Full text of "Journal of Computer Science and Information Security June 2010"

See other formats

UCSIS Vol. 8 No. 3, June 2010
ISSN 1947-5500

International Journal of
Computer Science
& Information Security

Editorial
Message from Managing Editor

The International Journal of Computer Science and Information Security is an
English language periodical on research in general computer science and information
security which offers prompt publication of important technical research work,
whether theoretical, applicable, or related to implementation.

Target Audience: IT academics and business people concerned with computer science
and security; university IT faculties; industry IT departments; government departments;
the financial industry; the mobile industry and the computing industry.

Coverage includes: security infrastructures, network security: Internet security,
content protection, cryptography, steganography and formal methods in information
security; multimedia, image processing, software, information systems, intelligent
systems, web services, wireless communication, networking and technologies.

Thanks to authors who contributed papers to the June 2010 issue and the reviewers,
for providing valuable feedback comments. IJCSIS June 2010 Issue (Vol. 8, No. 3) has
an acceptance rate of 30 %.

Available at http:/ / sites-qooqie-com/ site/ iicsis/

IJCSIS Vol. 8, No. 3June2010 Edition
ISSN 1947-5500 © IJCSIS 2010, USA.

Abstracts Indexed by (among others):

i^ocKJle scholar = n]if^^^f:,f^ DUIlUa

b&rLUincn tm*^ r^r Ecienc*

IJCSIS EDITORIAL BOARD

Dr. Gregorio Martinez Perez

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

Dr. M. Emre Celebi,

Assistant Professor, Departnnent of Connputer 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

Departnnent of Connputer Enigneering, Shahid Beheshti University, Tehran, Iran

Dr. Sanjayjasola

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

Dr Rilctesh Srivastava

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

Dr. Siddhivinayalc Kullcarni

University of Ballarat, Ballarat, Victoria, Australia

Professor (Dr) JNIoichtar Beldjehem

Sainte-Anne University, Halifax, NS, Canada

Dr. Alex Pappachen J ames, (Research Fellow)

Queensland Micro-nanotechnology center, Griffith University, Australia

TABLE OF CONTENTS

1. Paper 27051043: Implementation of SPIN Model Checker for Formal Verification of Distance
Vector Routing Protocol (pp. 1-6)

Kashif Javed, Department of Information Technologies, Abo Akademi University, Joukahaisenkatu, Turku,

FIN-20520, Finland

Asifa Kashif, Department of Electrical Engineering, National University- Foundation for Advancement of

Science and Technology, A.K. Brohi Road, H-11/4, Islamabad, 44000, Pakistan

Elena Troubitsyna, Department of Information Technologies, Abo Akademi University, Turku, FIN-20520,

Finland

2. Paper 31051062: Integrated Queuing based Energy-Aware Computing in MANET (pp. 7-10)

Dr. P. K. Suri, Dean and Professor, Faculty of Science, Deptt. ofComp. Sci. &Appl. Kurukshetra
University, Kurukshetra, Haryana, India

Kavita Taneja, Assistant Professor, M. M. Inst, of Computer Tech. & Business Mgmt. Maharishi
Markandeshwar University, Mullana, Haryana, India

3. Paper 30051057: A Review of Negotiation Agents in e-commerce (pp. 11-20)

Sahar Ebadi, Department of Information System, Faculty of Computer Science and Information

Technology, University Putra Malaysia, Serdang, Malaysia

Md. Nasir Sulaiman, Department of Information System, Faculty of Computer Science and Information

Technology, University Putra Malaysia, Serdang, Malaysia

Masrah Azrifah Azmi Murad, Department of Information System, Faculty of Computer Science and

Information Technology, University Putra Malaysia, Serdang, Malaysia

4. Paper 31051071: Customized Digital Road Map Building using Floating Car GPS Data (pp. 21-29)

G. Rajendran, Assistant Professor of Computer Science, Thiruvalluvar Government Arts College,

Rasipuram-637401 , Tamilnadu, India

Dr. M. Arthanari, Director, Bharathidasan School of Computer Applications, Ellispettai-6381 16,

Tamilnadu, India

M. Sivakumar, Doctoral Research Scholar, Anna University, Coimbatore, Tamilnadu, India

5. Paper 31051077: Robust stability check of fractional control law applied to a LEO (Low Earth
Orbit) Satellite (pp. 30-36)

Ouadid EL Figuigui, Noureddine Elalami,

Laboratoire dAutomatique et Informatique Industrielle EMI, Morocco

6. Paper 30051054: Performance Evaluation of Genetic Algorithm For Solving Routing Problem In
Communication Networks (pp. 37-43)

Ehab Rushdy Mohamed, Faculty of Computer and Informatics, Zagazig University, Zagazig, Egypt,

Mahmoud Ibrahim Abdalla, Faculty of Engineering, Zagazig University, Zagazig, Egypt,

Ibrahim Elsayed Zidan, Faculty of Engineering, Zagazig University, Zagazig, Egypt,

Ibrahim Mahmoud El-Henawy, Faculty of Computer and Informatics, Zagazig University, Zagazig, Egypt

7. Paper 25051033: Testing Equivalence of Regular Expressions (pp. 44-46)

Keehang Kwon, Department of Computer Engineering, Dong-A University, Busan, Republic of Korea
Hong Pyo Ha, Department of Computer Engineering, Dong-A University, Busan,
Republic of Korea

8. Paper 31051079: CRS, a Novel Ensemble Construction Methodology (pp. 47-51)

Navid Kardan, Computer Engineering Dep. lUST, Tehran, Iran
Morteza Analoui, Computer Engineering Dep., lUST, Tehran, Iran

9. Paper 31051072: Routing Optimization Technique Using M/M/1 Queuing Model & Genetic
Algorithm (pp. 52-58)

Madiha Sarfraz, M. Younus Javed, Muhammad Almas Anjum, Shaleeza Sohail

Department of Computer Engineering, College of Electrical & Mechanical Engineering, Pakistan

10. Paper 31051083: Architectural Description of an Automated System for Uncertainty Issues
Management in Information Security (pp. 59-67)

Haider Abbas, Department of Electronic Systems, Royal Institute of Technology, Sweden
Christer Magnus son. Department of Computer and System Sciences, Stockholm University, Sweden
Louise Yngstrom, Department of Computer and System Sciences, Stockholm University, Sweden
Ahmed Hemani, Department of Electronic Systems, Royal Institute of Technology, Sweden

11. Paper 14041018: Driving Architectural Design through Business Goals (pp. 68-71)

Lena Khaled, Software Engineering Department, Zarqa Private University, Amman, Jordon

12. Paper 11051005: Distributed Information Sharing Cooperation In Dynamic Channel Allocation
Scheme (pp. 75-82)

Mr. P. Jesu Jayarin, Sathyabama University, Chennai-119, India.
Dr. T. Ravi, KCG college of Technology, Chennai-97, India.

13. Paper 15051008: Key Generation For AES Using Bio-Metic Finger Print For Network Data
Security (pp. 83-88)

Dr. R. Seshadri, Director, University Computer Center, Sri Venkateswara University, Tirupati,

T. Raghu Trivedi, Research Scholar, Department of Computer Science, Sri Venkateswara University,

Tirupati.

14. Paper 18051018: Classification of Five Mental Tasks Based on Two Methods of Neural Network
(pp. 89-95)

Vijay Khare, Jaypee Institute of Information Technology, Dept. of Electronics and Communication,
Engineering, Nioda, India.

Jayashree Santhosh, Indian Institute of Technology, Computer Services Centre, Delhi, India.
SnehAnand, Indian Institute of Technology, Centre for Biomedical Engineering Centre, Delhi, India.
Manvir Bhatia, Sir Ganga Ram Hospital, Department of Sleep Medicine, New Delhi, India

15. Paper 25051047: Sixth order Butterworth Characteristics using LV MOCCII and Grounded
Components (pp. 96-100)

T. Parveen, Electronics Engineering Department, Z H. College of Engineering & Technology, AMU,
Aligarh, India

16. Paper 27051042: A Lightweight Secure Trust-based Localization Scheme for Wireless Sensor
Networks (pp. 101-107)

P. Pandarinath, Associate Professor, CSE, Sir C R, Reddy College of Engineering, Eluru-534001, Andhra

Pradesh

M. Shashi, Head of the Department, Dept. OfCS&SE, Andhra University, Visakhapatnam- 530 003,

Andhra Pradesh

Allam Appa Rao, Vice Chancellor, JNTU Kakinada, Kakinada, Andhra Pradesh

17. Paper 30051051: Mechanism to Prevent Disadvantageous Child Node Attachment in HiLOW (pp.
108-113)

Lingeswari V.Chandra, Kok-Soon Chai and Sureswaran Ramadass, National Advanced IPv6 Centre,

Universiti Sains Malaysia

Gopinath Rao Sinniah, MIMOS Berhad, 57000 Kuala Lumpur

18. Paper 30051052: Rough Entropy as Global Criterion for Multiple DNA Sequence Alignment (pp.
114-121)

Sara El-Sayed El-Metwally, Demonstrator, Computer Science Departement, Faculty of Computer and

information Science, Mansoura University, Egypt.

Dr. ElSayed Foad Radwan, Lecturer, Computer Science Departement, Faculty of Computer and

information Science, Mansoura University, Egypt.

Ass. Prof. Taker Tawfek Hamza, Vice Dean for Graduate Studies and Research, Assistant Professor,

Computer Science Departement, Faculty of Computer and information Science, Mansoura University,

Egypt.

19. Paper 30051053: Weighted Attribute Fusion Model for Face Recognition (pp. 122-128)

S. Sakthivel, Assistant Professor, Department of Information Technology, Sona college of Technology,
Salem, India

Dr. R. Lakshmipathi, Professor, Department of Electrical and Electronic Engineering, St. Peter's
Engineering College, Chennai, India

20. Paper 30051055: A DNA and Amino Acids-Based Implementation of Playfair Cipher (pp. 129-136)

Mona Sabry, Mohamed Hashem, Taymoor Nazmy, Mohamed Essam Khalifa

Computer Science department. Faculty of Computer Science and information systems, Ain Shams

University, Cairo, Egypt.

21. Paper 30051061: Ultra Wideband Slot Antenna with Reconfigurable Notch bands (pp. 137-142)

/. William and R. Nakkeeran, Department of Electronics and Communication Engineering
Pondicherry Engineering College Puducherry, India . 605014.

22. Paper 31051073: UWB Slot Antenna with Rejection of IEEE 802.11a Band (pp. 143-148)

/. William and R. Nakkeeran

Department of Electronics and Communication Engineering, Pondicherry Engineering College,

Puducherry, India . 605014.

23. Paper 31051086: A Study Of Various Load Balancing Techniques In Internet (pp. 149-156)

M. Azath, Research Scholar, Anna University, Coimbatore.

Dr. R.S.D. Wahida banu. Research Supervisor, Anna University, Coimbatore.

24. Paper 30051058: Laboratory Study of Leakage Current and Measurement of ESDD of
Equivalent Insulator Flat Model under Various Polluted Conditions (pp. 157-161)

A^. Narmadhai, Senior Lecturer, Dept of FEE Government College of Technology Coimbatore, India

S. Suresh, PG Scholar, Dept of FEE, Government College of Technology, Coimbatore, India

Dr. A. Ebenezer Jeyakumar, Director (Academics), SNR Sons Charitable Trust, SREC Coimbatore, India

25. Paper 31051076: SSL/TLS Web Server Load Optimization using Adaptive SSL with Session
Handling Mechanism (pp. 162-167)

R. K. Pateriya, J. L. Rana, S. C Shrivastava

Department of Computer Science & Engineering and Information Technology, Maulana Azad National

Institute of Technology, Bhopal, India

26. Paper 15051011: An Enhancement On Mobile TCP Socket (pp. 168-171)

S. Saravanan, Research Scholar, Sathyabama University, Chennai- 119, India.

Dr. T. Ravi, Prof & Head, Dept of CSE ,KCG College of Technology , Chennai, India

27. Paper 15051017: Modern Computer Graphics Technologies Used at Educational Programs and
Some Graphical output screens (pp. 172-174)

A^. Suresh Kumar, S. Amarnadh, K. Srikanth, Ch. Hey ma Raju,

GIT, GIT AM University, Visakhapatnam

D.V. Rama Koti Reddy, College of Engineering, Andhra University, Visakhapatnam

R. Ajay Suresh Babu, Raghu Engineering College, Visakhapatnam

K. Naga Soujanya, GIS, GITAM University, Visakhapatnam

28. Paper 20051023: Impact of language morphologies on Search Engines Performance for Hindi and
English language (pp. 175-178)

Dr. S.K Dwivedi, Reader and Head, Computer Science Dept., BBAU, Lucknow, India.
Rajesh Kr. Gautam, Research Scholar, Computer Science Dept., BBAU, Lucknow, India.
Parul Rastogi, Research Scholar, Computer Science Dept., BBAU, Lucknow, India.

29. Paper 20051029: Comparison of Traffic in Manhattan Street Network in NS2 (pp. 179-182)

Ravinder Bahl, Rakesh Kumar, Department of Information and Technology, MMEC, Muallana, Ambala,
Haryana, India

Rakesh Sambyal, Information and Technology, MBS College of Engineering and Technology, Babliana,
Jammu, Jammu and Kashmir, India

30. Paper 25051036: An Evolving Order Regularized Affine Projection Algorithm, suitable for Echo
Cancellation (pp. 183-187)

Shifali Srivastava, Electronics Deptt., JUT, Noida, India
M.C. Srivastava, Electronics Deptt., JUT, Noida, India

31. Paper 30051060: Design and Implementation of Flexible Framework for Secure Wireless Sensor
Network Applications (pp. 188-194)

Inakota Trilok, Department of Computer Science & Engineering, National Institute of Technology
Warangal, India

Mahesh U. Patil, National Ubiquitous Computing Research Centre, Centre for Development of Advanced
Computing, Hyderabad, India

32. Paper 31051070: Optimizing the Application-Layer DDoS Attacks for Networks (pp. 195-200)

P. Niranjan Reddy, K. Praveen Kumar, M. Preethi,
KITS, Warangal, A. P. , India

33. Paper 08051002: Survey - New Routing Technique for Grid Computing (pp. 201-206)

R. Rameshkumar, Research Schlolar, J.N.T. University, Kukatpally, Hyderabad.

Dr. A. Damodaram , Director/ U.GC Academic Staff College, J.N.T. University, Kukatpally, Hyderabad.

34. Paper 27051037: A Forager Bees Behaviour Inspired approach to predict the forthcoming
navigation pattern of online users (pp. 207-215)

V. Mohanraj, Assistant Profess or/IT, Sona College of Technology, Salem, Tamilnadu, India

Dr. R. Lakshmipathi, Professor/EEE, St. Peters Engineering College (Deemed University), Chennai,

Tamilnadu, India

J Senthilkumar, Assistant Professor/IT, Sona College of Technology, Salem, Tamilnadu, India

Y. Suresh, Assistant Professor/IT, Sona College of Technology, Salem, Tamilnadu, India

35. Paper 27051038: Quality of Service Issues in Wireless Ad Hoc Network (IEEE 802.11b) (pp. 216-
221)

Mohammed Ali Hus sain. Research Scholar, Dept. ofCSE, Acharya Nagarjuna University, Guntur, A.P.,

India.

Mohammed Mastan, Research Scholar, Dept. ofCSE, JNT University, Kakinada, A. P., India.

Syed Umar, Research Scholar, Dept. ofCSE, Dravidian University, Kuppam, A.P., India.

36. Paper 27051049: Collaborative Web Recommendation Systems based on Association Rule Mining
(pp. 222-227)

A. Kumar, Research Scholar, Department of Computer Science & Engineering, Sathyabama University,
Chennai, India.

Dr. P. Thambidurai, Department of Computer Science & Engineering, Pondicherry Engineering College,
Puducherry, India.

37. Paper 31051065: Similarity Based Imputation Method For Time Variant Data (pp. 11%-l^T)

Dr. E. Sagayaraj Erancis, Saranya Kumari Potluri, Vinolin Deborah Delphin, Vishnupriya. B
Pondicherry Engineering College, Pondicherry, India

38. Paper 31051075: Efficient Node Search in P2P Using Distributed Spanning Tree (pp. 233-240)

P. Victer Paul, T. Vengattaraman, M. S. Saleem Basha, P. Dhavachelvan

Department of Computer Science, Pondicherry University, Puducherry, India.

R. Baskaran, Department of Computer Science and Engineering, Anna University, Chennai, India.

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

Vol. 8, No. 3, 2010

Implementation of SPIN Model Checker for Formal
Verification of Distance Vector Routing Protocol

Kashif Javed

Department of Information Technologies

Abo Akademi University

Turku, FIN-20520, Finland

Kashif.javed@abo.fi

Asifa Kashif

Department of Electrical Engineering

National University of Computer and

Emerging Sciences, Islamabad, Pakistan

asifa.ilyas 85 @ gmail.com

Elena Troubitsyna

Department of Information Technologies

Abo Akademi University

Turku, FIN-20520, Finland

Elena.Troubitsyna@ abo.fi

Abstract - Distributed systems and computing requires routing
protocols to meet a wide variety of requirements of a large
number of users in heterogeneous networks. DVR is one of many
other employed protocols for establishing communication using
routes with minimum cost to different destinations from a given
source. Research work presented in this paper focuses on
implementation of DVR in SPIN and provides formal verification
of correctness of DVR behaviour covering all required aspects.
Simulation results clearly show a proof of the established paths
from each router to different destinations in a network consisting
of six routers and a number of links.

Keywords: Formal Verification, DVR Protocol, SPIN Model
Checker, Distance Vector Routing, Implementation in PROMELA

Introduction

A computer network consists of a number of routers which
have the capability to communicate with each other. Routing
Information Protocol (RIP) is widely used for routing packets
from a source to its destination in computer networks. RIP
requires information about distance and direction from source
to destination. Each router, in the Distance Vector Routing
(DVR) methodology, keeps updated record of distances and
hops of its neighbours. Various techniques are used to gather
useful routing table information for each router. In one
approach, special packets are sent by each router and are
received back after having time- stamped by the receivers.
Chromosomes have been employed in the Genetic Algorithm
[1] to select the most optimal path by utilizing its fitness
function, selection of next generation and crossover operation
for updating the routing tables in an efficient manner. Thus, all
routers keep refreshing their routing tables and maintain latest
information about other neighbouring routers in order to
provide optimized performance in the available network [1-3].

Mahlknecht, Madni and Roetzer [4] has presented an
efficient protocol that uses hop count and cost information in
its Energy Aware Distance Vector (EADV) routing scheme
and makes use of shot-multi-hop routing for consuming lesser
energy in the wireless sensor networks. EADV can do well for
long lasting battery-powered sensor nodes while using the
lowest cost path towards the selected sink node. An algorithm
is considered the most effective if it contains the correct and
latest information about its neighbours in its DVR table. An

effort has been made by Liwen He by devising a computational
method to protect a network from internal attacks (such as mis-
configuration and compromise) through the use of verifying
routing messages in the DVR protocols [5]. Formal verification
of standards for DVR protocols has also been comprehensively
presented by Bhargavan, Gunter and Obradovic [6] using three
case studies. The researchers have used HOL ( an interactive
theorem prover and SPIN (model checker) to verify and prove
salient properties of DVR protocols. HOL and SPIN have been
employed by these researchers for providing a proof of
convergence for the RIP [7 ].

The remaining paper is organized as follows. DVR protocol
is presented in Section II and Section III describes the use of
SPIN tool and PROMELA language for formal verification.
System design and implementation has been discussed in
Section IV covering network topology, implementation details
and operation of DVR protocol. Formal verification of
simulation results has been illustrated in Section V and finally
conclusions and future work is given in Section VII.

IL Distance Vector Routing Protocol

A. General Methodology

A routing table is required to be maintained for each router
in the network for the purpose of working of a DVR scheme.
Routing table information is used to determine the best path
(i.e. having minimum cost in terms of distance or hops) from a
source to destination. Links are needed to connect concerned
routers for establishing communication. An optimal DVR
protocol has to exchange frequent messages in order to update
the routing table of each router. So, exchanging information
among neighbours is carried out on regular intervals.

Routing table of every router keeps necessary information
(i.e. id of neighbouring routers, most suitable outgoing link to
be used for the destination, distance, hops (number of routers
on the route), time delay, number of queued messages on the
link). The process of making forwarding decision for selecting
the best optimal path from source to destination is based on a
combination of these parameters. The objective of routers is to
send packets to hosts connected to the networks for
heterogeneous requirements of a large number of users. In this
way, efficient DVR schemes ultimately establish good global

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

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

Vol. 8, No. 3, 2010

paths by connecting hosts in a distributed environment
covering very long distances. Those routers are taken as
neighbours which have links/interfaces to a common network.

B. Routing Information Protocol

RIP [8,9] is a widely used protocol for finding the optimal
path to the destination in a network. Each router has a routing
table and all routers periodically updated their routing tables by
using advertising approach. All routes of a router are advertised
through the mechanism of broadcasting RIP packets to all the
neighbouring routers in the network. Every router checks the
advertised information of neighbouring nodes and changes
information only in its routing table if the new route to the
same destination further improves the existing route length. In
other words, the updated routing table information now takes to
the best available route so far for the relevant destination.

The number of hops in the RIP are kept low (up to 15) for
the route length for faster convergence [6,7]. RIP methodology,
however, prevents formation of loops between pairs of routers
in order to minimize convergence time as well as permitted
route length. Timer expiry record is also maintained in every
routing table and is normally set to 180 seconds whenever a
routing table is updated. As routers advertise after every 30
seconds, the destination is considered unreachable if a router is
not refreshed for 180 seconds. It further waits for another 120
seconds. If the router remains un-refreshed during this time as
well, then its route is removed from the routing tables of the
concerned routers. This requirement is incorporated to cater for
broken links, faulty networks and congestions.

III. USE OF SPIN AND PROMELA

Formal Verification

A number of new systems and methodologies are being
devised by the researchers in different areas of science,
technology and engineering as a result of meaningful R&D
work being undertaken by academic and research institutes all
over the world. Every proposed system requires a proof of its
correctness by gathering results using simulation and testing
techniques. Formal verification terminology [10,11] is in fact a
process of actual demonstration of the system in order to check
its correctness under the defined boundaries and valid
conditions of used parameters/variables.

Precision and accuracy of the system is verified by running
the programming modules by employing required algorithms in
the model checking approach. Errors occurred (if any) are
properly identified under varying conditions so that such errors
can be easily located by the users and are later on
repaired/tackled by adjusting specifications of the model.
Afterwards, the model description is fine tuned to achieve
required model specifications for verification of correct results
of the system.

B. SPIN Tool and PROMELA High Level Language

SPIN [12,13] is a open-source software tool and is widely
used for the formal verification of software systems working in

the distributed environment. Inspiring applications of SPIN
include the verification of the control algorithms for various
applications, logic verification of the call processing software
for a commercial data communication, critical algorithms for
space missions, operating systems, switching systems,
distributed & parallel systems and formal verification of
various routing protocols. This tool also supports interactive,
random and guided simulations for a wide variety of
applications. Spin can be used in four main modes (i.e. as a
simulator, as an exhaustive verifier, as a proof approximation
system and as a driver for swarm verification).

Spin provides efficient software verification and supports
the PROMELA (PROcess MEta LAnguage) high level
language to specify systems descriptions [14]. It is a SPIN'S
input language which is used to build detailed PROMELA
models for complete verification of system designs. It provides
a way for making abstractions of distributed systems. Different
assumptions are used in SPIN to verify each model. After
checking correctness of a model with SPIN, it can then be used
to build and verify subsequent models of the system so that the
fully developed system produces the required behavior.
PROMELA programs consist of processes, message channels,
and variables.

IV. System Design and Implementation

A. Network Topology

The network topology shown in Figure 1 has been used for
implementation of DVR protocol. There are six routers (A,
B, C, D, E & F) and seven links (edges). Each link
connects two routers. Weight values range from 2 to 23 for
different links and these values indicate distances between
routers. Integer values have been used and distance units
can be chosen during actual implementation of the
network. For example, the distance between routers A and
C via B is 6 using 2 hops and via D, E and F is 33 using 4
hops.

5 2

Figure 1 : Network Topology

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

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

Vol. 8, No. 3, 2010

TimeO

IsitariAW
Shore «5i

Prinrfno
new

pTimf
path

Distance calcul a(ti on
ace or di ng to ti m «

PHntf
paih

Printf

Printfno
n*w

Printt

Figure 2: System Flowchart

B. System Implementation

spin's PROMELA language has been used to construct
complete model of DVR protocol on a Pentium machine.
Packets from the source to destination travel using links
provided by routers by making use of their routing tables for
the given distributed environment of the network. After
initialization of the variables, distance is calculated from each
router at time period T=0, T=l, T=2, T=3, T=4 and T=5. At
each stage it checks whether the measured distance forms a
new shortest path or not. Whenever the shortest path is found
from the source to destination, routing table entry for the
concerned router is automatically updated to make good
forwarding decision in order to ensure optimal path, having
minimum distance, for faster communication. Thus, each router
updates its routing table after each time period. The main
objective of the DVR protocol is to provide the current best
route (path) from source to destination for each

communication. Flowchart of the
SPIN/PROMELA is shown in Figure 2.

modeled system in

For the given network, the PROMELA program has six
processes (one for each time period) to find distance based
upon the time period conditions (0 to 5). The found distance
from a particular source to destination for each time period is
compared with all the available alternate routes. Router's table
is only updated if the new distance is minimum between the
selected source and destination. The new shortest path is
recorded after each calculation. If the determined route does
not find minimum distance during the given time period, then it
ignores its path without updating any entry in the routing table.
Routers improve their routes whenever a router advertises its
routing table to its neighbours. So, new routes are determined
purely based on their length measured in distance. For timely
convergence, the number of hops involved in the length are
limited to 15 as already highlighted by Bhargavan et. al. [7].

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

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

Vol. 8, No. 3, 2010

Via

Via 1

Via

From A

FromB

FromC

T=0

T=l

T=2

T=3

T=4

T=5

Table 1: Calculated Distance from Routers A, B and C for Different Destinations at Time Periods T=0 to T=5

C. Operation ofDVR Protocol

DVR protocol works independently for every destination
and it is assumed that there is no topology change for
protocol's convergence during every time period. The router
broadcasts after every 30 seconds and the destination is taken
as inaccessible if it is not refreshed for 180 seconds. The route
is removed from the tables of concerned routers if the
particular router fails to refresh itself for 300 seconds.

Although the PROMELA's built model can be used for any
number of routers but its operation is restricted only to the

topology given in Figure 1 . For the purpose of explanation of
the model, it is assumed that every router operates without any
problem and updates its routing table during regular intervals
of time.

At Time=0, it calculates distances to neighbouring routers
from each router having maximum one hop. Thus, distance
from A to B is 3 & A to D is 23from router A; from router B it
is 3, 3 & 5 for routers A, C & E respectively; and distances are
5, 5 & 2 for routers B, D & F respectively from router E. These
distances can be observed in Tables 1 and 2. Now two hops
from the current router are taken for T=l. So, distance from A

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

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

Vol. 8, No. 3, 2010

Via

FromD

FromE

T=0

T=l

T=2

T=3

T=4

T=5

Table 2: Calculated Distance from Routers D, E and F for Different Destinations at Time Periods T=0 to T=5

to C via B is 6; A to E via D 28; and A to E via B is 8 as
given in Table 1 .

When T is taken as T=2, three hop lengths are counted for
determining the distance from each router. From router D,
measured distances are 13 via E to A, 29 via A to C, 10 via E
to C and 31 via A to E. Same can be seen in Table 2. Hop
length is four when T=3, distance covered to B via E, D via
C and D via E is 33, 16 and 33 respectively from router F as
shown in Table 2. Similarly, routes have distances of 36 (B
via F), 36 (D via F) and 36 (F via B) from router C (for T=4)
as given in Table 1. Both Tables 1 and 2 clearly indicate that
no routes are available from any router when T=5 (six hops)
for network configuration of Figure 1 .

V. Formal Verification of Simulation Results

The implemented system in PROMELA programming
language has been tested exhaustively and obtained
simulation results are shown in Tables 1 and 2. Spin model
checker has been used to verify all the results. The developed
model ensures that all the routers correctly maintain and
update their tables as and when new routes are searched and
visited. The broadcast mechanism works well at different
time periods and the system provides correct and optimized
results from each router to various destinations depending
upon network topology, layout of routers and links
connecting different routers in the network.

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

The spin's verification model successfully checks all the
available routes via different routers and permits only the
shortest path from the available options. It is evident from
the following decisions (only four out of many are presented
here):

1) At T =1, the route length from E to C via B is 8
where as it is 5 via F. So, E router adopts F router's
path to reach C.

2) The distance between routers B &E via A and via C
is 31 and 8 respectively. SPIN's checker confirms
that minimum distance is covered for reaching to C
from E when T=2.

3) When T=3, the path cost determined by the model is
13 from C to A via F, E & B but another path for
connecting the same two router via B, E & D is 36,
each path makes use of four hops. Of course, the
longer path is simply ignored.

4) Similarly, route length from F to D through C, B &
A is 32 and it is 16 via routers C, B &E. A saving of
16 is noted while using the most economical path.

A careful analysis of the simulation results shown in
Tables 1 & 2 clearly indicates that the modeled system in
PROMELA operates correctly and provides the best possible
routes involving minimum distances using DVR protocol on
the given network environment. The system works
efficiently under all conditions and the SPIN model checker
has guaranteed correctness of all results. It means that all the
routing tables are timely updated while messages are being
sent to various destinations from a particular source. Now,
this can be extended to bigger networks in the distributed
environment for efficient and correct functioning using SPIN
tool.

VI. Conclusions and Future Work

Many researchers have implemented DVR protocols for
various applications. In this research work, PROMELA
language has been used to implement DVR protocol on a six
router model. Formal verification of DVR protocol
properties has been shown through the use of SPIN checker
model. The simulation results amply demonstrate correctness
and reliability of DVR protocol under varying conditions.

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

Vol 8, No. 3, 2010
Performance of the implemented has been extremely well
and it can further be improved to make it more efficient in
terms of reducing storage space requirements, incorporating
security mechanism for safer communication, minimizing
congestion at peak loads and making it fault-tolerant for
enhancing its reliability and flexibility.

References

[I] M. R. Masillamani, A. V. Suriyakumar, R. Ponnurangam and and
G.V.Uma, "Genetic Algorithm for Distance Vector Routing
technique", AIML International Conference, 13-15 June 2206, Egypt,
pp. 160-163.

[2] Andrew S.Tanenbaum, "Computer Networks", 4 "" Edition,. Prentice-
Hall Inc., 2005.

[3] G. Coulouris, J. DoUimore and T. Kindberg, "Distributed Systems :
Concepts and Design, 4th Edition, Addison- Wesley, 2005.

[4] S. Mahlknecht, S. Madani and M. Rotzer, "Energy Aware Distance
Vector Routing Scheme for Data Centric Low Power Wireless Sensor
Networks," Proceedings of the IEEE International Conference on
Industrial Informatics INDIN 06, Singapore, 2006.

[5] Liwen He, "A Verified Distance Vector Routing Protocol for
Protection of Internet Protocol", Lecture Notes in Computer Science,
Networking - ICN 2005, Volume 3421, Springer, pp. 463-470.

[6] K. Bhargavan, D. Obradovic and C. A. Gunter, "Formal Verification
of Standards for Distance Vector Routing Protocols", Journal of the
ACM, Vol. 49, no. 4, July 2002, pp. 538-576.

[7] K. Bhargavan, C. A. Gunter, and D. Obradovic, "Routing Information
Protocol in HOL/SPIN", Proceedings of the 13* International
Conference on Theorem Proving in Higher Order Logics
2000, August 14-18, 2000, London, UK, pp. 53-72.

[8] C. Hendrick, "Routing Information Protocol", RFC 1058, IETF, June
1988.

[9] G. Malkin, 'RIP Version Carrying Additional Information', IETF
RFC 1388, January 1993.

[10] J. Katoen, "Concepts, Algorithms and Tools for Model Checking",
Lecture Notes 1998/1999, Chapterl: System Vahdation.

[II] N. A. S. A. Larc, "What is Formal Methods?",
http://shemesh.larc.nasa.gov/fm/fm-what.html, formal methods
program.

[12] R. de Renesse and A. H. Aghvami "Formal Verification of Ad-Hoc
Routing Protocols using SPIN Model Checker", Proceedings of IEEE
MELECON'04, Croatia, May 2004.

[13] G. J. Holzmann, "The Model Checker SPIN", IEEE Transactions on
Software Engineering, Vol. 23, No. 5, May 1997, pp. 279-295.

[14] G. J. Holzmann, "Design and Validation of Computer Protocols",
Prentice Hall, November 1990.

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

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

Vol 8, No. 3, 2010

Integrated Queuing based Energy-Aware Computing

in MANET

Dr. P.K. Suri, Dean, Faculty of Sciences, Dean,

Faculty of Engineering, Professor, Deptt. of Comp.

Sci. & AppL, Kurukshetra University

Kurukshetra, Haryana, India

pksurikuk@rediffmail.com

Kavita Taneja, Assistant Professor

M. M. Inst, of Computer Tech. & Business Mgmt.

Maharishi Markandeshwar University

Mullana, Haryana, India

kavitatane@gmail.com

Abstract — Mobile Computing has witnessed a flare-up of
applications in mobile and personal communication. It is a
phrase that embodies on-the-go business initiatives. By marrying
today's dominant office computing environment with
increasingly compact-but-powerful handheld devices, mobile
computing makes it possible for millions of workers to conduct
business on their feet and from the road. Energy efficiency is an
important design consideration due to the limited battery life of
mobile devices. In order to minimize energy consumption and
maximize the network life time, the proposed simulator opt for
intelligent routing through substitute routes instead of
conventional routing through shortest route. To reduce energy
consumption only devices in the traversed route are active, other
mobile devices of network are switched off. Also simulator
implements clusters for efficient sleep and active mobile device
mechanism and reflects on MANET in terms of queuing network
and consider the packets arrival rate in terms of poisson
distribution.

Keywords- Queuing theory, poisson distribution, clustering,
mobile unit, mobile ad hoc network.

I. Introduction

MANET (Mobile Ad Hoc Network) is a group of mobile units
(MUs) that instantly form a network among themselves
without any fixed infrastructure. The mobile computing offers
users in such open networks, the on the fly access to
information. Due to the explosion of wireless and portable
devices, such as cellular phones, PDAs (Personal Digital
Assistants), palm computers, user is spoilt to enjoy the
freedom and convenience in their daily lives [1]. In addition,
the race against time and enhancements in the current
MANETs make user more expected to do things through the
Internet, such as online banking, online shopping etc. Then
there is no doubt user's demand for information access in
mobile environment increases spectacularly [2]. Since the
mobile computing is relatively novel to its big brother-
desktop computing, the resource in the mobile computing
world is relatively restricted. Also constraints, such as limited
hardware resources especially bounded battery life, stochastic
topology hence uncertain device location, etc, make
challenges for optimally utilizing resources. Increasingly,
power consumption within such open networks is becoming a

core issue for the low-power mobile devices [3]. For
example, soldiers in a battlefield can have handheld mobile
devices to communicate with each other, or in emergency
situations like earthquakes where the existing infrastructure
has been destroyed, an ad hoc network can instantly be
deployed to aid in disaster recovery, meetings or conference in
which persons wish to quickly share information, and data
acquisition operations in inhospitable terrains. In order to
crack shortages in mobile computing for smooth satisfaction
of the emerging needs, the enhanced computing needs to be
adaptive [4]. The primary goal of routing protocol in MANET
is correct and efficient route establishment between a pair of
MUs so that packets can be delivered in a timely manner with
minimum battery consumption for longer network
connectivity. In the evaluation of any algorithm for energy
conservation, an estimate of energy consumption is necessary
[5].

The energy is consumed per MU in two phase as
shown in Eq. (1), the energy used in route discovery and
amount of energy used in packet transmission.

-Total .

-'Route-Discovery

-'Packet- Transmission

T^ioiai _ T^KOuie-uiscovery , x^i-acKei-iransmission /i \

During packet transmission a typical MU may exist in any one
of the four modes:

a) Sleep mode: Sleep mode has very low power
consumption. The network interface at a MU in sleep
mode can neither transmit nor receive packets; the
network interface must be woken up to idle mode first
by an explicit instruction from the MU.

b) Idle mode: In this mode MU is neither transmitting nor
receiving a packet. This mode still consumes least
amount of power because the MU has to listen to the
wireless medium continuously in order to detect a
packet that it should receive, so that the MU can then
switch into receive mode.

c) Receive mode: In this mode a MU actually receives a
packet.

d) Transmit mode: In transmit mode a MU uses its
maximum power level to transfer packet to another
MU.

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

Receive and idle mode require similar power, and transmit
mode requires slightly greater power [6]. Sleep mode
requires more than an order of magnitude less power than
idle mode. These measurements show that the network
interface expends similar energy, whether it is just listening
or actually receiving data. Hence, switching to sleep mode
whenever possible is a wise decision on the part of MU that
will lead to momentous energy savings. The sources of
power consumption, with regard to network operations, can
be classified into two types: (a) communication related, and
(b) computing related. Communication involves the cost of
sending data along with control packet from source MU,
routed through intermediate MU and cost at receiving at
destination end. It mainly deals with the cost of routing in
network or in route establishment. Whereas computing
mainly involves the cost of using CPU and main memory,
disk drives and other components of computer for
calculating computing related cost [7]. If we take the
shortest route to deliver the message it is not the good way
of energy conservation in network. As a group of MUs
coming under the shortest route is used rapidly as compared
to other MUs, so their power decreases rapidly and may a
situation come that they are having no power. In that
situation we refer the substitute path routing [8]. The
substitute routes and clustering approach ensures that the
energy of a group of MUs is used at a particular instant in a
respective way [9]. By this all MUs of a network take active
participation in route selection; overall energy expenditure
of a network is minimized, causing maximization of
network life. The remainder of this paper is organized as
follows. The basic concept of queuing theory and MANETs
is explained in section II. Section III gives the Simulator for
finding substitute route and clustering approach in MANET.
Section IV gives the probabilistic approach of finding the
number of sleep MU and active MU. Finally we conclude
our paper in section V.

II. Queuing Theory And Manet

We consider that inter arrival times of packets at a MU follow
a discrete poisson distribution in which an event occurring
exactly k times during an interval t is given by a probability
mass function

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

Vol 8, No. 3, 2010
We generate Ti's according to Eq. 4 and keep on adding them
till their sum exceeds 1 and the count gives the Poisson sample
(k). At any given time t, the probability of a MU busy in
packet processing is given by "(5)"-

gk(t) = (t^)^(l/k!)e-^

(2)

where \ is the average number of times the event occurs in a
unit period. In our network the packet is always transferred
from source to destination, we consider our source as a server.
With single server, service times follow negative exponential
distribution. Packets are served in FIFO order. Pseudo
random numbers are generated and using x^ test, we have
samples (T) from an exponential distribution with specified
expected value \l\ as

RN=RNDY1(DUM),
= -1.0/}i*ALOG(RN).

(3)
(4)

p = P / a.

(5)

<~Packet queue- ^

Figure 1 Queuing theory embedded in MANET

Considering computation of Packet queue length of each MU
as shown in figure 1 , if more packets passed through Packet
queue it is lost, here Packet queue limit is n. Because packet
arrival at a MU is highly stochastic, hence simulator design
assumes that the Packet queue length is such that it is 99% of
the time sufficient to hold the packets waiting to be processed.
Thus the probability of having more than n packets in wait for
MUis

(6)

p"^^=0.01.

In "(6)", n gives the long enough Packet queue length with 99
per cent assurance. For efficient routing in order to lessen the
number of packets loss we suggest that the packet to be
delivered in substitute routes instead of shortest route. So
initially the numbers of all possible substitute routes are to be
calculated. Then while packets traveling through a particular
route the MUs of other routes to be switched off if they are not
doing any useful work in order to minimize energy
consumption. The algorithm describing this substitute route
finding procedure is given in section III.

III. Algorithm for substitute route And Clustering
Approach.

In this section we propose our algorithm to find out substitute
routes in a network to maximize network life if direct route
between source and destination does not exist [10]. Initially
the MUs are set up in an area to establish network
connectivity. Then we go for substitute route finding

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

procedure, by running this algorithm in a network we keep
track of all substitute routes which does not contain any
duplicate MUs. By taking the substitute routes traffic load is
shared and congested routes are avoided which may cause in

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

Vol 8, No. 3, 2010
required amount of energy it can elect the MU with huge
resource as CH in order to maintain network connectivity.

Algorithm for substitute route ( ):

1. Initialize all MUs of network to READY state from
SWITCH OFF state at time of deployment of network.

2. Put the source MU (S) in Packet queue.

3. Repeat step 4 until Packet queue is empty.

4. Process front MUs of Packet queue by adding its
neighbor's to Packet queue where each entry is unique
along with that keep track of their parent MU.

5. End of step 3.

6. As destination (D) is reached stop and find the route
traversing from 'D' in a reverse order by tracking the
origin till 'S' at origin is reached.

7. SWITCH OFF all MUs coming in the route for finding
another route without any duplicate MU.

7. Go to step 4

8. Exit.

Figure 2. Algorithm: Substitute Route

retransmission due to collision. Since no MU is duplicated, all
MUs take in active participation in route selection. No MU is
penalized more as compare to other MUs in network. So there
is no network partition which causes maximization of network
life [4, 11, 12]. Before going to the algorithm find out hst of
neighbors of all MUs. For evaluation of algorithm a priority
Packet queue is considered. Considering that the front MU of
a Packet queue has to be processed first as it has highest
priority than other. The network is divided level wise. Each
level is forming a cluster [6, 13]. Each cluster is assigned with
a cluster head (CH). A MU is selected as a CH having more
battery energy [14] which can communicate to its neighbor in
one hop distance. The lower level of network which is nearer
to the source MU keeps track of all substitute routes. All the
CHs which are formed in a network can exchange their
messages in between them at any point of time by
broadcasting messages in between the CHs. The CH in each
cluster broadcasts the control message to its entire MUs. This
control message contains the information about MU ID, and
residual energy. When packet starts transferring in one route,
the CH of initial level make the MU taking part in route
selection as ACTIVE and make other MUs to be in SLEEP
mode. It broadcasts the message to all the CHs. All the CH
makes the MUs in that path active and other to be in sleep
mode. When the next packet starts traversing it transfer
through another route. Then the CH positively play their role
to make the MU falling in that route to ACTIVE and other
MUs of the cluster to be SLEEP mode. Using this when
packets transferring in a particular route, the MUs of that route
are in ACTIVE state where all other MUs of a network are in
SLEEP mode. So this lessens the overall energy consumption
of whole network. At any point of time if CH is not having

CH: Cluster Head
D: Destination
S: Source

Figures Clustering in MANET

IV. The Probabilistic Approach

To find out the number of MUs to be ACTIVE and the number
of MUs to be in SLEEP MU, we take the following
probabilistic approach. Let T be the total number of MUs, A
be the number of ACTIVE MUs and S be the number of MUs
in SLEEP mode. Now we find the joint probability of 'a' MUs
among the total MUs active and 's' MUs among total MUs in
sleep mode. The packet arrival distribution follows Poisson's
distribution (k). Let probability of a MU being ACTIVE
mode= P. The probability of a MU being in SLEEP mode= 1-
P. For the discrete random variable, the joint probability mass
function

P {A=a and S=s} = P{S=s | A=a}*P{A=a}

= P{A=a|S=s}*P{S=s}. (6)

Also, S S P{ A=a and S=s}=l, Since A and S are independent
P{A=a and S=s}= P{A=a}* P{S=s}

= (re-Va!)*(:^VVs!). (7)

It shows that probability of 'a' number of MUs in ACTIVE
state and 's' number of MUs in SLEEP state. By putting the
value of probability to 1 and specifying the rate of arrival X we
find the ratio of number of ACTIVE MUs and SLEEP MUs,
which effectively measure the energy consumption rate. As by
the MU going into SLEEP state it consume less energy,
therefore as the number of MUs not doing any useful work in
the system going into SLEEP state, then the energy
conservation will increase, which enhance our system
performance with maximized network lifetime.

V. Conclusion

In this paper we give a brief study about role of queuing
theory in MANET. As all MUs are battery operated, and
battery power is restricted we go for proficient use of this in
order to minimize energy consumption and maximize network

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

life [15]. The today's user rely on mobile computing for a
multitude of operations as is equipped with range of devices
like notebook computers, personal digital assistants and other
communication hungry portable stuff, even kitchen of
homemakers is gripped with spectrum of wireless interfaces
for networked communication [16]. By using the substitute
route and clustering technique we try minimize energy
consumption of overall network. Also switching some of MUs
to SLEEP mode we try to further minimize energy
consumption in network. To calculate the number of ACTIVE
MUs and SLEEP MUs in network we are using the
probabilistic approach. This model promises to provide
stochastic handling equipped design platform for extension to
an energy efficient protocol for MANET.

References

[I] Sarkar, S. K., T.G. Basavaraju, and C. Puttamadappa, "Ad Hoc Mobile
Wireless Networks: Principles, Protocols and Applications", Auerbach
Publications, 2007.

[2] Do-hyeon Lee , Song Nan Bai , Jae-il Jung, "Enhanced next hop
selection scheme for QoS support in multi-hop wireless networks".
Proceedings in the 2009 International Conference on Hybrid Information
Technology, pp.587-593, August 27-29, 2009, Daejeon, Korea.

[3] D. Kim, J. Garcia and K. Obraczka, "Routing Mechanisms for Mobile
Ad Hoc Networks based on the Energy Drain Rate", IEEE Transactions
on Mobile Computing. Vol 2, no 2, 2003, pp.161-173.

[4] Thomas Kunz, "Energy-efficient routing in mobile ad hoc networks: a
cautionary tale". International Journal of Autonomous and Adaptive
Communications Systems, Vol.2 no.l, pp. 70-86, March 2009.

[5] L. Feeney and M. Nilsson, "Investigating the Energy Consumption of a
Wireless Network Interface in an Ad Hoc Networking Environmenf ,
Proceedings in INFOCOM 2001, Vol. 3, pp. 1548-1557, Anchorage,
Alaska, April 2001.

[6] C. E. Jones, K. M. Sivalingam, P. Agrawal and J. C. Chen, "A survey of
Energy Efficient Network Protocols for Wireless Networks", Journals in
Wireless Networks, Vol. 7, no. 4, pp. 343-358, August 2001.

[7] M. Tarique, K. E. Tepe and M. Naserian, "Energy Saving Dynamic
Source Routing for Ad Hoc Wireless Networks", Proceedings in Third
International symposium on Modelling and Optimization in Mobile Ad
Hoc, and Wireless Networks, Canada, pp. 305-310, April 2005.

[8] C.K. Toh. "Maximum Battery Life Routing to Support Ubiquitous
Mobile Computing in Wireless Ad Hoc Networks", IEEE
Communications Magazine, Vol 39, no 6, 2001, pp. 138-147.

[9] E. M. Royer and C. K. Toh, "A Review of Current Routing Protocol for
Ad Hoc Mobile Wireless Networks", Proceedings in IEEE Personal
Communication Magazine, Vol. 6, no. 2, pp. 46-55, April 1999.

[10] M. Tarique, K. E. Tepe and M. Naserian, "Energy Saving Dynamic
Source Routing for Ad Hoc Wireless Networks", Proceedings in Third
International symposium on Modelling and Optimization in Mobile Ad
Hoc, and Wireless Networks, Canada, pp. 305-310, April 2005.

[II] K. Pappa, A. Athanasopoulos, E. Topalis and S. Koubias,
"Implementation of power aware features in AODV for ad hoc sensor
networks. A simulation study". Proceedings in IEEE conference on
Emerging Technologies and Factory Automation, Rome, pp. 1372-1375,
September 2007.

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

Vol 8, No. 3, 2010

[12] C. Jie, C. Jiapin and L. Zhenbo, "Energy- efficient AODV for Low
Mobility Ad Hoc Networks", Proceedings in IEEE International
Conference on Wireless Communications, Networking and Mobile
Computing, Shanghai, pp. 1512-1515, September 2007.

[13] S. Soro and W. B. Heinzelman, "Cluster head election techniques for
coverage preservation in wireless sensor networks". Proceedings in
Elsevier Journal Ad Hoc Networks, Vol. 7, no. 5, pp. 955-972, July
2009.

[14] X. Hou, D. Tipper and S. Wu, "A Gossip-Based Energy Conservation
Protocol of Wireless Ad Hoc and Sensor Networks", Journals of
Networks and Systems Management, Computer Science, Springer New
York, Vol. 14, no. 381-414, September 2006.

[15] Zhao Cheng, Wendi B. Heinzelman, "Discovering long lifetime routes in
mobile ad hoc networks". Ad Hoc Networks, Vol.6 no. 5, pp. 661-674,
July 2008.

[16] Helmut Hlavacs , Karin A. Hummel, Roman Weidlich , Amine M.
Houyou, Hermann De Meer, "Modelling energy efficiency in distributed
home environments". International Journal of Communication Networks
and Distributed Systems, Vol.4 no. 2, pp. 161-182, January 2010.

Dr. P. K. Suri received his Ph.D. degree from
Faculty of Engineering, Kurukshetra University,
Kurukshetra, India and Master's degree from Indian
Institute of Technology, Roorkee (formerly known as
I Roorkee University), India. Presently He is Dean,
Faculty of Sciences, Dean, Faculty of Engineering,
Kurukshetra University and is working as Professor in
the Department of Computer Science & Applications,
Kurukshetra University, Kurukshetra, India since Oct.
1993. He has earlier worked as Reader, Computer Sc. & Applications, at
Bhopal University, Bhopal from 1985-90. He has supervised six Ph.D.'s in
Computer Science and thirteen students are working under his supervision. He
has more than 110 publications in International / National Journals and
Conferences. He is recipient of 'THE GEORGE OOMAN MEMORIAL
PRIZE' for the year 1991-92 and a RESEARCH AWARD -"The Certificate
of Merit - 2000" for the paper entitled ESMD - An Expert System for
Medical Diagnosis from INSTITUTION OF ENGINEERS, INDIA. His
teaching and research activities include Simulation and Modeling, SQA,
Software Reliability, Software testing & Software Engineering processes.
Temporal Databases, Ad hoc Networks, Grid Computing and Biomechanics.

Kavita Taneja has obtained M.Phil(CS) from
Alagappa University, Tamil Nadu and Master of
Computer Applications from Kurukshetra
University, Kurukshetra, Haryana , India. Presently,
she is working as Assistant Professor in M.C.A. at
M.M.I.C.T & B.M., M.M. University, Mullana,
Haryana, India. She is pursuing Ph.D in Computer
., ^ Science from Kurukshetra University.

She has published and presented over 10
papers in National/International journals/conferences and has bagged BEST
PAPER AWARD, 2007 at International Conference for the paper entitled
"Dynamic Traffic -Conscious Routing for MANETs" at DIT, Dehradun. She
has supervised five M.Phil scholars in Computer Science. Her teaching and
research activities include Simulation and Modeling and Mobile Ad hoc
Networks.

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

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

Vol 8, No. 3, 2010

A Review of Negotiation Agents in e-commerce

Sahar Ebadi

Department of Information System,

Faculty of Computer Science and

Information Technology, University

Putra Malaysia, Serdang, Malaysia

Sah_ebadi@yahoo.com

Md. Nasir Sulaiman

Department of Information System,

Faculty of Computer Science and

Information Technology, University

Putra Malaysia, Serdang, Malaysia

nasir® fsktm.upm.edu. my

Masrah Azrifah Azmi Murad

Department of Information System,

Faculty of Computer Science and

Information Technology, University

Putra Malaysia, Serdang, Malaysia

masrah. azrifah @ gmail .com

Abstract — With the growth of World Wide Web and the
increasing human demand on onHne trading, there is a pressing
need for complex systems that are capable of handling the client
needs in e-commerce. In recent years, numbers of Multi Agent
System (MAS) developers arise to fulfill this mission by
performing a huge number of studies on agent negotiation
systems. However, far too little attention has been paid to
provide a rich review as a repository for developers to distinguish
the aspect and scope of MAS. The purpose of this paper is to do
a review of progressing agent negotiation technology in e-
commerce. In order to achieve our aim we propose different
classification schemata and interpret different models according
to the proposed classifications. Popular methods for optimizing
negotiation agents have been introduced and the effect of relative
techniques has been analyzed. The result of analysis shows that
genetic algorithm is the most effective learning technique in
optimizing negotiation models. Moreover, we interpret the most
prominent negotiation models according to the main parameters
on which any negotiation agent model depends. The result of
these analysis supplies a resource of differentiating competing
alternatives for the area of negotiation agent's models to exploit.
Finally, a range of open issues and future challenges are
highlighted.

Keywords-component; Artificial Intelligence; Agent; Multi-
Agent System; Negotiation; e-Commerce

I. Introduction

With the rapid growth of the World Wide Web and huge
demand of online trading every day, there is a need for
complex systems that are capable of addressing the online
trading needs of human. Such systems must be capable of
establishing communications, making decisions and handling
customer's requirements. Many researchers in Multi Agent
System (MAS) and agent negotiation systems succeed to fulfill
these obligations on e-commerce.

As the number of research conducted on agent negotiation
development has rapidly increased, the need of conducting a
comprehensive review on negotiation agent in e-commerce has
increased. So far, however, there has been little review about
agent negotiation models specifically in e-commerce. The
purpose of this paper is to review studies conducted on
negotiation agent systems in e-commerce (online trading). In
general, lack of universally accepted definitions in negotiation
agent systems is one of the difficulties in this area. Therefore,
we initially clarify the negotiation agent system's

characteristics and concepts used in this field. Then we
explicate the most popular methods in optimizing negotiation
agent models. Related prominent techniques will be explained
and some exemplar stand out models will be interpreted.

This paper presents two novel classifications, namely
classification according to the type of agents, and classification
according to the attitude of agents; in addition to the
Wooldridge's popular classification according to the number of
agents involved in negotiation. These classifications help
developers and researchers to have a better understanding of
aspects and scope of agent negotiation systems. These features
can facilitate future developments on negotiation mechanism in
MAS. Finally, we will trace the progress of negotiation
systems generations by analyzing the most prominent works
during the last decade, from early 1996 to late 2009. In this
work, we will interpret the whole negotiation agent model
according to four important characteristics of the system so-
called negotiation protocol, negotiation strategy, agent
characteristics and negotiation setting. In addition, the most
effective learning technique will be verified through
interpreting the final utility of exemplar models. The remainder
of this article is structured as follows. Section 2 presents
essential characteristics and concepts in the scope of our work.
Section 3 presents popular methods of optimizing negotiation
models and a brief description of the relative techniques. In
section 4, proposed classifications are discussed. Section 5
draws together the two later strands. It discusses some
exemplar negotiation agent-based applications and highlights
new direction to the future works. Section 6 presents the
conclusion of this paper.

II. Concepts AND Characteristics

As mentioned earlier, one of the difficulties in agent
negotiation systems is the lack of universally accepted
definitions. In order to draw a clear picture of concepts and
characteristics of negotiation agents, it is important to recap
some key concepts and definitions which are accepted by some
experts in this field.

Agent: an agent is a computer system that is situated in
some environment, and is capable of autonomous action in this
environment in order to meet its designed objective [1].
According to Wooldridge and Jennings [1], the term agent is
most generally used to denote a hardware or (more usually)

* Responsible author

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

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

software-based computer system that enjoys the following
properties:

autonomy: agents operate without the direct
intervention of humans or others, and have some kind
of control over their actions and internal state [2]

social ability, agents interact with other agents (and
possibly humans) via some kind of agent-
communication language [3]

reactivity: agents perceive their environment, (which
may be the physical world, a user via a graphical user
interface, a collection of other agents, the
INTERNET, or all of these combined), and respond in
a timely fashion to changes

• pro-activness: agents do not simply act in response to
their environment, they are able to exhibit goal-
directed behavior by taking the initiative.

In most of the real world problems, agents need to interact
with other agents to achieve their objectives. Many problem
cases are innately multi party or social such as negotiation
scenarios. Many are more complex to be solved by an agent
(e.g. monitoring and performing an electronic marketplace). In
these cases, multi agent systems are designed to address these
issues.

Multi Agent System: generally Multi Agent Systems (MAS)
refers to such a system that many intelligent agents involved
interact with each other in a selfish or cooperative manner. In
MAS, agents are autonomous entities and can cooperate to
reach a common goal (e.g. an ant colony) or just follow their
own preferences (as in the free market economy).

According to Sycara [4], "the characteristics of MASs are
that (1) each agent has incomplete information or capabilities
for solving the problem and, thus, has a limited viewpoint; (2)
there is no system global control; (3) data are decentralized;
and (4) computation is asynchronous."

Since trading domains are often bounded by limited
resources and abilities, negotiation has become an essential
activity in e-commerce applications.

Negotiation: negotiation is a process in which two or more
parties with different criteria, constraints, and preferences,
jointly reach to an agreement on the terms of a transaction [5].

Negotiation Agents : according to Raymond [6], the notion
of agency can be applied to build robust, fast and optimal
architecture for automated negotiation systems within a group
of software agents communicating and autonomously making
decisions on behalf of their human users .

After a clear understanding of what agent-based concepts
are, it is necessary to emphasize the key concepts on
optimizing negotiation agent systems on e-commerce.

III.

Vol. 8, No. 3, 2010
NEGOTIATION AGENT'S CLASSIFICATION

Negotiation Agents can be categorized by severed
orthogonal dimensions. In this work, we present three
classifications. The first one is a popular classification
introduced by Wooldridge [5]. The second and third
classifications are proposed by this work. Proposed
classifications schemata are done according to i) the types of
agents involved in negotiation and ii) the attitudes of agents
involved in negotiation.

A. Number of Agents Involved in the Negotiation

This category is one of the most clear and popular
categories which are divided into three groups named one-to-
one negotiation, one-to-many negotiation and many-to-many
negotiation.

One-to-one negotiation is suitable in situations where one
agent is negotiating with another agent (e.g. Kasbah model
proposed by Anthony et al. [7] and Ryszard Kowalczyk [8]) in
case where one agent is involved in the negotiating with other
agent. This is a simple but basic kind of negotiation on e-
commerce.

One-to-many negotiation is when one agent negotiates with
several agents as its opponents. This kind of negotiation, in
fact, is originated from several combinatorial one-to-one
negotiations. A practical example of a one-to-many
negotiation is auction system where several bidders participate
in an auction at the same time. Many researchers [9, 10] on
online trading propose their models based on this classification.

Many -to-many negotiation as described by Wooldridge [5],
is when many agents negotiate with many other agents
simultaneously. Jiangbo many-to-many negotiation framework
is one of the best examples of this category [11-13].

B. Type of Agents Involved in Negotiation

We believe finding the suitable classification of opponent
agents type will help agents to choose the best possible tactics
to deal with their opponents over a particular good. We think
this classification will decrease the negotiation cost by
improving the accuracy of agent's behefs. Many researchers
[10, 14, 15] have discussed types of agents involved in a
negotiation. According to the variety of application different
concepts are offered. Nguyen [10] divided agents involved in
negotiation into two groups according to the amount of the
concede which they are willing to give. The first type is
conceder and the second is non-conceder. The conceder is
referred to the agents that are willing to concede with the aim
of selling. In contrast, non-conceder agents are those who just
deal with a situation where there is some amount of benefit.
Bayesian classification was employed to identify the probable
types of agents. Then, agents try to modify appropriate
strategies according to the type of the opponent agents.

A new direction in this area is to apply some sense of
machine learning techniques to increase accuracy of the
opponent agent classification. Therefore, agent is able of
making a better deal that improves the performance of the
model.

Ros and Sierra [16] defined 5 types of agents that have been
experienced by the model and finally evaluated by utility

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

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

product and utility difference. These 5 types of agents so-
called NegoTo agent, Random agent, Alternate agent, TOagent
and Nego agent are differentiated by tactics and behaviors they
obey. These 5 types of agents are explained In a nutshell.
NegoTo agent, before reaching deadlock applies trade-off
tactic, followed by negoEngin tactic. Random agent chooses
the next tactic randomly (e.g. negoEngin, trade-off, trade-off
and negoEngin). The above-mentioned agents alter negoEngin
and trade-off tactics one at a time. TOagent obeys trade-off
tactics when utility of new offer is higher than the previous
one, otherwise, the aspiration level is decreased and new offer
is proposed. Nego agent only follows negoEngin tactics during
the negotiation.

We think that applying a decision making model that
discovers the highest utility among all tactics can increase the
performance of the system. In addition, applying the previous
history or knowledge extracted from past experience (e.g. what
happened in the past when we used a specific tactic in the same
situation?) in the same situation can help agents decision
making. This will result in a more accurate choice of tactics or
type of agents to negotiate and end in a higher chance of
reaching agreement.

C. Attitude of Agents Involved in Negotiation

Internet is populated by many agents coming from different
sources with different attitudes and goals. Some researchers
mentioned that, although agents can be categorized by their
behaviors, they can also be categorized according to their
attitudes toward their goals. In a clear word, agent's attitude
defines how an agent selects beneficial opponent in a given
different situation[17]. Many researchers are working on this
area in order to reach to a better efficiency in their models [10,
15, 17, 18] . In Ikpeme proposed model [15] agents are divided
according to their social attitudes into three different groups so-
called helpful, reciproactive, and selfish. Helpful agents are
those who are willing to help. This group benefits in the
homogenous groups of helpful agents. Reciproactive agents
evaluate the request and will accept opponent's request if the
agent meets a balance of cost and saving, otherwise they reject
the request. This group has the best performance when situated
in an open group of agents. Selfish agents never help the other
agents. So, they always benefit when situated among helpful
agents but rarely benefit from reciproactive agents. Finally, the
results show that in an open environment success rate is many
times better for reciproactive agents than selfish agents.

In the work presented by Jaesuk et al [17], agents attitude
defines the priority that an agent places on various choices it
may have regarding member selection. So, agent's attitude is
(1) toward rewards or (2) toward risk. Agent attitude toward
reward is the agent's point of view toward finding the best
opponent. The attitude considers the opponent's quality of
service. Agent attitude toward risk is agent sensitivity to the
possible risk of opponent agent (which depends on unreliability
and unavailability of agents). The result of the research shows
that agents with strong attitude toward risk are more beneficial
when there is a higher chance of failing jobs due to tight time
or low availability and reliability of opponent. But, agents with
the high attitude toward quality situated in a case where time is
enough and the penalty value is small are able to earn more
benefit. Also, in some cases agent's classification is done

Vol. 8, No. 3, 2010
according to the agent's behavior as proposed by Ikpeme et al.

[15].

The proposed classifications will clear the roadmap for
developers in establishing new research in the area of
negotiation agent systems. Once developers define the agent
classification schema, automatically dimensions and scopes of
research area are clarified. Simply, by following previous
works conducted in the specified class, the gaps and
disadvantages of research area will be highlighted. This will
assist researchers to develop and maintain innovative research
in the area of negotiation agent models.

IV. Popular Methods of Optimizing Negotiation

Systems

Designing an optimized agent negotiation system is one of
the important issues addressed recently by many researchers
[16, 19-23]. In recent years, a great deal of effort has been
devoted toward optimizing negotiation agents [24-27]. Mostly
this aims at improving agent i) adaptability, ii) intellectuality,
iii) applied strategies, and iv) gathering information.

A. Adaptability

Negotiation agents are situated in open environment where
new agents may come and some agents may leave the
environment. Agents are characterized by deadlines, volatile
preferences, and incomplete information. In such situations,
agents must survive by changing their strategies, preferences
and even learning opponent's preferences and behaviors. This
attempt to change agent's behaviors, preferences, and strategies
toward reaching an agreement is called adaptability. This
method aims to find the highest satisfactory offer for agents as
well as being acceptable for opponent agent. Many researchers
[6, 27, 28] applied this methods for optimizing negotiation
systems, as it is effective and necessary to assist the autonomy
aspect of agents. Zhang et al. [27] proposed a new adaptive
negotiation strategy which assists negotiation models by
enhancing the adaptability of agents. In this scenario, an agent
uses adaptive strategy to estimate the strategies and preferences
that opponent agents used in the last few offers in negotiation.
So, an agent can choose appropriate negotiation factors to
adjust its strategy. Even after choosing the strategy, there is a
chance to change the parameters of chosen strategy to adapt
dynamically to the strategy of another agent. Also, Raymond
[6] proposed a negotiation mechanism in which an agent can
adapt itself by changing its preferences and behavior using
dynamic models. This learning model is applied by using
Genetic Algorithm (GA) in process of decision making, so that
agent can gradually acquire appropriate negotiation knowledge
based on its negotiation history with that opponent. Thus,
Raymond negotiation model aims to reach the same property
but by using machine learning techniques.

B. Intellectuality

As intellectual capability is one of the substantial
characteristics in agent technology, learning appears as one of
the most important and effective methods in improving
negotiation agents. Recently, many researchers enriched their
negotiation agents and models by using different kinds of
learning algorithms or machine learning techniques [6, 9, 26,
29-32]. Former attempts in this area were applied by using
some sort of machine learning techniques (e.g. GA, fuzzy, re-
enforcement learning, simulated annealing and data mining),
distribution probabilistic analysis or by applying some pre-

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

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

programmed strategies. In Intelligent Trading Agents(ITA)
model [9] four different types of strategies were introduced to
help agent's to choose the next action to be taken. E-
Commerce Negotiation (E_CN) [10] also presents complex
negotiation tactics combined with a sort of distribution
probabilistic analysis to predict the opponent's type which will
end to choose the best possible strategy for the next round.
Other researchers tried to employ machine learning techniques
directly to the agent's decision making engine. For example,
Raymond [6] applied a genetic algorithm on the space of
possible offer for agent A to propose to agent B. Fig. 1
represents an example of how agent offer encodes to the
chromosomes and draws a tow-point crossover operation
according to Raymond Lau[6].

Double Crossover

Parent Offer 1

Parent Offer2

Child Offer

20-40

10-20

—

10-30

Quality

Price Quality

Crossover

10-20

Quality

^ 10-30

Figure 1. Encoding Candidate Offers [6]

In this scenario agent A will generate a random pool of
possible offer to propose to agent B. Then GA is applied on
mating pool which is using three standard genetic operators:
cloning, crossover, mutation. The outcome is a list of
generated possible offers which are ranked by their utility
values. The first offer in the list is the offer to be proposed to
agent B by agent A. Proposed GA-based negotiation agents
assist agents by learning their opponent's preferences. This
improves the speed of the search in finding the best possible
offer for both agents. Although this learning ability increases
the overall utility rate to 10.3%, still there are some open issues
to increase the performance of the system. One of the open
issues is to acquire knowledge extracted from the recorded
history or a trusted third party in order to setup the GA
initiahzation accurately (e.g. possible range of value for genes).

Guo et al. [32] suggested an algorithm to extract knowledge
from a user and then inject it into the solution population.
Then, a simulated annealing was applied to render the solution
to make sure that the best possible offer will be proposed.

Isabel et al [26] proposed a multi-agent market simulator.
In this market, agents have the possibility to negotiate through
the pool which is regulated by a market operator (market
administrator). Three data mining techniques are proposed to
mine the administrator's transaction history, which contains all
previous interactions and transactions among agents. The

Vol 8, No. 3, 2010
Tow-Step clustering algorithm is used to cluster buyers

according to agent's characteristics. After that, a rule-based

modeling technique, C5.0 classification algorithm [26] is used

to extract the consumption pattern of each cluster population.

Finally, Apriory algorithm [26] is used to discover association

between buyer details and purchases. A future direction for

this research is to employ some sense of artificial intelligence

to the seller agent which will result in a higher overall

performance of the system.

C. Applied Strategies

The outcome of negotiation depends on several parameters
including agent's strategy. Strategies are divided into three
groups: (1) strategies which depend on time called time-
dependent strategies, (2) strategies which depend on agent's
behavior, called behavior-dependent strategies and (3)
strategies which depend on how a specific resource is
consumed. Recently there have been huge amounts of research
on agent's strategies [9, 10, 26, 27, 29].

In IT A model [9], some time-dependent strategies were
proposed, namely Desperate, Patient, Optimized patient and
strategy Manipulation. Desperate strategy accepts the first
acceptable offer that is suitable. In this strategy agent aims to
reach a deal as soon as possible. Patient strategy waits until all
negotiation threads reach a deal and then choose the best offer.
This strategy guarantees the best possible deal but does not
consider time constraint which is one of the most important
factors in real market places. In Optimized patient strategy, the
outcome of one sub -negotiation affects the performance of
other sub-negotiations. Manipulation is a combination of
above-mentioned strategies. A drawback of such strategies is
the lack of adaptability as they are preprogrammed.

E-CN model [10] used time-dependent strategies called
Conceder, Linear and Tough. Conceder strategy means that an
agent quickly lowers its preference values until it reaches its
minimum reserved value. Linear strategy is when agent drops
its reserved values but in a gradual manner. Tough strategy
deals in a tough manner, meaning that it keeps its values until
agent is close to deadline then suddenly drops the values to its
reserved values. However, these sorts of strategies are sub-
optimal in which using a sense of learning can improve the
efficiency and robustness of system.

Isabel et al. [26] proposed two types of behavior-dependent
strategies named Composed Goal Directed (CGD) and
Adaptive Derivate Following (ADF). CGD is based on two
objectives which should follow sequentially. The first objective
is to be sure that all needed goods are sold or purchased. The
second tries to reduce the pay off of the deal or increase the
benefit. ADF is based on the revenue earned in the previous
period as a result of changes in the price. If the change of price
by the last period produced more revenue than the previous
period, then the strategy takes a similar action otherwise it will
take a counter action.

Although, time-dependent strategies seem simple, with
concern to the time dependent of negotiation agent, they have a
significant effect on system's outcome. A combination of
alternatives of different types of tactics were proposed in this
area [16, 23].

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

D. Gathering Information

The lack of information about environment and opponent
agents is one of the major issues on negotiation systems [18,
26, 32-35] since these information assist the agents to choose
suitable agents and strategies. Such information can be
gathered via agent's recorded negotiation histories or via
trusted third party agents and referral systems. The referral
mechanism allows agents to find their required resources if
there is any agent with the required expertise close to the
location of the neighboring agents [35]. Many researchers [18,
26, 32-36] have focused on this issue with the aim to improve
their negotiation agent models. Guo [32] in his proposed
algorithm referred to information as an important factor which
assists suggested model in learning multi- attribute utility
function. Gathering information is mentioned as the second
step of the Guo algorithm Fig. 2.

Utility Elicitation Algorithm

Applying Evolutionary Operations

Selection, Crossover and Mutation

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

Vol 8, No. 3, 2010
In that second step called "applying external knowledge",

knowledge gathered from outside is obtained directly from user

input or derived from user observation. The base solution is

now modified by injecting external knowledge into the solution

population. The knowledge will be accepted based on

correctness of the knowledge and a certain probability [32] .

Guo et al. [32] showed that taking this knowledge into account

while generating solution population, has a significant effect on

learning user preferences in multi- attribute negotiation.

Applying External Knowledge

Knowledge Acquisition and Integration

Apply Local Search

Local Refinment with Simulated Anealing

Choo et al. [33] also presented a form of optimizing
negotiation agents by employing genetic algorithms. This
model attempts to learn the agent opponent's preferences
according to the history of the counter offers based upon the
stochastic approximation. One of the further discussions in the
agent models is employing trust on negotiation agents

Trust can help agents to follow their aims by gathering
useful information about their opponent's preferences and their
attitudes toward their goals. Multi Dimensional Trust (MDT)
introduced by Griffiths et al [37] opened a future direction in
the area of negotiation agent for finding trustworthy opponent.
Griffiths et al [37] applied a weighting factor concept which
enables agents to combine decision factors according to agents
current preferences. Later on, Gujral [38] provided its agents
with a model of recognizing the best trusted third party to
obtaining information from. In that model, agents should
consider the trustworthiness of the potential opponents in order
to maximize the agent's rewards, inasmuch as the more reliable
the trusty agent is, the higher the chance of reaching agreement.

There are some other important factors in multi agent
negotiation systems which, taken into account, can affect the
outcome of negotiation's systems. These include the number
and type of issues considered in agent services [16, 39], agent
attitude [17], one-sided or two-sided commitment [11, 40],
bilateral [39] or multilateral [6] negotiation.

The standout research and relative optimization techniques and
methods of negotiation agent models are summarized in table 1
below.

Figure 2. Utility elicitation algorithm[32]

TABLE I.

Popular Techniques and methods for optimizing agent negotiation systems

Optimization IMethods Te chnique s

Exempular description

Stand out Researches

Gathering information

Referral Systems

Trust

History

Asking information from neighbors
Based on trust ^vorthy of opponent
based on recorded experienced

Ebadiet.al. (2008)

Griffiths (2005),Gujral et.aL(2007)

Choo et.aL(2009), Guo et.aL(2003)

A daptability

adaptive learning techniques
Flexible models
Dynamic models

learnin adaptive factor during trading
overcoming pre-programmed tactics
Dynamic methods and programming

Raymond (2009),]VIagda et.al.(2009)
Zang et.al(2007),chung-wei(2008)
Raymond (2009)

Inte lie ctuality

IVlarchine learning techniques:

Data mining techniques
Others

Genetic Algorithm

Bayesian Learning

Reinforcement Learning

Fuzzy Logic

Evolutionary Learning

Apriory algo/Classification(C5)Algo

Simulated Annealing

Choo et.aL(2009), IVIagda et.al.(2009)

Duong e.aL(2004),Choo et.al(2009)

Sen et.aL(2007)

Cheng et.al.(2005)

Raymond(2009)

Isabel et.al.(2008)

Isabel et.aL(2008)

Applied Strategies

Time dependant tactics

imitative tactics

Resource dependant tactics

B oulware/Conc eder/Line ar

Relative tit-for -tat/Random absolut tit-

for-tat/Averaged tit-for-tat

Dynamic deadline /Re source estimation

lyad et.aL(2002),Doung et.aL(2004)

Isabel et.al.(2008)
not common

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

V. APPLICATIONS

After specifying the domain of classification and
optimization of negotiation agents, some outstanding works
carried out during recent years will be reviewed. Following the
new generation of the models we will track the progressing
flow of agent generations. In addition, the whole negotiation
systems proposed in the models will be analyzed to elaborate
on advantages and drawbacks of these systems. These systems
are analyzed in terms of desirable negotiation protocol,
negotiation strategy, agent characteristics and negotiation
setting.

1) Desirable negotiation protocol represents the rules that
govern the interaction between negotiators [22]. According to
Nguyen and Jennings. [41], desirable properties for a
negotiation protocol include pareto efficiency and Guarantee of
success.

2) Negotiation strategy is the arrangement of a series of
actions that the agents plan to take through the negotiation.
Negotiation tactics specify whether it is time dependent,
resource dependant or imitative. In some cases, negotiation
tactics are a combination of aforementioned tactics proposed by
models.

3) Agent characteristics specify agent's knowledge and
experience, learning and adaptation capabilities.

4) Negotiation setting deals with factors which are related
to problem domain. It includes number of negotiation issues
and number of parties involved.

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

Vol 8, No. 3, 2010
desperate, patient and optimal patient. Optimal patient helps

agents to assist agent's autonomous behavior in dynamic

environments. In this case, preferences of agents will be

marked by weighting factor which represents the degree of

importance of every issue.

Weighting factor was firstly introduced by Griffiths [37].
Later on, this concept under the term of Multi-dimensionality
[38] was used by many agent researchers [43, 44] to evaluate
their standard measurements and finding the more appropriate
deal or opponent. However, ITA's strategies help agent to gain
higher performance but still the act of agent is bounded by
premier choice of strategies before each round of negotiation.
This will reduce the adaptability of negotiation agent.

)ynamic Environme

/ Buyer agent )

Sub-buyer 1

Seller 1

/ \ /

/ V

Coordinator gl^

\ / \

Sub-buyer 2

Seller 2

\ / \

Sub-buyer n

Seller n

The analysis will deal with the evolution that the primary
Kasbah model [7] underwent from 1996 to 2009, leading to
GAMA model (table2)

Kasbah model [7] is a simple one-to-one negotiation
framework. Its application is on e-commerce where the agent
technology will meet the web-based systems and try to
overcome the need of online trading by applying some
autonomy on trading. Kasbah model is single issue and
considers price as the most important issue in negotiation.
Therefore, agents are searching to make a deal with appropriate
price even before they meet their deadline. Lack of adaptation
and intellectuality is obvious draw-back of Kasbah' s
negotiation agents which is overcome by defining new versions
of Intelligent Trading Agents such as ITA and e-CN.

IT A [9] is a one-to-many negotiation framework which is
an improved version of negotiation systems in terms of number
of issues considered in negotiation and in terms of
communicating as it follows bilateral negotiation. In bilateral
negotiation, agents have the ability to send offer and counter-
offer in both ways. ITA presents new system architecture
represented in Fig. 3.

ITA buyer agent includes two components, namely
coordinator and sub-buyers. Buyer agent will establish several
one-to-one negotiations between sub-buyers of buyer agent and
sellers. In this work, agent preferences are represented as a
constraint satisfaction problem as described by Vipin [42].
Moreover, this model proposes different sort of strategies like

Figure 3. System architecture adapted from ITA system [9]

The new version of ITA called e-CN was proposed by
Nguen and Jennings [10]. This method uses several number of
agent services' issues considered in negotiation. Also, a factor
is introduced as "probability of agent distribution" which
represents the probability of allocating types of agent in the
environment. Nguyen [10] believes that agents can be divided
by their behavior into two groups of agents called Conceder
and Non-conceder. Conceder stands for agents who are willing
to concede in order to reach a deal while Non-conceder stands
for agents who are not willing to sell, otherwise there is a
special amount of benefit for them, so they act in a tough
manner.

In e-CN model, agents choose their strategies based on a
method of predicting the expected utility of chosen strategy
considering the current situation. This Expected utility will be
evaluated according to 3 important probability factors:
probability of agent's distribution, possibility of reaching
agreement and average utility value if agent reaches the
agreement.

The disadvantage of proposed concession strategies in e-
CN and ITA is that in every round of offering new proposal,
agents are conceding while there is a possibility to find a
mutually acceptable proposal with the same utility level but
different values of the issues. Future research should be done

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

to investigate a sort of similarity function that reveals the
negotiable issues with higher importance. Finding important
negotiable issues assists agents to reach to an agreement using
the slightest possible amendments. We believe this will result
in higher optimization and also faster deals in terms of time
consuming. Aforementioned future developments will result in
decreasing negotiation cost.

Although these models support bilateral multiple
concurrent negotiation, there is still much more need to assist
the agent decision making methods. Helping agent to predict
its opponent next action or finding the opponent preferred
issues is another future work. As we discussed in section IV
this objective could be achieved by enhancing the
intellectuality of agents. Referring to table 1 there are many
possible techniques to investigate as future studies. An
appropriate learning technique could lead this model to find the
pareto optimal solution.

Cheng [23] proposed a heuristic, multiple issue, one-to-
many model for negotiations in a third-party-driven e-market
place. These negotiation agents employ trade-off tactics using
fuzzy inference systems to generate new offer in each round.
Trade-off tactics are navigated from time-dependent and
imitative tactics.

Although, Cheng's proposed model is pareto-efficient and
highly adaptive revealing the importance level of issues to the
other agents, it violates the privacy of information. These sorts
of assumptions are highly inappropriate since they are hardly
acceptable in the real world environments such as e-markets.

A negotiation meta strategy for one-to-one bilateral
negotiation was proposed by Ros and Sierra[16]. Meta strategy
is a combinatorial sequence of concession and trade-off tactics
which will try to keep the aspiration level, otherwise there is no
possible offer by the current aspiration level. Combining
tactics allows agents to outperform better in different situation
which fulfills the adaptive capability of agents. In addition,
detection of opponent agent preferences helps agents to
propose mutually acceptable offers. As a consequence, the
system's final success rate increases.

Although this model is placed in a satisfactory level of
adaptabihty, it is recommended that further research be
undertaken in order to learn the opponent agent's type. Finding
the correct classification for type of agent could increase the
chance of reaching agreement.

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

Vol 8, No. 3, 2010
Raymond [22] performed a similar series of experiments in

terms of models characteristics. Raymond negotiation agents
are enhanced as they are promising in supporting real-world e-
market places environment. This application sustains multi-
party, multi-issue, many-to-many negotiation which are based
on parallel and distributed decision making model. Moreover,
Raymond [22] introduced his novel genetic algorithm in this
experiment. The final result of the experiment showed that an
evolutionary negotiation agent guarantees pareto optimal
solutions underneath dynamic negotiation situation, for
example in the incidence of time limitations.

However, as we mentioned in section IV D, embedding a
sort of strategies could enhance the success rate of the two
aforementioned negotiation systems.

Magda [45] introduced an agent mediated shopping system
called Genetic Algorithm driven Multi- Agents (GAM A).
GAMA is a multi issue bilateral negotiation system enhanced
by learning ability. GAMA studies the effect of participating
opponent agent's preferences in decision making of agents. In
order to do that, one of the offers from the opponent agent's
previous offers (or list of offers) is chosen as one of the parents
and the other parent is chosen from agent's own preferable
proposals. Then, the mutated offspring is generated. The new
generated offspring is a potential satisfactory offer as it is
mutually acceptable for both agents. Experimental results
demonstrated that GAMA achieved to a higher satisfaction rate
while reaching to the higher numbers of the deal in comparison
with traditional GA methods.

One of the advantages of this work considering these parent
selections is increasing the adaptability of the system, since
every change of opponent agent's preferences effects the
decision making of agents. In addition, proposing mutually
admissible offer causes to reach an agreement in fewer rounds
of negotiation, thus reducing the cost of negotiation. In future
investigations, it might be possible to experiment this model
under qualitative issues as an alternative of quantitative issues.
This could result in higher optimization and also faster deals in
terms of time consumption.

In order to have a better understanding on the overview of
applications discussed, table2 illustrates the general
characteristics of the applications discussed.

Choo et al [19] conducted a research on one-to-many
bilateral negotiation with multiple issues (quantitative issues).
The system architecture of the study was based on ITA's
system architecture. They investigated two different machine
learning approaches genetic algorithm and Bayesian learning
called GA improved-ITA and Bayesian improved-ITA. The
result obtained from the final analysis showed that GA-
improved-ITA outperforms, Bayesian-improved-ITA in
maximizing the joint utility and negotiation payoff at the same
time as it increases the negotiation cost.

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

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

Vol. 8, No. 3, 2010

Table 2. A COMPREHENSIVE analysis over system application progress during the last thirteen years.

Negotiation system characteristics

Kasbah(1996)

ITA(2001)

e-CN(2004)

Cheng et.al.(2005)Ros et.al.(2006)

Choo et.al.(2009)

Raymond(2009)

GAMA(2009)

Cardinality of Negotiation Domain

single-issue

multiple-issue

Cardinality of Communication

unknown

bilateral

unknown

bilateral

mutilateral

bilateral

Number of agents involved in negotiaioi

one-to-one

one-to-many

one-to-one

one-to-many

many-to-many

one-to-many

Qualitative negotiation value of issues

Quantitative negotiation value of is sues

Privacy of model

Privacy of information

not-private, see Des

Negotiation tactics includes:

TIme/REsource dependant or IMitative

Time dependant

mix(TMM)

Trade-off(TI-FlM)

Mix(TMM)

Time dependant

Intellectuality

Bayesian learning

Fuzzy inference

Genetic Algorithm,
Bayesian learning

Genetic
Algorithm

Pare to efficiency

Preto Optimal

Guarantee success

Type of agents involved

unknown

conceder
non-conceder

NegoTo, Random,
aLtemate,
TOAgent,
NegoAgent

unknown

Adaptive ability

non-adaptive

semi-adaptive

adaptive

non-adaptive

adaptive

This table shows that embedding strategies in negotiation
agent models can increase the final system outcome. This
enhancement could be in terms of increasing adaptability or
assuring the guarantee of success. As illustrated, most of the
models empowered by machine learning techniques are fully
adaptive. As discussed in section IV A, adaptability is an
important characteristic for negotiation system embedded in
open environment. Also, as we can see in e-CN and Cheng et
al models [10, 23] an accurate classification on agent's type is
driving the negotiation model to a desirable level of
adaptabihty.

In order to reach to the pareto optimal result, agents most be
equipped with an effective learning method or a suitable
strategy. These cases show that learning ability helps agents to
predict opponent's characteristics (e.g. preferences, reservation
value, attitude and type) accurately. Therefore, by choosing
the best possible strategy (action) we can reach to a pareto
optimal result.

Every negotiation mechanism is desirable to meet some
important requirements. These include pareto efficiency and
guarantee of success. The power of applying suitable strategy
is revealed by assuring the guarantee of success only by
employing the appropriate strategy.

We believe tables 1 and 2 provide a good resource for
future developments. Since table 2 highlights the shortcomings
and loopholes of the above-mentioned models (represented by
"-" or "non"), developers can easily investigate future studies
in order to overcome these gaps by the help of the optimization
methods introduced in table 1.

As this study shows, artificial intelligence is an important
characteristic of multi agent systems which has been used as a
popular enhancing technique to optimize the final outcome of
many negotiation agent based systems. As Jennings [46]
admits" undoubtedly the main contributor to the field of
autonomous agents is artificial intelligence".

Artificial intelligence techniques such as GA [6, 19, 32] ,
fuzzy logic [23, 25], simulated annealing [32] ,neural network
[47, 48] and re-enforcement learning [30] have been used
broadly to improve agent's intellectuality in recent years.
According to our review, GA is the most popular learning
technique among others as usually models that employ GA in
negotiation systems reach to a higher final utility than other
techniques (e.g. [22, 49]) as shown in Fig. 4.

Figure 4. Comparison of final utility value of different machine learning
methods

This is due to the pattern of common problems defined in
negotiation which match GA technique characteristics, since, in
typical negotiation problems, we face a big space of feasible
solution and the goal is to find the best solution. Furthermore,
specifying the fitness function for population ranking is a
straightforward task as generally it equals to utility function for
evaluating offers.

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

VI. CONCLUSION

In the last few years we have witnessed an incredible
growth of multi agent systems on e-commerce. One of the
most important driving forces behind MAS research and
development is the internet. By increasing the human demand
on online trading, negotiation has become one of the most
important topics in MAS. In this area, many researchers tried
to improve the performance of the negotiation agent's model.
This is achieved by using adaptive methods, employing
learning abilities, applying strategic reactions and gathering
information. To sum up, negotiation agents can act more
efficiently when they are empowered with effective methods
for gathering information which assists the agents in employing
strategic actions to reach their goals. In addition, agents must
be adaptive by changing their attitudes and learning their
opponents' characteristics and preferences to improve the
overall performance of the system.

In order to improve agent negotiation systems, we need to
understand the dimensions and range of options in these areas.
To set up the foundation, we have developed a classification
scheme which is specially aimed at negotiation agent systems
on e-commerce. Negotiation agents can be categorized into
different groups with respect to (1) number of agents involved
in negotiation, (2) type of agents involved in negotiation, and
(3) according to negotiation agent attitude involve in
negotiation.

This classification system was demonstrated on an assorted
range of outstanding negotiation model and the outcome is
summarized in table 2. The purpose of this classification was
to present a complete and systematic source to objectively
compare and contrast different negotiation models. Such a
classification method is essential for developers as it supplies a
resource of differentiating competing alternatives for the area
of negotiation agent's models to exploit.

VII.

REFERENCES

[1] M. Wooldridge and N. R. Jennings, "Intelligent Agents: Theory and

Practice, " Knowledge Engineering Review, vol. 10, pp. 115-152, October

1995.

[2] C. Castelfranchi, "Guarantees for autonomy in cognitive agent

architecture, " Proceedings of the workshop on agent theories, architectures,

and languages on InteUigent agents, vol. 890, pp. 56-70, 1995

[3] S. P. K. Michael R. Genesereth "Software Agents, " Communications of

the ACM, vol.37, pp. 48-53, 1994

[4] K. P. Sycara, "Multiagent Systems, " AI Magazine vol. 19(2), pp. 79-92,

1998.

[5] M. Woolridge and M. J. Wooldridge, An introduction to multi agent

systems,10 ed., John Wiley & Sons, Inc. New York, NY, USA 2001,

[6] R. Y. K. Lau, "Adaptive negotiation agents for e-business, "Proceedings

of the 7th international conference on Electronic commerce, pp. 2005

[7] C. Anthony and M. Pattie, "Kasbah: An agent marketplace for buying

and seUing goods, " In Proceedings of the first international Conference on the

Practical Apphcation of InteUigent Agents and Multi -Agent Technology, pp.

1996

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

Vol 8, No. 3, 2010
[8] V. B. Ryszard Kowalczyk "On Constraint-Based Reasoning in e-

Negotiation Agents, " Lecture Notes In Computer Science, vol. 2003, pp.31-
46, 2001

[9] I. Rahwan, R. Kowalczyk and H. H. Pham, "Intelligent agents for
automated one-to-many e-commerce negotiation, "Proceedings of the twenty-
fifth Australasian conference on Computer science , vol.4, pp. 197-203, 2002
[10] T. D. Nguyen and N. R. Jennings, "Coordinating Multiple Concurrent
Negotiations, "Proceedings of the Third International Joint Conference on
Autonomous Agents and Multiagent Systems , vol.3, pp. 1062- 1069, 2004
[11] J. Dang and M. N. Huhns, "Concurrent Multiple-Issue Negotiation for
Internet-Based Services, " IEEE Internet Computing, vol. 10, pp. 42-49, 2006
[12] D. Jiangbo and H. M. N., "Coalition deal negotiation for services, "
Rational, Robust, and Secure Negotiation Mechanisms in Multi-Agent
Systems, 2005, pp. 67-81, 2005

[13] J. Dang, J. Huang and M. N. Huhns, "Workflow coordination for
service-oriented multiagent systems, "Proceedings of the 6th international
joint conference on Autonomous agents and multiagent systems, pp. 2007
[14] W. Danny, S. Kurt, H. Tom and G. Olivier, "Towards Adaptive Role
Selection for Behavior-Based Agents, "Adaptive Agents and Multi-Agent
Systems III, pp. 295-312, 2005

[15] I. Erete, E. Ferguson and S. Sen, "Learning task-specific trust decisions,
"Proceedings of the 7th international joint conference on Autonomous agents
and multiagent systems , vol.3 pp. 2008

[16] R. Ros and C. Sierra, "A negotiation meta strategy combining trade-off
and concession moves, " Autonomous Agents and Multi-Agent Systems, vol.
12, pp. 163-181,2006

[17] J. Ahn, D. DeAngelis and S. Barber, "Attitude Driven Team Formation
using Multi-Dimensional Trust, "Proceedings of the 2007 lEEEAVIC/ACM
International Conference on Intelligent Agent Technology, pp. 229-235,2007
[18] E. Toktam, P. Maryam and P. Martin, "Partner Selection Mechanisms
for Agent Cooperation, " Web InteUigence and InteUigent Agent Technology,
2008. WMAT '08. lEEE/WIC/ACM International Conference on, vol. 3, pp.
554-557, 2008

[19] M. N. S. a. M. H. S. S.C. Ng, "Machine learning approach in optimizing
negotiation agents for e-commerce, " Information Technology Journal, vol. 8,
pp. 801-810,2009

[20] L. Alessio, W. Michael and J. R. Nicholas, "A Classification Scheme for
Negotiation in Electronic Commerce, " Group Decision and Negotiation, vol.
12, pp. 31-56,2003

[21] P. Faratin, Carles Sierra and N. R. Jennings, "Negotiation Decision
Functions for Automated Agents, " Elsevier Sience, vol.24 pp. 159-189, 1997
[22] R. Y. K. Lau, "An Evolutionary Approach for InteUigent Negotiation
Agents in e-Marketplaces, " Intel. Agents in the Evol. of Web & Appl, vol.
167, pp. 279-301,2009

[23] C.-B. Cheng, C.-C. h. Chan and K.-C. Lin, "InteUigent agents for e-
marketplace: Negotiation with issue trade-offs by fuzzy inference systems, "
Decis. Support Syst., vol. 42, pp. 626-638, 2005

[24] M. A. Lopez-Carmona, I. Marsa-Maestre, J. R. Velasco and E. d. 1. Hoz,
"Using Clustering Techniques to Improve Fuzzy Constraint Based Automated
Purchase Negotiations, " Springer, Advances in Agent-Based Complex
Automated Negotiations, pp. 89-117, 2009

[25] J. Richter, "Multistage Fuzzy Decision Making in Bilateral Agent
Negotiation, " 3sd PHD symposium, pp. 71-73, 2010

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

[26] I. Pra, M. Jo, o. Viamonte, Z. Vale and C. Ramos, "Agent-based
simulation of electronic marketplaces with decision support, "Proceedings of
the 2008 ACM symposium on Applied computing, pp. 2008
[27] Z. Junyan, T. Jiang and D. Gang, "Agent-based multi-factors adaptive
negotiation in E-Commerce, " Grey Systems and InteUigent Services, 2007.
GSIS 2007. IEEE International Conference on, pp. 1528-1532, 2007
[28] C.-W. Hang, Y. Wang and M. P. Singh, "An adaptive probabihstic trust
model and its evaluation, "Proceedings of the 7th international joint
conference on Autonomous agents and multiagent systems, vol.3, pp. 1485-
1488, 2008

[29] L.-k. Soh and J. Luo, "Combining individual and cooperative learning
for multi-agent negotiations, "Proceedings of the second international joint
conference on Autonomous agents and multiagent systems, pp.1 122-1 123
,2003

[30] S. Sen, A. Gursel and S. Airiau, "Learning to identify beneficial
partners, " AAMAS07 '07 Honolulu, HI USA, 2007

[31] S. Saha, A. Biswas and S. Sen, "Modeling opponet decision in repeated
on-shot negotiations, " AAMAS'05, ACM, vol. july,pp. 25-29.2005
[32] G. Yutao, M. Jorg and W. Christof, "Learning User Preferences for
Multi-attribute Negotiation: An Evolutionary Approach, "Multi-Agent
Systems and Applications III, pp. 1067-1067, 2003

[33] n. s. choo, machine learning approach for optimizing negotiation agents,
UniversityPutraMalaysia, pp.167, 2007,

[34] T. D. Nguyen and N. R. Jennings., " A heuristic model for concurrent bi-
lateral negotiations in incomplete information settings, " Proceedings of the
Eighteenth International Joint Conference on Artificial Intelligence 2003, vol.
pp. 1467-1469, 2003

[35] T. Ebadi, M. Purvis and M. Purvis, "Finding Interaction Partners Using
Attitude-Based Decision Strategies, "Proceedings of the 2008 Ninth
International Conference on Parallel and Distributed Computing, Apphcations
and Technologies, vol. pp. 2008

[36] S. Kraus, "Beliefs, time and incomplete information in multiple
encounter negotiations among autonomous agents, " Annals of Mathematics
and Artificial InteUigence vol. 20, pp. 1-4, 1997

[37] N. Griffiths, "Task delegation using experience-based multi-dimensional
trust, "Proceedings of the fourth international joint conference on Autonomous
agents and multiagent systems,vol.ll, pp. 489-496, 2005
[38] N. Gujral, D. DeAngehs, K. K. FuUam and K. S. Barber, "Modeling
Multi-Dimensional Trust, " AAAI07_Materials, vol. 6, pp.35-41, 2007
[39] S. S.Fatima, M. Wooldridge and N. R. Jennings, "Approximate and
online Multi-issue negotiation, " AAMAS'07, pp. 947-954, 2007

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

Vol 8, No. 3, 2010
[40] T. D. Nguyena and N. R. Jennings, "Managing commitments in multiple

concurrent negotiations, " Electronic Commerce Research and Apphcations,

vol. 4, pp. 362-376, 2006

[41] P. F. NR Jennings, AR>Lomuscio , S Parsons, M. Wooldridg, C. Sierra,

"Automated Negotiation: prospects, method and challenges, " Group Decision

and Negotiation, vol.10 pp.199-215 2001

[42] V. Kumar, "Algorithms for constraint-satisfaction problems: a survey, "

AI Mag., vol. 13, pp. 32-44, 1992

[43] D. J. Kim, Y. I. Song, S. B. Braynov and H. R. Rao, "A

multidimensional trust formation model in B-to-C e-commerce: a conceptual

framework and content analyses of academia/practitioner perspectives, "

Decis. Support Syst., vol. 40, pp. 143-165, 2005

[44] J. Ahn, X. Sui, D. DeAngehs and K. S. Barber, "Identifying beneficial

teammates using multi-dimensional trust, "Proceedings of the 7th international

joint conference on Autonomous agents and multiagent systems, vol. 3pp.

2008

[45] M. B. Fayek, I. A. Talkhan and K. S. EL-Masry, "GAMA(Genetic

Algorithm driven Multi-Agents) for E-commerce Integrative Negotiation, vol.

pp. 1845-1846,2009

[46] N. R.Jennings, K. Sycara and M. Wooldridge, "A Roadmap of Agent

Research and Development, " Autonomous Agents and Multi-Agent Systems,

vol. 1, pp. 7-38, 1998

[47] V.-W. Soo and C.-A. Hung, "On-Line incremental learning in bilateral

multi-Issue negotiation, " AAMAS'02, vol. pp. July 15.2002

[48] S. Park and S.-B. Yang, "An Efficient Automated Negotiation System

Using Multi-attributes in the Online Environment, " lecture Notes In

Computer Science, vol. 3140, pp. 544-557, 2004

[49] P. M. C. Samuel, L. Jiming and S.-P. Chan, "Evolutionary Negotiation

in Agent-Mediated Commerce, " lecture Notes In Computer Science, vol.

2252, pp. 224-234,2001

AUTHORS PROFILE

Sahar Ebadi is currently doing her Master degree in
Faculty of Computer Science and Information
Technology, UPM. Sahar has received her B.Sc in
software computer engineering field in 2006 from
Iran Azad University. Her research interest includes
Artificial Intelligence and Autonomous Agents and
Data mining.

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

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

Vol 8, No. 3, 2010

Customized Digital Road Map Building using

Floating Car GPS Data

G. Rajendran

Assistant Professor of Computer Science,

Thiruvalluvar Government Arts College,

Rasipuram-6 37401, Tamilnadu, India

guru. rajendran @ yahoo.com

Dr. M. Arthanari

Director,

Bharathidasan School of Computer Applications,

Ellispettai-638116, Tamilnadu, India

arthanarims vc @ gmail . com

M. Sivakumar

Doctoral Research Scholar,

Anna University, Coimbatore, Tamilnadu, India

sivala @ gmail . com

Abstract — The vehicle tracking, navigation and road guidance
appHcations are becoming more popular but the presently
available digital maps are not suitable for many such
applications. Among the drawbacks are the insufficient accuracy
of road geometry and the delayed time in loading the unwanted
data. Most of the commercial applications in vehicle tracking
require digital maps which have only roads and places of interest
whereas the currently available maps show all available roads
and places. A simplified map building process to construct
customized high-precision digital maps from floating car data
obtained from Global Positioning System (GPS) receivers is
presented in this paper. The data collected from the GPS receiver
fixed in a moving car are used to construct the customized digital
road maps. The approach consists of six successive steps:
Collecting floating car data (FCD) for desired road segments in a
log file; refining the log file; constructing the road segments using
the data present in refined log file; merging the segments which
has negligible slope; refining the road intersections; and labeling
the points of interest. The quality of outcome of the map making
process is demonstrated by experimental results and the results
indicate that customized road maps of required routes with good
accuracy can be built with the help of the proposed map making
process.

Keywords- digital map; global positioning system; floating car
data; road network

I. Introduction

Map is a total or partial depiction of the structure of the
earth (or sky) on a plane, such that each point on the map
corresponds to an actual point on the earth (or in the sky).
Digital maps are used to produce this depiction in electronic
form. The digital maps are usually represented as graphs,
where the nodes represent intersections and the edges are
unbroken road segments that connect the intersections [1].
Each segment has a unique identifier and additional associated
attributes, such as the number of shape points that approximate
its geometry roughly, the road type (e.g., national highway,
state highway, city road, street etc.), name, speed information,
and the like. Digital map building is a process that utilizes the
information supplied by external equipments in terms of

physical features of an environment [2, 3, 4]. This information
is taken from different positions along the path followed by a
moving object. Once the map is obtained, it can be used to
improve the quality of the paths, to locate a target, to help
object recognition, to define expectations in the trajectory or to
replay the travelled path of a moving object.

Digital maps of required roads with good accuracy are
needed in a number of commercial applications. But the
presently available digital maps are fully populated with dense
roads and other information, most of which are irrelevant to the
requirement. Because of the presence of these unwanted data,
the loading time of the map in computer memory is also more
which results in slower execution of the application. Hence the
need for building customized digital road maps is essential and
such a map building process will also eliminate the expenses
involved in buying digital maps. This paper presents a
customized digital road map building process which can be
used to construct high quality maps of desired roads and
locations. The source code for the intermediate processing
steps is written in Matlab 7.6.

Map building process has already been discussed by some
of the researchers, but with limitations like complexity in map
building and insufficient accuracy. These limitations have been
addressed in this work. The remainder of this paper is
organised as follows. Section 2 of this paper describes road
network model and the collection of floating car GPS data.
Related work in this area is discussed in Section 3. In Section
4, the map making process is discussed. The experimental
results are dealt in Section 5. The work is concluded and the
possible improvements are discussed in Section 6.

II. The Road Network Model and the Roating Car
Data

A. The Road Network Model

The road network data are the basis for vehicle tracking and
related applications. The road network model is represented
with two-dimensional line segments.

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

The road network model [5] is represented by the equation

Rn=(N,S),

N={n I n=(x, y), x, y G Coordinates},
S= {s I s = <m, n>, m, n e N},

wherein Rn represents the road network, N represents the node
set that indicates the coordinate point set of the road in the road
network which is a pair of longitude and latitude (x,y), S
represents the road segment set of the road network which is
composed by the sequence <m, n>. S represents one directional
road that has the beginning node m=begin(s) and the
termination node n=end(s). Fig. 1 shows a road network with
nodes, segments and an intersection point.

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

Vol 8, No. 3, 2010
The Roating Car (Probe Car) technique is one of the key
technologies adopted by the ITS to get the traffic information
in recent years [9]. Its basic principle is to periodically record
the location, direction, date, time and speed information of the
traveling vehicle from a moving vehicle with the data of the
Global Position System (GPS) as shown in Fig 2. The
information can be processed by the related computing model
and algorithm so that the floating car data can be associated
with the city road in real time [10]. This data can also be used
as a source of data for creating research and commercial
applications on vehicle tracking and road guidance systems.

Ruad Intellection

node 1

segment 1 segment 2

node 2

nodes

node 4
Segments segment 4

Figure 1. Basic elements of the road network model.

If there is a road segment sequence <Si, Sj,..., Sk> in the
network Rn=(N, S), the termination node of every road segment
is the beginning node of the next road segment, this sequence is
called one directional road of the road network.

B. Floating Car Data

Nowadays, the main research focus in the community of
Intelligent Transport Systems (ITS) is how to acquire real-time
and dynamic transportation information. This information can
be applied in the transportation area like vehicle tracking,
navigation, road guidance and so on. GPS is one of the system
which is used to provide real time information on moving
objects. It is a Satellite Navigation System which is funded and
controlled by the U. S. Department of Defense [6, 7]. The
system consists of three segments viz., satellites that transmit
the position information, the ground stations that are used to
control the satellites and update the information, and finally
there is the GPS receiver that computes its location anywhere
in the world based on information it gets from the satellites [8].

GPS Satellites

Figure 2. A moving vehicle (Floating car) fixed with a GPS receiver

III. Related Work

The history of map making process starts with the
Egyptians who for the first time constructed a map for revenue
collection three thousand years ago. The digital map building is
a new concept developed after the revolution in Information
Technology. Though some work has been done in this area, a
number of map building techniques are being proposed to suit
the emerging requirements.

Y.L. Ip et al., have presented a technique for on-line
segment-based map building in an unknown indoor
environment from sonar sensor observations [4]. In thefr
approach, the environment data is first obtained from a rotating
sonar sensor, analyzed and fed to the Enhanced Adaptive
Fuzzy Clustering (EAFC) module to extract the line segments
within the workplace of robot. The basic motive is to use full
data set to obtain an initial approximation cluster centers via
Fuzzy c-mean. This initial approximation helps reducing the
number of iterations requfred for Adaptive Fuzzy Clustering
(AFC). This approach is somewhat similar to Fast Fuzzy
Clustering (EEC) [11] which is a strategy to speed up the Fuzzy
c-mean (FCM). In order to facilitate the map building, the
workplace of the robot is divided into squared areas as cells in
order to extract the line segments. This mechanism reduces the
computation time when extracting the line segments within the
world frame. EAFC uses the Noise Clustering (NC) technique
proposed in [12] to extract the line segments. EAFC also uses
adaptive fuzzy clustering algorithm [13] and fast fuzzy
clustering [11]. These algorithms are combined into a single
algorithm with enhanced characteristics such as improvement
in the computational burden and reduction of the effect of noisy
data in fuzzy clustering algorithm. Besides, the authors have
proposed a compatible line segment merging technique to
combine the similar line segments to a single long line segment
as a mechanism to reduce the number of segments in the world
model and further improve the quality of the map. This
technique is applicable for constructing maps of indoor
environment related to robotic applications.

Stefan Schroedl et al., have contributed to map-building by
introducing a system that generates digital road maps that are
significantly precise and contain descriptions of lane structure,
including the number of lanes and thefr locations, along with
detailed intersection structure[l]. The authors combine a large
quantity of possibly noisy data from GPS for a fleet of vehicles,
as opposed to a small number of highly accurate points
obtained from surveying methods. It is assumed that the input
probe data is obtained from vehicles that go about their usual

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

business unrelated to the task of map construction, possibly
generated for other applications based on positioning systems.
The work of authors include the development of a spatial
clustering algorithm for inferring the connectivity structure of
the map from scratch, the development of a lane clustering
algorithms that can handle lane splits and merges and forming
an approach to inferring detailed intersection models. This
system requires data from hundreds of vehicles already
connected to tracking systems for constructing a single segment
of the road.

Thus a few number of digital map building techniques are
available but they suffer from high complexity of map building
process, requirement of more data and dependence on technical
skills of the person who is working with map building. The
map building process proposed in this work addresses these
problems. The process discussed here is a very simple one and
even a novice user without technical skills can easily create
route maps according to his requirements.

IV. Map Building Process

A. Collecting Floating Car GPS data in a log file

GPS receiver communication is defined with National
Marine Electronics Association (NMEA) specification. The
NMEA has developed a specification that defines the interface
between various pieces of marine electronic equipments. The
NMEA standard permits marine electronics to send information
to computers and to other marine equipments [14] in
predefined formats. Most computer programs that provide real
time position information recognize data that are in NMEA
format which includes the complete latitude, longitude,
velocity and time computed by the GPS receiver. In NMEA
specification system, the collected GPS data is converted into a
line of text, called a sentence, which is totally self contained
and independent from other sentences. The commas act as
terminators for the sentences and the programs that read the
data should only use the commas to determine the end of a data
item.

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

Vol 8, No. 3, 2010
receiver module was fixed in a moving car and the generated
NMEA sentences are stored in a log file in a laptop kept in the
moving car. This GPS receiver generates $GPGGA, $GPGSA,
$GPRMC, $GPVTG and $GPGSV sentences at regular time
interval of one second. A list of NMEA sentences produced by
the GPS receiver and stored in a log file when travelled in a
road is given in Fig. 3.

5PGGA,112537.613,1123.5703.N.07739.6095,E, 1,10,00.8, 191. 4, M.-92.6,M,.MF

5PGSA,A,3,03,06,07,1 1,13.16, 19,20,23,32, ,,1.8,0.8,1. 6*39

5PRMC,1 12537.613.A,1123.5703.N.07739.6095,E, 12.58, 193.19,240410„,A*57

5P\/TG,193.19.T,., 12. 58,N, 23.29, KA*77

5PGGA,1 12538.61 3.1 123.5664.N.07739.6085,E, 1,10,00.8, 191. 4, M.-92.6.M,.*41

5PRMC, 11 2538.61 3.A,1123.5664.N.07739.6085,E,1 4.67, 193.98,240410„,A^5A

;P\/TG,193.9S.T,.,14.67,N,27.18,K.A*72

5PGGA,1 12539.613.1 123.5625.N.07739.6074,E, 1,10,00.8, 191. 5, M.-92.6.M,.*4A

5PGSV,3,1, 10,3,43,1 18.45,6,35.099,45,7, 16,285,34.1 1,12,204,4r4E

5PGSV,3,2,10,13,20,332,38,16,31,031,40.19,30,151,44.20,62.248,42V3

3PGSV,3,3,10,23,40,350,42,32,59,185,46 *7E

5PRMC,112539.613.A,1123.5625.N.07739.6074,E,14.41,195.11,240410,„A*53
5PVTG,195.11.T,., 14.41, N,26.70,K.A*7E

5PGGA,112540.613.1123.5584.N.07739.6062,E,1, 10,00.8,191 .7,M.-92.6,M..*49
5PGSA,A,3,03,06,07,11,13.16,19,20,23,32„,1.8,0.8,1.6*39
5PRMC,112540.613.A,1123.5584.N.07739.6062,E,15.43,196.10,240410,„A*53
5PVTG,196.10.T,.,15.43.N.28.57,K.A*74

3PGGA,112541.613,1123.5539.N.07739.6047,E,1, 10,00.8,191 .8,M.-92.6,M,.M6
5PRMC,1 12541 .613.A,1123.5539.N.07739.6047,E,17.12,197.57,240410,„A*57
5PVTG.197.57,T„.17.12.M,31.70,K.A*7D

Figure 3. Log file of floating car GPS data with $GPGG A, $GPGSA,
$GPRMC, $GPVTG and $GPGSV sentences

In order to collect the floating car GPS data, Wonde-X
series GPS receiver (ZX4125) module was used. This GPS

B. Refining the log file to get $GPRMC sentences

The log file contains a number of different types of
sentences but the recommended minimum sentence C,
$GPRMC, provides the essential GPS PVT (Position, Velocity
and Time) data. This data is used to locate moving objects in
terms of latitude and longitude. The $GPRMC data format is
given in Table I. The moving object, if attached with a GPS
receiver, can be located with the help of this NMEA sentence.

TABLE I.

$ G P R M C DATA FORMAT

Data Item

Format

Description

Message ID

$GPRMC

RMC protocol header.

UTC Time
(Coordinated
Universal Time)

hhmmss.sss

Fix time to 1ms accuracy.

Status

Char

A Data Valid.
V Data invalid.

Latitude

Float

Degrees * 100 + minutes.

N/S Indicator

Char

N=north or S= south.

Longitude

Float

Degrees * 100 + minutes.

EAV Indicator

Char

E=EastorW=West.

Speed over Ground

Float

Speed Over Ground in knots

Course over Ground

Float

Course Over Ground in
Degrees

Date

ddmmyy

Current Date

Magnetic Variation

Blank

Not Used

EAV Indicator

Blank

Not Used

Mode

Char

A Autonomous

Checksum

*xx

2 Digits

Message
Terminator

ASCII 13, ASCII 10

An example of $GPRMC NMEA sentence is given below:

$GPRMC, 120642.206, A, 1118.4253,N,07742.4325,E,3 1.6,
317.52,140510„,A*62

Where

$GPRMC : Recommended Minimum sentence C

120642.206 : Fix taken at 12:06:42.206 UTC

A : Status A=active or V=Void.

1118.4253,N : Latitude 11 deg 18.4253' N
07742.4325,E : Longitude 77 deg 42.4325' E

31.6 : Speed over the ground in knots

3 17.52 : Course over the ground

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

140510

*62

: Date -14th of May 2010

: Autonomous mode

: The checksum data, always begins with

$GPRMC.112537.613A1123.5703,N,07739.6095.E, 12.58,193.19,240410,
$GPRMC.112538.613A1123.5664,N,07739.6085.E, 14.67,193.98,240410,
$GPRMC.112539.613A1123.5625,N,07739.6074.E, 14.41, 195.1 1,240410,
$GPRMC.112540.613A1123.5584,N,07739.6062.E, 15.43,196.10,240410.
$GPRMC.1 12541. 613A1123.5539,N,07739.6047.E, 17. 12, 197.57,240410.
$GPRMC.1 12542. 612A1123.5489,N,07739.6032.E, 18. 72, 196.60,240410.
$GPRMC.112543.512A1123.5434,N,07739.601S.E,20.23,194.56,240410.
$GPRMC.112544.612A1123.5380,N,07739.6004.E,20.16,194.21, 240410.
$GPRMC.112545.512A1123.5327,N,07739.5991.E, 19.48,193.18,240410.
$GPRMC.112546.611,A,1123.5275,N,07739.5979.E,18.70,193.55,240410.
$GPRMC.112547.511,A,1123.5231,N,07739.5968.E,16.84,193.56,240410.
$GPRMC.112548.611A1123.5192,N,07739.5958.E,14.45,193.67,240410,
$GPRMC,112549.611A1123.5159,N,07739.5951.E,12. 11,192.77,240410,
$GPRMC.112550.611A1123.5132.N,07739.5945.E,9.81, 191.68,240410,,
$GPRMC.112551.610A1123.5113.N,07739.5941.E,7.23,191.44,240410„
$GPRMC.112552.610A1123.5101,N,07739.5939.E,4.31, 192.94,240410,,
$GPRMC.112553.610A1123.5096,N,07739.5937.E,1. 85,193.33,240410,,
$GPRMC.112554.610A1123.5095,N,07739.5937.E,0.00,193.33,240410„
$GPRMC.112555.610A1123.5095,N,07739.5937.E,0.00,193.33,240410„
GPRMC,112556.609,A,1123.5095,M,07739.5937.E.0.00,193.33,240410,

Figure 4. Refined Log file of $GPRMC sentences

Hence the next step in map making process is to refine the
log file by removing other sentences in such a way that it
contains the $GPRMC sentences only as shown in Fig 4. This
refined log file now contains the path of the probe car in terms
of latitude and longitude at an interval of one second per
sentence.

C. Segment Extraction

The road segment extraction is done by considering the
locations of the probe car at a fixed time interval. Since in most
cases there is no significant distance between the probe car and
the road centre line, it is assumed that the path of the vehicle is
along the road centre line. The idea is to pick out 'k' locations in
road centre line continuously by interleaving 'n' $GPRMC
sentences. When this process is iterated, the i* location of the
moving vehicle Xi,yi (longitude Xj and latitude y) for all the
values of 'i' ranging from 1 to 'k' along the road centre line at
'm' seconds interval is obtained.

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

Vol 8, No. 3, 2010

After obtaining the points at road centre lines after

interleaving m sentences in the middle, the slope between two

^ adjacent points say Xi+i,yi+i and Xi,yi is calculated. Now an

M3a imaginary line perpendicular to the slope is drawn through Xi,yi

and they are made to intersect an imaginary circle of radius 'r'

on both sides as shown in Fig. 5. The points of intersection on

either side (Ixj, ly and rxj, ry) are recorded and they act as the

nodes for the left line of the road and right line of the road,

forming road segments. This process is repeated for all values

of 'i' ranging from 1 to k-1. It is to be noted that the radius of

the imaginary circle determines the width of the road segment.

Roads with different widths can be drawn by altering the radius

'r' of the imaginary circle.

The following algorithm is used to extract segments on
either side of the road centre line. This algorithm extracts
segments by computing road vectors on either side of the road
and adds lines in vectors. The output of this algorithm for a
sample data is given in Fig 5.

Input : Refined log file; sentence count m; radius of
imaginary circle r; number of sentences to be interleaved n.

Output : Line segments on the left and right side of the road
forming a road segment; line vector Ix, ly for road left line and
rx, ry for road right line.

Step 1 : Initialize data items: m=0; i=0; n=40, r=0.5.
Step 2: Repeat step 3 to step 4 till the end of log file.
Step 3: Read the next sentence from the log file
Step 4: Ifm>=n

Read the sentence from the log file and
store longitude into Xj and latitude into y.
i=i+l
m = l
Else

m = m+l
:k = i-l
: Repeat step 7 for i = 1 ,2, . . . .,k- 1

Step 5 :
Step 6 :
Step 7 :

dx=Xi+i -Xi

dy=y+i -y

slope_radians =tan^(dy/dx)

9i =slope_radians / (tt/ISO)

a= 01 + 90

left_plot_radians = (tt /1 80) * a

Ixi = Xi + r*cos(left_plot_radians)

ly = y + r*sin(left_plot_radians)

PlotdXi, ly)

p=a + 180

right_plot_radians = (tt/ISO) * P

rxi = Xi + r*cos(right_plot_radians)

ry = y + r*sin(right_plot_radians)

Plot(rXi, ry)
Step 8 : line(lx,ly) - Add the line in vectors Ix and ly to the
current axes.

Step 9 : line(rx,ry) - Add the line in vectors rx and ry to the
current axes.
Step 10 : Stop.

Figure 5. Obtaining Points for Left and Right Road lines using a Point of Road

Centre line

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

QdJi ^ ^ \03iK'^-|l|nil|"a

(IJCSIS) International Journal of Computer Science and Information Security,
__ Vol 8, No. 3, 2010

™ process is iterated for subsequent segments until all the
segments are processed.

The algorithm used for segment merging is given below.

Input : Negligible slope (angle) \ road segments (vectors
for left line lx,ly and right line rx,ry).

Step 1 : Compute the slope(angle) of the line connecting
Xi,yi and Xi+i,yi+i of two adjacent segments (6i and 6i+i) with
respect to 'x' axis.

Step 2 : Repeat step 3 till all the segments are processed

(i=l,2,...,k).

Step 3 : if abs(6i-6i+i) < \ merge the two segments i.e.,
remove intermediate node; otherwise do not merge.

Figure 6. Road segments extracted on the left and right sides of road centre
line (vehicle trajectory)

D. Segment Merging

The road segment merging technique is used to merge the
similar basic road segments together to form a single road
segment. It is observed from the outcome of the segment
extraction algorithm that a number of adjacent road segments
are similar in direction and they can be merged together to
form a single segment. The primary advantage of segment
merging is that it reduces the number of nodes in the road
network and hence simplifies the map. It also results in faster
generation of the map because of the less number of segments
in the map.

Figure 8. Road segments with left and right road lines after removal of vehicle

trajectory

Figure 7. Road segments extracted on the left and right sides of floating car
GPS data after segment merging

The criteria used to merge the road segments is the slope
(angle) 6i of the line connecting Xi,yi and Xi+i,yi+i with respect
to X axis. A threshold limit is set for the negligible slope
(angle) \ and if this slope(angle) is within the threshold, say 5°
for two adjacent segments, then both the segments are merged
together, otherwise the segments are left as they were. This

Figure 9. Road Map after removing the plots for nodes

Thus the segment merging algorithm is based on negligible
slope and it merges the similar segments together which results
in faster production of digital maps by eliminating unwanted
nodes in the road network. The extracted segments are shown

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

in Fig. 6 and the merged segments are shown in Fig 7. It is
observed that different threshold values for negligible slope
(angle) X can be used to get required accuracy or width of the
route map. The merged segments after removing vehicle
trajectory are shown in Fig. 8. The final map of the road after
removing the plots for nodes is shown in Fig. 9.

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

Vol 8, No. 3, 2010
• Removal of a segment: Sometimes, a segment at the
end of one road and a segment at the beginning of the
next road may intersect and cross one another. In that
case, one of the segments is deleted.

E. Refining Intersection of Road Segments

The obtained road network after segment merging is still
unrefined at intersections of road segments. The refinement of
road intersection is done in the following steps.

Figure 10. Digital Map with unrefined intersection of road segments

Fie dt VleA :nsett Tocb Dsttop rtidcw -dp

^^^^

^^^^^"

^HMa

naii<^\>\''\-\i'j^'^/.-\^\nm\mm

^^B

^^^^~

■

1122.5

■

112\5
112'

/ /

1 1120.5
3

^^=^^=r

^^^^^'

__H

^—-^-Zl~^~~^

1120

^ ^

1119

^<::^

■

i 11.8.5

"^^36 773:.: 7737 ^37 5 7738
LoxrudE

773E.E 7^9

m%.i

Figure 11. Digital Map with refined intersection of road segments

► Adding a segment: Many a times, there may be a
necessity to add one segment either at the end or at the
beginning of a road so that the road is connected to
another road. In that case, a new segment is added.
Sometimes, instead of adding an entire segment, only
the left or right line of the road segment is extended.

• Extending a segment: When two or more segments do
not join at the intersection, the segments are extended
based on the previous slope till they form refined
intersection. It is to be noted that extending a segment
is different from adding a segment.

A map with unrefined road segments intersection is shown
in Fig. 10. The road intersection is refined and Fig. 11 shows
map with refined intersection of road segments. This process is
repeated for entire set of road intersections till a fine-tuned map
is obtained.

F. Labeling Points of Interest

A map contains points of interest which means places
which are important and noted down on the maps. Placing text
on a map is a particularly difficult challenge in digital maps
because the axis of the digital maps can be changed
dynamically. This is the final step in this map making process
and points of interests are noted down from the probe car data
and the text is placed so that it is readable and easily located.
Care has also been taken that the text does not interfere with
the map's data or design. Different font types, styles, sizes, and
colors are used to establish clear association between text and
map features like telephone post, petrol bunks and toll gates.
Fig. 12 displays a legible point of interest.

^Qffi

Qd a ii u- 1 '% % os«! ^ -I Si n s 1 ■ Q

1122.5

/ /'
/ /

112-i

1 I

/ J

ii;o

~~~~~~'~~-~

"-^

" ^^

' ,n Nasiysnur

"— — — -

1119.5

^/^^

■

1119

^^^^

■

I1f,5

x^>^"^^

■

1110

-£ 77Tz E 7737

7737 5

7738

773E.5 7^9 7739.5

LoigruJE

Figure 12. Digital Map with Labeled Points of Interest

V. Experimental Results

The outcome of the map making process is a list of
segments drawn on the map with refined intersections of roads.
For the purpose of demonstration, the floating car data was
collected in different roads. Segments for the roads are
extracted using the segment extraction algorithm. Thereafter
segment merging was done based on the slope of the adjacent
road segments. Finally, the road intersection points are refined
and points of interest are labeled to get the final map. The

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

following experiments demonstrate the simplicity and accuracy
of the map making.

A. Simplicity

A comparison of the map generated by the proposed map
making process as shown in Fig. 13 versus a digital map
available in the web as shown in Fig. 14 indicates that the
proposed process is simple and the map includes only the
desired routes and points of interest. The generated map can be
easily interpreted because of its simplicity.

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

Vol 8, No. 3, 2010
more accurate segments depending on the necessity. Two
different segment extractions for the same floating car data
with different values of 'r' according to varying requirement is
shown in Fig. 15, 16. It is to be noted that the nodes are shown
only to differentiate segments and they are removed in the final
map. A more accurate and thinner road is shown in Fig. 16
compared to the one shown in Fig. 15.

jj^i,;',\\?!?ii«^'is|Dni|ia

Figure 13. Map generated by the digital map building process discussed in
this paper

Figure 14. A previously available digital Map with dense routes and places

B. Accuracy

One of the input parameter of segment extraction algorithm
is the radius of the imaginary circle 'r'. By adjusting the value
of 'r' the road segment may be made thicker or thinner to get

Fie EJt VisA Insert Tocis Ds^lop jlrAivi -elp

UEid^ii^|v^a^i/-|'5|BB|i.a;:

Figure 15. Map with thick roads with less accuracy

e EJt VieA Insert TocIs Dsttop i^iT±w -dp

J^:^ > ^ -•.'i:)%v^- a D B I ■ g

Figure 16. Map with thin roads for the same data with more accuracy

Accuracy can also be obtained by making the sampling
sentences interval as minimum. Table II shows a Comparison
of segments extracted and segments generated after merging at
different intervals of sampling sentences. Accuracy increases
with more number of extracted segments and merged segments.
The graph shown in Fig. 17 gives a comparison between the

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

number of extracted and merged segments for a floating car
data at different sampling intervals.

TABLE II. Comparison of segments extracted and segments

GENERATED AFTER MERGING AT DIFFERENT INTERVALS OF SAMPLING
SENTENCES.

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

Vol 8, No. 3, 2010
This process can be used to create maps but the process
depends on a probe car for data collection. In future, this
drawback can be eliminated by collecting GPS data from the
GPS enabled vehicles already connected with commercial
applications. This work can also be extended to handle roads
with multiple lanes.

Interval

of
sampling
sentences
(in No. s)

No of

Points

generated

by probe

car data

No. of
segments
extracted

No. of
segments

after
merging

Percentage of

reduction in

segments after

merging

413

52.63

413

50.00

413

50.00

413

33.33

o 10

20 40 60 80

Sampling Sentences Interleaved

- No. of Extracted Segments

- No. of Segments after Merging

Figure 17. Number of segments generated at different sampling intervals

The accuracy of the map is inversely proportional to the
radius of the imaginary circle which is used to draw the road,
negligible slope and interval of sampling sentences. Desired
accuracy can be obtained by adjusting these values according
to the need.

VI. Conclusion and Future Work

This paper introduces a customized digital road map
building process which can be used to build digital maps of
desired routes. By changing the input parameters, the accuracy
of the route can be altered to required level and route maps for
different types of roads, for instance, national highways, state
highways, city roads and streets can be drawn. The results of
the map building process discussed in this paper are compared
with other maps generated by other commercial making
software and it is found that the proposed process is simple, yet
powerful. This map making process has eliminated the
complexity of the previous works carried out in this area and
customized road map can be built for commercial and other
applications like vehicle tracking, navigation and route
guidance systems.

References

[I] Stefen Schroedl, Kiri Wagstaff, Seth Rogers, Pat Langley and
Christopher Wilson., "Mining GPS traces for Map Refinement," Data
Mining and Knowledge Discovery, vol. 9, pp. 59-87, 2004.

[2] D. Lee, "The Map-Building and Exploration Strategies of Simple
Sonar-Equipped Mobile Robot," Cambridge Univ. Press., Cambridge,
1996.

[3] P. Weckesser, and R. Dillmann, "Modeling unknown environment with
a mobile robot,". Robotics Autonomous Systems, vol. 23, pp. 293-300,
1998.

[4] Y. L. Ip, A.B. Rad, K.M. Chow and Y.K. Wong, "Segment-based map
building using enhanced adaptive fuzzy clustering algorithm for mobile
robot applications," Journal of intelligent and robotic systems, vol. 35,
pp. 221-245, 2002.

[5] Li) WeiFengl,ZHU TongYu, WU DongDong, DAI Hong andHUANG
Jian, "A heuristic path-estimating algorithm for large-scale real-time
traffic information calculating," Science in China Series E:
Technological Sciences, vol. 51, pp. 165-174, Apr. 2008.

[6] B.W. Parkinson, and J.J. Spilker, "Global Positioning System: Theory
and Applications," American Institute of Aeronautics and Astronautics,
Washington, 1996.

[7] Interface control document. Navstar GPS Space Segment (Navigation
User Interfaces), 2000.

[8] Asoke K Talukder and Roopa R Yavagal. "Mobile Computing-
Technology, Applications and Service Creation," Tata McGraw Hill
Publishing Company, 2005.

[9] X. W . Dai, M.A. Ferman, "A simulation evaluation of a real-time traffic
information system using probe vehicles," IEEE ITSC, vol. 1, pp. 12-15,
2003.

[10] R. Kuehne, R.P. Schaefer and J. Mikat, "New approaches for traffic
management in metropolitan areas," IFAC CTS, 2003.

[II] T.W. Cheng, D.B. Goldgof, and L.O. Hall, "Fast fuzzy clustering,"
Fuzzy Sets Systems, vol. 93, No. pp. 49-56, 199 8.

[12] R.N. Dave, "Characterization and detection of noise in clustering,"
Pattern Recognition Lett. vol. 12, pp. 657-664, 1991.

[13] R.N. Dave, "Use of adaptive fuzzy clustering algorithm to detect lines in
digital images," Intelligent Robots Computer Vision, vol. VIII, pp. 600-
611, 1989.

[14] National Marine Electronic Association, http://www.nmea.org,
accessed on 20.04.2010.

AUTHORS PROFILE

G. Rajendran is a second-year Doctoral

Research Scholar in the Research and

Development Centre of Bharathiar University.

He received his Masters degree in Computer

Applications and M.Phil degree in Computer

Science from Bharathiar University. He passed

the National Eligibility Test for lectureship

conducted by UGC, the apex body of higher

education in India. He is also working as an

Assistant Professor of Computer Science at

Thiruvalluvar Government Arts College,

Rasipuram, India. His research interests include

Mobile Computing, Data Mining and programming-language support for

massive-scale distributed systems.

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

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

Vol. 8, No. 3, 2010
Dr. M. Arthanari holds a Ph.D. in Mathematics
from Madras University as well as Masters
Degree in Computer Science from BITS,
Pilani. He was the professor and Head of
Computer Science and IT Department at Tejaa
Shakthi Institute of Technology for Women,
Coimbatore, India. At present he is the
Director, Bharathidhasan School of Computer
Applications, EUispettai, Erode, Tamilnadu. He
holds a patent issued by the Govt, of India for
his invention in the field of Computer Science.
He has directed teams of Ph.D. researchers and

industry experts for developing patentable products. He teaches strategy,
project management, creative problem solving, innovation and integrated new
product development for last 36 years.

T^^

M. Sivakumar has 10+ years of experience in

the software industry including Oracle

Corporation. He received his Bachelor degree

in Physics and Masters in Computer

Applications from the Bharathiar University,

India. He is currently doing his doctoral

research in Anna University, Coimbatore. He

holds a patent for his invention in embedded

technology. He is technically certified by

various professional bodies like ITIL, IBM

Rational Clearcase Administrator, OCP -

Oracle Certified Professional lOG, PRINCE2 and ISTQB. His research

interests include Embedded Technology, Ubiquitous Computing and Mobile

Computing.

29 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

Robust stability check of fractional control
law applied to a LEO
(Low Earth Orbit) Satellite

Ouadia EL Figuigui \ Noureddine Elalami^

^ Labor atoire d'Automatique et Inform atique Industrielle
EMI, Morocco

( elalaini@eini . ac.ma ^ elfiguigui@gmail . com )

Abstract: The use of the differentiation and integration of
fractional order or non-integer order in systems control is
gaining more and more interests from the systems control
community. In this paper we will briefly describe the LEO (Low
Earth Orbit) satellite systems and recall the theoretical aspects
of robust stability check procedure. This procedure will be
applied to a LEO satellite that is under the fractional control
law. Numerical examples will be analyzed and presented at the
end of this document

Keywords: fractional control^ LEO satellite^ robust stability
L .INTRODUCTION

Recently, a lot attention was given to the problem of
fractional calculus. There were several works in this area
[11], [12], [13], [23]. .etc and the author in [1] is presenting for
the very first time the robust stability checking procedure for
uncertain fractional order linear time invariant (FO-LTI)
systems with interval coefficients described in state form. The
application of such procedure to LEO satellite was new idea.

The role of an attitude control system for an Earth-pointing
spacecraft is to maintain the local-vertical/local-horizontal
(LV/LH) attitude with the presence of different environmental
disturbances. Most of the time in order to ensure a precise
pointing, the satellite requires reaction wheel system to
counteract the attitude drifts caused by those perturbations,
especially the seculars ones, like torques caused by passive
gravity gradient, aerodynamic and solar forces. The reaction
wheels are governed by control laws which dictate the
amount of torques required to eliminate the drift caused by
external factors [9].

In the attitude control design, different approaches have
been used, i.e. Proportional Integral Derivative (PID) [18],
[25], LQR [17], pole placement techniques [14], etc. All those
methods, expressed in different attitude error terminologies
are using, very often, Euler angles for small attitude
commands while for large attitude maneuvers it makes use of
quaternion [7] and direction cosine errors [18].

In [9], the author introduce a fractional control law
aiming to stabilize the attitude movement of an earth pointing
satellite under the effect of external disturbances, using 3 -axis
reaction wheels as actuators. The dynamics of the satellite is
described by a quasi-bilinear multivariable coupled system. In
this study, we apply the robust stability check procedure to
the fractional control law system presented in [9].

This paper is organized as follows: In the next section, the
general nonlinear equations model of an Earth-pointing
satellite attitude dynamics is developed. Then, in section III,
the attitude equations are linearized with the nadir attitude
position as the origin. This leads to a quasi_bilinear
multivariable coupled system. Then, for small maneuvers, the
quasi_bilinear term is neglected in order to obtain a linear
system. In section IV, we recall some theoretical aspects of
robust stability checking procedure. Forward, in section V, we
apply this procedure to the LEO satellite system which is
subject to fractional control law and we share the related
Numerical results. The conclusion is provided in the last
section.

II. NONLINEAR MODEL OF THE SATELLITE
ATTITUDE DYNAMICS

The attitude motion of the satellite is represented by the
Euler equations for the rigid body motion under the influence
of external moments, such as the control moment generated by
the actuators. Attitude control requires coordinate
transformation from LV/LH to The Satellite Coordinate
System (SCS) system defined as follows: The LV/LH
coordinate system {X^^Y^.Z^) is a right orthogonal system
centred in the satellite's centre of mass (SCM). The roll
axis, Xq , points along the velocity vector, the pitch axis, Fq ,
points in the direction of the negative orbit normal and the
yaw axis, Zq , points in the nadir direction. The SCS system

(X^,F^,Z^) is a right orthogonal system centred in the
SCM, parallel to principal moment of inertia axle of satellite.
Z^ is parallel to the smallest moment of inertia axis; Y^ is

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

parallel to the largest moment of inertia axis. X^ completes

the right orthogonal system.

Consider a satellite with three reaction wheels. The general
nonlinear attitude dynamics model can be described as [18],
[27] and [21]:

I(hUt)= hjt) coUt)McoUt)

ojUt)AhJt) + Ml(t) + P(t) ^^^

where

> / : Total moment of inertia matrix for the satellite

without reaction wheels inertia (3x3).

> CO J (t) : Inertial angular velocity vector in SCS (3x1).

> h^(t) : Angular momentum vector of the wheel

cluster(3xl).

> M^ (0 : Torques due to Earth's gravity gradient (3x1).

> P(t ): Disturbance torque due to aerodynamics, solar

pressure and other environmental factors. It is assumed

to be [18], [21]:

P(t)--

4xlO"^ + 2xlO"^sin(6>oO
6x10"' +3x10"' sin(6>oO
3x10"' +3x10"' sin(6>oO

(2)

where cDq is the orbital angular rate.

To keep the satellite attitude earth pointing, the SCS axes
must remain aligned with LV/LH axes. The transformation
matrix, expressed with Euler Angles (^, 0, i//) , respectively,
roll, pitch and yaw angles, is given by [18], [27]:

(3)

where: S and C are respectively the sine and the cosine.
The gravity gradient torque M^ (t) is given by [18], [5]:

c,c^

CeC,

-Se

~^(j)\ ^^^^^^(//

cf,+s^s,s^

Sf,

_S^Sf + C^S,C^,

-sf,+c^s,s,

Cf^

<(h

-Iy)sm{2^)cos^{0)

3 9
: - (Dq (I^ - 1^ ) sin(26>) cos(^)

(4)

' gz

To describe the satellite kinematics, two important factors are
to be taken into account: angular velocity of the body axis
frame (SCS) with respect to the reference LV/LH

frame cOyf^ (p,q,r) , and the angular velocity of the body frame
with respect to inertial axis frame cOj {co^ ,cOy,co^) . These

quantities are related to the derivative of the Euler angles as
follows [18]:

and

cOj =c4H+TyH/s(0 -% 0/

(5)

III. LINEARIZED EQUATIONS OF MOTION

Assuming small variations of the Eulerian angles (^, 6, y/) ,
then the transformation matrix becomes:

-0~

-y/

-^ 1

(6)

On the other hand, one obtains that:

(j) ^ p ,6 ^ q ,if/ = r
and

(O^-^- CO^Xj/, CO -9-co^, co^ = ^ + co^^

(7)

(8)

Then the equations of motion (1) and (5) can be linearized
about the origin, giving a quasi_bilinear multivariable system:

x(t)^ Ax(t)-\-Bu(t)

3 t

+ ^2 ^.(^di (C.x(t))

i=l

;x{Q) = x.

(9)

+ G ^(^)d^ +BP(t)

where:

> 5^ Ji/,.(^)J^(C,.x(0) + Gfi/(^)J^

: Quasi-bilinear term.

> u{t) = -h^ (t) : Control action,

> x(t) = {(/), 0, y/, (j), 0, ij/) : State vector.
System matrices A, B, C- , G defined as

c,=

-0,

-1

c^o

-ffl

-0,

-1

B =

l/Iy

-1 0_

1//

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

M^i

36^00-2

(oIg,

-mAU

1
1

G)^(\-G,)

B. Robust Stability Check of Fractional System with
Interval Uncertainties [1];

We consider the following FO-LTI system with interval
uncertain:

(ojlx

-(0,1 Iz

X^''^(t) = AX(t) + Dw(t)

where:

> a is non integer number;

(13)

> A^A

= \a^a\= If -AAA' -\-AA\wii\\

A + AA
A = is a center matrix (normal plant without

where a.=[lj-lJ/l.foY the (/,j,^) index sets (l,2,3),
(2,3,1), and (3,1,2).

Moreover, assuming that the angular velocity
components p , q and r are also small, and for slight
manoeuvres, one can neglect the quasi bilinear term.

The equation of motion (9) can then be written in the
standard form of a linear equations system:

x(t) = Ax(t) + Bu(t) + BP(t);x(0) = Xo (10)

IV. ROBUST STABILITY CHECK OF
FRACTIONAL SYSTEM WITH INTERVAL
UNCERTAINTIES - MATHEMATICAL
ASPECTS

In this section, we're recalling the definition of fractional
system; we're presenting the robust stability checking
procedure afterwards.

A. Definition

In this paper, we consider the Riemann-Liouville
definition, in which the fractional order integrals are defined
as

uncertainties)

> AA = -

A-A

■ is a radius matrix correspondence interval

uncertainties.

The stability condition for system (13) is:
min\arg(l,(A))\>a^ ; i = l,2,...,N, W^A'

In the following subsection, we describe briefly the
procedure of checking robust stability using the minimum
argument phase. We proceed by introducing two important
lemma:

1. Lemma 1 [1]:

Define a sign calculation operator evaluated at A such as:

p' := sgn Lr vr - i^i"'vi"' T\ (W)

whQTQu^\v-\ul'^ Sind v-'^are eigenvectors corresponding to
ith eigenvalue of A'' .If P' is constant for all A ,A ^A^ , then
the lower and upper boundaries of the real part of ith interval
eigenvalue are calculated as:

:0;'{A -AAoP')

(15)

Dff(t)=j--iHrf(^)d(^)fi>o

r(fi)

(11)

where ^•'^^ (•) is an operator for selecting the ith real eigenvalue
= a,^jb,^j ,and as:

2!-; =o;'{a +aaop') (16)

and C = Ao 5 are c^^ = ^kpkj '^^^ ^^•

While the definition of fractional order derivatives is

Df,fit) = j^Dj' "^fity

1. Lemma 2: [1]

Defining a sign calculation operator evaluated at A such

as:

1 d
r(l fi)dt

Jt a'K^dia

(12)

where r( x ) = ^y^ ^ ^ ^y 1^ ^^ Gamma

function, {a,t) g IR^ with a<t and < // < 1 is the order of
the operation.

For simplicity we will note D^f{t) or f\t) for D^f(t)

QA=sgri^rvf^ufvry] (17)

if 2' is constant for SillA,A ^A^ , then the lower and upper
boundaries of the imaginary part of ith interval eigenvalue are
calculated as:

Af =Oi"'{A -AAoQ')

(18)

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

where ^-""Qis an operator for selecting the ith imaginary
eigenvalue, and as:

lf=ei"'{A'^AAoQ')

(19)

So, using Lemmas 1 and 2, it is easy to calculate the lower
and upper boundaries of interval eigenvalue separately in real
part and imaginary part. From above lemma, if P' and Q' ,
/=1,..., N are calculated, then, interval ranges of
eigenvalues are finally calculated as:

4 ^r/ := |£, A7 J+ AXr^, XrW (20)
where j represents imaginary part. We define

(j) = infi; min|arg/l- ( A)|) i = l,..N

Since the stability condition is given as ^ > aTi / 2 , if we
find sufficient condition for this, the stability can be checked.
For calculating ^* , the following procedure can be used:

PI. Calculate P. and g.for/ =1,..., A^ .

P2. Calculate A'^ , A'^ , if , and if for all / g {1,2, ...,n}.
P3. Find arguments of phase of four points such as

in the complex plane.

F4.Fmd</>: =inf\</>:[\</>^[\</>f\,\^;'\}.

P5. Repeat procedures P3 and P4 for i=l,...,N .

P6. Find/=inf{ ^*,/ = 1,..., A^}

P7. If (p* >an 12 , then the fractional interval system is
robust stable. Otherwise, the stability of system cannot be
guaranteed.

V. ROBUST STABILITY OF LEO SATELLITE

The aim of this section is to apply the robust stability
checking procedure subject of the section IV. This procedure
will be used to proof that our system (10) under the control
law presented in [9] is robust stable.

A. Fractional control law

As mentioned above, the LEO satellite atitude dynamics is
described, when neglecting the quasi-bilinear term, by the
system (10):

x(t) = Ax(t) + Bu(t) +w(t);x(0) = XQ (21)

where w(t) = BP(t) is the perturbation term, and u(t) is the
fractional control law applied in order to stabilize the system
(10), given by:

u(t)= K{xit)-x^i''^ (22)

where x^ is the attitude reference, equal to zero for nadir
pointing.

The linear fractional system is obtained in the form:

x(t ) = Ax(t )-BKx(t j""^ +w(t ) \ x{{)) = x^ (23)

In the following, only the fractional orders such as
a = l//7,;7GN* will be considered. Then

x{t) = Ax(t) - BKxitf^P^ + Dwit)

(24)

The equation (24) can be written into the following form:

X^''P\t) = AX(t) + Dw(t) (25)

We note:

(26)

x^/^M (t)=x(t),

*p-i

>M (r) = / ^^it)

and

X{t) =

(0,

V J

(f),...,U

p-i

it)

(27)

Id
Id
W

fo\

;D =

Jd)

(28)

Id
A -BK

As mentioned above in the subsection IV.2, the system (26)
is stable, if and only if:

minarg/l.(yl) >

(29)

The problem of stabilization by state feedback is equivalent to
find a matrix K which stabilizes (10); i.e. which checks the
stability condition given in (29).

B. Robust stability of the fractional control law
In this section, we suppose that our system is submitted to
some perturbations which affect his parameters:

1. Case 1: Perturbation of co^ orbital angular rate

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

We consider that, due to the external perturbation, co^

varies between (o.

■ cOq - AcOq

and

<^n

cOq +A(ifQ.

Consequently, A varies between A and A . So our system (26)

is transformed to FO-LTI system with ^ g [yl ylj .

For checking the robust stability of system (26), we apply
the procedure described in section IV.

lf<ff' >a7r/2 the system (26) is robust stable. Otherwise,
the stability of system cannot be guaranteed.

■ Numerical application
The simulation parameters are the orbital rate
^Q =0.00104 rod / sec and the total moment of inertia
matrix for the spacecraft.

0,95

0.85

0.75

12%

16%

20%

4,020

3,989
3,010

Kg.m^

Fig 1: (p* versus

Aco^

The fractional control law which stabilizes the system (10)
with these numerical values is u(t ) = -Kx"^ with a = 0,5 and

1 83 0,758 3,530

9 23,900 2900 7
3,200 55

We give below the results of robust stability checking
procedure applied to the system (26). These results justifies
the robustness of fractional control laws with respect to the
variation of 0)q which can reach ± 20%.

We note also that the curve of the evolution have a linear

behaviour and ^ decrease when increase.

Table I: Results Of Robust Stability Checking Procedure
For Variation Of 0)^

Aco^

Aco^/ 00^

COg

(P*

0.0010400

0.9334

0.0000416

0.0009984

0.0010816

0.9060

0.0000832

0.0009568

0.0011232

0.8772

0.0001248

12%

0.0009152

0.0011648

0.8476

0.0001664

16%

0.0008736

0.0012064

0.8178

0.0002080

20%

0.0008320

0.0010400

0.7888

2 .Case 2 : Intrinsic Parameters of LEO Satellite are
Perturbed

■ Variation of I ^ :
We suppose that /^varied between /^ -A/^et

/^ + M ^ due to external perturbation which modify the form

of the satellite.

For the simulation, we consider the same numerical values
and fractional control law of subsection V-2-1.

The following table resume the results of robust stability
checking procedure.

The system is still robust stable until a variation of ± 20%

ACOr.

of I ^ . The (p decrease when increase.

Table Ii:Results Of Robust Stability Checking
Procedure For Variation Of I^

AIx

AIx/Ix

Ixmin

Ixmax

0.00%

4.0200

0.9334

0.1608

4.00%

3.8592

4.1808

0.8909

0.3216

8.00%

3.6984

4.3416

0.8527

0.4824

12.00%

3.5376

4.5024

0.8244

0.6432

16.00%

3.3768

4.6632

0.8062

0.8040

20.00%

3.2160

4.8240

0.8020

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

0.95

4% 8% 12% 16%

AIx/Ix
Fig 2: cp' versus AIx/Ix

■ Variation of /

Using the same values as before, we give bellow the results
of robust stability checking procedure.

The system is still robust stable until a variation of

± 10,53%. The (p decrease when increase.

Table Hi: Results Of Robust Stability Checking
Procedure For Variation Of I

Mil

0.9334

2.00%

0.8889

4.00%

0.8457

6.00%

0.8165

8.00%

0.8163

10.53%

0.7897

0% 2.00% 4,00% 6,00% 8,00% 10.53%

AI/I

Fig 3: (j) versus AI/I

VI. CONCLUSION

In this paper, we assume that the orbital angular rate is
subject to some perturbations and, consequently, the LEO
satellite system become uncertain. The added value from this
work is to prove that the fractional order control present in [9]
is robust stable. We used as method the robust stability
checking procedure developed in [1]. The mathematical
aspects of this procedure were recalled. We have studied also
the case of the perturbation of intrinsic parameters of LEO
satellite due to the external perturbation which mean that the
total moment of inertia is uncertain.

REFERENCES

[I] Y.Q.Chon, H.S.Ahn, I.Podlubny, Robust stability check of fractional order
linear time invariant systems with interval uncertainties, signal processing 86
(2006)2611-2618.

[2] F.A.Devy Vareta, Pseudo-Invariance Sous Groupe de Transformation :un
nouveau Concept pour la Commande Robuste, Seminaire Toulousain
« Representation Diffusive et Application »- N°l-nov. 2000.

[3] L. Dorcak, I. Petras, I. Kostial and J. Terpak, State space controller design
for the fractional-order regulated system, ICCC 2001, Korynica, Poland, pp.

15-20.

[4] M. M. Dzhebashyan, Integral transforms and representation of functions
in the complex plan, Nauka Moscou 1966.

[5] P.C. Hughes, Spacecraft Attitude Dynamics, John Wiley & Sons, USA
New York, 1986

[6] B. Kim, E. Velenis, P. Kriengsiri & P. Tsiotras, A Spacecraft Simulator
for Research and Education, AAS 01-367.

[7] B.J. Kim, H. Lee and S.D. CHOI, Three-axis Reaction Wheel Attitude
Control System for KITSAT-3 Microsatellite, IFAC Conference in
Autonomous and Intelligent Control in Aerospace, Beijing, 1995.

[8] A. Kailil, Architecture et Analyse Dynamique par la Methode des
Elements Finis de la plate forme d'un microsatellite, DESA, CRES,
Mohammedia Engineers School, Maroc, 2000.

[9] A. Kailil, N. Mrani, M. Abid, M. Mliha Touati, S. Choukri, N. Elalami,
Fractional regulators for spacecraft attitude stabilization, accepted in the 22nd
AIAA-ICSSC, Monterey, Cahfomia 9-12 May 2004.

[10] S.Ladaci,JJ.Loiseau,A.Charef,Stability Analysis of Fractional Adaptive
High-Gain Controllers for a class of Linear Systems General case,2006,
ffiEE.

[II] S.Ladaci,A.Charef, Mit Adaptive Rule with Fractional Integration, in
Proceedings CESA 2003 MACS Multiconference Computational
Engineering in Systems Applications, Lille, France.

[12] S.Ladaci, A.Charef, On Fractional Adaptive Control, nonhnear
Dynamics, vol.43 ,n°4, March 2006.

[13] S.Ladaci, A.Charef, An Adaptive Fractional PI D. Controller, in
Proceedings TMCE 2006 international Symposium series on Tools and
Methods of Competitive Engineering, Ljubljana, Slovenia, April, Aprill8-22.

[14] CH. Lubich, Discretized fractional calculus, SIAM J. Math. Anal. Vol.
17 No. 3 May 1986. P

[15] D. Matignon, Fractals et loi d'echelle, Edition Hermes 2002.

[16] D. Matignon, Stability results for fractional differential equations with
apphcations to control processing, in Computational Engineering in Systems
and Application multiconference, vol. 2, pp. 963-968, MACS, lEEE-SMC.
Lille, France, July 1996.

[17] K.L. Musser, W.L. Elbert, Autonomous Spacecraft Attitude Control
using Magnetic Torquing only. Proceeding of flight mechanics estimation
theory symposium, NASA, 1986, pp. 23-38.

[18] M. J. Sidi, Spacecraft Dynamics and control, Cambridge University
Press, Cambridge, UK, 1997.

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

[19] S.G. Samko, A. A. Kilbas & O.I. Maritcheu, Fractional integrals

and derivative: theory and application, Gorden & Breach, 1987.

[20] A. SkuUestad, J. Gilbert, H^ Control of a Gravity Gradient Stabihsed

Satellite, Control Engineering Practice 8 (2000) 975-983.

[21] P. Tsiotras, H. Shen and C. Hall, SateUite Attitude Control and Power
Tracking with Momentum Wheels, AAS 99-317.

[22] C. Valentin-Charbonnel, G. Due and S. Le Ballois, Low-order Robust
Attitude Control of an Earth Observation Satellite, Control Engineering
Practice 7 (1999) 493-506.

[23] B.M.Vinagre,I.Petras, I.Podlubny and Y.Q.Chen, Using Fractional Order
Ajustement Rules and Fractional Order References Models in Model-
Reference Adaptive Control, Nonlinear Dynamics, vol.29.

[24] B. Wie, Space Vehicle Dynamics and Control, AIAA Education Series,
1998.

[25] C. Witford, D. Forrest, The CATSAT Attitude Control system.
Proceeding of the 12* Annual AIAA/USU Conference on Small Satellite,
1996.

[26] C. H. Won, Comparative study of various methods for attitude control of
LEO satellite. Aerospace Science and Technology. 1270-9638. 99/05/
Elsevier, Paris.

[27] J. R.Wertz, Spacecraft Attitude Determination and control, Kluwer
Academic Publishers, Dordrecht, Holland, 1978.

[28] K. Zhou, JC. Doyle & R. Glover, Robust and Optimal control. Prentice
Hall, 1996

36 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

PERFORMANCE EVALUATION OF

GENETIC ALGORITHM FOR SOLVING

ROUTING PROBLEM IN

COMMUNICATION NETWORKS

Ehab Rushdy Mohamed

Faculty of Computer and Informatics

Zagazig University

Zagazig, Egypt

ehab. rushdy (a),smail. com

Ibrahim Elsayed Zidan

Faculty of Engineering

Zagazig University

Zagazig, Egypt

ibrahim.zidan. 123(a),smaiL com .

Mahmoud Ibrahim Abdalla

Faculty of Engineering

Zagazig University

Zagazig, Egypt

mabdalla201 0(a),smail. com

Ibrahim Mahmoud El-Henawy

Faculty of Computer and Informatics

Zagazig University

Zagazig, Egypt

. /. m. elhenawv(a),smail. com .

Abstract — There has been an explosive growth in both
computer and communication networks since the last
three decades. Communication networks pervade our
everyday life, from telephony system, to airline
reservation systems, to electronic mail services, to
electronic bulletin boards, to the internet. Routing
problem is one of the most important issues facing the
development, improvement and performance of
communication networks. Recently, there has been
increasing interest in applying genetic algorithms to
problems related to communication networks. This study
evaluates the genetic algorithm used for finding the
shortest path in communication network. The paths
result from applying genetic algorithm could be used in
establishing routing table for network protocols. The
genetic approach is thought to be an appropriate choice
since it is computationally simple, provide powerful search
capability, and has the ability to move around in the
solution space without a dependence upon structure or
locality.

The performance of the genetic algorithm is compared
to Dijkstra algorithm, which is widely used in most
network protocols. This is realized using different
simulated networks to clarify the advantages and
deficiencies of each algorithm. The relative performance
of the two algorithms is judged on the basis of delay and
adaptation.

I. Introduction

One of the most common problems encountered in
analysis of networks is the shortest path problem:
finding a path between two designated nodes having
minimum total length or cost. In many applications,
however, several criteria are associated with traversing
each edge for a network. For example, cost and time
measures are both important in most networks, as are
economic and ecological factors. As a result, there has
been recent interest in solving the shortest path problem
and many algorithms are proposed to solve this problem
[1], [2], [3], [4]. Shortest path problem is a classical
research topic. It was proposed by Dijkstra in 1959 and
has been widely researched [5].

A. Dijkstra Algorithm

The Dijkstra algorithm is considered as the most
efficient method. It is based on the Bellman
optimization theory. Dijkstra routing algorithm is
widely used in many applications because it is very
simple and easy to understand. This algorithm finds the
shortest paths from a source to all other nodes. To do
this it requires global topological knowledge (the list of
all nodes in the network and their interconnections, as

1
37

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

well as the cost of each link). In most general the
weight of each link could be computed as a function of
the distance, bandwidth, average traffic, communication
cost, mean queue length, average delay, and other
factors. In most general the weight of each link could
be computed as a function of the distance, bandwidth,
average traffic, communication cost, mean queue length,
average delay, and other factors [6] .

The parameter D(v) is considered as the distance
( sum of links weights along any path) from source
node 1 to node v, and the parameter L(i,j) is
considered as the given cost between node / and nodey.
There are then two parts to the algorithm: an
initialization step, and a step to be repeated until the
algorithm terminates:

1 . Initialization

Set N = {I}. For each node v not in Set TV, D(v) is set
to L(l,v). The value oo is used for nodes that are not
connected to node 7; any number larger than maximum
cost or distance in the network would suffice.

2. At each subsequent step

Find a node w not in N for which D(w) is a minimum
and add w to N. Then D(v) is updated for all nodes
remaining that are not in N by computing
D(v)= Min [D(v) , D(w)+L(w,v)]

Step 2 is repeated until all nodes are in N.

The Dijkstra algorithm is widely used in most
popular routing protocols because of its simplicity and
efficiency. But when the network is very big, then it
becomes inefficient since a lot of computations need to
be repeated. The efficient set of paths may be very large,
possibly exponential in size. Thus the computational
effort required for solving the problem can increase
exponentially with the problem size [7].

B. Genetic Algorithm Mechanism

The genetic approach is a special kind of stochastic
search algorithm, it based on the concept of natural
selection and genetics. It provides a powerful search
capability and has the ability to move around in the
solution space without the structure and locality. This
approach identifies solutions that are closest to the ideal
solution as determined by some measures of distance.
Genetic algorithm constitutes the increasingly large part
of evolutionary calculation techniques, which form the
artificial intelligence .As it's obvious in its name,
genetic algorithm, forming evolutionary popular
technique, inspired from evolution theory of Darwin.
Any kind of problems which involves genetic algorithm
is solved through the application of artificial evolution
technique. Genetic algorithm is used to solve problems

that are hard to be solved by applying conventional
methods. In general terms, genetic algorithm has three
field of application. They are; optimization, practical
industrial applications, and categorization systems [8],
[9].

Genetic Algorithm starts with a set of solution,
which is identified with chromosomes and known as
population. Resolutions that have come out from a
population are applied to the next one with the
expectation of positive improvements. The selected
group is used for creation of a new population
according to their compatibility. Nevertheless, it's
likely that the compatible ones produce better solutions.
This would be continued until the expected solutions
are obtained. A simple genetic algorithm consists of the
following steps:

1. Initialization: a random population of n
chromosomes (appropriate solution of the problem) is
created.

2. Fitness: each x chromosome is evaluated using the
fitness f (x).

3. New population: a new population is created; this is
done by repeating the following steps until the new
population is complete

a. Selection: two parental chromosomes are selected
from the population according to their fitness.

b. Crossover: a new member is created; parents are
cross-fertilized according to possibility of crossover. If
cross-fertilization is not applied, new member is a copy
of a mother or father.

c. Mutation: the place of the new member is
changed according to the possibility of fission.

d. Addition: the new member is added to the new
population.

4. Alteration: new generated population is used when
algorithm is re-applied.

5. Test: If the result is convincing, algorithm is
concluded and the last population is presented as the
solution.

6. Cycle: Return to the second step.

As seen above, the structure of the genetic algorithm
is quite general and can be applied to any kind of
problem. Identification of chromosomes is generally
done by the numbers in double set. Members that are
used for the crosswise must be selected among the best
ones. There is no completion criterion of GA. Having a
satisfactory result or guaranteeing the convergence
could be used as criteria for completion of the
algorithm.

The most important parts of GA are the processes of
crossover and mutation. These processes are started
with a unit of probability, and in most of the cases

2
38

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

applied randomly. This helps to get satisfactory results.
A chromosome should include information on solution
that it represents. Each chromosome is set up with
binary series. Each number that named bit in this series
can represent a characteristic of the solution. Or, a
serial, on its own, would indicate a number. Expressing
the chromosomes with the set of numbers in the binary
series is the most common representation form;
however, also integer and real numbers can be used [9] .
The reason for selecting of binary series is as the
following: first of all it is simple, and secondly, it is
processed by the computer easier and faster. The
reproduction process is a process which is applied
according to certain selection criteria to reproduce new
generation. A selection criterion takes the compatibility
as a basis and selects the compatible members. At later
stage, it is possible that more compatible new members
emerge from those members that are subjected to
crosswise and fission. All members may be selected in
terms of compatibility or some are selected randomly
and transferred to next generation. Crossover can be
applied after the decision for representation of
chromosomes taken. Crossover is a process which is
applied through the deduction of some genes from
parents to create new members. There is a need of
selection of individuals to constitute parents. According
to theory the fittest individuals must be survive to leave
descendants. This selection can be based on several
criteria. Examples are Roulette selection, Boltzmann
selection, tournament selection, sorted selection [8].

Genetic algorithms, as powerful and broadly
applicable stochastic search and optimization
techniques, are the most widely known types of
evolutionary computation methods today. Recently, the
genetic algorithm community has turned much of its
attention to optimization problems in the field of
communication and computer networks, resulting in a
fresh body of research and applications [10], [11], [12],
[13], [14], [15], [16].

II. Genetic Approach for Solving the Shortest
Path

Routing problem is one of the most important issues
facing the development, improvement and performance
of communication networks. Many ideas and methods
have been proposed to solve routing problems [17],
[18], [19], [20], [21]. One of the most common
problems encountered in networks is the shortest path
problem. The shortest path problem is defined as that of
finding a minimum length or cost path between a given
pair of nodes. Shortest path problem is a classical
research topic.

A. Problem Formulation

An undirected graph G = (V,E) comprises a set of

nodes V={1,2, ,n} and a set of edges ^G VxV

connecting nodes in V. For each edge in the graph, there
is a nonnegative number Cij represents the cost, distance,
and others of interest, from node / to node j. A path
from node / to node j is a sequence of edges
{i,l),(l,m),....,(k,j) from E in which no node appears
more than once. A path can also be represented using a
sequence of nodes (i,l,m...k,j). Xy is an indicator
variable. It is equal 1 if the edge (i,j) is included in the
path and zero otherwise.

The shortest-path problem is formulated as follows:

minimize

/w^ZEs-^.

subject to ^x.. <% VieV

(1)

(2)

X X, > X,, , V(i,k) e £, Vi e V\{ 1 ,n} (3)

E^iy=E^y«=l' ^(i'J)^

(4)

where constraints (2) and (3) together imply that any
node other than nodes land node n has either or 2
nonzero incident edges. Constrain (4) makes node land
n the endpoints of the path [9].

B. Encoding The Path Of The Graph

The difficult part of developing a genetic algorithm
for this problem is how to encode a path in graph into a
chromosome. A priority based encoding method can
potentially represent all possible paths in a graph.
Besides, the capability of representing a path for
undirected graph is one of the most significant features
of the priority based encoding method. This difficulty
of finding a representation for the paths for a graph in
genetic approach arises because of the existence of
variable number of nodes in the paths, and the available
number is n-1 for an ^-order graph. Also, a path
consists of consecutive prepared number of edges. A
random sequence of edges doesn't correspond to a path.
These problems were solved by encoding some guiding
information for establishing a path in a chromosome but
not a path itself

Any chromosome is defined by two factors. The first
is locus that represents the position of gene in the
structure, while the second is allele represents the value

3
39

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

of gene. Simply, the node is represented by locus and
the priority of representing the node in constructing a
path with respect to other ones.

The mapping between encoding and path is many-to-
one, which means that different chromosomes may
produce an identical shortest path. But it is easy to
prove that the probability of occurrence of many-to-one
mapping is very low [8]. Therefore, in most cases, there
are no trivial genetic operations associated with the
encoding. It is also easy to verify that any permutation
of the encoding corresponds to a path, so that most
existing genetic operators can easily be applied to the
encoding. Also, any path has a corresponding encoding;
therefore, any point in solution space is accessible for
genetic search.

A path growth procedure is considered to generate a
path from chromosomes. The procedure generates a
path from initial node to end node by appending
eligible edges into the path consecutively. At each step,
several edges are considered. The edge added to a
partial path is always the edge incident to the node with
the highest priority, which extends the partial path from
the terminal node [9] .

C. Genetic Algorithm Procedure

Chromosomes are evaluated using a measure of fitness
during each iteration of a genetic algorithm. There are
three major steps included in the evaluation:

1 . Convert the chromosome to a path.

2. Calculate the objective values.

3. Convert the objective values to fitness values.

The roulette wheel approach, a type of fitness-
proportional selection, was adopted as the selection
procedure. The elitist method was combined with this
approach to preserve the best chromosome in the next
generation and overcome stochastic errors of sampling.
With the elitist selection, if the best individual in the
current generation is not reproduced in the new
generation, one individual is removed randomly from
the new population and the best one is added to the new
population [7], [8].

Genetic approach uses the comprise approach based
fitness assignment in solving the shortest path problem.
The compromise approach is regarded as a type of
mathematical formulation of goal-seeking behavior in
terms of a distance function. The compromise approach
identifies solutions that are closest to optimum. The
overall genetic approach is illustrated as follows:

• Step 1 : The data are entered and genetic parameters
are set.

• Step 2: The initial population is generated
randomly.

• Step 3: The paths are encoded into chromosomes.

• Step 4: The parent chromosomes are selected

• Step 5: The objectives are calculated and evaluated
for each individual.

• Step 6: The next generation is produced by
applying crossover and mutation to the parent
chromosomes.

• Step 7: The new generations replaced the current
ones.

• Step 8: If the maximum generation is reached then
procedure is stopped; otherwise; go to step 4.

III. Simulation and Results

Simulation experiments are carried out for a number
of networks with different size. The objective is to
investigate and evaluate the performance of the genetic
algorithm and Dijkstra algorithm as well using different
simulated networks .The sizes of simulated networks
considered are 30, 50, 100, 120 and 150 nodes.
Assumptions for cost links are considered for the
simulated networks. The topology of 30-node simulated
network used is illustrated in Figure 1 .

Fig. 1. The topology of 30-node simulated network

It should be pointed out that the performance of the
genetic algorithm depends on the choice of the
parameters like population size, crossover rate and
mutation rate. In this simulation, various values of these
parameters are experimented. The optimal values are
found to be 800, 0.6 and 0.1 for population size, cross
over rate and mutation rate respectively as illustrated in
Table I.

TABLE I. The Initial Values for Genetic Algorithm

Parameter

Value

Population size

800

Mutation rate

0.6

Cross over rate

0.1

Number of generation

100

4
40

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

As an example, Table II shows the source and
destination for each pair of nodes in the 30-node
simulated network to be applied on both genetic
algorithm and Dijkstra algorithm.

TABLE II. The Four Source-destination Pairs

FOR 30-NODE SIMULATED NETWORK

TABLE IV. The Results of Applying the Genetic Algorithm
ON THE 2 ^° Route in 30-node Simulated Network

Case

Source
node

Destination
node

Nil

N12

N25

N29

N22

Path

Cost

Running

time in

sec

Generation
number

N8-N6-N24-N23-N2-
N22-N4-N25

67.16

10.034

200

N8-N24-N6-N28-N14-
N11-N25

62.39

15.786

300

N8-N28-N14-N11-N4-
N25

57.70

21.983

400

N8-N2-N15-N4-N25

43.70

25.556

500

N8-N6-N22-N11-N25

25.54

34.984

700

N8-N6-N22-N11-N25

25.54

45.658

900

N8-N6-N22-N11-N25

25.54

48.772

1000

Table III shows the results of implementing the
genetic algorithm for the first route in the 30-node
simulated network. Table IV shows the results of
implementing genetic algorithm for the second route in
the 30-node simulated network. The results of
implementing genetic algorithm for the third route in
the 30-node simulated network is illustrated in Table V.
Table VI shows the results of implementing genetic
algorithm for the fourth route in the 30-node simulated
network. Figure 2 illustrates the variation of cost with
respect to number of generation for the four routes in
the 30-node simulated network.

Table VII illustrates the results of applying Dijkstra
algorithm on the four pairs of nodes in the 30-node
simulated network.

TABLE V. The Results of Applying the Genetic Algorithm
ON THE 3^° Route in 30-node Simulated Network

Path

Cost

Running
time in sec

Generation
number

N6-N8-N28-N16-N27-
N12-N19-N29

71.71

10.656

200

N6-N8-N16-N27-N9-
N12-N19-N17-N29

53.43

14.331

300

N6-N8-N16-N27-N12-
N19-N17-N29

50.26

19.711

400

N6-N8-N16-N27-N9-
N12-N29

49.24

25.000

500

N6-N8-N28-N16-N27-
N12-N29

46.07

29.886

600

N6-N8-N28-N16-N27-
N12-N29

46.07

33.223

700

N6-N8-N28-N16-N27-
N12-N29

46.07

38.119

800

TABLE III. The Results of Applying the Genetic Algorithm
ON THE 1^^ Route in 30-node Simulated Network

TABLE VI. The Results
ON THE 4™ ROUTE

OF Applying the Genetic Algorithm
IN 30-NODE Simulated Network

Path

Cost

Running
time in sec

Generation
number

N11-N20-N18-N7-N1-
N10-N9-N12

114.82

4.449

100

N11-N22-N14-N18-N10-

N1-N12

91.30

10.556

200

N11-N22-N4-N25-N3-
N10-N1-N12

88.66

14.008

250

N11-N22-N2N-N8-N16-
N27-N12

77.46

19.892

400

N11-N3-N10-N9-N12

59.18

29.327

600

N11-N3-N10-N9-N12

59.18

34.563

700

N11-N3-N10-N9-N12

59.18

39.437

800

Path

Cost

Running

time in

sec

Generation
number

N0-N21-N13-N7-N18-
N14-N28-N6-N22

116.67

15.980

300

N0-N1-N19-N29-N12-
N27-N16-N28-N14-N22

96.65

22.879

400

N0-N13-N5-N7-N18-
N14-N11-N22

79.94

27.870

500

N0-N13-N7-N18-N14-
N11-N22

71.63

40.676

800

N0-N1-N10-N8-N6-N22

66.06

51.922

1000

N0-N1-N10-N8-N6-N22

66.06

54.398

1100

N0-N1-N10-N8-N6-N22

79.94

60.933

1200

5
41

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

150.00 I
J 00.00 ■

"* 50.00 ■
0.00 ■■

st versus generation number for the first route

100 200 250 400 600 700 800
Generation number

Cost versus generation number for tiie tiiird route

200 300 400 500 600 700 800
Generation number

Cost versus generation number for the second route
80.00 T

200 300 400 500 700 900 1000
Generation number

Cost versus genertaion number for the 4th route

300 400 500 800 1000 1100 1200
Generation number

Fig. 2. Variation of cost with respect to number of generation
for 30-node simulated network

network size. On the other side, Dijkstra algorithm is
not affected with the increase of the network size.

TABLE VIII. Comparison Between the Genetic Algorithm
AND Dijkstra Algorithm in Accordance to Running Time

Simulated
Network

Average running time in sec

Genetic
algorithm

Dijkstra
algorithm

30-node network

36.530

0.126

50-node network

43.830

0.093

100-node network

48.440

0.112

120-node network

55.135

0.137

1 50-node network

69.055

0.106

TABLE Vll. The results of applying Dijkstra Algorithm on
THE 4™ Route in 30-node Simulated Network

Case

Path

Cost

Time
in sec

N11-N3-N10-N9-N12

59.18

0.125

N8-N6-N22-N11-N25

25.54

0.058

N6-N8-N28-N16-N27-
N12-N29

46.07

0.086

N0-N1-N10-N8-N6-N22

66.06

0.172

It is pointed out that the genetic algorithm can find
the optimum path that is exactly obtained by applying
Dijkstra algorithm. Furthermore, various candidate
paths close to optimum are obtained using the genetic
algorithm. Also, the associated time to get various
paths and generation number are illustrated in these
tables. It is noted that increasing the generation number
leads to get paths more close to optimum as shown in
Figure 2. Also, paths more close to optimum are
consumed more time than others.

The same results are achieved using simulated networks
with size of 50, 100, 120 and 150 nodes.
The only concern during implementing the genetic
algorithm is that genetic algorithm needs much running
time compared to Dijkstra algorithm, this is illustrated
in Table VIII. It is noticed that Dijkstra algorithm is
working efficiently and it is implemented in the
permitted time with network with size till 150 nodes.
Also, the relation between the average running time of
the genetic algorithm and the network size is illustrated
by applying the genetic algorithm on different
simulated networks of 30, 50, 100, 120 and 150 nodes.
It is noticed from the results shown in Figure 3 that the
running time needed to implement the genetic
algorithm is increased as result of increasing the

Average running time of genetic algorithm versus

Average running time of Dijkstra algorrthm versus

network size

<D 0.160 n

1 0.140 ■

ID 80.000 ■

ningt

^^^^

^ 0.120 ■
c 0.100 ■

*\^.--^"'^^^^

^ S 40.000 ■

r-""^^^^^

§ S 0.080 ■
^ .e 0.060 ■

'i 20.000 ■

^ 0.040.
^ 0.020 ■

1 0.000 \

< 0.000 \

30 50 100 120 150

node node node node node

Metv^orksiie

Network size

Fig. 3. The average running time of the genetic algorithm and
Dijkstra algorithm using five simulated networks with different
sizes

Therefore, the genetic algorithm is considered to be a
promising algorithm that produces multiple paths close
to optimum in addition to the optimum path itself and
also can combine with other algorithm to form a hybrid
technique that can be used in multiple applications and
environments.

IV. Conclusion

In this study, the analysis and performance
evaluation of the genetic algorithm that used to solve
the shortest path problem are presented. The genetic
algorithm is experimented using different simulated
networks with sizes of 30, 50, 80, 100,120 and 150
nodes. A comparison between both the genetic
algorithm and Dijkstra algorithm is illustrated .It should
be pointed out that the performance of the genetic
algorithm depends on the choice of the parameters like
population size, crossover rate and mutation rate. In the
simulation, various values of these parameters are
experimented before selecting the ones that achieve the
desired results. The desired values of population size,
crossover rate and mutation rate parameters are 800, 0.6

6
42

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

and 0.1 respectively. From the testing and results
analysis, the following results are achieved:

1. Genetic algorithm is able to find the optimum
solution achieved by Dijkstra algorithm.

2. Increasing the number of generation leads to
obtaining paths more close to optimum

3. Genetic algorithm achieves the desired results with
much running time comparing to Dijkstra algorithm

4. The running time of implementing the genetic
algorithm is increased as a result of increasing the
network size.

5. Dijkstra algorithm is working efficiently and it is
implemented in the permitted time with network with
size till 150 nodes

6. Genetic algorithm is able to find alternative paths
close to the shortest path; consequently, these results
can be used in establishing routing table for nodes.
These alternative paths can be generated using different
genetic operators, which can be invoked at a specific
probability. As a result, the load and utilization of the
paths in communication networks will be slightly
reduced and load balance can be achieved using
multiple paths close to optimum to route traffic between
two nodes.

References

[1] D. Awduche, A. Chiu, A. Elwalid, I. Widjaja, and X. Xiao,

"Overview and principles of Internet traffic engineering",

Internet RFC 3272, May 2002.
[2] [2]A. Montes, "Network shortest path application for optimum

track ship routing", California, 2005.
[3] S. F. Wu, F. Y. Wang, Y. F. Jou, and F. Gong, "Intrusion

detection for link-state routing protocols". In IEEE Symposium

on Security and Privacy, 1997.
[4] M. T. Goodrich, "Efficient and Secure Network Routing

Algorithms", Johns Hopkins University, 2001.
[5] E. Dijkstra, "A note on two problems in connection of graph".

Numerical Mathematical, 1:269-271, 1959.
[6] M. Pioro, D.Medhe, Routing, Flow, and Capacity Design in

Communication and Computer Networks, Morgan Kauffman

series, 2004.

[7] Y. Li, R. He, Y. Guo, "Faster Genetic Algorithm for Network

Paths", The Sixth International Symposium on Operations

Research and Its Applications (ISORA'06), China, 2006.
[8] M. Gen, R. Cheng, Genetic Algorithms & Engineering

Optimization, Wiley Series in Engineering and Automation,

2000.
[9] D. Goldberg, Genetic Algorithms in Search, Optimization, and

Machine learning, Addison Wesley, 1989.
[10] P. Sateesh, S. Ramachandram," Genetic Zone Routing Protocol"

Journal of Theoretical and Applied Information Technology,

2008.
[11] M. R. Masillamani, A. V. Suriyakumar, G. V. Uma, "Genetic

Algorithm for Distance Vector Routing Technique", AIML 06

International Conference, Sharm EL Sheikh, Egypt, 2006.
[12] N. Selvanathan and W. Jing, "A genetic algorithm solution to

solve the shortest path problem in OSPF and MPLS", Malaysian

Journal of Computer Science, Vol.16 No. 1, 58-76, 2003.
[13] J. Cunha, "Map algorithm in routing problems using genetic

algorithm, IFORS Triennial Conference - Edinburgh/Scotland,

2002.
[14] B. Fortz, M. Thorup, "Internet traffic Engineering by

Optimization OSPF Weights", Proc. IEEE Infocom, 2000.
[15] M. Ericsson, M.G.C. Resende, P.M. Pardalos, " A Genetic

Algorithm for the weight setting problem in OSPF routing".

Journal of Combinatorial Optimization, 2001.
[16] O. Akbulut, O. Osman, O. N. Ucan, "Computer Network

Optimization Using Genetic Algorithm", The journal of

Electrical & Electronics Engineering, Istanbul University, Vol.

6, 245-250, 2006.
[17] D. Staehle, S. Koehler, U. Kohlhhass, "Towards an optimization

of IP Routing by Link Cost Specification ", University of

Wurezburg, 2000.
[18] J. Milbrant, S. Koehler, D. Staehle, L. Berry, " Decomposition

of Large IP Networks for Routing Optimization", Technical

Report 293, University of Wurezburg, 2002.
[19] G. L. Li and P. W. Dowd , " An analysis of network

performance degration induced by workload fluctuations",

IEEE/ACM Trans, on networking, 3, No. 4, 433-440,1995.
[20] F. Xiang, L. Junzhou, W. Jieyi, G. Guanqun, " QoS routing

based on genetic algorithm". Computer Communication 22,

1392-1399, 1999.
[21] K. Vijayalakshmi, S. Radhakrishnan, " Dynamic Routing to

Multiple destinations in IP Networks using Hybrid Genetic

Algorithm" International Journal of Information Technology,

2008.

7
43

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

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

Vols, No.3, 2010

Testing Equivalence of Regular Expressions

Keehang Kwon

Department of Computer Engineering

Dong-A University

Busan, Republic of Korea

khkwon@dau.ac.kr

Hong Pyo Ha

Department of Computer Engineering

Dong-A University

Busan, Republic of Korea

hompoyo@hotmail.com

Abstract — We propose an algorithm that tests
equivalence of two regular expressions. This
algorithm is written in the style of a sequent proof
system. The advantage of this algorithm over
traditional algorithms is that it directly captures the
real essences regarding language equivalence. As a
consequence, our algorithm extends easily to other
larger languages with variables.

Keywords- regular expression, equivalence,
language, algorithm.

l.INTORDUCTION

easy to understand, nondeterministic and some
resemblance to the proof theory of intuitionistic
linear logic [2].

In addition, it is a simple matter to observe that this
algorithm extends well to more general grammars.
Our algorithm thus captures the essence of
language equivalence. In this paper we present our
algorithm, show some examples of its workings, and
discuss further improvements. The remainder of this
paper. The remainder of this paper is structured as
follows. We describe our algorithm in the next
section. In Section 3, we present some example.
Section 4 concludes the paper with some
considerations for further improvements.

Regular expressions [1] have gained much interest
for applications such as text search or compiler
components. One central test related to regular
expressions are to test, given two regular
expressions, whether two regular expressions are
equivalent. The equivalence of regular expressions
is useful for simplifying regular expressions. For
example, + 01* can be simplified to 01*.
Unfortunately, the traditional algorithms [4] focus
too much on the equivalence of regular expressions
and finite automatas. As a consequence, the
resulting algorithm is not so intuitive, as they
convert regular expressions to finite automatas for
testing equivalence.

Furthermore, this technique does not extend well to
other grammars such as context-free grammars such
as context-free grammars.

This paper introduces an algorithm for testing
equivalence of regular expressions. It is simple.

2.THE LANGUAGE

The -free regular expression is described by G-
formulas given by the syntax rules below:

G::=s\a\G^G\G+G\G"

In the rules above, a represents the set {a}, s
represents { s }. represents the empty set. F*G
represents the concatenation of two sets F and G.
The Kleene closure of G-G^- indicates there are
many number of G, i.e., G*...*G. We write GG in
place of G*G.

The regular expressions have a number of laws for
their equivalence.

For instance, is the identity for union: 0-\- L and
L +0= L where L is any regular expression.

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

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

Vol.8,No.3,2010

Similarly, s is the identity for concatenation: sL = w is in a regular expression r can be converted to

L6=L. the question of whether w is a subset of r.

On the other hand, is the annihilator for

concatenation : 0L=L 0= 0.

The question of whether two regular expressions are
equivalent is quite complex. We will present an
algorithm for this task in the style of a proof system.
Let G be a regular expression and /"i,. . .,/"n be a list

of regular expressions. Then A, • • • , ^n \- G - i\\^
notion that G is subset of the concatenation of
A,. . ., A - is defined as follows :

3. Example

This section describes the use of our algorithm.
An example of the use of this construct is
provided by the following equivalence: (/ )' =
/'.This equivalence follows from the facts that
(/')' is a subset of /' and vice versa. These are
derived below.

Algorithm for Subset Relation

r,p ■ ^\-A

\-€

(// h/*-

The proof of this as follows:

I \- I - Axiom

■Axiom — =r- — r^ — ^^
Php—ti;

r,v\-A ,

p\-p
r,p-\-^'\-A

W K

r,P,P^^^ I:P^OL

r,p \-A

^^^ WL ^±^PR

r,p^A

r \-p

In the above rules, F, A denote a list of regular
expressions and p,(p denote a single regular
expression. In proving p • (p from F, it splits F
into two disjoint parts.. In dealing with p
construct, the proof system can either discard it,
or use p at least once.

The above algorithm also tests membership in a
regular language: the question of whether a string

(1)/* |- / -DL
(2) / ' |- / * -PR

(3)/*h (/*/ -PR

The other direction, / ' |- (/ ' ) ' is proved below.

/ |- /

Axiom

(1)/* |- / -DL
{2)V\- V -PR

0)1' h(/*/ -PR

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

As a second example, we will show that r-\-s = s-\-r,
To do this,we have to show two things.

(a) (r+s) D (s+r)

(b) (s+r) D (r+s)

We have a proof of (a) below:

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

Vols, No.3, 2010

new rules which introduce universal quantifications...

shs +^_r + >^+^

r + sh s " r + 5l- r

T + S\-S+T

-R

Regarding the performance of our algorithm,
nondeterminism is present in several places of this
algorithm. In particular, there is a choice concerning
which way the text is split in the^goal. Hodas and
Miller[3] dealt with the goal rs by using lO-model in
which each goal is associated with its input resource
and output resource. The idea used here is to delay
this choice of splitting as much as possible. This
observation leads to a more viable implementation.

Our ultimate interest is in a procedure for
carrying out computations of the kind described
above. It is hoped that these techniques may lead to
better algorithms.

We have to proof of (b) below

5 .Acknowledgment

r\-r

s + s

s + rh r

s + rhs

s-i-r\-T-\- s

+ E

This paper was supported by Dong- A University
Research Fund.

6.REFERENCES

4.CONCLUSION

[1] S.C. Kleene, Introduction to Metamathematics,
North Holland,Amsterdam,1964.

We have described an algorithm for testing
equivalence of 0-free regular expressions. The

advantage of this algorithm is that it directly
captures the real essences regarding language
equivalence. As a consequence, it extends easily to
other larger language classes such as context-free
languages. For example, our algorithm extends
easily to the one that deals with algebraic laws, i.e.,
regular expressions with variables. Two regular
expressions with variables are equivalent if whatever
expressions we substitute for the variables, the result

are equivalent. For example,VZVM(Z+M =M +Z).

Algebraic laws are a useful tool for simplifying
regular expressions. To deal with algebraic laws, it
is a simple matter to extend our algorithm with some

[2] J.Y. Girard, "Linear
Computer Science,vol.50,pp

loffic". Theoretical
.r-10i,1987.

[3] J.Hodas and D. Miller "Logic programming in a
fragment of intuitionistic linear logic '' Journal
of Information and Computation, r992. Invited
to a speical issue of submission to the 1991
Lies conference.

[4] J.E.Hopcroft, R.Motwani, and J.D. UUman,
Automata Theory.Languages and Computation,
Addison Wesley, zOOo:

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

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

Vol 8, No. 3, 2010

CRS, a Novel Ensemble Construction Methodology

Navid Kardan

Computer Engineering Dep.

lUST

Tehran, Iran

n kardan@comp.iust.ac.ir

Morteza Analoui

Computer Engineering Dep.

lUST

Tehran, Iran

analoui @iust.ac.ir

Abstract — Constructing ensemble classifiers that are both
accurate and diverse is an important issue of research and
challenging task in machine learning. In this paper, we proposed
Class-based Random Subspace (CRS) method; a new ensemble
construction method based on the random subspace (RS)
strategy, and tested it on a number of standard data sets from
UCI machine learning repository. Our results show that CRS is
at least as good as RS, and outperforms it in datasets with strong
correlation between their classes.

Random subspace method; Feature selection; Classifier
ensemble; Classification

Introduction

The general methods for constructing ensemble classifiers
can be applied to any classification algorithm. These methods
try to create feature spaces that are as diverse as possible, from
the original data set. The most successful of these methods are
Boosting [1], Bagging [2], Random subspace [3], and decorate
[4]. Another possibly promising approach in this area is the
nonlinear transformations of the original data set [5].

In Boosting, every data instance is picked with a probability
according to its hardness or misclassification on earlier
classifiers of the ensemble. This way the harder instances that
need more effort can gain more notification and we can get
more diverse ensembles.

In bagging, we make new data sets by sampling with
replacement from the original data set. This way we have new
data sets that have enough data points and introduce some
degree of diversity in them, as well.

In decorate, some artificial instances are added to the
original data set for training a new member. The class label of
these new instances is set with a value that is different from
current ensemble decision. This way we strive for a better
community with increasing diversity.

The random subspace method samples features of the
original data set, in order to improve diversity, instead of
sampling from instances. This way, we have new data sets that
have the same amount of data as the original, therefore, can get
better classification accuracies. In some applications, such as
biometric data processing, this method has proven very good
performance [6]. In this paper we improve this approach by
taking into account the classes that data items come from. We

show that an ensemble created this way, permits to obtain
better accuracies than that obtained by a classic RS method.

II. Methods

An ensemble of classifiers combines different learners in
order to obtain a better overall performance. These methods
build new data sets from the original data and build the
classifiers on these modified data sets. Then the final decision
is obtained by combining all of the individual votes.

Suppose a data set with N instances and F features. In the
original RS method we build K new data sets, where K is the
number of base classifiers in the ensemble, by sampling
randomly from the F features and obtain K new training sets
each one with N training instances. Then we train each learner
with one of these training sets.

The main idea behind this method is the possible
redundancy in the original feature space. With sampling from
this space we select a subset of the base space and so it is likely
to find a space that can lead to better generalization. This
improvement in accuracy is obtained from combining these
different projections of the feature space. Another advantage of
this method is the reduced feature space that can tackle the
problem of curse of dimensionality. In fact, Vapnic [7] showed
that reducing the feature space is one method for gaining better
generalization ability.

Ho in [3] showed that RS can improve accuracy of the tree
classifiers considerably, by building a random forest.

RS method treats all features the same. It is also a top-down
approach, i.e. it selects the new features at the start of training
phase but it does not consider each instance individually. We
compensated the original idea of RS according to the fact that
the redundancy of data set can be class-based. This means,
although there might be better subsets of the feature space that
can lead to better classification, but if we consider the data as a
whole, we might lose some of these redundancies that are
between some parts of the original feature space. These parts of
feature space, in our work, correspond to the different classes
of the original data set.

For having a better understanding of the situation, consider
the case that we have three classes; say A, B, C in the data set.
Class A can be best discriminated from other classes with

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

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

Vol. 8, No. 3, 2010

feature subset As and the same is true for classes B and C. in
the RS method, we suppose that As, Bs and Cs are identical and
we can get better results by selecting the subsets that can better
approximate the best subset. But in the broader case when these
subsets are different, we can get stuck in the situation that leads
to no better performance. By taking into account this difference
between feature subsets we can get better ensembles and higher
accuracies.

Fig. 1 shows a data set with five features and three class
types that is sorted according to the class label.

Qassl

aass2

aass3

QassB

Fig. 1. A sample data set with five features and three class
types.

In figure 2 one possible data set according to RS strategy is
presented. In this figure the omitted feature is shown by ///
mark. Figure 3 present one possible feature subset according to
our method. As depicted in this figure, here we use different
feature subsets for each class.

Based on this idea we introduce the idea of Class-based
Random subspace (CRS) that can be used to build classifier
ensembles that are more likely to have better performance. In
CRS every classifier is responsible for one class only and the
final decision is obtained by combining different decisions
according to their corresponding class.

One important issue about our strategy is the way that we
make it possible to have different subsets for different classes.
Let we name a specific feature subset as /, feature space as F
and number of class types as k. in our method, for each
ensemble member we create k sub-feature sampling i.e.
m={fj,f2, ... , fk}- Each sub-feature/ will be used to train a
classifier that identifies one class. According to this one-vs.-all
strategy [8] we will make k classifiers for each unit of our
ensemble. If our ensemble consists of 1 members, M = {mi, m2,
..., m/}, we will need kxl random samples from F. In test
phase we will count the votes for each class and select the most
popular class.

Qassl

Class 1

Qassl

Qass2

Class 2

Qass2

QassB

Class 3

QassB

Fig. 2. Selecting three random features to construct a new
data set according to RS strategy

III

Class 1

III

Class 1

III

Class 1

III

Class 2

III

Class 2

III

Class 2

III

Class 3

III

Class 3

III

Class 3

Fig. 3. Selecting three random features to construct a new
data set according to our strategy

Our algorithm can be described as follows:

• Build k data sets from the original data set, where k is
the number of classes and each set is built for one of these
classes; i.e. each set is exactly the original data set except for
the class label that can only get two values, one for belonging
to the corresponding class and another for belonging to other
classes.

• Using each of these new data sets, choose a subset of
features randomly and project each instance to this space to
build a new data set for each base classifier.

• Train each classifier with its training data and
remember its corresponding class.

• For classification of a new instance, any combining
method can be used to obtain the final decision of the
ensemble. For instance, majority vote can add any vote to its
corresponding class and choose the class with most votes.

III. Experiments

We used 5 datasets from UCI machine
[9, 10] to evaluate our proposed method,
method with the original RS method for
sizes and the results show that our method
that are at least as good as the RS
improvements in some cases.

learning repository
We compared our
different ensemble
leads to ensembles
and show major

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

We used Vehicle, Vowel, Soybean, Glass and DNA
datasets as our test benchmarks. Vehicle dataset has 846
instances and 19 features including class label. It is used for
identifying four types of vehicles from their silhouette. Vowel
dataset consists of 990 instances, each with 10 features. It has
11 class types that are the steady state vowels of British
English. Soybean dataset has 683 instances with 35 features for
each and a class label which may have any of the 19 different
values. Glass dataset has 214 observations of the chemical
analysis of 7 different kinds of glass. It has 10 features
including class label. DNA data set has 181 features including
class label and has 3186 instances.

All the experiments are done in R software. Our base
classifier is the tree classifier implemented in the tree package

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

Vol 8, No. 3, 2010
of this software [11]. For combining decisions of ensemble
members, we used SUM rule [12]. This method computes sum
of the decisions of ensemble members for each class and select
the class with most value.

For estimating testing error, we used leaving-one-out
method [13]. We repeated this procedure for thirty times and
averaged the results. Tables I-IV show statistical properties of
the number of correctly classified instances for each data set.
Each column represents the result of one method for a
particular ensemble size. The ensemble size is written in
parentheses. Table V represents statistical properties of the two
ensemble classification accuracies on the DNA data set.

TABLE I.

Comparison of RS and CRS on Glass dataset using number of instances that classified correctly

TABLE II.

TABLE III.

TABLE IV.

Glass

CRS(5)

RS(5)

CRS(IO)

RS(IO)

CRS(15)

RS(15)

CRS(20)

RS(20)

Mean

156.9

152.93

159.17

152.13

160.27

151.83

160.43

150.77

MAX

168

163

167

162

167

162

165

156

MIN

150

144

152

146

150

142

154

143

Std. Dev.

4.41

4.62

3.33

4.83

3.62

4.92

2.91

3.57

Comparison of RS and CRS on Soybean dataset using number of instances that classified correctly

Soybean

CRS(5)

RS(5)

CRS(IO)

RS(IO)

CRS(15)

RS(15)

CRS(20)

RS(20)

Mean

522.2

513.93

522.4

512.23

525.47

511.87

521.97

511.17

MAX

554

542

534

522

543

521

534

514

MIN

508

506

512

503

511

505

510

504

Std. Dev.

11.49

6.90

6.51

2.88

8.26

3.05

6.54

2.69

Comparison of RS and CRS on Vehicle dataset using number of instances that classified correctly

Vehicle

CRS(5)

RS(5)

CRS(IO)

RS(IO)

CRS(15)

RS(15)

CRS(20)

RS(20)

Mean

614.6

614.47

618.6

616.17

619.87

619.03

623.7

618.53

MAX

638

626

630

626

632

629

636

628

MIN

594

599

606

600

605

600

609

610

Std. Dev.

13.01

7.27

7.57

7.87

6.18

6.85

5.26

4.99

Comparison of RS and CRS on Vowel dataset using number of instances that classified correctly

Vowel

CRS(5)

RS(5)

CRS(IO)

RS(IO)

CRS(15)

RS(15)

CRS(20)

RS(20)

Mean

788.13

558.23

812.1

574.53

819.37

578.5

825.67

578.8

MAX

809

605

834

662

833

634

843

626

MIN

764

499

788

525

800

524

814

532

Std. Dev.

12.14

30.63

11.54

29.53

9.10

24.58

7.55

21.33

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

TABLE V.

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

Vol. 8, No. 3, 2010
Comparison OF RS AND CRSonDNAdataset using classification accuracy

DNA

CRS(5)

RS(5)

CRS(IO)

RS(IO)

CRS(15)

RS(15)

CRS(20)

RS(20)

Mean

0.9341

0.9060

0.9393

0.9086

0.9376

0.9087

0.9386

0.9111

MAX

0.9523

0.9303

0.9517

0.9222

0.9504

0.9253

0.9523

0.9297

MIN

0.9134

0.8901

0.9259

0.8901

0.9165

0.8769

0.9259

0.8883

Std. Dev.

0.0090

0.0101

0.0067

0.0096

0.0094

0.0107

0.0062

0.0096

Figures 4-8 show the average ensemble accuracy vs.
ensemble size for each data set. Here we can see that CRS was
always at least as good as RS. In Vowel data set, CRS performs
much better. In soybean glass and DNA it is considerably
better. And in vehicle data set, two methods are not so
different.

Fig. 4. The two methods are compared on Vehicle dataset using

ACCURACY VS. ENSEMBLE SIZE

0-775

u./b
D.755

0.7S
0.745

0,74
0.735

-RS

Fig. 6. The two methods are compared on Soybean dataset
using accuracy vs. ensemble size

[}.y

0.7€
n 7";

.R5

0.72

0.71

0.7

0.fi9 -h

o.es

-CftS

Fig. 5. The two methods are compared on Glass dataset
using accuracy vs. ensemble size

0,55
0.5

Fig. 7. The two methods are compared on Vowel dataset
using accuracy vs. ensemble size

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

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

Vol. 8, No. 3, 2010

0.95

0".94

0.92
0,91

Fig. 8. The two methods are compared on DNA dataset
using accuracy vs. ensemble size

IV. Conclusion

We proposed a new ensemble construction strategy based
on Random subspace method. In this method we tried to
compensate classic RS method by trying to reduce the
redundancy that might rise from intra-class correlations. If we
have a brief comparison with the RS method and the fact that
there can be intra-class redundancy in feature space, we can
conclude that CRS have the potential to improve overall feature
space in the same way that RS can do it in an unsupervised
manner. We tested our method on four standard data sets and
compared it to the RS method. The results suggest that our
method is at least as good as original RS method, but in some
cases there is a major superiority. These results are according
to the fact that in some data sets there is a stronger correlation
between classes and this can lead to better results.

References

[I] Y. Freund, R.E. Schapire, "Experiments with a new boosting algorithm",
Proceeding of the thirteenth international conference on Machine
Learning Bari, Italy, July 3-6, 1996, 148-156.

[2] L. Breiman, "Bagging predictors", machine learning, 24, 1996, 123-140.
[3] T.K. Ho, "the random subspace method for constructing decision
forests", IEEE Trans. Pattern Anal. Mach. Intell. 20 (8) (1998) 832-844.

[4] P.Melville, R.J. Mooney, "creating diversity in ensembles using artificial
data", in information fusion: Special issue on diversity in multiclassifier
systems", vol. 6 (1), 2004, pp. 99-111.

[5] NicoLas Garc'ia-Pedrajas, Cesar Garc'ia-Osorio, Cohn Fyfe,
"Nonlinear Boosting Projections for Ensemble Construction", Journal of
Machine Learning Research 8 (2007) 1-33.

[6] L. Nanni, A. Lumini, "An experimental comparison of ensemble of
classifiers for biometric data", Neurocomp. 69 (2006) 1670-1673.

[7] V. Vapnic, "the nature of statistical learning theory", springer-verlag,
1995.

[8] E. Alpaydin, "Introduction to machine learning". Second edition, MIT
press, 2010.

[9] D.J. Newman, S. Hettich, C.L. Blake, C.J. Merz, UCI Repository of
machine learning databases

rhttp://www.ics.uci.edu/~mlearn/MLRepository.html 1. Irvine, CA:
University of California, Department of Information and Computer
Science (1998).

[10] A. Frank, A. Asuncion, UCI Machine Learning Repository
rhttp://archive.ics.uci.edu/ml 1 . Irvine, CA: University of California,
School of Information and Computer Science (2010).

[II] B. D. Ripley, "Pattern Recognition and Neural Networks", Cambridge
University Press, Cambridge, 1996.

[12] J. kittler, M. Hatef, R. Duin, J. Matas, "On combining classifiers", IEEE
Trans. Pattern Anal. Mach. Intell. 20 (3) (1998) 226-239.

[13] R.O. Duda, P.E. Hart, D.G. Stork, "Pattern Classification", second ed.,
Wiley, New York, 2000.

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

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

Vol 8, No. 3, 2010

Routing Optimization Technique Using M/M/1
Queuing Model & Genetic Algorithm

Madiha Sarfraz, M. Younus Javed, Muhammad Almas Anjum, Shaleeza Sohail

Department of Computer Engineering

College of Electrical & Mechanical Engineering

Pakistan

madya.khan@gmail.com, myjaved® ceme.edu. pk, almasanjum@yahoo.com, shaleezas@hotmail.com

Abstract — Optimization Approaches have been appHed to
various real Kfe issues in communication and networking. In this
research a new approach has been proposed for network path
optimization using Genetic Algorithm. The path which is best
fitted in the population is considered as the optimal path. It is
obtained after qualifying the fitness function measuring criteria.
The fitness function measures the best fitted path based on
constraints; bandwidth, delay, link utilization and hop count.
Population is composition of valid and invalid paths. The length
of the chromosome is variable. So the algorithm executes
competently in all scenarios. In this paper the comparison of this
approach with the fitness function; measuring delay and
bandwidth factor, has also been catered. This work has been
performed on smaller network; work is in progress on large
network. Thus, the results proved our affirmation that proposed
approach finds optimal path more proficiently than existing
approaches.

I. Introduction

OPTIMIZATION in the field of Genetic algorithm is
gaining massive magnitude. GA is globally used
optimization technique [1] based on natural selection
phenomenon [2]. It is considered as an important aspect in
networking. From source node to destination node essential
solution is optimization which is needed. The network traffic
flow is mounting rapidly. So the balance in Quality of Service
and Broadcast of traffic should be maintained. Increased in
load of network traffic will cause the delay in traffic and affect
the QoS as well. The routing problem scenarios can be
resolved through optimization [3].

In the research the approach which has been introduced is of
network optimization routing strategy using Genetic
Algorithm. It involves bandwidth, delay and utilization
constraints. All these will be in different catering in scenarios.
It digs out the most optimal path from the population lot on the
basis of Fitness Function. The fitness function selects the path
which has less delay, less utilization factor, more bandwidth
availability and less number of hops to be travel. The hop
count is used as the decision making factor when there are
more than one path with same strength. The chromosome
represents the path and the population is showing collection of

feasible and infeasible paths. Each chromosome has variable
number of nodes. This routing strategy is not efficient and
robust merely but it also congregates swiftly.

The paper is formatted as: The related work is done under
section II. Overview of Genetic Algorithm has been structured
in section III. In this the outline of GA and its operators are
explained. Section IV explains the optimization strategy for
routing using GA. In this two fitness functions are compared.
The proposed fitness function includes bandwidth, delay,
utilization and hop count. It is compared with the fitness
function which is catering bandwidth and delay factor. The
results and analysis is illustrated in section V.

II. Related Work

Genetic Algorithm which is a versatile technique designed for
optimization and searching network planning control in Anton
Riedl [4] work. It is also in planning of integration of packet
switched network.

Yinzhen Li, Ruichun He, Yaohuang Guo [20] work on finding
out optimal path with fixed length chromosomes. This is
following priority-based mechanism. In this work Mitsuo. Gen
technique's loophole has also been indicated.

In Carlos A. Coello Coello's tutorial [26] multiple objectives
optimization has been discussed. The pros and cons of these
approaches are done by them. Also illustrate research done
and their implications in the respective field.

Introduction of key-based technology in optimization is done
by Mitsuo Gen and Lin Lin [16]. They have combined
different operators [16]. They high level of search paradigm
leading to improvement in computational time and path
optimization.

Basela S. Hasan, Mohammad A. Khamees and Ashraf S.
Hasan Mahmoud [24] used Heuristic mechanism for Genetic
Algorithm. They have considered single source shortest path.
For searching heuristic approach is used for crossing over also

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

called as recombination and mutation.

In other work of Anton Riedl [19] title represents the research
done is upon path optimization in traffic engineering scenario.
It discusses the implications of network optimization and
traffic engineering. It main focus is routing with multiple
delay and bandwidth constraints for optimization.

Andersson and Wallace [25] proposed a GA which is robust
and requires few numbers of parameters. It main emphasis is
how multiple objectives GA works out on real life scenarios.

An approach has been developed for reducing congestion in
the network [17] by M. Ericsson, M.G.C. Resende and
P.M.Pardalos.

Ramon Fabregat, Yezid Donoso, Benjamin Baran, Fernando
Solano and Jose L. Marzo [23] presented traffic-engineering
load balancing classification. They have not work on packet
loss and other factors like backup paths. They introduced
GMM model.

Diverse and versatile genetic algorithm is proposed in the
work of Norio Shimamoto, Atsushi Hiramatsu and Kimiyoshi
Yamasaki [18]. In this approach last result of the iteration is
used for the next generation. Performance level of the
algorithm is good.

Abdullah Konak, David W. Coit, Alice E. Smith [21] shows
that investigation for solutions is the coherent response for
multiple objectives. A real life scenario entails immediately
multiple objectives for optimization. It also give overview of
GA which are developed for multipurpose objectives and
maintaining the diversity. They have introduced GR (Greedy
Reduction) technique. In the case of worst scenario GR
technique executes in the linear time of framework [22].

For bandwidth allocation Hong Pan and I. Y. Wang [15]
proposed GA for optimization. In this the average delay
network is lessened. Bandwidth can be reassigned again as per
this algorithm.

III. Overview Of Genetic Algorithm (GA)

Genetic Algorithm is a search paradigm. It follows
principles which are based on Darwin Theory of evolution. In
this population data fights for survival and the 'fittest' one
survives. This algorithm is mainly based on natural selection
phenomenon. GA is introduced by John Holland [7] [5]. It is
effective technique as its not only encountering mutation but
also use crossing over technique (or genetic recombination)
[10]. Crossing over technique improves the proficiency of
algorithm for having the optimum outcome. Holland had a
dual aim [11]:

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

Vol 8, No. 3, 2010
• To improve the concept of understanding of natural
selection process

• To develop artificial structure having functionality
analogous to natural system

Genetic Algorithm has been broadly used for solving MOPs
because it works on a population of solutions. [13]. Objective
function plays an important role for obtaining optimum result.
For having "Optimum" result does not mean that the result is
"Maximum". It's the best and the most appropriate value as
per the objective function criteria. Genetic algorithm is best
for optimization and is useful in any state.

A. Population

The Genetic algorithm starts up with a set of solutions which
is taken as a 'population' with the assumption that next
generation will be better than the previous one. The cycle is
terminated when the termination condition is satisfied.

B. Operators ofGA

The efficiency of Genetic algorithm is dependent on the way
the operators are used, that constitutes Genetic algorithm
course of action [9]. The GA operators are as follows [8] [4]:

1) Selection & Reproduction: Chromosomes are selected
according to their Objective Function (also called as fitness
function). The ultimate node for chromosome survival is
Objective Function. It works on Darwinian Theory for
survival of the fittest. This is an artificial version of natural
selection for survival. Chromosomes having higher fitness
have greater likelihood of being into next generation. There
are numerous methods for chromosome selection. For
selection of chromosomes several methods are in practice.

2) Crossover or Recombination: It is the distinguishing
feature from other techniques of optimization. On selected
parents the technique is applied as per their crossing-over
mechanism.

• 1 -point Crossing over

• 2-point Crossing over
Crossing over results into new off-springs.

3) Mutation: In GA maneuvering mutation has secondary
role [6]. It is required after crossing over segment because
there is a probability of information loss at this stage [12]. It is
done by flipping by bit as per requirement. The population
diversity maintenance is purpose of mutation operator.

C Steps to follow in GA

1) Population Generation: Generate population randomly
of'n'.

2) Fitness Function: Launch function for evaluation of
fitness in population.

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

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

3) Applying Operators: Create new population by
applying operators of GA until the new population is
complete. At the end of iteration another generation is
attained. The Rank Based Selection is applied.

a) Selection: Parents selection is on the basis of their
fitness function creteria.

b) Recombination: Recombination also called as
crossing over is than applied over the selected parents. It result
into offsprings.

c) Mutation: The resulting offsprings are than mutated.
After that their fitness function is measured. On the basis of
this value, their survival in the population is based.

4) Terminating Condition: If the terminating condition is
reached, the loop breaks and the best result is obtained.

5) Resulting Generation: Newly generated population is
used for further generation.

IV. Proposed Optimization Routing Approach

In this proposed optimization approach basic unit of a
chromosome is gene. Genes constitute to form a chromosome,
which in turn constitutes the population. Thus the
chromosome in population is represented by string of number
as in Figure 1 .

Figure 1 . Genetic Representation of Path

The gene represents the node while the chromosome
represents the network path. Population is the collection of all
possible paths. The chromosome 1-5-3-6 shows network path
which is constituted by nodes. The first node is the source
node while the last node is the destination node. The hop count
in this will be:

Hop Count = Chromosome Length - 1

The length of the chromosome is variable. So the number of
nodes in the path is directly proportional to number of hop the
data has to travel. The chromosome which have source and
destination same as defined for the path, they constitute the
population lot. The rest unfeasible paths will be discarded after
they are generated. From those feasible paths, the population
is generated randomly.

A. Strategies

The proposed strategy is explained in Fitness Function II and
is compared with Fitness Function I as well. The proposed
strategy is more efficient as it is handling more constraints
than the other approach [27].

Fitness Function I

Vol 8, No. 3, 2010

This fitness function is for finding the delay of the path using
the delay and bandwidth constraints [27]. The algorithm is as
follows:

• It checks the bandwidth availability of link.

• After that network delay of link is find out

• Than delay average of path is calculated.

All the paths are valid, as they have passed through the
bandwidth availability check before calculating the average
packet delay as [27]:

AveragePacketDelay = W

no. of links

The delay of the path is calculated as [27]:

delay = ^

bw Available,

DataSize
The optimum path selection is based

(2)
mainly on the
bandwidth and delay in this approach. Thus, if two or more
paths have same fitness value than the path with less number
of hops will be taken as the optimum path. Optimum path
leads to a path which has more bandwidth, less delay and less
number of hops.
Now the proposed strategy involving is illustrated as follows:

Fitness Function II

The proposed strategy is founded on M/M/1 queuing model
[14] by using GA for Path Optimization. In our research it is
used for handling Bandwidth, Utilization and Delay
constraints for finding Optimum path. The fitness function is
based mainly on (3). If there is a conflict in making decision
among these constraints than path with lesser hops is taken as
the optimum path. Constraints defined are Bandwidth
Availability, Delay, Utilization and Hop count. The fitness
function using M/M/1 queuing model [14] is as follows:

T =

1-p

(3)

T = Delay mean (sec)

l/|i = mean packet size (bits)

C = capacity (bps)

p = utilization factor

1) Capacity V: It is the bandwidth C available over the
network path.

2) Utilization factor 'p': After checking of bandwidth
availability, the network utilization factor is obtained as:

utili, =

DataSize
bwAvailable,

(4)

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

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

After summing up of utilization factor a links, utilization
factor value for a single path is obtained. Than mean of the
utilization factor of path is obtained by dividing the sum of the
utilization factor value for a single path with the total number
of links in that path as:

Mean Utilization = (X Utilization)/total number of links

(5)

3) Delay Mean: The equation (3) is precisely formulized
as follows:

Delay Utilization Mean = {Mean Utilization)! (1- Mean
Utilization)

(6)
It constitutes fitness function of this algorithm:

Fitness Function

Delay Utilization Mean
(7)

This utilization factor is calculated for every path. After each
generation the mean of all paths is also calculated. All the
paths are valid path as their bandwidth availability is checked
before finding out their fitness. Optimum path selection is
based on the link utilization, Bandwidth, Delay Factor and hop
count. The more the bandwidth, the less the delay, utilization
and the hop count the more optimum the path will be. When
there are two or more paths with same fitness value than hop
count will be playing the decision making role.

D. Proposed GA for Path Optimization
Proposed Routing Strategy using GA for network path
optimization is as follows:

Vol 8, No. 3, 2010
applied on the selected chromosomes, leading towards the

segregation of source and destination node. After this 1 -point

crossing over technique is done over the resultant

chromosome. In this the crossing over point is 2 [27].

4) Mutation: After crossing over, mutation is being done
on the offsprings. Mutation has been done as per the scenarios
which are [27]:

a) Scenario 1 (repeating node):

• The location of repeating node is traced out.

• Any of the missing nodes is find out.

• Place the missing node at traced location.

b) Scenario 2 (no missing or repeating node):

• Randomly pick two nodes from chromosome

• Swap those nodes

c) Scenario 3 (minimum chromosome length):

• Minimum length of chromosome is 2

• It has only source and destination nodes

• There will be no mutation

d) Scenario 4 (length of chromosome is one more than
the minimum):

• Length of chromosome one more than minimum
constitutes of 3 nodes

• Only middle node is flipped with missing node

5) Evaluation of Mutated Offsprings: The mutated
offspring's fitness is evaluated. The chromosome with worst
fitness is replaced by the offspring having better fitness. The
population size will remain unaffected by this replacement.
The worst chromosomes are discarded. The network path
survival is based on their Fitness Function Criteria.

1) Initialization of Population :T\\q population is randomly
generated [27]. There also exists the probability of valid and
invalid paths. The source and destination nodes are fixed. The
randomly selected population is 33% of the generated
chromosomes. So this 33% constitutes the whole population
and 30 generations are produced [27].

V. Experiments, Results & Analysis

A. Experiments and Testbed

1) Network Formation: The delay, bandwidth and
utilization factor has been evaluated on the following network:

2) Selection of Parents: Parent's selection is based on
Rank-based Selection. In this the parents are ranked on the
basis of their fitness function value. They are not selected
randomly except the first generation (in which parents are
selected randomly). The parents with high fitness will be
higher in ranking and the ones with low fitness will be lower
in ranking. Both parents which are best in the population are
used for crossing over as they have more tendency of survival
in nature.

3) Recombination (or Crossing Over): Crossing over of
the best fitted chromosome results into best offsprings. The
crossing over technique used is "2 -point over 1 -point
crossover" [27]. In this first 2-point crossing over technique is

Figure 2. Network Formation

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

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

Main factors of algorithm are availability of bandwidth,
utilization, delay and hop count in network traffic. When the
packet is passed through a network, two parameters are
involved. One is time packet takes to reach next node and
other is bandwidth. Time also shows distance packet travels.
The variety of time and bandwidth measures is included in the
network, leading to better result in a complicated network.

2) Population Selection Criteria: The population is
selected randomly. The result of randomly selected
population's result is compared with whole population and
with results of the [15] also. The obtained results are better
than both of them. Hence, proved that random selection is best
choice [27]. The criterion for population size and population
generation is as shown in the following table [27]:

TABLE I.

TABLE SHOWING POPULATION SIZE & NUMBER OF
GENERATIONS

Population Size

33%

Generations

3) Experimental Practice: The population is selected
randomly. First bandwidth availability is checked and then
calculates utilization and delay factor as mentioned in fitness
function. The hop count is being considered for resolving the
conflict between paths with same fitness value. The paths with
less hop count in that case will be considered as an optimum
path. Even if the hop count is same, than select any of the path
which is not used often.

B. Analysis & Results

Fitness I is applied over the whole population. The population
is generated 50 times. The minimum average Delay is taken
till 50 iterations and is shown in the graph below. So, there
will be a delay in getting the optimum result.
The horizontal axis is showing the generation and the vertical
axis is showing the delay average (fitness function value)
against every generation.

delay avg full popu

L4
12

1
0.8
0.6
0.4
0.2

-delay avg full popu

1 4 7 1013161922Z52S31343740434649

Figure 3.

Vol 8, No. 3, 2010
Average Delay (whole population)

In this case number of hops is constant. There is no alteration
in hop count till the 50 generation. It might show change later.
The graph is constant showing no variation.

Now the population is randomly selected and fitness function I
is implemented over it. The results are compared with the full
population selection results. The 33% of the population is
taken and is generated 30 times as per Table I. The minimum
of delay average of every generation is shown in the Figure 4.
Hence, it is showing better results than shown in the Figure 3.
There is a variation in this graph and is reaching the minimum
level of delay.

The horizontal axis is showing the generation and the vertical
axis is showing the delay average (fitness function value)
against every generation with random selection of population.

DelayAvg

0.95

0.85

13 5 7 9 111315 17 19212325 2729313335373941434547 4951

•DelayAvg

Figure 4. Average Delay (randomly selected population)

As per the results obtained the number of hops is also
reducing. The hops are more at the start of the generation, but
as the generations are increasing their number of hops is also
reducing. Therefore, it is involving traffic engineering of the
network as well. So in this case it is sharing the load of
network through delay and bandwidth constraints. In this
utilization factor is not involved in fitness function criteria.

At start the hops count of the paths are more as per the
network scenario, but in final result the number of hops is
lesser. Thus, it is also handling the traffic engineering problem
of the network. This is thus sharing and balancing the network
load as per bandwidth and delay criteria.

Now the proposed strategy (fitness function II) is applied over
the network. Population selection is random [27]. It involves
the delay, bandwidth and utilization constraints. It also
involves hop count.

The horizontal axis is showing the generation and the vertical
axis is showing the delay and utilization measure (fitness

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

function II) against every generation with random selection of
population.

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

Vol 8, No. 3, 2010
Figure 6 is showing comparison of the proposed algorithm

with the other algorithm mentioned in this research.

Utilization n mean delay

■utilization n mean
delay

1 4 7 1013161922252831343740434649

Figure 5. Proposed Algorithm involving utlization and Delay constraints
with randomly selected population

In this graph there is varitaion in fitness value involving delay
and utilization factors of the network. Thus it is reaching the
minimum value more earlier than the previous algorthims
discussed. It is reaching getting value below 1 at ealier stage
which fitness function I graph is showing after mid generation.

In Figure 6 the fitness function which is involving more
constraints are showing better results than compared to the
other fitness function results mentioned in this research paper.
So for calculating the optimum path of the network the Fitness
Function II should be applied over the network for congestion
control and traffic engineering problems.

1.4

■DelayAvg

■utilization n mean delay

•delay avg full popu

1 4 7 10 13 16 19 22 25 28 3134 37 40 43 46 49

Figure 6 . Comparison of both fitness function and respective population
selected (whole and random selection)

VI. Conclusion

Bandwidth scaling is elementary driver of popularity and
growth among interconnected computer networks. Increase in
bandwidth accompanies lesser delay. In this research, the
proposed technique is for finding optimized path with delay,
bandwidth and utilization measures. In this the comparison is
done between the fitness function catering just bandwidth and
delay with the fitness function handling bandwidth, delay and
utilization. Both of them are tackling the hop count. The
results prove our affirmation that proposed algorithm shows
better results than the earlier proposed algorithms.

References

[I] David E. Goldberg, "Genetic Algorithms in Search, Optimization &
Machine Learning", Addison-Wesley Longman Publishing Co., Inc.
Boston, MA, USA, 1989.

[2] A.W.W.NG and B.J.C. Perera, "Importance of Genetic Alrothim
Operators in River Water Quality Model Parameter Optimisation",
school of the Built Environment, Victoris University of Technology,
Melbourne.

[3] Zhao-Xia Wang, Zeng-Qiang Chen and Zhu-Zhi Yuan, "QoS routing
optimization strategy using genetic algorithm in optical fiber
communication networks". Journal of Computer Science and
Technology, Volume 19 , Year of Publication: 2004, ISSN: 1000-9000,
Pages: 213 -217

[4] Anton Riedl, "A Versatile Genetic Algorithm for Network Planning",
Institute of Communication Networks Munich University of
Technology.

[5] Darrell Whitley, "A genetic algorithm tutorial". Statistics and
Computing, 1992, volume 4, number 2, pages 65-85.

[6] David E. Goldberg, "Genetic Algorithms in Search, Optimization &
Machine Learning", Addison-Wesley Longman Publishing Co., Inc.
Boston, MA, USA, 1989.

[7] Darrell Whitley, "A genetic algorithm tutorial". Statistics and
Computing, 1992, volume 4, number 2, pages 65-85.

[8] David E. Goldberg, "Genetic Algorithms in Search, Optimization &
Machine Learning", Addison-Wesley Longman Publishing Co., Inc.
Boston, MA, USA, 1989.

[9] A.W.W.NG and B.J.C. Perera, "Importance of Genetic Alrothim
Operators in River Water Quality Model Parameter Optimisation",
school of the Built Environment, Victoris University of Technology,
Melbourne.

[10] Emmeche C, "Garden in the Machine. The Emerging Science of
Artificial Life", Princeton University Press, 1994, pp. 114.

[II] Goldberg D., "Genetic Algorithms", Addison Wesley, 1988.

[12] David M. Tate, Alice E. Smith, "Expected Allele Coverage and the Role
of Mutation in Genetic Algorithms", Proceedings of the 5th International
Conference on Genetic Algorithms Pages: 31 - 37 Year of
Publication: 1993, ISBN: 1-55860-299-2

[13] Dinesh Kumar, Y. S. Brar, and V. K. Banga, Multicast Optimization
Techniques using Best Effort Genetic Algorithms, World Academy of
Science, Engineering and Technology 50 2009

[14] Yantai Shu, Fei Xue, Zhigang Jin and Ohver Yang, " The Impact of
Self-similar Traffic on Network Delay", Dept. of Computer Science,
Tianjin University, Tianjin 300072, P.R. China, School of Information
Technology and Engineering, University of Ottawa,Ottawa, Ontario
Canada KIN 6N5, 0-7803-43 14-X/98/$10.0081998 IEEE

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

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

[15] Hong Pan, Irving Y. Wang, "The Bandwidth AUocation of ATM
through GA", Global Telecommunications Conference, 1991.
GLOBECOM '91, 1991, pages 125-129.

[16] Mitsuo Gen, Lin Lin , "A New Approach for Shortest Path Routing
Problem by Random Key-based GA", Genetic And Evolutionary
Computation Conference Proceedings of the 8th annual conference on
Genetic and evolutionary computation, 2006.

[17] M. Ericsson, M. Resende, P. Pardalos, "A Genetic Algorithm for the
weight setting problem in OSPF routing". Journal of Combinatorial
Optimization, 2002, volume 6, pages 299-333.

[18] Norio. Shimamoto, Atsushi Hiramatsu, Kimiyoshi Yamasaki, "A
dynamic Routing Control based on a Genetic Algorithm", IEEE
International Conference on Neural Networks, 1993, pages 1123-1128.

[19] Anton Riedl, "A Hybrid Genetic Algorithm for Routing Optimization in
IP Networks Utilizing Bandwidth and Delay Metrics", Institute of
Communication Networks, Munich University of Technology, Munich,
Germany.

[20] Yinzhen Li, Ruichun He, Yaohuang Guo, "Faster Genetic Algorithm for
Network Paths", Proceedings of The Sixth International Symposium on
Operations Research & Its Applications (ISORA'06) Xinjiang, China,
August 8-12, 2006. Pages 380-389.

[21] Abdullah Konak, David W. Coit, Alice E. Smith, "Multi -objective

optimization using genetic algorithms: A tutorial"
[22] Vandana Venkat, Sheldon H. Jacobson and James A. Stori, "A Post-

Optimality Analysis Algorithm for Multi -Objective Optimization",

Vol 8, No. 3, 2010
Journal Computational Optimization and Apphcations, Volume 28,

2004.

[23] Ramon Fabregat, Yezid Donoso, Benjamin Baran, Fernando Solano,
Jose L. Marzo, "Multi-objective optimization scheme for multicast
flows: a survey, a model and a MOEA solution". Proceedings of the 3rd
international IFIP/ACM Latin American conference on Networking,
2006, pages 73-86.

[24] Basela S. Hasan, Mohammad A. Khamees, Ashraf S. Hasan Mahmoud,
"A Heuristic Genetic Algorithm for the Single Source Shortest Path
Problem", IEEE/ACS International Conference on Computer Systems
and Applications, 2007, pages 187-194.

[25] Johan Andersson, "Applications of a Multi-Objective Genetic Algorithm
to Engineering Design Problems", Department of Mechanical
Engineering, Linkoping University 581 83 Linkoping, Sweden

[26] Carlos A. Coello Coello, "A Short Tutorial on Evolutionary
Multiobjective Optimization"

[27] Madiha Sarfraz, Younus Javed, Almas Anjum, Shaleeza Sohail,
"Routing Optimization Strategy Using Genetic Algorithm Utihzing
Bandwidth and Delay Metrics", 2010 the 2nd International Conference
on Computer and Automation Engineering (ICCAE 2010), Singapore

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

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

Vol. 8, No. 3, 2010

Architectural Description of an Automated System for
Uncertainty Issues Management in Information

Security

Haider Abbas

Department of Electronic

Systems,

Royal Institute of

Technology, Sweden

haidera@kth.se

Christer Magnusson

Department of Computer and

System Sciences,

Stockholm University,

Sweden

cmagnus@dsv.su.se

Abstract — Information technology evolves at a faster pace giving
organizations a limited scope to comprehend and effectively react to
steady flux nature of its progress. Consequently the rapid
technological progression raises various concerns for the IT system
of an organization i.e. existing hardware/software obsoleteness,
uncertain system behavior, interoperability of various
components/methods, sudden changes in IT security requirements
and expiration of security evaluations. These issues are continuous
and critical in their nature that create uncertainty in IT
infrastructure and threaten the IT security measures of an
organization. In this research, Options theory is devised to address
uncertainty issues in IT security management and the concepts
have been deployed/validated through real cases on SHS
(Spridnings-och-Hdmtningssystem) and ESAM (E-Society)
systems. AUMSIS (Automated Uncertainty Management System in
Information Security) is the ultimate objective of this research
which provides an automated system for uncertainty management
in information security. The paper presents the architectural
description of AUMSIS, its various components, information flow,
storage and information processing details using options valuation
technique. It also presents heterogeneous information retrieval
problems and their solution. The architecture is validated with
examples from SHS system.

Keywords: Information Security,
Options Theory

Uncertainty Issues,

I. Introduction

Technological uncertainty due to rapid development and
innovation in IT, continuously impacts security measures of an
organization. The development is desirable that could facilitate
business organizations with innovative hardware, novel
methods and state of the art technologies. While on the other
hand, technological progression also requires business
organizations to adapt new methods and technologies to secure
their information system (storage, retrieval, communication

Louise Yngstrom

Department of Computer

and System Sciences,

Stockholm University,

Sweden

louise@dsv.su.se

Ahmed Hemani

Department of Electronic

Systems,

Royal Institute of

Technology, Sweden

hemani@kth.se

etc) processes. The objective could be achieved by deploying
new security methods and by evaluating their validity,
serviceability and interoperability using re-evaluation. But the
service acquisition and validation process for IT security
mechanisms is victimized by uncertainty due to new
unforeseen threats and technological advancements appearing
from time to time. Also these newly acquired security
services/features may affect other interacting systems, this is
referred to as externalities [1][2]. We addressed three major
concerns in information security management due to
technological uncertainty i.e. dynamically changing security
requirements [3], IT security externalities [4] and obsoleteness
of security evaluations [5]. We have utilized options theory
from corporate finance [6]; known due to significance of
providing effective guidance during uncertain investments. The
options theory has been transformed using adaptability model
[7] to tailor the IT security processes. The options theory
methods were manually applied to illustrate and validate the
concepts using real cases on ESAM (E-Society) [8] and SHS
(Spridnings-och- Hamtningssystem) [9] systems. The ultimate
objective of this research is to develop an automated solution
(AUMSIS: Automated Uncertainty Management System in IT
Security) for uncertainty issues management in IT security.
The solution can be deployed in an organization and will be
capable of providing system generated reports for; i)
requirement change summary and suggested solutions ii)
externalities report and internalization parameters and iii) re-
evaluation strategy/guidance based on actual system state. In
this paper, we will present the architectural description of the
AUMSIS system which consists of its various components,
architectural styles, information flow between components,
storage details and heterogeneous information processing
description.

The paper is organized as follows: Next in section 2, the
related work will be highlighted, section 3 presents the holistic
view of the IT security uncertainty issues and section 4

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

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

Vol 8, No. 3, 2010

presents the concept of automated uncertainty management
solutions and elaborates its various constituents. Section 5
describes the information processing and flow in AUMSIS.
Section 6 elaborates heterogeneous information processing
problem and the proposed solution for this issue. Section 7
presents the discussion about the analysis and validation of the
AUMSIS framework using SHS example. Section 8 presents
conclusion and the future intention of this research.

II. Related Work

Automated information processing systems have been
emphasized from various researchers in many domain areas.
For example, Wilson, D. et al. has discussed various issues in
automated inspection and representation of uncertainty for the
real world issues [10]. McVicker, M. et al. has presented the
infrastructure that collects statements of security-related
statistics from the World Wide Web for source reliability
verifications [11]. The work presented in this paper addresses
the automated solution of uncertainty issues that might
suddenly appear during IT security requirements/evaluation
management and require a cumbersome solution exploration
process with significant resources [12]. The ultimate outcome
of this research will benefit organizations to have system-
generated reports for IT security management i.e. i) changing
requirements solutions ii) internalization guidance, iii) re-
evaluation strategies and iv) security investment related
suggestions/decisions.

III. Information Security and Uncertainty Issues

Most of the businesses today rely on IT infrastructures and
have to deploy various security mechanisms to protect their
work processes. Technological uncertainty strongly impacts
those security mechanisms, which become obsolete with the
rapid technological progression. The research emphasizes three
critical concerns caused by technological uncertanity for an
organization in IT security perspective as depicted in Figure 1 .

Problems Caused by Technological Uncertainty

1- Dynamically Changing Security Requirements

2- IT Security System's Externalities

3- Continuous Security Evaluation /Re-evaluation of IT
products/Mechanisms

An organization continuously has to go through a cumbersome
procedure to deal with uncertainty issues and to keep their IT
system up-to-date and according to new technological
standards. The research aims for an infrastructure that will help
to avoid the resource-hungry procedures and frame the system
state, organizational needs, system's externalities issues and re-
evaluation requirements analysis. The next section presents the
architectural details of such an automated system (AUMSIS)
that can be deployed in an organization. The system will
automatically generate uncertainty solution reports for the
issues depicted in Figure 1 .

IV. Automated Uncertainty Management Solutions
IN Information Security

AUMSIS is aimed to provide system-generated strategic
guidance for above-mentioned issues described in section III.
Decision-makers can use this information to formalize current
and future IT security management strategies based on actual
system state, which consists of organizational policies, up-
coming technologies, vulnerability logs, attack history and
available budget. Figure 2 depicts the abstract view of the
AUMSIS architecture as follows:

^ Internet A

Organizational
Policies/ Budget
Information

v-^

Up-coming
Technologies

Knowledgebase

Security
System
Vulnerability
Reports

Attack
Histories

Externality
Reports

Option
Analysis
Data

SW!

Security
Requirements

Figure 1 . Uncertainty issues addressed in AUMSIS

Figure 2. AUMSIS Architecture

The various components of AUMSIS architecture depicted in
Figure 2 are elaborated as follows:

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

A. Knowledgebase

Information related to system state during a specified time
period is named as historical data and organized in a
structured repository; knowledgebase. It consists of following
components:

i) System vulnerability reports

It contains malfunctioning reports of the security system and

the corresponding affected security components. The

information can be used to track the actual service/component

causing vulnerability and provides details to determine system

state.

ii) Attack history

Attack history data contains information about the

exploitation of a particular security service/component by

authorized/unauthorized sources. It will reveal shortcomings

in existing security mechanisms that need to be factored in.

iii) Externality reports

IT security system of an organization may also leave positive
or negative effects to other interacting systems/sub -systems
referred to as externalities [1]. Externalities of a security
system [2] can be identified by internal/external
malfunctioning reports from affected systems/partners.
Externality reports provide a holistic view of the IT security
system and help to determine system's desired functionality.

iv) Options analysis data

AUMSIS generates results using options technique that are
reusable by subsequent analysis. Options analysis data
contains information about already executed options and
results from a previous analysis. Option cards were used to
store data about the options analysis outcomes [3].

v) Security Requirements

Security requirement change reports from various
stakeholders or security requirements from any external
enforcing authority. This is continuously updated to factor in
new/changed requirements.

B. Up -coming Technology and Organizational
policies/Budget information

It is the prime objective of AUMSIS to provide contemporary
guidance about requirement solutions, internalization factors
and evaluation strategy. Therefore the AUMSIS has to interact

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

Vol. 8, No. 3, 2010

and extract information from Internet and organization's
policy database. These factors are considered as a separate
component in AUMSIS due to their evolving nature during
analysis.

Above-mentioned historical data, information about
upcoming-technologies and organizational policies/budget are
accumulated over time and readily available for options
analyst agent (OAA) for processing.

C. Options Analyst Agent (OAA)

Options analysis agent is a piece of software [13] that

formalizes requirement solutions, internalization results and

evaluation strategy using options technique. It extracts system

state from knowledgebase, Internet (for up-coming

technologies) and organizational policy/budget database.

OAA generates the strategic information for decision makers

i.e.

i) Analyze alternative solutions for a security requirement and

provides recommendations based on contemporary system

state.

ii) Internalizing solutions for externalities according to

organizational policies.

iii) Deterministic test plans strategies for the evaluation

process of each security service considering its

malfunctioning report and service exploitation history.

AUMSIS provides up-to-date strategic guidance for the
uncertainty issues in information security management
process. It considers uncertainty elements caused by changing
environment and helps to devise respective optimal IT
security strategy. Next section describes the information
processing and flow in AUMSIS system.

V. AUMSIS Information processing and flow

AUMSIS provides strategic guidance for three main areas of
information security management affected by uncertainty
issues. The uncertainty management process using AUSMIS
for these issues follows slightly different mechanism due to
the nature of problems. But the data are maintained in a single
repository i.e. knowledgebase. As the AUMSIS addresses
three uncertainty concerns in IT security, each one is
elaborated individually in Module 1, Module 2 and Module 3.
Figure 3 below depicts the information flow of these three
modules using information flow diagram as follows:

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

(IJCSIS) International Journal of Computer Science and Information Security,
Start J Vol. 8, No. 3, 2010

Module 1

Dynamic Requirements
Management

Module 2

Externality
Management

Modules

Re-evaluation of
Security System

Identification

Stakeholders/Requirements/I nternalization Parameters/Security system

Formulation in Options Theory Context
(Data from Knowledgebase)

Options Analysis for alternative solutions

Each Selected Solution

Up-coming

technologies/Budget

information/Uncerta

inty Revelation

Organizational

Strategy/

Uncertainty

Revelation

Test Results

Interdependence/

Uncertainty

Revelation

Option Selection
Opt/Delay/Abandon/Switch

Figure 3. Information flow for Module 1, Module 2 and Module 3

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

i) Module 1: Information processing flow of dynamic
security requirements management

Dynamically changing security requirement management
process starts with the identification of requirement change in
an organization. It proceeds with examining all possible
solutions for this particular requirement. Each solution is
divided into parts and analyzed/compared with system state
(determined by the data from knowledgebase), organizational
policies/budget and up-coming technologies. The significance
of the approach followed by AUMSIS for solution exploration
is the options theory. That concurrently analyzes all solutions
and decides about each solution to delay, abandon or opt in
existing scenario. It provides decision makers analysis reports
for the requirement under-consideration, its possible solutions
and the pros and cons of each solution according to their
organization's information system state.

ii) Module 2: Externality management information
processing and flow

AUMSIS generates internalization recommendations for the
externalities caused by a security system. The security system
is already described in knowledgebase according to security
mechanisms/services it offers. Internalization process starts
with identifying externalities by analyzing system data (from
knowledgebase). The next phase is to list all possible solutions
(internalization parameters) according to organizational
policies and available budget/resources. Each solution is then
divided into parts and analyzed using options technique to
build organization's internalization strategy considering
current system state and organization's future plans. AUMSIS
generates internalization results for each internalization
parameter to delay, opt and abandon according to existing
scenarios as depicted in Figure 3.

iii) Module 3: Re-evaluation of security services
/mechanisms information processing and flow

AUMSIS helps to build re-evaluation strategy for IT security
services/mechanisms considering the uncertain factors i.e.
requirements/polices change, vulnerability appearance and
interoperability issues that adversely impact evaluation
process. The process starts with identifying the boundaries of
system for evaluation. It could be the newly adapted solutions

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

Vol. 8, No. 3, 2010

in a security system or already evaluated components that
need to be re-evaluated. Tests are classified according to the
nature of the system under consideration. Uncertainty issues
during re-evaluation are dealt using options technique in
AUMSIS as tests are prioritized based on pervious evaluation
results and vulnerability reports from knowledgebase.

VI. HETEROGENEOUS INFORMATION ISSUE

As the AUMSIS retrieves information from various
information sources (i.e. knowledgebase, Internet,
organization's policies database) and therefore varies in their
structure, syntax and semantics. It is not directly
comprehendible by the Option Analyst Agent (OAA).
Therefore it is desirable to store information in uniformly
accessible and extractable manner. Without considering the
operating systems used and the hardware running these
softwares. To overcome the issue of heterogeneous
information retrieval we have proposed the use of ontologies
[14] [15] that provide a shared conceptualization of a system or
domain. The language used will be Web Ontology Language
(OWL) for the development of ontologies. Which is based on
strong constructs of description logic and is thus useful to
represent any set of rules that are options concepts,
organizational policies, internalization parameters etc in case
of AUMSIS.

With the help of the options analyst agent these ontologies can
be traversed to find the useful information models and to
resolve the semantic heterogeneity issues in AUSMIS
components. These issues are raised due to the merger of
information from various domains i.e. policy database,
technological information and vulnerability/malfunctioning
reports. It is worth mentioning here that the knowledgebase
contains all the organizational polices and rules. This
information plays a key role when OAA accesses information
from various information sources and formalizes
decisions/strategy. Figure 4 depicts the heterogeneous
information retrieval framework as follows:

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

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

Vol. 8, No. 3, 2010

Information Source 1 Information source 2 Information source n

Domain Ontology Domain Ontology Domain Ontology

Rast^ving semantic
eterogeneity

Retrieve information
processing rulesc^

Option analyst agent

Figure 4. Option analyst agent's communication with ontologies and
knowledgebase

VII. AuMSis Analysis and Validation

AUMSIS architecture presented in previous sections is based
on an in-depth study of its methodological details and manual
deployment to SHS and ESAM system in past [2] [3] [4]. The
current AUMSIS design/information flow is about the
automated version of options technique's concept for
uncertainty management in information security. The
architecture currently addresses three main uncertainty issues
but is flexible to opt any other problem's mechanism for
uncertainty management in information security. Example
given below presents the SHS uncertainty management
process using AUMSIS in a nutshell. It is worth mentioning
here that AUMSIS will be deployed into the organization that
interacts and extracts required information about the target
system i.e. SHS in this example.

A) Changed requirement request

The process is initiated when a change requirement is
identified for SHS system. This could be initiated by an
internal source (stakeholders, management and implicit
system's request) or by some external source (government
enforcing authority) to adapt new standards. Once a change
request is identified; it acts as a stimulus to AUMSIS process
for the SHS system.

i) Data collection

AUMSIS decides about the optimal solution for a requirement
change based on actual system state and within existing
circumstances. Knowledgebase provides data to determine
system state using vulnerability/malfunctioning reports and
system exploitation history with respect to affected SHS
services. Dynamic factors like uncertainty revelation, budget
information and up-coming technology information are also
continuously accessed/considered in solutions formulation
process as described in next section.

ii) Options analysis

OAA retrieves information about input data of requirement
change for SHS and lists all available solutions for the certain
requirement. Each solution is then assigned priorities
determined by the up-coming technology, budget information
and uncertainty involved in current state. Options theory
provides various alternatives to opt, delay or abandon a
solution based on uncertainty revelation; also during solution
formulation process. AUMSIS analyzes each possible solution
by staging its deployment process and wait for additional
information that becomes available with the time during
exploration and analysis. This additional information normally
requires altering the requirement selection strategy; and this
facility is factored in as a core feature of AUMSIS. Thus it
provides optimal solution about a requirement considering all
possible factors that cause uncertainty in determining a
solution. The output information of a solution evolution
process is stored in knowledgebase that can be used later and
provides guidance to examine future strategy.

The newly opted solutions for SHS from a requirement
management process may cause positive or negative effects
for other interacting partners. Next phase elaborates how these
effects are addressed as externalities in AUMSIS.

B) Externality management process

Externalities are the effects borne by the systems that are not
involved in a direct communication with the SHS security
system. These effects could be positive (that might bring in
benefits) or negative (that might cause vulnerabilities) and
may appear anytime during the life cycle of SHS system.
AUMSIS initiates externality management process by

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

specifying internalization parameters that describes solutions
in case of externality occurrences due to SHS system.
Organizations (responsible for controlling security system)
specify possible internalization parameters according to their
security objectives and are stored in knowledgebase.

i) Options Analysis

When an externality is reported/detected for SHS either
positive or negative, OAA lists possible internalization
parameters for each externality and compares with
organizational constraints, which include budget information
and organizational policies. These factors are uncertain that
may change and affect externality management process. It is
also uncertain if a solution will work appropriately. OAA
stages each solution into sections and analyzes them
individually. All solutions for the externalities of SHS are
decided using various options to delay, abandon or alter
decision with respect to uncertainty revelation, rational
analysis, budget and organizational policies. These factors can
be determined using the data from knowledgebase. AUMSIS
mechanism of externality management helps to
deterministically consider variable factors and to respond
accordingly for a specific scenario.

System up-gradation in case of newly installed services for
requirement management or externalities solutions (that
recommended modifications) requires to re-evaluate the
security system to test individual functionality and as a whole
interoperability. This factor is also addressed in AUMSIS as a
part of a complete solution and described in following section:

C) Initiation of Re-evaluation process

Re-evaluation is performed particularly when new solutions
are devised. For example in case of SHS system when the
existing system was reconfigured/modified. It is also
recommended as periodically scheduled analysis for the
complete system. AUMSIS classifies evaluation tests into two
major categories i.e. assurance and criticality [7]. Assurance
class contains tests to validate performance, serviceability and
functionality. Criticality class contains tests that may alter
testing strategy and are directly affected by uncertain
outcomes/uncertainty issues those are interoperability,
technological innovation and budget.

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

Vol. 8, No. 3, 2010

i) Options Analysis

OAA customizes and organizes evaluation strategy for a
particular service of SHS based on its history of service
failure, vulnerability reports and exploitation history. The
information is extracted from the stored data from
knowledgebase; which becomes readily available to OAA.
Tests are prioritized based on this information and system
state. AUMSIS using options theory; provides a deterministic
approach to generate evaluation strategy and the ability to
alter the evaluation directions. It helps to avoid unnecessary
tests that can be determined by the information from uncertain
outcomes and uncertainty revelations.

VIII. Conclusion & Future Work

Organizations need to overcome uncertainty issues in their
information security management progress due to obvious fact
of rapid technological development. They continuously
require significant changes in their existing security
infrastructure to meet the organizational security objectives
and security standards. Organizations also have to invest huge
resources and have to go through a cumbersome procedure to
keep their system up-to-date. The paper introduced AUMSIS,
the infrastructure of an automated system for uncertainty
management issues at organizational level based on an in-
depth study and manual validation of these concepts in past.
The system is capable of managing dynamic issues using
options theory mechanism from corporate finance that helps to
generate appropriate strategies according to system state. The
paper presented the architectural details and information flow
for AUMSIS system and its various components. The future
intention of this research is the deployment of AUMSIS
framework into a software architecture style.

References

[1] Richard Cornes, Todd Sandler, "The Theory of Externalities, Public Goods and
Club Goods", Cambridge University Press, June 1996

[2] Ann Cavoukian, "Privacy as a Negative Externality The Solution: Privacy by
Design" Workshop on the Economics of Information Security, London, June 24,
2009

[3] Abbas Haider, Yngstrom Louise and Hemani Ahmed, "Empowering Security
Evaluation of IT Products with Options Theory", in 30th IEEE Symposium on
Security and Privacy 2009, Oakland, Cahfornia, USA

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

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

Vol. 8, No. 3, 2010

[4] Abbas Haider, Magnusson Christer, Yngstrom Louise and Hemani Ahmed, "A
Structured Approach for Internalizing Extemahties Caused by IT Security
Mechanisms", In Proceesdings of IEEE International Workshop on Education
Technology and Computer Science (ETCS 2010), March 2010, Wuhan, China

[5] Abbas Haider, Yngstrom Louise and Hemani Ahmed, "Option Based
Evaluation: Security Evaluation of IT Products Based on Options Theory", In
Proceddings of IEEE Eastern European Regional Conference on the Engineering
of Computer Based Systems 2009, Novi Sad, Serbia, Pages.134-141

[6] J. Mun, "Real Options Analysis - Tools and Techniques for Valuing
Strategic Investments and decisions", Wiley, Finance, 2002

[7] Abbas Haider, Yngstrom Louise and Hemani Ahmed, (2009), "Adaptability
Model Development for IT Security Evaluation Based On Options Theory" in
proceedings of IEEE/ACM 2nd International Conference on Security of
Information and Networks (SIN 2009), North Cyprus

[8] Abbas Haider, Raza Asad , Louise Yngstrom, Ahmed Hemani, "Evaluation of
ESAM using Architectural Tradeoff Analysis Method", Project Report -VERVA,
December 2008

[9] Kurt Helenelund, Stephan Urdell, Bo Sehlberg, Anders Bremsjo, Anders
Lindgren, Jan Lundh, Christer Marklund, "SHS Version 1.2 Protocols", VERVA -
Swedish Administrative Development Agency, 2007

[10] Wilson, D. ; Greig, A. ; Gilby, J. ; Smith, R., "Intelligent automated
inspection, representing the uncertainty of the real world", lEE Colloquium on
Intelligent Sensors (Digest No: 1996/261), 19 Sep 1996, pages 11/1 - 11/6

[11] McVicker, M.; Avellino, P.; Rowe, N.C., "Automated Retrieval of Security
Statistics from the World Wide Web" IEEE SMC Information Assurance and
Security Workshop, 2007, 20-22 June 2007, pages 355 - 356

[12] Abbas Haider, Yngstrom Louise and Hemani Ahmed, "Security Evaluation of
IT Products: Bridging the Gap between Common Criteria (CC) and Real Option
Thinking" in proceedings of World Congress on Engineering and Computer
Science (WCECS 2008), 22-24 October, 2008, San Francisco, USA

[13] Nick Jennings, Michael Wooldridge, "Software Agents", lEE Review,
January 1996, pp 17-20

[14] Thomas R. Gruber: Automatically Integrating Heterogeneous Ontologies
from Structured Web Pages. Int. J. Semantic Web Inf. Syst. 3(1): 1-11 (2007)

[15] Xiaomeng Su, Mihhail Matskin and Jinghai Rao. "Implementing
Explanation Ontology for Agent System". In Proceedings of the 2003
lEEE/WIC International Conference on Web Intelligence, Wr2003, Halifax,
Canada, October, 2003. IEEE Computer Society Press

AUTHORS PROFILE

Haider Abbas has been working as doctoral student at
Department of Electronic Systems, KTH, Sweden. Mr. Abbas
has authored more than 10 international publications and has

been working for various governmental and private projects in
Pakistan and Sweden.

Christer Magnusson is Senior Lecture at the Department of
Computer and Systems Sciences, Stockholm University,
specialized in IS/IT Security and IS/IT Risk Management.
Before joining SecLab, Christer was Head of Corporate
Security and Risk Manager at Sweden Post and CEO of
Sweden Post Insurance AB and Sweden Post Reinsurance
S.A. He has also held the position as Head of Corporate
Security in the Ericsson group. In 1999, Christer was awarded
the SIG Security Award by the Swedish Computer Society
and in 2000 the Security Award by the Confederation of
Swedish Enterprise as recognition of the models and the
integrated processes regarding IS/IT Risk Management, that
he developed as a part of his research studies. Dr. Christer is a
member of the advisory committee of the Swedish Emergency
Management Agency. He serves on the risk and security board
of the Confederation of Swedish Enterprise (NSD). He is also
an adviser in Corporate Governance, Compliance, Risk
Management, and Information and ICT Security to
government agencies as well as trade and industry.

Louise Yngstrom is a professor in Computer and Systems
Sciences with specialization in Security Informatics in the
department of Computer and Systems Sciences at Stockholm
University. Her research base is Systems Science which she
since 1985 has applied within the area of ICT security forming
holistic approaches. Her research focuses various aspects on
how ICT security can be understood and thus managed by
people in organizations, but also generally on criteria for
control. She has been engaged in various activities of the
International Federation of Information Processing, IFIP, since
1973; the Technical Committee 3(TC3) with an educational
scope, the TC9 with focus on social accountabilities of ICT
structures and the TCll with focus on ICT security. She
founded the biannual conference WISE (World Conference on
Security Education) in 1999. She was engaged in European
networking for curricula developments within ICT security
and the Secured Electronic Information in Society working for
e-Identities during the 1990's. Since 2000 Dr. Louise is
involved in introducing ICT security in academic and business
life in African countries through her research students who
simultaneously with their research are academic teachers in
their home countries. Over the years she has traveled and

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

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

Vol 8, No. 3, 2010

networked extensively with international peers. Presently she Sweden. Dr. Hemani has authored more than 100 international

is the principal advisor of seven PhD students. publications. He is participating in and leading some national

and EU projects.
Ahmed Hemani has been working as professor and head of

post-graduate studies at Dept. of ES, School of ICT, KTH,

67 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

Driving Architectural Design through Bussiness Goals

Lena Khaled

Software Engineering Department

Zarqa Private University

Amman, Jordon

lenaumleen@yahoo.com

Abstract - Architectural design must encompass changes
to business goals, their relations to quality attributes
overtime and their results upon building the final
specific systems. This paper discusses the effect of
business goals on building the architectural design on
any system. It describes the relationship between
business goals and system qualities, and how these
qualities are met during architectural design. This paper
also describes how the qualities have an effect on the
decisions of building the architectural design on any
system. The role of the agent is described through the
process of building.

Keywords- Bussiness Goals; Qualities; Architectural Design
Decisions.

Introduction

An important part in the design phase and construction of
any complex system is its architecture. A good architecture is
that system which satisfies key requirements such as
performance, reliability, portability, interobelability,
availability, security, safety, scalability and other attributes.
A bad architecture causes a lot of damage. According to
IEEE, 'Architectural design is the fundamental part in any
system which includes components of a system, their
relationships to each other and to the environment and the
basic principles which guide the design and evolution of that
system.'' [5, 7]

On the other hand, business goals are the foundation on
which any architectural design is built upon. Architectural
design is built to achieve mission goals. Business goals come
in many types and at many levels of abstraction; therefore,
architectural design is a bridge between business goals and
the achieved system.

This paper describes how business goals cause the
progress to build architectural design to any system; many
processes need to be applied in order to reach the final
system. This paper is organized as the following: section 3
defines business goals and attributes and then it defines the
relation between them. Section 4 describes the architectural
design decisions and its major phases when defining the
design process. Section 5 defines the main result of this
paper; it talks about how the final architecture is built, the
role of software agent through building and software quality
metrics. You can find the conclusion in section 6.

II. Related Works

The first piece of related works is reported in the most
depth in Perry. Perry Dewayne and Wolf Alexander worked
on studying the foundations of software architecture, they
developed an intuition for software architecture by applying
to several disciplines of architecture, and they presented a
model of software architecture which consists of three
components: elements, form and rationale [12].

The second related works is [5]. Garlan examined
important trends of software architecture in research and
practice, and speculated on the important emerging trends:
aspiration and challenges.

The third paper in related works is [1]. Ozkaya, Kazman
and Klein worked on a case that represented the architectural
patterns which carries the economic value in the form of real
options. They summarized their observations in evaluating
the relative value of patterns using real option value models
on a model problem and they looked carefully on how
economics-informed approaches can provide better insights
for the selection of situated design strategies.

The fourth paper in related works is [1]. This paper
describes how architectural decisions are made from quality
attributes. It presents a set of steps that allowed moving from
quality attribute requirement to design fragment based on
achieving that requirement. All of these are demonstrated
through an application of an embedded system as in
example.

This paper works on the aim of business goals and their
relationships to the quality attributes, and how this relation
can drive the architectural design to any complex system.

III. BUSINESS GOALS AND QUALITIES

Understanding bussiness goals and their relations to the
qualities is a critical part of building the architectural design
of any system; we cannot easily use architectural design or
other solutions without understanding the concepts of both
bussiness goals and qualities. Therefore, these goals and
qualities drive the architectural design of the system [3].

Driving architectural design through goals need an early
method used to generate and refine qualities, which is called
QAW (Quality Attribute Workshop). This gets qualities that
are mapped to business goals scenarios for the qualities
which are built by stakeholders according to the main goals.
All these scenarios specify whether a system satisfies the

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

user's requirements or not [3,2]. The quality attributes must
be well understood and expressed early in the development
of the life cycle of the system, so the architect can design an
architecture that will satisfy these qualities. QAW is a
method to discover, document, prioritized the system's
quality attributes early in its life cycle [2].

A. Bussiness goals

Business goals are the parts that drive the methods of
design, and are the elements that shape the architecture. The
important thing is that all business goals that correspond to
quality attributes will view the end of the system [3,6].

According to [10], we can define a goal as a state of
events in the world that users would like to achieve. Goals
can be rather business goals or system goals. Business goals
are states that an individual or an organization wants to
reach.

B. Quality attributes

In manufacturing, the concept of the quality means that
the product should meet its requirements, but the popular
vision of the quality is that it is an intangible attribute.
Terms of bad or good quality represent how people talk
about something vague which they don't propose to define.
Quality attributes describe the property of the system that
refers to its fitness for use. The term, non-functional
requirement, is a synonym for quality attribute or attribute.
[13,9,4]

The international standard on software product qualities
classifies software quality as six main attributes:
functionality, reliability, usability, efficiency,

maintainability, and portability. Despite the fact that there
are many quality attributes, reliability and maintainability
are the main quality criterions and many of these attributes
are created at business levels and are better viewed as
business goals [6, 8]. Figure 1 illustrates how goals and
attributes affect each other.

Business goals

Affected by

Quality
Attributes

IV. ARCHITECTURAL DESIGN DESCISIONS

Architectural design decisions play an important
position in designing the architecture for any system. We can
define an architectural design decision as: a description of
the set of architectural modifications to the software, the
principles and rules to the design, design constraints and
additional requirements that realize requirements on a given
architecture. During the process of architectural design, the
software designer has to make a number of decisions that
affect the system and its development process [13].

Figure 1 shows how the decisions of a designer affect
the design process. According to the experience of the
designer, the main decisions are the answer to the following
questions [13]: What is the main approach that has been used
to structure the system? What is the strategy that has been
used to control the operations in the system? How is
distribution across the occurring system? What is the
appropriate style for the system? How do the evaluations of
the architectural design occur? How should the architecture
of the system be documented?

Specification

Designers Decision

Design process

Implementation

Figure 2. The role of designer's decisions on a design process

When developing systems, the process of the design is
divided into two distinct phases. Figure 3 describes these
phases [14].

Figure 1 . The relation between goals and attributes

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

Figure 3. The major phases of the software design process

1) In the first phase, the designer develops a high level
of solutions (the architectural or logical design) in which
only the abstraction view of the model is described without
details. Here, the system is described as black-box system.

2) In the second phase, the abstract view is going into
details(the details or physical design) . Turning the black
box into white box , the output from this phase provides the
specifications for the programmer.

Making a decision to any architecture is based on the
relation between goals and qualities.

V. FINALIZING ARCHITECTURAL DESIGN

The architectural design of any system can be defined as
the structure of the system, which consists of software
components, the external properties of these component and
their relationships. The architectural design affects the
performance, maintainability, robustness. The particular style
and structure chosen for an application may therefore depend
on non functional requirements [14, 13].

A. The role of the agent

Software agents are autonomous software entities that
navigate through environments and can either work alone or
with other agents to achieve the goal.

The software agent plays an important role in the cycle
that is represented in figure 3. The fundamental reason for
using an agent to build the architectural design for any

system is based on the concept that many users
(stakeholders) within an organization have many different
business goals. Therefore, by linking software agent to this
cycle, it becomes possible to give the main impact for
choosing a specific architectural design decisions that arise
through building the architecture [6]. The architect needs
many decisions to build the final system. Then, the software
agent helps the architect to choose the main decisions that
make the architect achieving his goal to meet the user
requirements.

Architect

Needs

Figure 4. Driving Achitectural Design through bussiness goals

Figure 4 represents the result of this paper; it describes
how business goals are affected by the attributes and how
this relation affects on the decision of the architect and in
which from these decisions the final architecture is built. The
role of the agent is described through the figure. The agent is
especially raised when we want to move from old to new
goals.

Creating goals is an important thing to the products and
process. These goals should be considered by the architect
through mechanisms, which are driven by models of
business. There are many types of mechanisms such as the
SQM that appears in figure 4 and is defined as a mechanism
for measuring goals. Measurements enable the organization
to make its process better than before, in addition to the
controlling of the software project and assessing the software
quality. Other measurement can be used instead of SQM
such as the Quality Function Deployment approach (QFD),
the Goal/Question/Metric paradigm (GQM).

Management is done better by measurement,
measurement enables the organization to improve the

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

activities of the quality management: planning, control and
asses the quality of the software that produced.

The quality management activities check the project
output to ensure that they are dependable with the
organizational standards and goals. The development team
should be independent from the quality assurance team so
that they can take a goal view of the software. An
independent team should be responsible for managing the
quality to maintain the project schedule and budget. As a
result, this team should not be associated with any particular
development group. An independent team ensures that the
organizational goals of quality are not compromised by
schedule consideration and short-term budget [13].

The evaluation of the overall work will reflect its
relationship to the quality of work that is expected from
decisions which were made by the architect before. It is
defined as the process of verifying satisfaction of
requirements. The final architectural design is very difficult
to be evaluated because the correct test of the architecture is
in how it meets the functional and non-functional
requirements after it has been deployed [13].

VI. CONCLUSION

Architectural design is one of the most important phases
in the life cycle of any system. Building an architectural
design for any system to any organization must include the
changes that occurred to business goals. This paper presents
how an architectural design was drove from business goals
and this is done through explaining the relation between
business goals and non-functional requirements (also called
quality attributes or attributes). It describes how an architect
needs decisions to build the architecture and how these
decisions are built upon the previous relation. The role of an
agent is shown through this paper to show the movement
from old business goals to new business goals.

REFERENCES

[I] Bachman F,Bass l,klein M, "Moving From Quality Attribute
Requirment to Architectural Decisions," Software
Engineering Institute,USA.

[2] Barbacci M,Ellison R,Lattanze A,Stafford J,Weinstock C,and

Wood W,"Quality Attribute Workshops(QAW),technical

report CMU/SEI,2003.
[3] Bass L,Clements P, Kazman R,Nord R, "Architectural

Business Cycle Revisted," Software Engineering

Institude,Caregie Mellon,2009.

[4] Berenbach B,PaulishD,KazmeierJ,Rudorfer A, "Software

and systems requirments engineeering in

practice,"McGrawHll,2009.
[5] David Garland/'Software Architecture: a Roadmap," ACM

Press,2000.
[6] Gross D, Yu E,"Evolving System Architecture to Meet

Changing Bussiness Goals: an Agent and Goal -Oriented

Aapproach," University of Toronto.

[7] http://www.sei.cmu.edu/ Carnegie Mellon,2007.
[8] Jalote P,"A concise introduction to software engineering,"
Sprmger,2008.

[9] Kan S, "Metrics and models in software quality
engineering," Addison Wesely,2003 .

[10] Liu l,yu eric, "From Requirment to Architectural Design -
using Goals and Scenarios," University of Toronto.

[II] Ozkaya I, Kazman R, Klein M,"Quality attribute based
economic valuation of architectural patters," SEI,9TH
international conference on software engineering,IEEE,2007.

[12] Perry D., Wolfs A., "Foundations of the study of software
architecture," ACM Sigsoft, vlo.l7, no. 4, 1992.

[13] Sommerville I,"Software Engineering, "Pearson education,
2007.

[14] Widhani A,Boge S,BarteltA,LamersdofW," Software
architectural and patterns for electronic commerce
systems, "University of Hamburg,2002.

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

DISTRIBUTED INFORMATION SHARING COOPERATION IN
DYNAMIC CHANNEL ALLOCATION SCHEME

Mr.PJesu Jayarin,
Research Scholar,
Sathyabama University,
Chennai-119, India.
jjayarin@gmail.com .

Dr.T.Ravi,

Prof&Head ,Dept of CSE,
KCG college of Technology
Chennai-97, India.

Abstract

The Channel allocation method is an
Cooperative Asynchronous multichannel MAC
which introduces an Distributed Information
Sharing (DISH) to be used in a distributed flavor
of control-plane cooperation, as a new approach
to wireless protocol design and then apply it to
multichannel medium access control(MAC) to
solve the MCC problem. The basic idea is to
allow nodes to share control information with
each other such that nodes can make more
informed decisions in communication. Medium
access control (MAC) protocols play a major
role to create a wireless communication
infrastructure. In Wireless network, transmitter-
receiver pairs make independent decisions,
which are often suboptimal due to insufficient
knowledge about the communication
environment. So, the new concept DISH is
introduced and overcomes the problem occurred
in the MAC protocol. The DISH concept avoids
collision and re-transmission among nodes. The
notion of control-plane cooperation augments
the conventional understanding of cooperation,
which sits at the data plane as a data relaying
mechanism. In a multichannel network, DISH
allows neighboring nodes to notify transmitter-
receiver pairs of channel conflicts and deaf
terminals to prevent collisions and
retransmissions. Based on this, we design a
single-radio cooperative asynchronous

multichannel MAC protocol called CAM-MAC
When the CAM-MAC is used in illustration
purposes, we choose a specific set of parameters
for CAM-MAC. First we analyze the throughput
to 91% of the system bandwidth to 96%, then
saturate 15 channels and compare the result, this
provides an good result in implementing
hardware.

Keywords-Distributed information sharing
(DISH), control-plane cooperation, CAM-MAC,
multichannel coordination problem, MAC
protocol, ad hoc networks.

1. INTRODUCTION

The Cooperative asynchronous
multichannel MAC is to allow nodes to share
control information with each other such that
nodes can make more informed decisions in
communication. This notion of control-plane
cooperation augments the conventional
understanding of cooperation, which sits at the
data plane as a mechanism for intermediate
nodes to help relay data for source-destination
pairs Asynchronous Multichannel MAC is to
allow nodes to share control information with
each other such that nodes can make more
informed decisions in communication.

The new approach DISH is
implemented first and then medium access
control to solve MCC problem. In a multichannel
network, DISH allows neighboring nodes to
notify transmitter-receiver pairs of channel
conflicts and deaf terminals to prevent collisions
and retransmissions. Based on this, we design a
single-radio cooperative asynchronous

multichannel MAC protocol called CAM-MAC.
This is used to provide a good result in
implementation. The heartily focused on giving a
distributed solution in minimal cost, efficient and
sufficient solution in the real world application is
to allow the nodes to share the control
information with all the other nodes in
communication. Example for instance, transfers
of data from one node to another node in a
distributed environment without collision and re-
transfer of data. DISH enables nodes to store
channel usage information at their neighbors, and
retrieve this information whenever it is needed. It
is not a compulsory coordination mechanism but
it provides an cooperation with other nodes and
operates well at all condition. In this network
does not rely on coordination and performs well.
Most wireless LANs are single channel systems.

However, as the number of nodes
communicating increases, systems with a single
channel suffer declining performance.
Contributing to the problem are the well-known

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

hidden and exposed terminal problems. To
combat these problems there is growing interest
in multi-channel systems. Indeed, the IEEE
802.11 standard already has multiple channels
available for use. The IEEE 802.11a physical
layer has 12 channels, 8 for indoor and 4 for
outdoor use. IEEE 802.11b has 14 channels,
5MHz a part in frequency. To avoid channel
overlap, the channels should have at least 30MHz
guard bands; typically, channels 1, 6 and 11 are
used for communication. In a multi-channel
system, the transmitter and receiver must both
use an agreed upon channel for communication.
This introduces a channel coordination problem.
As well, the hidden and exposed terminal
problems remain in the multichannel setting.

1^ V

X"^^ .P'^

- neighbor relationship
■•■► new communication

(§)

(a)

D
®

ongoing data transmission
cooperation

1 X

10K

1 X

2 OK

3 OK

3 X

U^"

Lv^

(b)

Fig. 1. An illustration of the DISH idea, (a) A
multichannel scenario.

(b) Knowledge at individual nodes. By consolidating
the knowledge at nodes C and D, or acquiring
knowledge from node E, it shows that the conflict-free
channel is channel 2.

Based on the idea of DISH, we design a
single-radio cooperative asynchronous

multichannel MAC protocol called CAM-MAC
for ad hoc networks. We evaluate CAM-MAC
from both theoretical and practical perspectives,
where we choose a specific set of protocol
parameters for illustration and evaluation
purposes:

1 .We show that its throughput upper bound
is 91 percent of the system bandwidth and its
average throughput approaches this upper
bound with a mere gap of 4 percent,
2.We show that it can saturate 15
channels at maximum and 14.2 channels on

average, which indicates that, although CAM-
MAC uses a control channel, it does not
realistically suffer from control channel
bottleneck,

3. To investigate the value of cooperation we
compare MAC and observe a throughput ratio of
2.81 and 1.70 between them in single-hop and
multihop networks, respectively,
4. We compare CAM-MAC with three recent and
representative multichannel MAC protocols,
MMAC, SSCH, and AMCP , and the results
show that CAM-MAC substantially outperforms
all of them.

For a further and more realistic
validation, we implemented CAM-MAC on
COTS hardware and conducted experiments. To
the best of our knowledge, these prototypes are
the first full implementation of single-radio
asynchronous multichannel MAC protocols.

We review literature in Section 2, and
identify new challenges to designing a
cooperative protocol in Section 3. Then, we
present the protocol details in Section 4.
Following that. Section 5 provides performance
results in various scenarios, and Section 6
describes our implementation and experiments,
and, finally, give concluding remarks in Section
7.

2. MULTICHANNEL MAC
PROTOCOLS

Multichannel MAC protocols for ad hoc
networks can be categorized into two general
classes as below.

2.1 Single-Radio Solutions

2.1.1 Control-Data Window Schemes

MMAC assumes the IEEE 802.11
power saving mode and divides time into beacon
intervals. Each beacon interval is 100 ms and
consists of a 20-ms ATIM window and an 80-ms
data window. Nodes negotiate and reserve
channels in the ATIM window on a common
channel, and transmit data in the data window on
multiple channels concurrently. The data
window size is fixed, and hence, it has to be set
according to the maximum data packet size,
leading to inefficiency. By contrast, MAP varies
the data window size and avoids this problem.
However, like MMAC, its reservation interval
(i.e., control window) is still fixed, and thus,
both protocols suffer from the inflexibility to
different node densities: at low density, the
control window has long idle time; at high
density, the control window cannot

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

accommodate all negotiations and some nodes
have to wait for the next control window.
Further-more, MM AC and MAP requires time
synchronization over the entire network, which is
a notoriously hard task involving considerable
overhead and complexity, and compromises
scalability. LCM MAC, on the other hand,
allows each neighborhood to negotiate the
boundaries of control-data windows
independently, in order to avoid time
synchronization. However, the negotiated
window size can hardly fit for all nodes in the
neighbor-hood, and this window negotiation,
plus a fine-tuning mechanism, considerably
complicates the protocol. Besides, it has a
starvation problem lacking in an appropriate
solution. Finally, all these control-data window
schemes have a common problem: during each
control window, all channels other than the
common channel cannot be utilized, resulting in
channel underutilization.

2. 1 .2 Channel Hopping Schemes

In SSCH , each node hops among all
channels according to a pseudorandom sequence
such that neighboring nodes will have channels
overlap in time periodically. Since a trans-mitter
can only communicate to a receiver when they
hop. In CAM-MAC, each node stays on a
common channel and only switches channel
when a data exchange is established successfully
or finished. This avoids switching channel too
often and, due to the common channel, does not
incur large delay. Besides, again, no clock
synchronization is required.

2.1.3 Routing and Channel Assignment
Schemes

CBCA combines channel assignment
with routing. It proposes to assign each set of
intersected flows, called a component, with a
single channel, in order to avoid channel
switching delay, node synchronization, and
scheduling overhead at flow-intersecting nodes.
CAM-MAC uses a control channel, which
automatically avoids the problem of node
synchronization and scheduling overhead.
Regarding channel switching delay, its effect on
network performance is much less than MCC
problems: the channel switching delay is 40-150
|is but a channel conflict can collide at least one
data packet whose delivery several and even tens
of milliseconds.

In fact, CBCA shifts complexity from
the MAC layer to the routing layer. Also,

compared to packet, link, and flow-based
channel assignments, it has the least flexibility in
exploiting multichannel diversity. Each
component, which spans all intersecting flows,
can only use one channel. As a consequence, any
two nodes in a common component cannot
transmit simultaneously unless they are at least
three or four hops apart (depending on the
interference range). In a single-hop network,
since all flows are intersected, a multichannel
network degrades to a single-channel network.

2.2 Multiradio Solutions

Using multiple radios can easily solve
MCC problems by dedicating one radio for
monitoring channel usage information.DCA uses
two transceivers, one for exchanging control
packets and the other for exchanging data
packets. The control packets are used to allocate
the channels on the data transceiver on demand.
A multichannel CSMA protocol assumes the
number of channels to be equal to the number of
transceivers per node, so that all channels can be
used simultaneously. This is very expensive. A
protocol similar to DCA in that it also dedicates
a transceiver for control purposes, but the
difference is that channel selection is done at the
receiver end based on signal-to-noise ratio. MUP
also uses two transceivers but it allows both
transceivers to exchange control messages and
data packets. xRDT extends RDT, which uses a
(possibly different) quiescent channel for each
node to receive packets, by adding a busy-tone
radio to each node in order to inform the
neighborhood of ongoing data reception, in order
to avoid collision and deafness. It proposed link-
layer protocols for routing in multiradio
multichannel ad hoc networks. Each node is
assigned a fixed interface for receiving packets
and multiple switchable interfaces for
transmitting packets. This is similar to the idea
of quiescent channel but uses more radios to
simplify overcoming MCC problems. Obviously,
the key drawback of multiradio protocols is the
increase of device size, cost, and potentially
energy consumption.

3. CAVEATS TO COOPERATIVE
PROTOCOL DESIGN

We identify three major issues in
designing a cooperative MAC protocol, which
will adversely affect protocol performance unless
properly addressed.

3.1 Control Channel Bottleneck

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

Using a dedicated control channel can
facilitate the design of a cooperative protocol,
because a control channel provides a unique
rendezvous for nodes to disseminate, gather, and
share information. However, this design scheme
may come with a drawback: When a large
number of channels and nodes are present, the
single control channel which is used to set up
communications can be highly congested and
become a performance bottleneck.

3.2 Cooperation Coordination

An MCC problem can be identified by
multiple neighboring nodes, and hence, their
simultaneous response of sending cooperative
messages will result in collision. This creates an
issue of cooperation coordination. One solution
is to make neighbors sequentially respond via a
priority-based or slot-based mechanism, thereby
ensuring all cooperative messages to be
transmitted without collision. However, this is
very inefficient because 1) there can be many
wasted (idle) intervals because not all neighbors
may identify the problem and 2) cooperative
messages pertaining to the same MCC problem
carry redundant information, and hence,
receiving all of them is not necessary. Another
solution is to let each neighbor send such
messages probabilistically, in order to reduce the
chance of collision. However, an optimal
probability (optimal in the sense of minimizing
the chance of collision) is hard to determine, and
such a scheme can result in no response which
essentially removes cooperation. Therefore, a
simple yet effective coordination mechanism is
needed.

I®

cooperaiion p, w

1/2 interference

Fig. 2. Cooperation interference

3.3 Cooperation Interference

This issue means that the cooperative

messages sent by neighbors for a transmit-
receive pair can unconsciously cause interference
to another (nearby) transmit-receive pair. This is
a new type of interference created by the
introduction of cooperation, and our simulations
found that it frequently happens and consider-
ably intensifies channel contention. As such, a
mechanism needs to be devised to address this
deleterious side effect.

4. PROTOCOL DESIGN AND
ANALYSIS

Our assumption is that each node is
equipped with a single half-duplex transceiver
that can dynamically switch between a set of
orthogonal frequency channels but can only use
one at a time. We do not assume specific channel
selection strategies; CAM-MAC runs on top of
any such strategy. For quantitative performance
evaluation, we will consider two strategies in
simulations and experiments: 1) RAND
selection, where a node randomly selects one
from a list of channels that it deems free based
on its knowledge, and 2) most recently used
(MRU) selection, where a node always selects its
MRU data channel unless it finds the channel to
be occupied by other nodes, in which case
RAND selection strategy is used.

We do not assume equal channel
bandwidth or channel conditions such as noise
levels; these can be taken into account by
channel selection strategies (e.g., choose the
channel with the highest SNR) which are not in
our assumptions. We also do not assume any
(regular) radio propagation patterns, nor assume
any relationship between communication ranges
and interference ranges.Intuitively,none of the
nodes is responsible for providing cooperation; a
node cooperates if it can (it is idle and overhears
a handshake that creates an MCC problem), and
simply does not cooperate . Actually, there often
exists at least one neighboring node that can
cooperate, and even in the worse where no one
can cooperate, the protocol still proceeds (as a
traditional noncooperative protocol).

4.1 Protocol Design

One channel is designated as the control
channel and the other channels are designated as
data channels. A transmitter and a receiver
perform a handshake on the control channel to
set up communication and then switch to their
chosen data channel to perform a DATA/ACK

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

handshake, after which they switch back to the
control channel. A transmitter sends a PR A and
its receiver responds with a PRB, like IEEE
802.11 RTS/CTS for channel reservation.
Meanwhile, this PRA/PRB also probes the
neighborhood inquiring whether an MCC
problem is created (in the case of a deaf terminal
problem, it is probed by PR A only). Upon the
reception of the PRA or PRB, each neighbor
performs a check and, if identifying an MCC
problem, sends an INV message to invalidate the
handshake (the receiver can also send INV after
receiving PRA, since it is also one of the
transmitter's neighbors). If no INV is sent and
the transmitter correctly receives PRB, it sends a
CFA to confirm the validity of PRA to all its
neighbors (including the receiver), and the
receiver will send a CFB to confirm the validity
of the PRB if it correctly receives CFA. This
marks the end of a control channel handshake. If
any INV is sent, the handshake will not proceed
and the transmitter will back off. The NCF is
merely used by the transmitter to inform its
neighbors that the PRA and CFA are invalid
when it fails to receive CFB (the receiver gets
INV after sending PRB).

cooperative collision avoidance period

SIFS
CCA

Transmitter.

PRA

CFA

CFB TMOUT

NCF

Transmitter's
neighbor"

Receiver -

Receiver's
neighbor "

CFB

PRA/PRB: probe If
CFA/CFB: confirmation NCF: negative

others in its vicinity cancel sending their INVs (a
receiver can also cancel its PRB).

6 6 11

PRA/PRB I

INV

CH|seq|

1 2

|cH|duration |

CFA/CFB seq duration

NCF/ACK seq

TA: transmitter address
RA: receiver address
CH: channel index

Fig. 4. A possible set of frame formats

Channel usage table, shown in Fig. 5.
Note that the until column does not imply clock
synchronization: It is calculated by adding the
duration in a received CFA/ CFB/INV message
to the node's own clock. Similarly, when
sending INV, a node does a reverse conversion
from until to duration using a subtraction.

until

11:30:52

11:30:56

Fig. 5. Channel usage table.

Also note that this table is by caching overheard
information while not by sensing data channels.
This is because sensing data channels often
obtains different channel status at the transmitter
and the receiver, and resolving this discrepancy
adds protocol complexity. In addition, this may
lead to more channel switching's and radio mode
(TX/RX/IDLE) changes and thus incurs longer
delay.

4.2 Caveats Revisited

Now, we explain how we address the
caveats stated in Section 3 in the design of
CAM-MAC.

Fig. 3. The CAM-MAC control channel handshake.

The cooperative collision avoidance
period is for mitigating INV collision caused by
multiple neighbors sending INVs

simultaneously. It is a simple CSMA-based
mechanism where each neighbor schedules to
send INV at a random point in this period and
continues sensing the channel. Once the node
that schedules at the earliest time starts to send.

4.2. 1 Cooperation Coordination

Recall that this issue is to coordinate
multiple neighbors to send cooperative messages
as efficiently as possible. We address this using
the cooperative collision avoidance period
described in Section 4.1. It ensures that only one
node will send out a cooperative message (INV)
in each single-broad-cast region, assuming that
propagation delay is negligible. We found via
simulations that this can reduce 70 percent-85

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

percent collisions between cooperative messages.
In case that collisions still happen (due
to propagation delay or because not all
cooperative nodes can hear each other), it is not a
serious problem because CAM-MAC makes
such collisions meaningful by using negative
feedback only. That is, since INV always means
invalidation, a collision resulting from INV still
conveys that the hand-shake should not proceed.
Actually, using negative feed-back is a logical
design. First, a node expects a binary feedback
since it selects one channel, instead of selecting a
list of channels which needs multibit feedback
indicating busy/free channels. Second, sending a
positive feedback can be misleading because
ensuring no MCC problem requires full
information while a cooperative node cannot
guarantee to have.

5. PERFORMANCE EVALUATION

We evaluate and compare five
protocols, namely IEEE 802.11, CAMMAC-
RAND, CAMMAC-MRU, UNCOOP-RAND,
and UNCOOP-MRU, using a discrete-event
simulator which we developed on Fedora Core 5
with a Linux kernel of version 2.6.9. In these
five protocols, IEEE 802.11 is used as a baseline
in comparison, X-RAND and X-MRU are two
versions of protocol X using RAND and MRU
channel selection strategies, respectively. The
protocol UNCOOP is identical to CAM-MAC
except that the cooperation element is removed,
i.e., neighboring nodes do not participate in
communication by sending INV messages. This
comparison will enable us to investigate the
value of cooperation.

We use three performance metrics: 1)
aggregate (end-to-end) throughput, 2) data
channel conflict rate, defined as the packet
collisions on data channels per second over all
nodes, and 3) packet delivery ratio, defined as
the number of data packets successfully received
by destinations. There is one control channel and
five data channels with bandwidth 1 Mbps each.
PHY and other MAC layer parameters, i.e.,
PLCP, SIFS, and retry limit, are the same as in
IEEE 802.11 . Each source generates data
packets with 2-Kbyte payload according to a
Poisson point process. The cooperative collision
avoidance period is 35|lis . In the comparison of
CAM-MAC and UNCOOP, we ignore channel
switching delay as both protocols use the same

handshake. However, in comparison to the other
protocols, namely MMAC, SSCH, and AMCP,
we use the parameters that they, respectively,
use, including channel switching delay. The
evaluation is based on 1, Impact of traffic load 2,
Impact of data payload size 3, Impact of the
number of nodes.

6. IMPLEMENTATION

There are 30 nodes forming 15 disjoint
flows in a single-hop network. In the first set of
simulations, the flows are always backlogged
and the number of channels varies from 2 to 12.
We see that CAM-MAC achieves a throughput
of 11.86 Mbps while AMCP achieves 8.5 Mbps
when there are 12 channels, which indicates a
ratio of 1.40. Furthermore, AMCP saturates at 10
channels whereas CAM-MAC still exhibits a
growing trend beyond 12 channels. In the second
set of simulations, there are four channels and
the traffic generation rate varies from 8 Kbps to
8 Mbps.Both CAM-MAC and AMCP have equal
throughput at light traffic load, but apparent
difference appears at medium to high load, and
finally, CAM-MAC saturates at 5 Mbps while
AMCP saturates at 4.2 Mbps.

120

A— ■^~

•■■■:

-*-SSCH
-■-CflM-HAC

5 10

Number olflowB

5 10 15

Number oi Flows

(b)

Fig. 6. Comparison with SSCH. (a) Disjoint flows, (b)
Nondisjoint flows

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

We implemented CAMMAC-RAND, CAMMAC-
MRU, UNCOOP-RAND, and UNCOOP-MRU on
COTS hardware implementation of single-radio
asynchronous multichannel MAC protocols for ad
hoc networks and a multichannel time
synchronization protocol. It periodically exchanges
beacon packets but data handshaking was not
implemented. It provide a test bed for routing and
channel assignment via statical manual
configuration instead of hardware implementation.
In a multichannel MAC protocol, which is designed
for sensor network data collection applications
assuming the many-to-one traffic pattern.

(a)

10' t^f 1D*

Tiatfic Geneialinn Rate psr Flaw (W&H^)

IQ"

Fig. 7. Comparison with AMCP. (a) Throughput
versus number of channels, (b) Throughput versus
traffic load. Four channels.

7 .CONCLUSION

Here, we have introduced DISH, which
is a distributed flavor of control-plane
cooperation, as a new approach to wireless
protocol design. It enables transmitter-receiver
pairs to exploit the knowledge at individual idle
neighbors to make more informed decisions in
communication. Applying DISH to multichannel

ad hoc networks, we propose a cooperative
multichannel MAC protocol called CAM-MAC,
where idle neighbors share control information
with transmitter-receiver pairs to overcome MCC
problems. This protocol uses a single transceiver
and, unlike many other protocols, is fully
asynchronous.

The simple idea of DISH turns out to be
very effective. In the comparison of CAM-MAC
with and without DISH, we observe remarkable
performance difference. In the comparison with
three recent and representative multichannel
MAC protocols, MMAC, SSCH, and AMCP,
CAM-MAC significantly outperforms all of
them. Our implementation on COTS hardware
and experiments further validates the advantages
of CAM-MAC and the DISH idea.

In a sense, DISH enables nodes to store
channel usage information at their neighbors, and
retrieve this information when it is needed. We
also highlight that this is not a compulsory
coordination mechanism, a network does not rely
on cooperation and still operates when
cooperation is not available. Ultimately, we
believe that control-plane cooperation merits due
consideration in the future design of wireless
network protocols.

8. ACKNOWLEDGEMENT

We take immense pleasure in thanking
our chairman Dr. Jeppiaar M.A, B.L, Ph.D, the
Directors of Jeppiaar Engineering College
Mr. Marie Wilson, B.Tech, MBA, (Ph.D),
Mrs. Regeena Wilson, B.Tech, MBA, (Ph.D)and
the principal Dr. Sushil Lai Das M.Sc(Engg.),
Ph.D for their continual support and guidance.
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 true. Above all, we would like to
thank God for making all our efforts success.

9. REFERENCES

[1] T. Luo, M. Motani, and V. Srinivasan,
"CAM-MAC: A Coopera-tive

Asynchronous Multi-Channel MAC
Protocol for Ad Hoc Networks," Proc. IEEE
Third Int'l Conf. Broadband Comm.,
Networks and Systems (BROADNETS '06),
Oct. 2006.

[2] J. So and N. Vaidya, "Multi-Channel MAC

7
78

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

for Ad Hoc Networks: Handling Multi-
Channel Hidden Terminals Using a Single
Transceiver," Proc. ACM MobiHoc, 2004.

[3] P. Bahl, R. Chandra, and J. Dunagan,
"SSCH: Slotted Seeded Channel Hopping
for Capacity Improvement in IEEE 802.11
Ad-Hoc Wireless Networks," Proc. ACM
MobiCom, 2004.

[4] J. Shi, T. Salonidis, and E.W. Knightly,
"Starvation Mitigation through Multi-
Channel Coordination in CSMA Multi-Hop
Wire-less Networks," Proc. ACM MobiHoc,
2006.

[5] S.-L. Wu, C.-Y. Lin, Y.-C. Tseng, and J.-P.
Sheu, "A New Multi-Channel MAC
Protocol with On-Demand Channel
Assignment for Multi-Hop Mobile Ad Hoc
Networks," Proc. Int'l Symp. Parallel
Architectures, Algorithms and Networks
(ISPAN), 2000.

[6] A. Nasipuri, J. Zhuang, and S.R. Das, "A
Multichannel CSMA MAC Protocol for
Multihop Wireless Networks," Proc. IEEE
Wireless Comm. and Networking Conf.
(WCNC), 1999.

[7] A. Nasipuri and J. Mondhe, "Multichannel
CSMA with Signal Power-Based Channel
Selection for Multihop Wireless Networks,"
Proc. IEEE Vehicular Technology Conf.
(VTC), 2000.

[8] N. Jain, S.R. Das, and A. Nasipuri, "A
Multichannel CSMA MAC Protocol with
Receiver-Based Channel Selection for
Multihop Wireless Networks," Proc. 10th
Int'l Conf. Computer Comm. and Networks
(ICCCN), 2001.

[9] A. Adya, P. Bahl, J. Padhye, and A.
Wolman, "A Multi-Radio Unification
Protocol for IEEE 802.11 Wireless
Networks," Proc. IEEE First Int'l Conf.
Broadband Comm., Networks and Systems
(BROADNETS), 2004.

Dr. T. Ravi, B.E, M.E, Ph.D
is a Professor & Head of the
Department of CSE at
KCG college of

Technology, Chennai. He
has more than 18 years of
teaching experience in
various engineering

institutions .He has published more than 20
papers in International Conferences & Journals.

AUTHORS PROFILE

P. Jesu Jayarin B.E., M.E.,
(Ph.D) working as an
Assistant Professor at
Jeppiaar Engineering

College, Chennai and he
has more than 5 years of
teaching experience. His
areas of specializations are

Mobile Computing, Computer Networks,

Network security and TCP/IP.

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

KEY GENERATION FOR AES USING BIO-METIC FINGER
PRINT FOR NETWORK DATA SECURITY

1. Dr.R.Seshadri

Director
University Computer Center
Sri Venkateswara University, Tirupati
ravalaseshadri @ gmail. com

2. T.Raghu Trivedi

Research Scholar,
Department of Computer Science
Sri Venkateswara University, Tirupati.
tamirisajtl @y ahoo.com

Abstract

Encryption is one of the Essential security
technologies for computer data, and it will go a
long way toward securing information. The
unauthorized thefts in our society have made
the requirement for reliable information
security mechanisms. Even though information
security can be accomplished with the help of
a prevailing tool like cryptography, protecting
the confidentiality of the cryptographic keys is
one of the significant issues to be deal with.
Here we proposed a biometric-crypto system
which generates a cryptographic key from the
Finger prints for encrypting and decrypting the
information the popular biometric used to
authenticate a person is fingerprint which is
unique and permanent through out a person's
life. Hence, the biometric is gone eternally and
possibly for all the applications. If your
information traverses on net to reach
destination a number of attacks may be done.
To protect from attacks we proposed a system
which will encrypt the data using AES with
biometric based key generation technique.

Key Words! Decryption, Encryption,
Histogram Equalization, Minutiae points.
Morphological Operation.

1. Introduction

Here we are using the AES for the
encryption and Decryption process. For AES
Key is important. Protecting the confidentiality
of the cryptographic keys is one of the
significant issues. We generate the key from
the receiver's finger print template. The
security to that is provided with the help of
finger print of the sender.

Biometric Cryptographic Key
Generators, or BKGs, follow a similar design:
during an enrollment phase, biometric samples
from a user are collected; statistical functions,
or features, are applied to the samples; and
some representation of the output of these
features is stored in a data structure called a
biometric template. Later, the same user can
present another sample, which is processed
with the stored template to reproduce a key [1].

Initially, the fingerprints are
employed to extract the minutiae points which
are transformed in an efficient manner to
obtain deformed points. Subsequently, the
deformed points are employed to generate the
cancelable templates which are utilized for the
extraction of irrevocable keys.

One way to protect privacy is to encrypt the
information we used a system which will
encrypt the data using AES with Novel method
of biometrics based key generation technique.
Biometric crypto systems can operate in one of
the following three modes l.Key Release
2. Key binding 3. Key generation.

In the key Release mode Biometric
authentication is decoupled from the key
release mechanism. The biometric template
and key are stored as separate entities and the
key is released only if the biometric matching
is successful

.In the key binding mode the key and
the templates are monolithically bound with in
a cryptographic frame work. It is
computationally infeasible to decode the key
or template without any knowledge of the
user's biometric data. A crypto biometric
matching algorithm is used to perform
authentication and key release in a single step.

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

In the key generation mode the key is
derived directly from the biometric data and is
not stored in the data base.

Though it is easy to implement a
biometric crypto system in the key release
mode such a system is not appropriate for a
high security application because the biometric
template is not secure template security is
critical issue because stolen templates cannot
be revoked. Key binding mode are more
secure but difficult to implement due to large
intra class variations in biometric data i.e.
samples of the same biometric trait of user
obtained over a period of time differ
substantially.

2.AES

Encryption is one of the securities
Technique for the information/data traversing
through network. Encryption is process of
transforming information using an algorithm to
make it unreadable (cipher) to any one except
sender and receiver. An encryption algorithm
along with a key used in encryption and
decryption of data .AES is one of the popular
Algorithms used in symmetric key
cryptography [3, 6]. It is symmetric block
cipher that can encrypt/decrypt information. It
is used at top secret level. AES supports key
sizes 128,192 and 256 bits and will serve as
replacement for the DES, which has a less key
size. AES can encrypt data much faster than
DES enhancement (Triple DES)..

• Cipher text=EK(plain text)

• Plain text= Dk (cipher text)

• Dk (Ek (plain text))=plain text

3. Cryptography and biometrics

Cryptography provides security to the
information which is transferring over the
insecure channel. Here key will play a major
role, because it is used for encryption as well
as for decryption. Lengthy key used to encrypt
and decrypt sending and receiving the
messages respectively. These keys can be
guessed/ cracked. Maintaing and sharing the
lengthy keys is critical problem. This can be
overcome with the help of biometric system
[5].

There are different biometric techniques. We
concentrate on fingerer prints to generate key.

Key vector Is formed based on minutiae points
(ridge ending and ridge bifurcation)

Finger Print Image

Minutiae feature extraction

Minutiae point location
Set

Key Vector Generation

Figure 1: Generating key vector from finger
print

4. Key Generation from Finger Print

Here we have to extract the minutiae points
from the fingerprint to generate the key. The
process is as below

Finger Print

Preprocessing

^ r

Binarized

Morphological
Operation

Minutiae Extraction

False Minutiae Removal

Key Generation

Figure2: Key generation from minutiae
points

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

4.1 Extraction of Minutiae Points from
Fingerprints

The extraction of minutiae points
from the fingerprint image is discussed in this
section. It is supposed that fingerprints are
distinct across individuals and across the
fingers of a particular individual [1] . Since
many existing fingerprint authentication
systems are based on minutiae points, which
are feature points extracted from a raw
fingerprint image, we have employed the
minutiae points in our scheme as well.

A fingerprint can be defined as a pattern of
ridges and valleys on the tip of the finger. A
fingerprint is therefore described by the
distinctiveness of the local ridge features and
their relationships. Minutiae points denote
these local ridge characteristics that appear
either at a ridge ending or a ridge bifurcation.
The point where the ridge comes to an abrupt
end is known as ridge ending and the ridge
bifurcation is denoted as the point where the
ridge divides into two or more branches [1].
The major steps involved in the minutiae
points' extraction are as follows:

• Segmentation

• Minutiae Extraction

4.1.1Segmentation

The first step in the minutiae points' extraction
is segmentation. The input fingerprint image is
segmented from the background to actually
extract the region comprising the fingerprint,
which ensures the removal of noise.
Segmentation of an image represents the
division or separation of the image into regions
that have similar attributes. At first, the image
is pre-processed. The pre-processing phase
includes the following: histogram equalization.
Later, the pre-processed image is divided into
blocks and segmentation is carried out. The
sample fingerprint images are shown

Figure 3: Two Sample Fingerprint Images
4.1.1.1 Pre-Processing

The pre-processing of fingerprint images
includes Histogram Equalization

4.1.1.1. a. Histogram Fqualization

Histogram equalization amplifies the local
contrast of the images, particularly when they
are represented with very close contrast values.
It is possible to distribute intensity through the
histogram with the aid of this regulation.
Histogram equalization utilizes a mono tonic,
non-linear mapping that re-assigns the
intensity values of pixels in the input image in
such a manner that the output image comprises
a uniform distribution of intensities (i.e. a flat
histogram). The original histogram of a
Fingerprint image is of bimodal type, the
histogram after the histogram equalization
transforms all the range from to 255 which
results if enhanced visualization effect [7]. The
results of histogram equalization are depicted
in Figure 4.

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

Figure 4: Fingerprint Images after
Histogram Equalization

binarization process. Then, the pixel value is
set to a binary value one when the value is
greater than the global threshold, or else a zero
is set as the pixel value. The foreground ridges
and the background valleys are the two level of
information held by the ensuing binary image.
Removal of distortions present in the image is
performed followed by the retrieval of the
exact skeleton image from the image.

4.1.2.2 Morphological Operation

The binary morphological operators are
applied on the binarized fingerprint image.
Elimination of any obstacles and noise from
the image is the primary function of the
morphological operators. Furthermore, the
unnecessary spurs, bridges and line breaks are
removed by these operators. Then thinning
process is performed to reduce the thickness of
the lines so that the lines are only represented
except the other regions of the image. Clean
operator, Hbreak operator. Spur operator and
Thinning are the morphological operators
applied.

4.1.2 Minutiae Points Extraction

4.1.2.3 Minutiae points' extraction

Finally, the minutiae points are extracted from
the enhanced fingerprint image. The steps
involved in the extraction of minutiae points
are as follows:

• Binarization

• Morphological Operations

• Minutiae points' extraction

Initially, the enhanced image is binarized.
After binarization, morphological operations
are performed on the image to remove the
obstacles and noise from it. Finally, the
minutiae points are extracted using the
approach discussed.

4.1.2.1 Binarization

The binary images with only two levels of
interest: The black pixels that denote ridges
and the white pixels that denote valleys are
employed by almost all minutiae extraction
algorithms. A grey level image is translated
into a binary image in the process of
binarization, by which the contrast between the
ridges and valleys in a fingerprint image is
improved. Hence, the extraction of minutiae is
achievable. The grey-level value of every pixel
in the enhanced image is analyzed in the

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

Figure 5:
Points

Fingerprint Images with Minutiae

Ridge Thinning is to eliminate the redundant
pixels of ridges till the ridges are just one pixel
wide, uses an iterative, parallel thinning
algorithm. In each scan of the full fingerprint
image, the algorithm marks down redundant
pixels in each small image window (3x3). And
finally removes all those marked pixels after
several scans. After the fingerprint ridge
thinning, marking minutia point is relatively
easy .For each 3X3 window, if the central
pixel is 1 and has exactly 3 one-value
neighbors, then the central pixel is a ridge
branch, if the central pixel is 1 and has only 1
one -value neighbors , then the central pixel is
ridge ending. Suppose both the uppermost
pixel with value 1 and the rightmost pixel with
value 1 have another neighbor outside the 3X3
window, so the two pixels will be marked as

branches too. But actually only one branch is
located in the small region. So a check routine
requiring that none of the neighbors of a
branch are branches is added.

4.1.2.4 False Minutia Removal:

The preprocessing stage does not totally heal
the fingerprint image .For example, false ridge
breaks due to insufficient amount of ink and
ridge cross-connections due to over inking are
not totally eliminated. Actually all the earlier
stages themselves occasionally introduce some
artifacts which later lead to spurious minutia.
These false minutias will significantly affect
the accuracy of matching if they are simply
regarded as genuine minutia. So some
mechanisms of removing false minutia are
essential to keep the fingerprint verification
system effective.

Key Generation From Minutiae Points

In this section we explain the key Generation
Algorithm [2] Assumptions

Length of the AES key

Minutiae point set

key length

Size of Minutiae pint set

Seed value

seed limit

(x, y) -co -ordinate of a minutiae point

Key Vector

Step 1: The Extracted minutiae pints are
represented

Mp={mi} i=l Np

Step2: The initial key vector is defined as
follows

Kv= {xi:p(xi)}i=l Kl

Where p(x) = Mp [1% Np] -f Mp [(i-^l) % Np]

-fS

i=l.

.Kl

step3: Initial value of S is equal to total
Number of Minutiae pints . The value of S will
be dynamically changed as follows

S=Kv (i) %Sl,-l<i<Kl

Step4: Initial key vector (Kv) is converted in to
a

Matrix Km of size Kl / 2 * Kl / 2

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

Km= (aij) Kl / 2 * Kl / 2
Step 5:
A intermediate key vector is generated as

follows KIV= {Ki: (m (ki)} i=l Kl

Where

M (k) =IAijl,

Aij=Km i, j: i + size + j + size, -l<i<kl/2

Aij is a sub matrix formed from the key matrix

Step 6: Final key vector is formed is

Sc=

l,ifKlV[i]>mean(KlV)
0, other wise

5. Conclusion

From the above discussion I have proposed a
method for providing security to the
information transferring on the network using
Encryption and a novel approach for
fingerprint based cryptography system. Here
we used fingerprint patterns, which are stable
through out person's lifetime. Since it creates
more complexity to crack or guess the crypto
keys. This approach has reduced the
complicated sequence of the operation to
generate crypto keys as in the traditional
cryptography system.

6. References

1. N.Lalithamani, K.P. Soman Irrevocable
Cryptographic Key Generation from
Cancelable Fingerprint Templates: An
Enhanced and Effective Scheme". European
Journal of Scientific Research ISSN 1450-
216X Vol.31 No.3 (2009), pp.372-387

2. P.Arul, Dr.A.Shanmugam "Generate a
Key for AES Using Biometric For VOIP
Network Security" Journal of Theoretical and
Applied Information Technology 2009.107-
112.

3. "Advanced Encryption Standard "from
http://en.wikipedia.org/wiki/Advanced_Encryp
tion_ Standard

4. Jain, A.K., Ross, A. and Prabhakar, S,
"An introduction to biometric recognition,"
IEEE Transactions on Circuits and Systems for
Video Technology, Vol. 14, No. 1, pp: 4- 20,
2004.

5. Umut Uludag, Sharath Pankanti, Salil
Prabhakar, Anil K .Jain "Biometric
Cryptosystems Issues and Challenges"
Proceedings of the IEEE 2004.

6. Announcing the "ADVANCED
ENCRYPTION STANDARD (AES)" -Federal
Information, Processing Standards Publication
197, November 26, 2001

7. Jain, A.K.; Prabhakar, S.; Hong, L.;
Pankanti, S., "Filter bank-based fingerprint
matching", IEEE Transactions on Image
Processing, vol. 9, no. 5, pp: 846-859, May
2000, Doi:10.1 109/83.841531.

Authors Profile

— Dr.R.Seshadri was born

0^^ in Andhra Pradesh, India,

^TB(*|| in 1959. He received his

1"- B.Tech degree from

\ J Nagarjuna University in

' ^ ^ 1981. He completed his

Masters 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 published

number of papers in national and international

journals. At present 12 members are doing

research work under his guidance in different

areas.

Wkm

Mr.T.RaghuTrivedi received
MCA degree from Andhra
University, Vizag He received
his M.Tech in Computer
Science from Nagarjuna
University. He is a research
scholor in S.V.University
Tirupati, Andhra Pradesh.

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

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

Vol 8, No. 3, 2010

Classification of Five Mental Tasks Based on Two
Methods of Neural Network

Vijay Khare^

Jaypee Institute of Information Technology

Dept. of Electronics and Communication, Engineering

Nioda, India

Email : vijay .khare@jiit.ac.in

Sneh Anand^

Indian Institute of Technology,
Centre for Biomedical Engineering Centre
Delhi, India
Email : sneh@iitd.emet. in

Abstract — In this paper performance of two classifiers based on
Neural Network were investigated for classification of five
mental tasks from raw Electroencephalograph (EEG) signal. Aim
of this research was to improve brain computer interface (BCI)
system applications. For this study, Wavelet packet transform
(WPT) was used for feature extraction of the relevant frequency
bands from raw electroencephalogram (EEG) signals. The two
classifiers used were Radial Basis Function Neural Network
(RBFNN) and Multilayer Perceptron Back propagation Neural
Network(MLP-BP NN) . In MLP-BP NN five training methods
used were (a) Gradient Descent Back Propagation (b) Levenberg-
Marquardt (c) Resilient Back Propagation (d) Conjugate
Learning Gradient Back Propagation and (e) Gradient Descent
Back Propagation with movementum.

Index Terms — Electroencephalogram (EEG), Wavelet Packet
Transform (WPT), Radial Basis Function Neural
Network(RBFNN), Multilayer Perceptron back propagation
Neural Network(MLP-BP NN), Brain computer interfaces (BCI).

Introduction

Brain signals extracted through EEG carry information
needed for the design and development of brain computer
interface (BCI) systems. It is well documented that proper
feature extraction and classification methods are the key
features deciding the accuracy and speed of BCI systems [1-
5]. ANN has been more widely accepted as one of the best
classification method to distinguish various mental states from
relevant EEG signals.

Past two decades have witnessed the importance of innovative
BCI with voice, vision and a combination of these, as a
communication platform [6-9]. Effective attempts have been
made to achieve successful BCI systems based on bioelectric
signals. They were mainly to help patients with various
neuromuscular disorders by providing them a way of
communication to the world, through extracting information
from their intensions. So far the accuracy of the classification
has been one of the main pitfalls of the existing BCI systems,
since it directly affects the decision made as the BCI output.
The speed &accuracy could be improved by implementing

Jayashree Santhosh^
Indian Institute of Technology,
Computer Services Centre
Delhi, India
Email : jayashree@cc.iitd.ac.in

Manvir Bhatia"^

Sir Ganga Ram Hospital,

Department of Sleep Medicine,

New Delhi, India

Email : manvirbhatial@yahoo.com

better methods for feature extraction and classification [10-
16]. In this study, wavelet packet transform (WPT) method
was used to capture the information of mental tasks from eight
channel EEG signals of nine subjects. The coefficients of
wavelet packet transform (WPT) were used as the best fitting
input vector for clssifiers [17]. The two classifiers (RBFNN
and MLP-BP NN) were used to compare the performance to
discriminate five mental tasks.

II. METHODOLOGY
A. Subjects

Nine right-handed healthy male subjects of age (mean:
23yr) having no sign of any motor- neuron diseases were
selected for the study. A pro-forma was filled in with detail of
their age & education level as shown in table 1. The
participants were student volunteers for their availability and
interest in the study. EEG data was collected after taking
written consent for participation. Full explanation of the
experiment was provided to each of the participants.

Table 1 : Clinical Characteristics of Subjects

S.No.

Subject

Age

Educational
status

Subject 1

Subject 2

Subject 3

Subject 4

M.TECH

Subject 5

Subject 6

Subject 7

M.TECH

Subject 8

Subject 9

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

B. EEG Data Acquisition

EEG Data used in this study was recorded on a Grass
Telefactor EEG TwinS Machine available at Deptt. of
Neurology , Sir Ganga Ram Hospital, New Delhi. EEG
recording for nine selected subjects were done for five mental
tasks for five days. Data was recorded for 10 sec during each
task and each task was repeated five times per session per day.
Bipolar and Referential EEG was recorded using eight
standard positions C3, C4, P3, P4, 01 02, and F3, F4 by
placing gold electrodes on scalp, as per the international
standard 10-20 system of electrode placement as shown in
figure 1. The reference electrodes were placed on ear lobes
and ground electrode on forehead. EOG (Electooculargram)
being a noise artifact, was derived from two electrodes placed
on outer canthus of left and right eye in order to detect and
eliminate eye movement artifact. The settings used for data
collection were: low pass filter IHz, high pass filter 35 Hz,
sensitivity 150 micro volts/mm and sampling frequency fixed
at 400 Hz.

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

Vol 8, No. 3, 2010
Data collected from nine subjects performing five mental tasks
were analyzed. The following mental tasks were used to
record the appropriate EEG data.

Figure 1:- Montage for present study
C. Experiment Paradigm

An experiment paradigm was designed for the study and
the protocol was explained to each participant before
conducting the experiment. In this, the subject was asked to
comfortably lie down in a relaxed position with eyes closed.
After assuring the normal relaxed state by checking the status
of alpha waves, the EEG was recorded for 50 sec, collecting
five session of lOsec epoch each for the relaxed state. This
was used as the baseline reference for further analysis of
mental task. The subject was asked to perform a mental task
on presentation of an audio cue. Five session of lOsec epoch
for each mental task were recorded, each with a time gap of 5
minute (as shown in figure2). The whole experiment lasted for
about one hour including electrode placement.

>^K

Mental

Figure 2: Timing of the Protocol

Sec.

• Movement Imagination: -The subject was asked to
plan movement of the right hand.

• Geometric Figure Rotation: -The subject was given 30
seconds to see a complex three dimensional object,
after which the object was removed. The subject was
instructed to visualize the object being rotated about
an axis.

• Arithmetic Task: -The subject was asked to perform
trivial and nontrivial multiplication. An example of a
trivial calculation is to multiply 2 by 3 and nontrivial
task is to multiply 49 by 78. The subject was
instructed not to vocalize or make movements while
solving the problem.

• Relaxed: - The subject was asked to relax with eyes
closed. No mental or physical task to be performed at
this stage.

D. Feature Extraction

The frequency spectrum of the signal was first analyzed
through Fast Fourier Transform (FFT) method. The FFT plot
of signals from the most relevant electrode pairs were
observed along with average change in EEG power for each
mental tasks as shown in figure (2-6).

For relaxed, the peaks of power spectrum almost
coincide (or difference of (0-10 %) for central and occipital
area in the alpha frequency range (8-13Hz). EEG recorded
with relaxed state is considered to be the base line for the
subsequent analysis. Mu rhythms are generated over
sensorimotor cortex during planning a movement. For
movement imagery of right hand, maximum upto 50% band
power attenuation was observed in contralateral (C3 w.r.t C4)
hemisphere in the alpha frequency range (8-13Hz). For
geometrical figure rotation, the peak of the power spectrum
was increased in right hemisphere rather than left in the
occipital area for the alpha frequency range (8-13Hz). For both
trivial and nontrivial multiplication, the peak of the power
spectrum was increased in left hemisphere rather than right
hemisphere in the frontal area for the alpha frequency range
(8-13Hz).

The data was preprocessed using Wavelet packet
transform to extract the most relevant information from the
EEG signal. [18-19]. By applying Wavelet packet transform
on the original signal wavelet coefficients in the (8-13Hz)
frequency band at the 5^^ level node (5, 3) were obtained. We
were able to reduce 1 second of EEG data to 21 coefficients.
The signal was reconstructed at node (5,3). These coefficients
are scaled and used as the best fitting input vector for
classifiers.

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

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

Vol. 8, No. 3, 2010

Relax B-C4 R-C3

Fig 2:- Power Spectra for Relax state at C3 and C4 channel

Right Hand movement B-C4 R-C3

g 3:- Power Spectra for planning of right hand movement at C3 and C4

rotation B-02 R-01

Power Spectra for visual rotation at Ol and 02 channel

simple arthemetic B-F4 R-F3

complex arithmatic B-P4 R-P3

Fig 6:- Power Spectra for complex Arithmetic P3 and P4 channel

E. Classifier

The main advantage of choosing artificial neural
network for classification was due to fact that ANN's could be
used to solve problems, where description for the data is not
computable. ANN could be trained using data to discriminate
the feature. We have compared two classifiers Radial Basis
Function Neural Network (RBFNN) and Multilayer
Perceptron Back propagation Neural Network (MLP-BP NN).
In MLP-BP NN five training methods used were (a)
Gradient Descent Back Propagation (b) Levenberg-Marquardt
(c) Resilient Back Propagation (d) Conjugate Learning
Gradient Back Propagation and (e) Gradient Descent Back
Propagation with movementum.

• Radial Basis Function Neural Network

For this (RBFNN) classifier, a two layer network was
implemented with 21 input vectors, a hidden layer consisting
as many as hidden neurons as there are input vectors with
Gaussian activation function and one neuron in the output
layer [20-23]. RBFNN produce a network with zero error on
training vectors. The output neuron gives 1 for a mental task
and for relax task.

• Multilayer
Network

Perceptron Back Propagation Neural

For this classifier, a two layer feed forward neural
network was used with topology of {10, 1}. 10 neurons in
hidden layer and 1 neuron in output layer. The neural network
was designed to accept a 21 element input vector and give a
single output. The output neuron was designed to give for
baseline (relax task) and 1 for mental task. The five different
training methods used for this classifier were Gradient
Descent, Resilient Back propagation, Levenberg-Marquardt,
Conjugate Gradient Descent and Gradient Descent back
propagation with movementum [24-27]. Parameter used for
five training methods of neural network for classification of
five mental tasks as shown in the table 2.

Fig 5:- Power Spectra for Simple Arithmetic F3 and F4 channel

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

Table 2
Parameter used for different algorithms
with topology { 10, 1}

Gradient descent with momentum (GDM)

Topology {10,1}

A=01.Mu = 0.01

MSE= le-5

Epoach=5000

Gradient descent method(GDBP)

Topology {10,1}

A=01

MSE=lexp-(5)

Epoach=5000

ResiUent Back propagation(RBP)

Topology {10,1}

A=01

MSE=lexp-(5)

Epoach=5000

B=0.75 and P 1=1.05

Conjugate gradient descent(CGBP)

Topology {10,1}

A=01

MSE=lexp-(5)

Epoach=5000

Levenberg-Marquardt(LM)

Topology {10,1}

Mu=.01

MSE=lexp-(5)

Epoach=5000

Mu_dec=0.1 and Mu_inc=10

F. Performance

The study evaluated the performance of two
classifiers (RBFNN and MLP-BP NN) for classification of
five mental tasks. For MLP-BP NN classifier five different
training methods used were Gradient Descent, Resilient Back
propagation, Levenberg-Marquardt, Conjugate Gradient
Descent and Gradient Descent back propagation with
movementum. 60% of entire EEG data (five sessions, five
mental tasks with nine subject) was taken as training data.
Remaining 40% of EEG data was taken as test data and the
performances were recorded. The entire analysis of the
recorded data was carried out using Matlab® 7.0 from
Mathworks Inc., USA.

Performance (Re) is calculated in percentage (%) as ratio
between correctly classified patterns in the test set to the total
number of patterns in the test set [28].

_ Number of correctly classified test patterns
Total number of patterns in the test set

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

Vol 8, No. 3, 2010

classification of each mental task. WPT is an excellent signal
analysis tool, especially for non stationary signals. Hence in
the present study, WPT was used for feature extraction [29].

As per literature most prominent area in brain for domain of
information during five mental tasks was shown in tableS.
Amplitude of power spectrum almost coincides in central and
occipital area at a particular base frequency (8-13Hz) for
relaxed states [12].

The frequency spectrum of the signal was observed
that the amplitude of the power spectrum for alpha frequency
range (8-13Hz) had an attenuation in contralateral area for
movement imagery task [12 17].

For geometrical figure rotation. It was observed that
the amplitude of the power spectrum increased in the right
occipital region for alpha frequency range (8-13Hz) [30 31].

For trivial multiplication, it was observed that the
amplitude of the power spectrum increased in the left frontal
region for alpha frequency range (8-13Hz).For nontrivial
multiplication, it was observed that the amplitude of the power
spectrum increased in the left parietal region for alpha
frequency range (8-13Hz) [31].

III. Result and Discussion

Nine right-handed male subjects participated in the
experiments. The subjects were asked to perform five mental
tasks namely relaxed, movement imagery, geometrical figure
rotation and arithmetic task (trivial and non trivial
multiplication). Out of 50 sec data recorded data the most
relevant one second epoch of signal were used for

For MLP-BP NN classifier. Resilient back
propagation training method showed better performance than
other back propagation training methods (Gradient Descent
method Levenberg-Marquardt, Conjugate Gradient Descent
and Gradient Descent back propagation with movementum)
for classification of five mental tasks (As shown in table 4).

Table 3: Domain of information

Tasks

Domain of
information

(Contralateral/
Ipsilateral)

Type of

change in

amplitude of

alpha

rhythm(8-

13Hz)

Base line

Occipital,
Central

Contralateral

Coincide

Movement
Imagination

Central

Contralateral

Decreased

Geometrical figure
rotational

Occipital

Ipsilateral

Increased

Arithmetic' s(trivial
and non trivial )
operation

Frontal,
parietal

Ipsilateral

Increased

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

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

Vol. 8, No. 3, 2010

The present study was a comparison of two classifier, MLP-
BP NN with Resihent back propagation training method and
RBFNN to discriminate five mental tasks effectively. From
figure(7-12) we can say that RBF Neural Network has better
performance as compare to MLP-BP NN with Resilient back
propagation method for classification of five mental tasks
movement imagination(M), figure rotation(R), simple
arithmetic (SA) task and complex arithmetic (CA) task, w.r.t
baseline. Average accuracy was obtained 100% by using
RBFNN classifier.

IV. Conclusion

For various applications of BCI systems, it is necessary that
EEG feature related to the human intentions were to be
uniquely identified as accurate as possible. In this paper, the
two classifiers used were, MLP-BP NN with resilient back
propagation training method and RBFNN. Radial basis
function neural network was found to be most suitable for
classification of five mental tasks.

Radial basis function networks showed a better
capability for solving larger data size problems at fast learning
speed, because of their capability of local specialization and
global generalization. Various other classification methods
could be analyzed in future for better performance.

Gradient Descent back Propagation

Conjugate Gradient BP

o 98]

2|94-
g o 92

CA R
Mental tasi^s

Fig 9: Classification accuracy CGBP training methods

Gradient descent method with
movementum

O 96

94
_ 92
J. 3 90
g 8 88
86

SA CA

Mental tasks

*S 100

95
90
85
80

>»

*->

-J

SA CA R M

mental tasks

Fig 7: Classification accuracy using GD BP training methods

Fig 10: Classification accuracy GDM training methods

Levenberg-Marquardt

n=^ 1

SA CA R M

Mental tasks

Fig 11: Classification accuracy LM training methods

= 3
e So

Resilient Back propagation

98 1

SA CA R M

Mental tasks

1 >•

0)
Q.

150 n

100

Radail Basis Function

SA CA R M
Mental Task

Fig 8: Classification accuracy RBF training methods

Figure 12: Classification accuracy for RBFNN

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

Acknowledgment

The authors would Uke to acknowledge their gratitude to the
scientific and technical staff of EEG Laboratory of Sir Ganga
Ram hospital, New Delhi for the help in carrying out the
experiment.

References

[I] Lotte P., M. Congedo, A. Lecuyer, F. Lamarche, B Arnaldi A
Review of Classification algorithms for EEG bases brain
computer interface Journal of neural Engineering, vol. 4, R1-R13,
2007.

[2] .Wolpaw J.R , N.Birbaumer, D.J Mc Farland, G.Plurtscheller,
T.M.Vaughan, Brain computer Interfaces for communication and
control Clinical Nurophys. , 113,767-791,2002.

[3] Pfurtschelle G. , D.Flotzinger, and J.Kalcher Brain Computer interface-
A new communication device for handicapped people J Microcomput.
Applicate., vol. 16,. pp.293- 299,1993.

[4] Wolpaw J.R, T.M.Vaughan and E.Donchin EEG Based Communication
prospects and problems IEEE,Trans.Rehab.Eng.vol.4,.pp. 425-430,
Dec.1996.

[5] Keirn Z. A. and J. I. Aunon, A new mode of communication between
man and his surroundings IEEE Trans. Biomed. Eng., vol. 37, no. 12,
pp. 1209-1214, Dec. 1990.

[6] Wolpaw J.R., G.E. Leob, B.Z. Allison, E.Donchin, J. N. Turner BCI
Meeting 2005-Wokshop on Signals and Rerecording Methods IEEE
transactions on Neural systems and rehabilitation Engineering,
vol.14, no 2, June 2006.

[7] Anderson Charles W., Erik A. Stolz, Sanyogita Shamsunder
Multivariable Autoregressive Model for classification of Spontaneous
Electroencephalogram During Mental Tasks IEEE Trans. Biomed. Eng.,
vol. 45 no-3, pp.277-278, 1998.

[8] Elean A, Curran and Jamaica Strokes learning to control brain activity A
Review of the production and control of EEG components for driving
Bain computer interface systems Brain and cognition, 51,326-336, 2003.

[9] Wolpaw J. R., D. J. Mc Farland, T.M. Vaughan The wads worth Centre
Brain Computer interface research and Development Program IEEE
Trans, on Neural System and rehab. Eng., 1 1(2)) 204-207, 2003.

[10] Liu Hailong, Jue Wang, C.Zheng and Ping He Study on the Effect of
Different Frequency band of EEG Signal on the Mental Tasks
Classification IEEE Engineering in medicine and biology 27th annual
conference shanghai china, sept. 1-4, and 2005.

[II] Boostani R. , B. Graimann, M.H.Moradi, G. Plurfscheller Comparison
approach toward finding the best feature and classifier in cue BCI Med.
Bio. Engg., Computer 45,403-413, 2007.

[12] Pfurtscheller G., C. Neuper, A. Schlogl, and K. Lugger Separability
of EEG signals recorded during right and left motor imagery using
adaptive auto regressive parameters IEEE. Trans, on rehabilitation
Engineering, 6(3), pp.3 16-325, 1998.

[13] Palaniappan R. Brain computer interface design using band powers
extracted during mental task proceeding of the 2ndInternational IEEE
EMBS Conference on Neural Engineering 321- 324, 2005.

[14] Kouhyar Tavakolian, Faratash Vasefi, Siamak Rezaei, Mental
Task Classification for Brain Computer Interface Application, first
Canadian student conference on biomedical 2006.

[15] Palaniappan Ramaswamy Utilizing Gamma Band to Improve
Mental task based Brain -Computer Interface Design IEEE Trans, on
Neural systems and rehabilitation Engg. , vol.14, no 3, September
2006.

[16] Lee J.C, D.S Tan Using a low cost Electroencephalograph for the
Mental task classification in HCI Research UIST 06, October 15-18,
montreux, Switzerland 2006.

[17] Santhosh Jayashree, Manvir Bhatia, S.Sahu, Sneh Anand Quantitative
EEG analysis for assessment to plan a task in ALS patients, a study of
executive function (planning) in ALS, Cognitive brain research 22, 59-
66, 2004.

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

Vol 8, No. 3, 2010
[18] Akay M Wavelet in biomedical engineering, annalsin Biomedical
Engineering 23 (5), 531-542, 1995.

[19] Strang Gilbert and Troung Ngugen. Wavelet and Filter Banks Wellesley
Cambridge press, 1997.

[20] Larsson Elisabeth, Krister Ahlander, and Andreas Hall Multi-
dimensional option pricing using radial basis functions and the
generalized Fourier transform. In J. Comput. Appl. Math., 2008.

[21] Pettersson Ulrika, Elisabeth Larsson, Gunnar Marcusson, and Jonas
Persson Improved radial basis function methods for multi-dimensional
option pricing. In J. Comput. Appl. Math., 2008.

[22] Warwick K. , J. Mason ., and Sutanto E Centre Selection for Radial
Basis Function Network", Artificial Network and Genetic algorithms
proceeding of the international conference ,Ales, France ,D.
Pearson,N.Seetle and R. Albrecht(Eds) 309-312, 1995.

[23] Chen S. , C.F.N. Cowan, and P. M. Grant Orthogonal Least Squares
Learning Algorithm for Radial Basis Function Networks IEEE
Transactions on Neural Networks, vol. 2, pp. 302-309, no. 2, March
1991.

[24] Ravi K. V. R. and R. Palaniappan Neural Network Classification of Late
gamma band electroencephalogram features Soft Comput, vol. 10, no.
2, pp. 163-169, 2006.

[25] Haykin S. Neural Network- A Comprehensive foundation, 2nd
Edition, Prentice Hall, 2000.

[26] Hagen M., H. Demuth, and M. beale. Neural Network design Boston
MA., PWS Publishing, 1996.

[27] Zhou Shang- Ming, John Q.Gan, Francisco Sepulved Classifying mental
tasks based on features of higher-order statistics from EEG signals in
brain-computer interface Information Sciences: an International Journal,
Volume 178 ,Issue 6, Pages 1629-1640,2008.

[28] Cheng M., Xiarong Gao, S.Gao, D.Xu Design and implementation of
a brain computer interface with high transfer rates IEEE, Transactions
on Biomedical Engg., vol-49, no. 10 October 2002.

[29] Ting Wu , Yan Guo-zhenga, Yang Bang-huaa, Sun Hong EEG feature
extraction based on wavelet packet decomposition for brain computer
interface Measurement, Elsevier journal 41 2008, 618-625, 2008.

[30] Nikolaev A.R. and A.P. Anokhin EEG frequency ranges during
reception and mental rotation of two and three dimensional objects.
Neuroscience and Bheaviour physiology , v. -28, no-6, 1998.

[31] Osaka M. Peak alpha frequency of EEG during a mental task: task
difficulty and hemisphere difference Psychophysiology,pp 101-
105,vol.21,1984.

Biographies

Vijay Khare is currently pursuing his PhD in Bio Signal Processing at the
Indian Institute of Technology, Delhi. He did his M.Tech in
Instrumentation & Control, from NSIT Delhi. He is currently,
with the Dept. Electronics and Communications Engineering at
the Jaypee Institute of Information Technology. His research
interests are Neural Networks, Brain Computer Interfacing, and
Control Systems.

Dr. Jayashree Santhosh completed her B.Tech in Electrical
Engineering from University of Kerala, M Tech in Computer &
Information Sciences from Cochin University of Science and
Technology, Kerala and Ph.D from IIT Delhi. She is a Fellow
member of lETE, Life member of Indian Association of Medical
■ I Informatics (lAMI) and Indian Society of Biomechanics (ISB).

1^:1 m, y[qx research interests include IT in Healthcare Systems and was
associated with a project on IT in Health Care at City University of Hong
Kong. She is also associated with various projects with Centre for Bio-
Medical Engineering at IIT Delhi in the area of Technology in Healthcare.
Her research interests focus on Brain Computer Interface Systems for
the Handicapped and in Neuroscience.

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

Prof. Sneh Anand is a professor and head, Center for
Biomedical Engineering, Indian Institute of Technology, Delhi.
She did B.Tech in Electrical Engg, from Punjab University,
Patiala, and M.Tech in Instrumentation & Control from IIT
Delhi and Ph.D. in Biomedical Engg. from IIT Delhi. Her
research interests include biomedical instrumentation,
rehabilitation engineering, biomedical transducers and
Sensors.

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

Vol 8, No. 3, 2010
completed her MBBS in 1981, and Doctor of Medicine in 1986 from Christian
Medical College and Hospital, Ludhiana. DM in Neurology 1993, from All
India Institute of Medical Sciences. She is a member of Indian Academy of
Neurology, Indian Epilepsy Society, Indian Sleep Disorders Association,
World Association of Sleep Medicine, International Restless Legs Society
Study Group and American Academy of Electrodiagnostic Medicine. Dr.
Manvir Bhatia has been invited to deliver lectures in National & International
workshops, conferences on topics related to Neurology, Epilepsy, Sleep
Medicine and has sleep published papers in leading journals.

Dr. Manvir Bhatia is the Chairperson of Dept. of
SleepMedicine at Sir Ganga Ram Hospital, New Delhi and is
also a Senior Consultant Neurologist.Dr. Manvir Bhatia

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

(IJCSIS) InternationalJournal of Computer Science and Information Security,

Vol. 8, No. 3, 2010

Sixth order Butterworth Characteristics using LV
MOCCII and Grounded Components.

T. Parveen

Electronics Engineering Department, Z. H. College of Engineering & Technology

AMU, Aligarh, INDIA
tahiraparveen2@gmail.com

Abstract — This paper introduces an active realization of the sixth
order current mode Butterworth filter function using low
voltage(LV) Multioutput current conveyors (MOCCII) and
grounded passive components. The proposed realization is based
on cascading an insensitive single input multi output (SIMO)
current mode universal biquadratic filter (UBF). The UBF is
constructed employing only two MOCCIIs, four grounded
components, that lead to simple structure, easy to design and
suitable for IC fabrication. The proposed UBF can realize all
standard biquadratic responses without any matching conditions
and has current outputs at a high impedance terminal, which
enable easy cascadablity. An example of eighth order current
mode Butterworth filter has been considered. The filter has the
advantages of minimum requirement of active and passive
component count, low sensitivity, and high performance. The
performance of the filter is verified through PSPICE simulation
using low supply voltage.

Keywords-component; Current mode circuits, High order
Butterworth filter, Multioutput current conveyors (MOCCIIs),
Universal biquadratic filter.

Introduction

The current mode signal processing techniques have been
received a wide attention due to its wide band width, improved
linearity, wide dynamic range, low voltage operation as
compared to voltage mode signal processing. By introducing
Multi output current conveyors (MOCCIIs) in active filter
designs, new more advantageous topologies [5] have been
obtained, by which current outputs and current feedback can be
developed when multiple current outputs are used [7].

Recently design of current mode universal biquadratic
filters (UBF) with single input multi output (SIMO) have
received considerable attention due to their convenience, high
performance and greater functional versatility in terms of signal
processing for practical applications [1-7]. Interconnections of
relevant output currents provides the low pass, band pass, high
pass, band elimination and all pass filter responses from the
same circuit.

This paper presents a minimum active-RC circuit
realization of eighth order current mode Butterworth filter
function using low voltage Multioutput current conveyors
(MOCCII) and grounded Passive components. The proposed
realization is based on cascading a novel single input multi
output (SIMO) current mode universal biquadratic filter (UBF),

which uses two Multi output current conveyors (MOCCII)
along with only four grounded passive components. The
presented UBF can realize the low pass, high pass, band pass,
band elimination and all pass filter responses through
appropriate selection of input and output terminals without any
matching conditions. The filter circuit is simple in structure and
has high impedance outputs enables easy cascading in current
mode operations, and can be used for the realization of any
type of high order filter function. The proposed UBF circuit
also has the additional advantage of low component count, and
high performance at low supply voltage, over previously
reported literature [1-7, 10]. The eighth order current mode
realization with Butterworth coefficients has not yet been
reported. The design may be extended for maximally flat
Butterworth characteristics by cascading another MOCCII
UBF stages.

II. Realization of universal biquadratic filter:

The presented universal biquadratic filter uses only two
MOCCIIs along with two grounded capacitors and two
grounded resistors as shown in Figure 2. The Multi Output
Current Conveyors (MOCCIIs) is shown in Figure 1. It is
characterized by

'. = ^.=^

y h, =+h

I = —I

(1)

where i= 1, 2, 3...

Analysis of the circuit yields the following current transfer
functions.

T,A^) =

Tbp(s) =

'LP

'IN

'BP

'IN

7?j 7?2 Ci C2
D{s)

(2)

(3)

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

(IJCSIS) InternationalJournal of Computer Science and Information Security,

Vol. 8, No. 3, 2010

i -0

^ z-

MOCCII ^
Z"

7*
X

, '^<

, '-1

. +

, '^'^

. +

- '-3

Figure 1. Symbol of MOCCII
1

^^+-

Tbe(s) =

7vj 7v2 Cj C2

where, the denominator is given by:
S 1

Z)(^) = ^' +

R,C,

■ + ■

7vj7v2CjC2

(4)

(5)

From eqn. (2), (3) and (4) it is seen that inverting low pass,
inverting band pass and non- inverting band elimination filters
are realized at three outputs. Non- inverting high pass filter is
realized just by connecting high impedance outputs Ilp and Ibe.
And all pass filter is realized by connecting the high impedance
outputs Ibp and Ire. The realized high pass and all pass filter
responses respectively are given by the following equations.

T^.(s) = ^^

- HP^

Tap(^) =

'UN

'AP

'UN

D(s)

+ -

7vj7v2CjC2

Dis)

(6)

(7)

The pole frequency cOo and the quality factor Q of the filters
are given by

0)^ =

7?j7?2CiC2

Q =

\ RA

R2C2

(8)

III. Realization of Butterworth higher order
filters:

Here we consider the realization of 6^ order Butterworth
low pass filter by cascading three sections of current mode
universal filter biquadratic filter (UBF) employing MOCCIIs
shown in Figure 2.

The normalized Butterworth transfer function for the
resulting 6th order CM low pass filter is given by [9]

Zi"'

2;'

MOCCII

z;-

Figure 2. Cascading stage (CM MOCCII UBF) for sixth order low pass
Butterworth filter

The transfer function can be de-normalized by replacing S -^
s/cOo to give the required sixth order filter fUnction at given
pole- COo and pole-Q. The pole-Q of an individual biquadratic
filter section is simply the reciprocal of the coefficients of s in
eqn. (9) [8]. The values of pole-Q are 0.518, 0.707, and 1.932,
respectively, for the three UBF to be cascaded. The UBF can
be designed using these values of Q and for a given pole
frequency. It is seen that no additional buffers are employed in
the realization. The filter's pole frequency cOo and the quality
factor Q of the current mode MO-CCII-based UBF are given
by:

(o„ =

R^RjC^C^

Q =

\ Rf^

R2C2

(10)

IV. Non Ideal Effects

Taking the non-idealities ai and Pi, i = 1 and 2, into
consideration, the current transfer functions of the UBF are
given then by:

T,As) = f^ = -

'IN

Tbp(s) =

' BP

'IN

D'is)

D'(s)

(11)

(12)

Tbe(s) =

'BE

'UN

^2 ^ c^ajMi

D\s)

(13)

T(.s) = — ^-

{s^ +l.932s + l){s' +lAl4s + l)(s' + 0.518s + 1)

The normalized pole frequency is at q)o=i.

(9)

'HP

Thp(s) = Iiin DXs)

(14)

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

(IJCSIS) InternationalJournal of Computer Science and Information Security,

Vol 8, No. 3, 2010

' AP

^2 sa,p, ^ a,a^p,p^

TApi^)J

D'{s)

(15)

and

D'{s) = s' +

2 sa^P^ a^a^PxPi

R,C,

■ + -

K^K2^\^2

(16)

where the pole frequency (q)o )and the quality factor (Q') of
the filters obtained from D'(s) are given by:

CO' =

xvjXV2CjC2

Q' =

(17)

At low to medium frequencies (f <\^ MHz), the circuit
continues to provide standard second order responses. The
pole- cOo is slightly lowered, but the pole-Q remains unaffected
by the non- idealities.

V. Sensitivity Study

The sensitivity of filter parameters are evaluated with
respect to active and passive elements and are given below.

SZ'

^1, '^2'Q^Q

SZ\

H^^l^Hl^Hl

Q —

S i?,,C,,a2,/?2 = 2

Q _ 1

i?2A,«i,A 2

(18)

From the above calculation it is evident that the sensitivities
of Oo and Q with respect to passive components are not more
than half in magnitude. This shows the attractive sensitivity
feature oftheUBF.

VI. Design and simulation

To demonstrate the performance of CM universal
biquadratic filter , the circuit is simulated using PSPICE level 3
parameters in 0.5|Lim CMOS process with supply voltages Vdd
= -Vss = 0.75V, using MOCCII+ model derived from CCII
[9]is shown in Figure 3. The dimensions of the MOS
transistors of CCII are listed in Table 1.

TABLE I.

MOSFET Dimensions of the CCII.

MOSFET

W(//m)

L(//m)

Ml, M2

0.5

M3, M9

M4, Me, My

0.5

Ms, Mio

IVCU
L I.

F p-g i ggg g ^[yM, t \gtl

2,*

z; 1:

i: i:

e i E I SE I E IS SH&EEI5

urn

Figure 3. CMOS Circuit for MOCCII

Initially the UBF was designed for /, = 1 MHz and Q =
0.707. For Ci = C2 = 1 1 pF, equation (8) yields Ri = 10 K^, and
R2 = 20 Kr2. The simulated LP, BP, BE response of current
mode UBF response is shown in Figure 4 shows good
agreement with the theory.

The frequency tuning of the BPF response at Q = 5 is next
investigated by changing the center frequency /> of the band
pass filter through resistor R2. The BP response curves
corresponding to/o = 300 KHz,/o = 500 KHz, and/o = IMHz
are given in Figure 5, which exhibit good agreement with the
theory.

30 KHz 100 KHz 300 KHz l.OMHz 3.0MHz IC

D ILP/IIN D IHP/IIN nIBP/IIN IBE/ TIN d IAP/ TIN

Frequency

Figure 4. The simulated UBF response at y ^ = 1 MHz

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

(IJCSIS) InternationalJournal of Computer Science and Information Security,

Vol 8, No. 3, 2010

loe

Gain

^^^>v,^^^

(DB)

^^'^^^,^^

-100

^^"-\

-200

ILP/IIN - —

ILP/IIN

Simulated

Theoretical

l.OMHz
Frequency

Figure 6. Frequency response of sixth order CM LPF in DB

30 KHz 100 KHz 300 KHz l.OMHz 3.0 MHz lOM

Frequency

Figure 5. Frequency tuning of BPF at Q = 5

Next we presents an example for the realization of 6^^ order
Butterworth low pass filter using low pass filter of UBF of
Figure 2. To evaluate the performance of the sixth order
Butterworth low pass filter, the circuit is designed for a pole
frequency /> =lMHz. The values of capacitors are selected
equal for convenient in IC implementation and are assumed to
be equal to 1 IpF. The resistors for each section are designed to
satisfy the equation (8). The designed values for each section
are given below.

Section-I : Ri
0.518

5.18 KQ, R2 = 19.32 KQ, for pole Q

Section-II : Ri

0.707

7.076 KQ, R2= 14.15 KQ, for pole Q

Section-Ill : Ri = 19.31 KKQ, R2= 5.18 KKQ, for pole Q
= 1.932

It is seen that the simulated pole frequency of 1.04 MHz, is
obtained from the simulation, which verifies the design. Figure
6 gives the stop band attenuation of 120 DB/decade, verifying
the 6*^ order low pass response. Through the entire range, the
simulated and theoretical responses overlap, showing close
agreement with theory.

The proposed circuit can also be used to realize other
higher order responses, such as, band pass, high pass and band
elimination filters, through a simple electronic switching
arrangement, for selecting the desired response of the UBF.
The result of the 6th order band pass filter is shown in Figure 7,
with a simulated pole frequency/, =1.02 MHz and a pole-Q of
1.84. The slopes below/) and above/ are each 60 DB/decade,
thus verifying the 6^ order band pass response. At the pole
frequency/ =1 MHz, the gain is equal to unity.

Figure?. Frequency response of sixth order CM BPF in DB

VII. Conclusion:

The MOCCII based current mode universal biquadratic
filter is used to realize sixth order Butterworth low pass filter
by cascading biquadratic filter sections of UBF of Fig. 2,
without using any additional current followers. With this
cascade approach the realization of higher order filter is
reduced to a much simpler realization of only second order
filters. The proposed circuit uses grounded capacitors hence is
suitable for IC implementation. The proposed circuit also has
low sensitivity, low component count, at low supply voltage of

± 0.75V.

References

[1] S. Ozoguz and C. Acar, "Universal current mode filter with reduced
number of active and passive elements", Electron. Lett., Vol.33, pp. 948-
949, 1997.

[2] A.M. Soliman,"Current Conveyor filters: classification and review".
Microelectronics. J., Vol.29, pp. 133-149, 1998.

[3] A. Toker, and S. Ozoguz, "Insensitive current mode universal filter
using dual output current conveyors,"Int. J. Electronics, Vol 87, No. 6
pp.667-674, 2000.

[4] H. Y. Wang and C. T. Lee, "Versatile insensitive current mode universal
filter implementation using current conveyors", IEEE Trans. Circuits
Sys.IL, Analog and Digital Signal Processing, Vol.48, No. 4, pp. 411-
413,2001.

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

(IJCSIS) InternationalJournal of Computer Science and Information Security,

Vol 8, No. 3, 2010

[5] E. Yuce, S. Minaei and O. Cicekoglu," Universal current mode active-C
filter employing minimum number of passive elements", Analog Integ.
Circ. Sig. Process., Vol.46, pp. 169-171, 2006.

[6] J. W. Horng, C. L. Hou, C. M. Chang, J. Y. Shie and C. H. Chang,"
Universal current filter with single input and three outputs using
MOCCIIs, Int. J. Electronics, Vol.94, No.4, pp.327-333, 2007.

[7] N. Herencsar and K. Vrba, "Tunable Current-Mode Multifunction Filter
Using Universal Current Conveyors", Third International Conference on
Systems, ICONS 08, pp. 1-6, 2008.

[8] R. Schaumann, M. E. Van Valkenburg, "Design of analog filters",
Oxford University press, 2003

[9] Rajput, S. S., Jamuar, S. S., "Advanced Applications of Current
Conveyors: A Tutorial", J. of Active and passive Electronic Devices,
Vol.2, pp.143-164, 2007.

[10] M. Kumar, M.C. Srivastava, U. Kumar, "Current Conveyor Based
Multifunction Filter", International Journal of Computer Science and
Information Security, Vol. 7, No. 2, pp.104-107, 2010.

[11] S.S. Rajput, and S.S. Jamuar, "A current mirror for low voltage, high
performance analog Circuits", Analog Integrated Circuits and Signal
Processing, Vol.36, pp 221-233, 2003.

AUTHORS PROFILE

Tahira Parveen received the B.Sc.
Engineering and M.Sc. Engineering degrees
from Z.H. College of Engineering &
Technology, A.M.U. Aligarh, India in 1984
and 1987 respectively. She has obtained her
Ph.D degree in 2009 from Electronics
Engineering Department, Z.H. College of Engineering &
Technology, A.M.U. , Aligarh, India. She is currently an
Associate professor in the department of Electronics
Engineering, Z.H.College of Engineering & Technology,
A.M.U. Aligarh, India. She has over 24 years of teaching and
research experience. Her research interests are Electronic
circuits and system design. Analog filters and Analog signal
processing. She has published over 24 research papers in
National and International Journals.

She has reviewed a book on "Basic Electronics" of McGraw-
Hill Publisher. She is a Fellow member of lETE (India). She
received Vijay Rattan Award for outstanding services
achievements & contributions. Her biography is published in
Marquis Who's Who in the World (USA) 2007. She is
Reviewer for the Journal: International Journal of Circuit
Theory and Applications, National and International
conferences. Her book entitled" A Textbook of Operational
Transconductance Amplifier & Analog Integrated Circuits" is
published by I.K. International publisher in 2009.

97 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

A Lightweight Secure Trust-based Locahzation
Scheme for Wireless Sensor Networks

P. Pandarinath

Associate Professor, CSE, Sir C R

Reddy College of Engineering

Eluru-534001, Andhra Pradesh

pandarinathphd @ gmail.com

sriramS 1 @ gmail.com

M. Shashi

Head of the Department, Department

of CS&SE, Andhra University,

Visakhapatnam- 530 003,

Andhra Pradesh
smogalla2000 @ yahoo .com

Allam Appa Rao
Vice Chancellor,
JNTU Kakinada,

Kakinada,

Andhra Pradesh

apparaoallam@gmail.com

Abstract — Location based security plays an important role in the
trustworthiness of wireless sensor networks and the results that
are obtained from them. Enforcement of location-aware security
policies requires trusted location information. As more of these
location-dependent services get deployed, the mechanisms that
provide location information will become the target of misuse and
attacks. In this paper, we propose to design a protocol that
validates the reliability of location information associated with
event reports. The protocol depends on the collaborative
interaction of the network nodes to find compromised nodes.
Nodes in the network record information of routing paths taken
by packets through the network. Upon receiving the route
request packets from the nodes, the sink checks whether their
route matches a historically expected behavior by packets from
the same claimed location. A trust value for the location claim of
this request is then created by the sink. The attached trust values
will be used to certify the truthfulness of the packets location
information. Since this scheme does not involve any complex
cryptographic operations, it has less overhead and delay. By
simulation results, we show that our proposed scheme attains
good delivery ratio with reduced delay and overhead.

Keywords- Localization; Wireless Sensor Networks; Trust;
Security issues; Lightweight Secure Trust-based Localization
(LSTL)

I. Introduction

A. Sensor networks and its applications

Wireless Sensor Networks (WSNs) are a specific kind of
ad hoc networks, highly decentralized, and without
infrastructure. They are building up by deploying multiple
micro transceivers, also called sensor nodes that allow end
users to gather and transmit environmental data from areas
which might be inaccessible or hostile to human beings. The
transmission of data is done independently by each node,
using a wireless medium. The energy of each node is limited
to the capacity of its battery. The consumption of energy for
both communication and information processing must be
minimized [1].

Wireless sensor networks are an area of great interest to
both academia and industry. They open the door to a large
number of military, industrial, scientific, civilian and
commercial applications. They allow cost-effective sensing

especially in applications where human observation or
traditional sensors would be undesirable, inefficient,
expensive, or dangerous [2].

B. Localization in Sensor Networks

In sensor networks, without earlier knowledge of their
location, nodes are organized into an unintentional
infrastructure dynamically. Localization or position estimation
problem refers to the problem of finding the positions of all
the nodes provided a few location aware nodes, relative
distance and angle information between the nodes. The
essential and vital problem in wireless sensor network
operation is to determine the physical positions of sensors for
following reasons.

• It is always necessary to have the position information
of sensors attached, in order to use the data collected
by the sensors. For instance, in sensor networks, the
physical location of each sensor should be known in
advance for identifying the position of the detected
objects in order to detect and track objects.

• With the knowledge of the geographic positions of
sensors, many communication protocols of sensor
networks are built. In majority of cases, there is no
supporting infrastructure available to locate the
sensors as they are deployed devoid of their position
information known in advance.

Hence it is necessary to find the position of each sensor in
wireless sensor networks after deployment [3].

C. Phases of Localization

The localization phase is a very critical step that must be
secured in order to ensure the integrity of the WSN and its
associated services.

• Firstly, this process allows the sensor to set up the
necessary parameters to establish the paths that will
lead their data towards end users.

• The knowledge of their position is also an essential
prerequisite for the final application that processes the
data collected by sensors, i.e., the user needs to know
the origin of collected data before using it.

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

• Finally, the end users might want to query some nodes
by sending the position where information needs to be
collected [1].

Many localization schemes have been proposed for sensor
networks in recent years without depending on expensive GPS
devices. The majority of existing schemes assume some
unique nodes, called beacon nodes, which have the potential to
know their own locations through either GPS receivers or
manual configuration. The remaining nodes or the non-beacon
sensor nodes can be equipped with somewhat cheap measuring
devices for directionality, signal strength, or time of arrival,
etc. The non-beacon nodes find their own locations using these
measurements and the locations of three or more beacon
nodes. This method is known as the beacon-based technique,
which involves two stages in location discovery.

i. A sensor node calculates its distances to each neighbor
of the node using the received signal information.

ii. With all these distance calculations, sensor nodes
calculate the actual location of the node. [4]

D. Security Issues on Localization

An important concern for various applications of WSNs is
the ability to validate the integrity of the sensor network as
well as the retrieved data. Various types of security attacks
include

• The injection of false information into the regular data
stream,

• The alteration of routing paths due to malicious nodes
advertising false positions (sink holes and worm
holes), and

• The forging of multiple identities by the same
malicious node.

Thus, location based security plays an important role in the
trustworthiness of WSNs and the results that are obtained from
them [5].

Enforcement of location-aware security policies requires
trusted location information. As more of these location-
dependent services get deployed, the mechanisms that provide
location information will become the target of misuse and
attacks. In particular, the location infrastructure will be
affected by many localization- specific threats that cannot be
tackled through traditional security services. Therefore, as we
move forward with deploying wireless systems that support
location services, it is sensible to integrate appropriate
mechanisms that protect localization techniques from these
new forms of attack [6] .

The wormhole attack is a typical kind of secure attacks in
WSNs. It is launched by two colluding external attackers
which do not authenticate themselves as legitimate network
nodes to other network nodes. One of the wormhole attackers
overhears packets at one point in the network, tunnels them
through the wormhole link to another point in the network,
and the other wormhole attacker broadcasts the packets among
its neighborhood nodes. This may cause a severe impact on
the routing and localization procedures in WSNs [7].

E. Problems Identified and Proposed Scheme

The proposed solutions to mitigate some of these
localization attacks always involve traditional security
techniques. But, it is unlikely that traditional security will be
able to remove all threats to wireless localization. We
therefore consider that instead of providing solutions for each
attack, it is essential to achieve robustness to unpredicted and
non-filterable attacks. Particularly, localization must function
properly even in the presence of these attacks [6].

Digital signatures can prevent bogus seeds from injecting
bogus location messages by authenticating seeds'
transmissions to nodes. This could be done by distributing
public keys corresponding to the seeds' private keys to each
node before deployment. But public key encryption operations
are often too computationally expensive for sensor nodes,
however, and the long messages required drain power
resources [8].

Another approach would be to use the mTesla protocol, by
preloading each node with the initial hash chain value and
each seed with the initial secret. This would save the expense
of public key operations, but would delay localization until the
next key in the hash chain is released. It would also require
loose synchronization among the seeds [8].

Moreover, since distance measurements are susceptible to
distance enlargement/reduction, such techniques may not be
used to infer the sensor location [9].

In this paper, we propose to design a protocol that
validates the reliability of location information associated with
event reports. The protocol depends on the collaborative
interaction of the network nodes to find compromised nodes.
Each active node automatically has some knowledge of the
activity within the network which can be used in determining
the anomalous behavior. Nodes in the network record
information of routing paths taken by packets through the
network. Upon receiving a packet, nodes check whether their
route matches a historically expected behavior by packets
from the same claimed location. A trust value for the location
claim of this packet is then created and propagated to the sink.
The attached trust values will be used by the sink to certify the
truthfulness of the packets location information.

II. Related Work

J. G. Alfaro, M. Barbeau and E. Kranakis [1] have
provided three algorithms that enable the sensor nodes of a
Wireless Sensor Network to determine their location in
presence of neighbor sensors that may lie about their position.
Their algorithms minimize the number of trusted nodes
required by regular nodes to complete their process of
localization. Also their algorithms always work for a given
number of neighbors provided that the number of liars is
below a certain threshold value, which is also determined. The
three algorithms that they have presented guaranteed that
regular nodes in the WSN always obtain their position
provided that the number of liars in the neighborhood of each
regular node is below a certain threshold value, which they
determine for each algorithm. Their three algorithms allow the
regular nodes to identify and isolate nodes that are providing
false information about their position. Moreover, their

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

algorithms minimize the necessary number of trusted nodes
required by regular sensors to complete their process of
localization. They also guarantee a small exchange of data
between nodes, minimizing in this manner the impact that the
localization process has in terms of energy and battery life of
sensors.

Kaiqi Xiong and David Thuente [4] have proposed novel
schemes for secure dynamic localization in sensor networks.
Their proposed schemes can tolerate up to 50% of beacon
nodes being malicious, and they have linear computation time
with respect to the number of reference nodes. They also
showed that their schemes are applicable and resilient to
attacks from adversaries. They proposed several methods for
secure location discovery in sensor networks which are
developed through the technique of beacon-based localization.

Honglong Chen et al [7] have investigated the impact of
the wormhole attack on the localization and they proposed a
novel consistency-based secure localization scheme against
wormhole attacks, which includes wormhole attack detection,
valid locator's identification and self-localization. They
developed an enhanced identification approach which obtains
better performance than the basic identification approach.

Yanchao Zhang et al [10] have first analyzed the security
of existing localization techniques. They then developed a
mobility-assisted secure localization scheme for UWB sensor
networks. They didn't intend to provide brand-new
localization techniques for UWB sensor networks. Instead,
they focused on analyzing and enhancing the security of
existing approaches when applied in adversarial settings. In
addition, they have proposed a location-based scheme to
enable secure authentication in UWB sensor networks.

Srdjan C apkun et al [11] have proposed a secure
localization scheme for sensor networks based on the received
signal strength (RSS) ranging techniques. Their scheme
enables the network authority to obtain locations of sensor
nodes in the presence of an attacker. Also their proposed
scheme uses a small number of anchor nodes with known
locations that provide points of reference from which the
sensors locations are computed. Their scheme also makes use
of robust localization and time synchronization primitives
which, appropriately combined, enable the detection of attacks
on localization, within a realistic attacker model.

III. Proposed Lightweight Secure Trust-based
Localization (LSTL) Scheme

A. System Design and Overview

Sensors will start generating event reports corresponding
to their respective location, once the network becomes
activated. A malicious sensor might then try to generate
illegitimate event reports for locations other than its own. This
can be detected when the nodes along the path from the source
to destination, attaches trust values to passing data packets,
ensuring the correctness probability of the declared source
locations. These trust values evaluated based on gathered past
traffic patterns combined with the claimed source location.
When receiving packets, if any of the routing information
diverges from projected traffic patterns, then the nodes have
the chance to decrease the trust values associated with these

packets. These decreased trust values replicate the appearance
of an attacker in the routing pattern.

We assume that the network at the start is free from
attackers for a short period of time. The novelty in this scheme
lies in the efficient way of shortening the history of traffic
pattern and capability of using this history to authenticate the
accuracy of future packets.

As a part of the routing protocol, each sensor node will
maintain a history and normalized count of each previously
seen source destination pair for routed packets. New incoming
packets from rarely seen sources will then be considered more
suspicious and associated with a low trust value.

B. Distance Estimation

We first describe the meaningful way of comparing packet
routes efficiently. Based on the sequence of nodes a packet has
visited, we define a distance metric to measure the distance
between the two paths. The distance is designed such that fake
claimed locations for packets will result in large distances
between real and expected paths. There are many generic ways
to measure the distance between two curves in space.

Given a path R , we take k samples {/?i , /?2 ' * * * ' ^y^ } on R.

we define the distance between two paths /? , 7? as the sum of
squared distance between corresponding sample points.

n^^^)

Ri - Ri

(1)

C. Trust Based Incentive Scheme

An additional data structure called Neighbor's Trust
Counter Table (NTT) is maintained by each network node. Let
{rCl,rC2,---} be the initial trust counters of the nodes

{Ni , N2 '"} along the route from a source S to the sink D.

Since the node does not have any information about the
reliability of its neighbors in the beginning, nodes can neither
be fully trusted nor be fully distrusted. When a source S wants
to send data to the sink D, it sends route request (RREQ)
packets. It contains the source and destination ids and location
of the source and a MAC computed over the accumulated path
with a key shared by the sender and the destination

Each time, the sink estimates the distance between the two

paths R and R towards the sink, using (1). If it is more than
a maximum threshold Thl, then the trust counter value is
decreased by a penalty of 5.

(ie) TCi=TCi-S,

(2)

Then the NTT of node Ni is modified with the values of
TCi When the subsequent RREQ message reaches the

destination, it checks the trust values of the intermediate
nodes. The nodes are considered as well behaving nodes if the
trust values are equal or greater than a trust threshold TC^/^ .
On the other hand, the nodes are considered as misbehaving

100

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

nodes if the trust values are less than TQ/^ . Also nodes with
trust values less than TQ/^ are prohibited from further
transmissions.

D. Route Discovery Process

In the proposed protocol, once a node S want to send a
packet to a sink D, it initiates the route discovery process by
constructing a route request RREQ packet. It contains the
source and destination ids and location of source and a MAC
computed over the accumulated path with a key shared by
each node.

When an intermediate node receives two RREQ packets
from the same claimed locations, it retrieves the corresponding
path information from their respective MAC and appends its
id to the path and recreates the MAC with a key which is
shared with the destination. It then forwards the RREQ to its
neighbors.

Then the route request process is illustrated as below:

> D

Figure 1. Route Request Process
From the figure 1 , we can see that there are two paths

R: S ^ D and

R\S > D

for the sink D.

For the path /? , let A/^^ , A/^2 ' * " ' ^4 t)e the intermediate
nodes between S and D. Then the route request process of path
R is given by

RREQ: [S,D, MAC(S)]

RREQ: [S, D, MAC {S, Ni), MAC(]
Ni ^ N2

RREQ: [S, D, MAC (S, N, N2)]

N2 > N,

RREQ: [S, D, MAC (S, Ni N2 Ni)]

N3 ^ N4

For the path R , let N\, N2,---N/\. be the intermediate
nodes between S and D. Then the route request process of
path R is given by

RREQ: [5, D, MAC (5)]

5 > Ml

RREQ: [5,D,MAC(S,iVi)]

RREQ: ['S, D, MAC (S^N,. Ni}]
77: ^ Nb

RREQ [S^D^MACiS.Ni^Ni^N:.)]

N,.

RREQ:[5,D,MAC(5,A^i,A^2,A^3,iV4)]

When RREQs of both R and R reaches the sink, from the
received MAC values, it calculates

V = D(Ni,Ni) where

D(Ni,Ni) = ^\\Ni-Ni 11^ by

(1)

i=l

Then it checks the value of V , based on which the trust
values are increemented or decreemented for the
corresponding nodes.

IfV<thi then

CCNi = CCNi + J,
Else

CCNi = CC^i - ^'
End if

where th^ is the minimum threshold value for V and S is
the scale factor for increement or decreement.

The process is repeated for various time intervals and
finally the value of credit counter is checked.

If CCNi > th2 then

RREP is sent
Else

The source is considered malicious,

RREQ is discarded
End if.

RREQ: [S, D, MAC (S, Ni. N3.N3 N4)]

N4 ^ — ^ — -^ D

Where th2 is the minimum threshold value for CCN

In our scheme, only nodes which are stored in the current
route need to perform these cryptographic computations. So
the proposed protocol is efficient and more secure.

101

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

IV. Performance Evaluation

A. Simulation Parameters

We evaluate our Light weight Secured Trust based
Localization (LSTL) Algorithm through NS2 simulation. We
use a bounded region of 1000 x 1000 sqm, in which we place
nodes using a uniform distribution. We assign the power levels
of the nodes such that the transmission range and the sensing
range of the nodes are all 250 meters. In our simulation, the
channel capacity of mobile hosts is set to the same value: 2
Mbps. We use the distributed coordination function (DCF) of
IEEE 802.11 for wireless LANs as the MAC layer protocol.
The simulated traffic is Constant Bit Rate (CBR).

The following table summarizes the simulation parameters
used

TABLE I. Simulation Parameters

No. of Nodes

25,50,75 and 100

Area Size

1000 X 1000

Mac

802.11

Simulation Time

50 sec

Traffic Source

CBR

Packet Size

512

Transmission Range

250

Routing Protocol

AODV

Speed

Mobility model

Random way point

Attackers

5, 10, 15, 20 and 25

B. Performance Metrics

We compare the performance of our proposed LSTL with
the SeRLoc [11]. We evaluate mainly the performance
according to the following metrics:

Average end-to-end delay: The end-to-end-delay is
averaged over all surviving data packets from the sources to
the destinations.

Control overhead: The control overhead is defined as the
total number of control packets exchanged.

Estimation Error: It is the estimation error, which
indicates how close the estimated location is to the actual
location.

Average Packet Delivery Ratio: It is the ratio of the
number .of packets received successfully and the total number
of packets transmitted.

A. Based On Nodes

In order to test the scalability, the number of nodes is varied as
25, 50, 75 and 100.

Nodes Vs Delay

-LSTL
-SeRLoc

Figure2. Nodes Vs Delay

Nodes Vs Overhead

50 75 100

Nodes

Figures. Nodes Vs Overhead

Nodes Vs Error

Figure4. Nodes Vs Error

Figure 2 show the end-to-end delay occurred for both
LSTL and SeRLoc. As we can see from the figure, the delay is
less for LSTL, when compared to SeRLoc.

Figure 3 shows the overhead for both LSTL and SeRLoc.
As we can see from the figure, the overhead is less for LSTL,
when compared to SeRLoc.

Figure 4 shows the error occurred for both LSTL and
SeRLoc. As we can see from the figure, the error is less for
LSTL, when compared to SeRLoc.

B. Based On Attackers

The number of attacker nodes is varied as 5, 10, 15, 20 and 25
in a 100 nodes scenario.

102

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

Attackers Vs Delay

0.6.

n R

A.^_

"Tn A

•"^^l^

^ n "^ -

^s?* -—A.

Q) rj p

■ *^^

□ U.^

n 1

10 15 20 25

Attackers

- SeRLoc
-LSTL

Figures. Attackers Vs Delay

Attackers Vs Overhead

- SeRLoc
-LSTL

10 15 20
Attackers

Figure6. Attackers Vs Overhead

Attackers Vs Error

10 15 20

Attackers

Figure?. Attackers Vs Error

Attackers Vs Delivery Ratio

-LSTL
- SeRLoc

10 15 20
Attackers

Figure 8. Attackers Vs DelRatio

Figure 5 show the end-to-end delay occurred for both
LSTL and SeRLoc. As we can see from the figure, the delay is
less for LSTL, when compared to SeRLoc.

Figure 6 shows the overhead for both LSTL and SeRLoc.
As we can see from the figure, the overhead is less for LSTL,
when compared to SeRLoc.

Figure 7 shows the error occurred for both LSTL and
SeRLoc. As we can see from the figure, the error is less for
LSTL, when compared to SeRLoc.

Figure 8 shows the delivery ratio for both LSTL and
SeRLoc. As we can see from the figure, the delivery ratio is
high for LSTL, when compared to SeRLoc.

V. Conclusion

In this paper, we have designed a protocol that validates
the reliability of location information associated with event
reports. The protocol depends on the collaborative interaction
of the network nodes to find compromised nodes. When a
source S wants to send data to the sink D, it sends route
request (RREQ) packets. It contains the source and destination
ids and location of the source and a MAC computed over the
accumulated path with a key shared by the sender and the
destination. When the RREQ packets reach the sink, it

estimates the distance between the two paths R and R
towards the sink. If it is more than a maximum threshold Thl,
then the trust counter value is decreased by a penalty of 5.
When the subsequent RREQ messages reach the destination, it
checks the trust values of the intermediate nodes. The nodes
are considered as well behaving nodes if the trust values are
equal or greater than a trust threshold TC^^ . On the other
hand, the nodes are considered as misbehaving nodes if the
trust values are less than TC^/^ . Also nodes with trust values

less than TC^/^ are prohibited from further transmissions.

Since this scheme does not involve any complex cryptographic
operations, it has less overhead and delay. By simulation
results, we have shown that our proposed scheme attains good
delivery ratio with reduced delay and overhead.

References

[1] J. G. Alfaro, M. Barbeau and E. Kranakis, "Secure Localization of
Nodes in Wireless Sensor Networks with Limited Number of Truth
Tellers", Proceedings of the Seventh Annual Communication Networks
and Services Research Conference, 2009.

[2] Eric Sabbah et al, "An Application Driven Perspective on Wireless
Sensor Network Security", Proceedings of the 2nd ACM international
workshop on Quality of service & security for wireless and mobile
networks, 2006.

[3] V. Vijayalakshmi et al, "Secure Locahzation Using Elliptic Curve
Cryptography in Wireless Sensor Networks", UCSNS International
Journal of Computer Science and Network Security, VOL.8 No. 6, June
2008.

[4] Kaiqi Xiong and David Thuente, "Dynamic Localization Schemes in
Malicious Sensor Networks", Journal of Networks, Vol. 4, no. 8,
October 2009.

[5] E. Ekici et al, "Secure probabihstic location verification in randomly
deployed wireless sensor networks". Ad Hoc Networks, Volume 6, Issue
2, April 2008.

[6] Zang Li et al, "Robust Statistical Methods for Securing Wireless
Localization in Sensor Networks", Proceedings of the 4th international
symposium on Information processing in sensor networks, 2005.

[7] Honglong Chen et al, "A Secure Locahzation Approach against
Wormhole Attacks Using Distance Consistency", EURASIP Journal on
Wireless Communications and Networking, Volume 2010 (2010).

103

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

[8] Lingxuan Hu and David Evans, "Localization for Mobile Sensor
Networks", Proceedings of the 10th annual international conference on
Mobile computing and networking, 2004.

[9] Loukas Lazos and Radha Poovendran, "SeRLoc: Robust Locahzation
for Wireless Sensor Networks", ACM Transactions on Sensor Networks
(TOSN),2005.

[10] Yanchao Zhang et al, "Secure Locahzation and Authentication in Ultra-
Wideband Sensor Networks", IEEE Journal on Selected areas in
communications, 24(4), 829-835, 2006.

[11] Srdjan C apkun et al, "Secure RSS-based Locahzation in Sensor
Networks", Technical Reports 529, ETH Zurich, 09, 2006.

a P. Pandarinath his B.Tech in CSE in 1994 and M.Tech in

CSE in 2001. He is presently pursuing his doctorate. He
has a total experience of 15 years. He is working as a
Associate Professor in Sir C R Reddy college of
Engineering. He has completed 80 projects and guided 100
projects. His areas of interests include Network Security
and Cryptography, Advanced Computer Architecture,
DataBase Management System, Computer Organization,
Computer Networks and Bio-informatics.

Dr. M. Shashi completed her B.E. in EEE, and M. E. in
Computers Engineering. She received Doctorate in
Engineering. She is working as Head of the Department in
Computer Science and Systems Engineering, Andhra
University, Visakhapatnam. She has a total experience of
24 years. She presented 17 international journals and 3
national journals. She presented papers in 10 international
conference and 8 national conferences. She attended 7 international and 25
national conferences. She guided 85 M.Tech projects. She received AICTE
Career Award for Young Teachers in 1996-1998 at National Level and
Best Ph.D. Thesis Prize for 1994 & 1995 at University Level. Her areas of
specializations include Data Warehousing & Mining, Al, Data Structures, Soft
computing and Machine Learning.

Dr. AUam Appa Rao received his B.Sc. (M.P.C.) in 1967
from Andhra University. He completed M.A. in
Economics: Mathematical Economics and Econometrics,
from Andhra University. He received the Ph.D. in
Computer Engineering from Andhra University, in 1984.
He is working as Vice Chancellor for JNTU Kakinada. He
has presented 150 research articles. Dr AUam holds two
patents as a co inventor for Method(s) of stabilizing and potentiating the
actions and administration of brain-derived neurotrophic factor (BDNF) patent
Number 20080234197 dated 25th September 2008 ( Refer
http://www.faqs.org/patents/inv/83318) and Method(s) of preventing,
arresting, reversing and treatment of atherosclerosis Patent Number
20080279925 dated 13th November2008(Refer

http://www.faqs.org/patents/inv/171637). He achieved best Researcher in
Engineering in recognition of commendable record of research in Engineering,
Andhra University, Visakhapatnam, hidia, 2003.

104

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

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

Vol 008, No. 003, 2010

Mechanism to Prevent Disadvantageous Child Node

Attachment in HiLOW

Lingeswari V Chandra, Kok-Soon Chai and

Sureswaran Ramadass

National Advanced IPv6 Centre, Universiti Sains Malaysia

{ lingeswari , kschai, sures }@nav6.org

Gopinath Rao Sinniah

MIMOS Berhad, 57000 Kuala Lumpur

gopinath. rao @ mimos.my

Abstract — Vast research is being conducted in the area of
Wireless Sensor Network in recent years due to it foreseen
potential in solving problems covering many aspects of daily,
industrial and ecological areas. One of the biggest challenges in
the power and memory constrained sensor network is in
establishing reliable network and communication among the
nodes. . IP-based 6L0WPAN was introduced to give a new
dimension to sensor network by enabling IPv6 to be applied to
the wired as well as wireless sensors. An extendable and scalable
Hierarchical Routing Protocol for 6L0WPAN (HiLOW) is one of
three routing protocols which has been introduced specially for
6L0WPAN. HiLOW was designed by exploiting the dynamic 16
bit short addresses assignment capabilities featured by
6L0WPAN. HiLOW clearly defines the network setup process,
address allocation method and routing mechanism. However
there are shortcomings or issues pertaining HiLOW that make it
less efficient. One of the major issues identified in HiLOW is in
the process of selecting the parent node to attach with during the
network tree setup. Disadvantageous parent selection could lead
to significant shorter life span of the network which affects the
reliability and stability of the network. In this paper we review
the HiLOW routing protocol, highlight the issues revolving
HiLOW and suggest a mechanism to prevent disadvantageous
child node attachment in HiLOW. The proposed mechanism
takes into consideration the LQI value, the potential parents'
depth in the network and the average energy level of the parent
in selecting the suitable parent node in order to provide a more
reliable wireless sensor network.

Keywords- 6L0WPAN, routing protocol, HiLOW, WSN,
Hierarchichal routing protocol

Introduction

Wireless Sensor Network (WSN) is an area which is being
vastly researched on in recent years due to its foreseen
capability in solving many existing problems, ranging from day
to day problems, industrial problems and up to ecological
problems. WSN initially started as a military network where it
was used to detect enemies, land mines and identifying own
man. Now WSN usage has been extended to general
engineering, agriculture monitoring, environmental monitoring,
health monitoring and also home and office monitoring and
automation.

The rapid growth and penetration of sensors to other areas
are due to engineering contribution where more types of
sensors has being developed and introduced while decreasing
the size of sensor nodes, and power consumption and price of

microprocessors while increasing the memory size of the
nodes. Even though vast improvement has been witnessed from
engineering perspective the nodes until today still faces the
limitation in power and computational capacities and memory
[1].

One of the most important elements after sensing activity
but most energy costly element of the WSN is the
communication part. Radio communication is typically the
most energy consuming activities [2] and the reception energy
is often as high as the transmission energy thus the network
protocol introduced as well as the routing protocol needs to
take into consideration the energy usage in setting up the
network as well as complexity of computation during routing.
A disadvantageous setup of network could lead to request of
retransmission to occur and this would lead towards wastage of
precious energy.

Many communication protocols have been introduced to
WSN prior to the introduction of 6L0WPAN namely 802.15.1
Bluetooth [3], WirelessHart [4], ZWave [5], ZigBee [5] and
others. Compared to the named communication network
6LoWPAN[6] was the first to introduce IPv6 to be applied to
not only wireless but also wired sensor network. 6L0WPAN
defines the network layer and also the transport layer and is
able to be deployed on any sensors which are IEEE 802. 15. 4 [7,
8] compliant. The 6L0WPAN stack is 30KB minimum in size
which is smaller compared to the named protocols. The routing
protocol for 6L0WPAN is an open area where it is open to
introduction of new protocols. Till today there are three
prominent routing protocols which have been introduced
specifically for 6L0WPAN namely Hierarchical Routing
Protocol (HiLOW) [10, 11], Dynamic MANET On-demand for
6L0WPAN (DYMO Low) [12] and 6L0WPAN Ad Hoc On-
Demand Distance Vector Routing (LOAD) [13].

The remainder of this paper is organized as follows. Section
2 reviews the HiLOW protocol in detail and the issues in it and
works done to improve HiLOW. Section 3 suggests a
mechanism to avoid a disadvantageous parent child attachment
during the routing tree set up. Section 4 presents the
conclusion.

II. Related Works

A hierarchical routing protocol (HiLow) for 6L0WPAN
was introduced by K. Kim in 2007 [10]. HiLOW is a routing
protocol which exploits the dynamic 16-bits short address

105

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

assignment capabilities of 6L0WPAN. An assumption that the
multi-hop routing occurs in the adaptation layer by using the
6L0WPAN Message Format is made in HiLOW. In the rest of
this section we will discuss the operations in HiLOW ranging
from the routing tree setup operation up to the route
maintenance operation while highlighting issues revolving
HiLOW. Other works done to solve some of the issues in
HiLOW is also reviewed here.

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

Vol 008, No. 003, 2010
has same depth, same energy level and has different number of
existing child. Their mechanism also does not take into
consideration the quality of the link established between the
parent node and child node. Therefore the suggested
mechanism does not solve the arising issue completely. So in
this paper we will be suggesting a mechanism to address this
first issue by taking into consideration the link quality, the
existing energy of the potential parent as well as the depth of
the parent.

HiLOW Routing Tree Setup, Issues and Works Done

The process of setting up the routing tree in HiLOW
consists of a sequence of activities. The process is initiated by a
node which tries to locate an existing 6L0WPAN network to
join into. The new node will either use active or passive
scanning technique to identify the existing 6L0WPAN network
in its Personal Operation Space (POS).

If the new node identifies an existing 6L0WPAN it will
then find a parent which takes it in as a child node and obtain a
16 bit short address from the parent. The parent will assign a 16
bit short address to a child by following the formula as in (1).
An important element of HiLOW is that the Maximum
Allowed Child (MC) need to be fixed for every network and all
the nodes in the network is only able to accept child limited to
the set MC. In the case where no 6L0WPAN network is
discovered by the node then the node will initiate a new
6L0WPAN by becoming the coordinator and assign the short
address by 0.

Future Child Node's Address

MC : Maximum Allowed Child Node

N : Number of child node inclusive of the new node.

AP : Address of the Parent Node

Second issue revolves around the MC value which is being
fixed for all nodes. The current scenario works well if all the
nodes have the same power conservation method; meaning if
all the nodes in the network are either battery powered or
mains-powered. In the case where some nodes are battery
powered and the others are mains-powered, then this method is
not advantageous. The main-powered nodes could support
more nodes as their child node as they are affluent in energy.
This is an open issue to be addressed in HiLOW, for present
time the assumption that all nodes having same energy
conservation have to be made. The activity of disseminating the
MC value to joining nodes is also left in gray. This issue is not
addressed in this paper.

B. Routing Operation in HiLOW

Sensor nodes in 6L0WPAN can distinguish each other and
exchange packet after being assigned the 16 bits short address.
HiLOW assumes that all the nodes know its own depth of the
routing tree. The receiving intermediate nodes can identify the
parent's node address through the defined formula (2). The '[]'
symbol represents floor operation

AC : Address of Current Node
MC : Maximum Allowed Child

FC = MC * AP -F N ( < N <= MC )

(1)

AP=[(AC-1)/MC]

(2)

Two potential issues have been identified in this process.
First issue would be when the child node gets respond from
more than one potential parent. There is no clear mechanism
rolled out in selecting the suitable parent to attach with. If the
new node chooses to join the first responding parent node, it
could be bias to the parent as some parent might be burdened
with more parents meanwhile other parents which is in the
same level has less child or none at all. Selecting the parent
based on first responded potential parent could also lead to fast
depletion of energy to certain parent causing the life span of the
network to be shorter and the stability to be jeopardized.
Selection of parent without considering the link quality could
cause towards high retransmission rate which will consume
energy from the child node as well as parent node.

In [15] a mechanism to overcome the issue was suggested.
Their mechanism suggests the potential parent node to provide
the new child with its existing child node count
(child_number). By issuing the child_number the node could
select suitable parent which has less child nodes. The suggested
mechanism performs well only when the potential parent node

The receiving intermediate nodes can also identify whether
it is either an ascendant node or a descendant node of the
destination by using the above formula. When the node
receives a packet, the next hop node to forward the packet will
be calculated by the following three cases (3) which is defined
in [10]. No issues have been identified so far in this process.

S A : Set of Ascendant nodes of the destination node

SD : Set of Descendant nodes of the destination node

AA(D,k): The address of the ascendant node of depth D of the
node k

DC : The depth of current node

C : The current node

Case 1 : C is the member of S A

(3)

106

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

The next hop node is A A (DC+1, D)

Case 2: C is the member of SD

The next hop node is AA (DC-1, C)

Case 3 : Otherwise

The next hop node is AA (DC-1, C)

C. Route Maintenance in HiLOW

Each node in HiLOW maintains a neighbor table which
contains the information of the parent and children node.
When a node loses an association with its parent, it should to
re-associate with its previous parent by utilizing the
information in its neighbor table. In the case of the association
with the parent node cannot be recovered due to situation such
as parent nodes battery drained, nodes mobihty, malfunction
and so on, the node should try to associate with new parent in
its POS [11]. Meanwhile if the current node realizes that the
next-hop node regardless whether its child or parent node is not
accessible for some reason, the node shall try to recover the
path or to report this forwarding error to the source of the
packet.

Even though a route maintenance mechanism has been
defined in HiLOW, the mechanism is seen as not sufficient to
maintain the routing tree. An Extended Hierarchical Routing
Over 6L0WPAN which extends HiLOW was presented by in
[16] in order to have better maintained routing tree. They
suggested two additional fields to be added to the existing
routing table of HiLOW namely, Neighbour_Replace_Parent
(NRP) and Neighbour_Added_Child (NAC). This NRP doesn't
point to the current parent node but to another node which can
be its parent if association to current parent fails. Meanwhile
NAC refers to the newly added child node. More work need to
be done on this mechanism on how many nodes allowed to be
adapted by a parent node in addition to the defined MC and
whether this mechanism will have any impact on the routing
operation, however this topic is beyond the scope of this paper.

in. DiSADVANTEGOUS CHILD NODE ATTACHMENT
AVOE)ANCE MECHANISM

A disadvantageous child parent attachment avoidance
mechanism for HiLOW is being suggested in this paper. The
suggested mechanism is able to overcome the bias child node
phenomena that could shorten the life span of the network as
well as affect the reliability and the stability of the network. We
are suggesting a mechanism where the new child node is
provided with three data; one is the Link Quality Indicator
(LQI) value, secondly the depth of the potential parent node
and thirdly the average amount of energy the potential parent
node has. The suggested mechanism is an enhancement work
of the mechanism suggested in [17].

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

Vol 008, No. 003, 2010
LQI value can be measured by either parent node or by
child node itself. The prior measurement by potential parent
node was more preferred compared to the latter which is
measurement by child. The measurement by child was not
selected as the child node needs to use energy to measure LQI
for every potential parent node and this would make the total
energy usage to be higher compared to energy used when
parent node measures the LQI only for one time.

Therefore in this mechanism the LQI value will be
measured by the potential parent node and the value is provided
to the new node which is looking for a parent node; this is in
contrast to the mechanism suggested in [18]. LQI is selected
compared to Received Signal Strength Indication (RSSI) as
LQI is more accurate to measure the quality of the link and the
delivery ratio especially when obstructions or noise exist [18].
In previous mechanisms quality of the link is not considered in
selecting the parent node. The quality of the link is important in
making association as bad links would cause retransmission of
data to occur and this causes nodes to use more energy.

The node which is able to accept the new node as child
node will calculate the average amount of energy it has
according to the mathematical equation in (4) which is defined
earlier in [17]. The average amount of energy represents the
energy the potential parent node can equally use for itself as
well as use to forward the data of existing child nodes and
potential child. So the value 2 in the equation represents itself
and the potential new child node.

Avg : Average Amount of Energy

CBP : Current Energy Level of the Potential Parent

EC : Existing Child Node

Avg= CBP / (EC-F2)

(4)

In situation where there is more than one potential parent
the child node will then make decision on which potential
parent node to associate with according to steps as displayed in
Fig. 1. First the child node will calculate the average LQI based
on the equation (5). The '[]' symbol represents floor operation
The average is calculated by summing up all the LQI value
received, then dividing it by the number of parent node which
responded and lastly flooring the value . In [18] a threshold was
to be set and the LQFs will be compared to this threshold, this
method needs intervention from human in determining the
threshold and a method to communicate the threshold to all
nodes; the communication process will again consume energy,
due to this factor the new method is introduced.

ALQI : Average LQI of all the potential links
TLQI : Total LQI of all the potential links
Count : Count of potential parent node

ALQI= [TLQI / Count]

(5)

107

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

>1 pp

Average out all
the PPs' LQI

PPs above or equal
to the average LQI

PPs having lowest
depth

, Select the PP with\
Only 1 parent ^( ^^e highest

average energy

with the highest
average energy

PPs Having Highest
Average Energy

with highest
LQI

/Select PP with the \
highest LQI

PPs Having Highest
Level of LQI

' Select the first PP which ^
y responded ,

Legend :

PP - Potential Parent Node

PPs - Potential Parent Nodes

Figure 1 . Parent Node Selection Mechanism

The child node then compares all the potential parents' LQI
with the average LQI calculated. In the case only one parent
node's LQI is higher than average LQI then it w^ould be
selected as parent node. If there is more than one parent above
the average LQI, then the child node will compare the depth of
all the potential parent nodes which is above the average. If
there is only parent node with the lowest depth, then that
particular parent node is selected and process of associating is
started. In the case where there is more than one potential
parent with the lowest level of depth, then the average energy
of these nodes are compared. The child selects the parent with
the highest average energy and associate with it if there is only
one particular potential parent with highest energy. In a
situation where more than one parent shares the highest level of

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

Vol 008, No. 003, 2010
energy and lowest level of depth; the child node will then
compare the LQI value of these nodes. The node with highest
LQI value will be selected as parent node to be associated with
if only one node qualifies. If more than one parent has the
lowest level of depth, highest level of energy and highest level
of LQI then the child node will try to establish association with
the first responded potential parent in this category. In [12] the
LQI value is not considered when more than one parent shares
the highest level of energy and lowest level of depth, the child
straight away selects the first responded parents which is not
advantageous if the first responded parent has lower LQI
compared to the other potential parent.

The first scenario will be described with the assistance of
Fig. 2. When three potential parent node(8), node(4) and
node(3) responds; according to the previous mechanism the
child node should attach to parent node(3) as it has no child
node compared to node(4) which has 2 child node and node(8)
which has 1 child node. According to our mechanism the child
will not consider the node(3) for attachment as the link quality
is below the average, this represent the quality is bad compared
to node(4) and node(8) and this could lead to high
retransmission rate compared to node(4) and node(8) link. If an
assumption that the node(3) has also two child nodes is made,
then according to the previous mechanism the node(8) will be
selected as suitable parent and this is also disadvantageous.
Meanwhile according to the suggested mechanism in both
scenarios where the node(3) has none or two child nodes,
node(4) will still be selected as the parent node as it above the
average LQI and it only goes through 1 hop to sink node
compared to 2 hops if attached to node(8). By attaching to
node(4), only node(X)'s and node(4)'s energy will be used in
transmitting the data to the sink node, meanwhile if attachment
in made to node(8) then energy of node(X)'s, node(8)'s and
node(l)'s will be consumed in transmitting the data to the sink
node.

The second scenario is when there are two or more
potential parent nodes with different number of existing child
nodes as represented in Fig. 4. According to the previous
mechanism the node(X) should associate itself with node(17)
or node(3) based on which node responded first as both nodes
have no child node compared to node(8) which has one child
node. In the case the first responding parent is node(3) then it is
disadvantageous if the child join node(3) as the link quality is
not good compared to node(17). Meanwhile if the first
responded parent node is node(17) the child will choose
node(17) to attach with. This attachment is acceptable if all the
nodes have same level of energy. In the case node(8) has
abundant amount of energy compared to node(17) then this
association is disadvantageous. Our mechanism suggests the
node(X) to take into consideration the average amount of
power the potential parent node has. In the case parent node(8)
has more average power than parent node(17) then node(X)
will join parent node(8). Meanwhile if the parent node(17) has
more power then it will attach itself with parent node(17).

The third scenario is when all potential parents have the
same number of child nodes. The previous mechanism didn't
anticipate such a situation will occur. Following our
mechanism node(X) will first compare potential parent depth
from sink node in this case node(4), node(8) and node(17) (Fig.

108

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

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

Vol. 008, No. 003, 2010

3). In this case the node(X) will try to associate with node(4)
as it nearer to the sink node and will not consider node(3) has
LQI below average. Meanwhile if the potential parent node
above average which responded is only node(8) and node(17),
node(X) will join the node which has highest average energy.
In the case both node(8) and node(17) has the same amount of
average energy it will associate itself with the node which has
highest LQI. In the case node(8) and node(17) has the same
depth, same average energy and same level of LQI then
node(X) will attach to the first potential parent which
responded.

Depth 1

Legend

Potential Parent with LQI below Average LQI

Potential Parent with LQI above or equal to
Average LQI

Figure 2. Two Potential Parent Node with LQI above average, with different
depth level and different number of existing child node

Depth

Depth 1

Depth 2

Depth 3

Legend

Potential Parent with LQI below Average LQI

Potential Parent with LQI above or equal to
Average LQI

Figure 3. Two potential Parent node with LQI above average and with same
depth level but different number of existing child node

Depth

Depth 1

Depth 2

Depth 3

Legend

Potential Parent with LQI below Average LQI

Potential Parent with LQI above or equal to
Average LQI

Figure 4. Potential parents with same number of child nodes with some from
different depth and some sharing same depth

IV. Conclusion

We provides review on HiLOW routing protocol, issues
revolving HiLOW and other works done in improving HiLOW.
In this paper we have suggested a new mechanism to overcome
disadvantageous parent and child node attachment in HiLOW;
disadvantageous attachment could jeopardize the reliability of
the network, shorten the life span of the network and also cause
wastage of energy due to retransmission. Previous mechanism
used seems to be less advantageous as it is not considering
potential link quality, the potential parent depth and also the
energy level. To verify our mechanism, we identify a number
of scenarios that need to optimize LQI, average energy level
and parent-child nodes parameters. By analyzing these
scenarios, the proposed new mechanism is shown to optimize
the parameters better that leads to establishing of a more
reliable network and enhancing the lifetime of the network.
However, the implementation of the proposed mechanism in an
actual network is a future work. Our future research will be
focused on validating the suggested mechanism as well as
solving the other issues highlighted in this paper.

Acknowledgment

The author would like to acknowledge Universiti Sains
Malaysia (USM) for funding of USM Fellowship Scheme
2009/10.

References

[1] Ian F.Akyildiz, Weihan Su, Yogesh Sankarasubramaniam and Erdal
Cayirci, "A Survey on Sensor Networks", Communication Magazine,
IEEE, Volume 40

[2] A. Dunkels, F. "Osterlind, N. Tsiftes, and Z. He. Software-based online
energy estimation for sensor nodes. In Proceedings of the Fourth IEEE

109

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

Workshop on Embedded Networked Sensors (Emnets IV), Cork,
Ireland, June 2007.

[3] L.Martin, B.D Mads, B.Philippe "Bluetooth and sensor networks: a
reality check", Proceedings of the 1st international conference on
Embedded networked sensor systems, 2003

[4] S.Jianping, et al., "WirelessHART: Applying Wireless Technology in
Real-Time Industrial Process Control", In Proceedings of IEEE Real-
Time and Embedded Technology and Apphcations Symposium, 2008.

[5] B.Chiara, C.Andrea, D.Davide, V.Roberto, "An Overview onWireless
Sensor Networks Technology and Evolution", Sensors 2009, Sensors
2009, 9, 6869-6896; doi:10.3390/s90906869

[6] N. Kushalnagar, et al., "Transmission of IPv6 Packets over IEEE
802.15.4 Networks", rfc4944, September 2007.

[7] IEEE Computer Society, "802. 15.4-2006 IEEE Standard for Information
Technology- Telecommunications and Information Exchange Between
Systems- Local and Metropohtan Area Networks- Specific
Requirements Part 15.4: Wireless Medium Access Control (MAC) and
Physical Layer (PHY) Specifications for Low-Rate Wireless Personal
Area Networks (WPANs)"

[8] K. Kim, S.Yoo, S.Daniel, J.Lee, G.MuUigan, "Problem Statement and
Requirements for 6L0WPAN Routing", draft-ietf-61owpan-routing-
requirements-02, March 2009

[9] K. Kim, S.Yoo, S.Daniel, J.Lee, G.MuUigan, "Commisioning in
6L0WPAN", draft-61owpan-commisioning-02, July 2008

[10] K. Kim, et al., "Hierarchical Routing over 6L0WPAN (HiLOW)", draft-
daniel-61owpan-hilow-hierarchical-routing-01, June 2007.

[11] K. Kim, et al., "Hierarchical Routing over 6L0WPAN (HiLOW)", draft-
daniel-61owpan-hilow-hierarchical-routing-00, June 2005.

[12] K. Kim, G.Montenegro, S.Park, I.Chakeres, C.Perkins, "Dynamic
MANET On-demand for 6L0WPAN (DYMO-low) Routing", draft-
montenegro-61owpan-dymo-low-routing-03, June 2007.

[13] K.Kim, S.Daniel, G.Montenegro, S.Yoo, N. Kushalnagar, "6L0WPAN
Ad Hoc On-Demand Distance Vector Routing (LOAD", draft-daniel-
61owpan-load-adhoc-routing-02, March 2006

[14] Martin Haenggi, "Opportunities and Challenges in Wireless Sensor
Network", Sensor Network Protocol, Taylor & Francais Group pp. 1-
1,1-7.

[15] Hun-Jung-Lim, Tai-Myoung Chung, "The Bias Routing Tree Avoiding
Technique for Hierarchical Routing Protocol over 6L0WPAN", 2009
Fifth International Joint Conference on INC, IMS and IDC.

[16] C.Nam, H.Jeong, D.Shin, "Extended Hierarchical Routing Protocol over
6L0WPAN", MCM2008, September 2008.

[17] V.C.Lingeswari et al., "Bias Child Node Association Avoidance
Mechanism for Hierarchical Routing Protocol in 6L0WPAN",
Proceedings of the Third IEEE International Conference on Computer
Science and Information Technology (In Press)

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

Vol 008, No. 003, 2010
[18] Zhu Jian, Zhao Lai, "A Link Quahty Evaluation Model in Wireless
Sensor Networks", Proceedings of the 2009 Third International
Conference on Sensor Technologies and Apphcations.

AUTHORS PROFILE
Lingeswari V.Chandra was bom in Penang,

Malaysia. She obtained her BIT with Management degree
from AIMST University in 2008. She is the university gold
medalist. She obtained her software engineering foundation
training from Infosys, Bangolore. She is currently pursuing
her PhD in National Advanced IPv6 Center, Universiti
Sains Malaysia.

Kok-Soon Chai was bom in Penang, Malaysia. He is
a certified Project Management Professional, Project
Management Institute, USA. He received his MSc and
Ph.D. (2003) degrees from the University of Warwick, UK.
He worked for more than seven years as a senior R&D
software engineer, embedded software manager, and CTO
at Motorola, Agilent, Plexus Corp., Wind River in
Singapore (now a division of Intel Corp.), and
NeoMeridian. He holds one US patent, with two US patents pending. His
main interests are wired and wireless sensor networks, green technology,
embedded systems, consumer electronics, and real-time operating systems.
Dr. Chai is a senior lecturer at the National Advanced IPv6 Centre of
Excellence (NAV6) in Universiti Sains Malaysia.

^^^ Gopinath Rao Sinniah obtained his BComp Science

^1^^^ and MSc (Comp Science) from the Universiti Sains

■rj^B Malaysia in 1999 and 2004 and currently pursuing his

' yH^B Ph.D at the same university. He has involved in the IPv6

V^^L^ development work since 1999 and currently working at

^^^^^^ MIMOS Berhad as Senior Staff Researcher focusing on

^^^^^^^H IPv6 specifically on wireless sensors network. Currently he

^^^^^^^^ holds 10 patents filed locally and 1 at WIPO. He is also the

chairman of MTSFB IPv6 working group.

^L I Sureswaran Ramadass obtained his BsEE/ce
^^ (Magna Cum Laude) and Master's in Electrical and
^^E Computer Engineering from the University of Miami in
^P 1987 and 1990, respectively. He obtained his Ph.D. from
. V Universiti Sains Malaysia (USM) in 2000 while serving as

JH a full-time faculty in the School of Computer Sciences. Dr.
^^^ Sureswaran Ramadass is a Professor and the Director of
the National Advanced IPv6 Centre of Excellence (NAV6)
in Universiti Sains Malaysia.

110

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

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

Vol. 8, No. 3, 2010

Rough Entropy as Global Criterion for Multiple DNA

Sequence Alignment

Sara El-Sayed El-Metwally

Dept. of Computer Science

Faculty of Computer and Information Science

Mansoura, Egypt

Sarah_almetwallv4@vahoo.com

ElSayed Radwan Taher. Hamza

Dept. of Computer Science Dept. of Computer Science

Faculty of Computer and Information Science Faculty of Computer and Information Science

Mansoura, Egypt Mansoura, Egypt

elsfradwan @ vahoo. com Taher_hamza @ vahoo. com

Abstract-T\)is paper presents a new method for multiple sequence
alignment using rough entropy as a global criterion to measure
the quality of alignment. This method collects DNA sequences in
clusters based on rough sets indcernibility relation. Rough
entropy is used to maximize the total number of sequences inside
each cluster with respect to the total number of sequences being
aligned. The method terminates when all aligned sequences are
located in all clusters and hence the problem is near optimally
solved.

Keywords-Multiple Sequence Alignment MSA, Rough Set
Theory, Indcernibility relation, Rough Entropy, Similarity relation

SIM(P),Sp(x).

I. Introduction

Multiple Sequence Alignment (MSA) is a fundamental and
challenging problem in computational molecular biology. It
plays a key role in computing similarity and in finding highly
conserved subsequences among set of DNA sequences .MSA is
one of the most commonly used methods for inferring
biological structures and functions. Moreover, it is the first
step of many tasks in computational biology involving
fragment assembly, evolutionary tree reconstruction, and
genome analysis.

For pairwise alignment, algorithms such as Needleman-
Wunsch and Smith- Waterman [1] which use dynamic
programming approach, guarantee finding the optimal solution
[2] .For multiple alignments, methods such as
multidimensional and repeating pairwise dynamic
programming, try to find near optimal solution in reasonable
amount of time. Finding optimal solution in reasonable
amount of time becomes difficult when the number of
sequences and the length of each sequence increase. The
problem of computing minimum cost for MSA has been shown
to A^P-hard. As a result, all multiple alignment algorithms
currently in use depend on different kinds of heuristics [2, 4, 7,
and 8].

The most common current solution to multiple DNA sequence
alignment problem is to use an evolutionary computation (EC)
where genetic algorithms find better alignment of sequences

with better similarity score by Darwin's theory about
evolution. Genetic algorithms use an objective function to
evaluate the quality of alignment. The objective function
summarizes the biological knowledge that is intended to be
projected into the alignment.

An alignment is considered to be correct if it reflects
evolutionary history of the species of sequences being aligned.
But, at the time of assessing the quality of alignment, such
evolutionary information is not frequently available or even
more not known [3].

One of existing measures for assessing the quality of
alignment is obtained by computing similarity using
substitutions matrices (for proteins). A substitution matrix
assigns a cost for each possible substitution or conservation
accordingly to the probability of occurrence computed from
data analysis. In this approach insertions and deletions are
weighted using affine gap penalties model. This model assigns
weight cost for each gap opening and gap extension in order to
favor alignments with smaller numbers of indels (each gap can
be regarded as an insertion-deletion event). The main
disadvantage of these substitution matrices is that they are
intended to rate the similarity between two sequences at a time
only. In order to extend them to multiple sequences, they are
scaled by adding up each pairwise similarity to obtain the
score of multiple sequence alignment [3, 4]. Another existing
measure is to use weighted sum of pairs, is the objective
function used by MSAs that associates a cost for each pair of
aligned bases in each column of the alignment (substitution
cost) and another similar cost for gaps (gap cost), sum of these
costs yields a global cost of the alignment[8,7 and 11].
This paper presents a new approach for multiple sequence
alignment using Rough Entropy as Global Criterion, REGC, to
measure the quality of alignment. This approach doesn't use
any traditional methods that compute similarity between all
sequences using scoring functions. Rather than rough sets
indcernibility relation used to collect DNA sequences in
clusters .Each cluster contains set of sequences that are similar
to each other. Gaps are inserted to each DNA sequence
according to specified length then sequences are aligned

111

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

randomly by change offsets of gaps in each sequence. In each
epoch, the random alignment is converted to suitable
representation for rough set analysis, matrix form, where each
sequence represents one row and each position of nucleotide
inside each sequence represents one column. The value of
each cell in the matrix can be a nucleotide base or gap
according to random alignment. Indcernibility definition for
missing values is used to compute the maximum clustering of
sequences due to the presence of gaps. REGC used to measure
the quality of alignment by maximize total number of
sequences inside each cluster with respect to the total number
of sequences. Each cluster has a name of one sequence and
objects inside it are set of sequences that are indiscernible by
knowledge P , set of attribute-value pairs which are
corresponding to positions and values of bases inside each
sequence.

This paper introduces a global criterion to measure the quality
of alignment by filling each cluster with the maximum number
of sequences in alignment problem. The rough entropy value
of knowledge P increases if the total number of sequences
inside each clusters increases. When one cluster doesn't
contain the maximum number of sequences, rough entropy
approach finds similar missing sequences by dropping
attributes from the knowledge P . If entropy value of
knowledge P after dropping attributes increases, this means
that the columns corresponding to those attributes must be
realigned. The process repeat until the entropy value remains
constant and hence the maximum numbers of sequences that
are similar to each other are located inside each cluster.
Paper is organized as follows: Section II gives an overview
about preliminaries of DNA sequence alignment and rough
sets. Section III presents the constructed hybrid model to align
DNA sequences based on Rough sets and Rough Entropy.
Section IV show simple experimented results of proposed
model and section V conclude this paper.

II. Preliminaries

A. DNA Sequence Alignment

DNA consists of two long strands of simple units called
nucleotides, with backbones made of sugars and phosphate
groups joined by ester bonds. These two strands run in
opposite directions to each other and are therefore anti-
parallel. Attached to each sugar is one of four types of letters
called bases. The four bases found in DNA are adenine
(abbreviated A), cytosine (C), guanine (G) and thymine (T).
Each type of base on one strand forms a bond with just one
type of base on the other strand. This is called complementary
base pairing, with A bonding only to T, and C bonding only to
G. The sequence of these four bases carries out the genetic
information that is used in the development and functioning of
all known living organisms. Thus DNA is the long-term
storage of genetic information. DNA is copied and transmitted
from parent to child, but from one generation to the next errors
in copying can occur, with one nucleotide being replaced by
another, or a new nucleotide being inserted, or an existing
nucleotide being deleted [10,12 and 13]. Over many

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

Vol 8, No. 3, 2010
generations, organisms with a common ancestor can end up
with fairly different DNA sequences.

To align two or more sequences, they are put together in a
S ' C matrix, where S is the number of sequences, and C is
the maximum number of bases in a sequence (positions in the
alignment); shorter sequences are filled at the end with gap
("-") to fit the matrix perfectly. The goal of alignment process
is to try to line up as many letters or portions of sequences that
are the same, to minimize the number of substitutions,
insertions, and deletions that would have had to happen if the
sequences came from a common ancestor. In order to choose
best alignment, the process of alignment has an objective to
align homologous residues (having the same evolutionary
origin) .The number of possible alignments of two sequences
grows exponentially as the length of sequences increases
[3]. So, with more sequences involved in the alignment
process, the number of possible alignments grows faster and
the problem of find optimal alignment becomes difficult to
solve [3].

B. Rough Set Theory

Rough set theory proposed by Pawlak [9] is an effective
approach to imprecision, vagueness, and uncertainty. Rough
set theory overlaps with many other theories such that fuzzy
sets, evidence theory, and statistics. From a practical point of
view, it is a good tool for data analysis. The main goal of the
rough set analysis is to synthesize approximation of concepts
from acquired data. The starting point of rough set theory is an
observation that objects having the same description are
indiscernible (similar) with respect to the available
information. Determination of similar objects with respect to
the defined attributes values is very hard and sensible when
some attribute values are missing. This problem must be
handled very carefully. The indiscernibility relation is a
fundamental concept of rough set theory which used in the
complete information systems. In order to process incomplete
information systems, the indiscernibility relation needs to be
extended to some equivalent relations.

The starting point of rough set theory which is based on
data analysis is a data set called information system ( IS ). IS
is a data table, whose columns are labeled by attributes, rows
are labeled by objects or cases, and the entire of the table are

the attribute values. Formally, 75 = (C/,A7'j, where U and
AT are nonempty finite sets called "the universe" and "the
set of attributes," respectively. Every attribute a G AT , has a
set of Y of its values called the "domain of ^ ". If y

contains missing values for at least one attribute, then S is
called an incomplete information system, otherwise it is
complete [5, 6, 8].
Any information table defines a function p that maps the

direct product U x AT into the set of all values assigned to
each attribute. The example of incomplete information system
depicted in Table I where set of objects in the universe
corresponding to set of DNA sequences and set of attributes
corresponding to set of bases inside each sequence. The values

112

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

of attributes are corresponding to the values of bases inside
each sequence such as the value of Sequence 1 at BaseO is
defined by /)(Seqi, Basco) = G

TABLE 1 : Example of Incomplete Information System

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

Vol 8, No. 3, 2010

Thus entry ^.. is the set of all attributes which discern objects

Baseo

Basel

Seq,

Seq2

Seq,

The concept of the indiscernibility relation is an essential
concept in rough set theory which is used to distinguish
objects described by a set of attributes in complete information
systems. Each subset A of AT defines an indiscernibility
relation as follows:

IND{A) = {(x,y)eUxU: p(x, a) = p(y, a)

\/aeA,A^AT} ^^^

The family of all equivalence classes of IND(A) is denoted
by U/IND(A)or UIA [5, 6 and 8].
Obviously IND(A) is an equivalence relation and:

IND(A) = n IND(a) where aeA (2)

A fundamental problem discussed in rough sets is whether
the whole knowledge extracted from data sets is always
necessary to classify objects in the universe; this problem
arises in many practical applications and will be referred to as
knowledge reduction. The two fundamental concepts used in
knowledge reduction are the core and reduct. Intuitively, a
reduct of knowledge is essential part, which suffices to define
all basic classifications occurring in the considered
knowledge, whereas core is in a certain sense the most
important part. Let A set of attributes and let a G A , the
attribute a is dispensable in A if:

IND(A) = IND(A - {a}) (3)

Otherwise a is indispensable attribute. The set of-y
attributes B , where B <^A is called reduct of A if:

IND(B) = IND(A) (4)

A may have many reducts. The set of all indispensable
attributes in A will be called the core of A , and will be
denoted as CO/?£( A).-

CORE (A) = n RED (A) (5)

Recently A.Skowron [8] has proposed to represent knowledge
in a form of discernibility matrix .this representation has many
advantages because it enables simple computation of the core
and reduct of knowledge.
Let K = (U,A) be a knowledge representation system with

U = {j^A^ j^2^"">Xn) ^^ ^ discernibility matrix of

^ denoted by M (/:), which means nxn matrix defined by:

(C>) " i^ ^A:p(x,a) ^ p(y,a)}for ij = l,2....,n. (6)

Identify applicable sponsor/s here, (sponsors)

xi ^n^jcj-

The core can be defined now as the set of all single element
entries of the discernibility matrix, i.e.

CORE (A) = {a^ A: q = (a) for some ij}, (7)

It can be easily seen that B (^ A is the reduct of A if 5 is
the minimal subset of A such that Bf^c ^ (^ for any

nonempty entry c(c^(^) in M (k) .In other words reduct is
the minimal subset of attributes that discerns all objects
discernible by the whole set of attributes. Let C,D(^A be
two subsets of attributes, called condition and decision
attributes respectively. KR-system with distinguished
condition and decision attributes will be called a decision table
and will be denoted T = (U,A,C,D) . Every xgU

associate a function ^j. : A^^y , such
that Jj^(^) = a(x) , for every ae C\jD , the function
^ will be called a decision rule, and x will be referred to
as a label of the decision rule j [5, 6, and 9].

III. Hybrid Model of Rough Sets and Rough Entropy

The hybrid model discussed in this paper is considered to
be a combination system that contains two methodologies
which are rough sets and rough entropy. DMA sequences are
converted to suitable representation for rough set analysis. The
rough set indcernibility relation used to collect DMA sequences
in clusters .Each cluster contains set of sequences that are
similar to each other. Rough Entropy is used to measure the
quality of alignment by maximize the total number of
sequences inside each cluster with respect to the total number
of sequences being aligned.

A. Rough Sets analysis for Sequence Alignment

The first part of the model consists of rough set analysis for
DMA sequence alignment. Rough set approach used here is
modified to deal with incomplete information system,
where IIS = (U, AT) , where U and AT are nonempty
finite sets called "the universe" and "the set of attributes,"
respectively. Every attribute a G AT , has a set of values
called Y and this set contains missing values for at least one
attribute. In order to process incomplete information
systems (7/5) , the indiscernibility relation has been extended
to some equivalent relations such as similarity relation.
Similarity relation 5/M(P) denotes a binary relation between

objects that are possibly indiscernible in terms of values of
attributes and in the case of missing values the modified
relation is defined by equation (8):

113

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

SIM{P) = {(x,y)eUxU, ae P,p(x,a) = p{y,a)
or p(x,a) = ^ or p(y,a) = ^]

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

Vol 8, No. 3, 2010

(8)

Sp{x) = {y e [/ : (jc, y) e SIM (PIP cz AT} (9)

Sp(x) Denotes maximal set of objects which are possibly

indiscernible by P with x [5, 6]. The indcernibility relation
in rough sets, as depicted in equation (8), used to collect DNA
sequences in clusters .Each cluster contains set of sequences
that are similar to each other. Gaps are inserted to each DNA
sequence according to specified length as depicted in Figure 1
then the sequences are aligned randomly by change offsets of
gaps in each sequence as depicted in Figure 2.

GCATGCTA

AGCTGC

TAGCAA

G C A CA T T "

Figure 1 : Gap insertion for Length=20 bases

GG~ft-TG~C T-fl

fl— G CTG— C

-TftGC~fl — n

~G-C-ft~CftTT

Figure 2: Random Alignment of DNA Sequences

In each epoch, the random alignment is converted to suitable
representation for rough set analysis, matrix form, where each
sequence represents one row and each position of nucleotide
inside each sequence represents one column. The value of
each cell in the matrix can be a nucleotide base or gap
according to random alignment as depicted in Table 2.

TABLE 2: Snapshot of Rough Representation Table

(a)

(b)

Indcernibility definition of missing values is used here to
compute the maximum clustering of sequences due to the
presence of gaps as depicted in Figure 3.

Figure 3: Set of Clusters corresponding to one Epoch

B. Rough Entropy for assessing the Quality of Alignment

The second part of the model uses REGC to measure the
quality of DNA sequence alignment. The definition of rough
entropy of knowledge in incomplete information system has
been introduced as:

Let IIS = (U,AT) , P^AT the rough entropy of
knowledge P is defined by the following equation:

I""' K (r)
E{P) = -2^ ' ,^^, ' log

\u\

(10)

Where:

U = \xvX2^X,^ 'Xf/}' ^^t of objects in

i.l'^2

the universe.

|[/ 1 is the cardinality of set U .
Log X = log2 X.

\Sp(x)\

represents the probability of tolerance

class C (j^.) within the universe [/ .

• -. r denotes the probability of one of

\Sp(x)\

values in tolerance class S p^X-^'

The maximum value of rough entropy for knowledge P is
computed by equation (11):

E(P) = \u\log\u\ (11)

This value is achieved only by the equation (12):

U/SIM{P) = {s^{x) = U\xgu} (12)

The minimum of rough entropy for knowledge P is . This
value is achieved only by the equation (13):

U/SIM(P) = {Sp(x) = [x}\xGu}[6]. (13)

REGC is used to measure the quality of alignment by
maximize the total number of sequences inside each cluster
with respect to the total number of sequences. Each cluster has
a name of one sequence such as depicted in Figure 3 and the
objects inside it are set of sequences that are indiscernible by
knowledge P , set of attribute-value pairs which are
corresponding to positions and values of bases inside each
sequence, such as depicted in Table 2.

The aim of our approach is to make a global criterion to
measure the quality of alignment by filling each cluster with

114

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

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

Vol 8, No. 3, 2010

the maximum number of sequences in alignment problem as
depicted in Figure 4 where the total number of sequences
being aligned is 4 sequences and the total number of clusters is
4 clusters one for each sequence .REGC try to fill each cluster
with maximum number of sequences similar to the sequence
corresponding to that cluster. The alignment process ends
when the total number of sequences located inside each cluster
is equal to the total number of sequences being aligned.

the Similar Sequences for Se0= 12 3

the Similar Sequences for Sel= 12 3

the Similar Sequences for Se2= 12 3

the Similar Sequences for Se3= 12 3

Figure 4: Maximum Clusters Set

When the total number of sequences inside each cluster
increases, the rough entropy value of knowledge P increases
such as depicted in Figure 5 where the rough entropy value of
P in Figure 5-b is higher than in 5- a because clusters in 5-b
are filling with the maximum number of sequences that are
similar to each other.

the Similar Sequences for Se0= 12

the Similar Sequences for Sel= 12

the Similar Sequences for Se2= 12

the Similar Sequences for Se3= 3
E<A>=3-5661656266226

(a)

the Similar Sequences for SeO=

the Similar Sequences for Sel=

the Similar Sequences for Se2=

the Similar Sequences for Se3=

12 3

(b)

Figure 5: Rough Entropy Value for Knowledge P

When one cluster doesn't contain the maximum number of
sequences, rough entropy approach finds similar missing
sequences by dropping attributes from the knowledge P . If
the entropy value of knowledge after dropping attributes
increases, this means that the columns corresponding to those
attributes must be realigned. The process repeat until the
entropy value remains constant and hence the maximum
number of sequences that are similar to each other are located
inside each cluster.

REGC Algorithm for MSA

• Input:

N : Total number of sequences to be aligned.
M : Max Length.
R : Randomization Times (optional).

Si'S2'S3'"''Sn' ^^^^^^^^^ being aligned.

• Output:

Alignment ofDNA Sequences

1. [Start] insert gaps to all sequences until all have the
same length M .

2. [Alignment]

o If ( /? =1) create an alignment by randomly
change the offset of gaps in each
sequence.

o If ( /? =2,3,.. R ) create an alignment by
randomly change the offset of gaps in
columns indexed as realigned from
attribute analysis step in previous epoch

3. [Rough representation] convert the resulting random
alignment to a suitable representation for rough set
analysis , matrix form, where each sequence represent
one row and each position of nucleotide inside each
sequence represent one column. The value of each
cell in the matrix can be a nucleotide base or gap
according to random alignment. Rows corresponding
to set of objects in the universe and columns are set
of attributes, knowledge P

4. [Rough Entropy] compute the entropy value of
knowledge P that satisfies equation 10.

5. [Attribute Analysis]REGC analysis the entropy value
by check each cluster if contains the total number of
sequences being aligned and hence for any

sequence X S p^"^^ ~ ^ -When one cluster doesn't

contain the maximum number of sequences, REGC
finds similar missing sequences by dropping

1,2, ,M — lattributes from the knowledge P. If

the entropy value of knowledge after dropping
attributes increases, this means that the columns
corresponding to those attributes must be realigned.

6. [Loop] go to step 2 until the maximum value of

rough entropy of knowledge P is C/logC/

reached, this value is achieved only

by U/Sm(P) = {^^(x) = U\xgU], or until

the maximum number of randomization times
encountered.

IV . Experimented Re s ult s

All DMA fragments supplied to our model are obtained
from Gene repository over the internet i.e.
http://www.ncbi.nlm.nih.gov/Genbank/ [14] Genbank, is the
NIH genetic sequence database, an annotated collection of all

115

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

publicly available DNA sequences. There are approximately
106,533,156,756 bases in 108,431,692 sequence records in the
traditional GenBank divisions and 148,165,117,763 bases in
48,443,067 sequence records in the WGS division as of
August 2009. GenBank is part of the International Nucleotide
Sequence Database Collaboration, which comprises the DNA
Databank of Japan (DDBJ), the European Molecular Biology
Laboratory (EMBL), and GenBank at NCBI. These three
organizations exchange data on a daily basis.
Consider these four input fragments of DNA sequences which
are given from Genbank.

GCATGCTA

AGCTGC

TAGCAA

GCACATT
Where the other input parameters supplied to our program are
7V=4,M=20and7?=100.

1. [Start] insert gaps to all sequences until all have the
same length 20.

GCATGCTA

AGCTGC

TAGCAA

GCACATT

2. [Alignment] create an ahgnment by randomly change
the offset of gaps in each sequence as depicted in
Figure 6.

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

Vol 8, No. 3, 2010
(b)

GC~fl-T G~C T -fl

ft~G CT G~C

-TfiGC— fl — n

k-G-C-fl~CflTT

the Similar Sequences

for

Se0=

the Similar Sequences

for

Sel =

the Similar Sequences

for

Se2 =

the Similar Sequences

for

Se3 =

Figure 6: Random Alignment of DNA Sequences

TABLE 3: Rough Representation of DNA Sequence Alignment

(a)

>!^

Pio

Pii

P12

Pl3

Pl4

Pl5

P16

Pl7

P18

Pl9

for

See=

for

Sel =

for

Se2 =

for

Se3 =

4. [Rough Entropy] compute the entropy value of
knowledge P that satisfies equation 10 as depicted
in Figure 7.

— G-C-fl— CfiTT

GC— A-TG— C T-ft

I— G CTG— C

-TfiGC— A A

— G-C-A— CATT

;he Similar Sequences
;he Similar Sequences
;he Similar Sequences
;he Similar Sequences
;<A>=3. 5661656266226

Figure 7: Rough Entropy Value of Knowledge P

5. [Attribute Analysis]whQn one cluster doesn't contain
the maximum number of sequences, rough entropy
approach finds similar missing sequences by
dropping one, two... M— 1 attributes from
knowledge P . Dropping attributes leads to increase
the number of sequences inside each cluster and
hence the value of entropy increases. The increasing
value of entropy after dropping the attributes
indicates that some columns in the alignment are

needed to realign and ^p{^)^U for any
sequence X .When dropping one attribute from
knowledge P as depicted in Figure 8, the value of
entropy remains constant. Every cluster doesn't
contain all sequences in the alignment and the

equation ^ (x) ^ U satisfied for any sequence X .
When dropping the

attributes /^2'/^9'/^io^^^/^ii ' ^^^ ^^^^^ ^^
entropy increased to 8 as depicted in Figure 9
.Increasing the value of entropy indicates that all
columns corresponding to those attribute must be
realigned to satisfy this value.

116

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

GC — A -T G — C T -fl

A — G CT G— C

-T A GC— fl A

— G-C- A — C A T T

the Similar Sequences for

the Similar Sequences for
E<A> =3.5661656266226

the Similar Sequences for

the Similar Sequences for
E<fl-P0>=3 -5661656266226

the Similar Sequences for

the Similar Sequences for
E<A-P1>=3 -5661656266226

the Similar Sequences for

the Similar Sequences for
E<fl-P2>=3 .5661656266226

Figure 8: Relation between Dropping Attributes of Knowledge P and Rough

Entropy

6. [Loop] go to Step 2 until the maximum value of

rough entropy of knowledge P is |C/| log|C/| reached

as depicted in Figure 9 , this value is achieved only

by the U/SIM(P) = {^^(x) = U\xgU] , or

until the maximum number of randomization times
encountered.

Se0 =

Sel =

Se2 =

Se3 =

Se0 =

Sel =

Se2 =

Se3 =

Se0 =

Sel =

Se2 =

Se3 =

Se0 =

Sel =

Se2 =

Se3 =

— GCftTGCTA-

-AGC-TGC

IflGCfl fl-

— GC A — C- A T T

bhe Similar

Sequences

for

Se0 =

the Similar

Sequences

for

Sel =

the Similar

Sequences

for

Se2 =

the Similar

Sequences

for

Se3 =

E<fl>=8

the Similar

Sequences

for

Se0 =

the Similar

Sequences

for

Sel =

the Similar

Sequences

for

Se2 =

the Similar

Sequences

for

Se3 =

E<fl-P0>=8

the Similar

Sequences

for

Se0 =

the Similar

Sequences

for

Sel =

the Similar

Sequences

for

Se2 =

the Similar

Sequences

for

Se3 =

E<A-P1>=8

the Similar

Sequences

for

Se0 =

the Similar

Sequences

for

Sei =

the Similar

Sequences

for

Se2 =

the Similar

Sequences

for

Se3 =

E<fl-P2>=8

the Similar

Sequences

for

Se0 =

the Similar

Sequences

for

Sel =

the Similar

Sequences

for

Se2 =

the Similar

Sequences

for

Se3 =

E<fl-P3>=8

the Similar

Sequences

for

Se0 =

the Similar

Sequences

for

Sel =

the Similar

Sequences

for

Se2 =

the Similar

Sequences

for

Se3 =

E<A-P4>=8

the Similar

Sequences

for

Se0 =

the Similar

Sequences

for

Sei =

the Similar

Sequences

for

Se2 =

the Similar

Sequences

for

Se3 =

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

Vol 8, No. 3, 2010
V. Conclusions

This paper presents a new approach for multiple sequence
alignment using REGC to measure the quality of the
alignment. This approach doesn't use any traditional methods
that compute similarity between all sequences using scoring
functions, Rather than rough set indcernibility relation used to
collect DNA sequences in clusters where each cluster contains
set of sequences that are similar to each other. Gaps are
inserted to each DNA sequence according to specified length
then sequences are aligned randomly by change offsets of gaps
in each sequence. REGC is used to measure the quality of
alignment by maximize the total number of sequences inside
each cluster with respect to the total number of sequences
being aligned. Each cluster has a name of one sequence and
the objects inside it are set of sequences that are indiscernible
by knowledge P , set of attribute-value pairs which are
corresponding to positions and values of bases inside each
sequence. REGC used to measure the quality of alignment by
filling each cluster corresponding to one sequence with the
maximum number of sequences in the alignment problem .The
rough entropy value of knowledge P increases when the total
number of sequences inside each cluster increases. If one
cluster doesn't contain the maximum number of sequences,
rough entropy approach finds similar missing sequences by
dropping attributes from knowledge P . If the entropy value
of knowledge after dropping attributes increases, this means
that the columns corresponding to those attributes must be
realigned. The process repeat until the maximum value of

rough entropy of knowledge P is C/ logC/ reached; this

value is achieved

by U/SIM(P) = {^^(x) = U\xgU\

sequenced .

for

only
any

Figure 9: Relation between Dropping Attributes of Knowledge P and Rough

Entropy

Acknowledgment

I would like to thank my father Dr.El-Sayed El-Metwally
and my mother Dr.Hemmat El-Shik for their moral support I
required in my life at all.

I am heartily thankful to my supervisor. Dr. ElSayed
Radwan, whose encouragement, supervision and support from
the preliminary to the concluding level enabled me to develop
an understanding of the subject.

Lastly, I offer my regards and blessings to all of those who
supported me in any respect during the completion of the
paper.

References

[1] Arthur M.Lesk, Introduction to Bioinformatics, University
of Cambridge, USA, Oxford University Press,2002

[2] Bryan Bergeron, Bioinformatics Computing, University of
British Columbia, USA, Prentice Hall,2002

[3] Edgar D. Arenas-D'iaz, Helga Ochoterena, and Katya
Rodriguez-Vazquez, Multiple Sequence Alignment Using
a Genetic Algorithm and GLOCSA, Universidad Nacional
Aut'onoma de Mexico, Genetic and Evolutionary
Computation Conference, Proceedings of the 2008 GECCO

117

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

conference companion on Genetic and evolutionary
computation, pp. 1795-1798, USA, ACM, 2008.

[4] Edward Keedwell and Ajit Narayanan, Intelligent
Bioinformatics, School of Engineering and Computer
Sciences, University of Exeter, UK, Wiley, 2005.

[5] E. A. Rady, M. M. E. Abd El-Monsef, and W. A. Abd El-
Latif, A Modified Rough Set Approach to Incomplete
Information Systems, Journal of Applied Mathematics and
Decision Sciences, Article ID 58248, Egypt, Hindawi
Pubhshing Corporation, , Volume 2007

[6] JIYE LIAN and ZONGBEN XU, The Algorithm on
Knowledge Reduction in Incomplete Information
Systems, Shanxi University, Taiyuan, International journal of
uncertainty, Fuzziness and Knowledge based systems.
Volume 2002.

[7] Koji Tajima, Multiple Sequence Alignment Using Parallel
Genetic Algorithms, Institute for social information Science
,FUJITSU LABORATORIES LTD, Japan ,Genome
Informatics Workshop IV, Volume 4, 1993

[8] Narayanan, E.C. Keedwell and B. Olsson, Artificial
Intelligence Techniques for Bioinformatics, School of

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

Vol 8, No. 3, 2010
Engineering and Computer Sciences, University of Exeter,
Exeter EX4 4QF, UK, Journal of Applied Bioinformatics ,
2005.

ZDZISLAW PAWLAK, Rough Sets Theoretical Aspects of
Reasoning about Data, Institute of Computer Science,
Warsaw University of Technology, Australia, Kluwer
Academic Publishers, 1991

http://en.wikipedia.org/wiki/Multiple sequence alignment
10-7-2009

http://oreillv.com/news/bioinformatics 040 1 .html 2-3-2008
http://www.ebi.ac.uk/2can/tutorials/transcription.html
4-4-2008

http://en.wikipedia.org/wiki/Genetic code 6-5-2008
http://www.ncbi.nlm.nih.gov/Genbank/ 5-1-2010

[9]

[10]

[11]
[12]

[13]
[14]

118

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

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

Vol 8, No. 3, 2010

Weighted Attribute Fusion Model for Face

Recognition

S.Sakthivel

Assistant Professor, Department of Information

Technology

Sona college of Technology, Salem, India

sakthits @rediffmaiLcom

Dr.R.Lakshmipathi

Professor, Department of Electrical and Electronic

Engineering

St.Peter's Engineering College, Chennai, India

drrlakshmipathi@vahoo.com

Abstract — Recognizing a face based on its attributes is an easy
task for a human to perform as it is a cognitive process. In recent
years, Face Recognition is achieved with different kinds of facial
features which were used separately or in a combined manner.
Currently, Feature fusion methods and parallel methods are the
facial features used and performed by integrating multiple
feature sets at different levels. However, this integration and the
combinational methods do not guarantee better result. Hence to
achieve better results, the feature fusion model with multiple
weighted facial attribute set is selected. For this feature model,
face images from predefined data set has been taken from
Olivetti Research Laboratory (ORL) and applied on different
methods like Principal Component Analysis (PCA) based Eigen
feature extraction technique. Discrete Cosine Transformation
(DCT) based feature extraction technique. Histogram Based
Feature Extraction technique and Simple Intensity based
features. The extracted feature set obtained from these methods
were compared and tested for accuracy. In this work we have
developed a model which will use the above set of feature
extraction techniques with different levels of weights to attain
better accuracy. The results show that the selection of optimum
weight for a particular feature will lead to improvement in
recognition rate.

Keywords- Face Recognition, Feature Fusion Method, Parallel
Method, PCA, DCT, Histogram Matching

Introduction

Face recognition is an important part of today's emerging
biometrics and video surveillance markets. Face Recognition
can benefit the areas of: Law Enforcement, Airport Security,
Access Control, Driver's Licenses & Passports, Homeland
Defense, Customs & Immigration and Scene Analysis. Face
recognition has been a research area for almost 30 years, with
significantly increased research activity since 1990[16] [15].
This has resulted in the development of successful algorithms
and the introduction of commercial products. But, the
researches and achievements on face recognition are still in its
initial stages of development. Although face recognition is still
in the research and development phase, several commercial
systems are currently available and research organizations
are working on the development of more accurate and
reliable systems. Using the present technology, it is

impossible to completely model human recognition system and
reach its performance and accuracy. However, the human

brain has its shortcomings in some aspects. The benefits of a
computer system would be its capacity to handle large amount
of data and ability to do a job in a predefined repeated manner.
The observations and findings about human face recognition
system will be a good starting point for automatic face
attribute.

A. Early Works

Face recognition has gained much attention in the last two
decades due to increasing demand in security and law
enforcement applications. Face recognition methods can be
divided into two major categories, appearance-based method
and feature-based method. Appearance-based method is more
popular and achieved great success [3].

Appearance-based method uses the holistic features of a 2-
D image [3]. Generally face images are captured in very high
dimensionality, normally which is more than 1000 pixels. It is
very difficult to perform face recognition based on original face
image without reducing the dimensionality by extracting the
important features. Kirby and Sirovich first used principal
component analysis (PCA) to extract the features from face
image and used them to represent human face image [16]. PCA
seeks for a set of projection vectors which project the image
data into a subspace based on the variation in energy. Turk and
Pentland introduced the well-known Eigenface method [15].
Eigenface method incorporates PCA and showed promising
results. Another well-known method is Fisher face. Fisher face
incorporates linear discriminant analysis (LDA) to extract the
most discriminant features and to reduce the
dimensionality [3]. when it comes to solving problems of
pattern classification, LDA based algorithms outperform PCA
based ones, since the former optimizes the low dimensional
representation of the objects with focus on the most
discriminant feature extraction while the latter achieves simply
object reconstruction [9][10][1 1]. Recently there has been a lot
of interest in geometrically motivated approaches to data
analysis in high dimensional spaces. This case is concerned
with data drawn from sampling a probability distribution that
has support on or near a sub manifold of Euclidean space [5].

Let us consider a collection of data points of n-dimensional
real vectors drawn from an unknown probability distribution.
In increasingly many cases of interest in machine learning and
data mining, one is confronted with the situation which is very

119

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

large. However, there might be reason to suspect that the
"intrinsic dimensionahty" of the data is much lower. This leads
one to consider methods of dimensionality reduction [1][2][8]
that allow one to represent the data in a lower dimensional
space. A great number of dimensionality reduction techniques
exist in the literature.

In practical situations, where data is prohibitively large, one
is often forced to use linear (or even sub linear) techniques.
Consequently, projective maps have been the subject of
considerable investigation. Three classical yet popular forms of
linear techniques are the methods of PC A [6] [14] [1][2],
multidimensional scaling (MDS) [6] [14], and LDA [14] [11].
Each of these is an eigenvector method designed to model
linear variability's in high-dimensional data. More recently,
frequency domain analysis methods [3] such as discrete
Fourier transform (DFT), discrete wavelet transform (DWT)
and discrete cosine transform (DCT) have been widely adopted
in face recognition. Frequency domain analysis methods
transform the image signals from spatial domain to
frequency domain and analyse the features in frequency
domain. Only limited low-frequency components which
contain high energy are selected to represent the image.
Unlike PCA, frequency domain analysis methods are data
independent [3]. They analyse image independently and do not
require training images. Furthermore, fast algorithms are
available for the ease of implementation and have high
computation efficiency.

In [3] new parallel models for face recognition were
presented. Feature fusion is one of the easy and effective ways
to improve the performance. Feature fusion method is
performed by integrating multiple feature sets at different
levels. However, feature fusion method does not guarantee
better result [3]. One major issue is feature selection. Feature
selection plays a very important role to avoid overlapping
features and information redundancy. New parallel model for
face recognition utilizes information from frequency and
spatial domains, addresses both features and processes in
parallel way. It is well- known that image can be analysed
in spatial and frequency domains. Both domains describe
the image in very different ways. The frequency domain
features [3] are extracted using techniques like DCT, DFT and
DWT methods respectively. By utilizing these two or more
very different features, a better performance is guaranteed.

Feature fusion method suffers from the problem of high
dimensionality because of the combined features. It may also
contain redundant and noisy data. PCA is applied on the
features from frequency and spatial domains to reduce the
dimensionality and extract the most discriminant information
[3]. It is surprising that until recent study demonstrated
that colour information makes contribution and enhances
robustness in face recognition [4].

II. MATERIALS AND METHODS

In statistics, dimension reduction is the process of reducing
the number of random variables under consideration, and can
be divided into feature selection and feature extraction.

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

Vol 8, No. 3, 2010
A. Extracting Eigen Features Fl

In previous work [1] five algorithms are evaluated, namely
PCA, Kernel PCA, LDA [9] [10], Locality Preserving
Projections (LPP) [1] [8] and Neighbourhood Preserving
Embedding (NPE) [1][2] [5] for dimensionality reduction and
feature extraction and found that Kernel PCA was the best
performer. This work uses PCA based algorithm to show some
improvements in its performance due to the use of Multiple
Weighted Facial Attribute Sets. The basic idea of PCA [5] is to
project the data along the directions of maximal variances so
that the reconstruction error can be minimized. Given a set of
data points xl... xn, let a be the transformation vector and yi =

a X;. The objective function of PCA is as follows:

a ^argmaxV (y. -y)^ = argmax a' Ca

-(1)

i=\

In equation y = ~/_, Yi and C is the data covariance the
n
eigen. The basic functions of PCA are the eigenvectors of the
data covariance matrix corresponding to the largest
eigen values. While PCA seeks direction that are efficient for
representation.

B. Extracting DCT Features F2

The DCT [2] can be used to Create DCT feature Set of the
Face. The Discrete Cosine Transform is a real domain
transform which represents the entire image as coefficients of
different frequencies of cosines (which are the basis vectors for
this transform). The DCT of the image is calculated by taking
8x8 blocks of the image in Figure 1, which is then transformed
individually. The 2D DCT of an image gives the result matrix
such that top left corner represents lowest frequency coefficient
while the bottom right corner is the highest frequency.

1—

1 —

Figure 1 The frequency domain representation of an image
The 1-D discrete cosine transform (DCT) is defined as

jc=0

C{u) = a{u)^f{x) • cos

(2)

Similarly, the inverse DCT is defined as

M=0

/(x) = ^a(i/)c(i/)-cos

(3)

for x= 0, 1 ,2, . . . ,A/^ 1 . In both equations (2) and (3)
(u) is defined as

120

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

a{u) =

for w = l,2,...,A^-l

The corresponding 2-D DCT, and the inverse DCT are
defined as

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

Vol 8, No. 3, 2010
major learning paradigms, each corresponding to a particular
abstract learning task. These are supervised learning,
unsupervised learning and reinforcement learning. Usually any
given type of network architecture can be employed in any of
those tasks.

■(4)

NANA

dii4,vj=4i^ci^yyA^y)'c^^

j^ro

{2x-^iju7r

cos^

"M'

■(5)

for u,v =0,1,2, ...,N 1 and (u) and (v) are defined in (4).
The inverse transform is defined as

/(x,};)=^^a(w)a(v)c(w,v)-cos

{2x + l)u7r

{2y + l)v7r

(6)

The advantage of DCT is that it can be expressed without
complex numbers. The DCT transform Equation (5) can be
expressed as separable, (like 2-D Fourier transform), i.e. it can
be obtained by two subsequent 1-D DCT in the same way as
Fourier transform. Equation (6) shows the Inverse
transformation.

C. The Histogram Feature Vector F3

The distribution of gray levels occurring in an image is
called gray level histogram. It is a graph showing the frequency
of occurrence of each gray level in the image versus the gray
level itself. The plot of this function provides a global
description of the appearance of the image. The histogram of a
digital image with gray levels in the range [0, L-1] is a discrete
function .

Hk/n

(7)

P (rk) ■■

Where,

Vi -^ Kth gray level

n^ -^ No of pixels in the image with that gray level.

n -^ total number of pixels in the image.

K =0,1, 2... L-1.

L = 256. (For 256 level gray images)

In Equation 7 V(ji) gives an estimate of the probability of
occurrence of gray level Vi If we use L value of small size,
then Wi will contain a range of nearest values in L number f
bins. So for constructing Histogram Based Feature, the set n^
and the mid values of the bin n^ were combined.

D. The Intensity Based Feature F4

Intensity Feature set can be formed by using values like
mean Median Mode of the 256 gray level images. We can use
one or many of this values to represent the average intensity of
the face image.

E. Neural Networks and Learning Paradigms

In principle, the popular neural network can be trained to
recognize face images directly. However, a simple network can
be very complex and difficult to train [12] [17]. There are three

F. Learning Algorithms

Training a neural network model essentially means
selecting one model from the set of allowed models (or, in a
Bayesian framework, determining a distribution over the set of
allowed models) that minimizes the cost criterion. There are
numerous algorithms available for training neural network
models; most of them can be viewed as a straightforward
appHcation of optimization theory and statistical estimation.
Most of the algorithms used in training artificial neural
networks are employing some form of gradient descent. This is
done by simply taking the derivative of the cost function with
respect to the network parameters and then changing those
parameters in a gradient-related direction. Evolutionary
methods simulated annealing, and Expectation-maximization
and non-parametric methods are among other commonly used
methods for training neural networks.

G. Support vector machines (SVMs)

Support vector machines are a set of related supervised
learning methods used for classification and regression.
Viewing input data as two sets of vectors in an n-dimensional
space, an SVM will construct a separating hyper plane in that
space, one which maximizes the margin between the two data
sets[7][13]. To calculate the margin, two parallel hyper planes
are constructed, one on each side of the separating hyper plane,
which is "pushed up against" the two data sets. Intuitively, a
good separation is achieved by the hyper plane that has the
largest distance to the neighbouring data points of both classes,
since in general the larger the margin the better the
generahzation error of the classifier. For the linearly separable
case, a hyper-plane separating the binary decision classes in the
three-attribute case can be represented as the following
equation:

(8)

In Equation (8), y is the outcome Xj, are the attribute values,
and there are four weights Wj to be learned by the learning
algorithm. In the above equation, the weights Wj are parameters
that determine the hyper-plane. The maximum margin hyper-
plane can be represented as the following equation in terms of
the support vectors:

■-b + Ya^tytK^^(t).x)

(9)

In Equation (9) the function K(x(t),x) is defined as the
kernel function. There are different kernels for generating the
inner products to construct machines with different types of
nonlinear decision surfaces in the input space.

SVM is selected as the classifying function. One distinctive
advantage this type of classifier has over traditional neural
networks is that SVMs can achieve better generalization
performance. Support vector machine is a pattern classification
algorithm developed by Vapnik [13]. It is a binary

121

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

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

Vol 8, No. 3, 2010

classification method that finds the optimal linear decision
surface based on the concept of structural risk minimization. As
shown by Vapnik, this maximal margin decision boundary can
achieve optimal worst-case generahzation performance. Note
that SVMs are originally designed to solve problems where
data can be separated by a linear decision boundary

III. TheModel of proposed system

Given a set of feature vectors belonging to n classes, a
Support Vector Machine (SVM) finds the hyper plane that
separates the largest possible fraction of features of the same
classes on the corresponding space, while maximizing the
distance from either class to the hyper plane. Generally a
suitable transformation is first used to extract features of face
images and then discrimination functions between each class of
images are learned by SVMs. Figure 2 shows the feature set
creation for face images from ORL database whereas Figure 3
shows the architecture for training and testing the SVM with
weighted attribute set.

Load n Training Face Images from ORL Face

Database

Resize the Ima

ges in to 48x48
1

Reshape the Images in to Ix 2304 and Prepare

n X 2304 Matrix 'M'

V V ^r

Find the

Find

Find the

Eigen

DCTof

Histogra

Average

Vector

all the

m of all

Intensity

Matrix

Row

the Row

of Each

'V and

elements

Row of

Project

'M' and

the

Matrix

forma

Matrix

'M' to

Intensity

MonV

form

form A

Feature

to find

DCT

Histogra

Matrix

the

feature

Eigen

Matrix

feature

Feature

'D' and

Matrix

Find

'F3'

PCAto

forma
Feature
Matrix
F2

^ ^

r ^

Feature

SetFl

SetF2

SetF3

SetF4

Feature
SetFl

Feature
SetF2

Multiply
with
Weight
Wl

Feature
SetF3

Multiply

with

Weight

Feature
SetF4

Multiply
with
Weight
W3

Multiply
with

Weight
W4

F1*W1

F2*W2

Iz.

F3*W3

±2.

F4*W4

The Input Layer of SVM Neural Network

The Final Trained SVM Neural Network

Figure 2: The Feature Set Creation

Figure 3: Architecture for Training and Testing the SVM with Weighted
Attribute Set

A. Steps Involved in Training

1) Load a set of 'n' ORL Face Images For Training

2) Resize the Images in to 48x48 pixel size to reduce the
memory requirements of the overall application

3) Reshape the Images in to Ix 2304 and prepare an n x 2304
Feature Matrix representing the training data set.

4) Apply a Feature Extraction, DCT, Histogram and
Dimensionality Reduction technique and find the Four of
above said Features sets Fl, F2, F3 and F4.

5) Multiply the Weights Wl, W2, W3 and W4 with Fl, F2,
F3 and F4.

6) Create an SVM network with "fl+f2+f3+f4" inputs where
fl, f2, f3 and f4 are the corresponding feature lengths.

7) Train a SVM using the Weighted Feature Set.

B. Steps Involved in Testing

1) The first three steps of the above procedures will be
repeated with test image set of ORL database to obtain a
Feature Matrix representing the testing data set.

2) Project the matrix using Previous Eigen Vector Matrix and
find the input Eigen Feature iFl.

3) Similarly find other input feature sets iF2, iF3 and iF4 of
the input Image.

4) Classify the feature set [iFl iF2 iF3 iF4] with previously
trained SVM network.

5) Calculate Accuracy of Classification

122

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

IV . Implementation results and analysis

The performance of proposed face recognition model was
tested with the standard set of images called "ORL Face
Database". The ORL Database of Faces contains a set of face
images used in the context of a face recognition project carried
out in collaboration with the Speech, Vision and Robotics
Group of the Cambridge University Engineering Department.
There are ten different images, each of 40 distinct subjects. For
some subjects, the images were taken at different times,
varying the lighting, facial expressions (open / closed eyes,
smiling / not smiling) and facial details (glasses / no glasses).
All the images were taken against a dark homogeneous
background with the subjects in an upright, frontal position
(with tolerance for some side movement).

Set of images from ORL databases were used for Training
and Testing. The accuracy of recognition with multiple
weighted attribute sets as well as a single attribute sets has been
evaluated. The following table shows the overall results of
these two types of techniques with different number of input
face images.

Table 1 Accuracy of recognition with single attribute set

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

Vol 8, No. 3, 2010
well as multiple attribute sets with same priority or weight will
lead to poor recognition results. For example, all the results of
Table 1 which used a single attribute at a time for recognition,
is in some what poor than Table 2 which is using combined
multi attributes. Further, if we note the fourth row (40 images)
corresponding to the weights (Wl=l, W2=l, W3=l, W4=l)
which is using all the attributes with same weight, the result is
poor while comparing it with column 2 (Wl=0.5, W2=l,
W3=0, W4=0). The following Graphs show the performance of
the two different approaches. In Figure 4 Line charts shows the
Performance with single attribute set, but in Figure 5 Line
charts shows the performance with multiple weighted attribute
sets.

No. of

Faces used

for

Training

and

Testing

Accuracy of Recognition with different
Weight Sets (%)

Wl=l
W2=0
W3=0
W4=0

W1=0
W2=l
W3=0
W4=0

W1=0
W2=0
W3=l
W4=0

W1=0
W2=0
W3=0
W4=l

90.00

40.00

80.00

90.00

80.00

25.00

60.00

65.00

83.33

16.67

56.67

76.67

80.00

12.50

52.50

70.00

Average

83.33

23.54

62.29

75.42

Table 2 Accuracy of recognition with multiple attribute set

No. of

Faces

used for

Training

and

Testing

Accuracy of Recognition with different
Weight Sets (%)

Wl = l
W2=l
W3=0
W4=0

Wl=.5
W2=l
W3=0
W4=0

Wl=.5
W2=l
W3=0
W4=l

Wl=l
W2=l
W3=l
W4=l

Wl=.12
W2=0
W3=l
W4=0

90.00

100

80.00

85.00

90.00

86.67

83.33

86.67

82.50

85.00

82.50

Average

84.79

89.17

88.54

87.71

89.79

As shown in the above table and the following graphs, the
performance of recognition while using Single Attribute Set as

100
90

; 80

i 70

! eo

O 50

40
30
20
10

Performance Wth Different Attribites

-^ Bgen Featines

-■- Intensity
G^l-fstogram
DCri^eatiJies

10 20 30 40

iSJLiinber of l^aoe Images

Figure 4 Performance with single attribute

Rerfommoe V\Ath Dflereit MribUtes and M\lsi^rts

20 30

NLirber cf Faoe Images

-wi=i,v\e=i,v\iB=o,v\«da -■-wida^v\2=i,v\Gd(^wida

WI=0.^V\e=1,V\G=0,WI=1. ^^ WI=1,V\fi=1,V\lB=1,Wfc1.
-WI^12,V\G=0,V\Q=1,WI=0.

Figure 5 Performance with multiple weighted attribute sets

123

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

The Following Two Charts shows the average performance
of the two approaches. In Figure 6, column charts shows the
average performance with single attribute set.

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

Vol 8, No. 3, 2010
V. CONCLUSION

Qo nn -,

C7 u .u u

83.33

o u .u u
"7 n nn -

/i)AZ

/ U .UU

62.29

dU .UU

OU .UU
/I n n n -

4U .UU
"^Ci r\r\ -

oU .UU

23.54

^U .UU
^ r\ r\r\ _

1 U .UU
r\ r\r\

U .UU -\

■ W1=1,W2=0,W3 = 0,W4=0. ■W1=0.5,W2=1,W3=0,W4=0.
□ W1=0,W2 = 0,W3 = 1,W4=0. □W1=0,W2 = 0,W3 = 0,W4 = 1.

Figure 6 average performance with single attribute set

In Figure 7, column charts shows the average performance
with multiple attribute sets.

91.00

■ W1 =1 ,W2=1 ,W3=0,W4=0.
D W1 =0.5,W2=1 ,W3=0,W4=1 .

■ W1 =.1 2,W2=0,W3=1 ,W4=0.

■ W1 =0.5,W2=1 ,W3=0,W4=0.
nW1=1,W2=1,W3=1,W4=1.

Figure 7 Average performances with multiple attribute set

Complicated Face recognition techniques constantly facing
very challenging and difficult problems in several research
works. In spite of the great work done in the last 30 years, it is
sure that the face recognition research community will have to
work for at least the next few decades to completely solve the
problem. Strong and coordinated efforts between the computer
visions signal processing and psychophysics and neurosciences
community is needed. The proposed Multiple Weighted
Feature Attribute Sets based training provided significant
improvement in terms of performance accuracy of the face
recognition system. With ORL data set, a significant 5%
performance improvement was observed during various tests.
In this work, we have selected PCA as the main feature
extraction technique. The weights of the used feature sets were
decided based on trial and error method. This will be applicable
for systems with predefined data sets. In future works, one may
explore different techniques like Kernel PCA, LDA for better
performance. So, future works may address methods for
automatic estimation of the weights of the feature sets with
respect to the application.

References

[1] S.Sakthivel, Dr.R.Lakshmipathi and M.A.Manikandan "Evaluation of
Feature Extraction and Dimensionality Reduction Algorithms for Face
Recognition using ORL Database", The Paper published in Proceedings
of The 2009 International Conference on Image Processing, Computer
Vision, and Pattern Recognition (IPCV'09) , Las Vegas, USA, 2009.
ISBN: 1-60132-117-1, 1-60132-118-X (1-60132-119-8) Copyright 2009
CSREA Press.

[2] S.Sakthivel, Dr.R.Lakshmipathi and M.A.Manikandan "Improving the
performance of machine learning based face recognition algorithm with
Multiple Weighted Facial Attribute Sets" The paper published in
proceedings of The Second International Conference on the Applications
of Digital Information and Web Technologies, 2009 Volume, Issue, 4-6
Aug. 2009 Page(s):658 - 663, ISBN: 978-1-4244-4456-4, INSPEC
Accession Number: 10905880, Digital Object Identifier:
10.1 109/ICADIWT.2009. 5273884

[3] Heng Fui Liau, Kah Phooi Seng, Li-Minn Ang and Siew Wen Chin,
"New Parallel Models for Face Recognition" University of Nottingham
Malaysia Campus, Malaysia - Recent Advances in Face Recognition,
Published by In-Teh, 2008 Recent Advances in Face Recognition, Book
edited by: Kresimir Delac, Mislav Grgic and Marian Stewart Bartlett,
ISBN 978-953-7619-34-3, pp. 236, December 2008, I-Tech, Vienna,
Austria.

[4] Khalid Youssef and Peng-Yung Woo, "A Novel Approach to Using
Color Information in Improving Face Recognition Systems Based on
Multi-Layer Neural Networks", Northern Illinois University USA -
Recent Advances in Face Recognition, Published by In-Teh, 2008,
Source: Recent Advances in Face Recognition, Book edited by: Kresimir
Delac, Mislav Grgic and Marian Stewart Bartlett, ISBN 978-953-7619-
34-3, pp. 236, December 2008, I-Tech, Vienna, Austria

[5] Xiaofei He; Deng Cai; Shuicheng Yan; Hong-Jiang
Zhang, "Neighborhood preserving embedding". Tenth IEEE International
Conference on Computer Vision, Volume 2, Issue, 17-21 Oct. 2005.
Proceedings of the Tenth IEEE International Conference on Computer
Vision (ICCV'05) 1550-5499/05

[6] Shuicheng Yan; Dong Xu; Benyu Zhang; Hong-Jiang Zhang, "Graph
embedding: a general framework for dimensionality reduction", IEEE
Computer Society Conference on Computer Vision and Pattern
Recognition, Volume 2, June 2005. Proceedings of the 2005 IEEE
Computer Society Conference on Computer Vision and Pattern
Recognition (CVPR'05) 1063-6919/05.

124

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

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

Vol 8, No. 3, 2010

[7] S. Canu, Y. Grandvalet, V. Guigue, and A. Rakotomamonjy. (2005).
SVM and kernel methods Matlab toolbox Perception Systems at
Information http://asi.insa-uen.fr/~arakotom/toolbox/index.html

[8] Xiaofei He, Partha Niyogi, "Locahty Preserving Projections (LPP)",
Computer Science Department, The University of Chicago, In Advances
in Neural Information Processing Systems, Volume 16, page 37,
Cambridge, MA, USA 2004, The MIT press.
http://books.nips.cc/papers/Files/nipsl6/NIPS2003_AA20.pdf

[9] Lu, J.; Plataniotis, K.N. & Venetsanopoulos, A. N. (2003) Face
Recognition Using LDA based Algorithm", IEEE trans. Neural Network,
vol.14. No 1, pp.195-199, January 2003. Digital Object Identifier
10.1 109/ TNN.2002.806647.

[10] Yu, Hu. & Yang, J. (2001) A Direct LDA algorithm for high-dimension
data with application to face recognition. Pattern Recognition, vol.34,
pp. 2067-2070, 2001. DOI: 10.1016/S0031-3203(00)00162-X

[11] Chen, L.F.; Mark Liao, H.Y.; Ko, M.T.; Lin, J.C. & Yu, G.J. (2000) A
new LDA-based face recognition system which can solve the small
space size problem. Pattern Recognition, vol.33, pp. 1703-1726, 2000.
DOL 10.1016/ S0031-3203(99)00139-9

[12] H.A. Rowley and T. Kanade, "Neural network -based face detection",
IEEE Trans. on PAMI, vol. 20, no. 1, pp. 23-

38,Janl998. http://www.informedia.cs. cmu.edu/documents/ rowley-

ieee.pdf

[13] E. Osuna, R. Freund, and F. Girosi, "Training support vector machines:
an application to face detection," in Proc. IEEE Conf. Computer Vision
Pattern Recognition, 1997.

http://doi.ieeecomputersocietv.org/10.1109/CVPR.1997.609310

[14] Imola K. Fodor, Center for Applied Scientific Computing, Lawrence
Livermore National Laboratory "A survey of dimension reduction
techniques" June2002.
https://computation.llnl.gov/casc/sapphire/pubs/148494.pdf

[15] Turk, M. & Pentland, A. (1991) Eigenfaces for recognition. Journal of
Cognitive Neuroscience, vol. 3, no. 1, pp. 71-86, Mar 1991.
http://portal.acm.org/ citation.cfm?id=1326894

[16] Kirby, M. & Sirovich, L. (1990) Application of the Karhunen-Loeve
procedure of the characteristic of human faces, IEEE Trans. Pattern
Anal. Machine Intell, vol.12, pp 103-108, Jan, 1990. DOI
10.1109/34.41390

[17] Shang-Hung Lin, "An Introduction to Face Recognition Technology",
Information science special issue on multimedia informing Technologies
- part2 volume 3 No 1, 2000. http://inform.nu/Articles/Vol3/v3nlp01-
07.pdf

AUTHORS PROFILE

^ S. Sakthivel received his M .E Computer science from Sona College of
Technology, Affiliated to Anna University, Chennai, India in the year
2004. Currently, pursuing his Ph.D., in Anna University,
Chennai,Tamilnadu.He has a work experience of 10 years. At present
working as a Assistant Professorin the department of Information Technology.
He has published paper in international journal.He has participated and
presented research papers in various national and international seminars and
conferences. He is an Life member of ISTE.

^ Dr.R.Lakshimpathi received the B.E degree in 1971 and M.E degree in
1973 from College of Engineering, Guindy, and Chennai.He received his PhD
degree in High Voltage Engineering from Indian Institute of Technology,
Chennai, India. He has 36 years of teaching experience in various Government
Engineering Colleges in Tamilnadu and he retired as Principal and Regional
Research Director at Alagappa Chettiar College of Engineering and
Technology, Karaikudi.He is now working as Professor in Electrical and
Electronics Engineering department at St.Peters Engineering College,
Chennai.His areas of research include HVDC Transmission, Power System
Stability and Electrical Power Semiconductor Drives.

1 25 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

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

Vol 8, No. 3, 2010

A DNA and Amino Acids-Based
Implementation of Playfair Cipher

Mona Sabry(i), Mohamed Hashem(2), Taymoor Nazmyd

Mohamed Essam Khalifa(3)

Faculty of Computer Science and information systems,

Ain Shams University,

Cairo, Egypt.

E-mail: mona. sabrv@ hotmail.com

ABSTRACT— The DNA cryptography is a new and very promising
direction in cryptography research. Although in its primitive stage,
DNA cryptography is shown to be very effective. Currently, several
DNA computing algorithms are proposed for quite some
cryptography, cryptanalysis and steganography problems, and they are
very powerful in these areas.

This paper discusses a significant modification to the old Playfair
cipher by introducing DNA-based and amino acids-based structure to
the core of the ciphering process.

In this study, a binary form of data, such as plaintext messages, or
images are transformed into sequences of DNA nucleotides.
Subsequently, these nucleotides pass through a Playfair encryption
process based on amino-acids structure.

The fundamental idea behind this encryption technique is to enforce
other conventional cryptographic algorithms which proved to be
broken, and also to open the door for applying the DNA and Amino
Acids concepts to more conventional cryptographic algorithms to
enhance their security features.

KEY WORDS: DNA, amino acids, encryption, decryption,
cryptography, security, Playfair cipher.

I. INTRODUCTION

As some of the modern cryptography algorithms (such as DES,
and more recently, MD5) are broken, the new directions of
information security are being sought to protect the data. The
concept of using DNA computing in the fields of cryptography
and steganography is a possible technology that may bring
forward a new hope for powerful, or even unbreakable,
algorithms.

The main purpose behind our work is to discover new fields of
encoding the data in addition to the conventional used
encryption algorithm in order to increase the concept of
confusion and therefore increase security.

In our work, we applied the conversion of character form or
binary form of data to the DNA form and then to amino acid
form. Then the resulting form goes through the encryption
algorithm which we chose for example; the classical Playfair
cipher.

It is Adleman, with his pioneering work [5]; set the stage for
the new field of bio- computing research. His main idea was to
use actual chemistry to solve problems that are either
unsolvable by conventional computers, or require an enormous
amount of computation. By the use of DNA computing, the
Data Encryption Standard (DES) cryptographic protocol can be
broken [6]. In DNA steganography, A DNA encoded message
is first camouflaged within the enormous complexity of human
genomic DNA and then further concealed by confining this
sample to a microdot[3]. Recent research considers the use of
the Human genome in cryptography. In 2000, the Junior Nobel
Prize was awarded to a young Romanian American student,
Viviana Risca, for her work in DNA steganography. [3]

The one-time pad cryptography with DNA strands, and the
research on DNA steganography (hiding messages in DNA),
are shown in [2] and [3].

However, researchers in DNA cryptography are still looking at
much more theory than practicality. The constraints of its high
tech lab requirements and computational limitations, combined
with the labor intensive extrapolation means. Thus prevent
DNA computing fi-om being of efficient use in today's security
world.

Another approach is lead by Ning Kang in which he did not use
real DNA computing, but just used the principle ideas in
central dogma of molecular biology to develop his
cryptography method. The method only simulates the
transcription, splicing, and translation process of the central
dogma; thus, it is a pseudo DNA cryptography method. [4]

There is another investigation conducted by [1] which is based
on a conventional symmetric encryption algorithm called "Yet
Another Encryption Algorithm" (YAEA) developed by Saeb
and Baith [1]. In this study, he introduces the concept of using
DNA computing in the fields of cryptography in order to
enhance the security of cryptographic algorithms. This is
considered a pioneering idea that stood behind our work in this
paper. [1]

(1) Computer Science Dept.

(2) Information Systems Dept.
126

(3) Basic Science Dept.

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

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

Vol 8, No. 3, 2010

Although Playfair cipher is believed to be an old, simple and an
easily breakable cipher, we believe our new modifications can
make it a more powerful encryption algorithm. This is done by
introducing concepts of confusion and diffusion to the core of
the encryption process in addition to preserving the cipher's
simplicity concept.

In addition shortage in security features the plaintext message
is restricted to be all upper case, without J letters, without
punctuation, or even numerical values. Those problems can be
easily handled in any modern cipher as handled in our new
algorithm [8].

The character form of a message or any form of an image can
be easily transformed to the form of bits. This binary form can
be transformed to DNA form through many encoding
techniques implemented in previous work and summarized in
[7].

Playfair is based on the English alphabetical letters, so
preserving this concept, we will use the English alphabet but
from an indirect way. DNA contains four bases that can be
given an abbreviation of only four letters (adenine (A),
cytosine (C), guanine (G) and thymine (T)). On the other side,
we have 20 amino acids with additional 3 codons to represent
the Stop of coding region. Each amino acid is abbreviated by a
single English character. So we are able to stretch these 20
characters to 26 characters, we will be able to represent the
English alphabet.

Then, we have to convert the DNA form of data to amino acid
form so that it can go through a classical Playfair cipher.
Through this conversion process, we have to keep in mind the
problem of ambiguity; that most amino acids are given more
than possible codon.

The rest of the paper is organized as follows: section two will
give a brief explanation of what is DNA and process of
Transcription and translation. Section three introduces our new
algorithm followed by a detailed explanation of the
encryption/decryption processes. Section four shows the
experiment steps, results. Section five introduces some
additional security features. Section six shows conclusion and
future work.

II. OVERVIEW OF DNA

A. What is Deoxyribonucleic acid ^DNA '?

DNA is a nucleic acid that contains the genetic instructions
used in the development and functioning of all known living
organisms and some viruses. The main role of DNA molecules
is the long-term storage of information. DNA is often
compared to a set of blueprints or a recipe, or a code, since it
contains the instructions needed to construct other components

of cells, such as proteins and RNA molecules. The DNA
segments that carry this genetic information are called genes,
but other DNA sequences have structural purposes, or are
involved in regulating the use of this genetic information.

The DNA double helix is stabilized by hydrogen bonds
between the bases attached to the two strands. The four bases
found in DNA are adenine (abbreviated A), cytosine (C),
guanine (G) and thymine (T). These four bases are attached to
the sugar/phosphate to form the complete nucleotide, as shown
for adenosine monophosphate.

B, The genetic code

The genetic code consists of 64 triplets of nucleotides. These
triplets are called codons. With three exceptions, each codon
encodes for one of the 20 amino acids used in the synthesis of
proteins. That produces some redundancy in the code: most of
the amino acids being encoded by more than one codon.

The genetic code can be expressed as either RNA codons or
DNA codons. RNA codons occur in messenger RNA (mRNA)
and are the codons that are actually "read" during the synthesis
of polypeptides (the process called translation). But each
mRNA molecule acquires its sequence of nucleotides by
transcription from the corresponding gene.

The DNA Codons is read the same as the RNA codons Except
that the nucleotide thymidine (T) is found in place of uridine
(U). So in DNA codons we have (TCAG) and in RNA codons,
we have (UCTG).

C Transcription and translation

A gene is a sequence of DNA that contains genetic information
and can influence the phenotype of an organism. Within a
gene, the sequence of bases along a DNA strand defines a
messenger RNA sequence, which then defines one or more
protein sequences. The relationship between the nucleotide
sequences of genes and the amino-acid sequences of proteins is
determined by the rules of translation, known collectively as
the genetic code. The genetic code consists of three-letter
'words' called codons formed from a sequence of three
nucleotides (e.g. ACT, CAG, TTT).

In transcription, the codons of a gene are copied into
messenger RNA by RNA polymerase. This RNA copy is then
decoded by a ribosome that reads the RNA sequence by base-
pairing the messenger RNA to transfer RNA, which carries
amino acids. Since there are 4 bases in 3 -letter combinations,
there are 64 possible codons (4^ combinations). These encode
the twenty standard amino acids, giving most amino acids more
than one possible codon. There are also three 'stop' or
'nonsense' codons signifying the end of the coding region; these

127

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

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

Vol 8, No. 3, 2010

are the TAA, TGA and TAG codons.RNA codon table,
WIKIPEDIA:

http://en.wikipedia.org/wiki/Genetic_code#cite_note-
pmidl 9056476-8

III. DNA-B ASED PLAYFAIR ALGORITHM

A, Encryption algorithm ofDNA-hased Play fair cipher:

Playfair used to be applied to English alphabet characters of
plaintext. It was unable to encode any special characters or
numbers which is considered a severe drawback that enforces
the sender to write everything in the English letters. This
problem appears while sending numerical data, equations or
symbols.

On the contrary, in our algorithm, we can use any numbers,
special characters or even spaces (not preferred) in or plaintext.
The encryption process starts by the binary form of data
(message or image) which is transferred to DNA form
according to Table 1 . Then the DNA form is transferred to the
Amino acids form according to Table 2 which is a standard
universal table of Amino acids and their codons representation
in the form of DNA [RNA codon table, Wikipedia:
http://en.wikipedia.org/wiki/Genetic_code#cite_note-
pmidl 9056476-8 1.

Note that each amino acid has a name, abbreviation, and a
single character symbol. This character symbol is what we will
use in our algorithm.

Table I: DNA Representation of bits.

Bit 1

Bit 2

DNA

y^ Plaintext

INPUT

/ Secret Key

Preprocessing

Convert To Binary

Convert To DNA

Convert To Amino
Acids

Playfair

Convert To DNA

OUTPUT

Fig. 1: flowchart of the DNA-based Playfair algorithm

B, Constructing the alphabet table:

In the table, we have only 20 amino acids in addition to 1 start
and 1 stop. While we need 25 letters to construct the Playfair
matrix (note that I/J are assigned to one cell).

The letters we need to fill are (B, O, U, X, Z). So we will make
these characters share some amino acids their codons.
The start codon is repeated with amino acid (M) so we will not
use it. We will assign to (B) the 3 stop codons. We have 3
amino acids (L, R, S) having 6 codons. By noticing the
sequence of DNA of each, we can figure out that each has 4
codons of the same type and 2 of another type. Those 2 of the
other type are shifted to the letters (O, U, X) respectively.
Letter (Z) will take one codon from (Y), so that Y: UAU, Z:
UAC. Now the new distribution of codons is illustrated in
Table 3.

Counting the number of codons of each character, we will find
the number varies between 1 and 4 codons per character. We
will call this number 'Ambiguity' of the character [AMBIG].

Now we have the distribution of the complete English alphabet,
so a message in the form of Amino Acids can go through
traditional Playfair cipher process using the secret key.

128

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

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

Vol 8, No. 3, 2010

Table II: Amino acids and their 64 codons

Ala/A

GCU, GCC,
GCA, GCG

Leu/L

UUA, UUG,
GUU, GUG,
GUA, GUG

Arg/R

CGU,CGC,
CGA,CGG,
AGA,AGG

Lys/K

AAA, AAG

Asn/N

AAU, AAC

Met/M

AUG

Asp/D

GAU, GAG

Phe/F

UUU, UUG

Cys/C

UGU, UGC

Pro/P

GGU,GGG,
GGA, GGG

Gln/Q

CAA, GAG

Ser/S

UGU, UGG,
UGA, UGG,
AGU, AGG

Glu/E

GAA, GAG

Thr/T

AGU, AGG,
AGA, AGG

Gly/G

GGU,GGC,
GGA,GGG

Trp/W

UGG

His/H

CAU, GAG

TyrA^

UAU, UAG

Ile/I

AUU, AUG,
AUA

VaW

GUU, GUG,
GUA, GUG

START

AUG

STOP

UAA, UGA,
UAG

Table III: New distribution of the alphabet with the corresponding new codons:

STOP

from

. ■ K

GCU

UAA

UGU

GAU

GAA

UUU

GGU

CAU

AUU

AAA

UUA

AUG

AAU

UUA

ecu

CAA

CGU

UCU

ACU

AGA

GUU

UGG

AGU

UAU

UAC

GCC

UAG

UGC

GAC

GAG

uuc

GGC

CAC

AUC

AAG

UUG

AAC

-ttJjG..

ccc

CAG

CGC

UCC

ACC

_AGG_

GUC

AGC

UAC

GCA

UGA

GGA

AUA

CUU

CCA

CGA

UCA

ACA

GUA

GCG

GGG

cue

CCG

CGG

UCG

ACG

GUG

CUA

AGA

AGU

CUG

AGG

AGC

129

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

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

Vol 8, No. 3, 2010

Table IV: New Distribution for codons on English alphabet

GCU, GCC,
GCA, GCG

<:%%

;%^

W/'.

^m.

AAA, AAG

AAU, AAC

AUG

GAU, GAG

UUU, UUC

UGU, UGC

CCU,CCC,
CCA, GCG

CAA, GAG

$$$$^

;^^

GAA, GAG

AGU, AGG,
AGA, AGG

GGU,GGC,
GGA,GGG

UGG

CAU, CAC

SXXW

SKNXX^

AUU, AUG,
AUA

GUU, GUG,
GUA, GUG

UAA, UGA,
UAG

UUA, UUG

AGA,AGG

AGU, AGC

UAC

of plaintext in DNA form which can be transferred to binary
form and then the final character form.

D, Pseudo-code

Input:

[P] Plaintext (characters with spaces, numbers or any
special characters).

[K] Secret key (English characters without any
number or special characters).

Algorithm body:

Preprocessing:

1- Prepare the secret key:

- Remove any spaces or repeated characters from [K].

- Put the remaining characters in the UPPER case
form. [K]^UPPER[K].

The output form is the amino acid form of cipher text. DNA
form of cipher text can be demonstrated also from Table 4 by
choosing random codons accompanied to each character. The
concept that one character can have more than one DNA
representation is itself an addition to confusion concept that
enhances the algorithm strength. Table\ IV shows the new
distribution of codons on the amino acids and additional
alphabetical English letters according to our algorithm.

C. Decryption and Ambiguity problem

The decryption process is simply the inverse of the encryption
process unless that we will find a problem in constructing the
DNA form of plaintext from the amino acid form which is of
length (L). The problem is that we are unable to choose which
codon to put in accordance to each amino acid character. This
is simply the problem of codon-amino acid mapping problem
arised with other algorithm based on the concept of Central
Dogma like [4]. The way Nang handled this problem is to put
this codon-amino acid mapping in the secret key to be sent
through a secure channel [4] . This idea is not efficient since it
increases the size of the key in relation to size of the plaintext.

The solution in our algorithm is located in two additional bits

for each amino acid character to demonstrate which codon to

choose.

We said before that each amino acid has 1 , 2, 3 or 4 codons to

represent it. This is a number that can be put in 2 bits from

0^3.

These 2 bits can be converted to DNA form from Table 1.

That is why the final cipher text is both the DNA form of
cipher text of length (3L) and the array carrying the ambiguity
of length (L).

In decryption, the amino acid form of plaintext with the
assistance of the ambiguity array can construct the correct form

2- Prepare the plaintext:
- Remove the spaces from [P] (done to avoid
attacker's trace to a character which is repeated many
times within the message)

Processing:

1- Binary form [BP] = BINARY [P] (Replace each
character by its binary representation-8 bits-)

2- DNA form [DP] = DNA [BP] (Replace each 2 bits
by their DNA representation)

3- Amino acids form [AP] = AMINO [DP] (Replace
each 3 DNA characters by their Amino acid character
keeping in track the ambiguity of each Amino acid
[AMBIG].

4- Construct the Playfair 5X5 matrix and add [K] row
by row, then add the rest of alphabet characters.

5- Amino acid of cipher text [AC]= PLAYFAIR [AP].

6- DNA form of cipher text [DC] = DNA [AC].

Output:

Add [DC] and [AMBIG] together in the suitable
form^ final cipher text [C].

130

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

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

Vol 8, No. 3, 2010

E, Samples of the program steps and output

Ambigaih'

,0 Embedded Ambigid^
O After Atnbiguitj

Alphabet Distmbatioa ■

©|Emj|§h
O Protem

Key EGYPT VICTORY

Plaintext

Binary

DNA

Proteins

After Playfair

Ambiguity

DNA

attack starts at 2:00 PM.

RPHSRPIGLSHDLAHYRPQBMDNOITNIXACE

OVAZOVCEKUAFKBDVOVLHWMXFCOKOSFIG

2*122310 20112 01020010*1121110010

UU A G GUU A GCU C U AC G UU A G GUU U UGU C G AA A AAA G AG A A GCU C UUU C AAA G U AA A G AU C GUU A UU A G GUU A CUU A C AU

Ciphertest c ugg a aug a agu c uuu c ugu g uua c aaa c uua c ucu a uuu a auu c ggu a

Figure 2: sample of steps of encryption implementation

- Ambignity

0: Embedded Ambigiity ;
O After AmbigBity

Alphabet Distrabntioa

Englisi
jO Protein

r Alpnaoet

!0 English

Key EGYTT VICTORY

Ciphertest

DNA

Ambiguity

Proteins

After Playfair

DNA

Binaiy

201223102*1120102**100112111001*

OVAZOVCEKUAFKBDVOVLHWMXFCOKOSFIG

RFHSRPIGLSHDLAHYRPQBMDNOITNIXACE

attackstartsat2:0~0PM.

Figure 3: sample of steps of decryption implementation

131

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

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

Vol 8, No. 3, 2010

IV. EXPERIMENT AND PERFORMANCE ANALYSIS

A. Experiment

1- Experiment inputs and attributes

We led our experiment on tlie famous novel 'A Tale of Two
Cities' by Charles Dickens found on

http://www.literature.org/authors/dickens-charles/two-cities/ .
We will take paragraphs from the beginning of the novel
according to the estimated storage size in Kilobytes (from 1
KB and increasing till 150 KB).

2- System Parameters

The experiments are conducted using Intel(R) Core (TM)
2CPU T5300, 1.73 GHz, 32 bit processor with 1GB of
RAM. The simulation program is compiled using the default
settings in .NET 2005 visual studio for C# windows
applications under WINDOWS XP as the operating system.
The experiments will be performed several times to assure that
the results are consistent and valid.

3- Experiment Factors

The chosen factor here to determine the performance is the
algorithm's speed to encrypt data blocks of various sizes.
Suppose we will use the original sequence of English alphabet
and embed the ambiguity inside the message not after it.
The secret key used is "CHARLES DICKENS" which results
in 11 Bytes key.

4- Experiment steps:
Experiment preprocessing:

1- Loading the table of the 64 amino acids with their
DNA Encodings and number of ambiguous encodings.

2- Formatting the secret key by removing spaces,
repeated characters and non English letters.

3- Formatting the plaintext by removing spaces between
words and separating the repeated doubles by the
character '-' which chosen to be a rarely used
character.

Processing:

This includes:

1- Converting characters to binary form.

2- Converting binary to DNA

3- Converting the DNA to amino acids and recording
ambiguity.

4- Do Playfair encryption.

5- Convert the amino acid form of cipher text to DNA
form in addition to embedding the ambiguity in the
DNA format.

5- Experiment Results

The next table illustrates the experiments and time taken to
encrypt each piece of plaintext (each is of different data loads)
in milliseconds.

The time taken by loading the amino acids table and preparing
the secret key is ignored because it is comparatively small to
processing time.

Table 4: performance results of DNA-based Playfair algorithm

Input size of
plaintext
(in KB)

Plaintext after
preprocessing

Preprocess

ing
plaintext

From Binary

to Amino Acids

form

playfair

Prepare
ciphertext

Total

processing

time

Bytes/Second

1 (1,022 B)

846B

15.625

65.408

10 (9,757 B)

8124B

62.500

15.625

125.000

203.125

48.034

20 (20,023 B)

16599B

203.125

15.625

171.875

390.625

51.259

50 (50,432 B)

41781B

1062.500

46.875

15.625

437.500

1562.500

32.276

100 (97,072 B)

83910B

4687.500

78.125

31.25

859.375

5656.250

17.162

150 (153,418 B)

127098 B

11390.625

140.625

31.25

1343.750

12906.25

11.887

132

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

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

Vol 8, No. 3, 2010

V. Additional security features

We have illustrated the main core of the algorithm and now we
are going to suggest some additional features to the algorithm
which can enhance its security and strength.

A- The key:

It is quite clear that the more random and long the key is, the
more the difficulty to break the cipher will be.

B- Use Amino acids alphabet sequence instead of English
alphabetical sequence:

The standard table of amino acids has a special sequence
defined in the matrix [4X4] (UCAG) X (UCAG). This
sequence of acids can be used instead of the sequence of
English alphabet letters to fill the rest of the 5X5 matrix after
adding the secret key.

C- Combine the total resulting message into long strand of
DNA to be inserted in a microdot (steganography):

One of the advantages of this algorithm is the variety of ways
we can use to write down the cipher text. It can be written in
DNA form, binary form or even character form which is more
confusing. The advantage of DNA form is that it can make use
of several steganography techniques developed for DNA
messages [3]. It can also be prepared in biological labs like in
[2] in which DNA message goes through a biological DNA
encryption process using one time pad or substitution.

D- Ambiguity a problem that contains useful confusion
feature:

Some characters in table 2 can have 6 codons representing the
problem of ambiguity. The way we handled the preparation of
table 3 made each character have in maximum 4 codons. The
number 4 can be represented by 2 bits and therefore ban be
represented by one DNA character. That was a benefit that
made us able to write the cipher text with ambiguity in the
form of DNA.

E- Use of conventional XOR-ing procedure:

Another way to increase security is defining another key that
can be XOR-ed with the amino acid form or DNA form of
cipher text. It was a pioneer idea by [1] to choose the key as the
DNA strand of a certain organism. This idea assures the key
randomness and variety in length according to the length of the
message.

VI. CONCLUSION AND FUTURE WORK

The fundamental idea behind this technique is to open the door
for the idea of applying the DNA and Amino Acids encoding
concepts to other conventional cryptographic algorithms to
enhance their security -vulnerability- features.
Our algorithm initially succeeded in overcoming some main
problems in "Playfair cipher" like restriction of plaintext to
"English Alphabet". As in our algorithm the plaintext is to be

converted to its binary value before encryption, it now clear
that the plaintext message can be written in upper or lower
case, with any punctuation, and numerical values.

Other papers conducted the idea of amino acids way of
representation from the point of view of the central dogma
design [4]. But they were unable to clearly handle the problem
of ambiguity as performed by our algorithm. Our algorithm
made few preprocessing steps to handle this problem and the
result was quite accurate (same input message obtained after
decryption). This feature is very important when regarding an
encryption algorithm in order to verify the concept of data
integrity or in other words, to assure that data after decryption
to be the same input data before encryption.
Finally, our algorithm provides different forms of the cipher
text like: Binary form, DNA form, Amino Acid form or
character form. Those various forms can match different used
applications.

Our future work is dedicated to implementing this encoding on
other known algorithms and measuring its performance and
security. Also, Experiments should be conducted to implement
the algorithm on different applications to ensure its feasibility
and applicability.

REFERENCES

[1] Sherif T. Amin, Magdy Saeb, Salah El-Gindi, "A DNA-based
Implementation of YAEA Encryption Algorithm," lASTED
International Conference on Computational Intelligence (CI 2006), San
Francisco, Nov. 20, 2006.
http://www.actapress.com/PaperInfo.aspx?PaperID=29058 .

[2] Ashish Gehani, Thomas LaBean and John Reif. DNA-Based
Cryptography. DIMACS DNA Based Computers V, American
Mathematical Society, 2000.

[3] TAYLOR Clelland Catherine, Viviana Risca, Carter Bancroft, 1999,

"Hiding Messages in DNA Microdots". Nature Magazine Vol.. 399, June
10, 1999.

[4] KANG Ning, "A Pseudo DNA Cryptography Method", Independent
Research Study Project for CS5231, October 2004.

[5] Leonard Adleman. "Molecular Computation of Solutions to

Combinatorial Problems". Science, 266:1021-1024, November 1994.

[6] Dan Boneh, Cristopher Dunworth, and Richard Lipton. "Breaking DES
Using a Molecular Computer". Technical Report CS-TR-489-95,
Department of Computer Science, Princeton University, USA, 1995.

[7] Dominik Heider and Angelika Barnekow, "DNA-based watermarks
using the DNA-Crypt algorithm", Pubhshed: 29 May 2007 BMC
Bioinformatics 2007, 8:176 doi:10.1 186/1471-2105-8-176,
http://www.biomedcentral.eom/l 47 1 -2 1 05/8/1 76 ,© 2007 Heider and
Barnekow; licensee BioMed Central Ltd.

[8] William Stallings. "Cryptography and Network Security", Third
Edition, Prentice Hall International, 2003.

133

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

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

Vol. 8, No. 3, 2010

Ultra Wideband Slot Antenna with Reconfigurable

Notch bands

J. William* and R.Nakkeeran

Department of Electronics and Communication Engineering

Pondicherry Engineering College

Puducherry, India . 605014.

Email id: wills.susan@gmail.com, rnakeeran@pec.edu

^Corresponding author: wills.susan@gmail.com

Abstract — An Ultra Wideband (UWB) slot antenna with
reconfigurable notch bands is presented in this paper. The basic
UWB antenna consists of a rectangular slot with triangular
structure that acts as a tuning stub with CPW feed. The CPW
feed is designed for 50 il impedance. The notch band is achieved
by inserting a rectangular slot in the ground plane with a
effective length of >^/2. The reconfigurable rejection of the bands
3.1 GHz - 3.9 GHz, 4 GHz-5.3 GHz and 4.1 GHz-5.9 GHz are
achieved by switching the diodes placed over the slot in the
ground plane. The characteristics of the designed structure are
investigated by using MoM based electromagnetic solver, IE3D.
The return loss (Sn) of the antenna is measured and that are
comparable with the simulation results. The proposed antenna
covers the entire UWB range 3.1 GHz to 10.6 GHz with
reconfiguration. The low profile and simple configuration of the
proposed antenna leads to easy fabrication that may be built in
any wireless UWB device applications where reconfigurable
rejection bands are required. The rejection of WiMax, IEEE
802.11a and HYPERLAN/2 bands can be achieved by using the
proposed antenna design.

Keywords- coplanar waveguide; notch band; slot antenna;
reconfiguration; ultra wideband

I. Introduction

In the year 2002, Federal Communications Commission
(FCC) released the unhcensed UWB spectrum 3.1 GHz to 10.6
GHz for the commercial purposes. After the release of UWB, it
gains much attention by the researchers due to its inherent
properties of low power consumption, high data rate and
simple configuration [1]. With the rapid developments of such
UWB systems, a lot of attention is being given for designing
the UWB antennas. Designing an antenna to operate in the
UWB band is quiet challenging one because it has to satisfy the
requirements such as ultra wide impedance bandwidth, omni
directional radiation pattern, constant gain, high radiation
efficiency, constant group delay, low profile, easy
manufacturing etc [2]. Interestingly the planar slot antennas
with CPW fed posses the above said features with simple
structure, less radiation loss, less dispersion and easy
integration of monolithic microwave integrated circuits
(MMIC) [3]. Due to inherent nature of UWB system sharing
the same spectrum with other systems. The interference from
other system should be considered when you design a UWB

system. For example, the other wireless communication
systems such as WiMax (3.3 GHz - 3.7 GHz), IEEE 802.11a
(5.15 GHz - 5.35 GHz and 5.725 GHz - 5.825 GHz) and
HYPERLAN/2 (5.15 GHz - 5.35 GHz and 5.47 GHz -
5.825 GHz) which are also operated in the portion of UWB
band. Therefore, the UWB antenna design could play an
important role when interference with other wireless
applications such as WiMax, IEEE 802.11a and HYPERLAN/2
systems which are also coexisting with UWB band (3.1 GHz -
10.6 GHz), which degrades the UWB system performance.
Therefore, a UWB antenna having reconfigurable frequency
band notch characteristics is enviable.

To mitigate the above electromagnetic interference from
nearby communication systems many antenna designs have
been proposed in the literature [4-13]. Many techniques also
used to introduce notch band for rejecting the interference in
the UWB slot antennas. It is done either by inserting half
wavelength slits, stripes in the tuning stub [14], or inserting
stub in the aperture connected to the ground planes [15], or
inserting square ring resonator in the tuning stub [16], or
inserting 'L' branches in the ground plane [17], or with
complementary split ring resonator [18], or inserting strip in the
slot [19]. All the above methods are used for rejecting a fixed
band of frequencies. But to effectively utilize the UWB
spectrum and to improve the performance of the UWB system,
it is desirable to design the UWB antenna with reconfigurable
notch band. It will help to minimize the interference between
the systems and to improve the performance of the UWB
systems. In general, reconfiguration are popular in antenna
engineering for their frequency agility, bandwidth enhancement
and polarization diversity [20-23]. In [24] RF MEMS switches
are used for the reconfigurabilty of rejection band between the
frequencies 5 GHz to 6 GHz. In this paper a new method is
proposed to obtain the reconfiguration in the frequency notch
bands for WiMax, WLAN 802.1 la and HYPERLAN/2 systems
by making 'on' and 'off the diodes placed over the slot that is
introduced on the ground plane. This intern changes the tuning
length of the slot, which is responsible for the desired
frequency notch band. Through our simulation it is also found
that the desired frequency notch band can be obtained by
inserting slot in the tuning stub with appropriate length rather
than on the ground plane. However this paper mainly focuses

134

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

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

on reconfigurable notch band through the introduction of slot in
the ground plane. The simulation software used for this
analysis is IE3D [25]. The paper is organized as follows:
Section II brings out the geometry of the antenna. In section
III, simulation results and analysis are presented. Obtained
experimental results are given in Section IV. Section IV
concludes the work.

2.4 mm

SubsTiat* li-1.6niiii,^j-^.4

Fig. 1 Geometry of the proposed slot antenna with biasing circuit.

II. Antenna Geometry

The structure of the antenna is shown in Fig 1 . The antenna
consists of rectangular slot with width 'Wi' and length 'Li'.
The tuning stub comprises a triangular patch with height 'H'.
The distance between the tuning stub and feed line is 'd', 'W
and 'L' are the overall width and length of the antenna
respectively. In this study, the dielectric substance (FR4) with
thickness of 1.6 mm with relative permittivity of 4.4 is chosen
as substrate to facilitate printed circuit board integration. The
CPW feed is designed for 50 ^2 characteristic impedance with
fixed 2.4 mm feed line width and 0.5 mm ground gap. Slits are
introduced to avoid the short circuit between the ground
planes. Fast switching diode LL4148 is used as a switching
device. The proposed antenna is designed to cover the entire
UWB band with reconfiguration capability. The placement of
the diodes 'L^' are desired by the effective wavelength ^^eff?
The effective wavelength of the slot is.

S+\

'eff

(1)

The placement of the diodes 'Lj'

Vol. 8, No. 3, 2010

L. =

(2)

where 'fn' is the centre frequency of the notch band. The
placement of the diodes is desired by the effective wavelength
for the different notch frequencies.

III. Simulated Results and Analysis

The analysis and performance of the proposed antenna is
explored by using IE3D for the better impedance matching.
The detailed parametric analysis of the UWB antenna is
carried out and presented in our paper [26]. In order to
evaluate the performance of the proposed antenna with
reconfigurable slot in the ground plane, the optimal parameter
values of the antenna (without slot) suggested in that paper are
considered. The final optimal parameter values of the antenna
are listed in the Table 1 . However to study the impact of the
slot, the slot length 'L2' and width 'W2' are varied by keeping
one of them as constant. In the simulation switching diodes
were simulated as capacitor for the 'Off state and as a resistor
in the 'On' state. The current distribution, gain and group
delay are also studied. The simulated return loss, with
different states of the diode are shown in Fig. 2, clearly
indicate that the notch frequency and notch bandwidth is
varied when the slot length is varied.

TABLE I.

Optimal Parameter Values of the Antenna

Parameter

Description

Optimal
Value

Length of the antenna

28 mm

Width of the antenna

21 mm

Length of the slot

15 mm

Width of the slot

16.8 mm

Feed gap distance

1.6 mm

Height of the patch

9.3 mm

3 4 5 G 7 8 9 10 11 12
Frequency in GHz

Fig. 2 Simulated response curves for different biasing condition.

135

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

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

Vol 8, No. 3, 2010

A. Effect of Slot in the Ground Plane

The length of the ground slot is adjusted by switching the
diodes placed appropriately over the slot. The switching
diodes 'Di' and 'D2' are placed over the slot along the ground
plane. The Table 2 shows the states of the diode switches and
notch frequency with notch band width.

TABLE 11.

Description of Switching of the Diodes

Notch Freq. (GHz)

Notch band

Off

3.6 GHz

3.1-3.9 GHz

Off

5.1GHz

4 -5.3 GHz

Off

5.6 GHz

4.1-5.9 GHz

When both diodes are in open state the effective slot
length is longer to achieve the notch frequency at 3.6 GHz
with frequency notch band of 3.1 GHz to 3.9 GHz, which is
the WIMAX band of frequency. When Di is 'Off and D2 is
'On' the length of the slot is tuned to eliminate the band of
4 GHz to 5.3 GHz with notch center frequency of 5.1 GHz
which covers the lower band of IEEE 802.11a. When Di is
'On' and D2 is 'Off the effective wavelength of the slot is
adjusted such that to eliminate the upper band of IEEE
802.11a with a notch frequency of 5.6 GHz with the notch
frequency band of 4.1 GHz to 5.9 GHz.

B. Simulated Current Distribuition

The simulated current distribution at different notch
frequencies according to the switching condition of the diodes
is presented in Fig.3, it is witnessed from the figure, the current
distribution around the radiation slot is disturbed by the
introduction of ground slot, which is responsible for the notch
in the frequency band. If the slot length is longer the destructive
interference with the main radiating slot is high which causes
the notching in that band is good. When the ground slot length
is decreased, the current distribution of the radiating slot is
disturbed by the tuning slot is less effective compared with the
longer slot size.

(a) (b) (c)

Fig. 3 Simulated current distribution for different switching conditions of
diodes, a) 'Di and 'D2' 'Off b) 'Di' 'On' and 'D2' 'Off c) 'Di' 'Off and
'D2' 'On'

C Gain

The computed gain of the proposed UWB antenna for
different tuning lengths of the slot in the ground plane and
without slot is compared in Figure 4.

■ without slot in the ground pia
i --- D1,D2=off

-j- ...... Di= on & 02= off -; ;-

; D1=off&D2=on

J \ \ \ \ \ L

3 4 5 6 7 8 9 10 11 12

Frequency in GHz

Fig. 4 Simulated gain for different biasing conditions of diodes.

It is observed that there is a gain variation at the notched
frequencies when the length of the slot is tuned, which is
-7.5 dBi for both diodes are 'off state and 0.4 dBi when Di is
'Off and D2 is 'On'. Where as it is 0.6 dBi for Di is 'On' and
D2 is 'Off. The antenna without slot and the gain varies from
2 dBi to 6 dBi across the UWB spectrum.

D. Radiation pattern

The radiation pattern for the E plane and H plane at
frequencies 3.4 GHz, 5.2 GHz and 5.6 GHz are simulated for
all the three cases of diode states are compared and displayed
in Fig.5, which disclose that the directivity gain of the
radiation pattern is reduced at the notch frequencies without
affecting the shape of the radiation pattern. In the E plane, it is
bidirectional pattern and in H plane, it is omni directional
pattern.

D1 off D2off
D1 off D2 on
D1 on D2 off

Dioff D2off
DI off D2 on
DI on D2 off

(b) 5.2 GHz

136

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

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

Vol 8, No. 3, 2010

D1 ofrD2ofr

--D1 ofrD2on
^ D1onD2off

D1off D2off
---D1 off D2on
—^ D1 on D2 off

E. Group delay

The group delay 'x' of the antenna is calculated from the
phase of the computed 'S21' by using the following equation
and plotted in Fig. 6,

x^-^ (2)

where ' ^ ' is phase of S21 in radians /sec and T is frequency
in GHz. From the Fig. 6, it is noticed that the variation in the
group delay for the antenna with and without slot is around 2
ns for the frequency range from 3.1 GHz to 10.6 GHz. There
is a variation in the group delay at the notch band in the
response which is due to notch behavior of the antenna.

S-2f

«4

-10

1 1

[ i i L...li

^v^^fH^^

i i

] 1 1 p 1

i i i i i

\ I

D1 off D2 on

i i

— — -D1 on D2 off

i i i i i

G 7 8 9 10

Frequency in GHz

Fig. 6 Group delay response.

H 12

IV. EXPERIMENTALL RESULTS AND DISCUSSION

The prototype of the proposed antenna shown in Fig. 1 was
fabricated for different parameters with their optimal values
and tested. Using Hewlett Packard Network Analyzer
(HP8757D), the VSWR is measured and compared with the
simulation result is shown in Fig.7. There is a discrepancy
between the measured and the simulated ones might be the
effect of soldering the SMA connector or fabrication
tolerance. The simulation result was obtained by assuming
coplanar as input port, whereas practically SMA connector
was used, the imperfect transition between SMA feed to
coplanar may introduce losses [27] and also the capacitances
can lead to shift in the frequency.

£
(A

-25

-30

D1 off D2 off-
D1 off D2 on '
D1 on D2 off -

- Measured
-Measured
-Measured

- Simulated
-Simulated
-Simulated

4 5 S r 8 9 10 11

Frequency in GHz

Fig. 8 Comparison of measured and simulated
VSWR.

V. Time Domain Analysis

In ultra wideband systems, the information is transmitted
using short pulses. Hence, it is important to study the temporal
behavior of the transmitted pulse. The communication system
for UWB pulse transmission must provide as minimum as
possible distortion, spreading and disturbance. The channel is
assumed to be linear time invariant (LTI) system to verify the
capability of the proposed antenna for transmission and
reception of these narrow pulses. The transfer function of the
entire system is computed using simulated value of 'S21'
parameter [28]. The received output pulse is obtained by taking
the Inverse Fourier Transform (IFT) of the product of transfer
function and spectrum of the test input pulse. While computing
'S21', two identical antennas are placed face to face at a
distance of 75 cm that is greater than the far-field distance of
the antenna. The cosine modulated Gaussian pulse is
considered for this analysis with centre frequency of 6.85 GHz
and pulse width of 220 picoseconds, whose spectrum is shown
in Fig.7. It satisfies the requirement of FCC mask for UWB
indoor emission. The comparison of input and output responses
of the system for the antenna with different notches are shown
in Fig. 8, which ensures the distortion less pulse transmission
and also guarantees that the designed antenna is capable of
transmitting and receiving short pulses.

9 1

2 4 G a 10 12 ' _. .

Frequency in GHz Time in nanosec.

(a) (b)

Fig. 7. a) Spectrum of the test input pulse with FCC mask
b) input pulse in time domain

137

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

0.8
0.G
0.4
0.2

-0.2
-0.4
-0.G
-0.8
-1

Input pulse

— — - Received pulse-DI Off D 2 off
Received pulse-DI Off D 2 on

— ■ — Received pulse-Pi on P2 off

1.5

3.5

2 2.5 3

Time in ns

Fig. 8. Comparison of input pulse and received pulses in time domain.

VI. Conclusion

This paper describes the simulated analysis of UWB slot
antenna with reconfigurable rejection bands. By switching the
diodes, the effective length of the ground slot is tuned to the
notch center frequency. The simple configuration and low
profile of the proposed antenna leads to easy fabrication that
may be built for any wireless UWB device applications where
reconfigurable rejection of notch bands is required. The
rejection of WiMax, IEEE 802.11a and HYPERLAN/2 bands
can be achieved using the proposed antenna.

References

[I] FCC NEWS(FCC 02-48), Feb. 14,2002. FCC News release.
[2] M. Ghavami, L.B. Michael and R. Kohno, Ultra Wideband Signals and

Systems in Communication Engineering, New York: John Wiley and
Sons, USA, 2004.

[3] K.L.Wong, Compact and Broadband Microstrip Antenna, John Wiley
and sons. Inc., NY, USA, 2001.

[4] I. -J. Yoon, H. Kim et al, "Ultra- wideband tapered slot antenna with band
cutoff characteristic," Electronics Letters, vol. 41, no. 11, pp. 629-630,
May 2005.

[5] A. M. Abbosh, M. E. Bialkowski et al, " A Planar UWB antenna With
Signal Rejection Capability in the 4-6 GHz Band," IEEE Microwave
and Wireless Comp. Letters, vol. 16, no. 5, pp. 278-280, May 2006.

[6] Keyvan Bahadori and Yahya Rahmat-Samii, "Miniaturized Elliptic-
Card UWB antenna with WLAN Band Rejection for Wireless
Communications," IEEE Trans. Antennas and Propag., vol.55, no. 11,
pp.3326-3332,Nov. 2007.

[7] Chong-Yu Hong, Ching-Wei Ling et al, " Design of a Planar Ultra-
wideband Antenna With a New Band-Notch Structure ," IEEE
Trans. Antennas and Propag. vol.55, no. 12, pp. 3391-3397, Dec. 2007.

[8] Q.-X. Chu and Y.-Y. Yang ," 3.5 / 5.5 GHz dual band-notch ultra-
wide band antenna," Electronics Letters, vol. 44, no. 3, pp. 172-174, Jan.
2008.

[9] L. Wang, . Wu, X.-W. Shi et al , "Design Of A Novel monopole UWB
Antenna with a notched ground," PIER, vol. 5, pp. 13-20, 2008.

[10] Jia-Yi Sze and Jen-Yi Shiu, " Design of Band-Notched Ultrawideband
square aperture antenna with a hat- Shaped Back-Patch," IEEE Trans.
Antennas and Propag., vol.56, no. 10, pp.3391-3397,Oct. 2008.

[II] S. Zhou, J. Ma and J. Deng, " A Novel dual band-notched ultra
wideband," Journal of Electromagn. Waves and Appln., vol. 23, pp.
57-63, 2009.

[12] J.Y. Deng, Y.Z. Yin, J. Ma and Q.Z.Liu, "Compact Ultra -wideband
with dual band-notched characteristics," Journal of Electromagnetic
Waves and Appln., vol. 23, pp. 109-116, 2009.

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

Vol. 8, No. 3, 2010

[13] X. Li, L.Yang, S.X. Gong and Y.J Yung, " A Novel tri band-notched
monopole antenna," Journal of Electromagnetic Waves and Appln., vol.
23, pp. 139-147,2009.

[14] Yi-Cheng Lin and Kuan- Jung Hung, "Compact Ultrawideband
Rectangular Aperture Antenna and Band-Notched Designs," IEEE
Trans. Antennas Propag., vol.54, no.l 1, pp. 3075-3081, Nov. 2006.

[15] Hyung Kuk Yoon, Yohan Lim et al, " UWB Wide Slot antenna with
Band -notch Function," Proc. IEEE Antennas and Propag. society
Intl Symposium, pp. 3059-3062, July 2006.

[16] Wen-jun Lui, Chong-hu Cheng and Hong-bo Zhu, "Improved
Frequency Notched Ultra wideband Slot Antenna Using Square Ring
Resonator," IEEE Trans. Antennas Propag., vol.55, no. 9,

pp.2445-2450, Sept.2007.

[17] Yunlong Cai and Zhenghe Feng, "A UWB Antenna with novel L
branches on ground for Band-Notching Application ," Proc. Of IEEE
Intl. Conference on Microwave and Millimeter wave Tec, vol. 4,
pp.1654-1657, April 2008.

[18] The-Nan Chang and Min-Chi Wu , " Band-Notched Design for
UWB Antennas , " IEEE Antennas Wire I. Propag. Letters, vol.7,
pp.636- 639, 2008.

[19] C- Y. Huang, S.- A. Huang and C- F. Yang , " Band-notched ultra-
wideband circular slot antenna with Inverted C-shaped parasitic
stripJ'Electronics Letters , vol. 44, no. 15, pp. 891-892, July 2008.

[20] Abdel-Fattah Sheta and Samir F. Mahmoud, "A Widely Tunable
Compact Patch Antenna, " IEEE Antennas Wirel. Propag. Letters,
vol.7, pp.40- 42, 2008.

[21] M.-I. Lai,T. Y . W u,J.-C. Hsieh,C.-H. Wang and S.-K. Jeng, "Design of
reconfigurable antennas based on an L-shaped slot and PIN diodes for
compact wireless devices," lET Microw. Antennas Propag., vol. 3, no. 1,
pp. 47-54,2009.

[22] Julien Sarrazin , Yann Mahe, Stephane Avrillon, and Serge Toutain,
"Pattern Reconfigurable Cubic Antenna," IEEE Trans. Antennas
Propag, vol.57, no.2, pp.3 10-3 17, Feb. 2009.

[23] Yevhen Yashchyshyn, Jacek Marczewski, Krzysztof Derzakowski, Jozef
W. Modelski , and Piotr B. Grabiec, "Development and Investigation of
an Antenna System With Reconfigurable Aperture," IEEE Trans.
Antennas Propag., vol.57, no.l, pp. 2-8, Jan. 2009.

[24] Symeon Nikolaou Nickolas D. Kingsley , George E. Ponchak, John
Papapolymerou, and Manos M. Tentzeris, "UWB Elliptical Monopoles
with a Reconfigurable Band Notch Using MEMS Switches Actuated
without Bias Lines," IEEE Trans. Antennas Propag., vol.57, no. 8,
pp.2242-2250, Aug. 2009.

[25] IE3D 14, Zeland Software, Ins., Fremont, USA.

[26] J. William and R. Nakkeeran , "A new compact CPW-fed wideband slot
antenna for UWB applications," Proc. of IEEE First Himalayan Intl.
Conference on Internet , Nov. 2009. OI 10. 1 109/ AHICI.2009.5340282.

[27] Kuang-ping ma, Yongxi Qian and Tatsuo Itoh, "Analysis and
applications of a New CPW- slot line Transition," IEEE Transactions on
Microwave theory and Techniques, vol. 47, pp. 426-432, April 1999.

[28] Stanislas Licul and William A Davis, "Ultra-wideband(UWB) antenna
measurements using vector Network analyzer," IEEE Antennas and
Propagation International Symposium, pp. 1319-1322,2004.

AUTHORS PROFILE

J.William received his B.E. degree in Electronics and Communication from
Bharathidasan University, Tamilnadu, India, and the M. Tech. degree in
Communication Systems from National Institute of Technology (N.I.T),
Trichy,India, in 1991 and 2006 respectively. He is currently working towards
the Ph.D. degree at Pondicherry Engineering College, Pondicherry, India. He
is a life member of ISTE and IE (I) and member of lEICE and EurApp. His
current research interest is in the area of coplanar waveguide feed antennas
and printed slot antennas for UWB.

R. Nakkeeran Received BSc. Degree in Science and B.E degree in
Electronics and Communication Engineering from the Madras University in
1987 and 1991 respectively and M.E degree in Electronics and
Communication Engineering (diversification in Optical Communication) from
the Anna University in 1995. He received Ph.D degree from Pondicherry
University in 2004. Since 1991, he has been working in the teaching
profession. Presently, he is Assistant Professor in Pondicherry Engineering
College. He is life member of lETE, ISTE, OSI and IE(I). Also he is member

138

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

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

Vol. 8, No. 3, 2010
of OS A, SPIE and IEEE. He has published seventy five papers in National and
International Conference Proceedings and Journals. He has co-authored a
book, published by PHI. His areas of interest are Optical Communication,
Networks, Antennas, Electromagnetic Fields and Wireless Communication.

1 39 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

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

Vol. 8, No. 3, 2010

UWB Slot Antenna with Rejection of IEEE 802.11a

Band

J. William* and R.Nakkeeran

Department of Electronics and Communication Engineering

Pondicherry Engineering College

Puducherry, India . 605014.

Email id: wills.susan@gmail.com, rnakeeran@pec.edu

^Corresponding author: wills.susan@gmail.com

Abstract — A compact coplanar waveguide (CPW) fed slot
antenna for ultra wideband (UWB) with notched band from
5.1 GHz to 5.9 GHz is presented in this paper. By inserting a
rectangular slot with the particular length and width in the
ground plane, a desired notch in the frequency band can be
achieved. The characteristics of the designed structure are
investigated using an electromagnetic solver, IE3D. The overall
size of the antenna comes around 28(L) x 21(W) x 1.6(T) mm^
For the developed antenna the VSWR is measured and compared
with the simulated results. The measured parameter is in good
agreement with the simulation and the antenna covers entire
UWB band ranging from 3.1 GHz to 11.4 GHz with band
notching between 5.1 GHz and 5.9 GHz. Time domain analysis of
the antenna is also investigated and presented, which ensures that
the antenna is capable of working effectively in the UWB
environment. This type of antenna configuration would be quiet
useful for UWB indoor applications with no interference from
WLAN and HYPERLAN/2 systems when they coexist.

Keywords- band notch; coplanar waveguide; slot antenna; time
domain analysis; ultra wideband

I. Introduction

To support variety of applications to the mobile users needs
wireless communication systems with higher capacity and data
rate. Ultra wideband (UWB) is a short pulse communication
system offers both of the above requirements. After the release
of UWB by Federal Communications Commission (FCC), it
has become one of the interesting technologies in indoor
wireless communications system and receives much attention
by the industries and academia due to its properties of low
power consumption, support of high secured data rate and
simple configuration [1]. Designing an antenna to operate in
the UWB band is quiet challenging one because it has to satisfy
the requirements such as ultra wide impedance bandwidth,
omni directional radiation pattem, constant gain, high radiation
efficiency, constant group delay, low profile, compact and easy
manufacturing [2]. Interestingly the planar slot antennas with
CPW fed possesses the above said features with simple
structure, less radiation loss, less dispersion and easy
integration of monolithic microwave integrated circuits
(MMIC) [3]. Most of the UWB antennas have a wide
bandwidth, covering bands used for other wireless
communication applications. Therefore, the UWB antenna

design could play an important role when interference with
other wireless applications such as IEEE 802.11a and
HYPERLAN/2 systems which are also coexisting with UWB
band (5.1 GHz - 5.9 GHz), which degrades the UWB system
performance.

Therefore, a UWB antenna having frequency band notch
characteristics is enviable. To mitigate this electromagnetic
interference from nearby communication systems many
antenna designs have been proposed in the literature [4-10].
Many techniques also used to introduce notch band for
rejecting the interference in the UWB slot antennas. It is done
either by inserting half wavelength slits, stripes in the tuning
stub [11], or inserting stub in the aperture connected to the
ground planes [12], or inserting square ring resonator in the
tuning stub [13], or inserting 'L' branches in the ground plane
[14], or with complementary split ring resonator [15], or
inserting strip in the slot [16]. In this paper, a new fype of
UWB antenna and its notched design is proposed by inserting a
slot in the ground plane at a half wavelength size at the desired
center notch frequency of 5. 5 GHz, which is an average of
notch band 5.1 GHz to 5.9 GHz. The proposed antenna is
simulated and analyzed by using simulation software, IE3D 14
[17]. The details of the proposed design and its experimental
results are presented and discussed in the following sections.
The paper is organized as follows: Section II brings out the
design and geometry of the antenna. In Section III simulation
results and analysis are presented. Obtained experimental
results are given in Section IV. Section V explains time domain
analysis of the antenna. Section VI concludes the paper.

II. Antenna Geometry

In designing this fype of antennas, the width 'W' and
length 'L' play a crucial role in determining the resonant
frequency of the system. The initial values of these parameters
are calculated by using the equations given in [18] for the
substrate height (h), dielectric constant (e^) and for the lower
frequency. The designed values of the antenna are optimized
with IE3D tool. The optimization was performed for the best
impedance bandwidth. The proposed antenna with slot in the
ground plane in this paper is a dimensional modification of
structure given in [19]. The notched design for the particular

140

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

band is obtained by introducing a slot in the ground plane
nearer to the radiating slot.

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

Vol 8, No. 3, 2010
this study, the dielectric substance (FR4) with thickness of
1.6mm with relative permittivity of 4.4 is chosen as substrate to
facilitate printed circuit board integration. The CPW feed is
designed for 50 ^ characteristic impedance with fixed 2.4 mm
feed line width and 0.5 mm ground gap.

III. Simulated Results and Analysis

The analysis and performance of the proposed antenna is
explored by using IE3D for the better impedance matching.
The parametric analysis of the antenna carried out by varying
one parameter and keeping other parameters constant. The
simulated return loss of the proposed antenna is shown in
Fig. 2, which clearly indicates that the impedance bandwidth
of the antenna is 8.3 GHz (3.1GHz -11.4 GHz) for return loss
(Sii ) < -10 dB ( VSWR < 2). The ultra wideband is due to
multiple resonances introduced by the combination of the
rectangular slot and the tuning stub. The resonant frequency
and bandwidth are controlled by the size of the rectangular
slot, antenna and tuning stub. Detailed parametric analysis has
been carried out and the final optimal parameter values of the
antenna are listed in the Table 1 . However to study the impact
of the slot, the slot length 'L2' and width 'W2' are varied by
keeping one of them as constant. The current distribution, gain
and group delay are also studied.

Fig. 1 Geometry of the proposed slot antenna and its notched design.

The effective length of the slot width is found to be
approximately 0.5 ^^eff for the notch in the frequency band at
the center frequency of 5.5 GHz. The introduction of the slot
in the ground plane causes destructive interference in the
current distribution which is responsible for the elimination of
the specific frequency band.
The effective wavelength of the slot width is,
C S +1

^eff

"eff

(1)

where '4' is the centre frequency of the notch band.

The structure of the basic UWB slot antenna and its
notched design is shown in Fig 1. The antenna consists of
rectangular slot with width 'Wi' and length 'Li'. The tuning
stub comprises a triangular patch with height 'H'. The distance
between the tuning stub and feed line is 'd'. 'W2' and 'L2' are
the width and length of slot in the ground plane. ' W and 'L'
are the overall width and length of the antenna respectively. In

-5

CO
■0-10

E -20

1-25
-30

\ ^

V /Tx

M p

(v/

\/-

]

-35

jif\

3 4 5 6 7 8 9 10 11 12

Frequency in GHz

Fig. 2 Simulated return loss of the proposed UWB antenna.

TABLE I.

Optimal Parameter Values of the Antenna

Parameter

Description

Optimal
Value

Length of the antenna

28 mm

Width of the antenna

21 mm

Length of the slot

15 mm

Width of the slot

16.8 mm

Length of the slot in the
ground plane

1.8 mm

Width ofthe slot in the
ground plane

16.2 mm

Feed gap distance

1.6 mm

Height ofthe patch

9.3 mm

141

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

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

The simulated VSWR, with and without slot in the ground
plane is shown in Fig. 3 and it is inferred that the antenna with
slot has a impedance bandwidth of 3-.1 GHz to 11.4 GHz for
VSWR < 2 (Sii < -10 dB) with notch band from 5.1 GHz to
5.9 GHz.
12

E£ 8

5
> G

i : i i

; II : : : : : :
■ with slot

...

without slot

T--1-1 1 T 1 T-

1 l¥ 1 1 1 1 1 1

1 . 1 .1 . J 1 : 1 : 1 . . .

^ '^JJ ^ ^/

1 1 1 1 1 I I

"3 4 5 6 7 8 9 10 11 12
Frequency in GHz

Fig. 3 Simulated VSWR with and without
slot in the ground plane.

A. Effect of Slot in the Ground Plane

The slot is introduced in the ground plane as shown in
Fig. 1 to obtain the notch in the specified band. Fig. 4a shows
the VSWR variation for different slot lengths 'L2' (for a
particular width). It is observed that only a very small
variation in the notch band for 28% of length variation with
respect to the ground plane. Also it is noticed that the shift in
notch band is directly proportional to slot length. Fig. 4b
depicts VSWR variation for different widths' W2' (L2=1.8 mm
optimal value). It is noticed that the notch frequency is shifted
towards higher frequency when slot width is decreased and
vice versa. The rate of notch frequency shift per unit width is
0.36 GHz. The optimal width value is 16.2 mm.
12

tf 8

> 6

2 V

Lj = 0,5 mm

■-2= 1 mm

La = 1-8 mm

L2 = 3 mm

4 5 6 7 8 9 10
Frequency in GHz

(a)

11 12

12
10

E£ 8

CO
> 6

Vol 8, No. 3, 2010

W2 = l4mm
W2 = 15 mm
W2 = 16.2 mm
W2 = 17 mm

"3 4 5 6 7 8 9 10 11 12
Frequency in GHz

(b)

Fig.3 Simulated VSWR a) diffemet slot lengtlis'L2' b) different slot

widths 'W2'

B. Simulated Current Distribuition

The simulated current distribution at frequency 5.5 GHz
with and without slot in the ground plane is presented in Fig. 4.
The current distribution of the proposed antenna is obtained by
accounting the optimal design parameter values. In Fig. 4a the
current distribution is mainly around the rectangular radiating
slot and in the tuning stub. From the Fig. 4b, it is witnessed that
the current distribution around the radiation slot is disturbed by
the introduction of slot in the ground plane, which is
responsible for the notch in the frequency band.

(a) (b)

Fig. 4 Current distributions at 5.5 GHz. a) without slot (b) with slot in the
ground plane.

C Simulated Gain

The computed gain of the UWB antenna with and without
slot in the ground plane is compared in Fig. 5. It is observed
that there is a gain variation at the notched frequency which is
0.5 dBi where as it is 3.5 dBi for the antenna without slot and
the gain varies from 2 dBi to 6 dBi across the UWB spectrum.

142

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

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

Vol. 8, No. 3, 2010

= 4

"5,

;;;::;

1 * *

■/ *- - -V

1* *

, { \ J. . I.* . . .1. .V- - - -L I.

, ; ■ -;-/^- --; r?Sw^ i

■ : ■ ''5 ^without slot

: ^ ■■■ with slot,

1 1 1 1 1 1 1

6 7 8 9
Frequency in GHz

10 11 12

Fig. 5. Comparison of simulated gain responses.

D. Group delay per/romance

The group delay 'x' of the antenna is calculated from the phase
of the computed 'S21' by using the following equation and
plotted in Fig. 6,

d(/)
T =

(2)

where ' ^ ' is phase of S21 in radians /sec and T is frequency
in GHz.

1 1 1

ff¥f^

1 1

-without si

-with slot

1 1 1 ■ ■
1 1 1

■0-2

0.
3

2-3

3 4 5 6 7 8 9 10 11 12
Frequency in GHz

Fig. 6. Computed group delay.

From the Fig. 6, it is noticed that the variation in the group
delay for the antenna with and without slot is around 2 ns for
the frequency range from 3.1 GHz to 10.6 GHz, but there is
small variation in the group delay at the notch band in the
response for the antenna with slot which is acceptable one.

IV. EXPERIMENTALL RESULTS AND DISCUSSION

The prototype of the proposed antenna shown in Fig. 1 was
fabricated for different parameters with their optimal values
and tested, which is depicted in Fig. 7. Using Hewlett Packard
Network Analyzer (HP8757D), the VSWR is measured and
compared with the simulation result is shown in Fig. 8.

Fig. 7 Fabricated UWB slot antenna and its notch antenna.

There is a discrepancy between the measured and the
simulated ones might be the effect of soldering the SMA
connector or fabrication tolerance. The simulation result was
obtained by assuming coplanar as input port, whereas
practically SMA connector was used, the imperfect transition
between SMA feed to coplanar may introduce losses [20] and
also the capacitances can lead to shift in the frequency.

C06

1 1

1 1 1 1

' ^^^b|^^^S|

jred i *-

3ted : :**

■

: 1 \}

E 1 Simula

i i

i i i i

11 12

4 5 6 7 8 9 10

Frequency in GHz

Fig. 8. Comparison of measured and simulated
VSWR.

V. Time Domain Analysis

In ultra wideband systems, the information is transmitted
using short pulses. Hence it is important to study the temporal
behavior of the transmitted pulse. The communication system
for UWB pulse transmission must provide as minimum as
possible distortion, spreading and disturbance. The channel is
assumed to be linear time invariant (LTI) system to verify the
capability of the proposed antenna for transmission and
reception of these narrow pulses. The transfer function of the
entire system is computed using simulated value of 'S21'
parameter [21]. The received output pulse is obtained by
taking the Inverse Fourier Transform (IFT) of the product of
transfer function and spectrum of the test input pulse. While
computing 'S21', two identical antennas are placed face to face
position at a distance of 70 cm that is greater than the far- field
distance of the antenna.

143

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

-5

Q -10

i -IS
^■20

1.:

o
Z

-30

-35

-40

^ [

I •

---i-i.

::4jz*

.......

A\\ \

« ■ al

FCC Indoor Emission Mask

1 1 1 1

4 6 8

Frequency in GHz

Fig. 9. FCC mask with the spectrum of the test input pulse.

The cosine modulated Gaussian pulse is considered for this
analysis with centre frequency of 6.85 GHz and pulse width of
220 picoseconds, whose spectrum is shown in Fig. 9. It
satisfies the requirement of FCC mask for UWB indoor
emission. The comparison of input and output received pulse
for the antenna with notch and without notch is shown in
Fig. 10, which ensures the distortion less pulse transmission
and also guarantees that the designed antenna is capable of
transmitting and receiving short pulses. The ringing effect in
the waveform may be due to the transmission properties of the
system.

0) 0.:

-0.5

'^f^^^m

^^^ Input pulse

~ ~ ■ Received pulse from UWB antenna

' Received pulse frcm UWB notch antenna

12 3 4

Time in nanosec.

Fig. 10. Comparison of the received pulses from the UWB antenna without
and with notch with input pulse.

VI. Conclusion

In this paper, a CPW fed UWB slot antenna with novel band
notched design is proposed. The triangular stub is introduced
at the anterior portion of the feed to enhance the coupling
between the slot and feed. By embedding a slot in the ground
plane, which is nearer to the radiating slot is responsible for the
notch in the undesired band. With the above structural features
the overall dimension of the proposed antenna comes around
28x21x1.6 mm^. The antenna covers the entire UWB system

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

Vol 8, No. 3, 2010
ranging from 3.1 GHz to 11.4 GHz with band notching
between 5.1 GHz and 5.9 GHz. Time domain analysis of the
proposed antenna ensures the suitability of the antenna for
UWB applications. Hence, this antenna configuration would be
quiet useful for UWB indoor applications with no interference
from WLAN and HYPERLAN/2 systems when they coexist.

Refepiences

[I] FCC NEWS(FCC 02-48), Feb. 14,2002. FCC News release. New
public safety applications and broadband internet access among uses
envisioned by FCC authorization of ultra wideband technology.

[2] M. Ghavami, L.B. Michael and R. Kohno " Ultra Wideband Signals

and Systems in Communication Engineering", John Wiley and sons.
Inc., NY, USA, 2004.

[3] K.L. Wong, " Compact and Broadband Microstrip Antenna", John

Wiley and sons. Inc., NY, USA, 2001.

[4] I. -J. Yoon, H. Kim et al, "Ultra- wideband tapered slot antenna with

band cutoff characteristic," Electronics Letters, vol. 41, no. ll,pp.
629-630, May 2005.

[5] A. M. Abbosh, M. E. Bialkowski et al, " A Planar UWB antenna

With Signal Rejection Capability in the 4-6 GHz Band," IEEE
Microwave and Wireless Comp. Letters, vol. 16, no. 5, pp. 278-280,
May 2006.

[6] Key van Bahadori and Yahya Rahmat-Samii, "Miniaturized

Elliptic-Card UWB antenna with WLAN Band Rejection
for Wireless Communications," IEEE Trans. Antennas and
Propag., vol.55, no. 11. pp.3326-3332,Nov. 2007.

[7] Chong-Yu Hong, Ching-Wei Ling et al, " Design of a Planar

Ultra- wideband Antenna With a New Band-Notch Structure ,"
IEEE Trans. Antennas and Propag. vol.55, no. 12, pp.339 1-
3397,Dec. 2007.

[8] Q.-X. Chu and Y.-Y. Yang ," 3.5 / 5.5 GHz dual band-notch

ultra- wide band antenna," Electronics Letters, vol. 44, no. 3, pp.
172-174, Jan. 2008.

[9] L. Wang, . Wu, X.-W. Shi et al , "Design Of A Novel monopole

UWB Antenna with a notched ground," PIER, vol. 5,pp. 13-20,
2008.

[10] Jia-Yi Sze and Jen-Yi Shiu, " Design of Band-Notched

Ultrawideband square aperture antenna with a hat-Shaped Back-
Patch," IEEE Trans. Antennas and Propag., vol.56, no. 10. pp. 3391-
3397,Oct. 2008.

[II] Yi-Cheng Lin and Kuan- Jung Hung, "Compact Ultrawideband
Rectangular Aperture Antenna and Band-Notched Designs," IEEE
Trans. Antennas Propag, vol.54, no.l 1, pp.3075-3081, Nov.2006.

[12] Hyung Kuk Yoon, Yohan Lim et al, " UWB Wide Slot antenna

with Band -notch Function," Proc. IEEE Antennas and Propag.

society Intl Symposium, pp. 3059-3062, July 2006.
[13] Wen-jun Lui, Chong-hu Cheng and Hong-bo Zhu, "Improved

Frequency Notched Ultra wideband Slot Antenna Using Square

Ring Resonator," IEEE Trans. Antennas Propag., vol.55, no. 9,

pp.2445-2450, Sept.2007.
[14] Yunlong Cai and Zhenghe Feng , " A UWB Antenna with

novel L branches on ground for Band-Notching Application ,"

Proc. Of IEEE Intl. Conference on Microwave and Millimeter

wave Tec, vol. 4, pp.1654-1657, April 2008.
[15] The-Nan Chang and Min-Chi Wu , " Band-Notched Design for

UWB Antennas , " IEEE Antennas Wirel. Propag. Letters, vol.7,

pp.636- 639, 2008.
[16] C- Y. Huang, S.- A. Huang and C- F. Yang , " Band-notched

ultra-wideband circular slot antenna with Inverted C-shaped

parasitic strip,'" Electronics Letters , vol. 44, no. 15, pp. 891-892, July

2008.
[17] IE3D 14, Zeland Software, Ins., Fremont, USA.

[18] C. Balanis ,Antenna Theory Analysis and Design, 3rd edition. New

York , Wiley, 2005
[19] J. William and R. Nakkeeran , "A CPW-fed wideband slot antenna

with triangular patch," Proc. of IEEE Intl. Conference on Computing,

144

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

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

Vol. 8, No. 3, 2010

Communication and Networking , Dec. 2008. DOI. current research interest is in the area of coplanar waveguide feed antennas

10. 1 109ICCCNET.2008.4787770. and printed slot antennas for UWB.
[20] Kuang-ping ma, Yongxi Qian and Tatsuo Itoh, "Analysis and

applications of a New CPW- slot line Transition," IEEE R. Nakkeeran Received BSc. Degree in Science and B.E degree in

Transactions on Microwave theory and Techniques, vol. 47, pp. Electronics and Communication Engineering from the Madras University in

426-432, April 1999. 1987 and 1991 respectively and M.E degree in Electronics and

[21] Stanislas Licul and William A Davis, " Ultra wide band (UWB) Communication Engineering (diversification in Optical Communication) from

antenna measurements using vector Network analyzer," IEEE the Anna University in 1995. He received Ph.D degree from Pondicherry

Antennas and Propagation International Symposium, pp. 1319- University in 2004. Since 1991, he has been working in the teaching

1322, June 2004. profession. Presently, he is Assistant Professor in Pondicherry Engineering

AUTHORS PROFILE College. He is life member of lETE, ISTE, OSl and 1E(1). Also he is member

T„,.„. . ,,. ^^ , . ^, ^ . , _ • .• r. of OS A, SPIE and IEEE. He has published seventy five papers in National and

J.Wilham received his B.E. degree in Electronics and Communication irom ^ ^ ^- ^ r^ n t^ j- j t i t t i ^i j

^, ^, . , ^^ . .^ ^ ., , ^ ,. . . ^^rr. , , • International Conierence Proceedings and Journals. He has co-authored a

Bharathidasan University, Tamilnadu, India, and the M. Tech. degree in , , ii- i j i t^ttt tt- x^ ■ ^ ^ ^ ^- i r^ • ^•

„ . ^. „ ^ -^ ' ^T .• 1 T .V . r ^ A 1 7t.ttt^x book, published by PHI. His areas of interest are Optical Communication,

Communication Systems irom National Institute oi Technology (N.l.T),

Trichy,India, in 1991 and 2006 respectively. He is currently working towards Networks, Antennas, Electromagnetic Fields and Wireless Communication,
the Ph.D. degree at Pondicherry Engineering College, Pondicherry, India. He
is a life member of ISTE and IE (1) and member of lElCE and EurApp. His

145 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

A STUDY OF VARIOUS LOAD BALANCING TECHNIQUES IN

INTERNET

M.Azath\ Dr.R.S.D.Wahidabanu ,

^Research Scholar, Anna University, Coimbatore.

^ mailmeazath @ gmail.com

^Research Supervisor, Anna University, Coimbatore.

^ drwahidabanu @ gmail.com

Abstract

One of the most important applications of
traffic engineering is load balancing.
Successful implementation of load balancing
depends on the underlying routing protocol
that provides connectivity through the
Internet by determining the routes used by
traffic flows. But the load-balancing
problem is not yet solved completely; new
applications and architectures are required to
meet the existing or incoming fastest
Internet world. And, for greatest impact,
these new capabilities must be delivered in
toolkits that are robust, easy-to-use, and
applicable to a wide range of applications.
For balancing traffic in internet, packets
should be reorder, reordering also having a
problem for flows in internet. In Internet,
unresponsive flows easily occupy the
limited buffers, there by reducing the
Quality of Service (QoS). In this paper,
various techniques that are adopted for load
balancing in Internet are analyzed.

Keywords:

Traffic engineering. Load Balancing,
Internet Services, unresponsive flows, QOS,
Buffer, Traffic splitting and Router.

1. Introduction

traffic has to be carried in the network in
such a way that performance objectives are
fulfilled. In computer networking, load
balancing is a technique to spread work
between two or more computers, network
links, CPUs, hard drives, or other resources,
in order to get optimal resource utilization,
throughput, or response time.

One of the most common applications of
load balancing is to provide a single Internet
service from multiple servers, sometimes
known as a server farm. Commonly load
balanced systems include popular web sites,
large Internet Relay Chat networks, high
bandwidth File Transfer Protocol sites,
NNTP servers and DNS servers.

The idea of load balancing is to move traffic
from congested links to other parts of the
network in a well-controlled way. Traffic
engineering seeks to effectively balance
traffic load throughout existing networks,
thus achieving QoS demands and
minimizing typical costs of adding hardware
and software implementations, common to
network engineering. When dealing with
real-world cases of load balancing, both
network and traffic engineering is general
purpose tools used throughout all steps of an
implementation [12].

Traffic engineering refers to the
performance optimization of operational
networks. On one hand, traffic offered
between origin and destination nodes loads
the network and on the other hand, this

2. Motivation

Load balancers are an integral part of today's
Web infrastructure. They're also complex
and under-documented pieces of hardware.

146

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

Today's Web sites are complex beasts.
Every component must work together to
create a site that is greater than the sum of
its parts.

site. This is why a firewall is often a suspect,
too, but to a lesser degree since it is
generally a simpler device than load
balancers [1].

^ Internet

^^^B

^H Routers

^^^^1

\ ^^^^^^^^^^^^^^^1

^^^1

^H Firewalls

\ ^^^^^^^^^^^^^^^1

^^^1

^U Load Balancers

\ ^Hm^^fflRSn^^^^^H

IKIIjjIjS

^U Web Servers

\ ^^^^^^^^^^^^^^^1

VHHH

^U App Servers

\ ^^IHIHIHHI^^^H^I

'^^ 1

^M DB Sen/ers

Figure 1 : Traffic flow for a load balancer

The Internet is connected to the routers,
which pass traffic through a firewall to the
load balancers, which distribute the traffic to
the Web servers, which pass information to
the application server bone, and the
application server bone is connected to the
database server bone. We get the picture. If
one component or piece of the process fails,
it can take down the entire site.

Load balancers are also in the direct path of
all traffic to a particular Web site. By
looking at Figure 2 below, we can see that if
the load balancer stops working, the entire
site stops working. This critical position in
the infrastructure can make it appear as
though the load balancer is the problem,
even in cases where it is not (such as a
firewall issue, a back-end database problem,
someone tripping over a cable, etc.). Unlike
a broken or malfunctioning Web server, a
miss configured or malfunctioning load
balancer will result in a dead-to-the-world

Internet Users

Web Sen/er 1

I WebS«

Figure 2: Load Balancer implementation

3. Needs of load balancing

Load balancing is common in ISP networks.
If the traffic demands are known, the load
balancing can be formulated as an
optimization problem. However, knowledge
of traffic demands is often lacking. Recent
trends in load balancing points toward
dynamic protocols [13]. These protocols
map the traffic of router onto multiple paths
and adapt the share of each path in real-time
to avoid hot-spots and cope with failures.
Dynamic load balancing needs schemes that
split traffic across multiple paths at a fine
granularity.

Since the power of any server is finite, a
web application must be able to run on
multiple servers to accept an ever increasing
number of users. This is called scaling.
Scalability is not really a problem for
intranet applications since the number of
users has little chances to increase.

However, on internet portals, the load
continuously increases with the availability
of broadband Internet accesses. The site's
maintainer has to find ways to spread the

147

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

load on several servers, either via internal
mechanisms included in the application
server, via external components, or via
architectural redesign.

Further complication increases due to
various factors such as:

• Sizes of objects might not be the
same.

• Object IDs might not be chosen at
random.

• Heterogeneity in the capabilities of
nodes.

4. Types of load balancing and its
approaches

Load balancing algorithms can be classified
into three main classes: static algorithms,
dynamic algorithms, and adaptive
algorithms. Static algorithms decide how to
distribute the workload according a prior
knowledge of the problem and the system
characteristics. Dynamic algorithms use
state information to make decisions during
program execution. Finally, Adaptive
algorithms are a special case of dynamic
algorithms. They dynamically change its
parameters in order to adapt its behavior to
the load balancing requirements [10].

As though there are three main algorithms
for load balancing, there are different load
balancing approaches available. They are:

1. Client-side load balancing: Client-side
load balancing is not a normal practice, but
it is indeed possible. For some times,
Netscape incorporated a simple balancing
algorithm in their Navigator browser,
making it choose a random Netscape web-
server when visiting www.netscape.com.
[2].

2. Core Internet routing uses protocols and
agreements that allow for automated load
balancing and fail-over mechanisms. These

are commonly based on the Border Gateway
Protocol, used for data-exchange between
large Internet operators.

3. DNS -based load balancing is a popular
way of distributing traffic amongst a set of
Internet addresses by returning a list of
active addresses to the requesting client.
These addresses can point to a set of servers
or even a set of geographically separate
sites.

4. Sites can connect to the net through
several links, a practice known as multi-
homing. This enables both incoming and
outgoing load balancing, in addition to the
increased redundancy.

5. Dispatcher-based load balancing is used
within a site to balance load between a set of
real servers. Generally, the dispatcher
assumes a virtual address for a service and
receives requests which it then redirects to
an appropriate server based on given
criteria.

6. The real servers can again operate some
form of balancing mechanism to decide
whether to handle the request or redirect it to
a more suitable server or site.

7. Content servers could access back-end
servers - typically running databases and
low-level services - in a load-balanced
fashion.

8. Back-end servers could also incorporate
balancing amongst themselves to avoid
over-utilization.

When talking about varying levels of load
balancing, it is fair to identify the level to be
proportional to the distance from the content
served - long distance equals high level, and
vice versa. In an informal manner, we can
designate steps 1 through 4 in the figure as
high-levels, and steps 5 through 8 as low-
levels of load balancing [12].

5. Adopted techniques

There are various papers which explain
about the load balancing in Internet. Each of

148

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

them is routed in different directions and
with various considerations.

Router mechanisms designed to achieve fair
bandwidth allocations, like Fair Queueing,
have many desirable properties for
congestion control in the Internet. In [6],
architecture is proposed that significantly
reduces this implementation complexity yet
still achieves approximately fair bandwidth
allocations. This architecture has two key
aspects:

1. To avoid maintaining per flow state at
each router, we use a distributed algorithm
in which only edge routers maintain per
flow state, while core (non-edge) routers do
not maintain per flow state but instead
utilize the per-flow information carried via a
label in each packet's header. This label
contains an estimate of the flow's rate; it is
initialized by the edge router based on per-
flow information, and then updated at each
router along the path based only on
aggregate information at that router.

2. To avoid per flow buffering and
scheduling, as required by Fair Queueing,
we use FIFO queueing with probabilistic
dropping on input. The probability of
dropping a packet as it arrives to the queue
is a function of the rate estimate carried in
the label and of the fair share rate at that
router, which is estimated based on
measurements of the aggregate traffic. The
limitation here is in the destination, packet
reorganizing should be there.

comparable performance to 16-bit CRC. As
though hashing methods provide best
performance, new hash based algorithms are
needed that have less computational
complexity.

In [7], they proposed a per-class queue
management and adaptive packet drop
mechanism in the routers for Internet
congestion control. An active queue
management is modeled as an optimization
problem and the proposed mechanism
provides congestion control and fairness for
different types of traffic flows. An optimal
packet drop rate is obtained to maintain a
relatively small queue occupancy, which
provides a less queue delay delivery of
packets. Moreover, the queue occupancy
and the packet drop rates obtained are both
upper bounded, which is meaningful for
providing the class-based guaranteed delay
services for real-time multimedia
applications. They model the general AQM
as an optimization problem, and try to obtain
a minimal packet drop rate that results in
low queue occupancy. Compared with RED
that controls the average queuing delay in
the router, the per-class queue management
and optimal packet drop mechanism can
obtain the minimal queuing delay and hence
the end-to-end delay. The major drawback
in this paper is that the packet drop rate is
minimal and the resulting queue occupancy
is also kept minimal.

In [18], they evaluated 5 direct hashing
methods and one table-based hashing
method. While hashing schemes for load
balancing have been proposed in the past,
this is the first comprehensive study of
performance using real traffic traces. They
find that hashing using 16-bit CRC over
TCP five tuples gives excellent load
balancing performance. Load-adaptive table-
based hashing uses the exclusive-OR of the
source and destination IP addresses achieve

In [19], a novel packet scheduler is proposed
and is called Stratified Round Robin, which
has low complexity, and is amenable to a
simple hardware implementation. In
particular, it provides a single packet delay
bound that is independent of the number of
flows. This property is unique to Stratified
Round Robin among all other schedulers of
comparable complexity.

149

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

An important component of the many QoS
architectures proposed is the packet
scheduUng algorithm used by routers in the
network. The packet scheduler determines
the order in which packets of various
independent flows are forwarded on a shared
output link. One of the simplest algorithms
is First Come First Served (FCFS), in which
the order of arrival of packets also
determines the order in which they are
forwarded over the output link. FCFS
clearly cannot enforce QoS guarantees, as it
allows rogue flows to capture an arbitrary
fraction of the output bandwidth.

Stratified Round Robin operates as a two-
step scheduler:

1. The first step uses the flow class
mechanism to assign slots to each
flow fi in proportion to its
approximate weight as defined by
the flow class Fk to which it belongs.

2. The second step uses the weight-
proportional credit mechanism to
ensure that each flow fi receives
service in proportion to its actual
weight wi.

The advantage of this approach is that it
considerably simplifies the scheduling
decision to be made. But we are having the
difficulty of considerable packet
reorganizing in the destination.

In this paper [11], based on measurements of
Internet traffic, they examined the sources of
load imbalance in hash-based scheduling
schemes. They proved that under certain
Zipf-like flow-size distributions, hashing
alone is not able to balance workload. They
introduced a new metric to quantify the
effects of adaptive load balancing on overall
forwarding performance. To achieve both
load balancing and efficient system resource
utilization, they proposed a scheduling
scheme that classifies Internet flows into

two categories: the aggressive and the
normal, and applies different scheduling
policies to the two classes of flows. They
have stated that their work is unique in
exploiting flow-level Internet traffic
characteristics.

In [13], a new mechanism called Flow-let
Aware Routing Engine (FLARE) is
proposed in which a new traffic splitting
algorithm that operates on bursts of packets,
carefully chosen to avoid reordering. Using
a combination of analysis and trace-driven
simulations, it is shown that FLARE attains
accuracy and responsiveness comparable to
packet switching without reordering packets.
FLARE is simple and can be implemented
with a few KB of router state. Highly
accurate traffic splitting can be implemented
with little to no impact on TCP packet
reordering and with negligible state
overhead. Owlets can be used to make load
balancing more responsive, and thus help
enable a new generation of real-time
adaptive traffic engineering.

Static load balancing is presented in [3]. The
static problem is possible to be formulated
and solved only if we have precise
information on all the traffic demands.
However, such information is not available
or traffic condition change unexpectedly. In
dynamic load balancing, dynamically
changing network status information are
utilized [5],[8],[14],[4],[15].

Traffic Engineering (TE) of dynamic
methodologies is classified into two basic
types: time-dependent and state-dependent.
In time-dependent TE, traffic control
algorithms are used to optimize network
resource utilization in response to long time
scale traffic variations. In state-dependent
TE, traffic control algorithms adapt to
relatively fast network state changes. State-
dependent load balancing is a key technique
for improving the performance and

150

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

scalability of the Internet. The fundamental
problem in dynamic load balancing in
distributed nodes involves moving load
between nodes. Each node can transfer load
to at most one neighbor also, any amount of
load can be moved along a communication
link between two nodes in one step. The
dynamic load balancing is formed with
incomplete information. More precisely, it is
assumed that the traffic demands are
unknown but the link loads are periodically
measured using a measurement system as in
[9], [16] [17].

Floyd et at. [20] study congestion collapse
from undelivered packets. This situation
arises when bandwidth is continuously
consumed by packets at the upstream that
are dropped at the downstream. Several
ways to detect unresponsive flows are
presented. It is suggested that routers can
monitor flows to detect whether flow is
responsive to congestion or not. If a flow is
not responsive to congestion, it can be
penalized by discarding packets to a higher
rate at the router. According to the authors
there are some limitations of these tests to
identify non-'TCP-friendly flow". It does
not help to save bandwidth at the upstream if
the flow sees the congestion at the
downstream because this solution does not
propagate the congestion information from
downstream to upstream [21].

In [22], architecture which contains an
adaptive packet scheduler with a bursty
traffic splitting algorithm is proposed for
better load balancing. The scheduler has a
classifier which classifies the flows into
aggressive and normal flows. Aggressive
flows are treated as high priority flows.
Based on the buffer occupancy threshold, a
trigger handler checks for load un-balance of
the network and automatically triggers the
load adapter. The load adapter reroutes the
high-priority aggressive flows into the least
loaded best path, using the bursty traffic

splitter algorithm. As though it is adaptive
algorithm for load balancing, it does not
satisfy the needs when we are coming across
the unresponsive flows.

6. Conclusion

Static load balancing is suffering from lack
of all the available information while we
perform for load balancing. Similarly
dynamic load balancing needs schemes that
split traffic across multiple paths at a fine
granularity. In adaptive load balancing the
performance of the network can be
improved by parameter adjustments of the
routes, traffic splitting, and scheduling. Thus
in this current fast Internet scenario, a new
adaptive load balancing algorithm is needed.
The adaptive load balancing is also not
sufficient for some situations like
unresponsive flows of traffics. Hence there
is a need that for load balancing algorithm
with the additional feature of detecting
unresponsive flows of the incoming traffic
in Internet.

References

[1]

http : //www . oreilly net . com/pub/a/oreilly /net

working/news/slb 0301.html Feb 2010.

[2] Dan Mosedale, William Foss, and Rob
McCool, "Lessons learned administering
netscape's internet site." IEEE Internet
Computing, l(2):28-35, 1997.

[3] Grenville Armitage, 2000. MPLS the
magic behind the myths. IEEE Commun.
Mag., 38: 124-131.

[4] Dinan, E., D. Awduche and B. Jabbari,
"Analytical framework for dynamic traffic
partitioning in MPLS networks", IEEE
International Conference on

Communications, (ICC 00), 18-22 June
2000, New Orleans, Volume-3, pp: 1604-

151

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

1608.

[5] Elwalid, A., C. Jin, S. Low and
I.Widjaja, MATE: MPLS adaptive traffic
engineering. IEEE Infocom., Twentieth
Annual Joint Conference of the IEEE
Computer and Communications Societies,
Volume-3, April 22-26 2001, Anchorage,
Alaska, USA. ISBN 0-7803-7016-3, 1300-
1309.

[6] Core-Stateless Fair Queuing: A Scalable
architecture to approximate fair bandwidth
allocations in high speed networks. Ion
Stoica, Scott Shenker, Hui Zhang, 2003.

[7] Mei-Ling Shyu, Shu-Ching Chen,
Hongli Luo, "Per-class queue management
and adaptive packet drop mechanism for
multimedia networking". Proceedings of the
2003 International Conference on
Multimedia and Expo - Volume 3 (ICME
'03) -Volume 03.

[8] Song, J., S. Kim, M. Lee, H. Lee and T.
Suda, "Adaptive load distribution over
multipath in MPLS networks". IEEE
International Conference on

Communications (ICC ' 03 ) , Anchorage ,
Alaska, Date: 11-15 May 2003 pp: 233-237.

[9] Butenweg, S., "Two distributed reactive
MPLS traffic engineering mechanism for
throughput opt'imization in Best effort
MPLS networks. In the Eighth IEEE
International Symposium on Computers and
Communications (ISCC'03). 379-384 vol.1,
30th June-3rd July 2003.

[10] Mohammed Aldasht, Julio Ortega,
Carlos G. Puntonet; Antonio F. Diaz, "A
Genetic Exploration of Dynamic Load
Balancing Algorithms", IEEE 2004.

[11] Shi, W.
Gburzynski, P.

Macgregor, M.H.
'Load Balancing For

Parallel Forwarding" Networking, leee/Acm
Transactions On , Aug. 2005.

[12] Sven Ingebrigt UUand, "High level load
balancing for web services", University of
Oslo, 20'^ May 2006.

[13] Srikanth Kandula, Dina Katabi,
Shantanu Sinha, Arthur Berger "Dynamic
Load Balancing Without Packet Reordering"
Acm Sigcomm Computer Communication
Review ,Volume 37 , Issue 2 (April 2007).

[14] Murugesan, G. and A.M. Natarajan,
"Adaptive granularity algorithm for
effective distributed load balancing and
implementation in multiprotocol label
switching networks". IEEE, International
Conference on Advanced Computing and
Communication (ADCOM'07) 18-21
December 2007, pp: 626-631.

[15] Dengyin Zhang, Zhiyun Tang and
Ruchuan Wang, "Automatic traffic balance
algorithm based on traffic engineering". J.
Network Syst. Manage., 14: 317-325.

[16] Ashwin Sridharan, Roch Guerin and
Christophe Diot, "Achieving near-optimal
traffic engineering solution for current
OSPF/IS-IS networks". In the IEEE/ACM
Trans. Network., 13: 234-247.

[17] G. Murugesan, A.M. Natarajan and C.
Venkatesh, "Enhanced Variable Splitting
Ratio Algorithm for Effective Load
Balancing in MPLS Networks" Journal of
Computer Science 4 (3): 232-238, 2008
ISSN 1549-3636 2008 Science Publications.

[18] Zhiruo Cao, Zheng Wang, Ellen
Zegura, "Hashing-based traffic splitting
algorithms for Internet load balancing",
Georgia Institute of Technology 1999.

[19] S. Ramabhadran and J. Pasquale,
"Stratified Round Robin: A Low

152

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

Complexity Packet Scheduler with
Bandwidth Fairness and Bounded Delay,"
Proc. Acm Communications Architectures
and Protocols Conf. (Sigcomm), Karlsruhe,
Germany, Pp. 239-249, Aug. 2003.

[20] S. Floyd and K. Fall. Promoting the use
of end-to-end congestion control in the
Internet. IEEE/ ACM Transactions on
Networking, Aug. 1999.

[21] Ahsan Habib, Bharat Bhargava,
"Network Tomography -based Unresponsive
Flow Detection and Control", Department of
Computer science, Purdue University, IN
47907-1398.

[22] M.Azath, Dr.R.S.D.Wahida banu,
"Load balancing in Internet Using Adaptive
Packet Scheduling and Bursty Traffic
Splitting", International Journal of
Computer Science and Network Security,
Vol.8, No.lO, Oct 2008.

1 53 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

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

Vol 8, No. 3, 2010

Laboratory Study of Leakage Current and

Measurement of ESDD of Equivalent Insulator Flat

Model under Various Polluted Conditions

N. Narmadhai

Senior Lecturer, Dept of EEE

Government College of Technology

Coimbatore, India

narmadhai @ get. ac. in

S. Suresh

PG Scholar, Dept of EEE

Government College of Technology

Coimbatore, India

suresh. sundararaju® gmail.com

Dr. A.Ebenezer Jeyakumar

Director (Academics)

SNR Sons Charitable Trust, SREC

Coimbatore, India

ebeyjkumar® rediffmail.com

Abstract — The phenomenon of flashover in polluted insulators
has been continued by the study of the characteristics of
contaminating layers deposited on the surface of insulators in
high voltage laboratories. This paper proposed the Equivalent
insulator flat plate model for studying the flashover phenomena
due to pollution under wet conditions even at low voltage.
Laboratory based tests were carried out on the model under AC
voltage at different pollution levels. Different concentrations of
salt solution has been prepared using sodium chloride, Kaolin
and distilled water representing the various contaminations.
Leakage current during the experimental studies were measured.
A conductivity measuring instrument (EQ-660 A) is used to
measure the conductivity of the salt-solution. Salinity and
Equivalent salt Deposit Density (ESDD) were calculated. Test
results in terms of conductivity and ESDD are plotted against salt
concentration and the relationship between the conductivity and
ESDD is examined.

Reported results of this preliminary study on the insulator
model simulates the distinctive stages of development of flashover
due to the pollution and it could be easily identified from the
contamination level of ESDD and from the magnitude of leakage
current.

Keywords-Conductivity, ESDD, Flashover, Insulator model.
Leakage Current, Salt solution.)

I. Introduction

Insulators used in outdoor electric power transmission lines
are exposed to outdoor environmental contaminations.
Contamination on outdoor insulators enhances the chances of
flashover. Depending on the nature and duration of exposure,
deposits of wind-carried industrial, sea and dust contaminants
build up on the insulator surface as a dry layer. The leakage
current path through a layer of dry contaminants on an
insulator surface is capacitive wherein the current amplitude is
small and sinusoidal. The dry contaminant layer becomes
conductive when exposed to light rain or morning dews. As
wetting progresses, the leakage current path changes from
capacitive to resistive with simultaneous increase in current
amplitudes. The increase in leakage current dries the
conducting layer and forms the dry bands around the areas
with high current density. These dry bands interrupt the
current flow and most of the applied voltages are impressed

across these narrow dry bands. If the dry bands cannot
withstand the voltage, localized arcing develops and the dry
bands will be spanned by discharges. The arcs merge together
and form a single arc, which triggers the surface flashover [1].

The contamination severity determines the frequency and
intensity of arcing and, thus the probability of flashover. In
favourable conditions when the level of contamination is low,
layer resistance is high and arcing continues until the sun or
wind dries the layer and stops the arcing. Continuous arcing is
harmless for ceramic insulators. The mechanism described
above shows that heavy contamination and wetting may cause
insulator flashover and service interruptions. Contamination in
dry conditions is harmless. B. F. Hampton investigated the
voltage distribution along the wet, polluted surface of a flat
insulating strip and the method of dry band formation, with
subsequent growth of discharges on the polluted surface [2].
Verma measured the peak leakage current and correlated the
current with the flashover voltage. He suggested that the
flashover is imminent if the leakage current peak exceeds
100mA [3]. Karady observes the same [4].

In practice, there are various contaminant types that settle
on outdoor insulators. These contaminants can be classified as
soluble and insoluble. Insulators located near coastal regions
are typically contaminated by soluble contaminants, especially
salt (or sodium chloride). Insulators located near cement or
paper industries are typically contaminated by non- soluble
contaminants such as calcium chloride, carbon and cement
dust. Irrespective of the type of contaminant, flashover can
occur as long as the salts in the contaminant are soluble
enough to form a conducting layer on the insulator's surface.
In order to quantify the contaminants on the surface of the
insulators, the soluble contaminants are expressed in terms of
Equivalent Salt Deposit Density (ESDD), which correlates to
mg of NaCl per unit surface area. Non- soluble contaminants
are expressed in terms of Non-Soluble Deposit Density
(NSDD), which correlates to mg of kaolin per unit surface
area.

Many researchers studied that the leakage current due to the
contamination level is the main cause for flashover.
M.A.M.Piah and Ahmed Darus [5] modelled the leakage

154

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

current in terms of conductivity and other environmental
parameters. I.R. Vazquez et al [6] have tested the non ceramic
insulators by non standard method where they expressed the
contamination level by ESDD. F. V. Topalis et al [7] studied
the critical flashover voltage of the insulator with the variation
of surface conductivity and ESDD. In Guan Zhicheng et al [11]
the maximum value of leakage current has the definite
relationship with the flashover voltage of the polluted insulator
used to express the pollution degree of insulator.

In this paper, flat plate model for studying the insulator
flashover phenomena has been presented. Experimental tests
have been conducted on the model under various polluted
conditions. The study on the leakage current is important in
order to diagnose the insulator condition by monitoring of the
leakage current. Here both ESDD and surface conductivity for
various quantities of contaminated salt deposits are measured
using the lEC standard [8].

II. Flat Plate Model and Experimental Setup

A. Equivalent Insulator Model

The simplified geometrical models equivalent to actual
insulator are being widely used for the purpose of flashover
analysis. Among these models, the basic flat trough model has
merited extensive attention in the context of pollution
flashover. So the proposed model, equivalent to standard disc
insulator made of an insulating glass material with two copper
terminals, one on cap and another at the pin. A simplified plan
of the insulator model is shown in Fig. 1

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

Vol 8, No. 3, 2010
current is shown in Fig 2. The slurry is poured so that it rolls
off uniformly in the trough.

6cni

-18»5cm

Fig. L Flat Trough Model

B. Experimental Setup

A proposed equivalent insulator trough model [1] of
dimension 1 8. 5x 0.6x 0.2cm is used for the contamination
flashover experiments. The principal application of this
equivalent model would be to help simulate as much as
possible the practical conditions of high voltage insulators in
the application of low voltage itself. In artificial testing, a
contaminant is usually substituted by a dissolved mixture of an
inert binder-Kaolin and NaCl salt. The inert binder is
supposedly non-conducting and the quantity of salt represents
the level of contamination. Contamination salt solution was
prepared for various NaCl values of 15g, 20g, 25g, and 30g.
The mixture, usually dissolved in distilled water is known as
slurry which is thoroughly mixed as per lEC standard [8].
Before coating, the trough is initially washed and wiped clean
and dry. The experimental setup to measure the leakage

©.

{■)-2,5-)ii:A)

Fig. 2. Experimental Setup

A test voltage of 230V, 50 Hz was applied across the
terminals and the leakage current is monitored through the
suitable measuring meter from the instant of application of
voltage till the formation of dry band. The dry band was
precisely located on the model. Its shape, contour of growth
and locations were physically measured. The test results either
in a flashover or a withstand. The conductivity and ESDD has
also been calculated from the deposited contaminations. The
contamination can thus be classified as light, medium or heavy
according to the lEC standard [9].

III.EsDD Calculation

In any insulator severity of pollution is characterized by
the Equivalent Salt Deposit Density (ESDD). The procedure
for calculating the ESDD [8] is as follows: After the test has
been completed, the deposits were collected by a small brush
from the contaminated plate and mixed with 1 litre of distilled
water to get the solution for specific area of the glass plate.
This process is repeated for the other samples of salt solutions.
The conductivity of each collected salt solution is measured
using a conductivity meter which is initially calibrated using
O.IN KCl solution. At the same time temperature is also
recorded. The conductivities at different temperature are
converted to 20° temperatures by the expression [8] as.

ijjo = [Je[l-&(^-20]l

(1)

Where,

9 is the solution temperature, °C

Oe is the volume conductivity at a temperature 9°C (S/m)

O20 is the volume conductivity at a temperature 20°C (S/m)

b is the factor depending on the temperature 6 as given in

Table I.

Table I values of b at different temperatures

e°c

0.03156

0.02817

0.02277

0.01905

The salinity Sg of the solution is determined by the following
expression [8] as.

155

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

S^ = (S.7tr,o)^

Finally, the equivalent salt deposit density can be determined
by the following expression [8] as,

ESDD =

Where,

SgXV

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

Vol 8, No. 3, 2010

(2) current caused the heating effect which leads to rapid dry band
formation and partial discharges across these dry bands. Due
to the higher resistance at pin and cap end the heat dissipated
in that location may be greater and therefore moisture dried
rapidly. After 10 minutes dry bands could not sustain the

(3) applied voltage cause the scintillations to occur which
ultimately leads to flashover. The scintillations have been
physically seen and captured using high speed camera.

V is the volume of the solution, cm
A is the area of the cleaned surface, cm^

According to lEC 60815, pollution site severity classification
are shown in Table II [9]

Table II Pollution Site severity (IEEE definitions)

Description

ESDD

Very Light

0-0.03

Light

0.03-0.06

Medium

0.06-0.1

Heavy

>0.1

IV. Artificial Contamination Test

A. Light Contamination

Insulators are mostly affected by flashover due to the
deposition of NaCl salt particles. Therefore the equivalent
model was uniformly sprayed with the slurry solution
consisting of 15g NaCl, 40g Kaolin and 1 litre of distilled
water. Leakage current started to flow on the surface of
insulator due to the pollution. It is observed that there is an
increase in leakage current magnitude when compared with
clean surface condition, which is mainly because of increase in
surface conductivity. Dry bands have started to form on the
polluted surface after reaching the maximum leakage current
of 42mA. The complete dried condition of pollution layer with
complete dry band was physically seen after 60 minutes
approximately. No flashover could be seen and the test results
in a withstand.

Similarly for 20g NaCl, the leakage current increased to
90mA and the time taken for the formation of dry band is
reduced compared to ISg.The test results in a Withstand.

B. Medium Contamination

Experiments were repeated for 25g NaCl. It is noticed that
the magnitude of maximum leakage current increased to
130mA. It is because that the current magnitude depends on
the level of contamination and the amount of moisture on the
insulator surface. The test results in withstand but the field
exceeds the withstand capability and it initiates the arc
discharge.

C. Heavy Contamination

Finally the insulator model is contaminated by 30g NaCl
solution and experiments were repeated in a similar way. Due
to the high contamination of NaCl the magnitude of leakage
current goes upto 220mA. The high magnitude of leakage

V. Results AND Discussions

A. ESDD

The correction conductivity, salinity and ESDD have been
calculated for various tests as per IEC-507. The measured
Salinity, ESDD and conductivities for 15g, 20g, 25g and 30g
of NaCl salt are shown in Table III.

Table 111 Values of conductivity, Salinity & ESDD using salt solutions

NaCl

<520

ESDD

0.114

0.0966

0.000439

0.0396

0.165

0.1398

0.000643

0.0579

28.5

0.205

0.1718

0.000795

0.072

0.315

0.2669

0.001252

0.1128

The correction conductivity and ESDD obtained from the
present measurements by varying the amount of salt
concentration is presented in Fig. 3.

0.4

20 25 50

Salt couceDtiatLoii in g/li-

Fig. 3. Variation of ESDD and Conductivity with salt concentration
0.1 i

o.os

0.04

O.OZ

0.0$

0.1»

D.17 0.26

CDDdiicii-^'iiy

Fig. 4. Variation of ESDD with conductivity

156

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

In Fig. 3, for the salt concentration from 15g to 30g, the
variation of conductivity and ESDD is almost linear. The
values of ESDD are also plotted versus the conductivity which
is shown in Fig. 4. It is observed that the relationship between
the conductivity and ESDD is linear [10].

B. Leakage Current

Table IV Experimental test result

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

Vol 8, No. 3, 2010
reach the critical value, the flashover is imminent
[3] [11].

Applied
voltage(V)

NaCl
Qty(g)

Leakage Current(mA)

Status of
Flashover

Initial
Value

Maximum
Value

230

Withstand

230

Withstand

230

130

Withstand

230

106

220

Flashover

Table V Test results for various pollution levels

NaCl
Qty(g)

Leakage Current(mA)

ESDD

(mg/cm^)

Pollution
Level

Initial
Value

Maximum
Value

0.0396

Light

0.0579

Light

130

0.0725

Medium

106

220

0.1128

Heavy

vi. Using the proposed model, the leakage current
were measured and found similar [3] [4] from
experiment. These revealed the equivalent model
with 230V supply could be used for flashover
prediction and analysis in the place of the
standard high voltage insulator.

VI. Conclusion

To simplify the mathematical analysis, tests were done on a
flat trough model of simple geometry. The maximum leakage
current and the conductivity of the contaminant solution on the
surface of the model and the corresponding ESDD of various
degree of pollution were also determined. The laboratory
model test results either in a flashover or a withstand. The
result also showed that the leakage current strongly correlated
with insulator polluted level and to assess the condition of the
insulation system which indicates the occurrence of a
flashover situation and the need for cleaning of the insulators
when it exceeds the medium level of ESDD.

Even though the model presented above still needs
modifications, the extent of study of leakage current
magnitude for the analytical case is satisfactory. At low
voltage, the model tests can be standardized easily; its use in
studying contamination flashover should be encouraged. In
this way, fairly accurate results could be obtained eliminating
the need of site testing.

The following points were observed from the results shown
above in Table IV and V.

i. For light contamination of 15g NaCl, the
maximum leakage current measured is 42 mA.
The calculated ESDD from the contamination
deposit is 0.0396 which indicates the light degree
of pollution.

ii. For 20g NaCl, maximum leakage current is
96mA which results in withstand and the
corresponding ESDD of 0.0579 also indicates the
low level of pollution.

iii. For 25g the leakage current reaches a peak of
130mA the field exceeds the withstand capability,
initiates an arc discharge and extends several
arcing which is actually preceded before the
flashover. ESDD value is 0.0725 shows the
medium level of pollution.

iv. For 30g the leakage current measured is 106mA
and reached a peak of 220mA. It results in flash
over after 10 minutes of wetting. ESDD shows
the heavy pollution level for 0.1 128.

V. The leakage current shows that the pollution
severity can be correlated with the surface
conductivity and ESDD. When these quantities

References

[1] Ebenezer Jeyakumar, "Development of verisimilar juxtaposition model
and study of physical phenomena on polluted insulators," PhD
Dissertation, Department of Electrical Engineering, Anna University
Madras, India, June 1991.

[2] B. F. Hampton, "Flashover mechanism of polluted insulation,"
Proc.IEE, vol.111, pp.985-990, 1964.

[3] M.P. Verma, "Die quantitative erfassung von fremdschichichteintlusse, "
ETZ-A97, pp. 281-285, 1976.

[4] George Karady, Felix Amarh, Raji Sundararajan, "Dynamic modeling of
ac insulator flashover characteristics, " 1 1th International Symposium of
High Voltage Engineering, Vol. 4, pp. 107-110, London, England,
August 1999.

[5] M.A.M. Piah, Ahmad Darus, "Modeling leakage current and electric
field behavior of wet contaminated insulators," Power Engineering
Letters, IEEE Transactions on Power Dehvery 19 (1) 432-433, January
2004.

[6] I.R. Vazquez, G.M. Tena, R.H. Corona, "Nonstandard method for
accelerated aging tests of non ceramic insulators, " lEE Proceedings -
Generation Transmission and Distribution 149 (4), 439-445, July 2002.

[7] F. V. Topahs, I. F. Gonos and I. A. Stathopulos, "Dielectric behavior of
polluted porcelain insulators, " lEE Proc.-Gener. Transm. Distrib., Vol
148, No. 4, pp. 269-274, July 2001.

[8] lEC 60507, "Artificial pollution tests on high voltage insulators to be
used in ac system, " Switzerland, 1991.

[9] IEC60815, "Guide for the selection of insulators in respect of polluted
conditions, " 1986.

[10] M. A. Salam, N Mohammad, Zia Nadir, Ah Al Maqrashi, A Al Kaf,
"Measurement of conductivity and equivalent salt deposit density of

157

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

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

Vol 8, No. 3, 2010
contaminated glass plate, " IEEE conference TENCON, Vol.3 pp 268-
270, 2004.

[11] Guan Zhicheng, Cui Guoshun, "A study on the leakage current along
the surface of polluted insulator," Properties and applications of
dielectric materials, Proceedings of the 4* international conference on
volume 2, 3-8 July, pp.495-498,1994

1 58 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

(IJCSIS) InternationalJournal of Computer Science and Information Security,

Vol. 8, No. 3, June 2010

SSL/TLS Web Server Load Optimization using
Adaptive SSL witli Session Handling Meclianism

R.K.Pateriya*, J.L.Rana^, S.C. Shrivastava^
^Department of Computer Science & Engineering and Information Technology

Maulana Azad National Institute of Technology, Bhopal, India
Emails: pateriyark@gmail.com , jl_rana@yahoo.com , scs_manit@yahoo.com

Abstract — Secure Sockets Layer (SSL) is the world standard for
web security, SSL provide authentication ,data integrity and ensure
message confidentiality using cryptography. This paper proposes
an approach for load management by applying Adaptive SSL
(ASSL) policy with session handling mechanism for enhancement
of the security and performance of the server , ASSL policy
negotiate session security at runtime by adapting more secured and
comparatively costly cryptographic algorithm at runtime if load is
under safe limit otherwise change to less secure algorithm ,Session
handling mechanism limit the active session running on the server
,This self-adaptive security policy offers great potential in providing
timely fine grained security control on server and therefore
enhance performance and security of e-commerce sites.

Keywords- Admission control,
Security, SSL Session,

E-commercCy Overload control^

I. Introduction

Security between network nodes over the Internet is
traditionally provided using HTTPS i.e. SSL (Secure Socket
Layer).It perform mutual authentication of both the sender and
receiver of messages and ensure message confidentiality. This
process involves certificates that are configured on both
sides of the connection. Although providing these security
increases remarkably the computation time needed to serve a
connection, due to the use of cryptographic techniques. The
disadvantage of the basic SSL approach is that security needs
to be preconfigured on only predefined static information,
namely the data and its location, that can be utilized when
making renegotiation decisions. Adaptive SSL extends SSL
beyond these limitations.

Adaptive security model was proposed in [1] which
provide appropriate security mechanisms for SSL sessions at
any particular moment in time as the environment changes.
The adaptation controller for SSL called Adaptive SSL.
concerned with adapting the choice of cryptographic
algorithms applied to client-server interactions. [2]
experimentally investigated the effects of security adaptation
in various client-server scenarios .

In [3,4,5,6,7,8] various session based admission control
(SBAC) mechanism for load management are discussed.
Admission control is based on reducing the amount of work
the server accepts when it is faced with overload. On most of

the prior work, overload control is performed on per request
basis, which may not be adequate for session-based
applications, such as e-commerce applications. Session

integrity, load management and security are the critical issue
for the success of e-commerce site.

This paper proposed a fi-ame work for enhancement of
server and e-commerce application performance through load
management using Adaptive SSL (ASSL) policy with session
handling mechanism .This Adaptive SSL facilitates runtime
adaptation of the SSL protocol by effectively intercepting
requests between the client and SSL module and based on a
set of conditions, may decide to renegotiate the session
security i.e. selecting appropriate cryptographic algorithm at
runtime. This self adaptive security will enhance performance
of server as it reduces some of the overhead related to
cryptographic algorithm used for secure connection.

The rest of the paper is organized as follows: Section 2
introduces the SSL protocol. Section 3 presents Overview of
session based admission control techniques. Sections 4
introduces ASSL and Section 5 define proposed approach and
Section 6 provide experimental environment detail and
section 7 presents the conclusions of this paper.

//.

SSL PROTOCOL

The SSL .protocol provides communications privacy over
the Internet. It uses combination of public-key and private-key
cryptography algorithm and digital certificates to achieve
confidentiality, integrity and authentication. The SSL protocol
increases the computation time necessary to serve a
connection remarkably due to the use of cryptography to
achieve their objectives. The study concludes that SSL
connections is 7 times lower than when using normal
connections as discuss in[7,8].

The four protocol layers of the SSL are Record Layer,
ChangeCipherSpec Protocol, Alert Protocol, and Handshake
Protocol ,they encapsulate all communication between the
client machine and the server.

159

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

HHP

SMTP

IMAP

flP

MM*

4lHtNK«l

AppllUllDI

Record Layer P^Dtocol

TCP

(IJCSIS) InternationalJournal of Computer Science and Information Security,

Vol. 8, No. 3, June 2010

handshake is negotiated when a chent estabhshes a new
HTTP connection with the server but using an existing SSL
connection. As the SSL session ID is reused, part of the SSL
handshake negotiation can be avoided, reducing considerably
the computation time for performing a resumed SSL
handshake. Notice that their is big difference between
negotiate a full handshake respect to negotiate a resumed SSL
handshake (175 vs. 2 ms) as discussed in[7,8].

SSL Connection

Fig 1 SSL Protocol
The SSL handshake allows the server to authenticate itself to
the client using public-key techniques like RSA and then
allows client and the server to cooperate in the creation of
symmetric keys used for rapid encryption, decryption.
Optionally the handshake also allows the client to authenticate
itself to the server.

l:ClientHello

2: ServerHello

3: Certificate (optional)

4: Certificate Request (optional)

■^

^ 5: Server Key Exchange (optional)
6: ServerHelloDone

Client

7: Certificate (optional)

"^ Server

8: Client Key Exchange

9: Certificate Verify (optional)
10: Change Cipher Spec

11: Finished

12: Change Cipher Spec

13: Finished

14: Encrypted Data

Fig 2 Handshake Protocol

Two different SSL handshake types can be distinguished:
The full SSL handshake and the resumed SSL handshake. The
full SSL handshake is negotiated when a client establishes a
new SSL connection with the server and requires the complete
negotiation of the SSL handshake, including parts that spend a
lot of computation time to be accomplished. The SSL resumed

This is a logical client-server link, associated with the
provision of a suitable type of service. In SSL terms, it must
be a peer-to-peer connection with two network nodes. Every
connection is associated with one session .

SSL Session

This is an association between a client and a server that
defines a set of parameters such as algorithms used, session
number etc. An SSL session is created by the Handshake
Protocol that allows parameters to be shared among the
connections made between the server and the client and
sessions are used to avoid negotiation of new parameters for
each connection. This means that a single session is shared
among multiple SSL connections between the client and the
server.

///.

OVERVIEW OF SBAC TECHNIQUE

Following section give comparative study of various
session based admission control mechanism for load
management as discussed in [3,4,5,6,7,8]

CPU utilization based implementation presented in [3,4] is
the simplest implementation of session based admission
control but can break under certain rates and not work
properly, reason is that the decision , whether to admit or reject
new sessions, is made at the boundaries of ac-intervals and
this decision cannot be changed until the next ac-interval.
However, in presence of a very high load, the number of
accepted new sessions may be much greater than a server
capacity, and it inevitably leads to aborted sessions and poor
session completion characteristics

Hybrid admission control strategy covered in [5] which
tunes itself to be more responsive or more stable on a basis of
observed quality of service. It successfully combines most
attractive features of both ac-responsive and ac-stable policies.
It improves performance results for workloads with medium to
long average session length.

Predictive admission control strategy also covered in [5]
which estimates the number of new sessions a server can
accept and still guarantee processing of all the future session
requests. This adaptive strategy evaluates the observed

160

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

workload and makes its prediction for the load in the nearest
future. It consistently shows the best performance results for
different workloads and different traffic patterns. For
workloads with short average session length, predictive
strategy is the only strategy which provides both: highest
server throughput in completed sessions and no (or, practically
no) aborted sessions.

Session-based adaptive overload control mechanism based
on SSL connections differentiation and admission control
presented in [7,8] prioritizes resumed connections maximize
the number of sessions completed and also limits dynamically
the number of new SSL connections accepted depending on
the available resources and the number of resumed SSL
connections accepted, in order to avoid server overload.

IV.

ADAPTIVE SSI

SSL protocol used to secure the communication channel
between an application or web server and a client. The basic
SSL secure transport layer connection, is established through a
handshake mechanism where algorithms are selected based on
those available to both the client and the server, for providing
the three security properties confidentiality, authentication
and data integrity. This process is commonly known as SSL
negotiation and the resulting connection is called a session.
Once established, the session can conduct a renegotiation.

'^M\yEmwm

1 Wftf't

Z^jsi

39L

IVi^U^

(IJCSIS) InternationalJournal of Computer Science and Information Security,

Vol. 8, No. 3, June 2010
The disadvantage of the basic SSL approach is that security
needs to be preconfigured only on predefined static
information, namely the data and its location, that can be
utilized when making renegotiation decisions.

Adaptive SSL extends SSL beyond these limitations.
Previous work on Adaptive SSL (ASSL) covered in [1,2]
facilitates runtime adaptation of the SSL protocol. ASSL aims
to provide appropriate security mechanisms for SSL sessions
at any particular moment in time as the environment changes.
Attributing environmental factors could include the threat
level, server load or transaction type. Client attributes such as
processing power, bandwidth or type of client can also be
considered. These factors are monitored by specialized third
party applications that inform ASSL of affected clients and
appropriate security measures. In [1,2] introduces a generic
design for controlling the renegotiation within SSL and the
resulting system called as 'Adaptive SSL'.
Design of Adaptive SSL

ASSL introduces a flexible approach to session
management where renegotiation logic is decoupled fi'om the
main SSL implementation and web server conflguration.
Separating these concerns enables us to build a more powerful
adaptive security model since the renegotiation rules can be
determined and deployed independently and parallel to the
web server and its components.

The ASSL module described in [1,2] effectively intercepts
requests between the client and SSL module and based on a
set of conditions, may decide to renegotiate the session
security. The set of conditions are specified and altered at
runtime by 'third parties' that is, other programs such as
firewalls, system performance monitors, network monitors,
server administrators, etc.. ASSL is used to create a self-
adaptive security solution based on fiexible use of
renegotiation in SSL.

Fig 3 Basic SSL Model

(enc=encrypted, req=request, resp=response, conf=configuration)

Fig. 3 taken fi-om [1,2] depicts a standard web server
request-response processing cycle during an SSL secured
session. The numbers in the figures indicate the event order
and the labels the interaction type. Events with the same
number indicate a decision point and only one of the events
take place. It shows the client sending an encrypted request to
the server in step 1. The request is initially passed to SSL
which then queries the web server's configuration file (where
the renegotiation decision rules are typically stored) and
decides either to process the request or renegotiate the session.

Fig 4 Adaptive SSL

Figure 4 taken fi-om [1,2] explain how ASSL handles SSL
requests. It effectively takes over the SSL negotiation and
renegotiation logic by intercepting requests and evaluating
them against the third party input. Decoupling the negotiation
logic in this way fi'om the main server logic, which is pre-
configured in the "Web Server Configuration" allows to build

161

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

a more powerful and flexible adaptive security model.
Adaptive Policy

Adapting security based on current system performance is
explored in [1,2] . In [1] it was shown that Adapting the
security, i.e. changing the security algorithm at runtime has a
server performance impact .It has investigated the different
aspects of client load patterns.

In [2] a policy is defined to trade-off security and
performance and shows how it performs in the scenario and
finally experiment shows that the performance implications
of adapting the security is highly dependant on the SSL
session length and the requested file size. In [2] it was shown
that performance cost incurred by the server through
encryption and decryption is significantly infiuenced by the
particular software implementation ,more costly algorithms
should result in lower server throughput i.e 3 DBS is the most
costly algorithm and is about 16 times more expensive than
RC4 which is the least secure .

Table 1 Average number of bytes processed per second

(IJCSIS) InternationalJournal of Computer Science and Information Security,

Vol. 8, No. 3, June 2010
server could cope with the same load at a higher security level.
When security is adapted only new clients start with the higher
level of security and so it takes some time before the server is
serving all clients at this new level .

Cryptographic
Algorithm

Throughput

RC4

112000

AES 128

36000

AES 192

30600

AES 256

26500

DES

19300

3DES

6900

Based on this fact following adaptive policy is discussed in[2]

Pseudo code

#comment

load = getCpuLoad

PolicyTable(row, column)

curSec #current Security level

newSec #new Security level

#Move to X

Move to PolicyTable(curSec,curSec)

#Check if security should be increased

#Check all values to the left

IF load < PolicyTable(curSec, curSec - n)

RECORD newSec

#Check if security should be decreased

#Check all values to the right

IF load >= PolicyTable(CurSec, CurSec - n)

RECORD newSec

IF newSec THEN adapt

#end

This adaptation policy is very robust to client behavior.
Firstly, security is reduced as soon as the server reaches its
maximum load.. Secondly, security is only increased if the

Application Area

Applications for adaptive security have been proposed in a
number of diverse areas. Examples include mobile ad-hoc
networks where current network conditions play a role in
choosing relevant security protocols at runtime . Media
streaming research proposes the use of additional contextual
information, in the form of an intelligent routing
infi-astructure, to allow for fiexible security levels . There have
also been proposals to provide adaptive system security
through system event monitoring .

This Adaptive SSL research is preparatory work to create a
self-adaptive security solution based on fiexible use of
renegotiation in SSL. Adaptive SSL policy is applied in
different realistic scenario and establish strategies and
optimizes the trade-off between security and performance
(overhead as well as server performance) when renegotiating
security level. The proposed work is discussed in following
section.

PROPOSED APPROACH

The proposed work utilizes session handling mechanism
with adaptive SSL policy for load management. This policy
provides more fiexible approach to session management where
renegotiation logic is decoupled fi'om the main SSL
implementation and web server configuration. Runtime
adaptation of cryptographic algorithm on online e-commerce
site based on current number of active session help in
increasing the number of completed session and enhancing
server and application performance.
Phases of proposed framework

Server Authentication

Session Monitoring

Application Of Adaptive SSL Policy

Testing of Server and Application Performance

Phase 1: Server Authentication

Security between network nodes over the Internet is
traditionally provided using Secure socket Layer .For
communicating through SSL, server authentication is
mandatory, the client tries to confirm the identity of the server
based on the server's certificate. To accomplish this, on
receiving a request fi-om the client, the server sends its
certificate to the client. This certificate contains information
such as: server's public key, certificate's serial number,
certificate's validity period, server's distinguished name.

162

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

issuer's distinguished name, and issuer's digital signature (a
message signed using the issuer's private key). The chent, on
receiving this certificate, authenticates the server..

For performing the above authentication, the server must
have a Pubhc Certificate and Key store containing the key. A
certificate can be self-signed when authentication over the
internet is not really a concern, that is only data privacy and
integrity are important .So self signed certificate is generate or
certificate request is sent to CA for trusted certificate .So our
work starts with generation of private and public key and self
signed certificate and export it into server configuration file
and hence configuring secure socket layer on server.

Phase 2: Session Monitoring

Session integrity is the critical metrics for e-commerce. So
load management through session handling is performed
which will enhance application performance. Firstly create a
web application and deploy it on server. Then monitor the
number of active session on that application and check against
maximum limit as new session is created, if the maximum
limit is reached , then do not allow the new session..

Pliase 3 : Application Of Adaptive SSL Policy

The proposal is to design a system for client server
interaction which will monitor active sessions on the server
and when the number of session reaches to alarming level then
change highly secure cryptographic algorithm used for https
communication to the algorithm which is less secure and
consume less time in encryption and decryption process. This
approach help in reduction of some load on the server due to
less time consumption during cryptographic process. We are
using security trade off between 3 DBS and RC4cryptographic
algorithm. 3DES is more secured but gives least throughput
and RC4 is the weak steam cipher but with good throughput
capability.

Triple- Data Encryption Standard (3DES)

Triple-DES Using standard DBS encryption, encrypts data
three times and uses a different key for at least one of the three
passes giving it a cumulative key size of 112-168 bits. The
Triple-DBS is considered much stronger than DBS, however,
it is rather slow compared to some new block ciphers hence
give poor throughput as compared to other cryptographic
algorithm.

RC4 Stream Cipher

RC4 is a cipher invented by Ron Rivest, co-inventor of the
RSA Scheme.RC4 has two advantages over other popular
encryption algorithms. First, RC4 is extremely fast. Second,
RC4 can use a broad range of key lengths. For most ciphers,
longer key length is better. However, RC4 was widely used
primarily because its shortest optional key length is 40 bits.
Unfortunately, RC4 is a dangerous cipher to use. If it is not
implemented perfectly, its protection is minimal.

(IJCSIS) InternationalJournal of Computer Science and Information Security,

Vol 8, No. 3, June 2010
Phase 4: Testing of Server and Application Performance.

This phase will analyze our approach and show its
importance in client /server paradigm. The main aim is to
compare the throughput and response time of the application
under different load pattern and in two different environment ,
which uses different cryptographic algorithm for security. This
phase also provides visualization of the various performance
metrics which can be utilized for further result analysis. In
this phase there are two measures for evaluating performance
(i) Throughput and (ii) Response Time

VI.

Conclusion

This proposal of runtime adaptation of cryptographic
algorithm using session handling mechanism will enhance
throughput and response time of server . Session handling
combines with Adaptive SSL is good alternative for load
management. This work will provide self adaptive security
control on server and optimizes trade off between security and
performance.

ACKNOWLEDGMENT

The Success of this research work would have been
uncertain without the help and guidance of a dedicated group of
people in our institute MANIT Bhopal. We would like to
express our true and sincere acknowledgements as the
appreciation for their contributions, encouragement and
support. The researchers also wish to express gratitude and
warmest appreciation to people, who, in any way have
contributed and inspired the researchers.

REFERENCES

[1] C. J. Lamprecht , Aad P. A. van Moorsel, "Adaptive SSL:

Design, Implementation and Overhead Analysis", prdc,

pp.289-294. First International Conference on Self-Adaptive

and Self Organizing Systems (SASO '07), 2007.

[2] C.J. Lamprecht, Aad P. A. van Moorsel, "Runtime Security

Adaptation Using Adaptive SSL," prdc, pp.305-312, 14th

IBBB Pacific Rim International Symposium on Dependable

Computing, 2008

[3] L. Cherkasova, P. Phaal "Session Based Admission

Control: a Mechanism for Improving the Perfor-mance of an

Overloaded Web Server." HP Laboratories Report No. HPL-

98-119, June, 1998.

[4] L. Cherkasova, P. Phaal "Session Based Admission

Control: a Mechanism for Improving Performance of

Commercial Web Sites." prcd. Seventh International

Workshop on Quality of Service, IBBB/IFIP event, London,

1999.

[5] L. Cherkasova, P. Phaal "Session Based Admission

Control: a Mechanism for Peak Load Management of

Commercial Web Sites." IBBB J. Transactions on Computers,

Vol. 51, No. 6, June 2002.

[6] M. Arlitt, "Characterizing Web User Sessions", ACM

SIGMBTRICS Performance Bvaluation Review, Vol. 28, No.

2, pp. 50-56, September 2000.

163

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

(IJCSIS) InternationalJournal of Computer Science and Information Security,

Vol. 8, No. 3, June 2010
[7] Jordi Guitart , David Carrera , Vincenc Beltran , Jordi
Torres , Eduard Ayguade, "Session-Based Adaptive Overload
Control for Secure Dynamic Web Applications" Proceedings
of the 2005 International Conference on Parallel Processing,
pp.341-349, 2005.

[8] Jordi Guitart , David Carrera , Vicen9 Beltran , Jordi
Torres , Eduard Ayguade, "Designing an overload control
strategy for secure e-commerce applications," Computer
Networks: The International Journal of Computer and
Telecommunications Networking, v.51 n.l5, p.4492-4510,
October, 2007

[9] P. Lin, "So You Want High Performance" (Tomcat
Performance), September 2003, online available URL:
http ://j akarta. apache . org/tomcat/articles/performance .pdf
[10] Sun Microsystems, "Java Secure Socket Extension
(JSSE)." online available URL:

http://j ava.sun.com/products/j sse/.

[11] Jakarta Project. Apache Software Foundation, Jakarta
Tomcat Servlet Container. URL:

http://jakarta.apache.org/tomcat.

[12] A.O. Freier, P. Karlton, C. Kocher, "The SSL Protocol
Version3.0" November 1996. available online URL:
http://wp.netscape.com/eng/ssl3/ssl-toc.htm

R K Pateriya MTech & B.E. in Computer
Science & Engg. and working as Associate
Professor in Information Technology
Department of MANIT Bhopal . Total 17 Years
Teaching Experience ( PG & UG ). Guided
twentv MTech Thesis .

Dr. J. L . Rana Professor & Head of Computer
Science & Engg deptt. in MANIT Bhopal .He
has received his PhD from NT Mumbai & M.S.
from USA (Huwaii) .He has Guided Six PhD.

Dr. S. C. Shrivastava Professor & Head of
Electronics Engg. department of MANIT
Bhopal. He has Guided three PhD , 36
M.Tech and presented nine papers in
international & twenty papers in national
conference in India.

164

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

An Enhancement on Mobile TCP Socket

S.Saravanan,
Research Scholar
Sathyabama University
Chennai, India

Dr.T.Ravi

Professor & Head, Dept. of CSE
KCG College of Technology,
Chennai, India

Abstract - A TCP session uses IP addresses (+ IP port) of
both end points as identifiers. Therefore when a mobile
handover to a new AP that belong to a different
subnet/domain, the IP address will changes and ongoing
TCP connections are reset. Several approaches have been
proposed to solve this problem, and one of which was to
modified the TCP/IP stack to update the changes of the IP
address for the ongoing connections [5] [6]. However, these
proposals causes unnecessary processing when TCP is used
in applications which have already employed some kinds of
security measures, such as SIP. This paper proposes the
Mobi Socket, which specifically supports TCP mobility for
intrinsic secure applications without unnecessary overhead.

1. INTRODUCTION

TCP/IP was developed when all network nodes
were stationary, and connection to a network is through
cable, therefore it is unthinkable that a node will move to
another subnet while ccessing to the Internet, and the IP
address of an end host is assumed to stay unchanged while a
computer is running. As a result, IP address (together with IP
port) is used as identifiers for TCP session, and the TCP layer
at the end host maintains TCP control blocks (TCBs), which
hold the IP addresses and IP ports of both ends for each TCP
connection to fmd the right socket for each datagram it
receives from the IP layer.

But with the introduction of wireless access
technologies such as Wi-Fi, it is possible for a mobile node to
handover to a new AP that belong to a different
subnet/domain while actively connecting to the Internet. This
causes an IP address change, and for current implementation
of TCP/IP stack, all ongoing TCP connections are reset. This
will cause problem for long-live TCP sessions.

There are two general approaches to solve the
problem of changing host IP address for TCP session. The
first one uses the split-connection approach, which introduces
a fix middle agent between the mobile host (MH) and
correspondent host (CH) [4]. The connection between CH
and MH is broken into two parts, the fixed part between the
CH and the agent remains unchanged regardless of the
position of the MH, and the TCP connection between the
agent and the MH will be re-established whenever the MH
handovers to a new address. In this sense only the TCP at the
MH is affected, while at the CH the TCP session is not
disturbed. The problems of these approaches are non-
transparent end-to-end operation of TCP session, as well as
the requirement of new infrastructure entities (the middle
agent) and triangle overhead.

The other approach modify the TCP stack so that when
the mobile host changes the connection to the internet, the
TCP stacks at both ends preserver the TCP connection and
update the TCBs with the new IP address at both ends
accordingly.

In [5], when the MH changes its location, the proposal
in [5] introduced new states to the TCP specification. When
the address of MH changes, MH and CN will exchange
information and update the new IP address accordingly.
Both sides will prepare in advance a share- secret, and use
this sharesecret to authenticate each other during the update
process.

The proposal in [6] employs a similar concept, but
instead of changing the TCP stack, it uses kernel extensions
and a userlevel redirect daemon process (this was the design
of the prototype in BSD). The daemon process will monitor
the wireless network interface for changes of IP address,
and if one is detected, the daemon at the MH will inform the
counterpart at the CH to update the new IP address together.
To secure the update process from malicious acts, MH and
CH also need a share- secret in advance.

The problem with [5] is that both sides has to perform
additional works to exchange a share- secret in advance,
regardless of whether the MH will actually performs the
handover to a new Access Point (AP) or not, or whether the
TCP session lives long enough to experience a handover.
The proposal in [6] relieves this matter by initiating the
preparation process only if the TCP connection exists longer
than a threshold. However, if the MH does not perform a
handover, then all of the preparations for the long-live
connections are wasted.

One more problems with [5] and [6] is that processing
the share- secret for authentication will requires a lot of
processing, which in tum consumes battery power at the
MH. If many TCP connections are used (such as if the user
constantly browsing the Internet) then battery life will be
shortened considerably. Moreover, both [5] and [6] are not
applicable in the case where both ends perform handover
simultaneously.

In the next parts of this paper, we propose a new type of
socket called the TCP MobiSocket, that remains connected
even if the concerned IP address changes. It works like
normal TCP socket, but does not get reset when the IP
address at either end changes, and with an additional
updateTCBO member function to update the TCBs with the
new address. All of the security issues that are required to
secure the update of the new address will be handled by the
calling applications. This new socket is dedicated to support

165

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

mobile TCP session for intrinsically secure application,
without all the above mentions problems of [5] and [6].

2. DESIGN OF THE MOBISOCKET

Logically, there are two phases when mobile device
handover. First, the Network interface/card disconnects from
the old AP. Then it connects to the new AP. In traditional
TCP stack, the network stack at the MH will close all TCP
connections in cleaning-up activities, as well as reset the
TCBs during these phases.

On the other hand, all of the ongoing MobiSocket will
remain in ESTABLISHED state, when the IP address
changes, waiting to be updated by the application.

We design a new socket that allows the application to
update the change of PoA at both end hosts. The socket is
designed based on the following assumptions/requirements:

• There are cases when the TCP connection needs explicit
handling before communicating using the new IP
address (Re-establishment/update of Security
Association for VPN, sending the PATH message of
RSVPforQoS, etc... )

• The application takes care of security activities regarding
the update of the new address. The reasons for this are
(1) if the connection needs to be secure, the applications
have already shared some kind of security, and (2) if the
connection is not important to the extend that it requires
a shared security association between both end host, then
it might not important enough to be hacked by others.

• Compatibility with applications using legacy TCP/IP
stack is desired to promote deployment. It means that in
the case the other end does not support the features of
mobile socket, connection will work according to that of
legacy TCP specification.

• Being able to provide handover of TCP session between
different network interfaces of the same mobile device.
The requirement is that not only IP address but change
of TCP port also must be supported, because the same
port of the other network interface might be in use by a
different application at the time of the request for
handover.

The application which uses the MobiSocket will call the
MobiSocket's updateTCB() member function to update the
TCB with the new destination address. To satisfy the above
requirements, the mobile socket will provide the following
APIs to the applications:

• acvMobi(socket_id)

socket id: the handler of the socket

■ The application will call this function to explicitly
activate the mobile feature of the socket

■ If this function is not called, then the MobiSocket
will work like normal TCP socket

■ When this function is called, the TCP connection
will not be abolished if the concerned wireless
interface changes to a new IP address

• updateTCB(socket_id, direction, newIPaddress,

newPort)

socket id: handler of the socket

direction: update the source or the destination address

newIPaddress, newPort: the new IP address and new port

to update to TCB/PCB (TCP Control Block/Protocol

Control Block). If the port is then keep the existing port

value

■ The application will call this function to update the
TCBIPCB (TCP Control Block/Protocol Control
Block) with the new source/destination address and
port

■ The MobiSocket will start a new congestion
control algorithm called the mobile congestion
control

• copyTCB(new_socket_id, old_socket_id)
old socket id: handler of the old socket

new- socket- id : handler of the new socket

■ The application will call this function to update the
TCBIPCB (TCP Control Block/Protocol Control
Block) of the newly created socket with the
information of the old socket. This is used when
the application want to handover from old interface
to new interface.

■ This function will copy all information of the old
socket (include current states, CWND, AWND,
RTO etc., except the source IP address and source
Port) to that of the new socket, and then delete the
old socket without sending FIN to the other end
(i.e., application at the CH).

Apart from the above two new APIs/member
functions, the MobiSocket also introduces two new
message, the AddChange and AddConfirm.

The AddChange contains (I) A shared token between
Mobile Host (MH) and Correspondent Host (CH), (2) the
old IP address and the new IP address encrypted by the
private key of the MH, (3) the new port address and (4) The
old IP address of the MH in plain-text.

The AddCorifirm contains (1) the shared token between
MH and CH, (2) the new IP address encrypted by the
private key of the CH.

If the two messages above are implemented as TCP
header options, then these header options must be sent to the
applications, but currently there is no mechanism to perform
such action. Therefore, it might be better to send this as
OOB (out-of-band) data using the TCP Urgent Pointer .

3. WORKING PROCEDURE OF THE MOBISOCKET

Let's consider the use of MobiSocket for a SIP
application. Suppose that a TCP connection is established
between MH and CH (the thick, solid line), which have
established a SIP session through the SIP server. The
MobiSocket will work as follows (see figure 1):

• First the application creates the TCP socket for the SIP
session, and calls the acvMobi () to activate the mobile
feature for the socket

166

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

In step <1>, the MH moves from Subnet 1 to Subnet 2,

and in the process its address change from IPaddress 1 to

IPaddress2

In step <2>, the SIP appHcation at the will call the

updateTCB () function to replace IPaddressl with

IPaddress2 at the TCB table. Then it issues a SIP INFO

message to ask CH update the new IP address of the

MH.

Upon receiving this INFO message in step <3>, the SIP

application at the CH will authenticate the message

using SIP security associations, and all the updateTCB ()

function to replace IPaddressl with IPaddress2 at the

TCB table.

Then in step <4> the SIP application at the CH will send

back the INFO message back the the MH to confirm the

change of address. Note that both INFO messages may

contain other parameters of the concerned TCP session,

such as new window size, MSS etc ...

CH and MH will start sending data using the TCP

connection when they receives the INFO message from

the other end, and they will start receiving data after they

send the INFO message to the other end.

TCB table

IP Src: MN's address 1

IP Dest: CN address

TCB table

IP Src: MN' address 2

IP Dest: CN address

MH
@ipaddress

TCB table(before calling

updateTCB)

IP Src: M N 's address 1

IP Dest: MN's IP addressl

O c^

TCB table (after calling

updateTCB)

IP Src: MN's address 1

IP Dest: MN's IP address2

Figure 1 : Working procedure of the mobisocket .

4. DISCUSSION AND OPTIMISATION
The merits of the MobiSocket are:

• Inherit intrinsic security feature of SIP

■ Less processing overhead for security issues
(conserve power)

• Depending on the security requirements, the application
can decide whether to allow the handover of TCP
connection

■ More suitable for application with strict
security requirements

• Still work when both ends handover simultaneously

■ Reach-ability through SIP Registration
functionality

We can further optimize the operation of the MobiSocket
as follows:

When the MN receives the INFO message from CN, both
ends might already time out (due to handover, NOT due to
congestion), so even if the TCB is updated, no data
exchange is possible until the time out is over (can be very
long). We can provide a new function to reset the timer after
the updateTCB O function, which is the reset'I'imensocket
id). This function will reset the TCP socket to the state as if
it has just received a datalACK packet from the other
machine.

Furthermore, if SIP proxy is used, then normally the MN
has to finish re-Registration with the SIP proxy first before
it can send SIP INFO message to the other end. This creates
further delay for the TCP session. To solve this, we note
that the MN and CN can share public key with each other
during the initial INVITE process, therefore after the MN
handover to a new IP address, it can use the public key of
the CN to send the SIP INFO message to the CN right away.
However, this solution cannot be used if both ends handover
simultaneously (therefore they do not know the IP address
of each other), in this case they must contact through SIP
proxy server (after the re-Registration process)

5. CONCLUSION

In this paper we propose the MobiSocket to support
TCP mobility for secure application such as SIP. This
socket causes no overhead if handover does not take place
like previous proposal, and moreover it still works when
both side handover simultaneously.

In this socket, there is no need for per-TCP connection
authentication, because the authentication is left to
application. Depending on the real situation, the application
can also control whether to keep the TCP session or not,
which is more appropriate for application which is applied
with other application level constrains such as security and
QoS policy .

In the future, we would like to carry out the
implementation of the MobiSocket to confirm the design of
the system, as well as to measure the delay and throughput
parameter when the reset'Iimen) function is (1) called and
(2) not called, and compare the results with that of [5] and
[6].

167

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

We also would like to measure the delay in the case of
SIP application, when we send the INFO message before and
after re-Registration, as well as when two end hosts handoff
together.

We also plan to update the proposal in [1] with this
new MobiSocket.

REFERENCES

[I] Vu Truong Thanh, Yoshiyori Urano, "Agent based LLMA

handover scheme for SIP communication - The case for UDP

traffic" , The II th International Conference on Advanced

Communication Technology (rCACT), Feb. 2009

[2] C. Perkins, "IP Mobility Support for IPv4", Request for

Comments: 3344, IETF, August 2002

[3] Huei-Wen Ferng et. ai, "A SIP-Based Mobility Management

Architecture Supporting TCP with Handover Optimization", Proc.

Of Vehicular Technology Conference, pp. 1224-1228, Apr. 2007

[4] Milind Buddhikot et. ai, "MobileNAT: a new technique for

mobility across heterogeneous address spaces", Proc. the 1st ACM

international workshop on Wireless mobile applications and

services on WLAN hotspot, pp. 75-84, Sept., 2003

[5] FUNATO D., "TCP-R: TCP mobility support for continuous

operation", Proc. IEEE International Conference on Network

Protocols, pp.229 -236 ,Oct. 1997

[6] Vassilis Prevelakis and Sotiris loannidis, "Preserving TCP

Connections Across Host Address Changes", Lecture Notes in

Computer Science, Springer Berlin / Heidelberg, pp. 299-310 Oct.,

2006

[7] Rosenberg, et. al., "Session Initiation Protocof, Request for

Comments: 3261, IETF June 2002

[8] D. Yon et. ai, "TCP-Based Media Transport in the Session

Description Protocol (SDP)", Request for Comments: 4145, IETF,

September 2005

AUTHORS PROFILE

S. Saravanan B.E., M.E., (Ph.D) working
as an Assistant Professor at Jeppiaar
Engineering College, Chennai and he has
more than 9 years of teaching
experience. His areas of specializations
are Computer Networks, Network
security and TCP/IP.

Dr. T. Ravi,B.E,M.E,Ph.D is a Professor
& Head of the Department of CSE at
KCG college of Technology, Chennai.
He has more than 18 years of teaching
experience in various engineering
institutions .He has published more than
20 papers in International Conferences
& Journals.

168

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

Technology in Education

Modern Computer Graphics Technologies Used at
Educational Programs and Some Graphical output screens

^N. Suresh Kumar, ^D.V. Rama Koti Reddy, ^S. Amarnadh, "^K. Srikanth, ^Ch. Heyma Raju,

^ R. Ajay Suresh Babu, ^K. Naga Soujanya

1,3,4,5 QYY^ GUAM University, Visaldiapatnam

^ College of Engineering, Andhra University, Visakhapatnam

^Raghu Engineering College, Visakhapatnam

^ GIS, GUAM University, Visakhapatnam

^nskgitam2009@gmaiLcom

Abstract In This paper a new technique is implemented to
teach microprocessor and to clarify the doubts in the
subject microprocessor. Although a lecturer has many aids
to explain the topic in the class room, but a graphical
environment is more power full environment in the present
education scenario, which can improve the student level of
understanding. The graphical interface develops the
concepts of the student graphics concepts and also a student
can easily grasp any level of task. The lecturer shows a
visual object or animated show to the student to explain the
particular topic in the class room. This work includes the
framework of graphics programming; students can
concentrate on the technical subject. Thus they acquire a
method to construct computer graphics programs in many
ways and gain knowledge in the concerned technical paper.
The project have used for six years, and convinced of the
positive effect.

I Introduction

E-leaming substantially improves and expands the
learning opportunities for students [3]. The modem
computer information technologies, which are widely
used both at educational programs for conducting of
effective lecture, conducted scientific researches, and
forming of practical and laboratory works with the
students of technical and computer-based special[2].
Every teacher comes to understand that successful
imparting of information and skills lies in the ability
to incorporate a variety of technologies that, directly
or indirectly, help communication between student
and teacher [4]. Advances in learning objects and
other technologies "that will optimize interoperability
with other institutions and organizations in areas such
as the creation of learning objects databases,
information databases such as libraries,
administrative systems and learner support strategies
as well as the facilitation of interactions among
learners and teachers", will continue to expand the
scope of possibilities with which educational
institutions will have to grapple [5].

II Experimental Idea

The project was to develop software delivered
diagnostic and remedial system for Diploma and
Graduate students taking subjects in microprocessors
and embedded systems. The system is comprised of a
module to diagnose a student's background

microprocessor knowledge and modules to which the
student is directed to remedy any deficiencies that are
found. The remedial modules focus on common areas
of weakness in microprocessor architectures and
interfacings and revise the basic peripherals that are
needed to interface. For example, in memory
interfacing the number of address lines and
microprocessor type are incorporated with the graphic
tools such as colors, sounds, messages and
animations as shown in figure 1, 4, 5. When a student
select any part on the selected architecture it will link
to the particular file which give the details of the unit
in the architecture as shown in figure 2, 3. The
Interfacing devices like 8255 and it modes can be set
in control register (CR) by an interactive message box
as shown in the figure6. The software also
demonstrates some of the bio-Medical applications
and some interfacing designs such as traffic light
controllers, stepper motor controllers, seven segment
interfacing etc. This software allows the user to
develop any interfacing circuit interactively as shown
in figure 4. For maximum accessibility and ease of
implementation, the software was designed to run in a
Windows environment using graphics application in
C language. This allows the software to be used off-
campus, and makes the system equally accessible to
students who are studying by distance education,
open learning or who are located on an overseas
campus [1].

Figure 1 8086 Architecture

169

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

Figure 2 Segment address

Figure SAccess rights and Permissions

Figure 4 Dialog box permits to set the memory
settings by user

1 'l!S3!S5S:KfSS2

1 1

1 M M 1 1

1 1 — ^ LSB of port C rtct« if bit is'0'->0/V or 'I'->ixp

' > port B Acts if bit is^V->a^ or 'I'-M^t.

> none Oporation for enOUP-B if bit i. ■V-^moOmO or 'l--;^»ci>

YOO UMIT TO IMTI

1 ^ port A Acts if bit is'O'-X.^ or 'I'->i^

Figure 6 8255 CR msg box which allow the
user to set the mode operations

Figure 5 Memory Interfacing

Figure 8 Bio medical Application, Heart Function

Deficiencies in background knowledge and skills
need to be diagnosed early before they become
barriers to students' further learning in
microprocessor subjects, it is not possible to offer
individual tuition. The problem is exacerbated with
off campus students taking Distance Education or
Open Learning subjects. The software also help
teaching staff relieved from from his duties for some
time where he can attend other problems and enabling
them to concentrate on presenting the primary subject
material. The anticipated outcome will be improved
student attitudes, and lower dropout and failure rates
in a subject area which is difficult for many. Wade
and Power [6] suggest the following "General
Requirements for WWW Based Listructional
Design:"

170

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

1) The presentation of material should support a
range of sensory experiences incorporating
interactivity and multimedia elements.

2) Students should be provided with the opportunity
to experiment with the knowledge they have
learned.

3) Testing and checkpoints are important from the
point of view of repetition and student retention.

4) Educational software should motivate the student.

5) The learning environment should support the
cognitive structures of the student;

6) Facilities for synchronous communication and
collaboration should be supported where possible.

7) A well-designed interface will enable the student
to interact with the material without the complex
intermediaries and will aid in the understanding of
the knowledge domain and structure.

8) The development of a Tele-Educational course
requires the support and cooperation of faculty
and administration.

9) WWW-based educational courses must be
integrated into a well understood and explicitly
specified curriculum this includes clear objectives,
content description, methods of teaching, student
learning, and student assessment and course
evaluation.

Li the present paper the work includes two stages
Stage 1:

• Develop soft ware to diagnose the student

• Develop assignments

• Seek the student feed back

• Ask the student to create his own
assignments

• Conduct quiz and descriptive test
Stage 2:

• Make interactive session to assess the
student

• Seek the student feed back

• Propose alterations and development in the
software if any

Preferably, these procedures are conducted early
enough at the end of each unit or chapter in the
semester to allow weak students to take remedial
action before being challenged. Conducting these
stages most of the student requirements for WWW
Based Listructional Design are satisfied which are
proposed by wade and power [6].

Ill Features

• The structure of the software especially the
presentation, animation, colors, fonts and
sound affects made interest in students to
diagnose him.

• The software structure is very interactive
and attractive.

• By further development of the modules in
the soft ware, it can use for research and real
time application problems.

IV Requirement

Our software package is developed in graphics using
C - language in Turbo C++ editor. The software is
developed on Microsoft window - 95 based system.
The main aim is, in poor countries low economic
colleges and lagging in new technologies can also use
this software. The systems require a graphic card and
program is developed for screen resolution 640X480.

V Conclusion

On the basis of comparative analysis the software
packages is marked as versatile direction in
conducting lectures, lab work and diagnose weak
student performance. E-leaming substantially
improves and expands the learning opportunities for
students. The modern computer

information technologies, which are
widely used both at educational
programs for conducting of effective
lecture, conducted scientific researches,
and forming of practical and laboratory
works

References

[1] Max King, Ian Kirkwood, Clive McCann, Department of
Econometrics, Monash University," The soft(ware) approach
to learning problems in quantitative methods" 0-7803-3173-
7/96/$5.00 @ 1996 EEE

[2] Vasyl Zayats, Vasyl Kogut, "Role of Information Technologies
in Progress of Science and Education" MEMSTECH'2009, 22-
24 April, 2009, Polyana-Svalyava (Zakarpattya), UKRAINE.

[3] Robert S. Friedman and Fadi P. Deck, "Innovation and
Education in the Digital Age: Reconciling the Roles of
Pedagogy, Technology, and the Business of Learning" IEEE
Transactions on engineering management, VOL. 50, NO. 4,
NOVEMBER 2003, pg 403

[4] R. Friedman and F. P. Deck, "Problem-based learning and
problemsolving tools: synthesis and direction for distributed
education environments," J. Interact. Learn. Res., to be
published

[5] F. P. Deck, M. Deck, and R. Friedman, "The virtual classroom
experience: Viewpoints from computing and humanities," J.
Interact. Learn. Environ., vol. 7, no. 2/3, pp. 113-136, 1999.

[6] V. Wade and C. Power, "Evaluating the design and delivery of
WWW based educational environments and courseware," in
Proc. ACM-ITiCSE, Dublin, Ireland, 1998, pp. 243-248.

171

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

Impact of language morphologies on Search
Engines Performance for Hindi and English

language

Dr. S.K Dwivedi

Reader and Head, Computer Science Dept.

BBAU

Lucknow, India

skd200@yahoo.com

Parul Rastogi

Research Scholar, Computer Science Dept.

BBAU

Lucknow, India

parul_rastogi@yahoo.com

Abstract - Search Engines are the basic tools of
Information Retrieval on the web. The performance of
the search engines are affected by various morphological
factors of the language. The paper covers the
comprehensive analysis and also the comparison of the
impact of morphological factors and other language
structure related factors (like sense ambiguity, synonyms)
on the performance of Hindi and English language
information retrieval on the web.

Keywords: Hindi language morphology; English language
morphology; web searching; morphological structure;
precision; Hindi language Search Engines.

I. INTRODUCTION

Hindi language is the national language of India,
with roughly 300 million native speakers. Another 100
million or more use Hindi as a second language. It is
the language of dozens of major newspapers,
magazines, radio and television stations, and of other
media.

In Hindi language, nouns inflect for number and
case. To capture their morphological variations, they
can be categorized into various paradigms based on
their vowel ending, gender, number and case
information.

Hindi language Adjectives may be inflected or

uninflected, e.g., '■cl<Hcp|cHr {chamkiila} (shiny).

'3T^^t5T {acchaa} (nice), 'et^ {lambaa} (long)
inflect based on the number and case values of their

Rajesh Kr. Goutam

Research Scholar, Computer Science Dept.

BBAU

Lucknow, India

Rajeshgoutam82@gmail.com

head nouns while '^t^ {sundar} (beautiful), ^^^U^

{bhaarii} (heavy) etc. do not inflect. Hindi Verbs
inflect for the following grammatical properties
Gender, Number, Person, Tense, Aspect and Modality
(GNPTAM).

The morphemes attached to a verb along with their
corresponding analyses help identify values for
GNPTAM features for a given verb form [1]. Hindi
language has core distinctive structure which affects
the results of the search queries on the web.

English is also the popular language in India and
across the globe. Though, the morphological structure
of English language is not complicated as Hindi but it
also follows some of the similar morphological
constraints.

Morphology is the study of the structure of words.
The structure of words can also be studied to show
how the meaning of a given morpheme, or its relation
to the rest of the word, varies from one complex word
to another. Consider how sun works in the following
words: sunbeam, sunburn, sundial, sunflower,
sunglasses, sunlight, sunrise, sun-spot (scientific
sense), sun-spot (tourist sense), suntan. Inflection does
not really yield "new" words, but alters the form of
existing ones for specific reasons of grammar.
Derivation, on the other hand, does lead to the creation
of new words.

Sometimes a word is changed in its form to show the
internal grammar of a sentence. Examples would be
plural forms of nouns (dog + s ^^ dog-s) or past
(imperfect) tenses of regular verbs (want + ed ^^ want-

172

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

ed). The study of such changes is inflectional
morphology (because the words in question are
inflected - altered by adding a suffix).

Other compound or complex words are made by
adding together elements without reference to the
internal grammar of a sentence. For example, the verb
infect suggests a new verb disinfect (=to undo the
action of infecting). New words are often formed by
noun + -ize, noun + ism, or verb + able (scandalize,
Stalinism, disposable). The study of such words,
"derived" from existing words or morphemes is
derivational morphology. Like Hindi language the
morphological structure of the English language also
has an impact on the performance of the search
engines.

II. SEARCH ENGINES SUPPORTING HINDI
LANGUAGE

With India emerging as a highly competitive Internet
market, more and more people are turning towards the
Internet. The number of internet users in India is
increasing day by day. Now it is high time that India
should have its own search engine leaning towards
Indian language contents. Few of the search engines
are already there in the market like- Google, Raftaar
and Guruji. Google [2] is capable of searching
information in many Indian languages successfully.
Google is far better then other Hindi supporting search
engines. Raftaar[3] is developed by Indicus Labs. It is
said completely Hindi language search engine. Funded
by Sequoia Capital, developed by students of IIT,
Delhi Guruji [4] is also one of the popular search
engines for Indian languages. Guruji Search can be
described in one word - fast. The search results are
thrown up pretty dam quickly with all the search
results pointing towards note worthy Indian sites.

III. FACTORS AFFECTING PERFORMANCE OF
SEARCH ENGINES

The information retrieval on the web in any
language faces numerous challenges. Besides all the
technical factors the morphological structure of the
language is one of them which also affect the
performance of the information retrieval on the web.

A. Morphological factors

Morphological factors are related to the morphology
of any language. As discussed in previous sections the
morphological structure of Hindi and English language
is somewhat similar to an extent. The Hindi as well as
English language faces the similar challenges in

Natural Language Processing tasks, which also
includes Information Retrieval on the web. Following
are the challenges which both the language faces while
information retrieval on the web.

1) Root word of the keywords: Every language use
some markers like (English language use s, es, ing and

■-■, -'"I", "m in Hindi language) are used with a root

word and new words are constructed. These new
words are called morphological variants of the stem.

For ex. <rii^i|l {Nadiyaan} (rivers), <riR^l {Nadiyon}

(rivers) are morphological variant of root word

5T^{Nadi} (river). While searching on the web users

give numerous queries which consist of words which
are not used in their root form.

Similarly in English language also some end markers
are used to constuct new words as already discussed in
this paper. For ex. Exam and Examination, Percent and
Percentage, River and Rivers etc.

The process of stemming of word which converts
the word into its canonical form/ root word is
obligatory. It is observed from th results that
sometimes it improves the performance of the search
engines.

We had taken a sample set of 50 TREC queries for
English language and 50 Hindi language queries
collected from the various users to test the affect of the
root word on the performance of the search engines.

Following Table 1 and 2 are the set of randomly
selected queries from this set. We compared the
performance of the English and Hindi language search
engines in the light of the root word. For Hindi
language we had tested our queries on Google, Raftaar
and Guruji. For English language, we had tested
queries on Google, Altavista and Yahoo.

Query# 1 ^H^TT^T ^ f^i^ # f^^rf^ {Samaaj mein

striyon ki sthiti} (women's situation in community)
Morphological variants of Query#l are as follows -

#1.1 WTUs^ ^ f^i^ # f^^rf^r, #1.2 ^e^^rra- ^ ;^

Query#2 trrf^ f^^RTl" ^ 31^=rr {Dhaarmik vivadon

kaa ant} (end of religious disputes)
Morphological variants of Query#2 are as follows -

#2.1 trrf^ f^m^ ^ 3iF^, #2.2 trrf^ f^mi^ ^

Query#3 ^ "H<|ui1 {Dharm puraadon} (Religious

Pandora)

Morphological variants of Query#3 are as follows -

173

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

#3.1 nA TOWf, #3.2 trt" TOW

Query#4 §TRcT % TK?^ {Bhaarat ke Rajyon} (States

of India)

Morphological variants of Query#4 are as follows -

#4.1 §TRcT % ;rn5^, #4.2 §TRcT % lj:5^

From the results it is evident that only Google indexes
the documents keyword in their root form. Raftaar and
Guruji do not index in that form that is the reason
number of document retrieved is also less in
comparison to Google. The overall comparison of the
results from the three search engines show that in
general the quantity of result retrieved increased when
the keywords are used in their root form.

In the case of search engines the quality of results is
more important than the quantity of the results. Table 1
shows the comparison of the precision value of the
three search engines. The precision value is calculated
by taking top ten results of the search engines. On
closely observing the result, we can say that the
precision value in the case of Google is high in almost
all the queries. We observed that relevancy of the
results is also high in Google in comparison to other
two search engines, Raftaar and Guruji which denotes
that not only quantity but the quality of the result is
also affected by the morphological variations of the
keywords.

Table I. Precision values of Google, Raftaar and Guruji
AGAINST Hindi language sample set queries

Que

Results for
(Google)

Results for
(Raftaar)

Results for
(Guruji)

Doc.

Retrieve

Pre.

Doc.
Retrieve
d

Pre.

Doc.

Retrie

ved

Pre.

1.1

1,87,000

0.90

1,322

0.50

1,87,0
00

0.80

1.2

34,200

1.0

3,230

1.0

34,20

0.80

2.1

1100

0.20

647

0.20

2.2
3.1

1000
19,100

0.20

1379
2,903

0.50
0.5

844

7760

0.4

3.2

24,100

2,609

0.8

21,07

0.5

4.1

6,92,000

61,691

1,55,5
43

0.3

4.2

6,06,000

2,24,911

6,16,1
88

0.3

Query#l.l Civil service exam 1.2 Civil service

examination

Query#2.1 Water wastage in India 2.2 Water waste in

India

Query#3.1 Funding and grants institution 3.2 Funds

and grants institution

Query#4.1 Mercury levels in birds 4.2 Mercury level

in birds

Table II. Precision values of Google, Alta vista and Yahoo
AGAINST English language sample set queries

Results for (Google)

Results for (Altavista)

Results for
(Yahoo)

Doc.
Retrieved

Precisio
n

Doc.
Retrieved

Precisio
n

Doc.
Retrieved

Precision

1.1

1,340,000

0.71

31,400,000

0.57

30,500,000

0.64

1.2

5,590,000

0.66

31,400,000

0.57

1,600,000

0.68

2.1

43000

0.68

698,000

0.47

1,660,000

0.44

2.2

2,390,000

0.52

29,400,000

0.64

3,260,000

0.68

3.1

62,300,000

0.26

25,800,000

0.67

25,600,000

0.36

3.2

78,800,000

0.54

21,800,000

0.46

21,600,000 0.45

4.1

4,860,000

0.56

1,790,000

0.53

1,640,000

0.67

4.2

9,180,000

0.59

3,480,000

0.45

2,850,000

0.46

Similarly we had taken five sample queries for English
language also to test the affect of Root Word on the
performance of the Search Engines.

It is observed that all the three search engines returned
almost same top results links with major differences in
coverage area. Only AltaVista sometimes shows the
same coverage area in morphological variant words.
However Yahoo and Google did not return similar
results in morphologically variant words. The
relevance level of four morphological variant queries
is calculated for the three search engines. It is found
that Google outperforms the other two.

2) Part of Speech (POS) Disambiguation: Part of
Speech Disambiguation is also one of the problems
while searching in Hindi language on the web. In
Hindi language there are words that can behave as
noun in some cases and the same word can behave as
verb in some other case. It is desirable from the search
engines to disambiguate the POS of the keyword in
particular results. Ambiguous POS of a word will have

a drastic affect on the results. For ex. Query '"^Tm

^T]?T' {gaya shahar} {Gaya city) in which '^aryzn"' is an

ambiguous POS it is used as verb in the sense of
'gone' and also as noun in the form of city name. All
the three search engines (Google, Raftaar and Guruji)
are returning maximum results in the sense of 'gone'.
It is desired from the search engines to disambiguate
the POS of the keywords which are ambiguous in
nature.

The similar problem was noticed in English
language information retrieval on the web. For English

174

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

language we had tested the following queries on the

three search engines

Query#lCode of conduct (noun)

Query#2 Conduct (verb) inaugural ceremony

It is observed that only Google is able to differentiate
between noun and verb 'conduct', whereas Altavista
and Yahoo both of the search engines give mix results
in both the queries which show that Yahoo and
Altavista are not able to differentiate between noun
and verb sense of the word 'conduct'.

3) Phonetic Tolerance: There are numerous words
which are phonetically equivalent but in writing they
appear distinctively. Search engines should be capable
of retrieving the results against any phonetically
equivalent word of a keyword entered by the user.
Indian languages alphabet contains many characters
which give same sound i.e they are phonetically
equivalent. These characters are used interchangeably

in many modem Indian languages. e.g. ^ra", ^%T, ^usi

and ^o-Sl are phonetically equivalent words.

We had observed that Google and Guruji are not
phonetic tolerant whereas Raftaar is not able to
identify different phonetic words. The apparent
variations are found in the relevancy of results on the
variation of the phonetic keywords. English language
is not at all phonetic in nature. So such problem does
not arise with the English language based search
engines.

B) Other factors

Besides the above mentioned factors in context of
morphological structure of Hindi language. There are
other critical challenges for the Hindi language and
English language searching on the web, which are as
follows-

Query#3 5:^ ts^ {Dand Baithak} (Dand Exercise)
Query#4 3T^ [5>7ll<ri {Ank Vigyan} (Numerology}

Query#5 ^^ RRT ^t^tfcT {Kam Priy Sangeet}

(Melodious Music)

Table III. Precision of Google, Raftaar and Guruji in

CONTEXT OF SENSE AMBIGUITY FOR HINDI LANGUAGE

Query#

Precision on
Google

Precision on
Raftaar

Precision on
Guruji

0.23

0.20

0.10

0.40

0.50

0.20

0.44

0.30

0.50

0.40

0.20

0.40

0.10

English language also faces the similar problem of

ambiguous keywords like 'bank', 'bat' etc. We had

selected the five sample queries from the complete set

of queries which are as follows-

Query#l built a bat house

Query#2 Capital tours

Query#3 Electrical current in Canada

Query#4 Free computer class.

Query#5 Clip art of light bulb

Table IV. Precision of Google, Altavista and Yahoo in

CONTEXT OF SENSE AMBIGUITY FOR ENGLISH LANGUAGE

Query#

Precision on
Google

Precision on
Altavista

Precision on
Yahoo

0.56

0.64

0.67

0.45

0.43

0.56

0.71

0.56

0.45

0.64

0.39

0.57

0.46

0.44

1) Ambiguous keywords: Many words are
polysemous in nature. Finding the correct sense of the
words in the given context is an intricate task. Various
researchers Eric Brill [5], Argaw [6], Navigili [7] and
Christopher Stokoe and John Tait [8] and many others
have justified the role of Word Sense Disambiguation
in the improvement of performance of web searching.
Ambiguous keywords deflate the relevancy of the
results. The example mentioned below shows this
aspect very clearly-

Query#l ^^[Pi\ 3l^ ^F^TT^"^ {Sona aur Swasthya}

{Sleep and Health}

Query#2 ^MR # W^om {Gulaab ki Kalam} (Rose

Branch)

It is quiet obvious from the results mentioned in Table
3 and 4 that in both languages, Search Engines are not
capable enough to cope up with this problem. The
results show the low precision values which justify that
the performance of the search engines is affected by
the sense ambiguity.

2) Word Synonyms: Every language has words and
its synonyms. It is observed while working on English
and Hindi language search engines that any word can
express a myriad of implications, connotations, and
attitudes in addition to its basic 'dictionary' meaning.
Choosing the right word can be difficult for people, as
well as for the Information Retrieval System. It is seen
that most of the times when we alter the keywords with

175

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

their synonyms the performance of the search engines
varies.

Query #1.1 STRSefW Tt ^1^1 cil {Arakshan se faida}

(Benefits of reservation)

Query #1.2 aTTTSercT q^ orr§T {Arakshan se laabh}

(Benefits of reservation)

Query #2.1 §TRcT % Tn5^-{Bhaarat ke rajya} (States

of India)

Query #2.2 §TRcT % y^^f {Bhaarat ke Pradesh}

(States of India)

Query #3.1 ^iM 3fr # 7^ {Gaandhi ji ki Mrityu}

(Death of Gandhi ji)

Query #3.2 ^mtft 3fr ^ f^TtR {Gandhi ji ka Nidhan}

{Death of Gandhi ji}

Query # 4.1 q^cTT ^i^8^ -^[P[ {Pehla Antriksh

Yaan}

(First Space ship)

Query #4.2 W^W 3Trrft£er ^TR" (Pratham Antriksh

Yaan) (First Space ship)

Table V. Precision of Google, Raftaar and Guruji in

CONTEXT OF SYNONYM PROBLEM FOR HINDI LANGUAGE

Query

Results for
Google

Results for
Raftaar

Results for
Guruji

Doc.

Pre.

Doc.

Pre.

Doc.

Pre.

1.1

3720

.70

4489

.50

.30

1.2

189,000

.90

15,408

.80

1625

.90

2.1

3,590,00

.70

319,903

.10

29,03
8

.40

2.2

2,670,00

.80

309,138

.10

37,75

.30

3.1

44,000

.60

12,844

516

.30

3.2

22,200

.10

18,455

517

.10

4.1

27,300

.80

2349

.40

6285

.70

4.2

13,000

.20

1314

308

.60

The variation in results is found in Query#l and
Query#2 when alternative synonyms are used. In

Query#l 'cTOT' in place of 'WMc^V and in Query#2

'^W in place of '^lo-^i' changed the result set. It is

justified that replacing some words with their
synonyms some times improves the results against the
query in any language.

The problem of synonyms lies in every language.
English language also faces the similar problem.
Changing the keywords of the query with its synonyms
in general vary the relevancy level of the results. From
the complete set of 50 queries 5 queries are used to

justify the impact of varying synonyms on the web

search results in English language.

Query# 1 . 1 bed sharing with children

Query# 1.2 bed sharing with kids

Query# 2.1 school bus safety

Query# 2.2 school bus security

Query# 3.1 aircraft safety act of 2000

Query# 3.2 aircraft protection act of 2000

Query # 4.1 Freedom of information act forms.

Query# 4.2 Liberty of information act forms.

Table VI. Precision of Google, Raftaar and Guruji in

CONTEXT OF SYNONYM PROBLEM FOR ENGLISH LANGUAGE

Results for

ery

(Google)

(AltaVista)

(Yahoo)

Doc.

Prec

Retrieve

Prec

Retriev

Prec

Retrieved

ision

1,070,000

82,200,0

77,300,

1.1

0.44

0.65

000

0.47

1,180,000

68,500,0

62,900,

1.2

0.56

0.48

000

0.43

9,460,000

47,400,0

48,700,

2.1

0.67

0.45

000

0.49

12,500,00

51,400,0

48,000,

2.2

0.68

0.57

000

0.67

231,000

6,480,00

39,200

3.1

0.68

0.35

0.68

420,000

6,480,00

77,300

3.2

20,600,00

0.66

62,900,0

0.36

725,00

0.65

4.1

908,000

0.47

00
19,200,0

0.48

253000

0.57

4.2

0.39

0.35

0.47

From the results mentioned in Table V and VI it is
evident that the variation of query terms with their
synonyms varies the precision level of the results.

IV. DISCUSSION

We have done comprehensive comparison of the
performance of the English language and Hindi
language based search engines in respect to their
morphological structure and also other factors. On
comparing the results of the query sample set it is
concluded that when the query is given in its root form
it returns into the maximization of results in Hindi
language but only sometimes in English language. But
in both the cases it is crystal clear from the results in
Table 7 that the precision values are better when the
key terms of the queries are in their root form.

From the results it is quiet evident that Google
indexes the keyword in its root form for Hindi
language but not for the English language. It is capable
of listing the documents consisting of all

176

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

morphological variants of the keywords which justifies
that the Google do not require stemmer for Hindi
language web information retrieval. The other two
search engines Raftaar and Guruji are not listing all the
morphological variants of the query and hence they
entail to have stemmer. For English language Web IR
AltaVista and Yahoo also show some improvement of
results when query terms are used in their root forms.
The results in Table VII justify that we do require
converting the keywords in their root form to improve
the relevancy of the results.

All the three search engines are not able to
disambiguate the POS as mentioned in section 3. For
English language Web IR only Google is capable of
disambiguating POS whereas Altavista and Yahoo are
not able to do that. Only Raftaar is upto some extent
Phonetic tolerant. Google and Guruji are not phonetic
tolerant for Hindi language. English language does not
face the problem of phonetic terms.

None of the three (Google, Raftaar and Guruji) are
attuned with ambiguity problem. Even English
language search engines are also not attuned with
ambiguity problem. On comparing the performance of
the English and Hindi language search engines we
found that precision values of English language is
somewhat better than the Hindi language though the
affect of ambiguity is quite visible on the performance
of the Hindi as well as English language search
engines. The Search Engines performance is degraded
because of the sense ambiguity problem in Hindi as
well as English language.

Sense / Synonym management is one of the
challenges mentioned. It is marked that all the three
search engines do not implement sense disambiguation
or synonym management. Sense disambiguation
improves the relevancy of results in web searching, as
various researchers have successfully justified the
application of WSD and synonym management in web
searching.

It is apparent from the results that relevancy of the
results retrieved by the search engines is dependent on
the morphological structure and also, senses and
synonyms in English as well as Hindi language.

An overall comparison we conclude that the
performance of the English and Hindi language search

engines affected by their morphology to some extent.
Since the English language search engines are grown-
up enough so the percentage of this affect is less in
English language search engines as compared to Hindi
language search engines.

V. CONCLUSION

We have compared the performance of the two
language search engines i.e. English and Hindi in the
light of their morphological structure and other factors
also.

Our results conclude that the performance of the
search engines is quiet affected by the morphological
issues as well as sense ambiguity and synonym
problems. This affect is much obvious in the Hindi
language search engines in comparison to the English
language.

The morphological structure of Hindi language is
more critical in comparison to the English language.
This is the reason the performance of the Hindi
language is more affected by such issues.

177

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

Table VII. Precision of Google, Raftaar and Guruji in context of Synonym Problem for English Language

Parameters

Hindi language Search Engines
Performance

English language Search
Engines Performance

Google

Raftaar

Guruji

Google AltaVista

Yahoo

Root Word

With Root Word

0.80

0.57

0.40

0.59

0.53

0.55

Without Root
Word

0.77

0.30

0.37

0.54

0.56

0.53

Sense Ambiguity

0.39

0.36

0.18

0.56

0.50

0.54

Synonym
Management

Variation

Precision

Variation

Precision

Variation

Precision

Variation in
Precision

POS
Disambiguation

Do not
support

Support to
some extent

Do not
Support

Phonetic
Tolerance

Not

Phonetic

Tolerant

Not

Phonetic

Tolerant

To some
extent
Phonetic
Tolerant

Not a phonetic Language

REFERENCES

[1] P. Bhattacharya, S. Singh, K. Gupta, and M. Srivastava,

"Morphological richness offsets resource demand- experiences in

constructing a POS tagger for Hindi" in Proceedings of COLING,

06, Sydney, Australia, pp.- 779-786, 2006

[2] http://www.google.com

[3] http://www.raftaar.com

[4] http://www.guruji.com

[5] E. Bril and S. Vassilvitskii, "Using WebGraph Distance for

Relevance Feedback in Web Search" in Proceedings of SIGIR'06,

Seattle, Washington, USA, pp. -147-153, 2006

[6] A. A. Argaw, "Amharic-English Information Retrieval with
Pseudo Relevance Feedback", in Proceedings of 8th Workshop of
the Cross-Language Evaluation Forum, CLEF 2007, Budapest,
Hungary, pp. 119-126,2007

[7] R. Navigili and P. Velardi, "Structural Semantic
Interconnection: a knowledge-based approach to Word Sense
Disambiguation", in Journal of Pattern Analysis and Machine
Intelligence, Volume 27, Issue 7, pp. 1075 - 1086, July 2005
[8] C. Stokoe and J. Tait, "Towards a Sense Based Document
Representation for Internet Information Retrieval", in Proceedings
of SIGIR'03, July 28- August 1, Toronto, Canada, pp. 791-795,
2003

178

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

Comparison of Traffic in Manhattan Street

Network in NS2

Ravinder Bahl (Author )

Information and Technology

M.M.E.C

Muallana, Ambala, Haryana, India

ravindra_ibm@yahoo . com

Rakesh Kumar (Author )

Information and Technology

M.M.E.C

Muallana, Ambala, Haryana, India

raakeshdhiman@gmail.com

Rakesh S3mhysil(Author )

Information and Technology

MBS College of Engineering and Technology

Babliana, Jammu, Jammu and Kashmir,India

rakeshsambyal@rediffmail.com

Abstract — The paper presents the Comparison analysis of traffic
in Manhattan Street Network (MSN). The behaviour of
Manhattan Street Network for constant bit rate (CBR) and
exponential traffic sources is demonstrated. The results are
produced using NS2 simulator. It is concluded that the
performance in multipath networks like MSN can be improved
by taking account of appropriate buffering for each traffic source
at the link node.

Keywords — Comparison Analysis, Traffic, Constant bit rate
(CBR), Exponential, MSN, Buffering,

I. Introduction

The selection of buffer is very important as it helps in
reducing the congestion and drop in packets .It helps in
balancing the load across the multipath and multihop
networks like Manhattan Street Network (MSN). The traffic
sources like constant bit rate (CBR) and exponential offer
different type of load in the network. In the large networks like
Manhattan Street Network the traffic is routed form two input
links to two outgoing links having equal bandwidth and delay.
The packet is stored in buffers upon arrival and is deflected on
the other. It cannot be accommodated at a later stage even if
the buffer allows admission to new arrivals at that stage. In
voice network and private line data networks the capacity
planning is very simple and straight forward with problem of
congestion and packet drops. For the selection of buffer
capacities, the growing traffic volume is to be calculated. As
the traffic increases so is the demand for more bandwidth. The
appropriate size and type buffer selection can reduce the
growing demand for more bandwidth. In this paper, attempt
has been made make provision for reducing congestion and
packet drops in different traffic sources like constant bit rate
(CBR) and exponential. In order to reduce the congestion and
packet drop , the drop tail queue is provided at each input node
which also help in reducing the deflections.

Further regardless of the destination, the buffering structure is
so designed to store the packets and deflection occurs only
when all the buffer slots are full. The distance vector
routing(RIP) and link state (OSPF) play a very important role
in today's internet [l].The simulation comparison of buffers in
OSPF and RIP algorithm conclude that the link state routing
performs better than distance vector routing in case of large
networks like Manhattan Street Network topology where large
volume of traffic flows form source to destination.

II. Background of routing algorithm

In case there are more than two paths from source to the
destination. The path selected between source and destination
represents the outgoing link to be used. The routing algorithms
based on cost metrics [5] helps in calculating the path between
the nodes. The cost is defined in terms of number of hops or
bandwidth, number of links, distance. This depends upon the
metric supported by different of buffer capacities. Many
different cost metrics [5] can be used to judge the shortest path
between the source and destination.

III. Concept of multipath routing algorithm

In multipath networks where data transactions in large volume
take place from source to destination, the path with the
optimum cost is to be selected to route the data. The selection
of improper path leads to over utilization of the network
resources resulting increasing delays and congestion, core
networks where there is a huge amount of data transactions
and there are more than one equal cost route possible from a
source node to destination node with large volume of traffic,
the multipath routing algorithm [6] [7] [9] may be used, which
helps in improving the available resources utilization and
helps in reducing congestion and packet drops and thus helps
in shaping the traffic between equal cost multi paths. The links
utilization [2] can be improved. By having improper queues
which in turn implies increasing delays, so sort of trade off is
required for selection of appropriate queue. The simulation has

179

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

proved that appropriate size queue for particular traffic source
results in better resource utilization .Results help in deciding
the size of queue at the link for better performance.
The selection of appropriate size queue avoids congestion at a
particular node including cross-traffic as given below.

(A) Departure of traffic from a node should be equal to the
traffic arriving at that node.

(B) It is essential in case of multipath networks that the load
should be balanced in such a way so that none of the outgoing
link of a router is over utilized.

(C) The appropriate queue with optimum storage capacity may
be maintained at the bottleneck links for better performance.

IV. Manhattan Street network (msn) topology

(A) The topology used in the paper for the purpose of
demonstration has 16 nodes and 32 links of equal bandwidth
and delay.

(B) The two traffic sources, constant bit rate and
exponential are used at the link node one at a time.

TABLE I : SHOWING SIMULATION Result with queue size

OF NO PACKET.

Fig.(l) : Experimental Topology (MxN) , Where (M=4,N=4)

V. Simulation results

The simulation study was performed and analysis were drawn
using Network Simulator (NS2) [8]. The Simulation results are
evaluated for in different buffering capacities required for each
traffic source using topology of Fig. (1) on case to case basis.

Packet
Size(bytes)

Traffic Type

Total Packet

Non

Dropped

Packets

Dropped
Packets

CBR

2002

Exponential

3110

TABLE I : SHOWING SIMULATION Result with queue size

OF 1 PACKET.

Packet
Size(bytes)

Traffic Type

Total Packet

Non

Dropped

Packets

Dropped
Packets

CBR

2002

Exponential

3110

TABLE IIL Showing Simulation Result with queue

SIZE OF 2 PACKETS.

Packet
Size(bytes)

Traffic Type

Total
Packet

Non

Dropped

Packets

Dropped Packets

CBR

2002

Exponential

3110

All links set to type simplex having bandwidth (1Mbps), delay
(10ms) the offered load was observed between nl to nl6 with
two constant bit rate and two exponential traffic sources
starting at nl and nl6 as destination. The drop tail queue was
attached between nl and nl6.

VI. Bottleneck link cases

The traffic from node nl to nl6 was observed where the link
bandwidth was set to 1Mbps, delay of 10 ms, and packet size
of 64 bytes. Initially two constant bite rate traffic sources and
later two exponential traffic sources with interval of 0.005
seconds and drop tail queue was introduced at a time. Initially
queue size was set to no packet and later was changed to the
capacity of more packets and performance was analyzed. The
simulation was run for 10 seconds and traffic was introduced
for 5 Seconds and the Following Results were obtained.

VII.

PERFORMANCE GRAPHS

The performance graph for bottleneck link, when one drop
tail queue with FIFO (First in First out) discipline is
maintained at the bottleneck link with two different traffic
sources as shown in Fig.(2). , Fig. (3), Fig. (4) and Fig. (5)
Respectively.

180

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

0.2

'CBRTfafficl
'CBR Traffic 2

1 ^ 7 10 13 le 19 22
Tifne

Fig. (2): CBR Traffic in Packets with Drop Tail Queue of size

1200

1000

-CBR Traffic
1

-CBR Traffic
2

14 7 10131619
Time

Fig. (3): CBR Traffic in Packets with Drop Tail Queue of size 2

Chart Title

0.9 -

1 a.e -

1 fiK

Jr n J.

J ■

n J ,

/1 1 J

fl-j

^^^

^^^_^_

3 5

7 9

11 1!
TiS*

1^ 1? IS n

Fig. (4): Exponential Traffic in Packets with Drop Tail Queue of size

^fimi

iSVWt

'6iij

T^jOUi

/ BMIVIIIIllBfl

a -^^

y^ j^

^ -BUD

/jt

^^* Lipjriiffiriil TrarBel

J^in

Hi-DpOituiiUiJ|¥rOfrpt2

/ If

JQD

y If

V\ii

1 Z 3 a 5 6 7 ft 5X0ail3JJ14]516171BI5M31

Fig. (5): Exponential Traffic in Packets with Drop Tail Queue of size 2

The simulation result helps in deciding the type of buffer at
the bottleneck links. Case study of buffer types for a
particular case is explained. The decision can be made based
on the performance graph that the appropriate buffer is
implemented at the bottleneck links for a particular case such
that the congestion in the network may be reduced and
performance may be improved.

VIII. Conclusion

The demonstration of buffer selection at bottleneck links in the
Manhattan Street Network Topology is done.

The paper demonstrated with different types of buffer , packet
size of 64 bytes and two constant bit rate (CBR) traffic sources
starting form node nl for destination nl6 the following effects

Case (1) Simulation of multipath network topology (MSN)

with drop tail node queue having storage capacity of packets

with two constant bit rate traffic Sources led to total packet

drop.

Case (2) Simulation of multipath network topology (MSN)

with drop tail node queue having storage capacity of packets

with two exponential traffic Sources led to total packet drop.

Case (3) Simulation of multipath network topology (MSN)

with drop tail node queue having storage capacity of one

packet with two constant bit rate traffic Sources led to total

packet drop

Case (4) Simulation of multipath network topology (MSN)

with drop tail node queue having storage capacity of one

packet with two constant bit rate traffic Sources led to total

packet drop

Case (5) Further it was concluded that Simulation of multipath

network topology (MSN) with drop tail node queue having

storage capacity of two packet with either two constant bit rate

or exponential traffic Sources led to no packet drop

181

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

IX. References

[1] Routing Basics,http:// www.cisco.com/ univercd / cc / td /doc/cisintwky

itodoc/ routing.htm.
[2] James Irvine, David Hade, "Data Communication and Networks", John

Wiley & Sons Ltd.,New York, USA,2002.
[3] William Stallings ,"Data and Computer Communications", PHI Pvt. Ltd.

N.Delhi, 7* Edition,2003.
[4] Alberto Leon-Garcia, Indra Widjaj a, ["Communication Networks,

Fundamental Concepts and Key Architectures",Tata McGraw-Hill

Publishing Company Ltd.,N.Delhi,2nd Edition, 2005.
[5] Brian Hill , "The Complete Reference,CISCO", Tata McGraw-Hill

Publishing Company Ltd., N.Delhi, 3rd reprint 2004.
[6] Johnny Chen, Peter Druschel, Devika Subramanian,

"An Efficient Multipath Forwarding Method",Proceedings of IEEE

INFOCOM , San Francisco, CA,March 1998.
[7] Israel Cidon, Raphael Rom, Yuval Shavitt, "Analysis of Multi-path

Routing", IEEE/ACM Transanctions on Networking,

Vol.7,No.6,Dec.l999.
[8] The Network Simulator ns-2, http://www.isi.edu/nsnam/ns/ns-

documentation.
[9] Ivan Gojmerac, Thomas Ziegler, Fabio Ricciato, Peter Reichl,

"Adaptive Multipath Routing for Dynamic Traffic Engineering", IEEE

GLOBECOM-2003,http:// userver.ftw.at /-reichl /publications

/GLOBECOM03.pdf

1 82 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

An Evolving Order Regularized Affine
Projection Algorithm^ Suitable For Echo

Cancellation

Shifali Srivastava
Electronics Deptt.
JUT,

Noida, India
shifalihbti2Q04@gmail.com

M.C. Srivastava
Electronics Deptt.
JUT

Noida, India
m.c.srivastava@iiit.co.in

Abstract — In this paper, a regularized Affine Projection algo-
rithm with Evolving Order is proposed. This algorithm auto-
matically determines its projection order, derived in the con-
text of acoustic echo cancellation (AEC). The simulation results
indicate that the proposed algorithm yields better performance
with small steady state error as compared to existing evolving
order affine projection algorithm (APA) and has fast conver-
gence speed.

Keywords — Acoustic echo cancellation (AEC), affine projec-
tion algorithm (APA), evolving order affine projection algorithm
(EO-APA), Evolving order regularized affine projection algo-
rithm (EO-RAPA),double talk(DT),echo path change(EPC).

I. INTRODUCTION

In acoustic echo cancellation (AEC) contexts the basic
approach is to build a model of the impulse response of the
echo path using an adaptive filter, which provides replica of
the echo at its output [1]. The adaptive filter output is sub-
tracted from the microphone signal to cancel the echo. Sev-
eral challenges are associated with AEC applications.
Firstly, the echo path can be extremely long and it may ra-
pidly change. Secondly, the background noise that appears
at the near-end side can be strong and non- stationary in na-
ture. Further, the involved signals (i.e., speech) are non-
stationary and highly correlated.

For echo cancellation several adaptive algorithms [1],
[2] have been applied. The normalized least-mean square
(NLMS) algorithm and the affine projection algorithm
(APA) are preferred due to their simplicity and robustness.
The affine projection algorithm (APA) [4], [5] updates the
weights based on the last input vectors. The convergence
speed of APA for correlated input signal is improved by
employing an updating-projection scheme of an adaptive
filter on a P-dimensional data-related subspace, but the con-
vergence speed of APA decreases, in the presence of noise.

Recently, EO-APA has been proposed to improve per-
formance in the presence of noise, with fast convergence
speed and small steady state error by varying the number of
input vectors [7]. An evolutionary method is employed in
EO-APA to determine necessary number of input vectors. In
this algorithm order of vector increases or decreases from
the previous one by comparing the output error with the

threshold involving the information of the steady-state
mean-squared error (MSE) [8].

The convergence speed of EO-APA gets degraded during
unvoiced speech signal and/or during silences when signal
value is either zero or close to zero. This paper proposes
solution of such problems by employing an evolving order
regularized Affine Projection Algorithm (EO-RAPA). In the
proposed algorithm, regularization is obtained by adding a
regularization factor matrix prior to taking inverse of the
correlation matrix. With the suitable choice of regularization
factor, the performance of the algorithm can be improved
with decreased computational complexity.

A major concern associated with the behavior of the
algorithm during double-talk, that may affect the overall
performance of proposed algorithm, is also addressed in the
proposed scheme. The standard procedure can be used for
double talk detection in order to slow down or completely
halt the adaptation process during double-talk periods.
Several algorithms have been proposed for detection of
double talk (DT) [9]-[ll]. The simplest double talk detec-
tion algorithm is the well-known Geigel DTD [11], which
provides a low-complexity solution. Since it is not efficient
to distinguish between echo path change and double talk, a
modified version of it is proposed in this paper.

This paper is organized as follows. Section II describes
evolving order APA (EO-APA). In section III, we propose a
regularized version of evolving order APA (EO-RAPA) for
AEC and performance improvement during double talk. The
experimental results which illustrate the convergence per-
formance of the proposed algorithm are discussed in section
IV. Finally, conclusions are presented in Section V.

II. EO-APA

fir end signal j(k)

Echo

Canceller

Stnictiire

LoHOspeake
(NonluKar)

ze:

Response

,11^)

<Zr

McToplione
Rspase

echoDfii(k)inid/{tr
liackgrQund and/or
near end speech

Desired signal d(k)
Figure 1 Acoustic echo canceller structure

183

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

A general acoustic echo canceller structure is depicted in
Fig. l.The aim of this structure is to model the acoustic echo
path (i.e. room impulse response) using an adaptive filter.
As shown in figure the far end signal x(k), on being
processed by the room impulse response, produces echo
signal. This echo signal is picked up by the microphone
and/or the near end signal v(k) and/or background noise
n(k) to produce a desired signal d(k).T\\Q far end signal that
also acts as the input to the adaptive filter produces a output
y(k) , the replica of echo signal.

The error signal e (k) = dk-y(k) , is used for weight adapta-
tion of the filter. With the far end signal denoted by Xk in the
matrix form, the desired dk may be expressed as:

dk= W\,,Xk + rik (1)

where Xk = [xk , Xk-i, . . ., Xk-L+iJ ,the subscript k
represents the time index, Wopt is an unknown L x l weight
adaptation column vector and nj, is a zero-mean Gaussian

independent sequence with variance of a^ .

An estimate Wy^ for Wopt at iteration k may be computed as
follows [4] by using an affme projection algorithm (APA):

H ,

,//,-!

W^=W^_1+//U-(U,U-) e^

(2)

where // is a constant step size .The error signal Cy^ may
therefore be expressed as:

,T (3)

H=^k

■^kn

where

dk = [dk,dk-i, - , dk-p+ij^ (4)

The matrix in (3) U /^ is the collection of the P most recent
input vectors [XkXk.j, • • • , X k.p+j] ^.
The order of the APA is defined by the projection order P,
,the number of the input vectors used to determine Uk , that
should be less than or equal to filter length L [l]-[3]. In EO-
APA the projection order P, at any iteration that varies in
accordance with adaptation, may be represented as [7] :

Since the number of input vectors at any iteration depends
on the output error and the previous number of input vectors,
the projection order adjusts itself if error exceeds a particular
threshold that can be determined by approximating steady-
state MSE.

Following Shin and Sayed [8], for each iteration the thre-
shold error may be written as:

(6)

a^{ju(P-l) + 2}
2-jU

s(P)

If square of the error signal e (k) is smaller than^(P ) ,

k - 1

projection order P^ at the Ji^ iteration in EO-APA should be
reduced by one from P^.j (its previous value), for smaller
steady state error. Whereas, Pk should be increased by one

from Pk.j when e(k) is larger than, s{Pj^_^ + 1) for faster
convergence speed. Therefore, the upper and lower thre-
sholds rj and 6>^ at k^^ iteration respectively may be ex-
/c

pressed for EO-APA can be expressed as:

Vk=s{Pj^_^+\)-

2- ju

and

0k=^(Pk-i^=

2-M

(7)

(8)

With these thresholds bound the projection order at any

iteration may be determined as:

2
MiriP^_^ + 1, Pjjjax }^i\<e (k)

4-1

(9)

ifOj^<ep)<77^

Ma4P^_ylA} ifel<0j^

The upper threshold controls the projection order of the EO-
APA to track the increased output error due to variation in
environment. Similarly the lower threshold acts as a switch-
ing point to decrease the projection order. Thus EO-APA is
expected to provide fast convergence speed with decrease in
steady-state error.

III. EO-RAPA

H. Rey, L. Rey Vega, S. Tressens, and J. Benesty in
[13], employed a variable explicit regularization factor in
their work on APA. They suggested an optimal value of re-
gularization factor (S) chosen such that minimizing differ-
ence of weight error vector of the consecutive iterations re-
sulted in lower steady state error. Such a choice maximizes
speed of convergence and minimizes steady-state mismatch.
Under simplifying assumption they define S

as:

P.a^.cT^.L

(10)

-Pa'

In the proposed work a regularization factor has been
added in the EO-APA. During unvoiced speech or during
silences, the signal level is either very low or zero and there-
fore inverse of (VkV^k) gets ill conditioned in weight upda-
tion in (2). Further, with the increase in the projection order
of the EO-APA, computational effort for determining in-
verse of the matrix increases. To avoid these problems a
regularization factor is added before taking inverse in the
proposed algorithm. The regularization factor also helps in
suppressing the effect of noise, which may be the back-
ground noise or/and the near-end speech corrupting the out-
put of the echo path. The proposed algorithm referred as
evolving order regularized APA (EO-RAPA), may be ex-
pressed as:

The identity matrix 1= Pk'^Pk and ^is the regularization
factor.

Performance improvement during Double Talk:

When the speech signal v (k) is zero and the near-end
noise n (k) is assumed to be insignificant, weights of the
adaptive filter would converge to successfully cancel the
echo, and successfully cancel the echo. However, when both

184

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

v(k) and x(k) are not zero, that is in double-talk (DT) situa-
tion, the near end speech v(k), acts as an uncorrected noise
to the adaptive algorithm, and may allow excessive un-
cancelled echo to pass. Solution to this problem is to slow
down or completely stop the filter adaptation when the pres-
ence of the near-end speech is detected.

A simple approach to detect DT is the well-known Gei-
gel algorithm [11]. The Geigel algorithm compares the mag-
nitude of the near-end received signal d (k) with the maxi-
mum magnitude of L most recent samples of the far-end
signal x(k), where L is the adaptive filter's length. The Gei-
gel algorithm computes its detection variable and makes
decision. If detection variable is larger than the threshold ,
DT is declared otherwise not. However, for an AEC envi-
ronment, the echo path characteristics are time varying.
Therefore, for the time-varying echo path, DT can be falsely
detected by the Geigel algorithm when a change of the echo
path occurs. As a result, the adaptive filter stops updating
the coefficients when the coefficient update is actually
needed.

In the proposed algorithm EO-RAPA a detection scheme
is presented that detects and distinguishes between echo path
change and double talk. Simulation results show that this
detection method provides adequately reliable performance
with lower complexity. This method is based on computa-

tion of detection variable ^ at any iteration:
mean power of microphone signal ^(k)

mean power of far end signal X(k)

(13)

If ^ > threshold Th and at the same time projection order is

above the certain value say Pth, double talk is detected. Since
projection order is directly related to the output error that
corresponds to the cross correlation between microphone
signal and far-end signal. Thus, our detection method is
power and correlation based method.

In the case of EPC condition, projection order is above
a certain value Pth but ^ < threshold Th. We therefore employ
two thresholds one on ^ and one on projection order, to

reduce the probability of false detection of DT or EPC con-
ditions. Flow chart of DT detection scheme is shown in
Fig. 2.

Monitor ^

AND Function

IV. SIMULATION RESULTS

The performance of the proposed algorithm is verified by

carrying out experiments in context of echo cancellation

with speech input sampled at 8 KHz and echo path length of

512.

ERLE, an important parameter to measure of convergence

speed and misadjustment that may be expressed in dB by

(13), is employed for performance evaluation.

£7?Z£ = 10*log

(13)

where g^ is the power of the microphone signal and

the power of the residual echo.

In Fig. 3, by ensemble averaging over 20 independent
speech samples, ERLE plot is obtained for proposed algo-
rithm with variable regularization factor, EO- APA and clas-
sical APA for different projection orders (P=8 and 16),
SNR=30 dB. As shown by simulation results the perfor-
mance of proposed algorithm is better in terms of conver-
gence speed and low steady state error.

-5

-10

LiJ -20

^-25

-30

-35

-40

EO-RAPA

EO-APA

APA(P=16)

APA(P=8)

1000 2000 3000 4000 5000 6000 7000 8000 9000
No. of iterations

Figure 3 ERLE plot of proposed algorithm, EO- RAPA and classical APA (echo
path length =512, speech input sampled at 8000 Hz, step-size=0.5)

Fig. 4 Shows ERLE plot for different values of fixed regula-
rized factor {S). Simulation results show the practical justi-
fication that as 5 increases, steady state mismatch reduces

2
but at the cost of lower convergence speed. For J =0.5* a ,

convergence speed is faster but steady state error increases.

2
On the other hand for d = 500 * (T slower convergence

speed is obtained with lower steady state error.

Double Talk
Detected

Figure 2 Flow chart of DT detection scheme

185

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

-5
-10
-15

! "^°|

-25!

-30

-35 [W-
40

del=0.5*sigma

del=5*sigma

del=20*sigma

del=50*sigma

del=500*sigma

1.5 2 2.5

no of iteration

3.5

<10^

Figure 4 ERLE plot of proposed algorithm with different values of fixed regularization
factor (where del is regularization factor and sigma is mean power of far end signal),
SNR=30dB

Double talk scenario is depicted in Fig. 5, ERLE plot
of proposed algorithm when DT occurs from 19,000 to
21,000 with proposed double talk detection (DTD) scheme
and without DTD scheme. Simulation results indicate that
deterioration in performance during double talk (DT) is im-
proved by employing proposed DTD scheme.

0,5

1.5

2.5

20-

h i

lUii

0,5

1,5

2
0, of iterations

3 3.5
xlO*

0-
0-

qII, „1, I i„

i ..1

.Jij.L

;L. 1.

J 1

1,5

2
oof iterations

3 3
x10*

^ 1 1 1 1 1

,, , i M Ill y, .L...

f 1

no of iterations ^^^^

Figure 6 Microphone signal, Far end signal, projection Order, Detection variable signal (Zeta)

and error plot of proposed algorithm when EPC occurs at iteration 10,000 and DT occurs

between 17,000 tol9,000 with DTD scheme, SNR=20dB

Fig 7 shows the performance of proposed algorithm by em-
ploying DT detection scheme when EPC and DT occurs at
different iterations.

^-400
o|200

1.5 2

no. of iterations

x10

— wttiout DTD sctieme
— wtti DTD scheme

J20

]-40

1 1

0.5

2.5

1 1.5

no.of iterations j^^g^

Figure 5 Microphone signal, far end signal, projection Order, Detection variable signal and

ERLE plot of proposed algorithm when DT occurs between 19,000 to 21,000

Fig. 6 shows plots of different signals and it can easily seen
from simulation results that proposed DTD scheme detect
and distinguish EPC and DT.

O 0.5 1 1.5 2 2.5 3 3.5

no of iterations ^ ^q^

Figure 7 ERLE Plot of the proposed algorithm with DTD scheme when EPC occurs at itera-
tion 10,000 and DT occurs between iterations 17,000 to 19,000, SNR=20dB

V. CONCLUSIONS

In this paper an evolving order regularized affme pro-
jection algorithm (EO-RAPA) has been evolved that is suit-
able for AEC applications. The performance of the proposed
algorithm is verified by carrying out experiments in context
of echo cancellation with speech input sampled at 8KHz.
Simulation results indicate that proposed algorithm has fast-
er convergence speed and lower steady state mismatch com-
pared to existing EO-APA. Further the proposed algorithm
provides an improvement in the performance during double
talk by employing a new DTD scheme.

REFERENCES:

1. S. Haykin, Adaptive Filter Theory, 4th ed. Upper
Saddle River, NJ: Prentice Hall, 2002.

2. A.H.Sayed, Fundamentals of Adaptive Filtering.
New York, Wiley, 2003.

3. Simon S. Haykin, Bernard Widrow, Least-mean-
square adaptive filters, Wiley ;2003.

186

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

4. S.L. Gay and S. Tavathia, "The Fast Affme Projec-
tion Algorithm," Proc. 1995 IEEE Int. Conf.
Acoustics, Speech, and Signal Process, May
1995, vol. 5, pp. 3023-3026.

5. K. Ozeki, T. Umeda, "An Adaptive Filtering Algo-
rithm Using an Orthogonal Projection to an
Affme Subspace. and Its Properties," Electronics
and Communications in Japan, Vol.67-A, no.
5, pp. 19-27, 1984.

6. Hyun-Chool Shin, Ali H. Sayed, and Woo- Jin
Song, "Variable Step-Size NLMS and Affme
Projection Algorithms", IEEE, Signal Processing
Letters, Vol.1 l,No.2, Feb. 2004.

7. Seong-Eun Kim, Se-Jin Kong , Woo-Jin Song ,"An
Affine Projection Algorithm With Evolving Order
", IEEE Signal Processing Letter, Vol. 16,
No. 11, Nov 2009.

8. H.-C. Shin and A. H. Sayed, "Mean- square perfor-
mance of a family of affme projection algo-
rithms," IEEE Trans. Signal Processing, vol. 52,
no.l,pp. 90-102, Jan.2004.

9. J. Benesty, D.R. Morgan, and J. H. Cho, "A new
class of doubletalk detectors based on cross-
correlation," IEEE Trans. Speech Audio Process.,
vol. 8, no. 2, pp. 168-172, Mar. 2000

10. Thien-An Vu, Heping Ding, and Martin Bouchard
"A survey of double talk detection scheme
for echo cancellation applications"

11. D.L. Duttweiller, "A twelve-channel digital echo
canceller," IEEE Trans. Comm., vol. 26, pp. 647-
653, May 1978.

12. I.Yamada, K.Slavakis, and K.Yamada, "An effi-
cient robust adaptive filtering algorithm based on
parallel subgradient projection techniques," IEEE
Trans. Signal Process., vol. 50, no. 5, pp. 1091-
1101, May 2002

13. H. Rey, L. Rey Vega, S. Tressens, and J. Benesty,
"Variable explicit regularization in affme projec-
tion algorithm: Robustness issues and optimal
choice," IEEE Trans. Signal Process., vol. 55, no.
5, pp. 2096-2108, May 2007.

AUTHORS PROFILE

Shifali Srivastava received her B. Tech degree
from H.B.T.I Kanpur, and now perusing
M.Tech from JUT, Noida. She has done pro-
jects on all optical networks. Her area of
interest is signal processing and communica-
tions.

M.C. Srivastava received his B.E. degree from
Roorkee University (now IIT Roorkee),
M.Tech from Indian Institute of technology
Mumbai and Ph. D form IIT Delhi in 1974.He
was associated with I.T. BHU, Birla institute of
Technology and Science Pilani, Birla institute
of Technology Ranchi and ECE Dept. JUT
Sector 62 Noida. He has published around 60
research papers. His area of research is signal
processing and communications. He was
awarded with Maghnad Saha Award for his
research paper.

187

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

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

Vol. 8, No. 3, 2010

Design and Implementation of Flexible Framework
for Secure Wireless Sensor Network Applications

Inakota Trilok

Department of Computer Science & Engineering

National Institute of Technology

Warangal, India

itrilok @ hotmail . com

Mahesh U.Patil
National Ubiquitous Computing Research Centre
Centre for Development of Advanced Computing

Hyderabad, India

maheshp@cdac.in

Abstract — Secure communications is an interesting and
challenging research area in Wireless Sensor Networks (WSN)
fundamentally because of the low power constraints and small
memory footprints inherent in the technology. In this context,
there are many hardware platforms like TelosB, MicaZ and
Mote2 which implement a security layer in hardware, supporting
multiple modes of operation like encryption, integrity or
combinations of both. However, not all hardware platforms
support hardware security which creates avenues of research in
designing low power security algorithms in software. As with the
development of security algorithms for WSN applications, there
is an urgent requirement to create a unified approach for
application developers by which they can integrate and use
existing security algorithms thereby maintaining an abstraction
from the intricacies of the algorithm.

This paper introduces a flexible framework which
implements a unified API to add new security algorithms to a
security library suite. This library integrates existing security
algorithms like TinySec, MiniSec etc. We also bring out the
implementation of Advanced Encryption Standard (AES) in
software supporting its various modes of operation. We have
integrated this implementation with the unified framework and
demonstrated its performance and our results. We compare our
software AES implementation with the Hardware AES
implementation, in all the supported mode settings.

Key words- Wireless Sensor Networks; Mote; Link Layer
Security; Network Layer Security; Hardware Level Security^
Integrity; Encryption; Authentication.

Introduction

Wireless Sensor Networks (WSN) is a collection of
distributed autonomous systems called sensor nodes that
monitor and collect physical data for assessment and
evaluation. These sensor nodes are very small in size, and are
limited in resources like CPU, memory and network
bandwidth. Moreover they are powered through small batteries.
All these make wireless sensor networks vulnerable to security
attacks and this is a crucial aspect of the sensor network. Much
of security is application specific and in applications like
physical intrusion detection or perimeter protection these are of
utmost importance. As sensor nodes are powered through
batteries, security techniques must ideally consume less energy.
Ironically, some sensor networks need high security which

leads to higher consumption of energy. In such cases, high-
energy and rechargeable batteries are used. There are some
contradictions between the communication energy cost and
cryptographic cost for WSN (See [1, 2]). So, the security
technique needed depends on the application that is to be
deployed into the network. Thus, there needs to be a balance
between the amount of security that can be provided to these
networks and resources on the mote. This is different from
conventional security solutions, since in WSN the security is
tightly coupled to the application's need.

TinyOS [22] is a free and open source embedded operating
system which is specifically designed for wireless sensor
network application development. It follows a component
based architecture which enables application developers to
integrate their application requirements with existing network
communication protocols. The application and operating
system is bundled into a final image which is burnt onto the
hardware thereby creating two tier architecture. Application
developers should wire a customized network stack for which
knowledge of low level details of each of the algorithms is
required. These algorithms could spawn diverse areas like
network communication, dissemination, time synchronization,
and security, etc. Moreover, to modify the stack inorder to test
performance, knowledge of the interfaces provided by alternate
algorithms also have to be understood. This requirement
imposes an additional burden on the developer.

There are number of security algorithms available in
TinyOS communication protocol stack. Some of them like
TinySec [3], SenSec [4] and CC2420/CC2430/CC243 1 Radio
AES [5] operate at the link layer, while other algorithms like
MiniSec [6] exist at the network layer. As is seen above there is
wide diversity in the implementation and detail of the
algorithms. This renders migration from one security algorithm
to another a point of bother for the application developer. A
uniform access method for all security algorithms is desirable.
The main contributions of this paper are:

• Introduction to an adaptive framework for WSN
applications.

• A general purpose security library suite composed of
existing security algorithms for popular sensor node
hardware platforms.

188

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

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

• An application developer's perspective in creating a
custom network stacks specific to his application
requirements.

• Support for existing versions of TinyOS providing the
application developer an abstraction to low level
implementation of the desired security algorithm
chosen. The application developer is provided with a
common API's for setting modes and updating keys for
all available security algorithms.

• Implementation and integration of software AES and
hardware AES with same mode settings like encryption
only, CBC-MAC integrity-only, counter encryption
only and counter encryption - CBC-MAC integrity
with dynamic and static key support.

The paper is outlined as follows. Section 2 presents the
background of various security protocols at different levels
and their limitations. Section 3 formulates the
implementation of flexible fi-amework and its design goals,
and introduces software AES and various modes of
operation used by us and Section 4 describes the results and
analysis of framework with software AES. Finally Section
5 concludes with a peek into the future work.

II. Background

A. Existing security protocols:

TinySec [3] which is implemented at Link layer ensures
low energy consumption but the algorithm is vulnerable to
replay attacks. MiniSec [6] at Network layer ensures low
energy consumption but operates only a fixed level of security
which supports both encryption and authentication. Many
applications require a combination of both confidential and
non-confidential data which is not supported by MiniSec.
However hardware AES security hosts ensures low power
consumption and combinational levels of security. This feature
is available for hardware that contains IEEE 802.15.4
comphant RE transceiver like CC2420/CC2430/CC243 1 chips
[6] and some of the motes that support this radio are TelosB,
MicaZ and Mote2. Not all motes like IRIS [19] etc support
hardware security. A software implementation of the AES
could be a way satisfies these requirements widely. We have
chosen AES for both efficiency and security reasons (see [8,
9]).

B. Supported block ciphers:

There are many block ciphers available namely Skipjack
[16], RC5 [15], RC6 [11] and Rijndael [17]. Each cipher is
chosen based on the need of applications security, memory and
energy efficiency of the cipher [10]. We have selected Rijndael
in the configuration of 128/128/10 (keysize/blocksize/rounds)
but still our library suite supports [192^256]/128/[ 12^14]
configurations. As, in WSN power is more of concern,
choosing 128 bit key is more appropriate.

C. Modes of Block cipher:

CTR encryption: This is counter mode encryption. To make
compatible with hardware AES we provided this option [18].

Vol 8, No.3,
mode provides

2010
only

CBC-MAC Authentication: This
authentication of the payload [7].

CCM Mode: This is Counter mode encryption and CBC-MAC
authentication mode that features authenticated encryption
[20].

OCB Mode: This Offset CodeBook, is a block-cipher mode of
operation that features authenticated encryption for arbitrary
length of data [7].

Only-Encryption/Only-Decryption: In this mode, simple
encryption/decryption operations are performed without any
mode settings.

III. The Adaptive Framework For Secure Wsn
Applications And It's Design Goals

We propose a flexible framework for programming
TinyOS. This framework is divided into multiple components
as shown in Figure 1 and each component contains group of
protocols: routing protocols. Time- synchronization protocols,
localization and security protocols. The application developer
is requfred to wfre a combination of these protocols for specific
application using the framework. For example, an application
can use TinyHop routing algorithm. Flooding Time
Synchronization Protocol (FTSP) [9] with Hardware or
Software security. In such a case this framework can be
configured depending on the apphcation's need and the level of
security required. So, we grouped these protocols into multiple
components by providing abstraction to the application
developer for easier access.

WSN Application

Common API ' s

Adaptive Framework f

[>r WSM Application

TinyHop,
MultiHop. . .

FTSP, TPSN. . .

Routing
Component

Time Synchronization

Component

TinySec, MiniSec,

Ha rdwa re AES

Software AES. . .

Localization
Protocols . . .

Security Component

Localization Component

Ha rdwa re

Fig. 1. Adaptive Framework

Since one of the objectives of this paper is in providing a
unified security access API to application developers, we
introduce a use case of the framework with the security
component described. This security library lists available
security algorithms, based on the version of TinyOS and the
hardware platform chosen. There are two actors in the
framework, the application developer and the system
component developer. A system component developer

189

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

integrates security algorithms into the hbrary suite in
conjunction with the unified access API which we have
standardized. The system component developer is required to
create glue logic between the standardized API and the
algorithm's internal functions. The application developer on the
other hand is required to use the unified access API to wire his
application's logic with the required security algorithm.
Algorithms implementing various modes of block ciphers
discussed earlier are integrable into the security suite. We have
integrated our software AES with IEEE 802.15.4 specification
and supported all modes for existing security algorithms is
outlined in Table V. We have also integrated security
algorithms like TinySec [3], MiniSec [6], and Hardware
Security [5] in the suite. This framework is useful for security
related experimentations in perspective of both application
developer as well as System developer. Any new cipher can be
easily plugged-in and plugged-out. The unified APIs provided
for existing security algorithms removes the need of changing
the existing application code. The framework takes care of
mode settings based on chosen security layer and algorithm.
This framework is simple in ease of use, flexible and adaptive.

TABLE I
Wrapper APIs Provided By Framework

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

Vol 8, No. 3, 2010
algorithm. For example, applications may use TinySec or
MiniSec, Hardware AES or Software AES etc.

command error t AFWAupdateKey(uint8_ t * key, set KEY);
event void AFWAupdateKeyDone(uint8_ t * key);
command error t AFW As etTransmitMode( uintl6_ t ctrlO,

uint8_ t ten );
command error t AFWAsetReceiveMode ( uintl6_ t ctrlO,

uint8_ t len );
command error t AFWAsend( uintl6_ t addr,

message_ t *msg, uint8_ t len );
event void AFWAsendDone( message _ t "^msg,

uint8_ t error);
command error t AFWAreset(uint8_ t Type);
command error t AFWAget(uint8_ t "^key, get KEY);

Ease of use: - It provides complete abstraction to the
application developer and introduces a GUI through which the
application developer can select components of the lower
layers. The framework will internally wire and create a
template for application development. For example, if
application developer uses IRIS mote and needs software AES
with various mode settings, a project can be created for IRIS
mote using WSN IDE [21] and then call commands for mode
settings. Now the framework makes a setup and loads AES
library.

Flexible: - Our framework is flexible, because it has feature
of plug-in and plug-out facility i.e. any security algorithm can
easily be integrated. Also, there are several schemes for key
settings. In such a case dynamic key support is more robust
than the pre-configured key. We used Java Cryptographic [23]
functions with mouse movement and keyboard random-key
generation. In this framework it supports both randomly
generated dynamic key and static key setting.

Adaptive: - Level of security is application specific. So, it is
a choice of application developer to choose type of security
needed. The framework is adaptive so that it can switch
between levels of security and provides corresponding security
layer to the application based on selection of security

A. Hardware Independent AES with IEEE 802. 15.4
Specification

In this section, implementation of the Software AES
security architecture with IEEE 802.15.4 specification which is
provided in the framework security component is outlined.
According to IEEE 802.15.4 specification it has eight different
security suites [13]. The Table IV, gives five modes and in last
two modes each have different variants based on the chosen
MAC value.

Fig. 2. Configuration file fi3r Software AES

The above Figure 2 depicts the configuration for software
AES. We have configured in such a way that n-number of
modes can be integrate with AES module but only suitable
mode with AES module is loaded during the compile time.
This configuration makes developer easier to integrate AES
module with any new mode.

We will see each AES security suite in-detail:

• CTR: This is counter mode encryption. It uses counter
value and it consists of sender's address and 4-byte
frame counter. We have not appended any flags and
block counter to the data payload and hence this
minimizes the power of mote. The block counter is the
number of blocks splited into 16 byte blocks within the
packet. This value can be calculated based on the size
of the packet that is to be sent. The frame counter is
maintained by the software AES and it is incremented
for each packet automatically by software. The send
API includes frame counter and encrypted payload into
the data payload of the packet. Below is the code
snippet for CTR mode encryption.

uint8_t *payload = call AMSend.getPayload(msg, len);

memcpyi smsg.data, payload, len );

call Ctrl.AESctr encrypt((uint8_ t ^^)smsg.data, len,

sec txl, nonceValue );
memcpyi&fcl, &nonceValue[3], 4 );
smsgfc =fcl;

memcpyipayload, &smsg, fcLen + len );
return call AMSend.sendj addr, msg,fcLen + len );

190

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

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

Vol 8, No.3, 2010

For example, if h || pi,...,Pn is frame format then after
CTR mode encryption, packet sent will be h || zi ||c,,
where, c, =pi @ AESi(Xi), h = header of Tiny Hop, Zi =
Frame counter, Ci = encrypted pay load, x, = nonce
value or counter or initial vector.

CBC_MACJ|SEC_M [0-7]: This mode provides
only authentication. And the size of MAC value can
vary between 4, 6, 8, 10, 12, 14, 16 byte. Here
CBC_MAC II SEC_M[0- 7] is a macro setting that tells
library to load AES with CBC_MAC mode. In the
below code snippet truncateTagO command will
truncates MAC in the range {4,6,8,10,12,14,16} byte
based on the SEC_M[0- 7] macro setting i.e. if the
macro is set to one among the following SEC_M1,
SEC_M2 ... SEC_M7 then each value maps to the
range {4,6,8,10,12,14,16} byte while SEC_MO is
reserved for future use. By default in CBC-MAC
mode, length field is authenticated by software.

In this it supports two protecting modes:

(a) It protects header of TinyHop routing algorithm as
well as the data payload. Suppose if h\\pi,...,pn is frame
format then after CBC_MAC authentication, packet
that is sent will be

h\\pi,...,pn II TruncsEc_M[o 7jfauth(h\\pi,...,pn)}.

(b) Protects only data payload. Suppose if h\\pi,...,pn is
frame format then after CBC_MAC authentication,
packet that is sent will be

h\\pi,...,pn \\TruncsEc_M[o 7jfauth(pi,...,pn)}.

Where, h = TinyHop header, pi= plain text.

call Ccml.AESccm auth((uint8_ t "^jsmsg.data, len,

sec txl,KeySizeB );
call CcmI.truncateTag((uint8_ t '^)smsg.data,len,

appLen);
call Ccml.AESccm encrypt((uint8_ t ^)smsg.data, len,

sec txl,nonceValue );
memcpyi &fcl, &nonceValue[3], 4 );
smsgfc =fcl;

memcpyi payload, &smsg, fcLen + len + appLen );
return call AMSend.send( addr, msg,fcLen + len +
appLen );

For example. If h \\pi, . . . , Pn is frame format then after
CCM mode, packet sent will be

h \\zi \\cj,...,Cn \\ENC(MAC)
where, MAC = TruncsEc M[0-7J{auth(h\\pi,
ENC(MAC) = MAC @AESi(xO,

Ci = Pi @AESk(Xil Zi = Frame Counter.

Pn)h

• AES_ENC: This is simplest mode. It provides simple
AES encryption operation without any mode settings.
Below is the code snippet that takes key size and
pointer to input array as an argument and produces
encrypted output in the same input array.

call AesI.startAES((uint8_ t)KeySize, (uint8_ t "^jinPut);

• AES_DEC: This is simplest mode. It provides simple
AES decryption operation without any mode settings.
Below is the code snippet that takes key size and
pointer to input array as an argument and produces
decrypted output in the same input array.

call Aesl.startAES((uint8_t)Key Size, (uint8_t *)inPut);

uint8_t "^payload = call AMSend.getPayload(msg, len);

memcpyi smsg.data, payload, len );

call Ctrl.AESctr encrypt((uint8_ t ^)smsg.data, len,

sec txl, nonceValue );
memcpy(&fcl, &nonceValue[3], 4 );
smsgfc =fcl;

memcpyipayload, &smsg, fcLen + len );
return call AMSend.send( addr, msg,fcLen + len );

• CCM||SEC_M [0- 7]: This is AES counter mode
encryption and CBC-MAC authentication. Here
CBC_MAC||SEC_M[0 - 7] is a macro setting that tells
library to load AES with CCM mode and functionality
of SEC_M[0- 7] is same as previous mode setting. In
this mode first it authenticates header of TinyHop
routing algorithm and data payload using CBC-MAC
and then encrypts both data payload and MAC using
AES-CTR mode. Below is the code snippet for CCM
mode implementation.

uint8_t ^payload = call AMSend.getPayload(msg,

len);
memcpyi smsg.data, payload, len );
call Ccml.AESccm nonce(nonceValue);

Figure 3, depicts the complete flow that takes changes in
the length field of the payload while sending a packet.

AFWAsend (addr, msg,

len); (^

start j

>1.

1 t

1 Address

Message, len

-^f<-

^^^ ^ If CTR or

,^>\ MO

CCM

CBC_MAC

1. Append FC

2. len = fcLen + len

+ sec_m/2+2

1. Append FC

2. len = fcLen

len

1. len

= len +
sec_nn/2+2

1. len = len

Fig. 3. AES Security suite

FC = frame counter, fcLen = frame counter length
len = length of the payload

sec_m = Number of bytes in authentication field for CBC-
MAC, encoded as (M-2)/2.

191

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

B. Common API 's for Security algorithms

Since, each security algorithm has its own parameter
settings; the developer has to know which API is to be used.
We propose Security interface that contains common wrapper
API's to set and get parameters for various security algorithms.
Before choosing security algorithm, application developer will
choose version of tinyos and for this we created framework
[21] that lays above wrapper API's. Some of the common
API's that we provided are:

a) command error t AFWAupdateKey(uint8_t * key,

uintSj setKEY);

event void AFWAupdateKeyDone(uint8_t "^ key);

AFWAupdateKeyO updates old key value to new key value
where KEY parameter defines which key is to be updated in
ACL entry. uintSj ^key is a pointer to an array containing 8-bit
unsigned integers. KEY is 8-bit unsigned integer that takes
macro value for ACL entries i.e. KEYO or KEYl or so on.

b) command error t AFWAsetTransmitMode( uintl6_t ctrlO,

uintSjt len );
command error t AFWAsetReceiveMode ( uintl6_t ctrlO,

uintSj len );

AFWAsetTransmitModeO / AFWAsetReceiveMode() sets
transmission and receiver mode for any security algorithm.
Parameter ctrlO value can be a combination of macros given in
Table IV and parameter len has value zero if the total payload
value is to be encrypted/decrypted otherwise len value
specifies number of bytes to be encrypted or decrypted or
number of bytes not to be encrypted or decrypted based on the
context and mode setting of ctrlO.

c) command error t AFWAsend( uintl6_t addr, messagejt

^msg, uintSj len );
event void AFWAsendDone( message _t ^msg,
uintSjt error );

AFWAsend( ) command is similar to AMSend() command.
This command will set correct length of the transmitted
message when Hardware/Software AES security is used and
this command does nothing for other security algorithms.

d) command error t AFWAreset(uint8_t Type);

This command is used to reset MAC or Encryption
initialization vectors of security algorithm.

e) command error t AFWAget(uint8_t ^key, KEY );

This command is used to get key value from ACL entry. KEY
is 8-bit unsigned integer that takes macro value for ACL
entries i.e. KEYO or KEYl or so on and the final result is
fetched to key.

IV. Results and Analysis

We have tested our proposed framework and software
Advanced Encryption Standard (AES) with various mode

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

Vol 8, No. 3, 2010
settings using two different motes i.e. IRIS and MicaZ. We
have taken application that sends encrypted packets using
TinyHop routing algorithm to the Base Station. And then Base
Station decrypts the received packets and forwards the packet
to serial forwarder where the user can view the original packet.
We sniffed the packets using sniffer to check whether packet is
encrypted or not in various mode settings. Tables II results are
obtained after installing software Advanced Encryption
Algorithm using TinyOS-2.1.0 without TinyHop routing
algorithm and without our framework into MicaZ and IRIS
motes.

TABLE II
Memory Utilization of Software AES USING TINYOS-2.1.0

Framework Without Tinyhop Routing Algorithm

Mote

ROM

occupied

in bytes

(percentage)

RAM

occupied

in bytes

(percentage)

Name of

Cipher

and its

configuration

MicaZ

22256(17.38%)

2196(54.9%)

AES128/128/10,
CCM mode
with 16- byte

22258(17.38%)

2196(54.9%)

AES 128/128/10
CBC-MAC with
16-byte MAC

22254(17.38%)

2196(54.9%)

AES 128/128/10
CTR mode

22242(17.37%)

2196(54.9%)

AES 128/128/10
only AES
Encryption/
Decryption

IRIS

21308(16.64%)

2404(30.05%)

AES 128/128/10
CCM mode
with 16- byte

21306(16.64%)

2404(30.05%)

AES 128/128/10
CBC-MAC with
16-byte MAC

21304(16.64%)

2404(30.05%)

AES 128/128/10
CTR mode

21292(16.63%)

2404(30.05%)

AES 128/128/10
Only AES
Encryption/
decryption

192

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

Table III results are obtained after installing software
Advanced Encryption Algorithm using TinyOS-2.1.0 with
TinyHop routing algorithm and with our framework into
MicaZ and IRIS mote. The TinyHop routing algorithm
occupies more memory as to manage routing information. Also
we have removed distinction between key and frame counter
[13].

TABLE III

Memory Utilization of Software AES USING TIN YOS -2. 1.0

Framework WITH TINYHOP Routing Algorithm

Mote

ROM

occupied

in bytes

(percentage)

RAM

occupied

in bytes

(percentage)

Name of Cipher

and its

configuration

MicaZ

29848(23.31%)

3739(93.4%)

AES 128/128/10
CCM mode with
16-byte

29850(23.32%)

3663(91.5%)

AES 128/128/10
CBC-MAC with
16-byte

29774(23.26%)

3435(85.87%)

AES 128/128/10,
CTR mode

29676(23.18%)

3359(83.97%)

AES 128/128/10,
Only AES
Encryption/
Decryption

IRIS

28758(22.46%)

3923(49.03%)

AES 128/128/10,
CCM mode with
16-byte MAC

28760(22.46%)

3831(47.88%)

AES 128/128/10,
CBC-MAC with
16-byte MAC

28682(22.40%)

3555(44.43%)

AES 128/128/10,
CTR mode

28650(22.38%)

3359(41.98%)

AES 128/128/10,
only AES
Encryption/
Decryption

V. Conclusion And Future Work

The framework implemented is unique for WSN
applications. It has general purpose security library suite

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

Vol 8, No.3, 2010
composed of existing security algorithms for popular sensor
node hardware platforms. The framework also supports
existing versions of TinyOS providing the application
developer an abstraction to low level implementation of the
desired security algorithm chosen. We also Implemented and
integrated software AES and hardware AES driver for
CC2420/CC2430/CC2431 radio chip [5] with same mode
settings like encryption-only, CBC-MAC integrity-only,
counter encryption only and counter encryption - CBC-MAC
integrity with dynamic and static key support. We have tested
the code using with and without TinyHop routing algorithm.

The work presented in this paper can be applied to various
mote platforms. Future extensions are also possible to integrate
various security modes with simple plug-in and plug-out
configuration file. Currently the framework is in-built with
WSN IDE [21] that creates templates for application and then
application developer has to wire the modules manually. In the
next design we extend this framework and security modules
wiring with drag and drop options.

References

[I] K. Piotrowski, P. Langendoerfer, S. Peter,"How public key cryptography
influences wireless sensor node lifetime," Proc. of fourth ACM
workshop on Security of adhoc and sensor networks {^A^'H 06), pages
169-176,2006. ACM..

[2] A. Wander, N. Gura, H. Eberle, V. Gupta, S. C. Sliantz,"Energy analysis
of public-key cryptography for wireless sensor networks," Proc. of Third
IEEE International Conference on Pervasive Computing and
Communications (PerCom 05), pages 324-328, March 2005.

[3] Chris Karlof, Naveen Sastry, David Wagner, "TinySec: Link Layer

Encryption for Tiny Devices", ACM Conference on Embedded

Networked Sensor Systems, 2004.
[4] Tieyan Li, Hongjun Wu, Xinkai Wang, Feng Bao,"SenSec Design, I2R

Sensor Network Flagship Projecf ; Technical Report TRvl.O.
[5] 2.4 GHz IEEE 802.1 5. 4/ZigBee -ready RE Transceiver, Chipcon

Products from Texas Instruments, http://www.ti.com .

[6] Mark Luk, GhitaMezzour, Adrian Perringm, Virgil Gligor,"MiniSec: A
Secure Sensor Network Communication Architeture", ACM
International Conference on Information Processing in Sensro
Networks, April 2007.

[7] Phillip Rogaway, Mihir Bellare, John Black, "OCB: A block-cipher
mode of operation for efficient authenticated encryption", ACM
Transactions on Information and System Security (TISSEC), Volume 6,
Issue 3, pp.365-403, August 2003.

[8] Devesh Jinwala, Dhiren Patel, K S Dasgupta,"Optimizing the Block
Cipher Modes of Operations Overhead at the Link Layer Security
Framework in the Wireless Sensor Networks," Proceedings of the 4^^
International Conference on Information Systems Security, LNCS,
pp.258-272, Springer Berhn/Heidelberg, 2008.

[9] Miklos Maroti, Gyula Simon, Branislav Kusy, and Akos Ledeczi,"The
flooding time synchronization protocol," in Proceedings of the 2nd
international conference on Embedded networked sensor systems,
Baltimore, MD, USA, Nov. 2004, pp. 3949.

[10] Law, Y.W., Doumen, J., Hartel, P, "Survey and benchmark block
ciphers for wireless sensor networks"; ACM Transactions on Sensor
Networks, 2006.

[II] RC6 cipher - http://people.csail.mit.edu/rivest/Rc6.pdf

[12] Mingbo Xiao, Xudong Wang, Guangsong Yang,"Cross-Layer Design
for the Security of Wireless Sensor Networks", Proceedings of the 6th
World Congress on Intelligent Control and Automation, Jnue 21-
23,2006 Dalian, China,pp(104-108).

[13] Naveen Sastry and David Wagner, "Security Considerations for IEEE
802.15.4 Networks". ACM Workshop on Wireless Security WiSe 2004,
October 2004.

193

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

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

Vol 8, No.3, 2010

[14] Kalpana Sharma, M.K. Ghose, Kuldeep, "Complete Security Framework
for Wireless Sensor Networks", (IJCSlS) International Journal of
Computer Science and Information Security, Vol. 3, No. 1, 2009.

[15] Rivest R; "The RC5 Encryption Algorithm", Proceedings of the Second
International Workshop on Fast Software Encryption, 1994.

[16] Skipjack - a representative of a family of encryption algorithms as part
of the NSA suite of algorithms;

http://csrc.nist.gov/groups/STM/cavp/documents/skipiack/skipiack.pdf

[17] J.Daemen, V.Rijmen, "AES Proposal: Rijndael",

http://www. esat.kuleuven . ac .be/rij men/rij ndael/rij ndaeldoc V2 . zip .

[18] Helger Lipmaa, Phillip Rogaway and David Wagner. Comments to
NIST Concerning AES-modes of Operations: CTR-mode Encryption. In
Symmetric Key Block Cipher Modes of Operation Workshop,
Baltimore, Maryland, USA, October 20, 2000.

[19] IRIS motes - http://www.xbow.com/Products/wproductsoverview.aspx .

[20] Whiting, D., Housley, R. and N. Ferguson, "AES Encryption

Authentication Using CTR Mode CBC-MAC," IEEE P802.ll doc

02/001r2, May 2002.

[21] http://www.ubicomp.in/afwa/

[22] http://www.tinvos.net/

[23] http://iava.sun.eom/i2se/l.4.2/docs/guide/securitv/CrvptoSpec.html

TABLE IV

Mode Settings For Various Security Algorithms

Algorithm

ctrlO

len

AES:

Stand Alone

AES STANDALONE ||KEY0

AES STANDALONE ||KEY1

AES:

In-line

AES INLINE ||[T||R]XKEY[0I1]
II CBC_MAC II SEC_M[0 - 7]

AES INLINE ||[T||R]XKEY[0I1]
II CTR

AES INLINE ||[T||R]XKEY[0I1]
IICCM

AES_INLINE [[T || R]XKEY[OI 1 ]
II CBC MAC 1 SEC M[0 - 7]

TinySec

TINYSEC_AUTH_ONLY

TINYSEC_ENCRYPT_AND_AUTH

TINYSEC_DISABLED

TINYSEC_RECEIVE_
AUTHENTICATED

TINYSEC_RECEIVE_CRC

TINYSEC_RECEIVE_ANY

MiniSec

MINISECU

TABLE V

Mode Settings For Various Security Algorithms

Algorithm

Macro

Description

AES

AES_ENC

only AES encryption

AES_DEC

only AES decryption

CTR

AES Counter-Mode

Encryption

This mode is not secure,

to make compatible with

hardware

AES we provided this

option.

CBC_MAC

SEC_M[0-7]

4, 6, 8,10,12,14, 16 byte
-MAC. AES CBC-MAC,
it provides only
Authentication.

CCM

SEC_M[0-7]

4,6,8,10,12,14,
16-byte MAC. AES
Counter mode
encryption and CBC-
MAC authentication

194

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

Optimizing the Application-Layer DDoS
Attacks for Networks

P.Niranjan Reddy
Head, Dept. of CSE
KITS, Warangal
A.P. , INDIA.
npolala@vahoo.co.in

K.Praveen Kumar

Lecturer, Dept, of CSE

KITS, Warangal

A.P. , INDIA.

praveen kumar35@vahoo.co.in

M .Preethi

Lecturer, Dept, of CSE

KITS, Warangal

A.P. INDIA.

preethi Q290@vahoo. CO. in

Abstract - The main aim of the proposed
framework is to implement the Application-
Layer DDoS Attacks Optimizing for Popular
Websites that employing legitimate HTTP
requests to flood out victim resources and to
implement an effective method to identify
whether the surge in trafflc is caused by App-
DDoS attackers or by normal Web surfers.

Keywords: Terms - Application-layer, distributed
denial of service (DDoS), popular website

INTRODUCTION

Distributed Denial of Service (DDoS)
attack is an attempt to make a computer
resource unavailable to its intended users. This
attack has caused severe damage to servers and
will cause even greater intimidation to the
development of new Internet services.
Traditionally, DDoS attacks are carried out at
the network layer, such as ICMP flooding,
SYN flooding, and UDP flooding, which are
called Net-DDoS attacks. The intent of these
attacks is to consume the network bandwidth
and deny service to legitimate users of the
victim systems. Among these floodings
another attack is Botnet[21] which is a network
of compromised hosts or bots, under the
control of a human attacker known as the
botmaster. Botnets are used to perform
malicious actions, such as launching DDoS
attacks, sending spam or phishing emails and
so on. Thus, botnets have emerged as a threat
to internet community. Peer to Peer (P2P) is a
relatively new architecture of botnets. These
botnets are distributed, and small. So, they are
difficult to locate and destroy.

Since many studies have noticed this type
of attacks and have proposed different schemes
(e.g., network measure or anomaly detection)
to protect the network and equipment from
bandwidth attacks, it is not as easy as in the
past for attackers to launch the DDoS attacks
based on network layer. To implement DDoS,

a worm like program is created to simulate
self-propagation onto many hosts on a
network.

When the simple Net-DDoS attacks fail,
attackers shift their offensive strategies to
application-layer attacks and establish a more
sophisticated type of DDoS attacks.

To overreach detection, the attackers
attacking the victim web servers by HTTP
GET requests (e.g., HTTP flooding) and
pulling large image files from the victim server
in overwhelming numbers. In another instance,
attackers run a massive number of queries
through the victim's search engine or database
query to bring the server down [4]. Such
attacks called as application-layer DDoS (App-
DDoS) attacks. The MyDoom worm [23] and
the CyberSlam [3] are all instances of this type
attack.

On the web, "flash crowd" [6], [7] refers to
the situation when a very large number of
users simultaneously access a popular
website[13], which produces a surge in
traffic[8] to the Website and might cause the
site to be virtually unreachable. Because burst
traffic and high volume are the common
characteristics of App-DDoS attacks and flash
crowds, it is not easy for current techniques to
distinguish them merely by statistical
characteristics of traffic.

II. RELATED WORK

The researchers made an attempt to detect
DDoS attacks from three different layers: IP
layer, TCP layer, and application layer. From
all of these views, researchers are looking into
various approaches to differentiate normal
traffic from the attack one.

Maximum DDoS -related research has
concentrated on the IP layer. These techniques
attempt to detect attacks by analyzing specific
features, e.g., arrival rate or header
information. For example, Cabrera et al [9]
used the management information base (MIB)
data which include parameters that indicate

195

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

different packet and routing statistics from
routers to achieve the early detection. Yuan et
al [14] used the cross-correlation analysis to
capture the traffic patterns and then to decide
where and when a DDoS attack possibly
arises. Mirkovic et al [15] monitored the
asymmetry of two-way packet rates and to
identify attacks in edge routers. Other
statistical approach for detection of DDoS
attacks includes IP addresses [16] and time-to-
live (TTL) values [17].

One of the most important research area is
TCP layer for detecting DDoS attack. For
example, authors [9] mapped ICMP, UDP, and
TCP packet statistical abnormalities to specific
DDoS attacks based on MIB. Wang et al [18]
used the TCP SYN/FIN packets for detecting
SYN flooding attacks. In [18], DDoS attacks
were discovered by examining the TCP packet
header against the welldefined rules and
conditions and differentiated the difference
between normal and abnormal traffic. Noh et
al [19] attempted to detect attacks by
computing the ratio of TCP flags (including
FIN, SYN, RST, PSH, ACK, and URG) to
TCP packets received at a Web server.

Ranjan et al [11] used statistical methods
to detect characteristics of HTTP sessions and
employed rate-limiting as the primary defense
mechanism. Yen et al [12] defended the
application DDoS attacks with constraint
random request attacks by the statistical
methods. Other researchers combated the App-
DDoS attacks by "puzzle," see, e.g., [20]. Jung
et al's work [7] he used two properties to
distinguish the DoS and normal flash crowd: 1)
a DoS event is due to an increase in the request
rates for a small group of clients while flash
crowds are due to increase in the number of
clients and 2) DoS clients originate from new
client clusters as compared to flash crowd
clients which originate from clusters that had
been seen before the flash event.

All^^r-s IMc-hil^

Ag&ms

Attack c<:i4iwn3FKE&
Attack iraflic

III. App-DDoS ATTACKS

In our opinion, the DDoS attack detection
approaches in different scenario can be
clustered as:

• Net-DDoS attacks versus stable
background traffic.

• Net-DDoS attacks versus flash crowd.

• App-DDoS attacks versus stable
background traffic.

• App-DDoS attack versus flash crowd.

The first two scenarios have been well
studied and can be dealt with by most existing
DDoS detection schemes while the other two
groups are quite different from the previous
ones.

This is a simple comparison between the
existing system and proposed system.

Existing System

Proposed System

Consume the network
bandwidth and deny

Bandwidth is
effectively used

Service to legitimate
users.

Service to all users if
and only if the
resource is available.

Abnormalities are
identified and denied

Identifying
abnormalities and
serve them in
different priority
queues.

Large amount of data
is required to train.

Identifies

abnormalities with
small amount of
training data

Only positive data's
are used to train

More accurate
identification

Identifying abnormal
traffic and filter the
network

Identifying most
abnormal traffic and
filter when the
network is heavily
loaded.

IV.

DETECTION PRINCIPLE

Fig 1 . How the attacker can perform attacks on
App-layer.

We can cluster the Web surfers and
evaluate their contributions to the anomalies in
the aggregate Web traffic. Here the DDoS
attack is caused only by the authenticated users
of the Website. Then, different priorities are
given to the clusters according to their
abnormalities and serve them in different
priority queues. The most abnormal traffic
may be filtered when the network is heavy
loaded. Here the priority level of the cluster is
given based on the access time only. The

196

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

different modules in the implementation are
given below in the next section.

Web Server Module

MODULES

• Login

• Registration

• Database Design

• Application Design
Attacker Module

• Normal User

• Abnormal User
Flash crowd dismisser

• Data preparation

• Training

Monitoring

Web Server Module

Web servers are computers on the internet
that host website serving pages to viewers
upon request. This service is referred to as
web hosting. Every web server has a unique
address so that other computers connected to
the internet know where to find it on the vast
network. When your request reaches its
destination, the web server that hosts website
sends the page in HTML code to your
ipaddress [5]. This return communique travels
back through the network. Your computer
receives the code and your browser interprets
the HTML code then displays the page for you
in graphic form.

Login module is general for all kinds of Web
application to authenticate and authorize the
user's access to the site. To make valid users
only can access the site, preventing the
unauthorized access.

Registration:

This module is also common to all the web
application. Making the users to access the
site based upon the registration. It may be free
or cost. In order to authenticate and authorize
a user, registration is must.

Application Design:

An application which suits for our project
is designed using the HTML code and the
relevant technologies.

Database Design:

Once the application has designed then
Database has to be designed. Here creation of
the tables related to our project is created.

Number of tables needed for the application
has to be decided and the tables are created for
that.

F. Attacker Module:

This module consists of webpage through
which Attackers attack the victim Web servers
by HTTP GET requests (e.g., HTTP Flooding)
and pulling large image files from the victim
server in overwhelming numbers. In another
instance, attackers run a massive number of
queries through the victim's search engine or
database query to bring the server down. Very
large number of attackers simultaneously
accesses a popular Website, which produces a
surge in traffic to the website and might cause
the site to be virtually unreachable.

16000

12000

6 8000

n
E

4000

request num

2000 4000 6000 8000 10000
index of time unit (5sec)

Fig 2. Simple network Attack path

Normal User: The user login in and acts as
a normal user there is no abnormality in his
behaviors.

Abnormal User: The user login in and acts
as the abnormal user, the behaviors of the users
are found to be abnormal (.e.g., attacker who is
causing the DDoS attack over the target site).

Flash crowd dismisser: This model is first
trained by the stable and low- volume web
workload whose normality can be ensured by
most existing anomaly detection systems, and
then it is used to monitor the following web
workload for a period of 10 min. When the
period is past, the model will be updated by the
new collected web workload whose normality
is ensured by its entropy fitting to the model.
Then, the model is used in anomaly detection
for the next cycle. If some abnormalities
hiding in the incoming web traffic are found,
the "defense" system will be implemented.

197

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

VI.

ARCHITECTURE

VII.

APPLICATION

The process is divided into three phases:

1 . Data preparation.

2. Training

Web servers application DoS attacks allow
for efficient DoS with only little resources at
hand, and thus pose a Serious threat to
organization.

Monitoring

Data preparation: The main purpose of
data preparation is to compute the AM by the
logs of the web server. Various user data are
collected while accessing the sites.

Training: Train the collected data for the
abnormalities. Check the user behaviour with
the predefined threshold. If the user exists the
threshold are named as the abnormal users
(.eg., attacker). Likewise all the user data are
trained and found out the abnormality.

Monitoring: In the Monitoring phase,
checks for the resource availability. If the user
found to be attacker then the resource is
available means allows that user to access the
sites (Simply allow the attacker also if and
only if the resource is available). If the
resource is not available means, temporarily
deny that user to access the site.

Hide speed internet.

Mobility tracking
networks.

in wireless

IMay

jimpori java.awt.BorderLayout;

limport java.awt. Color;

import java.awt. Container;

import java.awt.Font;

^import java.awt.Image;

Wimport java.awt.Too]ldt;

import java.awt. event. ActionE vent;

import java.awt. event. AdionListener;

import j ava. awt. e vent.KeyEvent ;

M • :~ . . : I

Figure 4. Time delay while transferring
the file with out attack.

Figure 3. Proposed Architecture

198

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

aaaaaaaaaaaaaaaaaa -

=
Health Information System

Banldng System

EEIS

Digital Image Processing

Chat Express

Customer Relationship Support

Network Banking System

Magazine Management System

Survey of Organisation(SIIRVEY TOOL S

ecure Net Auction System ^

^1 III 1 ^^

Delay

2SS60milli sec

Figure 5. Time delay while transferring the
file with attack.

rile Edr wj=,j Fs.ettes Tools Help

l^iH

■

IJ^^^"

O^aol > ",^ S^ *

Search N"'_)Y''3"°f''^^

6 0-

P*,

E-DBfi'S

Addtes. [^ hffp //localhosf SOSO/dCuJJS/iride i

Moiiitt>mig Application layw DDOS Attacks

Enter Your Search ;Swaroop_

^ http; //localhost ; 808Q/DDOS5/Gontroller?name=swaroop

Figure 6. Session closing when the
Attacking is found.

VIII. CONCLUSION

Creating defenses for attacks requires
monitoring dynamic network activities in order
to obtain timely and signification information.
While most current effort focuses on detecting
Net-DDoS attacks with stable background
traffic, we proposed detection architecture in
this paper aiming at monitoring web traffic in
order to reveal dynamic shifts in normal burst
traffic, which might signal onset of App-DDoS
attacks during the flash crowd event. Our
method reveals early attacks merely depending
on the document popularity obtained from the
server log.

REFERENCES

[1]. IEEE/ACM Transaction on Networking, Vol. 17,
No. 1, February, 0209.

[2]. Http ://www. linuxsecurity.com/resource_files/intrus
iondetection/ddos-whitepaper.html.

[3]. http://en.wikipedia.org/wiki/Denial-of-serviceattac
k.

[4]. K. Poulsen, "FBI Busts Alleged DDoS Mafia,"
2004.[Online].Available:
http://www.securityfocus.com/news/94 1 1

[5]. T. Peng and K. R. M. C. Leckie, "Protection from
distributed denial of service attacks using history-
based IP filtering," in Proc. IEEE Int. Conf.
Commun., May 2003, vol. 1, pp. 482-486.

[6]. I. Ari, B. Hong, E. L. Miller, S. A. Brandt, and D.
D. E. Long, "Modeling, Analysis and Simulation of
Flash Crowds on the Internet," Storage Systems
Research Center Jack Baskin School of
Engineering University of California, Santa Cruz
Santa Cruz, CA, Tech. Rep. UCSC-CRL-03-15,
Feb. 28, 2004 [Online]. Available:
http://ssrc.cse.ucsc.edu/, 95064.

[7]. J. Jung, B. Krishnamurthy, and M. Rabinovich,
"Flash crowds and denial of service attacks:
Characterization and implications for CDNs and
web sites," in Proc. 11th IEEE Int. World Wide
Web Conf., May 2002, pp. 252-262.

[8]. W. Leland, M. Taqqu, W. Willinger, and D.
Wilson, "On the selfsimilar nature of ethernet
traffic (extended version)," IEEE/ACM Trans.
Networking, vol. 2, no. 1, pp. 1-15, Feb. 1994.

[9]. J. B. D. Cabrera, L. Lewis, X. Qin, W. Lee, R. K.
Prasanth, B. Ravichandran, and R. K. Mehra,
"Proactive detection of distributed denial of service
attacks using MIB traffic variables a feasibility

199

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

study," in Proc. lEEE/IFIP Int. Symp. Integr. Netw.
Manag., May 2001, pp. 609-622.

[10]. S. Noh, C. Lee, K. Choi, and G. Jung, "Detecting
Distributed Denial of Service (DDoS) attacks
through inductive learning," Lecture Notes in
Computer Science, vol. 2690, pp. 286-295, 2003.

[11]. S. Ranjan, R. Swaminathan, M. Uysal, and E.
Knightly, "DDoS-resilient scheduling to counter
application layer attacks under imperfect
detection," in Proc. IEEE INFOCOM, Apr. 2006
[Online]. Available:http://www-ece. rice.edu/netwo
rks /papers/ dos-sched.pdf

[12]. W. Yen and M.-F. Lee, "Defending application
DDoS with constraint random request attacks," in
Proc. Asia-Pacific Conf. Commun., Perth, Western
Australia, Oct. 3-5, 2005, pp. 620-624.

[13]. C. Roadknight, 1. Marshall, and D. Vearer, "File
popularity characterisation," ACMSIGMETRICS
Performance Eval. Rev., vol. 23, no. 4, pp. 45-50,
Mar. 2000.

[14]. J. Yuan and K. Mills, "Monitoring the
macroscopic effect of DDoS flooding attacks,"
IEEE Trans. Dependable and Secure Computing,
vol.2, no. 4, pp. 324-335, Oct.-Dec. 2005.

[15]. J. Mirkovic, G. Prier, and P. Reiher, "Attacking
DDoS at the source," in Proc. Int. Conf. Network
Protocols, 2002, pp. 312-321.

[16]. T. Peng and K. R. M. C. Leckie, "Protection from
distributed denial of service attacks using history-
based IP filtering," in Proc. IEEE Int.Conf
Commun., May 2003, vol. 1, pp. 482-486.

[17]. B. Xiao, W. Chen, Y. He, and E. H.-M. Sha, "An
active detecting method against SYN flooding
attack," in Proc. 11th Int. Conf. Parallel Distrib.
Syst., Jul. 20-22, 2005, vol. 1, pp. 709-715.

[18]. H.Wang, D. Zhang, and K. G. Shin, "Detecting
SYN flooding attacks," in Proc. IEEE INFOCOM,
2002, vol. 3, pp. 1530-1539.

[19]. S. Noh, C. Lee, K. Choi, and G. Jung, "Detecting
Distributed Denial of Service (DDoS) attacks
through inductive learning," Lecture Notes in
Computer Science, vol. 2690, pp. 286-295, 2003.

[20]. S. Kandula, D. Katabi, M. Jacob, and A. W.
Berger, "Botz-4-Sale: Surviving Organized DDoS
Attacks that Mimic Flash Crowds,"MlT, Tech. Rep.
TR-969, 2004 [Online]. Available:

http://www.usenix.org/events/nsdi05/tech/kandula/
kandula.pdf

[21]. James Binkley and Suresh Singh. An algorithm for
anomaly-based botnet detection. In Proceedings of

Steps to Reducing Unwanted Traffic on the Internet

Workshop (SRUTI '06), 2006.

[22]. Zhichun Li, Anup Goyal, and Yan Chen.

Honeynet-based Botnet Scan Traffic Analysis

Northwestern University, Evanston, IL 60208

{lizcag o210,ychen}@cs.northwestern.edu.

[23]. "Incident Note lN-2004-01 W32/Novarg. A
Virus," CERT, 2004. [Online]. Available:
http://www.cert.org/incident_notes/ IN-2004-

01. html

AUTHORS PROFILE

P.NIRANJAN REDDY received
the B.E Computer Science from
Nagpur University in 1992 and
M.Tech (Computer Science and
Engineering) from NIT,

Warangal in the year 2001. He
worked as a Lecturer and
Assistant Professor in the
department of CSE of KITS,
Warangal, Since 1996. He is
doing a part-time research in
Kakatiya University, Warangal
since 2007. He authored two text
books. Theory of computation
and Computer Graphics in the
field of Computer Science. He
published 3 papers inlnternational
Journals and 6 papers in
International Conferences.

K.PRAVEEN KUMAR has

been working as a lecturer in
Dept. of CSE, KlTS,Warangal
in Andhra Pradesh,lNDlA for
the last 2 years. He has
completed his B.tech and
M.tech from KITS warangal.
He has published a research
paper at a National level
Conference

M.PREETHI has been
working as a lecturer in
Dept of CSE in KITS,
Warangal in Andhra
Pradesh, INDIA for the last
3years. She took her
M.Tech degree from
KlTS,Warangal.

200

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

Survey - New Routing Technique for Grid Computing

R.RAMESHKUMAR,

Research Schlolar,

J.N.T.University,

Kukatpally,

Hyderabad.

rrameshl968@gmail.com

Dr. A.DAMODARAM
Director/ U.G.C Academic Staff College,.
J.N.T.University,
Kukatpally,
Hyderabad.
adamodaram(%intuap.ac.in

Abstract- Trust plays an indispensable role in grid
computing. Trust-management systems provide applications
with a standard interface for getting answers to such
questions and provide users with a standard language for
writing the policies and credentials that control what is
allowed and what isn't. Using a trust-management system
for controlling security-critical services frees the application
developer from a number of often difficult design and
implementation issues and allows users to take advantage of
a flexible, standard, application-independent language for
specifying policy. In this paper, we develop trust
management architecture for trust enhanced Grid security
incorporating a novel trust model which is capable of
capturing various types of trust relationships that exist in a
Grid system and providing mechanisms for trust evaluation,
recommendations and update for trust decisions. The
outcomes of the trust decisions can then be employed by the
Grid security system to formulate trust enhanced security
solutions. Here we put forth ant algorithm for
implementation. The ant colony algorithm is an algorithm
for finding optimal paths that is based on the behavior of
ants searching for food. At first, the ants wander randomly.
When an ant finds a source of food, it walks back to the
colony leaving "markers" (pheromones) that show the path
has food. When other ants come across the markers, they are
likely to follow the path with a certain probability. If they
do, they then populate the path with their own markers as
they bring the food back. As more ants find the path, it

gets stronger until there are a couple streams of ants
traveling to various food sources near the colony.

Key words - Grid computing, security, trust. Ant Colony,
Service Request.

I. INTRODUCTION

Routers use routing algorithms to find the best
route to a destination. When we say "best route," we
consider parameters like the number of hops (the trip a
packet takes from one router or intermediate point to
another in the network), time delay and communication cost
of packet transmission. Based on how routers gather
information about the structure of a network and their
analysis of information to specify the best route, we have
two major routing algorithms: global routing algorithms and
decentralized routing algorithms. In decentralized routing
algorithms, each router has information about the routers it
is directly connected to ~ it doesn't know about every router
in the network. These algorithms are also known as DV
(distance vector) algorithms. In global routing algorithms,
every router has complete information about all other
routers in the network and the traffic status of the network.
These algorithms are also known as LS (link state)
algorithms.
• Routing Components

Routing involves two basic activities: determining
the optimal routing paths for destination networks and

201

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

transporting information groups, also known as packets,
through an internetwork. Within the context of routing, the
latter can be referred to as switching.

• Path Determination

A metric is a standard of measurement, such as
path length, that is used by routing algorithms to determine
the optimal path to a destination. To aid in this process of
path determination, routing algorithms initialize and
maintain routing tables, which contain route information.
This information can vary widely depending on which
routing algorithm generated the routes. Routing algorithms
fill routing tables with a list of networks and its
corresponding "next hop" on the way its destination. When
a router receives an incoming packet, it checks the
destination address and attempts to associate this address
with a next hop.
Algorithm Types

• Static versus dynamic

• Single-path versus multi-path

• Link state versus distance vector

• Dynamic vs. Static

Static routing algorithms are hardly algorithms at
all, but are table mappings established by the network
administrator prior to the beginning of routing. These
mappings do not change unless the network administrator
alters them. Algorithms that use static routes are simple to
design and work well in environments where network traffic
is relatively predictable and where network design is
relatively simple.

Because static routing systems cannot react to
network changes, they generally are considered unsuitable
for today's large, changing networks. Most of the dominant
routing algorithms in the 1990s are dynamic routing
algorithms, which adjust to changing network circumstances
by analyzing incoming routing update messages. If the
message indicates that a network change has occurred, the

routing software recalculates routes and sends out new
routing update messages. These messages permeate the
network, stimulating routers to rerun their algorithms and
change their routing tables accordingly.
Dynamic routing algorithms can be supplemented with
static routes where appropriate. A router of last resort (a
router to which all unroutable packets are sent), for
example, can be designated to act as a repository for all
unroutable packets, ensuring that all messages are at least
handled in some way.

• Single-Path vs. Multipath

Some sophisticated routing protocols support
multiple paths to the same destination. Unlike single-path
algorithms, these multipath algorithms permit traffic
multiplexing over multiple lines. The advantages of
multipath algorithms are obvious: They can provide
substantially better throughput and reliability.

• Link State vs. Distance Vector

Link-state algorithms (also known as shortest path
first algorithms) flood routing information to all nodes in
the internetwork. Each router, however, sends only the
portion of the routing table that describes the state of its own
links. Distance- vector algorithms (also known as Bellman-
Ford algorithms) call for each router to send all or some
portion of its routing table, but only to its neighbors. In
essence, link- state algorithms send small updates
everywhere, while distance- vector algorithms send larger
updates only to neighboring routers.

Because they converge more quickly, link- state
algorithms are somewhat less prone to routing loops than
distance- vector algorithms. On the other hand, link- state
algorithms require more CPU power and memory than
distance- vector algorithms. Link-state algorithms,
therefore, can be more expensive to implement and support.
Despite their differences, both algorithm types perform well
in most circumstances.

202

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

Trust Management Architecture

Using a trust-management system for controlling
security-critical services frees the application developer
from a number of often difficult (and subtle) design and
implementation issues and allows users to take advantage of
a flexible, standard, application-independent language for
specifying policy. Before trust management, every
application had to provide its own mechanisms for
specifying policy, interpreting credentials, and binding user
authentication with the authorization to perform
"dangerous" operations. Trust-management systems, on the
other hand, provide a simple interface that takes care of all
of these things. All the application designer has to do is
identify the trust management questions in the application
and formulate appropriate queries to the trust-management
system.

Remote (Untrustedji

cttmt [^^

client/
peer/
usatf
etc

Appllcatkjn

"TT'

appficalion
pohcy

diani/
peer/

USBTf

etc

Local

KeyNote

Intvrpralw-

"X.

Apui

c:atlon

KeyNcte
API

disnl/
peerf
user/
etc

appiicaiion
policy

dienl/

user/
etc

fteywote
API

applicalion
pdicv

\ 1

Crsdmtlal
tPKl) 1

OSflf/
etc

.—Irad

jKiundBTy

Fig. 1 . Keynote trust Management Architecture

II. Impact of ant algorithm on grid computing

Grid computing is a term used to describe both a
platform and type of application. A Grid computing platform
dynamically provisions, configures, reconfigures, and de
provisions servers as needed. Servers in the Grid can be
physical machines or virtual machines. The grid computing
environment typically include other computing resources
such as storage area networks (SANs), network equipment,
firewall and other security devices. Grid computing also
describes applications that are extended to be accessible

through the Internet. These Grid applications use large data
centers and powerful servers that host Web applications and
Web services. Anyone with a suitable Internet connection and
a standard browser can access a cloud application. The
primary components of Grid architecture are:

• Users/Brokers: Users or brokers acting on their
behalf submit service requests from anywhere in
the world to the Data Center and Grid Servers to be
processed.

• SLA (Service Level Agreements) Resource
Allocator: The SLA Resource Allocator acts as the
interface between the Data Center/Grid service
provider and external users/brokers. It requires the
interaction of the defined scheduled mechanisms to
support SLA-oriented resource management.

To access the grid resources and execution, it can
be divided in to three phases like resource recovery,
scheduling, and executing. In the second phase find the best
match between the set of jobs and available resources. The
second phase is a NP-hard Problem [6]. The computational
grid is a dynamic and unpredictable behavior. They are:

Computational performance of each resource varies
from time to time.

• The connection between computers and mobile
phones may be unreliable.

• The resources may join or give up the grid at any
time

• The resource may be occupied without a
notification.

The scheduling of grid architecture is dynamic in
nature and moreover Grid middleware and applications are
using local scheduling and data co-scheduling. The approach
of replication has been also applied and assisted in
scheduling and optimization of replication. There are
different existing algorithms like the Genetic algorithm (GA)
is used for searching large solution space. On other hand.

203

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

simulated Annealing (SA) is an iterative technique that
considers only one possible solution for each meta-task at a
time.

ACO algorithm can be interpreted as parallel
replicated Monte Carlo (MC) systems. MC systems are
general stochastic simulation systems, that is, techniques
performing repeated sampling experiments on the model of
the system under consideration by making use of a stochastic
component in the state sampling and/or transition rules.
Experimental results are used to update some statistical
knowledge about the problem. In turn, this knowledge can
also be iteratively used to reduce the difference in the
estimation of the described variables and directing the
simulation process toward the most interesting state space
regions. Analogously, in ACO algorithms the ants sample the
problem's solution space by repeatedly applying a
stochastic decision policy until a feasible solution of the
considered problem is found. The sampling is realized
concurrently by a collection of different instantiated replicas
of the same ant type. Each ant "experiment" allows to
adaptively modifying the local statistical knowledge on the
problem structure. The algorithm is recursive in nature.

III. PROPOSED Algorithm

The classic ant colony algorithm can be described as
follows:

Step 1 . Initialize
Step 2. Loop

/* An iteration */

Step 3. Each ant is positioned on a starting node.
Loop /* A step */

Step 4. Each ant applies a state transition
rule to incrementally build a solution and a local
pheromone updating rule until all ants have built a
complete solution

Step 5. Global pheromone updating rule is applied until
end condition.

Step 6. Stop further iterations

Each edge between node (r, s) has a distance or cost
associate 8 (r, s) and a pheromone concentration T (r, s).
The equation 1 is the state transition rule, which is a
probabilistic function for each node u, which has not been
visited by each placed ant on node r.

_ [r(...)lbKr,.)]-^

(1)

The parameter P determine the relevance of the
pheromone concentration compared with the distance or
cost, T (r, s) Global pheromone updating rule can be applied
as:

7-{r, s) — (1 - ct)r(rj s) + irzlrjt(r. s)

(2)

Where a is the pheromone evaporation factor
between and 1 and Ax, (r, s) is the reverse of the distance or
cost done by ant k, if (r, s) is its path and is if it is not in the
path. The steps can be modified to manage grid architecture.
The grid is visualized is the collection of clustered services,
hence the live services of grid behaves like an ant, when it
find its file object, the ant died. Subsequently, considering the
prime component of grid computing, the compute grid and
storage grid can be modeled as virtual services of grid.
Every time a request is processed on a grid cluster site, T is
updated for all the site connections and thus the "(2)" can be
modified by associating a parameter t .

t(r,s) = (1-oc )T(r,s) + 2AT, (r,s)

(3)

The dot operator represents time for each grid scheduling
service. Therefore, the a is introduced which expresses
the evaporation factor under time slot of grid service. The
heuristic can be divided into two categories for grid-based
services e.g. on-line mode service and the batch mode service.
In online mode, whenever a request arrive, it immediately

204

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

allocate to the first free resource allocator. The arrival order of
the request in grid is important in this method. Here, each
service request is considered only once for matching and
scheduling. In batch mode, the requests are collected; the
scheduler considers the approximate execution time for each
task and use heuristic approach to possibly make better
decision. The function free [j] - return time, when the
resource allocators Mj will be free. We consider,

free[j] = lA+ETij,
where, Ia is the initial time slot of request of service made on
the grid architecture and ETy is the execution time matrix of
request Vi on resource allocator m.

The scheduling of resource allocator on the grid
service proposes the probability of servicing the request:

phij Tiij ( 1/ETij )

PiJ =

I phii Tiii ( 1/ETij )

Where, - rjij is the attractiveness of the move as
computed by heuristic information indicating a prior
desirability of that move, phy fast and accuracy of the grid
service in the past (with lower a) to make that particular
move (it represents therefore a posterior service
accomplishment indication of the desirability of that request)
ETij Execution Matrix of service and resource allocator. In
this proposed model, we select the highest probability's 'i'
and 'j' are the next request of service ri executed on the
resource allocator].

IILCONCLUSION

This paper is the first to develop trust management archi-
tecture for Grid security solutions based on Subjective
Logic. We have identified the requirements of trust
management for Grid computing from security point of
view. We then develop trust management architecture to
meet the requirements defined. This trust management
architecture is designed to be transparent to the Grid

platforms. It thus can easily be instantiated in a practical
application as a separate layer, and thus allows seamless
integration to different Grid computing platforms. Once
instantiated this architecture allows explicit trust policies to
be defined and managed. In this paper, a heuristic algorithm
based on modified ant colony optimization has been proposed
to initiate the service load distribution under grid computing
architecture. The simulation doesn't consider the fault
tolerance issues. Due to absence of any restore time in service
and resource allocator distribution, it is expected that
continuous ant colony with other modified parameters could
demonstrate better results compared to other optimization
models, even in faulty service request and disrupted resource
allocator.

REFERENCES

[1] F. A. Maheswaran and M., "Evolving and managing
trust in grid computing systems," in Proceedings of the
2002 IEEE Canadian Conference on Electrical
Computer Engineering, 2002.

[2] L. Shen and J., "A mission-aware behavior trust model
for grid computing systems," in Intl Workshop on Grid
and Cooperative Computing (GCC2002), Sanya, China,
Dec. 26, 2002.

[3] S. Song and K. Hwang, "Fuzzy trust integration for
security enforcement in grid computing," in
International Symposium on Network and Parallel
Computing(NPC2004), submitted March 22, 2004.

[4] C. Lin and V. Varadharajan, "Trust relationship in
mobile agents - a reflexive approach," in Proceedings of
Internation Conference of Agent-Based Technologies
and Systems 2003 (ATS03), Calgary, Canada, August
2003, pp. 81-88.

[5] A. Abdul-Rahman and S. Hailes, "Using
recommendations for managing trust in distributed
systems," In Proceedings of IEEE Malaysia Inter-
national Conference on Communication' 97 (MICC'97),
Kuala Lumpur, Malaysia, 1997.

[6] M. R. Thompson, D. Olson, R. Cowles, S. Mullen, and
M. Helm, "Ca-based trust model for grid authentication

205

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

and identity delegation," in

http://www. sridcp. es. net/Documents/GGF6/TrustModel
-finalpdf . Grid Certificate Policy WG, October 2002.

[7] J. Broch, D. Maltz, D. Johnson, Y. Hu and J. Jetcheva,
"A Performance Comparison of Multi Hop Wireless
AdhocNetwork Routing Protocols", Carnegie Mellon
MONARCH Project, October 1998,

http://www.monarchxsxmnu.edu/ .

[8] C.E.Perkins and E.M.Royer, "Ad-Hoc On Demand
Distance Vector Routing", Proceedings of the IEEE
Workshop on Mobile Computing Systems and
Applications WMCSA), February 1999.

Prof. R.Rameshkumar is pursuing his
PhD at JNT University, Hyderabad
under the guidance of

Prof.Dr.A.Damodaram, Director of
UGC Academic Staff College of JNT
University Hyderabad. He has
obtained his Bachelor Degree in
Computer Science and Engineering
from Mookamibigai College of
Engineering (Bharathidasan

University) and Master Degree in
Computer Science and Engineering from Arulmigu
Kalasalingam College of Engineering(M.K.University).

[9] C.E.Perkins and E.M.Royer, "Ad-Hoc On Demand
Distance Vector Routing", Proceedings of the IEEE
Workshop on Mobile Computing Systems and
Applications (WMCSA), February 1999.

[10] E. Bonabeau, M. Dorigo and G. Theraulaz, Swarm
Intelligence: From Natural to Artificial Systems,
OxfordUniversity Press, 1999.

[11] M. Dorigo and G. DiCaro, "Ant Colony Optimization:
a New Meta-Heuristic", Proc. 1999 Congress on
Evolutionary Computation, July 6-9, 1999, pp. 1470-

1477.

J He joined as Faculty of Computer

^^ Science and Engineering in 1989 at

^B JNTU, Hyderabad. He worked in the

■^^ JNTU in various capacities since

% ^ 1989. Presently he is a professor in

Wf Computer Science and Engineering

^^^^^ Department. In his 19 years of
^^^^1 service Dr. A. Damodaram assumed
i^l^^^l office as Head of the Department,
^^^1 Vice-Principal and presently is the
Director of UGC Academic Staff
College of JNT University Hyderabad. He was board of
studies chairman for JNTU Computer Science and
Engineering Branch (JNTUCEH) for a period of 2 years. He
is a life member in various Professional bodies.
He is a member in various academic councils in various
Universities. He is also a UGC Nominated member in
various expert/advisory committees of Universities in India.
He was a member of NBA (AICTE) sectoral committee and
also a member in various committees in State and Central
Government

206

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

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

Vol 8, No. 3, 2010

A Forager Bees Behaviour Inspired approach to
predict the forthcoming navigation pattern of

onUne users

V.Mohanraj

Assistant Professor/IT

Sona College of Technology

Salem, Tamilnadu, INDIA

b vmohanraj @yahoo . in

J Senthilkumar

Assistant Professor/IT

Sona College of Technology

Salem, Tamilnadu, INDIA

rajkrishcounty@gmail.com

Dr.R.Lakshmipathi

Professor/EEE

St.Peters Engineering College (Deemed University)

Chennai, Tamilnadu, INDIA

drrlakshmipathi@yahoo.com

Y.Suresh

Assistant Professor/IT

Sona College of Technology

Salem, Tamilnadu, INDIA

shuresh_22@yahoo.co.in

Abstract — The World Wide Web is continuously growing and
become de facto place to conduct online business. In the current
internet world, peoples are more attracted towards participating
in the e-commerce sites. The real challenge for the web master of
any such website is to find the users need in advance and provide
the resources pages that keep them interested in browsing their
site. It is easy for any unsatisfied user to reach out the
counterpart site in a single click. Many Web usage mining
methods were adopted to work on web server log and predict the
forthcoming navigation pattern of user. However, the accuracy of
the methods can't satisfy the user especially in huge site.

This paper presents the forager bees behaviour inspired Forager
agent based architecture that uses its collective intelligence for
predicting the forthcoming navigation pattern of user. Our
practical implementation shows that accuracy and coverage
measures are very much improved than existing methods.

Keywords-Web Usage Mining; Web Personalization; Artificial
Bee Colony.

Introduction

In the current internet world, there are many e-commerce
sites compete with each other to attract the user. It becomes
mandatory for web master to predict the future navigation of
user and recommend those to users. This makes the user to
browse the site with lot of satisfaction. In case of unhappiness,
it's easy for online user to switch over to another e-commerce
site that provides the same kind of service.

All the e-commerce sites are focusing on how to provide
the excellent personalized access to users on their sites. The
solution of the problem is web usage mining (WUM). WUM
[8] is part of web mining which deals with the extraction of
knowledge from server log file [4] [5]. Source data mainly
consists of the logs, that are collected when user access web
server and might be represented in standard format. WUM has
become very critical for website management, creating

adaptive website, business and support services,
personalization and network traffic flow analysis.

Typically, the WUM based forthcoming navigation
pattern capturing process can be divided into FrontEnd and
BackEnd with respect to the web server activity [6]. The
activities of BackEnd component is focused on building the
knowledge base by analyzing server log file which records
user web usage data. The activities of the FrontEnd component
are classifying the current user navigations to any one of
cluster formed in BackEnd Phase and infer the useful pattern
to predict the future navigations of user. A particular feature of
our paper is that achieving the accuracy of excellence in
predicting the forthcoming navigation pattern of user using the
collective intelligence of Artificial Bee System [15]. This
system is relatively new member of swarm intelligence. It tries
to model natural behaviour of real honey bee in food foraging.
Honey bee use several mechanisms like waggle dance, round
dance and tremble dance to exchange the information about
location and profitability of food source. This makes them a
good candidate for developing new intelligent search
algorithms. Artificial Bee system has three area of study:
Foraging Behaviour, Marriage Behaviour and Queen Bee
concept. Our paper focuses on the usage of foraging behaviour
of bee in the FrontEnd Phase of Forager architecture to
achieve excellence in capturing the forthcoming browsing
pattern of online users.

The paper is organized as follows. In section II, we review
the related automatic recommendation systems and reported as
literature survey. In section III describes the different phases
of Forager agent based architecture and focuses on the
Greatest Common Subsequence detection algorithm which is
used by foragers and Intuition Deductive inference engine
used by the onlooker in the FrontEnd phase of the architecture.
In section IV, the illustration of Forager Agent system is
explained. In section V, the results of our practical
implementation are reported. Finally, section VI concludes
our work.

207

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

11. Related Works

Recently, a number of studies have been proposed to
capture the forthcoming navigation pattern of web users. We
have conducted investigation on different WUM system [10]
and architecture that can be matched with our proposed
system.

Analog [1] is one of the first web usage mining systems. It
consist of online and offline component. The offline
component builds session clusters by analyzing user
navigation pattern recorded in the log file. In the online
component part, active user session is classified according to
the generated model. The classification allows identifying the
pages matches with the active session and returning the
requested page with list of suggestion. Clustering approach of
system affected by several limitation especially scalability and
accuracy. There is variety of clustering algorithms available
for usage. Each approach could have different type of cluster
[Exclusive (K-Means), Overlapping (Fuzzy C Means) and
Hierarchical]. It's difficult to compare the performance of
algorithm on large dataset like web log. In addition. Clustering
approach used in all recommender system needs to be back up
by excellent classification method. Analog did not have the
proper classification approach over the overlapping cluster.

A Web personalizer system [11] provides dynamic
recommendation, as a list of hyperlinks to users. In the Web
personalizer system, analysis is based on the usage data
combined with structure formed by the hyperlinks of site.
Aggregated usage profile is obtained by applying data mining
technology [i.e clustering, association rule] in pre-processing
phase [12]. In this phase web server logs are converted into
cluster made up of set of pages with the common usage
characteristic. The online phase considers the active user
session in order to find match among user activities and
discover usage profile. Matching entries are then used to
compute a set of recommendations which will be inserted into
the last requested page as a list of hypertext links.
Webpersonalizer is good example two tier architecture for
personalized system. However the accuracy of the
Webpersonalizer is affected by association rule mining [16]
used for discovery of frequent item set in web log data. The
main problem with the association rule mining method is
discovery of contradictory association rules. As a result of
inconsistent rules, predicting accuracy of system is degraded.
Even the non redundant association rule mining algorithm
does not help the system because of the web log data nature
where the number of page hits is high.

Another WUM system called SUGGEST [2] provides
useful information to optimize web server performance and
make easier the web user navigation .SUGGEST adopts a two
level architecture composed by an offline creation of historical
knowledge and online component that understands the users
behaviour .SUGGEST uses the markov model for calculating
the probability of a page the web user visit in future after
visiting pages in the same session. This system uses the high
order markov model [7] to improve the accuracy. However,
the system can't be used for web site made up of large number

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

Vol 8, No. 3, 2010
of pages due to high space complexity. The limitation of
system might be a) the memory required to store web server
pages is quadratic in the number of pages. It is a sever
limitation in the huge web site, b) SUGGEST does not permit
us to manage web site made up of pages dynamically
generated.

Our survey reveals that there is a race for finding
architecture [3] [13] and classification algorithm to improve
the accuracy of capturing forthcoming navigation pattern of
online users. But still the accuracy does not meet the
satisfaction. In our work, we propose advanced Forager agent
based architecture and novel user navigation classification
approach in the architecture for improving accuracy and space
complexity. Our Forager agent based architecture is the
inspiration from artificial bee colony introduced by [14] [15].

1 . Each employed bee determines a food source, which
is also representative of a site, within the neighbourhood of the
food source in its memory and evaluates its profitability.

2. Each employed bee shares its food source
information with onlookers waiting in the hive and then each
onlooker selects a food source site depending on the
information taken from employed bees. Each onlooker
determines a food source within the selected site by herself
and evaluates its profitability using the collective intelligence.

3. Employed bees whose sources have been abandoned
become scout and start to search a new food source randomly
(Fluctuation).

The step (2) of the algorithm is implemented in the
FrontEnd phase of Forager agent architecture. In the FrontEnd
Phase, fleet of forager is originated by onlooker agent on
number of clusters formed in the BackEnd Phase. Each forager
search the web pages in the cluster based on reinforcement
given from onlooker agent. Each forager executes the Greatest
Common Subsequence detection algorithm on its cluster of
web pages and also runs the scoring algorithm. The final score
of each forager is received by onlooker agent. In case of more
than one profitable cluster, it is the onlooker agent that runs
the intuition deductive inference engine to choose the best one
among the alternatives. Finally, recommends the selected one
as forthcoming browsing pattern of the user.

III. Different Components Of Two Tier Forager
Agent Based Architechutre

According to different functionality, our proposed
architecture can be divided into two main phase Back End and
Front End. Both these phases are tightly coupled with each
other and work closely together. The Figure 1 and 2 depicts
the BackEnd and FrontEnd architecture of two tier Intelligent
Forager Agent respectively. In the Back End Phase there are
two main module. Data pre-processing and user navigation
mining. The main modules of the Front End phase are
onlooker agent, forager agent and Intuition deductive
inference engine.

A. BackEnd Phase of Architecture

Two main major modules of the BackEnd Phase are
Data pre-processing and user navigation pattern mining. In

208

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

this phase, we perform Data pre-processing on server log to
capture navigation session and after that we apply algorithm to
mine user navigation pattern. The Detailed module of
BackEnd Phase is shown in Figure 3.

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

Vol 8, No. 3, 2010
about actual page. We employ the content based information to
enrich web log data in the form of maintaining ontology for each web
page in the Array data structure where each index corresponds to
page number of web site. This information will be used by the
Intuition deductive engine of Front End for the choosing the best
classification.

Log Files

Data Pre Processing

User Navigation Mining

Navigation Profile O/P

Figure 1 . BackEnd Phase of Two Tier Forager Agent Architecture

Navigation Profile
Output

Forager Agents

Intuition Deductive
Inference Engine

Captured forthcoming

Browsing pattern of

user

Knowledge
Base

Log Files

Data Pre-processing

Content & structure
retrieving

Session Identification

Data Cleaning

Data collection

ser Navigation Mining

Navigation Pattern Modelling

Clustering

Navigation pattern
Profile

Figure 3. Components of BackEnd

b) User Navigation Mining

After the data pre-processing, we perform a navigation
pattern mining on the identified session. We perform
clustering which aims to group session into clusters based on
their sharable properties. These patterns will be further used to
facilitate the user profiling process of the system. It includes
two main modules:

1) Navigation Pattern Modeling

Figure 2. FrontEnd Phase of Two Tier Forager Agent Architecture

a) Data Pre-processing

The pre-processing of web logs is usually complex and time
demanding. It comprises of four different tasks 1) Data collection: A
flat file was constructed from original weblog file. Each record of the
file consists of time, ip address, name, requested resource (URL) and
HTTP Status code. 2) Data Cleaning: In this step, we perform the
removal of all the data tracked in web log that are useless for mining
purpose such as Navigation session performed by robots and web
spider. 3) Session Identification and reconstruction: it involves i)
Identifying the different users session from usually very poor
information available in log files and ii) Reconstructing the user's
navigation path within the identified session. 4) Content and
Structure Retrieving: Mostly all WUM uses the visited URL as the
poor source of information. They do not convey any information

In this step, web pages accessed are modeled as undirected
Graph G= (V, E). The set Vertex (V) identifies the different
web pages hosted on the web server model. The edge weights
are determined by the following equation

WPij =

_a/_

Max { a , Cj }

(1)

Where Cy is the number of session containing both pages i
and j. Ci and Cj are respectively the number of sessions
containing only pages i or page j. Dividing by the maximum
between single occurrences of two pages has the effect of
reducing the relative importance of links involving index
pages. Such pages are those that generally do not contain

209

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

useful content and are used only a starting point for a
browsing session. The edge weights (WPy) are kept in the
adjacency matrix WA where each entry WAy contains the
value computed according to equation (1). To limit the number
of edge in such graph, elements of WAy whose value is less
than a threshold are known to be less correlated and thus
discarded.

2) Clustering

We apply a graph partitioning algorithm to find groups of
strongly correlated pages by partitioning the graph according
to its connected components. Clusters are formed by starting
from a Vertex a DFS on the graph induced by WM is applied
to search for the connected component reachable from this
vertex. Once the component has been found, the algorithm
checks if there are any nodes not considered in the visit. If it
so, it means that a previously connected has been split and
therefore, it needs to be identified. To do this, DFS is again
performed by starting from one of the nodes not visited. In the
worst case, when the entire URL in the same cluster, the cost
of the algorithm will be linear in the number of edges of the
complete graph G. Before the clusters are put into navigational
pattern profile, the clusters are ranked based on values store in
the WM matrix. It will be used for classification performed by
foragers based on the Greatest Common Subsequence
Detection algorithm and also used for knowledge eliciting.

B. FrontEnd Phase of Architecture

In the FrontEnd Phase of our system, URL request of
the user is processed by the Onlooker Agent (OA) and Forager
Agents (FA). In the case of multiple profitable outputs, the
best option is choosen by the Intuition Deductive Inference
Engine (IDIE) which is run by Onlooker Agent. Finally,
captured imminent browsing pattern is suggested to the user
who initiated the URL request to web server.

a) Work of Onlooker and Forager agent in FrontEnd

Phase
The critical component of our system is Onlooker agent
and Forager agent. Main inputs for these agents are

1. Navigation pattern profile: It consists of clusters
formed in the BackEnd Phase of our system.

2. Live session window: A Sequence
LSW={lwpi,lwp2,....,lwpni} is the current size of live session
window where m is the size of the current active session
window.

On receiving URL's in the form of Live Session Window,
Onlooker initiates the Foragers that correspond to number of
clusters in the navigation profile. Each Forager agent (FA)
acts on the Navigation profiles by executing the novel Greatest
Common Subsequence Detection to discover the subsequence
which may be considered as imminent browsing pattern of
web user. Each FA submits the profitable score along the
discovered subsequence to the onlooker.

When the Onlooker Agent (OA) receives the profitable
score along the sub sequences, it starts to decide the best
profitable source of navigation profile. In the case of close

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

Vol 8, No. 3, 2010
race between the sub sequences. Intuition deductive
inference engine (IDIE) plays a crucial role in selecting the
right cluster. The main objective of IDIE is to test the each
cluster of discovered navigation pattern against already built
knowledge base and choose the cluster which attains the
maximum number of matches with the knowledge base.
Finally, Onlooker suggests the output of IDIE as the imminent
browsing pattern of web user. The cooperation between
Onlooker Agent (OA) and Forager Agents (FA) is depicted in
the Figure 4. The OA listens to FA. It is just similar to honey
bee dancing area where Onlooker listens to dances of different
foragers about the profitable food source.

Live
Session
Window

Intuition

Deductive

Inference Engine

(IDIE)

Cluster 1

Cluster 2

Cluster n

User

Captured

forthcoming

Navigation pattern

Figure 4. Onlooker Agent and Forager Agents

b) A Igorithm for predicting for the om ing navigation
pattern of online users

The following Algorithm 1 depicts the working behaviour
of onlooker and forager agents to capture the forthcoming
browsing pattern of user in the FrontEnd phase of the
Architecture.

Algorithm 1

1 . Live Session Window (LSW) is given as input to the

Onlooker Agent.LSW is the set of web page visited by
user in the live session. LSW is represented as {LWPi,
LWP2, ., LWPn) where 'n' is the size of session
window.

2. Onlooker Agent (OA) initiates the number of Forager
Agent (FAi) that corresponds to each cluster in
Navigation Profile (NPn) where'n' is number of cluster.

210

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

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

•th

FAi - Forager Agent works on i cluster in NP
and i G n.

3. Initially onlooker agent assigns the arbitrary profitable
score of value 100 to each Forager Agent (FAj). The
score is denoted as score(FAi).This score is updated by
the respective Forager Agent on the discovery of
subsequence in the step 4.

4. For each Forager Agent on its assigned cluster of
Navigation Profile does the following

i. Each FAi executes the Greatest Common subsequence
Detection (described in c) on its cluster in respect to
live session window (LSW) which produces the
highest degree of GCD as the discovered subsequence
which could be the candidate of online user's imminent
browsing pattern,
ii. Discovered sequence is denoted as IBP = {IBPi, IBP2,.
., IBPnj.Each Forager Agent updates its initial score
using Equation (2) and with the help of adjacency
matrix WA built in the BackEnd Phase where each
entry WAy contains the value computed according to
equation (1).

IF IPS,

Vol 8, No. 3, 2010

■PSJ or IPS1-PS3I or iPSi-PSj

< P Where P is Uncertain Profitable
Threshold value then

There is race between discovered subsequence of PSi or
PS2 or PS3 to become an imminent browsing pattern of web
user. Onlooker Agent (OA) sends the cluster of competing sub
sequences to the Intuition Deductive Inference Engine
(IDIE).The main objective of IDIE (described later) is to test
the each clusters navigation pattern against already built
knowledge base and choose the cluster which attains the
maximum number of matches with the knowledge base.

ELSE

Onlooker chooses the PSl's sequence as best
discovered subsequence.

6. Finally, Onlooker Agent reports the predicted
forthcoming browsing pattern as PS 1 to user Or best
Sequence selected by the IDIE in the case of
competition.

7. Suppose, if the next user activity in live session window
different from the suggested captured list then the
system has to restart once again to classify the new user
activities.

c) Greatest Common Subsequence Detection

^score{V Ai) = score(FAi) + J Z WAj^p^^wpj
Where WA

(2)

1=1 j=i
^iBPiLWPj = ^^1^^ i^ Adjacency
matrix between the each page in Imminent Browsing
Pattern (IBP) discovered by Forager and pages in the
Live Session Window (LSW).
iii. After each FAi executed the steps i and ii. Forager
Agent sends its updated score and discovered
Subsequence to the Onlooker Agent (OA).
5. After receiving the profitable scores from each Forager
agents (FA^) It selects the first 3 High scored Forager
Agent's output. The scores are denoted as PSi, PS2
and PS3.

i. Onlooker Agent computes the absolute difference
between the PSi, PS2 and PS3 to find the
closeness.

Every Forager Agent initiated by Onlooker Agent should
perform the similarity comparison between set of pages in
Live Session Window and web pages in the cluster to discover
the subsequence that could be the forthcoming browsing
pattern of online user. It's clear that every forager agent has to
perform some kind of pattern matching.

In the pattern matching [9], comparing the similarity between

the two sequences a and b are fundamental problem. One
of the fundamental problem is to determine^ the Greatest

Common Subsequence (GCS) between a and b . The GCS is
a String comparison metric that measures the subsequence of
maximum length common to both the sequences. Main
objective of Forager Agent (FA) is to find the Greatest
Common Subsequence among the sequence of paths in the

form of page visits A= {al,a2,a3,a4, an) ,

B={bl,b2,b3,b4,...bn).

Theorem 1 Let a= {al,a2,a3,a4, an) and

b={bl,b2,b3,b4,...bm) be the sequences and Let
d={dl,d2,d3,...dn} be any GCD of a and b.

1. If an=bni then di=an=bni and dn-i is a GCD of sl^.i and

bn-l-

2. If anT^bm then di^^an implies d is a GCD of an-i and b.

3. If anT^bm then di^^bm implies dn is a GCD of a and bm-i.

211

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

We have implemented the GCD with added module that
outputs the subsequence of indices of the two sequences that
match in getting the Greatest Common subsequence. For
example, If A={wpl,wp2,wp3,wp2,wp4,wp2,wpl} and
B={wp2,wp4,wp3,wpl,wp2,wpl}. Their GCD is GCD=
{ wp2,wp4,wp2,wp 1 ).

d) Combined Effort of Greatest Common subsequence
Detection and Intuition Deductive Inference Engine (IDIE)

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

Vol 8, No. 3, 2010
against the knowledge base to find the semantic valid
matching count. Intuition Deductive Inference Engine is
designed to infer the facts by following Top Down inference
mechanism. The excerpts of our IDIE knowledge base are
shown in Table I.The fact Next (wp7(cse
research),wp6(research)) says that wpv is the semantically the
next page of wp6 and values in the bracket are ontological
terms gathered from OM array. The rule (1) can directly apply
on the facts where as rule (2) is recursive nature.

After performing the clustering algorithm discussed in
A.b.2 of BackEnd Phase, We have a set of cluster wnp =
{wnpi,wnp2,wnp3,...,wnpn} where wnpi={wpi,wp2,...,wpk} is a
set of K pages as a users navigational pattern for each 1< i < n.
Here, wnp is set of web navigational pattern in the cluster.

We used the navigation profile which has web
navigational pattern of different cluster as facts for building
the knowledge base of Intuition deductive Inference Engine.
In addition, OA an ontology array data structure built in the
Content retrieving step of BackEnd phase is used to form the
meaningful facts. We used the unique ontological term for
each web page in the site. We get the ontology term of each
page from Meta tag of each page which conveys the actual
content of the page. Also, Structure of Website is used to input
the additional facts for knowledge base. In our WUM system.
Knowledge base building is considered as critical part.

A Sequence LSW={wpi,wp2,....,wpni} is the current size of
live session window where m is the size of the current active
session window. Before Onlooker Agent initiates the Forager
Agents to execute GCD on its cluster, we need to order the Isw
sequence based on their adjacency weight matrix (WM)
constructed in the navigation pattern modeling. Also, we rank
all the clusters based on their weight values. Each cluster
weight is computed as sum of all its edges weight. After this
step, each Forager Agent initiated by OA with the arbitrary
profitable score applies the Greatest Common Subsequence
Detection on the assigned cluster in respect to live session
window (LSW) which produces the highest degree of GCD.
Each Forager Agent sends its updated score and discovered
subsequence to the Onlooker Agent (OA).

After receiving the profitable scores from each Forager
agents (FA^) OA selects the first 3 High scored Forager
Agent's output and finds whether absolute difference between
them lesser then P (Uncertain Profitable Threshold value).

In the case of competition between discovered sequences by
forager agents. Intuition deductive inference engine (IDIE)

plays a crucial role in selecting the right cluster among various
options.

The main objective of IDIE is to test the each clusters
navigation pattern against already built knowledge base and
choose the cluster which attains the maximum number of
matches with the knowledge base. The rules of knowledge
base are written in InterProlog notation. Our inference engine
runs on the facts to check how many of web page navigational
sequence in each cluster are semantically correct. The rule
base is written in such a way that it checks out each cluster

As we discussed earlier, clusters of competing discovered
sequences selected by Onlooker Agent (OA) is given as input
to IDIE. The IDIE infers the count for each cluster that shows
how many sub sequences in each cluster are semantically
matching with the knowledge base. Finally, IDIE reports a
cluster of maximum valid match count. This is used for
preparing the intuition captured list and provided to the user as
a recommendation. Suppose, if the next user activity in live
session window different from the suggested captured list then
onlooker agent has to restart the algorithm once again to
identify the forthcoming user activities on their site.

TABLE 1.

Excerpts of IDIE Knowledge Base

Facts

Next (wpvCcse research),wp6(research))
Next (wpio(it research),wp6( research))
Next (wpi7(cse),wp24(course))
Next (wpvCcse research),wpi7(cse))
Next (wp27(cse staff details),wpi7(cse))

Rules:

Subsequence(x,y):- Next(x,y) (I)

SuperSubsequence(x,y):- Next(x,z) , Next(z,y) 2)..

IV. Illustration of Forager Agent Based System

For our illustration, consider the navigation profile of
BackEnd Phase as shown in Table II. Assume the Live
Session Window size as 3 and set of web pages visited by user
in the live session as LSW= {wp37,wp27,wpi8}. As stated in
our algorithm 1, LSW is given as input to the Onlooker Agent
(OA). OA initialize FA, (FAi, FA2, FA3, FA4 and FA5) that
works respectively on Navigation Profile (NPi, NP2, NP3,
NP4 and NP5) with the initial arbitrary profitable score of
100. Each Forager Agent (FAO executes the Greatest Common
subsequence (GCD) on its own controlled cluster in respect to
Live Session Window (LSW).

While Forager Agent discovers the subsequence that could
become Imminent Browsing Pattern (IBP), update its score
according to equation (2) i.e. Sum of adjacency matrix values
for pages between LSW and IBP is added to initial profitable
score. Consider the Table III that depicts the score of each
Forager Agent (Score (FA,)) and its discovered sequence
(IBPi).Each Forager Agents (FA,) sends its Final Profitable
Score (FAi) and Discovered Sequence (IBP,) to Onlooker
Agent (OA). After receiving the profitable score from all
initialized FA,, OA selects the first three high scored FA,

212

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

In this example, only FA3 and FA5 produced the updated
profitable score. The remaining Forager Agents were not
updated their initial score.

TABLE 11.

Navigation Profile of BackEnd Phase

Navigation Profile

Clustered Navigation Pattern

NPi

{WP2,WP10,WP15,WP20,WP8}

NP2

{wp3,Wp27,Wp54,Wpioo,Wpi2l}

NP3

{WP2,WP19,WP37,WP27,WP30,WP18,WP60}

NP4

{WP5,WP15,WP23}

NP5

{WP7,WP37,WP31,WP27,WP29,WP26,WP18}

TABLE m. SCORE(FAi) AND ITS DISCOVERED SEQUENCE (IBPi) IN

RESPECT TO Live Session Window

Forager
Agents

(FAi)

Initial
Score

Navigation
Profile
(NPi)

Discovered
Sequence (IBPi)

Final

Profitable

Score

(FAi)

FAi

100

NPi

100

FA2

100

NP2

100

FA3

100

NP3

{WP19,WP30,WP60}

172

FA4

100

NP4

100

FA5

100

NP5

{WP7,WP31,WP2,
WP26}

213

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

Vol 8, No. 3, 2010
cluster which attains maximum valid match count in respect to
LSW. In this example, FA5 is reported by the IDIE for
attaining the more number of valid matches with knowledge
base than FA3. From the FA5, OA suggest the following list as
perfect Imminent Browsing Pattern of user

IBP = {WP7,WP31,WP29,WP26}.

V. DATA Analysis Report

Our proposed Forager Agent based system was tested on
the web log dataset of Sona College. This site was deployed in
the IBM's Web sphere Application Server. All the algorithms
of BackEnd and FrontEnd phase were implemented in the
JAVA. The knowledge base used in the IDIE of FrontEnd
phase was implemented using the InterProlog. InterProlog
provides us the ability to call prolog goal through a prolog
object and for prolog to invoke any JAVA method through a
Java Message Predicate. The BackEnd phase of our system
was tested on the weblog entries of 120 students over a period
of 8 weeks. There were approximately 52,745 entries in the
log file. For our input dataset, BackEnd phase had produced
the navigation profile output which consists of 15 clusters. The
performance of the proposed system was analyzed based on
the two metrics. They are namely Accuracy and Coverage.

a) Accuracy based analysis of Forager Agent based

System
Accuracy measure is defined as a degree to which captured
imminent browsing pattern as suggested by the system
matches with the actual browsing pattern of user. It is given by

\p{IBP^^,LSW)[^Ormnal^

NO- No Output.

In the next step. Onlooker Agent (OA) finds the absolute

difference between \Score{FA^ ) - Score(FA^ )| and

compares the value with the /3 (Uncertain Profitable

Threshold value). Here the value of /^ is assigned to be 50. In

this example, the value of \Score(FA^) - Score(FA^)\ is

41 which lesser thany^. This situation is the typical case of
competition of who to become the Imminent Browsing pattern
of user between IBP3 and IBP5 which are discovered sequence
of FA3 and FA5 Now, Onlooker Agent uses the Intuition
Deductive Inference Engine (IDIE) to choose the best one
among the alternatives IBP3 and IBP5

Onlooker Agent feds the clusters of competing discovered
sequence NP3 and NP5 to the IDIE. The main function of IDIE
is to check each page of clusters with knowledge base and find
whether they are semantically valid next page in respect to
pages in Live Session Window. Onlooker Agent chooses the

Accuracy=

\P(IBP^^,LSW)

(3)

LSW - Live Session Window P(IBP^^,LSW) -

Navigation Pattern in predicted imminent browsing pattern of

usQY. Original - Original Navigational pattern of user

The Figure 5 depicts the accuracy of our Forager Agent
based system as Live Session Window (LSW) size is
increased. Our results show that increase of LSW size gives
more wisdom to system that improves the accuracy.

The Figure 6 depicts the comparison of our Forager Agent
based system with the other two recommendation system
namely Web personalizer and SUGGEST. Our results show
that Our Forager Agent based system outperforms the
recommendation system with the excellent behaviour of
Forager Agents which uses Greatest Common Subsequence
Detection to predict the forthcoming browsing pattern of user.
In the case of Competition between alternative. Onlooker
Agent uses the Intuition Deductive Inference Engine to choose
the best one and thereby completely avoids the
misclassification in finding the forthcoming browsing pattern
compared with other system. Our Forager Agent based system
outperforms the other recommendation system and achieves
the accuracy of 92% compare to the Web personalizer (81%)
and SUGGEST (83%) when Live size window is 10.

213

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

b) Coverage based analysis of Forager Agent based

System
Coverage measure is defined as the ability of two tier
Forager agent based system to produce all page views that are
most likely visited by the user. It is given by

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

Vol 8, No. 3, 2010
inference engine to choose the best one among the
alternatives. Finally, recommends the selected one as
forthcoming navigation pattern of the user. The practical
implementation shows that our approach really improves the
accuracy of predicting the forthcoming browsing pattern of
user and satisfies the users compared to other system.

Coverage =

\p{IBP^^,LSW ) [\ Original ^
\Original „^ |

(4)

The Figure 7 depicts the coverage of our Forager Agent
based system as Live Session Window (LSW) size is
increased. The practical implementation of our proposed two
tier Forager agent based system on sona college dataset prove
that there is increase in the accuracy of predicting the
navigation pattern of online user and also Coverage measure is
excellent.

Accuracy of Forager Agent based System

100

_ 80

& 70

2 60

I 50

-92

-1 1 1 1 1 1 1 1

3 4 5 6 7
LSW Size

9 10

Figure 5. Accuracy of Forager Agent Systems

VI. Conclusions

Our proposed two tier Forager agent based system
presented in this paper was inspired from onlooker bee making
a decision of profitable food source using a Collective
intelligence of Foraging Behaviour in Bee's Hive. In our
system, fleet of forager is originated by onlooker agent on
number of clusters formed in the BackEnd Phase. Each forager
executes the Greatest Common Subsequence detection on its
cluster of web pages and also runs the scoring algorithm to
discover the subsequence that could be the imminent browsing
pattern of user. Each forager agent sends the final score to
onlooker agent. In case of more than one profitable cluster, it
is the onlooker agent that runs the intuition deductive

100 ■

90 "

80 "

70 ■

? 60 ■

X 50 "

1 40

1 30 "

< 20 "

10 "

Forager Agent based system Vs others

Forager Agent

Web

personalizer

SUGGEST

^^^^w y^

..--r^f:4f ?J M

-^?if7^ ^i ^^ ''

^59 ^^

3456789 10
LSW Size

Figure 6. Accuracy of Forager Agent based System Vs Other Systems

90 "

80 "

70 ■

_ 60 ■

^ 50 "

OX)

i 40 "

" 30-
20 ■

10 ■

Coverage of Forager Agent based System

-^.

~^*'~'^*-^

^^^56

1 1 1 1 1 1 1 1

3456789 10
LSW Size

Figure 7. Coverage of Forager Agent based System

References

R. Agrawal and R. Srikant, "Mining sequential patterns", International
Conference on Data Engineering (IDCE), 1996, Taiwan, pp. 3-11.

R. Baraglia and Palmerini, "SUGGEST: A Web usage mining system".
Proceeding of International Conference on Information Technology:
Coding and Computing, 2002, pp.282-287.

R. Baraglia and F. Silversti,"Dynamic Personalization of Web Sites
without User Intervention", Communication of the ACM, 2007, pp.63-
67.

[4] R. Baragila and F. Silvertri,"An online recommendation system for large
web sites", Web Intelligence,IEEE/WIC/ACM,2004,pp.20-24.

[1]

[2]

[3]

214

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

[5] F.H. Chanchary, I. Haque and Md.Khalid,"Web Usage Mining to
Evaluate the Transfer of Learning in a Web based Learning
Environment", Knowledge Discovery and Data Mining of IEEE, 2008,
pp.249-253.

[6] R. Cooley, J. Srivastav and B.Mobasher, "Automatic Personalization
based on Web Usage Mining", Communication of the ACM,2000,
Volume 43,issue 8,pp. 142-151.

[7] Deshpande and G. Karypis,"Selective Markov models for predicting
web page access". Transactions on Internet Technology, 2004, Vol.4,
No.2,pp.l63-184.

[8] E. Frias-Martinez, V. Karamcheti,"Reduction of user perceived latency
for a dynamic and personalized site using web mining technique",
WebKDD,2004,pp. 12-22.

[9] D.S. Hirschbergand and J.D. Aho.Ullman,"Bounds on the Complexity of
the longest Common subsequence problem", J.Assoc. Comput. Mach.
ACM,1976,pp.l-12.

[10] R. Liu and V. Keselij, "Combined mining of Web Server logs and Web
contents for Classifying user navigation pattern and predicting users
future requests". Data & Knowledge Engineering,Elsevier,2008,pp.304-
330.

[11] B. Mobasher,"Web personalizer: A Server Side Recommender System
Based on Web Usage Mining", 1991, In. Technical Report TR-01-004.

[12] Nakagawa and B. Mobasher,"A hybrid web personalization Model based
on site connectivity", WebKDD,2003, pp. 5 9-70.

[13] Jalali, N. Mustapha, A. Mamat and N. Sulaiman.Md "OPWUMP -An
Architecture for online Predicting in WUM-based Personalization
system". In 13* International CSI Computer Science, 2008 Springer
Verlag.

[14] D. Karaboga and B. Basturk ," On the Performance of Artificial Bee
Colony (ABC) Algorithm, Applied Soft Computing,2008, Volume 8,
Issue 1, Pages 687-697.

[15] V. Tereshko, A. Loengarov,"Collective decision-making in honey Bee
foraging dynamics, Comput. Inf Syst. J., 2005, pp. 1352-1372.

[16] M. Yan, H. Jacobsen, Garcia-Molina, and U. Dayal, "From User Access
Patterns to Dynamic Hypertext Linking," Comp. Networks and
ISDN Sys, 1996, vol. 28, pp. 1007-14.

AUTHORS PROFILE

V.Mohanraj received his ME Computer science and Engineering
from Anna University, Chennai in 2004. He is currently working as
Assistant professor in IT department of Sona College of Technology.
He is pursuing PhD degree in Anna University, Chennai. His research
area includes web mining, database and intelligent system.

Dr.R Lakshmipathi received his BE from College of
Engineering, Anna University, India, in 1971, the ME and PhD in
Electrical Engineering from College of Engineering, Anna
University, India in 1973 and Indian Institute of Technology (IIT),
Chennai in 1979, respectively. He has 36 years of teaching
experience at UG degree level out of which 10 years in PG degree
level. He worked as principal in Govt. College of Engineering and
held the various prestigious posts like Dean, Regional Research
Director , Chairman for board of BE exams. Member of Academic
auditing committee, AICTE, University and State Govt, expert
committee member. He is currently a professor of Electrical
Engineering, St. Peters Engineering College (Deemed University),
Tamilnadu. His research interest includes Electrical power semi
conductor drives. Signal processing and Web Mining.

J.Senthilkumar received his ME Applied Electronics from Anna
University, Chennai in 2004. He is working as assistant professor in
IT Department of Sona College of Technology and pursuing PhD
degree in Anna University, Chennai. He is active in the research area
of data mining and Mobile computing.

Y.Suresh received his ME Applied Electronics from Anna
University, Chennai in 2004. He is working as assistant professor in

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

Vol 8, No. 3, 2010
IT Department of Sona College of Technology and pursuing PhD
degree in Anna University, Chennai. He is active in the research area
of data mining, control system and Mobile computing.

215

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

Quality of Service Issues in Wireless
Ad Hoc Network (IEEE 802.11B)

Mohammed Ali Hussain\ Mohammed Mastan^ Syed Umar^

^Research Scholar, Dept.of CSE, Acharya Nagarjuna University, Guntur, A.P., India.

hussain_ma2k@ yahoo.co.in

^Research Scholar, Dept.of CSE, JNT University, Kakinada, A.P., India.

mastanmohd @ gmail.com

^Research Scholar, Dept.of CSE, Dravidian University, Kuppam, A.P., India.

umar3 32 @ gmail.com

Abstract — A wireless Ad-hoc network consists
of wireless nodes communicating without the
need for a centralized administration, in which
all nodes potentially contribute to the routing
process. In this paper, we report Fluctuations in
channel quality effect the QoS metrics on each
link and the whole end-to-end route. The
interference from non-neighboring nodes affects
the link quality. QoS is an essential component
of ad-hoc networks. The most commonly studied
QoS metrics are throughput, bandwidth, delay
and jitter. Bandwidth is the QoS metric that has
received the most attention in the QoS literature.
The QoS requirements are typically met by soft
assurances rather than hard guarantees from the
network. Most mechanisms are designed for
providing relative assurances rather than
absolute assurances.

Keywords: QoS, Ad-hoc, Throughput,
Bandwidth, Delay, Jitter, 802.11.

I. INTRODUCTION

Wireless Ad-hoc network consists of
wireless nodes communicating without
the need for a centralized
administration. The idea of such
networking is to support robust and
efficient operation ad-hoc wireless
networks in which all nodes potentially
contribute to the routing process, the
fluctuations in channel quality effect
the QoS metrics on each link and the
whole end-to-end route. In ad-hoc

networks, Quality of Service support is
becoming an inherent necessity rather
than an "additional feature" of the
network. Wireless channel fluctuates
rapidly and the fluctuations severely
effect multi-hop flows. As opposed to
the wired network, the capacity of the
wireless channel fluctuates rapidly due
to various physical layer phenomena
including fading and multi-path
interference. In addition, background
noise and interference from nearby
nodes further effect the channel
quality. In ad-hoc networks, the end-
to-end quality of a connection may
vary rapidly as change in channel
quality on any link may effect the end-
to-end QoS metrics of multi-hop paths.
The Packets contend for the shared
media of the same stream at different
nodes impacts the QoS metrics of a
connection. Such contention arises as
the wireless channel is shared by nodes
in the vicinity. Interference effects are

216

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

pronounced in ad-hoc networks where
typically a single frequency is used for
communication in the shared channel.
In Single hop infrastructured wireless
networks frequency planning is mostly
used where nearby base stations can be
configured to function at different
frequencies for reducing interference.
Transmissions in the wireless media
are not received correctly beyond the
transmission range. But even beyond
the transmission range, the remaining
power may be enough to interfere with
other transmission. So, interference
from nonneighboring nodes may result
in packet drops. In order to support
QoS on multi-hop paths, QoS must be
designed for the end-to-end path as
well as for each hop. The physical and
MAC layers are responsible for QoS
properties on a single-hop. The routing
layer is responsible for QoS metrics on
an end-to-end route.

II. OVERVIEW OF IEEE 802. 11
PHYSICAL LAYER

One of the fundamental challenges in
wireless networks is the continuously
changing physical layer properties of
the channel. The physical layer of
802.11b can support multiple data
rates. Depending on the channel
quality the data rate can be altered to
keep the bit error rate acceptable, as

high data rates are also prone to high
bit error rates.

The 802.11b standard operates in the
2.4 GHz band and supports 1, 2, 5.5
and 11 Mbps. For efficient use of a
multi-rate physical layer, there have
been several algorithms proposed at
the physical layer. One of the
algorithm which is closely tied to the
MAC layer is Opportunistic Auto Rate
(OAR) for improving throughput in the
presence of multi-rate links in ad-hoc
networks. The key idea is to send
multiple packets when the channel rate
is higher.

III. IMPORTANCE OF MEDIUM
ACCESS LAYER

The original IEEE 802.11 [1] standard
specifies the physical layer and the
medium access layer mechanisms and
provides a data rate up to 2 Mbps.
Further the standards IEEE 802.11b
modifies the physical layer part of the
standard and increases the maximum
data rates to 11 Mbps and 54 Mbps
respectively. In this paper we discuss
the basic 802.11 MAC layer
functionality called Distributed
Coordination Function (DCF) for
distributed access to the shared
medium. DCF is a natural choice for
ad-hoc networks, as there is no
centralized controller such as an

217

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

access-point. However, PCF can
support QoS metrics in single-hop
wireless networks due to its centralized
design. Both DCF and PCF are
enhanced in the upcoming standard
802. lie [2] that are designed for
supporting QoS in WLANs.

IV. 802.11 DISTRIBUTED
COORDINATION
FUNCTION (DCF)

The DCF protocol attempts to provide
equal access (in terms of number of
packets) to all backlogged nodes that
share a channel. In an ad-hoc network
the throughput that a node obtains
using DCF is a function of the number
of neighbors that it has and the state of
their queues (backlogged or not).

SIF

Sourc

DATA

NAV(C

window used for backoffs. Initially cw
is set to cwmin . In the chosen slot the
node sends a MAC layer control
packet called RTS (request-to-send), to
the receiver. If the receiver correctly
receives the RTS and is not deferring
transmission, it responds with CTS
(clear-to-send). This is followed by
transmission of the data packet by the
sender, and a subsequent

acknowledgment from the receiver.
The transmissions of these four packets
are separated by short durations called
SIFS (Short Inter-Frame Space). The
SIFS allows time for switching the
transceiver between sending and
receiving modes. The sequence of
transmission of these four packets. The
MAC header of all these packets
contains a "duration" field indicating
the remaining time till the end of the
reception of the ACK packet. Based on
this advertisement, the neighboring
nodes update a data structure called
NAV (Network Allocation Vector).
This structure maintains the remaining
time for which the node has to defer all
transmissions.

Figure 1: IEEE 802.11 DCF

Each node that has a packet to send
picks a random slot for transmission in
[0, cw], where cw is the contention

If the packet transmission fails, the
sender doubles its contention window
(cw ^[2*cw-l]) and backs off before
attempting a retransmission. The
number of retransmissions is limited to

218

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

4 for small packets (including RTS
packets) and 7 for larger (typically
DATA) packets. If these counts are
exceeded, the data packet is dropped
and cw is reset to cwmm if the data
packet is successfully delivered, both
the sender and the receiver reset cw to

CWmin.

V. PROPOSED QOS SUPPORT USING DCF
BASED SERVICE DIFFERENTIATION

As it is difficult to provide absolute
QoS guarantees, relative QoS
assurance can be provided by service
differentiation. However, to provide
differentiated services, the 802.11
protocol needs to be modified. [3]
proposes three ways to modify the
DCF functionality of 802. 1 1 to support
service differentiation. The parameters
that need to be modified to achieve
service differentiation are.
1. Backoff increase function'. Upon an
unsuccessful attempt to send an RTS
or a data packet, the maximum backoff
time is doubled. More specifically the
backoff time is calculated as follows:

(Baciofftime = [2 (^^')xrand:()] XSCotume

Where i is the number of consecutive
backoffs experienced for the packet to
transmitted. To support different

priorities, the backoff computation can
be changed as follows:

(Baciofftime = [ Pj (^^'^XrancfQ ] XSCotume

where pj is the priority of node j
2. DIFS: As shown in Fig.l, this is the
minimum interval of time required
before initiating a new packet
transmission after the channel has been
busy. To lower the priority of a flow
we can increase the DIFS (Distributed
Coordination Function Inter Frame
Spacing) period for packets of that
flow. However, it is difficult to find an
exact relation between the DIFS period
for a flow and its throughput. Fig. 2
shows the different DIFS values and
the corresponding relative priorities.

Priority j+l

(high)

DIFS j+l

Priority j

(intermediate)

DIFSj

defer

Priority j-1 (low)

DIFS j 1

defer

Figure 2: Service Differentiation using
different DIFS values

3. Maximum Frame Length: Channel
contention using the DCF functionality
is typically used to send a single frame.
By using longer frames, higher
throughput can be provided to high-
priority flows.

219

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

VL CONCLUSION

REFERENCES

In this paper, the QoS issues discussed
at various networking layers for ad-hoc
networks. The physical layer and the
MAC layers are primarily responsible
for QoS metrics on each link and the
whole end-to-end route. The DCF
functionality of 802.11 is being
extended and specifically designed for
QoS support in multi-hop networks.
The algorithm which is needed to be
adapted for use in multi-hop ad-hoc
networks is Opportunistic Auto Rate
(OAR) for improving throughput in the
presence of multi-rate links in ad-hoc
networks.

QoS is currently an active research
area in ad-hoc networks. However,
there are several avenues that require
further exploration for designing a QoS
enabled ad-hoc network. For packets
that traverse multiple hops, the end-to-
end QoS is a function of the QoS
metrics at each intermediate link. End-
to-end QoS properties can be improved
by designing a MAC layer that
coordinates with other intermediate
nodes on a multi-hop path. We find
that QoS is an inherent component of
ad-hoc networking and that there are
several unsolved challenges that need
to be addressed to design QoS enabled
ad-hoc networks in future.

[1] IEEE Std. 802.11. Wireless LAN Medium
Access Control (MAC) and Physical Layer
(PHY) Specifications, 1999.

[2] S. Mangold, S. Choi, G. R. Hiertz, O.
Klein and B. Walke. Analysis of IEEE
802. lie for QoS Support in Wireless
LANs .IEEE Wireless Communications
Magazine, Special Issue on Evolution of
Wireless LANs and PANs, Jul. 2003.

[3] I. Aad and C. Caselluccia. Differentiation
mechanisms for IEEE 802. 1 1 . In Proc.
IEEE Infocom, volume 2, pages 594-602,
1996.

[4] A.Veres, A. T. Campbell, M.Barry, and L.
H. Sun. Supporting service differentiation
in wireless packet networks using
distributed control. IEEE Journal on
Selected Areas in Communications,
October 2001.

[5] B.Sadeghi,V.Kanodia,A.Sabharwal, and E.
Knightly. Opportunistic Media Access for
Multirate Ad- hoc Networks. In Proc.
ACMMOBICOM, 2002.

AUTHORS PROFILE

Mohammed Ali Hussain

received the Master's
degree M.Sc Computer
Science from Alagappa
University in 2003. He
received Master's degree
M.Tech in Information
Technology from Allahabad Deemed
University in 2005. He received Ph.D. degree
In Computer Science from Magadh University,
Bihar, India in 2008. He is doing Post Doctoral
degree in Computer Science & Engineering
from Acharya Nagarjuna University, Guntur,
Andhra Pradesh, India. He is currently an
Associate Professor in the Department of
Computer Science in Nimra College of
Engineering & Technology, Vijayawada,
Andhra Pradesh, India. He had published
several papers in National and International
Conferences & International Journals. His
research interests are Wireless Networks with
specialization in Quality of Service (QoS) in
IEEE 802.11 Wireless LANs. & Ad-Hoc
Networks. He is a member of lACSIT and
ISTE.

220

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

.^■M'

Mohammed Mas tan

received the Master's
degree in Computer
Applications from

Kakatiya University,

Warangal in 2006. He
received Master' s degree
in M.Tech Computer Science & Engineering
from JNT University, Hyderabad in 2008. He
is pursuing Ph.D. in Computer Science &
Engineering from JNT University, Kakinada,
Andhra Pradesh, India. He is currently as
Asst.Professor in Department of Computer
Science & Engineering in Nimra College of
Engineering & Technology, Vijayawada,
Andhra Pradesh, India. He has published
several papers in National and International
Conferences. His research interests are
Computer Networks & Wireless Networks.

'^^L

Syed Umar received the
B.Tech degree Electronics
and Communication

Engineering from JNT
University, Hyderabad in
2003. He received
Master's degree M.Tech
in Computer Science &
Engineering from JNT University, Hyderabad
in 2008. He is pursuing Ph.D. in Computer
Science from Dra vidian University, Kuppam,
Chittoor Dist, Andhra Pradesh, India. He is
currently an Asst.Professor in Department of
Computer Science & Engineering in Nimra
College of Engineering & Technology,
Vijayawada, Andhra Pradesh, India. He had
published several papers in National and
International Conferences. His research
interests are Computer Networks & Wireless
Networks.

221

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

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

Vol 8, No. 3, 2010

A New Collaborative Web Recommendation Systems
based on Association Rule Mining

A. Kumar

Research Scholar, Department of Computer Science &

Engineering , Sathyabama University, Chennai, India

email:akr2020av@ yahoo. in

Dr. P. Thambidurai

Principal, Perunthalaivar Kamarajar Institute of

Engineering & Technology, Karaikal, India.

email: ptdurai58@yahoo.com

Abstract — Massive development of internet in recent years
necessitate the development of recommender systems that turn
out to be user friendly in web applications. Recommender
systems make an effort to outline user preferences over items,
and model the relation between users and items. There are two
elemental approaches that can be applied when generating
recommendations systems. They are content based web
recommender system and the other is collaborative web
recommender system. This proposed paper presents a method of
developing a collaborative web recommendation systems using
association rule mining. The association rules were applied to
personalization based on web usage data. The method utilize
apriori algorithm to generate association rules. In general
association rule mining is a technique common in data mining
that attempts to discover patterns of products that are
purchased together. The greater part of web page recommender
systems that were proposed earlier utilized collaborative
filtering. Web Content Recommendation has been an active
application area for Information Filtering, Web Mining and
Machine Learning research. The future work explains some of
the modifications using other algorithms to generate the
association rules that can be adopted on existing web
recommendation system to make them functionally more
effective. In order to explore the performance of the proposed
web recommendation system experiments were conducted on
available dataset. The performance of the proposed approach is
best illustrated by comparing it with K-nearest neighboring
algorithm.

Keywords — Association Rules,
Collaborative Recommender System,
Machine Learning, and Web Mining.

Apriori Algorithm,
Information Filtering,

Introduction

The extent of the Internet is getting larger and larger in
modern years. Therefore it is obligatory that a user need to
expend much time to select indispensable information from
large amount of web pages created every day. Addressing this
problem, several web page recommender systems are
constructed which automatically selects and recommends web
pages suitable for user's support. The majority of web page
recommender systems that was proposed earlier utilized
collaborative filtering [1], [2], and [3]. Collaborative filtering
is often used in general product recommender systems, and

consists of the following stages. The foremost stage in
collaborative filtering is to analyze users purchase histories in
order to extract user groups which have similar purchase
patterns. Then recommend the products that are commonly
preferred in the user's group [4].

In general the Recommender Systems (RS) uses the opinion
of members of a community to facilitate individuals identify
the information most likely to be interesting to them or
pertinent to their needs. This can be achieved by drawing on
user preferences and filtering the set of feasible options to a
more manageable subset. Every Web Recommendation
Systems have its own advantages and limitations [5].
Moreover the assignment of recommender systems is to
recommend items that fit a user's taste, in order to help the
user in selecting/purchasing items from a devastating set of
choices. Such systems have immense importance in
applications such as e-commerce, subscription based services,
information filtering, web services etc.

There are two fundamental approaches that can be applied
when generating recommendations. Content based approaches
profile users and items by identifying their characteristic
features, such us demographic data for user profiling, and
product information/descriptions for item profiling. The
profiles are used by algorithms to unite user interests and item
descriptions when generating recommendations [6]. Web
Content Recommendation has been an active application area
for Information Filtering, Web Mining and Machine Learning
research. This proposed paper presents a method of developing
a collaborative web recommendation systems using
association rule mining. The method utilize apriori algorithm
to generate association rules. It also explains some of the
baseline algorithms that are used in developing the web
recommendation systems. The future work explains some of
the modifications using other algorithms to generate the
association rules that can be adopted on existing web
recommendation system to make them functionally more
effective.

The remainder of this is organized as follows. Section 2
discusses various collaborative web recommendation systems
that were earlier proposed in literature. Section 3 explains the
proposed work of developing a web recommendation system

222

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

(IJCSIS)

using association rules generated from apriori algorithm.
Section 4 illustrates the results for experiments conducted on
different dataset in evaluating the performance of the proposed
web recommendation system. Section 5 concludes the paper
with fewer discussions.

II. Related Work

In general, the collaborative recommendation systems can
be grouped into four categories. On the basis of its temporal
and spatial characteristics, each system can be either
synchronous or asynchronous, and either local or remote.
Conversely, the most significant difference between these
different collaborative Web recommendation systems is the
method used to extort user preferences from personal
information. This section of the paper discusses various
methods proposed earlier in literature for a collaborative web
recommendation system.

Chen et al. in [7] proposed a Gradual Adaption Model for a
Web recommender system. The model is used to track users'
center of attention and its transition by analyzing their
information access behaviors, and recommend appropriate
information. The web pages admittance by users are classified
by the concept classes, and grouped into three terms of short,
medium and long periods, and two categories of significant
and incomparable for each concept class, which are used to
describe users' focus of interests, and to institute reprocess
probability of each concept class in each term for each user by
Full Bayesian Estimation as well. According to the reuse
probability and period, the information that a user is likely to
be interested in is recommended. They proposed a new
approach by which short and medium periods are determined
based on dynamic sampling of user information access
behaviors.

Niwa et al. in [8] described a web page recommender
system based on Folksonomy mining. They projected a way
to assemble a new type of web page recommender system
covering all over the Internet, by using Folksonomy and
Social Bookmark which are getting very well-liked in these
days, a new way to express users' preference to web pages
was formulated by mining tag data of Folksonomy.
Folksonomy is a new classification technique which may take
place of past taxonomy. Social Bookmark (SBM) is a variety
of web services on which users can divide up their
bookmarks. Anyone can see anyone's bookmark on SBM. In
order to solve some problems faced by conventional
recommender systems, they expressed users' web page
preference by "affinity level between each user and each tag."
By this approach, users' preferences are abstracted and it
becomes easier to find similar users. Clustering can also solve
the problem of "tag redundancy in Folksonomy."

A hybrid web recommender system was described by
Taghipour et al. in [9]. They exploit an idea of combining the
conceptual and usage information to enhance a reinforcement
learning framework, primarily devised for web
recommendations based on web usage data. Moreover the
combination can improve the quality of web
recommendations. A hybrid web recommendation method is

International Journal of Computer Science and Information Security,

Vol 8, No. 3, 2010
proposed by making use of the conceptual relationships

among web resources to derive a novel model of the problem,

enriched with semantic knowledge about the usage behavior.

With their proposed hybrid model for the web page

recommendation problem they revealed the pertinent and

flexibility of the reinforcement learning framework in the web

recommendation domain, and demonstrated how it can be

extended in order to incorporate various sources of

information. Their test results suggested that the method can

improve the overall quality of web recommendations.

An intelligent recommender system was projected by
kavitha devi et al. in [10]. They designed and implemented
an Intelligent Collaborative Recommender System (ICRS) to
map users' needs to the items that can persuade them. A
methodology is used to animatedly modernize the accuracy
factor based on user intelligence. The diverse approaches for
recommendation are categorized as memory-based and
model-based approaches. Memory-based systems suffer from
data sparsity and scalability problems, whereas model-based
approaches are liable to bind the range of users. Therefore
they integrated these approaches to overcome their
limitations. They applied the collaborative filtering approach
for recommendations. Recommendations are made more
accurate by applying regression to weighted aggregated
predictions. Mean Absolute Error Metrics was considered for
evaluating the performance of their proposed system. This
approach thus alleviates scalability and sparsity problems and
offers accurate recommendations.

Cheng et al. in [11] developed a two stage collaborative
recommender system. They proposed a chronological pattern
based collaborative recommender system that predicts the
customer's time- variant acquisition behavior in an e-
commerce environment where the customer's purchase
patterns may change gradually. A new two-stage
recommendation process is developed to envisage customer
behavior for the selection of different categories, as well as
for product items. Their study is the first to recommend time-
decaying sequential patterns within a collaborative
recommender system. Their experimental results revealed that
the proposed system outperforms the traditional collaborative
system.

Lin et al. in [12] described an efficient Adaptive -Support
Association Rule Mining for Recommender Systems. They
investigated the utilization of association rule mining as an
underlying technology for collaborative recommender
systems. Association rules have been used with sensation in
other domains. Nevertheless, most currently existing
association rule mining algorithms were designed with market
basket analysis in mind. They described a collaborative
recommendation technique based on a novel algorithm
distinctively designed to excavate association rules for this
rationale. The main advantage of their proposed approach is
that their algorithm does not require the minimum support to
be specified in advance. Rather, a target range is given for the
number of rules, and the algorithm adjusts the minimum
support for each user in order to obtain a rule set whose size is
in the desired range. Moreover they employed associations
between users as well as associations between items in
making recommendations. The experimental evaluation of a

223

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

system based on their algorithm revealed that its performance
is significantly better than that of traditional correlation-based
approaches.

Jung in [13] described a user-support method based on the
distribution of knowledge with other users through the
collaborative Web browsing, focusing exclusively on the
user's interests extracted from users' own bookmarks. More
prominently, they focused on those items of information
which are associated to the user's interests. In collaborative
Web browsing, they considered that recognizing the user's
interests is an extremely essential mission. Furthermore,
asking applicable information for other users, filtering the
query results, and recommending them are additional most
important tasks that have to be unreservedly conducted by
them in [13]. Based on the personalized TF-IDF proposal they
introduced the extended application of a BIS Agent, which is
a bookmark sharing agent system. Moreover they
implemented an ontological supervisor which can perform the
semantic analysis of the Web sites pointed to by these
bookmarks. They also designed a multi-agent system that
consists of a facilitator agent and many personal agents. The
main limitation of this system is that it does not consider the
privacy problems related with sharing personal information of
the user.

A novel recommender system was formulated by Marko et
al. in [14]. Their approach named, "Fab" is a recommendation
system designed to help users sift through the mammoth
amount of information obtainable in the World Wide Web.
Their proposed approach is the combination of content-based
filtering and collaborative filtering methods. The combination
exploits the advantages of both the methods thereby avoiding
the shortcomings. Tab's hybrid structure allows for automatic
recognition of emergent issues relevant to various groups of
users. It also enables two scaling problems, pertaining to the
rising number of users and documents, to be addressed. In
general the content-based approach to recommendation has its
pedigree in the information retrieval (IR) community, and
utilizes many of the same techniques. The collaborative
approach computed the similarity of the users rather than
computing the similarity of the items. They maintained user
profiles content analysis and directly compared these profiles
to determine similar users for collaborative recommendation.
The process of recommendation can be partitioned into two
stages: collection of items to form a manageable database or
index, and subsequently selection of items from this database
for particular users. The experimental results using the hybrid
Fab system achieved higher accuracy.

III. Proposed Approach
A. Association Rule Mining

In general association rule mining is a technique common
in data mining that attempts to discover patterns of products
that are purchased together. The proposed approach adapts the
Apriori algorithm [15] to collaborative filtering in an attempt
to discover patterns of items that have common ratings.
Association rules capture relationships among items based on
patterns of co-occurrence across transactions. The association
rules were applied to personalization based on web usage data

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

Vol 8, No. 3, 2010
in [16]. Considering each user profile as a transaction, it is

possible to use the Apriori algorithm [15] and generate

association rules for groups of commonly liked items.

Given a set of user profiles U and a set of item sets I = {Ii,
I2... Ik}, the support of an item set Ij g I is defined as a(Ii) =
l{u G U : Ii ^ u}l / lUI. Item sets that satisfy a minimum
support threshold are used to generate association rules. These
groups of items are referred to as frequent item sets. In
addition, association rules that do not satisfy a minimum lift
threshold are shortened. If there is not adequate support for a
particular item, that item will never come into view in any
recurrent item set. The suggestion is that such an item will
never be recommended. The subject of coverage is a tradeoff.
Lowering the support threshold will make sure that more
items can be recommended, but at the hazard of
recommending an item without enough evidence of a pattern.

Ahead of performing association rule mining on a
collaborative filtering dataset, it is indispensable to discretize
the rating values of each user profile. Therefore first subtract
each user's average rating from the ratings in their profile to
attain a zero -mean profile. Next, give a discrete category of
"like" or "dislike" to each rated item in the profile if it's
rating value is > or < zero, respectively. As a result of
discretizing the dataset the total number of features used in
the analysis is doubled. It is clear that a collaborative
recommender must take such preference into description or
risk recommending an item that is rated often, but disliked by
consensus. Another possibility is that one association rule
may add item 'i' to the candidate set with a "like" label, while
another rule may add the identical item with a "dislike" label.
There is not an ultimate solution in this case, but we have
chosen to presuppose that there are opposing forces for the
recommendation of the item. This implementation subtracts
the confidence value of the "dislike" label from the
confidence value of the "like" label. The search for item sets
is facilitated by storing the frequent item sets in a directed
acyclic graph, called a Frequent Item set Graph [16].

Given a objective user profile 'u', it is necessary to execute
a depth-first search of the Frequent Item- set Graph. When we
arrive at a node whose repeated item set In is not enclosed in
'u', the item i g In not found in 'u' is added to the candidate
set 'C' and search at the current branch is finished. Note that
the item set of the parent node Ip to In must be restricted in 'u'
by definition, and because In is of size d -f 1 where Ip is size
'd', there can be only one item i g In that is not contained in
'u'. It follows that In = Ip u {i} and the two nodes correspond

to the rule L

[ij. Then calculate the confidence of the rule

as a(In)/a(Ip). The candidate i g C is stored in a hash table
along with its confidence value. If it already exists in the hash
table, then the highest confidence value takes precedent.

After finishing point of the depth-first search, all
promising candidates for the target user 'u' are enclosed in
'C', including items labeled "dislike". Consecutively to
accurately characterize an estimated negative implication,
items labeled "dislike" is given a recommendation score that
is the negation of the confidence value. As a final step, the

224

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

(IJCSIS)

candidate set 'C is arranged according to the
recommendation scores and the top N items are returned as a
recommendation.

B. Baseline Algorithms

K Nearest Neighbour

The customary k-nearest neighbor algorithm is broadly
used and reasonably precise [17]. Resemblance is calculated
using Pearson's correlation coefficient, and the k most
analogous users that have rated the target item are selected as
the neighborhood. This make note that a target user may have
a different neighborhood for each target item. It is also
general to filter neighbors with similarity below a specified
threshold. This prevents prophecy being based on very remote
or negative correlations. After identifying a neighborhood,
Resnick's algorithm is used to calculate the prediction for a
target item T and target user 'u'.
K Means Clustering

A standard model-based collaborative filtering algorithm
uses k-means to cluster similar users. Given a set of user
profiles, the space can be partitioned into k groups of users
that are close to each other based on a measure of similarity.
The discovered user clusters are then applied to the user based
neighborhood formation task, rather than individual profiles.
To make a recommendation for a target user 'u' and target
item 'i' it is essential to select a neighborhood of user clusters
that have a rating for T and whose aggregate profile is most
similar to 'u'. This neighborhood represents the set of user
segments that the target user is most likely to be a member,
based on a measure of similarity. Pearson's correlation
coefficient is implemented effectively to perform this task.
Probabilistic Latent Semantic Analysis

In general the Probabilistic latent semantic analysis (PLSA)
models [18] present a probabiHstic approach for characterizing
latent or hidden semantic associations among co -occurring
objects. PLSA can be applied to the creation of user clusters
based on web usage data. The proposed approach has adapted
this technique to the context of collaborative filtering [19].
Moreover the Expectation-Maximization (EM) algorithm is
used to perform maximum likelihood parameter estimation. In
the expectation step, posterior probabilities are computed for
latent variables based on current estimates. In the
maximization step, Lagrange multipliers [20] are used to
obtain the re-estimated parameters. Therefore iterating the
expectation and maximization steps monotonically increases
the total likelihood of the observed data L (U, I), until a local
optimum is reached.

IV. Experimental Results

In order to evaluate the robustness of our recommendation
algorithm based on association rule mining a data set is taken
into account. This dataset consists of 100,000 ratings on 1682
movies by 943 users. All ratings are integer values between
one and five, where one is the lowest (dis-liked) and five is
the highest (liked). Initially the accuracy of the proposed
association rule mining based recommender system is

International Journal of Computer Science and Information Security,

Vol 8, No. 3, 2010
analyzed. To estimate the recommendations, 10-fold cross-
validation on the entire dataset and without attack profiles
was performed. Since Apriori selects recommendations from
only among those item sets that have met the support
threshold, it will by necessity have lower coverage than our
baseline algorithms. There will be some items that do not
appear in the Frequent Item set Graph, and about which the
algorithm cannot make any predictions. This may the issues
that arise in most of the baseline algorithms.

The Apriori algorithm would therefore lend itself best to
scenarios in which a list of top recommended items is sought,
rather than a rating prediction scenario in which the
recommender must be able to estimate a rating for any given
item. The selectivity of the algorithm may be one reason to
expect it will be relatively robust - it will not make
recommendations without evidence that meets the minimum
support threshold. However, the performance of Apriori and
PLSA are superior to k-means at large attack sizes.
Robustness of the Apriori algorithm may be moderately due
to lower coverage. However, this does not account for the flat
trend of hit ratio with respect to attack size.

Only the Apriori algorithm holds steady at large filler sizes
and is essentially unaffected. As with attack size, the reason
that filler size does not affect the robustness of the algorithm
is because adding more filler items does not change the
probability that multiple attack profiles will have common
item sets. The fact that a profile's ratings are discretized to
categories of "like" and "dislike" means that an attack profile
with 100% filler size will cover exactly half of the total
features used in generating frequent item sets. Therefore, it is
very unlikely that multiple attack profiles will result in mutual
reinforcement. Apriori has also exhibited improved
robustness compared to the other algorithms against certain
attacks.

The Apriori algorithm appears to have the same robustness
as the other model-based algorithms at small attack sizes.
Although the performance of Apriori is not ideal against a
segment attack, certain scenarios may minimize the
performance degradation in practice. In particular, a
recommender system with a very large number of users is
somewhat buffered from attack. The algorithm is quite robust
through a 5% attack, and is comparable to both k-means and
PLSA. The robustness of Apriori is not drastically reduced
until attack size is 10% or greater. Table 1 shows the results
of normalized mean absolute error evaluated for proposed
approach and K nearest neighbor. Similar table 2 represents
the coverage comparison of the proposed approach using
apriori algorithm and K nearest neighbor. Figure 1 (a) shows
the comparison of the proposed approach and k-nn algorithm
in terms of their mean absolute error. Figure 1 (b) represents
the comparison of apriori and k-nn algorithm in terms of
coverage. The results revealed that the proposed approach
using association rule mining performed well in
recommendation by determining the user' needs.

225

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

Table.1. Normalized mean absolute error

Approach

Run 1

Run 2

Run 3

Mean

Apriori

0.3293

0.3292

0.3287

0.3291

K-nn

0.3332

0.3337

0.3339

0.3336

Table. 2. Coverage

Approach

Runl

Run 2

Run 3

Mean

Apriori

0.4701

0.4716

0.4718

0.4712

K-nn

0.9942

0.9941

0.9942

Normalized Mean Absolute Error

0.335
0.334
0.333
0.332
0.331
0.33
0.329
0.328
0.327
0.326

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

Vol 8, No. 3, 2010
large amount of web pages created every day. Addressing this

problem, several web page recommender systems are

constructed which automatically selects and recommends web

pages appropriate for user's support. The greater part of web

page recommender systems that were proposed earlier utilized

collaborative filtering. This proposed paper presents a method

of developing a collaborative web recommendation systems

using association rule mining. In general association rule

mining is a technique common in data mining that attempts to

discover patterns of products that are purchased together. The

association rules were applied to personalization based on

web usage data. The method utilize apriori algorithm to

generate association rules. The Apriori algorithm would

therefore provide itself best to scenarios in which a list of top

recommended items is required, rather than a rating prediction

scenario in which the recommender must be able to

approximate a rating for any given item. The results revealed

that the proposed approach using association rule mining

performed well in recommendation by determining the user'

needs. Future work mainly concentrates on determining the

mutual reinforcement between common item sets thereby

ehnancing the accuracy of the recommender system.

- Apriori
-K-nn

Run 1

Run 2

Run 3

(a)

Coverage

1.2 n

A _

0.8 -
0.6 -
0.4 -

— ♦ — Apriori
-■—K-nn

A A

0.2 -

Run 1

Run 2 Run 3

(b)

Figure. 1 (a) & (b) represents comparison of Apriori and K-nn

Algorithm in terms of Normalized Mean Absolute Error and

Coverage respectively

V. Conclusion

The scope of the Internet is getting larger and larger in
recent years. Therefore it is compulsory that a user need to
disburse much time to decide on necessary information from

References

[1]

J. Li, and O. Zaiane, "Combining Usage, Content, and Structure Data to
Improve Web Site Recommendation," Proceedings of WebKDD-2004
workshop on Web Mining and Web Usage, In 5th International
Conference on Electronic Commerce and Web Technologies, pp. 305-
315,2004.

P. Kazienko, and M. Kiewra, "Integration of relational databases and
Web site content for product and page recommendation," International
Symposium on Database Engineering and Apphcations Symposium,
IDEAS '04, pp. 111-116, 2004.

N. Golovin, and E. Rahm, "Reinforcement Learning Architecture for
Web Recommendations," Proceedings of the International Conference
on Information Technology: Coding and Computing (ITCC'04), vol. 2,
p. 398, 2004.

Jonathan L. Herlocker, Joseph A. Konstan, Loren G. Terveen, and John
T. Riedl, "Evaluating collaborative filtering recommender systems,"
ACM Transactions on Information Systems, vol. 22, no. 1, pp. 5-53,
2004.

Punam Bedi, Harmeet Kaur, and Sudeep Marwaha, "Trust based
Recommender System for the Semantic Web," Proceedings of the 20*^
international joint conference on Artificial intelligence, pp. 2677-2682,
2007.

Gabor Takacs, Istvan Pilaszy, Bottyan Nemeth, and Domonkos Tikk,
"Scalable Collaborative Filtering Approaches for Large Recommender
Systems," The Journal of Machine Learning Research, vol. 10, pp. 623-
656, 2009.

Jian Chen, Roman Y. Shtykh, and Qun Jin, "A Web Recommender
System Based on Dynamic Sampling of User Information Access
Behaviors," Ninth IEEE International Conference on Computer and
hiformation Technology, vol. 2, pp. 172-177, 2009.
Satoshi Niwa, Takuo Doi, and Shinichi Honiden, "Web Page
Recommender System based on Folksonomy Mining," Proceedings of
the Third International Conference on Information Technology: New
Generations, IEEE Computer Society, pp. 388-393, 2006.
Nima Taghipour, and Ahmad Kardan, "A hybrid web recommender
system based on Q-learning," Proceedings of the 2008 ACM
symposium on Applied computing, pp. 1164-1168, 2008.
[10] M.K. Kavitha Devi, and P. Venkatesh, "ICRS: an intelligent
collaborative recommender system for electronic purchasing,"

[2]

[3]

[4]

[5]

[6]

[7]

[8]

[9]

226

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

International Journal of Business Excellence 2009, vol. 2, no. 2, pp.

179-193, 2009.
[11] Cheng-Lung Huang, and Wei-Liang Huang, "Handling sequential

pattern decay: Developing a two-stage collaborative recommender

system," Elsevier, Electronic Commerce Research and Applications,

vol. 8, no. 3, pp. 117-129,2009.
[12] Weiyang Lin, Sergio A. Alvarez, and Carolina Ruiz, "Efficient

Adaptive-Support Association Rule Mining for Recommender

Systems," Journal on Data Mining and Knowledge Discovery, vol. 6,

no. 1, pp. 83-105,2004.
[13] Jason J. Jung, "Collaborative Web Browsing Based on Semantic

Extraction of User Interests with Bookmarks," Journal of Universal

Computer Science, vol. 11, no. 2, pp. 213-228, 2005.
[14] Marko Balabanovic, and Yoav Shoham, "Fab: content-based,

collaborative recommendation," Communications of the ACM, vol. 40,

no. 3, pp. 66-72, 1997.
[15] R. Agrawal, and R. Srikant, "Fast algorithms for mining association

rules," In Proceedings of the 20th International Conference on Very

Large Data Bases (VLDB'94), Santiago, Chile, September 1994.
[16] M. Nakagawa, and B. Mobasher, "A hybrid web personalization model

based on site connectivity," In Web KDD Workshop at the ACM

SIGKKDD International Conference on Knowledge Discovery and Data

Mining, Washington, DC, August 2003.
[17] J. Herlocker, J. Konstan, A. Borchers, and J. Riedl, "An algorithmic

framework for performing collaborative filtering," In Proceedings of the

22nd ACM Conference on Research and Development in Information

Retrieval (SIGIR'99), Berkeley, CA, August 1999.
[18] T. Hofmann, "Probabilistic latent semantic analysis," In Proceedings of

the Fifteenth Conference on Uncertainty in Artificial Intelligence,

Stockholm, Sweden, July 1999.
[19] B. Mobasher, R. Burke, and J. Sandvig, "Model-based collaborative

filtering as a defense against profile injection attacks," In Proceedings

of the 21st National Conference on Artificial Intelligence, pp. 1388-

1393, AAAI, July 2006.
[20] T. Hofmann, "Unsupervised learning by probabihstic latent semantic

analysis," Machine Learning Journal, vol. 42, no. 1, pp. 177-196, 2001.

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

Vol 8, No. 3, 2010
research interests include Natural Language Processing and

Data Compression. He has delivered lectures in the areas of

NLP, Information Security and Image Compression in

conferences. He organized and acted as a chair person for

various seminars and conferences. He introduced many

research programmes in Pondicherry University. He is a life

member of Computer Society of India and Indian Society for

Technical Education.

Author Biography

r _ A. Kumar has around 21 years of

0^^\ experience in Information Technology

\ ^ ^1 and its Applications with expertise in

^g. Data mining, Information and

^1^^ Knowledge Management and Web

"^^T*^ Technology. He has published a number

of papers in peer-reviewed journals and
conferences. He has actively served as a reviewer for many
leading International Journals and conferences. He has
delivered lectures in the areas of Web Recommendation
Systems, Knowledge Management for Seminars and
conferences. Currently serving as Head in Computer
Science and Engineering, Perunthalaivar Kamarajar Institute
of Engineering and Technology, India and Pursuing Ph.D.,
in the area of Web Recommendation Systems at
Sathyabama University, Chennai.

Dr. P. Thambidurai is Principal,
Perunthalaivar Kamarajar Institute of
Engineering & Technology, Karaikal, India.
He received the M.E (Computer Science).,
from College of Engineering, Guindy,
Chennai and Ph.D. in Computer Science
from Alagappa University, Karaikudi, India
in 1984 and 1995 respectively. He has published more than
100 papers in peer-reviewed journals and conferences. His

227

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

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

Vol 8, No.3, June 2010

SIMILARITY BASED IMPUTATION
METHOD FOR TIME VARIANT DATA

F.Sagayaraj Francis, Vishnupriya.B, Vinolin Deborah Delphin, Saranya Kumari Potluri
Pondicherry Engineering College, Pondicherry, India

Abstract — The intent of any analysis is to make valid inferences
regarding a population of interest. Missing data threatens this
goal if it is missing in a manner which makes the sample different
than the population from which it was drawn, that is, if the
missing data creates a biased sample. Therefore, it is important
to respond to a missing data problem in a manner which reflects
the population of inference. This paper deals with the proposal of
an efficient method of filling missing data called Similarity based
Imputation Method (SIM). SIM processes the target segment by
extracting its features and based on the extracted features the
target segment is classified into its appropriate cluster which has
complete data segments, which are similar to the target segment.
Now from the similar segments within the identified cluster, the
most identical segment is found using similarity measure and the
values substituted from that complete segment

Keywords— Imputation, Time variant Multi-dimensional data,
Clustering, Feature Extraction, Similarity measure. Segment
Matching

I. Introduction

A time series may be defined as a collection of
readings belonging to different time periods, of some
economic variable or composite of variables. It is a set of
observations of a variable usually at equal intervals of time
where time may be yearly, monthly, weekly, daily, hourly or
even minute data. Hourly temperature reading, daily sales,
monthly production, Earth's magnetic field variations are
examples of time series. Time series analysis comprises
methods for analyzing time series data in order to extract
meaningful statistics and other characteristics of the data.

The primary purpose of the analysis of time series is
to discover and measure all types of variations which
characterize a time series. For efficient analysis of data,
complete datasets are required. There is a possibility to miss
out several observations due to unexpected events such as
equipment failure or unexpected disturbances. Moreover,
occurrence of missing observations in datasets is an actual yet
challenging issue confronted in machine learning and data
mining. Missing values may generate bias and affect the
quality of the supervised learning process or the performance
of classification algorithms. However, most learning
algorithms are not well adapted to some application domains
due to the difficulty with missing values (for example, Web
applications) as most existing algorithms are designed under
the assumption that there are no missing values in datasets.
That implies that a reliable method for dealing with those
missing values is necessary. Generally, dealing with missing
values means to find an approach that can fill them and

maintain the original distribution of the data as closely as
possible. There are many approaches to deal with missing
values described in, for instance:

1. Ignore objects containing missing values;

2. Fill the missing value manually

3. Substitute the missing values by a global constant or
the mean of the objects

4. Obtain the most probable value to fill in the missing
values

The first approach usually loses too much useful
information, whereas the second one is time consuming and
expensive and hence infeasible in many applications. The third
and fourth approaches assume that all missing values are with
the same value, probably leading to considerable distortions in
data distribution. The method of imputation, however, is a
popular strategy. In comparison to other methods, it uses as
much information as possible from the observed data to
predict missing values.

The approach used in this paper fills the missing data
by imputation using a similarity measure. The method
involves finding a similar segment that matches the segment
with missing data best using similarity measures of the
segments.

II. Existing Methodologies

In general there can be two approaches to handle
missing data. The first approach to find missing or lost data is
by computation. These methods are statistical or numeric in
nature, such as Line/Parabola of best fit or
Interpolation/Extrapolation. The methods heavily depend on
the formation of a function/model/system to fill missing data.
The formation of function/model/system is inherently complex
and requires voluminous computations.

The second approach to find missing or lost data is by
imputation. Imputation is the substitution of missing values in
an incomplete dataset (target dataset) with values from a
similar but complete dataset (source dataset). This approach
uses similarity/distance/entropy measures to solve the
problem. They can also be extended to predict patterns that are
likely to occur. Traditional missing value imputation
techniques can be roughly classified into two types.

• Parametric imputation (e.g., the linear regression) and

• Non-parametric imputation (e.g., non-parametric
kernel-based regression method. Nearest Neighbor
method).

The parametric regression imputation is superior if a
dataset can be adequately modeled parametrically, or if users

228

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

can correctly specify the parametric forms for the dataset. For
instance, the linear regression methods usually can treat well
the continuous target attribute, which is a linear combination
of the conditional attributes. However, when the actual
relation between the conditional attributes and the target
attribute is not known, the performance of the linear regression
for imputing missing values is very poor. In real application, if
the model is incorrectly specified, the estimations of
parametric method may be highly biased and the optimal
control factor settings may be miscalculated. Non-parametric
imputation algorithms that can provide superior fit by
capturing structure in the dataset if the actual distribution of a
dataset is known. Using a non-parametric algorithm is
beneficial when the form of relationship between the
conditional attributes and the target attribute is not known a-
priori.

While nonparametric imputation method is of low-
efficiency, the popular nearest neighbor (NN) methods face
two issues: (i) Each instance with missing values requires the
calculation of the distances from it to all other instances in a
dataset; and (ii) There are only a few random chances for
selecting the nearest neighbor.

A. Feature Extraction

Time variant datasets are voluminous and hence
handling raw time variant data is not viable. A solution to this
would be to extract features from time series segments and
deal with features rather than the raw data as such.

Generalized feature extraction relies on a collection
of feature extractors that function independently of domain
and application. For time series data, such feature extractors
must be able to identify generally useful structures that emerge
from the relationships between consecutive measurement
values over time.

For traditional tabular data the similarity is often
measured by attribute value similarity or even attribute -value
equality [1]. For more complex data, e.g., geo- time series
data, such simple similarity measures do not perform very
well. So similarity of time series data should be based on
certain characteristics of the data rather than on the raw data
itself. Feature Extractor is characterized by its ability to
capture fundamental trends and relationships, generate
accurate approximations, represent the extracted structures
compactly, support subsequent classification, and being
domain independent [2]. But this technique only deals with
finding the nature of a particular segment and does not extract
the uniqueness of each segment and it does not maintain
originality of the target segment. Therefore a more general
time series feature extraction technique is to be adopted which
would represent any time series.

A well established feature extraction technique using
Discrete Fourier Transformation (DFT) for time series use
only the first k coefficients, discarding the rest. This
corresponds to saving only a rough sketch of the time series,
because the first coefficients represent the overall nature of the

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

Vol 8, No.3, June 2010
segment [6,8]. If N is the number of observations, in single
dimension.

N=0

. im

where yl=0,...,A^-l (1)

Since the data set would be a bunch of time segments, the
DFT would thus transform them from the time domain into
frequency domain. It is easy and fast to compute. It preserves
the distance between two objects. DFT ensures completeness
of feature extraction [7].

B. Similarity Search

Given a set of time series segments, there are two
types of similarity searches: (i) sub segment matching finds all
the data sub segments of a larger segment that are similar to
the given smaller segment and (ii) whole matching finds those
segments that are similar to one another.

Two key aspects for achieving effectiveness and
efficiency, when managing time series data are representation
methods and similarity measures. Time series are essentially
high dimensional data and dealing with raw data directly is
very expensive in terms of processing and storage cost. Unlike
canonical data types like ordinal or nominal variables, where
the distance definition is straightforward, the distance between
time series needs to be carefully defined in order to reflect the
underlying similarity of such data. This is particularly
desirable for similarity-based retrieval, classification and
clustering of time series.

Many similarity measures are efficient in accurately
quantifying the similarity between any two distinct objects.
Some of the efficient similarity measures to state can be Bray-
Curtis distance, Canberra Distance, Cosine Distance,
Correlation Distance and Chessboard Distance. However these
measures were not able to capture the similarity perfectly
between any two time variant segments. These methods were
lacking in way that they compare only the corresponding
points of the two data sets, but a time varying dataset is unique
in its nature of being recorded at discrete and equal intervals
of time and hence similarity measures exclusively for time
variant data are to be considered.

Past research, on the choice of distance/similarity
function to obtain similarity measure between two segments
that are time variant, reveal that it can be divided into two
classes. The first and straightforward method is the Euclidean
distance and its variants like Manhattan distance. The
complexities of these measures are linear, they are easy to
implement and parameter free.

However, since the mapping between the points of
two time series is fixed in the above two methods, these
distance measures are very sensitive to noise and
misalignments in time, and are unable to handle local time
shifting i.e., similar segments that are out of phase.

The second class includes the Dynamic Time
Warping (DTW) [3]. Continuity is less important in DTW than
in other pattern matching algorithms. DTW is an algorithm
particularly suited to matching segments with missing
information, provided there are long enough segments for

229

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

matching to occur. Optimal alignment [4,10] i.e., minimum
distance time warp path is obtained by allowing assignment of
multiple successive values of one time series to a single value
of the other time series and therefore it can be calculated on
time series of different lengths.

Dynamic time warping is an algorithm for measuring
similarity between two segments which may vary in time.
DTW aligns two time series in the way some distance measure
is minimized as shown in Fig. 1. It can efficiently handle noise
and misalignments in time, and are able to handle local time
shifting i.e., similar segments that are out of phase [9].

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

Vol 8, No.3, June 2010
Thus DTW gives a quantified measure of similarity
between any two instances and smaller the value more is the
similarity between them. So to impute any segment with
missing data, it is required to compute DTW between the
target segment and every other complete segment in the
dataset. But since the dataset is very large it is cumbersome to
calculate similarity measure with every other segment. Thus
there arises a need for clustering wherein the raw data and
their extracted features can be grouped as clusters each of
which contains similar instances.

\ \

/
/

"^^

Time

Figure. 1 A warping between two time series

In Fig. 1, each vertical line connects a point in one
time series to its correspondingly similar point in the other
time series. The lines actually have similar values on the y-
axis but have been separated so the vertical lines between
them can be viewed more easily. If both of the time series in
Fig. 1 were identical, all of the lines would be straight vertical
lines because no warping would be necessary to 'line up' the
two time series. The warp path distance is a measure of the
difference between the two time series after they have been
warped together, which is measured by the sum of the
distances between each pair of points connected by the vertical
lines in Fig.l. Thus, two time series that are identical except
for localized stretching of the time axis will have DTW
distances of zero. Given two time series X, and Y, of lengths
IXI and lYI; X= xi,X2,...,Xi,...,X|xi and Y= yi,y2,...,yi,...,yiyi, a
warp path W is given by

W=Wi,W2,...,Wi,...,Wk, where max(|X|,|Y|)<y^<IXI+IYI)
k is the length of the warp path and the li"^ element of the warp
path is Wk=(i,j),Wk+i=(ik,jk) and / is an index from time series
X, and j is an index from time series Y. The warp path must
start at the beginning of each time series at Wi = (1, 1) and
finish at the end of both time series at Wk = (IXI, lYI). This
ensures that every index of both time series is used in the warp
path. There is also a constraint on the warp path that forces /
and 7 to be monotonically increasing in the warp path, which is
why the lines representing the warp path in Fig. 1 do not
overlap. Every index of each time series must be used. More
formally, the optimal warp path is the warp path is the
minimum-distance warp path, where the distance of a warp
path W is

Dist(w)=i:^-fDi5t(Wfc,WfcyJ

(2)

Dist(W) is the distance, of warp path W, and Dist(Wki,
Wij) is the distance between the two data point indexes in the
Ji"^ element of the warp path.

C Clustering

Clustering is a division of data into groups of similar
objects. Each group or a cluster consists of objects that are
similar between themselves and dissimilar to objects of other
groups. When the dataset under consideration is very large,
manual handling of data proved to be inefficient. So clustering
techniques are adopted to handle the data and thereby
minimize the computations.

The common clustering techniques include
hierarchical clustering, partitioning clustering, grid based
clustering and constraint based clustering. Some of the
hierarchical clustering techniques are agglomerative clustering
and divisive clustering, k-medoids, k-means, probabilistic and
density based clustering techniques fall under the category of
partitioned clustering.

III. SIMILARITY BASED IMPUTATION METHOD

The key objective of this paper is to fill out missing
data that may be lost due to some inevitable reasons. Every
data point adds to the machine learning or data mining process
which implies that no data point can be ignored. So the need to
fill missing data by some efficient method is needed that
involves less computation and would fill in the missing
observations more logically and accurately. This section
discusses a new method to fill missing data called Similarity
based Imputation Method (SIM) which involves filling by
similarity search. The idea is to fill missing data by finding
similar complete segments for the segments with missing data
and impute them.

SIM takes as input a large time variant dataset with
missing segments. The imputation process consists of two
phases. The first phase collects the significant parameters of
the dataset and the second uses these values to impute. The
steps in the first phase are:

1 . Divide the entire dataset into segments of length /.

2. Except for the segments with missing data, compute
segment parameters for each of the segment in the
dataset.

3. Cluster the segments based on their segment
parameters.

The second phase that imputes the missing values consists of
the following steps. The steps are applied on each segment
with missing data.

230

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

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

Vol 8, No.3, June 2010

1 . Compute segment parameters and classify the segment into
an appropriate cluster.

2. Divide each segment into sub segments.

3. Extract DFT coefficients of all the sub segments.

4. Using DFT coefficients find the most similar segment to
the segment with missing data. Impute values from the
complete segment into the segment with missing data

A lot of segment parameters may be used based on
which the segments can be clustered. A few possible
parameters may be count of local maxima, count of local
minima, position of global maxima, position of global minima,
global maxima value and global minima value. For example
for the sample dataset (223.9, 223.7, 223.6, 223.4, 223.3,
223.3, 223.2, 223.1, 223.2, 223.4, 223.5, 223.6, 223.8, 223.7,
223.8, 223.8, 223.8, 223.7, 223.7, 223.9, 223.9, 224.0, 224.1,
224.2), the above said parameters are 7, 6, 24, 8, 224.2, 223.1,
respectively.

Phase 1

Phase 2

RAW
DATA

SEGMENT WITH
MISSING DATA

PARTITIONING
CRITERIA

SEGMENT
PARAMETERS

DISCRETE
SEGMENTS

CLASSIFICATION
OF SEGMENT

FEATURE
EXTRACTION

SEGMENT
PARAMETERS

PREPROC
ESSING

K-MEANS
CLUSTERING

SIMILARITY
MEASURE (DTW)

CLUSTERS WITH NO
MISSING DATA

COMPARISON WITH THE DFT
OF OTHER SEGMENTS

IMPUTATION FROM THE
MOST SIMILAR SEGMENT

Figure. 2 Phase 1 and Phase 2 of imputation process

To compute the DFT coefficients, each segment is
divided into three sub segments. Let a and b be the positions
from and to in the incomplete segment where the data are

missing. The data from the first to (a-1) position forms the
first sub segment; data from d"^ to Z?* position forms the
second sub segment and the rest form the third sub segment of
the segments. If (223.9, 223.7, 223.6, 223.4, 223.3, 223.2, -1, -
1, -1, -1, -1, -1, 223.7, 223.8, 223.8, 223.8, 223.8, 223.7,
223.7, 223.9) represents the segment and -1 represents the
missing value, then a and b of all the segments would be 7 and
12. The DFT coefficients of the upper and lower sub segments
of the dataset are (591.283, -0.041,-0.091, 0.097, 0.312, 0.663)
and (592.044, -0.252, -0.25, -0.178,0.039, 0.507)

The identification of the similarity of the sub
segments involves the computation of similarity measure
DTW between every pair of sub segments. The candidate for
the imputation is the segment that is most similar to upper and
lower sub segments of the incomplete segment. The Fig. 2
explains the steps pictorially.

IV. EXPERIMENTAL EVALUATION

The dataset considered for experimental evaluation of
the method proposed in this paper is one of the three measures
of earth's magnetic field variations which are measured along
three axes as H- the horizontal intensity, Z- the vertical
intensity and D the dip. The frequency of observation is one
minute. The parameter that was used for evaluating the result
is Relative Error Percentage (REP). If AV is the actual value
and IV is the imputed value REP is given by

REP =

AV-IV
AV

(3)

Table 1 Relative Error Percentage of various experimental
setups

%of
Missing
Points

Region

of
Missing
Points

Number of Segments

100

200

500

1000

5000

10000

20000

Upper

0.42

0.17

Middle

0.25

0.128

0.085

Lower

0.553

0.043

Upper

0.128

0.06

Middle

0.17

0.128

Lower

0.46

0.045

Upper

0.06

Middle

0.17

0.128

0.085

Lower

0.468

0.043

Upper

0.426

0.06

Middle

0.17

0.128

0.085

Lower

0.468

0.043

Upper

0.298

0.213

Middle

0.043

Lower

0.553

0.043

231

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

The segment lengths were fixed at 20. The segment
parameters considered for the experiments were count local
maxima, count of local minima, position of global maxima,
position of global minima, value of global minima, value of
global minima, number of local minima before global minima,
number of local maxima after global minima, number of local
minima before global maxima, number of local maxima after
global maxima, number of local minima between global
minima and global maxima and number of local maxima
between global minima and global maxima. The various
scenarios for which the experiment was conducted and the
REP obtained are given in Table 1. The regions of missing
data are indicators of the variations in the lengths of upper and
lower sub segments of the segments.

The following observations about the experiment and
results are worth mentioning.

• The repletion of the values in the table is a good indicator
of the identification of same segment repeatedly
irrespective of the increase in number of segments.

• The best results were obtained as the numbers of segments
were increased.

• DTW similarity measure is an effective similarity measure
for time variant data.

• The increase in segment size also minimizes the error.

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

Vol 8, No.3, June 2010

Measures", Proceedings of the ninth ACM SIGKDD

international conference on Knowledge discovery and data

mining", Pages: 216 - 225,2003
[4] Blaz Strle, Martin Mozina, Ivan Bratko:" Qualitative

Approximation To Dynamic Time Warping Similarity

Between Time Series Data", 23rd International Workshop

on Qualitative Reasoning 2007
[5] M.Sridharan, N.Gururajan and A.M.S.Ramasamy: "Fuzzy

Clustering Analysis to study geomagnetic coastal effects"
[6] M. D. Morse and J. M. Patel. "An efficient and accurate

method for evaluating time series similarity ".In SIGMOD

Conference, 2007.
[7] I. Popivanov and R. J. Miller. "Similarity Search Over

Time-Series Data Using Wavelets". In ICDE, 2002.
[8] K.-P. Chan and A.-C. Fu." Efficient Time Series Matching

by Wavelets". ICDE, pages 126-133, 1999
[9] D. Berndt and J. Clifford." Using Dynamic Time Warping

to Find Patterns in Time Series". In AAAI-94 Workshop

on Knowledge Discovery in Databases, pages 359-370,

1994.
[10] B.-K. Yi, H. V. Jagadish, and C. Faloutsos. "Efficient

Retrieval of Similar Time Segments Under Time

Warping". In ICDE, pages 201-208, 1998.

AUTHORS PROFILE

V. CONCLUSION

This proposed methodology can be applied to any
dataset that suffers from potential loss of data due to
instrumental and other unanticipated faults. It cuts out a
solution for the problem of missing data by the strategy of
Imputation, which involves finding the segment which is most
similar to the one with missing data. Proficient feature
extraction techniques are adopted to handle the voluminous
data which represent continuous time variant data. This
minimizes huge computations which need to be done when
raw data is used for analysis. SIM performs better than other
existing imputation and computational methods because the
time complexity is minimized as the use of feature extraction
and clustering techniques minimizes the search space to a
single cluster. The relative percentage error is also very less
which adds to the efficiency of the method. It can perform
even better by improvising the clustering algorithms.

F.Sagayaraj Francis is currently working as an Associate Professor in the
Department of Computer Science and Engineering, Pondicherry Engineering
College. He obtained his M.Tech. degree in Computer Science and
Engineering in the year 1997 and Ph.D. degree in Computer Science and
Engineering in the year 2008, both from Pondicherry University. His areas of
research and interest are Data Management and Information Systems.

B.Vishnupriya has finished her B.Tech in Computer Science and Engineering
degree in the current year from Pondicherry University.

Vinolin Deborah Delphin has completed her B.Tech in Computer Science and
Engineering degree in the academic year from Pondicherry University.

Saranya Kumari Potluri has done her B.Tech in Computer Science and
Engineering degree in the current year from Pondicherry University.

REFERENCES

[1] Hui Ding, Goce Trajcevski, Peter Scheurmann, Xiaoyue
Wang, Eamonn Keogh: "Querying And Mining Of Time
Series Data: Experimental Comparison Of Representations
And Distance Measures", Proceedings of VLDB
Endowment, Vol. 1, Pages 1542-1552, 2008

[2] Robert T.Olszewski: "Generalised Feature Extraction for
Structural Pattern Recognition in Time Series Data"

[3] Michail Vlachos, Marios Hadjielefttheriou, Dimitrios
Gunopulos, Eamonn Keogh: "Indexing Multi-Dimensional
Time-Series with Support for Multiple Distance

232

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

Efficient Node Search in P2P Using Distributed

Spanning Tree

^ ^ ^ ^ 44-

P. Victer Paul T. Vengattaraman M.S.Saleem Basha P. Dhavachelvan R. Baskaran
Department of Computer Science, Pondicherry University, Puducherry, India.
^Department of Computer Science and Engineering, Anna University, Chennai, India.

(victerpaul vengat. mailbox, smartsaleeml979, dhavachelvan, baskaran.ramachandran j @ gmail.com

Abstract

Peer to Peer (P2P) networks are the
important part of the next cohort of Internet, so
how to search the node in the P2P networks
efficiently is the key problem of the perception of
the P2P network. However, the node search
process in unstructured P2P is not efficient
because the same search message may go through
a node multiple times. To ease the complex search
and improve the search efficiency, we propose a
mechanism using the interconnection structure
called Distributed Spanning Tree (DST) which
facilitates the P2P Network into a layered
structure to improve the node search technique.
The performance evaluations of simulation
demonstrate that the proposed mechanism can
improve the node search efficiency ofP2P systems.

Keywords: Peer to Peer, Distributed Spanning
Tree, Node Search, Ant colony optimization

1. Introduction

P2P networks are important part of the next
generation of Internet, so how to search the
resources in the P2P networks efficiently is the
key problem of the realization of the P2P network
[1]. The advantages of the unstructured P2P
networks are that they have lower maintenance
overhead and can better adapt to node
heterogeneity as well as network dynamics.
However, the node search process in unstructured
systems is not as efficient because the same search
message may go through a node multiple times [2].
For communication in network, operation for
search of a particular node is performed often
which consumes more message passes even the
node identification factor like IP address are
known. Since routing information stored by each

node is limited that makes the search operation
complex in large scale networks. To facilitate the
complex search and improve the search efficiency,
we propose a novel approach of node search using
DST technique. This reduces the message passes
required to identify a node and prevent nodes
from receiving duplicate search messages and
retains the low maintenance overhead for the
unstructured system.

From these perspectives, in this paper, it is
aimed at developing an optimized node search
system using Distributed Spanning Tree (DST),
which will reduce the number of message passes
required in deterministic environment. Various
requirements to reduce the message pass can be
achieved by formulating DST in the network.
Performance of this improved mechanism is
simulated and analyzed using a theme specific
environment for Node search. The paper is
organized as follows: Section 2 defines the
Interconnection Structure used and the respective
constrains. Section 3 explains about Formulation
of DST in P2P network. Section 4 discusses about
proposed mechanisms for Node search in P2P
network. Section 5 provides statistical Analysis
over result obtained in the Simulation and Section
6 concludes the proposed work with its merit.

2. Interconnection Structure Used

Distributed Spanning Tree (DST) [9, 10] is the
interconnection structure we follow to reduce the
number of message passes required for node
search. DST organizes P2P network into a
hierarchy of groups of nodes. The nodes are put
together in groups and groups are gathered in
higher level recursively. This organization is built
on top of routing tables allows the instantaneous

233

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

creation of spanning trees rooted by many nodes
and keeps the load balanced between the nodes [3].
The DST is an overlay structure designed to be
scalable [4]. The DST is a tree without
bottlenecks which automatically balances the load
between its nodes. From [3] and [4], it is possible
to achieve economic traffic optimization in Peer
Network using DST interconnection structure. So
we virtually convert the peer network into
distributed spanning trees and each tree should
have its root node we call it as Head Node (HN)
and others are Leaf Node (LN). Every LN will
hold the details of its own HN. Likewise every
HN will hold the complete details regarding its
LNs and all other HNs in the network. The details
stored in HNs and LNs in the DST is used to
enhance the node search operation efficiently with
minimum message pass. During the formulation
of DST in Peer network LNs and HNs are chosen
randomly and dynamically with some requirement
criteria which improve the Fault Tolerance of the
system.

To enhance the efficiency of DST in Node
Search, we focus on a kind of randomized search
heuristics, namely Ant Colony Optimization
(AGO). AGO [5, 6, 7, 8], is a powerful heuristic
approach to solve combinatorial optimization
problems such as the TSP, Routing in
telecommunication networks. So applying AGO
approach can enhance the effective routing of
message (at low cost) in the network which in-tem
reduces the number of message pass required.

3. Formulation of DST

In a large P2P network, formation of DST is
complex. We elucidate the DST formation in the
P2P network using five procedures.

Firstly initialize _D ST is a procedure which
initializes DST by creating Head Node (HN) in
Peer network based on some test criteria(s). The
criteria(s) to be checked can be user approval,
traffic on a particular region, etc., and the
procedure creates an array on each HN to hold its
LN details. If the criteria(s) fail a variable on the
node is created. Each HN is provided with unique
Priority Number (PN) to provide write priority
among the HNs. initialize _D ST is also a

procedure to set its HN id as their own id and then
it calls the ^roc^dnx^ probe_DST(),

The procedure probe_DST(), which is called
by every HN creates probe message and set 'id'
field of message as its own id and flood the
message to all peers it is connected.

On receiving a message every peer execute the
procedure msg_recieve_DST(msg) where 'msg' is
the received message. During DDST formation it
should be possible to get any one of the two types
of messages, tht probe message or reply message:

If there is a probe message, any one of the
following would be occurred:

Case-1: The message is received by a HN: It is
just discarded.

Case-2a: The message is received by a LN which
is not under any HN: LN stores the Head variable
as the id which it read from pmsg. Then call the
procedures probe_reply_DST(N(id)) and

probe J'orword_DST(pmsg).

Case-2b: The message is received by a LN, which
is under any HN: It is just discarded.

If there is a reply message, any one of the
following would be occurred:

Case-1: The message is received by a LN: It just
forwards it to the node bearing the id 'id'.

Case-2: If the message is received by a HN: It
reads 'dest' from 'rmsg', if 'dest' equals N(id) it
shows required HN is reached. It read 'id' from
'rmsg' and add it to its array, otherwise it is
forwarded to N(id).

Procedure probe_reply_DST(N(id)) is called
by LN to reply to its HN. The LN creates a reply
message. The 'id' and 'dest' fields of the reply
message is set to be, the 'id' of the LN and the
'id' of the HN respectively. After the reply
message sent to HN, the LN calls the Procedure
probe J'orword_DST(pmsg) to flood the probe
message to all the peers except the peer from
where it was received.

After the completion of these five procedures
the Peer Network will be in required DST
structure.

234

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

Definition 1. Let n{DST^^^^^J be the Number of

Message Pass Required to form DST in the Peer
Network and it can be defined as,

n(DST^^^^J = ((L/PrP^M)H(L/PrN) (1)

In equation (1), 'L/P' gives the number of
DST formed in the network which is also equal to
n(HN). So equation (1) can be rewritten as,

<DST^^^^^J:

(L*M) + (n(ffiV)*7V)

(2)

where,

• 'N' be the number of message pass between
one HN and another HN

• 'M' be the number of message pass between
HN and LN

• 'L' be the number of Peers in the Network

• T' be the number of LN under HN (consider
equal number of LN for all HN)

• '(L/Py be the number of DST in the Peer
Network (or equals n(HN))

• It can be interpretable that 1<M <N <L and
\<P<L

In words. Total number of Message Pass
required to form DST in the Peer Network
n(DST^^gp^^^ ) is equal to sum of products between

number of Peers in the Network and the number
of message pass between HN and LN and between
the number of HNs in the Network and the
number of message pass between one HN and
another HN in the Network.

4. Proposed Efficient Node Search in Large
scale P2P network using DST

P2P networks are an important part of the next
generation of Internet, so how to search the node
in the P2P networks efficiently is the key problem
of the realization of the P2P network. While
studying the search technology of different
mechanisms, this paper proposes a method to
improve the node search in a large scale P2P
network using the systematic DST approach. We
propose a Node search algorithm in fig. 1 which

consists of five procedures Request(),
toAUHeadsO, Found(), Reply() and Receive().

Procedure Request() is invoked by any node in
network which want to identify a node for
communication. This procedure has two
arguments the source node v and destination node
d and creates doFind message with source and
destination details and sent to HN of source.

Procedure toallHeads() is invoked by node v
(should be a HN) when it receives Request
message to propagate the message to all other
HNs in the network using its HN_ARR array.

Procedure Found() is invoked by the HN
which hold the requested destination node d. This
procedure create the foundDest message and set
fields 'Dest' and 'Head2' as id of destination node
d and id of current HN.

Procedure Reply() is called by the Requester
HN to forward the foundDest message to the
source node v.

Procedure Receive() is invoked the node v,
when it receives any message. The received
message should be any of three variants; doFind
message^ toallHeads message and foundDest
message.

If the received message by node v is a doFind
message, any one of the following would be
occurred:

Case-1: If the node v is HN and 'Head' field of
the message is 'id' of the node v, then it is
procedure toAllHeads() is invoked.

Case-2: If case-1 fails, then the node v forwards
the message to the node with id in the 'Head' field
of the message.

If the received message by node v is a toallHeads
message, any one of the following would be
occurred:

Case-1: If the node v is HN and 'Head2' field of
the message is 'id' of the node v, then node v
retrieve the 'd' field of message and matches it
with LN entry in its LN_ARR array. There may
two possible ways.

235

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

Case-la: If 'd' field in the message matches with
LN entry of array LN_ARR in v which means that
the search node d is under the current HN v. Then
node V invoke Found() procedure to intimate the
requester HN.

Case-1: If the node v is HN and 'Headl' field of
the message is 'id' of the node v, then it is
procedure Reply() is invoked.

Case-2: If the node v is LN and 's' field of the

Request(v,d)

Receive(msg,v)

Step 1 : create doFind message, dmsg

Step 1: check if msg = dmsg goto step 2, if

Set Sour field as v.ID

msg = hmsg goto step 5

Set Best field as d.ID

if msg =ftnsg goto step 8

Set Head field as v.Head

Step 2: check if v = HN and v.ID =

Step 2: send(Jm^g,v.Head)

dmsg.Head, goto step 3 else step 4

Step 3: call toAllHeads(v,/;ms^)

Step 4: forward(;7m^g,pmsg.Head)

Found(s,d,HNl,HN2)

Step 1: CYtattfoundDestmosssigQ,finsg

Step 5: check if v = HN and v.ID =

Set Sour field as dmsg.Sour

hmsg.Htadl, goto step 6 else step 10

Set Dest field as dmsg.Dest

Step 6: VlN such that LN € v.LNs, repeat

Set //^aJ7 field as HNl

step?

Set Head2 field as v.ID

Step 7: if hmsg A = LN.ID, goto step 8 else

Step2:send(/m^g,HNl)

step 9

Step 8: call Found(/ims^.s,LN.ID,
hmsg.Headl, v.ID)

toAllHeads(v,dmsg)

Step 1: VhN such that HN £ v.HNs

Step 9: dtlttt pmsg

repeat step 2 and step 3

Step 10: foYwardihmsg, hmsg. Head!)

Step 2: create toallHeads message, hmsg

Step 11: check if v = HN and v.ID =

Set Sour field as dmsg.Sour

/m^g.Headl, goto step 12

Set Dest field as dmsg.Dest

if V = LN and v.ID = fmsg.s, goto

Set Headl field as v.ID

step 13 else step 14

Set //^aJ2 field as HN.ID

Step 12: call Replyifmsg)

Step 3: send (/xm^g,HN.ID)

Step 13: // communication starts
Step 14: forward(/m^g,fmsg.s)

Reply(fmsg)

Step 1: send(fmsg,fmsg.Sour)

Figure 1 . Proposed Node Search algorithm using DST structure

Case- lb: If no matches found in LN_ARR array
of HN V, then it discards the message.

Case-2: If case-1 fails, then the node v forwards
the message to the node with id in the 'Head2'
field of the message.

If the received message by node visa foundDest
message, any one of the following would be
occurred:

message is 'id' of the node v, then it shows that
the destination d is found. The node v retrieves
'Headr, 'Head2' fields of message which it uses
for communication with destination d through
destination HN Head2.

Case-3: If both case-1 and case-2 fails, then the
node V forwards the message to the node with id
in the 's' field of the message.

Thus the Proposed Node search algorithm uses
the details stored by every HNs regarding its LNs.

236

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

This approach makes the node search operation in
a large scale P2P network more efficient and
economic with very minimum number of message
passes and reduced routing table entry in every
node in the network.

Definition 2. Let Pi be a peer in Peer Network.
Then the number of routing table entries required
for node Pi in the DST Peer network to route any
message it receives is n(P.(RT _ entry)) and
given as,

n(P.(RT _ entry)) =
n(HN) + {n(tlN,(LN_ARR))u r(P.)}

(3)

where,

• ' n(P.(RT _ entry)) ' is the number of routing

table entries required for node Pi to route any
message it receives.

• ' n(HN) ' is the number of HNs in the
Network

• 'n(tlN,(LN_ARR))' is the number LNs in
theLN_ARRofHNofPi.

• ' r(P. ) ' is the number of nodes adjacent to Pi.

Thus the number of routing table entries
required for node Pi to route any message it
receives n(P.(RT _ entry)) is sum of number of

HNs in the network and combination (without
duplicate) of number of LNs in the LN_ARR of
HN of Pi and number of nodes adjacent to node Pi.

5. Simulation and Analysis

This section describes the simulation results
obtained during the investigation phases. We used

OMNeT++, is an object-oriented modular discrete
event network simulator. A Peer Network of 100
peers (compl, comp2, comp3...complOO)
interconnected randomly and spread in some
distant geographical location to validate the
proposed technique. The proposed Node search
technique is implemented in the simulated Peer
network and performed various fine grained
analyses. . It is assumed that the medium have
propagation delay of 10 ms.

In our simulation setup of Peer Network with
one hundred systems, niDST^^^^^^^) is nearly

equals 391 messages which take nearly 5.83
seconds for the formation of complete DST (i.e.)
to organize P2P network into a hierarchy of
groups of nodes.

5.1 DST Routing Scheme

In this scheme, to route message from source
to destination we used Static Routing technique.
From the simulation we gathered different criteria
measures and tabulated in Table I which show that
the node search operations performed efficiently
and economically in DST Peer Network than that
of Typical Peer Network. Analyses of large P2P
networks (with more than 100 peers) are
cautiously derived from the simulated network.
On simulation proceeds, it is observed that for the
time period of 10 seconds number of node search
operations performed in Typical Peer network and
DST Peer network are 19 and 32 respectively.
Thus the efficiency of node search operation in
Peer network can be improved nearly 68.9% by
using DST technique.

Table I. Comparison of various criteria measures between Typical and DST Peer Network

S.No

No.

Peers

No.

HNs

Average

Routing

table

entries

(approx.)

Typical Peer Network

DST Peer Network

Efficiency

No. of

nodes

involved

No. of
node
search
operations

Time
taken

(sec)

No. of

nodes

involved

No. of
node
search
operations

Time
taken

(sec)

10^

68.9%

10^

84.6%

10^

130

202

80.5%

10^

350

601

178

89.3%

237

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

Table 11. Comparison of various criteria measures between DST and ACO optimized DST Peer Network

S.No

No.

Peers

No.

HNs

Average
Routing
table entries
(approx.)

DST Peer Network

ACO DST Peer Network

Efficiency

No.

nodes

invol

ved

No. of
node
search
operations

Time
taken

(sec)

No. of

nodes

involved

No. of
node
search
operations

Time
taken

(sec)

10^

24 + 6

37.5%

10^

67 + 23

41.6%

10^

130

202+121

131

42.3%

10'

350

601+321

178

214

20.2%

5.2 DST with Ant Colony Optimized Routing
Scheme

To route message from source to destination
Ant Colony Optimized routing technique is
followed which improve the message pass
efficiency of DST because of dynamically
identified optimal route between every HN and
LN Peers and alternate optimal route between
nodes. Table 2 shows the criteria measure
between DST Peer Network and ACO optimized
DST Peer Network and from the result analysis, it
is evident that the efficiency of node search
technique is improved by ACO optimization. On
simulation proceeds, it is observed that for the
time period of 10 seconds number of node search
operations performed in Typical Peer network and
DST Peer network are 19 and 32 respectively.
Thus the efficiency of node search operation in
Peer network can be improved nearly 37.5% by
using Ant Colony Optimized DST technique.
Analyses of large P2P networks (with more than
100 peers) are cautiously derived from the
simulated network.

From the Table I and II, the following observation
can be derived.

Observation 1.

Average number of routing table entries required
by each peer in DST is more than that of ACO
optimized DST.

Proof. In DST network, every peer maintain
single optimal path for a destination whereas in
ACO technique find more than one optimal path

between a single source and destination to
increase the reliability on channel failures. In
ACO, usually alternate optimal path is maintained
between each Peer and all HNs. the number
routing table entries required by any peer
n(P.(RT _ entry)) given in equation (3) can be
expressed as, for DST,

n(P.(RT_entr^)=
n(HN)-^{n(UmLN _ARR))u t(P,)}

For ACO optimized DST,

(4)

n(P.(RT _ entry)) =
{2 * n(HN)}+ { n(HNi (LN _ ARR)) u r(P, )} (5)

where,

• ' n(HN) ' is the number of HNs in the
network

• 'n(llN.(LN_ARR))' is the number LNs in
theLN_ARRofHNofPi.

• ' r(P. ) ' is the number of nodes adjacent to Pi.

From equation (4) and (5), we can conclude
that the number routing table entries required by
any peer in ACO optimized DST will be more
than that of DST.

Observation 2.

With increase in the size of the P2P network,

efficiency of DST technique in node search also

increases.

238

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

I Typical P2P DDSTP2P

■■'IriTLLl

comp02 compll connpl6 connp22 comp26 comp34 comp35 connp44 connp61

Figure 2. Number node search operations performed by nodes in Typical and DDST P2P

connp02 comiDll coin|Dl6 comp22 comp26 cornp34 comp35 coinp44 comp61

Figure 3 Number node search operations performed by nodes in DDST and ACO DDST P2P

Proof. Number of HNs in the network increases
with number of nodes. Consequently the increase
in the node search operations is distributed among
the HNs. So, large number of node search
operations is performed effectively without any
bottleneck which increases the overall efficiency
of the node search technique using DST. Thus
from Table 1, we can conclude that with increase
in the size of the P2P network, efficiency of DST
technique in node search also increases.

Figure 2 and 3 shows the comparison graph
between the number of node search operations
performed by nodes in P2P network (only few
among the total nodes involved in the search
operation). From which we can conclude that the
efficiency of node search can be improved using
DST technique and ACO technique.

From the simulation analysis, the proposed
node search technique optimizes the traffic in

efficient manner by reducing the message pass
required for search operation in P2P networks.

6. Conclusion

One of the important issues with large
P2P networks is that a node search message may
go through the same node multiple times, which
causes inefficiency of the node search process.
We proposed a mechanism to deal with the
problem by using interconnection structure DST
which will reduce the number of message passes
required for node search in the Peer Network. An
enhanced algorithm ACO is used to optimize DST,
which offered better performance, is also
presented. From the simulation analysis, it is
shown that in the Peer Network with DST, high
performance node search can be achieved than
ordinary Peer Network.

239

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

Reference

[1]. Mengkun Yang Zongming Fei, Assigning
Identifications to Nodes in Unstructured Peer-to-
Peer Networks: A Novel Approach to Improving
Search Efficiency, Global Telecommunications
Conference, 2007.GLOBECOM '07. IEEE.

[2]. Lei Xiao, Liangzhong Shen, ZhongYi Hu, Fei Zhou,
"Improving Search in P2P by Location Identification, "
Environmental Science and Information Application
Technology, International Conference on, vol. I, pp.
519-522, 2009 International Conference on
Environmental Science and Information Application
Technology, 2009.

[3]. Sylvain Dahan, Jean-Marc Nicod and Laurent
Philippe, The Distributed Spanning Tree:A Scalable
Interconnection Topology for Efficient and Equitable
Traversal, International Symposium on Cluster
Computing and the Grid, 2005 IEEE

[4]. S. Dahan, ''Distributed Spanning Tree
Algorithms for Large Scale Traversals, " Proc.
11th Int'l Conf Parallel and Distributed Systems
(ICPADS V5), pp. 453-459, July 2005.

[5]. Runtime Analysis of a Simple Ant Colony Optimization
Algorithm Frank Neumann • CarstenWitt, Springer
Science+Business Media, LLC 2007

[6]. Dorigo, M., Maniezzo, V., Colorni, A.: The ant
system: An autocatalytic optimizing process. Tech.Rep.
91-016 Revised, Politecnico diMilano, Italy (1991)

[7]. M. Dorigo, T StUtzle (2002) The ant colony
optimization metaheuristic: Algorithms,
applications and advances. In F. Glover, G.
Kochenberger (eds) Handbook of Metaheuristic s.
Kluwer Academic Publishers, Norwell, MA, pp 251-
285

[8]. A. Colorni, M. Dorigo, V. Maniezzo (1991)
Distributed optimization by ant colonies, In
Proceedings of EC AL 91 European Conference on
Artificial Life, Elsevier Publishing, Amsterdam,
The Netherlands, pp 134-142

[9]. Sylvain Dahan, Jean-Marc Nicod and Laurent
Philippe, The Distributed Spanning Tree ,IEEE
transactions on parallel and distributed systems,
vol. 20, no. 12, december 2009

[10]. A. Rowstron and P. Druschel, ''Pastry: Scalable,
Decentralized Object Location, and Routing for
Large-Scale Peer-to-Peer Systems, '' Proc.
IFIP/ACM Int'l Conf. Distributed Systems
Platforms, vol. 2218, pp. 329-350, 2001.

240 http://sites.google.com/site/ijcsis/

ISSN 1947-5500

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

Vol.8, No. 3, June 2010

IJCSIS REVIEWERS' LIST

Assist Prof (Dr.) M. Ennre 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. Arul 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 PILANI, 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. 3, June 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 Shukia, 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,

Mr. 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. 3, June 2010

Mr. Santosh K. Pandey, Departnnent 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. AtuI Gonsai, Saurashtra University, Gujarat, India

Mr. Angkoon Phinyomark, Prince of Songkia 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. 3, June 2010

Dr. M. P. Thapliyal, Departnnent of Connputer 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 TIemcen, Algeria

Prof. J atinder Singh, Universal I nstitutiion of Engg. & Tech. CHD, I ndia

Mrs. M. Anandhavalli Gauthaman, Sikkim Manipal Institute 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. 3, June 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. 3, June 2010

Mr. Mohannnnad Abu Naser, Dept. of EEE, lUT, 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,lndia

Dr. Ajay Goel, HI ET , Kaithal, I ndia

Mr. NayakSunil Kashibarao, Bahirji Snnarak 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. Annjad Rehman, University Technology Malaysia, Malaysia

Mr. Rachit Garg, L K College, J alandhar, Punjab

Mr. J . William, M.A.M college of Engineering, Trichy, Tamilnadu,lndia

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

CALL FOR PAPERS
International Journal of Computer Science and Information Security

IJCSIS 2010

ISSN: 1947-5500

http://sites,googlexom/site/iicsis/

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
locaHzation, Trust estabHshment and maintenance, ConfidentiaHty 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, WiringAVireless 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 micro actuators,
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, MobileAVireless 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 ijcsiseditor@ 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 and Information Security June 2010"

See other formats