research issues is
to minimize those duplicated tests and reduce cost of executing
tests. Although there are many proposed methods to resolve
that issue, that issue is still remaining. Thus, it is a challenge
for researchers to continuously improve the ability to remove
duplicated tests.

Last, test cases are designed to reveal faults during software
testing phase. The empirical studies [8], [10], [19], [20], [21],
[30], [32], [39] describe that reducing test cases may impact to
the ability of detect faults. Many reduction methods decrease a
capability of testing and reveal those faults. Therefore, one of
outstanding research challenges for researchers is to remove
tests while preserving the ability to defect faults.

V. PROPOSED METHOD

For evolving software, test cases are growing dramatically.
The more test cases software test engineers have, the more time
and cost software test engineers consume. The literature review
shows that regression testing activities consume a significant
amount of time and cost. Although, a comprehensive set of
regression selection techniques [8], [9], [10], [19] has been
proposed to minimize time and cost, there is an available room
to minimize size of tests and clean up all unnecessary test
cases. Thus, removing all redundancy test cases is desirable.

There are many approaches to reduce redundancy test cases
and applying an artificial intelligent concept in the test case
reduction process is an innovated approach. The literature
review [5], [27] shows that there are many areas of artificial
intelligent concept, such as artificial neutral network, fuzzy
logic, learning algorithms and CBR concept. Also, it reveals
that CBR has a same research issue as software testing has. The
issue is that cases in the CBR system will be consistency
growing bigger and larger all the time. There are four steps in
CBR that can uncontrollably grow a size of the system:
retrieve, reuse, revise and retain. Therefore, many CBR papers
aim to reduce all redundancy cases, known as "deletion
algorithms". The smaller size of CBR system is better and
desirable. Due to the fact that CBR has the same problem as
software testing and this paper focuses on reduction methods,
therefore, this paper proposes to apply CBR deletion
algorithms to the test case reduction techniques.

This paper introduces three reduction methods that apply
CBR deletion algorithms: TCCF, TCIF and PCF methods.

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Those techniques aim to reduce a number of test cases
generated by path-oriented test case generation technique. This
technique is used for white-box testing only. However, the
generation methods are out of the scope of this paper.

The limitation of the proposed deletion algorithms are: (a)
those methods are perfectly suitable for only white-box testing
techniques and (b) path coverage may not be applicable for a
large system that contains over million lines of code.

A. Example of Test Cases

Given a set of test cases generated, this study discusses the
use of a number of case maintenance techniques, which have
been investigated by CBR researchers in ensuring that only
small amount of cases are stored in the case base, thereby
reducing number of test cases should be used in software
testing. Similar to what happen to software testing, a number of
CBR researchers have focused on finding approaches
especially for reducing cases in the CBR systems' storages.

This paper proposes to use the path coverage criteria in
order to reduce redundancy test cases. This is because path
coverage has a huge benefit of required very thorough testing
activities. The following describes in details of the above path
coverage using in the software testing field. Let S = {si, s2, s3,
s4, s5} to be a set of stage in the control flow graph. The
control flow graph can be derived from the source-code or
program. It is a white-box testing. Thus, each state represents a
block of code. The techniques that aim to generate and derive
test cases from the control flow graph are well-known as path-
oriented test case generation techniques. These techniques are
widely used to generate test cases. There are many research
papers on this area. However, the test case generation
techniques are out of scope in this paper.

Figure 2. An Example of Control Flow Graph

From the above figure, this paper assumes that each state
can reveal a fault. Thus, an ability to reveal faults of five states
is equal to 5. Also, it is assumed that every single transaction
must be tested. This example is used in the rest of paper.

Let TCn = {si, s2, ...,sn} where TC is a test case and sn is
a stage or node in the path-oriented graph that is used to be
tested. From the above figure, a set of test cases can be derived
as follows:

TCj = {s h s 2 J

TC 2 = {s h s 3 J

TC 3 = {si, s 4 J

TC 4 = {si, s 2 , s 3 J

TC 5 = {s h s 3 , s 5 J

TC 6 = {sj, s 4 , s 3 j

TC 7 = {s h s 2 , s 3 , s 5 j

TC 8 = {s h s 4 , s 3 , s 5 j

TC 9 = {s 2 , s 3 J

TC 10 = {s 2 , s 3 , s 5 J

TC n = {s 3 , s 5 J

TC 12 = {s 4 , s 3 J

TC 13 = {s 4 , s 3 , s 5 J

From the figure 2, we assume the following: (a) each state
represents a block of source code (b) each state can reveal only
1 fault; the total ability to reveal faults is 5 and (c) every single
transaction in the control flow graph must be tested.

The following describes the proposed methods that apply
the concept of CBR in details:

B. Path Coverage for Filtering (PCF)

Code coverage analysis is a structural testing technique
(also known as white box testing). Structural testing compares
test program behaviour against the apparent intention of the
source code. This contrasts with functional testing (also
referred to black-box testing), which compares test program
behaviour against a requirements specification. Structural
testing examines how the program works, taking into account
possible pitfalls in the structure and logic. Functional testing
examines what the program accomplishes, without regard to
how it works internally. Structural testing is also called path
testing since you choose test cases that cause paths to be taken
through the structure of the program. The advantage of path
cover is that it takes responsible for all statements as well as
branches across a method. It requires very thorough testing.
This is an effective substitute of other coverage criteria. The
path coverage is used as coverage value in this technique. The
Coverage value is combined into the addition policy for adding
significant case [12]. Within the adding algorithm along with
the coverage weight value stated in the review, the concept of
deletion algorithm and the coverage have been proposed. The
coverage value can specify how many nodes that the test case
can cover. In other words, the coverage value is an indicator to
measure that each test case covers nodes. It means that the
higher coverage value is, the more nodes can be contained and
covered in the test case.

Let Cov(n) = value where Cov is a coverage value, value is
a number of test cases in each coverage group and n is a
coverage relationship.

The procedure of this method can be elaborated briefly as
the following steps.

The first step is to determine a coverage set. From figure 2,
each coverage set can be identified as follows:

Coverage (l) = {rC;}

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Coverage (2) = {TC 2 }

Coverage (3) = [TC 3 ]

Coverage (4) = {TC h TC 4 , TC 9 ]

Coverage (5) = [TC 2 , TC 5 , TC n ]

Coverage (6) = {TC 3 , TC 6 , TC 12 ]

Coverage (7) = {TC h TC 4 , TC 7 , TC 9 , TC 1(h TC n ]

Coverage (8) = {TC 3 , TC 6 , TC 8 , TC n , TC 12 , TC 13 }

Coverage (9) = {TC 9 }

Coverage (10) = {TC 9 , TC 10 , TC n }

Coverage (11) = {TC n }

Coverage (12) = {TC 12 }

Coverage (13) = {TC lh TC 12 , TC 13 }

The second step is also to determine a reachability set. The
reachability set can be figured out from the above coverage set,
based on the given definition in this paper. Therefore, the
reachability set can be identified as follows:

Reachability (TCj) = {1,4, 7}

Reachability (TC 2 ) = {2, 5}

Reachability (TC 3 ) = {3, 6, 8}

Reachability (TC 4 ) = {4,1}

Reachability (TC 5 ) = {5}

Reachability (TC 6 ) = {6, 8}

Reachability (TC 7 ) = {7}

Reachability (TC 8 ) = {8}

Reachability (TC 9 ) = {4, 7, 9, 10}

Reachability (TC 10 ) = {7, 10}

Reachability {TC n ) = {5, 7, 8, 10, 11, 13}

Reachability (TC 12 ) = {6, 8, 12, 13}

Reachability (7C i5 ) = {8, 13}

The next step is to calculate a coverage value. This paper
proposes to calculate a coverage value based on a number of
test cases in each coverage group. Therefore, the coverage
value can be computed as follows:

Cov{\) = 1, Cov{2) = 1, Cov(3) = 1, Cov{A) = 3, Cov(5) = 3,

Cov(6) = 4, Cov(l) = 6, Cov(8) = 6, Cov(9) = 1, Cov(lO) = 3,

Cov(ll) = 1, Cov(12) = 1 and Cov(l3) = 3

Afterward, the step is to determine potential removable test
cases. These test cases can be identified when their number of
members in the reachability set is greater than a number of
members in the coverage set. Therefore, the potential removal
test cases are: TC h TC 2 , TC 3 , TC 9 , TC n and TC 12 .

The last step is to removes all test cases with minimum
coverage value, in the potential removable test cases.
Unfortunately, TC h TC 2 , TC 3 , TC 9 , TC n and TC 12 are removed
due to that they have the minimum coverage value.

C. Test Case Complexity for Filtering (TCCF)

A complexity of test case is the significant criteria in this
proposed method [1], [16]. In this paper, the complexity of test
case measures a number of states included in each test case. We
define the test case complexity as follows:

Definition 1: Let Cplx(TC) = {High, Medium, Low J where
Cplx is a complexity of test case, TC is a test case and the
complexity value can be measured as:

• High when a number of states are greater than an
average number of states in the test suite.

• Medium when a number of states are equal to an
average number of states in test suites.

• Low when a number of states are less than an
average number of states in the test suites.

The procedures of this method can be described briefly in
the following steps. The first two steps are to identify coverage
and reachability set.

Next, the step is to define an auxiliary set. Test cases that
can be included in the auxiliary set have a greater number of
members in the reachability set than a number of members in
the coverage set. From figure 2, therefore, the auxiliary set can
be identified as follows:

Auxiliary set = {TC h TC 2 , TC 3 , TC 9 , TC n , TC 12 }

Afterward, the method computes a complexity value for all
test cases in the above auxiliary set. From figure 2 and test
suites that contain 13 test cases, the average number of states is
(2+2+2+3+3+3+4+4+2+3+2+2+3)/13, which is equal to 3.
Based on the average number of states, the complexity value
for each test case can be computed as follows:

Cplx(TCj) = Low, Cplx(TC 2 ) = Low, Cplx(TC 3 ) = Low,
Cplx(TC 4 ) = Medium, Cplx(TC 5 ) = Medium, Cplx(TC 6 ) =
Medium, Cplx(TC 9 ) = Low, Cplx(TC 10 ) = Medium, Cplx(TC n )
= Low, Cplx(TC 12 ) = Low and Cplx(TC 13 ) = Medium

Finally, the last step removes test cases with minimum of
complexity value from the auxiliary set. Thus, TC b TC 2 , TC 3 ,
TC 9 , TC]j and TC 12 are removed.

D. Test Case Impact for Filtering (TCIF)

The study [21] shows that software is error-ridden in part
because of its growing complexity. Software is growing more
complex every day. The size of software products is no longer
measured in thousands of lines of code, but it measures in
millions. Software developers already spend approximately 80
percent of development costs [18] on identifying and correcting
defects, and yet few products of any type other than software
are shipped with such high levels of errors. Other factors
contributing to quality problems include marketing strategies,
limited liability by software vendors, and decreasing returns on
testing and debugging, according to the study. At the core of
these issues is difficulty in defining and measuring software
quality. Due to the fact that defining and measuring a quality of
software is important and difficult, the impact of inadequate

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

testing must not be ignorance. The impact of inadequate testing
could be lead to the problem of poor quality, expensive costs
and huge time-to-market. In conclusion, software testing
engineers require identifying the impact of each test case in
order to acknowledge and understand clearly the impact of
ignoring some test cases.

In this paper, an impact value is an impact of test cases in
term of the ability to detect faults if those test cases are
removed and not be tested. We define the test case impact as
follows:

Definition 2: Let ImpiTC) = {High, Medium, Low} where
Imp is an impact if a test case is removed, TC is a test case and
the impact value can be measured as:

• High when the test case has revealed at least one
fault for many times.

• Medium when the test case has revealed faults for
only one time.

• Low when the test case has never revealed faults.

The procedure of this method is similar to the previous
method. The only different is that this method aims to use an
impact value instead of complexity value. Therefore, the fire
three steps are to: identify coverage set, define reachability set
and determine an auxiliary set. Afterward, the next step is to
compute and assign an impact value. The method computes the
impact value for all test cases in the above auxiliary set. From
figure 2, the impact value for each test case can be computed as
follows:

Imp(TCj) = Low, Imp(TC 2 ) = High, Imp(TC 3 ) = Medium,
Imp(TC 4 ) = Low, Imp(TC 5 ) = High, Imp(TC 6 ) = Medium,
Imp(TC 9 ) = Low, Imp(TC 10 ) = Low, Imp(TC n ) = Low,
ImpiTC 12) - Low and Imp(TC 13 ) = Low

Finally, the last step removes test cases with minimum of
impact value from the auxiliary set. Thus, TC h TC 4 , TC 7 , TC 9 ,
TC] , TC]], TC] 2 and TC ]3 are removed.

VI.

EVALUATION

This section describes an experiments design,
measurement metrics and results.

This paragraph designs an experiment used to evaluate and
determine the best reduction methods. This paper proposes the
following three steps. First, the experiment proposes to
randomly generate 2,000 test data used in the
telecommunication industry. In this experiment, the test data is
represented as test case. Second, the experiment executes
reduction methods with the generated test cases and compares
among the following reduction methods: RD, UD, FD, FUD,
ICF and three proposed methods (e.g. TCCF, TCIF and PCF).
This step randomly simulates defects for each test case in
order to determine an ability to reveal faults. Third, the
experiment aims to run the above methods for 10 times in
order to calculate the average value for each metric. The
metrics used in this experiment are described in details in next
section. Afterward, the experiment compares the values and

evaluates a result by generating a comparison graph in order to
determine the most recommended reduction approach.

The following table lists the description of each test data
that need to be generated randomly.

TABLE II.

A Form of Test Cases

Attribute

Description

Data
Type

Test Id

A unique index to reference

test data. The value is a

sequence number, starting

atl.

Numeric

A Set of Input Data

Full Name

A first and last name who
own the mobile phone.

String

Name

A mobile brand name. The
value is a range of iPhone,
BlackBerry, Nokia, LG,
Sony Ericsson and
Samsung.

String

Model

A mobile model.

String

Price

A price of mobile. The unit
of price is baht.

Numeric

Weight

A weight of mobile. The
unit of weight is gram (g).

Numeric

Height

A height of mobile. The
unit of height is centimeter
(cm).

Numeric

Graphics

A graphics mode option.
The value can be T or F

Boolean

WAP

A WAP mode. The value is
T or F

Boolean

Color

A mobile color. The color

can be Black, Gold, Silver,

Blue, and White

String

Game

A game mode. The value is
T or F

Boolean

Warranty

A mobile warranty. The
unit of warranty is month.

Numeric

The following table describes an approach to generate
random data using the above attributes respectively.

TABLE III.

Approach to Generate Random Test Cases

Attribute

Approach

Test Id

Generate randomly from the following
combination: t + Sequence Number.

For example, tl, t2, t3, ..., tn.

Name

Random from the following values:

iPhone, BlackBerry, Nokia, LG, Sony

and Samsung.

Model

Random from the following values:
iPhone - iPhone 2G and iPhone 3G.

BlackBerry - BlackBerry Bold 9000,
BlackBerry Bold 9700, BlackBerry

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Curve 8300, BlackBerry Curve 8520,
BlackBerry Curve 8900, BlackBerry
Pearl, BlackBerry Pearl Flip 8200 and
BlackBerry Storm.

Nokia - Nokia N97 Mini, Nokia E90, Nokia
E72, Nokia 8800, Nokia N86 and Nokia
E66.

LG - Lotus Elite LX610, Accolade VX5600,
Chocolate Touch VX8575, Arena
GT950 and eXpo GW820.

Sony Ericsson - Xperia X10, Vivaz Pro,
Xperia X10 Mini Pro, Elm and Aspen.

Samsung - Samsung Gravity 2, Behold II,
Comeback, SGH-T139, SGH-T659 and
SGH-T239.

Price

Random from the following values:
High: 30,000 baht - 50,000 baht

Medium: 10,000 baht -29,999 baht

Low: 3,000 baht - 9,999 baht

Weight

Random from the following values: 4-15

Height

Random from the following values: 5-15

Graphics

Random from the following values:
True or False

WAP

Random from the following values:
True or False

Color

Random from the following values:
Black, Gold, Silver, Blue and White

Game

Random from the following values:
True or False

Warranty

Random from the following values:
Long: 12 months - 18 months

Medium: 6 months - 1 1 months

Short: 1 month - 5 months

methods are: RD, UD, FD, FUD, ICF, TCCF, TCIF and PCF.

Additionally, this section shows a graph format. There are two
dimensions in the following graph: (a) horizontal and (b)
vertical axis. The horizontal represents three measurements
whereas the vertical axis represents the percentage value.

The paragraph lists the measurement metrics used in the
experiment. The first measurement is a number of test cases.
The large number of test cases consumes time, effort and cost
more than the smaller size of test cases. Many reduction or
minimization approaches [6], [7], [8], [9], [10], [11], [19],
[24], [25], [36], [37], [39] have been proposed to minimize
size of test cases. This has proven that size is one of important
metrics in software testing area. The second is an ability to
reveal faults. It aims to measure the percentage of faults
detection. One of the goals of test case with a set of data is to
find defects. Thus, this metric is important criteria to measure
and determine which reduction methods can preserve the high
ability to reveal faults. The last measurement is a total of
reduction time: It is the total number of times running the
reduction methods in the experiment. This metric is related to
time used during execution time and maintenance time of test
case reduction methods. Therefore, less time is desirable.

This paragraph discusses an evaluation result of the above
experiment. This section presents the reduction methods
results in term of: (a) a number of test cases (b) ability to
reveal faults and (c) total reduction time. The comparative

Figure 2. A Graph Comparison of Deletion Methods

The above graph presents that both of FD and PCF
minimize a number of test cases by far better than other
reductions methods, approximately over 15%. Meanwhile,
both of them are the worst methods for preserving an ability to
reveal faults. FUD, TCCF and TCIF are best top three
methods to reserve a capability to detect faults. They are
greater than other methods over 22%. Unfortunately, they are
also the worst three methods that require a lot of time during a
reduction process. In the mean time, both of RD and PCF take
the least total reduction time among other methods.

For a number of test cases, the maximum and minimum
percentage is 56% and 13%. The different between maximum
and minimum value is 43%. An interval value is equal to a
result of dividing the different values by 5. As a result, the
interval value is 8.6. Thus, it can be determined as follows: 5-
Excellent (since 13% to 21.6%), 4-Very good (between 21.6%
and 30.2%), 3-Good (between 30.2% and 38.8%), 2-Normal
(between 38.8% and 47.4%) and 1-Poor (from 47.4% to 56%).

For an ability to reveal faults, the maximum and minimum
percentage is 94% and 37%. The different value is 57%. The
interval value is 11.4. Therefore, it can be determined as
follows: 5-Excellent (since 82.6% to 94%), 4-Very good
(between 71.2% and 82.6%), 3-Good (between 59.8% and
71.2%), 2-Normal (between 48.4% and 59.8%) and 1-Poor
(from 37% to 48.4%).

For a total reduction time, the maximum and minimum
percentage is 64% and 21%. The different between maximum
and minimum value is 43%. An interval value is equal to a
result of dividing the different values by 5. As a result, the
interval value is 8.6. Thus, it can be determined as follows: 5-
Excellent (since 21% to 29.6%), 4-Very good (between 29.6%

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

and 38.2%), 3-Good (between 38.2% and 46.8%), 2-Normal
(between 46.8% and 55.4%) and 1-Poor (from 55.4% to 64%).
Therefore, the experiment result of those eight comparative
methods can be shown below:

TABLE IV.

A Comparison of Test Case Reduction Methods

Algorithm

Number of

Test Cases /

Size

Abili
tyto
Reveal
Faults

Total
Reduction
Time

Random Deletion
(RD)

Utility Deletion
(UD)

Footprint
Deletion (FD)

Footprint Utility
Deletion (FUD)

Iterative Case
Base Filtering (ICF)

Test Case
Complexity for
Filtering (TCCF)

Test Case Impact
for Filtering (TCIF)

Path Coverage
for Filtering (PCF)

The above result suggests that FD and PCF is perfectly
suitable for a scenario that does not directly concern about an
ability to reveal faults and total reduction time. Both of FD
and PCF are two of the most excellent methods to minimize a
number of test cases. Meanwhile, FUD, TCCF and TCIF are
the most recommended methods to delete tests while
preserving the ability to detect faults. In addition, both of RD
and PCF are excellent in case that total reduction time is
matter.

The evaluation result suggests that FD and PCF is
perfectly suitable for a scenario that does not directly concern
about an ability to reveal faults and total reduction time. Both
of FD and PCF are two of the most excellent methods to
minimize a number of test cases. Meanwhile, FUD, TCCF and
TCIF are the most recommended methods to delete tests while
preserving the ability to detect faults. In addition, both of RD
and PCF are excellent in case that total time is matter.

VII. CONCLUSION

This paper reveals that there are many research challenges
and gaps in the test case reduction area. Those challenges and
gaps can give the research direction in this field. However, the
research issues that motivated this study are: a large number of
test cases and a decrease of ability to reveal faults. This paper
combines an approach to maintain CBR and test case
reduction in order to minimize a number of redundancy tests
while maintaining an ability to detect faults.

In this paper, we proposed deletion algorithms to reduce a
number of test cases that generated by widely-used white-box
testing techniques. As part of our research, we conducted an
experiment with 2,000 test cases used in the
telecommunication industry in Thailand. Consequently, our
evaluation result reveal that the proposed method is one of the
most recommendation techniques to maintain ability to reveal
faults and minimize a number of redundancy test cases.
However, the limitation of the proposed techniques is that path
coverage may be not an effective coverage factor for a huge
system that contains million lines of code. This is because it
requires an exhaustive time and cost of identify coverage from
a huge amount of codes.

In future research, we plan to develop deletion algorithms
with other coverage factors, such as function coverage and
block-statement coverage. Also, we aim to implement deletion
algorithms for multiple test suites. Finally, the evaluation
experiment should be conducted for a large commercial
system

REFERENCES

[1] Barry Smyth and Keane. "Remembering To Forget: A
Competence Preserving Deletion Policy for Case-Based
Reasoning Systems." Proceedings of the 14th International Joint
Conference on Artificial Intelligence. Montreal, Quebec,
Canada: Morgan-Kaufman Inc., 1995. 377-382.

[2] Beizer, Boris. Software Testing Techniques. New York, USA:
Van Nostrand Reinhold Inc., 1990.

[3] BO Qu, Changhai Nie, Baowen Xu and Xiaofang Zhang. "Test
Case Prioritization for Black Box Testing." Proceeding with 31st
Annual International Computer Software and Applications
Conference (COMPSAC 2007). Beijing, China, 2007. 465-474.

[4] Boehm, B.W. "A spiral model of software development and
enhancement." IEEE Software Engineering (IEEE Computer
Society), 1988: 61-72.

[5] Daengdej, Jirapun. Adaptable Case Base Reasoning Techniques
for Dealing with Highly Noise Cases. PhD Thesis, The
University of New England, Australia: The University of New
England, 1998.

[6] Gregg Rothermel and Mary Jean Harrold. "A Safe, Efficient
Regression Test Selection Technique." ACM Transactions on
Softw. Eng. And Methodology, 1997: 173-210.

[7] Gregg Rothermel and Mary Jean Harrold. "Analyzing
Regression Test Selection Techniques." IEEE Transactions on
Software Engineering, 1996: 529-511.

[8] Gregg Rothermel, Mary Jean Harrold, Jeffery Ostrin and
Christie Hong. "An Empirical Study of the Effects of
Minimization on the Fault Detection Capabilities of Test
Suites." Proceedings of IEEE International Test Conference on
Software Maintenance (ITCSM'98). Bethesda, Maryland, USA:
IEEE Computer Society, 1998. 33-43.

[9] Gregg Rothermel, Mary Jean Harrold, Jeffery von Ronne and
Christie Hong. "Empirical Studies of Test-Suite Reduction."

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Journal of Software Testing, Verification, and Reliability 12, no.
4 (December 2002): 219-249.

[10] Gregg Rothermel, Roland H. Untch, Chengyun Chu and Mary
Jean Harrold. "Prioritizing Test Cases For Regression Testing."
IEEE Transactions on Software Engineering, 2001: 929-948.

[11] Jefferson Offutt, Jie Pan and Jeffery M. Voas. "Procedures for
Reducing the Size of Coverage-based Test Sets." Proceedings of
the Twelfth International Conference on Testing Computer
Software. Washington D.C, USA, 1995. 111-123.

[12] Jun Zhu and Quiang Yang. "Remembering To Add
Competence-preserving Case Addition Policies for Case Base
Maintenance." Proceedings of the 16th International Joint
Conference in Artificial Intelligence. Stockholm, Sweden :
Morgan Kaufmann Publishers Inc, 1999. 234-241.

[13] Kaner, Cem. "Exploratory Testing." Quality Assurance Institute
Worldwide Annual Software Testing Conference. Orlando,
Florida, USA: Florida Institute of Technology, 2006.

[14] Lehmann, E. and J. Wegener. "Test case design by means of the
CTE XL." Proceedings of the 8th European International
Conference on Software Testing,. Kopenhagen, Denmark: ACM
Press, 2000. 1-10.

[15] Mary Jean Harrold, Rajiv Gupta and Mary Lou Soffa. "A
Methodology for Controlling the Size of A Test Suite." ACM
Transactions on Software Engineering (ACM) 2, no. 3 (July
1993): 270-285.

[16] Nicha Kosindrdecha and Jirapun Daengdej. A Deletion
Algorithm for Case-Based Maintenance Based on Accuracy and
Competence. MS Thesis, Faculty of Science and Technology,
Assumption University, Bangkok, Thailand: Assumption
University, 2003.

[17] Nicha Kosindrdecha and Siripong Roongruangsuwan.
"Reducing Test Cases Created by Path Oriented Test Case
Generation." Proceedings of the AIAA Conference and
Exhibition. Rohnert Park, California, USA: NASA AIAA, 2007.

[18] NIST. The economic impacts of inadequate infrastructure for
software testing. Technical Report, USA: NIST, 2002.

[19] Rothermel, G., R.H. Untch, C. Chu and M.J. Harrold. "Test case
prioritization: An empirical study." Proceedings of the 15th
IEEE International. Oxford, England, UK: IEEE Computer
Society, 1999. 179-188.

[20] S. Elbaum, A. G. Malishevsky and G. Rothermel. "Prioritizing
Test Cases for Regression Testing." Proceedings of the
International Symposium on Software Testing and Analysis.
2000. 102-112.

[21] S. Elbaum, A. Malishevsky, and G. Rothermel. "Test Case
Prioritization: A Family of Empirical Studies." IEEE Trans, on
Software Engineering 28 (February 2002): 159-182.

[22] S. Elbaum, P. Kallakuri, A. G. Malishevsky, G. Rothermel, and
S. Kanduri. "Understanding the effects of changes on the cost-
effectiveness of regression testing techniques." Journal of
Software Testing, Verification, and Reliability 13, no. 2 (June
2003): 65-83.

[23] Saif-ur-Rebman Khan and Aamer Nadeem. "TestFilter: A
Statement-Coverage Based Test Case Reduction Technique."
Proceedings of 10th IEEE International Multitopic Conference.
Islamabad, Pakistan, 2006.

[24] Scott McMaster and Atif Memon. "Call Stack Coverage for GUI
Test-Suite Reduction." Proceedings of the 17th IEEE
International Symposium on Software Reliability Engineering
(ISSRE 2006). North Carolina, USA, 2006.

[25] — . "Call Stack Coverage for Test Suite Reduction." Proceedings
of the 21st IEEE International Conference on Software
Maintenance (ICSM'05). Budapest, Hungary, 2005. 539-548.

[26] — . "Fault Detection Probability Analysis for Coverage-Based
Test Suite Reduction." Proceedings of IEEE International
Conference on Software Maintenance (ICSM 2007). Paris,
France, 2007. 335-344.

[27] Siripong Roongruangsuwan and Jirapun Daengdej. Techniques
for improving case-based maintenance. MS Thesis, Faculty of
Science and Technology, Assumption University, Bangkok,
Thailand: Assumption University, 2003.

[28] Siripong Roongruangsuwan and Jirapun Daengdej. Test case
reduction. Technical Report 25521, Bangkok, Thailand:
Assumption University, 2009.

[29] Smyth, Barry. Case Based Design. PhD Thesis, Department of
Computer Science, Trinity College, Dublin, Ireland: Trinity
College, 1996.

[30] Sprenkle, S., S. Sampath and A. Souter. "An empirical
comparison of test suite reduction techniques for user-session-
based testing of web applications." Journal of Software Test.
Verificat. Reliabil., 2002: 587-596.

[31] Sreedevi Sampath, Sara Sprenkle, Emily Gibson and Lori
Pollock. "Web Application Testing with Customized Test
Requirements - An Experimental Comparison Study."
Proceedings of the 17th International Symposium on Software
Reliability Engineering (ISSRE'06). Raleigh, NC, USA: IEEE
Computer Society, 2006. 266 - 278.

[32] Todd L. Graves, Mary Jean Harrold, Jung-Min Kim, Adam
Porter and Gregg Rothermel. "An Empirical Study of
Regression Test Selection Techniques." ACM Transactions on
Software Engineering and Methodology (TOSEM), 2001: 184-
208.

[33] W. Eric Wong, J. R. Horgan, Saul London and Hira Agrawal.
"A Study of Effective Regression Testing in Practice."
Proceedings of 8th IEEE International Symposium on Software
Reliability Engineering (ISSRE'97). California, USA: IEEE
Computer Society, 1997. 264.

[34] W. Eric Wong, Joseph R. Horgan, Saul London and Aditya P.
Mathur. "Effect of Test Set Minimization on the Fault Detection
Effectiveness of the All-Uses Criterion." Proceedings of the 17th
International Conference on Software Engineering. Seattle,
USA: ACM, 1995. 41-50.

[35] Wilson, David C. A Case-Based Maintenance: The husbandry of
experiences. PhD Thesis, Computer and Science, Indiana
University, USA: Indiana University, 2001.

[36] Xiaofang Zhang, Baowen Xu, Changhai Nie and Liang Shi. "An
Approach for Optimizing Test Suite Based on Testing
Requirement Reduction." Journal of Software (in Chinese),
2007: 821-831.

[37] Xiaofang Zhang, Baowen Xu, Changhai Nie and Liang Shi.
"Test Suite Optimization Based on Testing Requirements
Reduction." International Journal of Electronics & Computer
Science 7, no. 1 (2005): 9-15.

[38] Xue-ying MA, Bin-kui Sheng, Zhen-feng HE and Cheng-qing
YE. "A Genetic Algorithm for Test-Suite Reduction."
Proceedings of IEEE International Conference on Systems, Man
and Cybernetics. Hangzhou, China: ACM Press, 2005. 133-139.

[39] Yanbing Yu, James A. Jones and Mary Jean Harrold. "An
Empirical Study of the Effects of Test-Suite Reduction on Fault
Localization." Proceedings of International Conference on
Software Engineer (ICSE'08). Leipzig, Germany: ACM, 2008.
201-210.

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

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

Vol. 8, No. 6, September 2010

Nat Traversal for Video Streaming Applications

Omar A. Ibraheem

National Advanced IPv6 Centre of

Excellence (NAv6)

Universiti Sains Malaysia

1 1800, Penang, MALAYSIA

Omer Amer Abouabdalla

National Advanced IPv6 Centre

Excellence (NAv6)

Universiti Sains Malaysia

1 1800, Penang, MALAYSIA

Sureswaran Ramadass

National Advanced IPv6 Centre

Excellence (NAv6)

Universiti Sains Malaysia

1 1800, Penang, MALAYSIA

Abstract — This paper presents a novel method that exploits the
strength features of two streaming protocols (Real-time
Transport Protocol (RTP) and Hypertext Transfer Protocol
(HTTP)) to overcome the Network Address Translation (NAT)
and firewall traversal problem. The proposed solution is able to
bypass the RTP over all kinds of NATs (including symmetric
NATs) by adding extra fields to the RTP/UDP packet at
transport layer in the sender side. The NAT and firewall will
detect these packets as TCP packets on the channel that
initialized the connection. The receiver side will then remove the
extra fields and recover the packets to their original content. The
proposed work involves adding two modules, one at the client
and the other at the video streaming server. The proposed work
also avoids any modification to the NAT or the RTP protocol
itself.

Keywords- NAT Traversal; Video Streaming; RTP; TCP; UDP;
Windows OS.

I. Introduction

Video streaming is considering one of the famous
technologies which is used today. It provides the ability to
playback video files remotely through computer networks.
The demand for this technology is rapidly increasing due to
wide spread of Internet and increasing of the network
bandwidthfl]

There are two main application layer protocols that are used
for video streaming: RTP and HTTP. Although the RTP
protocol is originally developed for video streaming and the
HTTP protocol is originally developed for browsing, which
has many weakness points when dealing with video streaming,
HTTP protocol is still used and more spread than RTP when
used with video streaming. This is due to the simplicity of
using the HTTP protocol and in order to avoid the problems
that faced the RTP protocol.

While HTTP protocol uses one TCP port at the transport
layer, RTP can use many ports. RTP can use UDPs or TCPs
ports at the transport layer depending on how much the packet
path is suffered from packet loss [2]. In low packets loss
environment, the use of RTP over UDP protocol is preferable,
since in media streaming, the small ratio of packets loss better
than packets delay. Hence, the higher reliability of the TCP is
not desired[3]. UDP/RTP has also the multicasting feature and
has the ability to deal with real time communication due to its
features in bandwidth, jitter, reliability and end node's
processing.

RTP/TCP can cause the video streaming to suffer from
discontinuity because the need to reordering,
acknowledgement, and retransmission and the packets,
whereas RTP/UDP can suffer from dropping the packets by
some filters (firewalls) in the Internet Service Provider (ISPs).
Some ISPs drop UDP packets because they are connectionless
hence unfair against TCP traffic. They also need high
processing power and memory to ensure security [4]. But the
main issue that can occur is when using the RTP with the
Network Address Translation (NAT). NAT drops any
RTP/UDP or RTP/TCP packets that are initialized from the
outside (Internet) when incoming to the end-systems (behind
the NAT).

The NAT is a technology that permits many computers on
the same network to share a public Internet Protocol (IP)
address for accessing the Internet. The main reason behind the
wide spread of using the NAT is the limited number of the
available IPv4 addresses [5].

The use of RTP/UDP or RTP/TCP video streaming is
started with a TCP connection that is established by a request
from the client to the server, after initial negotiation using the
RTSP protocol on the same established TCP channel, the
server starts video streaming through UDP or TCP ports
initialized from the server not through the established
RTSP/TCP channel [2].

The NAT permits to pass the outgoing connections requests
produced from a host behind the NAT into the outside
network (like Internet) [6], however it does not permit to pass
any connection request produced from the outside network
(like Internet) to any host behind the NAT [7]. This is because
the translation table entry is constructed only when a client
(behind the NAT) initializes a request to connect to a host on
the outside network (Internet) [8], [9]. If the initialized request
comes from a host outside the network of the NAT into the
inside network, the NAT cannot identify the destination host
for this request and the connection between the outside host
and the inside one cannot be occur [8], [10]. Regarding to the
RTP/UDP video streaming, the NAT will not allow the UDP
video streaming channels to pass to the client behind the NAT,
since the RTP/UDP channels are initially established from the
server (on the Internet).

Considering the RTP weakness points, the HTTP protocol,
is the preferable choice for video streaming. However, HTTP
protocol also has known weakness points: the user can suffer

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

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

from quality reduction and playback discontinuity due to the
probing behaviour of TCP protocol. This can also cause an
oscillating throughput and slow recovery of the packet rate.

In contrast, the UDP protocol provides a mean to keep the
desired sending rate constant. It also keeps streaming smooth
and eliminates the TCP related processing.

This paper presents a novel method to utilize the benefits of
both TCP and UDP. The proposed method enables the video
streaming to traverse the NAT by converting each RTP/UDP
and RTCP/UDP packet into fake TCP packet just before being
sent (at data link layer) by adding a fabricated TCP header
before each UDP video streaming packet and making the
necessary modifications to the length and checksums fields.

These fabricated TCP packets will pass the NAT, since
they will be transmitted on the channel (IP, TCP port) that
firstly initialized (RTSP/TCP channel) by the client behind the
NAT. In this paper, this channel is called the active channel.

The receiver, on the other side has to restore the original
UDP packet before being processed by the corresponding
transport layer. The restoration is based on a specific signature.
In order to restore the packets, every fabricated TCP packet
has to have a known signature. Depending on that signature,
the receiver will restore the original packet. All of the
previous changes are performed at the data link layer.

The rest of this paper is organized as follows: section II,
looks at some related work. In section III, the proposed
methodology and algorithm are presented. In section IV, the
experiments of the implemented proposed method and its
discussions are described. In section V, the evaluation of the
proposed method and comparisons between the proposed
method and the existing technologies are presented. The paper
is concluded in section VI.

II. Related work

Limited to our knowledge, no many similar works are
presented. However, [4] present a method to overcome the
RTP/UDP issues by putting a proxy server between the client
and the streaming server (at the global network). The proxy
receives a HTTP request (TCP) from the client and translates
it to a RTSP/RTP request to the server (TCP+UDP). The
proxy has two different connections (one for the client and the
other for the streaming serve). The main function of the proxy
is to translate the HTTP streaming protocol into RTSP/RTP
streaming protocol. This can overcome the NAT problem due
to that the HTTP request (TCP) is initialized by the client and
the reply will pass through the same TCP port. However a
third device is needed. In addition it is still using the
constraints of the TCP between the proxy and the client (e.g.
retransmission and reordering ...etc) (in addition to the
increase of traffic to the network). Another issue is that there
are too many operations in order to convert a complete
application protocol into another one. Beside, this method
loses the real time property that is needed for end to end
communication because all the packets must be forwarded at
the proxy server.

Vol. 8, No. 6, September 2010

III. Proposed Methodology

In this work, both the client and the server are assumed to
convert all the RTP/UDP streaming packets into fabricated
TCP packets that can be sent to the other side using the active
channel.

This fabrication process which is implemented for
Windows Operating System (OS) requires a full control of the
incoming/outgoing packets. However, there is the issue of
source code of the TCP/IP (non open source for Windows OS)
is not readily accessible and Windows does not allow the
manipulation of the packets in any TCP/IP protocol suite from
level above the TCP/IP driver layer.

To overcome the inaccessibility issue, a hooking technique
is used in order to control the (frame/packet) at the point that
links between the protocol driver and the NIC card(s), which
is represented by the Network Driver Interface Specification
(NDIS).

Hooking is a technique that can convert the calling of one
operating system function into a new one that in turn calls the
old one. The new function can do extra job before moving the
execution to the old one. This can be done without the need
for the source code of the old one [11].

The proposed modules is implemented and run in windows
user mode. When the module can hook the NDIS, it can
monitor, control, add, and modify the NDIS
incoming/outgoing packets easily.

The NDIS-Hooking driver inserts itself between TCP/IP
and all of the adapters that bind with it as shown in figure (1).

User Mode

Kernel Mode

TCP/IP Stack

NDIS

N^l^JiooJkm^ Driver

Network Adapter Deriver

Figure 1 . NDIS hooking driver with relation to user mode 1

When TCP/IP sends a packet, it reaches the NDIS-Hooking
driver (as a frame) before sending to the adapter. Likewise,
packets that are to be indicated (received) on TCP/IP will go
to the NDIS-Hooking driver first.

1 http://www.ntkernel.com/w&p.php?id=7

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

The fabricated TCP header is inserted/deleted in the data link
frame, this means that the original RTP/UDP protocol is used
without modification. Nonetheless the fabricated packets can
still bypass the NAT as authenticated ones.

Frame
Header

IP
Header

Fabricated
TCP Header

UDP
Header

RDP Header
-KPayLoad

CRC
Checksum

Figure 2. Proposed frame structure

As these extra bytes (fabricated TCP header) will be added
when the packet is in the data link layer, this may cause the
packet to exceed the Maximum Transfer Unit (MTU) of the
network. Since, no packet must exceed the Maximum Transfer
Unit (MTU) of the network [12], [13], therefore, the sender's
MTU must be decreased by length of the fabricated TCP
header length (20 bytes).

The whole proposed system is composed of two main
modules. The first module resides on the streaming client
while the second resides on the streaming server. Figure (3)
shows the video streaming network topology.

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

Vol. 8, No. 6, September 2010
the client's and server's streaming ports, then record this
connection's information into an array. This is happened
normally at the setup phase of the RTSP connection. Later
(when the video streaming data is transfered), the client will
check every TCP packet if it contains a specified signature. If
this signature is raised (in the TCP header), this mean that this
TCP packet is fabricated and it contains the original
RTP/UDP packet. The program will remove the TCP header
and recomputed the UDP and IP checksums. All these steps
are done before sending the packet to the rest of TCP/IP
protocol stack

Outgoing packet: If the packet is outgoing to the streaming
server and the outgoing packet was a RTP/UDP packet, then
insert a new fabricated TCP header before the UDP header.
This fabricated TCP header contains the TCP connection
information taken from the appropriate record from an array
containing all streaming connections' details. This TCP
header also contains a specified signature that has to been
recognized from the streaming server in order to return the
packet back to its original RTP/UDP packet. This operation
also needs to recompute the checksums. All these steps are
done before sending the packet to the adapter. Figure 4 shows
the flowchart of client side module.

Client NAT

Figure 3. Video streaming network topology

Each module consists of the following components:

A component (hooking function in Fig. 1) that provides a
way to access the frame at the data link layer. This component
accesses the frames in data link layer which is in the kernel
mode and moves it into the user mode and vice versa.

A component that finds the required frame based on its
content. This component extracts the specified packets from
the frames which have to be changed (fabricated/restored)
depending on sending direction (income/outcome).

A component that makes the required modifications
(fabricating/restoring) to the predetermined packets. This
component changes the predetermined packets depending the
sending direction (send/receive). In sending, the component
changes the RTP/UDP packet into fabricated TCP packet. In
receiving, the component restores the fabricated TCP packet
into its original RTP/UDP content. This component also re-
computes length and checksums.

A. Client Side Module

As mentioned earlier, the module has to access the kernel
(at data link layer). This is done by accessing the NDIS driver.
The module listens until a packet event has occurred. There
are two possible scenarios:

Incoming packet: If the packet is coming from the
streaming server, then the program will look for the TCP that
contains an RTSP packet. If this RTSP packet contains both

Access the Data
link layer (in
kernel mode)

Figure 4. Flowchart of the client side module

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

B. Server Side Module

In server side module, similar steps to the client are also
implemented. The difference is that the system gets the RTSP
connection's details from the outgoing TCP packet instead of
incoming TCP packet in the client. Figure (5) shows the
flowchart of the main steps of the server module.

Return back to it;

original UDP

packet

Recompute the

Move the packet
to the kernel mode

Send packet to

the upper

layer

Figure 5. Flowchart of the server side module

IV. Experiments and discussions

A. Experiments Setup

In this experiment, we use three PCs running windows XP.
Two PCs with one LAN card (client and the server). The other
PC (working as a NAT) contains two LAN cards.

RedMug streaming server commercial software is used on
the server site. The VLC media player (version 1.0.5) is used
on the client side. The VLC media player is set to use the
RTSP protocol by giving a URL of one movie on the
streaming server. The proposed method (client and server
modules) is implemented in VC++.Net Framework and it is
running in windows OS environment in user mode. A
windows device driver (Windows Packet Filter Kit
"winpkfilter 3.0" from NT Kernel Resources,
http://www.ntkernel.com) is used for the hooking purpose.

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

Vol. 8, No. 6, September 2010
B. Experimental Results and Discussion

In the first experiment (before using the proposed method),
the client tries to access the movie on the streaming server
using the above system configuration. The connection's
establishment and the video streaming negotiations between
the client and the server are established normally. However,
the connection fails at the stage of data streaming
transformation (see Fig. 6).

User- Agent: Ub VLC ,1.1.2 (LIVE555 Streaming Media v20 10.03. 16)

RTSP Session identifier: 5. Sent to client: "RTSP/ 1.0 200 OK

RTSP Session identifier: 5. Received RTSP message from 150.150.100.110:02452. Message: 'DESCRIBE rtsp://ll

User- Agent: Lib VLC/ 1.1.2 (LIVE555 Streaming Media v20 10.03. 16)

RTSPSession identifier: 5. Is the "C:\red MUG \ Contents, sample. mp 4" MP4file in cache? 'true']

RTSPSession identifier: 5. Video track name to stream: track ID/ 2, audio track name to stream: trackID/1]

RTSPSession identifier: 5. Sent to client: 'RTSP/ 1.0 200 OK

RTSPSession identifier: 5. Received RTSP message from 150. 150. 100.1 10:62452. Message: 'SETUP rtsp:// 150. 1

User- Agent: Lib VLC/ 1.1.2 (LIVE555 Streaming Media v20 10.03. 16)

RTSPSession identifier: 5. Sent to client: 'RTSP/ 1.0 200 OK

Session: 257655783;timeout=360

RTSPSession identifier: 5. Received RTSP message from 150. 150. 100.1 10:62452. Message: 'SETUP rtsp: ,, 150. 1

Session: 257655783

User- Agent: Lib VLC/ 1.1.2 (LIVE555 Streaming Media v20 10.03. 16)

RTSPSession identifier: 5. Sent to client: 'RTSP/ 1.0 200 OK

Session: 257655783;timeout=360

RTSPSession identifier: 5. Received RTSP message from 150.150.100.110:62452. Message: 'PLAY rtsp://150. 15

Session: 257655783

User- Agent: Lib VLC/ 1.1.2 (LIVE555 Streaming Media v20 10.03. 16)

RTSPSession identifier: 5. Called the seek method. Start time in sees: 0.000000.]

RTSPSession identifier: 5. Sent to client: 'RTSP/ 1.0 200 OK

Session: 257655783;timeout=360

Client 6 is opened

RTSPSession identifier: 5. Received RTSP message from 150. 150. 100.1 10:62452. Message: 'TEARDOWN rtsp://

Session: 257655783

User- Agent: Lib VLC/ 1.1.2 (LIVE555 Streaming Media v20 10.03. 16)

RTSPSession identifier: 5. Sent to client: 'RTSP/ 1.0 200 OK

RTSPSession identifier: 5. Connection from 150.150.100.110:62452 closed.]

RTSPSession identifier: 6. Connection arrived from 150.150.100.110:62453.]

RTSPSession identifier: 6. Received RTSP message from 150. 150. 100.1 10:62453. Message: 'OPTIONS rtsp:// 15(

User- Agent: Lib VLC/ 1.1.2 (LIVE555 Streaming Media v20 10.03. 16)

RTSPSession identifier: 6. Sent to client: 'RTSP/ 1.0 200 OK

RTSPSession identifier: 6. Received RTSP message from 150.150.100.110:62453. Message: 'DESCRIBE rtsp://ll

User- Agent: Lib VLC/ 1.1.2 (LIVE555 Streaming Media v20 10.03. 16)

RTSPSession identifier: 6. Is the "C:, red MUG ■.Contents, sample. mp4" MP 4 file in cache? "true"]

RTSPSession identifier: 6. Video track name to stream: trackID/2, audio track name to stream: tracklD/lj

RTSPSession identifier: 6. Sent to client: 'RTSP/ 1.0 200 OK

RTSPSession identifier: 6. Received RTSP message from 150. 150. 100.1 10:62453. Message: 'SETUP rtsp: ,, 150. 1

User- Agent: Lib VLC/ 1.1.2 (LIVE555 Streaming Media v20 10.03. 16)

RTSPSession identifier: 6. Sent to client: 'RTSP/ 1.0 500 Internal Server Error

RTSPSession identifier: 6. Connection from 150.150.100.110:62453 closed.]

Figure 6. Connection breakdown when data streaming transforming began
(server side)

The reason for the success of the initialization of the client-
server connection and all the negotiations needed to transfer
the video streaming are that the connection request is a TCP
and the initialization is coming from the client (behind the
NAT) and the video streaming negotiations are done by the
RTSP that uses the active channel. However, the client could
not receive the video streaming data since the NAT dropped
the RTP/UDP video streaming packets. The client then sends
a teardown command to inform the server that the negotiation
is over. The client starts one additional negotiation tries before
it close the connection.

In the second experiment, we used the proposed client and
server modules. After running, the two modules start
monitoring the data link frames. The client monitors the
outgoing streaming request while the server monitors the
incoming streaming request.

When the client request a video streaming from the server,
The connection's establishment and the video streaming
negotiations between the client and the server are established
normally and the client started to display the video streaming
data as shown if figure (7A and 7B).

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

uient n is opened

RTSP Session identifier: 9. Connection arrived from 130.130.100.110:62488.]

RTSP Session identifier: 9. Received RTSP message from 150.150.100.110:62488. Message: "OPTIONS rtsp://15i

User- Agent: UbVLC, 1.1.2 (LIVE5S5 Streaming Media V 2010.03.16)

RTSP Session identifier: 9. Sent to client: 'RTSP, 1.0 200 OK

RTSP Session identifier: 9. Received RTSP message from 150.150.100.110:62488. Message: 'DESCRIBE rtsp:, ,, 1"

User- Agent: LibVLC/1.1.2 (LIVE 5 55 Streaming Media v20 10.03.16)

RTSP Session identifier: 9. Is the 'C:\redMUG\Contients\sample.mp4' MP4file in cache? 'true']

RTSP Session identifier: 0. Video track name to stream: track ID/ 2, audio track name to stream: track ID/ 1]

RTSP Session identifier: 0. Sent to client: "RTSP/ 1.0 200 OK

RTS P Session identifier: 0. Received RTSP message from 150.150.100.110:62488. Message: 'SETUP rtsp:// 150.1

User- Agent: UbVLC, 1.1.2 (LIVE 5 55 Streaming Media v2010.03.16)

RTSP Session identifier: 9. Sent to client: 'RTSP/1.0 200OK

Session: 247776867;tjmeout=360

RTSP Session identifier: 0. Received RTSP message from 150.150.100.110:62488. Message: 'SETUP rtsp:// 150.1

Session: 247776867

User- Agent: UbVLC: 1. 1.2 (LIVE555 Streaming Media v2010.03.16)

RTSP Session identifier: 0. Sent to client: "RTSP/ 1.0 200 OK

Session: 247776867;tjmeout=360

RTS P Session identifier: 0. Received RTSP message from 150.150.100.110:62488. Message: 'PLAY rtsp:// 150. 15

Session: 247776867

User- Agent: UbVLC, 1.1.2 (LIVE555 Streaming Media v2010.03.16)

RTSPSession identifier: 0. Called the seek method. Start time in sees: 0.000000.]

RTSP Session identifier: 0. Sent to client: 'RTSP/ 1.0 200 OK

Session: 247776867;timeout=360

Figure 7A. Connection still active when the data streaming are transforming
(server side)

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

Vol. 8, No. 6, September 2010
Internet gateway. Streaming might fail at times even if the
gateway has a built-in RTSP NAT.

Reference [4] utilizes the two streaming protocols
separately by using a third device (proxy) between the client
and server (every side with whole streaming protocol
advantages and disadvantages), the proposed method utilizes
the benefits of the RTP and HTTP protocols without using any
extra device.

'^r

fjf /. m

*-'r^-t

ft,

.■■

Figure 7B. Video streaming is displayed in the client (behind the

NAT)

When negotiation is started, the host records the connection
details: IP, TCP port and the streaming UDP ports. The host
will insert the fabricated TCP header (after the UDP header)
in the video streaming packet before sending it.

The reason for the success of transforming the streaming
data is that the sending host converts each streaming UDP
packet into a fabricated TCP packet that bypasses the NAT
because it uses the active channel. The receiving host in turn
restores the fabricated TCP packet into the UDP streaming
data at the data link layer before sending it to the upper layer.

V. Evaluation

A comparison between our proposed method and the
existing technologies is presented in Table 1. The proposed
method has several advantages over the existing technologies,
although the new packet size is 20 bytes larger than the
normal RTP/UDP packet, but less compared with the HTTP.
This has a little impact on the network performance.

The proposed method can traverse the video streaming over
all types of NAT. It can also traverse the firewall that blocks
the UDP ports that RTP may use, commonly with home

Table I. CURRENT AND PROPOSED METHOD COMPARISON

FEATURE

HTTP

RTP/TCP

RTP/UDP

PROPOSED
METHOD

Directional

Bidirectional

Uniary

Playback
hiccups

Yes

Quality
Reductions

Yes

Oscillating
throughput

Yes

Slow recovery

Yes

ISP firewall

Traverse

Blocked

Traverse

NAT traversal

Yes

End-to-End
Delay

Long

Short

Window buffer
and reordering

Yes

Streaming
method

Downloading
or progressive

Streaming

VI. Conclusion

The two main transport layer protocols: TCP and UDP can
be used in streaming but with the whole advantages and
disadvantages of using that protocol. In this paper, a new
method is presented and implemented that can merge some
advantages of both protocols. It enables the client and server
to use UDP advantages in each side for streaming. Both client
and server gains scalability by not having to deal with some
TCP processing feature (e.g. Acknowledgement and window
buffering ...etc). In the other hand, utilize the benefit of the
TCP advantages to traverse the NAT and the firewall. In other
words, the stream is not discarded and traverses the NAT and
the firewall. The experimental results show that the new
method achieves the firewall traversal and Nat traversal even
with the most difficult NAT (symmetric NAT).

References

[1] Chu-Hsing, L., et al., Energy Analysis of Multimedia Video Streaming
on Mobile Devices, in Proceedings of the 3rd International Conference
and Workshops on Advances in Information Security and Assurance.
2009, Springer- Verlag: Seoul, Korea.

[2] Matthew Syme, P.G., Optimizing Network Performance with Content
Switching: Server, Firewall and Cache Load Balancing. 1st ed. 2004:
Prentice Hall. 288.

[3] Philip, W.F., et al., Server- efficient high-definition media dissemination,
in Proceedings of the 18th international workshop on Network and
operating systems support for digital audio and video. 2009, ACM:
Williamsburg, VA, USA.

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

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

Vol. 8, No. 6, September 2010
[4] H, et al., Transparent protocol translation for streaming, in

Proceedings of the 15th international conference on Multimedia. 2007,

ACM: Augsburg, Germany.
[5] Sourour, M., B. Adel, and A. Tarek. Security Implications of Network

Address Translation on Intrusion Detection and Prevention Systems, in

Network and Service Security, 2009. N2S '09. International Conference

on. 2009.
[6] Sanmin, L., et al. TCPBridge: A software approach to establish direct

communications for NAT hosts, in Computer Systems and Applications,

2008. AICCSA2008. IEEE/ACS International Conference on. 2008.
[7] Yeryomin, Y., F. Evers, and J. Seitz. Solving the firewall and NAT

traversal issues for SIP -based VoIP, in Telecommunications, 2008.

ICT 2008. International Conference on. 2008.
[8] Takabatake, T. A Scheme of Relay Server Selection Methods for NAT

Traversal through Home Gateways, in Telecommunication Networks

and Applications Conference, 2008. ATNAC 2008. Australasian. 2008.
[9] P.Srisurensh, K.E., Traditional IP Network Address Translator

(Traditional NAT). RFC 3022, 2001. IETF.
[10] Khlifi, H., J.C. Gregoire, and J. Phillips, VoIP and NAT /firewalls:

issues, traversal techniques, and a real-world solution.

Communications Magazine, IEEE, 2006. 44(7): p. 93-99.
[11] Ivanov, I., API Hooking Revealed. 2002.
[12] Hasegawa, T. and T. Ogishi. A framework on gigabit rate packet

header collection for low-cost Internet monitoring system, in

Communications, 2002. ICC 2002. IEEE International Conference on.

2002.
[13] Jelger, C.S. and J.M.H. Elmirghani. Performance of a slotted MAC

protocol for WDM metropolitan access ring networks under self-
similar traffic, in Communications, 2002. ICC 2002. IEEE

International Conference on. 2002.

AUTHORS PROFILE

Omar A. Ibraheem (PhD) is currently a
Post Doctoral Research Fellow in National
Advanced IPv6 Centre of Excellence
(NAv6) at Universiti Sains Malaysia
(USM). Dr. Omar obtained his bachelor,
master, and doctorate in computer science
from Mosul University in 1998, 2000, and
2006 respectively. He has joined NAv6
since January 20 10. Before that, Dr. Omar
was a senior lecturer at the computer
science department, College of Computer
Science and Mathematics of Mosul
University in Iraq. His research area include
the Network protocols, Routing protocols,
Network security and Multimedia
communications.

Omer Amer Abouabdallah (PhD) is a

senior lecturer and post graduate
coordinator of the National Advanced IPv6
Centre of Excellence (NAV6) in Universiti
Sains Malaysia. Dr. Omar obtained his
Bachelor of Computer Science from Al-
Fateh University, Tripoli, Libya in 1993.
He obtained his Master of Computer
Science and doctorate from Universiti Sains
Malaysia in 1999 and 2004 respectively.
Dr. Omar is heavily involved in researches
carried by NAv6 centre, such as IPv6 over
Fiber Object and the Multimedia
Conferencing System. The highlights of Dr.
Omar's achievements include the winner of
Sanggar Sanjung Award 2005 and 2006 by
USM on 2007 and the winner of Innovative
Product Award for NAT AND FIREWALL
TRAVERSAL SOLUTION for MSCv6 as
well as the gold medal winner in
International Invention Innovation
Industrial Design & Technology Exhibition
2006 (ITEX2006).

Sureswaran Ramadass (PhD) is a

Professor and the Director of the National
Advanced IPv6 Centre (NAv6) at
Universiti Sains Malaysia (USM). He is
also the founder of Mlabs Systems Berhad
(MLABS), a public listed company on the
MESDAQ. Prof Dr Sureswaran obtained
his BsEE/CE (Magna Cum Laude) and
Masters in Electrical and Computer
Engineering from the University of Miami
in 1987 and 1990 respectively. He
obtained his doctorate from USM in 2000
while serving as a full time faculty in the
School of Computer Sciences. His
research areas include the Multimedia
Conferencing System, Distributed Systems
and Network Entities, Real Time
Enterprise Network Monitoring, Real
Time Enterprise System Security, Satellite
and Wireless Networks, IPv6 Research,
Development and Consultancy, and
Digital Library Systems.

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

(IJCSIS) InternationalJournal of Computer Science and Information Security,

Vol 8, No. 6, September 2010

The Integration of GPS Navigator Device with
Vehicle Tracking System for Rental Car Firms

Omarah Omar Alharaki

Faculty of Information and

Communication Technology,

International Islamic University

Malaysia,

Kuala Lumpur, Malaysia.

Fahad Saleh Alaieri

Faculty of Information and

Communication Technology,

International Islamic University

Malaysia,

Kuala Lumpur, Malaysia.

Akram M. Zeki

Faculty of Information and

Communication Technology,

International Islamic University

Malaysia,

Kuala Lumpur, Malaysia. .

Abstract — the aim of this research is to integrate the GPS
tracking system (tracking device and web-based application)
with GPS navigator for rental cars, allowing the company to
use various applications to monitor and manage the cars.
This is enable the firms and customers to communicate with
each other via the GPS navigator. The system should be
developed by applying new features in GPS tracking
application devices in vehicles. This paper also proposes new
features that can be applied to the GPS Navigator. It also
shows the benefits that the customers and staff will get from
this system.

Keywords', GPS tracking system, GPS devices, GPRS, Garmin.

I. Introduction

The Global Positioning System (GPS) is a satellite-
based navigation system made up of a network of 24
satellites to specify a position on the surface of the earth
[11] [4]. It also provides highly accurate location with the
use of special GPS receivers and their augmentations
[18]. In 1973 GPS was intended for USA military
applications [7], but in the 1980s, the government made
the system available for civilian use. GPS works in any
weather conditions, anywhere in the world, 24 hours a
day. There are no subscription fees or setup charges to
use GPS [11].

From NASA (2009) : The uses of GPS have extended
to include both the commercial and scientific worlds.
Commercially, GPS is used as a navigation and
positioning tool in airplanes, boats, cars, and for almost
all outdoor recreational activities such as hiking, fishing,
and kayaking [7]. GPS is also playing an increasing role
in the tracking of motor vehicles [2].

General Packet Radio Service (GPRS) is an
enhancement of GSM networks to support packet
switched data services such as email and web browser in
addition to existing GSM data services such as Short
Message Service (SMS). GPRS operates on the existing
GSM network infrastructure that it utilizes available time
slots during each frame transmission. Thus, it does not

overload the existing GSM network traffic and can
efficiently provide data services. The GPRS can transfer
data at the maximum rate of 115.2 kbps. Due to a very
large coverage area of GSM networks around the world,
GPRS becomes the largest data service network available
and always-on; thus, it is most suitable for a real-time
tracking management system. [1].

GPS tracking system developed that transmit
vehicle's data in real time via cellular or satellite
networks to a remote computer or data centre [15] [10].
Vehicle tracking system signifies the monitoring and
management of vehicle, trucks, etc by using GPS system
that can get in real time the current location, situation ,
history, performance and emissions and control them. [1]
[16].

The web based tracking system allows users to
securely log in and track their cars in real- time over the
Internet. The user sees moving dots on a map in a web
browser. It also display all transmitted information to the
users along with displaying location of vehicle on a map
[15].

The tracking system allows users to locate any car at
any time of day. Also, it can replay a past trace of the cars
history. It can remotely control the car such as run alarms
and locking devices. It enable the users to keep track of
the vehicles without the intervention of the driver where,
as navigation system helps the driver to reach the
destination [12].

Many shipment companies in the world use GPS
tracking systems in their trucks. It is very important for
the fleets, especially when they have big number of
trucks and staffs, to manage this huge number of vehicles
and people. In addition, The shipments and conveying the
tracking information to customers are perceived to be
important customer service components and they are
often considered industry norms rather than a potential
competitive advantage for shipment service providers
[14].

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

(IJCSIS) InternationalJournal of Computer Science and Information Security,

Vol 8, No. 6, September 2010

However, the biggest challenge in the cars rental
companies are car thieves and delay car return. This will
cause, lost a lot of money and cars. However, the rental
cars firms tried to find the solution to save their cars from
these troubles from their clients and staff.

The vehicle tracking system presented in this paper as
a system that is designed to track and manage rental cars
that are used by the customer, using a GPS tracking
technology with GPS navigator.

The system comprises of vehicle GPS tracking
devices (Tracking device and GPS navigator), GPRS and
a web-based application. Through this system, the
company staff will have the facility to monitor the
movement and relevant information of each vehicle.
Moreover, to keep in touch with their customers by GPS
navigator technology [13].

The paper highlights how to apply new application in
GPS navigator that allows the firm staff and customers to
communicate between each other via GPS navigator in
the rental cars. Moreover, satisfying customer needs and
building high confidence between the firm and
customers.

This paper illustrates the integration of multiple
technologies to achieve a common goal. It shows how
such technologies can be synergistically combined to
address real rental cars firm problems [13].

II. GPS TRACKING SYSTEM

GPS device is a small unit that receive signals from
satellites and send other signals to antennas (GPRS). This
device is a major part of the system and it will be
installed into the vehicle which is responsible for
capturing the following information for the vehicle such
as the Current location of vehicle, Speed of vehicle, Door
open/close status, Ignition on/off status, etc. [15].

This device is also responsible for transmitting this
information to the Tracking Server located anywhere in
the world. Also, it has to install the unit in a hidden and
safe place inside the vehicle [15].

The information about the vehicle saved in this unit
will be sent to antennas by GPRS, and there are many
applications (Figure 1 shows one of the application)
connected to the internet that can calculate it and put it on
the map to integrate with it.

(Figure 1 : web based tracking application)

The information includes location, speed, fuel level,
engine situation, start driving point, and car situation.
Also via this application, the company staff can control
the cars, for example, on/off lights, on/off air
conditioning, on/off car engine, on/off security system
and others, by sending some codes to the tracker in the
car.

A. GPS Tracking System Framework
The GPS tracking system consists of client-server
architecture where the web browser is the client and the
server functions are shared between a web server, a
communication server, a database server and a map
server [13]. The process of web application tracking
consists of four parts: a location-aware web system,
location determination, location-dependent content query
and personalized presentation. The location-aware web
system is the underlying infrastructure. It allows the
exchange of the location information between the web
browser and the web server, and the automatic update of
the web pages in the web browser when the location
changes [17].

GPRS is the main method of communication between
the tracking device and the web server. GPRS, being a
2.5G mobile technology, is available all over the world. It
is also ideally suitable for data transfer over an always
on-line connection between a central location and mobile
devices. The cost is per kilobyte of data transferred, in
comparison to SMS where the cost is per message. [13].

The location information collected through the GPS in
real time is placed in a central database that is owned by
the firm's staff via GPRS antenna. Each user of the
system may access this information via the Internet [15].
Figure 2 shows the system framework.

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

(IJCSIS) InternationalJournal of Computer Science and Information Security,

Vol 8, No. 6, September 2010

(Figure 2 : GPS tracking system framework)

B. Features of GPS Tracking System:

Some rental car firms offer GPS devices for rent when a
customer comes to rent a car by adding charges for it per
day. This is beneficial for customers to locate their rental
cars, know the location direction and their current
location. Moreover, it also helps the rental car firms to be
informed how the customers are using the cars.

This system is also used to prevent theft and retrieve
stolen/lost vehicles. The signal sent out by the installed
device help the rental car firms to track the vehicle. These
tracking systems can be used as an alternative for
traditional car alarms or in combination of it. Installing
tracking systems can thus bring down the insurance costs
for these vehicles by reducing the risk factor [10].

Moreover, this system gives benefits to the rental car
firms and to their customers such as: [9].

• Real-time location of vehicles,

• Historical vehicle reports,

• Security code/pin,

• Trip computer,

• Engine idle and start/stop reporting,

• Real-time tracking alerts and reports after hour vehicle
monitoring,

• Create custom Geo-fences and landmarks,

• Historical movement of vehicle,

• Mileage reporting,

• Optional starter disable/enable,

• Notification of doors opening and closing

C. Case Study:

Some fleet firms that applied this tracking system which
are the following:

Delhi Transport Corporation is the one of the largest City
Road Transport Undertaking in India [3]. It has a fleet of
around 15,000 vehicles carrying on the business of
passenger transport in 800 routes from 33 depots all over
the state of Delhi with a product mix comprising of City
and Inter-city services. After they implemented the
tracking system in vehicles and other functions under
Automatic Fleet Management System, the benefits they
obtained are [3]:

• Better bus scheduling.

• Quick replacement in case of breakdown/accident en-
route.

• Effective control over the drivers & checking bus-
stops skipping.

• Check on over-speeding.

• Basic communication between driver and control
room in emergencies.

• Automation of Fleet Operations minimizing human
intervention.

• Improved Fleet utilization leading to better services
and thereby enhancing commuter satisfaction.

Another car rental Firm is in the United Arab Emirates
(UAE) that fitted their cars with high-tech C track
GPS/GPRS satellite tracking units to prevent thieves from
driving away with the cars [8].

Thieves in Dubai and Sharjah are increasingly posing
as clients who rent a car and then ship the car out of the
country, mostly to Russia, North Africa and Eastern
Europe. This prompted car rental firms to integrate
vehicle with GPS tracking system to protect their
business and save themselves time and money. Al
Mumtaz Rent-a-Car in Dubai gives their vehicles some
level of safety and it can monitor their vehicle all the
time. [8].

The system is cost-effective, as it saves time and
money in the recovering of stolen vehicles. According to
Diamond Lease, they can also monitor the movement of
all their vehicles and can thus establish whether a vehicle
is being misused. The GPS tracking device can keep track
of the car's engine condition by recording harsh braking,
speeding and even the removal of any of the car's parts,
thus saving us more money [8].

In Saudi Arabia from rental cars company report they
have more than 300 cars and their experience more than
25 years in this field. Moreover, they have more than 10
branches in East of Saudi Arabia. However, during this
time the rental cars faced many troubles from either
customers or employees such as car lost, personal car use
from the staff and delays car return. From these problems
the rental cars lose money, lose cars, low service quality,
and short cars use period, [one of authors experience]

On the other hand, after they installed the tracking
system into the rental cars they save almost all of their
money, time and cars. Moreover, the firm got
improvement of their services quality and maintained the
cars quality for a longer period.

III. Proposed method and Recommendation

This paper shows that this system is easy to apply,
very important to manage, save and control the cars. In

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

(IJCSIS) InternationalJournal of Computer Science and Information Security,

Vol 8, No. 6, September 2010

this section, the paper will propose some helpful and
useful features in the GPS system.

There are some new features recommended to be
implement in the system that will save Firm time,
money, and manage their cars safely which are as
following (Figure 3):

Integrate GPS navigator device with
tracking system

i GPS

Create new application system for GPS
navigator

Connect car electricity with GPS
navigator device

Connect GPS navigator device with web based
application

Figure 3 : The process of applying GPS navigator system

1) Integrate GPS navigator device with GPS tracking
system

This device usually work as a guide for travelers to show
them where they are , where they want to go , maps ,
roads, shops and other useful information .
The integration framework is between GPS navigator
device and GPS tracking system as shown in figure 4.

(Figure 4: framework for the integration of GPS devices)

2) Create new application system for GPS navigator
that can link customers with rental car company via GPS
tracking device. This system will contain new menus that
provide new services for customers in addition to its main
function as guidance. The proposed new menu can help
the customers to contact with staff for any help such as
lost directions, lost key or any other assistants as shown
in figure 5.

LOST
KEY

LOST
DIRECTION

I NEED
HELP

(Figure 5: Proposed new menu in GPS nav. Application)

GPS navigator device must be installed in the cars
because it will link with the GPS tracking device.

3) Connect car electricity with GPS navigator device
that controls the cars, so that nobody can drive the car
without logging in using the specific username and
password. When the customer wants to drive the car after
he/she has signed the contract, he/she has to enter the
name that is in the contract and the contract number as
shown in figure 6 and figure 7.

J Enter Your Name please

< >

► ^

A
H

B C

■ 1 '

D E F f G

P Q R S T U

x y z „ a

Back Mode 12 3 Done

(Figure 6: field for entering customer name)

jEnter your contract number 1

1 1 2 1 3

5 I 6

{

)

8 I 9

■_i

(Figure 7: field for entering contract number)

4) connect GPS navigator device with web based
application through GPS tracking device that can send
messages from the company to the customer, such as,
reminders of expiry date of contract, last promotions, and
other useful information (as shown in figure 8 & 9) .

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

(IJCSIS) InternationalJournal of Computer Science and Information Security,

Vol 8, No. 6, September 2010

Dear Mr./Ms. "Customer Name"
Your contract has almost expired, you are
required to return the car within 5 hours or if
you wish to extend the contract kindly contact
us immediately.

Phone

Volume Tools

(Figure 8: one of the proposed reminder messages from
the rental car firm to their customers)

Dear Mr./Ms. "Customer Name"

There is construction in

"Street Name" and "Street Name"

For your safety, kindly avoid driving there at

this time.

®O0

Phone Volume

Tools

(Figure 9: one of the proposed notice messages from the
rental car firm to their customers)

IV. Conclusion
In conclusion, this paper presents the development of
a Rental cars tracking system using GPS & GPRS
technologies. It is a typical example of how the
advantages may be forced for the efficient and effective
managing of rental cars firm. However, after
implementation of this proposed system may give
benefits for the rental cars firm customers such as built in
road maps, Route capability Touch screen access,
monitor fuel consumption, vehicle maintenance alerts,
route guidance, Speed limit display [6], last promotions
from the firm, warning reminders, renewal car rental
contract, and keep contact with the rental cars firm.
Moreover, this system show that the customers will
connected with the rental cars firm in whole day via GPS
navigator device with full protection of the rental cars
firms from any violations from the customers.

Acknowledgment

We would like to acknowledge our mentor, Dr. Akram
zeki for his patience, guidance, and knowledge in order to
reach our goal. Also, we would like to convey our
gratitude to our parents who support us to understand the
importance of knowledge and show us the best way to
achieve it.

References

[1] Chadil, N.; Russameesawang, A.; Keeratiwintakorn, P. (2008),
"Real-time tracking management system using GPS, GPRS and
Google earth", proceeding of ECTICON.2008, pp.393 - 396.

[2] Craig A. Scott (1994), "Improved GPS Positioning for Motor
Vehicles Through Map, Matching", University of

Technology, Sydney, Presented at ION-94, Salt Palace
Convention Center, Salt Lake City.

[3] DELHI Transportation Corporation case study (2007),

http : //www. cmcltd. com/casestudies/transportation/GPSSy stem
_Case_Study_DTC.pdf

[4] General Information Document of GPS-based Fleet Management

and Tracking Systems (2008), Exaterra Inc., Ottawa, Canada:

Canadian company based in Ottawa.
[5] Hariharan, Krumm, Horvitz (2005), "Web-Enhanced GPS",

School of Information and Computer Sciences, University of

California, Irvine, USA.

[6] http : //gpstrackit. com/gps-tracking-products/garmin-integration

[7] http://scign.jpl.nasa.gov/learn/gpsl .htm

[8] http://www.digicore.com/ , (UAE Press Release August,2007)

[9] http://www.imardainc.com/smarttrack-vehicle-tracking-system

[10] http://www.roseindia.net/technology/vehicle-
tracking/VehicleTrackingSy stems, shtml

[11] http : //www8 .garmin. com/aboutGPS

[12] http://www.gisdevelopment.net/technology/gps/techgp0044.htm

[13] M. Medagama, D. Gamage, L. Wijesinghe, N. Leelaratna, I.

Karunaratne and D. Dias (2008), GIS/GPS/GPRS and Web based
Framework for Vehicle Fleet Tracking, ENGINEER , No. 05, pp.
28-33.

[14] Mikko Ka'rkka'inen, Timo Ala-Risku, Kary Fra'mling (2004),
"Efficient tracking for short-term multi-company networks",
International Journal of Physical Distribution & Logistics
Management Vol. 34 No. 7, pp. 545-564

[15] Muruganandham , P.R.Mukesh (2010), " Real Time Web based
Vehicle Tracking using GPS", World Academy of Science,
Engineering and Technology 61 2010.

[16] Robert W. Bogue (2004), "New on-vehicle performance and

emission monitoring system", Sensor Review, Volume 24, No. 4,
pp. 358-360

[17] Rui Zhou (2008), "Enable web-based tracking and guiding by
integrating location-awareness with the world wide web",
Campus-Wide Information Systems Vol. 25 No. 5, pp. 311-328

[18] S.S.S. Prakash, MadhavN. Kulkarni (2003), " Fleet

Management: A GPS-GIS integrated approach", Department of
Civil Engineering, IIT Bombay, Mumbai, GISdevelopment.net.

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

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

Vol. 8, No. 6, 2010

Process Framework in Global extreme Programming

Ridi Ferdiana, Lukito Edi Nugroho, Paulus Insap

Santoso

Department of Electrical Engineering and Information

Technology

Gadjah Mada University (UGM)

Yogyakarta, Indonesia

Ahmad Ashari

Department of Computer Science and Electronics

Gadjah Mada University (UGM)

Yogyakarta, Indonesia .

Abstract — Software development life cycle works as a process
framework that underlying of the software engineering
framework. In multi-site software development, the impact of not
having a process framework is often quite disastrous. The multi-
site team can work anonymous and have no guidance to work
and collaborate with others. Therefore, several researches have
begun to introduce the important of the process framework
through a legitimate software development life cycle. The most
common process framework that introduced in multi-site
software development is called as a Global Software Development
process framework (GSD). However, many GSD
implementations are reported as an enterprise process
framework that highly iterative and model driven oriented. This
paper will show an alternative way to modify the existing GSD
SDLC into the simplified process framework by integrating the
process framework with an agile method like extreme
Programming. By simplifying the process framework, it will
provide an opportunity for small and medium enterprise to adopt
the proposed SDLC in multi-site development.

Keywords-process fram ework; software

lifecycle; agile; extreme Programming

development

Introduction

Global Software development (GSD) is defined as a
software development process that uses teams from multiple
geographic locations. The physical distant between team is the
key issue in GSD, The team may be from the same
organization, collaboration that involves different organization,
outsourcing to the other organization. The dispersed can
happen in same country or other sides of the world. Developing
software in the distant introduces complex and interesting
issues.

Further research tells us that distributed is done in order to
fulfill well-establish motivators include [3].

• Limited trained workforce in technologies that are
required to build today's complex systems.

• Differences in development costs favoring dispersing
team geographically.

• A "round-clock" work system facilitated by time zone
differentials allowing for shorter times to market.

• Advanced infrastructures (internet bandwidth, and
collaboration software).

• A desire to be close to a local market.

Although the motivations behind GSD often differ from the
other, the main problem is still constant. It is naturally more
difficult to coordinate projects where teams separated by
physical distance. While in collocated software development
there are generally a best practices standard, artifact, and ad-
hoc coordination to do "real-time" software development. It
does not really exist in distributed software development.

Best practices, artifacts, or ad-hoc communication are done
to supply sufficient "shared vision" between the team, and also
client. Shared vision is creating a common understanding about
what the team is trying to do, what the finished product look
like, what the basis of the product, and when they must deliver
the product if it is to have its intended effect [8]. Either the
system is developed in collocated or distributed; the shared
vision its necessity to limit the development risk.

Shared vision in distributed development faces some
challenges and issues. The first matter is commonality. It is
uncommon to have multiple divisions, organizations, cultures,
and languages on a project. Often the team members have not
known each other, may have different level experiences, and
may have motivation conflict. All these factor plots to make it
increasingly difficult to coordinate across teams, manage
evolution, and monitor progress. Past studies have shown that
tasks take about 2.5 times longer that the collocated one [6].

Sangwan et al. [11] creates a set of practices that can be
used on projects that are geographically distributed. The
research identified a crucial factor to the success of GSD
projects and construct best practices for these factors to enable
a successful outcome. The outcome of the research is a process
framework that is leveraged the best practices in the critical
success factors and is based on the agility model. However, the
research still argued about agile implementation as a tradeoff of
its discipline process [2].

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

II. GSD Process Framework

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

Vol. 8, No. 6, 2010
approaches will be

Process framework in GSD is divided in four major phases,
which are requirements engineering, project planning,
architecture design, and product development. Those phases
are known as a software development life cycle (SDLC). GSD
SDLC provides wide range phases to initiate the project
(planning and requirement) and to construct the product
(architecture design and product development). Along with the
process framework, GSD also provides organization process,
and monitoring control as parts in its process framework. This
section will discuss the concept in separated sub sections.
Figure 1 shows the GSD process framework.

Mdr*.e1 Intflflt'PwlyCl Vi90*l—

Process Framework

" Goals

RaouirfimflTils Engirwflnng

Ptafici Planning

ArchitaCLiff Design

- ar>d Module -
3£tCih£itont

F+oducL Ob wtopnsm

Figure 1. GSD process framework [11]

Each process can be threatened as a process component that
has several input and output. Although it shows the sequential
one, several implementation of the process framework is
described as highly iterative and incremental. The next section
will discuss the concept in separated sub sections.

A GSD requirements engineering

The major concern in requirement engineering is doing
analysis of the problem, and creating a system based on the
available solution. It can be composed as functional
requirements and non-functional requirements. Requirements
answer questions about what should be implemented, how the
system should behave, what system attributes and constraints.

Requirement engineering process is started by finding
business goals and market intent for the product. The product
should have a reasonable definition to build, the benefit for the
system, and have clear target markets. Without reasonable
definition, it can be difficult to define concretely the users,
overlooked the product, and create an assumption of the
product.

Sangwan, et al. [11] proposes the requirement engineering
process through three main activities; which are elicitation,
modeling, and review. Each activity provides output and
process one or several input.

The elicitation activity is to extract the domain knowledge
from the experts, identify features that are in scope and out of
scope, and document the features in the highly structured and
logical model. In GSD, that activity can be done through
several approaches starting from face-to face meeting (it's
meaning the team should plan to travel), using CSCW tools
like instant messaging, or phone conferences, or indirectly by

using email communication. Those
discussed further in case studies section.

The primary output of the elicitation activity is a feature-
level requirements model. Feature level requirements model
focuses in creating a high-level structure of the product feature
without detailing and completing the model itself. The creation
step to provide the high-level structure by doing function
decomposition and categorize the feature through work based
structure model. Through this way the GSD team will get start
with easy and common understanding through a simple model
like Unified Modeling Language.

The model itself provides indirect control between project
planning and the design model. However, the requirement
model can sometimes become complex, difficult to understand,
and have been conflicting with requirements. Therefore, the
review process is important to validate the project plan,
adjusting the design, and following the software process. Tools
are playing some rules to provide several syntactic reviews
(e.g. model completeness, violation in rules, etc.). The semantic
syntactic should be manually reviewed. In GSD, these step
called requirement review meeting. These meeting have three
goals, which are; ensuring the requirements were correctly
understood and translated into the model, verifying the
requirement model itself is readable and understandable, and
disseminating the requirements itself.

B. GSD Architecture Designs

The drivers for software architecture in GSD project are
shared vision. When the distributed teams have shared vision
about the architectures, the development and design will shape
by itself. The concerns are many people intuitively know how
to create shared vision, but make the intuition explicit about the
remote teams is challenging. Shared vision come when the
team has same domain knowledge, and is able to design the
architecture in well-defined and loosely coupled components.

Define work units is the first step in GSD architecture
design to decompose the whole system into functional units
that can be allocated distributed in development. The drivers
for how to decompose the functional units come from several
considerations such as previous experiences, architecturally
significant requirements, referenced patterns, organization
information, and resource location.

The second step is to identify module responsibility. The
functional unit results are combined together with the
requirement models. The result of this step is module
specification, which is described the detail function of each
module. As a structural result, each module will be categorized
into two main categories, which are static and dynamic
modules. Static modules are defined as a module that works
and dedicated for common purposes such as security, print,
utility, compression, and any modules that work loosely
coupled with the business process. Dynamic modules are
defined as a collection of modules that work tight coupled with
the business process, e.g. payment processing, inventory
checking modules, etc. Both module categories are displayed in
a relation through an UML interaction diagrams.

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

The third steps in architecture development are creating the
architecture documents. Architecture document works as a
reference guide in development, not as a manual to build the
system. Clements et al. [4] provides a guidance to build an
architecture document in a sufficient way. According to the
research, an architecture document consists of two main views,
which are; execution views and implementation views.

C. GSD project planning

Project planning in GSD is focused in allocating the amount
work, budget and time in distributed manners. Project planning
in GSD usually uses time-boxed development to approach
synchronous development among the various geographically
distributed development sites. Requirements and architecture
results are the main input for this process. Cost estimation and
project plan is the main output in project planning.

Feature release planning is the first phases in project
planning. It is designed from market intent and functional
specification. The feature release planning is an iterative
activity that can be revised in time constraints depend on the
market intent and requirements changes. Feature release
planning is dividing the features of the system through time
iteration. Iteration may vary from two weeks or months. The
goal of this activity is to identify the minimum set of features
that can be sold in the marketplace.

Development planning is proposed by mixing feature
release plan, design model, and critical path analysis. The result
of the development planning is a project plan. Project plan
consists of schedule planning and integration-test planning.

Schedule planning provides a fix date for features. It is
driven by the critical path analysis and design model that
proposed in architecture sessions. Thus, each development
team must release an operational version to integrate and tested
on the date specified. If the team can make the date, the release
would be skipped, and the team would release on the next fixed
date on the next sprint.

Integration and test planning provides a detail when the
production can hit the date. It is driven by feature release plan
and schedule planning. Iterations will be followed by test and
integration planning. The result from this step is a detail works
of schedule called as a project plan.

Project plan can be a baseline for the team to make budget
proposals. It will become to go or not go to decision. Although
it becomes a trade off in the budget versus feature. The project
plan can be a live artifact for future reference development.
Therefore, GSD proposes three phases in project planning,
which are planning in inception phase (requirement phase),
planning in elaboration phase (architecture design phase), and
planning in construction phase (product development).

D. GSD product development

GSD product development is preliminary started by
structuring the team. GSD proposes a hierarchical team
structure. It contains a central team, and site team. GSD
provides a basic composition of the central team and team site.
Basic composition provides the basic structure of the team.
There are two communication models between central team

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

Vol. 8, No. 6, 2010
and site team, which are vertical communication and horizontal
communication. Horizontal communication is a
communication model that exists between team members in
one site. Vertical communication happens between two
different teams (e.g. central team with the site team) through a
proxy communication model. Supplier manager is
representative of central team, and R&D manager is
representative of site team. Table 1 provides the casual team
role in GSD.

TABLE I.

GSD Team Strucures

Role

Responsibility

Product manager

create release planning and decide what feature
will be released in a milestone

Program manager

plans the entire lifecycle of a product, managing
the requirements, and partitioning component
development in team site

Supplier manager

Plans the entire components and manages a remote
component development team.

Subject matter expert

Provides expertise on the business that the project
is to support. Requirements engineer captures
customer requests, develop the analysis model and
glossary, and translates analysis model to the
requirements document

Requirements engineer

Captures customer requests, develop the analysis
model and glossary, and converts analysis model
to the requirements document.

Architect

Expresses, designates, and communicates
architecture solution, implements an architectural
prototype, develops specification, and acceptance
tests for product components.

QA Experts

Provisions the automated build/test, test
management, and defect tracking systems used by
the architecture project team.

R&D Manager

Plans the entire development lifecycle and manage
the module development plans.

Designer

Focuses to build user interface and designs
modules using appropriate design patterns.

Developer

Gears the module and increment the quality of the
codes.

The communication models exist to solve coordination
between team and inside the team. Solving a problem in GSD
need to know who solve the problem, adjust the schedule, and
shift task to the other's team. It works using traditional or the
internet based communication tools. In line with engineering
and planning activities, preparing an adequate project
infrastructure for distributed development is an essential
project success factor.

Infrastructure supports in GSD should support accessibility,
collaboration, concurrency, processes, awareness, and
integration. Accessibility is a characteristic that the tools must
be available at all development sites. Collaboration is
characteristic that the tools must support both real-time
collaboration and asynchronous collaboration. Concurrency
refers to the extent to which different parties can collaborate
explicitly or implicitly on a single artifact. The process
Characteristics focuses in a specific support tool for a software
engineering process (e.g. Rational Rose for RUP). Awareness
could, for example, be supported by making as much
information as possible available in a single location and
linking different artifacts for navigability. The last

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

characteristic which is integration is a nice to have that tools
support the integration between different tools through real-
time integration or file based integration. Conventional tools
may not address all the characteristics; therefore, selecting right
tools is the important factor in GSD. Table II provides previous
researches in the GSD researches.

TABLE II.

GSD Tools Related Researches

Authors

Research Topic

Fitzpatrick et al., 2006

Discussing the tools and its implementation for
GSD management process through notification
and chat.

Pillati et al., 2006

Discussing about software configuration
management tools in GSD.

Bass, 2006

Explains monitoring technique for GSD process
by using shared mental models approach

Taylor et al., 2006

Discussing the implementation opportunity to
adopt Agile in GSD through open collaborative
tools.

Thissen et al., 2007

Discussing several alternatives for GSD
communication tools such as email, web, and
webcasts.

GSD Product development is the last step of GSD process.
After this step, product should be delivered to the customer.
Product in GSD process is defined as software and its artifacts
(e.g. document, user manual, installation guide).

III. GSD Process Framework Current issues

The phase in GSD is over when the product delivered and
passed the user acceptance test. Although GSD provides a
comprehensive process, it does still require software
engineering methodology like RUP to provide technical how-
to. GSD also needs specific tools to make the process become
more effective, some research aware of provide tools like
notification chat, web email, open collaborative tools, and
software configuration models. Joining GSD with effective
software development methodology and sufficient tools
absolutely give advantages in the overall software project.

Effective software development is defined as a construction
method that focused to deliver the right software products in
the right time with the right tools. Developing nowadays
software faces challenges that are not those much different
from building medieval war ships. There is the "unknown"
factor, since many technical problems cannot be understood in
their entirety at beginning of the project.

Stober and Hansmann [12] identified those challenges such
as complexity of the infrastructure on which IT solutions are
built. The challenges arising from international and distributed
teams, not yet identify dependencies, exploding costs, and
rapidly approaching deadlines, requirements that come up late
in the game of decision, unexpected issues to mitigate, and the
required knowledge is not sufficient and to make it worse the
need for cost-effectiveness is going as a market competition.
How do we make software development become effective in
GSD, while the team is separated geographically?

Requirements are the key of the effective software
development [1]. Through the stable requirements, it is easy to

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

Vol. 8, No. 6, 2010
define distinct milestone, development phases, and acceptances
closing phases, although it separates on distant. Unfortunately,
requirement changes are usually happened even in the most
precisely elaborated software development planning.
Therefore, it is important to prepare the team to focus on the
ability to adapt quickly to face any kind of forthcoming
challenges. Involving the development team as stakeholders,
rather than suppliers is a great way to establish an open project
structure which responsibility and changes control is shared
[12].

Since the key of effective software development is
functional software, Agile promises a better way to developing
software with a lightweight and adaptive process. Agile
provides manifestos and twelve principles that can be seen
online at http://www.agilemanifesto.org/. Those principles
focus in delivering working software through individual and
interactions, customer collaboration and responding changes.
Agile process implemented through several methods like
extreme Programming (XP), Scrum, Dynamic Software
Development Method (DSDM), Agile modeling, Crystal, Lean,
etc. Every method has own unique approaches to build in an
agile way.

Relating the agile and GSD is somewhat contradiction
Agile needs intensive communication both within the customer
and the team, but GSD make a distribution of the team that also
contributes to the team dysfunction. Communication between
peers becomes important to the team's collective
understanding. Remote team members miss these, and
consequently, their understanding suffers. Miller [9] shows that
when the direct communication doesn't exist in their practices.
It will weaken the adopted method. Taylor, et al [13] shows the
agile adoption in GSD is just like reinventing the wheel, since
many of the Agile GSD experience reports are not giving any
additional value in the existing GSD guidance. They also
recommend the researcher to create a value or a framework to
integrate Agile in GSD context.

Although it has a contradiction in the term of conditions,
several researchers also provide field reports about successful
agile adoption in GSD. Miller [9] confidently said that his team
at Microsoft delivered sufficient software by integrating Scrum,
XP, and GSD. Hazzan and Dubinsky [5] states the diversity
that exists in GSD is naturally supported by agile software
development. Paasivaara et al [10] captures the Scrum practices
that successfully adopted in three GSD projects. The following
section will discuss how extreme Programming method is
integrated with the current GSD framework to provide a
process framework that the research called as the Global
extreme Programming process framework.

IV. Research methodology

In order to integrate between the GSD process and XP
methodology, the research did case studies that based on the
real project. The organization that we are selected is the small
independent software vendor that contains five development
members with two clients who dedicated to support the project.
The project length is six months and works to develop project
management that running on the web.

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

The team itself has two years average of experience in
multi-site development. The client and the team are separated
in the different countries with different time zone. The team
composition as like follows.

(IJCSIS) International Journal of Computer Science and Information Security,
Vol 8, No. 6, 2010

TABLE III.

Team Structures

Roles

Job Descriptions

Customer

Drive the product direction to the team, answer the
developer questions, write stories, declare stories complete,
write acceptance test, accept the release

Coach

A thorough understanding of the XP process, professional
development experience, leadership experience, get the
developers and testers to adhere to the values and practices,
assume a leadership role for the development and testing
teams, pair with other developers and testers as needed,
assist customers in the story writing process, assist
customers in the acceptance test writing process

Developer

Estimate stories, brainstorm tasks, develop unit tests,
develop the tasks, refactor the code, and communicate with
customers when questions arise. There are two developers
in this case study

Client Subject
matter expert

Help the customers define intelligent stories, act as a proxy
for the customer.

Tester

Ensure that a story is testable, assist the customers with
writing acceptance tests, run the acceptance tests

The Tracker

Collect development metrics, produce reports indicating
the team's progress, communicate the team's historical
velocity, and communicate the status of the team's
progress

The research will collect the data by exploring the case
study related artifacts in the proposed workflow. The research
procedures are come as follows.

• The team is incubating and does workshops to
implement the proposed GXP framework process.

• Based on the theory, the team designs the artifacts.

• The evaluator evaluates the execution and captures
several data that can be written as finding of the case
study research.

Based on the project execution the GXP process framework
finding is discussed the next sections.

V. GXP Process Framework

Understanding the GXP process framework understands the
hybrid approaches that integrated between conventional GSD
process and XP method. As we know, conventional GSD
process divides the SDLC into four phases, which are
requirements engineering, project planning, architecture design,
and production. XP method adopts exploration, planning,
iteration, production, and maintenance as a phase in a
development cycle. The GXP SDLC proposes the uses the XP
method as a baseline and integrates the GSD workflow on it.
Figure 2 shows the GXP process framework.

<iXF SULC

^ i

Prc-djd Viacn

(HOW)

1. L3J LtjlitllJ -x.

rtududHc-jpr

PrniiKt_

tjtpfcHE^icn

H'ljnnnj

llcialuiB K.im g

rfftafieilLitinr

l f 1Yj

Balaam Fla'ri-g

h™^ S

PluJj.L 1

■hkltL-HKC

[>.:d.].n.:i h.:.:dl. M*

cnuw>

A.»ipL....H: 1 C :J

Figure 2. GXP Process Framework

GXP process
such as.

framework provides several input and output

Product vision. Product vision works as an input that
shows a reasonable reason definition why the company
building the product, what system will be used for, and
what the target markets are.

Business goal. Business goal works as an input that
described a business background for the product. It can
be related with productivity enhancement, increasing
revenue, gaining more customers, or cutting costs.

Product scope. Product scope works as an input that
show what kind of feature that will be developed in a
timeframe. This input usually depends on available
resources, existing budget, and time limitation.

User stories. User stories work as an output from the
exploration phases and an input for planning phase.
User stories describe the feature level requirements
model. User story details are described in user story
estimation. The detail's version of user stories is called
as the detail requirements model.

Spike solution. Spike solution is a high level
architecture of the product. It works as an input for
iteration phase and an output from exploration phase.
Spike solution also describes the simple design for the
product.

Release planning. Release planning is a milestone
plan for the product. It described the feature that will
be delivered in iteration. Release planning is an output
from the planning phases and works as an input for
production phase.

Iteration planning. Iteration planning is a set of
activities to build features in iteration. The activity is
designed from the task card that structured based on
planning game activity. It works as an output from
planning phase and an input for iteration phase.

Customer feedback. Customer feedback is a customer
feedback based on iteration. It described validation and
verification for a customer request. Changes will
happen when the team built misleading feature. It
works as an input for iteration phase.

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

• Small releases are an output from iteration. Small
releases described potential working feature that
delivers in iteration. Small releases also work as an
input for production phase.

• Acceptance test is an input for production phase. It
describes a black box test for product features. The
black box test usually runs by the customer.

• Product is an output from production phase and an
input from maintenance phase. It works as a final
release of the feature.

A Exploration phase

Exploration phase is where discovery of vision and mission
occurs. This phase is equal with requirements engineering
process in the GSD. At this stage, the stakeholder with coming
up with the vision statement. Vision statement is a high-level
statement that describes the goal of the system. For example,
the purpose of this new system is to read books via the Internet.
Vision statement is not more than 20-30 words. Then, this
vision statement is described visually through unique XP
artifacts called system metaphor. System metaphor is how the
team conceptualizes the system and is typically written in
language that is relevant to the business domain.

The detail of the system metaphor is described through a
user story. It is the tool that captures user requirements. User
stories are similar to use cases and are written by the customer
in non-technical language. Most user stories are around 1-4
sentences long. For example, library member can borrow 10
books for a maximum. User stories are written in index card or
sticky note that posted on a whiteboard.

Since the GXP is working in multi-site development the
user stories and vision statement should be documented in an
artifact. GXP proposes the user stories' artifacts such as
follows.

• Users, persona or actor that will use the system.

• User stories, the lists of the user scenarios that
supported by the system.

• Task cards, it provides a detail tasks that should be
performed for each story.

• Software estimation. The software estimation can be
done through estimation techniques such as user stories
points, or early estimation by the developer.

The user stories' artifacts are composed by the project
manager and the clients. The artifacts will be uploaded and
maintained through the document repositories that run on the
internet.

B. Planning Phase

Planning phase starts with the activity called planning
game. Planning game is a short workshop that plays by
customer and development team. Customer purposes are
determining the value, and development team determines the
cost. There are two stages of planning game, the first one is
releasing planning game, and the second is iteration planning

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

Vol. 8, No. 6, 2010
game. In the release planning game, the goal is to define the set
of features in the next release. The customer writes the stories,
development team estimates the story point, and customer
plans the overall release. In iteration planning game, the
customer chooses the stories for the iteration, and the
development team estimate how long, and accepts the
corresponding task. Estimation is based on experiences or
intuition of the development team. The goal in iteration
planning game is to define how long the iteration and what
kind of feature that delivered in iteration.

The planning game execution actually runs parallel with the
exploration phase. Therefore, the information in user story
artifact is refined in the planning phase. This phase also
discusses about creates an artifact called Iteration and Release
Planning. This artifact is created with a composition as follows.

• Categorized user story. In this section team should
create user stories that already categorized based on its
urgency. GXP introduces four levels of priority, which
are high (essentially needed), medium (adding business
value), low (nice to have) and none (verbose). Team
can easily throw none level and negotiate for low level.

• Project plan summary. It contains how long is the
project, how long the iteration length, how much the
user stories, and how many iterations in a project.

• Iteration detail. It describes the user story that executed
in the iteration. Actually, the iteration detail discusses
the work assignment for developer.

The artifact is also stored and maintained by the project
manager. It will be updated by the developer when the user
story is completed

C. Iteration Phase

Iteration phase is the real work of development happens.
The development team selects the story in current iteration.
Since user stories are generally too high-level to execute as
programming tasks, so the next step is to break out the actual
development tasks needed to support the story. Development
team works together to investigate the tasks for each user story.
The tasks are written either on the back of the story, on
separate task cards, or on some other tracking mechanism.
Developers begin work on a new task by writing the test first
and adding it to their test framework. Programming continues
in pairs with each partner taking turns to "drive" the keyboard
from time to time. At the end of the iteration, customers
perform the acceptance tests they have written. Any user stories
that fail acceptance will be fixed during the next iteration; there
is why the iteration phase provides feedback to planning phase.

The result of the iteration is software that works in staging
environment. Staging environment is the customer's software
environment that mirrors, or replicates, their live environment.
Staging is used to perform final prerelease tests and
performance checks. Customer will evaluate the staging system
and then approve the system shift to the production
environment.

In the iteration phase, the artifacts that live in the internet
are stored in the source code control systems. The tools like

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

Team Foundation Server, Team City, or hosted source code
online in third party are the several options to store the most
important artifacts in the software development. In GXP the
developer should have several agreements based on the codes
such as follows.

• The coding standard and naming conventions of the
codes. This little agreement gives benefits for the
other's team to understand the code better.

• The comment agreement. This approach clearly makes
the codes speak what developer did with the codes.
Having automatic commenting software like
GhostDoc, will make it happen.

• The developer notes. This approach makes an
agreement how the developer communicate each other
through the special comment token. A word like
TODO, BUGBUG or DEBUG will make the developer
communication is better in term exchanging
information.

D. Production and Maintenance Phase

Production is the user acceptance test that happens with the
developed codes or modules. The developer set up the system
in the development system and tests it by the client. The
approved features is copied into the production system or
staging system. The approval model works as follows.

• Developers build the codes based on the user stories,
and then uploaded into the development system.

• The client and developer do online meeting and discuss
through shared screen and review sheets.

• The client fills the review sheet approves it or rejects it
for further modification.

After the production, the system goes into the maintenance
mode. In this mode, development team can create an upgrade,
patch, or changes code with confidence since the change's
monitoring can be easily defined through test case. The hard
part in this phase is creating data migration and system
migration. Since those parts are technology specific, XP does
not provide the sufficient techniques to overcome it. The
artifact that exists in this phase is.

• Review sheet for user acceptances tests

• Defects document that fulfilled also by the client

VI. DISCUSSION AND FUTURE WORKS

In our studies, we do a novel approaches to integrate the
extreme Programming with the Global Software Development
process. Although both are contradict in the first vision, both
can be integrated through the power of artifacts and the tools.
The gap in direct communication between client and customer
can be limited by using a formal process framework and the
tools like source code version document management, or
others. The research also notes several finding that related with
the artifacts that designed by the team to tackle a
communication gaps between of them.

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

Vol. 8, No. 6, 2010
In the future, evaluation can be further improvements of the
research. For example, it will be a good idea to compare the
proposed process framework with the existing one. The impact
of the process framework in terms of productivity, product
quality, and resources will make the benefit of the framework
can be easily quantified. We are currently in the process of
defining some empirical studies with real multi-site distributed
projects.

References

[I] Belford, P. C, Bond, A. F., Henderson, D. G., and Sellers, L. S. 1976.
Specifications a key to effective software development. In Proceedings
of the 2nd international Conference on Software Engineering (San
Francisco, California, United States, October 13 - 15, 1976).
International Conference on Software Engineering. IEEE Computer
Society Press, Los Alamitos, CA, 71-79.

[2] Boehm, B. and Turner, R. 2004. Balancing Agility and Discipline:
Evaluating and Integrating Agile and Plan-Driven Methods. In
Proceedings of the 26th international Conference on Software
Engineering (May 23 - 28, 2004). International Conference on Software
Engineering. IEEE Computer Society, Washington, DC, 718-719.

[3] Carmel, E. 1999 Global Software Teams: Collaborating Across Borders
and Time Zones. Prentice Hall PTR.

[4] Clements, P., Bachmann, F., Bass, L., Garlan, D., Ivers, J., Little, R.,
Nord, R., and Stafford, J. 2002. Documenting Software Architectures:
Views and Beyond, Addison- Wesley.

[5] Hazzan, O. and Dubinsky, Y. 2008. Agile Software Engineering.
Springer.

[6] Herbsleb, J. D. and Mockus, A. 2003. An Empirical Study of Speed and
Communication in Globally Distributed Software Development. IEEE
Trans. Softw. Eng. 29, 6 (Jun. 2003), 481-494.

[7] Herbsleb, J. D., Paulish, D. J., and Bass, M. 2005. Global software
development at Siemens: experience from nine projects. In Proceedings
of the 27th international Conference on Software Engineering (St. Louis,
MO, USA, May 15 - 21, 2005). ICSE '05. ACM, New York, NY, 524-
533.

[8] McCarthy, J. and McCarthy, M. 2006 Dynamics of Software
Development. Second. Microsoft Press.

[9] Miller, A. 2008. Distributed Agile Development at Microsoft Patterns
and Practices. Microsoft.

[10] Paasivaara, M., Durasiewicz, S., and Lassenius, C 2009. Using Scrum
in Distributed Agile Development: A Multiple Case Study. In
Proceedings of the 2009 Fourth IEEE international Conference on
Global Software Engineering (July 13 - 16, 2009). ICGSE. IEEE
Computer Society, Washington, DC, 195-204.

[II] Sangwan, R., Bass, M., Mullick, N., Paulish, D. J., and Kazmeier, J.
2007. Global Software Development Handbook (Auerbach Series on
Applied Software Engineering Series). Auerbach Publications.

[12] Stober, W, and Hansmann, U. 2010. Agile Software Development: Best
Practices for Large Software Development Projects. Springer.

[13] Taylor, P. S., Greer, D., Sage, P., Coleman, G., McDaid, K., and
Keenan, F. 2006. Do agile GSD experience reports help the
practitioner?. In Proceedings of the 2006 international Workshop on
Global Software Development For the Practitioner (Shanghai, China,
May 23 - 23, 2006). GSD '06. ACM, New York, NY, 87-93.

AUTHORS PROFILE
Ridi Ferdiana. Mr. Ridi Ferdiana was born in 1983. He is a doctoral student
at Gadjah Mada University, Yogyakarta since 2008. He earned his master
degree from the same university in 2006. In his professional area, he holds
several professional certifications such as MCP, MCTS, MCPD, MCITP,
MVP and MCT. In his daily research activities he really enjoys to learn about
software engineering, business platform collaboration, and programming
optimization.

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

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

Vol. 8, No. 6, 2010

Lukito Edi Nugroho. Born in 1966, Dr. Lukito Edi Nugroho is an Associate
Professor in the Department of Electrical Engineering and Information
Technology, Gadjah Mada University. He obtained his M.Sc. and PhD
degrees from James Cook University in 1995 and Monash University in 2002,
respectively. His areas of interest include software engineering, distributed
and mobile computing, and application of ICT in education.

Paulus Insap Santosa. Insap was born in Klaten, 8 January 1961. He
obtained his undergraduate degree from Universitas Gadjah Mada in 1984,
master degree from University of Colorado at Boulder in 1991, and doctorate
degree from National University of Singapore in 2006. His research interest
including Human Computer Interaction and Technology in Education.

Ahmad Ashari Place and date of birth: Surabaya, May 2 n 1963. Get
Bachelor's degree 1988 in Electronics and Instrumentation, Physics
department Gadjah Mada University, Yogyakarta. Master degree 1992 in
Computer Science, University of Indonesia, Jakarta Doctor Degrees 2001 in
Informatics, Vienna University of Technology. Major Field of study is
distributed system, Internet, Web Services, and Semantic Web.

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

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

Vol.8,No.6,2010

A Hybrid PSO-SVM Approach for Haplotype
Tagging SNP Selection Problem

Min-Hui Lin

Department of Computer Science and Information

Engineering, Dahan Institute of Technology,

Sincheng, Hualien County , Taiwan, Republic of China

Chun-Liang Leu

Department of Information Technology, Ching Kuo

Institute of Management and Health,

Keelung , Taiwan, Republic of China

Abstract — Due to the large number of single nucleotide
polymorphisms (SNPs), it is essential to use only a subset of all
SNPs called haplotype tagging SNPs (htSNPs) for finding the
relationship between complex diseases and SNPs in biomedical
research. In this paper, a PSO-SVM model that hybridizes the
particle swarm optimization (PSO) and support vector machine
(SVM) with feature selection and parameter optimization is
proposed to appropriately select the htSNPs. Several public
datasets of different sizes are considered to compare the proposed
approach with other previously published methods. The
computational results validate the effectiveness and performance
of the proposed approach and the high prediction accuracy with
the fewer htSNPs can be obtained.

Keywords : Single Nucleotide Polymorphisms (SNPs),
Haplotype Tagging SNPs (htSNPs), Particle Swarm Optimization
(PSO), Support Vector Machine (SVM).

I. INTRODUCTION

The large number of single nucleotide polymorphisms
(SNPs) in the human genome provides the essential tools for
finding the association between sequence variation and
complex diseases. A description of the SNPs in each
chromosome is called a haplotype. The string element of each
haplotype is or 1, where denotes the major allele and 1
denotes the minor allele. The genotype is the combined
information of two haplotypes on the homologous
chromosomes and is prohibitively expensive to directly
determine the haplotypes of an individual. Usually, the string
element of a genotype is 0, 1, or 2, where represents the
major allele in homozygous site, 1 represents the minor allele
in homozygous site, and 2 is in the heterozygous site. The
genotyping cost is affected by the number of SNPs typed. In
order to reduce this cost, a small number of haplotype tagging
SNPs (htSNPs) which predicts the rest of SNPs are needed.

The haplotype tagging SNP selection problem has become
a very active research topic and is promising in disease
association studies. Several computational algorithms have
been proposed in the past few years, which can be divided into

two categories: block-based and block-free methods. The
block-based methods [1-2] firstly partition human genome into
haplotype blocks. The haplotype diversity is limited and then
subsets of tagging SNPs are searched within each haplotype
block. A main drawback of block-based methods is that the
definition of blocks is not a standard form and there is no
consensus about how these blocks should be partitioned. The
algorithmic framework for selecting a minimum informative
set of SNPs avoiding any reference to haplotype blocks is
called block- free methods [3]. In the literature [4-5], feature
selection technique was adopted to solve for the tagging SNPs
selection problem and achieved some promising results.

Feature selection algorithms may be widely categorized
into two groups: the filter approach and the wrapper approach.
The filter approach selects highly ranked features based on a
statistical score as a preprocessing step. They are relatively
computationally cheap since they do not involve the induction
algorithm. Wrapper approach, on the contrary, directly uses the
induction algorithm to evaluate the feature subsets. It generally
outperforms filter method in terms of classification accuracy,
but computationally more intensive. Support Vector Machine
(SVM) [6] is a useful technique for data classification. A
practical difficulty of using SVM is the selection of parameters
such as the penalty parameter C of the error term and the kernel
parameter y in RBF kernel function. The appropriate choice of
parameters is to get the better generalization performance.

In this paper, a hybrid PSO-SVM model that incorporates
the Particle Swarm Optimization (PSO) and Support Vector
Machine (SVM) with feature selection and parameter
optimization is proposed to appropriately select the htSNPs.
Several public benchmark datasets are considered to compare
the proposed approach with other published methods.
Experimental results validate the effectiveness of the proposed
approach and the high prediction accuracy with the fewer
htSNPs can be obtained. The remainder of the paper is
organized as follows: Section 2 introduces the problem
formulation. Section 3 describes the PSO and SVM classifier.
In Section 4, the particle representation, fitness measurement,
and the proposed hybrid system procedure are presented. Three
public benchmark problems are used to validate the proposed

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

approach and the comparison results are described in Section 5
Finally, conclusions are made in Section 6.

II. PROBLEM FORMULATION

As shown in Figure 1, assume that dataset U consists of n
haplotypesjfy.}^.^ , each with p different SNPs{S j } 1 < j < p , U is

n*p matrix. Each row in U indicates the haplotype fy and each
column in U represents the SNPSj . The element d t . denotes
the 7-th SNP of z-th haplotype, d tj e{0,l} . Our goal is to
determine a minimum size g set of selected SNPs (htSNPs)
V = {v k }, k e {1, 2,..., p} , g = |v| , in which each random

variable v k corresponding to the /c-th SNP of haplotypes in U,

to predict the remaining unselected ones with a minimum
prediction error. The size of V is smaller than a user-defined
value R (g <R), and the selected SNPs are called haplotype

tagging SNPs (htSNPs) while the remaining unselected ones
are named as tagged SNPs. Thus, the selection set V of htSNPs
is based on how well to predict the remaining set of the
unselected SNPs and the number g of selected SNPs is usually
minimized according to the prediction error by calculating the
leave-one-out cross-validation (LOOCV) experiments [7].

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

Vol.8,No.6,2010
where d =1,2,...,D , z=l, 2,...,S , and D is the dimension of
the problem space, S is the size of population, k is the iterative
times; vf d is the z-th particle velocity, x k id is the current particle

solution, ptf d is the z-th particle best (p besf ) solution achieved
so far; gb. d is the global best ( g best ) solution obtained so far by
any particle in the population; rj and r 2 are random values in
the range [0,1], both of q andc 2 are learning factors, usually
q = c 2 = 2 , w is a inertia factor. A large inertia weight

facilitates global exploration, while a small one tends to local
exploration. In order to achieve more refined solution, a
general rule of thumb suggests that the initial inertia value had
better be set to the maximum w max = 0.9, and gradually down
to the minimum w . = 0.4 .

5,
d u

J 2

d i,l d i,2

d n-l,l d n-l,2

d„

i
d.

2,j

d-u

**.,

J p-1

d ^

d ,,"

d 2,p-l

d 2 , P

d l;P

Cl.p-l

d n-l,p

d -.,p-l

d n , P _

"n,l "n,2

Figure 1 The haplotype tagging SNP Selection Problem.

III. RELATED WORKS

A Particle Swarm Optimization

The PSO is a novel optimization method originally
developed by Kennedy and Eberhart [8]. It models the
processes of the sociological behavior associated with bird
flocking and is one of the evolutionary computation techniques.
In the PSO, each solution is a 'bird' in the flock and is referred
to as a 'particle'. A particle is analogous to a chromosome in
GA. Each particle traverses the search space looking for the
global optimum. The basic PSO algorithm is as follow:

-n-ipK ~4)+Ci - r i '(qK -*£)

(i)

According to the searching behavior of PSO, the gbest
value will be an important clue in leading particles to the global
optimal solution. It is unavoidable for the solution to fall into
the local minimum while particles try to find better solutions.
In order to allow the solution exploration in the area to produce
more potential solutions, a mutation-like disturbance operation
is inserted between Eq. (1) and Eq. (2). The disturbance
operation random selects k dimensions (1 < k < problem
dimensions) of m particles (1 < m < particle numbers) to put
Gaussian noise into their moving vectors (velocities). The
disturbance operation will affect particles moving toward to
unexpected direction in selected dimensions but not previous
experience. It will lead particle jump out from local search and
further can explore more un-searched area.

According to the velocity and position updated formula
mentioned above, the basic process of the PSO algorithm is
given as follows:

1.) Initialize the swarm by randomly generating initial
particles.
2.) Evaluate the fitness of each particle in the population.
Compare the particle's fitness value to identify the both

3.)
of A

5.)
6.)

and g hest values.

x ; ,

=vi, +i +**

Update the velocity of all particles using Equation (1).

Add disturbance operator to moving vector (velocity).

Update the position of all particles using Equation (2).

Repeat the Step 2 to Step 6 until a termination criterion
is satisfied (e.g., the number of iteration reaches the pre-defined
maximum number or a sufficiently good fitness value is
obtained).

The authors [8] proposed a discrete binary version to allow
the PSO algorithm to operate in discrete problem spaces. In the
binary PSO (BPSO), the particle's personal best and global
best is updated as in continuous value. The major different
between discrete PSO with continuous version is that velocities
of the particles are rather defined in terms of probabilities that a
bit whether change to one. By this definition, a velocity must
be restricted within the range [V mh ,V nax ] . If v, k d +1 £ (V min ,V max )
thenv^/ 1 = max(min(y max , v ( k d +1 ),V min ) . The new particle position

(2) is calculated using the following rule:

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

If rand( ) < S(v* +1 ) , then x* +1 = 1 ; else x*

, where S(v* +1 ) = -

1+e"

The function S(v ld ) is a sigmoid limiting transformation and
rand() is a random number selected from a uniform distribution
in [0, 1]. Note that the BPSO is susceptible to sigmod function
saturation which occurs when velocity values are either too
large or too small. For a velocity of zero, it is a probability of
50% for the bit to flip.

B. Support Vector Machine Classifier

SVM starts from a linear classifier and searches the optimal
hyper-plane with maximal margin. The main motivating
criterion is to separate the various classes in the training set
with a surface that maximizes the margin between them. It is an
approximate implementation of the structural risk minimization
induction principle that aims to minimize a bound on the
generalization error of a model.

Given a training set of instance-label pairs
(x ( , y. ), i = 1, 2, ..., m where x z . e R n and y. e {+1, -1} . The
generalized linear SVM finds an optimal separating hyper-
plane f(x) = (w-x) + b by solving the following optimization
problem:

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

Vol.8,No.6,2010

(3) Linear SVM can be generalized to non-linear SVM via a
mapping function O , which is also called the kernel function,
and the training data can be linearly separated by applying the

(4) linear SVM formulation. The inner product (0(x ( ) • 0(x j )) is

calculated by the kernel function k^x^Xj) for given training

data. By introducing the kernel function, the non-linear SVM
(optimal hyper-plane) has the following forms:

Minimize — w T w + C^% {

Subject to: y i (<w-x i >+b) + £. -1>0, £ >0

(5)

where C is a penalty parameter on the training error, and ^ is

the non-negative slack variables. This optimization model can
be solved using the Lagrangian method, which maximizes the
same dual variables Lagrangian L D (a) (6) as in the separable
case.

m I m

Maximize L D (a) = £ a t - - £ a i a j y i y j < x. • x . >

Subject to: < a. < C, z = 1, 2, ..., m and ^ a.y. =

(6)

To solve the optimal hyper-plane, a dual Lagrangian
L D (a) must be maximized with respect to non-negative a {

under the constraint ^a.y. = and < a ( < C . The penalty

;=1

parameter C is a constant to be chosen by the user. A larger
value of C corresponds to assigning a higher penalty to the
errors. After the optimal solution a* is obtained, the optimal
hyper-plane parameters w* and b* can be determined. The
optimal decision hyper-plane f(x,a,b*) can be written as:

f(x,a,b*) = ^ d y.a* <x i -x>+b* =(w -x) + b* (7)

f (x, a , b* ) = J] y, a* < O(x) • ® (x, ) > +5*

(8)

Though new kernel functions are being proposed by
researchers, there are four basic kernels as follows.

• Linear: k(x i ,x j ) = xjxj (9)

• Polynomial: k(x i ,x.) = (yxj x . + r) d , y > (10)

• RBF: k(x,,x j ) = exp(- r ||x,-x j || 2 ),/>0 (11)

Sigmoid: k(x ; . , x. ) = tanh(/x i T x j + r)

(12)

where y , r and d are kernel parameters. Radial basis function
(RBF) is a common kernel function as Eq. (11). In order to
improve classification accuracy, the kernel parameter y in the
kernel function should be properly set.

IV. METHODS

As the htSNPs selection problem mentioned above in
Section 2, the notations and definitions are used to present our
proposed method. In the dataset U of nxp matrix, each row
(haplotypes) can be viewed as a learning instance belonging to
a class and each column (SNPs) are attributes or features based
on which sequences can be classified into class. Given the
values of g htSNPs of an unknown individual x and the known
full training samples from U, a SNP prediction process can be
treated as the problem of selecting tagging SNPs as a feature
selection problem to predict the non-selected tagging SNPs in x.
Thus, the tagging SNPs selection can be transformed to solve
for a binary classification of vectors with g coordinates by
using the support vector machine classifier. Here, an effective
PS O- SVM model that hybridizes the particle swarm
optimization and support vector machine with feature selection
and parameter optimization is proposed to appropriately select
the htSNPs. The particle representation, fitness definition,
disturbance strategy for PSO operation and system procedure
for the proposed hybrid model are described as follows.

A Particle Representation

The RBF kernel function is used in the SVM classifier to
implement our proposed method. The RBF kernel function
requires that only two parameters, C and y should be set. Using

the RBF kernel for SVM, the parameters C , y and SNPs
viewed as input features which must be optimized

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

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

Vol. 8, No.6, 2010

simultaneously for our proposed PSO-SVM hybrid system. The
particle representation consists of three parts including: C
and y are the continuous variables, and the SNPs mask are the

discrete variables. For the feature selection, if n f features are

required to decide which features are chosen, then n f +2

decision variables in each particle must be adopted.

Table 1 shows the particle representation of our design.
The representation of particle z with dimension of n f + 2 ,

where n f is the number of SNPs (features) that varies from

different datasets. x u ~ x,. e {0,1} denotes the SNPs mask,

x t n +1 indicates the parameter value C , x in +2 represents the

parameter value y . If x ik =l,k=l,2,...,n f represents the k-th

SNP on the z-th particle to be selected, and vice versa.

TABLE I.

The particle

' representation.

Discrete -variables

Continuous-variables

SNPs mask

c r

X i,l X i,2 '" X i,n f

X i,n f +1 X i,n f +2

A random key encoding method [9] is applied in the PSO
algorithm. Generally, random key encoding is used for an
order-based encoding scheme where the value of random key is
the genotype and the decoding value is the phenotype. Note
that the particle in each {x ik \< k < n is assigned a random

number on (0, 1), and to decode in ascending order with regard
to its value. In the PSO learning process, the particle to be
counted larger tends to evolve closer to 1 and those to be
counted smaller tends to evolve closer to 0. Therefore, a repair
mechanism such as particle amendment in [5] to guarantee the
number of htSNPs after update process in PSO is not required.

B. Fitness Measurement

In order to compare the performance of our proposed
approach with other published methods SVM/STSA in [4] and
BPSO in [5], the leave-one-out cross validation is used to
evaluate the quality of fitness measurement. The prediction
accuracy is measured as the percentage of correctly predicted
SNP values on non-selected SNPs. In the LOOCV experiments,
each haplotype sequence is removed one by one from dataset U,
the htSNPs are selected using only the remaining haplotypes to
predict these tagged SNPs values for the removed one. This
procedure is repeated such that each haplotype in U is run once
in turn as the validation data.

C. The Proposed Hybrid System Procedure

Figure 2 shows the system architecture of our proposed
hybrid model. Based on the particle representation and fitness

Data Preprocessing

Dataset

Split dataset by LOOCV

_£

Training set

htSNPs (Feature) Selection

Testing set

Selected htSNPs (features) subset F
PSO parameter : htSNPs mask

Testing set tagged SNPs

Training set htSNPs

SVM parameter Optimization

Training SVM classifier
PSO parameters : C and r

Learned SVM classifier

PSO operation

Fitness calculation

PSO

operation

Optimized C , r , and feature subset F

Figure 2 The flowchart of the proposed PSO-SVM model.

measurement mentioned above, details of the proposed hybrid
PSO-SVM procedure are described as follows:

Procedure PSO-SVM hybrid model

1.) Data preparation

Given a dataset U is considered using the leave-one-out
cross-validation process to split the data into training and
testing sets. The training and testing sets are represented as
U JR andU TE , respectively.

2. ) PSO initialization and parameters setting

Set the PSO parameters including the number of iterations,
number of particles, velocity limitation, particle dimension,
disturbance rate. Generate initial particles comprised of the
features mask, C and y .

3.) Selected htSNPs (features) subset

Select input features for training set according to the feature
mask which is represented in the particle from 2), then the
features subset can be determined.

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

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

Vol.8,No.6,2010
TABLE II. Results to compare PSO-SVM with SVM/STSA [4] and BPSO [5] on four real haplotype datasets.

Datasets

Prediction accuracy

(num of SNPs)

5q31
(103)

SVM/STSA

BPSO

PSO-SVM

TRPM8
(101)

SVM/STSA

BPSO

PSO-SVM

LPL
(88)

SVM/STSA

BPSO

PSO-SVM

4.) SVM model training and testing
Based on the parameters C and y which are represented in
the particle, to train the SVM classifier on the training dataset,
then the prediction accuracy for SVM on the testing dataset by
LOOCV can be evaluated.

5.) Fitness calculation

For each particle, evaluate its fitness value by the prediction
accuracy obtained from previous step. The optimal fitness
value will be stored to provide feedback on the evolution
process of PSO to find the increasing fitness of particle in the
next generation.

6.) Termination check

When the maximal evolutionary epoch is reached, the
program ends; otherwise, go to the next step.

7.) PSO operation

In the evolution process, discrete valued and continuous
valued dimension of PSO with the disturbance operator may be
applied to search for better solutions.

V. EXPERIMENTAL RESULTS AND COMPARISONS

To validate the performance of the developed hybrid
approach, three public experimental SNP datasets [4] including
5q31, TRPM8 and LPL are used to compare the proposed
approach with other previously published methods. When there
are missing data exist in haplotype datasets, the GERBIL [4-5]
program is used to resolve them. The chromosome 5q31
dataset was from the 616 kilobase region of human
chromosome 5q31 and the SNPs were 103. The TRPM8 which
consists of 101 SNPs was obtained from HapMap. The human
lipoprotein lipase (LPL) gene was derived from the haplotypes
of 71 individuals typed over 88 SNPs.

Our implementation platform was carried out on the Matlab
7.3, a mathematical development environment by extending the
Libsvm which is originally designed by Chang and Lin [10].
The empirical evaluation was performed on Intel Pentium IV
CPU running at 3.4GHz and 2 GB RAM. Through initial
experiment, the parameter values of the PSO were set as
follows. The swarm size is set to 200 particles. The searching

ranges of continuous type dimension parameters are:
Ce[10" 2 ,10 4 ]and y e [10" 4 ,10 4 ] . The discrete type particle
for features mask, we set [V min ,V max ] = [-6,6] , which yields a
range of [0.9975,0.0025] using the sigmoid limiting
transformation by Eq. (4). Both the cognition learning factor q
and the social learning factor c 2 are set to 2. The disturbance
rate is 0.05, and the number of generation is 600. The inertia
weight factor w m in = 0.4 and w m ax = 0.9 . The linearly
decreasing inertia weight is set as Eq. (13), where i now is the
current iteration and z m ax is the pre-defined maximum iteration.

w=w„ iI -^(w liI -w lil )

(13)

To compare the proposed PSO-SVM approach with the
SVM/STSA in [4] and BPSO in [5] on the three haplotype
datasets by LOOCV experiments, the computational results of
prediction accuracy according to the numbers of selected
htSNPs are summarized in Table 2. As mentioned in [4], it is
astonished that only one SNP for the 80% prediction accuracy
in 5q31 and TRPM8 datasets can be achieved. In practice, if
one guesses each SNP as 0, the prediction accuracy of 72.5%
for 5q31 dataset and 79.3% for TRPM8 dataset would be
obtained. Therefore, the appropriate selection of one htSNPs to
correctly predict 80% on the rest of non-selected SNPs is
reasonable. It is obvious that the proposed PSO-SVM hybrid
model achieves higher prediction accuracy with fewer selected
htSNPs in the three haplotype datasets. In general, the
prediction accuracy is increased refers to the incremental
selected htSNPs number. From Figure 3 to Figure 5 show that
the numbers of selected htSNPs on haplotype datasets are
proportional to the prediction accuracy and the PSO-SVM
algorithm has very good performance for haplotype tagging
SNPs selection problem in the three testing cases.

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

I i 5 6 7

<S ]0 II L? B L4 15 L6 ]7 IS W 20
Number of lilSNfc

Figure 3 The comparison result of prediction accuracy associated with selected
htSNPs on 5q31 datasets.

Figure 4 The comparison result of prediction accuracy associated with
selected htSNPs on TRPM8 datasets.

i *

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

Vol.8,No.6,2010
VI. CONCLUSION

In this paper, a hybrid PSO-SVM model that combines the
particle swarm optimization (PSO) and support vector machine
(SVM) with feature selection and parameter optimization is
proposed to effectively solve for the haplotype tagging SNP
selection problem. Several public datasets of different sizes are
considered to compare the PSO-SVM with SVM/STSA and
BPSO previously published methods. The experimental results
show that the effectiveness of the proposed approach and the
high prediction accuracy with the fewer number of haplotype
tagging SNP can be obtained by the hybrid PSO-SVM system.

REFERENCES

K. Zhang, M. Deng, T. Chen, M. Waterman and F. Sun, "A dynamic

programming algorithm for haplotype block partitioning," Proc. Natl.

Acad. Sci., vol. 99, pp. 7335-7339, 2002.

K. Zhang, F. Sun, S. Waterman and T. Chen, "Haplotype block partition

with limited resources and applications to human chromosome 21

haplotype data," Am. J. Hum. Genet., vol. 73, pp. 63-73, 2003.

H. Avi-Itzhak, X. Su, and F. de la Vega, "Selection of minimum subsets

of single nucleotide polymorphisms to capture haplotype block

diversity," In Proceedings of Pacific Symposium on Biocomputing, vol.

8, pp. 466-477, 2003.

He Jingwu and A. Zelikovsky, "Informative SNP Selection Methods
Based on SNP Prediction," IEEE Transactions on NanoBioscience, Vol.
6, pp. 60-67, 2007.

Cheng-Hong Yang, Chang-Hsuan Ho and Li-Yeh Chuang, "Improved
tag SNP selection using binary particle swarm optimization," IEEE
Congress on Evolutionary Computation (CEC 2008), pp. 854-860, 2008.
V.N. Vapnik, "The nature of statistical learning theory," New York:
Springer- Verlag, 1995.

E. Halperin, G. Kimmel and R. Shamir, "Tag SNP selection in genotype
data for maximizing SNP prediction accuracy," Bioinformatics, Vol. 21,
pp. 195-203, 2005.

J. Kennedy and R. C. Eberhart, "A discrete binary version of the particle
swarm algorithm," in Proceedings of the World Multiconference on
Systemics, Cybernetics and Informatics, Piscataway, NJ, 1997, pp.
4104-4109.

-^-SVMCTSA

— EfSO

1 2 3 4 b G 7 8 9 10 11 12 13 14 li l& 17 1$ 19 20

Nuiiit^j of lilSNFs

Figure 5 The comparison result of prediction accuracy associated with
selected htSNPs on LPL datasets.

[2]

[3]

[4]

[5]

[6]

[7]

[8]

[9]

[10]

J.C. Bean, "Genetics and random keys for sequencing and optimization,"
ORSA J. Comput., Vol. 6, pp. 154-160, 1994.

C.C. Chang, and C.J. Lin, LIB SVM: a library for support vector
machines, Software available at:

http://www.csie.ntu.edu.tw/~cjlin/libsvm, 2001.

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

PAPR Reduction Technique for LTE SC-FDMA
Systems Using Root-Raised Cosine Filter

Md. Masud Rana, Jinsang Kim and Won-Kyung Cho

Deptartment of Electronics and Radio Engineering, Kyung Hee University

1 Seocheon, Kihung, Yongin, Gyeonggi, 449-701, Republic of Korea

Email: mamaraece28@yahoo.com

Abstract — Recently, mobile radio communications have de-
veloped rapidly due to the endless demand for broadband
multimedia access and wireless connection anywhere, and any
time. With the emergence of diverse fourth generation (4G)
enabling technologies, signal processing has become ever in-
creasingly important for small power, small chip resources,
and efficient physical implementations of potential multimedia
wireless communication systems. In this paper, we analytically
derive the time and frequency domain single carrier-frequency
division multiplexing (SC-FDMA) signals. Simulation results
show that the SC-FDMA sub-carrier mapping scheme has a
significantly lower peak-to average power ratio (PAPR) compared
to orthogonal frequency division multiplexing (OFDMA). In
addition, the interleave FDMA (IFDMA) sub-carrier mapping
scheme with root raised cosine filter reduced PAPR significantly
than localized FDMA (LFDMA) and distributed (DFDMA) sub-
carrier mapping scheme. As a results, improves the mean power
output from a battery driven terminal equipment and power
amplifier efficiency.

Index Terms— CCDF, IFDMA, OFDMA, PAPR, root-raised
cosine, SC-FDMA.

I. Introduction

The further increasing demand on high data rates in wireless
communication systems has arisen in order to support broad-
band services. The third generation partnership project (3 GPP)
members started feasibility study on the enhancement of the
universal terrestrial radio access (UTRA) in December 2004,
to improve the mobile phone standard to cope with future
requirements. This project was called long term evolution
(LTE) [1].

LTE uses single carrier frequency division multiple access
(SC-FDMA) for uplink transmission and orthogonal frequency
division multiple access (OFDMA) for downlink transmission
[6]. SC-FDMA is a promising technique for high data rate
transmission that utilizes single carrier modulation and fre-
quency domain equalization. Single carrier transmitter struc-
ture leads to keep the peak-to average power ratio (PAPR)
as low as possible that is reduced the energy consumption.
SC-FDMA has similar throughput performance and essentially
the same overall complexity as OFDMA [3], [10], [12]. A
highly efficient way to cope with the frequency selectivity
of wideband channel is OFDMA. OFDMA is an effective
technique for combating multipath fading and for high bit

rate transmission over mobile wireless channels. In OFDMA
system, the entire channel is divided into many narrow sub-
channels, which are transmitted in parallel, thereby increasing
the symbol duration and reducing the intersymbol-interference
(ISI) [4], [8]. Despite many benefits of OFDMA for high speed
data rate services, it suffer from high envelope fluctuation
in the time domain, leading to large PAPR. Because the
high PAPR is detrimental to user mobile equipment (UE)
terminals, SC-FDMA has drawn great attention as an attractive
alternative to OFDMA for uplink data transmission. It can be
regarded as DFT-spread OFDMA (DFTS-OFDM), where time
domain data signals are transformed to frequency domain by
a DFT before going through OFDMA modulation. The main
benefit of DFTS-OFDM compared to OFDM scheme, is re-
duced variations in the instantaneous transmit power, implying
the possibility for increased power- amplifier efficiency, low-
complexity high-quality equalization in the frequency domain,
and flexible bandwidth assignment [12].

In order to solve the high PAPR problem seen in the
uplink of OFDMA, research is now addressing techniques such
as a SC-FDMA. The most of the previous work related to
3 GPP LTE uplink has been mainly focused on implementation
problems in the physical layer [2], [5], [9], [13]. In [10], [12]
proposed raised-cosine pulse shaping method that compare
PAPR characteristics using the complementary cumulative
distribution function (CCDF) for different subcarrier mapping.

PAPR reduction is of the most importance performance
parameter in case of high amplitude signals subject to non
linear power amplification. This situation more and more occur
due to the ever-growing demand in high spectral efficiency
advanced mobile telecommunications systems implying multi
dimensional waveforms considerations for which the PAPR
is high. Pulse shaping is required for a single carrier sys-
tem to bandlimit the transmit signal. This paper addresses
a theoretical analysis of the PAPR reduction of LTE SC-
FDMA systems when root-raised cosine (RRC) filter is used.
RRC is used as the transmit and receive filter in a digital
communication system to perform matched filtering. The
combined response of two such filters is that of the raised-
cosine filter. In this paper, we analytically derive the time and
frequency domain SC-FDMA signals. Simulation results show

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

that the SC-FDMA has a significantly lower PAPR compared
to OFDMA system. In addition, we comparing the three forms
of SC-FDMA sub-carrier mapping scheme and find that the
interleave FDMA (IFDMA) sub-carrier mapping with root
raised cosine based pulse shapping method reduced PAPR
significantly than localized FDMA (LFDMA) and DFDMA
sub-carrier mapping scheme. As a results, improves the mean
power output from a battery driven terminal equipment and
power amplifier efficiency.

The rest of the paper is organized as follows. We describes
the 3GPP LTE and LTE SC-FDMA system model in sec-
tion II and III, respectively. In section IV, we describes the
different SC-FDMA sub-carrier mapping scheme. In section
V, we describes the PAPR reduction technique for LTE SC-
FDMA systems. In section VI, we simulated and compare
the proposed method with OFDMA for different sub-carrier
mapping scheme. Finally, conclusions are made in section VII.

II. 3GPPLTE

The main purposes of the 3GPP LTE are substantially
improved end-user throughputs, low latency, reduced user
equipment (UE) complexity, high data rate, and significantly
improved user experience with full mobility. First 3 GPP LTE
and LTE-advanced (LTE-A) specification is being finalized
within 3GPP release 9 and release 10, respectively [1].

One radio frame = 20 Slots =10 Sub-frames = 10 ms

1 Slot = 7 OFDM symbols = 0.5 ms

2 Slots = 1 Sub-frame = 10 ms

TT^

; ■

< l : 2

6 ; 7

A A

► Cyclic prefix

7 OFDM symbols

••••►

.§ N

j__ Resource block:

Short CP:
7 symbols x 12 sub-carriers

Long CP:
6 symbols x 12 sub-carriers

— 1 resource element
Pilot □

Fig. 2. LTE generic frame structure.

blocks fit in a carrier of 1.4 MHz and 100 resource blocks
fit in a carrier of 20 MHz. Slots consist of either 6 or 7
OFDM symbols, depending on whether the normal or extended
cyclic prefix (CP) is employed. The CP is added in front of
each block. The details transmission scheme parameters of
the 3GPP LTE system are shown in Table I [7]. LTE uses SC-
FDMA scheme for the uplink transmissions and OFDMA in
downlink transmission.

TABLE I

LTE SYSTEM PARAMETERS

2008

2009

2010

2011

-*- Release 7 study phase (HSPA+)

► Release 8 work phase (LTE)

► Release 9 test specs

t t t

-► Release 10 (LTE
advanced)

Release 6
1 1 SPA uplink

Core First
drafted specs

Fig. 1. LTE release timeline.

Specifically, the physical layer has become quite stable re-
cently for a first implementation. LTE supports multipule input
multiput output (MIMO) with one, two, four, and eight antenna
elements at base station (BS) and mobile terminal. Both closed
and open loop MIMO operation is possible. The target of LTE-
A is to reach and surpass the international telecommunication
union (ITU) requirements. One of the important LTE-A ben-
efits is the ability to leverage advanced topology networks;
optimized heterogeneous networks with a mix of macros with
low power nodes such as picocells, femtocells, ensures user
fairness, worldwide roaming, and new relay nodes [1].

In 3GPP LTE, the basic unit of a transmission scheme is a
radio frame which is ten msec long. They are divided into ten
sub-frames, each sub-frame one msec long. Each sub-frame is
further divided into two slots, each of half msec duration. Fig.
2, shows the basic LTE generic frame structure [11]. The sub-
carrier spacing in the frequency domain is 15 kHz. Twelve
of these sub-carriers together (per slot) is called a resource
block therefore one resource block is 180 kHz. Six resource

Trans, bandwidth (MHz)

1.25

2.5

FFT size

128

256

512

1024

1536

2048

Occupied sub-carrier

151

301

601

901

1200

Sampling frequency (MHz)

1.92

3.84

7.68

15.36

23.04

30.72

No. of available PRBs

100

User plane latency (ms)

< 5

PRB bandwidth (kHz)

180

Frame duration (ms)

0.5

Sub-carrier bandwidth (kHz)

Coverage (km)

5-30

Mobility (km/hr)

15-350

Peak data rates (Mbits/s)

DL: 100, and UL: 50

Antenna configuration

DL: 4x2, 2x2, 1x2, lxl, and UL: 1x2, lxl

Spectrum efficiency

DL: 3-4 x HSDPA, and UL: 2-3 x HSUPA Rel.6

Control plane latency (ms)

100 (idle to active), and 50 (dormant to active)

Radio resource

DL: 3-4 fold higher than Rel.6

III. LTE SC-FDAMA SYSTEM MODEL

The basic principle of a LTE SC-FDMA transmission sys-
tem is shown in Fig. 3.

At the transmitter side, a baseband modulator transmits the
binary input to a multilevel sequences of complex number
mi (q) in one of several possible modulation formats including,
quandary phase shift keying (QPSK), and 16 level-QAM.
These modulated symbols are perform a N-point discrete
Fourier transform (DFT) to produce a frequency domain
representation [3]:

si(ra)

N-l

^rai(g)e

(1)

q=0

where mi is the discrete symbols, q is the sample index, j is
the imaginary unit, and mi(q) is the data symbol. The output

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Input data

Output data

Modulation

Demodulation

mi(q)

Size-N DFT

I R(k)

Size-N IDFT

si(n)

Subcarrier
mapping

4 s 3( m )

Equalization

S2(q)

Subcarrier
demapping

Size-M IDFT

Channel

Noise

s(m)

r(m)

Size-M
DFT

Cyclic
prefix(CP)
insertion

■^

Remove CP

Fig. 3. LTE SC-FDMA transceiver system model [6].

of the DFT is then applied to consecutive inputs of a size-M
inverse DFT (M >N) and where the unused inputs of the IDFT
are set to zero. If they are equal (M=N), they simply cancel
out and it becomes a conventional single user single carrier
system with frequency domain equalization. However, if N is
smaller than M and the remaining inputs to the IDFT are set
to zero, the output of the IDFT will be a signal with 'single-
carrier' properties, i.e. a signal with low power variations, and
with a bandwidth that depends on N. The SC-FDMA is single-
carrier, not single frequency. The data signal of each user
consists of a lot of frequency. DFT of SC-FDMA is used to
filter the frequency items and maps them into IDFT to reform
single user waveform. This may justify the reduced peak- to-
average power ratio (PAPR) experienced in the IDFT output.
The details description of the sub-carrier mapping mode are in
section IV. PAPR is a comparison of the peak power detected
over a period of sample occurs over the same time period. The
PAPR of the transmit signal is defined as [14]:

max 0<m<T |s(ra)| 2

PAPR

l_ f TN

(2)

w . \s(m)\ 2 dm

where T is the symbol period of the transmitted signal s(m).
PAPR is best described by its statistical parameter, com-
plementary cumulative distribution function (CCDF). CCDF
measures the probability of signal PAPR exceeding certain
threshold [12], [14]. To further reduce the power variations
of the DFTS-OFDM signal, explicit spectrum shaping can be
applied. Spectrum shaping is applied by multiplying the fre-
quency samples with some spectrum- shaping function, e.g. a
root-raised-cosine function (raised-cosine-shaped power spec-
trum). The IDFT module output is followed by a CP insertion
that completes the digital stage of the signal flow. A CP is
used to eliminate ISI and preserve the orthogonality of the

tones. Assume that the channel length of CP is larger than the
channel delay spread [8].

The transmitted symbols propagating through the radio
channel can be modeled as a circular convolution between the
channel impulse response (CIR) and transmitted data blocks.
At the receiver, the opposite set of the operation is performed.
The CP samples are discarded and the remaining N samples
are processed by the DFT to retrieve the complex constellation
symbols transmitted over the orthogonal sub-channels. The
received signals are de-mapped and equalizer is used to
compensate for the radio channel frequency selectivity. After
IDFT operation, the corresponding output is demodulated and
soft or hard values of the corresponding bits are passed to the
decoder.

IV. SC-FDMA SUB-CARRIER MAPPING SCHEME

There are two principal sub-carrier mapping modes-
localized mode, and distribution mode. An example of SC-
FDMA transmit symbols in the frequency domain for two
user, three sub-carrier per user and six sub-carriers in total
is illustrated in Fig. 4 [6].

si(l)

Li 1 i T i ttt

In tttttt

Zeros
Complex weight

si(2)

(x)

(y)

(z)

si(3)

ttrttrtrt

User 1

User 3
User 2

Time domain

Frequency domain ■

Fig. 4. Multiple access scheme of SC-FDMA: (x) IFDMA mode, (y)
DFDMA mode, and (z) LFDMA.

In distributed sub-carrier mode, the outputs are allocated
equally spaced sub-carrier, with zeros occupying the unused
sub-carrier in between. While in localized sub-carrier mode,
the outputs are confined to a continuous spectrum of sub-
carrier [10], [12]. Except the above two modes, interleaved
sub-carrier mapping mode of SC-FDMA (IFDMA) is another
special sub-carrier mapping mode. The difference between
DFDMA and IFDMA is that the outputs of IFDMA are
allocated over the entire bandwidth, whereas the DFDMA' s
outputs are allocated every several sub-carriers. If there are
more than one user in the system, different sub-carrier map-
ping modes give different sub-carrier allocation [10], [12].
In order to accommodate multiple access to the system and
to preserve the constant envelope property of the signal, the

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

elements of transmitted signal s(m) are mapped, by using the
LFDMA or IFDMA sub-carrier mapping rule.

Here is output symbol calculation in IFDMA in time
domain. The frequency samples after IFDMA sub-carrier
mapping is [12], [16]:

i(q/C) = l 0<n<N-l
else

s i(q)

where q = Cn, and C = M/N and it is the bandwidth
expansion factor of the symbol sequence. If N = M/C
and all terminals transmit N symbols per block, the system
can handle C simultaneous transmissions without co-channel
interference. After M-point IDFT operation (M > N), time
domain signal can be described as follows. Let m = Nc + q,
where < c < C — 1 and < q < N — 1. The time domain
IFDMA transmitted signal can be express as [12], [16]:

M-

s(m)

S2(q)e~

-y

N-l

^si(n)

~NC

n=0

1 1 N ~ 1

y si[n)e N

C N

n=0

N-l

1 r 1 ^— V , . J27rqn

C V N

(3)

n=0

The square backed [.] of the above equation represent the N-
point IDFT operation. The above equation can be rewritten
as

s ( m ) = g m i(q)i

(4)

is the N-point IDFT

where rra(q) = ± E n =o si(n)e'
operation. Therefore in case of IFDMA, every output symbol is
simple a repeat of input symbol with a scaling factor of 1/C in
time domain. When the sub-carrier frequency allocation starts
from fth sub-carrier i.e. q = Cn + / then the frequency
samples after IFDMA sub-carrier mapping is

i(q/C-f) = l 0<n<N-l
else

s 2(q)

and the time domain transmitted signal can be express as

1 r 1

s(m)

N-l

C [ N ^ 5 i( n ) e ^

n=0

1 22,

— e *
C

~mi(q).

(5)

Thus, when the sub-carrier frequency allocation starts from
fth instead of zero then there is an additional phase rotation

of e J2irmf/M^

Here is output symbol calculation in LFDMA sub-carrier
mapping in time domain. After DFT operation the frequency
domain sub-carrier mapping signal can be written as

si(l) = 1 0<Z <N -1
N <l< M -1

s 2(q)

After IDFT operation, time domain signal can be described
as follows. Let m = Cq + c, where < c < N — 1 and
< c < C — 1. Then time domain transmitted signal can be
represent as [12], [16]:

M-l

5 ( m ) = m Yl s ^

j27rml

e M

1 1

N-l

£ s i(o

j2n(Cq + c)l

e CN

(6)

if c = 0, then

1 1 N ~ 1
5 ( m ) = 7^77 ^2 s i( l ) e

C l N

mi(q)

(7)

Since mi(q) = ^ p=0 mi(p)e n , and if c ^ 0, then

N-l N-l

j27Tlp j27T(Cq + c)l

5 ^ = ~nc E£ mi W e ^ P ] e

1=0 p=0
N-l N-l

= 4n E E m 1 (p)e^ 2 -^-^/ JV+c / cw )'

1 _ e J2K(q-p) e J2irc/C)

NC
1

Z=0 p=0

iV-1

NC 2^ mi W ! _ e (j2ir(q-p)/N+c/CN)

1 _ e J27Tc/C

p=0
N-l

^E^w r

(1-e-

7VC ^ "" 1V/ ' / 1 - e U^{q-p)/N+c/CN)
p=0

N-l
J2nc/C\±_ V^ _

J N ^ 1

n=0

™>l(p)

D (j27r(q-p)/N+c/CN)

(8)

So, the time domain LFDMA sub-carrier mapping signal has
exact copies of input time signals with a scaling factor of 1/C
in the N-multiple sample positions and in between values are
sum of all the time input symbols in the input block with
different complex- weighting (CW) [12].

Here is output symbol calculation in DFDMA in time
domain. The frequency samples after DFDMA sub-carrier
mapping is [12], [16]:

, x f sAq/C) = 1 0<n<N -1
52 ^ ) = {o else ~

where < c < C - 1, q = Cn, and < C < C. Let
m = Nq + c where 0<c<C-landO<g<iV- land
< q < N — 1 . The time domain DFDMA transmitted signal
can be express as [12], [16]:

M-l

5 0) = m Z^ S2 ( q ) e M

1=0

1 1 N ~ 1

1 1 V^ /7\ 327r(C q + c)l ~

CN^ Sl(l)e CN ° l

1=0

(9)

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

if c = and according to the previous procedure, we obtain

TABLE II
THE SYSTEM PARAMETERS FOR SIMULATIONS

(10)

Since mi(q) = ^2 p=0 mi(p)e n , if c ^ and according
to the previous procedure, we obtain

7V-1

sim) = ^l-e^'%Y.- l

mi(p)

n=0

e (j2Tr(Cq-p)/N+Cc/CN)

So, the time domain symbols of DFDMA sub-carrier mapping
have the same structure as those of LFDMA sub-carrier
mapping.

V. PAPR REDUCTION TECHNIQUE FOR LTE SC-FDMA
SYSTEMS

In case of high amplitude signals subject to non linear
power amplification, PAPR reduction is one of the most
importance performance parameter. This situation more and
more occur due to the ever-growing demand in high spectral
efficiency telecommunications systems implying multi dimen-
sional waveforms considerations for which the PAPR is high.
In a single-carrier communication system, pulse shaping is
required to bandlimit the signal and ensures it meets the
spectrum mask. In this paper, a root raised cosine (RRC) filter
is used to pulse shape the SC-FDMA signals. RRC is used
as the transmit and receive filter in a digital communication
system to perform matched filtering. The combined response
of two such filters is that of the raised-cosine filter. The raised-
cosine filter is used for pulse-shaping in digital modulation
due to its ability to minimize intersymbol interference (ISI).
Its name stems from the fact that the non-zero portion of the
frequency spectrum of its simplest form (3 = 1 (called the
roll-off factor) is a cosine function, 'raised' up to sit above
the / (horizontal) axis. The RRC filter is characterized by
two values; /3, and T s (the reciprocal of the symbol-rate). The
impulse response of such a filter can be given as:

1-/3 + 4/3/tt,£ =

P/V2[(l + 2/tt) sin(7r//54) + (1 - 2 /it) cos(tt//34)],

t = ±T s //?4

sin[7rt/T s S (l-/3)]+4/3t/T s cos[7rt/T 5 (l+/3)] ,

7rt/T s [l-(4/3t/T s )2]) i eibe

It should be noted that unlike the raised-cosine filter, the
impulse response is not zero at the intervals of ±T S . However,
the combined transmit and receive filters form a raised-cosine
filter which does have zero at the intervals of ±T S . Only in
the case of (3 = 0, does the root raised-cosine have zeros at
±T a .

VI. PERFORMANCE ANALYSIS

The performance of the aforementioned PAPR reduction
technique is explored by performing extensive computer sim-
ulations. All simulation parameters of the LTE SC-FDMA
systems are summarized in Table II [6].

The CCDF)of PAPR, which is the probability that PAPR
is higher than a certain PAPR value PAPRO, is calculated by
Monte Carlo simulation. We compare the PAPR value that is

h(t)

System parameters

Assumptions

System bandwidth

hMHz

Number of sub-carriers

512

Data block size

Roll of factor

0.0999999999

Overs ampling factor

Number of iteration

10 4

Sub-carrier mapping schemes

DFDMA, IFDMA, LFDMA

Modulation data type

Q-PSK and 16-QAM

Spreading factor for IFDMA

Spreading factor for DFDMA

exceeded with probability less than 0.1 percentile PAPR. The
PAPR calculation using various sub-carrier mapping schemes
for SC-FDMA and OFDMA system is shown in Fig. 5. The
modulation scheme used for the calculation of PAPR is QPSK.
It can be seen that SC-FDMA sub-carrier mapping schemes

o" 1

X \

\ v \ \

\ \ Y \\ \ :

\ \ \\ A \
\ \ \\ \ \

\ \ y \ \

- - Proposed DFDMA

— Existing DFDMA

Proposed LFDMA

Existing LFDMA

Proposed IFDMA

Existing IFDMA

OFDMA

\ \ \\ \ -
M V \\ \

i \ \ M \

\ \ \ l\ \

i Kl \ :

3 4 5 6 7

PAPRo dB

Fig. 5.
QPSK.

Comparison of CCDF of PAPR for SC-FDMA with OFDMA using

gives lower PAPR values as compared to OFDMA scheme.
In addition, the root raised cosine pulse shaping method has
lower PAPR than the case of existing pulse shapping method
by more than 3dB. Due to the complex- weighting of LFDMA
and DFDMA equation would increase the PAPR. Due to the
phase rotation it is unlikely that the LFDMA samples will
all add up constructively to produce a high output peak after
pulse shaping. But it is shown that LFDMA has a lower PAPR
than DFDMA when pulse shaping is applied. Another PAPR
simulation using various sub-carrier mapping schemes for LTE
SC-FDMA systems is shown in Fig. 6. The modulation scheme
used for the calculation of PAPR is 16-QAM. It show that
IFDMA has lowest value of PAPR at 7.8dB which is 6.7dB in
case of QPSK as modulation technique. Finally, we conclude

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

o" 1

o- 2
o- 3

- - Proposed DFDMA

— Existing DFDMA

Proposed LFDMA

Existing LFDMA

Proposed IFDMA

Existing IFDMA

— OFDMA

\ \ M \
\ \ \ \\\ \
l \ \ -\\ \
i \ \ iu \

\ \ \ \\ \ "

i \ \ \ \ ;

i V \ 1 \

1 \ v 1

|_ 1 iV 1

1 ' 1

Fig. 6. Comparison of CCDF of PAPR for SC-FDMA with OFDMA using
16-QAM.

that the higher values of PAPR by using 16-QAM which is
undesirable because they cause non linear distortions at the
transmitter.

VII. Conclusions

The efficiency of a power amplifier is determined by the
PAPR of the modulated signal. In this paper, we analysis
different sub-carrier mapping scheme for LTE SC-FDMA
systems. We derive the time and frequency domain signals
of different sub-carrier mapping scheme, and numerically
compare PAPR characteristics using CCDF of PAPR. We come
to the conclusion that the IFDMA sub-carrier mapping with
RRC pulse shaping method has lowest PAPR values compare
to the other sub-carrier mapping methods. As a results, im-
proves the mean power output from a battery driven terminal
equipment and power amplifier efficiency. Therefore, SC-
FDMA is attractive for uplink transmissions since it reduces
the high PAPR seen with OFDMA.

[3] B. Karakaya, H.Arslan, and H. A. Cirpan, "Channel estimation for LTE

uplink in high doppler spread," Proc. WCNC, pp. 1126-1130, April 2008.
[4] J. Berkmann, C. Carbonelli, F.Dietrich, C. Drewes, and W. Xu, "On

3G LTE terminal implementation standard, algorithms, complexities and

challenges," Proc. Int. Con. on Wireless Communications and Mobile

Computing, pp. 970-975, August 2008.
[5] A. Ancora, C. Bona, and D.T.M. Slock, "Down-sampled impulse response

least-squares channel estimation for LTE OFDMA," Proc. Int. Con. on

Acoustics, Speech and Signal Processing, Vol. 3, pp. 293-296, April 2007.
[6] M. M. Rana, M. S. Islam, and A. Z. Kouzani, "Peak to average power ratio

analysis for LTE aystems," Proc. Int. Con. on Communication Software

and Networks, pp. 516-520, February 2010.
[7] A. Ancora, C. B. Meili, and D. T. Slock, "Down-sampled impulse

response least- squares channel estimation for LTE OFDMA," Proc. Int.

Con. on Acoustics, Speech, and Signal Processing, April 2007.
[8] L. A. M. R. D. Temino, C. N. I Manchon, C. Rom, T. B. Sorensen, and

P. Mogensen, "Iterative channel estimation with robust wiener filtering in

LTE downlink," Proc. Int. Con. on Vehicular Technology Conference, pp.

1-5, September 2008.
[9] J. Zyren, "Overview of the 3GPP long term evolution physical layer," Dr.

Wes McCoy, Technical Editor, 2007.
[10] H. G. Myung, J. Lim, and D. J. Goodman, "Single carrier FDMA for

uplink wireless transmission," IEEE Vehicular Technology Magazine, vol.

1, no. 3, pp. 30-38, September 2006.
[11] M. Noune and A. Nix, "Frequency-domain precoding for single carrier

frequency- division multiple access," IEEE Commun. Magazine, vol. 48,

no. 5, pp. 86-92, May 2010.
[12] H.G. Myung, J. Lim, and D. J. Goodman, "Peak to average power ratio

for single carrier FDMA signals," Proc. PIMRC, 2006.
[13] S. Maruyama, S. Ogawa, and K.Chiba, "Mobile terminals toward LTE

and requirements on device technologies," Proc. Int. Con. on VLSI

Circuits, pp. 2-5, June 2007.
[14] S. H. Han, and J. H. Lee, "An overview of peak to average power ratio

reduction techniques for multicarrier transmission," IEEE Transction on

Wireless Communications, April 2005.
[15] H. G. Myung, "Introduction to single carrier FDMA," Proc. Int. Con.

on European Signal Processing (EUSIPCO), Poznan, Poland, September

2007.
[16] A. Sohl, and A. Klein, "Comparison of localized, interleaved and block-
interleaved FDMA in terms of pilot multiplexing and channel estimation,"

Proc. Int. Con. on PIMRC, 2007.

Acknowledgment

This research was supported by the Basic Science Research
Program through the National Research Foundation of Korea
(NRF) funded by the Ministry of Education, Science and
Technology (20100017118).

References

[1] Q. Li, G. li, W. Lee, M. II. Lee, D. Clerckx, and Z. li, "MIMO techniques

in WiMAX and LTE: a feature overview," IEEE Commun. Magazine, May

2010.
[2] E. Dahlman, S. Parkvall, J. Skold, and P. Beming, "3G evolution HSPA

and LTE for mobile broadband," Academic Press is an Imprint of Elsevier,

2007.

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

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

Vol. 8, No. 6, September 2010

Survey of Routing Protocols and Channel
Assignment protocols in Wireless Mesh Networks

Vivek M R athod, Suhas J M anangi, S atish E , S a u m y a H e g d e
National Institute of Technology Karnataka - Surathkal

Abstract: This paper is a survey on wireless mesh networks.
Here we mention the basics of wireless mesh network, their
purpose, channel assignment techniques and routing
protocols. This survey is prepared towards helping those
working on the relationship between channel assignment
and routing protocols.

Keywords: Wireless Mesh Networks, Routing protocols,
Channel Assignment, Multi Hop, Multi Radio.

INTRODUCTION

[1] : c

A wireless mesh network (WMN) is a communication
network made up of radio nodes organized in a mesh
topology. The nodes which constitute the WMN are in
adhoc mode so as to realize mesh topology.
Wireless mesh architecture is an effort towards providing
high-bandwidth network over a specific coverage area.
Wireless mesh architecture's infrastructure is, in effect, a
router network minus the cabling between nodes. It's built
of peer radio devices that don't have to be cabled to a
wired port like traditional WLAN access points (AP) do.
The traditional WLANs consist only of single hop end-to
end connection (i.e., between the client and access point).
In contrast, Mesh architecture sustains signal strength by
breaking long distances into a series of shorter hops.
Intermediate nodes not only boost the signal, but
cooperatively make forwarding decisions based on their
knowledge of the network, i.e. perform routing. Such
architecture may with careful design provide high
bandwidth, spectral efficiency, and economic advantage
over the coverage area.
This paper is organized in the following sections.

1. Types of wireless mesh networks (network
architectures).

2. Essential characteristics of WMN.

3. Components of WMN and their alternatives.

4. Routing purposes, problems and protocols.

5. Areas for research.

II. NETWORK ARCHITECTURE
The types of network structures being used for WMNs can
be classified into three types in a very broad sense.

1. Client wireless mesh networks

2. Infrastructure wireless mesh networks

3. Hybrid wireless mesh networks [8] .

A Client WMN:

Client mesh networks or simply ad-hoc networks are formed

by client devices with no supporting fixed infrastructure.

Each node plays same role and participates in packet

forwarding.

B. Infrastructure WMN:

In contrast to client WMN, infrastructure WMN consists of
routers and client devices. The routers are interconnected
via wireless links to form a multi-hop backhaul
infrastructure. One or more routers are connected to the
wired network and are called gateways of the WMN.
Generally mesh router has two or more radio interfaces. One
of which is an access interface for the clients to access the
network. The second radio interface is a relay interface for
forwarding and routing data packets. This is basically used
for inter-router communication. Client devices associate
themselves with the nearest mesh router to access the
network. They don't participate in routing or relaying of
packets. Therefore even if two clients are within the wireless
range of each other, they cannot directly communicate. It
has to happen through their respective routers.

C. Hybrid WMN [8] :

This architecture is the combination of Infrastructure and
client meshing; clients can access the network through
mesh routers as well as by directly meshing with other
mesh clients. A hybrid WMN is an extension to the
Infrastructure WMN. In a hybrid WMN the clients not
only connect to the wireless backhaul, but also serve as a
gateway to for the clients which are located too far from
the wireless mesh router. Therefore a hybrid WMN is
more robust and more scalable than the previous two. A
well-built hybrid WMN would enable fast, cheap and easy
deployment of networks, leading to interesting
applications such as emergency networks.

III. ESSENTIAL CHARACTERISTICS OF WMN
WMN has mainly following essential characteristics:

1. Multihop and Multi-pathing: multiple paths
between two point in a WMN leads to the increase
in bandwidth. This increase in the bandwidth is
because the RTT (round trip time) for shorter paths
(hops) is less than that of a single end to end path.
Multiple packets can travel simultaneously

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

between two ends. Multi-pathing strips the data to
be sent to a destination and sends these chunks via
multiple paths, which increases the throughput
significantly. It also provides robustness to paths,
because there is always an alternative unless the
destination itself is not connected to the network.

2. Self-healing, self-forming and self-organising:
Since most of the nodes of the WMN are mobile,
the WMN is always aware of its surroundings. It
dynamically changes the routing paths based on the
current state of the network. If a participating node
quits, the network is reconfigured so as to keep the
remaining nodes connected. Similarly the dynamic
changes in the network must also take place based
on the network traffic at different routes.

3. Compatibility and interoperability: The WMNs
built on the IEEE 802.11 standards must be capable
of supporting conventional Wi-Fi clients.

4. Cost Factor: WMNs can be very cost effective
because we can build and configure a WMN with
minimal existing resources. A WMN could provide
an effective and good internet bandwidth to the
group of users who share a single internet link.

IV. COMPONENTS OF WIRELESS MESH
NETWORKS
A WMN consists of two types of wireless nodes. Mesh
Routers and Mesh Clients. The Mesh Routers have
improved computational, communication and power
resources as compared to Mesh Clients. Mesh Routers are
generally static and form the multi-hop backhaul network
with optional access to other auxiliary networks. In addition,
Mesh Routers are also typically equipped with multiple
wireless network interfaces (IEEE 802.11 [3] ) and are
therefore able to establish high capacity connections. Mesh
Clients are mobile devices, which take advantage of the
existing communication infrastructure provided by the Mesh
Routers.

A 1-Radio VS Multi-Radio Approaches:
In 1-radio approach the participating nodes have only one
radio each. Consider a network where both the clients and
the mesh routers have only one radio, and then mesh
routers would not be able to listen to the backhaul and the
client simultaneously. Collisions would be very frequent.
This will result in very low throughput. Thus one radio
WMN is inferior to multi-radio infrastructure mesh
networks in Multihop situations. In the case of 1 radio ad
hoc mesh networks, available bandwidth is reduced by
50% with each hop: bandwidth available at the 3rd hop is
1/8 of the available capacity. However, while one-radio ad
hoc mesh networks are unsuitable for Multihop situations,
they are useful in one-hop situations for quickly
establishing p2p communications. Conversely, 2-radio
infrastructure meshes are ideal for Multihop situations
with no restriction on the number of hops. Thus One radio
mobile client mesh network combined with two or more

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

Vol. 8, No. 6, September 2010
radio routers' backhaul support provides the best hybrid
WMN; ubiquitous connectivity but with multiple levels of
redundancy built in.

V. MULTI RADIO MESH ROUTER
There are mainly 4 types of Multi Radio Mesh Routers:

1. Single unit mesh router

2. split wireless router

3. Multi-Channel Multi-Transceiver single radio

4. low cost mesh router construction

A Single Unit Mesh Router:

Single unit mesh router is a single package with multiple
radios in it. All these radios' operate in non overlapping
channels. Some of these could be used to relay packets
between routers, while the others to provide connectivity to
the clients or client adhoc network. Even though the radios
operate in non-overlapping channels, the practical results
have shown that there is a significant amount of interference
between them due to the near-field effect, resulting in
reduced throughput

B. Split Wireless Router:

Split mesh router is a network (wired) of two or more
single radio routers. This design has gained motivation
from the limitations of the single unit multi-radio routers.
We refer to the single radio routers which are part of split
router as nodes hereafter.

The commercially available single-radio routers often
provide multiple interface technologies like the Ethernet,
fiber or ATM. Two or more such units are connected via a
backhaul using one of the available interface options like
the Ethernet. Since the separation between these nodes is
determined by the cable length forming the backhaul, the
interference can be significantly reduced by increasing the
distance between them. This is an effective solution for
the interference due to near-field effect in the single unit
mesh router.

Since our mesh router unit is a combination of 3-single
radio routers, we need a software abstraction by which the
assembly appears like a single unit to the network. Each
single radio router must here be aware of the neighbors of
the other two.

C. Multi-Channel Multi Transceiver Single Radio [4] :

In this kind of routers, a radio includes multiple parallel RF
front-end chips and baseband processing modules to support
several simultaneous channels. On top of the physical layer,
only one MAC layer module is needed to coordinate the
functions of multiple channels. So far no multi-channel
multi-transceiver MAC protocol has been proposed for
WMNs.

D. Low Cost Mesh Router Construction

A low cost router can be set up using two USB or PCI radio
cards on a low-end computer. But this would also require a
MAC layer which supports multiple NICs simultaneously.

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

VI. CHANNEL ASSIGNMENT IN MULTI RADIO
ROUTERS [2] [4]
In case of a two radio router network, there is no much
flexibility, because one of the radios is used to
communicate with other mesh routers on same channel
and the other radio is used to communicate with clients.
But if we have more than two radios on each mesh router,
we could use one for communicating with clients and the
other radios could be intelligently assigned different
channels so that they form channel diversified routes
among mesh routers. Whether or not two routers are
neighbors is decided by the channels assigned to them.
The channel assignment can be made based on link quality
and the topology.

There are various algorithms proposed for the channel
assignment problem. They can be classified into two
categories;

1. Interference-aware channel assignment(IACA)

2. Traffic-aware channel assignment (TACA)
Some of the algorithms are:

1. Identical channel assignment

2. Hybrid channel assignment

3. Centralized channel assignment

4. Maxflow based channel assignment routing
(MCAR)

5. Topology and interference-aware channel
assignment (TIC)

A Identical Channel Assignment:

In this method first radio is assigned channel 1, second is
assigned next non overlapping channel and so on. Though
this preserves connectivity, this method in no way makes
any effort in reducing interference.

B. Hybrid Channel Assignment:

In this strategy some radios are statically assigned channels
while other radios are assigned channels dynamically.

C. Centralized Channel Assignment:

In this method the links are visited in some order and a
common channel is assigned to interfaces on both the ends.
If all the interfaces of the end node are already assigned a
channel and they don't share any common channel, then it is
necessary to replace one on the channel assignments. This
ends up in a recursive channel assignment procedure. The
visit can be in the decreasing order of the number of links
falling in the interference range and the least used channel in
that range is selected (interference aware). It could also be
based on the estimated link flow rates(traffic aw are). The
algorithm might then visit all the links in decreasing order of
expected link flow rate and select the channel which
minimizes the sum of expected flow rates of all the links in
the interference region that are assigned the same channel.

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

Vol. 8, No. 6, September 2010
D. Maxflow Based Channel Assignment Routing:
MCAR is an improvement over centralized channel
assignment algorithm. The interdependence among channel
assignments across the whole network is taken into account
by first identifying the groups of links that need to be
assigned the same channel in order for the number of
different channels on every router not to exceed the number
of radios. Then, the actual channel assignment stage exploits
the result of the first stage to assign channel in such a way
that no replacement of previously assignments are
necessary.

E. Topology and Interference aware Channel Assignment:
This algorithm undergoes two phases. One is Topology
discovery and the other is channel selection.
Topology discovery: Prior to the channel assignment the
topology is discovered. Topology discovery for every
router is the identification of band-specific set of
neighboring routers and the measurement of quality of
link to each of these neighbors. Each router tunes itself to
various channels on which band topology is to be
discovered. This activity is co-ordinated by the channel
management server. The link quality is measured by ETT
(estimated transmit time).

Channel selection: Dijkstra's shortest path algorithm is
used in TIC to discover frequency-diversified routes
between the gateway and routers. The interference
between mesh links is generated using conflict- graph
model. For generating the above model interfering mesh
links have to be identified in the first place. Thus the data
generated in the first phase (topology discover) can be
used to construct conflict graph. Thus the interfering links
are assigned non-overlapping channels.
Cross-layer work: In most of the situation the throughput
of configured WMN depends on both the channel
assignment and routing algorithm chosen. So there is a lot
of research in developing the cross-layer protocols which
deals with the channel assignment and routing jointly.

VII. Routing Protocols [9]
Routing protocols lie at the heart of designing a WMN
network. They, in simple terms, specify the relay routes for
packets in the network. Most of the protocols neglect the
traffic between the mesh nodes and only consider the
traffic between the nodes and the internet.
Network Asymmetry: This is the situation in which
forward direction of a network is significantly different
from the reverse direction in terms of bandwidth, loss rate,
and latency. Forward path routing protocols are effective
in routing of the packets from the mesh nodes to the
gateway of the WMN, backward routing protocols are
effective in routing the packets from the internet to the
mesh nodes.

Some of the most popular protocols being used are AODV
and OLSR.

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

A. Ad-hoc On Demand Distance Vector Routing (AODV):
The Ad hoc On Demand Distance Vector (AODV) routing
algorithm is a routing protocol designed for ad hoc mobile
networks. AODV is capable of both unicast and multicast
routing. It is an on demand algorithm, meaning that it
builds routes between nodes only as desired by source
nodes. It maintains these routes as long as they are needed
by the sources. Additionally, AODV forms trees which
connect multicast group members. The trees are composed
of the group members and the nodes needed to connect the
members. AODV uses sequence numbers to ensure the
freshness of routes. It is loop-free, self-starting, and scales
to large numbers of mobile nodes.

AODV builds routes using a route request / route reply
query cycle. When a source node desires a route to a
destination for which it does not already have a route, it
broadcasts a route request (RREQ) packet across the
network. Nodes receiving this packet update their
information for the source node and set up backwards
pointers to the source node in the route tables. In addition
to the source node's IP address, current sequence number,
and broadcast ID, the RREQ also contains the most recent
sequence number for the destination of which the source
node is aware. A node receiving the RREQ may send a
route reply (RREP) if it is either the destination or if it has
a route to the destination with corresponding sequence
number greater than or equal to that contained in the
RREQ. If this is the case, then it unicast a RREP back to
the source. Otherwise, it rebroadcasts the RREQ. Nodes
keep track of the RREQ's source IP address and broadcast
ID. If they receive a RREQ which they have already
processed, they discard the RREQ and do not forward it.
If the RREP propagates back to the source, then nodes set
up forward pointers to the destination. Once the source
node receives the RREP, it may begin to forward data
packets to the destination. If the source later receives a
RREP containing a greater sequence number or contains
the same sequence number with a smaller hop count, it
may update its routing information for that destination and
begin using the better route.

As long as the route remains active, it will continue to be
maintained. A route is considered active as long as there
are data packets periodically travelling from the source to
the destination along that path. Once the source stops
sending data packets, the links will time out and
eventually be deleted from the intermediate node routing
tables. If a link break occurs while the route is active, the
node upstream of the break propagates a route error
(RERR) message to the source node to inform it of the
now unreachable destination(s). After receiving the
RERR, if the source node still desires the route, it can
reinitiate route discovery.
Advantage and disadvantages:

The main advantage of this protocol is that routes are
established on demand and destination sequence numbers
are used to find the latest route to the destination. The
connection setup delay is less. One of the disadvantages of

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

Vol. 8, No. 6, September 2010
this protocol is that intermediate nodes can lead to
inconsistent routes if the source sequence number is very
old and the intermediate nodes have a higher but not the
latest destination sequence number, thereby having stale
entries. Also multiple RouteReply packets in response to a
single RouteRequest packet can lead to heavy control
overhead. Another disadvantage of AODV is that the
periodic beaconing leads to unnecessary bandwidth
consumption.

B. Optimized Link State Routing Protocol(OLSR):

It is a proactive protocol. The Optimized Link State

Routing Protocol (OLSR)

[1] [5]

is a proactive routing

protocol. Every node sends periodically broadcast
"Hello"-messages with information to specific nodes in
the network to exchange neighborhood information. The
information includes the nodes IP, sequence number and a
list of the distance information of the nodes neighbors.
After receiving this information a node builds itself a
routing table. Now the node can calculate with the shortest
path algorithm the route to every node he wants to
communicate. When a node receives an information
packet with the same sequence number twice he is going
to discard it. In these routing tables he stores the
information of the route to each node in the network. The
information is only updated:

1. A change in the neighborhood is detected.

2. A route to any destination is expired.

3. A better (shorter) route is detected for a
destination.

The difference from OLSR to LSR (Links State Protocol)
is that OLSR relies on multi-point relays (MPR). MPR is a
node which is selected by its direct neighbor (one
hop).The first idea of multipoint relays is to minimize the
flooding of broadcast messages in the network. An
information packet should not be sent twice in the same
region of the network. MPR helps to optimize and reduce
that problem. Each node informs its direct neighbors (one
hop) about its MPR set in the "Hello"-messages. After
receiving such a "Hello"-message, each node records the
nodes MPR Selector that selects it as one of their MPRs.
The second idea is that the size of the hello messages is
reduced. It includes only the neighbors that select node N2
as one of their MPR nodes. In this way partial topology
information is propagated. Node N2 can be reached only
from its MPR selectors.
Advantages:

1. Minimal latency.

2. Ideal in high density and large networks.

3. OLSR achieves more efficiency than classic LS
algorithms when networks are dense.

4. OLSR avoids the extra work of "finding" the
destination by retaining a routing entry for each
destination all the time, thus providing low
single-packet transmission latency.

5. OLSR can easily be extended to QoS monitoring
by including bandwidth and channel quality

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

information in link state entries. Thus, the quality
of the path (e.g., bandwidth, delay) is known
prior to call setup.
Disadvantages:

1. When the network is sparse, every neighbor of a
node becomes a multipoint relay. The OLSR then
reduces to a pure LS protocol.

2. High control overhead (reduced by MPR usage) .

3. Higher computation.

4. Storage.

5. Implementation complexity.

C. Backward Routing Protocol:

The above mentioned protocols are designed to route in
the direction from the mesh nodes to the internet.
However most services generate asymmetric traffic and
the amount of the downstream from the servers in the
internet to the mesh nodes far exceeds the upstream.
Therefore some routing protocols are proposed which take
care of this backward traffic. Backward path routing in
more involved than routing on forward path because the
data addressed at any host in the internet only needs to be
forwarded to the gateway, while the backward routing
protocol needs to address each node individually.
There are three main families of backward routing
protocols; reactive hop-by-hop routing, proactive hop-by-
hop routing, and proactive source routing.

D. AODV-CGA:

This extended AODV protocol allows the use of multiple
gateways to the internet. It shares most of the mechanisms
with the well-known AODV protocol. The addition to the
existing AODV which is made is that, all gateways are
connected to a dedicated router that acts as a proxy to the
internet. This router has two tasks:

1. On the forward path, it sends route on behalf of
hosts in the internet;

2. On the backward path, it initiates route requests for
nodes in the wireless mesh network.

E. Proactive Field-based Routing (PFR):

Wireless mesh nodes periodically exchange beacons.
These beacons contain a list of all known destinations with
their respective field value. When a new destination
appears, it announces its presence with beacons to its
neighbors in order to establish a field. With this
mechanism, a field on the network is constructed for every
destination. This field assigns a value to every node in the
network; the destination bears the maximum value.
Packets are then routed along the steepest gradient
towards the destination.
Advantages and disadvantages:

This protocol ensures loop freedom. This protocol enables
nodes to consider multiple routes to the destination. This
protocol also has a drawback. Since it proactively
maintains all routes, it incurs communication overhead
even if the traffic is too low.

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

Vol. 8, No. 6, September 2010
F. Gateway Source Routing (GSR):
In this protocol forward path information from the packets
that arrive at the gateway is reused. In the routing header
of every packet, the intermediate hops from the mesh node
to the gateway are recorded. These paths are then stored in
the gateways. To route packets to a mesh node, the mesh
gateway inverts the recorded forward path and copies it to
the packet header. The gateway then sends the packet to
the first node of the backward path. Each node updates the
path in the header by removing its entry and forward the
packet to the given next hop until the packets reaches the
destination.

GSR requires that a packet towards a host in the internet is
first sent by a mesh node in order to establish the
backward path. This should not be a problem when we
assume that majority of communication is initiated by the
mesh nodes.

If a mesh node has to act as a server, a dedicated
addressing mechanism will have to be used.

G. Hierarchical Routing Protocols:
Some of the protocols which differ from the above are
hierarchical routing and geography based routing.
In hierarchical routing, a certain self-organization scheme
is employed to group network nodes into clusters. Each
cluster has one or more cluster heads. Nodes in a cluster
can be one or more hops away from the cluster head.
Since connectivity between clusters is needed, some nodes
can communicate with more than one cluster and work as
a gateway. When the node density is high, hierarchical
routing protocols tend to achieve much better performance
because of less overhead, shorter average routing path,
and quicker set-up procedure of routing path. However,
the complexity of maintaining the hierarchy may
compromise the performance of the routing protocol.
In WMNs, a mesh client must avoid being a cluster head
because it can become a bottleneck due to its limited

capacity.

H. Geographical Base Routing:

Compared to topology-based routing schemes, geographic
routing schemes forward packets by only using the position
information of nodes in the vicinity and the destination
node. Thus, topology change has less impact on the
geographic routing than the other routing protocols. Early
geographic routing algorithms are a type of single-path
greedy routing schemes in which the packet forwarding
decision is made based on the location information of the
current forwarding node, its neighbors, and the destination
node. However, all greedy routing algorithms have a
common problem, i.e., delivery is not guaranteed even if a
path exists between source and destination. In order to
guarantee delivery, planar-graph-based geographic routing
algorithms have been proposed recently. However, these
algorithms usually have much higher communication
overhead than the single-path greedy routing algorithms.

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

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

Vol. 8, No. 6, September 2010
VIII. CONCLUSIONS
Here we mentioned the basics of wireless mesh network,
their purpose, various techniques involved and the area of
the research in wireless mesh networks. Further research can
be done studying the relationship between channel
assignment techniques and routing protocol. These two
areas will influence one another and an efficient
combination can be possibly found.

IX. REFERENCES

[1]. Optimization of routing algorithm in wireless mesh networks.

Gupta B.K, Acharya B.M, Mishra M.K. NaBIC 2009
[2]. Distributed channel assignment for multi-radio wireless mesh

networks. Makram S>A, Gunes M. ISCC 2008
[3]. Performance Analysis of IEEE802.il Wireless Mesh Networks. Ye

Yan, Hua Cai, Seung-Woo Seo. ICC 2008
[4]. Routing metrics for multi-radio wireless mesh networks. Guerin J,

Portmann M, Pirzada A. ATNAC 2007
[5]. Routing Packets into Wireless Mesh Networks. Baumann R,

Heimlicher S, Lenders V, May M. WiMOB 2007
[6]. Neighbor selection technique for multi hop wireless mesh

networks. Coll B, Gozalvez J.LCN 2009
[7]. A New multi channel MAC protocol combined with on demand

routing for wireless mesh networks. Guojun Shui, Shuqun Shen.

ICCSSE 2008
[8]. Hybrid routing with periodic updates in wireless mesh networks.

Damle A, Rajan D, Faccin S.M. WCNC 2006
[9]. Multipath routing algorithm based on traffic prediction in wireless

mesh networks. Li Zhi yuan, Wang Ru chuan, Bi Jun lei. ICNC

2009

77 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

An Approach For
Designing Distributed Real Time

Database

Dr. Dhuha Basheer Abdullah
Computer Sciences Dept./Computers Sciences and
Mathematics College /Mosul University
Mosul- Iraq

Ammar Thaher Yaseen

Computer Sciences Dept./Computers Sciences and

Mathematics College /Mosul University

Mosul- Iraq

Abstract- A distributed Real Time database system is a
transaction processing system that is designed to handle
workloads where transactions have service deadlines. The
emphasis here is on satisfying the timing constraint of transactions
(meet these deadlines, that is to process transactions before their
deadlines expire) and investigating the distributed databases. This
paper produces a proposed system named ADRTDBS.

In this work a prototype of client/server module and
server/server module for distributed real time database has been
designed. Server gets the data from direct user or a group of
clients connected with it, analyze the request; and broad updating
to all servers using 2PC (Two Phase Commit) and executing the
demand by using 2PL (Two Phase Locking). The proposed model
does not concern with data only, but provide a synchronize
replication, so the updating on any server is not saved unless
broadening the updating on all servers by using 2PC, and 2PL
protocols. The database on this proposed system is homogenous
and depend on full replication to satisfy real time requirements.

The transactions have been scheduled on the server by using a
proposed algorithm named EDTDF (Earliest Data or Transaction
Deadline First). This algorithm works to execute transactions that
have smallest deadline at the beginning, either this deadline
specific to the data or to the transaction itself. Implementing this
algorithm helps to execute greater rate of transactions before their
deadlines.

In this work two measures of performance for this system
(proposed model) were been conducted; first, computing the Miss
Ratio (rate of no. of executing transactions that miss their
deadline); second, computing the CPU utilization (CPU utilization
rate), by executing a set of transactions in many sessions.

Keywords: real time, databases, distributed, replication, Scheduling

I. INTRODUCTION

According to the definition provided by Coulouris,
Dollimore & Kindberg [4], a distributed system consists of a set
of autonomous processing elements that are connected via a
communication network and interact via message passing.

A database is a structured set of data maintained by a
database management system (DBMS) that interfaces with a set
of applications or clients that access and modify the data. In a
distributed database system, the data is distributed among
autonomous DBMS instances (nodes or sites) that communicate
via a network. The nodes, potentially along with a central

coordinator, are collectively referred to as a distributed
database management system (DDBMS) [1,7,8,15].

In a distributed database, replication of data objects(The
term object is used for the unit of replication; this could just as
well be a table in a relational database as an object) is often
used to improve fault tolerance and availability in the system
by maintaining several copies of data objects and placing those
copies close to the clients that want to use them [19].

In a real-time system (RTS), the value of a performed task
depends not only on its functional correctness, but also on the
time at which it is produced. For example, when an
autonomous vehicle detects an obstacle in its intended path, it is
crucial that it changes its path before a collision occurs. Real-
time systems are often embedded, meaning that they are a part
of (and interact heavily with) a physical environment.
Typically, embedded systems use specific-purpose rather than
general-purpose computers, such as in the embedded system
controlling fuel injection in a car engine [6,20].

It is paramount that real-time systems have predictable,
bounded and sufficiently low requirements on resources such as
memory, network bandwidth and processor execution time,
since failures due to unpredictable behavior and/or over
consumption of available resources may cause unacceptable
damage to humans or equipment. Real-time systems also need
to be highly and predictably available, meaning that when a
request is made to the system, it can be guaranteed that the
system is available to service that request within a predictable
and bounded time.

A distributed real-time system (DRTS) combines
characteristics of distributed and real-time systems. This means
that in such a system, issues related to distribution (such as
execution of distributed algorithms and network
communication) must be addressed with real-time requirements
in mind.

Real-time database systems (RTDBS) are often used to
manage data in real-time systems, since traditional databases
cannot meet the timeliness and predictability requirements of a
RTS. As many embedded applications with real-time
requirements are inherently distributed, RTDBS are often
distributed over a set of autonomous nodes, creating a need for
distributed real-time database systems (DRTDBS) [10,14,16].

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

• Replication in DRTDB S

Data replication can be used to increase availability,
predictability, and reliability of transaction processing in
DRTDB S. Common replication approaches for DRTDB S use
either a primary copy to deterministically apply updates to
replicated data, or use distributed concurrency control and
distributed commit protocols.

The distributed algorithms required to implement, e.g.,
distributed locking (to ensure serializability) and distributed
commit (to ensure mutual consistency and durability) are hard
to make predictable and sufficiently efficient due to their
reliance on correct message delivery. Furthermore, depending
on the replication approach a transaction may be forced to
either wait or roll back and restart due to concurrent execution
of transactions on remote nodes. Such behavior is problematic
in real-time systems, since potential blocking times and
rollbacks must be considered when determining worst-case
execution times of transactions. For this reason, optimistic
replication approaches, where transactions are allowed to
execute as if no concurrent transactions exist, are more suitable
than pessimistic replication approaches in real-time databases.
Optimistic replication increase the availability, predictability
and efficiency of transaction execution at the cost of transaction
conflicts that must be resolved [2,9].

II. RATED WORK

In 1994 the researcher Nandt Subakr and others discuss the
ways in which the Commit Protocol that could be adapted to
the environmental sensitivity of the cases required for real-time.
This protocol depends on the strategies and the installation
optimistic on local compensation [13].

In 1994 also provided a researcher Victor Fiy and other
researchers produce the basic rules to support the necessary
qualities to the environment of the account distributed to RT,
which is the modalities of distribution of time. It provides
general concepts to clarify the application of the expansion of
CORBA [18].

In 1998 the researcher Krayas Shihabi and others discuss
the experience to implement the 2- Server have the same DBMS
are linked through the Internet. Focusing on the intelligence
linking the researchers explained how the firm Optimal query
plan may choose the most expensive mistake. This takes
precedence over the lack of knowledge of the operational
environment [12].

In 2003 the researcher Yuan Wei and others discuss
produce a study on the extraction using real-time updating of
data and strategies on demand in DRTDB and the definition of
certain laws to choose the best policy of modernization. Based
on these laws, the researchers suggested an algorithm to derive

the updated data, the derivation policy of modernization
practical data sets automatically [17].

In 2005 the researchers Broheedi Marcos and steen Andler
illustrate how to bring forward the requirements in the
DARTDBS. It is possible to use a model requirements of the
modalities of information with RT[3].

In 2006 also provided a researcher Benoi Ravindran and
others Where they distributed scheduling algorithm Call CUA.
The parameters indicated it would satisfy for Thread time when
there is failure. Algorithm is the Best-Effort and the Thread of
the highest importance when they arrive at any time be the
possibility of implementing a very high [11].

In 2008 the researcher Alexander Zharkov discuss how to
use the material offers Materialized Views in DRTDBMSs. The
researcher offers an algorithm for building dynamic and
evaluation of the material cost. President difference this
algorithm from its predecessors is taken into consideration the
characteristics of time Temporal Properties of relations
president and data processing [21].

III. CONTRIBUTIONS

ADRTDBS is a real time distributed database management
system prototype that is designed to support distributed
transactions processing with timing constraints.

The ADRTDBS offers many contributions listed below:

• Database in main memory: disk access should be minimize
in a RTS, since reading from disk is both unpredictable and
orders of magnitude slower than memory access.
ADRTDBS is built to keep the entire database resident in
main memory.

• Full Replication: times for network messages are
unpredictable, and accessing data on remote nodes in much
slower than local access. So ADRTDBS employs a full
replication scheme which ensures that local replicas exist
for all objects accessed by a transaction removing the need
for remote object access.

• ADRTDBS Design and Implementation:

■ A structure to add support of executing real time

transactions in distributed environment.

■ Providing scheduling algorithm named EDTDF

Earliest Data or Transaction Deadline First for
Transaction execution.

■ Produce an approach for concurrency control by using

2PL (Two Phase Locking) protocol to managing
concurrent execution of transactions.

■ Execute data shipping and transaction shipping.

■ Provide synchronize replication and synchronization
updating by using 2PC (Two Phase Commit).

■ Provide backup and recovery approach to process
failure.

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

IV. THE PROPOSED SYSTEM

Given the important developments in computer and
software industry databases and the increasing use in different
areas of life (such as the management of banks, libraries,
companies, factories ... etc.) and because of its great importance
in a systematic compilation of data and processing, updating
and retrieval with pinpoint accuracy, speed and the urgent need
to provide such techniques in our country to keep pace with this
development software tremendous invaded the whole world,
this system was built to be a first step in the application of
modern techniques and contemporary distributed database
environment in real time. It was named Approach for designing
Distributed Real Time Database System (ADRTDBS). The
system ADRTDBS deals with Homogeneous distributed
databases (i.e. all computers linked to the network is made of
the same company (Pentium IIII) and contains the same version
of the operating system (Windows XP) as well as containing
the same version of the database management system ( DBMS
Oracle 9i), the same version of the program interfaces
(Developer 6i). Has the capacity to implement Soft Real Time
Transactions.

V. SYSTEM ARCHITECTURE

The system architecture consists of the following structure
shown in the figure (1):

-a

Server 1 I II Server 2 |

:3f

Figure (1) Architecture of ADRTDBS System

It contains two computers working as server having same
database, and two computers working as clients connected with
each server. Connection between computers is via HUB. The
database resident in each server connected with the network
and having same data and structure (replicas). The clients
contain interfaces that making connection with servers and
retrieve updating database. Any transaction will be executed on
the database will have one of two cases: either local transaction,
the implementation is the only current computer server. Or

Global transaction, the implementation to all servers linked in
the network.

VI. SYSTEM MODEL
• Database Model

A real time distributed system consists of two autonomous
computers system (sites) connected via a communication
network. Each site maintains a copy of database. In order for
transactions to be applied consistently to all replicas and give a
result within deadline time, a prototype units runs at each site.
Also this prototype architecture gives the distributed nature and
the increased communication burden of such a database system.

The smallest unit of data accessible to the user is called data
object. In this distributed database system with replicated data
objects a logical data object is represented by a set of one or
more replicated physical data object. The database is fully
replicated at all sites. The database consists of two types of data
objects: temporal and non-temporal. Temporal data object are
used to record the state of the object in the external
environment which its value changes frequently with time.

- Shipping Approaches

Two approaches for processing transactions in a
ADRTDBS system: query shipping and data shipping.

• Data Shipping

In the data shipping approach, a transaction initiated by a
client will be processed at the client. While the transaction is
processing, the client sends data requests, which are required by
the transaction, to the database server. The server responds to
the requests by sending the required data objects to the client.
The processing of the transaction will be completed at the
client.

• Query Shipping

In the query shipping approach, the client sends queries to
the database server for the transaction, instead of data requests.
Once the server receives a query, it processes the query and
sends the results back to the client. In the query shipping
approach, the communication cost and the buffer space required
at the client side are smaller than that in the data shipping
approach. Also, the query shipping approach provides a
relatively easy migration path from an existing single-site
system to the client-server environment since the database
engine can have a process structure similar to that of a single-
site database system. On the other hand, the data shipping
approach can off-load functionality from the server to the
clients. This may improve the scalability of the system and
balance the workload in the system. Figure (2) illustrates
flowcharts for query shipping from the point of view for server
and client.

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

| S«uJ. mue l&Ajay-tXf TTl*= Sg^qr |

C^Z^ 1

Figure (2) : a- Query Shipping in Server, b- Query Shipping in Client

• Transaction Model

A transaction is a sequence of operations that takes database
from a consistent state to another consistent state. Two types of
transactions are used in this proposed system: query
transactions and update transactions. Query transactions consist
only of read operations that access data object and return their
values to the user. Thus, query transactions do not modify the
database state. Update transactions consist of both read and
write operations.

A transaction Ti in this proposed system characterized by
the following attributes: Ti = ( r i9 we i5 rd } , p } )

r z : released time for the transaction, which represent the

arrival time,
we,: the estimated worst case execution time.
rd { : the relative deadline, it indicates the requirement to

complete the transaction before the instant deadline.
Pi: the priority, of transaction, which depend on the
transaction relative deadline.

VII. ADRTDBS SYSTEM UNITS
An ADRTDBS system capable of executing transactions
with timely constraint in distributed environment. The system
consists of ten of working units, and each server contains copy
of program for these units. Figure (3) illustrate the prototype of
the ADRTDBS system.

• Transaction Admission Control (TAC) Unit:

This unit receives transaction request from database
servers and clients. This unit provides a database interface to
the application. This interface consists of a data manipulated
language, in which the user (application) can query and
manipulate data elements.

• Index Management (IM) Unit-:

It is used to maintain an index for all tuples in the
database. It is capable of transforming a database key into the
memory address of the tuple correspondent to the database
key.

• Memory Management (MM) Unit

This unit is responsible for memory allocation of tuples
and database indexes.

• Transaction Management (TM) Unit -:

This unit responsible of managing transactions coming
from admission control unit TAC and transmit it to scheduler
unit TS to schedule them according to the proposed algorithm.
This unit provides required data to each transaction and
controls and assembles results for each request. This unit also
controls and manages other units and calculate deadline for
each transaction.

A deadline function computes the execution time for each
transaction and predicts the deadline according to the
following equation:

TD = RL (T) + Pr_Ex(T) * SF

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Figure (3) Units of Proposed ADRTDBS System

Where

RL (T) : release time for transaction T
Pr_Ex(T) : Predict execution time for T and this time can
be computed by
Pr_Ex(T) = (T_operation + T_update) * N_op + T_cc
Where

T_operation : time to process an operation
T_update : time to update data
N_op : number of operation
T_cc : communication cost
SF : slack factor 15 >= SF >= 1

• Transaction Scheduler (TS) Unit

This is responsible for scheduling transactions. This unit
maintains the list of transactions in a ready queue and releases
the next transaction when the previous is completed and give it
to the CPU. The ready queue is organized according to the
transaction priority. Each transaction is characterized by a

deadline which defines its urgency with respect to the other
transactions of real time application. The higher priority is
given to transaction with minimum deadline according to the
scheduling algorithm EDTDF (Earliest Data or Transaction
Deadline First). If the system cannot complete transaction
before its deadline, the transaction is aborted.
The algorithm is work like this:

Receive transaction from TM unit.
Determine if the transaction contains temporal data.
Define the deadline for the temporal data (DD).
Compute the deadline for transaction (TD).
Compute the Final Transaction Deadline (FTD) by
if the transaction contain temporal data then

FTD = Min(TD,DD)
else FTD = TD
Put the transaction in the ready queue with
transaction with earliest deadline at the head of the
queue.

1.
2.
3.
4.
5.

» Transaction Deadline Test (TDT) Unit

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

This unit is responsible of decision that a transaction is
aborted whenever it is found to have missed its deadline. So,
for each transaction periodically checks whether or not the
transaction will be able to meet its deadline taking into
consideration the fact that the transaction has to update the data
object in its write-set at each database. If the system's current
time plus the time to update all data objects in a transaction's
write-set is greater than the transaction's deadline, it means that
this transaction will not be able to commit before its deadline is
reached. In order not to be waste any system resources, the
transaction will be aborted and removed from the system.
Figure (4) illustrate the work of this unit.

Fjw&rYftTmis'iCTMi from Queua-wili
IJiiIiiijim De-vlhii*

L+iYipm

ite- Current lutein Tin*

H-uinuin Transaction Tim* = Current tine- +
Ei«ut mi Tine- + Up'hte- Tun*

Hv. inu in Transaction Time*

^*»»- Transaction Dwdlii*

.rtbort Tmisiccion

lis Dwdliie- Couute-i .1H:D.= 1II:D+1

ExecutaTnns-iction

Fi»l

Figure (4) Transaction Deadline Test unit

• Concurrency Control (CC) Unit

This unit is responsible of synchronous execution for
more than one transaction that require execution on same
database at same time. In this work a 2PL (Two Phase
Locking) protocol was used to control concurrency. 2PL
protocol follows the following three rules:
- When the protocol receive a lock request, it tests whether the

requested lock conflicts with another lock that is already set.

If so, it queues the lock request. If not, it responds to the lock

request by setting the lock.

- The 2PL protocol locks a data item only once, it cannot
release the lock until the DM (Data Manager) has completed
processing of the lock's corresponding operation.

- The 2PL protocol releases a lock for a transaction, it may not
subsequently allow any lock for the same transaction.

• Network Management (NM) Unit

This unit is responsible of managing the transfer of data
between servers Depending on the TCP/IP Protocol which
consider the best protocol that provide high speed for sending
and receiving data over the network.

Database server communicates among each other via 1-to-n
communication which is consider as group communication.
Reliable broadcast parameter of this communication ensures
that a message sent by a correct database server, or delivered
by a correct database server, is eventually delivered to all
correct database servers.

• Replication Control (RC) Unit

This unit controls all updating on the local database. This
unit broad changes on database to remain copies of database on
servers connected by network in synchronizing manner by
using 2PC (Two Phase Committing).

Replication requires to have a specific site - the main copy
- associated with each data item. The clients must send their
requests to one particular server. This server is the main copy.
Because there is only one server executing the transactions,
there are no conflicts across the servers. Any update to the data
item must be first sent to the main copy where it is processed.
The main copy then propagates the update (or its results) to all
other sites. This approach is used in ADRTDBS system to
minimize conflicts among transactions executed over replicated
data.
The steps for this technique are the following:

1. The transaction starts at the primary copy site.

2. Read operations are executed locally.

3. The result of write operations are broadcast to the other

sites, (backups), (i.e. update every where).

4. The main copy site starts the Two Phase Commitment
Protocol (2PC).

5. The transaction is committed on all sites.

•Recovery and Backup (RB) Unit

This unit makes backups of database and recovering it when
required. This unit maintains information on database when an
error occur on computer or database or when we need copy the
database on more than one computer. The famous manner of
back up and recovery is export and import.
The steps of export and import are illustrated in figure (5).

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Ifrrior CtfriWiiaTul

Xton^t

I Erd^E" C CiXlTltlXul. bwij't

Eriio: I£i^(xriiuaxLaiiu^4f jvciri

SpcoCy £0.c:Lcic:xi.ciTL:iTLil£i±9i£kiTL= DMT

Choose Eifort Entire Datakue Or Ufa:* Or TiHe* |

IfrrieriiYiEoitiiFeErL»ftje^a£Fvwi4.

Sp «i£y GLe IxicaAfxii ami EdHuaim = JXMT

Erii^r Bi-rflar Sue

Ouoos e Content* rf IH.e Orilv

■ T ,

Choose Create Enroi: >lu.e to Olject Existence

1
I Specify Imfc-xt Grant* Or Not I

SfHiCy Imfc-xt Tal-le Ddi Or Hot

I Sjiedfy frnjioxt EritK'e Ejifoit iuc

Figure (5) : a- Export Data algorithm, b- Import Data algorithm

VIII. SYSTEM TEST
The system has been tested on Alrafidain bank. Many
transactions were take into consideration, (add customer , close
customer account, draw money, transfer funds, currency
change, etc).

. Transfer Fund Transaction: If we transfer fund between
customers the system demand many steps illustrated in the
following figure (6). The figure also shows that every step (or a
number of steps) of the unit within a system ADRTDBS. The
time parameters for this transaction are illustrated in table (1).

Table (1) Deadline Computation of Transfer Fund Transaction

No of Tables That Effected

No of Record That Effected

No of Fields That Effected

Type of Operation

Updating & Adding

T_operation

0.2778 mill sec.

T_update

0.1388 mill sec.

T_cc

2 mill sec.

N_op

6 mill sec.

Pr_Ex(T) = (T_operation + T_update) *
N_op+ T_cc

4.5 mill sec.

RL(T)

mill sec.

2.8

TD = RL(T) + Pr_Ex(T) * SF

12.6 ~ 12 mill sec.

IX. PERFORMANCE EVALUATIONS
In conventional distributed database systems, performance
is primarily measured by the number of transactions completed
within a unit time. In distributed real time database systems,
timing and criticality characteristics of transactions must be
taken into account. So performance depends on many other
criteria, which are related to real time. Some of these criteria
are the number of transactions that missed their deadline,
average tardy time, etc. In this work, the performance metric
employed is the percentage of transactions that missed their
deadlines (%miss) in the total number of transactions that were
submitted to ARTDDBS system during the session period :
Miss ratio = No. of missed deadline transactions/
Total transactions * 100
Also we measure the total CPU utilization. And develop
performance measurement take into consideration database
reside in main memory consisting of 72 organism data. (Table
2) shows the model parameters and their baseline values.

We take a sample of 70 transactions of this application
distributed as (add new customer, updating customer
information, close account, deposit money, query about
account, transfer fund, display customer information). The time
of execute this transactions are 293 millisecond, and the
transactions missed their deadline 14.285%.
Miss Ratio = 10/70*100 = 14.285

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Figure (6) : Transfer Fund Execution in ADRTDBS

Figure (6) : Cont. Transfer Fund Execution in ADRTDBS

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Table (2) Performance Evaluation

Parameter

Base line value

System

Number of servers

Number of clients for each server

Communication cost

2 mill sec.

Database

Number of local databases in each site

1 Database

Number of objects in local database in
each site

11 tables, and 5 views

Database size

16 data object per local
database

Concurrency control

2PL (two phase locking)

Fraction of temporal data object

0.1

Transaction

Transaction size

3 to 5 operation uniform
distributed

Proportion of write operations

0.35

Slack Factor Range

1...15 the slack factor is
uniformly distributed in the
slack range (we use 2.8)

CPU

CPU scheduling

EDTDF (earliest data or
transaction deadline first)

CPU time to process an operation

1... 6 mill sec.

CPU time to commit changes on
database

1 mill sec.

CPU time to rollback changes on
database

1 mill sec.

CPU time to update a data object

6 mill sec.

The figure (7) illustrates measuring of Miss Ratio of this
system and figure (8) illustrates the CPU utilization.

~7 14 ^T~

Figure: (7) Miss Ratio of this system.

no. of transaction

Figure (8) CPU utilization

CONCLUSIONS AND FURTHER WORKS
In this paper a model of proposed distributed real-time
database system was designed, as this system has the ability to
execute real time transactions in distributed environment. The
proposed system uses full replication method, replication of the
same database on all sites.

The replication of the whole database at every site in this
proposed system improves availability remarkably because the
system can continue to operate as long as at least one site is
functioning well. It also improves performance of retrieval for
global queries, because the result of such a query can be
obtained locally from any site, hence a retrieval query can be
processed at the local site where it is submitted, if that site
includes a server module. Replication of data from the remote
site to the local site makes the data available on the local site
and minimizes the response execution time which very suitable
to distributed real time databases environment. Also, by
maintaining multiple copies, it becomes possible to provide
better protection against corrupted data.

using 2PL (Two Phase Locking) protocol help to control
synchronize execution of transactions in this proposed system
in two cases:

- When there is a request to execute same transactions from
some clients on one server at same time.

- When there is a request to execute same transactions from
some servers on one server at same time.

As a further work we suggest to use new encryption
algorithms to increase the security of system. Use another
distributed manner of database like hybrid and then see how it
is compatible with distributed real time database.

REFERENCES

[1] Al-Kinany, E. A., "Heterogeneous System Design for

Distributed Database Systems", Msc, Alrasheed College,

University of Technology, 2005.

[2] Bouzefrane, S. S.; Kaiser, C, "Distributed Overload Control

for Real Time Replicated Database Systems", Cedric

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Laboratory, Conservatoire National des arts et Metiers,
FRANCE, 2002.
[3] Brohede, M; Andler, S. F., "Using Distributed Active Real
Time Database Functionality in Information Fusion
Infrastructures", University of Skode, SE-54128, Sweden.
http://cgi.omg.org/docs/ptc/01-08-34.pdf, 2005.
[4] Coulouris, G.; Dollimore, J.; Kindber g, T., "Distributed
Systems Concepts and Design", Fourth Edition, ADDISON-
WESELY PUBLISHING, 2005.
[5] Deborah, S.; Henry, R., "Oracle9i Net Services
Administrator's Guide", Release 2 (9.2), Oracle
Corporation, 2002.
[6] Deng, G.; Schmidt, D. C, Nechypurenko A.; Gokhale, A.,
"Developing Distributed Real Time and Embedded Systems
with Modeling Tools and component Middleware: A Case
Study", Department of EECS, Vanderbilt University,
Siemens Corporate Technology, Germany, 2005.
[7] Graham, M. H., "Issues in Real Time Data management",
Technical Report, CMU/SEI-91-TR-017, ESD-91-TR-017,
1991.
[8] Lindstrom, J.; Niklander, T.; Raatikainen, K., "A
Distributed Real Time Main Memory Database
Telecommunication",Department Computer Science,
University of Helsinki.

www.imtrg.me.metu.tr/publications/paper.pdf, 1998.
[9] Peddi, P.; Dipippo, L. C, "A Replication Strategy for
Distributed Real Time Object Oriented Database",
Proceedings of the Fifth IEEE International Symposium
on Object-Oriented Real-Time Distributed Computing,
Washington, D.C., 2002.
[10] Ramamritham, K., "Real Time databases", International
Journal of Distributed and Parallel Databases 1(2), 199-
226, 1993.
[11] Ravindran, B.; Anderson, J. S.; Jensen, E. D., "On
Distributed Real Time Scheduling in Network Embedded
Systems in the Presence of Crash Failures", ECE Dept.
Virginia Tech, VA 24061, USA, The MITRE
Corporation, Bedford, mA01730, USA, 2006.
[12] Shanker, U., " Some performance issues in distributed real
time database systems", M.sc, Department of Computer
Science, University of Virginia, TesiOnline. USA, 2000.
[13] Soparker, N.; Lery E.; Korth, H. F.; Silberschatz, A.,
"Adaptive Commitment for Distributed Real Time
Transactions", Work Partially Supported By NSF Grant
IRI-8805215, and by a Grant from IBM Corporation,
USA, 1994.
[14] Syberfeldt, S., "Optimistic Replication with forward
Conflict Resolution in Distributed Real Time Database",
Ph.D., Department of Computer and Information Science,

Linkopings University, Sweden. Printed by LiU-tryck,
Linkoping, 2007.

[15] Tao, J.; Williams, J. G., "Concurrency Control and Data
Replication Strategies for Large Scale and Wide
Distributed Database",University of Pittsburgh, IEEE, 0-
7695-0996-7/01,10.00, 2001.

[16] Wei, Y.; Prasad, V.; Son, S. H., "Qos Management of Real
Time Data Stream Queries in Distributed Environments",
Department of Computer Science, University of Virginia.
www.cs.virginia.edu/~radb/sch.html, 2006.

[17] Wei, Y.; Son, S. H.; Stankovic, J. A ., " Maintaining Data
Freshness in Distributed Real Time Databases", USA,
This Work was Supported in Part by NSF Grant IIS-
0208758,CCR-032,60, and CCR 0098269, 2003.

[18] Wolfe, V. F.; Black, J.; Thuraisingham, B.; Krupp, P.,
"Real Time Method Invocation in Distributed
Environment", USA, The MITRE Corporation Before
MA,USA, This Work is Partially supported by the US
National Science Foundation, Us NAVAL Undersea
Warfare Center, 1994.

[19] Woochul K., Sang H., and John A., "DRACON: QoS
Management for Large-Scale Distributed Real-Time
Databases", JOURNAL OF SOFTWARE, VOL. 4, NO. 7,
2009.

[20] Zharkov, A., "Performance Evaluation of Transaction
Handling Policies on Real Time Database Management
System Prototype", Proceeding of the Spring Yong
Researcher's Colloquium On Database and Information
Systems SYRCODIS, MOSCOW, Russia, 2007.

[21] Zharkov, A., "On Using Materialized Views for Query
Execution in Distributed Real Time Database
Management System", Proceedings of the Spring Young
Researcher's Colloquium On Database and Information
Systems SYRCoDIS, St. -Petersburg, Russia, 2008.

Dr. Dhuha Basheer Abdullah Albazaz /Asst.
Prof / computers Sciences Dept. / College of
Computers and Mathematics / University of
Mosul. She has a Ph.D. degree in Computer
Sciences since 2004. Specific Specialist in
Computer Architecture and Operating System.
Supervised many Master degree students in
operating system, computer architecture, dataflow machines,
mobile computing, real time, distributed databases. She has
three Phd. Students in FPGA field, distributed real time
systems, and Linux clustering. She also leads and teaches
modules at both BSc, MSc, and Phd. levels in computer
science. Also she teaches many subjects for Ph.D. and master
students.

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

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

Vol 8, No. 6, September 2010

Person Identification System using Static-dyamic

Signatures Fusion

Dr. S.A Daramola

Department of Electrical and Information Engineering

Covenant University

Ota, Nigeria

Prof. T.S Ibiyemi

Department of Electrical Engineering

University of Ilorin

Ilorin, Nigeria

Abstract — Off-line signature verification systems rely on static
image of signature for person identification. Imposter can easily
imitate the static image of signature of the genuine user due to
lack of dynamic features. This paper proposes person identity
verification system using fused static-dynamic signature features.
Computational efficient technique is developed to extract and
fuse static and dynamic features extracted from offline and online
signatures of the same person. The training stage used the fused
features to generate couple reference data and classification stage
compared the couple test signatures with the reference data
based on the set threshold values. The system performance is
encouraging against imposter attacker in comparison with
previous single sensor offline signature identification systems.

Keywords- fused static -dynamic signature; feature extraction;
forgeries

I. Introduction

Person identity verification is a problem of authenticating
individual using physiological or behavioral characteristics
like face, iris, fingerprint, signature, speech and gait. Person
identification problem can be solved manually or
automatically. Biometric systems automatically use biometric
trait generated from one or two sensors to validate the
authenticity of a person. Automatic person identity
verification based on handwritten signature can be classified
into two categories: on-line and off-line, differentiated by the
way signature data is acquired from the input sensor. In off-
line technique, signature is obtained on a piece of paper and
later scanned to a computer system while in on-line technique,
signature is obtained on a digitizer thus making dynamic
information like speed, pressure available while in offline only
the shape of the signature image is available [1] [2].

In this paper, combination of offline and online signatures
are used for person identification. The process involves
verification of a signature signed on both paper and electronic
digitizer concurrently. Therefore the physical present of the
signer is required during the period of registration and
verification. This type of system is useful particular in the
bank while the physical present of the holders of saving
current are required before money can be withdrawn. In
Nigeria many banks manually identify holder of saving current
using face and static signature, in the process genuine users

are rejected as imposters because of high intra-class variation
in signatures. Frauds as result of signature forgeries must be
prevented particular among closely resemble people. Fusion
of dynamic and static signature will strengthen the
identification of people physically in paper documentation
environment.

A detailed review of on-line signature verification
including summary of off-line work until the mid 1990's was
reported in [1] [2]. Alessandro et al. [3] proposed a hybrid
on/off line handwritten signature verification system. The
system is divided into two modules. The acquisition and
training module use online-offline signatures, while the
verification module deals with offline signatures. Soltane et al
[4] presented a soft decision level fusion approach for a
combined behavioral speech- signature biometrics verification
system. And Rubesh et al [5] presented online multi-parameter
3D signature and cryptographic algorithm for person
identification. Ross et al [6] presented hand-face multimodal
fusion for biometric person authentication while Kiskus et al
[7] fused biometrics data using classifiers.

From the approaches mentioned above, some of the
authors used online signature data to strengthen the system
performance either at registration or training stage, while
others combined online signature data with other biometric
modalities data like speech, face as means of person
identification. The system proposes in this novel frame work
is based on fusion of static and dynamic signature data at
feature level for person identification. The universal
acceptance of signature, compatibility of offline and online
signature features make the proposed system more robust,
accurate and friendly in comparison with other previous multi
biometric modalities systems or single sensor offline system
for person identification.

Section 2 provides the description of the system, the
signature preprocessing and feature extraction and fusion
technique. Also in section 2, the signature training, threshold
selection and classification are presented. Section 3 shows the
experimental results and finally, conclusions are drawn in
section 4.

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

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

Vol 8, No. 6, September 2010

II. PROPOSED SYSTEM

The system block diagram is shown in Fig. 1 . The offline
and online data are collected at the same time from the same
user during registration/training and verification exercises.
Also the offline signature are preprocessed to remove
unwanted noise introduced during scanning process whereas,
the online signatures are not preprocess in order to preserve
the timing characteristics of the signature. Discriminative
static and dynamic features are extracted separately from
offline and online signature respectively. At the feature level
the two signatures are fused together to obtain a robust static -
dynamic features. These features are used to generate couple
reference data during training and for signatures classification.

A. Data Acquisition

The signature database consists of a total number of 300
offline and 300 online handwritten genuine signature images
and 100 forged signatures. The genuine signatures are
collected from 50 people. Each of the users contributed 6
offline and 6 online signature samples. The 100 skilled
forgeries consist offline and online signatures, they are
collected from 25 forgers and each of the forgers contributed 4
samples. The raw signature data available from our digitizer
consists of three dimensional series data as represented by (1).

S(t) = [x(t),y(t),p(t)] T t = 0,1,2,.

(1)

where (x(t), y(t)) is the pen position at time t , mdp(t)
{0,1,... ,1024} represents the pen pressure.

Input offline
signature on paper

Input online
signature on digitizer

Preprocessing

Offline signature
feature extraction

On-line signature
feature extraction

Fusion of online and offline features

Couple test feature

Training

Couple reference feature

Classification stage

Result

Figure 1: System block diagram

B. Offline Signature Preprocessing

The scanned offline signature images may contain noise
caused by document scanning and it has to be removed to
avoid errors in further processing steps. The gray-level image
is convolved with a Gaussian smoothing filter to obtain a
smoothed image. The smoothed gray image is converted into
binary image and then thinned to one pixel wide.

C. Offline Feature Extraction

The feature extraction algorithm for the static signature is
stated as follows:
(1) Locate signature image bounding box.

(i) Scan the binary image from top to bottom to

obtain the signature image height.

(ii) Scan the binary image from left to right to obtain

the signature image width.
(2) Centralization of the signature image.

(i) Calculate centre of gravity of the signature image

using (2).

1 ^

^ N

(2)

(ii)Then move the signature image centre to coincide
with centre of the predefined image space.

(3) The image is partitioned into four sub-image parts.

(i) Through point X make a horizontal splitting

across the signature image.

(ii) Through point y make a vertical splitting across

the signature image.

(4) Partition each of the sub-image parts into four rectangular
parts.

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

(i) Locate the centre of each of the sub-image parts
using (2).

(ii) Repeat step 3(i) and 3(ii) for each of the sub-
image parts in order to obtain a set of 16 sub-image
parts.

(5) Partition each of the 16 sub-image parts into four signature
cells

(i) Locate the centre of each of the sub-image parts
using (2).

(ii) Repeat step 3(i) and 3(ii) for each of the sub-
image parts in order to obtain a set of 64 sub-image
cells.

(6) Calculate the angle of inclination of each sub-image centre
in each cell to the lower right corner of the cell.

(i) Locate the centre of each of the 64 sub-image

cells using (2)

(ii) Calculate the angle that each centre point makes

with the lower right corner of the cell.
The feature extracted at this stage constitutes the offline
feature vector, which is represented as: F = f b f 2 , f 3 ,

f 4 f 64 . The details and diagrams of the feature extraction

are given in [8].

D. Online Signature Extraction

Three on-line signature features are extracted at each
sampling point from the raw data. The features are Ap/Ax,
Ap/Ay and v. Ax corresponds to change of x between two
successive sampling points, Ay corresponds to change of y
between two successive sampling points, Ap corresponds to
change of p between two successive sampling points, Ap/Ay
corresponds to ratio of Ap to Ay, Ap/Ax corresponds to ratio
of Ap to Ax and v corresponds to change of speed between
two successive sampling points [9]. These features are
obtained using (3), (4) and (5).

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

Vol 8, No. 6, September 2010
done at decision level. While in [6] fusion of hand and face
was done at feature level. In this work offline feature is
combined with online feature at feature level. The
compatibility of the signature data from the same person, from
different sensors made the fusion possible without any lost of
information. The steps involve are stated as follows: given that

extracted static feature is F = f b f 2 , f 3 f 4 f 64 . The mean

and variance of the feature vector are calculated using (6) and
(7) respectively

v = JiAxf+(Ay) 2

Ap = pjt)-pjt-\)
Ay y(t)-y(t-l)

Ap = pit)- pit -\)

Ax x(t) - xit - 1)

(3)

(4)

(5)

I N

(6)

(7)

The fused features are obtained by normalized each of the

, Ap Ap x
extracted online features (v, , ) using the variance

Ay Ax

of the offline feature (& ofr ) • The three fused features (SF1,

SF2 and SF3) become:

Ap Ap

<j rr Aver r, Axcj

off j off (

off

F. Training and Threshold Setting

Each of the registered users submitted 12 genuine
signatures to the system, out of which 8 signatures are fused
together to generate 4 couple reference features. These
features are used to generate 6 distance values by cross-
aligned the couple reference features to the same length using
Dynamic Time Warping (DTW). These distance values are
used to measure the variation within each of the user's
signatures, so as to set user-specific threshold for accepting or
rejecting a couple test signatures. Given four couple reference
signature samples Rl, R2, R3 and R4, these features are cross
aligned to obtain 6 distance values as shown in Fig.2. The
mean (m k ) and standard deviation (o k ) of the distances: d i2 ,
di3, di4, d 2 3, d 24 and d 34 are calculated and used to set the
threshold (t k ) for each of the users based on each of the fused
features as given in (8).

0>t k <m k +a k

(8)

E. Fusion of Offline and Online features

This technique is designed to compute fused feature
vector, which contains information from both the offline and
online signatures and used this feature vector for subsequence
processing. Information from two input sensors can be
combined at data acquisition stage, feature extraction stage or
at decision level stage. The accuracy of the system also
depends on the level of fusion and the discriminative ability of
fused data. In [4] the fusion of voice and signature data was

Figure2. Cross-alignment of Couple reference

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

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

Vol 8, No. 6, September 2010

G. Classification of Couple Signature Features

Whenever a couple test signatures (offline and online)
come into the system, the fused feature vector of the couple
test signatures is pair-wise aligned with each of the four
couple reference features using DTW. Four distance values are
obtained as shown in Fig.3. The distance (d t ) of the couple test
feature (FT) from the four couple reference features Rl, R2,
R3 and R4 is calculated using (9).

TABLE 1: OFFLINE FEATURES IN COMPARISON WITH THE
PROPOSED FUSED FEATURES.

i f _ JT\ + JT2 + JT3 + JT4

(9)

If d t is within the assigned threshold value then the fused test
signature is assigned a pass mark otherwise it has no pass
mark. Finally a decision by the system in accepting or
rejecting a couple test signatures is based on total pass mark it
obtained based on the three fused features.

III. Experimental Results

Experiments have been conducted to evaluate the
discriminative ability of each of the fused features against
forgers attack. Also the proposed system is tested based on the
three new fused features. Total number of 150 fused signatures
made up of 100 genuine signature features and 50 skilled
forgery features are collected from 75 people are tested. The
performance evaluation is based on False Acceptance Rate
(FAR) and False Rejection Rate (FRR). Table 1 shows the
results of the performance of these fused features in
comparison with previous single offline features. Table 2
shows the proposed system FAR for skilled forgeries and the
FRR for genuine signatures.

Fused test features (FT)

Type

Feature

FRR

FAR

Some

offline

features in

[8][10]

Pixels normalized angle relative
to the cell lower right corner

1.250

2.500

Image centre angle relative to
the cell lower right corner

2.500

Vertical centre points

7.500

8.750

Horizontal centre points

6.250

7.500

Proposed
fused
offline-
online
features

SF1

0.150

0.120

SF2

0.120

0.052

SF3

0.100

0.080

TABLE2: FAR AND FRR RESULTS OF THE PROPOSED SYSTEM

Type

Total

Accepted

Rejected

FAR

FRR

Individual
genuine
fused
features

100

0.05

Individual
skilled fused
forgeries

0.02

Figure 3. Distance between couple reference features and couple
test feature

IV. CONCLUSION

This paper has proposed a person identification system
using fused signature features from two biometric sensors.
Fused signature feature is used to strengthen verification
system in paper documentation environment like banks where
the present of the account holders are required for transaction.
Signature is universally accepted, this make the proposed
system more friendly and acceptable in comparison with

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

others biometric traits combination. The experimental results
have shown that fused signature identification method is more
accurate in comparison with previous single sensor offline
signature identification techniques.

References

[1] R. Plamondon and S.N. Srihari "On-line and off-line handwriting
recognition: a comprehensive Survey", IEE trans, on Pattern Analysis and
Machine Intelligence, vol. 22, No.l, pp. 63-84, 2000.

[2] F. Leclerc and R. Plamondon, "Automatic verification and writer
identification: the state of the art 1989-1993", International Journal of Pattern
Recognition and Artificial Intelligence, vol. 8, pp. 643 - 660, 1994.

[3] A. Zimmer and L.L. Ling "A hybrid on/off line handwritten signature
verification system", Proc. of the seventh International Conference on
Document Analysis and Recognition, 2003.

[4]. S. Mohamed, G. Noureddine and D. Noureddine "Soft decision level

fusion approach to a combined behavioral speech-signature biometics

verification", International journal of Biometrics &Bioinformatics, vol. 4,

issue 1.2009.

[5] P.M. Rubesh, G. Bajpai and V. Bhaskar "Online multi-parameter 3D

signature verification through Curve fitting", International Journal of

Computer Science and Network Security, vol.9, No. 5, pp 38-44. 2009.

[6] A. Ross and R. Govindarajan "Feature level using hand and face

biometrics", Proc. of SPIE Conference on Biometric Technology for Human

Identification, vol. 5779, pp. 196-204, 2005.

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

Vol 8, No. 6, September 2010
[7] D. R. Kisku, P. Gupta and J. K. Sing "Offline signature identification by
fusion of multiple classifiers using statistical learning theory", International
Journal of Security and Its Applications, vol.4, No. 3, 2010.
[8] S. Daramola and S. Ibiyemi "Novel feature extraction technique for offline
signature verification", International Journal of Engineering Science and
Technology, vol (2)7, pp 3137-3143, 2010.

[9] S. A. Daramola and T.S Ibiyemi "Efficient On-line Signature verification"
International Journal of Engineering &Technology, vol 10, No. 4. pp 48-52.
2010.

[10] M. Banshider, R .Y Santhosh and B .D Prasanna "Novel features for off-
line signature verification" International Journal of Computers,
Communications & Control, vol. 1 , No. 1, pp. 17-24. 2006.

AUTHORS PROFILE

Dr. S.Adebayo Daramola obtained Bachelor of Engineering from University
of Ado-Ekiti, Nigeria, Master of Engineering from University of Portharcourt,
Nigeria and PhD from Covenant University, Ota, Nigeria. His research
interests include Image processing and Cryptography.

Prof. T.S Ibiyemi is a Professor in Computer Engineering. He has more than
30 years teaching and research experience; he has many papers in local and
international journals. His research interests include Image processing,
Multimedia and Processors architecture.

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

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

Vol 8, No. 6, 2010

Short term flood forecasting using RBF static neural
network modeling a comparative study

Rahul P. Deshmukh

Indian Institute of Technology, Bombay

Powai, Mumbai

India.

Abstract — The artificial neural networks (ANNs) have been
applied to various hydrologic problems recently. This research
demonstrates static neural approach by applying Radial basis
function neural network to rainfall-runoff modeling for the
upper area of Wardha River in India. The model is developed
by processing online data over time using static modeling.
Methodologies and techniques by applying different learning
rule and activation function are presented in this paper and a
comparison for the short term runoff prediction results
between them is also conducted. The prediction results of the
Radial basis function neural network with Levenberg
Marquardt learning rule and Tanh activation function indicate
a satisfactory performance in the three hours ahead of time
prediction. The conclusions also indicate that Radial basis
function neural network with Levenberg Marquardt learning
rule and Tanh activation function is more versatile than other
combinations for RBF neural network and can be considered
as an alternate and practical tool for predicting short term
flood flow.

Keywords-component; Artificial neural network; Forecasting;
Rainfall; Runoff;

Introduction

The main focus of this research is development of
Artificial Neural Network (ANN) models for short term flood
forecasting, determining the characteristics of different neural
network models. Comparisons are made between the
performances of different parameters for Radial basis function
artificial neural network models.

The field engineers face the danger of very heavy flow
of water through the gates to control the reservoir level by
proper operation of gates to achieve the amount of water
flowing over the spillway. This can be limited to maximum
allowable flood and control flood downstream restricting river
channel capacity so as to have safe florid levels in the river
within the city limits on the downstream.

By keeping the water level in the dam at the optimum
level in the monsoon the post monsoon replenishment can be
conveniently stored between the full reservoir level and the
permissible maximum water level. Flood estimation is very

A. A. Ghatol

Former Vice-Chancellor

Dr. Babasaheb Ambedkar Technological University,

Lonere, Raigad, India.

essential and plays a vital role in planning for flood regulation
and protection measures.

The total runoff from catchment area depends upon
various unknown parameters like Rainfall intensity, Duration
of rainfall, Frequency of intense rainfall, Evaporation,
Interception, Infiltration, Surface storage, Surface detention,
Channel detention, Geological characteristics of drainage
basin, Meteorological characteristics of basin, Geographical
features of basin etc. Thus it is very difficult to predict runoff
at the dam due to the nonlinear and unknown parameters.

In this context, the power of ANNs arises from the
capability for constructing complicated indicators (non-linear
models). Among several artificial intelligence methods
artificial neural networks (ANN) holds a vital role and even
ASCE Task Committee Reports have accepted ANNs as an
efficient forecasting and modeling tool of complex hydrologic
systems[22].

Neural networks are widely regarded as a potentially
effective approach for handling large amounts of dynamic,
non-linear and noisy data, especially in situations where the
underlying physical relationships are not fully understood.
Neural networks are also particularly well suited to modeling
systems on a real-time basis, and this could greatly benefit
operational flood forecasting systems which aim to predict the
flood hydrograph for purposes of flood warning and
control[16].

A subset of historical rainfall data from the Wardha
River catchment in India was used to build neural network
models for real time prediction. Telematic automatic rain
gauging stations are deployed at eight identified strategic
locations which transmit the real time rainfall data on hourly
basis. At the dam site the ANN model is developed to predict
the runoff three hours ahead of time.

In this paper, we demonstrate the use of Radial basis
function neural network (RBF) model for real time prediction
of runoff at the dam and compare the effectiveness of different
learning rules and activation function. Radial basis function
neural network is having a feed-forward structure consisting of
hidden layer for a given number of locally tuned units which
are fully interconnected to an output layer of linear units.

At a time when global climatic change would seem to
be increasing the risk of historically unprecedented changes in

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

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

Vol. 8, No. 6, 2010

river regimes, it would appear to be appropriate that
alternative representations for flood forecasting should be
considered.

II. Methodology

In this study different parameters like learning rule and
activation function are employed for rainfall-runoff modeling
using Radial basis function neural network model of artificial
neural network.

Radial basis functions networks have a very strong
mathematical foundation rooted in regularization theory for
solving ill-conditioned problems.

The mapping function of a radial basis function
network, is built up of Gaussians rather than sigmoids as in
MLP networks. Learning in RBF network is carried out in two
phases: first for the hidden layer, and then for the output layer.
The hidden layer is self-organising; its parameters depend on
the distribution of the inputs, not on the mapping from the input
to the output. The output layer, on the other hand, uses
supervised learning (gradient or linear regression) to set its
parameters.

Figure 1. The Radial basis function neural network

In this study we applied different learning rules to the
RBF neural network and studied the optimum performance
with different activation function. We applied Momentum,
Deltabar Delta, Levenberg Marquardt , Conjugate Gradient,
Quick prop learning rule with activation function Tanh, Linear
Tanh, Sigmoid and Linear Sigmoid.

Performance Measures:

The learning and generalization ability of the estimated
NN model is assessed on the basis of important performance
measures such as MSE (Mean Square Error), NMSE
(Normalized Mean Square Error) and r (Correlation
coefficient)

MSE (Mean Square Error):
The formula for the mean square error is:

MSE = ^°"°

... (1)
Where

P = number of output PEs,
N = number of exemplars in the data set,

ij = network output for exemplar i at PE j,

ij = desired output for exemplar i at PE j.

NMSE (Normalized Mean Square Error):

The normalized mean squared error is defined by
the following formula:

PNMSE

NMSE = -

Z i=0 y/=o
N
Iy ... (2)

Where

P = number of output processing elements,
N = number of exemplars in the data set,
MSE = mean square error,

ij = desired output for exemplar i at processing
element j.

r (correlation coefficient):

The size of the mean square error (MSE) can be used
to determine how well the network output fits the desired
output, but it doesn't necessarily reflect whether the two sets of
data move in the same direction. For instance, by simply
scaling the network output, the MSE can be changed without
changing the directionality of the data. The correlation
coefficient (r) solves this problem. By definition, the
correlation coefficient between a network output x and a
desired output d is:

zU - x \\ d i ~ d

x — — Yx

N
where i=1

... (3)

d=-Yd

and N ^

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

The correlation coefficient is confined to the range [-1, 1].
When r = 1 there is a perfect positive linear correlation between
x and d, that is, they co-vary, which means that they vary by
the same amount.

III. Study Area and Data Set

The Upper Wardha catchment area lies directly in the path
of depression movements which originates in the Bay of
Bengal. When the low pressure area is formed in the Bay of
Bengal and cyclone moves in North West directions, many
times this catchment receives very heavy intense cyclonic
precipitation for a day or two. Occurrence of such events have
been observed in the months of August and September.
Rainfall is so intense that immediately flash runoff, causing
heavy flood has been very common feature in this catchment.

For such flashy type of catchment and wide variety in
topography, runoff at dam is still complicated to predict. The
conventional methods also display chaotic result. Thus ANN
based model is built to predict the total runoff from rainfall in
Upper Wardha catchment area for controlling water level of the
dam.

In the initial reaches, near its origin catchment area is hilly
and covered with forest. The latter portion of the river lies
almost in plain with wide valleys.

The catchment area up to dam site is 4302 sq. km. At
dam site the river has wide fan shaped catchment area which
has large variation with respect to slope, soil and vegetation
cover.

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

Vol. 8, No. 6, 2010
Data: Rainfall runoff data for this study is taken from the
Wardha river catchment area which contains a mix of urban
and rural land. The catchments is evenly distributed in eight
zones based on the amount of rainfall and geographical survey.
The model is developed using historical rainfall runoff data ,
provided by Upper Wardha Dam Division Amravati,
department of irrigation Govt, of Maharashtra. Network is
trained by rainfall information gathered from eight telemetric
rain-gauge stations distributed evenly throughout the catchment
area and runoff at the dam site.

The data is received at the central control room online through
this system on hourly basis. The Upper Wardha dam reservoir
operations are also fully automated. The amount of inflow,
amount of discharge is also recorded on hourly basis. From the
inflow and discharge data the cumulative inflow is calculated.
The following features are identified for the modeling the
neural network .

Table I- The parameters used for training the network

Month I RG1 I RG2 I RG3 I RG4 I RG5 I RG6 I RG7 I RG8 I CIF

Month

Rainl to Rain8

Cum Inflow

- The month of rainfall

- Eight rain gauging stations.

- Cumulative inflow in dam

Seven years of data on hourly basis from 2001 to 2007 is
used. It has been found that major rain fall (90%) occurs in the
month of June to October Mostly all other months are dry
hence data from five months. June to October is used to train
the network

Figure 2- Location of Upper Wardha dam on Indian map

IV. Result

The different structures of neural network are
employed to learn the unknown characterization of the system
from the dataset presented to it. The dataset is partitioned into
three categories, namely training, cross validation and test. The
idea behind this is that the estimated NN model should be
tested against the dataset that was never presented to it before.
This is necessary to ensure the generalization. An experiment is
performed at least twenty five times with different random
initializations of the connection weights in order to improve
generalization.

The data set is divided in to training , testing
and cross validation data and the network is trained for all
models of Radial basis function neural network for 5000
epochs.

The performance results obtain on parameters by
applying learning rules Momentum, Deltabar Delta, Levenberg
Marquardt , Conjugate Gradient, Quick prop with activation
function Tanh, Linear Tanh, Sigmoid, Linear Sigmoid are
listed in Table II through Table VI.

Figure 3- The Wardha river catchment

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

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

Vol. 8, No. 6, 2010
Table II- RBF network performance with Momentum learning rule Table V -RBF network performance with Conjugate Gradient

LEARNING RULE

Param

MSE

Min

Max

eter

MSE

Abs

error

Tanh

0.106

0.124

0.034

0.465

0.534

Linear

0.097

0.105

0.024

0.212

0.639

Tanh

Sigmo

0.089

0.093

0.047

0.421

0.678

Linear

0.094

0.132

0.041

0.381

0.689

Sigmo

Param

eter

MSE

N
MSE

Min
Abs
error

Max
Abs
error

Tanh

0.094

0.165

0.051

0.312

0.646

Linear

Tanh

Sigmo

Linear

Sigmo

0.089
0.092
0.094

0.094
0.134
0.124

0.059
0.041
0.064

0.215
0.474
0.541

0.633
0.701
0.732

Table III - RBF network performance with De l t a b a r Delta

LEARNING RULE

Table VI -RBF network performance with Quick prop. learning

RULE

Param

eter

MSE

N
MSE

Min
Abs
error

Max
Abs
error

Tanh

0.093

0.141

0.051

0.564

0.651

Linear

Tanh

Sigmo

Linear

Sigmo

0.190
0.143
0.086

0.241
0.215
0.095

0.041
0.032
0.067

0.412
0.495
0.315

0.591
0.543
0.603

Table IV- RBF network performance with L.M. learning rule

MSE

Min

Max

eter

MSE

Abs

error

Tanh

0.076

0.064

0.018

0.143

0.854

Linear

0.086

0.094

0.028

0.298

0.732

Tanh

Sigmo

0.083

0.094

0.020

0.228

0.634

Linear

0.089

0.095

0.034

0.469

0.758

Sigmo

MSE

Min

Max

eter

MSE

Abs

error

Tanh

0.133

0.245

0.042

0.465

0.584

Linear

0.169

0.212

0.054

0.514

0.601

Tanh

Sigmo

0.106

0.256

0.059

0.329

0.563

Linear

0.098

0.112

0.046

0.311

0.609

Sigmo

The parameters and performance for RBF model with
different learning rule and activation function are compared on
the performance scale and are listed in the Table VII shown
below. The comparative analysis of the MSE and r (the
correlation coefficient) is done.

Table VII- Comparison of performance parameters

Tanh

Linear Tanh

Sigmoid

Linear
Sigmoid

MSE

Momentum

0.106

0.534

0.0975

0.639

0.089

0.678

0.094

0.689

Deltabar
Delta

0.0931

0.651

0.1906

0.591

0.143

0.543

0.086

0.603

L.M.

0.07629

0.854

0.0861

0.732

0.083

0.634

0.0894

0.758

Conjugate
Gradient

0.0946

0.646

0.0894

0.633

0.0921

0.701

0.0945

0.732

Quickprop

0.1331

0.584

0.1691

0.601

0.106

0.563

0.0986

0.609

After training the network the optimum performance is studied
and it is found that Levenberg Marquardt learning rule and
Tanh activation function produce optimal result. In the Table-
VIII the parameters and the best performances for Radial basis
function neural network are listed.

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

Table VIII- RBF network parameters

Parameter

Performance

MSE

0.07629

NMSE

Min Abs Error

0.06431
0.01943

Max Abs Error

0.14387

0.85437

Fig 4 shows the plot of actual Vs predicted optimum values for
Radial basis function neural network found with Levenberg
Marquardt learning rule and Tanh activation function.

Actual Vs Predicted Runoff by RBF NN Model

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

Vol. 8, No. 6, 2010
have fewer weights, these networks train extremely fast and
require fewer training samples.

V. Conclusion

An ANN-based short-term runoff forecasting system is
developed in this work. A comparison between five different
learning rules with four activation function for optimal
performance for Radial basis function neural network model is
made. We find that Radial basis function neural network with
Levenberg Marquardt learning rule and Tanh activation
function is more versatile than other approaches studied. Radial
basis function neural network with Levenberg Marquardt
learning rule and Tanh activation function is performing better
as compare to other approaches studied as far as the overall
performance is concerned for forecasting runoff for 3 hrs lead
time. Other approaches studied are also performing optimally.
Which means that static model of Radial basis function neural
network with Levenberg Marquardt learning rule and Tanh
activation function is powerful tool for short term runoff
forecasting for Wardha River basin.

-Actual Runoff -

-Predicted Runoff

Figure 4.- Actual Vs. Predicted runoff by RBF for L.M. and Tanh

The error found in the actual and predicted runoff at the
dam site is plotted for RBF network as shown in the Figure 5.

Error in prediction for RBF NN Model

2j£Tl

^S^

24x( (WW

UJ/vA 5

22A//VVM

i^xvA 6

21 1 C3^g^

|fy^| f] ] —♦—error

20 — \^HmI

\^fc^ W 8

is V X* f

\\ VX Yio

17< /\h~

-T\Ml

i>yj

-V 12

Fig 5 - Error graph of RBF Model for L.M. and Tanh

The main advantage of RBF is that it finds the input to output
map using local approximators. Each one of these local pieces
is weighted linearly at the output of the network. Since they

Acknowledgment

This study is supported by Upper Wardha Dam Division
Amravati, department of irrigation Govt, of Maharashtra, India

References

[1] P. Srivastava, J. N. McVair, and T. E. Johnson, "Comparison of process-
based and artificial neural network approaches for streamflow modeling
in an agricultural watershed," Journal of the American Water Resources
Association, vol. 42, pp. 545563, Jun 2006.

[2] K. Hornik, M. Stinchcombe, and H. White, "Multilayer feedforward
networks are universal approximators," Neural Netw., vol. 2, pp. 359-
366,1989.

[3] M. C Demirel, A. Venancio, and E. Kahya, "Flow forecast by SWAT
model and ANN in Pracana basin, Portugal," Advances in Engineering
Software, vol. 40, pp. 467-473, Jul 2009.

[4] A. S. Tokar and M. Markus, "Precipitation-Runoff Modeling Using
Artificial Neural Networks and Conceptual Models," Journal of
Hydrologic Engineering, vol. 5, pp. 156-161,2000.

[5] S. Q. Zhou, X. Liang, J. Chen, and P. Gong, "An assessment of the VIC-
3L hydrological model for the Yangtze River basin based on remote
sensing: a case study of the Baohe River basin," Canadian Journal of
Remote Sensing, vol. 30, pp. 840-853, Oct 2004.

[6] R. J. Zhao, "The Xinanjiang Model," in Hydrological Forecasting
Proceedings Oxford Symposium, 1ASH, Oxford, 1980 pp. 351-356.

[7] R. J. Zhao, "The Xinanjiang Model Applied in China," Journal of
Hydrology, vol. 135, pp. 371-381, Jull992.

[8] D. Zhang and Z. Wanchang, "Distributed hydrological modeling study
with the dynamic water yielding mechanism and RS/GIS techniques," in
Proc. of SPIE, 2006, pp. 63591M1-12.

[9] J. E. Nash and I. V. Sutcliffe, "River flow forecasting through
conceptual models," Journal ofHydrology, vol. 273, pp. 282290,1970.

[10] D. Zhang, "Study of Distributed Hydrological Model with the Dynamic
Integration of Infiltration Excess and Saturated Excess Water Yielding
Mechanism." vol. Doctor Nanjing: Nanjing University, 2006, p. 190.529

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

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

Vol. 8, No. 6, 2010

[11] E. Kahya and J. A. Dracup, "U.S. Streamflow Patterns in Relation to the
EI Nit'lo/Southern Oscillation," Water Resour. Res., vol. 29, pp. 2491-
2503,1993.

[12] K. J. Beven and M. J. Kirkby, "A physically based variable contributing
area model of basin hydrology," Hydrologi cal Science Bulletin, vol. 43,
pp. 43-69,1979.

[13] N. J. de Vos, T. H. M. Rientjes, "Constraints of artificial neural
networks for rainfall-runoff modelling: trade-offs in hydrological state
representation and model evaluation", Hydrology and Earth System
Sciences, European Geosciences Union, 2005, 9, pp. 111-126.

[14] Holger R. Maier, Graeme C. Dandy, "Neural networks for the perdiction
and forecasting of water resources variables: a review of modeling issues
and applications", Environmental Modelling & Software, ELSEVIER,
2000, 15, pp. 101-124.

[15] T. Hu, P. Yuan, etc. "Applications of artificial neural network to
hydrology and water resources", Advances in Water Science, NHRI,
1995, 1, pp. 76-82.

[16] Q. Ju, Z. Hao, etc. "Hydrologic simulations with artificial neural
networks", Proceedings-Third International Conference on Natural
Computation, ICNC, 2007, pp. 22-27.

[17] G. WANG, M. ZHOU, etc. "Improved version of BTOPMC model and
its application in event-based hydrologic simulations", Journal of
Geographical Sciences, Springer, 2007, 2, pp. 73-84.

[18] K. Beven, M. Kirkby, "A physically based, variable contributing area
model of basin hydrology", Hydrological Sciences Bulletin, Springer,
1979, 1, pp.43-69.

[19] K. Thirumalaiah, and CD. Makarand, Hydrological Forecasting Using
Neural Networks Journal of Hydrologic Engineering. Vol. 5, pp. 180-
189, 2000.

[20] G. WANG, M. ZHOU, etc. "Improved version of BTOPMC model and
its application in event-based hydrologic simulations", Journal of
Geographical Sciences, Springer, 2007, 2, pp. 73-84.

[21] H. Goto, Y. Hasegawa, and M. Tanaka, "Efficient Scheduling Focusing
on the Duality of MPL Representatives," Proc. IEEE Symp.
Computational Intelligence in Scheduling (SCIS 07), IEEE Press, Dec.
2007, pp. 57-64.

[22] ASCE Task Committee on Application of Artificial Neural Networks in
Hydrology, "Artificial neural networks in hydrology I: preliminary
concepts", Journal of Hydrologic Engineering, 5(2), pp. 115-123, 2000

Rahul Deshmukh received the B.E. and
M.E. degrees in Electronics Engineering from
Amravati University. During 1996-2007, he
stayed in Government College of Engineering,
Amravati in department of Electronics and
telecommunication teaching undergraduate
and postgraduate students. From 2007 till now
he is with Indian Institute of Technology (IIT)
Bombay, Mumbai. His area of reserch are
artificial intelligence and neural networks.

A. A. Ghatol received the B.E. from
Nagpur university foallowed by M. Tech
and P.hd. from IIT Bombay. He is best
teacher award recipient of government of
Maharastra state. He has worked as
director of College of Engineering Poona
and Vice-Chancellor, Dr. Babasaheb
Ambedkar Technological University,
Lonere, Raigad, India. His area of
research is artificial intelligence, neural
networks and semiconductors.

98 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Analysis of impact of Symmetric Encryption
Algorithms in Data Security Model of Grid Networks

N. Thenmozhi
Department of Computer Science
N.K.R. Govt. Arts College for Women
Namakkal-637 001, India.

M. Madheswaran
Department of Electronics and Communication Eng
Muthayammal Engineering College
Rasipuram-637 408, India.

Abstract— The symmetric and asymmetric encryption algorithms
are commonly used in grid software to provide necessary
security. The use of symmetric encryption algorithm will
significantly affect the network communication performance.

In this paper, the impact of using different popular and
commonly used symmetric key cryptography algorithms for
encrypting data in a typical grid commuting environment is
analyzed. It is obvious that the use of encryption and decryption
at application layer will certainly have an impact in the
application layer performance in terms of speed. In this work,
we have studied its impact at network layer performance in a
typical grid computing environment in the algorithms such as
DES, Triple DES, AES, Blow Fish, RC2 and RC6. The
performances are measured through simulation studies on ns2
by simulating these algorithms in GARUDA Grid Network
Topology.

Keywords- Grid Security; Encryption; ECGIN; ERNET;
GARUDA; PPlive; GridFTP;

INTRODUCTION

Internet and Grid computing applications are growing
very fast, so the needs to protect such applications have
increased. Encryption algorithms play a main role in
information security systems. On the other side, those
algorithms consume a significant amount of computing
resources such as CPU time, memory, and battery power.

The Globus Toolkit is the very commonly used software
for Grid computing. It provides different kinds of security for
grid computing. The Grid Security Infrastructure (GSI) of
Globus and a Public Key Infrastructure (PKI) provide the
technical framework (including protocols, services, and
standards) to support grid computing with five security
capabilities: user authentication, data confidentiality, data
integrity, non-repudiation, and key management.

A Security Issues

Authentication is the process of verifying the validity of a
claimed individual and identifying who he or she is.
Authentication is not limited to human beings; services,

applications, and other entities may be required to
authenticate also. Basic authentication is the simplest web-
based authentication scheme that works by sending the
username and password within the request. Generally
authentication is achieved through the presentation of some
token that cannot be stolen (forged). This can be either peer-
to-peer relationship (password for client and server) or
through a trusted third party (certification authority or
Kerberos server). Biometrics characteristics can also be used
to a service for authentication purpose, since a unique
identification of human being can give more security for
example a finger print scanner can be used to log into a local
machines. Trust can be defined as the assured reliance on the
character, ability, strength, or truth of someone or something.

Access Control is the ability to limit and control the
access to host systems and applications via communications
links. The process of authorization is often used as a synonym
for access control, but it also includes granting the access or
rights to perform some actions based on access rights.

Data integrity assures that the data is not altered or
destroyed in an unauthorized manner. Integrity checks are
provided primarily via hash functions (or "message digests").
Data confidentiality, Sensitive information must not be
revealed to parties that it was not meant for. Data
confidentiality is often also referred to as privacy. The
standard approach to ensure confidentiality is through
encryption, which is the application of an algorithm that
transforms "plaintext" to "cipher text" whose meaning is
hidden but can be restored to the original plaintext by another
Algorithm (the invocation of which is called decryption).

Key management deals with the secure generation,
distribution, authentication, and storage of keys used in
cryptography. Nonrepudiation refers to the inability of
something that performed a particular action such as a
financial transaction to later deny that they were indeed
responsible for the event.

Basically, security requires at least three fundamental
services: authentication, authorization, and encryption. A grid
resource must be authenticated before any checks can be done
as to whether or not any requested access or operation is

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

allowed within the grid. Once the grid resources have been
authenticated within the grid, the grid user can be granted
certain rights to access a grid resource. This, however, does
not prevent data in transit between grid resources from being
captured, spoofed, or altered [18]. The security service to
insure that this does not happen is encryption. Obviously, use
of data encryption certainly will have its impact on
application layer performance. But, in this work we will
examine its impact on total network performance. In this
paper, we will study the impact of four symmetric encryption
algorithms in a typical grid network.

The use of cryptography will certainly have an impact on
network performance in one way or another. So we decided to
model an application layer encryption -decryption scenario in
a typical grid computing environment and study its impact on
network performance through network simulations.

B. Security Methods

Symmetric encryption: Using the same secret key to
provide encryption and decryption of data. Symmetric
cryptography is also known as secret-key cryptography.

Asymmetric encryption: Using two different keys for
encryption and decryption. The public key encryption
technique is the primary example of this using a "public key"
and a "private key" pair. So it is referred as public-key
cryptography.

Secure Socket Layer/Transport Layer Security
(SSL/TLS): These are essentially the same protocol, but are
referred to one another differently. TLS has been renamed by
the IETF, but they are based on the same RFC.

Public Key Infrastructure (PKI): The different
components, technologies, and protocols that make up a PKI
environment. Grid security implementations are
predominantly built on public key infrastructure (PKI)
(Housely et al., 2002; Tuecke et al, 2004). In a PKI each
entity (e.g. user, service) possesses a set of credentials
comprised of a cryptographic key and a certificate.

Mutual Authentication: Instead of using an Lightweight
Distribution Access Protocol (LDAP) repository to hold the
public key (PKI), two parties who want to communicate with
one another use their public key stored in their digital
certificate to authenticate with one another.

C. The symmetric key Encryption Algorithms

Data Encryption Standard(DES), was the first encryption
standard to be recommended by NIST (National Institute of
Standards and Technology). It is based on the IBM proposed
algorithm called Lucifer. DES became a standard in 1974.
Since that time, many attacks and methods were recorded that
exploit the weaknesses of DES, which made it an insecure
block cipher[22].

Advanced Encryption Standard(AES), is the new
encryption standard recommended by NIST to replace DES.
Rijndael (pronounced Rain Doll) algorithm was selected in

1997 after a competition to select the best encryption
standard. Brute force attack is the only effective attack known
against it, in which the attacker tries to test all the characters
combinations to unlock the encryption. Both AES and DES
are block ciphers[20].

Blowfish is a variable length key, the block size is 64 bits,
and the key can be any length up to 448 bits block cipher.
This algorithm can be optimized in hardware applications
though it's mostly used in software applications. Though
it suffers from weak keys problem, no attack is known
to be successful against [8][23].

RC2 is a block cipher with a 64-bits block cipher
with a variable key size that range from 8 to 128 bits. RC2 is
vulnerable to a related-key attack using 234 chosen plaintexts
[20].

Authentication and authorization has been a basic and
necessary Service for internet transactions. Several new
standards have merged which allow dynamic access control
based on exchanging user attributes. Unfortunately, while
providing highly secure and flexible access mechanisms are a
very demanding task. Authentication and Authorization
Infrastructures (AAIs) can provide such integrated federations
of security services. They could, in particular, provide
attribute based access control (ABAC) mechanisms and
mediate customers' demand for privacy and vendors' needs
for information [10].

II.

LITERATURE SURVEY

The Globus Security Infrastructure (GSI) is one of the
most famous security architecture. GSI is based on Public
Key Infrastructure (PKI), which performs mutual
authentication via X.509 certificates. The author describes
present a password-based grid security infrastructure
(PBGSI), which authenticates clients by authenticated key
exchange (AuthA) methods and uses improved Chaffing and
Winnowing for secure data transfer. By using password-based
methods in authentication, authorization and delegation,
PBGSI provides convenient interface for the user. At the
same time, encryption-less secure data transfer improves the
performance; and mechanisms used in our scheme (time-
stamp etc.) enhance the security of the whole grid [11].

A grid environment is built to verify the feasibility and the
efficiency of the extended OCSP protocol. The paper deals
with the running requirement and the data description of the
client and each extended OCSP responder in detail. It
describes the processing algorithm of each responder. In order
to improve the efficiency of the system, the path length
constraint and time constraint of request transmitting are
designed specially. Theory and experiments all prove that the
extended OCSP system improves the efficiency of certificate
verification effectively [12].

Recently, Authentication protocol has been recognized as
an important factor for grid computing security. This paper
[20] described a new simple and efficient Grid authentication
system providing user anonymity. It is based on hash
function, and mobile users only do symmetric encryption and

100

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

decryption and it takes only one round of messages exchange
between the mobile user and the visited network, and one
round of message exchange between the visited network and
the corresponding home network.

There are number of projects investigating attribute-based
authentication such as the VO Privilege Project, GridShib,
and PERMIS. However, there are quite a few decision
dimensions when it comes about designing this scheme in
grid computing [10].

Authentication in the grid environment can be performed
in two ways either in the application layer part or in the
communication part. Cryptography plays a major role to
implement authentication. It is obvious that the use of
encryption and decryption at application layer will certainly
have an impact in the application layer performance in the
grid environment. In this paper, we have simulated the
encryption algorithms in a typical grid network scenario using
the results from the paper [1].

A Europe-China Grid Internetworking (EC-GIN) Project

The Internet communication infrastructure (the TCP/IP
protocol stack) is designed for broad use; as such, it does not
take the specific characteristics of Grid applications into
account. This one-size-fits-all approach works for a number
of application domains, however, it is far from being optimal
general network mechanisms, while useful for the Grid, and
cannot be as efficient as customized solutions. While the Grid
is slowly emerging, its network infrastructure is still in its
infancy. Thus, based on a number of properties that make
Grids unique from the network perspective, the project EC-
GIN (Europe-China Grid Internetworking) will develop
tailored network technology in dedicated support of Grid
applications. These technical solutions will be supplemented
with a secure and incentive-based Grid Services network
traffic management system, which will balance the conflicting
performance demand and the economic use of resources in
the network and within the Grid [30].

By collaboration between European and Chinese partners,
EC-GIN parallels previous efforts for real-time multimedia
transmission across the Internet: much like the Grid, these
applications have special network requirements and show a
special behavior from the network perspective.

B. The ERNET Project

ERNET[26] (Education and Research Network) was the
first dedicated and integrated step taken towards to enable the
research and education community in India to leverage the
benefits of ICTs. ERNET India aims at developing, setting up
and operating nationwide state-of-the-art computer
communication infrastructure and providing services to the
users in academic and research institutions, Government
organizations, and industry, in line with technology
developments and national priorities. Dissemi- nation,
training and knowledge transfer in the field of computer
communication and information technology are an integrating
part of ERNET mission.

ERNET also acts as a bridge for co-operation with other
countries in the area of computer com- munications,
information technology, computer networking and other
related emerging technologies.

The ERNET network has 15 Points of Presence spread
throughout India serving 1389 institutions, including 152
universities, 284 agricultural universities and many other
research organizations. It has 14 points of peering for Internet
bandwidth connectivity using submarine cables.

The network comprises a mix of terrestrial and satellite-
based wide area networks. It provides a wide range of
operation and application services. As of today, universities,
academic institutions, R&D labs and schools, etc. use ERNET
for a variety of applications and services including email, file
transfer, database access, world wide web , web hosting, mail
relaying, security solutions, distant learning and grids.

ERNET is the first network in the country to provide dual
stack access of Internet protocol version 6 (IPv6) and Internet
protocol version 4 (IPv4) test beds to its users to develop, test
and implement IPv6 based mail, Domain name Services, Web
applications and products.

ERNET has deployed many overlay networks over its
terrestrial and satellite network under different schemes.
Some examples are GARUDA (see below), UGC-Infonet,
interconnecting Indian universities, ICAR-Net,

interconnecting Agricultural Research centers, Universities
and Stations, and several pilot projects aiming at
interconnecting schools. Separate networks were
implemented to allow DAE institutes to connect to the
GEANT network and to participate in LHC activities.

ERNET Backbone and POPs

• Backbone Nodes & POPs
■ Proposed POPS

— ft' 34 Mbps

— 2/SMbps

Eh^- international Gateway
£ SATWAN Hub

Figure 1. The ERNET Topology [18]

C. Overview of GAR UDA Project

GARUDA[27] initiative is a collaboration of science
researchers and experimenters on a nation- wide grid of
computational nodes, mass storage and scientific instruments
that aims to provide the technological advances required to

101

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

enable data and compute intensive science of the 21st century.
One of GARUDA's most important challenges is to strike the
right balance between research and the daunting task of
deploying that innovation into some of the most complex
scientific and engineering endeavours being undertaken
today.

The Department of Information Technology (DIT) has
funded the Center for Development of Ad- vanced
Computing (C-DAC[27]) to deploy the nation-wide
computational grid 'GARUDA' which today connects 45
institutions across 17 cities in its Proof of Concept (PoC)
phase with an aim to bring "Grid" networked computing to
research labs and industry. In pursuit of scientific and
technological excel- lence, GARUDA PoC has also brought
together the critical mass of well-established researchers.

GARUDA Grid Component Architecture

Figure 2. GARUDA Grid Component Architecture [29]

C. Present Network Architecture

The GARUDA network is a Layer 2/3 MPLS Virtual
Private Network [VPN] connecting selected institutions at
10/100 Mbps with stringent quality and Service Level
Agreements. The network has been contracted as a man- aged
service with a peak capacity of 2.43 Gbps across 17 cities.
This network is a pre-cursor to the next genera- tion Gigabit
speed nation-wide Wide Area Network with high
performance computing resources and scientific instruments
for seamless collaborative research and ex- periments. The
PoC network was established at all the GARUDA partner
institutes in close collaboration with ERNET who are
responsible for the operation, mainte- nance and management
of this network.

D. Computational Resources in GARUDA

In this collaborative grid project, various resources such
as high performance computing sys- terns (HPC) and satellite
based communication systems have been committed by
different centers of C-DAC and GARUDA partners. It may
be noted that since the resources are diverse in nature, one of
the major challenges of GARUDA is to deploy appropriate
tools and middleware to enable applications to run seamlessly
across the grid.

<S1*

<S 3

^ST^

Figure 3. GARUDA topology - EU-I NDIA GRID [18]

E. Network Simulator

The Grid Computing paradigm has been widely adopted
within the research community for scientific computing. Grid
Computing is used as a method by which access is seamlessly
given to a set of heterogeneous computational resources
across a dynamic set of physical organizations, supplying
massive computing and storage capabilities. Within a Grid
environment, computational jobs are submitted to and run on
suitable resources and data is stored and transferred
transparently without knowing its geographic location. All of
this behavior will obviously show its impact on the
underling network infrastructure and the data generated
within a Grid environment may substantially affect the
network performance due to the volume involved.

We will use NS2 to simulate the network, but it is
well known that NS2 doesn't implement any security
features. Till now, there is no option for simulating security
things in NS2. The reasons for lack of security features in ns2
are:

> Security is a subtle thing related to many aspects, which
is much different from other kinds of network protocols.

> Generally there will not be any real data or packet to
encrypt or decrypt in ns2.

> The scope of a simulation will be minimizing the overall
simulation time. But if we do real encryption or
decryption in simulator, then it will go beyond the
concept of a simulator.

> Lack of support for sending real pay load in ns2.

> Lack of support for handling socket connection like real
TCP/IP scenario.

> Ns2 simulator has limitation in simulating simultaneous
threaded processes to mimic real socket connections.

Ns2 [16] is an object oriented simulator, written in C++,
with an OTCL interpreter as a frontend. The simulator
supports a class hierarchy in C++, and a similar class
hierarchy within the OTcl interpreter. The root of this

102

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

hierarchy is the class TclObject. Users create new simulator
objects through the interpreter. Applications sit on top of
transport agents in ns and there are two basic types of
applications: traffic generators and simulated applications.
Currently, there are four C++ classes derived from the traffic
generator class [20]. Traffic Generator: EXPOO_Traffic,
POO_Traffic, CBR_Traffic, TrafficTrace.

However, none of these classes match the traffic
characteristics of PPLive, and of GridFTP. So we decided to
simulate encryption in ns2 at application layer, by modeling a
new encrypted traffic generator.

III.

MODELING GRID AND GRID TRAFFIC IN NS2

Though there are different kinds of security requirements
or models for grid computing systems, the role of a symmetric
key encryption algorithm and its impact will be a significant
one when implemented in application layer that will affect the
performance in terms of time. In this work, we have simulated
the workload of different Symmetric Key Encryption
algorithms such as DES, Triple DES, AES, Blow Fish, RC2
and RC6 at application layer using Network Simulator tool.
The proposed traffic model is based on the model used in
ECGIN for symmetric key encryption and GridFTP as a cross
traffic. The proposed model is implemented on the Indian grid
network topology GARUDA, to study the impact of the
encryption based traffic model.

A Modeling Encrypted PPLive Traffic

Along with the rapid development of P2P file sharing and
IPTV video services, P2P streaming services have become
a core multi-user video sharing application on the
Internet. The focus of grid technology in the video area is
generally on the resource scheduling and replica
management aspects, while the service traffic
characteristics are still similar to the traditional video service.
In depth work has already been carried out in the areas of
monitoring and modeling video traffic[25]. Therefore,
exploring the developing trends of grid systems, video
sharing, monitoring and the analysis of P2P IPTV traffic are
interesting and promising topics of research.

The time interval between two packets and the size of each
packet waiting for sending out is very important when
modeling actual traffic. Therefore if the model can accurately
match these two characteristics, it can be said to generate
traffic that is similar to the actual data. The EC-GIN project
built a new traffic generator to model the actual traffic called
Lognormal Traffic, which is primarily responsible for
controlling the packets time interval and the packet sizes.

In this work, we extended the traffic model of PPLive
(Lognormal Traffic) to support a simulated encryption-
decryption scenario.

Based on traffic model of EC-GIN, an algorithm has been
put forward to control the packet generation sequence. First,
data initialization is performed as follows:

• Send a video packet when simulation begins.

• Compute the next video packet sending time. Put it
into a variable NextT.

Next, the time needed to send the next packet is computed.
To account for different packet sizes, different parameters are
used to calculate inter-video packet time (variable NextT) and
the inter-control packet time (array t_i). The values of t_l to
t_n are summed to variable SmallT. As long as the value of
SmallT is less than NextT, t_i is used as the inter- packet time
for sending small packets (control packets). Otherwise, a
large packet(video packet) is sent immediately with an inter-
packet time of NextT - (SmallT - t_i).

In addition to the above process, we have delayed the
packet transmission with respect to the size of the packet to
be sent and the selected encryption algorithm.

So the new Scheduled Transmission Time will be equal to
the sum of inter-packet time and the time taken for encrypting
the packet by the selected algorithm.

In our implementation we have simulated the encryption
algorithms in a typical grid network scenario just by including
the encryption delay at the traffic generator using the results
from the paper [1]. In the traffic model of ECGIN, they used
UDP in their design. We have decided to use TCP in our
design, because, TCP is the most commonly used transport
protocol in grid network communication.

B. Modeling GridFTP

The GridFTP tool of Globus Toolkit is one of the most
important components provided by Globus for moving large
amounts of data in bulk. GridFTP is based on FTP, the
highly- popular Internet file transfer protocol. Given the
characteristics of Grid traffic - often a mixture of short,
sporadic service calls and bulk data transfers - a GridFTP
simulation scenario differs from other traffic models and is
therefore important for testing Grid-specific network
mechanisms. The GridFTP simulator of EC-GEN was
developed with the OTCL language to mimic this GridFTP
traffic. The EC-GEN GridFTP is embedded in a gridftp.tcl
file. In this work we just used GridFTP as a background cross
traffic during evaluation the impact of encrypted PPLive
traffic. The three major parameters defined for the GridFTP
simulator are:

• Bandwidth: this parameter is used to set the total
bandwidth of the link. By default, this parameter is set to
1.0Mbps. With this and the ratio parameter, we can
determine the "rate_" parameter for each FTP instance.

• Parallel: this parameter is used to set the parallel
GridFTP streams. By default, this is set to 4. Since each
GridFTP stream can be simulated by FTP, this parameter
will actually set the number of FTP instances for the
GridFTP simulator.

• Ratio: this parameter is used to set the throughput ratio
among the parallel streams. By default, this is set to
1:1:1:1 which means each stream will transmit packets at
an equal speed.

103

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

The GridFTP simulator consists of two classes. One is the
GridFTP class and the other is the GridFTPSink class. We
also override two methods for the basic Simulator class,
attach-agent and connect, with which the GridFTP instance
can be attached to the network node and be connected to the
GridFTPSink instance.

C. The Simulation of GARUDA Network in ns2

The following NAM (Network Animator) output shows
the model of GARUDA network simulated on ns2. The
topology was derived from the information provided by the
ERNET and GARUDA projects [26][27].

Figure 5. The Simulated GARUDA Topology

> The links shown in green are 8/34Mbps links

> The links shown in red are 2/8 Mbps links

> Nodes shown as red hexagon are backbones and POPs

> Nodes shown as blue circles are the connected institutes

IV. SIMULATION RESULTS AND DISCUSSION

A simple model of GARUDA grid network has been
simulated in ns2 and the impact of different encryption
schemes on network performance has been evaluated. A
normal 2 GHz Pentium IV computer with 1 GB RAM was
used for this simulation.

A Traffic models

In order to create the different traffic scenarios files we
used different types of grid traffics mentioned in ECGIN
project. They are GridFTP Traffic and PPLive Traffic.

Some of the simulation parameters are

Number of Backbone and POP nodes 12

Number of Simulated Institution Nodes 36

Routing Protocol DV

Backbone Link Capacity 8/34 Mbps

Institution to Backbone Links 2/8 Mbps

Queue Type DropTail

We have simulated a encrypted PPLive traffic from one
node to another (in this topology, from Madras to Delhi) and
used some GridFTP cross traffic.

B. Performance

The following graph shows the performance of the
network with respect to different cryptography algorithms
used in application layer.

The Throughput

The following graphs show the comparison of throughput
in different encryption schemes over time.

UtouijIifiuUlO 3

Time VS Throughput ■ Comparison

5.0000 10.0000 15.0000

Figure 6. Time VS Throughput- Comparison

The following graph shows the average throughput. The
throughput in the case of Blowfish based scheme was good.

10.0000 15.0000 20.0000 25.QC

Figure 7. The Average Throughput

The Received Packets comparison

The following graphs show the comparison of time and
received packets in different encryption schemes.

104

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

[ciose] | Hdcpy | [^ouT|

Recieved Bytes ;«;1CP

Time VS Recieved Bytes - Comparison

Blown*

RC2

Figure 8. The Time VS Received Packets - Comparison

The End to End Delay

The following graphs show the comparison of end to end
delay in different encryption schemes over time

E2Edelay k1Ct 3

Time VS E2Edelay ■ Comparison

NONE

DES
AES

BlowFish
RC2

2.0000 3.0000 4.0000 5.0000 6.0000 7.0000 8.0000

Figure 9. Time VS E2Edelay - Comparison

The Average Delay

The following graphs show the average delay in different
encryption schemes.

The Average Delay

Encryption Method

Even though all the transmitted packets were received
successfully, the throughput and delay was much affected by
the retransmission of the packets during the packet loss or
drop. This retransmission of packet had an impact on
throughput. Faster the encryption algorithm, higher the
bandwidth it will try to use. So it will increase delay, packet
loss as well as drop at intermediate nodes.

CONCLUSION

Figure 10. The Average Delay

The security is a very important issue in grid network
design. Apart from authentication and authorization, the use
of symmetric encryption algorithm for grid data security is
also having significant impact on the design and performance
of grid networks. A model for grid security infrastructure has
been implemented on network simulator ns2 and the impact
of use of encryption algorithms in network performance has
been measured. We have simulated a simplified model of
GARUDA grid network in ns2 and simulated some of the
basic traffic types of grid network (proposed in ECGIN). As
shown in the graphs in previous section, the use of
cryptography at application layer has obvious impact on
network performance. Depending on the cryptographic
algorithms, the delay in delivery of packet is proportional
with respect to time. Due to queuing delay at the intermediate
node, the faster algorithm provides better throughput with a
little bit of delay in packet delivery.

Future works may address the issues of impact of
asymmetric encryption algorithms used in a grid network for
authentication and other purposes. Further, the work may be
extended for implementing some other traffic types of grid
network.

REFERENCES

[1] Diaa Salama Abd Elminaam, Hatem Mohamed Abdual Kader,
and Mohiy Mohamed Hadhoud, "Evaluating The Performance
of Symmetric Encryption Algorithms" International Journal of
Network Security, Vol.10, No.3, PP.216-222

[2] D. S. Abdul. Elminaam, H. M. Abdul Kader and M. M.
Hadhoud, Performance Evaluation of Symmetric

Encryption Algorithms, Communications of the IBIMA

Volume 8, 2009 ISSN: 1943-7765.

[3] Aamer Nadeem, "A Performance Comparison of Data
Encryption Algorithms", IEEE 2005.

[4] Earle, "Wireless Security Handbook,". Auerbach Publications

2005
[5] Priya Dhawan., "Performance Comparison: Security Design

Choices", Microsoft
[6] Edney, " Real 802.11 Security: Wi-Fi Protected Access and

802.11i",

Addison Wesley 2003.
[7] Hardjono, " Security In Wireless LANS And MANS ", Artech
House
Publishers 2005

[8] Bruce Schneier, "Applied Cryptography", John Wiley & Sons,
Inc 1996

[9] Ronald L. Rivest, M.J.B. Robshaw, R. Sidney, and Y.L. Yin, "
The

105

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

RC6TM, Block Cipher", Version 1.1 - August 20, 1998.
[10] Christian Schlager, Manuel Sojer, Bjorn Muschall, and Giinther
Pernul

, "Attribute-Based Authentication and Authorisation

Infrastructures for E-Commerce Providers", K. Bauknecht et al.
(Eds.): EC-Web 2006,

LNCS 4082, pp. 132 - 141, 2006.

[11] Zhun Cai,"A Password-based Grid Security Infrastructure"
10.1109 /ICDS.2008.39, Second International Conference on
The Digital Society, Institute of Digital Technology

AISINOInc.

[12] Shaomin Zhang, Baoyi Wang, Hebei Province, "Research on
An Extended OCSP Protocol for Grid", Proceedings of

the 7 th World Congress on Intelligent ontrol and

Automation, 25 - 27, 2008, China.

[13] Ronghui Wu, Renfa Li, Fei Yu ,guangxue, Cheng Xu,
"Research on User Authentication for Grid Computing

Security", Proceedings of the Second International Conference
on Semantics, Knowledge, and Grid (SKG'06) 0-7695-2673-
X/06 $20.00 © 2006.

[14] Anna Cinzia Squicciarini, Elisa Bertino and Sebastien
Goasguen, "Access Control Strategies for Virtualized
Environments in Grid Computing Systems", Proceedings of
the 11th IEEE International Workshop on Future Trends of
Distributed Computing Systems (FTDCS'07) 0-7695-2810-
4/07 $20.00 © 2007.

[15] Marty Humphery, Mary R. Thomson, and Keith R.Jackson,
"Security for Grids", Proceeding of the IEEE, Vol 93, No. 3,
pp.644-650, March 2005.

[16] Europe-China Grid InterNetworking, European Sixth
Framework STREP FP6-2006-IST-045256, Deliverable

D2.1, Ns2 code for Grid network simulation. The EC-GIN
Consortium, Europe-China Grid InterNetworking, Survey of
Grid Simulators, Network-level Analysis of Grid Applications,
The EC-GIN Consortium.

[17] International Technical Support Organization, "Introduction to
Grid Computing with Globus", September 2003, IBM
Corporation.

[18] http://partners.euindiagrid.eu/deliverables/D3. 1. html

[19] http://www.faqs.org/rfcs/rfc2828.html

[20] http://msdn2.microsoft.com/en-us/library/ms978415.aspx,
Developer Network October 2002.

[21] http://en.wikipedia.org/wiki/Block_cipher

[22] http://www.tropsoft.com/strongenc/des.htm

[23] http://www.eskimo.com/~weida^enchmarks.html

[24] Coder's Lagoon, http://www.hotpixel.net/software.html

[25] http://www.ec-gin.eu

[26] http://www.eis.ernet.in

[27] www.garudaindia.in

[28] http://www.euindiagrid.eu/

[29] www.cdac.in

[30] ttp://www.euindiagrid.eu/index.php/documents/doc_downloa
d/1 1- einfrastructures-across-europe-and-india

Mrs. N. Thenmozhi is working as Assistant Professor,
Department of Computer Science in N.K.R. Govt. Arts
College for Women, Namakkal. She obtained her Bachelor
degree in Statistics from Saradha College, Salem under
Madras University, Master's degree in Computer
Applications from Bharathiar University, Coimbatore,
Master's degree in Software Systems from BITS, Pilani,and
M.Phil From Manonmaniam Sundaranar University. She is
currently pursuing Ph.D. under Mother Teresa Women's
University, Kodaikanal. She has 18 years of Teaching
Experience and 2 years of Industrial experience. She has
published number papers in various national and international
conferences. She is life member of ISTE. Her field of interest
includes Grid Computing, Network Security and Image
Processing.

M.Madheswaran received the BE Degree from Madurai
Kamaraj University in 1990, ME Degree from Birla Institute
of Technology, Mesra, Ranchi, India in 1992, both in
Electronics and Communication Engineering. He obtained his
PhD degree in Electronics Engineering from the Institute of
Technology,Banaras Hindu University, Varanasi, India, in
1999. At present he is a Principal of Muthayammal
Engineering College, Rasipuram, India. He has authored over
Seventy five research publications in International and
National Journals and Conferences. Currently he is the
chairman of IEEE India Electron Devices Society Chapter.
His areas of interest are theoretical modeling and simulation
of high-speed semiconductor devices for integrated
optoelectronics application, Bio-optics and Bio-signal
Processing. He was awarded the Young Scientist Fellowship
(YSF) by the State Council for Science and
Technology,TamilNadu, in 1994 and Senior Research
Fellowship (SRF) by the Council of Scientific and Industrial
Research (CSIR), Government of India in 1996. Also he has
received YSF from SERC, Department of Science and
Technology, Govt, of India. He is named in Marquis Who's
Who in Science and engineering in the year 2006. He is a
Member of Institute of Electrical and Electronics Engineers,
Fellow of Institution of Electronics and Telecommunication
Engineers, Member of Indian Society for Technical Education
and Member of Institution of Engineers.

AUTHORS PROFILE

106

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Low Power and Area Consumption Custom Networks-On-Chip
Architectures Using RST Algorithms

^.Ezhumali 2 Dr.C.Arun

Professor, Dept of Computer Science Engineering

2 Asst. Professor, Dept of Electronics and Communication

Ralalakshmi Engineering College, Thandalam-602 105, Chennai, India

Abstract: Network-on-Chip (NoC)

architectures with optimized topologies have
been shown to be superior to regular
architectures (such as mesh) for application
specific multiprocessor System-on-Chip
(MPSoC) devices. The application specific NoC
design problem takes, as input the system-level
floorplan of the computation architecture .The
objective is to generate an area and power
optimized NoC topology. In this work, we
consider the problem of synthesizing custom
networks-on-chip (NoC) architectures that are
optimized. Both the physical links and routers
determine the power consumption of the NoC
architecture. Our problem formulation is based
on the decomposition of the problem into the
inter-related steps of finding good flow
partitions, and providing an optimized network
implementation for the derived topologies. We
used Rectilinear-Steiner-Tree (RST)-based
algorithms for generating efficient and
optimized network topologies. Experimental
results on a variety of NoC benchmarks showed
that our synthesis results were achieve reduction
in power consumption and average hop count
over different mesh implementations. We
analyze the quality of the results and solution
times of the proposed techniques by extensive
experimentation with realistic benchmarks and
comparisons with regular mesh-based NoC
architectures.

Index Terms — Multicast routing, network-on-
chip (NoC), synthesis, system-on-chip (SoC),
topology.

l.Introduction

Network-on-Chip (NoC) is an emerging

paradigm for communications within large
VLSI systems implemented on a single silicon
chip. The layered-stack approach to the design
of the on-chip intercore communications is the
Network-on-Chip (NOC) methodology. In a
NoC system, modules such as processor cores,
memories and specialized IP blocks exchange
data using a network as a "public
transportation" sub-system for the information
traffic. A NoC is constructed from multiple
point-to-point data links interconnected by
switches (a.k.a. routers), such that messages
can be relayed from any source module to any
destination module over several links, by
making routing decisions at the switches.

A NoC is similar to a modern
telecommunications network, using digital bit-
packet switching over multiplexed links.
Although packet switching is sometimes
claimed as necessity for a NoC, there are several
NoC proposals utilizing circuit-switching
techniques. This definition based on routers is
usually interpreted so that a single shared bus, a
single crossbar switch or a point-to-point
network is not NoCs but practically all other
topologies are. This is somewhat confusing
since all above-mentioned are networks (they
enable communication between two or more
devices) but they are not considered as network-
on-chips. Note that some erroneously use NoC
as a synonym for mesh topology although NoC
paradigm does not dictate the topology.
Likewise, the regularity of topology is
sometimes considered as a requirement, which
is, obviously, not the case in research
concentrating on "application-specific NoC
topology synthesis".

107

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

1F=%=%

■Core

^»=^=^=^i

Network Adapter
Routing Node

-Link

1<H=^=^H=^

figure. 1 Topological illustration of a
4-by-4 grid structured NoC.

The wires in the links of the NoC are shared
by many signals. A high level of parallelism
is achieved, because all links in the NoC can
operate simultaneously on different data
packets. Therefore, as the complexity of
integrated systems keeps growing, a NoC
provides enhanced performance (such as
throughput) and scalability in comparison
with previous communication architectures
(e.g., dedicated point-to-point signal wires,
shared buses, or segmented buses with
bridges). Of course, the algorithms must be
designed in such a way that they offer large
parallelism and can hence utilize the
potential of NoC.

the complexity of designing wires for
predictable speed, power, noise, reliability,
etc., because of their regular, well-controlled
structure. From a system design viewpoint,
with the advent of multi-core processor
systems, a network is a natural architectural
choice. A NoC can provide separation between
computation and communication; support
modularity and IP reuse via standard
interfaces, handle synchronization issues,
serve as a platform for system test, and, hence,
increase engineering productivity.

Although NoCs can borrow concepts and
techniques from the well-established domain
of computer networking, it is impractical to
blindly reuse features of "classical" computer
networks and symmetric multiprocessors. In
particular, NoC switches should be small,
energy-efficient, and fast. Neglecting these
aspects along with proper, quantitative
comparison was typical for early NoC
research but nowadays they are considered in
more detail. The routing algorithms should
be implemented by simple logic, and the
number of data buffers should be minimal.
Network topology and properties may be
application-specific. Research on NoC is now
expanding very rapidly, and there are several
companies and universities that are involved.
Figure 1 shows how a NoC, in comparison
with shared buses, could be occupied with
various components as resources

Traditionally, ICs have been designed with
dedicated point-to-point connections, with one
wire dedicated to each signal. For large
designs, in particular, this has several
limitations from a physical design viewpoint.
The wires occupy much of the area of the chip,
and in nanometer CMOS technology,
interconnects dominate both performance and
dynamic power dissipation, as signal
propagation in wires across the chip requires
multiple clock cycles. NoC links can reduce

2.EXISTING RELATED WORKS

So far, the communication problems faced
by System on chip were tackled by making use
of regular Network on chip architectures. The
following are the list of popular regular NoC
architectures:

Mesh Architecture.
Torus Architecture.
Butterfly Fat Tree Architecture.

108

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Extended Butterfly Fat Tree Architecture

The NoC design problem has received
considerable attention in the literature. Towles
and Dally [1] and Benini and De Micheli [2]
motivated the NoC paradigm. Several existing
NoC solutions have addressed the mapping
problem to a regular mesh-based NoC
architecture [3], [4]. Hu and Marculescu [3]
proposed a branch-and-bound algorithm for
the mapping of computation cores on to mesh-
based NoC architectures. Murali et ah [4]
described a fast algorithm for mesh-based NoC
architectures that considers different routing
functions, delay constraints, and bandwidth
requirements. On the problem of designing
custom NoC architectures without assuming
existing network architecture, a number of
techniques have been proposed [5]-[10]. Pinto
et ah [7] presented techniques for the
constraint-driven communication architecture
synthesis of point-to-point links by using
heuristic-based -way merging. Their technique
is limited to topologies with specific structures
that have only two routers between each
source and sink pair. Ogras et ah [5], [6]
proposed graph decomposition and long link
insertion techniques for application-specific
NoC architectures. Srinivasan et ah [8], [9]
presented NoC synthesis algorithms that
consider system-level floor planning, but their
solutions only considered solutions based on a
slicing floorplan where router locations are
restricted to corners of cores and links run
around cores. Murali et ah [10] presented an
innovative deadlock-free NoC synthesis flow
with detailed backend integration that also
considers the floorplanning process. The
proposed approach is based on the min-cut
partitioning of cores to routers. This work
presents a synthesis approach based on a set
partitioning formulation that considers
multicast traffic, Although different in
topology and some other aspects, all the above
papers essentially advocate the advantages of
using NoCs and regularity as effective means

to design high performance SoCs. While these
papers mostly focus on the concept of regular
NoC architecture (discussing the overall
advantages and challenges), to the best of our
knowledge, our work is better than previous
custom NoC synthesis formulations and
efficient way to solve it.

PROPOSED SYSTEM

3.1 PROBLEM DEFINITION

• We consider the problem of synthesizing
custom networks-on-chip (NoC)
architectures that are optimized for a
given application.

• We divide the problem statement into
the flowing interrelated steps:

Physical topology Construction.
Power and Area Comparisons

3.2 SYSTEM ARCHITECTURE

Input
Specification J

Networks-on-
chip Synthesis

♦

NoC Power -\
and Area

Estimation J

Topology
Generation

Figure. 2 Proposed System Architecture

Our NoC synthesis design flow is depicted in
Figure 2. The major elements in the design
flow are elaborated as follows.

Input Specification: The input specification
to our design flow consists of a list of
modules. As observed in recent trends, many
modern SoC designs combine both hard and
soft modules as well as both packet-based
network communications and conventional

109

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

wiring. Modules can correspond to a variety
of different types of intellectual property (IP)
cores such as embedded microprocessors,
large embedded memories, digital signal
processors, graphics and multimedia
processors, and security encryption engines,
as well as custom hardware modules. These
modules can come in a variety of sizes and
can be either hard or soft macros, possibly as
just black boxes with area and power
estimates and constraints on aspect ratios. To
facilitate modularity and interoperability of
IP cores, packet-based communication with
standard network interfaces is rapidly gaining
adoption. Custom NoC architectures are
being advocated as a scalable solution to
packet-based communication. In general, a
mixture of network-based communications
and conventional wiring may be utilized as
appropriate, and not all inter-module
communications are necessarily over the on-
chip network. For example, an embedded
microprocessor may have dedicated
connections to its instruction and data cache
modules. Our design flow and input
specification allow for both interconnection
models. Below is an example of a
communication demand graph:

Figure 3 Sample Input Specification

NoC Synthesis: Given input specification
information, the NoC synthesis step then
proceeds to synthesize a NoC architecture

that is optimized for the given specification.
Consider the above diagram that depicts a
small illustrative example. It only shows the
portion of the input specification that
corresponds to the network-attached modules
and their traffic flows. The nodes represent
modules, edges represent traffic flows, and
edge labels represent the length of the two
vertices. The NoC Synthesis generates
topologies based on the communication
demand graph and comparing with
parameters like power consumption and area
usage chooses the best architecture. Below is
an example of two architectures generated
based on the given CDG.

r\ 200

>v5

v2 I

r ^00

3v6

J PL

200 1

200 v ^

Figure 4 Sample Topologies Generated
NoC Power and Area Estimation: To

evaluate the power and area of the

110

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

synthesized NoC architecture, we use a state-
of the- art NoC power-performance simulator
called Orion that can provide detailed power
characteristics for different power
components of a router for different
input/output port configurations. It accurately
considers leakage power as well as dynamic
switching power, which is important since it
is well known that leakage power is
becoming an increasingly dominating. Orion
also provides area estimates based on a state-
of-the-artrouter microarchitecture.

MODULE DESCRIPTION

s~ ~ x

Flow Partitioning

Steiner Tree Based Topology Construction

Implementation Optimization

Figure 5 Formulation of Synthesis Problem

4.1 Flow Partitioning

Flow partitioning is performed in
the outer loop of our synthesis formulation to
explore different partitioning of flows to
separate subnetworks. We make use of the
following algorithm to implement flow
partitioning:

4.2 STEINER TREE BASED
TOPOLOGY CONSTRUCTION

For each flow partition considered, physical
network topologies must be decided. In current
process technologies, layout rules for
implementing wires dictate physical topologies
where the network links run horizontally or
vertically. Thus, the problem is similar to
Rectilinear Steiner Tree (RST) problem that has
been extensively studied for the conventional
VLSI routing problem. Given a set of nodes, the
RST problem is to find a network with the

shortest edge lengths using horizontal and
vertical edges such that all nodes are
interconnected. The RST problem is well
studied with very fast implementations
available. We create an RST solver in the inner
loop of flow partitioning to generate topologies
for the set partitions considered.

[ ii put: G"(V ? E. TV. A): communication demand graph

G: specified evaluation function for implementation cost
Output: T: synthesized network architecture

I: initialize P Q =

2: for all e k £ E

4: ccEtCK}) = EvaluatcOD5t(r({e fc }) ? C)

5: end for

& t -

7: while F*\ > 1 do

S: for all g u ,g v fc P* do

A«. = At U 0„

T{ff uv ) = SolvcRST^,)

cestui,) = EvaluafccCcB^X^,), C)

P($u>9v) = «*tCffi«0 + 5^e*» *?*„*, «»tCft)
end for

(n : j>) = airsmin^,^/* }%$^gj)

# = *+!
IS: end while

19: foralU G [0,^-1] do
20: <F*) = E, u ^™t(^)

21: #Ghi[P i ]=X) 9ue i-A9u)

22: end for

23: t — itfgMm^emn-ii rfP'-)

24: T ~ £ol?i[F t \

23: return T

JO:
11:
\2:
13:
14:
15:
J 0.
J7;

ena lor

Figure 6 Flow Partitioning Algorithm
IMPLEMENTATION RESULTS

5.1. EXPERIMENTAL SETUP

We have implemented our
proposed algorithm in C. In our
implementation, we have designed a
Rectilinear Steiner Tree solver to generate
the physical network topologies in the inner
loop of the algorithm. Simulator ORION 2.0
does the power and area estimates. The
Results obtained are shown in a line chart for
mere comparisons. A snapshot of the all the
results have been shown later in this chapter.
All experimental results were obtained on a
3.06-GHz Intel P4 processor machine with
512 MB of memory running Linux.

111

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

5.2. EXPERIMENTAL RESULTS

ALL FSTs: 64 Points
Figure 7 Snapshot of ALL The FSTs
Generated

— -i

V. L

■^-p

■ .

i ,

L-' ■

Steiner Minimal Tree: 64 Points, length =
56729

Figure 8 Steiner Minimal Tree Generated

Method of Evaluation: In all our experiments,
we aim to evaluate the performance of the
proposed algorithms. On all benchmarks with
the objective of minimizing the total area as
well as power consumption of the synthesized
NoC architectures. The total area as well as
power consumption includes all network
components. We applied the design parameters
of 1 GHz clock frequency, 4-flit buffers, and
128-bit flits. For evaluation, fair direct
comparison with previously published NoC

synthesis results is difficult in part because of
vast differences in the parameters assumed. To
evaluate the effectiveness of our algorithms, we
have the full mesh implementation for each
benchmark for comparison from previous
published papers have been taken. These
comparisons are signified to show the benefits
of custom NoC architectures.

Table 1. NOC Power Comparisons

S.No

V ertices

Custom Power

Mesh Power

Opt MeshPower

0.0416

0.0990

0.0430

0.0432

0.1000

0.0500

0.0494

0.1780

0.0600

0.0617

0.2570

0.1220

0.0663

0.2720

0.1520

0.0848

0.3760

0.1310

0.0987

0.4950

0.1540

0.1034

0.6330

0.2020

0.1034

0.6420

0.2600

0.1203

1.0080

0.3000

0.1353

1.4250

0.3440

Power Comparison

£ o.eooo

: :::

.* — -♦

__— — ~

_..-*■ ■

:=I_"II~~-»^-~i^-i

■— ■— ■

-tuaom Mnwar

Figure 9 NoC Power Comparisons

The area results, power results, the execution
times, and area as well as power
improvements of that algorithm are reported.
The results show the algorithm can
efficiently synthesize NoC architectures that
minimize power and area consumption as
compared with regular topologies such as
mesh and optimized mesh topologies.

112

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Table 2. NoC Area Comparisons

S.No

1
2

Vertice

Custom
Area
0.1543
0.1557

Opt.

Mesh

Area

0.31

0.43

0.1810

0.41

0.2252

0.48

0.3060

0.91

0.3563

1.03

0.3732

1.19

0.3753

1.81

0.4332

1.81

0.493(5

2.01

NoC Area Comparisons

-Custom Area
-opt. Mesh Area

6 7 8 12 14 20 24 25 36 44
Vertices

Figure 10. NoC Area Estimates

Thus, the above two line charts in
figure 9 and 10 clearly show a reduction in
power and area estimates of custom NoC
with mesh and optimized mesh topologies.
Mesh topologies was explained in chapter 2.
Eliminating router ports and links that are not
used forms optimized mesh topologies. The
power reduction is at an average of 83.43
percent and 50 percent as compared to mesh
and optimized mesh topologies respectively.
The area reduction is at an average of 70.95
percent as compared to optimized mesh
topologies.

6.CONCLUSION AND FUTURE WORK

In this research Works have been carried out
in context related to Regular topologies like
mesh, torus and etc. This work presented an
idea on building customizing network on
chip with the better flow partitioning and
also considered power and area reduction as
compared to the already presented Regular
topologies, we proposed a formulation of the
custom NoC synthesis problem based on the
decomposition of the problem into the inter-
related steps of deriving a good physical
network topology, and providing an
comparison in terms of area and power with
the well established regular topologies. We
used the algorithm called CLUSTER for
systematically examining different possible
set partitioning of flows, and we proposed
the use of RST algorithms for constructing
good physical network topologies. Our
solution framework enables the decoupling
of the evaluation cost function from the
exploration process, thereby enabling
different user objectives and constraints to be
considered. Although we use Steiner trees to
generate a physical network topology for
each group in the set partition, the final NoC
architecture synthesized is not necessarily
limited to just trees as Steiner tree
implementations of different groups may be
connected to each other to form non-tree
structures.

This work does not differentiate the
routers/switches (communication modules)
with the operating modules present in the
chip. In near future, the work of identifying
the best placement of routers and minimizing
the number of routers and also the effectives
of the customized Network on Chip in terms
of other parameters like throughput, latency.
Link utilization and buffer utilization can be
taken into account.

113

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

REFERENCES

[1] Shan Yan, Bill Lin, " Custom Networks-
on-Chip Architectures With Multicast
Routing," IEEE transactions on very large
scale integration (VLSI) systems, vol. 17, no.
3, march 2009.

[2] K. Srinivasan, K. S. Chatha, and G.
Konjevod, "Linear-programming based
techniques for synthesis of network-on-chip
architectures," IEEE Trans. Very Large Scale
Integr. (VLSI) Syst., vol. 14, no. 4, pp. 407-
420, Apr. 2006.

[3] K. Srinivasan, K. S. Chatha, and G.
Konjevod, "Application specific network-on-
chip design with guaranteed quality
approximation algorithms," in Proc.
ASPDAC, 2007, pp. 184-190.

[4] S. Murali, P. Meloni, F. Angiolini, D.
Atienza, S. Carta, L. Benini, G . De Micheli,
and L. Raffo, "Designing application-specific
networks on chips with floor plan
information," in Proc. ICCAD, 2006, pp.
355-362.

[8] D. Greenfield, A. Banerjee, J. -G. Lee,
and S. Moore, "Implications of rent's rule for
NoC design and its fault-tolerance," in Proc.
NOCS, May 2007, pp. 283-294.

[9] S. Yan and B. Lin, "Application-specific
network-on-chip architecture synthesis based
on set partitions and Steiner trees," in Proc.
ASPDAC, 2008, pp. 277-282.

[10] Xilinx, San Jose, CA, "UMC delivers
leading-edge 65 nm FPGAs toXilinx," Des.
Reuse, Nov. 8, 2006 [Online]. Available:
http://www.design-

reuse.com/news/14644/umc-edge-65nm-
fpgas-xilinx.html

[11] P. Gratz, K. Sankaralingam, H. Hanson,
P. Shivakumar, R.McDonald, S. W. Keckler,
and D. Burger, "Implementation and
evaluation of a dynamically routed processor
operand network," in Proc. NOCS, May
2007, pp. 7-17.

[12] N. Enright-Jerger, M. Lipasti, and L.-S.
Peh, "Circuit-switched coherence," IEEE
Computer. Arch. Lett. vol. 6, no. 1, pp. 193-
202, Mar. 2007.

[5] L. Zhang, H. Chen, H. Chen, B. Yao, K.
Hamilton, and C.-K. Cheng, "Repeated on-
chip interconnect analysis and evaluation of
delay, power, and bandwidth metrics under
different design goals," in Proc. ISQED,
2007, pp. 251-256.

[6] R. Mullins, "Minimizing dynamic power
consumption in on-chip networks," in Proc.
Int. Symp. Syst.-on-Chip, 2006, pp. 1-4.

[7] C. -W. Lin, S. -Y. Chen, C. -F. Li, Y. -
W. Chang, and C. -L. Yang, "Efficient
obstacle-avoiding rectilinear Steiner tree
construction," in Proc. Int. Symp. Phys. Des.
2007, pp. 127-134.

[13]. Shan Yan, Student Member, IEEE, and
Bill Lin, Senior Member, IEEE "Custom
Networks-on-Chip Architectures With
Multicast Routing" IEEE Transactions On
Very Large Scale Integration (VLSI)
Systems, Vol. 17, No. 3, Pp 342-355, March
2009.

S .J

(M.Tech.,)

Engineering

Hyderabad,

Ezhumalai Periyathambi

received the B.E degree in

Computer Science and

engineering from Madras

University, Chennai , India in

1992 and Master Technology

in computer science and

from J N T University,

India in 2006. He is currently

114

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

working towards the Ph.D degree in
Department of Information and

Communication, Anna University, Chennai,
India. He is working as Professor in the
Department of Computer Science and
Engineering , Rajalakshmi Engineering
College, Chennai, Tamilnadu, India. His
research in reconfigurable architecture, Multi-
Core Technology CAD - Algorithms for VLSI
Architecture. Theoretical Computer Science.
And mobile computing.

1 1 5 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Prediction of Epileptic form Activity in Brain

Electroencephalogram Waves using

Support vector machine

Tavithra Devi S T

M.Phil Research Scholar

PSGR Krishnammal College for Women

Coimbatore Tamilnadu, INDIA

2 VijayaMS

Assistant Professor and Head
GRG School of Applied Computer
Technology

PSGR Krishnammal College for Women
Coimbatore Tamilnadu, INDIA

ABSTRACT

Human brain is a highly complex structure composed of
millions of nerve cells. Their activity is usually well organized
with mechanisms for self-regulation. The neurons are
responsible for a range of functions, including consciousness
and bodily functions and postures. A sudden temporary
interruption in some or all of these functions is called a
seizure. Epilepsy is a brain disorder that causes people to have
recurring seizures. Electroencephalogram (EEG) is an
important diagnostic test for diagnosing epilepsy because it
records the electrical activity of the brain. This paper
investigates the modeling of epilepsy prediction using Support
Vector Machine, a supervised learning algorithm. The
prediction model has been employed by training support
vector machine with evocative features derived from EEG
data of 324 patients and from the experimental results it is
observed that the SVM model with RBF kernel produces 86%
of accuracy in predicting epilepsy in human brain.

Keywords

Support Vector Machine, Epilepsy, Prediction, Supervised

Learning.

1. INTRODUCTION
Epilepsy is a disorder characterized by recurrent
seizures of cerebral origin, presenting with episodes of
sensory, motor or autonomic phenomenon with or without
loss of consciousness. Epilepsy is a disorder of the central
nervous system, specifically the brain [1]. Brain is one of
the most vital organs of humans, controlling the
coordination of human muscles and nerves. Epileptic
seizures typically lead to an assortment of temporal
changes in perception and behavior. Based on the
physiological characteristics of epilepsy and the
abnormality in the brain, the kind of epilepsy is determined.
Epilepsy is broadly classified into absence epilepsy, simple
partial, complex partial and general epilepsy. Absence
epilepsy is a brief episode of staring. It usually begins
between ages 4 and 14. It may also continue to adolescence
or even adulthood. Simple partial epilepsy affects only a
small region of the brain, often the hippocampus. It can
also include sensory disturbances, such as smelling or

hearing things that are not there, or having a sudden flood
of emotions. Complex partial epilepsy usually starts in a
small area of the temporal lobe or frontal lobe of the brain.
In general epilepsy the patient becomes unconscious the
patient has a general tonic contraction of all their muscles,
followed by alternating colonic contractions. It affects the
entire brain.

Various diagnostic techniques like Computed
Tomography (CT), Magnetic Resonance Imaging (MRI),
Electroencephalogram (EEG), and Positron Emission
Tomography (PET) are commonly presented.
Electroencephalography (EEG) is the recording of
electrical activity along the scalp produced by the firing of
neurons within the brain. In clinical contexts, EEG refers to
the recording of the brain's spontaneous electrical activity
over a short period of time, usually 20^10 minutes, as
recorded from multiple electrodes placed on the scalp. The
Electroencephalograph (EEG) signal is one of the most
widely signal used in the bioinformatics field due to its rich
information about human tasks for epilepsy identification
because of its characteristics like frequency range, spatial
distributions and peak frequency. EEG waves are observed
by neurologists based on spectra waveform of the signal to
identify the presence of epilepsy.

Machine learning provides methods, techniques and
tools, which help to learn automatically and to make
accurate predictions based on past observations. Current
empirical results prove that machine learning approach is
well-matched for analyzing medical data and machine
learning techniques produce promising research results to
medical domains.

Forrest Sheng Bao carried out the work and developed
a neural network based model for Epilepsy diagnosis using
EEG [1]. Piotr Mirowski carried out the work and
implemented a model based on classification of patterns of
EEG synchronization for seizure prediction using neural
network [2]. Suleiman A.B. R. proposed a new approach
for describing and classifying the EEG brain natural
oscillations such as delta, theta, alpha, and beta frequencies

116

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

using Wigner-Ville analysis with Choi-Willians filtering
and neural network [3].

The motivation behind the research reported in this
paper is to predict the presence of epilepsy in human brain.
Supervised learning technique, a kind of machine learning
algorithm is used to model the epilepsy prediction problem
as classification task to assist physician for accurate
prediction of epilepsy in patients.

In this paper, the prospective benefits of supervised
learning algorithm namely support vector machine are
made use of for the computerized prediction of epilepsy.
The proposed SVM based epilepsy prediction model is
shown in Figure 1 .

Feature Extraction using
Wavelet Toolbox in
MATLAB

SVM Training

SVM Based
Prediction model

Prediction

Figure 1. Proposed SVM based epilepsy prediction model

2. DATA ACQUISITION
EEGs show continuous oscillating electric activity. The
amplitude and the patterns are determined by the overall
excitation of the brain which in turn depends on the activity
of the reticular activating system in the brain stem.
Amplitudes on the surface of the brain can be up to 10 mV,
those on the surface of the scalp range up to 100 mV.
Frequencies range from 0.5 to 100 Hz. The pattern changes
markedly between states of sleep and wakefulness. Distinct
patterns are seen in epilepsy and five classes of wave
groups are described as alpha, beta, gamma, delta and
theta.

• Alpha waves contain frequencies
between 8 and 13 Hz with amplitude less than 10
mV. It found in normal people who are awake and
resting quietly, not being engaged in intense

mental activity. Their amplitude is highest in the
occipital region. When the person is asleep, the
alpha waves disappear. When the person is alert
and their attention is directed to a specific activity,
the alpha waves are replaced by asynchronous
waves of higher frequency and lower amplitude.

• Beta waves have a frequency range of 14
to 22 Hz, extending to 50 Hz under intense mental
activity. It has their maximum amplitude (less than
20 mV) on the parietal and frontal regions of the
scalp. There are two types: beta I waves, lower
frequencies which disappear during mental
activity, and beta II waves, higher frequencies
which appear during tension and intense mental
activity.

• Gamma waves have frequencies between
22 and 30 Hz with amplitude of less than 2 mV
peak-to-peak and are found when the subject is
paying attention or is having some other sensory
stimulation.

• Theta waves have a frequency range
between 4 to 7 Hz with amplitude of less than 100
mV. It occurs mainly in the parietal and temporal
regions in sleep and also in children when awake,
and during emotional stress in some adults,
particularly during disappointment and frustration.
Sudden removal of something causing pleasure
will cause about 20 s of theta waves.

• Delta waves have frequency content
between 0.5 and 4 Hz with an amplitude less than
100 mV. It occurs during deep sleep, during
infancy and in serious organic brain disease. They
will occur after transactions of the upper brain
stem separating the reticular activating system
from the cerebral cortex. They are found in the
central cerebrum, mostly the parietal lobes.

Five sets of images namely Normal Epilepsy, Absence
Epilepsy, Simple Partial Epilepsy, Complex Partial
Epilepsy and General Epilepsy are taken into
consideration.

3. FEATURE EXTRACTION
Feature extraction process plays a very important role
on the classification. Fourier transformation method,
discrete transformation method and continuous
transformation methods are normally available to extract
features that characterize EEG signals. The wavelet
transform (WT) provides very general techniques which
can be applied to many tasks in signal processing. Wavelets
are ideally suited for the analysis of sudden short-duration
signal changes.

In the proposed model, EEG signal analysis and feature
extraction have been performed using Discrete Wavelet
Transform (DWT). The DWT is a extraordinary case of the
WT that provides a compact representation of a signal in
time and frequency that can be computed efficiently.

117

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

The DWT is defined by the following equation:

The energy is computed using E is given by

¥ (a , b) (t) = 2 a/2 ¥ (2 a/2 (t-b))

(1)

E=2X7N

(3)

where a is a scales and b is positions of the wavelet
mother v|/ (t) is a time function with finite energy. Choosing
scales and positions are based on powers of two, which are
called dyadic scales and positions (a j=2 J; bj k =2 J k) ( j and
k integers). Equation (1) shows that it is possible to build a
wavelet for any function by dilating a function \|/ (t) with a
coefficient 2 j, and translating the resulting function on a
grid whose interval is proportional to 2-j.

The selection of appropriate wavelet and the number of
decomposition levels is very important in analysis of
signals using the WT. The number of decomposition levels
is chosen based on the dominant frequency components of
the signal. The levels are chosen such that those parts of the
signal that correlate well with the frequencies required for
classification of the signal are retained in the wavelet
coefficients. The smoothing feature of the Daubechies
wavelet of order 2 (db2) made it more suitable to detect
changes of the signals. Thus, the wavelet coefficients are
computed using db2. The frequency bands corresponding
to different levels of decomposition for db2 with a
sampling frequency of 256 Hz. The discrete wavelet
coefficients are computed using the MATLAB wavelet
toolbox.

The purpose of feature extraction is to reduce the size
of the original dataset by measuring certain properties or
features that distinguish one input pattern from another.
The various measurements based on statistical features
from EEG are extracted. The extracted features provide the
characteristics of the input type to the classifier by
considering the description of the relevant properties of the
signals into a feature space.

The statistical feature of the wavelet coefficients in
each subband such as energy, entropy, Minimum subband,
maximum subband, mean, and standard deviation are used
to investigate the adequacy for the discrimination of normal
and abnormal patients. The following statistical features
have been derived using the following.

Entropy is the diminished capacity for spontaneous
changes in signals.

where Xi is signal value, values are present in waves is
denoted as n. Total number of signal is N

Maximum Subband - It generate maximum of the
wavelet coefficients in each subband is calculated using

M ax =Max( Xl ) (4)

where max (xi) is maximum number of signal value.

Mean - It is defined as average value of a distribution
of the wavelet coefficients in each subband which is given
by

E=£xi/N

(5)

where xi is signal and total number of signal is present
in the wavelet is N

Minimum Subband - calculate minimum of the
wavelet coefficients in each subband is defined as

M in =Min(xO (6)

Where min (xi) is minimum number of signal value.

Standard deviation - standard deviation of each
subband is defined as o .This feature provide information
about the amount of change of the frequency distribution.

°=I(x-u) 2

(7)

Entropy = J] P(i, j) log P(i 9 j)

(2)

Where P(i, j) reflects the distribution of the probability
of occurrence of each signal (i , j are integer).

Energy - Provides the sum of squared elements in the
wavelet. This is also known as uniformity or the angular
second moment.

where £ is sum of squared elements in the wavelet,x is
signal value and |i is a mean of the corresponding
signal(xi).

Thus a total of 21 statistical feature are extracted from
EEG signal for each subband for preparing dataset.

118

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

4. SUPPORT VECTOR MACHINE
Support Vector Machine (SVM) is a kind of learning
machine based on statistical learning theory. SVM is
basically applied to model pattern classification task. SVM
first, maps the input vectors into feature vectors in feature
space with a higher dimension, either linearly or non-
linearly. Then, within the feature space SVM constructs a
hyperplane which separates two classes. SVM training
always seeks a global optimized solution and avoids over-
fitting, thus it has the ability to deal with a large number of
features. The machine is presented with a set of training
examples, (xi, yi) where the xi is the real world data
instances and the yi are the labels indicating which class the
instance belongs to. For the two class pattern recognition
problem, yi = +1 or yi = -1. A training example (xi, yi) is
called positive if yi = +1 and negative otherwise. SVMs
construct a hyperplane that separates two classes and tries
to achieve maximum separation between the classes.
Separating the classes with a large margin minimizes a
bound on the expected generalization error.

The simplest model of SVM called Maximal Margin
classifier, constructs a linear separator (an optimal
hyperplane) given by w T x - y= between two classes of
examples. The free parameters are a vector of weights w
which is orthogonal to the hyper plane and a threshold
value. These parameters are obtained by solving the
following optimization problem using Lagrangian duality.

Ill II 2
Minimize = —\\w\\
2 II H

Subject to D ii(w T x i -y)>l 9 i = 1, , /.

(8)

where D^ corresponds to class labels +1 and -1. The
instances with non-null weights are called support vectors.
In the presence of outliers and wrongly classified training
examples it may be useful to allow some training errors in
order to avoid over fitting. A vector of slack variables £i
that measure the amount of violation of the constraints is
introduced and the optimization problem referred to as soft
margin is given below. In this formulation the contribution
to the objective function of margin maximization and
training errors can be balanced through the use of
regularization parameter C.

The following decision rule is used to correctly predict
the class of new instance with a minimum error.

f(x)= sgnfwVy]

The advantage of the dual formulation is that it permits
an efficient learning of non-linear SVM separators, by
introducing kernel functions. Technically, a kernel function
calculates a dot product between two vectors that have
been (non- linearly) mapped into a high dimensional
feature space. Since there is no need to perform this
mapping explicitly, the training is still feasible although the

dimension of the real feature space can be very high or
even infinite. The parameters are obtained by solving the
following non linear SVM formulation (in Matrix form),

Minimize L D( u)=l/2u T Qu - e T u (9)

d T u=0 0<u<Ce

where and K - the Kernel Matrix. Q = DKD.

The Kernel Function K (AAT) (polynomial or
Gaussian) is used to construct hyperplane in the feature
space, which separates two classes linearly, by performing
computations in the input space.

f(x)= sgn(K(x,Xi T )*u-y)

Where u - the Lagrangian multipliers. In general the
larger the margin the lower the generalization error of the
classifier.

5. EXPERIMENTAL SETUP

The data investigation and epilepsy prediction is carried
out using SVMlight 1 for machine learning. Five categories
of feature vectors are labeled as 1 for Absence, 2 for
General, 3 for Complex Partial Epilepsy, 4 for Normal
Epilepsy and 5 for Simple Partial Epilepsy, The training
dataset used for epilepsy prediction modeling consists of
about 324 images, where each category consists of about
65.

The dataset has been trained using SVM with linear,
polynomial and RBF kernel and with different parameter
settings for d, gamma and C-regularization parameter. The
parameters d and gamma are related with polynomial
kernel and RBF kernel respectively.

The 10 fold cross validation method is used for
evaluating the performance of the SVM based trained
models. The performance of the models is evaluated based
on prediction accuracy of the models and learning time.

6. RESULTS AND DISCUSSION

The cross validation outcome of the trained models
based on support vector machine with linear kernel is
shown Table I.

Table I.

SVM Linear kernel

Linear SVM

C=0.1

C=0.2

C=0.3

C=0.4

Accuracy (%)

Time(secs)

0.01

0.02

0.03

1 SVMlight is an open source tool.

http ://www. cs. Cornell. edu/people/tj/svm_light/

119

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

The outcome of the model based on SVM with
polynomial kernel and with parameters d and C are shown
in Table II.

Table II. SVM Polynomial kernel

C=0.1

C=0.2

C=0.3

C=0.4

Accuracy (%)

Time(secs)

0.2

0.1

0.2

0.6

0.3

0.1

0.3

0.4

The predictive accuracy of the non-linear support
vector machine with the parameter gamma (g) of RBF
kernel and the regularization parameter C is shown in
Table III.

Table III. SVM RBF kernel

C=0.1

C=0.2

C=0.3

C=0.4

Accuracy (%)

Time(secs)

0.2

0.3

0.4

0.5

1.5

1.6

1.2

The average and comparative performance of the SVM
based prediction model in terms of predictive accuracy and
learning time is given in Table IV and shown in Figure 1
and Figure 2.

Table IV. Overall performance of three models

Kernel type

Accuracy

Learning time

Linear

84.96%

0.027 sees

Polynomial

90.12%

0.362 sees

RBF

93.87%

0.787 sees

Prediction Accuracy

I Accuracy(%)

Linear Polynomial RBF

Figure 2. Prediction Accuracy

Learning Time

0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1

S-

I Learning
Time(secs)

Linear Polynomial RBF

Figure 3 : Learning Time

As far as the epilepsy predictions task is anxious,
accuracy plays major role in determining the performance
of the epilepsy trained model than considering the learning
time. From the above results, it is found that the predictive
accuracy shown by SVM with RBF kernel with parameters
C=0.2 and g=2 is higher than the SVM with linear and
polynomial kernel.

7. CONCLUSION

This paper elucidates the modeling of the epileptic
seizure prediction task as multi-class classification problem
and the implementation of supervised learning algorithm,
support vector machine. The performance of SVM based
epilepsy prediction models is evaluated using 10 fold cross
validation and the results are analyzed. The results indicate
that the support vector machine with RBF kernel provide
the high prediction accuracy compared to other kernels.
SVM is better than conventional methods and show good
performance in all experiments it is very flexible and more
powerful because of its robustness. It is hoped that more
interesting results will follow on further exploration of
data.

8. ACKNOWLEDGMENT
The author would like to thank the Management and
Hospital, Coimbatore for providing the EEG data.

9. REFERENCES

[1] Forrest Sheng Bao , Jue-Ming Gao, Jing Hu , Donald Y. C.
Lie , Yuanlin Zhang , and K. J. Oommen. "Automated
Epilepsy Diagnosis Using Interictal Scalp EEG". 31st
Annual International Conference of the IEEE EMBS
Minneapolis, Minnesota, USA, September 2-6, 2009.

[2] Piotr Mirowski MSc*, Deepak Madhavan Yann Le
Cun, uben Kuzniecky " Classification of Patterns of
EEG Synchronization for Seizure Prediction".

[3] A. R.Sulaiman, " Joint Time - Frequency Analysis and
Its pplication for Non - Stationary Signals", Ph.D. Thesis
Elect. Eng. Dept, University of Mosul, 2001.

[4] Webster, J. G., "Medical Instrumentation Application and
esign", 2 nd ed., New York: Wiley, 1995.

120

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

[5] Nello Cristianini and John Shawe - Taylor. "An
Introduction to Support Vector Machines and other
kernel - based earning methods" Cambridge University
Press, 2000.

[6] K. Crammer and Y. Singer. "On the Algorithmic
implementation of Multi - class SVMs, JMLR,
2001. Vojislav Kecman: "Learning and Soft Computing —
Support Vector Machines, Neural Networks, Fuzzy Logic
Systems", The MIT Press, Cambridge, MA, 2001.

[7] Chui, C.K. (1992a), "Wavelets: a tutorial in theory and
applications", Academic Press.

[8] Ian H. Witten, Eibe Frank, Len Trigg, Mark Hall, Geoffrey
Holmes, Sally.

[9] Ian H. Witten, Eibe Frank. : Data Mining - Practical
Machine Learning Tools and Techniques. 2nd edn. Elsevier.
(2005).

[10] Joachims T, SchoTkopf B, Burges C, Smola A,"Making
large-Scale SVM Learning Practical. Advances in Kernel
Methods - Support Vector Learning", 1999, MIT Press,
Cambridge, MA, USA.

[11] John Shawe-Taylor, Nello Cristianini, "Support Vector
Machines and other kernel-based learning methods", 2000,
Cambridge University Press, UK.

[12] Soman K.P, Loganathan R, Ajay V, "Machine Learning with
SVM and other Kernel Methods", 2009, PHI, India.

[13] Crammer Koby, Yoram Singer,"On the Algorithmic
Implementation of Multi-class Kernel-based Vector
Machines", Journal of Machine Learning Research, MIT
Press, Cambridge, MA, USA, 2001, Vol.2 Page 265-292.

121 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

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

Vol. 8, No. 6, September 2010

Deployment of Intelligent Agents in
Cognitive Networks

Huda Fatima
Dept. ofCS
Jazan University
Jazan, K.S.A

Dr.Sateesh Kumar Pradhan
Dept. of Comp.Engineering
King Khalid University
Abha, K.S.A

Mohiuddin Ali Khan Dr. G.N.Dash

Dept. of Comp. Networks Dept. of Comp. Science

Jazan University Sambalpur University

Jazan, K.S.A Orissa, India

Abstract — Every organization faces challenging
task in the designing of the communication
network in order to make its efficiency smoother
by the increasing complexities. Therefore, we
have to proposed a concept of cognitive network
and how the intelligent agents are deployed to
overcome the challenges.With the tremendous
expansion of networks across the globe, the
deployment of intelligent agents in cognitive
networks contributes as an efficient, reliable and
challenging task for the researchers. In this
paper, we survey the existing research work on
cognitive networks and later we provide the
artificial intelligent techniques that are
potentially suitable for the development of
cognitive networks.

Keywords: Artificial Intelligence, Cognitive
network, Intelligent agents.

INTRODUCTION:

One of the fastest growing areas is the information
and communication technologies. These changes
have an immediate impact on diverse aspects of the
modern society, which includes inter-human
relations, economy, education & entertainment. In
this respect, the development of reliable, flexible and
future-proof infrastructure should be capable of
increasing the users' quality of life by providing
services such as e-health, e-learning and e-payments.
In order to meet the demand of the increased
complexity, future networks should be easily

maintainable and their capabilities should be
continuously improved and upgraded by relying as
little as possible on human intervention. Therefore
the network research community proposed a new
concept of networking: The Cognitive Network.
What is a Cognitive Network and how are the
intelligent agents deployed is what we have
presented here.

Cognitive networks

In this section, we analyze several existing
definitions for cognitive networks, and we argue that
two elements are essential for developing a cognitive
network (CN): the knowledge representation and the
cognition loop. Next, we discuss the framework
proposed in [2] for introducing cognition to
communication networks. The main part of the
section focuses on methods from AI that seem
applicable for developing CNs. We provide a
summary of several types of intelligent agents (IAs),
map them to the functional states of the cognitive
loop. As we go along, we also refer to existing
research on CNs which makes use of the respective
type of IA, where available. How it started? The
word cognitive refers to an entity that is able to
perform some kind of conscious intellectual activity
such as thinking, reasoning, learning or remembering
in order to make sense of its surroundings. This word
was first used in communication networks to refer to
a technology by Mitola as he introduced the
cognitive radio [4].

122

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

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

Vol. 8, No. 6, September 2010

We would like to emphasize that, according to the
dictionary [9], the word cognitive used as an
adjective to a noun means: of, relating to, being, or
involving conscious intellectual activity (as thinking,
reasoning, or remembering); based on or capable of
being reduced to empirical factual knowledge.

In [2], the authors define the CN as a network with a
cognitive process that can perceive current network
conditions, plan, decide, act on those conditions,
learn from the consequences of its actions, all while
following end-to-end goals. This loop, the cognition
loop, senses the environment, plans actions
according to input from sensors and network
policies, decides which scenario fits best its end-to-
end purpose using a reasoning engine, and finally
acts on the chosen scenario as discussed in the
previous section. The system learns from the past
(situations, plans, decisions, actions) and uses this
knowledge to improve the decisions in the future.

This definition of CN does not explicitly mention the
knowledge of the network; it only describes the
cognitive loop and adds end-to-end goals that would
distinguish it from CR or so called cognitive layers
[2]. We consider this definition of CN incomplete
since it lacks knowledge which is an important
component of a cognitive system as discussed so far
in this paper and also in [2,4,6,8].

The cognitive process can operate in a centralized
way, spanning over a large network, or in a totally
distributed manner at a device level. In the first case,
it might be too expensive to centralize all the
network specific information that the cognition loop
requires, while in the second case there might be too
little knowledge available to pursue end-to-end
network goals. In reality, the deployment of the
cognitive functionality in a network will depend on
the network specific problems and will be an
engineering decision. However, it is important that
the cognitive framework is designed in such way as
to be modular, easily upgradeable and scalable in
order to be able to accommodate existing as well as
next generation technologies and applications.

The capabilities of a Cognitive Network can be
highly distributed or extremely centralized. In
general, a Cognitive network is formed of a set of

distributed cognitive entities (agents) which are
somehow "smart" as they have certain reasoning
capabilities to be connected to the network. The
entities in such a network interact with each other,
they can cooperate, act selfishly or a combination of
the two. While functioning in this environment, the
entities are able to learn and take decisions in such
way as to reach an end-to-end goal. These end-to-
end goals are dictated by the business and user
requirements [2,4]. Developing and maintaining such
a network is an extremely challenging task and has
enormous potential, especially in the area of network
management.

A Cognitive Network needs to evolve overtime: its
set of technologies has to be updated by removing
deprecated and adding new ones; its set of tools that
help managing complexity should be added and
removed in a plug and play fashion. Thus, the
architecture of cognitive network should be flexible
and should lead to a modular and highly scalable
infrastructure. Furthermore, the cognitive network
must be self aware : it should be able to determine
appropriate actions to achieve goals and to learn
while doing all these. It should be self-configuring,
self-optimizing, self-healing and self-protecting in a
cognitive way.

In this paper, we analyze some recent trends in the
development of communication networks and
investigate in more detail the concept of cognitive
network. Cognitive networks are promising to be the
major step towards efficient and automatic
management of increasing complexity of
communication networks.

Cyclic Process in Cognitive Network.

All systems that are able to adjust their functioning
according to changes in their environment are based
on feedback information. Cognitive networks are no
exception in this respect, so they will also use a
control loop, also called cognition cycle [7, p. 7],
feedback loop [2], context based adaptation loop [8].
According to Thomas et al. [2], the loop employed
by a cognitive network should be based on the
concept of the Observe-Orient-Decide-Act loop
originally used in the military, augmented by
learning and following end-to-end goals to achieve
cognition. In [8], the loop also has a communicating
capability for communicating with other loops in a
distributed environment.

123

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

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

Vol. 8, No. 6, September 2010

The cognition cycle as described by Mitola [7, p. 8]
features the following states: observe, orient, plan,
decide, act and learn. It uses the orient module for
classifying stimuli and does not explicitly encompass
policies.

Cycle management:

In [10], the authors investigate a cognitive agent for
wireless network selection which is designed to hide
the complexity of the wireless environment from the
user. The selection problem is decomposed

into four elements that enhance the agent to select
the network which is most suitable to user
preferences. First, user's feedback that the decision
making process will be used is captured. Second, the
available services are evaluated against learned user
preferences. Third, the agent decides when to change
services and which new service to select based on
user's preferences, context and goals.

Fourth, the value of previously unseen services is
predicted. Using this approach, the agent
continuously monitors the wireless environment and
selects the best service according to the current
model of user preferences. However, when the user
is unsatisfied (or changes preferences), The model is
updated and a new selection is made to satisfy
preferences.A Cognitive Resource Manager (CRM)
and its conceptual architecture are introduced in
[14]. The CRM's functioning is based on a cognition
cycle adapted from Mitola [7] and aims at enabling
autonomic optimization of the communication stack
as a whole, thus acting as an intelligent vertical
calibration(Fig.l). The intelligence would be based
upon methods from the field of AI.

rr[~

I!
U

»2

Fig 1. Open Systems Interconnection (OSI)
model

Loop for security

The CycSecure application [12] makes use of an
incomplete cognitive loop. It uses daemons installed
on machines in the network that collect local
information and send it to the server when polled. A
human operator can examine and modify the
network model, query and view network statistics.
The system is able to generate possible attack plans
based on the information gathered from the system
and the internal knowledge base. Based on these
attack plans, the human operator can decide for
remedy measures to increase the security of the
system.

Communication Requirements and research
directions

In the history of telecommunications, development
has always been driven by humans need to
communicate, i.e. reliably transmit ever increasing
amount of information across increasing distances.
However, communication networks became
increasingly complex and more difficult to manage,
requiring increasingly specialized tools and human
operators for their maintenance, configuration and
optimization.From the user's point of view the
necessities in the world of telecommunications, as it
is today, are : higher bandwidth or alternative
solutions capable of accommodating the traffic .
These necessities derive from the user's thirst for
digital content.

From the network operators' point of view, some of
the main necessities are: complexity, management,
security, scalability, fault tolerance, fast integration
of new technologies and a good business model [6].
The network operator has to create adequate
premises for delivering the digital content.

These user's and network operators necessities are
actually forming the basis for research activities
currently underway in the area of cognitive
networks. In general, research directions in
communications can be classified in 8 broad
categories: theory, signal processing, networks,
software, user satisfaction, security, management
and next generation protocols and architectures. In
an attempt to obtain an objective big picture of the
trends in research areas as well as quantitative
estimation of the ongoing work, we used ontogeny, a
semi-automatic ontology editor [6] to analyze the

124

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

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

Vol. 8, No. 6, September 2010

conference proceedings of IEEE Globecom 2006 and
2007, totaling 201 1 papers

Artificial intelligence:

Artificial intelligence is concerned with intelligent
behavior in artifacts. Intelligent behavior, in turn,
involves perception, reasoning, learning,
communicating, and acting in complex environments
.Artificial Intelligence has as one of its long-term
goals the development of machines that can do these
things as well as humans can, or possibly even
better. Another goal of AI is to understand behavior
whether it occurs in machines or in humans or other
animals

Intelligent agents developed in a couple of streams
of work, among them is cybernetics [Wiener 198],
cognitive psychology, Computational linguistics
[Chomsky 1914], and adoptive control theory
[Widrow & Hoff 1960], also contributed to the
intellectual matrix developed by Artificial
intelligence.

Intelligent Agents:

Intelligent agents in Artificial intelligence react,
plan, reason and learn in an environment more or
less compatible with its abilities and goals. Here we
shall see how the actions of other agents can be
anticipated in each agents own planning, and indeed,
how an agent can even affect the actions of other
agents in the service of its own goals. To predict
what another agent will do , we need methods for
one agent to model another ; to affect what another
agent will do. There are two kinds of models used by
agents, iconic and feature based. An iconic model of
the environment attempts to simulate relevant
aspects of the environment; a feature-based model
attempts to describe the environment-perhaps by
formulas in the predicate calculus. The agents that
we deploy can use either an iconic or a feature-based
model of the other agents cognitive structure. And
the other agent itself might be presumed to be using
either an iconic or feature-based model. The four
possibilities are shown in table 1 along with the
modeling strategy each one provokes.

Figure 2. Intelligent agents for Cognitive Networks

The starting point towards developing a CN is the
intelligent agent (IA). This section presents existing
and emerging AI techniques that can prove useful for
developing agents for CNs. According to Russell and
Norvig [13, p. 42], an agent is central to AI. It is an
entity that perceives the environment through
sensors and acts upon that environment through
actuators. This is the so called "weak" definition of
agency while "stronger" definitions take into
account functions and characteristics of the agent
[14, p. 8,13, p. 42]. Among different classifications
of agents, we will consider as a reference, the one
established at IBM, which uses three dimensions to
describe agents (see Fig.6). The first dimension is
the Agency, which determines the degree of
"autonomy and authority vested in the agent". The
second dimension is the Intelligence, which
describes the degree of reasoning and learned
behavior. Finally, the third dimension is Mobility,
which specifies the degree to which agents travel
through the network [14, p. 9]. Current networks
operate via message passing (i.e. IP packets between
two routers or primitives between TCP and IP)
where the receiver takes an action as a consequence

of the received message. This type of operation is
asynchronous and is characteristic to expert systems
[14, p. 9,15]. This approach permitted loose coupling
of complex systems (e.g. communication networks).
However, this approach permits the lowest degree of
autonomy according to Fig. 1. On the Intelligence
axis, some of the current communication systems do
not even reach the lowest level as they do not even
allow specification of preferences (e.g. QoS
specifications). In this respect, CNs are expected to
enhance the level of intelligenceof current
communication systemsby incorporating so called

125

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

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

Vol. 8, No. 6, September 2010

Intelligent Agents (IAs) in the KP. On the Agency
axis, IAs can perform actions on behalf of the user,

more specifically they can interact with data,
applications or services. On the Intelligence axis, IAs
can hold a model(i.e. user, system, environment,
etc.), perform reasoning, planning and learning.
These actions are exactly the same as the ones
desired from CN and can be found in the states of the
cognition loop (see Plan, Decide, Act, Learn and
Policy Fig.4).

Mg. i The ccrap rt ion loop.

Networks of the future will make use of agents to
improve their performance with respect to all three
axes in Fig 1 .

Eervic* inlermctivity
Application interactmty
Dala iMGraclMty
Representation of us&f
Asyncshrcny

■■£

ssss

Intelligence

In the case of CNs, the main improvement is
achieved with respect to the Intelligence axis.
Therefore, in the remainder of the section we focus
on describing utility of Intelligent Agents for these
networks. We also emphasize the correspondence
between Intelligent Agents and the states of the
cognition loop. From the intelligence point of view,

the minimal requirement for an Intelligent Agents in
general is to hold a model and be able to reason
based on this model. These IAs (Intelligent Agents
)are also called knowledge-based agents. Reasoning
can take place upon two types of knowledge: certain
(true, false and unknown) and uncertain. Reasoning
under certain knowledge is accomplished by logical
agents. In this respect, agents "can form
representations of the world, use a process of
[logical] inference to derive new representations
about the world, and use these new representations to
deduce what to do" [13, p. 191]. Logical agents use
symbolic knowledge representations, so called
artificial languages, and typically first-order logic to
infer new facts. These representations also support
semantic querying. Agents that have incomplete or
uncertain information use decision theory and are
also called decision theoretic agents. These agents
use knowledge representations specific for uncertain
domains (i.e. full joint distributions can constitute
the knowledge base) to reason. Then they perform
probabilistic inference, which is the computation of
posterior probabilities from the observed evidence.

Conclusions

The recently emerging CN concept is promising to
be the right answer to emerging challenges of the
network management. In this paper we surveyed
existing work on CNs. We first analyzed recent
research trends in communications. We mapped
existing AI techniques to the states of the cognition
loop and identified challenges for research in AI
from which CNs could benefit. We concluded the
paper with identification of standardization activities
related to or potentially benefiting from the research
in the area of CNs.

The discussions in this paper indicate that the way
forward in developing CNs is to bring together the
experts from the areas of communication networks
and AI. Communication networks are faced with
great complexity challenges and several AI
techniques proved to handle complexity well.
Furthermore, AI is searching for areas of
applications, and communication networks are
underexploited in this respect. However, due to the
vastness in Artificial Intelligence field, we hope to
upgrade more in terms of Cognitive Networks and
other methods & tools of AI.

126

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

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

Vol. 8, No. 6, September 2010

References :

[I] Nils J. Nilsson, Artificial Intelligence A New Synthesis.

[2] R.W. Thomas, L.A. DaSilva, A.B. MacKenzie, Cognitive networks, in: Proceedings of the First IEEE
International Symposium on New Frontiers in Dynamic Spectrum Access Networks, Baltimore, MD,
USA, November 8-11, 2006.

[3] J. Mitola, Cognitive Radio - An Integrated Agent Architecture for Software Defined Radio, Ph.D.
Dissertation, Royal Institute of Technology, Kista, Sweden, May 8, 2000

[4] Q. Mahmoud, Cognitive Networks - Towards Self-Aware Networks, John Wiley and Sons, 2007, ISBN
9780700999.

[5] D.D. Clark, C. Partrige, J.C. Ramming, J.T. Wroclawski, A knowledge plane for the internet, in:
Proceedings of the SIGCOMM 2004, Karlsruhe, Germany, August 26-29, 2004.

[6] R.W. Thomas, Cognitive Networks, Ph.D. Dissertation, Virginia Polytechnic and State University,
Blacksburg, VA, June 16, 2007.

[7] J. Mitola, Cognitive Radio - An Integrated Agent Architecture for Software Defined Radio, Ph.D.
Dissertation, Royal Institute of Technology, Kista, Sweden, May 8, 2000.

[8] P. Balamuralidhar, R. Prasad, A context driven architecture for cognitive nodes, Wireless Personal
Communications 16 (2008) 124- 110.

[9] FCC, ET Docket No. 04-422, Notice of Proposed Rule Making and Order, December 2004.
<http://www.scribd.com/doc/112914/ Federal-Corn munications-Commission-FCC04422Al>.

[10] QWL-QoS Ontology. <http://www4.ntu.edu.sg/home6/PG0487868/ OWLQoSOntology.html>
(visited on August 2008).

[II] P. Mahonen, M. Petrova, J. Riihijarvi, M. Wellens, Cognitive wireless networks: your network just
became a teenager, in: Proceedings of the INFOCOM 2006, Barcelona, Spain, April 24-29, 2006.

[12] B. Shepard, C. Matuszek, C.B. Fraser, W. Wechtenhiser, D. Crabbe, Z. Gundordu, J. Jantos, T.
Hughes, L. Lefkowitz, M. Witbrock, D. Lenat, E. Larson, A knowledge-based approach to network
security: applying Cyc in the domain of network risk assessment, in: Proceedings of the Innovative
Applications of Artificial Intelligence Conference, Pittsburgh, PA, USA, July 9-14, 2006.

[13] S. Russell, P. Norvig, Artificial Intelligence: A Modern Approach, second ed., Prentice Hall, 2002,
ISBN 0147901262.

[14] J. Bradshaw, Software Agents, AAAI Press/The MIT Press, 1997, ISBN 0262622412.

[15] P. Jackson, Introduction to Expert Systems, Addison-Wesley International Computer Science Series,
1986, ISBN 0201142246.

AUTHORS PROFILE

I am currently employed in Jazan University, Jazan, K.S.A Department of Computer Networks. My area of
Research is Artificial Intelligence, Data Mining, Network Security. I have published few papers in International
Journals. I wish to do research more into these fields.

1 27 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

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

Vol 8, No. 6, September 2010

A Performance Study on AES Algorithms

B.D.C.N.Prasad 1

Dept. of Computer Applications,

P V P Siddardha Institute of Technology

Vijayawada, India

P E S N Krishna Prasad 2

Dept. of Computer Science & Engineering

Aditya Engineering College

Kakinada, India

P Sita Rama Murty
Dept. of Computer Science & Engineering
Sri Sai Aditya Institute of Science & Technology
Kakinada, India

K Madhavi 4

Dept. of CSE

Dadi Institute of Technology

Anakapalli, India

Abstract — The Aim of this project is to find the performance
comparative analysis of AES algorithms such as MARS, RC6,
Rijndael, Serpent, Twofish algorithms in terms of speed,
memory, time, encryption and decryption, key setup time,
number of rounds, key sizes and also hardware considerations.
Most of the AES algorithms, especially symmetric block
ciphers, are based on the principle of substitution and
transposition to encrypt a plain-text message and to produce a
cipher-message. Those transformations are based on well-
understood Mathematical problems using non-linear functions
and linear modular algebra.

Implementation of cryptographic algorithms mainly uses bit-
level operations and table look-ups. Bit-wise operators (XORs,
AND/OR, etc.), substitutions, logical shifts and permutations
are quite common operations. Such operations are well suited
for their fast execution in hardware platforms. Furthermore,
currently abundant memory resources in hardware platforms
enhance encryption speed for the operations like substitution.
These operators play an important role in analysis and
comparison of the performance of the above mentioned AES
algorithms, to evaluate simple, effective and efficient outcomes
and also the information might be more secure.

Keywords-AES algorithms; Mars; RC6; Rijndeal; Sarpent;
Two fish;

Introduction

Security is a broad topic and covers a multitude of sins,
in its simplest form. It is concerned with making sure that
nosy people cannot read, or worse yet, modify message
intended for other recipients. It is concerned with people
trying to access remote services that they are not authorized
to use. Security also deals with people trying to deny that
they sent certain message.

Network security problems can be divided roughly into
four intertwined areas:

• Confidentiality,

• Authentication and Integrity control

• Denial of service

Cryptography, over the ages, has been practiced by many
who have devised ad-hoc techniques to meet some of the
information security requirements. The last twenty years
have been period of transition as the discipline to a broader
area. There are now several international scientific
conferences denoted exclusively to cryptography and also
and International Association for Crypto-logic Research
(IACR), aimed at fostering research in the area.

There are two general types of cryptographic algorithms.

1. Symmetric algorithms.

2. Asymmetric algorithms.

The current Digital Encryption Standard (DES) does no
longer satisfy the need for data security because of its short
56-bit key. Such short keys can today be broken by brute
force attacks. We are looking for newer and more flexible
algorithms to replace DES. Some of the candidates for the
Advanced Encryption Standard (AES) are MARS encryption
algorithm, RC6, Serpent, Rijndael, and Twofish. These are
symmetric key block ciphers use 128 bit blocks and supports
variable key sizes (from 128 to 1248 bits). These use
addition and subtractions, S -boxes, fixed and data dependent
rotations, and multiplications.

The final AES selection was made on the basis of several
additional characteristics:

• computational efficiency and memory
requirements on a variety of software and
hardware, including smart cards

• flexibility, simplicity
implementation

and

ease

The existing system consisted of files with literally no
file security standards like AES algorithms such as MARS,
RC6, Rijndael, Serpent, and Twofish. AES algorithms are
symmetric cipher algorithms which are far better than DES
algorithms, since DES algorithms are limited key size with
fixed number of blocks. So, we have chosen for finding the
comparison of AES algorithms to provide the security for
Data as well as networks and files. AES algorithms are to be

128

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

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

implemented due to the following factors against which
several security measures had to be taken up:

1. Reading data

2. Manipulating and modifying data

3. Illegal use of files

4. Corrosion of data files

5. Distortion of data transmission

The main issue of (1) is secrecy and confidentiality.
Confidentiality has always played an important role in
diplomatic and military matters. Often Information must
store or transferred from one place to another without being
exposed to an opponent or enemy. Key management is also
related to Confidentiality. This deals with generating,
distributing and storing keys. Items (2-4) are primarily
concerned with reliability. Often the expression integrity is
used as a measure of genuineness of data. Also computer
files and networks must be protected against intruders and
Unauthorized. Item 5 is different aspect of the security of the
information.

Vol 8, No. 6, September 2010
CBC mode within ESP. This mode requires an Initialization
Vector (IV) that is the same size as the block size. Use of a
randomly generated IV prevents generation of identical
cipher text from packets which have identical data that spans
the first block of the cipher algorithm's block size.

The IV is XOR'd with the first plaintext block before it is
encrypted. Then for successive blocks, the previous cipher
text block is XOR'd with the current plaintext, before it is
encrypted. For the use of CBC mode in ESP with 64-bit
ciphers.

2) Key Size
Some cipher algorithms allow for variable sized keys,
while others only allow specific, pre-defined key sizes. The
length of the key typically correlates with the strength of the
algorithm; thus larger keys are usually harder to break than
shorter ones. This article stipulates that all key sizes MUST
be a multiple of 8 bits.

The default key size that implementations MUST support
128 bits. In addition, all of the ciphers accept key sizes of
192 and 256 bits.

A. AES Algorithms

AES algorithms are symmetric cipher algorithms
with variable key sizes and blocks, also with number of
rounds to encrypt and decrypt the data than DES algorithms.
There are numerous algorithms in AES. From them we have
chosen the following algorithms for finding the performance
analysis on time, memory, key sizes, key setup time,
encryption, and decryption and so on.

The Chosen algorithms are as:

• MARS encryption algorithm

• RC6 Algorithm

• Rijndael Algorithm

• Serpent Algorithm

• Twofish Algorithm

TABLE II.

Key sizes

TABLE I.

General Structure

Cipher

Type

Rounds

Using

MARS

Extended
Feistel

Variable Rotation,
Multiplication
Non Cryptic Rounds

RC6

Feistel

Variable Rotation,
Multiplication

Rijndael

Square

10,12,14

Serpent

SP Network

Bitslice

Twofish

Feistel

I) Mode

No operational modes are currently defined for the
AES cipher. The Cipher Block Chaining (CBC) mode is
well-defined and well-understood for symmetric ciphers, and
is currently required for all other ESP ciphers. This article
specifies the use of the AES cipher and the other finalists in

Algorithm

Key Sizes(bits)

Default

MARS

128 - 448*

128

RC6

Variable up to 2040

128

Rijndael

128,192,256

128

Serpent

Variable up to 256**

128

Two fish

Variable up to 256***

128

MARS key lengths must be multiples of 32 bits.

** Serpent keys are always padded to 256 bits. The
padding consists of a "1" bit followed by "0" bits.

*** Twofish keys, other than the default sizes, are always
padded with "0" bits up to the next default size.

3) Weak Keys

Some cipher algorithms have weak keys or keys that
MUST not be used due to their interaction with some aspect
of the cipher's definition. If weak keys are discovered for the
AES or any of the other finalists, then weak keys SHOULD
be checked for and discarded when using manual key
management. When using dynamic key management, weak
key checks SHOULD NOT be performed as they are seen as
an unnecessary added code complexity that could weaken the
intended security.

4) Block Size and Padding

All of the algorithms described in this document use a
block size of sixteen octets (128 bits), mandatory for the
AES. Some of the algorithms can handle larger block sizes
as well. Padding is required by the algorithms to maintain a
16-octet (128-bit) blocksize. Padding MUST be added, such
that the data to be encrypted has a length that is a multiple of
16 octets. Because of the algorithm specific padding
requirement, no additional padding is required to ensure that
the cipher text terminates on a 4-octet boundary (i.e.
maintaining a 16-octet blocksize guarantees that the ESP Pad

129

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

Length and Next Header fields will be right aligned within a
4-octet word

5) Rounds
This variable determines how many times a block is
encrypted. While this variable MAY be negotiated, a default
value MUST always exist when it is not negotiated.

Algorithm

Negotiable

Default of Rounds

MARS

yes

RC6

yes

Rijndael

yes

10,12,14

Serpent

yes

Twofish

yes

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

Vol 8, No. 6, September 2010
1) RC6
RC6 is the submission of MIT (Massachusetts Institute of
Technology) and the RSA-Laboratories. Similar to MARS it
splits the 128 bit blocks into four words in the algorithm, but
the algorithm is designed in a way that you can easily change
to two 64 bit words in newer architectures. RC6 is also a
Feistel network. It uses the same type of operations except
from look-up tables and fixed rotations. The algorithm is
more flexible than MARS about blocksize and number of
rounds. The AES submission is optimized for 128 bit blocks
and 20 rounds. Several performance test showed that RC6 is
slower than MARS for encryption and for the key setup. But
it uses less memory because there are no look-up tables.

B. MARS Algorithm

MARS is a shared-key block cipher that works with
a block size of 128 bit and a variable key size. The algorithm
is a type-3 Feistel network which is word (32 bit) oriented.
The word orientation should bring a performance for
software implementations on most computer architectures
available today. A fully optimized implementation is
expected to run at lOOMbit/second and hardware can achieve
an additional lOx speedup factor.

1) Operations
MARS algorithm uses a big variety of different
operations:

Additions, subtractions and xors: These simple
operations are used to mix data and key values together.
Because xors are interleaved with additions and subtractions
these operations do not commute with each other.

Table look-up: Similar to the S-boxes in DES has also
MARS cipher a table look-up. It uses a single table of 512
32-bit words, simple called S-box. A problem of the table
look-up is the slow software implementation (at least 3
instructions per look-up). That's why S-box look-up is only
used sparely in MARS where fast avalanche of the key bits is
needed.

Fixed rotations: Data-dependent rotations: Data
dependent rotations may lead to differential weaknesses.
This problem is solved in MARS by combining these
rotations with multiplications.

Multiplications: All multiplications in MARS are modulo
232 which suits most modern computer architectures.
Multiplications used to be a problem in cryptographic
algorithms because they were slow. Today is this no longer
the case. Most architecture can complete a multiplication in 2
clock cycles. MARS algorithms uses 16 multiplications per
block. This should not be a big deal. For hardware
realizations we have the problem that a multiplicator needs
much more chip-space than adders or logical units.

C. Comparison with other AES Candidates

There are 4 other candidates for AES in the last
round. So they are all 128 bit block ciphers with variable key
length from 128 bit to at least 192 bit. All designs claim to be
secure against all known attacks like differential, linear,
known plaintext or cipher text and other attacks.

2) Rijndael

Rijndael is the submission of the Belgium Proton World
Int. and the Katholieke Universities Leuven, Belgium. This
algorithm is quite different from MARS. It works with
Galois Field GF(128) arithmetic and handles the input blocks
as matrices of bytes. They define several operations to these
matrices as Byte Sub, ShiftRow, MixColumn and
AddRoundKey. For detailed information about these
operations consult [Rijndael99]. Several combinations of
these operations define a round. Depending on the key length
which is in the range from 128 to 256 bits a fixed number of
rounds has to be executed. This cipher is not a Feistel
network. Several performance tests showed that Rijndael is
about the same speed in encryption and decryption as
MARS. But the key expansion for keys of the same length is
significant slower.

3) Serpent

Serpent is a submission from three universities
(Cambridge University, England; Technion, Haifa, Israel;
University of Bergen, Norway). Therefore it's the only
algorithm where no company stands behind. The algorithm is
pretty similar to DES, it uses permutations, xors, fixed
rotations and shifts and constant table look-up's. The first
version of the algorithm even used the same S-boxes as DES.
The key can vary from 128 to 256 bit. The algorithm works
internally also with 4 words as RC6 and MARS.
Performance tests that the encryption of Serpent is about
25% faster than the MARS encryption. But the key
expansion is significant slower. An implementation of
Serpent also uses a lot of memory because of the look-up
tables.

4) Twofish

Twofish is the submission from a company called
Counterpane. It is a 16 round Feistel cipher that works with
key dependent 8x8 bit look-up tables, 4 by 4 matrices over
the Galois field GF(128), a pseudo-Hadamard transform,
permutations and rotations. The detailed description of these
functions can be found in [Twofish]. The key length varies
also from 128 bit to 256 bit as in most other AES candidates.
Performance tests showed that the encryption speed of
Twofish is about the same as for MARS, but the Twofish key
setup is significant faster.

130

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

D. Performance Analysis

The performance analysis can be done with various
measures such as speed comparison with encryption and
decryption cycles, key setup and key initialization, analysis
of various key sizes and fair speed/security comparisons. The
performance analysis will be presented in the form of tables
and figures below.

1) Speed Comparisons

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

Vol. 8, No. 6, September 2010

TABLE III.

Speed

Key Setup

Cipher

Encrypt
(Cycles)

Decrypt
(Cycles)

Encrypt

Decrypt

Init

MARS

1600

1580

4780

5548

RC6

1436

1406

5186

5148

Rijndael

1276

17742

18886

Serpent

1800

2102

13154

12648

TwoFish

1254

1162

18846

18634

2000
1500
1000 f-
500

□ Encryption(Cycles)
1 ■ Decryption(Cycles)

MARS RC6 Rijndael Serpent TwoFish

Figure 1. Graph for Encryption and Decryption (Cycles)

20000
18000
16000
14000
12000
10000

6000
4000 -j-
2000 -

□ Encryption
■ Decryption

MARS RC6 Rijndael Serpent TwoFish

Figure 2. Graph for Key setup Encryption and Decryption

Initialization

30 -

20
15
10 -
5

D Initialization

MARS RC6

Rijndael Serpent

TwoFish

Figure 3. Key Initialization
2) Analysis on various Key Sizes

a) Encryption

TABLE IV. Encryption

Algorithm

Encryl28

Encryl92

Encry256

MARS

3738

3707

3733

RC6

4698

4740

4733

Rijndael

4855

4664

4481

Serpent

1843

1855

1861

Twofish

1749

1744

b) Decryption

TABLE V.

Decryption

Algorithm

Encryl28

encryl92

encry256

MARS

3965

3936

RC6

4733

4698

4740

Rijndael

4819

4624

4444

Serpent

1873

1897

1896

Twofish

1781

1775

1761

Encryption

5000
4000

s \

— ♦ — Algorithm

B 3000
in

— m — Encryl28
encryl92

* 2000
1000

encry256

12 3 4 5 6

Figure 4. Encryption

131

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

Decryption

r^ J ^ === \

tfj 3000

* 2000

^ n

-Encryl28

-encryl92

encry256

** 4? .*

4* *

x^°

Figure 5. Decryption

c) Fair Speed/ Security Comparisons

TABLE VI. Fair Speed/ Security Comparisons

Cipher

Original
(cycles)

Rounds

Minimal Rounds

Time
(Cycles)

MARS

1600

1000

RC6

1436

Rijndael

1276

1021

Serpent

1800

956

Twofish

1254

940

MARS RC6 Rijndael Serpent Twofish

Figure 6. Fair speed / security comparisons

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

Vol. 8, No. 6, September 2010
algorithms. But MARS is one among the chosen algorithms

is some what better as considered reports.

It won't be an easy decision to choose one of the

finalists as AES. There is no known weakness in all these

algorithms, so other factors as performance, needed

hardware or flexibility must be used for the decision. MARS

cipher is for sure a good candidate. It has the largest

available key length of all of them and it is expandable to

larger block sizes than 128 bit. Another advantage of MARS

is that it comes from a well known company that is in this

business for a long time which means they have a lot of

experience and have proven their trustworthiness.

F. References

[I] Cryptography and Network Security -"William Stallings" /Third
Edition.

[2] The Laws of Cryptography with JAVA Code -"Neal R.Wagner".

[3] MARS: C.Burwick, D. Coppersmith, E.DAvignon, R.Gennaro,
S.Halevi, C.Jutla, S.Matyas, L. O'Connor, M.Peyravian, D.Safford,
N.Zunic, "MARS - a candidate cipher for AES", IBM Corporation,
September 1999.

[4] TweakIBM99 - Shai Halevi, "Detailed discussion of the MARS
"tweak" for Round 2", IBM Corporation, Mai 1999.

[5] RC6: Ronald L. Rivest, M.J.B. Robshaw, R. Sidney, Y.L. Yin, "The
RC6 Block Cipher", M.I.T. Laboratory for Computer Science, RSA
Laboratories.

[6] Rijndael: Joan Daemen, Vincent Rijmen, "AES Proposal: Rijndael",
Proton World Int.l, Belgium, Katholieke Universiteit Leuven,
Belgium, September 1999.

[7] Serpent: Ross Anderson, Eli Biham, Lars Knudsen, "Serpent: A
Proposal for the Advanced Encryption Standard", Cambridge
University, England; Technion, Haifa, Israel; University of Bergen,
Norway.

[8] Twofish: Bruce Schneier, John Kelsey, Doug Whiting, David
Wagner, Chris Hall, Niels Ferguson, "Two sh: A 128-Bit Block
Cipher", Counterpane Systems, University of California Berkeley.

[9] E Biham, " A Note Comparing AES Candidates, NIST,1999.

[10] P. Preneel, V Rijmen and A Bosselaers, " Principles and Performance
of Cryptographic Algorithms", Dr. Dobb's journal.

[II] B. Schneier, J Kelsey, D. Whiting, D Wagner, C. Hall and N
Ferguson, "Performance Comparison of the AES Candidate
conference,1999.

E. Conclusion

A performance comparison can be made among various

AES Algorithms such as MARS, RC6, Rijndael, Serpent,

Twofish. The Performance analysis reports were presented

in the specified contents. It is concluded that all the above

specified algorithms have almost similar speed rate and

timings while using Java tool for execution of these

AUTHORS PROFILE

Dr. B D C N Prasad, currently is a Professor & Head of
Department of Computer Applications at Prasad V. Potluri
Siddardha Institute of Science and Technology,
Vijayawada, Andhra Pradesh, India. He received Ph.D. in
Applied Mathematics from Andhra University,
Visakhapatnam, India in 1984. His research interests

132

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

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

Vol. 8, No. 6, September 2010
includes Machine Intelligence, Data Mining, Rough Sets and Information Security in

Computer Science and Boundary value problems and Fluid Dynamics in Mathematics.

He has several publications in mathematics and computer science in reputed national

and international journals. He is a member of ISTAM , ISTE and also he is a national

executive member of Indian Society for Rough Sets.

Mr. P E S N Krishna Prasad, currently is a Research
Scholor under the guidance of Dr. BDCN Prasad in the
area of Machine Intelligence and Neural Networks. He is
working as Associate Professor in the Department of CSE,
Aditya Engineering College, Kakinada, Andhra pradesh,
India. He is a member of ISTE. He has presented and
published papers in several national and International conferences and journals. His
areas of interest are Artificial Intelligence, Neural Networks and Machine Intelligence.

Mr. P Sita Rama Murty, currently is a Research Scholor, in
the area of ATM networks and Information Secuirty. He is
working as Assistant Professor in the department of CSE,
Sri Sai Aditya Institute of Science and Technology,
Kakinada, Andhra Pradesh, India

133

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Hybrid Fingerprint Image compression and
Decompression Technique

*Dr.R.Seshadri, ,B.Tech„M.E,Ph.D
**Yaswanth Kumar.Avulapti ,M.C.A, M.Tech, (PhD)
***Dr.M.Usha Rani M.C.A,PhD

^Director, S.V.U. Computer Center S.V.University, Tirupati
**Research Scholar, Dept of Computer Science, S.V.University, Tirupati

***Associate Professor, Dept. of Computer Science,, SPMVV, Tirupati

Abstract

In this paper a biometric authentication
system based on Fingerprint. A fingerprint is
the representation of the epidermis of a finger.
It consists of a pattern of interleaved ridges
and valleys.

Like every thing in the body fingerprint
ridges form through a combination of genetic
and environmental factors. Finger prints are
fully formed at about seven months of fetus
development. Fingerprint ridges don't change
throughout the life of an individual except
incase of accidents such as cuts on the
fingertip (or) burns on the fingertip. In this
paper we proposed a hybrid model to
compress the fingerprints.

Keywords :Biometrics,Enrollment
Authentication, compression, Decompression.

Introduction

The term Biometrics is derived from
the Greek word bio (life) and metrics (to
measure). Basically it is a method of
identifying a person based on his/her

physiological or behavioral characteristics
such as Fingerprints, Iris, Face, Hand
geometry, Retinal scan,,DNA,Signature,Key
Stroke, Voice,Gait,Ear,Palm print, Dental
radiographs. Among all the biometric
techniques, fingerprint recognition is the most
popular method and is successfully used in
many applications

Biometrics is a rapidly evolving
wonderful technology which has been widely
used in forensics applications such as criminal
identification and prison security.
Biometrics(Fingerprints) can be used to
prevent unauthorized access to a computer.

The main objective of the fingerprint
image compression is to reduce the number of
bits as much as possible by keeping the
resolution and quality of the fingerprint while
decompressed as same as to the original
fingerprint image.

Fingerprint Identification System

A finger print system works in two
different modes they are Enrollment mode
and Authentication mode as shown in figure. 1

134

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Enrollment mode in which fingerprint system
is used to identify and collect the related
information about the person and his/her
fingerprint image.

Authentication mode in which fingerprint
system is used to identify the person who is
declared to be him/her.

Enrollment Mode

Finger print
Acquisition

Feature
Extraction

Template

Authentication Mode

Finger Print
Acauisition

Feature
Extraction

Matching

Film

Digital

Spatial
resolution

4
linepairs/mm

1024 by
1024 pixels

Data

capacity per
image

1MB

Data rate

30 images
per seconds

30 MB/Sec

Media

one film

4-CDs

Matching Score

Fig. 1. Enrollment and Authentication of a
Fingerprint system

Need for fingerprint compression

Let us see this scenario the spatial
resolution of the cinefilm is generally
assumed to be equivalent to a digital matrix of
at least 1000 by 1000 pixels, each with up to
256 gray levels (8 bit or one byte) of contrast
information.

The Figure. 2 derives from this principal
parameter some examples for requirements on
digital image communication and archival in a
catheterization laboratory with low to mediums
volume.

Fig.2.Replacement of cine film by digital
Imaging with high resolution

In this scenario the enormous data rate
of 30 Megabyte per second has to be
supported. This is much faster than even
advanced ATM networks (offering less than
20MB/s orl60Mbit/s).

Looking for existing off-line media real-time
display from CD-R would require a CD-R
player with a data rate of 200X, while the
fastest players available presently deliver 5 OX
(IX stands for a data rate of 150 KB/sec). The
total amount of data required in this

scenario is even more frightening. To store
these images and make them available over
network (e.g. the internet), compression
techniques are needed.

Steps involved in Fingerprint compression

1) Specifying the rate (bits available) and
distortion parameters for the target

fingerprint image

2) Dividing the fingerprint data into various
classes based on their importance

3) Dividing the available bit budget among
these classes such that the distortion is
minimum.

4) Quantize each class separately using the
bit allocation

5) Encode each class separately and write to
the file

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Steps involved in Fingerprint decompression Steps in Proposed Fingerprint compression

1) Read the fingerprint quantized data from

file using the decoder
2)Dequantize the fingerprint data(Recerse of

step 4)
3)Rebuild the Fingerprint Image(Reverse of

step 2)

Proposed Fingerprint Image Compression
System

We proposed a Fingerprint image
compression system in which the fingerprint
images are compressed by Lossless
compression technique called Run length
Encoding and Decompressed by Huffman
coding. It is also called as Hybrid technique.

A Fingerprint compression system
consists of two blocks namely Encoder and
Decoder as shown in the fig. 3

Fingerprint
Compression

Fingerprint
Decompression

Original
Finger print

Run Length
Encoding

Quantizer

Encoder

Reconstructed
Finger print

Huffman
Encoding

De-Quantizer

Decoder

1) Specifying the bits available and distortion
parameters for the target fingerprint image

2) Dividing the fingerprint data into various
classes based on their importance

3) Dividing the available bit budget among
these classes such that the distortion is
minimum.

4) Quantize each class separately using the bit
allocation

5) Encode each class separately using an Run
Length encoder and write to the file

Steps involved in Fingerprint decompression

1) Read the fingerprint quantized data from

file using the Huffman decoder

2)Dequantize the fingerprint data(Reverse of

step 4)

3 Reconstructed the Fingerprint

image(Reverseof step 2)

In the Run length encoding the

Fingerprint images with repeating grey values
along rows or columns can be compressed by
storing "runs" of identical grey values as
shown in the fig.4 & fig 4.b

Finger print
Grey value 1

Repetition 1

Finger print
Grey value 2

Repetition 2

Fig.4.aRun length encoding the Fingerprint
Images

This a very simple fingerprint
Fig.3.Fingerprint image"co!hpression using compression method used for chronological
Run length Encoding & Decompressed 36 data.The run leifff;^^^fflg /slt f^ces the
using Huffman coding sequences of identical pixels called the runs

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

by tiny symbols. The run length code for
fingerprint image is represented by a sequence
of {A,B} where a is the intensity of the pixel
and B is the number of consecutive pixels
with the intensity A as shown in the figure .If
both A and B are represented by 8-bits this
span of 12 pixels is coded using eight bytes
yielding a compression ratio 1:5.

For Example

23,23,23,23,23,67,67,89,89,78,78

{23,5}{67,2}{89,2}{78,2}

Fig.4.bRun length encoding the Fingerprint
Images

Huffman encoding is performed that is
mapping of the code words to the
corresponding symbols will result in a
compressed data. The original image is
reconstructed i.e. decompression is done by
using Huffman decoding.

Generate a tree equivalent to the
encoding tree. Read input character wise and
left table until last element is reached in the
table. Output the character encodes in the leaf
and returnto the root, and continues until all
the codes of corresponding symbols.

Advantages of Fingerprint Compression:

a) It reduces the storage space at the time of
processing the fingerprints

b) It not only reduces the storage requirements
but also overall execution time

c)It also reduces the probability of
transmission errors since fewer bits are
transferred

d) It also provides a level of security against
criminal monitoring

137

Conclusion

This paper presents different steps
involved in the development of fingerprint
authentication system. The proposed
fingerprint image compression and
decompression technique uses both the Run
length encoding and Huffman coding. This
hybrid fingerprint compression and
decompression technique are good for certain
applications like the security technologies.
These two compression and decompression
techniques are lossless ones. Using this hybrid
technique which leads to less storage of
memory and reducing the execution time.

References

l.A. K. Jain,Patrick Flynn,Arun A.Ross .
"Handbook of Biometrics".
2. The Henry Clas sification System Copyright
© 2003 International Biometric Group
3. Compression Using Fractional Fourier
Transform A Thesis Submitted in the partial
fulfillment of requirement for the award of the
degree of M.E Electronics &Communication.
By Parvinder Kaur.

4. RL-Huffman Encoding for Test
Compression and Power Reduction in Scan
Applications-MEHRDAD NOURANI and
MOHAMMAD H. TEHRANIPOUR, The
University of Texas at Dallas
5.A.B.Watson,"Image Compression using the
DCT" ,Mathematica Journal, 1995,pp.81-88.

6. DAVID A. HUFFMAN, Sept. 1991, profile
Background story: Scientific American, pp.
54-58.

7. Efficient Huffman decoding by MANOJ
Aggarwal and Ajai Narayana A NEW LOSSLESS
METHOD OF IMAGE COMPRESSION AND
DECOMPRESSION USING HUFFMAN CODING
TECHNIQUES by JAGADISH H. PUJAR, 2LOHIT
M. KADLASKAR Faculty, Department of EEE, B V
B College of Engg. & Tech. India

2 Student, Department of EEE, B V B College of
Engg. & Tech. India

8. D. Monro and ^MiWe-.c§^«ilpq uenc y
balance in biorthdpNMASWelets. transactions

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

of the IEEE Int.Conf. on Image Processing,
1:624(627, 1997.

9. A. Said and W. Pearlman. A new fast and
e_cient image codec based on set partitioning
in hierarchical trees. IEEE Transactions on
Circuits and Systems for Video Technology,
6:243(250, June 1996.

10. W. Sweldens. The lifting scheme: A
custom-design construction of biorthogonal
wavelets.

Authors Profile

T^»* i**f

Dr.R.Seshadri was

born in Andhra

Pradesh, India, in 1959.

He received his B.Tech

degree from Nagarjuna

University in

1981. He received his

M.E degree in Control

System Engineering

from PSG College of Technology,

Coimbatore in 1984. He was awarded with

PhD from Sri Venkateswara University,

Tirupati in 1998. He is currently Director,

Computer Center, S.V. University, Tirupati,

India. He has Published number of papers in

national and international conferences,

seminars and journals. At present 12 members

are doing research work under his guidance in

different areas

papers in national and international
conferences, seminars. He attends Number of
work shops in different fields.

Dr. M. Usha Rani is an
Associate Professor in the
Department of Computer
Science and HOD for
CSE&IT, Sri Padmavathi

Mahila
Viswavidyalayam(SPMVV
Womens' University),

Tirupati. She did her Ph.D. in Computer
Science in the area of Artificial Intelligence
and Expert Systems. She is in teaching since
1992. She presented more than 34 papers at
National and Internal Conferences and
published 19 articles in national &
international journals. She also written 4
books like Data Mining - Applications:
Opportunities and Challenges, Superficial
Overview of Data Mining Tools, Data
Warehousing & Data Mining and Intelligent
Systems & Communications. She is guiding
M.Phil and Ph.D. in the areas like Artificial
Intelligence, DataWarehousing and Data
Mining, Computer Networks and Network
Security etc.

Mr.YaswanthKumar
.Avulapati received

his MCA degree with

First class from Sri

Venkateswara

University, Tirupati. He

received his M.Tech

Computer Science and

Engineering degree

with Distinction from

Acharya Nagarjuna University, Guntur.He is a

research scholar in S.V.University Tirupati,38

Andhra Pradesh. He has presented number of

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Punctured Self -Concatenated Trellis Codes
with Iterative Decoding

Labib Francis Gergis

Misr Academy for Engineering and Technology
Mansoura City, Egypt

Abstract-A special concatenated code structure
called self-concatenated trellis code (SCTC) is
presented. This scheme based on only one
recursive convolutional code(# SC), followed by
a mapping modulator. The union bounds of
SCTC are derived for communications over
Additive White Gaussian Noise (AWGN) and
Rayleigh fading channels. Asymptotic results
for large interleavers are extended to M-ary
bandwidth efficient modulation schemes by
puncturing process. The combination of self-
concatenated codes with powerful bandwidth-
efficient component codes leads to a
straightforward encoder structure, and allows
iterative decoding. The scheme has been
investigated for 4-PSK, 8-PSK, 16-PSK, and
16-QAM modulation schemes with varying
overall bandwidth efficiencies. The choice
based on the rate of RSC and puncturer
encoder component.

key words ;S elf -Concatenated codes, trellis-coded
modulation, uniform interleaved coding,

convolutional coding, iterative decoding

1. INTRODUCTION

Trellis coded modulation (TCM) [1] was
originally proposed for transmission over
AWGN and fading channels due to its
attractive bandwidth efficiency.

Concatenated trellis-coded modulation is an
alternative to TCM. Different approaches to
concatenated trellis-coded modulations were
presented in [2], and [3]. The main principle in
the concatenated coding schemes is to use two
codes in series (or parallel) joined through one
or more interleavers. This means that the
information sequence is encoded twice, the

second time after a scrambling of the
information bits.

Concatenated trellis codes are classified as
serially concatenated convolutional codes
(SCCC), these codes were analyzed in [4].
Using the same ingredients, another type of
concatenated codes named parallel
concatenated convolutional codes (PCCC), was
described in [5]. A third choice is defined as a
hybrid concatenation of convolutional codes
(HCCC) was described in [4] and [6]. Self-
concatenated convolutional codes proposed in
[7], [8], and [9] constitute another attractive
iterative detection aided code-family for their
low complexity, since they invoke only a single
encoder and a single decoder.

Puncturing is the process of deleting some
parity bits from the codeword according to a
puncturer code rate. The redundant bits in
coding decrease the bandwidth efficiency.
Puncturing increases code rate without
increasing complexity and decreases free
distances of code. The advantage of punctured
codes for binary transmission is that the
encoders and decoders for the entire class of
codes constructed easily by modifying the
single encoder and decoder for the rate 1/2
binary convolutional code from which the high
rate punctured code was derived [10].

The construction of self-concatenated trellis
codes (SCTC) is described in section 2. Section

3, derives analytical upper bounds to the bit-
error probability of SCTC using the concept of
uniform interleavers. Factors that affect the
performance of SCTC are described in section

4. Finally results for some examples depicted in
section 4, have been stated in section 5.

139

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

2. SCTC MODEL

The basic concept of self-concatenated
scheme is shown in Figure. 1, the input bit
sequence {b^ of the self-concatenated encoder
is interleaved to yield the bit sequence {b 2 }.
After the parallel-to-serial (P/S) conversion, the
information sequence is defined as b (1) = {b 1}1
b 2 ,i b 1)2 b 2 ,2 .... }• The resultant bit sequences
are input to a recursive systematic
convolutional (RSC) encoder. At the output of
the encoder the interleaved bit sequence is
punctured. The encoder output is composed of
the combined systematic bit sequence and
parity bit sequence.

where E b /N is the bit energy to noise density
ratio, A c w>h for block code C represents the
number of codewords of the block code with
output weight h associated with an input
sequence of weight w, and N is the size of the
interleaver. The A c w/l is the input-output weight
coefficient (IOWC). The function Q (^2R h
E b /N ) represents the pairwise error
probability which is a monotonic decreasing
function of the signal to noise ratio and the
output weight ft.

For a fading channels, assuming coherent
detection, and perfect Channel State
Information (CSI), the conditional pairwise
error probability is given by

b (l)

RSC

Encoder
R^l/2

Puncturer

R 2 =l/2

,(i)

Fig 1. The Self -Concatenated
Code Encoder

The overall code rate, R, can be derived
based on [9] as:

R = R t /2R 2 = (1/2) / 2 (1/2) = 1/2

(1)
It can be observed that different codes can
be designed by changing R 2 .

3. PERFORMANCE OF

SELF-CONCATENATED

TRELLIS CODES

Consider a linear block code C with code
rate R, and minimum distance ft m . An upper
bound on the conditional bit-error probability
of the block code C over AWGN channels,
assuming coherent detection, maximum
likelihood decoding, can be obtained in the
form [4]

N/R N

Pb(e/p) < Z Z MV) A\ M •

h=d min w=l

Rh(E b /N )

(2)

Q(\ hREb/N^pt 2 )

(3)

i=l

The fading samples p are independent
identically distributed (i.i.d.) random variables
with Rayleigh density of the form

f(p)= 2 pe

P 2

(4)

The structure of a SCTC, as shown in
Figure .1, is composed of q-1 interleavers each
of size N bits, and a single systematic recursive
trellis code C with rate (bq/bq+1), where only
the b + 1 outputs of the encoder are mapped to
2 b+1 modulation levels.

The average input-output weight
coefficients A c w/l for SCTC with q-1
interleavers can be obtained by averaging
equation (2) over all possible interleavers. A
uniform interleaver is defined as a probabilistic
device that maps a given input word of weight
w into all its distinct pV ~1 permutations with

equal probability 1/ f^V~] •

Thus, the expression for IOWC of SCTC is
derived as [7]

A c

w,h

(5)

JV
w

where A C WW) i|W|h is the number of code words of
the trellis encoder of weight ft, which is
determined in [5], and

140

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

(6)

Substituting equation (6) in equation (2) yields
[3]

P b (e/p) = B m W +1 Q(< 2R h m E b /N )

(7)
where the constant B m is independent of N, and
is derived in [8], and h m is the minimum
Euclidean distance of the SCTC scheme .

4. SCTC: PERFORMANCE
FACTORS

i .e-t

1.e-5 -

1.e-6 -

■

1.e-7 -

T *

♦.. ■

' ■

1.e-8 -
1.e-9 -

'■♦.

'T .
''♦..

■
'▲
>

^»

■

•

N = 10

'♦.

■

1.e-10-

■
▲

N = 50
N = 100

N = 200

' ' ♦ .

'♦..

A
>

1.e-10-

♦

N = 300

'''♦

It is shown from equation (7), that there
are many factors that affect the performance of
SCTC. The most influential parameter is the
interleaver size N. The bit error probabilities
for self-concatenated trellis code with overall
rate R-l/2, is shown in Fig. 2, with various
interleaver lengths N= 10, 50, 100, 200, and 300
are plotted versus the signal-to-noise ratio
E b /N . The systematic and parity bits, b and b ly
are mapped to 4-ary Phase Shift Keying
(QPSK) modulation. The figure shows the
beneficial gain that can be achieved through
increasing N.

HD}^[D}^]Hp-

TCM constituent encoder

Mapper

Fig. 3. Upper Bounds to the Bit Error
Probability for SCTC with QPSK
using different Interleaver Lengths

Applying the upper bound of equation (7),
we obtain the results reported in Fig. 3.1t is
also clear from equation (7) that, the minimum
Euclidean distance of the SCTC code (h m ) is an
another main parameter affecting the
performance of SCTC. Different values of h m
could be obtained by a variety of modulation
schemes. Puncturing is used in order to
increase the achievable bandwidth efficiency.
Different codes could be designed by changing
the rates Ri and R 2 . The output of the encoder
is then mapped to the Gray-code mapping
function. The various coding schemes
considered in this paper are characterized in
Table 1, that defines both R h R 2 , the overall
code rate R, and the associated mapped
modulation scheme to R. The BER versus
E b /N performance curves of the various
QPSK, 8-PSK, 16-PSK, and 16-QAM are
shown in Fig.4.

Fig. 2. Self-Concatenated Trellis Encoder with
rate R = 1/2

141

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

R 2

Modulation
Scheme

1/2

QPSK

1/3

1/4

2/3

8-PSK

1/3

2/3

1/4

16-PSK
16-QAM

Table 1.
Various Modulation Schemes Obtained
from Varying fij and R 2 .

1.e-6

1.e-7

a: 1 .e-8

1.e-9 -

1.e-10- v

2
E h /N n

Fig. 4. Upper Bounds to the Bit Error

Probability for SCTC versus

Different Modulation Schemes

The choice of decoding algorithm and
number of decoder iterations also influences
performance.

A functional diagram of the iterative
decoding algorithm for SCTC is presented in
Fig. 5.

Fig 5. Self -Concatenated Trellis Decoder

The decoder is a self-concatenated scheme
using a soft-input soft-output (SISO) maximum
aposteriori probability (MAP) algorithm [9]. It
first calculates the extrinsic logliklihood Ratio
(LLR) of the information bits, namely L e (bi)
and L e (b 2 ). Then they are appropriately
interleaved to yield the a priori LLRs of the
information bits, namely L> a (bi) and L a (b 2 )> as
shown, in Fig. 5.Self-concatenated decoding
proceeds, until a fixed number of iterations is
reached.

The performance of SCTC with QPSK
modulation schemes considered are shown in
Fig .6. The SCTC has an overall rate R = 111,
the interleaver length N of this code = 100 bits.
The performance after various numbers of
iteration is shown. It is clear that performance
improves as the number of decoder iterations
increases.

5. CONCLUSIONS

In this paper, a channel coding scheme
(SCTC) that is bandwidth efficient and allows
iterative decoding of codes built around
punctured codes together with higher order
signaling.

142

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

1.e-6 -

QPSK Scheme

1.e-7 -

V ^

N = 100

1.e-8 -

1.e-9 -

□

1.e-10-

□

LU
00

1.e-11-
1.e-12-
1.e-13-
1.e-14-
1.e-15-
1.e-16-

•
□

No Iteration

2 Iteration

3 Iteration

4 Iteration

□

)

4 e

E h /N n

Fig. 6. Upper Bounds to the Bit Error
Probability for SCTC versus
Different Decoding Iterations

The SCTC schemes consists of binary RSC
codes and different puncturing rates. The
puncturer is used to increase the achievable
bandwidth efficiency. A search for good rates
was performed, taking into account the
puncturing at the transmitter. It is also
demonstrated the significant in the
performance and the decrease of the bit error
rate and probability of errors to SCTC within
increasing: the interleaver size N, and the
number of decoder iterations

REFERENCES

[1] G. Ungerboeck, "Channel coding
with multilevel phase signaling,"
IEEE Trans. Inf. Th., Vol. 25, pp. 55-
67, Jan. 1982

[2] F. Brannstrom, A. Amat, and L.

Rasmussen, " A General Structure for
Rate-Compatible Concatenated
Codes," IEEE Trans on Communication
Letters, Vol. 11, pp. 437-439, May 2007.

[3] A. Amat, G. Montorsi, and S.
Benedetto, " New High-rate
Convolutional Codes for Concatenated
Schemes", Proceeding of IEEE
International Conference on
Communication, ICC 2002,Vol. 3, pp.
1661-1666, April 2002.
[4] D. Divsalar, and F. Pollara, " Serial and
Hybrid Concatenated Codes with
Applications, " International Symposium
on Turbo Codes and Related Topics,
Brest, France 1997
[5] S. Benedetto, and G. Montorsi,"
Unveiling Turbo Codes: Some Results
On Parallel Concatenated Coding
Schemes, "IEEE Transactions on
Information Theory, Vol. 42, No. 2,
March 1996
[6] A. Amat, and E. Rosnes," Good
Concatenated Code Ensembles for the
Binary Erasure Channel", IEEE
Journal on Selected Areas in
Communications, Vol. 27, No. 6, pp.
928-943, August 2009.
[7] S. Ng, M. Butt, and L. Hanzo, " On the
Union Bounds of Self-Concatenated
Convolutional Codes", IEEE Signal
Processing Letters, Vol. 16, No. 16, No.
9, September 2009.
[8] S. Benedetto, D. Divsalar, G. Montorsi,
and F. Pollara, " Self-Concatenated
Codes with Self-Iterative Decoding for
Power and Bandwidth Efficiency"
International Symposium on
Information Theory, ISIT 1998,
Cambridge, MA, USA, August 1998.
[9] M. Butt, R. Riaz, S. Ng, and
L. Hanzo, Distributed Self-
Concatenated Codes for Low-
Complexity Power-Efficient
Cooperative Communication", IEEE
VTC 2009, Anchorage, Alasks, USA,
2009.
[10] R. Deshmukh, and S. Ladhake, "
Analysis of Various Puncturing
Patterns and Code Rates: Turbo
Code", International Journal of
Electronic Engineering Research
Volume 1, Number 2, pp.79-88, India,
2009.

143

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

AUTHOR PROFILE

Labib F.Gergis received the Bsc, Msc, and
Ph.D from faculty of engineering, Mansoura
University, Egypt, in 1980, 1990, and 2000,
respectively. He is presently in Misr Academy
for Engineering and Technology, Egypt. His
areas of interest include digital
communications, Coding, and Multiple Access.

144 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

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

Vol. 8„ No. 6, 2010

Application of Fuzzy Composition Relation For DNA

Sequence Classification

Amrita Priyam

Dept. of Computer Science and Engineering

Birla Institute of Technology

Ranchi, India.

Abstract — Abstract — This paper presents a probabilistic
approach for DNA sequence analysis. A DNA sequence
consists of an arrangement of the four nucleotides A, C, T
and G. There are various representation schemes for a
DNA sequence. This paper uses a representation scheme in
which the probability of a symbol depends only on the
occurrence of the previous symbol. This type of model is
defined by two parameters, a set of states Q, which emit
symbols and a set of transitions between the states. Each

transition has an associated transition probability, CZ Z/ - ,

which represents the conditional probability of going to
state j in the next step, given that the current state is i.
Further, the paper combines the different types of
classification classes using a Fuzzy composition relation.
Finally a log-odd ratio is used for deciding to which class
the given sequence belongs to.

Keywords-component; Transition Probability, Fuzzy Composition
Relation, Log-Odd ratio

Introduction

A DNA sequence is a succession of the letters A, C, T and G.
The sequences are any combination of these letters. A physical
or mathematical model of a system produces a sequence of
symbols according to a certain probability associated with
them. This is known as a stochastic process, that is, it is a
mathematical model for a biological system which is governed
by a set of probability measure. The occurrence of the letters
can lead us to the further study of genetic disorder. There are
various representation schemes for a DNA sequence. This
paper uses a representation scheme in which the probability of
a symbol depends only on the occurrence of the previous
symbol and not on any other symbol before that. This type of
model is defined by two parameters, a set of states Q, which
emit symbols and a set of transitions between the states. Each

transition has on associated transition probability, CL^ , which

represents the conditional probability of going to state jfrom
state i in the next step, given that the current state is z. Each

B. M. Karan + , G. Sahoo + +

+ Dept. of Electrical and Electronics Engineering

++ Dept. of Information Technology

Birla Institute of Technology

Ranchi, India

class has a set of transition probabilities associated with it.
This transition probability is the measure of going from one
state to another. Now, each class has a set of transition
probability associated with it. We further group the similar
classes and their respective transition probability is merged
using a fuzzy composition relation. Finally a log-odd ratio is
used for deciding to which class the given sequence belongs
to.

II. DNA SEQUENCES

DNA sequence is a succession of letters representing the
primary structure of a real or hypothetical DNA molecule or
strand, with the capacity to carry information as described by
the central dogma of molecular biology. There are 4
nucleotide bases (A - Adenine, C - Cytosine, G - Guanine, T
- Thymine). DNA sequencing is the process of determining
the exact order of the bases A, T, C and G in a piece of DNA
[3]. In essence, the DNA is used as a template to generate a set
of fragments that differ in length from each other by a single
base. The fragments are then separated by size, and the bases
at the end are identified, recreating the original sequence of
the DNA[8][9]. The most commonly used method of
sequencing DNA the dideoxy or chain termination method
was developed by Fred Sanger in 1977 (for which he won his
second Nobel Prize). The key to the method is the use of
modified bases called dideoxy bases; when a piece of DNA is
being replicated and a dideoxy base is incorporated into the
new chain, it stops the replication reaction.

Most DNA sequencing is carried out using the chain
termination method [4]. This involves the synthesis of new
DNA strands on a single standard template and the random
incorporation of chain-terminating nucleotide analogues. The
chain termination method produces a set of DNA molecules
differing in length by one nucleotide. The last base in each
molecule can be identified by way of a unique label.
Separation of these DNA molecules according to size places
them in correct order to read off the sequence.

145

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

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

Vol. 8„ No. 6,2010

III. A Probabilistic approach for sequence

REPRESENTATION

A DNA sequence is essentially represented as a string of four
characters A, C, T, G and looks something like
ACCTGACCTTACG. These strings can also be represented in
terms of some probability measures and using these measures
it can be depicted graphically as well. This graphical
representation matches the Markov Hidden Model. A physical
or mathematical model of a system produces a sequence of
symbols according to a certain probability associated with
them. This is known as a stochastic process [2]. There are
different ways to use probabilities for depicting the DNA
sequences. The diagrammatical representation can be shown
as follows:

O OG

Q.02

FIG 1: [The states of A, C, G and T.]

For example, the transition probability from state G to state T

is 0.08, i,e,

P( x . = T | x M = G) = 0.08

In a given sequence x of length L, x h x 2 , x L , represent the

nucleotides. The sequence starts at the first state x h and makes
successive transitions to x 2 , x 3 and so on, till x L . Using Markov
property [6], the probability of x L , depends on the value of
only the previous state, x L . ls not on the entire previous
sequence. This characteristic is known as Markov property [5]
and can be written as:

P(x) = P(x L I x^Pix^ I x L _ 2 ) P(x 2 1 x l )P(x l )

= P(x 1 )fl miVi)

(1)

In Equation (1) we need to specify PfxJ, the probability of the
starting state. For simplicity, we would like to model this as a
transition too. This can be done by adding a begin state,
denoted by 0, so that the starting state becomes x =0.

p M=FK,

(2)

If there are n classes, then we calculate the probability of a
sequence x being in all the classes. To overcome this
drawback we use Fuzzy composition relation. That is, we
divide the n classes into different groups based on their
similarities. So, if out of n classes, m are similar then they are
treated as one group and their individual transition probability
tables are merged using the fuzzy composition relation. The
remaining (n - m) classes are similarly grouped. Lets say, if
there are two classes Rl and R2, the Fuzzy composition
relation between Rl and R2 [6] [7] can be written as follows:

R 1 °R 2 = Max{Min{R 1 {x\R 2 {y)))

(3)

O O A A ,_

O A
O O

Different class representation

Grouping of similar classes

Fig 2: Grouping of similar classes

A table is then constructed representing the entire (n - m)
similar classes. From this table we compute the probability
that a sequence x belongs to a given group using the following
equation:

log

1*,-

P(x|+)

P(x I -) — a^x,

= 2>g—

(4)

Here "+" represents transition probability of the sequence
belonging to one of the classes using fuzzy composition
relation and "-" represents the transition probability of the
same for another class [1].

If this ratio is greater than zero then we can say that the
sequence x is from the first class else from the other one.

An Example:

Let us consider an example for applying this classification
method. We have taken into consideration the Swine flu
data. [11] The different categories of the Swine flu data are
shown as R h R 2 and R 3 .

R 1} R 2 and R 3 shows the Transition Probability of Type 1, Type
2 and Type 3 varieties of Avian Flu.

Now considering O x . x . , the transition probability we can
rewrite (1) as

146

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

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

Vol. 8„ No. 6,2010

"0.13

0.06

0.09

0.08

0.09
0.06

0.04
0.05

0.05
0.05

0.02
0.07

0.08

0.04

0.05

0.06

"0.13

0.06

0.09

0.08"

R 2

~ T

0.08
0.06

0.04
0.05

0.04
0.06

0.02
0.07

0.08

0.04

0.06

0.09

0.06

0.07
0.04

0.05
0.05

0.06
0.07

0.07
0.07

0.07

0.09

0.04

0.05

"0.07

0.05

0.07

0.04'

x 3 = c

0.05
0.03

0.06
0.06

0.06
0.04

0.08
0.09

0.06

0.09

0.06

0.07

"0.08

0.05

0.08

~ T

0.08
0.05

0.03
0.06

0.06
0.06

0.02
0.09

0.08

0.06

0.04

0.07

Using the Fuzzy composition relation technique on R x and R 2
and then the result of the application with relation R 3 we get
the final table for the Swine Flu class as:

Applying Fuzzy composition relation to these tables we get
the final table as

"0.07

0.06

0.05

0.06

0.08

0.09

0.07

0.06

0.04

0.03

0.07

0.06

0.05

0.08

0.04

0.02

Suppose we are given the sequence x which is to be classified
as either falling into any of the given classes and say x =
CGCG

"0.08

0.06

0.08

0.09

0.08

0.06

0.08

0.09

0.07

0.06

0.07

0.08

0.06

0.08

Similarly, we can repeat the same procedure for another class
Staphylococcus. X lf X 2 and X 3 shows the Transition
Probability of Type 1, Type 2 and Type 3 varieties of
Staphylococcus.

"0.08

0.05

0.07

0.06

0.06
0.04

0.05
0.06

0.06
0.08

0.08
0.07

0.08

0.09

0.03

0.05

From the final fuzzy composition table the log odds ratio of
this sequence is:

, 0.09 , 0.06 , 0.09 „__„
log + log + log = 0.032270

0.08

0.07

0.08

Now, since this ratio is greater than 0, we can conclude that
the input sequence x belongs to the class Avian Flu. If further
classification on the data is needed we will then consult the
individual transition probabilities for all the three types.

CONCLUSION

In this paper we have used a probabilistic function for the
Markov Property. We have applied this for probabilistic
determination in the case of Avian flu virus and
Staphylococcus. The paper also presented a way for
identifying particular classes of genes or proteins. A given
input sequence can belong to either of the given classes. By
using a transition probability measure, one had to determine a
value for each class even though they were similar. The paper
presented a scheme such that the similar classes were merged
by using the fuzzy composition relation and now instead of
calculating each individual probability measure, one measure

147

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

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

Vol. 8„ No. 6,2010

is sufficient to depict all the similar classes. This measure is
further used in the log odds ratio to finally predict the class of
the input sequence.

References

[I] Anna Loanova, "Introduction to (log) odds ratio statistics and
Methodology", University of Groningen, 2008.

[2] C. E. Shanon, "A Mathematical Theory of Communciation",

The Bell System Technical Journal, Vol. 27, pp. 379 - 423, 623
-656.

[3] D. W. Mount, " Bioinformatics, Sequence and Genome
Analysis", 2 nd edition, CSHL Press, (2004). (3)

[4] Durbin, Eddy, Krogh, Mitchison, "Biological Sequence
Analysis", Cambridge University Press, 1998.

[5] L. R Rabiner, "A tutorial on Hidden Markov Models and

selected Application in speech recognition" -Proceeding of the

IEEE, Vol.77, No.2 Feb.1989.
[6] Lee, Kwang Hyung, "First course on Fuzzy Theory and

Applications", Advances in Soft Computing, Vol. 27, Springer,

2005.
[7] Michael Hanss , "Applied Fuzzy Arithmetic : An Introduction

with Engineering Applications", Springer, 2005.

[8] T. Dewey and M. Herzel, "Application of Information Theory to

Biology", Pacific Symposium on Biocomputing, 5:597 - 598
(2000).

[9] W. J. Ewens, G. R. Grant, "Statistical Methods in
Bioinformatics: An Introduction", Vol. 13, 2 nd edition, Springer.

[10] Y. Ephraim, L. R. Rabiner, "On the Relations Between
Modeling Approaches for Speech Recognition", IEEE
transactions on Information Theory, vol. 36, no. 2, March 1990.

[II] A.Priyam, B.M.Karan, G.Sahoo, "A Probabilistic Model For
Sequence Analysis", Inter national Journal of Computer Science

and Information security vol7 No. 1, (2010) 244-247.

148 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Data Security in Mobile Ad Hoc Networks using
Genetic Based Biometrics

B. Shanthini, Research Scholar

CSE Department

Anna University

Chennai, India

S. Swamynathan, Assistant FPofessor

CSE Department

Anna University

Chennai, India

Abstract — A mobile ad hoc network (1 AINKI) -is^^a^setf
configuring, dynamic, multi hop radio network without any fixed
infrastructure. 1 AlNKIs -fffe^coHections of wireless mobile
devices with restricted broadcast range and resources and
communication is achieved by relaying data along appropriate
routes that are dynamically discovered and maintained through
collaboration between the nodes. The main challenge in the
design of such networks is how to prevent the attacks against
data such as unauthorized data modification, impersonation etc.
En&metrics provides possible solutions for this security problem
in 1 AINKI MHceJthas the direct connection with user identity
and needs little user interruption. So, researchers have been
investigating ways to use biometric features of the user rather
than memorable password or passphrase, in an attempt to
produce tough and repeatable cryptographic keys. In this paper
such a security system based on EE&metrics and dhetic

A. Security challenges in MANET

Wireless ad hoc networks are vulnerable to various attacks
[1]. Adversaries may attempt passive and active attacks to gam
unauthori^d access to classified information, modify the
information, delete the information or disrupt the information
flow. The best way to protect data information in a most fme-
g^anular way is by providing__security at the application layer. It
is highly desirable to handle data confidentiality and integrity
in application layer, since this is the easiest way to protect data
from altering^ fabrication and compromise. With the rapid
evolution of wireless technology the reliance of ad hoc
networks to carry mission critical information is rapidly
growing^ This is especially important in a military scenario
where strategic and tactical information is sent. Therefore the
ability to achieve a highly secure authentication is becoming

algorithm which is providing data security in 1 AINL11 tiP^-E 016 cn tical.
presented.

Keywords— Mobile Ad hoc Networks, Data Security,
Biometrics, Genetic Algorithm

I. INTRODUCTION

Mobile ad hoc networks are seen as autonomous that can be
quickly formed, on demand, for specific tasks and mission
support. Communication generally happens through wireless
links, in which nodes within a radio range, communicate and
coordinate to create a virtual and temporary communication
infrastructure for data routin^and data transmission. MANET
can operate in isolation or in coordination with a wired networks
through a gateway node participatin^in both networks. This
flexibility along^ with their self-organi^ng^ capabilities, are
some of their biggest strengths, as well as their biggest security
weaknesses.

The applications of MANET include the foremost
situations such as emergency/crisis management, military,
healthcare, disaster relief operations and intelligent
transportation systems. So message security plays a vital role in
data transmission in MANET. However, because of the
absence of an established infrastructure or centralis
administration, implementation of hard-cryptographic
algorithms is a challenging prospect. So, in this paper, we
present a novel security method using_genetic based biometric
cryptography for message security in mobile ad hoc networks.

Numerous countermeasures such as strong_,authentication,
encrypting and decrypting^ the messages using_, traditional
cryptographic algorithms and redundant transmission can be
used to tackle these attacks. Even though these traditional
approaches play an important role in achieving_eonfidentiality,
integrity, authentication and non-repudiation, these are not
sufficient for more sensitive and mission-critical applications
and they can address only a subset of the threats. Moreover,
MANETs [2] cannot support complex computations or higl}
communication over head due to the limited memory and
limited computation power of mobile nodes.

B. Necessity of Biometrics Security

For mission-critical applications such as a military
application may have higher requirements regarding data or
information security. In such a scenario, we may design the
security system combining_,both biometrics and cryptography.
Biometric based security scheme overcome the limitations of
traditional security solutions. Biometrics refers to the methods
for uniquely recogni^ng^ humans based upon one or more
intrinsic physical or behavioral traits like fingerprints, iris,
retina scans, hand, face, ear geometry, hand vein, nail bed,
DNA, palm print, signature, voice, keystroke or mouse
dynamics, and gajt analysis etc.

149

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Biometric technologies have confirmed its importance in
the fields such as security, access control and monitoring
applications. The tradeoffs among., these biometric
technologies really depend on the application and security
level involved. The best biometric technology [3] that can
easily be deployable in ad hoc networks is fingerprint
recognition. Fingerprints have been successfully used in
civilian identification for years because of their
unchanggability during_,the human life time and uniqueness of
each individual. As biometrics can't be borrowed, stolen, or
forgotten, and forging^ is practically impossible, it has been
presented as a natural identity tool that offers greater security
and convenience than traditional methods of personal
recognition.

Even though biometric has advantages, it also raises many
security and privacy concerns as giyen below:

i. Biometric is authentic but not secret,
ii. Biometric cannot be revoked or cancelled,
iii. If a biometric is lost once, it is compromised forever,
iv. Cross-matching^can be used to track^individuals without
their consent.

To overcome these disadvantages, instead of usmg^the
original biometric, a set of features are taken from it and
transformed using_, genetic alggrithm. If a biometric is
compromised, it can be simply reenrolled usmg_janother feature
set and another genetic operation, thus providing^revocability
and the privacy of the biometric is preserved.

C. Genetic Algorithms

Genetic algorithms [4] are a family of computational
models inspired by natural evolution. They belong_,to the field
of evolutionary computation and are based on three main
operators: Selection selects the fittest individuals, called
parents that contribute to the reproduction of the population at
the next generation, Crossover combines two parents to form
children for the next generation and Mutation applies random
changes to individual parents to form children. Two-point
crossover operator is used here which has the ability to generate,
promote, and juxtapose building blocks to form the optimal
strings.

This paper is organi^d into 4 sections. Section 1 introduces
the background and initiatives of the research. It also discusses
the challenges of message security, the necessity of biometric
security in MANET and Genetic alggrithms. Section 2 explains
the related research works that has been done to provide
security in MANET. Section 3 proposes a new security scheme
for MANET which combines genetic algorithm and biometrics.
Section 4 contains conclusion and suggestions for future
research.

II. RELATED WOIL

A few research works that has been done for data security
in MANET, the various approaches of biometric security and
Genetic alggrithms in security are briefly presented.

Qingljan Xiao [5] introduced a new strategy for
authentication of mobile users. Each user has a profile which
contains all the information of the ID holders. The group leader
also maintains the biometric templates of the group members.
Instead of a central authentication server, the group leaders act
as distributed authenticators. Each group has a shared
cryptographic key which is used for cryptographic
communication within the group. The proposed approach is
designed for hig^ security small group coalition operations and
may not be suitable for enterprise usage.

Jq Liu et al. [6] proposed an optimal biometric-based
continuous authentication scheme in MANET which
distinguished two classes of authentications: user-to-device and
de vice-to -networks This model focused on the user-to-device
class and it can optimally control whether or not to perform
authentication as well as which biometrics to use to minimis
the usage °f system resources.

B Ananda>Lrishna et al. [7] depicted a model which used
multiple alggrithms for encryption and decryption. Each time a
data packet is sent to the application layer it is encrypted using.,
one of these randomly selected alggrithms. When responses are
analy^d they giye a random pattern and difficult to know
neither algorithms nor keys. The proposed scheme worked
well for heavily loaded networks with hig^ mobility.

2Lr^ L et al. [8] explained the context of the study of
Genetic Algorithms as an aiding tool for generating and
optimi2ng^ecurity protocols. This paper explains how security
protocols can be represented as binary strings,, how GA tools
are used to define genome interpretation in optimi^tion
problems.

B. Shanthini et al. [9] explained Cancelable Biometric-
Based Security System (CBBSS), where cancelable biometrics
is used for data security in mobile ad hoc networks. Fingerprint
feature of the receiver is coupled with the tokened random
data by usin^ inner-product algorithm and this product is
discreti^d based on a threshold to produce a set of private
binary code which is actinias a crypto graphic key in this
system.

A. Jigadeesan et al. [10], proposed an efficient approach
based on multimodal biometrics (Iris and fingerprint) for
generating^ secure cryptographic key, where the security is
further enhanced with the difficulty of factoring large
numbers. At first, the features, minutiae points and texture
properties are extracted from the fingerprint and iris images
respectively. Then, the extracted features are fused at the
feature level to obtain the multi-biometric template. Finally, a
multi-biometric template is used for generating a 256-bit
cryptographic key.

III. FTtOroSED WOIL

In this proposed Genetic-Based Biometric Security System
(GBBSS), a genetic two-point crossover operator is applied on
biometric feature set and is used for data security in mobile ad
hoc networks. The main objective of the proposed security
scheme is to improvise the existing^data security approaches
for MANET to suit technology enhancements and to study the
networkzperformance.

150

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

A. Generation of Genetic-Based Biometric Key

In this model all the grpup members maintain the biometric
templates of the other group members. Suppose a member
wants to send a message, to any other member, the receiver's
fingerprint is divided into slices and feature set taken from the
slices is undergone a genetic two-point crossover operation and
the result is the cryptographic key in this system. Generation of
cryptographic key is shown in figure 1.

Hhgerprint

Parent

^^^■^^^^^^^^^^H

Crossover
Points

■

Children ^^^^^^

Cryptographic^ ey

Figure 1 . Generation of cryptographic key from the finger print features.

The same key is generated by the receiver by using^his
biometric and the same sort of cross over operations and is
used for decryption.

Example:
Rrent

Children

01011100
00110011

1010000011111010
1111000011110000

After Crossover
1111000011110000
1010000011111010

00110101
11001100

If this biometric based key is compromised a new one can
be issued by using_a different set of features and different cross
over operation and the compromised one is rendered
completely useless. It can also be an application specific that is
different sets of fingerprint features can be used with different
cross over operations to generate respective crypto graphic key
for different applications.

B. Securing the Data

Data is secured by applying^ this cryptographic key to
encrypt the actual message, using_, a simple cryptographic
algprithm say Fiestel algorithm. The encryption and decryption
processes are specified by the formulae:

C = % R (F) andF^D* R (C)

where F*- Hain Text
C - Cipher Text
^LR^Ley created by Receiver's Biometric
E - Encryption Algprithm
D - Decryption Algprithm

In Fiestal alggrithm, a blockzof si^ N is divided into two
halves, of length N/2, the left half called XL and rig^t half
called XR. The output of the ith round is determined from the
output of the (i-l)th round. The same key is used for all
iterations without generating^ sub keys. Also the number of
iterations performed is reduced to show that security can be
achieved by using_, simple algprithm. For example if the
plaintext is of 512 bytes, then encryption is performed for
every 64 bits and the process is repeated until all 512 bytes are
encrypted. Fiestel structure is given in figure 2. [1].

Pkirdfiid(2wrbits)

(M3-

Key Gene rat ion
Algorithm

Key

<EM3-

Key

G-&-

Key

Ciprerbeirt [A<v b its)

Figyre 2. Fiestel Algorithm

Algorithm for Encryption:

1 . Divide the plaintext into two blocks of si^, 32 bytes,
XL and XR

2. ForI=lto32

Do XL = XL XOR^ey
XR = F (XL) XOR XR
Swap XL, XR
jbin XL, XR

3 . Repeat step 2 until the entire plaintext is encrypted

Algprithm for Decryption:

Do the reverse operation of Encryption process.

151

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

C. Implementation ofGBBSS in MANET

The proposed scheme can be implemented over any unicast
routing_protocols like DSR or AODV which discover routes as
and when necessary and the routes are maintained just as long^
as necessary. A typical MANET is shown in figure 3.

Suppose User A wants to send the message, to User C, after
the forward and reverse paths are set up by the route discovery
method, the data will be sent through that path to the
destination C. Before sendmg_,the data through that path, the
data will be encrypted by Fiestel algorithm using_,the genetic
based biometric key. Once the cipher text is received by the
receiver, the cipher text is decrypted by using_Jhe same key.

Figure 3. MANET Structure

The security functions of the proposed system

Confidentiality: The privacy of the message, is protected
by this scheme. Suppose if the attacker wants to derive the
original message, from the encrypted text, he needs the
cryptographic key. The key can be obtained only by using^
the biometric of the receiver. Furthermore the biometric is
not used as such instead a cancelable version is used. So, it
is computationally infeasible to ge,t the key.

Authentication: In our proposed scheme, the members of
the ad hoc group can authenticate each other through their
biometric. If the receiver wants to verify whether the
message, is commg_from the genuine sender, the message,
can be encrypted by using_the sender's biometric and the
receiver can use the same biometric to decrypt the
message,. These processes can be specified by the
following_£brmulae:

C = % s ( F=) and F^ % s ( C )

where>Ls is th^Ley created by Sender's Biometric.

Integrity: In our proposed scheme, the recipient can verify
whether the received message, is the original one that was
sent by the sender. If the attacker changes the cipher text,
the original plain text can not be generated after decrypting^
with the key created by using_receiver's biometric. By the
property of one-way hash function, it is computationally
infeasible for the attacker to modify the cipher text.

Man-in-the-middle attackr An attacker sits between the
sender and the receiver and sniffs any information beings
sent between two ends is called man in the middle attackz

Even though the attacker can ge,t the cipher text he cannot
view the original message, since it is secured using_ge,netic
based biometric cryptography.

E. Security Analysis

This section reports the analysis of the security parameters
like time taken for key generation, encryption and decryption
for various algorithms like 3DES192, AES128, AES256 and
GBBSS64 in an ad hoc networks environment. The graphs
shown in figure 4 and figure 5 are generated by usin^the
values giyen in the folio wmgjable 1 :

Encryption
Algorithm

Parameters

Key
Size

Time taken

for Key
Generation

Time taken for
Encryption

Time taken for
Decryption

3DES192

192

0.08 ms

0.07 ms

AES-128

128

0.13 ms

0.1 ms

AES-256

256

0.13 ms

0.12 ms

0.11 ms

GBBSS-64

0.06 ms

0.04 ms

0.02 ms

Table L>Ley si^ and Timing_nieasurements for various algorithms

0.14 -
0.12 -

</> 0.1 -

c 0.08 -

| 0.06 -

•- 0.04 -

0.02 -

3DES-192 AES-128 AES-256 GBBSS-64
Algorithms Applied

- Key Generation — ■— Encryption Decryption

Figure 4. Timing_nieasurements for various algorithms

3DES192 AES-128 AES-256 GBBSS-64
Algorithms Applied

- Key size

- Security Level |

Figure 5.^Eey Si^ and Security Levels for various algorithms

From the above charts we can understand that our proposed
GBBSS achieves relatively higl} performance in terms of less
overhead and higl} security level. Since the key si^ is very
small compared to the other algorithms, the time taken to
generate the key, time taken to encrypt and decrypt are also
less.

152

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

IV. CONCLUSION AND FUTURE WO^

Although MANET is a very promising^ technology,
challenges are slowing its development and deployment.
Traditional security mechanisms are not sufficient for the
nodes roaming^in a hostile environment with relatively poor
physical protection. Therefore to strengthen the encryption
algorithm and key, first the advantages of biometric and
genetic algorithms are taken into our system. Secondly,
security should be achieved by using^ simple algorithms that
involve small inherent delays rather than complex algorithms
which occupy considerable memory and delay. Finally, ad hoc
networks may consist of thousands of nodes. So, security
mechanisms should be scalable to handle such a large networks

The method presented in this paper remains as a
preliminary approach to reali^ biometric security in ad hoc
networks which needs higjj security. This approach can be used
in very critical, crucial and vital applications where data
security is very important and members who have accessed that
data is limited in number like military officers at war-field,
scientists in a confidential conference, officers in the intelligent
buildings, etc. There are many security problems still persist in
these types of ad-hoc networks and as a future work^this paper
can be extended to solve those problems with different
biometrics and also with multimodal biometrics.

[5] Qin^an Xiao, "A Biometric Authentication Approach for
Higl} Security Ad hoc Networks", F?bceedmg§ of IEEE
Workshop on Information Assistance, pp. 250-256, Jme
2004.

[6] Je Liu, F. Richard Yu, Chung^Horng^Lung_,and Helen
Tang^ "Optimal Biometric-Based Continuous
Authentication in Mobile Ad hoc Networks", Third IEEE
International Conference on Wireless and Mobile
Computing^Networkmg-and Communications, pp. 76-81,
2007.

[7] B Ananda^Lrishna, S Radha an(M_ Chenna^Lesava
Reddy, "Data Security in Ad hoc Networks using_,
Randomi^tion of Cryptographic Algorithms", Jmrnal of
Applied Sciences, pp. 4007-4012, 2007.

[8] 2Lr^ L., Fe"gueroles J and Soriano M "Interpretation of
Binary Strings, as Security Frbtocols for their Evolution
by means of Genetic Algorithms", International
Conference on Database and Expert Systems
Applications, pp. 708-712, 2007.

[9] B. Shanthini and S. Swamynathan "A Cancelable
Biometric-Based Security System for Mobile Ad Hoc
Networks", International Conference on Computer
Technology (ICONCT 09), pp. 179-184, December, 2009.

[10] A. Jtgadeesan, T. Thillaikkarasi ancf^L. Duraiswamy,
" Crypto gr.aphic^Ley Generation from Multiple Biometric
Modalities: Fusing Minutiae with Iris Feature",
International Jmrnal of Computer Applications , Vol. 2,
No.6, pp. 0975-8887, .line 2010.

References

[l] Stalling^ W, "Cryptography and Networks Security-
FFmciples and Practices", 3 r Edition, F^arson Education,
2004.

[2] Animesh^L. Trivedi, Raj an Arora, Rishi^Lapoor, Sudip
Sanyal, Ajith Abraham, Sugata Sanyal, "Mobile Ad Hoc
NetworltSecurity Vulnerabilities", IGI Global, 2009.

[3] Maltoni D. Maio, Jiin A^L. and Frabhakar S, "Handbooks
of Fingerprint Recognition", Springer Verlag^ 2003 .

[4] Fessi B A, Ben Abdallah, S, Hamdi Mand Boudrig§, "A
new genetic algorithm approach for intrusion response
system in computer networks", IEEE Symposium on
Computers and Communications, pp. 342-347, 2009.

BShanthini is a research scholar in
Anna University, Chennai, India.
She received her Bachelor's degree
in C.S.E. from MsL.University,
Madurai and Master's degree in
C.S.E. from M.S. University,
Tirunelveli. Her research interests
include Networks Security, Web
Security, Wireless Communication,
Biometrics and Cloud Computing^

Dr. S. Swamynathan is an

Assistant Professor of Computer
Science and Engineering at Anna
University Chennai, India. He
received his Master's in Computer
Science and Engineering^ and
Doctorate in Reactive Web Services
from Anna University, Chennai. His
research interests include Web
Service, Security, Web Mining_,and
Automated Workflow Systems.

153

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Effective Multi-Stage Clustering for Inter- and
Intra-Cluster Homogeneity

Sunita M. Karad
Assistant Professor of Computer Engineering,
MIT, Pune, INDIA

V.M.Wadhai
Professor and Dean of Research, MITSOT,
MAE, Pune, INDIA

M.U.Kharat
Principle of Pankaj Laddhad IT,
Yelgaon, Buldhana, INDIA

Prasad S.Halgaonkar

Faculty of Computer Engineering

MITCOE, Pune, INDIA

Dipti D. Patil
Assistant Professor of Computer Engineering,
MITCOE, Pune, INDIA

Abstract - A new algorithm for clustering high-dimensional
categorical data is proposed and implemented by us. This
algorithm is based on a two-phase iterative procedure and
is parameter-free and fully-automatic. Cluster assignments
are given in the first phase, and a new cluster is added to
the partition by identifying and splitting a low-quality
cluster. Optimization of clusters is carried out in the
second phase. This algorithm is based on quality of cluster
in terms of homogeneity. Suitable notion of cluster
homogeneity can be defined in the context of high-
dimensional categorical data, from which an effective
instance of the proposed clustering scheme immediately
follows. Experiment is carried out on real data; this
innovative approach leads to better inter- and intra-
homogeneity of the clusters obtained.

Index Terms - Clustering, high-dimensional categorical
data, information search and retrieval.

I. INTRODUCTION

Clustering is a descriptive task that seeks to identify
homogeneous groups of objects based on the values of their
attributes (dimensions) [1] [2]. Clustering techniques have
been studied extensively in statistics, pattern recognition, and
machine learning. Recent work in the database community
includes CLARANS, BIRCH, and DBSCAN. Clustering is an
unsupervised classification technique. A set of unlabeled
objects are grouped into meaningful clusters, such that the
groups formed are homogeneous and neatly separated.
Challenges for clustering categorical data are: 1) Lack of
ordering of the domains of the individual attributes.
2) Scalability to high dimensional data in terms of

effectiveness and efficiency. High-dimensional categorical
data such as market-basket has records containing large
number of attributes. 3) Dependency on parameters. Setting of
many input parameters is required for many of the clustering
techniques which lead to many critical aspects.

Parameters are useful in many ways. Parameters
support requirements such as efficiency, scalability, and
flexibility. For proper tuning of parameters a lot of effort is
required. As number of parameters increases, the problem of
parameter tuning also increases. Algorithm should have as less
parameters as possible. If the algorithm is automatic it helps to
find accurate clusters. An automatic approach technique
searches huge amounts of high-dimensional data such that it is
effective and rapid which is not possible for human expert. A
parameter free approach is based on decision tree learning,
which is implemented by top-down divide-and-conquer
strategies. The above mentioned problems have been tackled
separately, and specific approaches are proposed in the
literature, which does not fit the whole framework. The main
objective of this paper is to face the three issues in a unified
framework. We look forward to an algorithmic technique that
is capable of automatically detecting the underlying interesting
structure (when available) on high-dimensional categorical
data.

We present Two Phase Clustering (MPC), a new
approach to clustering high-dimensional categorical data that
scales to processing large volumes of such data in terms of
both effectiveness and efficiency. Given an initial data set, it
searches for a partition, which improves the overall purity.
The algorithm is not dependent on any data-specific parameter
(such as the number of clusters or occurrence thresholds for
frequent attribute values). It is intentionally left parametric to

154

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

the notion of purity, which allows for adopting the quality
criterion that best meets the goal of clustering. Section-2
reviews some of the related work carried out on transactional
data, high dimensional data and high dimensional categorical
data. Section-3 provides background information on the
clustering of high dimensional categorical data (MPC
algorithm). Section-4 describes implementation results of
MPC algorithm. Section-5 concludes the paper and draws
direction to future work.

II. RELATED WORK

In current literature, many approaches are given for clustering
categorical data. Most of these techniques suffer from two
main limitations, 1) their dependency on a set of parameters
whose proper tuning is required and 2) their lack of scalability
to high dimensional data. Most of the approaches are unable to
deal with the above features and in giving a good strategy for
tuning the parameters.

Many distance-based clustering algorithms [3] are
proposed for transactional data. But traditional clustering
techniques have the curse of dimensionality and the sparseness
issue when dealing with very high-dimensional data such as
market-basket data or Web sessions. For example, the K-
Means algorithm has been adopted by replacing the cluster
mean with the more robust notion of cluster medoid (that is,
the object within the cluster with the minimal distance from
the other points) or the attribute mode [4]. However, the
proposed extensions are inadequate for large values of m:
Gozzi et al. [5] describe such inadequacies in detail and
propose further extensions to the K-Means scheme, which fit
transactional data. Unfortunately, this approach reveals to be
parameter laden. When the number of dimensions is high,
distance-based algorithms do not perform well. Indeed, several
irrelevant attributes might distort the dissimilarity between
tuples. Although standard dimension reduction techniques [6]
can be used for detecting the relevant dimensions, these can be
different for different clusters, thus invalidating such a
preprocessing task. Several clustering techniques have been
proposed, which identify clusters in subspaces of maximum
dimensionality (see [7] for a survey). Though most of these
approaches were defined for numerical data, some recent work
[8] considers subspace clustering for categorical data.

A different point of view about (dis)similarity is
provided by the ROCK algorithm [9]. The core of the
approach is an agglomerative hierarchical clustering procedure
based on the concepts of neighbors and links. For a given
tuple x, a tuple y is a neighbor of x if the Jaccard similarity
J(x, y) between them exceeds a prespecified threshold O. The
algorithm starts by assigning each tuple to a singleton cluster
and merges clusters on the basis of the number of neighbors
(links) that they share until the desired number of clusters is
reached. ROCK is robust to high-dimensional data. However,
the dependency of the algorithm to the parameter O makes
proper tuning difficult.

Categorical data clusters are considered as dense
regions within the data set. The density is related to the
frequency of particular groups of attribute values. The higher

the frequency of such groups the stronger the clustering.
Preprocessing the data set is carried by extracting relevant
features (frequent patterns) and discovering clusters on the
basis of these features. There are several approaches
accounting for frequencies. As an example, Yang et al. [10]
propose an approach based on histograms: The goodness of a
cluster is higher if the average frequency of an item is high, as
compared to the number of items appearing within a
transaction. The algorithm is particularly suitable for large
high-dimensional databases, but it is sensitive to a user
defined parameter (the repulsion factor), which weights the
importance of the compactness/sparseness of a cluster. Other
approaches [11], [12], [13] extend the computation of
frequencies to frequent patterns in the underlying data set. In
particular, each transaction is seen as a relation over some sets
of items, and a hyper-graph model is used for representing
these relations. Hyper-graph partitioning algorithms can hence
be used for obtaining item/transaction clusters.

The CLICKS algorithm proposed in [14] encodes a
data set into a weighted graph structure G(N, E), where the
individual attribute values correspond to weighted vertices in
N, and two nodes are connected by an edge if there is a tuple
where the corresponding attribute values co-occur. The
algorithm starts from the observation that clusters correspond
to dense (that is, with frequency higher than a user-specified
threshold) maximal k-partite cliques and proceeds by
enumerating all maximal k-partite cliques and checking their
frequency. A crucial step is the computation of strongly
connected components, that is, pairs of attribute values whose
co-occurrence is above the specified threshold. For large
values of m (or, more generally, when the number of
dimensions or the cardinality of each dimension is high), this
is an expensive task, which invalidates the efficiency of the
approaches. In addition, technique depends upon a set of
parameters, whose tuning can be problematic in practical
cases.

Categorical clustering can be tackled by using
information-theoretic principles and the notion of entropy to
measure closeness between objects. The basic intuition is that
groups of similar objects have lower entropy than those of
dissimilar ones. The COOLCAT algorithm [15] proposes a
scheme where data objects are processed incrementally, and a
suitable cluster is chosen for each tuple such that at each step,
the entropy of the resulting clustering is minimized. The
scaLable InforMation Bottleneck (LIMBO) algorithm [16]
also exploits a notion of entropy to catch the similarity
between objects and defines a clustering procedure that
minimizes the information loss. The algorithm builds a
Distributional Cluster Features (DCF) tree to summarize the
data in k clusters, where each node contains statistics on a
subset of tuples. Then, given a set of k clusters and their
corresponding DCFs, a scan over the data set is performed to
assign each tuple to the cluster exhibiting the closest DCF.
The generation of the DCF tree is parametric to a user-defined
branching factor and an upper bound on the distance between
a leaf and a tuple.

Li and Ma [17] propose an iterative procedure that is
aimed at finding the optimal data partition that minimizes an

155

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

entropy-based criterion. Initially, all tuples reside within a
single cluster. Then, a Monte Carlo process is exploited to
randomly pick a tuple and assign it to another cluster as a trial
step aimed at decreasing the entropy criterion. Updates are
retained whenever entropy diminishes. The overall process is
iterated until there are no more changes in cluster assignments.
Interestingly, the entropy-based criterion proposed here can be
derived in the formal framework of probabilistic clustering
models. Indeed, appropriate probabilistic models, namely,
multinomial [18] and multivariate Bernoulli [19], have been
proposed and shown to be effective. The classical
Expectation-Maximization framework [20], equipped with any
of these models, reveals to be particularly suitable for dealing
with transactional data [21], [22], being scalable both in n and
in m. The correct estimation of an appropriate number of
mixtures, as well as a proper initialization of all the model
parameters, is problematic here.

The problem of estimating the proper number of
clusters in the data has been widely studied in the literature.
Many existing methods focus on the computation of costly
statistics based on the within-cluster dispersion [23] or on
cross-validation procedures for selecting the best model [24],
[25]. The latter requires an extra computational cost due to a
repeated estimation and evaluation of a predefined number of
models. More efficient schemes have been devised in [26],
[27]. Starting from an initial partition containing a single
cluster, the approaches iteratively apply the K-Means
algorithm (with k = 2) to each cluster so far discovered. The
decision on whether to switch the original cluster with the
newly generated sub-clusters is based on a quality criterion,
for example, the Bayesian Information Criterion [26], which
mediates between the likelihood of the data and the model
complexity, or the improvement in the rate of distortion (the
variance in the data) of the sub-clusters with respect to the
original cluster [27]. The exploitation of the K-Means scheme
makes the algorithm specific to low-dimensional numerical
data, and proper tuning to high-dimensional categorical data is
problematic.

Automatic approaches that adopt the top-down
induction of decision trees are proposed in [28], [29], [30].
The approaches differ in the quality criterion adopted, for
example reduction in entropy [28], [29] or distance among the
prototypes of the resulting clusters [29]. All of these
approaches have some of the drawbacks. The scalability on
high-dimensional data is poor. Some of the literature that
focused on high dimensional categorical data is available in
[31], [32].

III. The MPC Algorithm

The key idea of Two Phase Clustering (MPC) algorithm is to
develop a clustering procedure, which has the general sketch
of a top-down decision tree learning algorithm. First, start
from an initial partition which contains single cluster (the
whole data set) and then continuously try to split a cluster
within the partition into two sub-clusters. If the sub-clusters
have a higher homogeneity in the partition than the original
cluster, the original is removed. The sub-clusters obtained by

splitting are added to the partition. Split the clusters on the
basis of their homogeneity. A function Quality(C) measures
the degree of homogeneity of a cluster C. Clusters with high
intra-homogeneity exhibit high values of Quality.

Let M be set of Boolean attributes such that M =

{a l9 , a m } and a data set D = {x l9 x 2 , , x n } of tuples which

is defined on M. a e M is denoted as an item, and a tuple x e D
as a transaction x. Data sets containing transactions are
denoted as transactional data, which is a special case of high-
dimensional categorical data. A cluster is a set S which is a
subset of D. The size of S is denoted by n s , and the size of M s
= {a|a C x, x C S} is denoted by m s . A partitioning problem is
to divide the original collection of data D into a set P =

{Ci, ,C k } where each clusters Cj are nonempty. Each

cluster contains a group of homogeneous transactions.
Clusters where transactions have several items have higher
homogeneity than other subsets where transactions have few
items. A cluster of transactional data is a set of tuples where
few items occur with higher frequency than somewhere else.

Our approach to clustering starts from the analysis of
the analogies between a clustering problem and a
classification problem. In both cases, a model is evaluated on
a given data set, and the evaluation is positive when the
application of the model locates fragments of the data
exhibiting high homogeneity. A simple rather intuitive and
parameter-free approach to classification is based on decision
tree learning, which is often implemented through top-down
divide and conquers strategies. Here, starting from an initial
root node (representing the whole data set), iteratively, each
data set within a node is split into two or more subsets, which
define new sub-nodes of the original node. The criterion upon
which a data set is split (and, consequently, a node is
expanded) is based on a quality criterion: choosing the best
"discriminating" attribute (that is, the attribute producing
partitions with the highest homogeneity) and partitioning the
data set on the basis of such attribute. The concept of
homogeneity has found several different explanations (for
example, in terms of entropy or variance) and, in general, is
related to the different frequencies of the possible labels of a
target class.

The general schema of the MPC algorithm is
specified in Fig. 1 . The algorithm starts with a partition having
a single cluster i.e whole data set (line 1). The central part of
the algorithm is the body of the loop between lines 2 and 15.
Within the loop, an effort is made to generate a new cluster by
1) choosing a candidate node to split (line 4), 2) splitting the
candidate cluster into two sub-clusters (line 5), and (line 3)
calculating whether the splitting allows a new partition with
better quality than the original partition (lines 6-13). If this is
true, the loop can be stopped (line 10), and the partition is
updated by replacing the original cluster with the new sub-
clusters (line 8). Otherwise, the sub-clusters are discarded, and
a new cluster is taken for splitting.

The generation of a new cluster calls STABILIZE-
CLUSTERS in line 9, improves the overall quality by trying
relocations among the clusters. Clusters at line 4 are taken in
increasing order of quality,
a. Splitting a Cluster

156

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

A splitting procedure gives a major improvement in
the quality of the partition. Choose the attribute that gives the
highest improvement in the quality of the partition.

GENERATE-CLUSTERS (D)

Input: A set D ={x 1/ ...,x N } of transactions;

Output: A partition P = {C l7 ...,C k } of clusters;

Let initially P = {D};

repeat

Generate a new cluster C initially empty;

for each cluster C, E P do

PARTITION-CLUSTERS(QC);

P' <— PU {C};

if Quality(P) < Quality(P') then

P <— P';

STABIUZE-CLUSTERS(P);

10.

break

11.

else

12.

Restore all x y gCintoC,;

13.

end if

14.

end for

15.

until no further cluster C can be generated

Figure 1 : Generate Clusters

PARTITION-CLUSTER(aO)

PI.

repeat

P2.

for all x ECiU C 2 do

P3.

if cluster(x) = CI then

P4.

C u ^ Ci; C v ^ C 2 ;

P5.

else

P6.

C u ^ C 2 ; C v ^ Ci;

P7.

end if

P8.

Qi <— Quality(C u ) + Quality(C v );

P9.

Q s <— Quality(C u -{x}) + Quality(C v U{x});

P10.

if Qs > Qj then

Pll.

C u .Remove(x);

P12.

C v .lnsert(x);

P13.

end if

P14.

end for

P15.

until C x and C 2 are stable

Figure 2: Partition Cluster

PARTITION-CL USTER

The PARTITION-CLUSTER algorithm is given in Fig.2. The
algorithm continuously evaluates, for each element x eQU
C 2 , to check whether a reassignment increases the
homogeneity of the two clusters.

Lines P8 and P9 compute the involvement of a: to the
local quality in two cases: either x remains in its original

cluster (C u ) or x is moved to the other cluster (C v ). If moving x
gives an improvement in the local quality, then the swapping
is done (lines P10-P13). Lines P2-P14 in the algorithm is
nested into a main loop: elements are continuously checked
for swapping until a convergence is met. The splitting process
can be sensitive to the order upon which elements are
considered: In the first stage, it could be not convenient to
reassign the generic Xi from Ci to C 2 , whereas a convenience
in performing the swap can be found after the relocation of
some other element Xj. The main loop partly smoothes this
effect by repeatedly relocating objects until convergence is
met. Better PARTITION-CLUSTER can be made strongly
insensitive to the order with which cluster elements are
considered. The basic idea is discussed next. The elements that
mostly influence the locality effect are either outlier
transactions (that is, those containing mainly items, whose
frequency within the cluster is rather low) or common
transactions (which, dually, contain very frequent items). In
the first case, C 2 is unable to attract further transactions,
whereas in the second case, C 2 is likely to attract most of the
transactions (and, consequently, Ci will contain outliers).

The key idea is to rank and sort the cluster elements
before line PI, which is on the basis of their splitting
effectiveness. To this purpose, each transaction x belonging to
cluster C can be associated with a weight w(x), which
indicates its splitting effectiveness, x is eligible for splitting C
if its items allow us to divide C into two homogeneous sub-
clusters. In this respect, the Gini index is a natural way to
quantify the splitting effectiveness G(a) of the individual
attribute value a 6 x. Precisely, G(a) = 1 - Pr(a|C) 2 -
(1 - Pr(a|C)) 2 , where Pr(a|C) denotes the probability of a
within C. G(a) is close to its maximum whenever a is present
in about half of the transactions of C and reaches its minimum
whenever a is unfrequent or common within C. The overall
splitting effectiveness of x can be defined by averaging the
splitting effectiveness of its constituting items
w(x) = avg a g x (G(a)). Once ranked, the elements x e C can be
considered in descending order of their splitting effectiveness
at line P2. This guarantees that C 2 is initialized with elements,
which do not represent outliers and still are likely to be
removed from C x . This removes the dependency on the initial
input order of the data. With decision tree learning, MPC
exhibits a preference bias, which is encoded within the notion
of homogeneity and can be viewed as the preference for
compact clustering trees. Indeed, due to the splitting
effectiveness heuristic, homogeneity is enforced by the effects
of the Gini index. At each split, this tends to isolate clusters of
transactions with mostly frequent attribute values, from which
the compactness of the overall clustering tree follows.

b. STABILIZE-CLUSTERS

PARTITION-CLUSTER improves the local quality
of a cluster. And STABILIZE-CLUSTERS try to increase
partition quality. It is carried out by finding the most suitable
clusters for each element among the ones which are there in
the partition.

Fig. 3 shows the pseudo code of the procedure. The
central part of the algorithm is a main loop which (lines S2-

157

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

SI 7) examines all the available elements. For each element x,
a pivot cluster is identified, which is the cluster containing x.
Then, the available clusters are continuously evaluated. The
insertion of x in the current cluster is done (lines S5-S6), and
the updated quality is compared with the original quality.

STABILIZE

-CLUSTERS^

SI.

repeat

S2.

for all x E0do

S3.

Cpi V ot **— cluster(x); Q *— Quality (P);

S4.

for all CEP do

S5.

C p/Vot .REMOVE(x);

S6.

CINSERT(x);

S7.

if Quality(P) > Q then

S8.

if C pivot = then

S9.

P.REMOVE(C p/Vot );

S10.

end if

Sll.

Cpivot +— Q Q +— Quality (P);

S12.

else

S13.

C pivot .INSERT(x);

S14.

C.REMOVE(x);

S15.

end if

S16.

end for

S17.

end for

S18.

until P is stable

Figure 3 : Stabilize Clusters

If an improvement is obtained, then the swap is accepted (line
Sll). The new pivot cluster is the one now containing x, and if
the removal of x makes the old pivot cluster empty, then the
old pivot cluster is removed from the partition P. If there is no
improvement in quality, x is restored into its pivot cluster, and
a new cluster is examined. The main loop is iterated until a
stability condition for clusters is achieved.

c. Cluster and Partition Qualities

AT-DC gives two different quality measures, 1) local
homogeneity within a cluster and 2) global homogeneity of the
partition. As shown in Fig. 1, it is noticed that partition quality
is used for checking whether the insertion of a new cluster is
really suitable: it is for maintaining compactness. Cluster
quality in procedure PARTITIONCLUSTER is done for good
separation.

Cluster quality is known when there is a high degree
of intracluster homogeneity and intercluster homogeneity. As
given in [35], there is strong relation between intracluster
homogeneity and the probability Pr(ai|C k ) that item ai appears
in a transaction containing in C k There is a strong relationship
between intercluster separation and Pr(x E C k , ai Ex). Cluster

homogeneity and separation is computed by relating it to the
unity of items within the transactions that it contains. Cluster
quality is equal to the combination of the above probability,

2 ff tt-VD Pi (a|£)Fr. (S|fl)Fr. (fli The last term is used

for weighting the importance of item a in the summation:
Essentially, high values from low-frequency items are less

relevant than those from high-frequency values. By the Bayes
theorem, the above formula is expressed as

MC)I^, ym Jr.(m- t 33 ]- Terms Fr(>|£?

(relative strength of a within C) and Pr(C) (relative strength of
C) work in contraposition. It is easy to compute the gain in
strength for each item with respect to the whole data set, that
is

Quality (CO = Pr(Q) Eh-s^Pr ft Cft* - Rr.tffl.fl^ ]

(1)

Where,

C k - cluster

Pr(C k ) - relative strength of C k

a C MC k - an item

M = {a 1} , a m } is set of Boolean attributes

Pr(a| Ck) - relative strength of a within C k

Pr(a|D) - relative strength of a within D

D = {x b , x n } is data set of tuples defined on M

Qualit y( C0^I y ^ t (^.(^]

(2)

where na and Na represent the frequencies of a in C and D,
respectively. The value of Quality (C k ) is updated as soon as a
new transaction is added to C.

IV. RESULTS AND ANALYSIS

Two real-life data sets were evaluated. A description of each
data set employed for testing is provided next, together with
an evaluation of the MPC performances.

UCI DATASETS [34]

Zoo: Zoo dataset contains 103 instances, each having 18
attributes (animal name, 15 Boolean attributes and 2
numerics). The "type" attribute appears to be the class
attribute. In total there are 7 classes of animals, that is, class 1
has 41 set of animals, class 2 has 20 set of animals, class 3 has
5 set of animals, class 4 has 13 set of animals, class 5 has 4 set
of animals, class 6 has 8 set of animals and class 7 has 10 set
of animals. Here is a breakdown of which animals are in
which type: (it is unusual that there are 2 instances of "frog"
and one of "girl"!). There are no missing values in this dataset.
Table 1 shows that in cluster 1, a class 2 is having high
homogeneity and in cluster 2, classes 3, 5 and 7 are having
high homogeneity.

Hepatitis: Hepatitis contains 155 instances, each having 20
attributes. It represents the observation of patients. Each
instance is one patient's record according to 20 attributes (for
example, age, steroid, antivirals, and spleen palpable). Some
attributes contains missing values. A class as "DIE" or

158

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

"LIVE" is given to each instance. Out of 155 instances, 32 are
"DIE" and 123 are "LIVE". Table 2 shows that in cluster 1
and cluster 2 are having high homogeneity. In cluster 2 and 4
there are 2 (DIE) and 1 (LIVE) instances which are
misclassified.

Table 1: Confusion matrix for zoo

Cluster No.

Classes

Table 2: Confusion matrix for Hepatitis

Cluster No.

Classes

DIE

LIVE

V. CONCLUDING REMARK

This innovative MPC algorithm is fully-automatic, parameter-
free approach to cluster high-dimensional categorical data.
The main advantage of our approach is its capability of
avoiding explicit prejudices, expectations, and presumptions
on the problem at hand, thus allowing the data itself to speak.
This is useful with the problem at hand, where the data is
described by several relevant attributes.

A limitation of our proposed approach is that the
underlying notion of cluster quality is not meant for catching
conceptual similarities, that is, when distinct values of an
attribute are used for denoting the same concept. Probabilities
are provided to evaluate cluster homogeneity only in terms of
the frequency of items across the underlying transactions.
Hence, the resulting notion of quality suffers from the typical
limitations of the approaches, which use exact-match
similarity measures to assess cluster homogeneity. To this
purpose, conceptual cluster homogeneity for categorical data
can be easily added to the framework of the MPC algorithm.

Another limitation of our approach is that it cannot
deal with outliers. These are transactions whose structure
strongly differs from that of the other transactions being
characterized by low-frequency items. A cluster containing
such transaction exhibits low quality. Worst, outliers could
negatively affect the PARTITION-CLUSTER procedure by
preventing the split to be accepted (because of an arbitrary
assignment of such outliers, which would lower the quality of
the partitions). Hence, a significant improvement of MPC can
be obtained by defining an outlier detection procedure that is

capable of detecting and removing outlier transactions before
partitioning the clusters. The research work can be extended
further to improve the quality of clusters by removing
outliers.

REFERENCES

[I] J. Grabmeier and A. Rudolph, "Techniques of Cluster Algorithms in
Data Mining," Data Mining and Knowledge Discovery, vol. 6, no. 4,
pp. 303-360, 2002.

[2] A. Jain and R. Dubes, Algorithms for Clustering Data. Prentice Hall,

1988.
[3] R. Ng and J. Han, "CLARANS: A Method for Clustering Objects for

Spatial Data Mining," IEEE Trans. Knowledge and Data Eng., vol. 14,

no. 5, pp. 1003-1016, Sept./Oct. 2002.
[4] Z. Huang, "Extensions to the K-Means Algorithm for Clustering Large

Data Sets with Categorical Values," Data Mining an Knowledge

Discovery, vol. 2, no. 3, pp. 283-304, 1998.
[5] C. Gozzi, F. Giannotti, and G. Manco, "Clustering Transactional Data,"

Proc. Sixth European Conf. Principles and Practice of Knowledge

Discovery in Databases (PKDD '02), pp. 175-187, 2002.
[6] S. Deerwester et al., "Indexing by Latent Semantic Analysis," J. Am.

Soc. Information Science, vol. 41, no. 6, 1990.
[7] L. Parsons, E. Haque, and H. Liu, "Subspace Clustering for High-
Dimensional Data: A Review," SIGKDD Explorations, vol. 6, no. 1, pp.

90-105,2004.
[8] G. Gan and J. Wu, "Subspace Clustering for High Dimensional

Categorical Data," SIGKDD Explorations, vol. 6, no. 2, pp. 87-94, 2004.
[9] M. Zaki and M. Peters, "CLICK: Mining Subspace Clusters in

categorical Data via k-Partite Maximal Cliques," Proc. 21st IntT Conf.

Data Eng. (ICDE '05), 2005.
[10] Y. Yang, X. Guan, and J. You, "CLOPE: A Fast and Effective

Clustering Algorithm for Transactional Data," Proc. Eighth ACM Conf.

Knowledge Discovery and Data Mining (KDD '02), pp. 682-687, 2002.

[II] E. Han, G. Karypis, V. Kumar, and B. Mobasher, "Clustering in a High
Dimensional Space Using Hypergraph Models," Proc. ACM SIGMOD
Workshops Research Issues on Data Mining and Knowledge Discovery
(DMKD '97), 1997.

[12] M. Ozdal and C. Aykanat, "Hypergraph Models and Algorithms for

Data-Pattern-Based Clustering," Data Mining and Knowledge

Discovery, vol. 9, pp. 29-57, 2004.
[13] K. Wang, C. Xu, and B. Liu, "Clustering Transactions Using Large

Items," Proc. Eighth IntT Conf. Information and Knowledge

Management (CIKM '99), pp. 483-490, 1999.
[14] D. Barbara, J. Couto, and Y. Li, "COOLCAT: An Entropy-Based

Algorithm for Categorical Clustering," Proc. 11th ACM Conf.

Information and Knowledge Management (CIKM '02), pp. 582-589,

2002.
[15] P. Andritsos, P. Tsaparas, R. Miller, and K. Sevcik, "LIMBO: Scalable

Clustering of Categorical Data," Proc. Ninth IntT Conf. Extending

Database Technology (EDBT '04), pp. 123-146, 2004.
[16] M.O.T. Li and S. Ma, "Entropy-Based Criterion in Categorical

Clustering," Proc. 21st IntT Conf. Machine Learning (ICML '04), pp.

68-75, 2004.
[17] I. Cadez, P. Smyth, and H. Mannila, "Probabilistic Modeling of

Transaction Data with Applications to Profiling, Visualization, and

Prediction," Proc. Seventh ACM SIGKDD IntT Conf. Knowledge

Discovery and Data Mining (KDD '01), pp. 37-46, 2001.
[18] M. Carreira-Perpinan and S. Renals, "Practical Identifiability of Finite

Mixture of Multivariate Distributions," Neural Computation, vol. 12, no.

l,pp. 141-152,2000.
[19] G. McLachlan and D. Peel, Finite Mixture Models. John Wiley & Sons,

2000.
[20] M. Meila and D. Heckerman, "An Experimental Comparison of Model-
Based Clustering Methods," Machine Learning, vol. 42, no. 1/2, pp. 9-

29,2001.
[21] J.G.S. Zhong, "Generative Model-Based Document Clustering: A

Comparative Study," Knowledge and Information Systems, vol. 8, no. 3,

pp. 374-384, 2005.
[22] A. Gordon, Classification. Chapman and Hall/CRC Press, 1999.

159

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

[23] C. Fraley and A. Raftery, "How Many Clusters? Which Clustering

Method? The Answer via Model-Based Cluster Analysis," The

Computer J., vol. 41, no. 8, 1998.
[24] P. Smyth, "Model Selection for Probabilistic Clustering Using Cross-
Validated Likelihood," Statistics and Computing, vol. 10, no. 1, pp. 63-

72,2000.
[25] D. Pelleg and A. Moore, "X-Means: Extending K-Means with Efficient

Estimation of the Number of Clusters," Proc. 17th Int'l Conf. Machine

Learning (ICML '00), pp. 727-734, 2000.
[26] M. Sultan et al., "Binary Tree- Structured Vector Quantization Approach

to Clustering and Visualizing Microarray Data," Bioinformatics, vol. 18,

2002.
[27] S. Guha, R. Rastogi, and K. Shim, "ROCK: A Robust Clustering

Algorithm for Categorical Attributes," Information Systems, vol. 25, no.

5, pp. 345-366,2001.
[28] J. Basak and R. Krishnapuram, "Interpretable Hierarchical Clustering by

Constructing an Unsupervised Decision Tree," IEEE Trans. Knowledge

and Data Eng., vol. 17, no. 1, Jan. 2005.
[29] H. Blockeel, L.D. Raedt, and J. Ramon, "Top-Down Induction of

Clustering Trees," Proc. 15th Int'l Conf. Machine Learning (ICML'98),

pp. 55-63, 1998.
[30] B. Liu, Y. Xia, and P. Yu, "Clustering through Decision Tree

Construction," Proc. Ninth Int'l Conf. Information and Knowledge

Management (CIKM '00), pp. 20-29, 2000.
[31] Yi-Dong Shen, Zhi-Yong Shen and Shi-Ming Zhang,"Cluster Cores -

based Clustering for High - Dimensional Data".
[32] Alexander Hinneburg and Daniel A. Keim, Markus Wawryniuk,"HD-

Eye-Visual of High-Dimensional Data: A Demonstration".
[33] http://en.wikipedia.org/wiki/Bayes'_theorem

[34] UCI Machine Learning Repository http://www.ics.uci.edu/~mlearn/
[35] D. Fisher, "Knowledge Acquisition via Incremental Conceptual

Clustering," Machine Learning, vol. 2, pp. 139-172, 1987.

AUTHORS PROFILE

Sunita M. Karad has received B.E. degree in
Computer Engineering from Marathvada University,
India in 1992, M.E. degree from Pune University in
2007. She is a registered Ph.D. student of Amravati
University. She is currently working as Assistant
Professor in Computer Engineering department in
MIT, Pune. She has more than 10 years of teaching
experience and successfully handles administrative
work in MIT, Pune. Her research interest includes
Data mining, Business Intelligence & Aeronautical space research.

1 1 "

Dr. Madan U. Kharat has received his B.E. from Amravati University, India
in 1992, M.S. from Devi Ahilya University (Indore), India in 1995 and Ph.D.
degree from Amravati University, India in 2006. He has
experience of 18 years in academics. He has been
working as a Principle of PLIT, Yelgaon, Budhana. His
research interest includes Deductive Databases, Data
Mining and Computer Networks.

Prasad

bachelor's

degree

Amravati
Computer

Science

Engineering

currently a

lecturer

research

interest

and Data

Mining,

databases

and

Halgaonkar received his
in Computer Science from
University in 2006 and M.Tech in
from Walchand College of
Shivaji University in 2010. He is
in MITCOE, Pune. His current
includes Knowledge discovery
deductive databases, Web
Semi- Structured data.

Dipti D. Patil has received B.E. degree in Computer
Engineering from Mumbai University in 2002 and M.E.
degree in Computer Engineering from Mumbai
University, India in 2008. She has worked as Head &
Assistant Professor in Computer Engineering
Department in Vidyavardhini's College of Engineering
& Technology, Vasai. She is currently working as
Assistant Professor in MITCOE, Pune. Her Research
interests include Data mining, Business Intelligence and

Body Area Network.

I Dr. Vijay M.Wadhai received his B.E. from

! Nagpur University in 1986, M.E. from Gulbarga

University in 1995 and Ph.D. degree from Amravati

University in 2007. He has experience of 25 years

| which includes both academic (17 years) and

I research (8 years). He has been working as a Dean

I of Research, MITSOT, MAE, Pune (from 2009) and

simultaneously handling the post of Director -

Research and Development, Intelligent Radio

Frequency (IRF) Group, Pune (from 2009). He is

currently guiding 12 students for their PhD work in both Computers and

Electronics & Telecommunication area. His research interest includes Data

Mining, Natural Language processing, Cognitive Radio and Wireless

Network, Spectrum Management, Wireless Sensor Network, VANET, Body

Area Network, ASIC Design - VLSI. He is a member of ISTE, IETE, IEEE,

IES and GISFI (Member Convergence Group), India.

J L

160

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

A Pilot Based RLS Channel Estimation for LTE
SC-FDMA in High Doppler Spread

M. M. Rana

Department of Electronics and Communication Engineering

Khulna University of Engineering and Technology

Khunla, Bangladesh

Abstract — Main challenges for a terminal implementation are
efficient realization of the inner receiver, especially for channel
estimation (CE) and equalization. In this paper, pilot based
recursive least square (RLS) channel estimator technique is
investigate for a long term evolution (LTE) single carrier-
frequency division multiple access (SC-FDMA) system in high
Doppler spread environment. This CE scheme uses adaptive RLS
estimator which is able to update parameters of the estimator
continuously, so that knowledge of channel and noise statistics
are not required. Simulation results show that the RLS CE
scheme with 500 Hz Doppler frequency has 3 dB better
performances compared with 1.5 kHz Doppler frequency.

Keywords— Channel estimation, LTE, RLS, SC-FDMA.

I. INTRODUCTION

The 3rd generation partnership project (3 GPP) members
started a feasibility study on the enhancement of the universal
terrestrial radio access (UTRA), to improve the mobile phone
standard to cope with future requirements. This project was
called long term evolution (LTE) [1], [2]. LTE uses
orthogonal frequency division multiple access (OFDM A) for
downlink and single carrier-frequency division multiple
access (SC-FDMA) for uplink transmission [1]. A highly
efficient way to cope with the frequency selectivity of
wideband channel is OFDMA. OFDMA is an effective
technique for combating multipath fading and for high bit rate
transmission over mobile wireless channels. Channel
estimation (CE) has been successfully used to improve the
system performance. It can be employed for the purpose of
detecting received signal, improve signal-to-noise ratio
(SNR), channel equalization, cochannel interference (CCI)
rejection, and improved the system performance [3-5].

In general, CE techniques can be divided into three
categories such as pilot CE, blind CE, and semi-blind CE [10],
[11]. Pilot CE techniques offer low computational complexity
and good performance [12]. The blind CE techniques exploit
the statistical behavior of the received signals and require a
large amount of data [13]. Semi-blind CE methods are used a
combination of data aided and blind methods [11]. The pilot
CE algorithm requires probe sequences; the receiver can use
this probe sequence to reconstruct the transmitted waveform

[6-8]. Pilot symbols can be placed either at the beginning of
each burst as a preamble or regularly through the burst. Pilot
sequences are transmitted at certain positions of the SC-
FDMA frequency time pattern, in its place of data.

Adaptive CE has been, and still is, an area of active research
topics, playing imperative roles in an ever growing number of
applications such as wireless communications where the
channel is rapidly time-varying. Signal processing techniques
that use recursively estimated, time varying models are
normally called adaptive. Different adaptive CE algorithms
have been proposed over the years for the purpose of updating
the channel coefficient. The least mean square (LMS) method,
its normalized version (NLMS), the affine projection
algorithm (APA), as well as the recursive least square (RLS)
method are well known examples of such CE algorithms. The
well known LMS/NLMS CE algorithms are attractive from a
computational complexity point of view but their convergence
behavior for highly correlated input signals is poor. The RLS
CE method resolves this trouble, but at the expense of
increased complexity. A very large number of fast RLS CE
methods have been developed over the years, but regrettably,
it seems that the better a fast RLS CE method is in terms of
computational efficiency and numerical stability. In addition,
the RLS algorithm has the recursive inversion of an estimate
of the autocorrelation matrix of the input signal as its
cornerstone, problems arise, if the autocorrelation matrix is
rank deficient.

In this paper, we investigate the adaptive RLS CE method
in the LTE SC-FDMA systems in high Doppler spread
environment. This CE method uses adaptive estimator which
is able to update parameters of the estimator continuously so
that knowledge of channel and noise statistics are not
required. Simulation results show that the RLS CE scheme
with 500 Hz Doppler frequency has 3 dB better performances
compared with 1500 Hz Doppler frequency.

We use the following notations throughout this paper: bold
face lower letter is used to represent vector. Superscripts x*
and x T denote the conjugate and conjugate transpose of the
complex vector x respectively.

The remainder of the paper is organized as follows: section
II describes wireless communication systems and LTE SC-
FDMA systems model is describes in section III. The RLS CE
scheme is presented in section IV, and its performance is

161

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

analyzed in section V. Finally, some concluding remarks are
given in section VI.

II. WIRELESS COMMUNICATION SYSTEMS

III. LTE SC-FDMA SYSTEMS DESCRIPTION

In this section, we briefly explain LTE SC-FDMA system
model, fading channel statistics, and received signal model.

Nowadays, cellular mobile phones have become an
important tool and part of daily life. In the last decade, cellular
systems have experienced fast development and there are
currently about two billion users over the world. The idea of
cellular mobile communications is to divide large zones into
small cells, and it can provide radio coverage over a wider
area than the area of one cell. This concept was developed by
researchers at AT & T Bell laboratories during the 1950s and
1960s. The initial cellular system was created by nippon
telephone & telegraph (NTT) in Japan, 1979. From then on,
the cellular mobile communication has evolved.

The mobile communication systems are frequently
classified as different generations depending of the service
offered. The first generation (1G) comprises the analog
communication techniques, and it was mainly built on
frequency modulation (FM) and frequency division multiple
accesses (FDMA). Digital communication techniques
appeared in the second generation (2G) systems, and main
access schemes are time division multiple access (TDMA) and
code division multiple access (CDMA). The two most
commonly accepted 2G systems are global system for mobile
(GSM) and interim standard-95 (IS-95). These systems mostly
offer speech communication, but also data communication
limited to rather low transmission rates. The concept of the
third generation (3G) system started operations on October,
2002 in Japan. The 3GPP members started a feasibility study
on the enhancement of the universal terrestrial radio access
(UTRA) in December 2004, to improve the mobile phone
standard to cope with future requirements. This project was
called LTE. LTE uses SC-FDMA for uplink transmission and
OFDMA for downlink transmission. Fig. 1 summarizes the
evolution path of cellular mobile communications systems.

Narrowband

Circuit
switched

Circuit

Wideband

cet switched

Broadband

Fully IP based core network

Analog
FDMA

Digital
TDMA,
CDMA

WCDMA,
CDMA

HSPA,
CDMA,
OFDMA

OFDMA,
SC-FDMA

OFDMA,
SC-FDMA,
SC-CDMA,
MC-CDMA

2G V 3G /3.50V 3.9G
MMT2000 Y 3GLTE

V 14kbpS A 64kbpS 1 2M PP S I 14Mbps A^lOOMl^ W

Voice voice, c

Data with video ) I Voice and data with vide

A. Baseband system model

A baseband block diagram for the communications system
under investigation is shown in Fig. 2.

Bit stream

Demodulation

Modulation

N-point IDFT

N-point DFT

Pilots signal
generation

— ►

CE and
equalization

Mapping

Mapping
restore

Multipath
fading
channel

AWGN
noise

M-point IDFT

M-point
DFT

1 !

i i

Add cyclic
prefix (CP)

-*&—+&>-+

Synchronization
and remove CP

Fig. 2. Block diagram of a LTE SC-FDMA system.

At the transmitter, a baseband multiple phase shift keying
modulator takes binary sequence and produces the signaling
waveforms

m^t) = x j — cos((Dt + aj, < t < T

[cos^) cos(cot) - sin^) sin(cot)]

= a i b(t) + c i d(t), (1)

where T is the symbol duration, E is the energy of m i (t),
cd = 27rf, f is the carrier frquency, phase anagle

a- , M is the alphabate size, a '. = yj E COS OC

M l

inphasse basis , b(t) =. \— COs((Dt), C i = V £'sina i , and

Fig. 1. Evolution path in mobile communication systems.

quadrature basis, d(t) = -A — COs((Ot). CE is often achieved

by multiplexing known symbols, so called, pilot symbols into
data sequences [1]. These modulated symbols and pilots
perform N-point discrete Fourier transform (DFT) to produce
a frequency domain representation:

162

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

-j2^mt

s 1 (t)=- ? =X m i( t ) e N '
VNtD

where j is the imaginary unit. It then maps each of the N-point
DFT outputs to one of the orthogonal sub-carriers mapping that
can be transmitted. There are two principal sub-carrier mapping
modes: localized mode, and distribution mode. In distributed
sub-carrier mode, the outputs are allocated equally spaced sub-
carrier, with zeros occupying the unused sub-carrier in
between. While in localized sub-carrier mode, the outputs are
confined to a continuous spectrum of sub-carrier. Interleaved
sub-carrier mapping mode of FDMA (IFDMA) is another
special sub-carrier mapping mode [13], [14]. The difference
between DFDMA and IFDMA is that the outputs of IFDMA
are allocated over the entire bandwidth, whereas the DFDMAs
outputs are allocated every several subcarriers [15], [16].

Finally, the inverse DFT (IDFT) module output is followed
by a cyclic prefix (CP) insertion that completes the digital stage
of the signal flow. The CP is used to eliminate ISI and preserve
the orthogonality of the tones. Assume that the channel length
of CP is larger than the channel delay spread [17].

B. Channel model

Channel model is a mathematical representation of the
transfer characteristics of the physical medium. These models
are formulated by observing the characteristics of the received
signal. According to the documents from 3GPP, a radio wave
propagation can be described by multipaths which arise from
reflection and scattering [17]. The received signal at the mobile
terminal is a superposition of all paths. If there are L distinct
paths from transmitter to the receiver, the impulse response of
the multipath fading channel can be represented as [17]:

CD(nvc) = ^©.(rrj) 8[m- x.(m)], (3)

where CO . (m) and T . (m) are attenuations and delays for each

path at time instant m, and 8(.) is the Dirac delta function. The
fading process for the mobile radio channel is given by

(D(v) = «) jA /l-(v/f d ) , (4)

where Doppler frquency f d = s/X, s is the speed of the mobile,

and X is the wavelength of the transmitted carrier. In order to
do simulations as close to the reality as possible, it is essential
to have a good channel model. This model is used to describe
the fast variations of the received signal strength due to
changes in phases when a mobile terminal moves. In case of
wideband modeling, each path of the impulse response can be
modeled as Rayleigh distributed with uniform phase except line
of sight (LOS) component cases [17].

C. Received signal model

The transmitted symbols propagating through the radio
channel can be modeled as a circular convolution between the
CIR and the transmitted data block i.e., [s(m)*&>(m,r)] .

Since, the channel coefficient is usually unknown to the
(2) receiver, it needs to be efficiently estimated. The impulse
response of multipath fading channel can be represented by a
tap-delayed line filter with time varying coefficients and
symbol-rate spaced coefficients.

Input
signal

s(m-l)

Delay

s(m-2)

Delay

s(m-3)

Wi(m)-*(X) w 2 (m)— Mj

Delay

s(m-L)

w 3 (m)

-G>

w L (m) — >{X)

(+>-

— -<£h-

Fig. 3. L-tapped delay line filter of a fading channel.

At the receiver, the opposite set of the operation is
performed. After synchronization, CP samples are discarded
and the remaining samples are processed by the DFT to
retrieve the complex constellation symbols transmitted over
the orthogonal sub-channels. The received signals are de-
mapped and equalizer is used to compensate for the radio
channel frequency selectivity. After IDFT operation, these
received signals are demodulated and soft or hard values of
the corresponding bits are passed to the decoder. The decoder
analyzes the structure of received bit pattern and tries to
reconstruct the original signal.

IV. RLS ADAPTIVE CE METHOD

An adaptive CE technique is a process that changes its
parameters as it gain more information of its possibly
changing environment. Among many iterative techniques that
exist in the open literature, the well-liked classes of
approaches which are achieve from the minimization of the
mean square error (MSE) between the output of the adaptive
filter and desired signal to perform CE as shown in Fig. 4.

Transmitted
sequence

AWGN noise
z(m)

s(m)

Multipath channel
model w(m)

Receive signal

AM.

e(m)

Mechanism for adaptive
update of weight coefficients

West(m)

Filter output

y(m)=w est T (m)s(m)

Fig. 4. Scheme for adaptive CE.

The signal s(m) is transmitted via a time-varying channel
w(m), and corrupted by an additive noise estimated by using
any kind of CE method. The main aim of most channel
estimation algorithms is to minimize the MSE i.e., between
the received signal and its estimate, while utilizing as little

163

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

computational resources as possible in the estimation process.
In the Fig. 4, we have unknown multipath fading channel, that
has to be estimated with an adaptive filter whose weight are
updated based on some criterion so that coefficients of
adaptive filter should be as close as possible to the unknown
channel. The RLS CE requires all the past samples of the
input and the desired output is available at each iteration. The
objective function of a RLS CE algorithm is defined as an
exponential weighted sum of errors squares:

c(m) = J r m e H (m)e(m) + SV w H (m)w(m), (5)

m= 1

where 8 is a positive real number called regularization
parameter, e(m) is the prior estimation error, and X is the
exponential forgetting factor with < X < 1. The X is used to
ensure that data in the distant past are paid less concentration
in order to provide the filter with estimating facility when it
operates in a time varying environment. When X = 1, the
algorithm has growing memory because the values of the filter
coefficients are a function of all the precedent input. In this
case, all the values of the error signal, from the time the filter
starts its process to the present, have the same influence on the
cost function. Consequently, the adaptive filter losses its
estimating ability, which is not important if the filter is used in
a stationary environment. In contrast, when < X < 1, the
algorithm has exponentially decaying memory as the recent
values of the observations have greater influence on the
formation of the filter coefficients and tends to forget the old
ones as shown in Fig. 5.

Weighting
factor (a)

8c(m)
8w(m)

= = -2 J r m s(m)e H (m) + 28^ n w(m)

m=l
n

= -2^ r m s(m)[h(m) - w H (m)s(m)] H + 28X n w(m)

w(m)[ J r m s(m)s H (m) + Sri] = £r m s(m)h H (m)

R s (m)w(m) = R sh (m)
w(m)= Rj^nOR^m)

(8)

where R s (m) is the transmitted auto-correlation matrix
R s (m) = Jr m s(m)s H (m) + SVI = XR s (m-l) + s(m)s H (m)

m=l

and R sh (m) is the cross correlation matrix
i.e.,

R sh ( m ) =J k n ' m s(m)h H (m) = ^R sh (m-1) + s(m)h H (m) .

m= 1

According to the Woodbury identity , the above R sh (m) can
be written as

/ /
/ /
/ /

/ /

/ '
/ 1

/ 1

' 1

/
/
/

\ «

, — 1 *►

(m-m) (m-1) ( m -0) (ra + 1)

Sample observation time

Fig. 5. Exponential weighting of observations at different time index.

The prior estimation error is the difference between the
desired response and estimation signal:

e(m) = h(m) - w H (m) s(m) (7)

The objective function is minimized by taking the partial
derivatives with respect to w(n) and setting the results equal to
zero.

Ri(m) = X- 1 Ri(m-l)-

^- 2 R;l(m-l)s(m)s H (m)R;l(m-l)
l+XV^RiCm-lJsCni)

(9)

For convenience of computing, let D(m) = R S h(m) and

K(m) = ^W (10)

l+}i 1 s H (m)D(m-l)s(m)

The K(m) is referred as a gain matrix. We may rewrite (9)
as:

D(m) = ^DCm-l) - r 1 K(m)s H (m)D(m-l) (11)

So simply (10) to

K(m) = D(m)s(m) = R^(m)s(m)

(12)

Substituting (11), (12) into (8), we obtain the following RLS
CE formula

w(m) = w(m-l) + K(m)[h(m) - w H (m-l)s(m)] H

= w(m-l) + K(m)s H (m), (13)

where e(m) is a prior estimation error as

e(m) = h(m) - w H (m-l)s(m) (14)

Therefore equation (13) is the recursive RLS CE algorithm to
update channel coefficient.

164

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

V. PERFORMANCE ANALYSIS

A. Complexity Analysis

The complexity of CE algorithm is of vital importance
especially for time varying wireless communication channels,
where it has to be performed periodically or even
continuously. Table I summarizes the computational
complexity of the RLS CE technique, where L is the channel
length, and real number indicates scalar operation. Here we
assume that each iteration requires the evaluation of the inner
product D(m)h(m) between two vectors of size L each. This
calculation requires L multiplications and L-l additions. Also
assumed that the evaluation of the scalar addition or
subtraction needs one real addition and multiplying the scalar
by the vector requires L multiplications.

TABLE I
COMPLEXITY PER ITERATION

versus SNR for the RLS CE method with different Doppler
frequencies 500Hz and 1.5kHz. One can observed that the
RLS CE method with 500 Hz Doppler frequency has 3 dB
better performances compared with 1.5kHz Doppler
frequency as desired. This CE scheme uses adaptive RLS
estimator which is able to update parameters of the estimator
continuously, so that knowledge of channel and noise
statistics are not required. The similar behavior can be
observed for BER performance in Fig. 7.

Operation

Complexity

Division

Multiplication

L 2 + 5L+1

Addition

L 2 + 3L

B. Experimental results

The error performance of the aforementioned iterative
estimation algorithm is explored by performing extensive
computer simulations. All simulation parameters of the LTE
SC-FDMA system in Doppler spread environments are
summarized in Table II.

□ For Doppler frequency lbuui-iz :
• For Doppler frequency 500Hz ~

= = = "^= S : =: q: : : : ;

X x j~" ~i~

-^s< +

^\ 4^

j _ _ X=X- - - - -

= = = = = 1 = = = =^Px= = =

^\^~

1 ^X \L

-^X

1 1

X_ ^\

- %V

10 15 20

SNR [dB]

Fig. 6. MSE performance comparisons of the LMS CE method.

Table II

THE SYSTEMS PARAMETERS FOR SIMULATION

Systems parameters

Assumptions

System bandwidth

5 MHz

Sampling frequency

7.68 MHz

Sub-carrier spacing

9.765 kHz

Modulation data type

BPSK

FFT size

Sub-carrier mapping scheme

IFDMA

IFFT size

512

Data block size

Cyclic prefix

4us

Channel

Rayleigh fading

Forgetting factor

0.99

Equalization

Doppler frequency

100, and 1.5 kHz

&— For Doppler frequency 1500Hz
For Doppler frequency 500Hz

In practice, the perfect channel coefficient is unavailable,
so estimated channel coefficient must be used instead. The
more correct estimated channel coefficient is, the better MSE
performance of the CE will achieve. Fig. 6 shows the MSE

10 15

SNR [dB]

Fig.7. BER performance comparisons of the LMS CE method.

165

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

VI. CONCLUSION

In this paper, we explore the performance of RLS CE
method for LTE SC-FDAM wireless communication systems
with different Doppler frequencies. The complexities, MSE
and BER performance of the RLS CE method, are analyzed
and compared with the different Doppler frequencies. We can
come to the conclusion that the RLS CE method with 500 Hz
Doppler frequency has 3 dB superior performances compared
with 1.5 kHz Doppler frequency.

REFERENCES

[I] B. Karakaya, H.Arslan, and H. A. Cirpan, "Channel estimation for
LTE uplink in high Doppler spread," Proc. Int. Con. on WCNC,
pp. 1126-1130, April 2008.

[2] J. Berkmann, C. Carbonelli, F.Dietrich, C. Drewes, and W. Xu,

"On 3G LTE terminal implementation standard, algorithms,

complexities and challenges," Proc. Int. Con. on WCMC , pp. 970-

975, August 2008.
[3] A. Ancora, C. Bona, and D.T.M. Slock, "Down-sampled impulse

response least-squares channel estimation for LTE OFDMA,"

Proc. Int. Con. ASSP, Vol. 3, pp. 293-296, April 2007.
[4] L. A. M. R. D. Temino, C. N. I Manchon, C. Rom, T. B. Sorensen,

and P. Mogensen, "Iterative channel estimation with robust wiener

filtering in LTE downlink," Proc. Int. Con. on VTC, pp. 1-5,

September 2008.
[5] S. Y. Park, Y.Gu. Kim, and C. Gu. Kang, "Iterative receiver for

joint detection and channel estimation in OFDM systems under

mobile radio channels, " IEEE Trans. On Comm., vol. 53, ilssue 2,

pp. 450-460, March 2004.
[6] S. Haykin, "Adaptive Filter Theory," Prentice-Hall International

Inc, 1996.
[7] J. J. V. D. Beek, O. E. M. Sandell, S. K. Wilsony, and P. O.

Baorjesson, "On channel estimation in OFDM systems," Proc. Int.

Con. on VTC, vol. 2, pp. 815-819, July 1995.
[8] O. Edfors, M. Sandell, J. V. D. Beek, and S. Wilson, "OFDM

channel estimation by singular value decomposition," IEEE Trans.

on Comm., vol. 46, no. 7, pp. 931-939, July 1998.
[9] M.H. Hsieh, and C.H. Wei, "Channel estimation for OFDM

systems based on comb-type pilot arrangement in frequency

selective fading channels," IEEE Trans, on Consumer Electronics,

vol. 44, issue 1, pp. 217-225, February 1998.
[10] P. Hoeher, S. Kaiser, and P. Robertson, "Two-dimensional pilot

symbol aided channel estimation by wiener filtering," Proc. Int.

Con. on ASSP, pp. 1845-1848, vol.3, April 1997.

[II] M. M. Rana, J. Kim, and W. K. Cho,"Low complexity downlink
channel estimation for LTE systems," in Proc. Int. Con. On
Advanced Commun. Technology, February 2010, pp. 1198-1202.

[12] M. M. Rana, J. Kim, and W. K. Cho," Performance Analysis of

Sub-carrier Mapping in LTE Uplink Systems," in Proc. Int. Con.

On COIN, August 2010.
[13] F. Adachi, H. Tomeba, and K. Takeda, "Frequency-domain

equalization for broadband single-carrier multiple access," IEICE

Trans, on Commun., vol. E92-B, no. 5, pp. 1441-1456, May 2009.
[14] S. Yameogo, J. Palicot, and L. Cariou, "Blind time domain

equalization of SC-FDMA signal," in Proc. Vehicular Technology

Conference, pp. 1-4, September 2009.
[15] S. H. Han and J. H. Lee, "An overview of peak to average power

ratio reduction techniques for multicarrier transmission," IEEE

Trans, on Wireless Commun., vol. 12, no. 2, pp. 56-65, 2005.
[16] H. G. Myung, J. Lim, and D. J. Goodman, "Peak-to-average power

ratio of sngle carrier FDMA signals with pulse shaping," in Proc.

Personal, Indoor and Mobile Radio Commun., September 2006.
[17] W. C. Jakes, Ed., Microwave mobile communications. New York:

Wiley-IEEE Press, Jan. 1994.

1 66 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Priority Based Congestion Control for
Multimedia Traffic In 3G Networks

Neetu Sharma 1 , Amit Sharma 2 , V.S Rathore 3 , Durgesh Kumar Mishra 4

1 3 Department of Computer Engineering, Rajasthan, India

l3 Rajasthan College of Engineering for women, Rajasthan, India

Shri Balagi College of Engineering & Technology, Rajasthan, India

Acropolis Institute of Technology and Research, Indore, MP, India

ABSTRACT- There is a growing demand for efficient
multimedia streaming applications over the Internet and next
generation mobile networks. Multimedia streaming services are
receiving considerable interest in the mobile network business.
As communication technology is being developed, the user
demand for multimedia services raises. The third generation
(3G) mobile systems are designed to further enhance the
communication by providing high data rates of the order of 2
Mbps. High Speed Downlink Packet Access (HSDPA) is an
enhancement to 3G networks that supports data rates of several
Mbit/s, making it suitable for applications like multimedia, in
addition to traditional services like voice call. Services like
person- to- person two way video calls or one way video calls, aim
to improve person- to- person communication. Entertainment
services like gaming, video streaming of a movie, movie trailers
or video clips are also supported in 3G. Many more of such
services are possible due to the augmented data rates supported
by the 3G networks and because of the support for Quality of
Service (QoS) differentiation in order to efficiently deliver
required quality for different types of services.

This paper present congestion control schemes that are suitable
for multimedia flows. The problem is that packet losses, during
bad radio conditions in 3G, not only degrade the multimedia
quality, but render the current congestion control algorithms as
inefficient. This paper proposed a solution that integrated the
congestion control schemes with a priority based multimedia
packets to increase the speed of multimedia data and reduce the
packet loss that is developed due to congestion in networks

Key words: UMTS, CN, BS, TFMCC, UTRAN, RNC

INTRODUCTION

The emerging multimedia application requires a fresh approach for
congestion control. A widely popular congestion control schemes are

TCP friendly rate control (TFRC) and Adaptive increase
multiplicative decrease (AIMD) used in networks. These algorithms
used for multimedia traffic but not much effective in packet loss.
TCP is the dominant transport protocol in the Internet, and the
current stability of the Internet depends on its end-to-end congestion
control, which uses an Additive Increase Multiplicative Decrease
(AIMD) algorithm. End-to-end congestion control of best-effort
traffic is required to avoid the congestion collapse of the global
Internet [11]. While TCP congestion control is appropriate for
applications such as bulk data transfer, some real-time applications
(that is, where the data is being played out in real-time) find halving
the sending rate in response to a single congestion indication to be
unnecessarily severe, For providing a better congestion control with
higher data rates a new effective scheme is used. Congestion control
is an important issue in both wired and wireless streaming
applications. Multimedia applications should use some form of
congestion control, both in wired and cellular networks, in order to
adapt the sending rate to the available bandwidth. Today's Internet
stability is due to TCP and its congestion control algorithm. TCP
represents a very efficient transport protocol in general and is
suitable for data transfer. However, it has been argued [13] that TCP
is unsuitable for video streaming because strict delay and jitter
requirements of video streaming are not respected by TCP.
Moreover, some TCP retransmissions are unnecessary for video
when data may miss the arrival deadline and become obsolete. This
has led researchers to look for alternative options. Most of the work
related to congestion control for video flows has either emulated TCP
or has used the TCP model. The well-known TCP-Friendly Rate
Control (TFRC) congestion control consists in an equation based rate
control mechanism [13] [14] [15], designed to keep a relatively steady
sending rate while still being responsive to congestion. When used
over wireless links, TFRC and TCP cannot distinguish between the
wireless losses and the congestion losses. They both may suffer from
the link underutilization if the connection traverses a wireless link.

167

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

This is because they consider dropped packets as a sure sign of
congestion and reduce the ending rate significantly. The inability to
identify a wireless loss followed by unnecessary reduction in sending
rate results in link underutilization.
A. UMTS Introduction

Universal Mobile Telecommunications System (UMTS) is a third-
generation (3G), wireless cellular network that uses Wideband Code
Division Multiple Access (WCDMA) as its radio interface
technology. UMTS offers higher data rates with respect to older 2G
and 2.5G networks and, with the Release 5 version, is evolving into
an all-IP, wireless network. The increased bandwidth provided by
UMTS allows for the deployment of a wide range of services, like
voice, data and multimedia streaming services. In wireless networks,
congestion control, alone, may not be enough to ensure good quality
of multimedia streaming and efficient utilization of the network.
Packet losses due to the high bit error rate not only degrade the
multimedia quality, but render the current congestion control
algorithms as inefficient: these algorithms back-off on every packet
loss even when there is no congestion. We integrate the congestion
control schemes with an adaptive retransmission scheme in order to
selectively retransmit some lost multimedia packets. Fig.l shows the
transmission of multimedia data over a wireless channel.

■ Overview

How to improve QoS & resource utilization ?

Network
■ Video Streaming over 3G
•Header Compression

User

How can applications adapt :
improve the multimedia quality ?

• Congestion Control for Video Flows

Fig. 1 Transmission of Multimedia data
B. GENERAL UMTS NETWORK:

UMTS, the successor of GSM, is evolving toward a future wireless
all-IP network. In this paper we present how it supports real-time IP
multimedia services, as these services are expected to drive the
adoption of wireless all-IP networks.

UMTS networking architecture is organized in two domains. The
user equipment (UE) and the public land mobile network (PLMN).
The UE is used by the subscriber to access the UMTS services.
PLAN is further divided into two land-based infrastructures

(i) UTRAN (UMTS terrestrial radio- access network)

(ii) CN (core network).

The UTRAN handless all radio-related functionalities and the CN is
responsible for maintaining subscribes data and for connections.
UTRAN contain two types of nodes Radio network controller (RNC)
and Node B. Node B is the base station and provides radio coverage
to one or more cells. Node B connected to UE via Uu interface and to
the RNC via Iub interface. Uu is a radio interface based on the
wideband code division multiple access (WCDMA) technology [7].

The CN consist by two types of general packet radio service support
nodes (GSNs). That is gateway GSN (GGSN) and serving GSN
(SGSN). SGSN provide the routing functionality. It manages a group
of RNCs and interacts with the home location register (HLR). HLR
permanently store the subscriber data. SGSN connected to GGSN via
the Gn interface. RNC connect to SGSN via Iu interface. Through the
GGSN the UNTS network connect to external packet data network
like the internet.

/Tub lub jfS|

Fig.2 General UMTS Network

C. 3G/UMTS Problems

• Problems due to the use of IP

o IP doesn't support real time streaming

requirements
o Overhead due to packet header

• Problems due to radio conditions

o Scarce and time varying bandwidth

o Congestion, wireless losses & large delay

168

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

D. UMTS QoS Classes

UMTS defines four QoS classes [2] and the classified traffic gets the
treatment inside the UMTS network according to its class. The four
QoS classes are:

• Conversational class: The traffic from the applications like person-
to-person video or voice call is classified into conversational class.
The delay and jitter requirements for this type of traffic are very
strict. This is because on the both end points there is a human
expecting the delivery of the voice and/or video data in very short
time after it is sent.

• Streaming class: Video on Demand (VoD) falls under this class.
The delay requirements are there but are not as strict as the
conversational class.

• Interactive class: The interactive traffic like interactive e-mail or
web browsing falls under this category. Though there is still some
delay requirement, it is less strict than the conversational and
streaming classes. Moreover, since the traffic mostly pertains to data
applications, the bit error rate should be very low.

• Background class: This class is the most insensitive to delay. It
includes the traffic from background applications like background
email and SMS. Though, the bit error rate, like the Interactive class,
should be very low.

D Congestion Control for Multimedia data

TCP-Friendly Rate Control (TFRC) is an end-to-end congestion
control mechanism, whose goal is to provide rate control for unicast
flows in IP networks. The main feature of TFRC is its ability to
smoothly adapt the sending rate of a flow to network conditions,
while competing for bandwidth with TCP flows in a relatively fair
manner. TFRC was designed to offer a more stable sending rate than
TCP on wired, best-effort networks, making it suitable for
applications like multimedia streaming. We evaluate the performance
of TFRC, compare it with that of TCP and new TFRC for different
multimedia classes, under different scenarios of MAC -layer
scheduling, radio conditions and background traffic.

TFRC [4] [10] is an end-to-end congestion control mechanism
suitable for applications with constraints on rate stability, like voice
or streaming media. It has been designed to adapt the sending rate of
a flow in a smooth manner, while trying to fairly share the available
bandwidth with competing TCP flows. TFRC is an Internet standard
[4], and it has been adopted at the IETF as one of the congestion

control profiles that may be used with the DCCP transport protocol
[10]; TFRC may also be implemented by UDP -based applications
wishing to perform congestion control. This paper presents a
simulation study of TFRC over UMTS networks supporting

HSDPA. Since we are interested in video streaming applications, we
evaluate the performance of TFRC in terms of rate stability over
different time scales, and compare it with that of TCP. Several
scenarios of MAC-layer scheduling, radio conditions and background
traffic are considered.

This paper proposed a more reliable algorithm that provides
congestion control for different multimedia classes. Priority assigned
to each of the packet according to multimedia classes. So whenever
the congestion occurs in the network the lowest priority packets are
dropped. If overall loss rate for lower priority packets is not very
high, then we can safely assume that the congestion loss rate for the
highest priority packets will be insignificant. In such a case, the loss
of highest priority packets will be mainly due to wireless errors.
Thus, it is to be expected that, in general, there is a good correlation
between wireless packet loss rate and the total loss rate of highest
priority packets.

2. THE PROPOSED SCHEME

This paper provides a mechanism of congestion control for the
multimedia transmission over UMTS. We analyze TCP friendly
multicast congestion control (TFMCC) over UMTS and generalize it
to different multimedia classes [5] [6]. We design a novel mechanism
for congestion control that is Content Sensitive TCP Friendly
Multicast Congestion Control (CSTFMCC). We perform a little
modification in UMTS network and the packet field. At various level
of network we provide the control mechanism that prevents the
network from the congestion. Multimedia Class I traffic includes
video and audio traffic from users equipped with an adjustable rate.
Class II traffic includes non-real time data traffic such as e-mail, file
transfer and web browsing traffic. These two classes contain different
multimedia traffic that is more delay sensitive or less delay sensitive.
So class I traffic support the real time applications and more delay
sensitive. Due to congestion, if any loss of the packet or the delay
between the packets can reduce the quality of received video/audio.
Whether in class II traffic, if congestion occur it is acceptable to
buffer non-real time data at a network node or at the user station and
transmit them at a slower rate. In a large multicast group, there will
usually be at least one receiver that has experienced a recent packet

169

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

loss. If the congestion control mechanisms require that the sender
reduces its sending rate in response to each loss, as in TCP, then
there is little potential for the construction of scalable multicast
congestion control.

In wireless communication systems like UMTS, the packet loss may
not mean network congestion. The quality of wireless link may be
degrading due to signal fading. During a fading period, the bit error
rate of wireless link may become very high but after that period the
wireless link is expected to recover. TFMCC uses a feed back
scheme which allows the receiver to calculating the slowest
transmission rate to always reach the sender.

A. Sender end: Fragmentation of data packets perform at the sender
end. The sender fragment the data packet with on bit of priority.
There are two parts of the data (i) packet header and (ii) payload. The
size of header part change by one bit shows the priority of packet. So
there is only one bit modification perform in the size of data packet
and it increases the speed of multimedia packets.

If (incoming request for higher priority packets)
If (there is a free channel) then

allocate the free channel
else
If (lower priority packets)

put in a buffer

If (there is free channel again)

allocate the free channel to lower

priority packets

else

Ignore request

endif
else

Ignore request
endif

endif
End.

ignore request
endif

Fig.3 Algorithm For proposed model

Fig. 4 shows the flow chart for the proposed model. Flow chart
shows the arrival of packet and priority check by the routers at
layer2. According to this priority the packets is being processed.

B. Multimedia Packet Size: Multimedia packet size depends on the
multimedia classes. The proposed scheme redesigns the multimedia
packets. It increases the multimedia packet size by one bit. This bit
shows the priority of multimedia packets. The highest priority
packets serve first by the routers at the layer 2. So the size of the
packet is increased by one bit.

C. Routing Scenario: For fast transmission of multimedia
information the proposed scheme give the priority to all multimedia
packets. When a user want to send multimedia data the data framing
perform at the sender end. The sender constructs the frame with a
priority bit. This information stored in the header of the packet for
priority access to the router. Sender sends the packets towards its
destination. Multimedia packets reach at the network. At layer two
the router checks the destination address and priority bit of the
packet. If a higher priority packet arrives then router serves first to
the packet which contains a highest priority. This increases the speed
of multimedia packets and decreases the congestion in the networks.

For implementing this scenario the changes perform in the size of
packets and in routers. Following algorithm shows the scenario for
routing the various packets according to priority. Fig. 3 shows the
algorithm for the proposed model.

Channel? ^^-i£2

> 1

assign' c tonne I* ^

pack".'" 6

Accept packet and

Fig. 4 Flow Chart for the proposed model

Receiving End: At the receiving end defragmentation perform. The
receiving data packet reaches at the destination and multimedia
information is available for the user respectively

A. Simulation Platform

The simulation that we use for this is EURANE (NS-2

Extension) [16]. Following fig. 3 shows the simulation topology to
increase the multimedia quality. This paper focuses on the problem
of evaluating the subjective video quality and presents the quality
estimation tool that we employed, a performance evaluation study
done with the well known ns-2 network simulator

170

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

I I

I I UMTS

IKTTERNET I jJOREJ4ETWORK_

§]j^X^v a;

Fig:3 Simulation Topology

the video packet trace file is fed to the ns-2 simulator (compiled with
the EURANE extensions). This trace file serves as a traffic generator
during the simulation. A simulation script allows defining the
particular scenario under consideration (network topology,
simulation parameters, and so on). When the simulation is run, an
output trace file is produced which contains delay- and loss-related
information for every video packet sent by the (simulated) video
server.

Conclusion

This paper represents to design a multicast group congestion control
mechanism over the UMTS networks. This mechanism is content
sensitive and optimized for multimedia traffic. This paper performs
some changes in existing UMTS architecture and the packet field of
data packet. User equipment (UE) has some additional function to
detect the multimedia class type. The additional functionality of UE
and the field of data packet has main target to remove the packet loss
and congestion.

Refrences

[1] Holma. H. and Toskala, A. "WCDMA for UMTS: Radio
Access for Third Generation Mobile Communications"

(3 rd Edition),Wiley,2004.

[2] Antonios Alexiou, Dimitrios Antonellis and Chritos
Bouras" A Study of Multicast Congestion Control for

UMTS" proc Int. J. Commun. Sys. (2009)

[3] Miiighua Clien and Avitleh Zaklior "Rate Control for
Streaming Video over wireless "proc of IEEE

INFOCOM 2004 Hong Kong China

[4] J. Widmer, R. Denda, M. Mauve "A Survey on TCP-
Friendly Congestion Control" IEEE Network, 15 "(3),
May- June 2001

[5] Cui-Quing Yang and Alapati V.S. Reddy "A Taxonomy
for congestion control algorithms in packet switching
networks"proc IEEE 1995

[6] Ljiljana Trajkovic and S. Jamaloddin Golestant"
Congestion Control for Multimedia Services" proc of
IEEE INFOCOM 1992

[7] Antonios Alexiou, Dimitrios Antonellis and Chritos Bouras
"Equation Based Congestion Control for Video
Transmission Over WCDMA Networks"proc of IEEE
AINA'06 Vienna, Austria, pp .445-450.

[8] Minghua Chen and Avideh Zakhor "Rate Control for
Streaming Video over Wireless" proc of IEEE
INFOCOM 2004

[10] Kamal Deep Singh*, Arpad Huszak., David Ros§, Cesar
Viho* and Gabor Jeney.

"Congestion Control and Adaptive Retransmission for
Multimedia Streaming over Wireless Networks" The

Second International Conference on Next Generation
Mobile Applications, Services, and Technologies

[11] Kamal Deep Singh ^, Julio Orozcof, David Ros/ and
Gerardo Rubino "Improving perceived streaming-video
quality in High Speed Downlink Packet Access" proc of
IEEE IEEE "GLOBECOM" 2008

[12] S. Floyd and K. Fall. Promoting the Use of End-to-end
Congestion Control in the Internet. IEEE/ACM
Transactions on Networking, Aug. 1999.

[13] R. Jain, K. Ramakrishnan, and D. Chiu. Congestion
Avoidance in Computer Networks with a
Connectionless Network Layer. Tech. Rep. DEC-TR-
506, Digital Equipment Corporation, August 1987.

[14] S. Floyd, M. Handley, J. Padhye, and J. Widmer .Equation
based congestion control for unicast applications., In
Proceedings of ACM SIGCOMM 2000, pages 43.56,
Stockholm, Aug. 2000.

[15] S. Floyd, E. Kohler, and J. Padhye, .Profile for Datagram
Congestion Control Protocol (DCCP) Congestion Control
ID 3: TCP-Friendly Rate Control (TFRC)., RFC 4342,
IETF, Mar. 2006.

[16] M. Handley, S. Floyd, J. Padhye, and J. Widmer. TCP
Friendly Rate Control (TFRC): Protocol Specification.
Internet Standards Track RFC 3448, IETF, Jan. 2003.

[16] Ns-2 . Network Simulator,

http:///www.isi.edu/nsnam/ns/index.html

171

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Authors Profile

Ms. Neetu Sharma, Reader

Biography: Mrs. Neetu Sharma obtained her
Engineering degree from University of Rajasthan
and Masters Degree from Rajasthan Vidyapeeth,
Udiapur securing First division with honors in both.
Currently she is pursuing Ph.D. (CSE) in
Congestion Control in 3G from Gyanvihar
University, Jaipur, India. She has been Reader and
HOD of the department of CSE at Rajasthan
College of Engineering for Women, Jaipur, India.
She has extensively worked in various field of
Computer Engineering. She has published many
national papers in the reputed journals and
conferences. She is an author of the book 'System
Software Engineering' for B.Tech. students. She is
the member of renowned societies like IEEE, IEEE
computer society, ISTE and CSI also.

Dr. Vijay Rathore, Associate Professor

Biography: Dr. Vijay Singh Rahore obtained his
MCA and Ph.D. (CSE) from University of
Rajasthan, India. He is an Associate Professor,
Shree Kami College, Jaipur, India. He has more
than 10 years of industrial and teaching experience.
His areas of interest are Computer Organization &

Architecture, Operating System Fundamentals,
DBMS & RDBMS (Oracle, DB2, SQL Server, MS-
Access, DBASE, etc.), Data Structures,
Programming Languages (C, C++, Java (J2SE,
J2ME, J2EE), VB, COBOL), Networking
Technologies (Data Communications, Internet &
Intranet, E-Commerce, Network Security,
Cryptology etc.), Software Engineering, System
Analysis & Design, Management Information
System, Decision Support System, Artificial
Intelligence, E-Governance, Computer Center
Management, UNIX, etc.. He is the member of
renowned society like ISTE.

Dr. Durgesh Kumar Mishra

Professor (CSE) and Dean (R&D),

Acropolis Institute of Technology and Research,
Indore, MP, India,

Ph - +91 9826047547, +91-731-4730038
Email: durgeshmishra@ ieee.org

Chairman IEEE Computer Society, Bombay Chapter
Vice Chairman IEEE MP Subsection

Biography: Dr. Durgesh Kumar Mishra has
received M.Tech. degree in Computer Science from
DAW, Indore in 1994 and PhD degree in
Computer Engineering in 2008. Presently he is
working as Professor (CSE) and Dean (R&D) in
Acropolis Institute of Technology and Research,
Indore, MP, India. He is having around 21 Yrs of
teaching experience and more than 7 Yrs of
research experience. He has completed his research
work with Dr. M. Chandwani, Director, IET-DAVV
Indore, MP, India in Secure Multi- Party
Computation. He has published more than 60
papers in refereed International/National Journal
and Conference including IEEE, ACM etc. He is a
Senior Member of IEEE, Chairman of IEEE

172

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Computer Society, Bombay Chapter, India. Dr.
Mishra has delivered his tutorials in IEEE
International conferences in India as well as other
countries also. He is also the programme committee
member of several International conferences. He
visited and delivered his invited talk in Taiwan,
Bangladesh, Nepal, Malaysia, Bali-Indonesia,
Singapore, Sri Lanka, USA and UK etc in Secure
Multi-Party Computation of Information Security.
He is an author of one book also. He is also the
reviewer of tree International Journal of
Information Security. He is a Chief Editor of
Journal of Technology and Engineering Sciences.
He has been a consultant to industries and
Government organization like Sale tax and Labor
Department of Government of Madhya Pradesh,
India.

Mr. Amit Sharma, Assistant Professor

Biography: Mr. Amit sharma obtained his MCA
from University of Rajasthan and aboout to
complete his M.Tech.(CSE) from Rajasthan
Technical University. He is an Assistant Professor
in Sri Balaji College of Engineering & Technology,
Jaipur. He has more than 5 years of industrial and
teaching experience. His areas of interest are Open
Source, Networking, Advance Computer
Architecture and Information Security. He is the
member of renowned society like ISTE.

1 73 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Adaptive Sub-block ARQ techniques for wireless networks.

A.N.Kemkar, Member, ISTE and Dr. T.R.Sontakke Member,ISTE

Abstract — Wireless channels are highly affected
by unpredictable factors such as co channel
interference, adjacent channel interference,
propagation path loss, and shadowing and multi
path fading. An adaptive ARQ scheme, we mean
an ARQ scheme with two or more different
transmission modes meant for different channel
conditions, which uses some channel sensing
mechanism to decide which transmission mode is
used. In this paper, we propose an adaptive
transmission of sub-blocks scheme, for wireless
networks. As the channel becomes increasingly
noisy, the data block is divided into smaller sub-
blocks for transmission. Each sub-block is
encoded for error control by a CRC code. The
received block is checked for errors sub-block by
sub-block. The propose sub-block retransmission
scheme provides improved throughput over
conventional ARQ schemes by retransmitting
only the same number of sub-blocks in the
occurrence of errors. .

Index Terms — Retransmission protocol; Hybrid
ARQ,CRC

1. INTRODUCTION:

In a mobile radio channel, burst errors frequently
occur because of Rayleigh fading and shadowing.
In particular, for a large cell- size system with the
radius of more than several km, shadowing often
becomes the predominant source of burst errors.
(Shown in Fig- 1)

Therefore, in order to provide reliable packet data
transmission in such a channel, an efficient
automatic repeat request (ARQ) protocol must be
employed, since data service can tolerate some

□ A.N.Kemkar 1 ,S.R.T.M.U,Nanded.

+91-9819150392, ankemkar@gmaiLcom

Dr.T.R.Sontakke 2

Ex.Director - S.G.G.S.I.T.E.- Nanded

Principal,Sidhant college of Engineering

Pune.+91 -98223 92766,trsontakke@gmail.com

delays but is sensitive to loss and errors. Many
researchers have devoted much effort to analyze
the throughput for various ARQ protocols in
Rayleigh fading channels. In the packet data
transmission, short packets are less likely to
encounter fades than long packets, but on the
other hand, they are more burdened by overheads.
In other words, the packet length to maximize the
instantaneous throughput is closely related to the
dynamic channel condition due to fading,
shadowing, and propagation path loss. Therefore,
if we choose the optimum packet length
adaptively by estimating the channel condition,
we can continuously achieve the maximum
attainable throughput.

Paper is organized as follows. We start by,
describes in detail about related work in section 2.
A Communication system model and proposed
method of adaptation in section 3. In section 4
we have presented system analysis 5. Simulation
parameters and results . Followed by conclusion
in section 6.

2. RELATED WORK:

A change of transmission mode can mean, for
example, a change of the packet size in the SR
scheme [1] or a change of the number of
transmitted copies of a packet in the GBN scheme
[2] or a change of the code rate in an HARQ-I
scheme [3]. In these schemes, the channel sensing
is usually done by observing the
acknowledgements sent by the receiver to the
transmitter. This can mean either estimation of
error rates, as in [4], or detection of channel state
changes, as in [5] and [6], which does not require
as long an observation interval (OBI) as reliable
error rate estimation.

In [2], an adaptive SR scheme was proposed,
where the packet size used in the current
transmission was selected from a finite set of
values based on a long-term BER estimate. This
estimate was obtained by counting the incorrectly
received packets over a time interval and

174

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

assuming that there can be at most one bit error in
an erroneous packet. Another adaptive SR
scheme with variable packet size was proposed in
[5], where the a posteriori distribution of the
BER was computed based on the number of
retransmissions during the OBI, and the packet
size was selected so that the expected efficiency
of the protocol was maximized. In [2], Yao
proposed an adaptive GBN scheme where the
transmitted number the transmitted number of
copies of a packet was variable.
Numerous adaptive HARQ schemes have been
suggested in the literature. Typically, the code
rate is varied according to the estimated channel
conditions. In [6] and [7], adaptive HARQ-I
schemes were studied with convolutional codes
used for error correction. Finite-state Markov
models were assumed for the channel. Switching
between transmission modes depended on the
number of erroneous blocks occurring during an
OBI. A similar adaptive HARQ-I scheme with
either block or convolutional codes were
proposed in [3]. In [4], sequential statistical tests
were applied on the acknowledgements to detect
channel state changes. An adaptive HARQ-II
scheme with variable packet size was proposed
for wireless ATM networks in [8]. This scheme
used rate compatible convolutional (RCC) codes
for error correction. In [9], three different
adaptive HARQ schemes are proposed using
Reed- Solomon codes for error correction.
Another adaptive HARQ scheme using Reed-
Solomon codes with variable rate for error
correction was proposed in [10]. In this scheme
short-term symbol error rate was estimated by
computing the bitwise modulo-2 sum of two
erroneous copies of a packet. This method was
originally proposed in [1 1].
3. A COMMUNICATION SYSTEM MODEL
AND PROPOSED METHOD OF
ADAPTATION:

3.1 A communication system model:
Fig. 1 shows the communication system model.
In the non cellular or large cell- size system, a
radio base station continuously transmits data
packets to a single mobile terminal with no
packet collision after the link connection is
established. Table I summarizes the digital
mobile communication system characteristics,
where we choose a binary frequency shift keying

non coherent detection (non coherent FSK)
scheme in terms of easy implementation, because
it requires no complicated carrier recovery ci rcuit .

► Buffer

Transmitter
Electronics

Adaptive Logic

Buffer

Receiver
Electronics

Fig.l. Communication System model.
Table 1 -Communication system parameters

Parameters

Description

Channel type

Contention Free, half duplex

Modulation
/Demodulation

Binary FSK,Non coherent
detection.

Packet structure

Information packet length
256,128,64,32,16 bits

3.2. Adaptation Policy

According to the variations of SNR, the receiver
channel may be consider to be in one of the states
at each instant 't\ We assume that the sender
knows the state at the transmission time for
receiver. Let's define the transmission status at
time t as the set of all channel state. Before
transmitting, the adaptive algorithm in the sender
must estimate the efficiency and packet loss rate
of the ARQ/FEC protocol using all the available
coding schemes as well as the ARQ protocol as a
function of the transmission status. It then tries to
find the protocol satisfying the desired packet loss
rate. The time is divided into transmission rounds.
Each transmission round corresponds to the
transmission c k' packets in case of ARQ. A
transmission round ends when the sender is
informed about the reception state of the receiver.
The adaptive algorithm is repeated at the end of

175

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

each transmission round. Basically, the algorithm
goes through the following steps:

1 . At the beginning of the algorithm, the sender
determines the desired packet loss rate (SNR) of
the session. It also determines the transmission
status.

2. The sender estimates the packet loss rate of the
ARQ protocol as well as the ARQ/FEC protocol
using all the available coding schemes, based on
the transmission status. It then adjusts its
parameters and starts the transmission of the
blocks.

3. At the end of a transmission round, the sender
again determines the transmission status. It then
repeats the step.

4. PERFORMANCE ANALYSIS OF THE
PROPOSED SCHEME:

The performance analysis of the scheme is
measured in terms of throughput of the proposed
scheme. Further we show the comparison of
throughput with sub block and without sub block
transmission schemes with Adaptive scheme.
Expression of throughput for ARQ for present
scheme:
K

V =

E[T]

(1)

Where ^information bits in a block. E[]

=Expectation of number of transmitted bits in a
given block.

T = Mn + Yji (2)

Where M =number of sub blocks, « =number of
bits in a sub block, Ti =number of transmitted bits
for i th transmission.

E[T]=Mn + f d E[T l ] (3)

1=1

Where E[T\ =Average number of transmitted

bits.

Out of M sub blocks if L sub blocks are
transmitted at the f 1 retransmission, then random
variable, Ti takes the value Ln ,if L out of M
sub-blocks are retransmitted at the f 1
retransmission. Our algorithm compute the value
of (3) to get the result from equation (l).In our
analysis we have consider the variable packet size

as a retransmission units with fixed rate of data
transmission. Hence we can send the packet of
specific size based on the estimated signal to
noise ratio (SNR). As shown in the Fig.l the
system configuration of ARQ techniques
combines with adaptive packet size modulation.
With an exact bit error rate equation for FSK at
certain signal to noise ratio '7\ we can decide
the value of packet size satisfying the required
BER (bit error rate) Assuming that we have ' M

'different block sizes { L1L2L3L4 Lm\ .

Let Ai',ie{0 9 M) with ^,-as the threshold

value of signal to noise ratio, being between the
i th level and i + 1 level.

Ao is the lowest possible signal to noise ratio and
Am is the highest possible signal to noise ratio.

5. SIMULATION RESULTS:

We evaluate the performance of the proposed
scheme implemented with Matlab. We run the
simulation for three schemes i.e. with sub block
transmission and without sub block transmission
and adaptive. The simulation parameters are
shown in the table 2.

Table 2- Simulation Parameters

Parameters

Notation

Values

Signal to Noise Ratio

Varied

Threshold values of
SNR

snr

8,7,6,5

Max. number of
Retransmissions

Number of sub blocks
retransmitted

Varied

Cyclic Redundancy
Check

CRC

Varied

Bit error rate

BER

Varied

Packet error rate

PER

Varied

Throughput efficiency

Varied

Data Rate

9.6kbps

Simulation runs for 5000 total blocks. Result
is the average of independent experiments where
each experiment uses different randomly
generated uniform parameters. We use mean
values which are obtained independent

176

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

experiments as a basic data to get the result.
Simulation results are shown Table 3.

For the packet error rate 0.3,0.5,0.7 the
throughput of the system is say VvVi^Vs-

Adaptive Scheme
parameters

Throughput efficiency

SNR

Sub
block

Packet
size

v 2

256

0.92

0.84

0.82

128

0.93

0.85

0.83

0.94

0.92

0.88

0.97

0.95

0.93

Below
this

0.98

0.94

0.96

Table 3.Simulation Results of adaptive
scheme

I h f o i) o \\ p i) t V s . P a c k

Li.

8 5

1 o jc r

-0-2 S

u 1

-0-t S

u 1 ____B_

1 o jc k

-0-8 S

u 1

1 o |c k

■ N. "X

-0-1 6

S b

B l!o c

k I X\

-0-3 2

S b

B Ijo c
i

k j

10 . 20 . 30 . 40 . 50 . 60 . 70 . 80 . 9 1
P a c k e t E r r o r R a

T h ro u g h p u t V s . P a c k e t E rro r R a

W it h Micro
W ill) o i) t
Adaptive A

R Q

j i i i i i i_

lo c k
o b Ic

Fig. 2. Comparison as per Table -2

.20.30.40.50.60.70.80.9 1
P a c k e t E rro r R ate

Fig. 3. Comparison of three scheme
6.conclusion: Table -2 and Fig.2 shows the
result of five modes of sub block
transmission .Proposed Adaptive scheme
will choose proper mode in accordance with
the channel parameters, here threshold
values of SNR (considered as channel
sensing mechanism).From the Fig. 3 the
simulation results showed that the simulated
throughput points for adaptive scheme
follows the ideal throughput curve(with
micro block transmission) very closely. The
proposed Adaptive sub-block retransmission
scheme improved the throughput and the
reliability by using dynamically adapting the
number of Sub block transmission according

177

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

to length to the varying channel packet error

rate..

REFERENCES :

[1] J.A.C. Martins and J.C. Alves. ARQ
protocols with adaptive block size
perform better over a wide range of
bit error rates. IEEE Transactions on
Communications, 38:73 7-73 9, June
1990.

[2] Y.-D. Yao. An effective go-back-N
ARQ scheme for variable-error-rate
channels. IEEE Transactions on
Communications, 43:20-23, January
1995.

[3] M. Rice and S.B. Wicker. Adaptive
error control for slowly varying
channels. IEEE Transactions on
Communications, 42:917-926,

February/March/April 1994.

[4] M. Rice and S.B. Wicker. A
sequential scheme for adaptive error
control over slowly varying
channels. IEEE Transactions on
Communications, 42: 1 533-1 543,
February/ March/April 1994.

[5] E. Modiano. An adaptive algorithm
for optimizing the packet size used in
wireless ARQ protocols. Wireless
Networks, 5:279-286, July 1999

[6] B. Vucetic. An adaptive coding
scheme for time- varying channels.
IEEE Transactions on

Communications, 39:653-663, May
1991.

[7] B. Vucetic, D. Drajic, and D. Perisic.
Algorithm for adaptive error control
synthesis

[8] I. Joe. A novel adaptive hybrid ARQ
scheme for wireless ATM networks.
Wireless Networks, 6:211-219, July
2000.

[9] S. Choi and K.G. Shin. A class of
adaptive hybrid ARQ schemes for
wireless links. IEEE Transactions on
Vehicular Technology, 50:777-790,
May 2001.

[10] H.Minn,M. Zeng, and V.K.

Bhargava. On ARQ scheme with

adaptive error control. IEEE
Transactions on Vehicular

Technology, 50:1426-1436,

November 2001. IEE Proceedings,
PartF, 135:85-94, February 1988
[11] S.S. Chakraborty, M.

Liinaharja, and E. Yli-Juuti. An
adaptive ARQ scheme with packet
combining for time varying
channels. IEEE Communications
Letters, 3:52-54, February 1999.

178

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

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

Vol 8, No. 6, September 2010

Trigon Based Authentication Service Creation with

Globus Middleware

Ruckmani V l Anitha Kumari K 2 Sudha Sadasivam G 3

Senior Lecturer , MCA, Lecturer, IT, , Professor ,CSE,

Ramakrishna Engineering College, PSG College of Technology, PSG College of Technology,

Coimbatore, India Coimbatore,India Coimbatore,India

Dhaarini M P 4
Lecturer , IT,

PSG College of Technology
Coimbatore,India .

Abstract — A Grid is built from multi-purpose protocols and
interfaces that address fundamental issues as authentication,
authorization, resource discovery, and resource access.
Security is of utmost importance in grid computing
applications as grid resources are heterogeneous, dynamic, and
multi- domain. Authentication remains as the significant
security challenge in grid environment. The proposed
approach uses a dual authentication protocol in order to
improve the authentication service in grid environment. The
protocol utilizes the fundamental concepts of trigon and based
on the parameters of the trigon the user authentication will be
performed. In the proposed protocol, the password is
interpreted and alienated into more than one unit and these
units are stored in two different servers, namely,
Authentication Server and Backend Server. Only when the
combined authentication scheme from both the servers
authenticates the user, the privilege of accessing the requested
resources is obtained by the user. The main advantage of
utilizing the dual authentication protocol in grid computing is
that an adversary user cannot attain the access privilege by
compromising a single consolidated server because of the fact
that the split password is stored in different servers. Grid
service is stateful and transient web service, which can be
invoked by clients, and is considered to be the mainstream of
future internet. The creation of Web Services standards is an
industry-led initiative, with some of the emerging standards in
various states of progress through the World Wide Web
Consortium (W3C). To achieve reuse of behaviors of this
authentication concept, operations are often grouped together
to form a trigon based authentication service.

Keywords — Trigonbasedauthentication, web services,
globus.

virtual organizations boundaries. A VO is a dynamic group
of individuals, groups, or organizations that have common
rules for resource sharing [8]. Confidentiality of information
in a VO Should also be ensured [28]. The necessity for
secure communication between grid entities has motivated
the development of the Grid Security Infrastructure (GSI).
GSI provides integrity, protection, confidentiality and
authentication for sensitive information transferred over the
network in addition to the facilities to securely traverse the
distinct organizations that are part of collaboration.
Authentication is done by exchanging proxy credentials and
authorization by mapping to a grid map file. Grid
technologies have adopted the use of X.509 identity
certificates to support user authentication. SOAP protocol
[12] is used by the emerging OGSA. This necessitates for
support message layer security using XML digital signature
standard and the XML encryption standard [11]. Globus
Toolkit [24] provides security services for authentication,
authorization, management of user credentials and user
information. Laccetti and G. Schmid [14] have introduced a
unified approach for access control of grid resources. PKI
(Public Key Infrastructure) and PMI (Privilege Management
Infrastructure) infrastructures were utilized at the grid layer
after authentication and authorization procedures.
Czajkowski [5] have explained about agreement based grid
management. Nagaratnam [18] have introduced security
architecture for open grid services. H.-L. Truong[26] define a
framework for monitoring and analyzing qos metrics of grid
Services. The proposed work aims at authenticating the users
by using trigon concept and to host this operation as a web
service.

I. INTRODUCTION

A. Globus Middleware

Grid computing has emerged as a significant new field,
distinguished from conventional distributed computing by its
concentration on large-scale resource sharing, innovative
applications, and, in some cases, high-performance
orientation . Grid computing is concentrating on large-scale
resource sharing and collaboration over enterprises and

Globus [25] provides a software infrastructure that
enables applications to handle distributed heterogeneous
computing resources as a single virtual machine. Globus is
constructed as a layered architecture in which high-level
global services are built upon essential low-level core local
services. Middleware is generally considered to be the layer

179

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

of software sandwiched between the operating system and
applications, providing a variety of services required by an
application to function correctly. Recently, middleware has
re-emerged as a means of integrating software applications
running in distributed heterogeneous environments. In a
Grid, the middleware is used to hide the heterogeneous
nature and provide users and applications with a
homogeneous and seamless environment by providing a set
of standardized interfaces to a variety of services.

B. Web Services

The term Web Services describes an important emerging
distributed computing paradigm that differs from other
approaches such as DCE, CORBA, and Java RMI in its focus
on simple, Internet-based standards to address heterogeneous
distributed computing. Web services define a technique for
describing software components to be accessed, methods for
accessing these components, and discovery methods that
enable the identification of relevant service providers. Once a
web service is created, it is advertised in a registry called
UDDI (Universal Description, Discovery and Integration)
[27], where it can be searched. The UDDI will provide the
location to the service provider's WSDL (Web Services
Description Language) [29] file that describes the methods
that can be invoked and the parameters required. Messages
are exchanged through the protocol SOAP (Simple Object
Access Protocol) [30].

The established standards include:

SOAP (XML Protocol). SOAP provides an envelope
which encapsulates XML data for transfer through the Web
infrastructure (e.g. over HTTP, through caches and proxies),
with a convention for Remote Procedure Calls (RPCs) and a
serialization mechanism based on XML Schema data types.
SOAP is being developed by W3C in cooperation with the
Internet Engineering Task Force (IETF).

Web Services Description Language (WSDL). Describes
a service in XML, using an XML Schema; there is also a
mapping to the Resource Description Framework (RDF). In
some ways WSDL is similar to an interface definition
language IDL. WSDL is available as a W3C note [WSDL].

Universal Description Discovery and Integration (UDDI).
This is a specification for distributed registries of web
services, similar to yellow and white pages services. UDDI
supports 'publish, find and bind': a service provider
describes and publishes the service details to the directory;
service requestors make requests to the registry to find the
providers of a service; the services 'bind' using the technical
details provided by UDDI. It also builds on XML and SOAP
[UDDI].

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

Vol 8, No. 6, September 2010
Web Services have certain advantages over other
technologies:

Web Services are platform-independent and language-
independent, since they use standard XML languages. This
means that my client program can be programmed in C++
and running under Windows, while the Web Service is
programmed in Java and running under Linux.

Service Processes: This part of the architecture generally
involves more than one Web service. For example, discovery
belongs in this part of the architecture, since it allows us to
locate one particular service from among a collection of Web
services.

Service Description: One of the most interesting features
of Web Services is that they are self-describing. This means
that, once you've located a Web Service, you can ask it to
'describe itself and tell you what operations it supports and
how to invoke it. This is handled by the Web Services
Description Language (WSDL).

Service Invocation: Invoking a Web Service (and, in
general, any kind of distributed service such as a CORBA
object or an Enterprise Java Bean) involves passing
messages between the client and the server. SOAP (Simple
Object Access Protocol) specifies how we should format
requests to the server, and how the server should format its
responses. In theory, we could use other service invocation
languages (such as XML-RPC, or even some ad hoc XML
language). However, SOAP is by far the most popular choice
for Web Services.

Transport: Finally, all these messages must be
transmitted somehow between the server and the client. The
protocol of choice for this part of the architecture is HTTP
(Hypertext Transfer Protocol), the same protocol used to
access conventional web pages on the Internet. Again, in
theory we could be able to use other protocols, but HTTP is
currently the most used one.

C. Web Service Definition Language(WSDL)

Web Services programmers usually only have to
concentrate on writing code in their favorite programming
language and, in some cases, in writing WSDL. SOAP code,
on the other hand, is always generated and interpreted
automatically for us. Once we've reached a point where our
client application needs to invoke a Web Service, we
delegate that task on a piece of software called a stub. Using
stubs simplifies our applications considerably. We don't have
to write a complex client program that dynamically generates
SOAP requests and interprets SOAP responses (and similarly
for the server side of our application). We can simply
concentrate on writing the client and/or server code, and
leave all the dirty work to the stubs (which, again, we don't
even have to write ourselves... they can be generated
automatically from the WSDL description of a web
service). The stubs are generally generated only once. In
general, we only go through the discovery step once, then
generate the stubs once (based on the WSDL of the service
we've discovered) and then reuse the stubs as many times as
we want (unless the maintainers of the Web service decide to

180

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

change the service's interface and, thus, its WSDL
description).

II. TRIGON BASED AUTHENTICATION
ARCHITECTURE

When legitimate entities (users) login, the trigon based
authentication server splits the password into its components
and stores the authentication information in two servers -
namely authentication and backend server. Users have to
register with the Authentication server, so that it can hold a
part of the interpreted password with itself and another part
in the Backend server. The block diagram illustrating the
registration process of the users is depicted in the Figure
5. As illustrated in Figure5, the users who require services
from the VO have to register initially with the Authentication
server using their username and password. The
Authentication server calculates the Pi as given in (1). Along
with this authentication server generates two large prime
numbers, namely, a and a', which are considered as the two
sides of a trigon. It is difficult to hack the values of a and a'
as they are large prime numbers (as per RSA Factoring
Challenge). Here, Pi is taken as the angle between the two the
two sides of the trigon a and a'. Now, the Authentication
server can easily determine the opposite side of the angle P i?
termed as a".

With these trigon parameters, a, Vaa' and Paa' are
found as

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

Vol 8, No. 6, September 2010

Vaa'=a-a'

(i)

User registers with username
and password in Authenticati on

f -\

Authentication file calculates
a. a" .a"

Determines variance and pro duct
c£ the sides V ££ ' and P^ ~~ and a

Saves fj. : username in authentication

file and forwards username, "VW and

P^ 'to Backend file

B acken d f il e saves VW an d P ££ Tor
the corresrjondine username

Fig 1 Flow Diagram

In (4), PAI is the ASCII-interpreted value of the given
password pwi , n is the total number of digits in PAI and PAI
( j) represents the first j digits of PAI . The PAI can be
calculated by the following steps.

Change the pwi into its corresponding ASCII value.

Paa' = a * a' (2)

a = 2Paa'-a" 2 (3)

where, a, a' and a' 'are the three sides of trigon. a is a
strengthening parameter used as the index . Vaa' and Paa' are
the variance and the product of the sides a and a'
respectively. With the parameters a , a' and a' ' as the
sides of trigon and Pi be the angle between the sides a
and a' the generated trigon will be assumed . After the
calculation of a , Vaa' and Paa' , the authentication server
stores the a value and its corresponding username in a
database and forwards Vaa' and Paa' to the Backend server
along with the username. Hence, the password is interpreted
and alienated into two units and stored in two separate
server. The authentication procedure is based on the
fundamental concepts of a trigon. Initially, the user who
wants the services of VO has to login to the Authentication
server using the username and password. Here, ui and
pwi refers to username and password of i th user. The
Authentication server calculates the Password index ( Pi )
from the password as

Pi =

P AI(i) / 10 pow n-2 ; if P AI (i) > 180

Calculate the three-fourth of total digits of the
ASCII value modulo 1 80, which results the first three digits
of PAI .

Append the remaining one- fourth of the ASCII digits to

PAI,

Then, from Pi the Authentication Server determines
the Authentication index ( AI ) for ui as

AI(i) = Pi/2

(5)

PAI(i)/10pown-3 ; else

(4)

Then, the Authentication Server searches for the
username index a i for the corresponding ui which has
already been stored in the server database during
the process of the registration. Subsequently, a i is
sent to the backend server along withui . When the
Backend server receives the index a i and the
username from the Authentication server, it
searches for Vaa' and Paa' the Variance and the product
of the sides a and a' respectively, which have been
saved in the backend server database during the
process of registration. From these values,

181

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

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

Vol 8, No. 6, September 2010
the Backend server calculates an Authentication Token
ATi and sends it to the Authentication server to xmlns:tns= M http://www.globus.org/namespaces/add/hello_in

authenticate the ui . The ATi can be calculated as stance"

AT(i) = ai + Vaa'I * 2Paa'I (6)

In (6), Vaa' and Paa' are pre-calculated

values computed during individual user registration. After
retrieval of ATi from the Backend server, the
Authentication server authenticates the user based on the
token from the Backend server and the index calculated at
the Authentication server. The authentication code (or)
condition which authenticates the ui is given by

SinAI(i) =(l-ATi /2) 1 / 2 (7)

The authentication process is performed by the
authentication condition given in (7). When the condition is
satisfied, the user is decided to be valid and the Server sends
a token called Token for VO access to the user.

III. IMPLEMENTATION - AUTHENTICATION AS
SERVICE

A service is an entity that provides some capability to its
clients by exchanging messages. A service is defined by
identifying sequences of specific message exchanges that
cause the service to perform some operation. By thus
defining these operations only in terms of message exchange,
we achieve great flexibility in how services are implemented
and where they may be located. A service-oriented
architecture is one in which all entities are services, and thus
any operation visible to the architecture is the result of
message exchange.

Prerequisites are:

build.xml
globus-build-service. sh

1 . Creation of auth. wsdl File

<?xml version= M 1.0" encoding="UTF-8"?>
<defmitions name-'auth"

targetNamespace= "hrtp ://www. globus . org/namespaces/add/h
elloinstance"

xmlns= "hrtp ://schemas .xmlsoap . org/wsdl/"

xmlns:xsd="http://www.w3 .org/200 l/XMLSchema">

<types>

<xsd: schema
targetNamespace= "hrtp ://www. globus . org/namespaces/add/h
ello instance"

xmlns:tns="http://www.globus.org/namespaces/add/hello_in
stance"

xmlns:xsd="hrtp://www.w3 .org/200 l/XMLSchema">
<xsd:element name="addition">
<xsd:complexType>
<xsd:sequence>

<xsd:element name="inputl" type="xsd:int"/>
<xsd:element name="input2" type="xsd:int"/>
</xsd:sequence>
</xsd: complexType>
</xsd:element>
<xsd:element name="response" type="xsd: string "/>

<xsd: element
type="xsd: string "/>

name-'additionrequest"

</xsd:schema>

</types>

<part name-'parameters"

element="tns:additionrequest"/>

</message>

</message>

</portType>

</definitions>

2. Create namespace2package.mappings for mapping
instances ,bindings and services.

182

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

3. Write Implementation program .

4.Create deploy-server.wsdd

[globus@g20 service]$ vi deploy-server.wsdd

5. Create deploy-jndi-config.xml

[globus@g20 service] $ vi deploy-jndi-config.xml

6.Build the service

[globus@g20 -example] $ sh globus-build-service.sh -d
org/add/service/ -s schema/add/hello.wsdl

7. After the successful building Grid Archive(GAR) file
has been created. Now we have to deploy the GAR file using
globus-deploy-gar command.

[globus@g20~example] $ globus-deploy-gar

orgaddser vice, gar

8. After successful deployment of the GAR file start the
globus container.

[globus@g20 ~]$ globus-start-container

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

Vol 8, No. 6, September 2010
[37]:
hrtp://192. 168. 100. 3 :8443/wsrtf services/Trigger Service

[38]:
hrtp://192. 168. 100.3 :8443/wsrf / services/gsi/AuthenticationS
ervice

i BJE3SSEH

0§§

File Edit View Terminal Help

Anitha

Authentication check

Anitha

Authentication check

Anitha

Authentication check

Anitha

Authentication check

Anitha

Authentication check

Anitha t

alpha value

Alpha : -5.1360B02760354224E11

Vp : -764080.0

Pp : 1.35901310201E11

Token is : 0.2583132520673121

Pi : 104.97

Authentication token <At> 0.2583132520673121

myToken : 0.3708433739663

UserToken : 0.3708433739663

RegToken : 0.3708433739663

VALID USER

Authentication call

[39]:
hrtp://192. 168. 100.3 :8443/wsrf7services/TestRPCService

[40]:
hrtp://192. 168. 100.3:8443/wsrFservices/ManagedMultiJobSe
rvice

9. Write Client Program for authentication.

10. Before running the compiler, make sure to run the
following:

source SGLOBUSLOC ATION/etc/globus-devel-env. sh

The globus-devel-env.sh script takes care of putting all
the Globus libraries into your CLASSPATH.

[globus@gcluster example] $ source /usr/local/globus-
4.0.7/etc/globus-devel-env.sh

1 1 . [globus@gcluster example] $ j avac
org/add/client/Client.java

12. [globus@gcluster example] $ Java
org/add/client/Client

iObus@anitt:~/TrigonBased Authenticate

glii

File Edit View Terminal Help

[globus@anitt TrigonBasedAuthentication] $ source /usr/local/globus-4. Q . 7/etc/glo
bus-devel-env.sh

[globus@anitt TrigonBasedAuthentication] $ javac org/add/client/Client.java

[globus@anitt TrigonBasedAuthentication]$ Java org/add/client/Client

https : //192 . 168 . 4 . 114 : 8443/wsrf /services/TrigonBasedAuthenticationService

Inputs to TrigonBasedAuthentication Service : Anitha p Anitha

Username and Password Authentication: INVALID USER

[globus@anitt TrigonBasedAuthentication] $ Java org/add/client/Client

https : //192 . 168 . 4 . 114 : 8443/wsrf /services/TrigonBasedAuthenticationService

Inputs to TrigonBasedAuthentication Service : Anitha p Anitha

Username and Password Authentication: VALID USER

[globus@anitt TrigonBasedAuthentication] $ \\

Fig 2. GUI - Server side authentication

Fig 3. GUI -Valid User

[35]:
hrtp://192. 168. 100.3:8443/wsrf7services/DefaultTriggerServi
ce

[36]:
http://192.168.100.3:8443/wsrf/services/TrigonBasedAuth
enticationService

183

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

The user is invalid since the username and password
haven't stored in authentication and backend file. First time
when user login his/her trigon value of the password gets
stored in the respective files. Next time when they login they
gets authenticated and token gets issued .

IV. CONCLUSION

The Internet is a reasonable model for the Grid, providing
both an early version of its services and a platform from
which to evolve. The authentication protocol, proposed here,
enhanced the grid security as the authentication mechanism
utilized two servers for authentication. This simple trigon
concept utilization in the authentication protocol introduced a
novel and revolutionary idea in the authentication
mechanism as well as in grid environment. The
implementation of our dual authentication protocol showed
its effective performance in pinpointing the adversaries and
paving the way to valid users for access with the VO for
resource sharing. So by hosting this authentication as a
service it make the grid environment more secure. In future
these services will be located by type instead by names.

ACKNOWLEDGMENT

Our thanks to Dr.R.Rudramoorthy,Principal,PSG
College of Technology and Mr. K.Chidambaram, Director,
Grid and Cloud systems group, Yahoo software
development, India Private Limited for their support. This
project is carried out in Grid and Cloud lab, PSG College of
Technology.

REFERENCES

[1] S. Albayrak, S. Kaiser, and J. Stender. Advanced grid management
software for seamless services. Multiagent Grid Syst, 1(4):263.270,
2005.

[2] Andrew Johnson, Carl Kesselman, Jason Leigh, and Steven Tuecke,
Application Experiences with the Globus Toolkit, Seventh IEEE
International Symposium on High Performance Distributed
Computing (HPDC-7 ... T).

[3] Antonioletti, M, and Jackson, M, OGSA-DAI Product Overview,
2003. Available at www.ogsa-dai.org.uk/downloads/docs/OGSA-
DAI-USER-M3 -PRODUCTOVERVIEW.pdf.

[4] Christensen, E., Curbera, F., Meredith, G., and Weerawarana, S., Web
Services Description Language (WSDL) 1.1, W3C, Note 15, 2001.
Available at www.w3.org/TR/wsdl.

[5] K. Czajkowski, A. Dan, J. Rofrano, S. Tuecke, and M. Xu.
Agreement-based grid service management (ogsiagreement),
Proceedings of 6th IEEE/ACM International Workshop on Grid
Computing (Grid2005).

[6] Della-Libera, G. et al. (2002) Web Services, Secure Conversation
Languages (WS-Secure Conversation). Version 1.0, available at
http://msdn.microsoft.com/library/default.asp7urW

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

Vol 8, No. 6, September 2010
lib/enus/dnglobspec/html/wssecureconversation.asp (accessed on
2002).

[7] Eastlake, D. and Reagle, J. (Eds.) (2002) XML Encryption Syntax
and Processing. W3C Recommendation, available at
http://www.w3.org/TR/xmlenc-core/ (accessed on December 2002).

[8] Foster. I., Kesselman. C. and Tuecke. S, "The Anatomy of the Grid:
Enabling Scalable Virtual Organizations", International Journal of
High Performance Computing Applications", vol. 15, no. 3, pp. 200-
222,2001.

[9] I. Foster, C. Kesselman, J. Nick, and S. Tuecke. The physiology of
the grid: An open grid services architecture for distributed systems
integration, 2002.

[10] I. foster, Argonne National Laboratory and the University of Chicago
,carl kesselman, Information Sciences Institute, University of
Southern California "THE GRID 2 blueprint for a new computing
infrastructure",2004

[11] Graham, S., Simeonov, S., Boubez, T. etc, "Building Web Service
with Java: Making Sense of XML, SOAP, WSDL and UDDI",
Indianapolis, IN: Sams Publishing, 2002.

[12] Gudgin, M., Hadley, M., Mendelsohn, N., Moreau, J-J. and Nielsen,
H.F. (2003) SOAP Version 1.2 Part 1: Messaging Framework. W3C
Recommendation, Available at http://www.w3.org/TR/soapl2-partl/
(accessed on June 2003).

[13] Grid Archive Creation: http://gdp.globus.org/gt4-

tutorial/multiplehtml/ch08s02.html.

[14] G. Laccetti and G. Schmid, "A framework model for grid security",
Future Generation Computer Systems, vol. 23, no. 5, pp.702-713,
June 2007.

[15] Li, Y., Jin, H., Zou, D., Chen, J. and Han, Z. (2007) 'A scalable
service scheme for secure group communication in grid', 31st Annual
International Computer Software and Applications Conference
(COMPSAC 2007).

[16] Li, Y., Jin, H., Zou, D., Liu, S. and Han, Z. (2008) 'An authenticated
encryption mechanism for secure group communication in grid', 2008
International Conference on Internet Computing in Science and
Engineering.

[17] Li Hongweia, Sun Shixina and Yang Haomiaoa, "Identity-based
authentication protocol for grid", Journal of Systems Engineering and
Electronics, Vol. 19, no. 4, pp.860-865, August 2008.

[18] Nagaratnam, N., Janson, P., Dayka, J., Nadalin, A., Siebenlist, F.,
Welch, V., Foster, I., and Tuecke, S., Security architecture for Open
Grid Services, 2002. Available at

www.globus.org/ogsa/Security/draft-ggf-ogsa-sec-arch-0 1 .pdf.

[19] Open Grid Services Architecture Data Access and Integration
(OGSA-DAI) Project: www.ogsa-dai.org.uk.

[20] K. Rochford, B. A. Coghlan, and J. Walsh. An agent-based approach
to grid service monitoring. In Proc. International Symposium on
Parallel and Distributed Computing (ISPDC 2006), July July, 2006.

[21] Shirasuna, S., Nakada, H., Matsuoka, S., and Sekiguchi, S.,
Evaluating Web services based implementations of GridRPC, in 11th
IEEE International Symposium on High Performance Distributed
Computing, Edinburgh, Scotland. IEEE Computer Society Press, Los
Alamitos, CA, 2001.

[22] Siebenlist, F., Welch, V., Tuecke, S., Foster, I., Nagaratnam, N.,
Janson, P., Dayka, J., and Nadalin, A., Roadmap towards a secure
OGSA. Global Grid Forum, draft, 2002.

[23] Shengxian Luo, Xiaochuan Peng, Shengbo Fan Peiyu Zhang, Study on
Computing Grid Distributed Middleware and Its
Application,International Forum on Information Technology and
Application,2009.

[24] Thatte, S., XLANG, Web services for business process design Web
site, Microsoft Corporation.

www.gotdotnet.com/team/xml_wsspecs/xlang-c/default.htm..

[25] Ian Foster ,Carl Kesselman,The Globus Project: A Status Report, 5th
IEEE Symp. on High. Performance Distributed Computing.

[26] H.-L. Truong, R. Samborski, and T. Fahringer. Towards a framework
for monitoring and analyzing qos metrics of grid services. In E-
SCIENCE '06: Proceedings of the Second IEEE International

184

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

[27]
[28]

[29]

[30]

Conference on e-Science and Grid Computing, page 65, Washington,

DC, USA, 2006. IEEE Computer Society.

UDDI. The UDDI technical white paper, http://www.uddi.org/, 2000.

Von Welch, Frank Siebenlist, Ian Foster, John Bresnahan, Karl

Czajkowski, Jarek Gawor, Carl Kesselman, Sam Meder, Laura

Pearlman and Steven Tuecke, "Security for Grid Services", in

proceedings of the 12th IEEE International Symposium on High

Performance Distributed Computing, pp.48- 57, June 2003.

W3C Note "Web Services Definition Language (WSDL) 1.1",

http://www.w3.org/TR/WSDL.

W3C Note "Simple Object Access Protocol(SOAP)
1 . 1",http://swww.w3 .org/TR/WSDL.

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

Vol 8, No. 6, September 2010
Cloud Computing. She has presented 1 paper in National Conference. She
is the Best Outgoing Student in MTech 2010-2011 in PSG College of
Technology. You may contact her at dhaarinimp@gmail.com

AUTHORS PROFILE

Dr G Sudha Sadasivam is working as a
Professor in Department of Computer Science and
Engineering in PSG College of Technology,
India. Her areas of interest include, Distributed
Systems, Distributed Object Technology, Grid
and Cloud Computing. She has published 20
papers in referred journals and 32 papers in
National and International Conferences. She has
coordinated two AICTE - RPS projects in
Distributed and Grid Computing areas. She is also the coordinator for PSG-
Yahoo Research on Grid and Cloud computing. You may contact her at
sudhasadhasivam@yahoo . com

in Sri Ramakrishna En
Grid Computing special
ruckmaniv@yahoo.com

V Ruckmani received B. Sc, MCA and M. Phil
degrees from the department of computer science,
Bharathiar University, India in 1994, 1997 and
2003 respectively. She is currently pursuing the
Ph. D degree, working closely with Prof. G.
Sudha Sadasivam. From 1997 to 2000 she worked
at PSG College of Arts and Science in the
department of Computer Science. Since
December 2000 she is working as a senior
lecturer in Department of Computer Applications
ineering College, India. She works in the field of
izing in the area of security. You may contact her at

EK Anitha Kumari received BE(CSE) from
Department of Computer Science ,Avinashilingam
Deemed University and ME(SE) from Department
of Computer Science ,Anna University. She is
working as a Lecturer in Department of Information
Technology in PSG College of Technology, India.
Her areas of interest include Grid and Cloud
Computing. She has published 1 paper in referred
international journal and 5 papers in National and
International Conferences. She is the Best Outgoing Student in ME 2010-
201 1 in PSG College of Technology. She awarded Gold Medal in BE(CSE)
in Avinashilingam Deemed University . You may contact her at
anitha.psgsoft@gmail.com

M P Dhaarini received BTech(IT) from
Department of Information Technology ,Anna
University and MTech(IT) from Department of
Information Technology, Anna University. She is
working as a Lecturer in Department of
Information Technology in PSG College of
Technology, India. Her areas of interest include

185

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

Dr. H. B. Kekre et. al.(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, 2010

Performance Evaluation of Speaker Identification for

Partial Coefficients of Transformed Full, Block and

Row Mean of Speech Spectrogram using DCT,

WALSH and HAAR

Dr. H. B. Kekre

Senior Professor,

MPSTME, SVKM's NMIMS

University

Mumbai, 400-056, India

Dr. Tanuja K. Sarode

Assistant Professor,
Thadomal Shahani Engg.

College, Bandra (W),
Mumbai, 400-050, India

Shachi J. Natu

Lecturer,
Thadomal Shahani Engg.

College, Bandra (W),
Mumbai, 400-050, India

Prachi J. Natu

Assistant Professor,

GVAIET, Shelu

Karjat 410201,

India

Abstract- In this paper an attempt has been made to provide
simple techniques for speaker identification using transforms
such as DCT, WALSH and HAAR alongwith the use of
spectrograms instead of raw speech waves. Spectrograms form a
image database here. This image database is then subjected to
different transformation techniques applied in different ways
such as on full image, on image blocks and on Row Mean of an
image and image blocks. In each method, results have been
observed for partial feature vectors of image. From the results it
has been observed that, transform on image block is better than
transform on full image in terms of identification rate and
computational complexity. Further, increase in identification rate
and decrease in computations has been observed when
transforms are applied on Row Mean of an image and image
blocks. Use of partial feature vector further reduces the number
of comparisons needed for finding the most appropriate match.

Keywords- Speaker Identification, DCT, WALSH, HAAR,
Image blocks, Row Mean, Partial feature vector.

I. INTRODUCTION

To provide security in a multiuser environment, it has
become crucial to identify users and to grant access only to
those users who are authorized. Apart from the traditional login
and password method, use of biometric technology for the
authentication of users is becoming more and more popular
nowadays. Biometrics comprises methods for uniquely
recognizing humans based upon one or more intrinsic physical
or behavioral traits. Biometric characteristics can be divided in
two main classes: Physiological which are related to the shape
of the body. Examples include fingerprint, face recognition,
DNA, hand and palm geometry, iris recognition etc.
Behavioral, which are related to the behavior of a person.
Examples include typing rhythm, gait and voice. Techniques
like face recognition, fingerprint recognition and retinal blood
vessel patterns have their own drawbacks. To identify an

individual by these methods, he/she should be willing to
undergo the tests and should not get upset by these procedures.
Speaker recognition allows non-intrusive monitoring and also
achieves high accuracy rates which conform to most security
requirements. Speaker recognition is the process of
automatically recognizing who is speaking based on some
unique characteristics present in speaker's voice [2]. There are
two major applications of speaker recognition technologies and
methodologies: speaker identification and speaker verification.

In the speaker identification task, a speech utterance from
an unknown speaker is analyzed and compared with speech
models of known speakers. The unknown speaker is identified
as the speaker whose model best matches the input utterance.
In speaker verification, an identity is claimed by an unknown
speaker, and an utterance of this unknown speaker is compared
with a model for the speaker whose identity is being claimed. If
the match is good enough, that is, above a threshold, the
identity claim is accepted. The fundamental difference between
identification and verification is the number of decision
alternatives [3]. In identification, the number of decision
alternatives is equal to the size of the population, whereas in
verification there are only two choices, acceptance or rejection,
regardless of the population size. Therefore, speaker
identification performance decreases as the size of the
population increases, whereas speaker verification performance
approaches a constant, independent of the size of the
population, unless the distribution of physical characteristics of
speakers is extremely biased.

Speaker identification can be further categorized into text-
dependent and text independent speaker identification based on
the relevance to speech contents [2, 4].

Text Dependent Speaker Identification requires the speaker
saying exactly the enrolled or given password/speech. Text
Independent Speaker Identification is a process of verifying the
identity without constraint on the speech content. It has no

186

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

Dr. H. B. Kekre et. al.(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, 2010

advance knowledge of the speaker's utterance and is more
flexible in situation where the individuals submitting the
sample may be unaware of the collection or unwilling to
cooperate, which presents more difficult challenge.

Compared to Text Dependent Speaker Identification, Text
Independent Speaker Identification is more convenient because
the user can speak freely to the system. However, it requires
longer training and testing utterances to achieve good
performance. Text Independent Speaker Identification is more
difficult problem as compared to Text Dependent Speaker
Identification because the recognition system must be prepared
for an arbitrary input text.

Speaker Identification task can be further classified into
closed set and open set identification.

In closed set problem, from N known speakers, the speaker
whose reference template has the maximum degree of
similarity with the template of input speech sample of unknown
speaker is obtained. This unknown speaker is assumed to be
one of the given set of speakers. Thus in closed set problem,
system makes a forced decision by choosing the best matching
speaker from the speaker database.

In the open set text dependent speaker identification,
matching reference template for an unknown speaker's speech
sample may not exist. So the system must have a predefined
tolerance level such that the similarity degree between the
unknown speaker and the best matching speaker is within this
tolerance.

In the proposed method, speaker identification is carried out
with spectrograms and transformation techniques such as DCT,
WALSH and HAAR [15-18]. Thus an attempt is made to
formulate a digital signal processing problem into pattern
recognition of images.

The rest of the paper is organized as follows: in section II
we present related work carried out in the field of speaker
identification. In section III our proposed approach is
presented. Section IV elaborates the experiment conducted and
results obtained. Analysis of computational complexity is
presented in section V. Conclusion has been outlined in section
VI.

II. RELATED WORK

All speaker recognition systems at the highest level contain
two modules, feature extraction and feature matching.

Feature extraction is the process of extracting subset of
features from voice data that can later be used to identify the
speaker. The basic idea behind the feature extraction is that the
entire feature set is not always necessary for the identification
process. Feature matching is the actual procedure of identifying
the speaker by comparing the extracted voice data with a
database of known speakers and based on this suitable decision
is made.

There are many techniques used to parametrically represent
a voice signal for speaker recognition task. One of the most
popular among them is Mel-Frequency Cepstrum Coefficients
(MFCC)[1].

The MFCC parameter as proposed by Davis and
Mermelstein [5] describes the energy distribution of speech
signal in a frequency field. Wang Yutai et. al. [6] has proposed
a speaker recognition system based on dynamic MFCC
parameters. This technique combines the speaker information
obtained by MFCC with the pitch to dynamically construct a
set of the Mel-filters. These Mel-filters are further used to
extract the dynamic MFCC parameters which represent
characteristics of speaker's identity.

Sleit, Serhan and Nemir [7] have proposed a histogram
based speaker identification technique which uses a reduced set
of features generated using MFCC method. For these features,
histograms are created using predefined interval length. These
histograms are generated first for all data in feature set for
every speaker. In second approach, histograms are generated
for each feature column in feature set of each speaker.

Another widely used method for feature extraction is use of
linear Prediction Coefficients (LPC). LPCs capture the
information about short time spectral envelope of speech. LPCs
represent important speech characteristics such as formant
speech frequency and bandwidth [8].

Vector Quantization (VQ) is yet another approach of
feature extraction [19-22]. In Vector Quantization based
speaker recognition systems; each speaker is characterized with
several prototypes known as code vectors [9]. Speaker
recognition based on non-parametric vector quantization was
proposed by Pati and Prasanna [10]. Speech is produced due to
excitation of vocal tract. Therefore in this approach, excitation
information can be captured using LP analysis of speech signal
and is called as LP residual. This LP residual is further
subjected to non-parametric Vector Quantization to generate
codebooks of sufficiently large size. Combining nonparametric
Vector Quantization on excitation information with vocal tract
information obtained by MFCC was also introduced by them.

III.PROPOSED METHODS

In the proposed methods, first we converted the speech
samples collected from various speakers into spectrograms
[11]. Spectrograms were created using Short Time Fourier
Transfer method as discussed below:

In the approach using STFT, digitally sampled data are
divided into chunks of specific size say 128, 256 etc. which
usually overlap. Fourier transform is then obtained to calculate
the magnitude of the frequency spectrum for each chunk. Each
chunk then corresponds to a vertical line in the image, which is
a measurement of magnitude versus frequency for a specific
moment in time.

Thus we converted the speech database into image
database. Different transformation techniques such as Discrete
Cosine Transform [12], WALSH transform and HAAR
transform are then applied to these images in three different
ways to obtain their feature vectors.

1. Transform on full image

2. Transform on image blocks obtained by dividing an
image into four equal and non-overlapping blocks

187

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

Dr. H. B. Kekre et. al.(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, 2010

3. Transform on Row Mean of an image and on Row
Mean of image blocks.

From these feature vectors, again identification rate is
obtained for various portions selected from the feature vector
i.e. for partial feature vector [15, 23, 24]. Two different sets of
database were generated. First set, containing 60% of the total
images as trainee images and 40% of the total images as test
images. Second set, containing 80% of the images as trainee
images and 20% of the total images as test images. Euclidean
distance between test image and trainee image is used as a
measure of similarity. Euclidean distance between the points
X(X1, X2, etc.) and point Y (Yl, Y2, etc.) is calculated using
the formula shown in equation. (1).

S(X i -Y i ) z
i=l

(i)

Smallest Euclidean distance between test image and trainee
image means the most probable match of speaker. Algorithms
for transformation technique on full image and transformation
techniques on image blocks are given below.

A Transformation techniques on full image[27, 28]:

In the first method 2-D DCT / WALSH / HAAR is applied
on the full image resized to 256*256. Further, instead of full
feature vector of an image only some portion of feature vector
i.e. partial feature vector is selected for identification purpose.
This selection of feature vector is illustrated in Fig. 1 and it is
based on the number of rows and columns that have been
selected from the feature vector of an image. For example,
initially first full feature vector (i.e. 256*256) has been selected
and then partial feature vectors of size 192*192, 128*128,
64*64, 32*32, 20*20 and 16*16 were selected from the feature
vector. For these different sizes, identification rate was
obtained.

135

192

Fig. 1: Selection of partial feature vector

Algorithm for this method is as follows:

Step 1. For each trainee image in the database, resize an
image to size 256*256.

Step 2. Apply the transformation technique (DCT /
WALSH / HAAR) on resized image to obtain its
feature vector.

Step 3. Save these feature vectors for further comparison.

Step 4. For each test image in the database, resize an
image to size 256*256.

Step 5. Apply the transformation technique (DCT /
WALSH / HAAR) on resized image to obtain its
feature vector.

Step 6. Save these feature vectors for further comparison.

Step 7. Calculate the Euclidean distance between feature
vectors of each test image with each trainee
image corresponding to the same sentence.

Step 8. Select the trainee image which has smallest
Euclidean distance with the test image and
declare the speaker corresponding to this trainee
image as the identified speaker.

Repeat Step 7 and Step 8 for partial feature vector
obtained from the full feature vector.

B. Transformation technique on image blocks[27, 29]:

In this second method, resized image of size 256*256 is
divided into four equal parts as shown in Fig. 2 and then 2-D
DCT / WALSH / HAAR is applied to each part.

Fig. 2: Image divided into four equal non-overlapping blocks

Thus when N*N image is divided into four equal and non-
overlapping blocks, blocks of size N/2*N/2 are obtained.
Feature vector of each block when appended as columns forms
a feature vector of an image. Thus size of feature vector of an
image in this case is of 128*512. Again Euclidean distance is
used as a measure of similarity. Here also using partial feature
vectors, identification rate has been obtained. Partial feature
vectors of size 96*384, 64*256, 32*128, 16*64 and 8*32 have
been selected to find identification rate. Detailed steps are
explained in algorithm given below:

Step 1. For each trainee image in the database, resize an
image to size 256*256.

Step 2. Divide the image into four equal and non-
overlapping blocks as explained in Fig. 2.

Step 3. Apply transformation technique (DCT/ WALSH
/HAAR) on each block obtained in Step 2.

Step 4. Append the feature vectors of each block one
after the other to get feature vector of an image.

Step 5. For each test image in the database, resize an
image to size 256*256.

Step 6. Divide the image into four equal and non-
overlapping blocks as shown in Fig. 2.

Step 7. Apply transformation technique (DCT /WALSH
/HAAR) on each block obtained in Step 6.

Step 8. Append the feature vectors of each block one
after the other to get feature vector of an image.

188

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

Dr. H. B. Kekre et. al.(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, 2010

Step 9. Calculate the Euclidean distance of each test
image with each trainee image corresponding to
the same sentence.

Step 10. Select the trainee image which has smallest
Euclidean distance with the test image and
declare the speaker corresponding to this trainee
image as the identified speaker.

Repeat Step 9 and Step 10 for partial feature vectors
selected from feature vector obtained in Step 4 and Step 8.
Selection of partial feature vector is similar to the one shown in
Fig. 1. But in this method, size of feature vector is 128*512,
96*384, 64*256, 32*128, 16*64 and 8*32.

C. Transformation techniques on Row Mean [16-18] of an
image and on Row Mean of image blocks [27, 29]:

In this approach, Row Mean of an image is calculated. Row
mean is nothing but an average of pixel values of an image
along each row. Fig. 3 shows how the Row Mean of an image
is obtained.

IV.

EXPERIMENTS AND RESULTS

Fig. 3: Row Mean of an image

1-D DCT / WALSH / HAAR is then applied on this Row
mean of an image to its feature vector and Euclidean distance is
used as measure of similarity to identify speaker. Detail
algorithm is given below:

Step 1: For each trainee image in the database, resize an

image to size 256*256.
Step 2: Calculate Row Mean of an image as shown in

Fig. 3.
Step 3: Apply 1-D transformation technique (DCT /

WALSH / HAAR) on Row Mean obtained in

Step 2. This gives the feature vector of an image.
Step 4: For each test image in the database, resize an

image to size 256*256.
Step 5: Apply 1-D transformation technique (DCT /

WALSH / HAAR) on Row Mean obtained in

Step 4. This gives the feature vector of an image.
Step 6: Calculate the Euclidean distance of each test

image with each trainee image corresponding to

the same sentence.
Step 7: Select the trainee image which has smallest

Euclidean distance with the test image and

declare the speaker corresponding to this trainee

image as the identified speaker.
For Row Mean of image blocks, first divide the image into
equal and non-overlapping blocks (of size 128*128, 64*64,
32*32, 16*16 and 8*8). Obtain Row Mean of each block as
shown in Fig. 3. Transformation technique is then applied on
Row Mean of each block and then combined into columns to
get feature vector of an image.

Implementation for the proposed approach was done on
Intel Core 2 Duo Processor, 2.0 GHz, and 3 GB of RAM.
Operating System used is Windows XP and softwares used are
MATLAB 7.0 and Sound forge 8.0. To study the proposed
approach we recorded six distinct sentences from 30 speakers:
11 males and 19 females. These sentences are taken from
VidTIMIT database [13] and ELSDSR database [14]. For every
speaker 10 occurrences of each sentence were recorded.
Recording was done at varying times. This forms the closed set
for our experiment. From these speech samples spectrograms
were created with window size 256 and overlap of 128. Before
creation of spectrograms, DC offset present in speech samples
was removed so that signals are vertically centered at 0. After
removal of DC offset, speech samples were normalized with
respect to amplitude to -3 dB and also with respect to time.
Spectrograms generated from these speech samples form the
image database for our experiment. In all we had 1800
spectrograms in our database.

From these spectrograms, two sets were created.

Set A: Contains six spectrograms as trainee images per
speaker and four spectrograms as test images per speaker. So in
all it contains 1080 trainee images and 720 test images.

Set B: Contains eight spectrograms as trainee images per
speaker and two spectrograms as test images per speaker. So in
all it contains 1440 trainee images and 360 test images.

Since our work is restricted to text dependent approach,
Euclidean distance for a test image of speaker say 'x' for a
particular sentence say 'si' is obtained by comparing the
feature vector of that test image with the feature vectors of all
the trainee images corresponding to sentence 'si'. Results are
calculated for set of test images corresponding to each
sentence.

A Results for DCT/WALSH/HAAR on Full image:

1) Results for DCT on full image

Table I shows the identification rate for six sentences si to
s6 when DCT is applied on full image in set A and partial
feature vectors are selected to find the matching spectrogram.

Table I. Identification rate for sentences si to s6 for full and

PARTIAL FEATURE VECTOR WHEN DCT IS APPLIED TO FULL IMAGE IN SET A

Portion of feature
vector selected

Sentence

256*256

54.16

59.16

56.66

68.33

62.50

192*192

58.33

67.5

73.33

69.16

128*128

65.83

64.16

71.66

67.5

74.16

72.5

64*64

70.83

71.66

72.50

77.50

75.83

32*32

73.33

74.16

77.5

20*20

78.33

75.33

78.33

71.66

81.66

16*16

72.5

76.66

74.16

76.66

79.16

Table II shows the identification rate for six sentences si to
s6 when DCT is applied on full image in set B and partial
feature vectors are selected to find the matching spectrogram.

189

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

Dr. H. B. Kekre et. al.(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, 2010

table ii. Identification rate for sentences sIto s6for varying
portion of feature vector when dctls applied to full image in set b

Portion of feature
vector selected

Sentence

256*256

63.33

66.67

76.67

192*192

73.33

76.67

78.33

128*128

78.33

73.33

78.33

81.67

64*64

78.33

86.67

83.33

88.33

32*32

86.67

20*20

86.67

88.33

16*16

86.67

91.67

Table III shows the comparison of overall identification
rate considering all sentences, for partial feature vectors of
different sizes when set A and set B is used. It also shows the
number of DCT coefficients used for identifying speaker for
corresponding selected portion of feature vector.

table in. Comparison of Overall Identification rate for

DIFFERENT NUMBER OF DCT COEFFICIENTS WHEN DCT IS APPLIED TO FULL
IMAGE IN SET A AND SET B

Portion of feature
vector selected

Number of DCT
coefficients

% Identification rate

Set A

SetB

256*256

65536

70.83

192*192

36864

66.38

75.27

128*128

16384

69.30

78.88

64*64

4096

73.19

82.77

32*32

1024

75.83

87.77

20*20

400

77.63

88.05

16*16

256

76.66

87.5

Table v. Identification rate for sentences s 1 to s 6 for full and

partial feature vector when walsh transform is applied to full

image from set b

Portion of feature
vector selected

Sentence

256*256

63.33

66.67

76.67

192*192

71.67

76.67

73.33

78.33

81.67

128*128

78.33

83.33

81.67

64*64

86.67

83.33

81.67

83.33

32*32

86.67

81.67

88.33

83.33

91.67

20*20

91.67

78.33

83.33

86.67

83.33

16*16

86.67

83.33

86.67

Table VI shows the overall identification rate considering
all sentences, for partial feature vectors. For set A, highest
identification rate is obtained for partial feature vector of size
64*64 i.e. 4096 WALSH coefficients. For set B, it requires
32*32 partial feature vector i.e. 1024 WALSH coefficients.

Table vi. Comparison of overall identification rate for varying

number of coefficients when walsh transform is applied to full

image from set a and set b

Portion of feature
vector selected

Number of Walsh
coefficients

% Identification rate

Set A

SetB

256*256

65536

70.83

192*192

36864

66.66

76.11

128*128

16384

70.69

64*64

4096

84.16

32*32

1024

73.33

85.55

20*20

400

72.91

84.72

16*16

256

71.94

2) Results for Walsh on full image

Results of Walsh transform on Spectrograms are tabulated
below. Table IV shows the identification rate for sentences si
to s6 for full and partial feature vectors when WALSH
transform is applied on full image and set A is used.

Table IV Identification rate for sentences sIto s6for varying

PORTION OF FEATURE VECTOR WHEN WALSH TRANSFORM IS APPLIED TO
FULL IMAGE FROM SET A

Portion of feature
vector selected

Sentence

256*256

54.16

59.16

57.5

68.33

63.33

192*192

59.16

66.66

65.83

63.33

73.33

71.66

128*128

65.83

66.66

70.83

73.33

72.5

64*64

74.16

73.33

76.66

75.83

32*32

70.83

71.66

70.83

78.33

76.66

20*20

70.83

69.67

71.67

70.83

76.67

78.33

16*16

70.83

71.67

66.67

77.5

Table V shows sentencewise identification rate for
WALSH transform on full image from set B. It can be
observed from Table IV and Table V that, identification rate
for each sentence is increased as more training is provided to
the system. From both the tables, it can be seen that as size of
the partial feature vector is decreased, the identification rate
also decreases, achieves its peak value and then again decrease.

3) Results for HAAR on full image

Table VII shows sentencewise identification rate when 2-D
HAAR transform is applied to full image with size 256*256
and partial feature vectors are selected from these feature
vectors. These results are for set A.

Table vii. Identification rate for sentences s1tos6for varying

portion of feature vector when haar transform is applied to full

image from seta

Portion of feature
vector selected

Sentence

256*256

54.16

59.16

57.5

68.33

63.33

192*192

59.16

66.66

65.83

63.33

73.33

71.66

128*128

65.83

66.66

70.83

73.33

72.5

64*64

74.16

73.33

76.66

75.83

32*32

70.83

71.66

70.83

78.33

76.66

20*20

65.83

73.33

71.67

76.67

16*16

70.83

71.67

66.67

77.5

Table VIII shows identification rate for HAAR transform
on full image when set B is used. From both the tables, it can
be seen that as the number of coefficients selected from the
feature vector is decreased, the identification rate also
decreases, achieves its peak value and then again decrease.
From the Table VII and Table VIII, it can also be noted that
when more training is provided to the system, identification
rate per sentence is increased.

190

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

Dr. H. B. Kekre et. al.(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, 2010

Table viii. Identification rate for sentences sIto s6for varying

portion of feature vector when haar transform is applied to full

image with training set of eight images for each speaker

Portion of feature
vector selected

Sentence

256*256

63.33

66.67

76.67

192*192

73.33

78.33

76.67

78.33

128*128

78.33

83.33

81.67

64*64

86.67

83.33

81.67

83.33

32*32

86.67

81.67

88.33

83.33

91.67

20*20

86.67

88.33

86.67

16*16

86.67

83.33

86.67

Table IX shows identification rate obtained by considering
all six sentences, for set A and set B, with different sized partial
feature vectors. Maximum identification rate is observed for
4096 and 400 HAAR coefficients with set A and set B
respectively.

Table ix. Comparison of overall identification rate for varying

number of coefficients when haar transform is applied to full

image from set a and set b

Portion of feature
vector selected

Number of HAAR
coefficients

Identification rate (%)

Set A

SetB

256*256

65536

70.83

192*192

36864

67.91

77.5

128*128

16384

70.69

64*64

4096

84.16

32*32

1024

73.33

85.55

20*20

400

72.08

86.39

16*16

256

71.94

Table X shows the comparison of identification rates for all
three transformation techniques on full image when set A and
set B are used per speaker.

increases, reaches its maximum value and then again decreases
or remains constant.

Table xi. Identification rate for sentences sIto s6for full and
partial feature vector using dcton image blocks for images from

SETA

Portion of feature
vector selected

Sentence

128*512

54.16

59.16

57.5

68.33

63.33

96*384

63.33

65.33

73.33

68.33

64*256

70.83

66.66

74.16

71.16

32*128

70.83

71.66

76.66

16*64

75.83

74.16

75.83

81.66

77.5

8*32

69.16

76.66

75.83

Table xii. Identification rate for sentences si to s6for full and

PARTIAL FEATURE VECTOR USING DCTON IMAGE BLOCKS FOR IMAGES FROM

SET B

Portion of feature
vector selected

Sentence

128*512

63.33

66.67

76.67

96*384

71.67

76.67

78.33

64*256

78.33

73.33

76.67

81.67

32*128

78.33

86.67

83.33

86.67

16*64

88.33

86.67

88.33

8*32

88.33

86.67

Table XIII shows the comparison of overall identification
rate considering all sentences, for partial feature vectors using
DCT on image blocks. For both the training sets, maximum
identification rate is achieved for partial feature of size 16*64
i.e. for 1024 DCT coefficients.

Table xiii. Comparison of overall identification rate for full and

partial feature vector portion using dcton image blocks for

images from set a and set b

Table x.

Comparison of identification rates when DCT, WALSH

AND HAARON FULL IMAGE FROM SET A AND SET B

Portion of
feature
vector
selected

Identification rate (%)
when set A is used

Identification rate (%)
when set B is used

DCT

WALSH

HAAR

DCT

WALSH

HAAR

256*256

70.83

192*192

66.38

66.66

67.91

75.27

76.11

77.5

128*128

69.30

70.69

78.88

64*64

73.19

82.77

84.16

32*32

75.83

73.33

87.77

85.55

20*20

77.63

72.91

72.08

88.05

84.72

86.39

16*16

76.66

71.94

87.5

B. Results for DCT/WALSH/HAAR on image block:

1) Results for DCT on image blocks:

Table XI shows the identification rate for sentences si to s6
when full and partial feature vectors are selected to identify
speaker using DCT on image blocks using set A. Table XII
shows the sentence wise identification rate for full and partial
feature vectors when DCT is applied on image blocks for set B.
It can be seen from the table that identification rate is improved
when more training is provided to the system. From both the
tables, it can be seen that, as the number of coefficients used
for identification purpose decreases, the identification rate

Portion of feature
vector selected

Number of DCT
coefficients

Identification rate (%)

Set A

SetB

128*512

65536

70.83

96*384

36864

65.97

64*256

16384

68.88

78.61

32*128

4096

72.63

82.22

16*64

1024

76.66

88.33

8*32

256

74.58

86.67

2) Results for WALSH on image blocks:

Table IVX on the next page shows the sentencewise
identification rate when WALSH transform is applied to image
blocks using images in Set A.

Table IVX. Identification rate for sentences sIto s6for varying

PORTION OF FEATURE VECTOR USING WALSHON IMAGE BLOCKS WITH
IMAGES FROM SET A

Portion of feature
vector selected

Sentence

128*512

54.16

59.17

57.5

68.33

63.33

96*384

59.17

66.67

65.83

63.33

73.33

71.67

64*256

65.83

66.67

70.83

73.33

72.5

32*128

74.17

73.33

76.67

75.83

16*64

70.83

71.67

70.83

78.33

76.67

8*32

70.83

71.67

66.67

77.5

191

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

Dr. H. B. Kekre et. al.(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, 2010

Table XV show the sentencewise identification rate when
WALSH transform is applied to image blocks using Set B.
From Table IVX and Table XV, it can be seen that, as the
number of coefficients used for identification purpose
decreases, the identification rate increases, reaches its
maximum value and then again decreases or remains constant.
Table XVI summarizes overall identification rate for both
training sets for various partial feature vectors.

Table xv. Identification rate for sentences sIto s6for varying

PORTION OF FEATURE VECTOR USING WALSHON IMAGE BLOCKS WITH
IMAGES FROM SET B

Portion of feature
vector selected

Sentence

128*512

63.33

66.67

76.67

96*384

71.67

76.67

73.33

78.33

81.67

64*256

78.33

83.33

81.67

32*128

86.67

83.33

81.67

83.33

16*64

86.67

81.67

88.33

83.33

91.67

8*32

86.67

83.33

86.67

Table XVI. Comparison of overall identification rate for varying

SIZE OF FEATURE VECTOR PORTION USING WALSHON IMAGE BLOCKS FOR
IMAGES IN SET A AND SET B

Portion of

feature vector

selected

Number of

WALSH

coefficients

Identification rate (%)

Set A

SetB

128*512

65536

70.83

96*384

36864

66.67

76.11

64*256

16384

70.69

32*128

4096

84.16

16*64

1024

73.33

85.55

8*32

256

71.94

It can be observed from Table XVI that the maximum
identification rate in case of Set A is obtained for 4096
WALSH coefficients i.e. for partial feature vector of size
32*128. The maximum identification rate in case of Set B is
obtained for 1024 WALSH coefficients i.e. for partial feature
vector of size 16*64.

used for identification purpose decreases, the identification rate
increases, reaches some peak value and then again decreases or
remains constant.

Table xviii. Identification rate for sentences s 1 to s 6 for full and
partial feature vector using haar on image blocks using set b

Portion of

feature vector

selected

Sentence

128*512

63.33

66.67

76.67

96*384

73.33

78.33

76.67

78.33

81.67

64*256

78.33

83.33

81.67

32*128

86.67

83.33

81.67

83.33

16*64

86.67

81.67

88.33

83.33

91.67

8*32

86.67

83.33

86.67

Table XIX shows overall identification rate for both
training sets obtained by considering the identification rate for
each sentence for various partial feature vectors. For Set A, the
maximum identification rate of 75.27% is obtained for 32*128
feature vector. Whereas, for Set B, the maximum identification
rate of 85.55% is obtained for 16*64 feature vector. Table XX
shows comparison of overall identification rates for all three
transformation techniques when applied on image blocks for
Set A and SetB.

Table XIX. Comparison of overall identification rate for varying

SIZE OF FEATURE VECTOR PORTION USING HAARON IMAGE BLOCKS USING
SETA AND SET B

Portion of feature
vector selected

Number of HAAR
coefficients

Identification rate (%)

Set A

SetB

128*512

65536

59.86

70.83

96*384

36864

65.97

76.39

64*256

16384

70.69

32*128

4096

75.27

84.44

16*64

1024

73.33

85.55

8*32

256

71.94

85.27

Table xx. Comparison of identification rates w hen dct, walsh
and haar are applied on image blocks for images in set a and set b

3) Results for HAAR on image blocks:

Table XVII shows identification rate for each sentence
when 2-D HAAR transform is applied on image blocks
obtained by dividing an image into four equal and non-
overlapping blocks as shown in Fig. 2. These results are for
training Set A.

Table xvii. Identification rate for sentences sIto s6for varying
portion of feature vector using haaron image blocks using set a

Portion of feature
vector selected

Sentence

128*512

53.33

59.17

57.5

68.33

63.33

96*384

58.33

61.67

65.83

68.33

72.5

69.17

64*256

65.83

66.67

70.83

73.33

72.5

32*128

74.17

73.33

76.67

75.83

16*64

70.83

71.67

70.83

78.33

76.67

8*32

70.83

71.67

66.67

77.5

Table XVIII shows results when Set B is used and 2-D
HAAR transform is applied on image blocks. For both the
training sets, it is observed that, as the number of coefficients

Portion of
feature
vector
selected

Identification rate (%)
When Set A is used

Identification rate (%)
When Set B is used

DCT

WALSH

HAAR

DCT

WALSH

HAAR

128*512

59.86

70.83

96*384

65.97

66.67

65.97

76.11

76.39

64*256

68.88

70.69

78.61

32*128

72.63

75.27

82.22

84.16

84.44

16*64

76.66

73.33

88.33

85.55

8*32

74.58

71.94

86.67

85.27

C. Results for DCT/ WALSH/ HAAR on Row Mean of an
image and Row Mean of image blocks :

1) Results for DCT on Row Mean of an image :

Table XXI shows sentence wise results obtained for Set A
when DCT of Row Mean is taken by dividing an image into
different number of non-overlapping blocks.

192

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

Dr. H. B. Kekre et. al.(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, 2010

Table xxi. Identification rate for sentences si to s6for DCT on

ROW MEAN OF AN IMAGE WHEN IMAGE IS DIVIDED INTO DIFFERENT NUMBER
OF NON-OVERLAPPING BLOCKS USING SET A

No. of blocks for
image split

Sentence

Full image
(256*256)

57.5

66.66

64.16

60.83

62.5

4 Blocks
(128*128)

60.83

70.83

63.33

65.83

16 Blocks
(64*64)

69.16

75.83

70.83

65.83

73.33

71.66

64 Blocks
(32*32)

76.66

75.83

78.83

75.83

256 Blocks
(16*16)

76.66

75.83

72.5

82.5

1024 Blocks
(8*8)

74.16

72.5

80.83

78.33

It can be seen from the Table XXI that, as the block size
chosen for calculating Row Mean reduces, better identification
rate is achieved. For block size 16*16, maximum identification
rate is obtained and then it decreases again.

Table XXII shows the identification rate for sentence si to
s6 and Set B of images. It can be seen from the Table XXII
that, as the block size chosen for calculating Row Mean
reduces, better identification rate is achieved. For block size
16*16, maximum identification rate is obtained and then it
decreases again.

Table xxii. Identification rate for sentences sIto s6for DCT on

ROW MEAN OF AN IMAGE WHEN IMAGE IS DIVIDED INTO DIFFERENT NUMBER
OF NON-OVERLAPPING BLOCKS USING SET B

No. of blocks for image
split

Sentence

Full image (256*256)

73.33

76.67

78.33

76.67

4 Blocks (128*128)

78.33

81.67

16 Blocks (64*64)

91.67

81.67

83.33

81.67

83.33

64 Blocks (32*32)

91.67

86.67

88.33

256 Blocks (16*16)

91.67

88.33

91.67

1024 Blocks (8*8)

88.33

83.33

86.67

88.33

The overall identification rates for both the sets, when DCT
of Row Mean is taken by dividing an image into different
number of non-overlapping blocks are tabulated in Table
XXIII.

Table xxiii. Comparison of overall identification rate for DCT on

ROW MEAN OF AN IMAGE WHEN IMAGE IS DIVIDED INTO DIFFERENT NUMBER
OF NON-OVERLAPPING BLOCKS WITH SET A AND SET B

No. of blocks for image
split

Number of

DCT
coefficients

Identification rate

(%)

For Set A

For Set B

Full image (256*256)

256

62.08

76.67

4 Blocks 128*128)

512

66.11

80.27

16 Blocks (64*64)

1024

71.11

84.17

64 Blocks (32*32)

2048

75.27

87.5

256 Blocks 16*16)

4096

77.08

89.17

1024 Blocks (8*8)

8192

75.55

86.11

2) Results forWALSH on Row Mean of an image :

Table IVXX shows the sentence wise identification rate
when Walsh transform is applied to Row Mean of an image

when it is divided into different number of non-overlapping
and Set A is used. Table XXV shows the sentence wise
identification rate when Walsh transform is applied to Row
Mean of an image when it is divided into different number of
non-overlapping and Set B.

Table ivxx. Identification rate for sentences si to s6 for Walsh

TRANSFORM ON ROW MEAN OF AN IMAGE WHEN IMAGE IS DIVIDED INTO
DIFFERENT NUMBER OF NON-OVERLAPPING BLOCKS WITH SET A

No. of blocks for image
split

Sentence

Full image (256*256)

57.5

66.66

64.16

60.83

62.5

4 Blocks (128*128)

60.83

70.83

63.33

65.83

16 Blocks (64*64)

69.16

75.83

70.83

65.83

73.33

71.66

64 Blocks (32*32)

76.66

75.83

78.83

75.83

256 Blocks (16*16)

76.66

75.83

72.5

82.5

1024 Blocks (8*8)

74.16

72.5

80.83

78.33

Table xxv. Identification rate for sentences si to s6 for Walsh

TRANSFORM ON ROW MEAN OF AN IMAGE WHEN IMAGE IS DIVIDED INTO
DIFFERENT NUMBER OF NON-OVERLAPPING BLOCKS WITH SET B

No. of blocks for image
split

Sentence

Full image (256*256)

73.33

76.66

78.33

76.66

4 Blocks (128*128)

78.33

81.67

16 Blocks (64*64)

91.67

81.67

83.33

81.67

83.33

64 Blocks (32*32)

91.67

86.66

88.33

256 Blocks (16*16)

91.67

88.33

91.67

1024 Blocks (8*8)

88.33

83.33

86.66

88.33

It can be seen from the Table IVXX and Table XXV that,
as the block size chosen for calculating Row Mean reduces,
better identification rate is achieved.

Table XXVI summarizes overall identification rate for both
training sets by considering all six sentences. For block size
16*16, maximum identification rate is obtained and then it
decreases again.

Table xxvi. Comparison of overall identification rate for Walsh
transform on row mean of an image when image is divided into
different number of non-overlapping blocks for set a and set b

No. of blocks for
image split

Number of Walsh
coefficients

Identification rate

(%)

Set A

SetB

Full image (256*256)

256

62.08

76.67

4 Blocks (128*128)

512

66.11

80.27

16 Blocks (64*64)

1024

71.11

84.17

64 Blocks (32*32)

2048

75.27

87.5

256 Blocks (16*16)

4096

77.08

89.17

1024 Blocks (8*8)

8192

75.55

86.11

3) Results for HAAR on Row Mean of an image :

Table XXVII shows identification rate for each sentence
when 1-D HAAR transform is applied to Row Mean of an
256*256 image and when image is divided into different
number of non-overlapping blocks for Set A. Similarly, Table
XXVIII shows identification rate for each sentence when 1-D
HAAR transform is applied to Row Mean of an 256*256 image

193

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

Dr. H. B. Kekre et. al.(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, 2010

and when image is divided into different number of non-
overlapping blocks for Set B.

Table xxvii. Identification rate for sentences s 1 to s6 for
haar transform on row mean of an image when image is divided
into different number of non-overlapping blocks using set a

No. of blocks for image
split

Sentence

Full image (256*256)

57.5

66.66

64.16

60.83

62.5

4 Blocks (128*128)

60.83

70.83

63.33

65.83

16 Blocks (64*64)

69.16

75.83

70.83

65.83

73.33

71.66

64 Blocks (32*32)

76.66

75.83

78.83

75.83

256 Blocks (16*16)

76.66

75.83

72.5

82.5

1024 Blocks (8*8)

74.16

72.5

80.83

78.33

Table xxviii. Identification rate for sentences sIto s6for
haar transform on row mean of an image when image is divided
into different number of non-overlapping blocks using set b

No. of blocks for
image split

Sentence

Full image
(256*256)

73.33

76.67

78.33

76.67

4 Blocks
(128*128)

78.33

81.67

16 Blocks
(64*64)

91.67

81.67

83.33

81.67

83.33

64 Blocks
(32*32)

91.67

86.67

88.33

256 Blocks
(16*16)

91.67

88.33

91.67

1024 Blocks
(8*8)

88.33

83.33

86.67

88.33

Table XXIX shows overall identification rate for the two
training sets when 1-D HAAR transform is applied to an image
divided into different number of equal and non-overlapping
blocks.

Table xxix. Comparison of overall identification rate for
haar transform on row mean of an image when image is divided
into different number of non-overlapping blocks for set a and b.

No. of blocks for
image split

Number of

HAAR
coefficients

Identification rate (%)

For Set A

For Set B

Full image
(256*256)

256

62.08

76.67

4 Blocks
(128*128)

512

66.11

80.27

16 Blocks
(64*64)

1024

71.11

84.17

64 Blocks
(32*32)

2048

75.27

87.5

256 Blocks
(16*16)

4096

77.08

89.17

1024 Blocks
(8*8)

8192

75.55

86.11

Overall identification rate for DCT, WALSH and HAAR
on Row Mean of an image and image blocks are summarized in
the Table XXX.

Table xxx. Comparison of DCT, WALSH and HAAR on Row

Mean of image and image blocks

No. of

blocks for

image

split

Identification rate (%)
when Set A is used

Identification rate (%)
when Set B is used

DCT

WALSH

HAAR

DCT

WALSH

HAAR

Full image
(256*256)

62.08

76.67

4 Blocks
(128*128)

66.11

80.27

16 Blocks
(64*64)

71.11

84.17

64 Blocks
(32*32)

75.27

87.5

256
Blocks
(16*16)

77.08

89.17

1024
Blocks
(8*8)

75.55

86.11

V. COMPLEXITY ANALYSIS

Complexity analysis of DCT, WALSH and HAAR on full
image:

For 2-D DCT on N*N image, 2N 3 multiplications are
required and 2N 2 (N-1) additions are required. For 2-D WALSH
on N*N image, 2N (N-l) additions are required. For 2-D
HAAR transform on N*N image where N=2m, number of
multiplications required are 2(m+l)N 2 and number of additions
required are 2mN 2 . Table XXXI summarizes these details
along with actual values of mathematical computations needed
for processing of 256*256 images.

Table xxxi. Comparison between DCT, WALSHand HAAR

WITH RESPECT TO MATHEMATICAL COMPUTATIONS AND IDENTIFICATION RATE
WHEN APPLIED ON FULL IMAGE

Parameter

Algorithm

DCT on full
image(N*N)

WALSH on full
image(N*N)

HAAR on
full image

(N*N)

Number of
Multiplications

2N 3

2(m+l)N 2

N=256

33554432

1179648

Number of
Additions

2N 2 (N-1)

2mN 2

N=256

33423360

1048576

Identification rate
(%) for Set A

77.63

Identification rate
(%) for Set B

88.05

85.55

86.39

From the above table, it can be seen that DCT on full image
gives the highest identification rate for both the training sets as
compared to WALSH and HAAR on full image. However this
outstanding performance is achieved at the expense of higher
computations.

Number of multiplications required by DCT on full image
is approximately 28 times more than the number of
multiplications required by HAAR on full image. Whereas
number of additions required by DCT on full image is
approximately 31 times more than the number of additions
required by HAAR on full image.

194

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

Dr. H. B. Kekre et. al.(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, 2010

Though WALSH on full image does not require any
multiplications, overall CPU time taken by it is more than that
of HAAR on full image. This is because the number of
additions taken by WALSH on full image is approximately 31
times more than the number of additions required by HAAR on
full image.

B. Complexity analysis ofDCT, WALSH and HAAR on
image blocks:

The number of multiplications required in case of 2-D DCT
on image blocks is N 3 and the number of additions required are

N 2 (N-2).

For 2-D WALSH on four image blocks of size N/2*N/2,
number of additions required are N 2 (N-2).

The number of multiplications required for 2-D HAAR on
image blocks is 2mN 2 . Similarly number of additions required
for 2-D HAAR on image blocks is 2(m-l)N 2 . Table XXXII
summarizes these details along with actual values of
mathematical computations needed for processing of 256*256
images.

Table xxxii Comparison between DCT, WALSH and HAAR with

RESPECT TO MATHEMATICAL COMPUTATIONS AND IDENTIFICATION RATE
WHEN APPLIED ON IMAGE BLOCKS

Parameter

Algorithm

DCTon
image blocks

(N/2*N/2)

Walsh on
image
blocks

(N/2*N/2)

HAAR on
image
blocks

(N/2*N/2)

Number of
Multiplications

N 3

2(m+l)N 2

N=256, four blocks

16777216

1048576

Number of Additions

N 2 (N-2)

2(m-l)N 2

N=256, four blocks

16646144

917504

Identification rate (%)
for Set A

76.66

75.27

Identification rate (%)
for Set B

88.33

85.55

From the Table XXXII, it can be seen that, in all the three
transformation techniques on image blocks, DCT on image
blocks gives best identification rate for both the training sets.
But this performance is achieved at the expense of higher
number of computations. DCT on image blocks takes 16 times
more multiplications and 18 times more additions than HAAR
on image blocks. Though WALSH transform does not need
any multiplications, still it takes more number of computations
than HAAR. This is because WALSH on image blocks requires
approximately 18 times more additions than HAAR on image
blocks.

C. Complexity Analysis of DCT, Walsh and HAAR on Row
Mean of an image and on Row Mean of image blocks:

Since Row Mean of an image is a one dimensional vector,
only 1-D DCT, WALSH and HAAR need to be applied on
Row Mean. This itself reduces the number of multiplications
and additions required for feature vector calculation. Row
Mean of an image of size N*N is a vector of size N*l. For 1-D
DCT on this N*l vector, N 2 multiplications and N(N-l)

additions are required. One dimensional Walsh on Row Mean
of an image takes N(N-l) additions and no multiplications.
Whereas, 1-D HAAR on Row Mean of an image of size N*N
requires (m+l)N multiplications and mN additions where
N=2 m . Following Table XXXIII summarizes this statistics in
case of each transformation technique applied for the Row
Mean of block size 16*16 which gives highest identification
rate.

Table xxxiii. Comparison between DCT, WALSH and HAAR

WITH RESPECT TO MATHEMATICAL COMPUTATIONS AND IDENTIFICATION RATE
WHEN APPLIED ON ROW MEAN OF AN IMAGE

Parameter

Algorithm

DCTon

Row Mean

of image

(N*l)

Walsh on

Row

Mean of

image

(N*l)

HAAR on

Row Mean

of image

(N*l)

Number of
Multiplications

N 2

(m+l)N

N=16, 256 blocks

65536

20480

Number of Additions

N(N-l)

N=16, 256 blocks

61440

16384

Identification rate (%)
for Set A

77.08

Identification rate (%)
for Set B

89.17

From the Table XXXIII, we can see that all three
transformation techniques result in same identification rate
when applied on the Row Mean of an image and on Row Mean
of an image blocks. For both the training sets, highest
identification rate is obtained when image is divided into 16*16
size blocks. However, in terms of computations, HAAR
transform is proved to be better one. Number of multiplications
required by HAAR is approximately three times less than the
number of multiplications required in case of DCT on Row
Mean. Also the number of additions required by HAAR is 3.5
times less than the number of additions required by DCT and
WALSH on Row Mean of an image.

Along with the different approaches of applying
transformation techniques on spectrograms, comparative study
of computational complexity of three transformation
techniques for each approach has been done and is presented
below.

D. Complexity Analysis of DCT transform on Full, Block and
Row Mean of Spectrograms:

For 2-D DCT on N*N image, 2N 3 multiplications are
required and 2N 2 (N-1) additions are required. For 2-D DCT on
four blocks of size N/2*N/2, N 3 multiplications are required
and N 2 (N-2) additions are required. For 1-D DCT on N*l
image, N 2 multiplications are needed and N(N-l) additions are
needed. These computational details are summarized in Table
XXXIV along with the actual number of computations for
256*256 image using three methods of applying DCT.

195

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

Dr. H. B. Kekre et. al.(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, 2010

Table xxxiv. Computational details for 2-DDCTon N*N

IMAGE, 2-D DCT ON N/2*N/2 IMAGE AND 1-DCT ON N*l IMAGE
RESPECTIVELY

Parameter — >

No. of
Multiplications

No. of Additions

Algorithm j

2-D DCT on N*N image

2N 3

2N 2 (N-1)

2-D DCT on 256*256
image

33554832

33424159

2-D DCT on four blocks of
size N/2*N/2 each

N 3

N 2 (N-2)

2-D DCT on four blocks of
size 256/2*256/2 each

16778240

16648191

ID DCT on N*l Row
Mean of N*N image

N 2

N(N-l)

ID DCT on N*l Row
Mean of 256*1 image

69632

69631

When all three methods of applying DCT are compared, it
has been observed that though number of coefficients used in
Row Mean method is higher, number of multiplications and
additions reduce drastically as compared to other two methods.
Number of multiplications in DCT on full image method is 480
times more than the number of multiplications in Row Mean
method whereas for DCT on image blocks it is 241 times more.
Number of additions needed in DCT on full image and DCT on
image blocks are also 480 times and 239 times more than the
additions required in Row mean method respectively. For the
Set A, the identification rate for DCT on Row Mean is almost
same as identification rate for DCT on full image. In case of
Set B, DCT on Row Mean gives better identification rate as
compared to DCT on full image and DCT on image blocks and
that too with reduced number of mathematical computations.

E. Complexity Analysis of WALSH transform on Full, Block
and Row Mean of Spectrograms:

For 2-D WALSH on N*N image, 2N 2 (N-1) additions are
required. For 2-D WALSH on four blocks of size N/2*N/2,
N 2 (N-1) additions are required. Whereas for 1-D WALSH on
N*l image, N(N-l) additions are needed as shown in table
XXXV. In all three cases number of multiplications required is
zero.

Table xxxv. Computational details for 2-D WALSHon N*N
IMAGE, 2-D WALSH ON N/2*N/2 IMAGE AND 1-D WALSH ON N*l IMAGE
RESPECTIVELY

Parameter — >

No. of
Multiplications

No. of
Additions

Algorithm J,

2-D WALSH on N*N image

2N 2 (N-1)

2-D WALSH on 256*256
image

33423360

2-D WALSH on four blocks of
size N/2*N/2 each

N 2 (N-2)

2-D WALSH on four blocks of
size 256/2*256/2 each

16646144

1-D WALSH on N*l size Row
Mean vector of image N*N

N(N-l)

1-D WALSH on 256*1 size

Row Mean vector of image

256*1

65280

WALSH transform is applied to Row Mean of an image. Also
the number of additions required when WALSH transform is
applied on image blocks is 255 times more than the number of
additions required when WALSH transform is applied to Row
Mean of an image. Thus number of additions is drastically
reduced for Walsh transform on Row Mean of an image.

F. Complexity Analysis ofHAAR transform on Full, Block
and Row Mean of Spectrograms:

For 2-D HAAR transform on N*N image where N=2 m ,
number of multiplications required are 2(m+l)N 2 and number
of additions required are 2mN 2 . For 2-D HAAR transform on
four blocks of size N/2*N/2 each, 2mN 2 multiplications and
2(m-l)N 2 additions are needed. Whereas for 1-D HAAR
transform on N*l image, number of multiplications required
are (m+l)N and number of additions are mN as shown in table
XXXVI.

Table xxxvi. Computational details for 2-DHAARon N*N
IMAGE, 2-D HAAR ON N/2*N/2 IMAGE AND 1-D HAAR ON N*l IMAGE
RESPECTIVELY

Parameter — >

No. of
Multiplications

No. of Additions

Algorithm [

2-D HAAR on N*N image

2(m+l)N 2

2mN 2

2-D HAAR on 256*256
image

1179648

1048576

2-D HAAR on four blocks

of size N/2 *N/2 each

2mN 2

2(m-l)N 2

2-D HAAR on four blocks
of size 256/2*256/2 each

1048576

917504

1-D HAAR on N*l image

(m+l)N

1-D HAAR on 256*1 size

Row Mean vector of

image 256*1

2304

2048

From table 7.5 it can be seen that number of additions
required when WALSH transform is applied on full image is
512 times more than the number of additions required when

From Table XXXVI, it can be seen that number of
multiplications required when HAAR transform is applied on
full image is 512 times more than the number of
multiplications required when HAAR transform is applied to
Row Mean of an image. Also the number of multiplications
required when HAAR transform is applied on image blocks is
455 times more than the number of multiplications required
when HAAR transform is applied to Row Mean of an image.
Thus number of multiplications is drastically reduced for
HAAR transform on Row Mean of an image. Number of
additions required is also reduced to a greater extent when
transformation technique is applied on Row Mean of an image.
Number of additions required when HAAR transform is
applied on full image is 512 times more than the number of
additions required when HAAR transform is applied to Row
Mean of an image. Also the number of additions required when
HAAR transform is applied on image blocks is 448 times more
than the number of additions required when HAAR transform
is applied to Row Mean of an image.

VI.COnclusion

In this paper, closed set text dependent speaker
identification has been considered using three different
transformation techniques namely DCT, WALSH and HAAR.
Each transformation technique is applied in three ways:

a) On full image

196

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

Dr. H. B. Kekre et. al.(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, 2010

b) On image blocks and

c) On Row Mean of an image.

For each method, two training sets were used as mentioned
earlier.

It can be clearly concluded from the results that as more
training is provided to the system, more accuracy is obtained in
the results in terms of identification rate.

Further for each method, Identification rates are obtained
for various numbers of coefficients from feature vectors of
images. It has been observed that as the number of coefficients
chosen is smaller up to a certain limit; better identification rate
is achieved in all three methods.

DCT on full image gives its best identification rate for only
20*20 portion of feature vector i.e. by using only 400 DCT
coefficients. DCT on image blocks gives highest identification
rate when 16*64 portion of its feature vector is considered
which has 1024 DCT coefficients. Finally DCT on Row Mean
gives highest identification rate for Row Mean of 16*16 size
image blocks i.e. for 4096 DCT coefficients. When these
highest identification rates in all three methods in DCT are
compared, it has been observed that DCT on image blocks
gives slightly improved results for training set of eight images
per speaker. Whereas, DCT on Row Mean, further improves
these results with drastically reduced computations. Though the
number of coefficients used in Row Mean method is higher,
overhead caused for its comparison is negligible as compared
to number of mathematical computations needed in other two
approaches.

Similarly, WALSH on Row Mean of image gives better
identification rates as compared to WALSH on full image and
WALSH on image blocks for both the training sets. These
better identification rates are obtained with the advantage of
reduced mathematical computations. For HAAR transform
also, identification rate for HAAR on Row Mean is better than
HAAR on full image and HAAR on image blocks.

From the results of DCT, WALSH and HAAR on full
image, it can be concluded that DCT on full image gives better
identification rate than WALSH and HAAR on full image but
at the expense of large number of mathematical computations.
In WALSH transform on full image, numbers of mathematical
computations required are greatly reduced as compared to DCT
since no multiplications are required in WALSH. These
computations are further reduced by use of HAAR transform
but at the slight expense of identification rate. Similar
conclusions can be drawn for DCT, WALSH and HAAR on
image blocks. So there is a trade off between better
identification rate and less CPU time for mathematical
computations.

However, in case of Row Mean approach of applying
transform, performances of all three transformation techniques
are same for a specific block size chosen for Row Mean. In that
HAAR transform proves to be better because it requires
minimum number of computations.

The overall conclusion is that Row Mean technique
requires less number of mathematical computations and hence
less CPU time for all three transformation techniques as

compared to transformation techniques on full image and on
image blocks. HAAR transform on Row Mean of an image
gives the best result with respect to identification rate as well as
number of computations required.

REFERENCES

[I] Evgeniy Gabrilovich, Alberto D. Berstin: "Speaker recognition: using a
vector quantization approach for robust text-independent speaker
identification", Technical report DSPG-95-9-001', September 1995.

[2] Tridibesh Dutta, "Text dependent speaker identification based on
spectrograms", Proceedings of Image and vision computing, pp. 238-
243, New Zealand 2007,.

[3] J.P.Campbell, "Speaker recognition: a tutorial", Proc. IEEE, vol. 85, no.
9, pp. 1437-1462, 1997.

[4] D. O'Shaughnessy, "Speech communications- Man and Machine", New
York, IEEE Press, 2nd Ed., pp. 199, pp. 437-458, 2000.

[5] S. Davis and P. Mermelstein, "Comparison of parametric representations
for monosyllabic word recognition in continuously spoken sentences,"
IEEE Transaction Acoustics Speech and Signal Processing, vol. 4, pp.
375-366, 1980.

[6] Wang Yutai, Li Bo, Jiang Xiaoqing, Liu Feng, Wang Lihao, "Speaker
Recognition Based on Dynamic MFCC Parameters", International
Conference on Image Analysis and Signal Processing, pp. 406-409, 2009

[7] Azzam Sleit, Sami Serhan, and Loai Nemir, "A histogram based speaker
identification technique", International Conference on ICADIWT, pp.
384-388, May 2008.

[8] B. S. Atal, "Automatic Recognition of speakers from their voices", Proc.
IEEE, vol. 64, pp. 460-475, 1976.

[9] Jialong He, Li Liu, and G"unther Palm, "A discriminative training
algorithm for VQ-based speaker Identification", IEEE Transactions on
speech and audio processing, vol. 7, No. 3, pp. 353-356, May 1999.

[10] Debadatta Pati, S. R. Mahadeva Prasanna, "Non-Parametric Vector
Quantization of Excitation Source Information for Speaker
Recognition", IEEE Region 10 Conference, pp. 1-4, Nov. 2008.

[II] Tridibesh Dutta and Gopal K. Basak, "Text dependent speaker
identification using similar patterns in spectrograms", PRIP'2007
Proceedings, Volume 1, pp. 87-92, Minsk, 2007.

[12] Andrew B. Watson, "Image compression using the Discrete Cosine
Transform", Mathematica journal, 4(1), pp. 81-88, 1994,.

[13] http://www.itee.uq.edu.au/~conrad/vidtimit/

[14] http://www2.imm.dtu.dkMf/elsdsr/

[15] H.B.Kekre, Sudeep D. Thepade, "Improving the Performance of Image
Retrieval using Partial Coefficients of Transformed Image",
International Journal of Information Retrieval (IJIR), Serials
Publications, Volume 2, Issue 1, pp. 72-79 (ISSN: 0974-6285), 2009.

[16] H.B.Kekre, Tanuja Sarode, Sudeep D. Thepade, "DCT Applied to Row
Mean and Column Vectors in Fingerprint Identification", In Proceedings
of International Conference on Computer Networks and Security
(ICCNS), 27-28 Sept. 2008, VIT, Pune.

[17] H.B.Kekre, Sudeep D. Thepade, Archana Athawale, Anant Shah,
Prathmesh Verlekar, Suraj Shirke,"Energy Compaction and Image
Splitting for Image Retrieval using Kekre Transform over Row and
Column Feature Vectors", International Journal of Computer Science
and Network Security (IJCSNS),Volume:10, Number 1, January 2010,
(ISSN: 1738-7906) Available at www.IJCSNS.org.

[18] H.B.Kekre, Sudeep D. Thepade, Archana Athawale, Anant Shah,
Prathmesh Verlekar, Suraj Shirke, "Performance Evaluation of Image
Retrieval using Energy Compaction and Image Tiling over DCT Row
Mean and DCT Column Mean", Springer-International Conference on
Contours of Computing Technology (Thinkquest-2010), Babasaheb
Gawde Institute of Technology, Mumbai, 13-14 March 2010, The paper
will be uploaded on online Springerlink.

[19] H.B.Kekre, Tanuja K. Sarode, Sudeep D. Thepade, Vaishali
Suryavanshi, "Improved Texture Feature Based Image Retrieval using
Kekre's Fast Codebook Generation Algorithm", Springer-International
Conference on Contours of Computing Technology (Thinkquest-2010),

197

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

Dr. H. B. Kekre et. al.(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, 2010

Babasaheb Gawde Institute of Technology, Mumbai, 13-14 March 2010,
The paper will be uploaded on online Springerlink.

[20] H. B. Kekre, Tanuja K. Sarode, Sudeep D. Thepade, "Image Retrieval
by Kekre 's Transform Applied on Each Row of Walsh Transformed VQ
Codebook", (Invited), ACM-International Conference and Workshop on
Emerging Trends in Technology (ICWET 2010),Thakur College of
Engg. And Tech., Mumbai, 26-27 Feb 2010, The paper is invited at
ICWET 2010. Also will be uploaded on online ACM Portal.

[21] H. B. Kekre, Tanuja Sarode, Sudeep D. Thepade, "Color-Texture
Feature based Image Retrieval using DCT applied on Kekre 's Median
Codebook", International Journal on Imaging (IJI), Volume 2, Number
A09, Autumn 2009,pp. 55-65. Available online at
www.ceser.res.in/iji.html (ISSN: 0974-0627).

[22] H. B. Kekre, Ms. Tanuja K. Sarode, Sudeep D. Thepade, "Image
Retrieval using Color-Texture Features from DCT on VQ Codevectors
obtained by Kekre's Fast Codebook Generation", ICGST-International
Journal on Graphics, Vision and Image Processing (GVIP), Volume 9,
Issue 5, pp.: 1-8, September 2009. Available online at http:
//www.icgst.com /gvip /Volume9 /Issue5 /P1150921752.html.

[23] H. B. Kekre, Sudeep Thepade, Akshay Maloo, "Image Retrieval using
Fractional Coefficients of Transformed Image using DCT and Walsh
Transform", International Journal of Engineering Science and
Technology, Vol.. 2, No. 4, 2010, 362-371

[24] H. B. Kekre, Sudeep Thepade, Akshay Maloo, "Performance
Comparison of Image Retrieval Using Fractional Coefficients of
Transformed Image Using DCT, Walsh, Haar and Kekre's Transform",
CSC-International Journal of Image processing (DIP), Vol.. 4, No. 2,
pp.:142-155, May 2010.

[25] H. B. Kekre, Tanuja Sarode "Two Level Vector Quantization Method
for Codebook Generation using Kekre's Proportionate Error Algorithm"
, CSC-International Journal of Image Processing, Vol.4, Issue 1, pp.l-
10, January-February 2010

[26] H. B. Kekre, Sudeep Thepade, Akshay Maloo, "Eigenvectors of
Covariance Matrix using Row Mean and Column Mean Sequences for
Face Recognition", CSC-International Journal of Biometrics and
Bioinformatics (IJBB), Volume (4): Issue (2), pp. 42-50, May 2010.

[27] H. B. Kekre, Tanuja Sarode, Shachi Natu, Prachi Natu, "Performance
Comparison Of 2-D DCT On Full/Block Spectrogram And 1-D DCT On
Row Mean Of Spectrogram For Speaker Identification", (Selected),
CSC-International Journal of Biometrics and Bioinformatics (IJBB),
Volume (4): Issue (3), pp. 100-112, August 2010, Malaysia..

[28] H. B. Kekre, Tanuja Sarode, Shachi Natu, Prachi Natu, "Performance
Comparison of Speaker Identification Using DCT, Walsh, Haar On Full
And Row Mean Of Spectrogram" , (Selected), International Journal of
Computer Applications, pp. 30-37, August 2010, USA.

[29] H. B. Kekre, Tanuja Sarode, Shachi Natu, Prachi Natu, "Speaker
Identification using 2-d DCT, Walsh and Haar on Full and Block
Spectrograms", International Journal of Computer Science and
Engineering, Volume 2, Issue 5, pp. 1733-1740, August 2010.

AUTHORS PROFILE

Dr. H. B. Kekre has received B.E. (Hons.) in Telecomm.

FEngg. from Jabalpur University in 1958,
M.Tech (Industrial Electronics) from IIT
Bombay in 1960, M.S.Engg. (Electrical Engg.)
from University of Ottawa in 1965 and Ph.D.
(System Identification) from IIT Bombay in

1970. He has worked Over 35 years as Faculty of Electrical
Engineering and then HOD Computer Science and Engg. at
IIT Bombay. For last 13 years worked as a Professor in
Department of Computer Engg. at Thadomal Shahani
Engineering College, Mumbai. He is currently Senior
Professor working with Mukesh Patel School of Technology
Management and Engineering, SVKM's NMIMS University,
Vile Parle (w), Mumbai, INDIA. He ha guided 17 Ph.D.s, 150

M.E./M.Tech Projects and several B.E./B.Tech Projects. His
areas of interest are Digital Signal processing, Image
Processing and Computer Networks. He has more than 250
papers in National / International Conferences / Journals to his
credit. Recently six students working under his guidance have
received best paper awards. Currently he is guiding ten Ph.D.
students.

Dr. Tanuja K. Sarode has received M.E. (Computer
Engineering) degree from Mumbai University
in 2004 and Ph.D. from Mukesh Patel School
of Technology, Management and Engg.
SVKM's NMIMS University, Vile-Parle (W),
Mumbai, INDIA, in 2010. She has more than
10 years of experience in teaching. Currently
working as Assistant Professor in Dept. of Computer
Engineering at Thadomal Shahani Engineering College,
Mumbai. She is member of International Association of
Engineers (IAENG) and International Association of
Computer Science and Information Technology (IACSIT).
Her areas of interest are Image Processing, Signal Processing
and Computer Graphics. She has 70 papers in National
/International Conferences/journal to her credit.

Shachi Natu has received B.E. (Computer) degree from
Mumbai University with first class in 2004.
Currently Purusing M.E. in Computer
Engineering from University of Mumbai. She
has 05 years of experience in teaching.
Currently working as Lecturer in department
' of Information Technology at Thadomal

Shahani Engineering College, Bandra (w), Mumbai. Her areas
of interest are Image Processing, Data Structure, Database
Management Systems and operating systems. She has 3 papers
in National / International Conferences /journal to her credit.

Prachi Natu has received B.E. (Electronics and
Telecommunication) degree from Mumbai
University with first class in 2004. Currently
Purusing M.E. in Computer Engineering from
University of Mumbai. She has 04 years of
experience in teaching. Currently working as
Lecturer in Computer Engineering department
at G. V. Acharya Institute of Engineering and Technology,
Shelu. Mumbai. Her areas of interest are Image Processing,
Database Management Systems and operating systems. She
has 3 papers in National / International Conferences /journal to
her credit.

198

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

A Research Proposal for Mitigating DoS Attacks

in IP-based Networks

Sakharam Lokhande

Assistant Professor

School of Computational Science,

Swami Ramanand Teerth Marathwada University, Nanded,

MS, India, 431606.

Dr. Santosh Khamitkar

Associate Professor
School of Computational Science,
Swami Ramanand Teerth Marathwada University, Nanded,
MS, India, 431606.

Parag Bhalchandra

Assistant Professor
School of Computational Science,
Swami Ramanand Teerth Marathwada University, Nanded,
MS, India, 431606.

Nilesh Deshmukh

Assistant Professor
School of Computational Science,
Swami Ramanand Teerth Marathwada University, Nanded,
MS, India, 431606.

Santosh Phulari

Assistant Professor
School of Computational Science, Swami Ramanand Teerth
Marathwada University, Nanded,
MS, India, 431606.

Ravindra Rathod

Assistant Professor
School of Computational Science, Swami Ramanand Teerth
Marathwada University, Nanded,
MS, India, 431606.

Abstract : This paper studies denial of service (DoS) attacks in
computer networks. These attacks are known for preventing
availability of network services from their legitimate users. After
careful review of literature, we wish to presents a structured view
on possible attack and defense mechanisms. An outline to
describe some new defense mechanisms is also presented in terms
of a research proposal .

Keywords- Denial of Service Attacks, Intrusion, Security

I. PROBLEM DEFINATION

Defending against DoS attacks is a task from network and
computer security. As scientific disciplines, network and
computer security are relatively primitive. An indication of
this fact is to be aware that the computer security terminology
is not yet stabilized [4]. Computer and network security
aspects were first studied in the early 1970s. As in some of the
earliest security papers listed and available in, the Denial of
Service attacks are timely and extremely important research
topic. According to the C SI/FBI computer crime and security
survey in the United States [1] for the year 2004, DoS attacks
are the second most widely detected outsider attack type in
computer networks, immediately after virus infections. A
computer crime and security survey in Australia[l] for the
year 2004, gives similar results. It is currently not possible to
prevent DoS attacks because many of these attacks are based
on using ordinary protocols and services in an overwhelming

manner. Specific security holes in the victim hosts or networks
are thus not necessarily needed. For this reason we can only
mitigate these attacks.

II. Overview of Denial of Service Attacks

Denials of Service (DoS) attacks have proved to be a
serious and permanent threat to users, organizations, and
infrastructures of the Internet [1]. The primary goal of these
attacks is to prevent access to a particular resource like a web
server [2]. A large number of defenses against DoS attacks
have been proposed in the literature, but none of them gives
reliable protection. There will always be vulnerable hosts in
the Internet to be used as sources of attack traffic. It is simply
not feasible to expect all existing hosts in the Internet to be
protected well enough. In addition, it is very difficult to
reliably recognize and filter only attack traffic without causing
any collateral damage to legitimate traffic.

A DoS attack can be carried out either as a flooding or a
logic attack. A Flooding DoS attack is based on brute force.
Real-looking but unnecessary data is sent as much as possible
to a victim. As a result, network bandwidth is wasted, disk
space is filled with unnecessary data (such as spam e-mail,
junk files, and intentional error messages), fixed size data
structures inside host software are filled with bogus
information, or processing power is spent for un useful
purposes. To amplify the effects, DoS attacks can be run in a
coordinated fashion from several sources at the same time

199

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

(Distributed DoS, DDoS).A logic DoS attack is based on an
intelligent exploitation of vulnerabilities in the target. For
example, a skillfully constructed fragmented Internet Protocol
(IP) datagram may crash a system due to a serious fault in
the operating system (OS) software. Another example of a
logic attack is to exploit missing authentication requirements
by injecting bogus routing information to prevent traffic from
reaching a victim's network. [5, 6]

There are two major reasons that make DoS attacks
attractive for attackers. The first reason is that there are
effective automatic tools available for attacking any victim, so
expertise is not necessarily required. The second reason is that
it is usually impossible to locate an attacker without extensive
human interaction or without new features in most routers of
the Internet. DoS attacks make use of vulnerabilities in end-
hosts, routers, and other systems connected to a computer
network. The size of a population having the same
vulnerability can be large. In July 2003 a vulnerability was
found from the whole population of Cisco routers and
switches running any version of the Cisco 10 S software and
configured to process Internet Protocol version 4 (IPv4)
packets. This vulnerability made it possible to block an
interface, which resulted in a DoS condition without any
alarms being triggered. Another example of a large population
is the Microsoft Windows Metafile (WMF) vulnerability
which was found in December 2005 from all versions of
Windows 98, 98SE, ME, 2000, and XP. This vulnerability
made it possible to install any malicious software on these
hosts, for example, to send DoS attack traffic. User interaction
was, however, required to exploit this vulnerability.

III. Research Problem

Mitigating DoS attacks is difficult especially due to the
following problems:

1) Very little has been done to compare, contrast, and
categorize the different ideas related to DoS attacks and
defenses. As a result it is difficult to understand what a
computer network user needs to do and why to mitigate the
threat from DoS attacks.

2) There are no effective defense mechanisms against
many important DoS attack types. There is no guidance on
how to select defense mechanisms.

3) Existing defense mechanisms have been evaluated
according to very limited criteria.

4) Often relevant risks have been ignored (such as in
[3]) or evaluations have been carried out under ideal
conditions.

5) No research publications exist for giving a
systematic list of issues related to defense evaluation

IV. Objective of the Research

The objective of this research proposal is to help any user
in any network for mitigating DoS attacks in IP-based
networks. This study concentrates especially on the following
areas:

1) One should understand existing attack mechanisms
and available defense mechanisms, and have a rough idea
about the benefits (best-case performance) of each defense
mechanism.

2) One should acknowledge possible situation
dependency of defense mechanisms, and be able to choose the
most suitable defense when more than one defense
mechanisms are available against a specific attack type.

3) One should evaluate defense mechanisms in a
comprehensive way, including both benefits and
disadvantages (worst-case performance), as an attacker can
exploit any weakness in a defense mechanism.

Knowledge of all of these issues is necessary in successful
mitigation of DoS attacks. Without knowing how a specific
defense mechanism works under different possible conditions
and what the real benefits and weaknesses are, it is not
possible to assure the suitability of a defense mechanism
against a certain type of a DoS attack.

V. Research Methodology

Research methodologies aimed to be used in this proposal,
are primarily based on simulating different attack scenarios,
but measurements, mathematical modeling based on game
theory, and requirement specification are also planned to be
used .

VI. Scope of the Research

Since this proposal studies DoS attacks in computer
networks using the Internet Protocol (IP), namely the Internet
and mobile ad hoc networks, is extremely useful for the
security concern. DoS attacks in the physical world will not be
studied here. Major work concentrate on the fixed (wired)
Internet, but most of the considered attack and defense
mechanisms will be applicable to wireless networks, too. The
emphasis of this research proposal is on DoS attacks in
general, and DDoS attacks are treated as a subset of DoS
attacks. DDoS attacks are based on the same mechanisms as
basic DoS attacks, but there is one exception during the
deployment phase .A DDoS tool needs to be installed on many
vulnerable hosts. The installation of DoS software on a single
vulnerable host is, however, a common prerequisite for most
DoS attacks. Thus attack and defense mechanisms described
in this dissertation are applicable to both DoS and DDoS
attacks.

VII. Possible Outcome
The main contributions of this proposed work include,

1) A comprehensive and well-structured description can
be given about what DoS attacks really are? How DoS attacks
can be carried out in IP networks? And how one can defend
against DoS attacks in IP networks. A good understanding of
existing attack mechanisms and available defense mechanisms
is a prerequisite for succeeding in mitigating these attacks cost
effectively.

2) An overview of an organized approach for selecting a
comprehensive set of defense mechanisms against DoS attacks
is given. This emphasizes the importance of basic security
mechanisms at every host in the Internet, the importance of
risk management in choosing additional defenses when basic
defenses are not enough, and the necessity of implementing
new defenses against such important DoS attacks for which
there are no existing defenses.

200

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

3) A new defense mechanism for protecting
organization-specific name servers will be described and
simulated.

4) Since knowledge about DoS and DDoS is in
primitive stage, we are hopeful to extend above objectives to
study DoS attack in mobile ad hoc networks. An earlier
attempt is found successful in some similar work [6].

CONCLUSION

This proposal aim to evaluate the DoS problems and
availability of defence mechanism. It is understood that the
existing defence mechanisms are mainly passive, in the sense
that the target host or network is impaired before the attack
source(s) can be found and controlled. We wish to propose a
novel concept of active defence against DoS attacks by
mitigating them in the Internet. This proposed style has
sufficient advantages over conventional passive defence
mechanisms. However, this is only the first step toward
realizing the secure Internet paradigm. The proposed work can
also be extended for designing of robust active defence
architecture, developing a sensitive and accurate surveillance
system, or for a powerful active trace back system and
deployment of such system in real Internet environment.

Authors

Dr. S.D.Khamitkar: He is PhD in computer science and has 15+ research
papers in International Conferences and journals. His interest area includes
ICT, Green computing and Network Security.

P.U.Bhalchandra , N.K. Deshmukh , S.N.Lokhande : These are SET-
NET qualified faculties and have 8+ years teaching experience. They have 5+
research papers in international conferences and journals. At present they are
also working on research related to ICT and Green computing. The present
paper is research topic of Mr. S.N.Lokhande

S.S.Phulari , R.P.Rathod : These are also faculties and have qualified
M.Phil in computer science . They have 2+ papers in international conferences
and journals.

REFERENCES

[1] L. Zhou and Z. Haas. Securing ad hoc networks. IEEE Network,

13(6):24~30, November/December 1999.
[2] Y. Zhang and W. Lee, "Intrusion detection in wireless ad hoc

networks," ACM MOBICOM, 2000.
[3] P.Papadimitratos and Z.J. Haas, "Secure Routing for Mobile Ad

Hoc Networks," SCS Communication Networks and Distributed

Systems Modeling and Simulation Conference (CNDS 2002), San

Antonio, TX, January 27-31, 2002.
[4] S.Marti, T.Giuli, K.Lai and M.Baker, "Mitigating Routing

Behavior in Mobile Ad Hoc Networks", Proceedings ofMobicom

2001, Rome,2001.
[5] X Zeng, R. Bagrodia, and M. Gerla. GloMoSim: a library for

parallel simulation of large-scale wireless networks. In Proceedings

of the 12th Workshop on Parallel and Distributed Simulations, May

1998. 11.
[6] Jean-Pierre Hubaux, Levente Butty an, Srdjan Capkun,

"The Quest for Security in Mobile Ad Hoc Networks",

In Proceedings of the ACM Symposium on Mobile Ad Hoc

Networking and Computing (MobiHOC), Long Beach, CA,

USA, October 2001.

201

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

An Efficient and Minimum Cost Topology Construction for Rural Wireless Mesh

Networks

Prof. V. Anuratha & Dr. P. Sivaprakasam

Abstract

Many research efforts as well as
deployments have chosen IEEE802.il as a
low-cost, long-distance access technology to
bridge the digital divide. IEEE 802.11 Wi-Fi
equipment based wireless mesh networks
have recently been proposed as an
inexpensive approach to connect far-flung
rural areas. To establish such network high-
gain directional antennas are used to achieve
long-distance wireless point-to-point links.
Some nodes in the network are called
gateway nodes and are directly connected to
the wired internet, and the remaining nodes
connect to the gateway(s) using one or more
hops.

In this paper the cost of constructing the
antenna towers required is investigated. The
problem is NP hard is shown and that a
better than 0(log n) approximation cannot
be expected, where n is the number of
vertices in the graph. To minimize the
construction cost a new algorithm is
proposed called constant time approximation
algorithm.

The results of proposed approximation
algorithm are compared with both the
optimal solution, and a naive heuristic.

INTRODUCTION

There has been a huge proliferation of
Internet and other communication based
services in the last two decades. However,
this spread is confined to developed
countries, and metropolitan pockets of
developing countries. This is really
unfortunate for developing countries like
India, where around 74% of the population

is rural and are on the wrong side of the
digital divide.

Bridging this divide necessitates, providing
internet connectivity to each and every
village. Providing the same by expanding
the current telephone network to rural areas
is infeasible because of the huge initial
infrastructure costs. Also, deployment of
cellular wireless would not be sustainable
because of its business model, which
demands more high-paying consumer
density.

Emerging technologies like 802.16
WMAN[12],[13], have not yet reached the
scale of competitive mass production, hence
the equipments are expensive. In this regard,
the 802.11 Wi-Fi has shown tremendous
growth and acceptance as a last hop access
solution, because of their low price.
Although 802.11 was primarily designed for
indoor operation, but [3] has established the
possibility of using 802.11 in long-distance
networking.

The diverse requirements are in provisions
of 1) Communication pattern which deals
with the mode of communication one-to-
one, one to- many, many-to-one, and many-
to-many, 2) Delay (real-time, non-real-time,
and delay-tolerant), 3) Service availability
(centralized, distributed, and location-aware)
that deals with the awareness of the
availability of different services, such as
Internet access, real-time communications,
content distribution, interactive gaming,
medical applications, and vehicular safety
applications. 4) Security and 5) Reliability.
An essential requirement to establish long-
distance links is that line-of-sight is
maintained between the radio antennas at the
end-points.

202

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

To ensure line-of-sight across such long
distances would require the antennas to be
mounted on tall towers. The required height
of the towers depends both on the length of
the link, and the height of the obstructions
along the link.

The cost of the tower depends on its
height and the type of material used. For
relatively short heights (10- 20 meters)
antenna masts are sufficient. For greater
heights, sturdier and much more expensive
antenna towers are required. In this paper,
several important contributions are made
towards developing efficient algorithms [14]
to solve this problem. First, the requirements
to establish a point-to-point 802.11 link
between two nodes of a given network graph
is described. Then the formal definition of
the Topology Construction problem (denote
by TC) is given. It's proved that the problem
to be NP hard by a reduction from the set-
cover problem. The approximation algorithm
is presented for this NP hard problem and
the establishment cost of the tower in rural
areas using constant time approximation
algorithm is presented. The rest of this paper
is organized as the following. Section II
gives the Related Works of this technique is
presented. In Section III, the Methodology
of proposed approach is given. Section IV
has the Experiment Results and this paper is
concluded in Section V.

RELATED WORKS

802.11 -based long-distance networks have
been proposed as a cost-effective option to
provide Internet connectivity to rural areas
in developing regions, to enable Information
and Communication Technology (ICT)
services [5].

Rural areas (especially in developing
regions) have populations with very low
paying capacities. Hence, a major factor in
network deployment is the cost of the
infrastructure and the network equipment. In
this context, efficient algorithms are
investigated for the minimum cost topology
construction problem in rural wireless mesh
networks. R. Ramanathan et.al,[4] discussed

on the most critical design issues in
multihop wireless networks. Topology
control has been investigated extensively in
the literature. Nevertheless, it is noted that
most existing studies do not consider the
requirements on upper layer applications or
services. In this article the author address the
topology control issues on service-oriented
wireless mesh networks. In particular, the
author provides a comprehensive survey of
existing works on topology control from a
service-oriented perspective. A general
framework for topology control in service-
oriented WMNs is proposed. To
demonstrate the effectiveness of the
framework, a case study is conducted in
which the main objective is to maximize the
overall throughput in a network with random
unicast traffic. The performance of this
topology control scheme is evaluated by
numerical results. In addition, it is illustrated
that the generated topology can support
advanced technologies, including network
coding and physical-layer network coding,
which can significantly improve the
throughput capacity of a network. The cost
of laying wire to rural areas is prohibitively
expensive.

Also, traditional wireless technologies such
as cellular data networks (e.g., EV-DO) and
upcoming technologies like IEEE 802.16
WiMAX have prohibitively expensive
equipment costs. As a result, there has been
considerable recent interest [6], [7], [8] in
the design of rural mesh networks using
IEEE 802.11 (Wi-Fi) equipment. The cost of
an 802.11 radio (»$50/PCMCIA card) is
orders of magnitude less than that of
cellular/WiMAX base stations. Thus, this
approach is an attractive option for building
low cost networks. D. S. Lun et.al,[9]
presented a distributed random linear
network coding approach for transmission
and compression of information in general
multisource multicast networks.

Network nodes independently and randomly
select linear mappings from inputs onto
output links over some field. The author
shows that this achieves capacity with

203

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

probability exponentially approaching 1
with the code length. Random linear coding
is demonstrated which performs
compression when necessary in a network,
generalizing error exponents for linear
Slepian-Wolf coding in a natural way.
Benefits of this approach are decentralized
operation and robustness to network changes
or link failures. The author shows that this
approach can take advantage of redundant
network capacity for improved success
probability and robustness. Some potential
advantages of random linear network coding
are illustrated over routing in two examples
of practical scenarios: distributed network
operation and networks with dynamically
varying connections. The derivation result
also yields a new bound on required field
size for centralized network coding on
general multicast networks.

METHODOLOGY

In this paper a Novel Topology
Control Scheme is used to identify a set of
semi-permanent highways, such that the best
throughput capacity of the network can be
obtained. Particularly, the wireless highways
are predicted to be rather similar to the
highway system in public transportation
system, which can efficiently provide
connectivity in real application.

A. Computing tower heights at the end-
points of a link

Consider two nodes, u and v that are
separated by a distance luv. The edge (u; v)
is considered to be covered if an 802.11
based point-to-point communication link can
be established between u and v. Assume that
the transmit powers [15] and the gains of the
antennas at both ends are sufficient to over
come the free-space path loss between the
two points. The first basic requirement to
cover the edge between u and v is that there
is a clear visual line-of-sight between the
antennas at the end-points (as shown in Fig.
4a). In other words, the line joining the
antennas mounted on the towers should clear
any obstructions along the path. Secondly, it
is also required that RF line-of-sight is
maintained between the two points. This is

determined by an elliptical area between u
and v termed the first Fresnel zone. To
establish RF line-of-sight, a significant area
of the Fresnel zone (> 60% of the radius of
the Fresnel zone at the location of the
obstruction [1]) should also clear all
obstructions between u and v. However, this
can be simply modeled by extending the
height of the obstruction to include the
radius of the Fresnel zone that has to be in
the clear.

(a)

h^n

■:!>:■

fe* o

tatu)

b(v}

Ol Oa

Figure 1 : Computing the height of towers
at the end-points of a link

In reality, there can be multiple
obstructions between u and v. As in Figure
4b, consider multiple obstructions, Ol, 02,
...,Ok between u and v. Now, let h(u) and
h(v) represent the tower heights at the nodes
of u and v. Covering edge (u,v) requires a
visual and RF line-of-sight connection
between the towers at its two terminal
nodes. This would imply that the straight
line fuv joining the top of the two towers (of
heights h(u) at u and h(v) at v) should clear
every obstruction in (u,v). Hence, it is noted
that given a particular pair of tower heights
at u and v, deciding whether these heights
covers edge (u,v) can be done in time linear
in the number of obstructions on that edge.2

B. Modeling tower costs
An important component in this problem is
the nature of the cost function that maps
tower heights to the cost of building the
tower.

204

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

There are two types of antenna towers that
are used. For heights less than 20 meters,
one can use the cheaper masts. For greater
heights, the more expensive steel towers is
used.

Further, there is an order of magnitude
difference between the cost of the cheaper
masts and that of the steel towers. Thus,
roughly speaking, the cost function is
constant as long as the cheaper masts can be
used and becomes linear in height once the
steel towers are needed, with a jump in cost
when we switch from masts to steel towers.

Let us denote the height at which the
material of the tower has to be switched as
hmin.

Further, there is a physical restriction on the
maximum possible height of a tower,
denoted by hmax. Thus, the cost function c
can be formally defined as

:m=l

h + B ifh mtn <k<h„

Where A, B and K are constants and
Ahmin + B » K. Although, in practice, the
cost function can be modeled as discussed
above, our algorithm works with a much
more general cost functions. Specifically,
we only require the cost function c to satisfy
the following two natural properties CI and
C2. CI

Given the tower costs at two neighboring
nodes u and v, it can be determined (in
polynomial time) whether the corresponding
tower heights cover the edge (u,v). This
simply requires that the corresponding tower
height can be computed (in polynomial
time) given the tower cost. As mentioned
earlier, determining whether the height of
the towers is sufficient to cover an edge can
be done in polynomial time.C2 the cost
function is monotonically increasing with
height, i.e., hl> h2 and c(hl) > c(h2) for
any values of hi and h2.

It is easy to verify that the cost function
c defined earlier in this section satisfies both

of the above properties. In the remainder of
this paper, when the height function h is
unambiguous, often the cost of the tower is
denoted at a node v as c(v) rather than
c(h(v)).

C. The Topology Control Scheme
In this framework, it is considered that
the highways can be partitioned into two
groups, horizontal and vertical. Highways in
each group can operate simultaneously
because they are mutually parallel and can
be placed away enough to reduce
interference below a certain threshold.
Consequently, horizontal and vertical
highways will partition the whole
geographical area into grids, in which nodes
will try to forward their traffic to the nodes
on neighboring highways. The combination
of the following parameters can be
considered for the Topology Control
Scheme

1) Transmission range: Transmission

range of each node in the network
is traditionally an important
design parameter in topology
control. In general, a smaller
transmission range will improve
the channel reuse but may
compromise the connectivity. A
larger transmission range will
improve the connectivity but
reduce the channel reuse.
Therefore, an appropriate range is
chosen as a trade-off between
connectivity and channel reuse.

2) Type of antenna: When directional

antenna or beam forming is used it
may improve the capacity of the
network by reducing the
interference and improve the
transmission quality.

3) Traffic pattern: Traffic pattern is

very important parameter to the
topology. In most studies
previously done suggest that the
traffic is broadcast. With such an
assumption, the problem is
formulated in a way such that the
overall transmission for each

205

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

message is minimized. However,
broadcast traffic may only be a
special case in the future service-
oriented WMN, in which a variety
of patterns may appear, from one-
to one to many-to-many [10].
4) Quality of service (QoS): To
achieve an efficient network a
crucial issue is to enable services
with certain QoS requirements,
such as bandwidth, delay, security
and reliability.

For the first step of this study, only
omnidirectional antenna and purely
random unicast traffic pattern are
considered. Moreover, the random
wireless network elaborated to gain
insights for the future investigation.

Figure 2: Approximation algorithm for NP-hard problem

Constant Time Approximation Aigori thrr.

1. F t- 8 Comment: Implicitly sett growth variables y s t- and cumulative growth variables
«- for all s c V.-w(jS~} denotes the total potential expended in component S including any ju&

components that were merged in creating S.Also implicitly setw v

(1 - a}fc

for all node v

■ r.

2. c *- [i? : v E V,v ^ r} .

For each v == V set d(v') *-

Q. Cement: d(v) denotes the distance vfnade v to the boundary of ths component containing v.

For each veV if v = r them l({v}] *- else A(M] «-
1 , Comment :lis') equal 1 if s active and Q otherwise.

5. while there are active components

6. Comment: Find the next event

7. Find edge e = {i,j)with i E C p E C,j E C fl E C, € p ^ C fl that minimizes

_ Co - d(iO - ri(/]
e= 2(c p ) + l(C fl )

8. Find C E C writh X(£) = 1 that mmimiiss e 2 = | tf"|.— — — — mjE)

9. c = min(e L .. fj)

10. cjCC] <— lcj(CD +■ e J(C] for all C e c

11. /or aft 1? E C r E c

12. d(v0 <— rift?] +■ e.A(C L J

13. ;'/ e = e 2 (i. e. , (T z"s deactiratsd before C^ and C fl mffrfj

14. A.((!") «-

15. Mark all unlabeled vertices of € with label tT, else (i.e., C and C q meet before C is deactivated"}

16. F^F\j[e]

17. C e- CU{c p U Cj- {C p }- {cj

18. ^(C p U C q ) ^- w(C p ) + w(C fl )

19. i/ r E C p UC fl tftm l(c p UC fl ) e- else A{c p UC^g- 1.

20. For ever\- unlabeled vertex v £ CV

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

EXPERIMENTAL RESULTS

In this section, extensive numerical
simulations are carried out to evaluate our
approximation algorithm with the optimal
solution and also a naive heuristic. For this
simulations, synthetic topologies are
generated that aim to match the geographical
structure of village clusters. Simulation
setup is described in more detail as follows.

A. Generating
topologies.

synthetic graph

A circular plane is considered with a
radius of 25Kms. Nodes are placed at
random locations on this plane. A link (u,v)
is considered between any two nodes, u and
v, and for these simulations, assume just one
obstacle, ouv, located on the middle of this
link.

The height of the obstruction (ho) is
selected randomly with a maximum value of
20 meters - the typical height of trees and
small houses in a rural setting. A weight
wuv is assigned to the link equal to twice the
effective height of the obstruction on this
link. As described earlier, the effective
height of an obstruction, is the sum of the
physical height (ho) and 60% of rf , the
radius of the fresnel zone.

B. Naive heuristic

In order to compare with the proposed
approximation algorithms, a naive heuristic
is described for selecting connected
subgraphs and assigning heights to the
nodes. As a first step, to select a connected
subgraph of an input graph G, the heuristic
computes the minimum spanning tree
(MST), T of G (using the link weights
computed as described above). Next, the
heuristic has to assign heights to the nodes
in G, so as to cover all the edges in T while
minimizing the total cost.

Given a set of links to be covered, the
height assignment problem is formulated as
a simple LP, and compute the heights
required on every node.

C. Comparing with the naive heuristic

Now the naive heuristic described above is
compared with the proposed approximation
algorithm. Graphs considered with number
of nodes n =10, 15, 20,..., 50. For each
value of n we generate 50 graph instances.
For each graph, Cnaive the cost of the
solution produced by the naive heuristic, and
Capprox cost of the approximation
algorithm is computed. With this value it is
observed that the proposed approximation
algorithm performs substantially better than
the naive heuristic.

On average, the solutions returned by the
naive heuristic range from 60% (for n = 10)
to as much as 225% more expensive (for n =
50) compared to the solution returned by the
naive algorithm.

D. Comparing with the optimal solution

The optimal solution is computed by
solving an ILP that models the topology
construction problem. The CPLEX LP-
solver [11] to solve this ILP. This approach
is, however, computationally very
expensive, and the LP-solver could return
solutions for graphs with at most 1 1 nodes.

The solution returned by our
approximation algorithm is compared with
the optimal solution for graphs with number
of nodes n = 8, 9, 10, 11. For

each value of n we generate 50 graphs. For
each graph Capprox, the cost of the solution
returned by proposed approximation
algorithm, and Copt, the cost of the optimal
solution.

207

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Then the mean and standard deviation is
computed over all graphs for different

values of n R

*- n t-.-t.-t n v t-n

K o j pt

L opt

Mean (std. dev.) of
Ropt

0.58 (0.30)

0.57 (0.25)

0.55 (0.23)

0.52 (0.25)

Table 1 : New Approximation algorithm
vs. Optimal Solution.

The results presented in Table 7 show
that the incorporated approximation
algorithm gives solutions that are 50 - 60%
more expensive than the optimal solution
(for small values of n). Thus, the new
approximation algorithm performs much
better than the worst case guarantee of 0(log
n) on the approximation factor.

While this gap between constant time
approximation algorithm and the optimal
solution is not small, its expected computing
the optimal solution (even if it has to be
done only once) is practically infeasible for
real-life networks. Moreover, this
approximation algorithm performs

substantially better in practice than the naive
heuristic and previous approximation
algorithms

CONCLUSION

In this paper an overview of
establishing a low cost wireless mesh
network for rural areas is presented. In rural
areas nodes are connected using long
distance 802.11 wireless links which are
established using high-gain directional
antennae. The main problem is with the
topology construction for long distance
wireless communication.

An efficient approximation

algorithm is proposed for the topology
construction problem in rural mesh
networks. This work introduces a number of

open research problems in the topology
construction.

One immediate problem is to
consider the case of k > 2 vertex or edge
connectivity, similar to the power optimal
network construction for k-connectivity.
Another important research direction is the
geometric version of this problem. In
practice, all nodes within a certain distance
of each other can establish a link.

In this paper, the location of the
towers is assumed to be fixed (within a
village). A variant of the problem would
make the location of the tower to be a
variable.

This method has added flexibility
than the previous method which would
result in reduced cost. The numerical
experiments demonstrate that the proposed
method with constant time approximation
algorithm performs well within its worst
case performance bounds, and outperforms
the naive heuristic by a substantial margin.

References

[1] I. Akyildiz and X. Wang, "A survey on
wireless mesh networks," IEEE
Communications Magazine, vol. 43, no.
9, pp. S23-S30, Sept. 2005.

[2] M. Lee, J. Zheng, Y.-B. Ko, and D.
Shrestha, "Emerging standards for
wireless mesh technology," IEEE
Wireless Communications, vol. 13, no.
2, pp. 56-63, April 2006.

[3] Pravin Bhagwat, Bhaskaran Raman, and
Dheeraj Sanghi," Turning 802.11 Inside-
Out", In HotNets-II, Nov 2003.

[4] R. Ramanathan and R. Rosales-Hain,
"Topology control of multihop wireless
networks using transmit

poweradjustment," vol. 2, 2000.

[5] Eric Brewer, Michael Demmer, Bowei
Du, Kevin Fall, Melissa Ho, Matthew
Kam, Sergiu Nedevschi, Joyojeet Pal,
Rabin Patra, and Sonesh Surana. The
Case for Technology for Developing

208

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Regions. IEEE Computer, 38(6):25-38,
June 2005.

[6] P. Dutta, S. Jaiswal, K. Naidu, D.
Panigrahi, R. Rastogi, and A. Todimala.
Villagenet: A low-cost, 802.11 -based
mesh network for rural regions. In
Wireless Systems: Advanced Research
and Development Workshop

(WISARD), 2007.

[7] R. Patra, S. Nedevschi, S. Surana, A.
Sheth, L. Subramanian, and E. Brewer.
WiLDNet: Design and Implementation
of high performance wifi based long
distance networks. InNSDI, 2007.

[8] B. Raman and K. Chebrolu. Design and
evaluation of a new MAC for long
distance 802.11 mesh networks. In
Mobicom, 2005.

[9] D. S. Lun, N. Ratnakar, M. Medard, R.
Koetter, D. R. Karger, T. Ho, E. Ahmed,
and F. Zhao, "Minimum-cost multicast
over coded packet networks," IEEE
Transaction on Information Theory, vol.
52, no. 6, pp. 2608-2623, June 2006.

[10] K. Lu, Y. Qian, H.-H. Chen, and S.
Fu, "WiMAX Networks: From Access
To Service Platform," IEEE Network,
2008, accepted.

[11] ilogcplex.

http://www.ilog.com/products/cplex/ .

[12] M. T. Hajiaghayi, N. Immorlica, and
V. S. Mirrokni. Power optimization in
fault-tolerant topology control

algorithms for wireless multi-hop
networks. In MOBICOM, 2003.

[13] M. T. Hajiaghayi, G. Kortsarz, V. S.
Mirrokni, and Z. Nutov. Power
optimization for connectivity problems.
In IPCO, 2005

[14] Z. Nutov. Approximating minimum
power covers of intersecting families
and directed connectivity problems. In
APPROX-RANDOM, 2006.

[15] S. Sen and B. Raman,"Long distance
wireless mesh network planning:
Problem formulation and solution", In
WWW, 2007.

[16] P. N. Klein and R. Ravi. A nearly
best-possible approximation algorithm
for node-weighted steiner trees. J.
Algorithms, 19(1), 1995.

Author Biographies

Mrs. V. Anuradha graduated with B.Sc
Computer Science in the year 1995 and
completed M.C.A at Madras University in
the Year 2000. Completed her M.Phil in the
year 2003 and also got Guide Approval for
M.Phil in Bharathiar University, Peiyar
University, Bharathiar University and
currently doing her Ph.d.Hear area of
interest is Networks and data mining. Mrs.
V. Anuratha have guided 20 M.Phil scholors
and she have participated and presented
many papers in the national and
international conferences and etc..

Currently she is working as a H.O.D - PG
Department of Computer Science at Sree
Saraswathi Thyagaraja College with a
decade a teaching experience

Co author biography

Dr. P. Sivaprakasam :

He have completed his M.Sc(c.s) in the year
1986, M.Phil in the year 1995 and Ph.d in
2005 on the topic "An Analysis of Web
performance and caching". He have 19 years
of teaching experience. He have published 6
papers at national level and 3 at international
level. He was also sanctioned with 2 UGC
research projects.

He is now currently working as a Associate
professor in Computer Science at Sri Vasavi
college of Arts and Science. His areas of
interest are Internet, Computer Networks,
Service Oriented Architecture.

209

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Reinforcement Learning by Comparing Immediate

Reward

Punit Pandey
Department Of Computer Science and Engineering,
Jaypee University Of Engineering And Technology

DeepshikhaPandey
Department Of Computer Science and Engineering,
Jaypee University Of Engineering And Technology

Dr. Shishir Kumar
Department Of Computer Science and Engineering,
Jaypee University Of Engineering And Technology

Abstract — This paper introduces an approach to Reinforcement
Learning Algorithm by comparing their immediate rewards
using a variation of Q-Learning algorithm. Unlike the
conventional Q-Learning, the proposed algorithm compares
current reward with immediate reward of past move and work
accordingly. Relative reward based Q-learning is an approach
towards interactive learning. Q-Learning is a model free
reinforcement learning method that used to learn the agents. It is
observed that under normal circumstances algorithm take more
episodes to reach optimal Q-value due to its normal reward or
sometime negative reward. In this new form of algorithm agents
select only those actions which have a higher immediate reward
signal in comparison to previous one. The contribution of this
article is the presentation of new Q-Learning Algorithm in order
to maximize the performance of algorithm and reduce the
number of episode required to reach optimal Q-value.
Effectiveness of proposed algorithm is simulated in a 20 x20 Grid
world deterministic environment and the result for the two forms
of Q-Learning Algorithms is given.

Keywords-component; Reinforcement Learning, Q-Learning
Method, Relative Reward, Relative Q-Learning Method.

Introduction

Q-Learning algorithm proposed by Watkins [2,4] is a model
free and online reinforcement learning algorithm. In
reinforcement learning selection of an action is based on the
value of its state using some form of updating rule. There is an
interaction between agent and environment where the agent has
to go through numerous trials in order to find out the best
action. An agent chooses that action which has maximum
reward obtained from its environment. The reward signal may
be positive or negative depends on the environment.

Q-learning has been used in many applications because it
does not require the model of environment and is easy to
implement. State-action value, a value for each action from
each state, converges to the optimal value as state-action pairs
are visited many times by the agent.

In this article we propose a new relative reward strategy for
agent learning. Two different form of Q-Learning method is
considered here as a part of study. First form of Q-Learning
method uses a normal reward signal. In this algorithm Q-value
evaluates whether things have gotten better or worse than

expected as a result of an action selection in the previous state.
The action selected by agents is most favorable which has
lower TD error. Temporal difference is computed on the basis
of normal reward gain by agents from its surroundings. An
estimated Q-value in the current state is than determined using
Temporal Difference. Agent actions are generated using the
maximum Q-values. The second form of Q-Learning
algorithm is an extension towards a relative reward. This form
of Q-Learning method utilizes the relative reward approach to
improve the learning capability of algorithm and decreases the
number of iteration. In this algorithm only those action is
selected which has a better reward from its previous one.

This idea comes from psychological point of views that
human beings tend to select only those action which has higher
reward value. However, this algorithm is not suitable for multi
agent problems. To demonstrate effectiveness of the proposed
Q-Learning algorithm, Java applet is utilized to simulate a robot
that reaches to a fixed goal. Simulation result confirms that the
performance of proposed algorithm is convincingly better than
conventional Q-learning.

This paper is organized as follows: Basic concept of
reinforcement learning is presented in section 2. Section 3
describes about the conventional Q-Learning method. Section 4
presents a new Relative Q-Learning in context of relative
immediate reward. Section 5 describes Experimental setup &
results and concluding remarks follow in Section 6.

II. Reinforcement Learning

Reinforcement learning (RL) is a goal directed learning
methodology that is used to learn the agents. In Reinforcement
learning [1,5,6,7,8,9] the algorithm decide what to do and how
to map situations to actions so that we maximize a numerical
reward signal. The learner is not advised which actions to take,
but instead it discover which actions provide the maximum
reward signal by trying them. Reinforcement learning is
defined by characterizing a learning problem. Any algorithm
that can able to solve the defined problem, we consider to be a
reinforcement learning algorithm. The key feature of
reinforcement learning is that it explicitly considers the whole
problem of a goal-directed agent interacting with an uncertain
environment. All reinforcement learning agents [3,10,11,12]
have explicit goals, can sense aspects of their environments,
and can choose actions to influence their environments. In

210

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

reinforcement learning agent prefer to choose actions that it has
tried in the past and found to be effective in producing
maximum reward. The agent has to exploit based on what it
already knows in order to obtain reward and at the same time it
also has to explore in order to make better action selections in
the future. Reinforcement learning has four elements policy,
reward function, value function and model of environment.

(stale)

: ■=««■: :

Learner

:;:lcn'i

Environment

agent because it alone is sufficient to take the decision on
further action.

III. Q-Learning

Q-learning is a form of model- free reinforcement learning [2]
(i.e. agent does not need an internal model of environment to
work with it). Since Q-learning is an active reinforcement
technique, it generates and improves the agent's policy on the
fly. The Q-learning algorithm works by estimating the values
of state-action pairs.

The purpose of Q-learning is to generate the Q-table, Q(s,a),
which uses state-action pairs to index a Q-value, or expected
utility of that pair. The Q-value is defined as the expected
discounted future reward of taking action a in state s, assuming
the agent continues to follow the optimal policy. For every
possible state, every possible action is assigned a value which
is a function of both the immediate reward for taking that
action and the expected reward in the future based on the new
state that is the result of taking that action. This is expressed by
the one-step Q-update equation [2,4,10,13,14].

Q(s, a)

Q(s, a) + a [r+ y * max a Q(s', a') - Q(s, a)]

(i)

Figure 1.1: .Reinforcement learning

S 2

Goal: learn to choose actions that maximize:
ro + 7 ri + f n + . . . , where < 7 < 1

Figure 1. 2 : Reinforcement learning

Model of the environment is an optional element because
reinforcement learning also supports the model free algorithms
like Q-learning.

A policy for our agent is a specification of what action to take
for every Input. In some cases policy may be a simple function
or look-up table or sometime it can be an extensive
computation. The policy is the core of reinforcement learning

AGENT

action seletor

Q- factor Table

Input

Reinforcement , ,

action

WORLD

Figure 2: Structure of the Q-Learning agent

Where a is the learning factor and y is the discount factor.
These values are positive decimals less than 1 and are set
through experimentation to affect the rate at which the agent
attempts to learn the environment. The variables s and a
represent the current state and action of the agent, r is the
reward from performing s' and a', the previous state and
action, respectively.

The discount factor makes rewards earned earlier more
valuable than those received later. This method learns the
values of all actions, rather than just finding the optimal
policy. This knowledge is expensive in terms of the amount of
information that has to be stored, but it does bring benefits. Q-
learning is exploration insensitive, any action can be carried
out at any time and information is gained from this experience.
The agent receives reinforcement or reward from the world,

211

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

and returns an action to the world round and round as shown
below:

A Elementary parts of Q-learning:

Environment:

Q-learning based on model-free mode of behavior i.e the
environment is continuously changing. Agent does need to
predict future state. Environment can be either deterministic or
non-deterministic. In deterministic environment application of
single state lead to a single state where as in nondeterministic
environment application of a single action may lead to a
number of possible successor states. In case of non-
deterministic environment, each action not only labeled with
expected immediate reward but also with the probability of
performing that action. For the sake of simplicity we are
considering deterministic environment in this thesis work.

Reward Function:

A reward function defines the goal in a reinforcement learning
problem, it maps each perceived state (or state-action pair) of
the environment to a single number, a reward, indicating the
intrinsic desirability of that state. A reinforcement learning
agent's sole objective is to maximize the total reward it
receives in the long run. The reward function defines what the
good and bad events are for the agent.

Action-value function:

The Q-learning learning is based upon Quality-values (Q-
values) Q(s,a) for each pair (s,a). The agent must cease
interacting with the world while it runs through this loop until
a satisfactory policy is found. Fortunately, we can still learn
from this. In Q-learning we cannot update directly from the
transition probabilities-we can only update from individual
experiences. In 1 step Q-learning, after each experience, we
observe state s', receive reward r, and update:

Q(s, a) = r+ y maxa' Q(s', a') (2)

B. Q-learning Algorithm:
Initialize Q(s, a) arbitrarily

Repeat (for each episode)

Choose a starting state, s

Repeat (for each step of episode):

Choose a from s using policy derived from Q
Take action a, observe a immediate reward r, next state s'
Q(s, a) <— Q(s, a) + a [r+ y * maxa' Q(s', a') - Q(s, a)]
s<— s' ;
Until state s' match with the Goal State

Until a desired number of episodes terminated

State

Action

Figure 3 : Q-Learning Architecture

IV. Relative Q-Learning

This section introduces a new approach Relative reward to
conventional Q-learning that makes Relative Q-Learning.
Conventional Q-learning has been shown to converge to the
optimal policy if the environment is sampled infinitely by
performing a set of actions in the states of the environment
under a set of constraints on the learning rate a. No bounds
have been proven on the time of convergence of the Q-learning
algorithm and the selection of the next action is done randomly
when performing the update. This simply mean that the
algorithm would take a longer time to converge as a random set
of states are observed which may or may not bring the state
closer to the goal state. Furthermore, it means that this function
cannot be used for actually performing the actions until it has
converged as it has a high chance of not having the right value
as it may not have explored the correct states. This is especially
a problem for environments with larger state spaces. It is
difficult to explore the entire space in a random fashion in a
computationally feasible manner. So by applying below
mention method and algorithm we try to keep the Q-learning
algorithm near to its goal in less time and less number of
Episode.

A Relative Reward

Relative reward is a concept that compares (current reward
with the previous received reward) two immediate rewards.
The objective of the learner is to choose actions maximizing
discounted cumulative rewards over time. Let there is an agent
in state st at time t, and assume that he chooses action at. The
immediate result is a reward rt received by the agent and the
state changes to st+1. The total discounted reward [2,4]
received by the agent starting at time t is given by:

r(t)=r t +yr t+1 +y 2 r t+ 2+ +y n r t+n +

Where y is discount factor in the range of (0: 1).

(3)

212

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

The immediate reward is based upon the action or move
taken by an agent to reach the defined goal in each episode.
The total discounted reward can maximize in less number of
episode if we select the higher immediate reward signal from
previous.

B. Relative Reward based Q-Learning Algorithm

Relative reward based Q-learning is an approach towards
maximizing the total discounted rewards. In this form of Q-
learning we selected the maximum immediate reward signal by
comparing it with previous one. This is expressed by the new
Q-update equation.

Q(s, a) = Q(s, a) + a [max(r(s,a),r(s\a'))+ y maxa' Q(s', a')
-Q(s,a)]

Algorithm:

Initialize Q(s, a) arbitrarily

Repeat (for each episode)

Choose a starting state, s

Repeat (for each step of episode):

Choose a from s using policy derived from Q

Take action a, observe a immediate reward r, and next state
s'

Q(s, a) = Q(s, a) + a [max(r(s,a),r(s',a'))+ y maxa' Q(s', a')
-Q(s,a)]

s<— s' ;

Until state s' match with the Goal State

Until a desired number of episodes terminated

V. Experiments & Results

The Proposed Relative Q-Learning was tested on 10 x 10
and 20 x 20 grid world environment. In the Grid World Square
There are four possible actions for the agent as it is a
deterministic environment given in figure 4.

IJtMSl

| Start | | Reset |

In order to consider the situation of encountering a wall, the
agent has no possibility of moving all the way in the given
direction. When the agent enters into goal states, it receives 50
as a reward. We are also providing the immediate reward value
by incrementing or decrementing the Q-value marked with S
represent the start state and G represent the goal state. The
purpose of the agent is to find out the optimum path to arrive at
the goal state starting from the start state, and to maximize the
reward it receives.

Conventional Q-Learning/ Random Strategy

60 1

1 38 75 112 149 186 223 260 297 334 371 408 445 482 519 556
Episode

Figure5: Conventional Q-Learning.

Relative Reward Basd Q-Learning/Random Strategy

— Seriesl

1 42 83 124 165 206 247 288 329 370 411 452 493 534 575 616
Episode

Figure4: A 10 x 10 Grid World Environment

Figure6: Relative Q-Learning

We have executed 500 episodes to converge the Q-value.
The grid world is a deterministic environment so the value of
learning a and discount rate Y were set to 0.8. Figure 5 &

213

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Figure 6 shows the relationship between Q- Values and the
number of episode where x axis represents the number of
episode and y axis represents the Q-values. Figure 5 represents
the result of conventional Q-Learning where we can see that Q-
value converges after executing 500 episodes but in figure 6
Relative Q-learning takes 300 episode

So we can say that convergence rate of relative Q-learning
is faster than conventional Q-learning.

VI. Conclusion & future work

This paper proposed an algorithm which compares the
immediate reward signal with its previous one. The agent will
immediately return back to previous state if it will receive the
lower reward signal for that particular move. If conventional
Q-learning was applied in the real experiment, a lot of
iterations were required to reach the optimal Q values. The
Relative Q-learning algorithm was proposed for environment
which used small amount of episodes to reach the convergence
of Q-values. This new concept allows the agent to learn
uniformly and helps in such a way so that it will not deviate
from its goal. Part of future work may be included to verify the
proposed algorithm in nondeterministic environment.

References

[I] J.F. Peters, C. Henry, S. Ramanna, Reinforcement learning with pattern-
based rewards, in proceding of forth International IAS TED Conference.
Computational Intelligence (CI 2005) Calgary, Alberta,Canada, 4-6 July

2005,267-272

[2] Technical Note Q,-Learning Christopher J.C.H. Watkins and Peter
Dayan Centre for Cognitive Science, University of Edinburgh, Scotland
Machine Learning, 8, 279-292 (1992)

[3] J.F. Peters, C. Henry, S. Ramanna, Rough Ethograms: Study of
Intelligent System Behavior. In:M.A.Klopotek, S. Wierzchori ,
K.Trojanowski(Eds), New Trends in Intelligent Information Processing
and Web Mining (IIS05), Gdansk, Poland, June 13-16 (2005),1 17-126.

[4] C. Watkins, "Learning from Delayed Rewards", PhD thesis, Cambridge
University, Cambridge, England, 1989

[5] J.F.Peters,K.S.Patnaik,P.K.Pandey,D.Tiwari, "Effete of temperature on
swarms that learn", In Proceeding of IASCIT-2007,Hyderabad,INDIA

[6] P.K.Pandey,D.Tiwari, " Temperature variation on Q-Learning",In
Proceeding of RAIT in FEB 2008,ISM Dganbad

[7] P.K.Pandey,D.Tiwari, " Temperature variation on Rough Actor-Critic
Algorithm", Global Journal Computer Science and Technology, Vol 9,
No 4 (2009), Pennsylvania Digital Library

[8] L.P. Kaelbling, M.L. Littman, A.W. Moore, Reinforcement learning: A
survey Journal of Artificial Intelligence Research, 4, 1996, 237-285.

[9] R.S. Sutton, A.G. Barto, and Reinforcement Learning: An Introduction
(Cambridge, MA: The MIT Press, 1998).

[10] C. Gaskett, Q-Learning for Robot Control. Ph.D. Thesis, Supervisor:
A.Zelinsky, Department of Systems Engineering, The Australian
National University, 2002.

[II] Thrun. S.and Schwartz.A.(1993),Issues in using function approximation
for reinforcement learning, in Proceeding of the 1993 Connectionist
Models Summer School,Erblaum Associates. Nj.

[12] Richard S. Sutton, Reinforcement Learning Architectures, GTE
Laboratories Incorporated, Waltham, MA 02254.

[13] Tom 0'Neill,Leland Aldridge,Harry Glaser, Q-Learning and Collection
Agents, Dept. of Computer Science, University of Rochester

[14] Vanden Berghen Frank, Q-Learning, IRIDIA, Universit Libre de
Bruxelles

214 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Information realization with statistical predictive
inferences and coding form

D.Mukherjee

Sir Padampat Singhania University

Udaipur-3 13601 ,Raj asthanjndia

P.Chakrabarti* , A.Khanna , V.Gupta

Sir Padampat Singhania University

Udaipur-3 13601 ,Raj asthan,India

Abstract — The paper deals with information realization in case of
grid topology. Nodal communication strategies with clusters has
also been cited. Information prediction has been pointed out with
relevant statistical method, forward sensing, backward sensing
and cumulative frequency form. Binary tree classifier theory has
been applied for information grouping. The paper also deals with
comparison analysis of information coding.

Keywords- grid topology forward sensing
binary tree classifier, information coding

backward sensing,

I. INFORMATION MERGING IN GRID IN
DIAGONAL APPROACH

In order to solve complex problems in artificial intelligence
one needs both large amount of knowledge & some
mechanisms for manipulating that knowledge to create
solutions to new problems .Basically knowledge is a mapping
of different facts with help appropriate functions for e.g. Earth
is a planet. Can be realized as a function -planet (Earth).

Information merging can be realized as combining different
pieces of information to arrive at a conclusion. The different
information elements can be related in different ways i.e.
either in hierarchy or in form of a graph or even a mesh.
Consider a mesh of size m X n i.e. m rows & n columns then
if each intersection point has a information element placed on
it then one way of merging element A with B can be covering
a path of length (5XN) (here m= 8 & n=9). If weight of
covering each path is considered same then in case of diagonal
approach we can find a path of diagonal nature of length 5 V2
and then travelling a length (N-5) in linear fashion thus finding
a shortest path the same can also be determined by graph
algorithms like Dijkstra's or kruskal's algorithm for
minimum spanning tree. If each path is considered to be of
zero weight then interestingly there is no sense travelling a
path from A to B i.e. we can directly merge the two points
i.e. we take point A &directly merge it with point B in such a
case we need to have some stack like mechanism to determine
the order in which the nodes arrive & are merged.

*
*
*
*

4 ►

Mrows (m- l)paths

* shows path of traversal

N columns (n-1) paths

Figl: Information merging in mesh/grid

The above concept can be realized in DDM(Distributed Data
Mining) where large amount of geographically scattered
knowledge is merged & is mined to derive conclusions &
make decisions for e.g. GIS i.e. the Geographical Information
System which uses cartography(art of making maps) with
various information elements(sources) to derive decision
support results like which route to choose for a given
destination.

II. INFORMATION MERGING IN CLUSTER NETWORKS

This section mainly focuses on the nodal communication
between the farthest node in a N*N structurefl] and
information realization indicates nodal message . Let us
assume each cluster to be consisting of 16 nodes and then try
to communicate between the source and the destination node
as described in the figl. The point to be noted here is that to
establish the communication link between the adjacent
elements or units of the cluster we have to have the
communication in just reverse order in the 2 adjacent
elements. The order of the communication is

215

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

The condition can be visually imagined as follows:

Now let us first talk about the case when there is only one
element i.e, 1*1. In this particular case if we want to
communicate between the farthest node then there will be only
1 node in between the source and the destination which can be
further visualized as follows:

If we denote it by using the function f(x)then the value of
f(x)will be l.f(x)=l; The intermediate node is 11. Now let us
consider the case 2*2 matrix the value here will be
f(x)= 1+2=3; The intermediate nodes are 1(2,3),2(4).

n n

For the case for the 3*3 matrix the value of the function

f(x)= 1+2+2=5;

i m x in

nodes, i.e, f(x)=3+4+4;In case of 4*4 matrix to communicate
between the farthest node we need 7 nodes, i.e,
f(x)=3 +4+4+4 ;In case of 16 elements in a ring , we can
proceed as follows. Let us consider the case of 1*1 matrix to
communicate between the farthest node we need 3 nodes, i.e,
f(x)=7.In case of 2*2 matrix to communicate between the
farthest node we need 7 nodes, i.e, f(x)=7+8;In case of 3*3
matrix to communicate between the farthest node we need 7
nodes, i.e, f(x)=7+8+8;In case of 4*4 matrix to communicate
between the farthest node we need 7 nodes, i.e,
f(x)=7+8+8+8;Now the total number of nodes can be derived
by the general formula as (N/2-l)+(M-l)*(N/2) where N =
number of nodes present in the unit or element, M =
dimension of the square matrix. The data can be represented in
the tabular form as follows:

No. of
nodes

1*1

2*2

3*3

4*4

Fig. 2: Nodal communication in cluster

The x-axis represents the M*M matrix where M varies from 1
to 3. The y-axis represents the number of optimum
communication nodes required in the establishing the path
between the source node and the farthest node. The number of
nodes per element is indicated by the 3 colors.

Similarly for the 4*4 matrix we can get the value of
f(x)=l+2+2+2.

Here in this case we were having only 4 elements in a ring
.Suppose we have 8 elements in the ring in that case we have
to compute the number of nodes required to communicate or
establish the connection between the farthest nodes.
Justification - Let us consider the case of 1*1 matrix to
communicate between the farthest node we need 3 nodes, i.e,
f(x)=3.In case of 2*2 matrix to communicate between the
farthest node we need 7 nodes, i.e, f(x)=3+4;In case of 3*3
matrix to communicate between the farthest node we need 7

III. STATISTICAL INFERENCE OF FUTURISTIC
VALUES

In statistical inferences the input & output of a situation are
related with a certain relation or function based on which we
infer futuristic values. Consider a real-time situation in which
a given input parameter is observed over time between instants
Tl & T2 given the relation [2]

M t = a.e* then M avg = V(M ti . M t2 )

216

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Case 1:

If we take observations at equal instants of time then

tl+k

M t i = a.e

M t2 = a.e

M t3 = a.e tl+2k

General term M tn = a.e tl+(n " 1)k i.e. the values of output M

forms a G.P. series of increasing order common ratio as e k .

C. Cumulative frequency based information sensing

OBSERVATIONS

INFORMATION INVOLVED

UMmag

13,15

14,15,16

12,13,15

11,12

11,12,13,16

Case 2:

Table 1 : Association of information against each observation

t\ =>M tl = a.e

tl + kl => M f2 =a.e'

tl+kl

t2 + k2

tl+kl+k2

tl + (kl + k2) => M t2 = a.e t2+k2 = M t2 =

tl+Ktotal

i.e.

If we take observation at unequal timing interval in that case

a.e

General term Tn = Tl+(kl+k2+k3+. . .+kn)

Tn = tn-1 + kn-1 = tl + (kl + k2 + k3+...+kn-l) =>

M tn = *.e tn - l+kn - 1 = M tn = a . e ( tl+kl+k2+k3+ - +kn - 1 > = a.e'

now any futuristic value say at instant tn is

M tn = a.e^.e^ '* 1 (observed value)

Given M t = a.e 1 , taking log on both sides we have,

ln(M t ) = ln(a) + 1

i.e. h^Mtn) = ln(a) + tn

ln(M ta ) = ln(a) + tl+kl+k2+k3+...+kn-l

Thus we have obtained a log linear model for the above

function M t = a.e t using which we can calculate or predict the

futuristic values for increased ranges.

Y = m.X + C

If we try to minimize the value of Ktotal we can do so by

making kl=k2=k3=...=kn-l which is same as Case 1.

IV. PROJECTION OF SENSED INFORMATION

Let 1= {ii,i 2 ,...i n } be the set of sensed information. In the
process of feature appropriate observation, forward selection ,
backward elimination and decision based induction methods
are applied.

A. Forward selection based information sensing

Let 1= {ii , i 2 ,....,i n }be the set of information estimates of
various trends noted after observation in respective timing
instants Y = {yi,y2,--.y n }- The accuracy measurement is to be
calculated first based on comparison analysis. The minimum
deviation reflects high accuracy level of prediction and that
information will be selected. In this manner, { } , {best
information} , {first two} . . . .will be selected.

B. Backward elimination based information sensing

Using backward elimination , in each stage each information is
eliminated and thereby after the final screening stage the
projected set reveals the final optimum information space.

Features

Initial
value

Count

Value

(Value) 2

0.1

0.3

0.09

0.2

0.6

0.36

0.3

1.2

1.44

0.4

0.8

0.64

0.5

1.5

2.25

0.6

1.8

3.24

Table 2 : Determination of count and value

Now CF = ( x , y , z )

where x = number of elements , y = linear sum of the elements

and z = sum of the square of the elements [3]

V. BINARY TREE BASED GAIN CLASSIFIER

In this section information represents gain analysis. A
search[4] can be formed based on the initial search term and
its gradual sub term while the process of matching. Thereby
the level is increased, in initial search term is the root and the
final term fully matching with the context of the users' desire
is a leaf node.

Gl,l Gl,2
G2,l G2,2 G2,3 G2,4

A A KK

G3,l G3,2 G3,3G3,4 G3,5 G3,6 G3,7 G3,8

LEVEL

LEVEL 1

LEVEL 2

LEVEL 3

Fig3 : Binary tree based gain classifier

In the above figure, GO is the root that is initial search term. If
a user wants to analyze further gain classification, then
identify each search term as a binary code and by giving the
code number he can analyze the position of gain estimate in
the model . The concept of coding is as follows:

217

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Value = if the search term is a left child of parent node
= 1 otherwise

C4 :

{10}

{11}

N
Theorem: In the process of coding, X l/2 L i =1, where

i=l
Li is the length of code of ith leaf node in the tree, N is total
number of leaf nodes and Ki<N.

Proof:

From fig.3 codes of leaf nodes are as follows:

Nodes

Respective code

G3,l

000

G3,2

001

G3,3

010

G3,4

011

G3,5

100

G3,6

101

G3,7

110

G3,8

111

So, N=8. Each leaf node has identical code length i.e. 3.
Therefore, 1/2^=1/2=1/8, 1/2= 1/8, ...1/2=1/8

We now design a binary tree based classifier taking some
parameters for examination purpose and represent each point
on the basis of a code generated by arithmetic coding. Finally,
represent the same on the basis of set theory .We assume that
the gain set available is G ={ gl,g2,g3,g4 }.The parameters
based on which the examination is to be carried are the
elements of the set P = { pl,p2,p3 }.The result of the
examination are denoted in the form of Boolean variables such
that the outputs are denoted as:
NO =
YES=1

At the initial timing instant, the parameter pi is applied for
testing purpose. Hence, in the initial stage, there will be at
least one class while a maximum of two classes. In the second
level, the parameter p2 is applied and accordingly the classes
are defined. In the final stage, the parameters p3 is applied.
If we assume the classifier as a binary tree representation, we
can apply arithmetic coding to each class such that a 'NO' of a
particular exam is denoted by '0' and a 'YES' is denoted by
T.In the initial stage, the class which contains the elements
for negative supply of pi is CI ={ dl,d2}, while, C2 = { d2,d4
}. In this manner, the tree is to be constructed such that the
code word for each class is denoted by ijk where i C { 0,1 } , j
G{0,1 } and kC{0,l }.

For pi:

Forp2:

CI
C2

Cl =
C2 =

= { dl,d2 }
= { d2,d4 }

{00}
{01}

For p3 :

Class

ijk

False

True

000

pl,p2,p3

001

pl,p2

P 3

010

pl,p3

011

P 2,p3

100

p2,p3

101

pl,p3

110

pl,p2

111

pl,p2,p3

In the initial stage, classes are CI, C2 based on the parameter
pl. In the second stage, the classes are C1,C2,C3 based on p2.
In the last stage, classes are C1,C2...C8 based on p3.This
means that if we assume that 'n' is the number of parameters
involved in the system for examination purpose. Then, the
maximum length of code word for a particular class is 'n'. The
number of classes is 2 n , provided that the classes are distinct in
nature.

VI. CODED INFORMATION SENSING

Let original message is "FATHER". For the first alphabet,
lvalue = Imposition of that)+ n /100). Hence it's offset value =
ceiling of (the product of |u va i ue and 10). The weight is given by
its position in alphabet string[5].

Therefore totalvalue = offset value * weight. From the next
character onwards, |u va i ue _next = l/(mod value of (position of
next - position of previous ) + 7i/100). Hence totalvalue is
calculated in similar manner. Now, bias value will be equal to
total number of characters in the message. Compute netvalue
as (totalvaluefirst char + total valuelast char)- (bias value)
and let it be x (say).

Mode Operation

0<x<100 Reverse the message.

1 00<x<l 50 Circular left shift of message by n/2 bits

where n= bias value.
1 50<x<200 Circular right shift of message by n/2 bits

Iteration 1: |u F = l/((position of 'F' in alphabet list) + 71 /100) =
l/((6)+ 71 /100) = 0.165798547. Offset value = ceiling of
(0.165798547*10) = 2. Weight = position of T' in alphabet list
= 6. Thus, totalvalue = 2*6 =12.

Iteration 2: [i A = l/(|(position of 'A' - position of 'F')|+ 7i/100)
= 1/(|(1-6)|+ tt/100) = 1/5.031415927 = 0.198751209. Offset
value = ceiling of (0.198751209*10) = 2. Weight = 1. Thus,
totalvalue = 2*1 =2.

Iteration 3: \i T = l/(|(position of 'T' - position of 'A')|+ ti/100)
= 1/(|(20-1)|+ tt/100) = 1/19.03141593 = 0.052544697. Offset
value = ceiling of (0.052544697*10) = 1. Weight = 20. Thus,
total value =1*20 = 20.

218

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Comparator

Encryption
system

Output
Cipher

Ci = offset value for i = 1 to n , wi = weight , wb = bias value

Fig 4: Coding Model

Iteration 4: ju H = l/(|(position of 'H' - position of T')|+ tt/100)
= 1/(|(8-20)|+ ti/100) = 1/12.03141593 = 0.083115736. Offset
value = ceiling of (0.083115736*10) = 1. Weight = 8. Thus,
total_value=l*8 = 8.

Iteration 5: \i E = l/(|(position of 'E' - position of 'H')|+ tt/100)
= 1/(|(5-8)|+ tt/100) = 1/3.031415927 = 0.32987885. Offset
value = ceiling of (0.32987885*10) = 4. Weight = 5. Thus,
totalvalue = 4*5 = 20.

Iteration 6: \i R = l/(|(position of 'R' - position of 'E')|+ tt/100)
= 1/(|(18-5)|+ tt/100) = 1/13.03141593 = 0.076737632. Offset
value = ceiling of (0.076737632*10) = 1. Weight = 18. Thus,
total_value=l*18 = 18.

Now, wb= bias value = number of bits in FATHER= 6. So
net_value= accumulated sum of all totalvalue - wb =
(12+2+20+8+20+18) - 6 = 74. It falls in the range 0<x<100.
So, "FATHER" is reversed.

Therefore resultant cipher is "REHTAF".

VII. CONCLUSION

The paper points out information merging in grid and cluster
network models. Statistical means of information prediction as
well as forward, backward and cumulative frequency based
schemes have been analyzed . Binary tree based information
classification and coded information have been justified with
relevant mathematical analysis.

REFERENCES

[1] A.Kumar , P.Chakrabarti , P.Saini , V.Gupta /'Proposed techniques of
random walk, statistical and cluster based node realization" communicated to
IEEE conf. Advances in Computer Science ACS 2010 , India , Dec 10

[2] P.Chakrabarti , S.K.De , S.C.Sikdar , "Statistical Quantification of Gain
Analysis in Strategic Management" published in IJCSNS ,Korea , Vol 9 Nol 1
,pp.3 15-3 18, 2009

[3]P.Chakrabarti, "Data mining- A Mathematical Realization and cryptic
application using variable key" published in International journal , Advances
in Information Mining , Vol 2 No 1, pp- 18-22,20 10

[4] P.Chakrabarti, P.S.Goswami, "Approach towards realizing resource
mining and secured information transfer" published in international journal of
IJCSNS, Korea , Vol 8 No.7, pp345-350, 2008

[5] P.Chakrabarti , "Attacking Attackers in Relevant to Information Security"
Proceedings of RIMT-IET, Mandi Gobindgarh. pp 69-71, March 29, 2008

About authors:

Debasis Mukherjee (20/08/80) is pursuing Ph.D. from USIT,
GGSIPU, Delhi, India from2010. He received the M. Tech.
degree in VLSI Design from CDAC Noida in 2008 and
bachelor degree in Electronics and Instrumentation
Engineering from BUIE, Bankura, West Bengal, India in
2003 .He achieved first place in district in "Science Talent
Search Test" 1991. He has some publications of repute in
IEEE conferences.

Dr.P.Chakrabarti(09/03/81) is currently serving as Associate
Professor in the department of Computer Science and
Engineering of Sir Padampat Singhania University,Udaipur.
Previously he worked at Bengal Institute of Technology and
Management , Oriental Institute of Science and Technology,
Dr. B.C. Roy Engineering College, Heritage Institute of
Technology, Sammilani College. He obtained his Ph.D(Engg)
degree from Jadavpur University in Sep09,did M.E. in
Computer Science and Engineering in 2005, Executive MBA
in 2008and B.Tech in Computer Science and Engineering in
2003 .He is a life member of Indian Science Congress
Association , Calcutta Mathematical Society , Calcutta

219

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Statistical Association , Indian Society for Technical
Education , Cryptology Research Society of India,
IAENG(HongKong), CSTA(USA), annual member of
Computer Society of India, VLSI Society of India ,
IEEE(USA), senior member of IACSIT(Singapore) and
selected member of The IAENG Society of Artificial
Intelligence , Computer Science , Data Mining. He is a
Reviewer of International journal of Information Processing
and Management (Elsevier) , International Journal of
Computers and Applications , Canada and International
Journal of Computer Science and Information
Security(IJCSIS,USA), editorial board member of
International Journal of Engineering and Technology,
Singapore and International Journal of Computer and
Electrical Engineering. He has about 100 papers in national
and international journals and conferences in his credit and
two patents(filed). He has several visiting assignments at BHU
Varanasi , IIT Kharagpur , Amity University,Kolkata , et al.

A.Khanna and V.Gupta are the third year students of
Information Technology and Computer Science & Engg.
branch respectively of Sir Padampat Singhania University.

220 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Scaling Apriori for Association Rule Mining
using Mutual Information Based Entropy

S.Prakash,Research Scholar
Sasurie College of Engineering
Vij ayamangalam,Erode(DT)
Tamilnadu, India.Ph.09942650818
Mail:prakash_ant2002@yahoo.co.in

Abstract - Extracting information from large datasets is
a well-studied research problem. As larger and larger
data sets become available (e.g., from customer
behavior data from organizations such as Wal-Mart) it
is getting essential to find better ways to extract
relations (inferences) from them. This thesis proposes
an improved Apriori algorithm to minimize the number
of candidate sets while generating association rules by
evaluating quantitative information associated with
each item that occurs in a transaction, which was
usually, discarded as traditional association rules focus
just on qualitative correlations. The proposed approach
reduces not only the number of item sets generated but
also the overall execution time of the algorithm. Any
valued attribute will be treated as quantitative and will
be used to derive the quantitative association rules
which usually increases the rules' information content.
Transaction reduction is achieved by discarding the
transactions that does not contain any frequent item set
in subsequent scans which in turn reduces overall
execution time. Dynamic item set counting is done by
adding new candidate item sets only when all of their
subsets are estimated to be frequent. The frequent item
ranges are the basis for generating higher order item
ranges using Apriori algorithm. During each iteration
of the algorithm, use the frequent sets from the previous
iteration to generate the candidate sets and check
whether their support is above the threshold. The set of
candidate sets found is pruned by a strategy that
discards sets which contain infrequent subsets. The
thesis evaluate the scalability of the algorithm by
considering transaction time, number of item sets used
in the transaction and memory utilization. Quantitative
association rules can be used in several domains where
the traditional approach is employed. The unique
requirement for such use is to have a semantic
connection between the components of the item-value
pairs. The proposal used mutual information based on
entropy to generate association rules from non-
biological datasets.

Key words- Apriori, Quantitative attribute, Entropy

Dr.R.M.S.Parvathi M.E.(CSE),Ph.D.

Principal

Sengunthar College of Engg.for Women
Tiruchengode. Tamilnadu, India
rmsparvathi @ india.com

I. INTRODUCTION

Data mining, also known as knowledge
discovery in databases, has been recognized as a new
area for dataset research. The problem of discovering
association rules was introduced in latter stages.
Given a set of transactions, where each transaction is
a set of items, an association rule is an expression of
the from X + Y, where X and Y are sets of items. The
problem is to find all association rules that satisfy
user- specified minimum support and minimum
confidence constraints.

Conceptually, this problem can be viewed as
finding associations between the "1" values in a
relational table where all the attributes are Boolean.
The table has an attribute corresponding to each item
and a record corresponding to each transaction. The
value of an attribute for a given record is "1" if the
item corresponding to the attribute is present in the
transaction corresponding to the record, "O" else.

Relational tables in most business and
scientific domains have richer attribute types.
Attributes can be quantitative (e.g. age, income) or
categorical (e.g. zip code, make of car). Boolean
attributes can be considered a special case of
categorical attributes. This thesis defines the problem
of mining association rules over quantitative attribute
in large relational tables and present techniques for
discovering such rules. This is referred as the
Quantitative Association Rules problem.

The problem of mining association rules in
categorical data presented in customer transactions
was introduced by Agrawal, Imielinski and Swami
[1][2]. This thesis work provided basic idea to several
investigation efforts [4] resulting in descriptions of
how to extend the original concepts and how to
increase the performance of the related algorithms.

221

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

The original problem of mining association
rules was formulated as how to find rules of the form
setl -> set2. This rule is supposed to denote affinity
or correlation among the two sets containing nominal
or ordinal data items. More specifically, such an
association rule should translate the following
meaning: customers that buy the products in setl also
buy the products in set2. Statistical basis is
represented in the form of minimum support and
confidence measures of these rules with respect to the
set of customer transactions.

The original problem as proposed by
Agrawal et al.[2] was extended in several directions
such as adding or replacing the confidence and
support by other measures, or filtering the rules
during or after generation, or including quantitative
attributes. Srikant and Agrawal describe a new
approach where quantitative data can be treated as
categorical. This is very important since otherwise
part of the customer transaction information is
discarded.

Whenever an extension is proposed it must
be checked in terms of its performance. The
algorithm efficiency is linked to the size of the
dataset that is amenable to be treated. Therefore it is
crucial to have efficient algorithms that enable us to
examine and extract valuable decision-making
information in the ever larger databases.

This thesis present an algorithm that can be
used in the context of several of the extensions
provided in the literature but at the same time
preserves its performance. The approach in our
algorithm is to explore multidimensional properties
of the data (provided such properties are present),
allowing to combine this additional information in a
very efficient pruning phase. This results in a very
flexible and efficient algorithm that was used with
success in several experiments using quantitative
databases with performance measure done on the
memory utilization during the transactional pruning
of the record sets.

II. LITERATURE REVIEW

Various proposals for mining association
rules from transaction data were presented on
different context. Some of these proposals are
constraint-based in the sense that all rules must fulfill
a predefined set of conditions, such as support and
confidence [6,7,8]. The second class identify just the
most interesting rules (or optimal) in accordance to
some interestingness metric, including confidence,

support, gain, chi-squared value, gini, entropy gain,
laplace, lift, and conviction [9,6]. However, the main
goal common to all of these algorithms is to reduce
the number of generated rules.

A)Existing Scheme

The thesis extend the first group of
techniques since it do not relax any set of conditions
nor employ a interestingness criteria to sort the
generated rules. In this context, many algorithms for
efficient generation of frequent item sets have been
proposed in the literature since the problem was first
introduced in [10]. The DHP algorithm [11] uses a
hash table in pass k to perform efficient pruning of
(&+7)-item sets. The Partition algorithm minimizes
I/O by scanning the dataset only twice. In the first
pass it generates the set of all potentially frequent
item sets, and in the second pass the support for all
these is measured. The above algorithm are all
specialized techniques which do not use any dataset
operations. Algorithms using only general purpose
DBMS systems and relational algebra operations
have also been proposed [9.10].

Few other works trying to solve this mining
problem for quantitative attributes. In [5], the authors
proposed an algorithm which is an adaptation of the
Apriori algorithm for quantitative attributes. It
partitions each quantitative attribute into consecutive
intervals using equi-depth bins. Then adjacent
intervals may be combined to form new intervals in a
controlled manner. From these intervals, frequent
item sets (c.f. large item sets in Apriori Algorithm)
will then be identified.

Association rules will be generated
accordingly. The problems with this approach is that
the number of possible interval combinations grows
exponentially as the number of quantitative attributes
increases, so it is not easy to extend the algorithm to
higher dimensional cases. Besides, the set of rules
generated may consist of redundant rules for which
they present a "greater-than-expected-value" interest
measure to identify the interesting ones.

Some other efforts that exploit quantitative
information present in transactions for generating
association rules[12]. In [5], the quantitative rules are
generated by discrediting the occurrence values of an
attribute in fixed-length intervals and applying the
standard Apriori algorithm for generating association
rules. However, although simple, the rules generated
by this approach may not be intuitive, mainly when
there are semantic intervals that do not match the
partition employed.

222

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Other authors [5] proposed novel solutions
that minimize this problem by considering the
distance among item quantities for delimiting the
intervals, that is, their "physical"' placement, but not
the frequency of occurrence as a relevance metric.

To visualize the information in the massive
tables of quantitative measurements we plan to use
clustering and mutual information based on entropy.
Clustering is an old studied technique used to extract
this information from customer behavior data sets.
This follows from the fact that related customer
purchase through word of mouth have similar
patterns of customer behavior. Clustering groups
records that are "similar" in the same group. It suffers
from two major defects. It does not tell you how the
two customer buyin behavior/clusters are exactly
related. Moreover, it gives you a global picture and
any relation at a local level can be lost.

B) Proposed Scheme

The proposed scheme comprises of two
phases. The first phase of the thesis concerns about
the quantitative association rule mining with the
enhancement on Apriori algorithm. The second phase
of the thesis deals with the reduction of memory
utilization during the pruning phase of the
transactional execution.

The algorithm for generating quantitative
association rules starts by counting the item ranges in
the data set, in order to determine the frequent ones.
These frequent item ranges are the basis for
generating higher order item ranges using an
algorithm similar to Apriori. Take into account the
size of a transaction as the number of items that it
comprises.

a) Define an item set m as a set of items of size m

b) Specify frequent (large) item sets by Fm

c) Specify candidate item sets (possibly frequent) by
Lm.

A n range set is a set of n- item ranges, and
each m-item set has a n-range set that stores the
quantitative rules of the item set. During each
iteration of the algorithm, the system uses the
frequent sets from the previous iteration to generate
the candidate sets and check whether their support is
above the threshold. The set of candidate sets found
is pruned by a strategy that discards sets which
contain infrequent subsets. The algorithm ends when
there are no more candidates' sets to be verified.

The enhancement of Apriori is done by
increasing the efficiency of candidate pruning phase
by reducing the number of candidates that are
generated to further verification. The proposed
algorithm use quantitative information to estimate
more precisely the overlap in terms of transactions.
The basic elements considered in the development of
the algorithm are number of transactions, average
size of transaction, average size of the maximal large
item sets, number of items, and distribution of
occurrences of large item sets.

The second phase of the thesis claimed
improvement over A priori by considering memory
consumption for data transaction. This part of the
algorithm generate all candidates based on 2-frequent
item sets on sorted dataset and already generates all
frequent item sets that can no longer be supported by
transactions that still have to be processed. Thus the
new algorithm no longer has to maintain the covers
of all past item sets into main memory. In this way,
The proposed level-wise algorithms accesses a
dataset less often than Apriori and require less
memory because of the utilization of additional
upward closure properties.

C)Mutual Information based entropy

The mutual information I (X, Y ) measures
how much (on average) the realization of random
variable Y tells us about the realization of X, i.e.,
how by how much the entropy of X is reduced if we
know the realization of Y .

I(X; Y ) = H(X) - H(XIY )

For example, the mutual information between a cue
and the environment indicates us how much on
average the cue tells us about the environment. The
mutual information between a spike train and a
sensory input tells us how much the spike train tells
us about the sensory input. If the cue is perfectly
informative, if it tells us everything about the
environment and nothing extra, then the mutual
information between cue and environment is simply
the entropy of the environment:

I(X; Y ) = H(X) - H(XIY ) = H(X) - H(XIX) = H(X).

In other words, the mutual information
between a random variable and itself is simply its
entropy: I(X;X) = H(X). Surprisingly, mutual
information is symmetric; X tells us exactly as much
about Y as Y tells us about X.

223

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

III. QUANTITATIVE ASSOCIATION

RULE MINING - MUTUAL
INFORMATION BASED ENTROPY

The proposal of this work use mutual
information based on entropy for generating
quantitative association rules. Apart from the usual
positive correlations between the customers, this
criterion would also discover association rules with
negative correlations in the data sets. It is expected to
find results of the form attrib 1/ attrib 2 -> A attrib3,
which can be interpreted as follows: Attrib 1 and
Attrib2 are co expressed and have silencing effect on
attrib 3. Then compare the results from our
experiments to those obtained from clustering.

First tune various parameters (like support,
support fraction, significance level), of the auto
performance dataset. This was because even with
binary data, 2468 attributes may lead to the power(2,
2468) relations (which the software was not designed
to handle). Here, it is needed to know that for the
problem under consideration, the auto are attributes,
as needed to find relationships among them. To
overcome this problem used another approach, in
which data attributes were already known to be
related, using the results obtained from clustering.
This decreases the number of attributes to
manageable levels (both for program). The proposed
work used the approach above to find the
relationships (positive, negative) among the
attributes.

Algorithm Steps

a. Find all frequent item sets (i.e., satisfy
minimum support)

b. Generate strong association rules from the
frequent item sets (each rule must satisfy minimum
support and minimum confidence).

c. Identify the quantitative elements

d. Sorting the item sets based on the
frequency and quantitative elements

e. Merge the more associated rules of item
pairs

f. Discard the infrequent item value pairs

g. Iterate the steps c to f till the required
mining results are achieved

Let I = { il, i2 ... im} be a set of items, and
T a set of transactions, each a subset of I. An
association rule is an implication of the form A=>B,
where A and B are non-intersecting The support of
A=>B is the percentage of the transactions that
contain both A and B . The confidence of A=>B is the
percentage of transactions containing A that also

contain B (interpret as P(BIA)). The occurrence
frequency of an item set is the number of transactions
that contain the item set.

IV. IMPLEMENTATION OF QUANTITATIVE
APRIORI

The function op is an associative and
commutative function. Thus, the iterations of the
foreach loop can be performed in any order. The
data- structure Reduc is referred to as the reduction
object. The main correctness challenge in
parallelizing a attribute like this on a shared memory
machine arises because of possible race conditions
when multiple processors update the same element of
the reduction object.

The element of the reduction object that is
updated in a loop iteration is determined only as a
result of the processing. In the a priori association
mining algorithm, the data item read needs to be
matched against all candidates to determine the set of
candidates whose counts will be incremented. The
major factors that make these loops challenging to
execute efficiently and correctly are as follows:

It is not possible to statically partition the
reduction object so that different processors update
disjoint portions of the collection. Thus, race
conditions must be avoided at runtime.

The execution time of the function process
can be a significant part of the execution time of an
iteration of the loop. Thus, runtime preprocessing or
scheduling techniques cannot be applied.

The updates to the reduction object are fine
grained. The reduction object comprises a large
number of elements that take only a few bytes, and
the for each loop comprises a large number of
iterations, each of which may take only a small
number of cycles.

{ * Outer Sequential Loop *}
While() {

{* Reduction Loop*}
Foreach(element e ) {
( i, val) = process (e);
Reduc (i) = Reduc (i) op val;

Fig 1 : Pseudo code

The consumer behavior auto databases
obtained data from UCI Machine Learning
Repository. The data obtained was about CPU-
performance and automobile mileage. The data was
discretized into binary values. For these data sets the

224

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

discretization was done in accordance with
interpretation required. This discretization was done
automatically using the written software. This
software also formatted the data into the format
required by the program. A finer level of
discretization (or supporting the real values) would
have been more appropriate, but the used approach
also gave much of the useful results.

V. EXPERIMENTAL RESULTS FROM APRIORI

The process of executing quantitative
Association rule mining for the auto manufacturer
power evaluation data is given below

a) Get data (file: auto-mpg.data): 9 attributes, 398
samples.

b) Remove unique attributes (IDs). Here, car_name
attribute has been removed.

c) Remove those samples (total 5) that contain "?"
(missing data) as a value for some of their attributes
(so, we are left with 8 attributes and 393 samples).

d) Discretize real-valued attributes based on their
average values (which is (maximum attribute value +
minimum attribute value) / 2)

e) Run the program to generate association rules
using mutual information based on entropy metric.

The experiment focused on evaluating all
quantitative a priori techniques. Since we were
interested in seeing the best performance, we used
banking data set samples. We used a 1 GB dataset. A
confidence of 90% and support of 0.5 is used.

Execution times using 1, 2, 3, and 4 threads
are presented on the processor. With 1 thread, Apriori
does not have any significant overheads as compared
to the sequential version. Therefore, this version is
used for reporting all speedup s. Though the
performance of quantitative a priori is considerably
lower than a priori, they are promising for the cases
when sufficient memory for supporting full
replication may not be available. We consider four
support levels, 0.1%, 0.05%, 0.03%, and 0.02%. The
execution time efficiency is improved for the
quantitative a priori on frequent item set evaluation
with the support count (Graph 1).

1 10

i.i.

1.4

1.2

a ■]

il.rt

II II

rmir

1 1 ofl

1 1

■'.I.-

11.11=..

u.u.?s

' '.' V :,

Support LevGl

Graph 1 : Support VS Time on Quantitative and qualitative A priori

The thread execution on the quantitative a
priori and qualitative a priori are evaluated for the
same data set (Graph 2). Here the initial thread
requires more time, however consequent threads
shows better scalable performance of quantitative
Apriori.

■i

.*;.■■■ ■

.-. 2500

if!
■I'

h 2000

;,.„,

fr-

man

1 1

fr-int

Dfl-

man

□ afl-

int

i i

csl

-man

■

■il

-int

1 thread

'■ WW- ■■)■[■

4 threads

No. of threads

Graph 2: Thread Vs Time for a priori execution on rule set
generation

225

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

By observing the output from the program it is seen
that a few relationships between the attributes had
high values of mutual information. Namely, the
highest Mi-values were obtained for:

a) displacement and horsepower. Further, by
observing the entropy values we may notice that
there are very few cars that have small displacement
and high horsepower.

b) displacement and weight. Further, by observing
the entropy values we may notice that there are very
few cars that have large displacement and light
weight.

c) cylinders and weight. Further, by observing the
entropy values we may notice that there are very few
cars that have small number of cylinders and heavy
weight.

d) horsepower and weight. Further, by observing the
entropy values we may notice that there are very few
cars that have large horsepower but heavy weight.

VI. CONCLUSION

The thesis have defined a new a rule set
namely the informative rule set that presents
prediction sequences equal to those presented by the
association rule set using the confidence priority. The
informative rule set is significantly smaller than the
association rule set, be especially when the minimum
support is small.

The proposed method has some merit in
extracting information from huge data sets by
pruning the initial information (to bring it down to
the manageable levels) and then finding the
association rules among the attributes. Further, the
approach is used to predict the relationships among
the silencer auto power, weight, model, year etc.,
could be extended to unknown function.

The proposed scheme have characterized the
relationships between the informative rule set and the
non-redundant association rule set, and revealed that
the informative rule set is a subset of the non-
redundant association rule set. The thesis considers
the upward closure properties of informative rule set
for omission of uninformative association rules, and
presented a direct algorithm to efficiently generate
the informative rule set without generating all
frequent item sets.

efficiency improvement results from that the
generation of the informative rule set needs fewer
candidates and dataset accesses than that of the
association rule set rather than large memory usage
like some other efficient algorithms.

REFERENCES

[I] R. Agrawal, T. Imielinski, and A. Swami. Dataset mining: A
performance perspective. In IEEE Transactions on Knowlegde and
Data Engineering, December 1993.

[2] R. Agrawal, T. Imielinski, and A. Swami. Mining association
rules between sets of items in large databases. In Proc. of the
ACM SIGMOD Washington, D.C., pages 207-216, May 1993.
[3] R. Miller and Y. Yang. Association rules over interval data. In
ACM SIGMOD Conference, Tucson, Arizona, pages 452 - 461,
May 1997.

[4] J. Park, M. Chen, and P. Yu. An effective hash based algorithm
for mining associative rules. In ACM SIGMOD Conference, San
Jose, CA, pages 175 - 186, May 1995.

[5] R. Srikant and R. Agrawal. Mining quantitative association
rules in large relational tables. In Proceedings of the ACM

SIGMOD Conference on Management of Data, pages 1-12,

Montreal, Canada, June, 1996.

[6] R. Agrawal, T. Imielinski, and A. Swami. Dataset mining: A
performance perspective. In IEEE Transactions on Knowlegde and
Data Engineering, December 1993.

[7] R. Agrawal, H. Mannila, R. Srikant, H. Toivonen, and
A.Verkamo. Fast discovery of association rules. In Advances in
Knowledge Discovery and Data Mining, San Jose, CA, pages
307-328,1996.

[8] R. Bayardo, R. Agrawal, and D. Gunopulos. Constraint- based
rule mining in large, dense databases. In 15th International

Conference on Data Engineering, Sydney, Australia, pages 188
- 197, March 1999.

[9] R. Bayardo and R. Agrawal. Mining the most interesting rules.
In 5th ACM SIGKDD International Conference on Knowledge,
San Diego, CA, Pages 145 - 154, August 1999.
[10] R. Agrawal, T. Imielinski, and A. Swami. Mining association
rules between sets of items in large databases. In Proc. of the

ACM SIGMOD Washington, D.C., pages 207-216, May 1993.

[II] J. Park, M. Chen, and P. Yu. An effective hash based
algorithm for mining associative rules. In ACM SIGMOD
Conference, San Jose, CA, pages 175 - 186, May 1995.

[12] Prakash.S and R.M.S.Parvathi. " Scaling Apriori for
Association Rule Mining Efficiency ".Proceedings of the Fourth
International Conference, Amrutvani College of Engineering,
Sangamner,Maharatra.pp 29, March 2009.

The informative rule set generated in this
thesis is significantly smaller than both the
association rule set and the non-redundant association
rule set for a given dataset that can be generated more
efficiently than the association rule set. The

226

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

AUTHORS PROFILE

Prof. S. Prakash has completed his M.E.,
(Computer Science and Engineering) in
K.S.R.College of Technology , Tamilnadu,
India in 2006. Now he is doing research in
the field of Association Rule Mining
algorithms. Currently, he is working as
Assistant Professor in the department of
Information Technology, Sasurie College of Engineering,
and Tamilnadu. India. He has completed 9 years of
teaching service.

Dr. R.M.S. Parvathi has completed her
Ph.D., degree in Computer Science and
Engineering in 2005 in Bharathiar
University, Tamilnadu, India. Currently she
is a Principal and Professor , Department of
CSE in Sengunthar College of Engineering
for Women, Tamilnadu, India, She has completed 20 years
of teaching service. She has published more than 28 articles
in International / National Journals. She has authorized 3
books with reputed publishers. She is guiding 20 Research
scholars. Her research areas of interest are Software
Engineering, Data Mining, Knowledge Engineering, and
Object Oriented System Design.

227

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Clustering of High- Volume Data Streams
In Network Traffic

M.Vijayakumar

Department of Computer Science and Engineering

Sasurie College of Engineering,

Vijayamangalam, Erode(Dt), Tamilnadu, India.

tovijayakumar @ gmail.com

Dr.R.M.S.ParvathiM.E.(CSE),Ph.D.

Principal,

Sengunthar College of Engineering, for Women,

Tiruchengode,Tamilnadu, India.

rmsparvathi @ india.com

Abstract — The thesis concerned with the problem of
mining network traffic data discovering useful
associations, relationships, and groupings in large
collections of data. Mathematical transformation
algorithms have proven effective at reducing the content
of multilingual, unstructured data into a vector that
describes the content. Such methods are particularly
desirable in fields undergoing information explosions,
such as network traffic analysis, bio-informatics, and the
intelligence community. In response, traffic mining
methodology is being extended to improve performance
and sufficiently scalable.The usage of data flow collected
from site routers for various analysis i.e., network
performance characterization, investigating computer
security incidents and their prevention, network traffic
statistics, and others. Currently, the data flow analysis is
built as a distributed system to collect data from multiple
routers, both at the edge of the site network as well as
from local routers and multilayer switches. Average per
day volume is about 2GBytes of raw data. Despite a high
volume of collected information, some analysis is
conducted in near real time to satisfy demands of users
communities for quick results.The proposed work present
an efficient clustering means to analyze experimental
results for traffic data streams nature (symmetric and
asymmetric). As summary, this paper describe a system
designed to satisfy three primary goals i.e., real-time
concept mining of high-volume data streams, dynamic
data flow into a relational hierarchy; and adaptive
reorganization of the traffic data hierarchy in response to
evolving circumstances and network traffic time to time.
The proposed clustering network traffic data flow
collection and analysis system describe traffic
characterization and network performance estimation for
the data flow centre. The system checks the traffic
consistency for End To End circuits and Policy Based
Routing and finally, profiling of host's traffic to keep
track of their typical behavior to prevent accidental
blocking by site IDS system.

Keywords — Traffic analysis, network management ,
clustering, frequent Item set , hierarchical clustering.

I. INTRODUCTION

As part of an ongoing research project,
developed a novel algorithmic approach for extracting
traffic data from voluminous data streams[l][14]. The
approach is applicable to internet data servers, which
can be automatically identified and converted into a
common structure.

Here, report on an extension of the traffic data
clustering work, which clusters data flow hierarchically
based on the nature requested by users, (demand, load).
The new method represents the streaming hierarchical
clustering of documents [2] [18], which can be seen as a
subfield of the nascent discipline of "streaming AI",
"evolutionary clustering", or "AI in hardware".

The proposed system is a High Speed Content
classification system that works in three stages to
classify flows of TCP traffic. A TCP flow is half of a
TCP conversation. A connection from a client to a Mail
server with SMTP (simple mail transfer protocol) is an
example of a flow[4][26]. The connection from the
Mail server back to the client is considered a separate
flow. It extracts words then builds a vector
representation and then scores against known nature of
the data flow. The scores of the completed flows are
passed out of the system for evaluation.

The base data list from stage one used for
counting in stage two. Each dimension is represented
by 4bits[3]. Counting for a dimension saturates at 15.
When the flow ends the count vector representing the
flow is passed on to stage three. In the third stage a
vector representing a flow is scored against vectors
representing known traffic data[10][24][25]. The data
vectors are called the Score Table (ST) and are
reconfigurable at run time like the WMT. The ST is
derived from a set of documents. The output of the

228

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

system is a set of scores and the count array of the
flow.

Evaluation of the scores determines the
classification of the flow against the known traffic data.
However, simply classifying the flow as the data with
greatest score is not adequate. A forced classification
of all flows will be undesirable in most applications. A
threshold provides a confidence level to the
classification of flows. Any traffic data that is not
classified is considered unknown to the system.
Clustering these unclassified flows is the focus of the
Streaming Clustering work in this thesis.

II. HIERARCHICAL PARTITIONING

The notion that natural categories tend to be
organized hierarchically as a general principle dates
back at least to Aristotle, and has been refined by
Simon[5][ll][12]. Considering that libraries,
governments, Internet newsgroups and taxonomies, and
the human neocortex all organize and process
information hierarchically, it is certainly a natural
methodology for organizing unknown content. A
logical assumption for document clustering is that like
documents will group together. Groupings with large
number of items are more general.

As the number of items starts to dwindle, the
topic of a grouping becomes more specific. For
example, the topic of "cars" is very general, while
"Cadillac" is more specific and would be part of the
general topic of "cars." Two standard approaches can
be taken to the problem organizing a collection of
documents represented as fixed length vectors of high
dimensionality hierarchically agglomerative (bottom-
up), and divisive (top-down) [6] [13] [22].

Agglomerative -Cluster Algorithm

Assign each document to a unique cluster

While more than one cluster exists

Merge the two closest clusters

Return the single remaining cluster as the root

Divisive-Cluster Algorithm

If the collection contains a single traffic data

Return it in a single cluster

Divide the collection into two parts

Call divisive-cluster on each part

The final result for both procedures is a binary
tree containing a single data in every leaf, where close
leaves (measured by tree-traversal distance) are related.
Both methods have their general advantages and
drawbacks, but for the specific problem of data

clustering where traffic data are represented via a flow
sequence approach, empirical studies have shown that
the top-down divisive approach produces consistently
superior results.

The generally accepted explanation for these
results is that local neighborhood information (which
pairs of documents contain most similar word
distributions) is often misleading, deceiving
agglomerative clustering into making bad merge
decisions early on. Divisive clustering can often avoid
these mistakes by first considering the global statistics
of collections of traffic data, which are more robust.
Thus, a divisive hierarchical clustering approach is
expected to give us the best results; the remaining
decision that must be made is how to divide a
collection of documents.

The centroid gives an indication of the overall
makeup of a cluster, and is used in many clustering
algorithms. Assuming that our set of documents, V,
represents some cohesive grouping (e.g., a collection of
postings from a single internet newsgroup), its centroid
provides us with an indication of what concepts the
"average" document refers to the value in some
dimension will be between zero and one, and denote
the probability of a random document from the
collection scoring a hit in that dimension. It is natural
to consider measuring the affinity of a traffic data to a
given collection by comparing its distribution of hits to
our expectations normalized by data flow vector
magnitude, as data with higher magnitudes have more
opportunities for hits.

The heuristic approach that is taken to attempt
this is quite simple. A set of data- vectors is randomly
partitioned (a fair coin is flipped for each document to
assign it to either cluster one or cluster two). Data are
iteratively transferred between the two clusters, one at
a time, as long as doing so strictly increases the
division quality. Requiring a strict increase in division
quality ensures that the partitioning procedure is
guaranteed to terminate [7] [22].

The system will compare hierarchical
partitioning to flat and hierarchical variants of the
popular k- means clustering algorithm[8], and to a
baseline hierarchical clustering algorithm (bisection k-
means ) from the synthetic dataset. These traffic data
sets consist largely of flow of data at the network
points, demand, and load etc.,

III. K-MEANS CLUSTERING

The k-means clustering algorithm separates
input data into K groups. The number of groups, or K,
is set prior to running the clustering algorithm. Each
document in the data is assigned to a cluster. These
assignments are used to calculate the cluster centroid or
center[9][21][23]. The cluster centroids are then

229

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

utilized to determine the distance between each
centroid and a data element. The algorithm seeks to
minimize the inner cluster distance (i.e. form tight
groups of similar data) and minimize the inter cluster
distance (i.e. the groupings are non-overlapping).

The distance calculation can be performed in
any number of ways. Three common methods of
distance calculation include the Minkowski,
Manhattan, and Euclidean Distance metrics. The cosine
theta distance has also been used with k-means in order
to cluster high dimensional data. The algorithm is a
cyclical algorithm that performs in the following
manner

a) Initially assign document in the data to K groups

b) Calculate the cluster centroids based on assignments

c) For each document in the data

i) Recalculate distances from document to all
centroids and find closest centroid

ii) Change document assignment to closest centroid
and update the centroids that the document
used to reside and currently resides

d) Repeat step 3 until either no changes are made to
document assignments or the epoch limit is reached.

Bisection k-means is a variant of the k-
means algorithm. It starts with a single cluster and
continually selects a cluster to split into 2 sub-clusters
until the requested number of clusters is achieved.

a) Pick a cluster to split.

b) From the selected cluster, use k-means to cluster
the elements into 2 sub clusters.

c) Repeat steps 1 and 2 until the desired number of
clusters is found.

The selection of the cluster to bisect can be
done in a number of ways. For choosing the cluster
with the most elements was sufficient to find good
clustering, use the same heuristic here. In order to
objectively compare the results of a hierarchical
clustering algorithm to a flat clustering, need a means
of automatically flattening a full binary cluster tree to a
set of k clusters (given some particular k). A simple
heuristic for doing so is to choose the (non-
overlapping) subtrees that make the highest quality
clusters, as defined in the previous section. So for k=2,
will choose the left and right subtrees of the root. For
k=3 we will take the lowest scoring of our current
clusters (that is not already a leaf) and expand it into
two clusters by replacing it with its two children (recall

that divisive hierarchical clustering always produces a
binary tree). For bisection k-means, stop tree creation
after k leaves have been formed and take them as our
clusters.

The ground-truth newsgroups (horizontal axis)
are ordered so that the far right column of the
confusion matrix is the chaff, and the ordering of the
remaining newsgroups is arbitrary. The clusters
(vertical axis) are ordered based by their most frequent
newsgroup, with color showing purity (blue is lowest,
red highest). A perfect clustering would hence be
denoted by a crisp diagonal red line. K-means is clearly
inferior to the two hierarchical approaches, placing the
majority of the documents into two large clusters (the
two horizontal red lines near the bottom of the plot).
Bisection k-means and hierarchical partitioning
produce comparable results; however half of the
clusters created by bisection k-means are dominated by
chaff, whereas hierarchical partitioning creates only ten
such "junk" clusters. This is preferable from a human
analyst's point of view, as it allows uninteresting
document sets to be identified and discarded more
quickly.

IV. EXPERIMENTAL RESULTS

This part of the thesis describes experimental
results with the streaming hierarchical partitioning
algorithm. Results are presented on ISP data, streamed
according to a regime designed to simulate nature of
the traffic data are randomly collected, but to begin,
only data from half of the ISP server are evaluated. At
uniform intervals, a ISP is gradually introduced into the
distribution (and hence the old newsgroup density
gradually reduced).

The traffic flow appears uniformly throughout
the entire data stream. In order to run streaming
hierarchical partitioning clustering, need to set two
parameters, the maximal number of traffic data that are
capable of collecting at a time, and how often the set of
data flowing in the server working memory will be
reclustered. In actual deployment of course, will want
to store as many traffic data as possible in fast memory,
and recluster as often as possible, and the bandwidth of
the data stream. Given that our dataset contains about
10GB of data should prove illustrative.

To analyze the quality of hierarchical
streaming clustering, there are two basic factors
consider, traffic quality as in the non- streaming case,
how meaningful are the clustered data discovered by
the algorithm, and traffic discovery as drift occurs ,
does the algorithm effectively identify new flow of
traffic data.

In a controlled experiment, where know the
ground-truth labeling of the data, traffic quality can be
measured by considering how many non-chaff data are

230

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

assigned to clusters that are nearly "pure" (at least 90%
of the documents originating in a single newsgroup).
Furthermore, only consider clusters with more than 10
documents in computing this measure, henceforth
referred to as a purity score.

Similarly, traffic discovery over time can be
measured by considering, at any given time, how many
pure clusters have been created corresponding to
unique non-chaff labels, up until this time. The
measure is hence cumulative, and will henceforth be
referred to as a discovery score. In order to understand
how effective the data insertion and removal heuristics
are at augmenting hierarchical partitioning for
streaming clustering. In this procedure, full hierarchical
partitioning clustering will still be carried out at the
same regular intervals. However, when new data are
congested, traffic will be chosen entirely at random to
be removed. This methodology will henceforth be
referred to as naive streaming clustering.

In order to provide a fair comparison between
naive and non-naive streaming clustering, scores are
only computed immediately after batch clustering has
been completed. That is, naive streaming clustering is
not penalized for having poor results in between batch
clustering. Both methods are equally effective in terms
of purity scores as expected, since this score is
essentially determined by the batch clustering, which is
identical.

V. PERFORMANCE EVALUATION

Analysis of flow data gathered from multiple
routers is the source of valuable information for various
tasks. We distinguish three major areas where that
information is used. These areas are computer security,
performance analysis for data movement applications
and verification of traffic consistence across site
network infrastructure. We have considerable
experience to use flow data in two first areas. The last
one, verification of traffic consistency is relatively new
to us. It is motivated by recent deployment End To End
circuits for LHC/CMS experiment. These circuits have
predictable performance characteristics and dynamic
capabilities to use alternative WAN path available for
the data centre.

Traffic is analyzed for different time intervals,
typically lmin, 5min, 15min, lhour, or 1 day. A start
date and time can be also specified to generate historical
results. Tagging of raw data can be done at any time, but
for most applications, it runs when new flow sets are
created and become available on NAS server. New
intervals can be added as needed. The next level is the
target, which represents defined list of IP blocks assigned
to particular users group or community. The system Tier
1 LAN is an example of a user group address block.

ill Data

Actual Traffic

Monthly

Daily

Monthly

Bord9r

600MB

17GB

15IBP)

300TB PTB)

StaUi

1MB

BOTPTB)

upep)

1,2GB

30GB

8GTB(120TB)

1.5PB(2.4PB)

3GB

40GB

Table 1 : Volumes of flow data and actual traffic

A target's traffic is going to be inspected based on
set of predefined filters. For example, analyze appropriate
traffic to specific destinations. For each target there may be
several filters defined. Filters have unique names and are
stored in a database with time stamp allowing to track
changes. While inspecting traffic, we may use some
specific names for destinations or sources. For example,
we don't need to known all details of traffic to each node
of the destination cluster. Instead, we need to know traffic
to whole cluster, and use a predefined name for the address
block. The next level is the static definition of rules
specified in separated XML files. This level is identified
by name of XML configuration file. The content of
configuration files may be changed over time, due
different reasons, i.e. readdressing of nodes, moving them
to different subnets and so on. To track these changes, the
content of configuration files is stored in database with
time stamp. Anything that is not statically defined will be
resolved based on DNS level, i.e. top level, second, third
and so on, as needed.

ipaddr name

hosiCouni

octets

Flows

packets

duration

mmm |kmq-iw-i .sun.coM

53677

5523542

67220

125026

34003

221.215.63.230

39694

1634658

39595

39946

34995

83.11 7.22.1 V. c53?5t&7(.cable.wanadoc-.nl

32763

327fi9

55019

34995

212.6I105.1S cd4J06912.cable.wanadoo.nl

32762

32792

56649

34995

66.1 40.1 05.98 |ad5l>BG*1 4a-105-9B.dsl.lbcPrtx.swholl.not

32757

3602024

33433

59323

34993

8230.176.81 |hosl&1 -1 7fi-sta1ic.30-fl2-b.busiiies5.telecoinilalia.it

32750

33223

50151

34995

90.1.63.17 ALille-153-1-100-17.w90-1.abo.wanadQO.fr

32735

3091255

33065

50815

34995

208.1 57.1 65.93 amg>bb<dsl<348.dsLairslruniconim.net

32733

6234200

62393

102200

34995

201.66,29.248 201^-29-Mclaje700.dSLbrasil:eleron.neLbr

3272&

3870511

32751

63451

34933

83.2B.73.1 75 |bgj175.neoplus.adsl.tpnstpJ

32724

3564D2Q

33161

SB544

34993

21 6. 1 71 .1 77.4S |acc-rssiel-2 16-171-1 77-46 . 1 77.1 71 .21 6.in -add r^rpa

32715

2981603

33021

49002

34993

Table 1: ISP Data traffic flow output

231

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Knowing transfer rates for data movement
applications is very important. Often, neither SNMP
monitoring of router/switch interfaces, nor applications
themselves can provide that information. For this purpose,
use passive monitoring based on net flow information.

First, on the regular basis, we generate reports
with topN senders, receivers and conversations for
specified interval of time between the entities that we need
to monitor. The reports are produced using the results of
traffic tagging described earlier. The output of the WEB
based interface is very similar to Top Scanning reports.

Our objective is to detect conditions automatically
and notify if it becomes a long term situation. After
preliminary investigation, determined that flow data
collected from routers at the edge of the campus network
could be used for that purpose by comparing flow rates at
various points of network. Flow rates for a symmetric path
should be very close in both directions, inbound and
outbound. The proposed cluster tool is used interactively to
show asymmetric traffic conditions. Currently, we are
working on automated alerts.

traffic at work group router is almost symmetric. However,
at the border router observed only outbound flows (second
graph) and the high impact infrastructure providing
alternative path, only inbound flows are observed. Thus,
by comparing flow rates at the different routers along
potential traffic path successfully detect asymmetry
conditions.

VI. CONCLUSION

The proposed model describe a system for
extracting traffic data based on the nature requested by the
network administrator with an effective clustering system.
The implemented system, streaming hierarchical
partitioning, hierarchically clusters data streaming content.

The performance predictions include the quality
of clustering, and traffic data discovery. The streaming
hierarchical clustering algorithm was able to improve the
ability to discover traffic data of required nature. The
system has been prototyped and tested on a Xeon
processor as well as on a PowerPC. To implement
additional streaming functionality, some of the same
circuitry can be reused in particular the computation of
similarity of traffic data insertion.

In the future, plan to additionally move towards a
system that, Integrates clustering into our classification
system, continually searches for new and emerging traffic
data sets on larger inter networks, allows the resolution of
data to fade over time to allow for streaming with infinite
flow of data sets.

REFERENCES

Tine

Graph 1 : Detection of traffic asymmetry

IDBD 1

0.00 L

'i-

Id. DC IS

ll"

4Q.Q0. M
00.00 mM

mm Ifl

1DD.DD ^-|

iiddd 1

[2]

I.I

[3]

Ifl

';:

■:■

■■■

frj

•->

F-

"■

*■

■-

'-'

■-

■■■■

[41

Time

Graph 2: Detection of traffic asymmetry

The graph 1 and 2 visualize the idea of steering
production traffic. The traffic switched into a high impact
path is almost always symmetric in all routers. However,
traffic is going to be asymmetric at the border and the
point of presence. From graphs at the right we can see that

[5]

[6]

Abdun Naser Mahmood, Christopher Leckie, and Parampalli
Udaya " An Efficient Clustering Scheme to Exploit
Hierarchical Data in Network Traffic Analysis ". In IEEE
Transactions on Knowledge and Data Engineering, Vol 20,
No 6, June 2008.

Cao J., D. Davis, S. Vander Weil, and B. yu, "Time- Varying
Network Tomography", J. Am. Statistical Assoc, 2000.

Chadi Barakat, Patrick Thiran, Gianluca Iannaccone,
Christophe Diot, and Philippe Owezarski. Modeling internet
backbone traffic at the flow level. IEEE Transactions on
Signal Processing, 51(8):21 11-2124, August 2003. [2] Graham
Cormode, S.Muthukrishnan, and Irina

Dai , B.-R. J.-W. Huang, M.-Y. Yeh, and M.-S. Chen,
"Adaptive Clustering for Multiple Evolving Streams," IEEE
Trans. Knowledge and Data Eng., vol. 18, no. 9, pp. 1166-
1180, Sept. 2006.

Goldschmidt O., "ISP Backbone Traffic Inference Methods to
Support Traffic Engineering", Proc. Internet Statistics and
Metrics Analysis Workshop (ISMA '00), Dec. 2000.

Guha S, A. Meyerson, N. Mishra, R. Motwani, and L.
O'Callaghan, "Clustering datastreams: Theory and practice,"
IEEE Transactions on Knowledge and Data Engineering, vol.
15, no. 3, pp. 515-528, 2003.

232

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

[7] Guha S., N. Mishra, R. Motwani, and L. O'Callaghan,

"Clustering Data Streams", 41 st Annual Symposium on
Foundations of Computer Science, 2000.

[8] Henzinger M. R. , P. Raghavan, and S. Rajagopalan,

"Computing on Data Streams", External Memory Algorithms,
Boston: American Mathematical Society, 1999.

[9] Huang Z, M. Ng, H. Rong, and Z. Li, "Automated Variable

Weighting in k-Means Type Clustering," IEEE Trans. Pattern
Analysis and Machine Intelligence, vol. 27, no. 5, pp.657-668,
May 2005.

[10] Hun-Jeong Kang, Hong-Taek Ju, Myung-Sup Kim and James

W. Hong, "Towards Streaming Media Traffic Monitoring and
Analysis", APNOMS 2002, September 2002, Jeju, Korea.

[11] Jacobus van der Merwe, Ramon Caceres, Yang-hua Chu, and

Cormac Sreenan,"mmdump- A Tool for Monitoring Internet
Multimedia Traffic," ACM Computer Communication Re-
view, Vol. 30, No. 5, 2000.

[12] Jensen C , D. Lin, and B.C. Ooi, "Continuous Clustering of

Moving Objects," IEEE Trans. Knowledge and Data Eng., vol.
19, no. 9, pp. 1 161-1 173, Sept. 2007.

[13] Kuman A, M. Sung, J. Xu, and J. Wang, "Data Streaming
Algorithms for Efficient and Accurate Estimation of Flow Size
Distribution", Proc. ACM SIGMETRICS, 2004.

[14] Lakhina A, K. Papagiannaki, M. Crovella, C. Diot, E.
Kolaczyk, and N. Taft, "Structural Analysis of Network
Traffic Flows", Proc. ACM SIGMETRICS '04, June 2004.

[15] Lakhina A, M. Crovella, and C. Diot, "Characterization of
Network- Wide Anomalies in Traffic Flows", Technical
Report BUCS-2004-020, Boston Univ., 2004.

[16] Lockwood J. W, S. G. Eick, J. Mauger, J. Byrnes, R. P. Loui,

A. Levine, D. J. Weishar, and A. Ratner, "Hardware
Accelerated Algorithms for Semantic Processing of Document
Streams", 2006 IEEE Aerospace Conference, March 4-11,
2006.

[17] Lockwood J. W, S. G. Eick, D. J. Weishar, R. P. Loui, J.
Moscola, C. Kastner, A. Levine, and M. Attig,
"Transformation Algorithms for Data Streams, 2005 IEEE
Aerospace Conference", March 5-12, 2005.

[18] Li W , W.K. Ng, Y. Liu, and K.-L. Ong, "Enhancing the
Effectiveness of Clustering with Spectra Analysis," IEEE
Trans. Knowledge and Data Eng., vol. 19, no. 7, pp. 887-902,
July 2007.

[19] Madina A , K. Salamatian, N. Taft, I. Matta, and C. Diot, "A

Two Steps Statistical Approach for Inferring Network Traffic
Demands", revision of Technical Reports BUCS- TR-2003-
003,Mar.2004.

[20] Maulik U and S. Bandyopadhyay, "Performance Evaluation of

Some Clustering Algorithms and Validity Indices," IEEE
Trans. Pattern Analysis and Machine Intelligence, vol. 24,
no.12, pp. 1650-1654, Dec. 2002.

[21] Mouratidis K , D. Papadias, S. Bakiras, and Y. Tao, "A
Threshold-Based Algorithm for Continuous Monitoring of K
Nearest Neighbors," IEEE Trans. Knowledge and Data Eng.,
vol. 17, no. 11, pp. 1451-1464, Nov. 2005.

[22] Patrikainen A and M. Meila, "Comparing Subspace
Clusterings," IEEE Trans. Knowledge and Data Eng., vol. 18,
no. 7, pp. 902-916,July2006.

[23] Su M.-C. and C.-H. Chou, "A Modified Version of the It-
Means Algorithm with a Distance Based on Cluster
Symmetry," IEEE Trans. Pattern Analysis and Machine
Intelligence, vol.23, no. 6,pp. 674-680, June 2001.

[24] Subhabrata Sen and Jia Wang, "Analyzing Peer-to-Peer Traffic
Across Large Networks", IMW2002 Workshop, 2002,
Marseille, France.

[25] Vijayakumar M and R.M.S.Parvathi. " Concept Mining of
High- Volume Data Streams in Network Traffic using
Hierarchical Clustering'Mn Proceedings of the Fourth
International Conference, Amrutvani College of Engineering,
Sangamner,Maharatra. Proc. ITECH '09 pp 69 , March 2009.

[26] Wang J, D. Miller, and G. Kes dis, "Efficient Mining of the

Multidimensional Traffic cluster Hierarchy for Digesting
Visualization, and Anomaly Identification", IEEE J. Selected
Areas ofComm. Vol. 24, no. 10, pp. 1929-1941, Oct. 2006.

[27] Xu R and D. Wunsch II, "Survey of Clustering Algorithms,"

IEEE Trans. Neural Networks, vol. 16, no. 3, pp. 645-678,
2005.

[28] Yip K.Y.L., D.W. Cheng, and M.K. Ng, "HARP: A Practical
Projected Clustering Algorithm," IEEE Trans. Knowledge and
Data Eng., vol. 16, no. 11, pp. 1387-1397, Nov. 2004.

AUTHORS PROFILE

M.Vijayakumar has completed
his Bachelor of Engineering in
Computer Science in Bharathiar
University, Tamilnadu, India and
Master of Engineering , in
Computer Science in Anna
University,Chennai , Tamilnadu,
India. He has started his teaching
profession in the year 2004 in
Muthayammal Engineering

College,Tamilnadu. India., At
present, he is working as an Assistant Professor in the
department of Computer Science and Engineering in Sasurie
College of Engineering, Tamilnadu. India. He has published
15 research papers in National and International journals and
conferences. Currently he is a part time Research Scholar in
Anna university of Technology, Coimbatore. His areas of
interest are Data mining, Knowledge Engineering, Clustering
algorithms and Network security. He is a life member of ISTE.

Dr. R.M.S. Parvathi has

completed her Ph.D., degree in
Computer Science and

Engineering in 2005 in

Bharathiar University,

Tamilnadu, India. Currently she
is a Principal and Professor ,
Department of Computer
Science and Engineering in
Sengunthar College of

Engineering for Women,
Tamilnadu, India, She has completed 20 years of teaching
service. She has published more than 28 articles in
International / National Journals. She has authorized 3 books
with reputed publishers. She is guiding 20 Research scholars.
Her research areas of interest are Software Engineering, Data
Mining, Knowledge Engineering, and Object Oriented
System Design. She is a life member of ISTE and Indian
Computer Society.

233

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

A.R.Q. techniques using Sub-block retransmission for wireless networks
A.N.Kemkar, Member, ISTE and Dr. T.R.Sontakke MemberJSTE

Abstract — In this paper we mainly focus our
investigation on the throughput performance in
conjugation with sub-block transmission scheme.
The throughput of a wireless data
communications system depends on a number of
variables, one of it is length of the message
blocks. Over a noisy communication medium like
wireless medium used for mobile ad-hoc network,
our propose scheme performs effectively. In
propose scheme random length of the message is
divided in to fixed length blocks and applying
ARQ techniques if the error occurs. A threshold
model is used for fading channel, estimation and
CRC detection codes are used. Comparison of
transmission efficiency of proposed scheme with
varying channel condition is shown.
.Index Terms— FEC, Hybrid ARQ, BER.

1. INTRODUCTION:

Wireless channels are highly affected by
unpredictable factors such as co-channel
interference, adjacent channel interference,
propagation path loss, shadowing and multi path
fading. The unreliability of media degrades the
transmission quality seriously. Automatic Repeat
ReQuest (ARQ) and Forward Error Correction
(FEC) schemes are frequently used in wireless
environments to reduce the high bit error rate of
the channel.

As we have seen, the throughput efficiencies of
all the basic ARQ schemes are functions of the
packet size n. [1],[2],[3],[4],[5]. Our main result
is a mathematical technique for determining the
block size as a function of the other variables like
BER, signal-to-noise ratio.

A.N.Kemkar 1 ,S.R.T.M.U,Nanded.

+91-9819150392, ankemkar@gmail.com

Dr.T.R.Sontakke 2

Ex.Director - S.G.G.S.I.T.E.- Nanded

Principal,Sidhant college of Engineering

Pune.+91-

9822392766,trsontakke@gmail.com

In an attempt to improve throughput
performance, we have included an analysis using
forward error correcting (FEC) block codes (used
in Hybrid ARQ). The optimum amount of FEC
coding was found to be dependent upon the Block
length. As the Block length increases, the number
of correctable errors to optimize the throughput
also increases, mathematical expression is shown
in 2.2.

The paper is organized as follows. In Section 2
.Summary on the Related work and basic concept.
In section 3. scheme description and system
model. We consider the performance analysis of
the proposed scheme for simulation study in
section 4. followed by conclusion in Section 5
2.RELATED WORK :

2.1: Related Work- The efficiency of HARQ
scheme is compared with GBN schemes using
different lengths of IP Blocks. Further show that
usage of smaller Blocks and hybrid schemes leads
to an improved throughput. Differences between
pure and hybrid GBN schemes are also discussed.

[i]

When the channel is quiet the sub-block
retransmission scheme behaves like a
conventional ARQ or hybrid ARQ scheme. As
the channel becomes increasingly noisy, the data
block is divided into smaller sub-blocks for
transmission. Each sub-block is encoded for error
control by an appropriate shortened code of
which the code length is adapted to the
corresponding channel BER. [2] Further optimum
block size in accordance with the channel
conditions [4] A single code HARQ scheme was
proposed in which transmitter is operating in any
one mode with the degree of errors encounter.
The operating state is selected based on the
channel BER. Data bits are divided in blocks and
are encoded with shortened codes. During the
retransmission new coded blocks are combined
and at the receiver end proper decoding
techniques are used to separate retransmitted
blocks from the new blocks.
2.1. BASIC CONCEPT -

234

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Analytical expression how Throughput

performance of the system varies with the size of

the block length and FEC. : Consider the

following two cases to verify the Throughput

performance of the system.

Case 1. Blocks are transmitted without FEC.

Case 2. Blocks are transmitted with FEC.

Our analysis includes the following simplifying

assumptions: 1. The CRC decoder detects all

errors in the output of the FEC decoder.2.

Transmission of acknowledgments from the

receiver to the transmitter is error free and

instantaneous.

System throughput (P) is the number of payload

bits per second received correctly:

„ K

RfiY)

(1)

where

b/s is the payload transmission rate

KR
L

Where f(Y) = Block success rate defined as
probability of receiving block correctly.
Probability is a function of signal to noise ratio.
Eb
No

(2)

p
In which Eb = —joules in received energy per bit.
R

where R= Transmission rate. Probability is a

function of signal to noise ratio.

= Eb = P_

N No*

(3)

Each Block, of length L bits, is a combination

of a payload [K] and overhead (L-K) . Because

the Block success rate,/(7) is a decreasing

function of,£ there is an optimum Block length,.

V . When

Lp l* excessive overhead in each Block limits

the throughput. When L > l* Block errors limits

the throughput.

For case 1. When there is no forward error

correction coding. In this case

f(7) = (l-Pe(r)) L (4)

where f(7) = block success rate defined as

probability of receiving block correctly.

Pe{ y) is block error rate.

Therefore, in a system without FEC, the

throughput as a function of L ,from (1)

T=^R(i-p e {y)) L (5)

Case 2.: Now instead of transmitting those L bits
with no error correction capability, we will now
add B error correcting bits and transmit a total of
bits L+B .Using a block code forward error
correction scheme, the minimum number of B
bits required to correct t errors is given by [5]

(6)

B > log 2

t
I

n=0

Now that we can correct 4 1 ' errors, our block
success rate,/(7) should be larger than its
previous value with no error correction. Recall
that /(/) with t= is given by:

f(y) = (l-p e (y)) L where P e (y) is the probability

of a bit error as a function of the SNR. Now, with
error correction capability, the Block success rate
for some arbitrary value oft is [7]

^-Pe[r)) L+B - n (7)

\P n e{7)

ft(r)= E

Our new equation for the throughput as a function
of the signal to noise ratio is:
P

T(Y):

L-C
L+B

ft(r)

(8)

From (5) and (8) it is clear that throughput of the
system is a function of message block length.
Further (5) and (8) are used for pure and hybrid
ARQ techniques.

3. SCHEME DISCRIPTION AND SYSTEM
MODEL :

This paper presents a sub-block retransmission
scheme for ARQ . The data block is divided into
smaller sub-blocks for transmission. Each sub-
block is encoded by an appropriate error detection
codes. The encoded block is then transmitted. The
received block is checked for errors sub-block by
sub-block. The proposed scheme provides
improved throughput by retransmitting only the
sub-blocks in the occurrence of errors.

3.1 SYSTEM MODEL :

235

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

We consider an ad-hoc network with V nodes
and assume that each node is equipped with only
one antenna. A Point to Point protocol is used at
the medium access control layer. A Selective
repeat request ARQ mechanism is used.
Particularly, the source node transmits a data
packet with a C-bit CRC attached. The
destination node detects CRC and then sends an
acknowledgement that is either positive (ACK) or
negative (NACK) back to the source node. If the
packet is correctly detected by the destination
node (with ACK feedback), the source node
continues to transmit a new data packet and the
above process is repeated. Otherwise,
retransmission will start. A threshold model for
channel characterization is used for fading
channel.

4. PERFORMANCE ANALYSIS OF THE
PROPOSE SCHEME:

Expression of throughput for ARQ for present
scheme:
K

V =

E[T]

(9)

where K =information bits in a block. E[.\

=Expectation of number of transmitted bits in a
given block.

T = Mn + ^Ti (10)

i=\

where M =number of sub blocks, n =number of
bits in a sub block, Ti =number of transmitted bits
for i th transmission.

E[T]=Mn + Y,E[T i ] (11)

1=1

where E[T\ =Average number of transmitted

bits.

Out of M sub blocks if L sub blocks are

transmitted at the f 1 retransmission ,then random

variable, Ti takes the value Ln ,if L out of M

sub-blocks are retransmitted at the f 1

retransmission.

SIMULATION RESULTS: We evaluate the
performance of the proposed scheme
implemented with Matlab. We run the simulation
for two schemes i.e. with sub block transmission
and without sub block transmission. The
simulation parameters are shown in the table
1. Simulation run for 5000 total blocks. Result is
the average of independent experiments where
each experiment uses different randomly
generated uniform parameters. We use mean
values which are obtained independent
experiments as a basic data to get the result.
Simulation results are shown Table 2
Table 1: System Parameters:

Parameters

Notaion

Values

Signal to Noise Ratio

Varied

Total number of blocks

5000

Total sub block

Information bits in a block

Packet length

5000*32*16

Max. number of
Retransmissions

Number of sub blocks
retransmitted

Varied

Cyclic Redundancy Check

CRC

Varied

Bit error rate

BER

Varied

Packet error rate

PER

Varied

Throughput efficiency

Varied

Table 2 : Simulation Results: Following
Simulation results shows the comparison of
Throughput efficiency verses varied block size
verses changing channel condition in terms of
PER.
Table 2:

Block length

Packet error rate

Throughput

Without
sub block

With

sub

block

Without

sub

block

With

sub

block

Without

sub

block

With

sub

block

Whole
block is
transmitted
with out
sub
division.

Whole

block

divided

into

sub 4

blocks

0.1

0.9

0.99

0.3

0.86

0.96

0.5

0.66

0.94

0.7

0.59

0.93

0.9

0.57

0.9

0.5

0.89

236

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

5. CONCLUSION:

From the Fig.l shown below performance of the
proposed scheme. This paper has presented a sub-
block retransmission scheme for

1. With micro block ARQ

2. Without micro block ARQ

Proposed sub-block retransmission schemes
showed better overall performance compared to
the other competitive schemes by retransmitting
without sub-blocks in the occurrence of errors.
From Table 2 and the Fig.l it is clear that as the
packet error rate increases i.e. channel condition
gets deteriorated the throughput performance
goes down by more than 40% as compare to
proposed scheme. There fore we proposed sub-
block retransmission scheme with ARQ is more
reliable than existing i.e. without sub block
transmission scheme.

6. REFERENCES:

[1] W.W. Chu. Optimalmessage block size
for computer communications with error
detection and retransmission strategies.
IEEE Transactions on Communications,
COM-22:1516- 1525, October 1974.

[2] J.S. DaSilva, H.M. Hafez, and S.A.

Mahmoud. Optimal packet length for
fading land mobile data channels. In
Proceedings of ICC 1980, pages 61.3.1-
61.3.5, June 1980.

[3] R.L. Kirlin. Variable block length and
transmission efficiency. IEEE

Transactions on Communication

Technology, COM-17:350-355, June
1969.

Fig.l: Performance of proposed scheme vs.
existing schemes.

Throughput Vs. Packet E rro r Rate

) block

[4] E. Modiano. An adaptive algorithm for
optimizing the packet size used in
wireless ARQ protocols. Wireless
Networks, 5:279-286, July 1999.

[5] J.M.Morris. Optimal block lengths for
ARQ error control schemes. IEEE
Transactions on Communications, COM-
27:488-493, February 1979

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.
Packet E rro r Rate

237

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

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

Vol. 8, No. 6, 2010

Performance Analysis of Delay in Optical Packet
Switching Using Various Traffic Patterns

A.Kavitha/Chettinad College of Engineering &
Technology

IT dept

Chettinad College of Engineering & Technology,

Karur, Tamilnadu, India

V.Rajamni/Indra Ganesan College of Engineering,

Indra Ganesan College of Engineering
Trichy, Tamilnadu, India

P.Anandhakumar/Madras Institute of Technology,

IT Dept

Chennai, Tamilnadu, India

Abstract — Quality of Service parameters are improved
for development of optical packet switching technology.
Delay is an important parameter in optical packet
switching networks and it affects the performance of the
network. In this paper, a mathematical model is presented
to evaluate the delay rate. Delay rates are analyzed for
fixed packet length and variable length packet for various
traffic patterns viz. Non-uniform, Poisson and ON-OFF
traffic models for various service classes using
Reservation Bit technique. The results are compared with
the existing port based First-Fit wavelength assignment
algorithm. Here delay rates are reduced by 29% in our
class based model than the port based model.
Keywords-component; Optical Packet Switching (OPS),
RB (Reservation Bit algorithm), FF (First-Fit Wavelength
assignment algorithm), Quality of Service (QoS), Packet
Loss Rate (PLR), BER (Bit Error Rate), WDM
(Wavelength Division Multiplexing).

I Introduction

Due to the explosive growth of internet applications in
recent years, data traffic has been exceeded the telephony
traffic and bandwidth demands have been continuously
increasing. It is also expected of the future networks to
transport heterogeneous traffic services including
multimedia and interactive applications necessitating
bandwidth guarantees, minimum delay, less PLR, controlled
jitter and etc. QoS provisioning seems therefore a mandatory
task. Optical networks offer an extremely high traffic
bandwidth capable of providing communication channels
for several hundred nodes. Thus, the network traffic requires
the network to evolve by increasing transmission capacity of
optical fibers as well as switching capability. There are three
switching schemes in optical networks namely optical
circuit switching, optical burst switching (OBS) and optical
packet switching (OPS). In the optical circuit switching, a
dedicated end-to-end light path is established for each
connection. Thus the transmission delay can be guaranteed

and there is no virtually any loss, but less utilization of
wavelength in this technique. In OBS, data is sent in bursts
and a burst control message is sent ahead of each data burst
to reserve a wavelength at each hop based on the expected
arrival time of the data burst. In OPS, messages are
transmitted in packets. At each switching node, the packet
head is processed in the electrical domain for routing
purpose and the packet data is kept in the optical domain.
Wavelengths can be efficiently used in OPS [1]. Thus the
optical packet switching has emerged as one of the most
promising technologies for future telecommunication
networks. OPS utilize very high bandwidth in the optical
fiber using WDM. WDM offers an aggregate throughput of
the order of terabits per second. WDM is widely becoming
accepted as a technology for meeting growing bandwidth
demands, and WDM systems are beginning to be deployed
in both terrestrial and undersea communication links. Thus,
WDM offers an excellent platform for carrying IP traffic.
Consequently, OPS technology has many advantages, attract
more intensive attention than ever. The next generation
telecom infrastructure definitely comprise of optical
networks with improved QoS.

In order to improve the performance of QoS in optical
packet switching network, a detailed study has been made in
this paper. The performance analysis of optical packet
switching consists of two important issues namely packet
loss and delay. Inorder to provide better QoS in optical
packet switching, PLR and delay should be reduced. Packet
loss and delay are not new issues in optical networks;
however minimum loss and delay provide better QoS in
Optical packet switching. In our earlier report the
performance of 8B/10B code, Systematic code and Viterbi
code in optical transmission in terms of Bit Error Rate has
been analyzed. In the physical layer, transmission is done on
bit by bit basis and Bit Error Rate has been reduced in the
physical layer. When the bit errors in the physical layer are
rectified, it will reduce the packet loss rates in the higher
layers [2,3]. We have already reported that PLR has been

238

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

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

Vol. 8, No. 6, 2010

reduced in non-uniform traffic pattern [4]. We have also
implemented a RB algorithm for minimizing PLR in
buffered non-uniform traffic pattern of optical packet
switching mechanism [5].

In this paper we have analyzed the asynchronous OPS in
terms of delay. Delay rate for fixed length packet and
variable packet length has been studied for various traffic
patterns viz. Non-uniform, Poisson and ON-OFF traffic
models.

This paper has been organized as follows: Section II
describes the architecture of asynchronous OPS. In section
III, the analysis of delay rate is carried out for various traffic
patterns in asynchronous OPS. In section IV, results along
with discussions are presented and Section V deals with the
conclusion and future work.

II. Description of Architecture

The architecture used in this paper is presented in [4,5]. It
has been reproduced for the reference. The size of switch
under consideration is N X N. The switch has F input fibers
and F output fibers. By utilizing WDM, each fiber provides
N wavelengths to transport data with a capacity of C bps.
Buffers with the size of 5 are used in the OPS switch
catering to each of the service classes. The switching
process in OPS can take one of the two main forms. It can
be synchronous (time slotted) with the fixed packet length or
asynchronous (non-slotted) with variable packet lengths. In
synchronous operation mode, all arriving packets have a
fixed size and their arrival on each wavelength is
synchronized on a time-slot basis, where a time slot is the
time needed to transmit a single packet.

The operations of optical packet switching can be briefly
described as follows: When a packet arrives at the switch,
the packet header is extracted and processed electronically
by the control module. While the header is processed, the
packet payload is buffered in the optical domain using FDL
processing buffers. Based on the destination, information
extracted from the packet header and the control module
decides to which output fiber/wavelength the packet is
switched and configures the switch accordingly. Contention
occurs when two or more packets are assigned to the same
output port on the same wavelength at the same time [6 and
7]. The network has d service classes, ranging from service
class to service class d- 1 . We assumed that the output on a
single fiber/wavelength as the tagged fiber/wavelength.
Delay is calculated for class i traffic at the tagged output
fiber. Figl shows a switch for packet arrivals to a tagged
output fiber must originate from one of the FN input
wavelengths.

Input Wavelengths
2

Output
Wavelengths

i
i
i

or"'

Fig 1. Model of the switch in OPS network under consideration

The following assumptions are made for the simulation:

Let ji denote the number of class i packets that arrive in a
time slot. The total number of packet arrivals at the fiber in a
time slot is k. That is j + j i +. . . +jd-i = k. In order to isolate
the service classes, the parameter li (0 < li < N) is
introduced, which is the number of wavelengths reserved for
Class i traffic in case of contention in a time slot. For a
service class i, if ji < li (the number of incoming packets are
less than the wavelengths assigned), it will result in ji - li
free slots. For a service class i, if ji > li (the number of
incoming packets are greater than the wavelengths
assigned), resulting in 1 ± - j ± overflow of packets.
The RB algorithm [5] is implemented by introducing
buffers. The RB algorithm is used for slotted OPS wherein
buffers are used to avoid overlapping of packets. This
algorithm is designed and used to improve the QoS of the
networks in terms of reduction in PLR. By using the same
algorithm delay rate is found and analyzed in slotted OPS
with fixed packet lengths for various traffic patterns. Hence
QoS is improved in terms of reduction in PLR and delay
[4,5]. In this paper the same algorithm is implemented in
asynchronous OPS for the packets of variable lengths.

III. Operating Principle

Operation of optical packet switch in a synchronous manner
with fixed length of packets is explained in [5]. Here we
assume that optical packet switch operates in an
asynchronous manner with variable length of packets.
Analysis of delay rate for variable packet sizes in
asynchronous OPS using three types of traffic patterns viz.
Non-uniform, Poisson and ON-OFF traffic models is
presented in this paper. In contrast to [8], immaterial of
packet size, allotted time slots remain the same and no
shifting of time slot allocation is done.
In non-uniform traffic pattern all nodes are not to receive
and send similar volumes of traffic [4, 9]. The number of
packets arrived and transmitted of packets is not equal.
There is an incoming packet for every slot. We assume that

239

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

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

Vol. 8, No. 6, 2010

each packet has equal probability of 1/N being addressed to
any given output port and successive packets that arrive at
the tagged output/wavelength are independent. Also, the
packet arrival time at the input and transmission time to the
output port are not equal. This is due to randomness of
packet arrival. This increases the occupancy of the free
slots. In Poisson traffic model, packet arrivals are random
and are mutually independent. The Poisson distribution can
be obtained for packet arrivals during an infinitesimal short
period of time X t, where X is called the arrival rate. Packet
arrival at the output fiber is a single Poisson process with X.
In Poisson model, the number of arrivals in non-overlapping
intervals is statistically independent. The arrival rate X is
expressed as the average number of arrivals during a unit of
time. The time distance between consecutive packet arrivals
is exponentially distributed [10]. However Poisson arrival
model is not an accurate model considering packet arrivals
at the tagged output fiber/wavelength in OPS. The Poisson
arrival model assumes an infinite number of sources, which
is not the case in a real switch. When using a Poisson arrival
model, there is the possibility of independency between the
service time and the packet inter-arrival time [11].
Bursty traffic is also known as ON-OFF traffic model. Both
the ON and OFF periods are distributed using exponential

distribution with the meanl/// andl//l respectively. It is
an alternating process where an OFF period follows an ON
period, and an ON period follows an OFF period. During
ON periods, series of packets are transmitted from the
source node and the time is called active periods. OFF
periods are called passive periods and no packets are
transmitted from the source. Active periods of a source are
exponentially distributed with one specific mean value, and
passive periods are exponentially distributed with another
mean value. During an active period, packets are generated
at regular periods. The most commonly used VOIP traffic
model is based on a two-state on-off model of a single voice
source. When a voice source is transmitted, it is in the ON
state, and when the voice source is silent, it is in the OFF
state; and the ON and OFF states appear alternatively. ON-
OFF sources, each of which exhibits a phenomenon called
the "Noah Effect" resulting in self similar aggregate traffic.
The Noah Effect for an individual ON-OFF source model
results in ON and OFF periods, i.e. "train lengths" and
"intertrain distances" that can be very large without
negligible probability [12 and 13]. The number of packet
arrivals divides the timeslot and produces the time slice
which is utilized for packet transmission. Hence the packet
arrival rate is random, time slice is determined according to
randomness of the packet arrival. ON and OFF periods are
used with their own mean values and using the same mean
values, various packets arrive and are transmitted during ON
period and idling occurs during the OFF period. During ON
period, packet is transmitted and there is no flow of packet
during OFF period. In this model, FN independent state
model generate packet arrivals at a tagged output fiber. In

this paper, at every node there should be arrival(s) of
different classes of packets. In ON-OFF model, ON periods
and OFF periods occur at each node for different classes.
When one class is in ON period, other classes may be in
OFF period at that particular node and OFF period of one
class is used by the other classes for transmitting packets.
Also, other possibilities are (1) all classes may be in OFF
period, (2) all classes may be in ON period and (3) some of
the classes may be in ON period and rest may be in OFF
period.

We assume that the switch is having buffers. In RB
algorithm, the packets use the wavelength according to their
service classes. If there is flow of packets into the ports
with a specified service class, the incoming packets with
assigned wavelengths occupy the ports as per the assigned
service class. When the assigned wavelengths in one service
class are occupied, it checks for the free wavelengths of
other service classes and the packets will occupy the free
slots in the other service classes. When the assigned
wavelengths are completely occupied in all the ports, the
packets overflow. By introducing buffers, the packets that
overflow are saved. When a free slot is not available for an
incoming packet, instead of dropping that packet, it will be
saved in the buffer which is provided in the switch. Before
entering the buffer, a bit is added in the packet header for
the purpose of reservation with respect to their service
classes. Whenever free slots are available, the packets in the
buffer occupy free slots. Buffered packets will have the
priority over the incoming packets. In the FF algorithm, all
wavelengths are numbered in a certain order, for example
ascending order from to W-l, where W is a number of
wavelengths. When the deciding port attempts to assign a
wavelength, it sequentially searches all wavelengths in an
ascending order and assigns the first available wavelength
[14]. In class based model, each node transmits the packets
according to their classes. Buffers are placed in the port for
every class. In port based model, wavelengths are placed in
a sequential order. Irrespective of the class, the available
wavelength is used by the incoming packets in a sequential
order and buffers are only placed in each port.
FF algorithm is implemented in port based and packets are
transmitted according to their wavelengths, whereas
transmission of packets is class based in RB algorithm and
packets are transmitted according to their service classes and
wavelengths. Thus the drop rates of packets are reduced in
optical networks resulting in improved QoS. The delay rate
is found and analyzed in asynchronous OPS for various
traffic patterns viz. Non-uniform, Poisson and ON-OFF
traffic models for various service classes with packets of
variable lengths. The delay rates for RB algorithm and FF
algorithm are found and compared.

For fixed size packets and variable size packets in OPS,
delay rate is encounted by implementing RB algorithm. The
packet includes payload and header. Fig 2 shows the packet
header. For fixed size packets, size considered is 512 bytes.
For variable packet size, the range of packet size is in the

240

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

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

Vol. 8, No. 6, 2010

range of 512 bytes and 2k bytes. 20 bytes is included as
header along with the payload invariably for any type of
packets. Variable size packets are used in VoIP applications.
In variable packet size, the size has been controlled by the
application. Packet sizes of the range of 1024 to 2048 bytes
show good efficiency in terms of bandwidth and reliability
in Digital Video Broadcasting [15]. Packet size is measured
using uniform min and max distribution. Packet size is
chosen depending on the application.

Source

IP
address

Desti
-nation

IP
address

Source
Port

Desti

-nation

Port

Packet

Sequence

Time
Stamp

Flow
id

4
bytes

2
bytes

4
bytes

2
bytes

Fig. 2 Structure of the packet header
When a packet is send from one node to other, the following
delays occur: (1) transmission delay (time required to send
all bits of packet into the wire), (2) propagation delay (the
time taken by the packet to travel through the wire), (3)
processing delay (the time taken to handle the packet in the
network system), and (4) queuing delay (the time taken is
buffering the packet before it can be sent). In most cases, the
delay (2) and (4) are considered in simulations and
measurements. The transmission delay (1) is usually small
for fast links and small packets and is therefore not
considered. Traditionally, the processing delay (3) has also
been negligible [16]. In our measurements, we consider the
delay (2) and (4). The mean delay for the above said traffic
pattern is found using equation (1).

T =T +T (U

2 D * Propagation ' * Queue v 1 /

Initially, we calculate the propagation delay that occurs
when a packet travels from the source to the destination.
Next, the queue delay experienced by a packet is calculated
using equation (2). This delay is due to waiting period of the
packet in the queue. A packet is in a queue, if a free
wavelength is not available at that particular time slot.

(2)

T =— Y(T)

A Queue AT /^ V i/

where Ti is the transmission time of class i packets at

particular time slice.

Summation of the waiting time in the buffer and the

transmission time between source node and destination node

through the switch is considered as delay and the same is

found for the above said traffic patterns.

IV. Results and Discussions

The delay values for the fixed length packets in slotted
OPS using RB algorithm is studied and is also compared to
the FF algorithm in our earlier paper [5]. In this paper, a
detailed analysis is carried out to find delay in asynchronous
OPS for variable length packet and is compared to FF
algorithm.

We consider 240 packets for the simulation purpose. Delay
for class i packets are calculated in the tagged fiber.
Wavelength assigned is 16 and total time slot chosen is 10,
hence this architecture can transmit 160 packets in a time

slot. Buffers are used along with RB technique, 240 packets
are chosen with an Erlang load of 1.5.

Delay in SC3(Class based)

Cl.iss

Fig 3. Delay rate for various traffic patterns for service class 3 using
reservation bit technique.

Delay in SC3(Port based)

Fi
g 4. Delay rate for various traffic patterns for service class 3 using First-
Fit wavelength assignment algorithm

Delay in SC4(Class based)

Fig 5. Delay rate for various traffic patterns for service class 4 using
reservation bit technique

241

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

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

Vol. 8, No. 6, 2010

Deay in SC4(Port based)

Delay Rate for various Traffic pattern

Fig 6. Delay rate for various traffic patterns for service class 4 using
First-Fit wavelength assignment algorithm

Delay in SC5(Class based)

Fig 7. Delay rate for various traffic patterns for service class 5 using
reservation bit technique

Delay in SC5(Port based)

□ NU
■ O-F
DP

Fi
g 8. Delay rate for various traffic patterns for service class 5 using First-
Fit wavelength assignment algorithm

X10

SC5

SC4

Port Based

Class Based

Fig 9. Delay rate comparison
Transmitted Bytes

SC5

Fig 10. Comparison of data transmission

In service class 3, RB technique has 15 buffers, each service
class has 5 buffers, but in FF algorithm, each port has 5
buffers and the total number of buffers is 20. 10.96ms,
10.88ms and 9.83ms are the delay values while employing
FF algorithm and 7.55ms, 7.79ms and 7.91ms are the delay
values while employing RB technique in asynchronous OPS
using Non-Uniform, ON-OFF and Poisson traffic model
respectively for Service class 3 and the same is shown in
figs 3 and 4.

In Service class 4, both the techniques have 20 buffers.
10.94ms, 10.56ms and 10.18ms are the delay values while
employing FF algorithm and 7.77ms, 8.01ms and 7.561ms
are the delay values while employing RB technique in
asynchronous OPS using Non-Uniform, ON-OFF and
Poisson traffic model respectively for Service class 4 and

242

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

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

Vol. 8, No. 6, 2010

the same is shown in figs 5 and 6. From these it is seen that
both the techniques have same amount of buffer. FF
algorithm exhibits more delay, but delay is less in our
approach.

While in service class 5, RB techniques have 25 buffers and
FF algorithm need 20 buffers. 10.09ms, 10.88ms and
9.50ms are the delay values while employing FF algorithm
and 7.1ms, 6.27ms and 6.48ms are the delay values while
employing RB technique in asynchronous OPS using Non-
Uniform, ON-OFF and Poisson traffic model respectively
for Service class 5 and the same is shown in figs 7 and 8.

For all the service classes under consideration, buffers are
more or less the same for RB technique and FF algorithm,
whereas the delay rate is slightly higher in FF algorithm.
The above statement is true when the class based
transmission is compared with port based transmission.

Simulation results show that for all service classes under
any type of traffic pattern, class based model produces 29%
reduction of delay rate when compared to port based model
and the same is shown in fig 9. In RB technique the buffered
packets with respect to their classes will occupy the free
slots of the corresponding service class wavelengths on First
Come First Serve basis which reduces the waiting time in
the buffer whereas in FF algorithm wavelength utilization is
sequential, so buffered packets wait until they are serviced
with the next sequential order of wavelengths. Thus delay
rate is less in asynchronous OPS when RB technique is
employed. The delays are improved much in our algorithm.
This is due to the class based transmission.
Fig 10 shows the total number of transmitted bytes per slot.
It is shown that more number of packets is transmitted in
class based model for all service classes for all traffic
patterns under consideration when compared to port based
model. Hence class based model produces lesser delay as
well as more number of bytes transmission when RB
technique is employed. It is also shown that non-uniform
traffic pattern produces lesser delay when compared to ON-
OFF and Poisson model.

V. Conclusion

The delay rates for the Non-uniform, Poisson and ON-OFF
traffic models for various service classes are analyzed.
Minimum delay has been achieved in all service classes
under consideration by using reservation bit technique. It
provides lesser delay for all service classes for fixed and
variable length packets when compared to first-fit
wavelength assignment algorithm. It is seen from our
simulation, Poisson arrival model which is assumed in the
analysis approximates a more realistic model wherein all
input wavelengths are modelled as independent on/off
processes with exponential holding time. Also we have
presented a comparative study of reservation bit technique
and first-fit wavelength assignment algorithm for
synchronous and asynchronous OPS. It is concluded that in
Optical packet switching reservation bit technique reduces
delay; hence QoS is improved and at the same time Non-

uniform traffic pattern result in better Quality of service
compared to ON-OFF model and Poisson model. At the
same time ON-OFF traffic pattern have better QoS when
compared to Poisson process if OFF periods of one service
classes are more efficiently utilized by other service classes.

[I] Biao Chen and Jianping , " Hybrid Switching and P-Routing for
Optical Burst Switching Networks", IEEE Journal on Selected areas in
communications, vol 21, no 7,pp. 1071-1080, Sep 2003.

[2] A.Kavitha, V.Rajamani and P.AnandhaKumar, "Performance
analysis of Coding Techniques to find BER in Optical Transmission",
IEEE 1 st International Conference on Advanced Computing- ICAC 2009,
13-15 Dec. 2009 Page(s):21 - 27

[3] A.Kavitha, V.Rajamani and P.AnandhaKumar, "Evaluation of BER
in Optical Packet Switching using Various Coding Schemes" IEEE
Transaction on Optical Communication and Networking (under review).
[4] A.Kavitha, V.Rajamani, "Performance Analysis of Slotted Optical
Switching Scheme using Non-Uniform traffic" - Journal of Optical
Communications, Vol. 29, July 2008, pp. 107-1 11.
[5] A.Kavitha, V.Rajamani and P.AnandhaKumar, "Performance
Analysis of Slotted Optical Packet Switching Scheme in Non-Uniform
Traffic Pattern Using Reservation Bit Technique"- Selected for
publication- INFOCOMP, Journal of Computer Science.
[6] H. Overby and N. Stol, "Evaluating and comparing two different
service differentiation methods for OPS: the wavelength allocation
algorithm and the preemptive drop policy," in Proc. 3rd Int. Conf
Networking, vol. 1, Feb. 2004, pp. 8-15.

[7] S.Bjornstad, N.Stol and D.R.Hjelme, "A highly efficient optical
packet switching node design supporting guaranteed service" in proc. Of
SPIE, Vol 4910, 2002, pp.63-74.

[8] BO Wen, Ramakrishna Shenai, and Krishna Sivalingam," Routing
Wavelength and Time-Slot-Assignment: Algorithms for Wavelength-
Routed Optical WDM/TDM Networks", Journal of Lightwave
Technology, vol 23, no 9, pp.2598-2609,Sep 2005.
[9] Adisak Mekkittikul, Nick McKeown "Scheduling VOQ Switches
under Non-Uniform Traffic", CSL Technical Report, CSL-TR 97-747,
Stanford University, 1997 Stanford University,Standford.
[10] Wuyi Yue, Yutaka takahashi,hideaki tagaki, "Advances in
Queueing theory and network application:, Springer,ISBN: 978-0-387-
09702-2, e-ISBN: 978-0-387-09703-9.

[II] Harald Overby and N.Stol " Quality of Service in synchronous
bufferless optical packet switched networks" Kluwer Telecomm. Sys.
Vol 27, 2004. pp. 151- 179.

[12]Mihails Kulikos and Ernests Petersons"Remarks Regarding
Queueing Model and Packet Loss Probability for the Traffic with self -
Similar Characteristics", Networks", International Journal of Computer
Science 3;2,Spring 2008,pp. 85-90.

[13] Eric W.M. Wong and Moshe Zukerman," Bandwidth and Buffer
Tradeoffs in Optical Packet Switching", Journal of Lightwave
Technology, vol 24, no 12, Dec 2006, pp 4790- 4798.
[14] Xuehong Sun, Yuunhao Li, Ioannis Lambadaris and Yiqiang
Q.Zhao, Performmance Analysis of First - Fit Wavelength Assignment
Algorithm in Optical Networks", IEEE 7th International Conference on
Telecommunications - ConTEL2003, Junell-13, 2003, pp. 403-409.
[15]Vadakital, V.K.M. Hannuksela, M.M. Razaei, M. Gabbouj,
M," Optimal IP Packet Size for Efficient Data Transmission in DVB-H",
Proceedings of the 7th Nordic ,IEEE Signal Processing Symposium, pp.
82-85, 7-9 June 2006.

[16] Ramaswamy Ramaswamy, Ning Weng and Tilman Wolf,
"Characterizing Network Processing Delay", Proc. in Globecom 2004,
IEEE Communication Society, pp.1629- 1634.

AUTHORS PROFILE

243

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

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

Vol. 8, No. 6, 2010

A.Kavitha received the B.E
degree in Electronics and
Communication Engineering
from Madurai Kamaraj
University, India in 1997. The
Master of Engineering degree
in Computer and

Communication Engineering
from Anna University, Chennai
in 2004 and currently pursuing
Ph.D degree in Optical
Communication Networks in
Anna University, Chennai. At
present she is working as a Senior Lecturer in Chettinad College
of Engineering and Technology, Karur. She has published more
than 10 papers in referred National and International
conferences/Journals. Her area of interest includes Networks,
Computer Architecture, Digital Communication and etc.

V.Rajamani received the B.E
degree in Electronics and
Communication Engineering from
National Engineering College, Anna
University, India in 1990. The
Master of Engineering in Applied
Electronics from Government
College of Technology, Bharathiyar
University Coimbatore in the year
1995 and the Ph.D. degree in
Electronics Engineering from Institute of Technology, Banaras
Hindu University, Varanasi in 1999. He started his carrier as
Lecturer in Mohamed Sathak Engineering College from 1991
onwards. He has held various positions in various Engineering
Colleges. At present he is working as a Principal in Indra
Ganesan College of Engineering, Tiruchirappalli. He has
completed a project under AICTE - RPS scheme successfully. He
has published more than 80 papers in referred National and
International Journals/ conferences. His area of interest includes
Device modeling, VLSI, Image processing, Optical
Communication and system.

P.Anandhakumar received the B.E degree in Electronics and
Communication Engineering from University of Madras, India in
1994. The Master of Engineering degree in Computer Science
and Engineering from Bharathiyar University in 1997 and Ph.D
degree in Computer Science and Engg. from Anna University,
Chennai in the year 2006.At present he is working as a Assistant
Professor in Madras Institute of Technology, Chennai. He
published more than 50 papers in referred National and
International conferences/Journals. His area of interest includes
Digital Communication, Soft Computing, Robotics and etc.

244 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

A Feedback Design for Rotation Invariant Feature
Extraction in Implementation with Iris Credentials

M. Sankari

Department of Computer Applications,

Nehru Institute of Engineering and Technology,

Coimbatore, INDIA.

R. Bremananth

School of EEE, Information Engg. (Div.),

Nanyang Technological University,

Singapore.

Abstract — Rotation invariant feature extraction is an essential
objective task in computer vision and pattern credentials
problems, that is, recognizing an object must be invariant in
scale, translation and orientation of its patterns. In the iris
recognition, the system should represent the iris patterns, which
is invariant to the size of the iris in the image. This depends upon
the distance from the sensors to subjects' eye positions and the
external illumination of the environments, which in turn make
the changes in the pupil diameter. Another invariant factor is the
translation, the explicit iris features should be a positional
independent even though eye present anywhere in the acquired
image. These two invariants are perfectly achieved by the weight
based localization approaches. However, the iris orientation
estimation is an important problem to avoid in preserving
selective orientation parameters. Multiple source points are used
to estimate the segmented objects orientations. After estimating
the deviation in angle of segmented object that can be rotated to
its principal origin and then the feature extraction process is
applied. A multi resolution approach such as wavelet transform
is employed for feature extraction process that provides efficient
frequency and spatial texture feature deviations present in the
irises. In this paper, we work on a feedback design with Radon
transform with wavelet statistical analysis of iris recognition in
two different ways. In order to check the viability of the proposed
approaches invariant features are directly compared with
weighted distance (WD) measures, in the first phase and second
phase is to train the Hamming neural network to recognize the
known patterns.

Keywords- Iris credentials; Invariant Features; Rotation
estimation; Multiresolution anlysis;

Introduction

In computer vision and pattern recognition, rotation
invariant feature extraction is an essential task, that is,
recognizing an object must be invariant in scale, translation and
orientation of its patterns. This paper emphasizes on invariant
feature extraction and statistical analyses. In the iris
recognition, the system should represent the iris patterns, which
is invariant to the size of the iris in the image. This depends
upon the distance from the sensors to subjects' eye positions
and the external illumination of the environments that make the
changes in the pupil diameter. Another invariant factor is the
translation where iris features should be a positional
independent of iris pattern, it could occur anywhere in the

acquired eye image. However, the iris orientation estimation is
an important problem to avoid in preserving selective
orientation parameters, for example, 7 relative orientations
were maintained for iris best matching process in the literature
[1] and seven rotation angles (-9, -6, -3, 0, 3, 6 and 9 degrees)
used by Li ma et al. [2]. In the real time imaging, due to the
head tilt, mirror angle and sensor positions, iris images are
captured in widely varied angels or divergent positions. We
estimate the rotation angle of iris portion within the acquired
image by using multiple line integral approaches, which
provide better accuracy in the real time capturing. Local binary
patterns, gray-level and auto-correlation features were used to
estimate orientation of the texture patterns. It projected the
angles that are locally invariant to rotation [3]. In [4], texture
rotation-invariant was achieved by autoregressive models. It
used several circle's neighborhood points to project the rotation
angle of the object. Aditya Vailaya et al. [5] had dealt with
Bayesian learning framework with small code features that are
extracted from linear vector quantization. Thus, these features
can be used for automatic image rotation detection. A hidden
Markov model and multichannel sub-band were used for
estimating rotation angles of gray level images in the study [6].
In this work, we propose Radon transform based multipoint
sources to estimate the rotation angle estimation for real-time
objects.

Classification is a final stage of pattern recognition system
where each unknown pattern is classified to a particular
category. In iris recognition system, a person is automatically
recognized based on his / her iris pattern already trained by the
system. This is done in a way of training a brain to teach
certain kind of sample patterns. In the testing process, system
recalls the trained iris patterns as a weighted distance specified
by the system. If threshold is attained then system genuinely
accepts a person, otherwise false alarm sounds. However, the
way to find the statistical level is a tedious work because it
makes decision to evaluate the pattern either genuine or fake.
Hence combinatorics of iris code sequence should be carried
out by means of statistical independence. Moreover, failure of
iris recognition is principally concerned with a test of statistical
independence because it absorbs more degree-of-freedom. The
test is nearly assured to be allowed whenever the extracted iris
code comparing from two different eyes are evaluated. In
addition, the test may exclusively fail when any iris code is
compared with another version of itself. The test of statistical
independence was implemented by the Hamming distance in

245

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

[1] with a set of mask bits to prevent non-iris artifacts. Li ma et
al. [7] proposed a classifier design that was based on exclusive-
OR operation to compute the match between pairs of iris bits.
In [2], authors worked with the nearest centre classifier to
recognize diverse pair of iris patterns. A competitive neural
network with linear vector quantization was reported for both
identification and recognition of iris patterns by Shinyoung
Lim et al. [8]. Our main contribution to this paper is a feedback
design (Fig. 1) to extract an appropriate set of rotation invariant
features based on Radon and wavelet transforms. An iteration
process is repeated until a set of essential invariant features is
extracted from the subject. We have done two different phases
of statistical analyses of rotation invariant iris recognition.
During phase I, wavelet features are directly compared with
weighted distance (WD) measures and in phase II invariant
features were trained and recognized by the Hamming neural
network.

Rotation

estimation

using multiple

sources

Rotation

Correction to

its principal

direction

Wavelet based
Rotation
Invariant
extraction

No,
Rotat

Fine

ion

i another suitable
invariant Feature

1
s

^ — ^Ts it prov

ide best^^

identification?,^ — ^

Yes ^^>£

Enroll the
Invariant
Features

Fig. 1. A feedback design of rotation invariant feature extraction.

The remainder of this paper is organized as follows: Section
II emphasizes on invariance and estimation of rotation angle.
Radon and wavelet based rotation invariant is described in
section III. Section IV depicts the results obtained based on the
proposed methodologies while Concluding remarks and future
research direction are accentuated Section V.

II. Invariance in Rotation

A 2D rotation is applied to an object by repositioning it along a
circular path. A rotation angle 6 and pivot point about which
the object to be rotated is specified for generating series of
rotation. In counterclockwise, positive angle values are used for
rotation about the pivot point and in contrast clockwise rotation
requires negative angle values. The rotation transformation is
also described as a rotation about an axis that is perpendicular
to the xy plane and passes through the pivot point. The rotation
transformation equations are determined from position
( Xi , y± ) to position ( x 2 , y 2 ) through an angle B relative to the
coordinate origin. The original displacement of the point from
the x-axis is, angle A. By trigonometric ratios, sin(A) = Vj Ir ,

sin(A + Z?) = y 2 lr , cos(A + B) = x 2 /r and
cos(A) = jci / r • From the compound angle formulae described
as

sin(A + B) = sin(A) • cos(B) + cos(A) • sin(B) . (1)
Substituting trigonometric ratios and obtain the following

(y 2 /r) = (y 1 /r)*cos(B) + (x 1 /r)*sin(B), (2)

y 2 = y\ cos(B) + x\ sin(5) , (3)

y 2 = x\ sin(Z?) + yy cos(B) , (4)

Likewise, substituting trigonometric ratios and derived as
cos(A + B) = cos(A) • cos(B) - sin(A) • sin(2?) , (5)

(jc 2 / r) = (jq / r) • cos(£) - ( y l I r) • sin(£) , (6)

x 2 = xi cos(B) - yi sin(i?) , (7)

Therefore, from Eqs. (5) and (10) we can get counterclockwise
rotation matrix and the new coordinate position can be found
as described in Eq. (8). The basics of rotation and line
integrals are incorporated together to form equations for
projecting the object in single and multi source points.

A. Multipoint source

Based on the basics of rotation, multipoint source method
computes the line integrals along parallel beams in a specific
direction. A projection of image f(x,y) is a set of line integrals
to represent an image. This phase takes multiple parallel-
beams from different angles by rotating the source around the
centre of the image.

y 2

cos(B) -sin(5)
sin(£) cos(B)

(8)

This method is based on Radon transform, which estimates
the angle of rotation using the projection data in different
orientations. A fusion of Radon transform and Fourier
transform had been performed for digital watermarking which
is invariant to the rotation, scale and translation invariant in the
literature [9]. A parallel algorithm for Fast Radon transform
and its inverse was proposed by Mitra et al. [10]. Radon
transform was employed for estimating angle of rotated texture
by Kourosh et al. [11]. Image object recognition based Radon
transform was proposed by Jun Zhang et al. [12], this method is
robust and invariant to rotation, scale and translation of image
object. Fig. 2 shows a multipoint source at a specified angle for
estimating rotation angle of a part of iris. This method projects
the image intensity with a radial line orientation at a specific
angle from the multipoint sources. Multipoint projection
computes any angle 9 by using the Radon transform R(x\6) of
f(x,y), it is the line integral of parallel paths to the y axis. After
applying the function of multipoint sources R( x \@) the

resultant data contain row and column. Column describes
projection data for each angle in 9 and it contains the respective
coordinates along the x' axis. The procedure for applying
multipoint source projection to estimate the angle is as follows:
Image is rotated to a specific angle in counterclockwise by bi-
cubic interpolation method.

246

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Rotation in counter
clockwise

Fig. 2. Multipoint estimation using multipoint sources.

Assume the rotation angle from 1° to 180° in order to find the
peak area of rotation angles. After applying the multipoint
sources, Radon transform coefficients have been generated for
each angle. The standard deviation of the Radon transform
coefficients is calculated to find the maximum deviation of
rotation angle. This is shown in Fig. 3. Then, using estimated
angle, object rotation is rotated to its original principal angle
using bi-cubic interpolation method. If the estimated angle 6 is
positive then rotate the object as -( + 90° ) in clockwise
direction else if the estimated angle is negative or above
90° then rotate the object as -(#-90°) in clockwise direction.

Multi point source Radon transform R(X',e)

10 20 30 40 50 60 70 fiO 90 100 110 120 130 140 150 160 170 160
Rotation e (degrees)

Fig. 3. Illustration of orientation angle estimation using multipoint.

III. Iris wavelt Feature analysis

In this phase wavelet based feature extraction process has
been employed to extract feature obscured in the iris patterns. It
is an essential task for recognising a pattern from others
because some features may produce same type of responses for
diverse patterns. It causes the hypothesis in pattern recognition
process to differentiate one from another. To overcome the
problem of uncertainty the system needs an efficient way to
extort quality features from the acquired pattern. Iris provides
sufficient amount of interclass variability and minimises intra-
class variability. Thus the characteristics of these patterns are
well efficiently taken out by the sense of using less
computational process. Among various feature extractors,

wavelet series approximate hasty transitions much more
accurately than Fourier series. Consequently, wavelet analysis
perfectly replicates constant measurements. It produces better
approximation for data that exhibit local variation and because
of its basis function each term in a wavelet series has a compact
support within a finite interval. The other sense to employ
wavelet is orthogonal. This means that information carried by
one idiom is independent of information conceded by the other.
Thus, there is no redundancy in the feature extraction. This is
fine when neither computational sequence time nor storage is
wasted as a result of wavelet coefficient computed or stored.
The next sense related with wavelet is multi resolution, which
is like biological sensory system. Many physical systems are
organised into divergent levels or scales of some variables. It
provides an economic structure and positional notion of
arithmetic whose computational complexity is O (N), where N
data points are to be accessed [13]. In the current literature
various computer vision and signal processing applications
have been based on wavelet theory [14] such as detecting self-
similarity, de-noising, compression, analysis and recognition.
This technique has proven the ability to provide high coding
efficiency, spatial and quality features. However, wavelets
features are not rotation invariant due to its directional changes.
Hence this approach initially estimates the extorted pattern
rotation angle and rotates to its principal direction. Afterwards
multi resolution wavelets have been employed to extort
features from the rotation corrected patterns. In the iris
recognition process, this approach has adopted Daubechies (db)
wavelet to decompose the iris patterns into multiple resolution
sub-bands. These sub-bands are employed to transform well-
distributed complex iris patterns into a set of one-dimensional
iris feature code. Decay is a process to divide the given iris
image into four sub-bands such as approximation, horizontal,
vertical, and diagonal coefficients. A 2D Daubechies wavelet
transform of an iris image (I) can be carried by performing two
steps, Initially, it performs ID wavelet transform, on each row
of (I) thereby producing a new image Ii. In second step it takes
Ii as an input image and performs ID transform on each of its
columns. A Level- 1 wavelet transform of an image can be
described as

a 1 ti

v 1 d 1

a 2 h 2

2 j2

v a

(9)

where the sub-images a 1 , h 1 v 1 and d 1 represent level- 1
approximation, horizontal, vertical and diagonal coefficients
a 2 , h 2 v 2 and d 2 level 2 coefficients. The approximation is
created by computing trends along rows of I followed by
computing trends along columns. Trends represent the running
average of the sub-signals in the given image. It produces a
lower frequency of the image I. The other sub-signals such as
horizontal, vertical and diagonal have been created by taking
fluctuation. It is a running difference of sub-signals. Each
coefficient represents a spatial area corresponding to one-
quarter of the segmented iris image size. The low and high
frequencies represent a bandwidth corresponding to

247

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

0<\cD\<nl2 and tt 1 2 <\co\ < n > respectively. Fig. 4 shows

frequency variation of Daubechies wavelets. The wavelet
transform is defined as

W(a,r x ,r v )= j J I{x,y) ¥a ^ t

>x*>y>

(x, y)dxdy

x v y

^a,T x T y

1 X - T

\a\ a

(10)

(ii)

where I(x,y) is a segmented iris image, W(a,x x ,T y ) is a

wavelet transform function,

^a,T x T y

(x,y)

the wavelet basis

function, a is a scaling factor, T x and T are translation factors

of x and y axes, respectively. The properties separability,
scalability, translatability of discrete wavelet transform is
performed as

<p(x, y) = <p(x)<p(y),i// (x, y) = \ff(x)<p(y) ,
y/ v =(p(x)i//(y),i// D =if/(x)ij/(y),

j M N

i M N

W H ¥ (l,m,n) = -j= X £/(*, y)w" m Jx, y) '

J M N

W v ¥ (l,m,n) = -j== X £/(x, y)y/l m>n {x, y) '

yIMN x=ly=l

W D v/ (l,m,n) =

J M N

^g, /(M) ^ (lj) '

(12)
(13)
(14)

(15)

(16)

(17)

where (p(x, y) and W(p{l^ , in, ri) are scaling function and
approximation coefficients of I(x,y) at scale L , respectively.
W H „ x W V si x W D ,i x are coefficients of horizontal,

YY y/(l,m,n) rr i//(l,m,n) rr i//(l,m,n)

vertical and diagonal details for scales / > / respectively.

Normally / = o, and assigning M = N = 2 so that

/ = 0,1,2..., L - land m = n = 0,1,2,...,2 J " - 1 .

The decomposition of signals produces sub-signals such as
low, middle and high frequency of the components, which
play a very important role in the feature extraction process. In
this approach Daubechies wavelet is employed for feature
extraction process. Its frequency distribution for different level
is illustrated in Fig. 5. The DWT (Discrete Wavelet
Transform) consists of log 2N stages if the given signal s is of
length N. Fig. 6 shows the scaling and wavelet functions of
Daubechies wavelets. Initially s produces two sets of
coefficients such as approximation coefficients cA b and detail

coefficients cDi. These coefficients are obtained by
convolving s with the low -pass filter Lo_D for approximation,
and with the high-pass filter Hi_D for detail coefficients. In
the case of images, a similar procedure is possible for 2D
wavelets and scaling functions obtained from one-dimensional
wavelets by tensorial products. This kind of 2D DWT leads to
a decomposition of approximation coefficients at level j in
four components: the approximation at level j + 1 and the
details in three orientations (horizontal, vertical, and diagonal).
Fig. 7 shows the decomposition process.

Daubechies(dbl> wavelet (rends and fluctuations

te^UJ4* ss *tt

Decomposition high-pass tiler

Reconstruction high-pass filter

Fig. 4. Daubechies (dbl) wavelets frequency variations.

| 0.5

i c

Fig. 5. Frequency distribution of Daubechies wavelets by different iterations.

In the feature extraction process of iris patterns four levels of
decompositions have been performed to obtain fine level of
frequency details from the pattern. The scaling factor is very
important for decomposing the given iris signals. At the first
level it produces 648 signals, second level has 162 signals,
third level 45 signals and finally it generates 15 signals for
each frequency. The MRA produces the frequency signals to
compact approximation of features which aid to generate an
efficient set of distinct features that are provided with less
intra class variability and more interclass variability in the iris
pattern recognition process. Fig. 8 shows four levels of
decomposition process for the given iris images. Low-pass

248

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

filter corresponds to an averaging operation and extract the
coarse information of the signals, whereas high-pass filter
corresponds to a dissimilarity operation that extracts the
detailed information of the signals.

Wavelet fund J on (1)

Wavelet function (2)

Wavelet function (3)

Fig. 6. Scaling and wavelet functions of Daubechies wavelets.

When iris signal passes through these low and high pass
filters, it generates the frequency variation occurring in a
pattern.

Approximation

Lo_D

Rows

Hi_D

Rows Down

sample

- 2 + 1

Columns Down

sample

Lo_D -*- 1+2 -*- ^i+i

Hi_D -+ 1+2 -^ cD h

-* 2+ 1 -

Lo D -^ 1+2 -** cD*

Hi D -*~ 1+2 -+ CD*

Fig. 7. Decomposition of wavelet signals in the feature extraction.

A. Iris feature selection

In this phase frequency variation of iris signals in divergent
levels are quantized into iris features. For that multi resolution
frequencies of low and high pass filters are taken for
quantization process of conversion of real signals into binary.
The mean and standard deviation of approximation and detail
coefficients vary in each level of the decomposition of iris
patterns which raises up to generate an efficient feature sets of
the given patterns. The horizontal, vertical and diagonal
coefficients wavelet features have middle and high frequencies
of the components of iris signals. The histogram analyses of
signals in divergent levels are illustrated in Fig. 9. The
frequency distribution of signals at level 1 ranges from -10 to
10 and from -100 to 100 at level 4 for horizontal coefficients.

Thus this approach quantizes these trends and fluctuation of
sub-signals into iris features. After performing the four level of
decay process the horizontal, vertical and diagonal coefficients,
the iris are used for iris feature encoding process. The
frequency variation occurring in these decomposition
coefficients are employed to extract iris feature codes. In order
to make an efficient set of features and reduce the
computational time of iris matching process the coefficient
values are converted into binary values which senses to create a
compact feature set.

Fig. 8. Four level of decomposition of iris patterns.

s image (72 x i6) analysed at level I witit dl

¥«r«itnl Pi;H*a

Fig. 9. Histogram of divergent levels of iris image.

In the current literature, Haar wavelets are used for iris image
feature extraction by decomposing the signals into four levels
[8]. It uses only high frequency of the components for
representing iris patterns. However, iris patterns are having
middle frequency of the components, which are essential for
recognizing iris patterns in large population. Moreover, in
their approach there is no transformation-invariant analysis.
When there is a rotation between a set of irises from the same
subject, it may produce false positives in the recognition

249

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

because these patterns produce different kinds of features for
diverse rotation and translations. Here, transformation
invariant analysis is performed before extracting features from
the iris patterns. In addition, middle and high frequency of iris
wavelet features were extracted for recognition. Thus, it
reduces less false positives in the recognition process using a
feedback design based on the rotation invariant features.
Though iris patterns are unique in nature, it is a difficult
process to generate identical template for the same subject.
This is mainly due to the changes in imaging position,
distance, illumination conditions, eyelashes / eyelid occlusion
and eyewear reflections. These factors may affect the efficacy
of the system. Thus this approach compensates the
deformation of these factors and recognizes the iris patterns,
which are independent of transformation factors and other
artifacts. The classification results of rotation invariant and
wavelet features are illustrated in Section V.

VI. Rotation Invariant classification

The different pairs of eye images were captured in diverse
distances and illuminations provide more challenges to this
approach. Experimentations were also performed with
different eye images in diverse criteria like normal, outdoors,
contact lens, spectacles, and diseased (Tumours, Tear,
Iridocyclities) eyes. The database of iris images has 2500
images captured from 500 different subjects as each has been
acquired as 5 different images with different real-time
conditions [18, 19]. In the iris matching process, inter and
intra class iris features are efficiently separated and they
prevent impostors from entering into the secure system. To
authenticate any genuine user, iris feature sets are treated as
trained sets and stored in the encrypted file. Verification
subjects' irises are represented as test sets. The same subject
iris feature codes could vary due to external noises, lighting,
illuminations and other factors such as closed eyelashes or
eyelids. This possibly will lead to different iris template for an
eye, even though iris is unique in nature. However, capturing
eye images with advanced biometric camera may solve this
problem. The process by which a user's biometric data is
initially acquired, validated, processed and stored in the form
of a template for ongoing use in a biometric system is called
enrolment. Quality enrolment is a critical factor in the long-
term accuracy of biometric system. Wavelet features of irises
are recognized using the weighted distance (WD). It
recognizes the various classes of iris codes by checking a
minimum distance between two iris codes. This is defined as,
imn4pOFC(x tmined ),iFC(r, .w , where

t))

WD(IFC(x ),IFC(x ))
i J

represents weighted distance in between two iris feature sets
as defined as

WD(iFC(x trained )jFC( Xtest )).

\ IFC ^trained^ IFC ^test)\ ? (1 8 )

where N denotes the number of bits in the iris feature set. The
weighted distance (WD) is used to determine the number of
error bits in between two iris classes. In the experimentation,

the weighted distance of the intra class feature set is
discriminated by the constraint < WD < 0.2 and inter class
iris features is abandoned with the constraint WD > 0.2 . These
distances are also evaluated based on the normal distribution
of mean, standard deviation and degree-of-freedom of the
wavelet iris codes. In addition the same candidate's iris image
may have more artifacts due to various deteriorates as stated
previously. Hence WD needs more discriminability range for
recognising the genuine subject. Conversely, if system
maintained large distance variation to allow the subjects, then
more FAR (False accept rate) might be encountered.
Moreover, if WD is reduced then more FRR (False rejection
rate) may be produced by the system. The system was tested
with normal and abnormal images and their mean of weighted
distance of genuine-class iris codes was u=0.10813, its
standard deviation was a= 0.0392 and degree-of-freedom V
was 62.621991. Impostor-class mean value was u=0.27104
and its standard deviation was g=0. 040730. During the
weighted distance computation, an identical iris pattern was
produced WD = and due to abnormal conditions the same
subject iris was assorted from to 0.19 WD. This is shown in
Fig. 10. If distributions are very large then system allows more
changes for impostors to access the system. This type of
limitation of distributions may be provided with more false
reject rate, but minimum false accept rates. In most of the
applications such as Bank-ATM and biometric voting
machines these type of constrained weighted distance are
essentially desirable in order to agree entire genuine subjects.
In the recognition phase, GAR (Genuine accept rate) was
99.3% and FAR was 0.7% and in confirmation MR (Matching
rate) was 99.94% and FRR was 0.06%.

Normal Distribution Analysis of Ins

Frequency Polygon of Ins codes

ins code Weighted distance

code Weighted distance

Fig. 10. Weighted distance distribution for wavelet iris features and frequency
polygon of the iris codes.

A. Hamming neural network (HNN)

Hamming neural network (HNN) is an alternative way to train
and test the extracted features [15]. This network is employed
to train for both iris and character patterns. Its input layer can
accept wavelet features. That is, it works with bipolar value of
the extracted iris wavelet features. Wavelet based iris feature
codes are fed for recognition process. HNN is used to
recognize iris features from the trained set. The aim of the

250

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

HNN is to decide which trained iris feature set is closest to the
test feature set [16]. HNN consists of two layers; the first layer
is called as a feed forward layer (FFL) that is used to calculate
a maximum score of the input patterns and a recurrent layer
(RL) is used to select the maximum score among the input
patterns. Each neuron in the FFL is set up to give maximum
response to one of the trained patterns. If test set is same as
trained set the maximum score is taken by the recurrent
network. The weights initialization process of the HNN is
described as

W ;;

2 '

Wi0 = n/2 ,

(19)

where Wy and x y are the weights value and input features of f 1
bit of the i th iris feature, Wi is a bias value and n is the number
of bits in the iris features. In order to incorporate HNN with
iris recognition, wavelet features are converted from binary to
its corresponding bipolar form. For example, the i th iris feature
set is {-1, +1, +1, -1, +1...+1}. The weight of i th neuron is set
to {Wio= 67.5, Wii= -0.5, w i2 = 0.5, w i3 = 0.5, w i4 = -0.5, w i5 =
0.5, ... , Wii35= 0.5}. The weighted sum is 135. Each of the
neurons in the FFL gives a maximum response of 135 for
exactly identical iris codes, and a minimum value to other
features. In HNN, the number of neurons in the FFL is same as
the number of neurons in the recurrent layer. When a test
feature is given to the FFL, the output from each of the
neurons in the FFL is measured by the Hamming distance
from the iris in the training set. The Hamming distance
between two iris patterns is a measure of the number of bits
that are different between the two iris patterns. For example, if
an input iris pattern of {+1, +1, +1, -1, -1...+1} is fed then the
output of FFL is 133 which has 2 less than the maximum of
135. This is because the given pattern has 2-bit difference.
Perhaps, if entire bits are changed in the iris patterns, the
neuron that corresponds to that pattern produces an output of
0. The function of RL is to select the neuron with the
maximum output. The final output of the RL contains a large
positive value at the neuron that corresponds to the nearest iris
pattern, and all other neurons produce value. The RL is
trained by setting the weight to 1, if the weight connection
corresponds to the same neuron and all other weight value are
small negative value less than -1/TI. The response of the RL is
described as

Y =

^ w { . x t if 2* w / • x i — A*

otherwise

(20)

where TI is the total iris patterns available in the trained set, X
is a threshold maintained in the iris recognition process. In this
process, the output is fixed to the value of the output of the
FFL. The RL is allowed to iterate, initially the outputs of the
RL is equal to the score produced by the FFL. Then, because
of the less than -1/TI weights, the output is gradually reduced.
After some iteration, all the outputs reach except the
recognized pattern with threshold, for example,^ =81, i.e.

weighted distance for HNN, WD h is 0.6. The testing process

of HNN is stopped if there is no change expected in the
iterations.

V. Experiments Invariant classification

Experiments were carried out on cases like left and right eye
evaluation, twins eye evaluation, eyewear and artifact
evaluation, hypothesis test, segmented iris, normalized iris,
Receiver operating characteristics curve (ROC) evaluations
and feature vector dimension variations. To evaluate these
phases, system was tested based on GAR, FAR and FRR
factors of the recognition.

A. Fusion of left and right eyes

Evaluating both the left and right eye combinations provide
better security in the application domains. However, the
recognition time is directly prepositional with the number of
entries in the iris database. A pair of 120 subjects' eye images
was acquired to test the algorithm, that is, a total of 240 iris
patterns were trained and tested by the ENDM, weighted
distance and HNN. The feature vector size is double the
dimension of normal vector. Thus, 270 wavelet iris features
were computed for each subject to test the weighted distance
and HNN. Table I depicts the recognition rates for evaluating
both left and right eye images. In the recognition process, a
system was set by a matching threshold level. It determines
the error tolerance of the system with respect to the features
for which the network is trained, and is used to determine
whether the final result is accepted / rejected. For any
recognition system that is used for security applications, this
error tolerance should be minimal and therefore the setting of
this matching threshold is a crucial factor. Recognition rates
were reported based HNN and WD. WD was better
recognition rates with minimal FAR. Furthermore, its FRR
was also an acceptable one, hence the system with wavelet and
WD produce good performance than the HNN.

TABLE I.

Comparison of classifiers accuracy rate

Feature

Types of
classifier

Left and
right iris
features

Matching

Threshold

Range

Recognition rate

GAR

FAR

FRR

Wavelet

270

[0.0-0.19)

99.4

0.6

1.2

Wavelet

HNN

270

[0.7-1.0)

99.32

0.68

2.7

B. Recognition of twins

Identical twins' irises were separately verified with different
methods. From 50 twins, 500 eye images were acquired. It
contained both left, right eye images with each subject having
10 eye images. The twins' iris code result was generated by
the classifiers as the same weighted distances as the regular
iris codes available in the database. The mean of WD in the
images acquired from twins is 0.086360 with the standard
deviation 0.044329. A confidence interval is a range of values
that have a chosen probability of containing the true

251

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

hypothesized quantity. The standard deviation of confidence
intervals is in the range 0.0417 to 0.0473. Fig. 11 shows the
normal distribution of twin's iris code weighted distance. In
the checking of twin's iris database WD was changed in the
range to 0.19, GAR was 99.3 and FAR was 0.7.

D. Iris hypothesis test

Hypothesis test plays very important role in the biometric
recognition system, i.e., making a decision is based on the
availability of data during the training or enrolment and testing
or verification processes.

C. Eyewear and Artefacts

Eyewear images are major problematic ones in the iris
recognition because these images may produce more false
localization and FAR or FRR of the system. To evaluate the
recognition rates of eyewear images, 50 subject's eye images
were acquired with white glasses from each of them 5 images
were captured, that is total of 250 images were utilized for the
recognition. As stated previously, an exact identical iris
pattern could be produced WD=0, but due to eyewear noises
and other artifacts its patterns require a certain WD range.
Hence system evaluated the same subjects' iris patterns before
and after wearing the eyewear. This also included soft contact
lens and white glass with different varieties. Table II shows
the WD on the image with and without wearing eyewear. In
that hard contact lens produced more FRR in the recognition.
Thus it was around 32 bits average error bits and WD was
0.237. Moreover, localization system may be disrupted by the
designed frames of eyewear in the hypothesis to locate the
ROI. However, in the recognition, it produced minimal
average error of 22-bits.

Identical twins Normal Distribution Analysis of Iris codes

TABLE II.

Eyewear noises and other artifacts assessment

1 ** 1

i i

1 '

A A A

■

£.

Inti a-clasi

Inter-class

r *

■

ji=0.269047

]A=O,0S6360

o=0.050261

f a=0.044329

■

•

A
A
A

0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45

Iris code Weighted distance

Fig. 11. Representation of weighted distance of twins' iris code.

The FRR and FAR was high when images were acquired with
eyewear and in diverse illuminations such as sunshine and
twilight conditions, eyewear at twilight the average of 34 bits
were corrupted, therefore WD was 0.251. As a consequence,
the system recommends the application domain while
enrolling a void eyewear because during enrolment iris
patterns could be signed up with minimum amount of error
bits. Therefore, it increases system recognition rates in order
to achieve better rotation invariant feature set.

Types of eye wears

Without eyewear

With eyewear

Average
Error bits
out of 135

Average

error in

Average
Error bits
out of 135

Average

error in

White glass

0.104

0.165

Soft contact lens

0.112

0.194

Hard contact lens

0.112

0.237

Sunshine

0.129

0.198

Twilight

0.138

0.251

This test may be neither true nor false. It could be dependent
on the feature extraction and classifier design of the system.
Thus this system makes iris images as transformation invariant
patterns to increase the performance of the system. The
biometric estimation is based on some terms of assumptions
that is, make a system as null hypothesis. The null hypothesis
is the original declaration. In iris recognition the null
hypothesis is specified by the WD range between 0.0 and 0.2
for the genuine subject. The significance level is another term
related to the degree of certainty that the system requires in
order to reject the null hypothesis in favor of the alternative
hypothesis. By taking a small sample the system cannot be
certain about the conclusion. So decide in advance to reject the
null hypothesis if the probability of observing the sampled
result is less than the significance level. A typical significance
level is 0.21. The p-value is the probability of observing the
given sample result under the assumption that the null
hypothesis is true. If the p-value is greater than the WD range,
then system rejects the null hypothesis. For example, if
WD= 0.2 and the p-value is 0.22, then the system rejects the
null hypothesis. The results of biometric for many hypothesis
tests also include confidence intervals. That is, a confidence
interval is a range of values that have a chosen probability of
containing the true hypothesized quantity. An illustrative
example, WD = 0.03 is inside a 97% confidence interval for
the mean. That is equivalent to being unable to reject the null
hypothesis at a significance level of 0.03. Conversely, if the
100(1 -WD) is confidence interval that does not contain
weighted distance range then the system rejects the null
hypothesis at the level of significance.

E. Receiver operating characteristics curve analysis

The ROC analysis of wavelet features with WD and HNN
is illustrated in Fig 12. The both WD and HNN classifiers
produced approximately the same amount of accuracy.
However, WD produced quite better exactness than the HNN
since it requires minimal error tolerance and threshold in

252

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

training and testing processes. Thus HNN loosely allows
impostors more than WD.

ROC of Wavelet WD

;© J ; "

I f ■■■■■-■ ■•■•-|--'

ROC of Wavelet HNN

0.98

B 0.96

1 ° 92

9
0.88

0.6i3

! ; i

; j] [ ■

005 0.1

False Accept Rate

0.O5 0.1
False Accept Rate

Normal Conditions : ' Outdoor Conditions Contact lens and spectacles Diseased iris |

Fig. 12. ROC analysis wavelet features with WD and HNN.

F. Performance comparison

In this work, a feedback design for extraction of rotation
invariant iris recognition based on local segmentation of iris
portions was suggested. It prevents misclassifications (FAR)
of iris patterns and limits the overall FRR of the system. As
per research work, 40% of iris images have been obscured by
eyelids / eyelashes and 35% of images hid the top portions of
iris. This system pulls out left, right and bottom local area of
iris for iris code extraction. It provides overall accuracy of
98.3% in the iris localization process. In [2], elastic
deformation has occurred in the iris portion due to
illumination changes. It was compensated to convert the
circular portion of the iris (including eyelids / eyelashes) into a
rectangle strip, which was used for convolution operations in
iris matching. In the present work these types of discrepancies
have been resolved by local segmentation process. In addition,
the previous method influenced by seven rotation angles (-9, -
6, -3, 0, 3, 6 and 9 degrees). But in our proposed system, the
rotation-invariance was achieved by on combination of Radon
transform and wavelet feature sets. In [1], 2048 feature
components were used to classify the diverse iris patterns. It
readily achieved scale and translation invariant pattern
analysis using integrodifferential operator. However, rotation-
invariant might be carried out by shifting of iris phase codes.
So, it inclined sequences of orientation of templates for the
recognition process. In our approach, we employed with
sequences of rotation estimation preprocessing based on the
Radon transform in order to extract the rotation invariant
features, which in turn, influence a distinctive template for
each subject in enrolled of the system. In [3], Shiny oung Lim
et al. suggested an approach based on Haar wavelet with linear
vector quantization method. This method worked with 87 high
pass filter of the wavelet transformation. However, middle
frequencies of the iris patterns are very useful in the
recognition. In our present work both middle and high

frequencies of wavelet component of iris patterns are used.
Additionally, transformation-invariant is efficiently achieved
prior to the feature extraction. Therefore, multiple iris features
or additional shift operation is completely avoided in the
proposed methodology. Thus, this paper provides better
accuracy with compact rotation invariant feature set than
previous methods.

IV. Conclusion and Future work

This paper processes a feedback design for rotation invariant
feature extraction in application with iris patterns using Radon
and wavelet analysis. After correcting rotation angle, rotation
invariant contours are processed by feature extractor
repeatedly until a suitable set was encountered. It increases
more recognition rate and rotation estimation with diverse
artifacts than the other methods since the previous methods
used redundant patterns of iris feature templates for different
angle of capturing or additional shift operation for
compensating the invariants. Suggested methods would be
possibly implemented with other applications of object
rotation estimation and recognition. This paper opens a new
direction of research in the vision and biometric committees.

Acknowledgment

Authors thank their family members and children for their
continuous support and consent encouragement to do this
research work successfully.

References

[1] Daugman J., 'How Iris Recognition Works', IEEE Transactions On
Circuits and Systems For Video Technology, vol. 14, no. 1, pp. 21-30,
2004.

[2] Li ma, Tieniu Tan, Yunhong Wang and Dexin Zhang, 'Personal
Identification Based on Iris Texture Analysis', IEEE Transactions on
Pattern Analysis and Machine Intelligence, vol. 25, no. 12, pp. 1519-
1533,2003.

[3] Pietikainen M., Ojala T. and Xu Z., 'Rotation- invariant texture

classification using feature distributions', Pattern Recognition, vol. 33,

pp. 43-52, 2000.
[4] Mao J. and Jain A. K., 'Texture classification and segmentation using

multiresolution simultaneous autoregressive models', Pattern

Recognition, vol. 25, no. 4, pp. 173-188,1992.

[5] Aditya Vailaya, Hong Jiang Zhang, Changjiang Yang, Feng-I Liu and
Anil K. Jain, 'Automatic Image Orientation Detection', IEEE
Transactions on Image Processing, vol. 1 1, no. 7, pp. 746-755, 2002.

[6] Chen J. L. and Kundu A. A., 'Rotation and Gray scale transformation
Invariant Texture Identification Using Wavelet Decomposition and
Hidden Markov Model', IEEE Transactions on Pattern Analysis and
Machine Intelligence, vol. 16, no. 2, pp. 208-214, 1994.

[7] Li ma, Tieniu Tan Yunhong Wang and Dexin Zhang, 'Efficient Iris
Recognition by Characterizing key Local variations', IEEE Transaction
on Image Processing, vol. 13, no. 6, pp. 739-750, 2004.

[8] Shinyoung Lim, Kwanyong Lee, Okhwan Byeon and Taiyun Kim,
'Efficient Iris Recognition through Improvement of Feature Vector and
Classifier', ETRI J., vol. 23, nNo. 2, pp. 61-70, 2001.

[9] Lian Cai and Sidan Du, 'Rotation, scale and translation invariant image
watermarking using Radon transform and Fourier transform',
Proceedings of the IEEE 6th Circuit and systems Symposium Emerging

253

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Technologies: Mobile and Wireless Communication, Shanghai, China,
pp. 281-284, 2004.

[10] Mitra Abhishek and Banerjee S., 'A Regular Algorithm For Real Time
Radon and Inverse Radon Transform', Proceedings of IEEE Acoustics,
Speech and Signal Processing (ICASSP), Montreal, Quebec, Canada, pp.
V.105-V. 108, 2004.

[11] Kourosh Jafari-Kkouzani and Hamid Soltaian-Zadeh, 'Rotation-
Invariant Multiresolution Texture analysis using Radon and Wavelet
Transforms', IEEE Transactions on Image Processing, vol. 14, no. 6, pp.
783-795,2005.

[12] Jun Zhang, Xiyuan Zhou and Erke Mao, 'Image Object Recognition
based on Radon Transform', Proc. of IEEE 5th World Congress on
Intelligent Control and Automation, Hangzhou, China, pp. 4070-4074,
2004.

[13] Haward L. Resnikoff and Raymond O. Wells , 'Wavelet Analysis-The
Scalable Structure of Information', Springer- Verlag, New York (ISBN:
81-8128-226-4), 1998.

[14] James S. Walker, 'A Primer on Wavelets and their Scientific
Applications', CRC Press LLC, USA, 1999.

[15] Phil Picton, 'Introduction to Neural Networks', The Macmillan Press
Ltd., First edition, Great Britain (ISBN:0-333-61832-7), 1994.

[16] Bremananth R., and Chitra A., 'A new approach for iris pattern analysis
based on wavelet and HNN' , Journal of CSI, vol. 36, no. 2, pp. 33-41
(ISSN: 0254-7813), 2006.

[17] Bremananth R., Chitra A., 'Real-Time Image Orientation Detection and

Recognition', International Conference on Signal and Image Processing

(ICSIP), Dec. 2006, pp. 460-461.
[18] Bremananth R., and Chitra A, 'Rotation Invariant Recognition of Iris',

Journal of Systems Science and Engineering, Systems Society of India,

vol.17, no.l, pp.69-78, 2008.

[19] Bremananth R., Ph.D. Dissertation, Anna University, Chennai, India,
2008.

AUTHORS PROFILE

recognition, Analysis of algorithms, Data structure, Computer graphics
and multimedia.

Bremananth R received the B.Sc and M.Sc.

f degrees in Computer Science from Madurai

Kamaraj and Bharathidsan University,
respectively. He obtained M.Phil, degree in
^ Computer Science and Engineering from
Government college of Technology, Bharathiar
University. He received his Ph.D. degree from
^^^^ ^^^^ Department of Computer Science and Engineering,
PSG College of Technology, Anna University, Chennai, India.
Presently, he is working as a Post-doctoral Research Fellow, at Nanyang
Technological University, Singapore. He received the M N Saha
Memorial award for the best application oriented paper in 2006 by
Institute of Electronics and Telecommunication Engineers (IETE). His
fields of research are acoustic imaging, pattern recognition, computer
vision, image processing, biometrics, multimedia and soft computing.
Dr. Bremananth is a member of Indian society of technical education
(ISTE), advanced computing society (ACS), International Association of
Computer Science and Information Technology (IACIT) and IETE.

Mrs. M. Sankari received her B.Sc. and M.Sc.
degrees in Computer science from Bharathidasan
University, respectively. She has completed her
Master of Philosophy degree in Computer science
from Regional Engineering College, Trichy.
Presently, she is a Head of the department of MCA
at MET and pursuing her doctorate degree in
computer science at Avinashilingam University,
She has published various technical papers at IEEE

Her field of research includes Computer vision, Pattern

254

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Empirical Mode Decomposition Analysis of Heart Rate Variability

C.Santhi.M.E., Assistant Professor,
Electronics and Communication Engineering, Government College of Technology, Coimbatore-64 1 013

N.Kumaravel Ph.D

Professor, Head of the Department,

Electronics and Communication Engineering, Anna University,Chennai-600 025.

Abstract

The analysis of heart rate variability (HRV)
demands specific capabilities not provided by either
parametric or nonparametric spectral estimation methods.
Empirical mode decomposition (EMD) has the possibility of
dealing with nonstationary and nonlinear embedded
phenomena, for a proper assessment of dynamic and
transient changes in amplitude and time scales of HRV
signal. In this work EMD and a non-linear curve fitting
technique are used to study half an hour HRV signal and its
intrinsic mode function obtained from 20 healthy young
control subjects, 20 healthy old control subjects and 20
subjects with long term ST. The intrinsic oscillations are
measured by means of its meanperiod and variance.
Significant meanperiod reduction is observed in the intrinsic
time scales of healthy old control subjects and subjects with
long term ST, which is used to classify the three groups of
HRV signal with high sensitivity and specificity. The
estimated slope using the non-linear curve fitting technique
represents the flexibility of the cardiovascular system. The
main advantage of this method is it does not make any prior
assumption about the HRV signal being analyzed and no
artificial information is introduced into the filtering method.

Index Terms- Empirical Mode Decomposition, Heart Rate
Variability, Intrinsic Mode Functions, RR intervals,
nonlinear curve fitting.

1. Introduction

Over the last 20 years there has been widespread interest in
the study of variations in the beat-to-beat interval of heart
known as heart rate variability (HRV) or RR interval
variations. HRV has been used as a measure of mortality
primarily with patients who have undergone cardiac
surgery. Clinical depression strongly associated with
mortality with such patients may be seen through a decrease
in HRV [1]. HRV is a non invasive measure of autonomic
nervous system balance. Heart rate is influenced by both
sympathetic and parasympathetic (vagal) activities of ANS.
The influence of both branches of the autonomic nervous
system (ANS) is known as sympathovagal balance reflected
in the RR interval changes. A low frequency (LF)
component provides a measure of sympathetic effects on the
heart and generally occurs in a band between 0.04 Hz and
0.15 Hz. A measurement of the influence of the vagus nerve
in modulating the sinoatrial node can be made in the high
frequency band (HF) loosely defined between 0.15 and 0.4
Hz known as respiratory sinus arrhythmia (RSA), and is a
measure of cardiac parasympathetic activity. The ratio of

power in the LF and HF bands (LF/HF) provides the
measure of cardiac sympathovagal balance. Empirical Mode
Decomposition (EMD) retains the intrinsic nonlinear
nonstationary property of the signal. Any intrinsic timescale
derived from the signal is based on the local characteristics
timescale of the signal [2-4]. EMD carries out layer upon
layer sifting and obtains ordered array components from
smallest scale (highest frequency) to largest scale (lowest
frequency) [4]. Empirical mode decomposition has the
possibility of dealing with nonstationary and nonlinear
embedded phenomena, and owing to its suitability for a
proper assessment of the dynamic and transient changes in
amplitude and in frequency of the HRV components [2& 3].
Application of EMD to half an hour HRV data
yields nine intrinsic mode functions (IMFs). The first scale
represents the highest frequency or the shortest period
component of the signal. The second scale represents the
lower frequency or the longer period component of the
signal. Similarly the last IMF represents the lowest time
scale present in the HRV signal. The first two scales contain
more than 85% of total signal power. The meanperiod and
variance of IMFs are computed as time domain measures.
The variance of IMF decreases exponentially with respect to
increasing timescales (meanperiods). Using nonlinear curve
fitting technique the IMFs variations are estimated. The
estimated parameter represents the flexibility of the
cardiovascular system.The methodology is applied to HRV
signal obtained from 20 healthy young control subjects, 20
healthy old control subjects and 20 subjects with long term
ST. The intrinsic time scale of IMF 2 classifies the three
groups HRV signal with high sensitivity and specificity.

2. Empirical Mode Decomposition (EMD)

EMD is a procedure oriented adaptive method for
decomposing non-linear non- stationary signals. The
components resulting from EMD are called Intrinsic Mode
Functions (IMFs) [2]. The IMFs are amplitude frequency
modulated intrinsic signals. The IMF's represents the
oscillatory modes imbedded in the signal. It should satisfies
the following two conditions. 1. In the whole data set the
number of extrema's and the number of zero crossings must
be either equal or differ by at most one. 2. At any point the
mean value of the envelope defined by the local minima and
the envelope defined by the local maxima is zero.

255

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

Fig. 1 .RR interval signal Fig.2. Intrinsic Mode Functions

tot

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Step 5: Check h(t) for the conditions of an Intrinsic Mode
Functions. [2]

If h(t) is an IMF compute residue r(t)=x(t)-h(t) and again
the process is repeated to extract the next IMF. If h(t) is not
an IMF x(t) is replaced with h(t) and the procedure is
repeated from step 1. Fig.6 shows all IMFs of the signal
x(t).

The process ends when the range of residue is below a
predetermined level or the residue has a monotonic trend. In
order to guarantee that the IMF components retains enough
physical sense in both amplitude and frequency
modulations, the sifting process is stopped by limiting the
size of standard deviation(SD) which is computed from two
consecutive sifting results.

Fig. 3. Reconstructed signal FigADetrended signal

Figs 1-4 explain the efficiency of EMD for RR
interval signal. The ECG data has been collected from the
biomedical website [7] http://www.physionet.org . The RR
intervals are derived from half an hour ECG signal by
identifying the QRS complexes. The signal is manually
edited and only noise free ectopic free segments are used for
the analysis. A real time RR interval signal and its EMD
decomposed IMFs are shown in Fig.l&2. Application of
EMD to real time RR interval signal identifies eight to nine
IMFs. The IMFs are zero mean amplitude frequency
modulated signal. The decomposition is adaptive and
lossless. The original RR interval signal is reconstructed
using decomposed IMFs (Fig. 3). The nonstationary trend is
removed when the residue or monotonic trend (last IMF) is
omitted while reconstructing the signal (Fig.4).

From the RR intervals the HRV signal or ARR
signal (Ri+i-Ri) is obtained by computing successive
difference between consecutive RR intervals. The obtained
HRV signal and its IMFs are shown in Fig. 5 and Fig.6.
Matlab 7.1 tools are used for the analysis.

3. Methodology

SIFTING ALGORITHM:

Fig.5.HRV signal Fig.6. Intrinsic Mode Functions

Step 1 : All the minima and maxima of the HRV Signal x (t),
are located.

Step 2: Spline Interpolate the minima and maxima points to
obtain lower and upper envelopes of the signal.

Step 3: Compute mean envelope

m (t)=(maxima's+minima's)/2.

Step 4: Subtract local mean from the original Signal to
obtain local details h(t)=x(t)- m(t).

sd = £[|V-d(0-M0| 2 /*» Vd(0] (i)

t=0
where k represents number of siftings.

The process of finding an intrinsic mode function
requires number of iterations and the process to find all the
IMFs requires further more iterations. As a result of this
iterative procedure finally yields many IMFs and a residue.
By summing up all the IMF functions and the residue, the
original signal is reconstructed, given by the mathematical
formulae

X(t) = £h,(t)+r(n)

(2)

1=1

Where each hi represents an intrinsic mode function and r(n)
either a mean trend or a constant.

For each IMF the meanperiod and variance are
computed. The meanperiod is the ratio of distance between
the first and last zero-crossings to number of zero-crossings
of IMF.

The obtained RR interval signal using ECG
represents the response of the cardiovascular system to ANS
activities not the ANS activities themselves. The
characteristics of cardiovascular system determine how the
system responds to ANS activity and can alter significantly
the characteristics of the HRV signal. The response
characteristics are often nonlinear in nature. The IMFs
capture the all the variations present in the HRV signal.
Plotting the variance of all IMFs against its meanperiods
gives a nonlinear function. The variance of IMF decreases
with increasing meanperiod and this behavior is
approximated using a geometric function

Y= aX b ~ (3)

where Y represents vector of IMF's variance,. X represents
vector of meanperiods of IMFs, a is constant and b is the
exponential decrease of the function. The IMFs meanperiod
and variance of healthy young control subjects, healthy old
control subjects and long term ST subjects vary
significantly. The variations in the IMF are quantified by
the parameter b. The parameter b represents the flexibility
of cardiovascular system to ANS activities. The parameter b
is estimated using nonlinear curve fitting technique
explained below.

256

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

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

Taking logarithm of equation (3), gives
lnY = lna + blnX

(4)

putting Y =ln Y , X =ln X , A = In a then the above
nonlinear equation becomes Y = A + bX which is a linear
equation in X. The corresponding normal equations are

2X = wi*+b5X (5 )

_*2

£jry*=A*jx+bjx

(6)

Fig. 7. Curve fitting
N represents number of IMFs. Solving the normal equations
using least mean square method the variables 'a' and 'b' are
estimated [5]. The simulated response function using the
estimated parameter is shown in Fig. 7.

4. Results and Discussion

EMD and curve fitting techniques are applied to half an
hour HRV signal of 20 healthy young control subjects, 20
healthy old control subjects and 20 subjects with long term
ST. Empirical mode decomposition adaptively decomposes
the half an hour HRV signal into number of Intrinsic mode
functions (Fig. 6). The analysis is done with ARR intervals.
ARR (R i+1 -Ri) represents the difference between successive
beat intervals. The IMFs are measured by their absolute
variance, relative variance and meanperiods. The
meanperiod is the ratio of distance between the first and last
zero-crossings to the number of zero-crossings of IMF. First
3 IMFs contains more than 92% of total variance. The IMF1
represents the highest frequency or the shortest period
component of the signal. The IMF2 contains the lower
frequency or the longer period component of the signal.
Since the 1 st and 2 nd IMF contains more than 85% of total
power they are very significant.

Relative powers are computed with respect to total power
considering all IMFs except the residue with zero
meanperiod. In healthy young subjects an increase in
relative power of IMF1 decreases the relative power of
IMF2 (Fig. 8). IMF 1 and IMF 2 are in phase opposition
representing different components of the HRV signal. The
original signal is interpolated to 2 Hz for a meaningful
frequency measure. The Welch periodogram (with window
width 1024 and window overlap of 512 samples) of IMFs of
a healthy young control subject are shown in Fig. (9). Table-
1 gives the peak frequency(Hz) and absolute spectralpower
(ms 2 -miliseconds square) of IMFs The figure shows the
frequency spectrum of the IMFs falls in the recognized
spectral bands of HRV signal: 1 .High frequency band from
0.1 5Hz to 0.5Hz; 2. Low frequency band from 0.04Hz to

0.15Hz; 3. Very low frequency band from 0.01Hz to
0.04Hz..

Relative powers of IMF 1 and IMF 2

Relative power

6
5

3
2

nlfl

ititfl ww\

■ Relative powers of IMF 1

■ Relative powers of IMF 2

1 3 5 7 9 11 13 15 17 19
Healthy young control records

Fig 8: Relative powers of IMF 1 and IMF 2

-A

:.A

Fig 9: Welch periodogram of IMFs

IMFs

Peak

frequency in
Hz

Peak power in
ms 2

IMF1

0.2891

0.01

IMF2

0.13

0.003

IMF3

0.068

0.002

IMF4

0.03

0.00069

IMF5

0.021

0.0007

IMF6

0.01

0.00062

Table- 1 Spectral values of IMFs

The meanperiod of IMF2 of healthy young controls subjects
are significantly higher compared to healthy old controls
subjects and subjects with long term ST. Considering
meanperiod of IMF2 (2.9724 sees) as threshold value, we
classified the healthy young control subjects and subjects
with long term ST with 95% sensitivity and 90% specificity.
The classification is shown in Fig. (10). A threshold value
of 2.8 sees classifies the healthy old controls subjects and
subjects with long term ST with 90% sensitivity and 70%
specificity shown in Fig .(11).

IMF 2 meanperiod of healthy young and subjects with long term

3.5
3

2.5
2

1.5
1

0.5

* v/A *S*~*^^/\ ^*-^ /

\ r\v ^^ : /O^/

\J ^r-m-mr-^ ^ ^ ~W \J/

—♦—Healthy young
—■—Sub. With long term st

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Fig. 10. Meanperiod comparison of healthy young subjects
and subjects with longterm ST.

257

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

IMF 2 meanperiod of healthy old and sub.with long term st

-Healthy old controls
-sub.with long term st

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Fig.l 1. Meanperiod comparison of healthy old subjects and
subjects with longterm ST.

The parameter b of IMFs of the three groups HRV signal
are estimated, average plots are shown in Fig. (12). The
estimated parameter b of healthy young control subjects,
healthy old control subjects and long term ST subjects are -
1.49, -1.43 and -1.39 (average values only). The more
negative value represents the flexibility of the system. The
healthy young control subject's cardiovascular system is
more flexible than healthy old subjects and longterm ST
subjects. The flexibility of the system decreases in healthy
old control subjects and longterm ST subjects. The absolute
powers of healthy young control subjects are significantly
higher compared to healthy old subjects and long term ST
subjects as shown in Fig. (13) (average values only). The
higher values of absolute power represent more fluctuating
power in the signal. The results show the HRV of healthy
young control subjects contains higher power, longer time
scales and more adaptive to ANS activities compared to
healthy old control subjects and subjects with long term ST.

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 6, September 2010

rate variability analysis", Med.Bio.Eng.Comput., 2001, 39,

471-479.

[3]E.P.Souza Neto, M.A.Custaud, J.C.Cejka,P.Abry,

J.Frutoso, C.Gharib, P.Flandrin, "Assessment of

Cardiovascular Autonomic Control by the Empirical Mode

Decomposition", Methods Inf Med 2004;43:60-5.

[4] N.E.Huang, Z.Shen, S.R.Long, M.C.Wu, H.H.Shih,

etal.1998. "The empirical mode decomposition and the

Hilbert spectrum for nonlinear and nonstationary time series

analysis" Proc.R.Soc.A, vol 454, pp.903-995.

[5] B.V.Ramana, "Higher Engineering Mathematics", Tata

McGraw-Hill Publishing Company Limited,New Delhi.

[6] HRV Analysis Software 1.1, developed by The

Biomedical Signal Analysis Group, Department of Applied

Physics, University of Kuopio, Finland.

http://venda.uku.fi/research/biosignal

[7] www.physionet.org.

[8] Jan W.Kantelhard, Stephan A, Armin Bunde, 2002,

Multifractal Detrended Fluctuation Analysis of

Nonstationary Time Series, Physica A 316, 87-1 14.

Absolute powerslMFlfi IMF2

0.004

Absolute 1
powers 1

0.9)1- '

-lOp[k

tlZoHMR

Young 2. Old 3. Long termST

Fig. 12 Correlation graphs Fig. 13. Absolute powers

5. Conclusion

In order to cope up nonlinearity and nonstationarity
issue of HRV signal EMD and nonlinear curve fitting
techniques are used in this work. The IMFs of HRV signal
are negatively correlated. The frequency spectrum of first
two IMFs falls in the recognized HF and LF spectral bands
of HRV signal. The meanperiod of IMF2 classifies half an
hour HRV signal of healthy young control subjects, healthy
old control subjects and subjects with long term ST with
high sensitivity and specificity. The nonlinear curve fitting
technique estimates the flexibility of cardiovascular system.
The method is simple, adaptive and no artificial information
is introduced in the analysis.

6. References

[1] R. M. Carney, J. A. Blumenthal, P. K. Stein, L.
Watkins, D. Catellier, L. F. Berkman, S. M. Czajkowski, C.
O'Connor, P. H. Stone, K. E.Freedland, "Depression, Heart
Rate Variability, and Acute Myocardial Infarction,"
Circulation, vol. 104, no. 17, pp. 2024 - 2028, 2001.
[2] J.C.Echeverria, J.A.Crowe, M.S.Woolfson, B.R.Hayes-
Gill, "Application of empirical mode decomposition to heart

258

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

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

Vol. 8, No. 6, September 2010

IJCSIS REVIEWERS' LIST

Assist Prof (Dr.) M. Emre Celebi, Louisiana State University in Shreveport, USA

Dr. Lam Hong Lee, Universiti Tunku Abdul Rahman, Malaysia

Dr. Shimon K. Modi, Director of Research BSPA Labs, Purdue University, USA

Dr. J ianguo Ding, Norwegian University of Science and Technology (NTNU), Norway

Assoc. Prof. N. Jaisankar, VIT University, Vellore,Tamilnadu, India

Dr. Amogh Kavimandan, The Mathworks Inc., USA

Dr. Ramasamy Mariappan, Vinayaka Missions University, India

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

Assist. Prof. Sugam Sharma, NIET, India / Iowa State University, USA

Dr. Jorge A. Ruiz-Vanoye, Universidad Autonoma del Estado de Morelos, Mexico

Dr. Neeraj Kumar, SMVD University, Katra (J&K), India

DrGenge Bela, "Petru Maior" University of Targu Mures, Romania

Dr. J unjie Peng, Shanghai University, P. R. China

Dr. Ilhem LENGLIZ, HANA Group - CRISTAL Laboratory, Tunisia

Prof. Dr. Durgesh Kumar Mishra, Acropolis Institute of Technology and Research, Indore, MP, India

Jorge L. Hernandez-Ardieta, University Carlos III of Madrid, Spain

Prof. Dr.C.Suresh Gnana Dhas, Anna University, India

Mrs Li Fang, Nanyang Technological University, Singapore

Prof. Pijush Biswas, RCC Institute of Information Technology, India

Dr. Siddhivinayak Kulkarni, University of Ballarat, Ballarat, Victoria, Australia

Dr. A. Ami Lawrence, Royal College of Engineering & Technology, India

Mr. Wongyos Keardsri, Chulalongkorn University, Bangkok, Thailand

Mr. Somesh Kumar Dewangan, CSVTU Bhilai (C.G.)/ Dimat Raipur, India

Mr. Hayder N. Jasem, University Putra Malaysia, Malaysia

Mr. A.V.Senthil Kumar, C. M. S. College of Science and Commerce, India

Mr. R. S. Karthik, C. M. S. College of Science and Commerce, India

Mr. P. Vasant, University Technology Petronas, Malaysia

Mr. Wong Kok Seng, Soongsil University, Seoul, South Korea

Mr. Praveen Ranjan Srivastava, BITS PI LAN I, India

Mr. Kong Sang Kelvin, Leong, The Hong Kong Polytechnic University, Hong Kong

Mr. Mohd Nazri Ismail, Universiti Kuala Lumpur, Malaysia

Dr. Rami J. Matarneh, Al-isra Private University, Amman, Jordan

Dr Ojesanmi Olusegun Ayodeji, Ajayi Crowther University, Oyo, Nigeria

Dr. Riktesh Srivastava, Skyline University, UAE

Dr. Oras F. Baker, UCSI University - Kuala Lumpur, Malaysia

Dr. Ahmed S. Ghiduk, Faculty of Science, Beni-Suef University, Egypt

and Department of Computer science, Taif University, Saudi Arabia

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

Vol. 8, No. 6, September 2010

Mr. Tirthankar Gayen, NT Kharagpur, India

Ms. Huei-Ru Tseng, National Chiao Tung University, Taiwan

Prof. Ning Xu, Wuhan University of Technology, China

Mr Mohammed Salem Binwahlan, Hadhramout University of Science and Technology, Yemen

& Universiti Teknologi Malaysia, Malaysia.

Dr. Aruna Ranganath, Bhoj Reddy Engineering College for Women, India

Mr. Hafeezullah Amin, Institute of Information Technology, KUST, Kohat, Pakistan

Prof. Syed S. Rizvi, University of Bridgeport, USA

Mr. Shahbaz Pervez Chattha, University of Engineering and Technology Taxila, Pakistan

Dr. Shishir Kumar, Jaypee University of Information Technology, Wakanaghat (HP), India

Mr. Shahid Mumtaz, Portugal Telecommunication, Instituto de Telecomunicagoes (IT) , Aveiro, Portugal

Mr. Rajesh K Shukla, Corporate Institute of Science & Technology Bhopal M P

Dr. Poonam Garg, Institute of Management Technology, India

Mr. S. Mehta, Inha University, Korea

Mr. Dilip Kumar S.M, University Visvesvaraya College of Engineering (UVCE), Bangalore University,

Bangalore

Prof. Malik Sikander Hayat Khiyal, Fatima J innah Women University, Rawalpindi, Pakistan

Dr. Virendra Gomase , Department of Bioinformatics, Padmashree Dr. D.Y. Patil University

Dr. Irraivan Elamvazuthi, University Technology PETRONAS, Malaysia

Mr. Saqib Saeed, University of Siegen, Germany

Mr. Pavan Kumar Gorakavi, IPMA-USA [YC]

Dr. Ahmed Nabih Zaki Rashed, Menoufia University, Egypt

Prof. Shishir K. Shandilya, Rukmani Devi I nstitute of Science & Technology, I ndia

Mrs.J .Komala Lakshmi, SNR Sons College, Computer Science, I ndia

Mr. Muhammad Sohail, KUST, Pakistan

Dr. Manjaiah D.H, Mangalore University, India

Dr. S Santhosh Baboo, D.G.Vaishnav College, Chennai, India

Prof. Dr. Mokhtar Beldjehem, Sainte-Anne University, Halifax, NS, Canada

Dr. Deepak Laxmi Narasimha, Faculty of Computer Science and Information Technology, University of

Malaya, Malaysia

Prof. Dr. ArunkumarThangavelu, Vellore Institute Of Technology, India

Mr. M. Azath, Anna University, India

Mr. Md. Rabiul Islam, Rajshahi University of Engineering & Technology (RUET), Bangladesh

Mr. Aos Alaa Zaidan Ansaef, Multimedia University, Malaysia

Dr Suresh Jain, Professor (on leave), Institute of Engineering & Technology, Devi Ahilya University, Indore

(MP) India,

Dr. Mohammed M. Kadhum, Universiti Utara Malaysia

Mr. Hanumanthappa. J . University of Mysore, I ndia

Mr. Syed Ishtiaque Ahmed, Bangladesh University of Engineering and Technology (BUET)

MrAkinola Solomon Olalekan, University of I badan, Ibadan, Nigeria

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

Vol. 8, No. 6, September 2010

Mr. Santosh K. Pandey, Department of Information Technology, The Institute of Chartered Accountants of

India

Dr. P. Vasant, Power Control Optimization, Malaysia

Dr. Petr Ivankov, Automatika - S, Russian Federation

Dr. Utkarsh Seetha, Data Infosys Limited, India

Mrs. Priti Maheshwary, Maulana Azad National Institute of Technology, Bhopal

Dr. (Mrs) Padmavathi Ganapathi, Avinashilingam University for Women, Coimbatore

Assist. Prof. A. Neela madheswari, Anna university, India

Prof. Ganesan Ramachandra Rao, PSG College of Arts and Science, India

Mr. Kamanashis Biswas, Daffodil International University, Bangladesh

Dr. Atul Gonsai, Saurashtra University, Gujarat, India

Mr. Angkoon Phinyomark, Prince of Songkla University, Thailand

Mrs. G. Nalini Priya, Anna University, Chennai

Dr. P. Subashini, Avinashilingam University for Women, India

Assoc. Prof. Vijay Kumar Chakka, Dhirubhai Ambani IICT, Gandhinagar , Gujarat

Mrjitendra Agrawal, : Rajiv Gandhi Proudyogiki Vishwavidyalaya, Bhopal

Mr. Vishal Goyal, Department of Computer Science, Punjabi University, India

Dr. R. Baskaran, Department of Computer Science and Engineering, Anna University, Chennai

Assist. Prof, Kanwalvir Singh Dhindsa, B.B.S.B.Engg. College, Fatehgarh Sahib (Punjab), India

Dr. Jamal Ahmad Dargham, School of Engineering and Information Technology, Universiti Malaysia Sabah

Mr. Nitin Bhatia, DAV College, India

Dr. Dhavachelvan Ponnurangam, Pondicherry Central University, India

Dr. Mohd Faizal Abdollah, University of Technical Malaysia, Malaysia

Assist. Prof. Sonal Chawla, Panjab University, India

Dr. Abdul Wahid, AKG Engg. College, Ghaziabad, India

Mr. Arash Habibi Lashkari, University of Malaya (UM), Malaysia

Mr. Md. Rajibul Islam, Ibnu Sina Institute, University Technology Malaysia

Professor Dr. Sabu M. Thampi, .B.S Institute of Technology for Women, Kerala University, India

Mr. Noor Muhammed Nayeem, Universite Lumiere Lyon 2, 69007 Lyon, France

Dr. Himanshu Aggarwal, Department of Computer Engineering, Punjabi University, India

Prof R. Naidoo, Dept of Mathematics/Center for Advanced Computer Modelling, Durban University of

Technology, Durban, South Africa

Prof. Mydhili K Nair, M S Ramaiah Institute of Technology(M. S.R.I. T), Affliliated to Visweswaraiah

Technological University, Bangalore, India

M. Prabu, Adhiyamaan College of Engineering/Anna University, India

Mr. Swakkhar Shatabda, Department of Computer Science and Engineering, United International University,

Bangladesh

Dr. Abdur Rashid Khan, ICIT, Gomal University, Dera Ismail Khan, Pakistan

Mr. H. Abdul Shabeer, l-NautixTechnologies,Chennai, India

Dr. M. Aramudhan, Perunthalaivar Kamarajar Institute of Engineering and Technology, India

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

Vol. 8, No. 6, September 2010

Dr. M. P. Thapliyal, Department of Computer Science, HNB Garhwal University (Central University), I ndia

Prof Ekta Walia Bhullar, Maharishi Markandeshwar University, Mullana (Ambala), India

Dr. Shahaboddin Shamshirband, Islamic Azad University, Iran

Mr. Zeashan Hameed Khan, : Universite de Grenoble, France

Prof. Anil K Ahlawat, Ajay Kumar Garg Engineering College, Ghaziabad, UP Technical University, Lucknow

Mr. Longe Olumide Babatope, University Of Ibadan, Nigeria

Associate Prof. Raman Maini, University College of Engineering, Punjabi University, India

Dr. Maslin Masrom, University Technology Malaysia, Malaysia

Sudipta Chattopadhyay, Jadavpur University, Kolkata, India

Dr. Dang Tuan NGUYEN, University of I nformation Technology, Vietnam National University - Ho Chi Minh

City

Dr. Mary Lourde R., BITS-PI LANI Dubai , UAE

Dr. Abdul Aziz, University of Central Punjab, Pakistan

Mr. Karan Singh, Gautam Budtha University, India

Mr. Avinash Pokhriyal, Uttar Pradesh Technical University, Lucknow, India

Associate Prof Dr Zuraini Ismail, University Technology Malaysia, Malaysia

Assistant Prof. Yasser M. Alginahi, College of Computer Science and Engineering, Taibah University, Madinah

Munawwarrah, KSA

Mr. Dakshina Ranjan Kisku, West Bengal University of Technology, India

Mr. Raman Kumar, Dr B R Ambedkar National I nstitute of Technology, J alandhar, Punjab, I ndia

Associate Prof. Samir B. Patel, Institute of Technology, Nirma University, India

Dr. M.Munir Ahamed Rabbani, B. S. Abdur Rahman University, India

Asst. Prof. Koushik Majumder, West Bengal University of Technology, India

Dr. Alex Pappachen James, Queensland Micro-nanotechnology center, Griffith University, Australia

Assistant Prof. S. Hariharan, B.S. Abdur Rahman University, India

Asst Prof. Jasmine. K. S, R.V. College of Engineering, India

Mr Naushad Ali Mamode Khan, Ministry of Education and Human Resources, Mauritius

Prof. Mahesh Goyani, G H Patel Collge of Engg. &Tech, V.V.N, Anand, Gujarat, India

Dr. Mana Mohammed, University of Tlemcen, Algeria

Prof. J atinder Singh, Universal I nstitutiion of Engg. & Tech. CHD, I ndia

Mrs. M. Anandhavalli Gauthaman, Sikkim Manipal I nstitute of Technology, Majitar, East Sikkim

Dr. Bin Guo, Institute Telecom SudParis, France

Mrs. Maleika Mehr Nigar Mohamed Heenaye-Mamode Khan, University of Mauritius

Prof. Pijush Biswas, RCC Institute of Information Technology, India

Mr. V. Bala Dhandayuthapani, Mekelle University, Ethiopia

Mr. Irfan Syamsuddin, State Polytechnic of Ujung Pandang, Indonesia

Mr. Kavi Kumar Khedo, University of Mauritius, Mauritius

Mr. Ravi Chandiran, Zagro Singapore Pte Ltd. Singapore

Mr. Milindkumar V. Sarode, J awaharlal Darda I nstitute of Engineering and Technology, I ndia

Dr. Shamimul Qamar, KSJ I nstitute of Engineering & Technology, I ndia

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

Vol. 8, No. 6, September 2010

Dr. C. Arun, Anna University, India

Assist. Prof. M.N.Birje, Basaveshwar Engineering College, India

Prof. Hamid Reza Naji, Department of Computer Enigneering, Shahid Beheshti University, Tehran, Iran

Assist. Prof. Debasis Giri, Department of Computer Science and Engineering, Haldia Institute of Technology

Subhabrata Barman, Haldia Institute of Technology, West Bengal

Mr. M. I. Lali, COMSATS Institute of Information Technology, Islamabad, Pakistan

Dr. Feroz Khan, Central Institute of Medicinal and Aromatic Plants, Lucknow, India

Mr. R. Nagendran, Institute of Technology, Coimbatore, Tamilnadu, India

Mr. Amnach Khawne, King Mongkut's Institute of Technology Ladkrabang, Ladkrabang, Bangkok, Thailand

Dr. P. Chakrabarti, Sir Padampat Singhania University, Udaipur, India

Mr. Nafiz Imtiaz Bin Hamid, Islamic University of Technology (I UT), Bangladesh.

Shahab-A. Shamshirband, Islamic Azad University, Chalous, Iran

Prof. B. Priestly Shan, Anna Univeristy, Tamilnadu, India

Venkatramreddy Velma, Dept. of Bioinformatics, University of Mississippi Medical Center, Jackson MS USA

Akshi Kumar, Dept. of Computer Engineering, Delhi Technological University, India

Dr. Umesh Kumar Singh, Vikram University, Ujjain, India

Mr. Serguei A. Mokhov, Concordia University, Canada

Mr. Lai Khin Wee, Universiti Teknologi Malaysia, Malaysia

Dr. Awadhesh Kumar Sharma, Madan Mohan Malviya Engineering College, India

Mr. Syed R. Rizvi, Analytical Services & Materials, Inc., USA

Dr. S. Karthik, SNS Collegeof Technology, India

Mr. Syed Qasim Bukhari, CI MET (Universidad de Granada), Spain

Mr. A.D.Potgantwar, Pune University, India

Dr. Himanshu Aggarwal, Punjabi University, India

Mr. Rajesh Ramachandran, Naipunya Institute of Management and Information Technology, India

Dr. K.L. Shunmuganathan, R.M.K Engg College , Kavaraipettai ,Chennai

Dr. Prasant Kumar Pattnaik, KIST, India.

Dr. Ch. Aswani Kumar, VIT University, India

Mr. Ijaz Ali Shoukat, King Saud University, Riyadh KSA

Mr. Arun Kumar, Sir Padam Pat Singhania University, Udaipur, Rajasthan

Mr. Muhammad Imran Khan, Universiti Teknologi PETRONAS, Malaysia

Dr. Natarajan Meghanathan, Jackson State University, Jackson, MS, USA

Mr. Mohd Zaki Bin Mas'ud, Universiti Teknikal Malaysia Melaka (UTeM), Malaysia

Prof. Dr. R. Geetharamani, Dept. of Computer Science and Eng., Rajalakshmi Engineering College, India

Dr. Smita Rajpal, Institute of Technology and Management, Gurgaon, India

Dr. S. Abdul Khader J ilani, University of Tabuk, Tabuk, Saudi Arabia

Mr. Syedjamal Haider Zaidi, Bahria University, Pakistan

Dr. N. Devarajan, Government College of Technology,Coimbatore, Tamilnadu, INDIA

Mr. R. Jagadeesh Kannan, RMK Engineering College, India

Mr. Deo Prakash, Shri Mata Vaishno Devi University, India

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

Vol. 8, No. 6, September 2010

Mr. Mohammad Abu Naser, Dept. of EEE, IUT, Gazipur, Bangladesh

Assist. Prof. Prasun Ghosal, Bengal Engineering and Science University, India

Mr. Md. Golam Kaosar, School of Engineering and Science, Victoria University, Melbourne City, Australia

Mr. R. Mahammad Shafi, Madanapalle I nstitute of Technology & Science, I ndia

Dr. F.Sagayaraj Francis, Pondicherry Engineering College,! ndia

Dr. Ajay Goel, HI ET , Kaithal, I ndia

Mr. NayakSunil Kashibarao, Bahirji Smarak Mahavidyalaya, India

Mr. SuhasJ Manangi, Microsoft India

Dr. Kalyankar N. V., Yeshwant Mahavidyalaya, Nanded , India

Dr. K.D. Verma, S.V. College of Post graduate studies & Research, India

Dr. Amjad Rehman, University Technology Malaysia, Malaysia

Mr. Rachit Garg, L K College, J alandhar, Punjab

Mr. J . William, M.A.M college of Engineering, Trichy, Tamilnadujndia

Prof. J ue-Sam Chou, Nanhua University, College of Science and Technology, Taiwan

Dr. Thorat S.B., I nstitute of Technology and Management, I ndia

Mr. Ajay Prasad, Sir Padampat Singhania University, Udaipur, India

Dr. Kamaljit I. Lakhtaria, Atmiya Institute of Technology & Science, India

Mr. Syed Rafiul Hussain, Ahsanullah University of Science and Technology, Bangladesh

Mrs Fazeela Tunnisa, Najran University, Kingdom of Saudi Arabia

Mrs Kavita Taneja, Maharishi Markandeshwar University, Haryana, India

Mr. Maniyar Shiraz Ahmed, Najran University, Najran, KSA

Mr. Anand Kumar, AMC Engineering College, Bangalore

Dr. Rakesh Chandra Gangwar, Beant College of Engg. & Tech., Gurdaspur (Punjab) I ndia

Dr. VV Rama Prasad, Sree Vidyanikethan Engineering College, India

Assist. Prof. Neetesh Kumar Gupta, Technocrats Institute of Technology, Bhopal (M.P.), India

Mr. Ashish Seth, Uttar Pradesh Technical University, Lucknow ,UP India

Dr. VV S S S Balaram, Sreenidhi I nstitute of Science and Technology, I ndia

Mr Rahul Bhatia, Lingaya's I nstitute of Management and Technology, I ndia

Prof. Niranjan Reddy. P, KITS , Warangal, India

Prof. Rakesh. Lingappa, Vijetha Institute of Technology, Bangalore, India

Dr. Mohammed Ali Hussain, Nimra College of Engineering & Technology, Vijayawada, A. P., I ndia

Dr. A.Srinivasan, MNM Jain Engineering College, Rajiv Gandhi Salai, Thorapakkam, Chennai

Mr. Rakesh Kumar, M.M. University, Mullana, Ambala, India

Dr. Lena Khaled, Zarqa Private University, Aman, Jordon

Ms. Supriya Kapoor, Patni/Lingaya's Institute of Management and Tech., India

Dr. Tossapon Boongoen , Aberystwyth University, UK

Dr . Bilal Alatas, Firat University, Turkey

Assist. Prof. Jyoti Praaksh Singh , Academy of Technology, India

Dr. Ritu Soni, GNG College, India

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

Vol. 8, No. 6, September 2010

Dr . Mahendra Kumar , Sagar I nstitute of Research & Technology, Bhopal, I ndia.

Dr. Binod Kumar, India

Dr. Muzhir Shaban Al-Ani, Amman Arab University Amman - J ordan

Dr. T.C. Manjunath , ATRIA Institute of Tech, India

Mr. Muhammad Zakarya, COMSATS Institute of Information Technology (CUT), Pakistan

Assist. Prof. Harmunish Taneja, M. M. University, India

Dr. Chitra Dhawale , SICSR, Model Colony, Pune, India

Mrs Sankari Muthukaruppan, Nehru Institute of Engineering and Technology, Anna University, India

Mr. Aaqif Afzaal Abbasi, National University Of Sciences And Technology, Islamabad

Prof. Ashutosh Kumar Dubey, Trinity Institute of Technology and Research Bhopal, India

Mr. G. Appasami, Dr. Pauls Engineering College, India

Mr. M Yasin, National University of Science and Tech, karachi (NUST), Pakistan

Mr. Yaser Miaji, University Utara Malaysia, Malaysia

Mr. Shah Ahsanul Haque, International Islamic University Chittagong (IIUC), Bangladesh

Prof. (Dr) Syed Abdul Sattar, Royal Institute of Technology & Science, India

Dr. S. Sasikumar, Roever Engineering College

Assist. Prof. Monit Kapoor, Maharishi Markandeshwar University, India

Mr. Nwaocha Vivian 0, National Open University of Nigeria

Dr. M. S. Vijaya, GR Govindarajulu School of Applied Computer Technology, India

Assist. Prof. Chakresh Kumar, Manav Rachna International University, India

Mr. Kunal Chadha , R&D Software Engineer, Gemalto, Singapore

Mr. Pawan J indal, J aypee University of Engineering and Technology, I ndia

Mr. Mueen Uddin, Universiti Teknologi Malaysia, UTM , Malaysia

Dr. Dhuha Basheer abdullah, Mosul university, Iraq

Mr. S. Audithan, Annamalai University, India

Prof. Vijay K Chaudhari, Technocrats Institute of Technology , India

Associate Prof. Mohd llyas Khan, Technocrats Institute of Technology , India

Dr. Vu Thanh Nguyen, University of Information Technology, HoChiMinh City, VietNam

Assist. Prof. Anand Sharma, MITS, Lakshmangarh, Sikar, Rajasthan, India

Prof. T V Narayana Rao, HITAM Engineering college, Hyderabad

Mr. Deepak Gour, Sir Padampat Singhania University, India

Assist. Prof. Amutharaj Joyson, Kalasalingam University, India

Mr. Ali Balador, Islamic Azad University, Iran

Mr. Mohitjain, Maharaja Surajmal Institute of Technology, India

Mr. Dilip Kumar Sharma, GLA I nstitute of Technology & Management, I ndia

Dr. Debojyoti Mitra, Sir padampat Singhania University, India

Dr. Ali Dehghantanha, Asia-Pacific University College of Technology and I nnovation, Malaysia

Mr. Zhao Zhang, City University of Hong Kong, China

Prof. S.P. Setty, A.U. College of Engineering, India

Prof. Patel Rakeshkumar Kantilal, Sankalchand Patel College of Engineering, India

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

Vol. 8, No. 6, September 2010

Mr. Biswajit Bhowmik, Bengal College of Engineering & Technology, India

Mr. Manoj Gupta, Apex Institute of Engineering & Technology, India

Assist. Prof. Ajay Sharma, Raj Kumar Goel Institute Of Technology, India

Assist. Prof. Ramveer Singh, Raj Kumar Goel Institute of Technology, India

Dr. Hanan Elazhary, Electronics Research Institute, Egypt

Dr. Hosam I. Faiq, USM, Malaysia

Prof. Dipti D. Patil, MAEER's MIT College of Engg. & Tech, Pune, I ndia

Assist. Prof. Devendra Chack, BCT Kumaon engineering College Dwarahat Almora, India

Prof. Manpreet Singh, M. M. Engg. College, M. M. University, India

Assist. Prof. M. Sadiq ali Khan, University of Karachi, Pakistan

Mr. Prasad S. Halgaonkar, MIT - College of Engineering, Pune, India.

CALL FOR PAPERS
International Journal of Computer Science and Information Security

IJCSIS 2010

ISSN: 1947-5500

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

International Journal Computer Science and Information Security, now at its sixth edition, is the premier
scholarly venue in the areas of computer science and security issues. IJCSIS 2010 will provide a high
profile, leading edge platform for researchers and engineers alike to publish state-of-the-art research in the
respective fields of information technology and communication security. The journal will feature a diverse
mixture of publication articles including core and applied computer science related topics.

Authors are solicited to contribute to the special issue by submitting articles that illustrate research results,
projects, surveying works and industrial experiences that describe significant advances in the following
areas, but are not limited to. Submissions may span a broad range of topics, e.g.:

Track A: Security

Access control, Anonymity, Audit and audit reduction & Authentication and authorization, Applied
cryptography, Cryptanalysis, Digital Signatures, Biometric security, Boundary control devices,
Certification and accreditation, Cross-layer design for security, Security & Network Management, Data and
system integrity, Database security, Defensive information warfare, Denial of service protection, Intrusion
Detection, Anti-malware, Distributed systems security, Electronic commerce, E-mail security, Spam,
Phishing, E-mail fraud, Virus, worms, Trojan Protection, Grid security, Information hiding and
watermarking & Information survivability, Insider threat protection, Integrity

Intellectual property protection, Internet/Intranet Security, Key management and key recovery, Language-
based security, Mobile and wireless security, Mobile, Ad Hoc and Sensor Network Security, Monitoring
and surveillance, Multimedia security , Operating system security, Peer-to-peer security, Performance
Evaluations of Protocols & Security Application, Privacy and data protection, Product evaluation criteria
and compliance, Risk evaluation and security certification, Risk/vulnerability assessment, Security &
Network Management, Security Models & protocols, Security threats & countermeasures (DDoS, MiM,
Session Hijacking, Replay attack etc,), Trusted computing, Ubiquitous Computing Security, Virtualization
security, VoIP security, Web 2.0 security, Submission Procedures, Active Defense Systems, Adaptive
Defense Systems, Benchmark, Analysis and Evaluation of Security Systems, Distributed Access Control
and Trust Management, Distributed Attack Systems and Mechanisms, Distributed Intrusion
Detection/Prevention Systems, Denial-of- Service Attacks and Countermeasures, High Performance
Security Systems, Identity Management and Authentication, Implementation, Deployment and
Management of Security Systems, Intelligent Defense Systems, Internet and Network Forensics, Large-
scale Attacks and Defense, RFID Security and Privacy, Security Architectures in Distributed Network
Systems, Security for Critical Infrastructures, Security for P2P systems and Grid Systems, Security in E-
Commerce, Security and Privacy in Wireless Networks, Secure Mobile Agents and Mobile Code, Security
Protocols, Security Simulation and Tools, Security Theory and Tools, Standards and Assurance Methods,
Trusted Computing, Viruses, Worms, and Other Malicious Code, World Wide Web Security, Novel and
emerging secure architecture, Study of attack strategies, attack modeling, Case studies and analysis of
actual attacks, Continuity of Operations during an attack, Key management, Trust management, Intrusion
detection techniques, Intrusion response, alarm management, and correlation analysis, Study of tradeoffs
between security and system performance, Intrusion tolerance systems, Secure protocols, Security in
wireless networks (e.g. mesh networks, sensor networks, etc.), Cryptography and Secure Communications,
Computer Forensics, Recovery and Healing, Security Visualization, Formal Methods in Security, Principles
for Designing a Secure Computing System, Autonomic Security, Internet Security, Security in Health Care
Systems, Security Solutions Using Reconfigurable Computing, Adaptive and Intelligent Defense Systems,
Authentication and Access control, Denial of service attacks and countermeasures, Identity, Route and

Location Anonymity schemes, Intrusion detection and prevention techniques, Cryptography, encryption
algorithms and Key management schemes, Secure routing schemes, Secure neighbor discovery and
localization, Trust establishment and maintenance, Confidentiality and data integrity, Security architectures,
deployments and solutions, Emerging threats to cloud-based services, Security model for new services,
Cloud-aware web service security, Information hiding in Cloud Computing, Securing distributed data
storage in cloud, Security, privacy and trust in mobile computing systems and applications, Middleware
security & Security features: middleware software is an asset on

its own and has to be protected, interaction between security-specific and other middleware features, e.g.,
context-awareness, Middleware-level security monitoring and measurement: metrics and mechanisms
for quantification and evaluation of security enforced by the middleware, Security co-design: trade-off and
co-design between application-based and middleware-based security, Policy-based management:
innovative support for policy-based definition and enforcement of security concerns, Identification and
authentication mechanisms: Means to capture application specific constraints in defining and enforcing
access control rules, Middleware-oriented security patterns: identification of patterns for sound, reusable
security, Security in aspect-based middleware: mechanisms for isolating and enforcing security aspects,
Security in agent-based platforms: protection for mobile code and platforms, Smart Devices: Biometrics,
National ID cards, Embedded Systems Security and TPMs, RFID Systems Security, Smart Card Security,
Pervasive Systems: Digital Rights Management (DRM) in pervasive environments, Intrusion Detection and
Information Filtering, Localization Systems Security (Tracking of People and Goods), Mobile Commerce
Security, Privacy Enhancing Technologies, Security Protocols (for Identification and Authentication,
Confidentiality and Privacy, and Integrity), Ubiquitous Networks: Ad Hoc Networks Security, Delay-
Tolerant Network Security, Domestic Network Security, Peer-to-Peer Networks Security, Security Issues
in Mobile and Ubiquitous Networks, Security of GSM/GPRS/UMTS Systems, Sensor Networks Security,
Vehicular Network Security, Wireless Communication Security: Bluetooth, NFC, WiFi, WiMAX,
WiMedia, others

This Track will emphasize the design, implementation, management and applications of computer
communications, networks and services. Topics of mostly theoretical nature are also welcome, provided
there is clear practical potential in applying the results of such work.

Track B: Computer Science

Broadband wireless technologies: LTE, WiMAX, WiRAN, HSDPA, HSUPA, Resource allocation and
interference management, Quality of service and scheduling methods, Capacity planning and dimensioning,
Cross-layer design and Physical layer based issue, Interworking architecture and interoperability, Relay
assisted and cooperative communications, Location and provisioning and mobility management, Call
admission and flow/congestion control, Performance optimization, Channel capacity modeling and analysis,
Middleware Issues: Event-based, publish/subscribe, and message-oriented middleware, Reconfigurable,
adaptable, and reflective middleware approaches, Middleware solutions for reliability, fault tolerance, and
quality-of-service, Scalability of middleware, Context-aware middleware, Autonomic and self-managing
middleware, Evaluation techniques for middleware solutions, Formal methods and tools for designing,
verifying, and evaluating, middleware, Software engineering techniques for middleware, Service oriented
middleware, Agent-based middleware, Security middleware, Network Applications: Network-based
automation, Cloud applications, Ubiquitous and pervasive applications, Collaborative applications, RFID
and sensor network applications, Mobile applications, Smart home applications, Infrastructure monitoring
and control applications, Remote health monitoring, GPS and location-based applications, Networked
vehicles applications, Alert applications, Embeded Computer System, Advanced Control Systems, and
Intelligent Control : Advanced control and measurement, computer and microprocessor-based control,
signal processing, estimation and identification techniques, application specific IC's, nonlinear and
adaptive control, optimal and robot control, intelligent control, evolutionary computing, and intelligent
systems, instrumentation subject to critical conditions, automotive, marine and aero-space control and all
other control applications, Intelligent Control System, Wiring/Wireless Sensor, Signal Control System.
Sensors, Actuators and Systems Integration : Intelligent sensors and actuators, multisensor fusion, sensor
array and multi-channel processing, micro/nano technology, microsensors and microactuators,
instrumentation electronics, MEMS and system integration, wireless sensor, Network Sensor, Hybrid

Sensor, Distributed Sensor Networks. Signal and Image Processing : Digital signal processing theory,
methods, DSP implementation, speech processing, image and multidimensional signal processing, Image
analysis and processing, Image and Multimedia applications, Real-time multimedia signal processing,
Computer vision, Emerging signal processing areas, Remote Sensing, Signal processing in education.
Industrial Informatics: Industrial applications of neural networks, fuzzy algorithms, Neuro-Fuzzy
application, biolnformatics, real-time computer control, real-time information systems, human-machine
interfaces, CAD/CAM/CAT/CIM, virtual reality, industrial communications, flexible manufacturing
systems, industrial automated process, Data Storage Management, Harddisk control, Supply Chain
Management, Logistics applications, Power plant automation, Drives automation. Information Technology,
Management of Information System : Management information systems, Information Management,
Nursing information management, Information System, Information Technology and their application, Data
retrieval, Data Base Management, Decision analysis methods, Information processing, Operations research,
E-Business, E-Commerce, E-Government, Computer Business, Security and risk management, Medical
imaging, Biotechnology, Bio-Medicine, Computer-based information systems in health care, Changing
Access to Patient Information, Healthcare Management Information Technology.
Communication/Computer Network, Transportation Application : On-board diagnostics, Active safety
systems, Communication systems, Wireless technology, Communication application, Navigation and
Guidance, Vision-based applications, Speech interface, Sensor fusion, Networking theory and technologies,
Transportation information, Autonomous vehicle, Vehicle application of affective computing, Advance
Computing technology and their application : Broadband and intelligent networks, Data Mining, Data
fusion, Computational intelligence, Information and data security, Information indexing and retrieval,
Information processing, Information systems and applications, Internet applications and performances,
Knowledge based systems, Knowledge management, Software Engineering, Decision making, Mobile
networks and services, Network management and services, Neural Network, Fuzzy logics, Neuro-Fuzzy,
Expert approaches, Innovation Technology and Management : Innovation and product development,
Emerging advances in business and its applications, Creativity in Internet management and retailing, B2B
and B2C management, Electronic transceiver device for Retail Marketing Industries, Facilities planning
and management, Innovative pervasive computing applications, Programming paradigms for pervasive
systems, Software evolution and maintenance in pervasive systems, Middleware services and agent
technologies, Adaptive, autonomic and context-aware computing, Mobile/Wireless computing systems and
services in pervasive computing, Energy-efficient and green pervasive computing, Communication
architectures for pervasive computing, Ad hoc networks for pervasive communications, Pervasive
opportunistic communications and applications, Enabling technologies for pervasive systems (e.g., wireless
BAN, PAN), Positioning and tracking technologies, Sensors and RFID in pervasive systems, Multimodal
sensing and context for pervasive applications, Pervasive sensing, perception and semantic interpretation,
Smart devices and intelligent environments, Trust, security and privacy issues in pervasive systems, User
interfaces and interaction models, Virtual immersive communications, Wearable computers, Standards and
interfaces for pervasive computing environments, Social and economic models for pervasive systems,
Active and Programmable Networks, Ad Hoc & Sensor Network, Congestion and/or Flow Control, Content
Distribution, Grid Networking, High-speed Network Architectures, Internet Services and Applications,
Optical Networks, Mobile and Wireless Networks, Network Modeling and Simulation, Multicast,
Multimedia Communications, Network Control and Management, Network Protocols, Network
Performance, Network Measurement, Peer to Peer and Overlay Networks, Quality of Service and Quality
of Experience, Ubiquitous Networks, Crosscutting Themes - Internet Technologies, Infrastructure,
Services and Applications; Open Source Tools, Open Models and Architectures; Security, Privacy and
Trust; Navigation Systems, Location Based Services; Social Networks and Online Communities; ICT
Convergence, Digital Economy and Digital Divide, Neural Networks, Pattern Recognition, Computer
Vision, Advanced Computing Architectures and New Programming Models, Visualization and Virtual
Reality as Applied to Computational Science, Computer Architecture and Embedded Systems, Technology
in Education, Theoretical Computer Science, Computing Ethics, Computing Practices & Applications

Authors are invited to submit papers through e-mail ij csiseditorfo) gmail. com . Submissions must be original
and should not have been published previously or be under consideration for publication while being
evaluated by IJCSIS. Before submission authors should carefully read over the journal's Author Guidelines,
which are located at http://sites.google.com/site/ijcsis/authors-notes .

Internet Archive Audio

Featured

Top

Images

Featured

Top

Software

Featured

Top

Books

Featured

Top

Video

Featured

Top

Mobile Apps

Browser Extensions

Archive-It Subscription

Save Page Now

Full text of "Journal of Computer Science IJCSIS September 2010"

See other formats