6-3confrence Report

Proceeding
1st National Conference on Security, Computing, & Communication
May 23-25, 2008
Organized By
Institute of Information Technology
Kohat University of Science & Technology
Kohat, NWFP, Pakistan
Sponsored By
Higher Education Commission
Islamabad, Pakistan
Copyright © 2008
Kohat University of Science & Technology
Kohat, NWFP, Pakistan
Table of Contents
S.No Content Page No
1. Conference Honorary Committee i

2. Conference Steering Committee i
3. Conference Organizing Committee ii
4. Conference Session Chairs iii
5. Conference Founder Message iv
Conference Chapter on Security
1. A Semi-Fragile Watermarking Scheme for Gray Scale Image Authentication Based on Self 1
Embedding Mechanism with Improved Recovery of Original Work
2. What Smart ! About Smart Cards Technology 5
3. A System for Multimedia Ownership Identification Using PIC 9
4. Personal Authentication System Using Biometrics Techniques 13
5. Securing Services in Multi-Agent Systems 20
6. DoS and DDoS Attacks; Analysis and Prevention 22
7. SAKE: Secure Authenticated Key Establishment in Wireless Sensor Networks 28
8. Overview of Spam for VOIP and IMS and Future Direction for Developing Spam Protection 33
Model for Video Conferencing
Conference Chapter on Computing
1. Assessing the Usability of Official Web Portal of Government of Pakistan; A Heuristics 37

Approach
2. Architecture of Reconfigurable Artificial Neural Network Co-Processor 41
3. Towards Terrain Modeling For Robot Navigation Using A Single Camera 44
4. Decrease Training Time of Feedforward Multilayer Perceptrons by Attribute Selection 49
Method
5. Classification in Associated Patterns of Stock Data 52
6. A Framework for Agent-based Earthquake Management System 55
7. Intelligent Information Retrieval 58
8. Decentralized Resource Discovery in User-Oriented Grid 63
9. MST and SFMST based Clustering 68
10. Association Rule Mining Towards Business Solution 72
11. Motion Invariant Multi-Focused Image Fusion for Video Processing 75
12. From Deep Web Up To the Surface Web 78
Conference Chapter on Communication
1. Analysis of VOIP by Using Different Vocoders in IEEE 802.11b Wireless Networks 84

2. Using Mobile Phones for Quizzes in m-Learning Environments 88
3. Load Balancing for Congestion Control in MPLS Networks 91
4. Integration between WLAN and MANET 94
5. Handoff Issues in Cellular Networks 99
6. Location Determination Technique for Locating Wireless Devices in Indoor Environment 104
7. RSSI Based Triangulation Method and Clustering Technique for Locating Wireless Devices in 109
Indoor Environment
8. Noise Cancellation in WLAN and Bluetooth 112
Conference Patron
Prof. Dr. Atta Ur Rehman
Chairman, Higher Education Commission
Conference Co-Patron
Prof. Dr. Lutfullah Kakakhel
Vice Chancellor, KUST
Conference Founder & Principle Organizer

Muhammad Hamad Hassan
Director IIT, KUST
Conference Steering Committee

Prof. Dr. Fida Younus Khattak KUST
Prof. Dr. Salikhov T.K. KUST
Prof. Dr. Shafiq Ur Rehman KUST
Dr. Muneeb Ur Rehman KUST
Dr. Syed Asif Mehmood Gilani GIKI
Dr. Ajmal Khan Bangash GIKI
Dr. Junaid Mughal GIKI
Prof. Dr. Anwar Majeed Mirza NUCES-FAST
Prof. Dr. Farooq Ahmed NUCES-FAST
Prof. Dr. Amjad Shah NUCES-FAST
Dr. Tariq Mahmood Jadoon LUMS
Dr. Sohaib Khan LUMS
Dr. Sharif Ullah Khan NUST
Dr. Hafiz Farooq NUST
Dr. Fauzan Mirza NUST
Dr. Raihan Ur Rasool NUST
Prof. Dr. Jamil Ahmed Iqra University
Prof. Dr. Ismail Shah Iqra University
Prof. Dr. Abid Khan Peshawar University
Dr. Saeed Mahfooz Peshawar University
Dr. Azhar Rauf Peshawar University
Dr. Shah Khisro Peshawar University
Dr. Rashid Khan Gomal University
Dr. Abdul Bais NWFP-UET
Dr. Siraj Ul Islam NWFP-UET
Engr. Ghulam Mubashar NWFP-UET
i
Conference Organizing Committee
Mr. Tariq Saleem Qasuria Registrar, KUST

Mr. Azad Khan Khattak Director Academics, KUST
Maj ® Haider Jahanzeb Director Administration, KUST
Mr. M. Masood Shah Assistant Director Finance, KUST
Mr. M. Hamad Hassan Director IIT, KUST
Mr. Rauf Khan IT Manager, KUST
Mr. Nadeem Ahmed Assistant Professor IIT, KUST
Mr. Sohail Shinwari Lecturer IIT, KUST
Mr. Abdur Rehman Lecturer IIT, KUST
Mr. Amjad Mehmood Lecturer IIT, KUST
Mr. Zeeshan Bangash Lecturer IIT, KUST
Mr. Faisal Baseer Lecturer IIT, KUST
Mr. Qadeem Khan Lecturer IIT, KUST
Mr. Abdul Wahid Lecturer IIT, KUST
Mr. Farhan Shafique Lecturer IIT, KUST
Mr. Asad Habib Lecturer IIT, KUST
Mrs. Saima Tariq Lecturer IIT, KUST
Ms. Rabia Khan Lecturer IIT, KUST
Mr. Zeeshan Iqbal Lecturer IIT, KUST
Mr. Jawad Ashraf Lecturer IIT, KUST
Mr. Ali Zeb Lecturer IIT, KUST
Mr. Athar Ali Qureshi Lecturer IIT, KUST
ii
Session Chairs
Session I Conference Chapter on Security
Chair Dr. Fauzan Mirza

School of Electrical Engineering & Computer Sciences
National University of Science & Technology, Pakistan
Session II Conference Chapter on Computing
Chair Prof. Dr. Mushtaq Ahmed

Faculty of Computer Science & Engineering
Ghulam Ishaq Khan Institute of Engineering Sciences & Technology, Pakistan
Session III Conference Chapter on Communication
Chair Prof. Dr. Salikhov Tagaimurod Khaitovich

Department of Physics
Kohat University of Science & Technology, Pakistan
iii
Conference Founder Message
The Institute of Information Technology (IIT) at Kohat University of Science & Technology (KUST) was
established in 2001, the pioneer teaching department providing various degree programs at undergraduate
and graduate level, professional certifications and IT services.
The IIT is equipped with state of the art IT infrastructure, qualified faculty and hardworking students. Our
aim is to teach the people about the IT and its applications to improve their living standards. Our vision is
to become one of the high tech institutes of Pakistan in providing a band of disciplines in IT and Engineering
at all level with promise of providing quality education to produce future leaders in the domain of Science
& Technology.
The IIT does arrange events that include lecture series, IT trainings, seminars, and workshops. At present
the IIT successfully managed to organize a national level conference at KUST by the title; “1st National
Conference on Security, Computing, & Communication” that was held on May 23-25, 2008.
The key aim of the event was to provide the formal platform to academicians, faculty members,
researchers, IT professionals and research students to share their contributions in the domain of Security,
Computing, & Communication.
In first phase, we sent a conference grant proposal to Higher Education Commission (HEC) office in late Sep
2007 that got sanctioned in Oct 2007 and consequently we started our job.
In second phase, we consulted the eminent Professors working in the domain of Security, Computing, and
Communication at various reputed Universities of Pakistan including GIKI, NUST, LUMS, NUCES-FAST,
Peshawar University, NWFP-UET Peshawar, Iqra University, Gomal University and of course at KUST and
after having their consent we constituted our Conference Steering Committee and we are very much
thankful to them for their extended and timely support in arranging this event at KUST.
In third phase, we finalized our conference toolkit, got it published from the press and sent it to almost all
public and private sector universities and R & D organizations of Pakistan.
In fourth phase, we got about 50+ research papers from all over the country, which were thoroughly
evaluated by the conference referees and finally about 28 papers, were recommended for the oral
presentation in the conference.
In fifth phase, we did start our work towards setting up the conference stage, conference hall, hanging the
conference banners, doing boarding & lodging arrangements to facilitate the participants who had to come
from all over the country to attend this conference either to deliver the key note lecture or to present the
research paper.
We are very much thankful to the Chairman HEC Prof. Dr. Atta Ur Rehman and the HEC officials who are
striving hard to promote the higher education and research culture in Pakistan.
We are heartedly thankful to our Vice Chancellor Prof. Dr. Lutfullah Kakakhel and the Dean of Sciences
Prof. Dr. Fida Younus Khattak for their administrative and moral support and encouragement in arranging
this national level event at KUST in a nice manner.
I am personally thankful to my faculty members, students, technical staff, and university administrative
staff, who remained supportive and helpful during this event.
At last but not the least, I am thankful to my family who remained supportive and patient throughout the
event.
Regards !
Hamad
iv
Proceeding of the First National Conference on Security, Computing, & Communications (1st NC-SCC 2008), KUST, NWFP, Pakistan May 23-25, 2008
Session I Conference Chapter on Security
Chair Dr. Fauzan Mirza

School of Electrical Engineering & Computer Sciences
National University of Science & Technology, Pakistan
A Semi-Fragile Watermarking Scheme for Gray Scale Image Authentication Based

On Self Embedding Mechanism with Improved Recovery of Original Work
Research & Development Cell
Kohat University of Science & Technology, Kohat, NWFP, Pakistan
m.hamad.hassan@gmail.com
Abstract — In this paper a semi-fragile watermarking scheme B. Semi-Fragile Watermarking Techniques

is proposed for gray scale image authentication based on
self-embedding mechanism with improved recovery of the The semi-fragile watermarking techniques are primarily designed to
original work. The host image that is to be watermarked for have some degree of intelligence to differentiate between the
authentication is first divided into blocks each of size 8x8 pixels. legitimate and illegitimate processing. Semi-fragile watermark is
Each block is then transformed into frequency domain using simply a mark that is unaffected by the legitimate processing like
the DCT and the desired DCT coefficients selected in zigzag compression but destroyed by the illegitimate processing like
pattern are subjected for quantization based on the 50% JPEG tampering.
compression standard. The watermark is generated by
encoding the DCT quantized coefficients into binary bits string In this paper a semi-fragile watermarking scheme has been proposed
of desired length and embedded into the selected LSBs of each based on the self embedding mechanism and is able to support the
block. The proposed scheme is semi-fragile in nature and thus legitimate operations like compression upto an acceptable level. In
supports the common legitimate operations like JPEG addition, our scheme is able to recover the original work if the work
compression, filtering and is able to detect the malicious has been affected by the illegitimate operations like feather
tampering like feather cropping and recover the original work. cropping.
The quality of work is quite reasonable both subjectively and
objectively as the PSNR of watermarked image and recovered The host image that is to be authenticated is first divided into blocks
image satisfies the standard threshold. each of size 8x8 pixels and each block is transformed into frequency
domain using DCT (Discrete Cosine Transform). The desired DCT
coefficients are selected in a zigzag pattern using the zigzag
Index Terms — Data Security, Data Authentication,
scanning table Z (see Table I) and are subjected for quantization
Multimedia Signal Processing, Multimedia Communication
using the quantization table Q (see Table II) that corresponds to the
50% quality of JPEG (Joint Photographic Experts Group)
I. INTRODUCTION compression. The watermark is generated for each block by
encoding the DCT quantized coefficients into desired number of bits
Data authentication verifies the originality of work by ensuring that using the bits length table L (see Table III). The watermark payload
the work has not been changed or transformed in any way since it and number of LSBs (Least Significant Bits) of each block used for
was last created. The work can be a text document, an audio clip, an embedding depends on the type of algorithm chosen to develop an
image or a video. To develop a data authentication system there are authentication system. The watermark is then embedded into the
two sister technologies that are in practice; Cryptography and destination block’s selected LSBs determined by the distance vector
Digital Watermarking. The former deploys its hash functions to for random distribution of a watermark in the image. The proposed
verify the integrity of work but it can tell us only; whether the work scheme is semi-fragile in nature therefore it supports common
has been changed or not. However, the latter can even tell us; where legitimate processing like JPEG compression, filtering etc. Also the
the changes have been made in the work and if so it can even recover scheme is able to detect the malicious tampering like feather
the original work by deploying its powerful authentication and cropping in the watermarked work. Our scheme is able to recover
recovery mechanisms. In recent decades, information security the original work without any loss. The quality of work is quite
researchers have shown their interests in deploying the digital reasonable both subjectively and objectively as the PSNR (Peak
watermarking technology in designing data authentication systems Signal to Noise Ratio) of watermarked image and recovered image
and such like applications. For ensuring data authentication there meets the threshold.
are two principle watermarking techniques that are discussed in this
paper. The rest of the paper is structured as: Section II discusses the related
work. Section III explains the Fridrich’s self-embedding scheme
A. Fragile Watermarking Techniques and its implementation results. Section IV explains the
Fragile watermarking techniques are primarily designed and are proposed scheme and its implementation results. Section V
very sensitive for all kind of transformations if applied to the work. demonstrates the authentication and recovery mechanism. Section
Although they cannot differentiate between the legitimate VI derives the conclusion and future directions.
processing and illegitimate processing but still they are applicable in
variety of scenarios where we do want to detect even if a bit is II. RELATED WORK
changed like patient information, legal documents, war plans etc.
Fragile watermark is simply a mark that does not survive if a work is The survey of fundamental authentication schemes based on the
modified in any way. watermarking technology can be found in the paper presented by T.
Liu. et al. [1].
Fragility is important and helpful for authentication purposes. If a
very fragile mark is detected in the work, we infer that the work has An early scheme for image authentication was presented by S.
not been changed since the watermark was last embedded. Walton [2] that uses a key based pseudo-random scanning of an
image to compute the checksum. The check-sum is obtained by
summing the numbers determined by the 7 MSBs (Most Significant
Bits) and taking a remainder with a large integer N. The computed
1
check-sum is then inserted in a binary form in the LSBs of the determined using the following equation.
selected pixels. The method is very fast and on average modifies P = [ 3 / 10 ] * Image Dimension ……… (1)
only half of the pixels by one gray level. The check-sum approach
provides a very high probability of tamper detection but cannot TABLE I
distinguish between an innocent change like brightness adjustment
Z: ZIGZAG SCANNING TABLE
or replacing person’s face modules.
Van Schyndel et al. presented their work in [3] that modifies the
LSBs of the pixels by adding extended m-sequences to the rows of
pixels. The sequence is generated with a linear feedback shift
register. For an NxN image a sequence of length N is randomly
shifted and added to the host image rows. The phase of the sequence
carries the watermark information. A simple cross-correlation
method is used to test the presence of the watermark.
Wolfgang and Delp [4] extended Van Schyndel’s idea and improved
their work by tuning the localization and robustness parameters.
J. Fridrich presented her schemes in the paper [5]-[8] where an

approximation of the image is embedded in the LSBs of the original
image for authentication and recovery of original work. TABLE II
Q: QUANTIZATION TABLE
I. Kostopoulos et al. [9] presented their scheme for color image
authentication based on self-embedding scheme in which a given 16 11 10 16 24 40 51 61
color image is initially transformed from RGB to YCbCr color
space in order to extract the luminance channel that holds 12 12 14 19 26 58 60 55
approximately 75% information of an image. The two LSBs are
14 13 16 24 40 57 69 56
discarded to have reduced image space. The watermark is generated
using special mapping function that swaps the given pixel’s 14 17 22 29 51 87 80 62
intensity value with a new value within the boundaries specified by
the mapping function. The watermark information is then embedded 18 22 37 56 68 109 103 77
in all the three channels, using a secret key in such a way that on
24 35 55 64 81 104 113 92
average one bit of information affects at most the four LSBs of each
channel. 49 64 78 87 103 121 120 101
The present work is based on Fridrich’s self-embedding scheme 72 92 95 98 112 100 103 99
presented in [7] with the objective to recover the original work with
relatively high quality.
TABLE III
L: BITS LENGTH TABLE
III. FRIDRICH’S SELF EMBEDDING SCHEME
7 7 7 5 4 3 2 1
This scheme was primarily presented by Fridrich et al. [7]. The
scheme is illustrated by the following standard algorithm. 7 6 5 5 4 2 1 0
A. Algorithm 6 5 5 4 3 1 0 0
An original image is first divided into blocks each of size 8x8 pixels. 5 5 4 3 1 0 0 0
The following three steps are carried out for each block of an image.
4 4 3 1 0 0 0 0
1) Step 1 Pre-Processing of an Image
3 2 1 0 0 0 0 0
Gray levels of all the blocks are set into the interval [-127, 128] and
the LSBs of all pixels of each block are set to zero. This step helps in 2 1 0 0 0 0 0 0
distinguishing between a tampered block and non-tampered block.
1 0 0 0 0 0 0 0
2) Step 2 Watermark Generation
Each block B of an image is transformed into the frequency domain B. Simulation Results
using DCT. The first 11 DCT coefficients are selected in zigzag The simulation of the above algorithm is developed in the Matlab
pattern using the zigzag scanning table Z (see Table I) and quantized (Matrix Laboratory) Version 7.0 using the IPT (Image Processing
using the quantization table Q (see Table II) that corresponds to the Toolbox) and some other built-in and user-defined functions where
50% JPEG compression standard. desired. The original image is shown in the Fig. 1(a) and the
watermarked image is shown in Fig. 1(b) with PSNR of 51 dB. The
The DCT quantized coefficients of each block are then encoded into tampered image is shown in Fig. 1(c) and tampering detection in
binary form using the bits length table L (see Table III). Coding Fig. 1(d). The recovered work is shown in Fig. 1(e) and zoomed
based on L table ensures that the first 11 DCT quantized coefficients version in Fig.1 (f). However the recovery is not perfect as
of each block are encoded into exactly 64 bits. This 64 bits string highlighted by the ellipse that shows patches in the recovered area.
yields a watermark for each block.
3) Step 3 Watermark Embedding
The 64 bits binary string obtained in Step 2 is then inserted into the
LSBs of the block B + P, where P is the position vector that can be
2
into frequency domain using the DCT and quantized in the same
manner as done before while implementing the Fridrich’s
self-embedding algorithm.
After quantization, the first 3 and 22-36 DCT quantized coefficients

are selected for onward binary encoding using the bits length table L
that yields a watermark payload of upto 190 bits in general. The 190
bits string can be now easily accommodated into 3 LSBs of each
block pixels after 0’s padding to have uniform bits string i.e. upto
(a) Original Image (b) Watermarked Image 192 bits.
(PSNR=51 dB)
With incorporation of more bits, our scheme is now able to survive
under all kind of legitimate operations like JPEG compression;
filtering and can deal with illegitimate operations like feather
cropping, tampering etc. In addition, it is now able to recover the
original work without compromising the quality of work. The
proposed scheme is illustrated by the following standard algorithm.
A. Algorithm
An original image is first divided into blocks each of size 8x8 pixels.
The following three steps are carried out for each block of an image.
(c) Tampered Image (d) Tampering Detection
1) Step 1 Pre-Processing of an Image
Gray levels of each block are set into the interval [-127, 128] but
Image Specification now the 3 LSBs of each block are set to zero to have large
Test Image: Saturn
embedding space and accommodate the maximum information of an
Format: tiff
Dimension: 256x256 image into image itself to have full recovery of original work in
Resolution: Gray Scale authentication and recovery phase.
Tampered Region: Surface
2) Step 2 Watermark Generation
Each block of an image is transformed into the frequency domain
using the DCT and the transformed coefficients are quantized using
the quantization table Q. After quantization, the first 3 and 22-36
(e) Recovered Image DCT quantized coefficients are selected in zigzag pattern for
onward binary encoding using the bits length table L that yields a
watermark payload of upto 190 bits. The 190 bits string can be now
easily accommodated into 3 LSBs of each block pixels after
applying 0’s padding to have uniform bit stream of 192 bits. The
reason for considering more bits is to store the maximum
information about image into the image itself that ensures the full
recovery of original work in the authentication and recovery phase.
3) Step 3 Watermark Embedding

The watermark bits string of 192 bits obtained in Step 2 is now
embedded into the 3 LSBs of each block’s pixels of an image. The
rest of procedure is same as discussed in Step 3 of Fridrich’s
algorithm.
B. Simulation Results
(f) Recovered Image (Zoomed Version)
The simulation of the above algorithm is developed in the Matlab 7
(PSNR=32 dB) using the IPT and other built-in and user-defined functions where
desired. The original image is shown in the Fig. 2(a) and the
Fig. 1 Fridrich’s Self Embedding Scheme Simulation Results
watermarked image is shown in Fig. 2(b) which is quite
imperceptible as the PSNR value of watermarked image is 42 dB.
IV. PROPOSED SCHEME The tampered image is shown in Fig. 2(c) and tampering detection
in Fig. 2(d). The recovery of original work is shown in Fig. 2(e) and
For improved recovery of original work Fridrich [7] proposed an
Fig. 2(f) for zoomed version which is close to that of original image
advanced version of above algorithm that deploys each block for
as the PSNR is about 37 dB.
DCT quantization and the first 3 and 22-36 DCT quantized
coefficients are selected for binary encoding using the bits length
table L that yields a watermark payload of 190 bits on average but V. AUTHENTICATION AND RECOVERY MECHANISM
only 128 bits are embedded into the 2 LSBs of each block’s pixels.
To check the performance of our authentication system, a
In proposed scheme, for improved recovery of original work, we watermarked image is tampered in Adobe’s Photoshop software.
increase the embedding space by using the 3 LSBs to accommodate For instance a block from the watermarked image is cropped (say
the entire information of each block represented by 190 bits string of block B (1)) and replaced with a different image block, the code C
a watermark. Each block of size 8x8 pixels is initially transformed (1) stored at B (1+P) will not match with the code C generated from
3
the content of B (1). In addition, the code stored in the block B (1) In such scenario, the region that has been tampered is marked by
LSB will not match the code generated from its respective source setting the affected pixels values to zero. After the tampered region
block say B (1-P). identification, the next phase is to recover the original work. For that
first we have to determine the source block for the affected block to
extract the watermark bits string for the recovery of original work.
Once the desired bits string is extracted, it is de-quantized using the

same quantization table as deployed in the quantization step of a
watermark generation phase. The de-quantized bits are transformed
using inverse DCT to have spatial values of affected pixels. In this
way the original work is recovered in the form of a reconstructed
image which is very close to original work as shown in Fig. 2(e) and
a zoomed version in Fig. 2(f) respectively.
(a) Original Image (b) Watermarked Image
(PSNR=51 dB) VI. CONCLUSION AND FUTURE DIRECTIONS
In this paper, some of the fundamental image authentication

techniques have been reviewed. A new semi-fragile watermarking
scheme has been proposed for gray scale image authentication that
uses JPEG compression model to create the watermark and embed it
into the image itself, that’s why called self embedding scheme. This
enables us to recover the portions of image that have been cropped,
replaced or severely modified. The increment in the watermark
payload helps us in the recovery of work without losses. The intent
to propose this technique is to maximize the quality of the recovered
(c) Tampered Image (d) Tampering Detection work.
In future we will extend our scheme for color image authentication

Image Specification by choosing a suitable color space to accommodate enough payload
Test Image: Saturn of watermark in each channel and with ability to recover the original
Format: tiff work. The choice of transformation domain also matters to extend
Dimension: 256x256 the scheme capability to support the range of legitimate and
Resolution: Gray Scale
illegitimate operations. For instance watermarking in wavelet
Tampered Region: Surface
domain will enable the proposed scheme to support JPEG2000
compression, filtering and operations like that and would be able to
survive under the attacks like feather cropping, tampering or
modification like that.
(e) Recovered Image REFERENCES
[1] T. Liu and Z.D. Qiu, “The Survey of Digital Watermarking Based
Image Authentication Techniques”, 6th ICIP, pp 1566-1559, 2002.
[2] S. Walton, “Image Authentication for a Slippery New Age”, Dr.
Dobb’s Journal of Software Tools for Professional Programmers, vol.
20, April 1995.
[3] R. G. Van Schyndel, A. Z. Tirkel, and C. F Osborne, “A Digital
Watermark”, Proc. of the IEEE Int. Conf. on Image Processing, vol. 2,
pp. 86–90, Austin, Texas, 1994.
[4] R. B. Wolfgang and E. J. Delp, “A Watermark for Digital Images”,
Proc. IEEE Int. Conf. on Image Processing, vol. 3, pp. 219–222, 1996.
[5] J. Fridrich, “Image Watermarking for Tamper Detection”, Proc.
ICIP 98, Chicago, Oct 1998.
[6] J. Fridrich, “Methods for Detecting Changes in Digital images”,
ISPACS, Melbourne, 1998.
[7] J.Fridrich and M.Goljan, “Protection of Digital Images using Self
Embedding”, Symposium on Content Security and Data Hiding in
Digital Media, Newark, NJ, USA, May 1999.
(f) Recovered Image (Zoomed Version) [8] J. Fridrich, “Methods for Tamper Detection in Digital Images”,
(PSNR=37 dB) Multimedia and Security Workshop at ACM Multimedia, Orlando,
Florida, USA, Oct, 1999.
[9] Kostopoulos, S.A.M.Gilani and A.N.Skodras, “Color Image
Fig. 2 Proposed Scheme Simulation Results Authentication Based on a Self-Embedding Technique” 14th
International Conference on Digital Signal Processing, Santorini,
Greece, 2002.
4
WHAT SMART! ABOUT SMART CARDs technology

to develop the specifications for the use of smart cards in payment

Abstract — Smart cards technology is perhaps the widely
cards. This let the banks to use smart cards systems to forecast the
used technology in defending people and systems against the
reduction in frauds, counterfeits, losts and steels.
front line attacks. Some of the notable features make the smart
cards really smart in terms of size, compactness, storage
The research and development in smart cards technology is in fancy
capacity, and computation power to run the powerful security
till date due to its valuable demand by the industry and
algorithms. The high profile use of smart cards technology for
organizations with billion of users incrementing annually for variet
applications like ID cards and passports, banking, transport,
y of applications ranging from money transits, mass transits,
mobile communication, access to internet, satellite TV, and
telephony, personal identification and authorization, upto adds on
services entitlement and applications like that make it appealing
services and enhancing IT infrastructure [2] - [3].
for the academia, security researchers, and the industry to
invest in this technology. Smart cards technology is relatively
new technologies that already affect and will of course affect the III. ARCHITECTURE AND COMPONENTS
life style of billions of people like the way we shop, use the
telephone, see the doctor, and even enjoy leisure!!! As per ISO standards, the size of standard smart card is 85.6mm
(length) x 53.98mm (width) x 0.76mm (thickness).
Index Terms— Smart cards, Chip cards, RFIDs,
53.98 mm
I. INTRODUCTION
Smart cards technology is one of the latest and smart additions to the
world of information security. Smart card is almost the same as
today’s plastic payment cards in terms of size but it contains
memory chip to store the desired information which cannot be 85.6 mm
accessed other than the real owner and forged or copied by the
attacker. In addition, it contains microprocessor to run a range of Fig. 1 ISO Smart Card Architecture
security algorithms and functions to implement the principles of
security like confidentiality, integrity, authentication, and access The architecture in Figure 1 is just to give an idea to the intended
control among others. All these features make the smart card reader. Infact, the size of the smart card and the components
“smart” and help it in a successful automated electronic transaction (electrical contacts, magnetic stripe etc) it contains vary from
[1]. application to application and on the manufacturer. [4]
A. Central Processing Unit
II. HISTORY
Unlike standard credit, debit and loyalty cards that contains
magnetic stripe, smart card uses microprocessor chip which are
Smart cards technology has roots back in the seventh century when available in 8, 16, and 32 bit architecture. However, none have
inventors in Germany, Japan, and France filed the patents and the multi-threading and other powerful features that are common in
industrialists in the U.S., Japan and Austria, were given privilege to standard computers. Smart card CPUs execute machine instructions
implement these patents. at a speed of approximately 1 MIPS. A coprocessor is often included
to improve the speed of encryption computations.
The chip card was historically invented by the German scientist
Helmut Gröttrup and his colleague Jürgen Dethloff in 1968 but their B. Memory
patent was approved in 1982. Roland Moreno patented his idea of The smart card contains the following type of memories:
the memory card in 1974. Michel Ugon at Honeywell Bull invented
the first microprocessor smart card in 1977. Bull patented the SPOM
(Self Programmable One-chip Microcomputer) that defines the
necessary architecture to auto-program the chip in 1978. Three
years later, the very first "CP8" based on this patent was produced
by Motorola. Today, Bull has almost 1200 patents related to smart
cards. Telecarte, was the first French pay phones company that used
cards for their payments in 1983. When paying with a Carte Bleue,
one inserts the card into the merchant's terminal, then types the PIN,
before the transaction is accepted. Only very limited transactions
(such as paying small autoroute tolls) are accepted without PIN.
Finnish manafactured smart cards by combining the features of
credit cards and debit cards. In Europe since from the mid of 90s,
systems using smart cards technology, used to store information on
the card chip rather on an externally recorded account, so that
machines accepting the card need no network connectivity. In
addition, SIM cards were used by the GSM mobile phone
equipments. In 1993, the international payment authorities
including MasterCard, Visa, and Europay agreed to work together Fig. 2 Smart Card Components
5
1) RAM H. Programming Language

The RAM (Random Access Memory) is of about 1kilobytes. This is Most smart cards are currently programmed in low-level languages
needed for fast computation and response. based on proprietary smart card operating systems. Some of the
2) EEPROM programming has been done in the chip's native instruction set
The EEPROM (Electrically Erasable Programmable ROM) is about (generally Motorola 6805, Intel 8051, or Hitachi H8). In 1998-
1 to 24 kilobytes. Unlike RAM, its contents are not lost when power 2000, a new type of card has shown up, sometimes called a
is. Applications can run off and write to it, but it is very slow and re-configurable card. These have a more robust operating system
one can only read/write to it so many (100 000) times. that permits the addition or deletion of application code after the
3) ROM card is issued. Such cards are generally programmed in Java and are
The ROM (Read Only Memory) is about 8 to 24 kilobytes. The therefore called Java Cards. Other relatively popular languages are
operating system and other basic softwares like encryption Windows for Smart Cards, MEL (Multos programming language)
algorithms are stored here. and Basic.
C. Input/Output Unit
This is via a single I/O port that is controlled by the processor to IV. TYPES
ensure that communications are standardized, in the form of APDUs
(A Protocol Data Unit). There different types of cards that come under the smart cards
D. Interface Devices (IFDs) technology and are discussed briefly in the following paragraphs.
Smart card needs power and a clock signal to execute its operating A. Magnetic Stripe Cards
system and application programs which is provided by the Interface Magnetic stripe card technology is widely used smart cards
Device - usually a Smartcard Reader - in contact with the card. The technology characterized for its low cost and easy processing like
IFD is also helpful in establishing a communication channel read/write operation. This card technology is normally used in range
between application software on the computer and the operating of applications like in banking; debit cards and credit cards are the
system on the card. The mode of smart card communication channel valid examples, in access control systems; entitlement cards and
is half-duplex. Generally, the smartcard reader serves both the tickets are valid examples.
purposes, that is, it allows an application to write to the card as well
as read from the card. The receiver is required to sample the signal Consider the Figure 3, we have a plastic card typical of size
on the serial line at the same rate as the transmitter sends it in order 3.375″x2.125″ and is being used to carry stripe of magnetic tape.
for the correct data to be received. This rate is known as the bit rate Apart from the embedded information in the magnetic stripe, this
or baud rate. Data received by and transmitted from a smart card is plastic card may carry useful information in the form of text or
stored in a buffer in the smart card’s RAM. Due to insufficient images so that people can easily interpret it. The stripe is almost the
RAM, relatively small packets (10 - 100 bytes) of data are moved in same as used in a cassette recorder i.e.
each message. magnetic field controls the alignment of magnetic dipoles into
E. Operating System various orientations along the length of the tape. The capacity of
magnetic stripe to hold information in quite limited but it can be
Microsoft Corporation released a miniaturized version of Windows extended by using multiple tracks along the stripe just like an audio
for smart cards in late 1998. Some versions of a Gnu O/S have also tape-recorder. On each track few bits of information related to
been released. The operating system found on the majority of smart identity can be stored using a method so called Wiegend [5] – [6].
cards implements a standard set of commands ranging from 20 – 30
commands, to which the smart card responds. The reader sends a
command to the smart card, the card executes the command and
returns the result to the reader accordingly and waits for another Magnetic Stripe
command thus operating in master-slave manner.
F. File System Signature Strip
Most operating systems also support a simple file system based on
the ISO 7816 standard. A Smart Card file is actually just a
Embossed Useful Data
contiguous block. Files are organized in a hierarchical tree format.
Once a file is allocated, it cannot be extended and so files must be
created to the maximum size that they are expected to be. Each file Fig. 3 Magnetic Stripe Card Architecture
has a list of which parties are authorized to perform which
operations on it. There are different types of files: linear, cyclic,
The typical examples of magnetic stripe cards are shown in the
transparent, SIM, etc. The usual create, delete, read, write and
following figures.
update file operations can be performed on all of them. Certain other
operations are supported only on particular types of files
G. Software
Smart cards are either Soft-Mask or Hard-Mask, depending on
whether the application is in EEPROM or in ROM. Hard-Mask
cards are more expensive and non-editable. Some
application-specific data with instructions set needs to be stored on
EEPROM. In practice, while programming a smart card, it is tested
on a simulator first for debugging. Banks normally use soft-mask
cards for pilot testing new applications and then to move on to more
customer-resistant hard-mask cards for larger deployments. Hard
masking is relatively expensive w.r.t. time and cost than soft
masking. .
Fig. 4 Author’s Credit Card (Front side)
Courtesy: Askari Bank Limited, Pakistan
6
information or update it until and unless various checks are satisfied

based on cryptographic algorithms.
In general, if the embedded microprocessor offers tamper-resistant

property as offered by other smart cards then theses microprocessor
chip cards fulfills the smart card basic requirements. Due to
tamper-resistant property it is very difficult to forge or copy the
sensitive data and it is rather only accessible in automated electronic
transactions, can store data security and can run range of security
protocols and algorithms [5]-[6].
D. Smart Tokens
Fig. 5 Author’s Credit Card (Back side) A smart token is a personal device that does have all the basic
Courtesy: Askari Bank Limited, Pakistan properties and security measures as that of smart card but it doesn’t
have any formal plastic format or size. The SIM (Subscriber
It is of worth mentioning that they lack the important characteristic Identification Module) used by the mobile service companies is the
that smart cards do have i.e. security of information. The reason simple example of a smart token. Smart tokens go through the same
being that magnetic stripe is a normal piece of audio tape and so it manufacturing process as smart cards but they come in plug-in
can be easily readable and re-writable. Measures have been taken to format as shown in the following figure [5]-[6].
make the plastic card harder to duplicate but merely can be done
about the magnetic stripe used in the automated transactions.
B. Chip Cards
The chip card is the solution to overcome the issues presented in the
case of magnetic stripe cards. The bottle neck was the magnetic
stripe in case of magnetic stripe cards and so is replaced by the
electronic chip in the chip-cards. Historically theses cards were
identical from their golden or silver color contacts. The typical
architecture of chip card is shown in the following figure.
Vcc GND
Fig. 7 Author’s Mobile Phone SIM (Front side)
RST Vpp Courtesy: PMTCL, Pakistan
CLK I/O
RFU RFU
Fig. 6 Chip Card Architecture
The Vcc, GND, RST, CLK and I/O together give basic functionality
and power to the chip card. The Vpp pin is used for re-programming
the chip card just like EEPROM. In past only above six pins were
used by some cards and readers but nowadays industry has planned
to also deploy the RFU and VPP contacts for adds on services. For
instance Mobile phones industry is planning to use these spare pins Fig. 8 Author’s Mobile Phone SIM (Back side)
Courtesy: PMTCL, Pakistan
for high speed USB interface for variety of applications.
A chip card is placed within card reader which physically touches In future mobile service companies may deploy SIM even smaller
the golden contacts, enabled the chip to power on for clocking and than the current size and may incorporate other modes of powering
for communication to offer services to user. and communicating with it.
E. Contact-less Smart Cards and RFIDs
The simplest chip card contains a fixed value and so it is easy for an
There is growing demand and usage for smart cards that do not
attacker to read the value from a valid card and produce a copy of
requires physical contacts but exploit the radios communication
this chip card for later on frauds.
techniques. This special property of smart cards helps the industry
in tagging products and later on tracking. RFID is a special kind of
Another type of card is the memory card keeps the record of
smart card since it meets all the basic requirements of smart card but
telephone call minutes in a count down manner. Such card normally
it is contact-less and communicate its ID with its associated reader
doesn’t need added security and so are easily read and copied. In
via RF i.e. radio frequency. Normally RFIDs do not imply any
addition, the memory may be re-written to disturb the application or
protocol security, prevention against sensitive information cloning
to change user information.
or tamper-resistance but nowadays they do incorporate these basic
security measures [5]-[6].
All these limitations make these chip cards useful for dedicated
systems rather general purpose smart cards [5]-[6].
V. WHY SMART CARDS?
C. Microprocessor Chip Cards
To overcome the limitations of chip cards, the industry shifted to If a portability of record of one or more applications is necessary or
have microprocessor chip cards that not only store the useful desirable, and records are likely to require updating over time,
information but can run powerful security protocols. The benefit of records will interface with more than one automated system,
protocol interface is that it is very impossible to access the security and confidentiality of records is important, then smart cards
7
are a feasible solution for making data processing and transfer more J. Entitlement Cards
efficient and secure. Smart cards are small, handy, light weight, Smart cards are now used in electronic benefits transfer systems to
durable and re-usable (i.e. they allow multiple read/writes). They carry food stamp and WIC food benefits in lieu of paper coupons
use chip operating systems that support multiple applications at the and vouchers. In addition, agricultural producer smart marketing
same time. The inbuilt storage facility equipped with security card to track quotas.
features of a smart card makes it robust against attacks and reduces
it dependency on vulnerable external resources. The sensitive data
VII. CONCLUSION AND FUTURE DIRECTIONS
to be stored in the card can be biometrics information of the card
owner, personal medical history, and cryptographic keys for
authentication. The smart cards technology works on the The important thing about Smart Cards is that they are everyday
international standards; ensures multiple vendor support at objects that people can carry in their pockets, yet they have the
competitive prices and hence established track record in the real capacity to retain and protect critical information stored in
world applications. electronic form. The “smartness” of Smart Cards comes from the
integrated circuit embedded in the plastic card. Embedding similar
VI. APPLICATIONS circuits in other everyday objects, such as key rings, watches,
glasses, rings or earrings, could perform the same electronic
function. The development of contactless card technology was the
Smart cards are nowadays deployed in public and private sector catalyst for what is known as tags. Tags function like contactless
departments that includes personal identification on real time smart cards but are in the form of a coin, a ring or even a baggage
systems or on internet, banking and finance, loyalty, telephony, label. They are generally attached to objects such as gas bottles, cars
healthcare, mass transit, services entitlement, physical access or animals and can hold and protect information concerning that
control, IT access control, and entertainment. object. This allows the object to be managed by an information
A. Finance system without any manual data handling.
Almost all the countries have either reduced or eliminated the coins
The use of Biometrics will soon mean that his/her hand, fingerprint
from the payment systems by incorporating the smart cards. Smart
and the retina of the eye or the sound of the voice can reliably
cards in the form of e-purse have replaced the traditional coins
identify a person. Soon it will be possible to authorize the use of
system even for small purchases in vending machines and at counter
electronic information in Smart Cards by using a spoken word or the
transactions. Securing money transactions using smart cards on
touch of a hand. Also Smart Card readers will be appearing on the
internet form the basis of electronic commerce.
PC and will enable the user to pay for goods purchased over the
B. Telephony Internet. This will be especially useful for small value purchases,
There are unlimited GSM mobile phones that use smart cards that which are not really appropriate for credit card transactions. If you
contain the mobile phone security and subscription information. The have products that have relatively low value - for example a few
handset is personalized to the individual by inserting the card, which pages of information about your product that customers may pay
contains its phone number on the network, billing information, and 50c for - they may well pay you in the future using a Smart Card.
frequently call numbers.
As a smart infrastructure for mobile computing, Smart Card
C. Personal Identification technologies will prove to be the killer application for the networked
All purpose student ID-Card contains variety of applications that is economy. The Smart Card will be "charged up" with money and you
electronic purse for vending machines, laundry machines, library will use it as you do cash or a phone card. In the near future, the
card, and meal card. traditional magnetic strip card will be replaced and integrated
together into a single card by using the multi-application Smart
D. 6.4 Physical Access Control Card, which is known as an electronic purse or wallet in the Smart
Employee access cards with secured ID and the potential to employ Card industry. It will be used to carry a lot of sensitive and critical
biometrics to protect physical access to facilities. data about the consumers ever more than before when compared
with the magnetic strip card.
E. IT Access Control
Employee access cards with secured passwords and the potential to Smart Cards are a relatively new technology that already affects the
employ biometrics to protect access to computer systems. everyday lives of millions of people. This is just the beginning; soon
it will influence the way we shop, see the doctor, use the telephone
F. Healthcare
and even enjoy leisure!!
Various countries with national health care programs have deployed
smart card systems. The Germany and Austria are the one that REFERENCES
deploy this technology to issue smart cards to every person in their
country in the form of consumer health card that contains insurance
eligibility and emergency medical data. [1] www.smartcardforum.org
[2] www.ewh.ieee.org/r10/bombay/news5/SmartCards.htm
G. Entertainment [3] www.wikipedia.org/wiki/Smart_card
Almost every small dish TV satellite receiver uses a smart card as its [4] www.smartcard.co.uk
[5] www.scc.rhul.ac.uk
removable security element and subscription information.
[6] Keith Mayes, Konstantinos Markantonakis “Smart Cards, Tokens,
H. Loyalty Security and Applications” Springer, 2008.
[7] www.cardwerk.com
Smart cards can be used in retail loyalty schemes and corporate staff [8] www.crisptelecom.com
systems. Consumer reward/redemption tracking on a smart loyalty
card, that is marketed to specific consumer profiles and linked to
one or more specific retailers serve that profile set.
I. Transport
Smart cards can be used for driver license, mass transit passes, fare
collection systems, electronic toll collection systems.
8
A System for Multimedia Owner Identification Using PIC

Muhammad Muneer Umar, Muhammad Hamad Hassan
Abstract— A new approach for multimedia owner time. In this way the owner will be able to pick his creation
identification is introduced. In this approach a unique code is anywhere and any time and can prove his ownership. The rest of the
calculated by the system first and then this code is embedded paper is structured as: Section II discusses about some basics
into the multimedia content. We used this code with images. relating to our system. Section III demonstrates our proposed
The code generated is called PIC (Personal Identification code) scheme. Section IV reflects the conclusion and future directions.
and one PIC can be allotted to a single person only. This new
approach is able to identify the exact owner of the multimedia II. BACKGROUND
content by producing the owner’s identification data. The
embedding points in the image are made adaptive according to
the owner’s identity so that it is difficult to detect the In this section we will discuss some basic elements that are
watermarks in the image by the hackers. necessary to understand our system. First of all we will discuss A.
Multimedia, its types and its classifications, then we will briefly
Index Terms — Multimedia, Ownership Identification, discuss B. Multimedia Security C. Digital Rights Management and
Digital Watermarking at the last we will discuss on D. Digital Watermarking in details.
A. Multimedia:
I. INTRODUCTION It is hard to clearly define multimedia in few words. Most of the
authors define multimedia as “It is media that uses multiple forms of
Use of multimedia contents in computer systems is increasing day information content and information processing.[5]” In general we
by day. Now people prefer to use a system showing information in can say that multimedia is the combination of text, still and
the form of graphics, displaying images and playing videos, music animated images, animations, videos and any type of medium where
and sounds. Due to recent growth in the multimedia tools and information can be represented and stored [1]. Multimedia
technologies it is now easy to create multimedia contents like generation can be categorized into two terms i.e. Captured which is
images, videos and music. But on the other hand it has also been taken from real world like images and videos and Synthesize that are
made easy to make pirated copies of multimedia data. Due to the developed through multimedia tools. e.g. animations, digital
availability of powerful multimedia processing tools, illegal graphics and text etc[5]. Multimedia content can be continuous or
distribution of the under subject multimedia is very easily which can be discrete in terms of signal processing. Different formats of
may effect the royalty, ownership and data authentication issues. multimedia can be defined as follows.
These copies are usually distributed through different mediums like
internet and mobile devices like CD's, DVD's, flash disks, cell Text: In early days of computers text alphabets were appeared white
phones and PDA’s. There is no way to stop the piracy except to of green on black background having size of 8 pixels wide and 8
monitor and keep an eye on the pirated contents. The owner can pixels high. Each pixel used single color and was monochrome.
only stop the piracy to claim his ownership if he is able to find some ASCII codes were used to display these letters. Later on Adobe
traces in the illegally distributed multimedia contents [1, 2]. introduced vector based true type and colored text. Now there are
different types of text available in different styles and colors [6].
There were many techniques introduced to stop the illegal Text has many attributes like face, size, color and formatting
distribution. The most common way is through Digital Rights (bold/italic/regular).
Management (DRM) Technology: instructions are embedded in the
multimedia data files which can limit their use [3]. This technology Images: Images are pictures captured from real world or synthesized
has many advantages but limitations on the other side. It is possible by computer. It is said that “A picture worths that thousand words
in this system that the content can’t be copied or modified so most can’t”. Images are more suitable for providing information than
important is that owner does not want to stop people to make one or reading text. Images can be either vector based or raster [6].
two copies but to avoid making of million copies. Videos: Combination of images is called video. Images are
To implement principles of security like confidentiality, integrity combined and showed in a timely fashion. Each image in the video
and authentication, there are various techniques like cryptography is called frame. The number of frames per second is called frame
and digital watermarking. For multimedia contents watermarking is rate. There are different frame rates for videos. For example if we
considered the best way for imposing DRM in the multimedia talk about TV Channels then they have 30-40 frames per second. In
contents. Many organizations use watermarking techniques in case of computer monitor the frame rate is 70 frames per second.
different ways [4]. The minimum frame rate for human vision system is 16 frames per
second. [6].
Our approach is to embed a unique watermark that can easily
identify the owner of a multimedia content. For this purpose a Animations: There is a minor difference in videos and animations
system is introduced which generates a PIC (Personal Identification and in most cases animations are called videos. Animations are
Code) for the owners. PIC is generated from the user’s daily life made by computers and different tools. Most of the common tools
aspects which identify him like name and identity card number. The are Macromedia Flash, Macromedia Director and Swish.
user has to remember a short code but the system will use a long Animations are broadly used in web pages, commercial movies and
code stored in the database. The generated code is then embedded games. [6].
into the multimedia content which is invisible and can’t be detected.
This content is called watermarked multimedia content. The Sound: Sound can be speech, voice, music and audio effects.
watermarked content can be easily used to identify it owner any Usually speech is referred to a living and live sound while voice is
9
considered as the recorded non living sound. Sound can be digital or watermarks are normally useful for copyright protection and are
analog. Sound is represented and stored in single dimension [7]. commonly used in paper currency. Invisible watermarking is the
process in which we put some data hidden into a multimedia
B. Multimedia Security
content. A watermark can be fragile of robust. A fragile watermark
The recent growth of multimedia systems has increased the need for is used to detect slight changes which can be made by anyone to the
the protection of digital media. In multimedia security we normally data. Robust watermarks are used for copy rights protection.
try to emphasis on the following security principles Robustness also leads to survival from all kinds of attacks on the
Confidentiality: By confidentiality we mean to protect the actual data [11]
contents of multimedia data from an unauthorized user. For example
we want to transmit a very sensitive image or video so there should To implement watermarking system we need to follow these steps
be a restriction to avoid use of such data by a third unauthorized 1. Watermark Generation Phase
person. [9]. 2. Embedding Phase
3. Extraction Phase
Integrity: Means that the actual data is not altered by any cracker or
hacker. With the advancement in computer applications now it is
easy to make changes in multimedia data. Now multiple images or
III. PERSONAL IDENTIFICATION CODE (PIC)
videos can be combined and make the new ones.
Authentication: The right person who is allowed to use the data or
its owner is called authenticated. The process of identifying the We use a new approach by putting a maximum of 23 bytes code as a
owner is called authentication. watermark. This code is named as Personal Identification Code.
The code can be embedded into the image or video. This code is
C. Digital Rights Management primarily generated from the name of the owner who wants to
For assuring multimedia security Digital Rights Management embed his ownership identity into the data. The owner has to
(DRM) system is introduced. DRM is defined as "A system which provide his full name and a small code of maximum 6 digits called
has the collective set of technologies and policies used to impose User Code (UCode). This small code is used for the purpose to avoid
rules on how content is being produced, distributed, and consumed". clash in case of same names of the two or more owners. These codes
Without DRM, digital content can be easily copied, altered, and are stored in the database, after the generating process. We define a
distributed to a large number of recipients which can cause revenue set of bits used for each alphabet and digit. The number of bits is 6
and honor loss to media companies by these activities. DRM for each symbol as shown in the table 1 and 2. Space and separator
provides secure delivery over the internet by using protection are also given bits patterns. Separator will be discusses later in this
technology against interception and modification and restriction of paper.
the content usage for the end user so user will not be able to tamper TABLE I
the image or video due to data-resistant mechanism [3]. SHOWING THE LIST OF ALL BIT PATTERNS FOR
ALPHABETS
D. Digital Watermarking
Digital watermarking is the process of embedding some useful Symbol Bits Symbol Bits
information in coded forms into the under subject multimedia for a 000001 n 001110
variety of reasons like copyright protection, copy control, data
b 000010 o 001111
authentication, ownership identification and multimedia indexing
etc. in our system we are focusing on the digital watermarking for c 000011 p 010000
ownership identification [4,10]. d 000100 q 010001
e 000101 r 010010
There is a difference between encryption and watermarking. f 000110 s 010011
Encryption is based on key mechanism while watermarking uses
g 000111 t 010100
marks which are embedded into the image or video. More over
encryption forms capsulation of data while watermarking is the h 001000 u 010101
embedding process. In watermarking process there is less i 001001 v 010110
degradation in the quality of the data due to embedding process. j 001010 w 010111
k 001011 x 011000
l 001100 y 011001
Encryption Watermarking
m 001101 z 011010
TABLE II
. SHOWING THE LIST OF ALL BIT PATTERNS FOR DIGITS
Symbol Bits Symbol Bits
• Key mechanism • Mark mechanism

0 011011 6 100001
• Capsulated Form • Embedded Form

1 011100 7 100010
• High Random Source • Less degradation of
2 011101 8 100011
3 011110 9 100100
Generation multimedia quality
4 011111 Separator 111111
5 100000 Space 000000
Fig. 1 Difference between encryption and watermarking The system can only use small alphabets with blank space and
digits. The system does not understand special symbols as there is
Watermarking can be classified as visible and invisible. Visible no need to specify the special symbols in names or UCodes.
10
The length of the PIC is variable and depends upon the length of the A sequence of bits is calculated for the name and UCode using the
owner’s name and the code length. The maximum number of defined table. Merger is used to merge these two sets of bits. A
characters in the name are allowed to be less than or equal to 20 and Separator is also added in between the two to specify the boundary
greater than 3 and the UCode should be less than or equal to 6 between the two sets of sequences of bits. This merged data and the
characters. The maximum size of a PIC is 23 bytes. The PIC timestamp of entry is stored in the database if UCode is available for
structure is defined in the following table 3. a same name. This code is further merged with the SCode at run
time to build a complete PIC.
Table III
B. PIC Validation Process
Bit Sizes of different fields of PIC
The PIC validation process is same as the above by replacing few
Name Separator UCode Separator SCode steps. The data storage process is omitted and a true return is added
120 bits 6 bits 36 bits 6 bits 16 bits in the system. The system at the end returns a false in case the entry
is not stored in the database and it returns the merged bits and
timestamp of entry in case the system founds the entry stored in the
Special Code (SCode) which is 16 bits is used for key purpose. First
database.
8 bits are used to show the initial column number while the rest are
used to determine the key row of the image. SCode is generated C. Special Code (SCode) Generation
through a function applied on the name and timestamp of entry. SCode is the last segment of PIC consisting of 16 bits. SCode is used
Separator is defined to show the end of the name and the UCode as for key purpose. It is calculated in a systematic way that uses both
we already said that PIC is a Variable code in length. the owner’s name bits and the timestamp of entry. Timestamp is
also converted to bit sequence accordingly. We calculate 4 bits from
We can take two examples of a PIC which are given to ‘muneer’ the timestamp and then rest from the name.
with UCode 304 and ‘hamad’ with same UCode. The PIC for Calculation is made in the following steps
‘muneer’ will be as follows.
1. Divide the month number of the timestamp by 2 and take a ceil
001101 010101 001110 000101 000101 010010 111111 011110 function of the result.
011011 011111 111111 1010101010101010 2. Convert the timestamp digits into bits according to the defined
table.
The PIC for ‘hamad’ will be as under 3. Now take the first bit from the timestamp using the calculated
001000 000001 001101 000001 000100 111111 011110 011011 number by counting from left to right. Second bit is taken by
011111 111111 1010101010101010 continuing counting after the first one. Same calculation is made for
the other 2 bits by scanning timestamp bits from right to let using the
The last 16 bits in both the above strings are generated at run time same calculated number.
and depend upon the stored time stamp with the user data in the
database. so these 16 bits are only typed for understanding purpose. We need to calculate 12 bits from the name bit. This can be done as.
These bits are not actual according to the system.
A. A. PIC Generation Process 1. Divide the number of name bits by 12 and take a floor function.
This number will be used as increment value.
The generation process is started by taking the owner’s name and
2. Take the first bit of the name as 5th bit in the SCode.
any UCode. The system will first of all check the availability in the
3. Count from the first bit using increment value and take the 2nd bit
database. If there is same entry stored in the database then the
from the name bits and so on.
process will be restarted by asking the user to change the UCode.
Process of initial code (PIC without SCode) generation is illustrated D. PIC Embedding Process
in the follow figure 2. Once this initial PIC is stored in the database, the owner needs to
submit his full name along with his UCode at next time for
verification. The system will first of all check the validity. If the
data entered is found valid then the system generates the SCode. A
whole PIC is calculated by merging the SCode bits with name and
UCode bits. Separator bits are also used to identify the boundaries.
The next step is to take the image in which the owner wants to
embed his identity. The first and foremost thing is that the image
should be enough in size to accommodate these 184 bits in its pixels,
one bit for each pixel. we suggest at least 256 x 256 pixels image for
best results. Then the coordinate system of the image is calculated
and the pixels are determined for embedding data accordingly
during embedding process.
In our system we use LSB [3] watermarking scheme. We alter the

last bit of all the selected pixels of the image. The embedding
process will start as
Calculate the jump size. Jump is increment value which is used for
the next location for embedding. This includes both the rows and
columns increment values.
Fig 2 Initial PIC generation process
11
W/ =Floor of (Image.Width/184)
H/= Floor of (Image.Height/184)
Divide SCode into two segments, one for specifying the current row
and the other one for current column.
W= Decimal of first 8 most significant bits of SCode

H = Decimal of first 8 least significant bits of SCode
Initialize a counter variable for locating the bits in the PIC array
J=1
XOR function of first most significant bit (MSB) of PIC with the
first least significant bit (LSB) of the image at location W, H of
image is performed to embed the first bit into the image.
LSB(Image[W,H]) XOR PIC[J]
Now loop is applied which is used for inserting PIC bits into the
image. In this process the current row and column values are
incremented by the jump values W/ and H/. In case the current value
of either row or column is increased than the image size then the Fig. 3 Watermark Detection
current value(s) are set at the backward positions in the image
accordingly. IV. CONCLUSION AND FUTURE WORK
LOOP I =1 to PIC.Length
Increment J by 1 In this paper we presented a new scheme for identification of
Increment W by W/ multimedia ownership. We described the process of watermarking
Increment H by H/ for images but it can be applied on other multimedia types like
videos and sound with minor modifications. The system is found
IF H is greater than Image. Height successful and produce results with less degradation of multimedia
Then H= H – Image. Height images. Because we use an adopting watermark key mechanism
[END IF] therefore it is difficult to identify the exact location and order of
IF W is greater than Image. Width watermarked pixels of the image. Our future work will be to extend
Then W= W – Image. Width our proposed system to sound and videos.
[END IF]
REFERENCES
LSB (Image [W, H]) XOR PIC [ J ]
[END LOOP] [1] http//en.wikipedia.org/wiki/Multimedia [ last visited on 10 April 2008]
[2] http://en.wikipedia.org/wiki/Piracy [ last visited on 12 April 2008]
[3] Bechtold, S. “The Present and Future of Digital Rights Management:
Using the above algorithm the PIC code can be successfully inserted
Musings on Emerging Legal Problems.” In: Digital Rights Management:
into the image. This code is inserted using Exclusive OR (XOR) Technological, Economic, Legal and Political Aspects, E. Becker W.
function. It can be re detected using the same parameters by XOR Buhse, D.Gunnewig, & N. (2003).
function. [4] Richard Barnett. “Digital watermarking: Applications, techniques, and
challenges.“, Electronics and Communication Engineering Journal,
E. Watermark Detection Process 11(4):173–183, August 1999
[5] R. J. Flynn, W. H. Tetzlaff “Multimedia: An Introduction”, IBM Journal
Same procedure is used to detect the ownership of a person. The of Research and Development, archive Volume 42 , Issue 2 (March 1998)
owner has to produce his name with UCode and the watermarked [6] http://www.adobe.com/aboutadobe/adobeventures/ [last visited on 05
image. The full name and the UCode is first of all checked in the January 2008]
database for availability and then calculated to build PIC using the [7] Ze-Nian Li, Mark S Drew, “Fundamentals of Multimedia” , Prentice Hall,
2004
entry timestamp available in the database. [8] P. Heracleous, S. Nakamura, and K. Shikano, “Multiple Sound Sources
Reconition by a Microphone Array-based 3-D N-best Search with
Once the PIC is calculated, then we need to scan the whole image Likelihood Normalization.” , in Proc International Workshop on
Hands-free Speech Communication, 2001, pp 103–107
for the watermarked bit. For this purpose we need to dig out the
pixels where the XOR function is made. The location can be easily [9] Oded Goldreich, “Foundations of Cryptography”, Cambridge University
searched out using the same algorithm of watermark generation with Press, 2004
minor modifications. Once again XOR function is applied with the [10] I.J. Cox, J. Killian, F.T. Leighton and T. Shamoon, “Secure
spread-spectrum watermarking for multimedia”, IEEE Trans. Image
same locations using the calculated PIC. After this process if we get
Processing, vol. 6, no. 12, pp. 1673–1687, Dec. 1997
the image equal to the actual image (non-watermarked) then the [11] http://en.wikipedia.org/wiki/Digital_watermark [last visited on 10 April
owner is identified else the given owner details are fake or wrong. 2008]
The process is show in the following chart. [12] Schyndel, R.G., Tirkel, A.Z., Osbome, C.F., “A Digital Watermark”,
Proceedings of the IEEE International Conference on Image Processing;
Austin, Texas, vol 2, pp. 86–90, 1994.
[13] B. Chen et al, 2001, “Quantization index modulation: A Class of Provably
Good Methods for Digital Watermarking and Information Embedding,”
IEEE Trans. Inform. Theory, vol. 47, pp. 1423–1443, May 2001.
12
Personal Authentication System Using Biometrics Techniques

Muhammad Altaf Khan, Muhammad Hamad Hassan
Abstract — Personal authentication is the process of accuracy, then assessing performance evaluation in a given
verifying an individual who is requesting an access to the system scenario, followed by operational evaluation before full operations
concern followed by pool of resources. Various techniques exist begin. To perform the evaluation of performance of different
for personal authentication system; knowledge-based biometrics techniques there are different quality measures such as
authentication system that uses passwords or PINs for FAR, FRR, FTE and EER.
authenticating users.,Token-based authentication system that
uses tokens or smart cards for authentication and Biometrics II. PERSONAL AUTHENTICATION SYSTEM
authentication system that uses physical and behavioral traits of
human for authentication. In knowledge-based authentication
system, If the user forgets required password or PIN then the Personal authentication refers to authenticating an individual who is
system will not be accessed by him. If the password or PIN is requesting an access to a pool of resources, to use them or update
shared with someone then the system will be insecure. them as desired. The purpose of personal authentication is to ensure
Token-based authentication system is made more secure by that the rendered services are being accessed only by a legitimate
using passwords or PINs in combination with the tokens. In this user. At the present, the personal authentication system is mainly
ever growing and ever-dynamic world old systems of used at close range, such as for laptops, cellular phones, ATMs,
authentication have become outdated. Every individual is in-house room-entry control, access to safes, and systems operation.
unique, while the overall human structure is same. This But future demands something differently as diversified
approach puts biometrics authentication system in a great applications effective for networks to control the access of
demand in the constantly updating field of authentication. information, to provide remote personal identification and
Though this system is still in its early days but in future it will e-commerce via the internet will be shaped. That’s why efforts on
play a critical role in applications used for authentication. Some standardization are being made from a global point of view [3].
quality measures like FAR, FRR, FTE and EER can be used to
A. Standard Authentication Systems
measure the performance level of a biometrics technique used
for personal authentication. Performance level of any There are three standard authentication systems, which are basically
biometrics technique is dependent on the environment in which used to implement principle of authentication.
application is deployed.
1) Knowledge-Based Authentication System
Knowledge-based authentication system is largely employed for
Index Terms — Authentication, Password, PIN, Token and
user authentication. Knowledge-based authentication system
Biometrics
consists of text-based that uses passwords or PINs and
graphics-based authentication that uses graphics for authentication.
I. INTRODUCTION But here we will focus only on text-based authentication. In
knowledge-based authentication system based on Password or PIN,
Personal authentication is a system that ensures that the services are user provides some information to authentication system then the
being provided to the legitimate user. The recent events of terrorism system process this information and suggests that either the user is a
and theft have invited the organizations to safeguard employees and legitimate user or not. Many new measures are introduced to
themselves against the theft of confidential information or property improve the security including inserting passwords’ options. By
[1]. Both the concept of Knowledge-based and Token-based obeying passwords’ options, user will select a password or PIN that
authentication systems are long in used but they are widely will not be predicted by imposters easily. But knowledge-based
fundamental security concept. If a person can prove that who he is authentication is not free from errors i.e. it relies heavily on the
and also knows something that only he could know, it is reasonable precise recall of secret information. If a minor error on the part of
to think that a person is he, who claims to be. However, changing user is committed during authentication process, the system fails to
environment has forced to adopt new system of security as old respond as desired. It would not be easy for the user to remember the
security systems lasted for a very short time. For instance, identity password for a long time or sometimes a user may have so many
theft through forged passports or documents and the skimming of accounts with different passwords. So it would be very difficult to
the information encoded in the magnetic strip on the back of plastic remember all these passwords all the time. To overcome this
cards is now very common. Though some organizations are still problem the user will either write his password somewhere or he
persistent with outdated Knowledge-based and Token-based will share it with someone. By doing so the user has overcome the
security systems but present day life has proved it almost entirely problem but the uncertainty has been increased [4].
insecure. Therefore, scientific community comes up with a
solution-oriented authentication in the form of ‘biometrics 2) Token-Based Authentication System
technique’. It is convenient in use, safe and reliable. The scientific Token-based authentication requires some unique information,
community is seriously putting efforts to improve biometrics designed personally, called token that he possesses during the
techniques to be accepted as alternative to other authentication authentication process. Each user is registered against that unique
techniques. A biometric system is essentially a pattern-recognition token which becomes his Identifying Label of the token in order to
system that recognizes a person based on features derived from a verify its verification. Tokens are named as Storage Token and
specific physiological or behavioral characteristic that the person Dynamic Token. In storage token, the stored information is
possesses [2]. presented to the system for authentication, for instance smart cards
and USB tokens. In Dynamic Token, some responses are stored
The accuracy of a biometric system is ascertained through a series which will be provided to system when the system will ask some
of tests beginning with an assessment of matching algorithm questions. But situation demands necessary changes to ensure
13
unaltered and flawless security for the system. Because in performs the desired processing on that translated digital signals to
token-based authentication, if same information is provided by the improve the signals quality by reducing noise and prepare the data
unauthorized user he will also be authenticated as the systems needs for pattern analysis. The pattern analysis module locates the most
only correct response or correct information to verify someone. In distinctive patterns of the biometrics data. Templates are generated
Today’s age, Token-based authentication is combined with the in extraction module by picking those distinctive patterns found in
knowledge-based authentication to strengthen the security systems. pattern analysis. The pattern matching module compares the newly
Multifactor token is used to prevent impersonation through theft or generated templates with those in the biometrics database, and
loss of token. The most common example of multifactor token is calculates match scores or quality scores for final decision. The
ATM authentication which requires a token (i.e. ATM card) as well decision module decides whether there is a match between the
as secret knowledge (i.e. PIN) for the user’s authentication [4]. matching score and a predetermined threshold. If the match is
found, the decision module performs the necessary action (service
3) Biometrics-based Authentication System granted) [8].
Any physical features; fingerprints, hand, face, iris, retina or voice
of humans can be used for authentication. Biometrics is used for
Data Pattern
both authentication and identification. By and large, biometrics is a Processing Analysis Pattern
Extraction
stronger means of authentication than other authentication method.
Following are the fundamental phases of a biometrics authentication
system [5].
a) Enrollment phase Pattern

Data
Matching
Acquisition
During enrollment phase the raw biometrics data is being captured
by using some sensors.
b) Feature Extraction phase

Biometrics Decision
In this phase, the raw data captured during enrollment is processed (Match?)
Sensor(s)
by using some efficient biometrics algorithms to locate and encode
No
the distinctive characteristics on which the system operates.
c) Template Creation phase

User
Service
A template is “a small digital file derived from the distinctive Raw
GrantedYes
Data
features of a user Biometrics data” [5]. A Template is of size from
250 to 1,000 bytes. Template can occur in two forms i.e.
“Enrollment Template “and” Match Template”. Enrollment Fig.1. Block diagram of Biometrics System [8]
template is generated during the user first interaction with the
system and stored in the enrollment database for future use. Match
Template is obtained when the user provides his data when he wants III. STANDARD BIOMETRICS TECHNIQUES
to be authenticated.
A biometrics is any reliable human characteristic that can be
d) Biometrics Matching measured and used for personal authentication which includes;
During this process the enrollment template is compared with the fingerprints, iris, face, voice, vein identification, retina, etc. But
match template to determine the degree of correlation. The there hardly exists any biometrics module that accurately fulfils all
matching process results in a score that is compared against a the implementations. So, it calls of considerable preventive
threshold. If the score exceeds the threshold, the result is a match; measures while implementing a biometrics module including
otherwise it is considered a mismatch [5]. location, security risks, task (identification or verification),
expected numbers of users, user circumstances, existing data, etc. A
For the last few years, researches have been conducted in order to fact remains that biometrics modalities are in different stages of
strengthen biometrics authentication system in the hope that these maturity e.g. fingerprints in use for centuries while iris recognition
will provide more secure alternatives. This research has been hardly a decade old. Maturity is an indicator proving which modules
triggered by the discovery of many discrepancies in fits well. Biometrics modules used for personal authentication can
Knowledge-based and Token-based authentication systems. The use iometrics modules used for personal authentication can be divided
of Biometrics for authentication or verification is often regarded as a into two categories
revolutionary concept, coming straight out of modern thoughts of A. Physiological Biometrics
science.
Physiological Biometrics is inborn human characteristics that define
physical features. It includes:
A few glimpses into past reveal that physical or behavioral
characteristics for identification existed even in ancient times [6]. In
1) Fingerprint Print
our modern world Biometrics applications are employed in large
A fingerprint is the pattern of ridges and so many very small points
scale identification systems, where they offer two important
called minutia available on the surface of a fingerprint, the
benefits: Fraud detection and Fraud Deterrence [7]. Biometrics is
information of which is determined during the first seven months of
being used in many locations to enhance the security and
fetal development. Fingerprints of individual twins are different and
convenience of the society. Example: Deployments within the
so are the prints on each finger of the same person.
United States Government include the FBIs IAFIS, the US-VISIT
Fingerprints, a century old personal authentication, was first time
program, the Transportation Workers Identification Credentials
used in 1888 for identification process by an Argentinean Police
(TWIC) program, and the Registered Traveler (RT) program etc.
officer, Jaun Vucentich. Personal authentication systems based on
Fingerprints are proved to be more accurate for verification system
In Fig.1, initially the biometrics sensor recognizes biometrics data
and small-to-medium scale identification system involving a few
from the user and translates into digital signals by the data
hundred users [9]. Sir Edward, who classified fingerprints,
acquisition module afterwards. Data processing module then
14
published a book “Classification and Use of Fingerprints”. Macro ̇ Fingerprints technique can also be employed in applications used
and Micro are two features of fingerprints identification. for controlling borders of the country.
̇ A national ID card, planned by the U.K government, will be
Macro-features of the Fingerprints can be seen unaided by the based on fingerprint and iris data [1].
human eye .It includes Ridge Patterns, Pattern Types, Core Point
Delta Point, Type Lines and Ridge Count. Micro-features of the 2) Face Print
Fingerprints include the features which can not been seen directly The face of the person is considered to be the most immediate and
by the human eye. It includes Ridge Endings, Ridge Bifurcation, transparent biometrics modality for physical personal authentication
Ridge Divergence, Dot or Island, Short Ridges, Orientation, Spatial application. 2D and 3D images of a face provides bases for the facial
Frequency and Curvature [4]. recognition scan for personal authentication. Facial recognition
features offer significant images that are different from other
Strengths individuals.
̇ It is a century old authentication security system and has
delivered high accuracy level. Strengths
̇ Its flexible nature makes it acceptable for a wide range of ̇ It can be easily integrated in an environment that already uses
environment. image acquisition equipment.
̇ Devices used in Fingerprints authentication system are very ̇ It can also be used to search against static images for example
user-friendly [7]. photographs used on driver license [7].
̇ It doesn’t always require the user’s cooperation to obtain the
Weaknesses necessary data. It is relatively economic module to implement
̇ In some cases hardware weaknesses and physiological reasons than the other modules like Iris, Retina etc.
prevent enrollment of some small number of users [7]
̇ Performance of the system based on fingerprints degrades over Weaknesses
time because Fingerprints changes due to many reasons like aging ̇ While taking snapshots of the face, many other variables are
and wear or tear [7]. captured which constitutes an implementation challenge due to
̇ According to T. Matsumoto, a Japanese mathematician which measurements like FAR and FRR of the system are greatly
Fingerprints device can be made fool easily by artifacts gelatin affected.
fingers [10]. ̇ Privacy of the user is at risk.
̇ Another problem with the facial recognition scan is that with the
Physiological Biometrics passage of time FAR increases. Because the facial expression of
the user changes significantly with the passage of time. The rate
of error in algorithm is 43% to 50% in one year [11]-[12].
Applications
Facial recognition scan technique is being used in many places for
authentication and/or verification. It includes banks, health care
centers, access controls and law enforcements.
Fingerprint Face Iris
Hand Scan Retina 3) Iris Print
Iris, only internal organ, can be utilized as a biometrics technique,
which does not undergo for a longer period of time. It is the best
biometrics technique when imaging is done at the distance less than
one millimeter. Especially in a case where false acceptance rate is
considered to be very low while searching a very large database. Its
DNA great mathematical advantage help avoiding many problems as iris
pattern varies widely among different persons. As a planar object its
image is relatively insensitive to angle of illumination, and changes
in viewing angle cause only affine transformations; even the
non-affine pattern distortion caused by papillary dilation is readily
reversible. Due to its unique annular shape eyes can be easily
localized in a face [13]. These features enable iris scan to be more
Hand Vein reliable and precise. The uniqueness of iris, discovered in 1987, was
Behavioral Biometrics later confirmed by two ophthalmologists that no two irises were the
Signature Gait same.
Strengths
̇ Iris recognition is the most accurate means of identification that’s
why it will be widely used for high-security identification by
governments and border controls.
Voice Scan ̇ Its user acceptance is very high due to its convenient use and
non-disturbing nature.
̇ Iris-scanning identification is a fast process.
Fig.2. Standard Biometrics Techniques [4] ̇ Active participation of people being identified not required.
Applications
̇ Fingerprints technique is being used by many consumer Weaknesses
electronic devices, personal digital assistant (PDA) and Personal ̇ Image acquisition process requires some proprietary devices and
Computers (PC). accurate positioning of capturing devices. So special training is
̇ The law enforcement agencies are using Automated Finger required.
Identification System (AFIS) for implementing security.
15
̇ Sometimes eye-based authentication represents a discomfort for to appear as a significant step towards real-time processing of
the users [5]. DNA samples.
Applications ̇ Thoughtful future predictions indicate that it will emerge as a
̇ In May 2004, the UK immigration service started project IRIS, major component of future personal authentication systems used
installing iris recognition systems in the UK’s five major airports. for criminal justice system [17]
The system will be used to help process visa holders through
immigration [1]. Weaknesses
̇ In 2002, an iris recognition system was implemented in the infant ̇ DNA matching is not done in real time, and currently not all
unit of the Hospital of Bad Reichenhall in Bavaria, Germany. stages of comparison are automated.
Individuals requiring permission to enter the baby unit, including ̇ DNA matching does not employ templates or feature extraction,
doctors, nurses and parents, are authorized using the system. but rather represents the comparison of actual samples
̇ By the end of 2005, the iris recognition patent holder, Iridian
Technologies, planned to introduce a chip that could be Applications
incorporated into laptops, PDAs and other handheld devices. The ̇ In the UK, forensic scientists tend to use DNA to link
chip would be low cost and of small size. When used in a person to a crime scene sample reliably.
conjunction with a small digital camera, the chip would provide a ̇ The UK National DNA Database, established in 1995, is an
basic iris recognition control system. Experts predicted that these important tool in the fight against crime and fear of crime.
chips would bring down prices and make iris recognition an
affordable identifier for handheld devices. 7) Vein Print
This technique is still experiencing development phase and has not
4) Hand Print attained maturity but its unmatched, high accuracy has made it an
Hand scan authentication works on extraction on hand pattern that integral part in commercial products like VeinID. It is non-intrusive
includes parameters such as length of fingers, width, thickness, to the user.
curvatures, or relative location [15]. CCD camera is the device that
B. Behavioral Biometrics
captures all these information and infrared illumination. Ordinary
and somewhat needless details are neglected during capturing above Behavioral biometrics includes the following traits of human:
features. Inter-finger pegs or locator pins are used to force the
specific hand positioning. 1) Voice Print
Vocal tracts, mouth, nasal cavity and lips constitute the shape and
Strengths science of the appendages of an individual Voice. Age factor,
̇ It includes Small template size of hand scan makes it very useful medical condition and emotional state are some of the reasons that
for personal authentication system involving very small target change voice considerably. Voice, A non distinctive feature, may
samples (such as 500 users). not be trusted for large scale identification.
̇ This technique is simple, easily used and cost effective.
̇ Its market share, according to 2004 survey, is above 11%. A Voice authentication system could either be a Text-dependent or
̇ From the user perspective, it is considered to be non-intrusive. Text-independent. A text-dependent voice authentication system is
̇ Physiological characteristics involved in this technique are based on the statement of a fixed predetermined phrase. A
considered to be more stable then other biometrics. text-independent voice authentication system recognizes the
speaker. In a Text-independent speaker authentication system we
Weaknesses don’t need to provide the password. The words that are spoken by
̇ According to Organization and Technology Research Group the user will only be processed.
(OTR) this technique is of limited accuracy and the ergonomic
design limits usage by certain population [1]. While in a text-dependent speaker authentication system password
is required. Text-independent speaker authentication does not see
Applications what the speaker speaks. Therefore a text-independent system is
̇ Right from 1993, above 18000 employees are authenticated by more difficult to design than a text-dependent system. As for as
this technique in San Francisco’s security is concerned, text independent system is more secure
international airport [1]. against frauds than text-dependent speaker authentication system
̇ It has been implemented for a frequent traveler programme at Tel [14].
Aviv’s Ben Gurion airport and for the US immigration and
Weaknesses
̇ Voice recording is affected by a number of factors; background
Naturalization Service Passenger Accelerated Service System
(INSPASS) programme which operates at nine US airports [16].
noise is one of them. Voice capturing devices are absolutely
5) Retina Print critical for gathering accurate data. Though voice authentication
This technique, used for Identification /Authentication, works on system best suits in phone-based applications but microphones in
communication deteriorate the quality of voice signals.
̇ Due to the large size of Voice templates it is not suitable for many
the distinctive features of the human retina. This technique, least
used in the field of Biometrics, restricted merely to government and
military facilities. Its difficult use and users’ discomfort have made applications [14].
it unpopular irrespective of its high level of accuracy. [5].
Applications
̇ In the mid-1990s, most of the voice recognition systems used
6) DNA Print
DNA (Deoxyribonucleic acid, the molecule that encodes genetic
software to recognize person’s voice and the words spoken.
̇ In late-1990s this technique was used in call centers for
information) is a true biometrics technique. It is relatively a new
technique that relies on the analysis of DNA sequences for
verification of speaker.
̇ Interactive voice response (IVR) systems predicted by the
identification and authentication.
Strengths
̇ Very soon requirements needed for the DNA sampling will Organization and Technology Research (OTR) Group in 1999
become affordable and small in size. Mobile DNA labs are likely [1]. IVR systems used with a touch-tone keypad to provide
automated self-service offerings to customers. It is mostly
adopted by call centers.
16
2) Keystroke (Typing Pattern) Print Where “N” represents total population of users and “n “represents
Typing pattern of a person, assisted by traditional password scheme, the users, presenting their biometrics for false authentication. The
is used for authentication and identification. It does not require any importance of the FAR is the strength of the matching algorithm. As
additional equipment as all the data entered through keyboard. But the algorithm is stronger, the chances of false authentication will be
all the discrepancies of password-based system have been inherited less.
by the system. [5].
Failure to Enroll (FTE)
3) Gait Recognition The FTE is defined as the probability that a user attempting to
It is a new biometrics concept presented for personal authentication. biometrically enroll himself/herself will be unable to do so [4]. The
It is based on the analysis of the “rhythmic patterns associated with FTE is normally defined by minimum of three attempts. This is
walking stride” [5] of a user. It is intrusive therefore It would be justified by the Rule of Three. The Rule of Three in this case
dislike by the user because it violates the privacy concerns. provides us with a confidence level for a given error rate for our
FTE. It also assume that each attempt to enroll is independent,
4) Electronic Signature Print identically distributed and that the used population is significantly
The user’s hand written signatures are used in this technique for large enough. Thus FTE can be calculated as
Event' (n) = Number of Unsuccessful Enrollment Events

authentication and identification of a user. So it is non-invasive to
the user and unlike other Biometrics Technologies it can be changed
by the user. Event (n) = Total Number of Enrollment Events
FTE (n) = Event' (n)/ Event (n)..................(A)
for n = 1
(7)
IV. QUALITY MEASURES FOR BIOMETRICS TECHNIQUE
FTE = Event ' / Event.................................( B )
(8)
∑ FTE (n) − − − − − − − − − −(3(9))

Reliable performance and recognition are the ultimate supreme
If we generalize the formula, we get:
measures of the utility of the biometrics systems for a specific
FTE = 1/N ×
N
application. Recognition rate is generally ascertained by the
n =1
following terms
The strength of FTE is the amount of coverage for the population
False Rejection Rate (FRR)
that the biometrics system has. The More coverage the biometrics
FRR also known as Type-1 error. It is the probability that a user’s
system has, the less likely that a user will experience an FTE.
verification template is not matched with his enrollment template
[4]. So it is likelihood of a valid user not being recognized. It is
calculated as Equal Error Rate EER
The EER is defined as the crossover point on a graph that has both
MatchUser' (n) = Number Of Match User Unsucessful Authentication
the FAR and FRR curves plotted. The EER can also be calculated
MatchUser(n) = Number of Match User' s Attempts to Authenticate
from a Receiver Operating Characteristic (ROC) curve, which plots
FRR(n) = MatchUser' (n) / MatchUser (n).................( A(1)

FAR against choice of using the crossover point of the FRR / FAR
for n = 1
) or using a ROC is a question of significance. An EER calculated
using the FRR and FAR is susceptible to manipulation based on the
FRR = Match User / Match User ..............................(B)
(2)
granularity of threshold value. A ROC based EER is not affected by
such manipulations because the FRR and FAR are graphed together.
∑
If we generalize the formula, we get
Thus, the EER calculated using a ROC is less dependent on scaling.
= 1 / N × ( n ) − − − − − − − ( 1(3)
N
FRR FRR )
n =1 To calculate the EER using the FRR/FAR crossover, the following
is done: for any given threshold value from 0 to 1, respective FAR
Where “N” represents total population of users and “n” represents and FRR are calculated and plotted on the same graph. Since the
the users, presenting their biometrics for authentication. The granularity used for the values between 0 and 1 can be selected, this
strength of the FRR is the robustness of the algorithm. As the introduces the possibility of manipulating the results. The EER will
matching algorithm will be more accurate probability of false occur at the point where the two lines of FRR and FAR will cross the
rejection will be less. each other.
False Acceptance Rate (FAR)

FAR also known as Type-II error is the probability that a system
will match a user’s verification template with the enrollment
template of other user” [4]. It can be understood as the likelihood of
an imposter being recognized as an authenticated user. In general it
is the most critical security metric, as it is imperative in most
er'(n)= Numberof ImpostersSuccessfulfalseAuthentica

application to keep the imposters out. It is calculated as
NonMatchUs tion
NonMatchUs er(n)= Numberof impostersAttemptstofalselyAuthentica ted
FAR (n) = NonMatchUser' (n)/ NonMatchUser (n)............(A)
(4)
for n = 1
Fig. 3 ROC-curve showing FRR Vs FAR [18]
FAR = NonMatchUser' / NonMatchUser...........................(

The strength of the EER is that it gives a comparison of different
E)
(5) biometrics systems. Procedure for using the ROC-curve to judge
performance in a particular application is as follows. Choose an
∑ FAR(n) − − − − − − − − − − − − − − − − −(6)
If we generalize the formula, we get acceptable level of FAR. On Figure shown below a dashed line is
FAR = 1 / N ×
N
shown at 0.1% FAR. The FRR corresponding to this choice is the
( 2) attainable FRR, in this example about 1.6% [18].
n =1
17
Modification to the parameters of the system, especially the to be much, but in a company with 500 employees, it means five of
threshold value, allow to lower either FRR or FAR but them won’t be able to enroll an alternative authentication method
unfortunately not both at the same time as research conducted by the would have to be provided, raising cost and potential security
BWG has shown that there is inverse relationship between the two problems. It will be interesting to see whether improved acquisition
criteria [14]. This is why many available systems tolerate enormous devices and algorithms will be able to overcome this problem in the
FRR just to keep the FAR as low as possible. The identifying near future.
method based on pattern information must find a compromise
between FAR and FRR. Another test conducted in 2001 by a UK-Based National Physics
Laboratory, which is a national standards laboratory in
Test Results measurement and materials science. It tested various biometrics
The Biometrics World Group (BWG) has conducted a large-scale systems. Six biometrics systems were evaluated: face, fingerprint,
comparative test [14]. Over a period of 3 months, a volunteer crew hand geometry, iris, vein and voice. These were tested with 200
of slightly over 200 participants has used authentication systems volunteers over a three-month period. The objectives were to show
based on 7 different biometrics technologies in a normal office the level of performance attainable [1].
environment. The test was back in late 2000, but it is still widely
referenced in the biometrics community. Except for retina scanning The test results, shown in Fig.4, demonstrate that iris recognition is
(difficult acquisition process, limits it mere military use) and DNA the most accurate biometrics method as it had:
matching (isn’t fully automated and in experimental stages), all
physiological biometrics mentioned above have been tested.
10
In the following, we present the most important results. The data is
cited from the final Test report [15]. For more detailed results and an
9
exact description of test scenario and methodology, please refer to
the report.
TABLE I 8
%
FAILURE-TO-ENROLL RATE [15]
7 Failure to Enrole
System Failure-to-Enroll Rate Rate
6 Failure to Acquire
Face 0.0% Rate
5 False Rejection
Fingerprint-chip 1.0% Rate
Fingerprint – Optical 2.0% 4 False Acceptance

Rate
Hand 0.0% 3
Iris 0.5%
2
Vein 0.0%
1
Voice 0.0%
0
Iris Hand Finger Finger Facial
TABLE.2 Optical Chip Recog
FAR VS FRR [15]
Fig.4. Biometrics Systems False Error Rates [1]
Courtesy: UK-National Physics Laboratory
FAR FAR FAR FAR
̇ Zero FAR in over 2 million comparisons (2,735,529
System
0.001% 0.01% 0.1% 1.0%
comparisons),
̇ A 0.0% FTE rate. The failure-to-acquire rate is the proportion of
FRR FRR FRR
Face --
40% 30% 15%
attempts for which a biometrics system is unable to capture an
Fingerprint-chi FRR FRR FRR FRR
̇ A FRR of 1.8%, the lowest of all the technologies tested, in
image of sufficient quality from a person,
p 2.7% 2.3% 2.1% 1.7%
comparison, four of the
Fingerprint – FRR FRR FRR
--
Optical 16% 12% 10%
V. CONCLUSION AND FUTURE DIRECTIONS
FRR FRR FRR FRR
Hand
13% 9.0% 1.2% 0.25% Personal authentication has become a dire need for organization not
Iris FRR 0.25% only to cope with security threats but also to develop policies,
FRR FRR FRR FRR procedures and mechanisms in order to provide administrative,
Vein physical and logical security. Whatever the authentication
13% 13% 12% 10%
FRR FRR FRR technique is employed, the main objective is to gain increasing level
Voice -- of assurance of valid users and to abandon the imposters to access
12% 4.5% 0.5%
We can conclude from the above-mentioned Table.1 and Table.2 the system. Knowledge-based authentication system is very
that iris scanning and capacitive fingerprint systems achieve good effective for a number of authentication applications but since
overall performance. Unfortunately, those technologies appear to passwords are liable to be forgotten or stolen that reduces their
have the highest failure-to-enroll rates. One percent does not seem credibility. Tokens may be used for personal authentication. The
18
level of security is increased in this technique by adding REFERENCES

password/PIN with tokens called multi-factor token. Again
problems related to password/PIN are inherited in token-based [1] http://www.otr-group.com/Biometrics Authentication Devices.pdf,
systems. Proved by experiments, strong biometrics authentication OTR Group members, Version 1.0, 2005.
systems ensure reliable and considerably high level of security in [2] Salil Prabhakar , Sharath Pankanti, Anil k. Jain , “Biometrics
comparison to any other technique being used for authentication. Recognition: Security and Privacy Concerns”, IEEE Security &
Privacy, March 2003.
Environment is the key-factor for the importance and strength of
[3] Seto Y, “Activities for Standardizing Biometrics Technologies”, In
biometrics technique. Therefore, this misconception must be Proceedings of symposium, IEICE’s ,Tokyo, May 2002.
completely eradicated that biometrics techniques produce the same [4] Paul Rovage, “Biometrics and Network Security”, Pearson Education,
level of performance and accuracy in all environments and 2004.
application. On the whole, fingerprints and Iris for personal [5] Uwe Bubeck, Dina Sanchez,” Biometrics Authentication :
authentication systems are convincing their users for their high Technology and Evaluation”, Term Project CS574, 2003.
degree of conviction. The performance of any biometrics technique [6] Ashbourn, Julian, “Biometrics: Advanced Identity Verification”,
can easily be measured by the quality measures i.e. FAR, FRR, FER Journal of Springer, London, 2000.
[7] Nunavuti, Samir et al,”Biometrics: Identity Verification in a
and EER. As the probability of FAR decreases the probability of
Networked World”,Wiley Computer Publishing, New York, 2002.
FRR will be increased. According to the research done by BWG, [8] Yingzi , Eliza.Du, “Multimodal Biometric System Using Low-cost
there is an inverse relationship between FRR and FAR. Therefore Sensors”, MURI Mentors Project Proposal Form for Fall, 2006.
the personal authentication system used for a higher secure [9] Anil K. Jain, Arum Ross, Salil Prabhakar, “An Introduction to
environment tolerates on FRR by keeping FAR as low as possible. Biometrics Recognition “, IEEE, 2004.
[10] Matsumoto, Tsutomu, “Importance of Open Discussion on Adversarial
The fact cannot be overlooked that maturity level and use of other Analyses for Mobile Security Technologies”, Yokohama National
authentication system is greater than the biometrics system. But the University, 2002.
coming events predict that soon individuals and organizations will [11] A Pentland and T. Choudhury, “Face recognition for smart
environments,” Computer, vol. 33, no. 2, pp. 50–55, 2000.
be significantly relying on the improved biometrics acquisition
[12] P. J. Phillips, H. Moon, S. A. Rizvi, and P. J. Rauss, “The FERET
algorithms and techniques and the availability of industry standards, evaluation methodology for face-recognition algorithms,” IEEE Trans.
which will guaranty the bright future of biometrics. Biometrics Pattern Anal. Machine Intel, vol. 22, no. 10, pp. 1090–1104, 2000.
provides a good trade-off between security and user convenience, so [13] John Daugman,”How Iris Recognition Works”. IEEE Transactions on
future personal authentication systems will be the mixture of Circuits and Systems for Video Technology, Vol. 14, No. 1, January
different technologies like token and biometrics for instance 2004.
Bio-smart cards. [14] Mansfield, T. et al, "Biometrics Product Testing Final Report", UK
Biometrics Working Group, 2001.
[15] A.K. Jain, R. Bolle, and S. Pankanti (Eds.), “Biometrics:
Personal Identification in Networked Society”, Norwell, MA: Kluwer,
1999.
[16] http://www.jrc.cec.eu.int.
[17] J.Elliott,“Biometrics roadmap for police applications”
[18] Lawrence O’Gorman, Veridicom Inc. “An Overview of
Fingerprints Verification Technology”. Elsevier Science
Ltd, 1998.
19
Securing Services in Multi-Agent Systems

Zeeshan Iqbal 1, Amjad Mehmood 1, Muhammad Hammad Hassan 1, Abdul Ghafoor 2,
1 Research & Development Cell
2
NUST Institute of Information Technology
Chaklala Scheme III, Rawalpindi, Pakistan
zn zeeshan@yahoo.com, amjadiit kust@yahoo.com, m.hamad.hassan@gmail.com, abdul.ghafoor@niit.edu.pk
Abstract — Agent oriented software engineering is the software and can be tolerate able in agent architecture.
rapid development area in distributed computing. Securing Heterogeneous applications and platform implementation adds the
agent platform is the limelight research area. Several aspects of motive behind having standard specification for security by FIPA.
security have been the major focus of research community e.g.
attacks from one agent to another, from agent to platform, from Security concern can be categorized as direct and indirect as follows
platform to agent. Less focus has been given to securing the [4]:
services provided by one agent and securing them at Directory
Facilitator (DF). A. Direct Security Concern
Managing security at service level will be the motivating factor Lack of authentication: an agent can masquerade as another agent
in this research issue. Providing security at service level will
B. Indirect Security Concerns
•
enhance interoperability regarding platform independence.
Direct & indirect security concern to agent will be addressed. Authorization problems/lack of access control:
This paper propose an authentication & authorization o a masquerading agent is authorized to change
approach in which some form of public key infrastructure another agent’s published service description
(PKI) & operating system concept of sand box will be utilized o any agent can also read any service provider agent’s
for direct & indirect security concern respectively. entry in the DF.
Index Terms — Distributed Computing, Multi-Agent III. DESIGN & IMPLEMENTATION

System, Security, Authentication, Interaction Protocol
FIPA model helps in moving towards interoperability and
I. INTRODUCTION globalization of agent industry. Platform to platform
communication is the major consideration in proposing all the
specification. High level security will not only limit the agent
Distributed systems are relatively autonomous building blocks
interaction with other agent especially the external platform
which work together to achieve some overall objective. This notion
communication will be reduced, which will be a break FIPA dream
of autonomy and synchronization are the basic of any distributed
about interoperability. It is indispensable to deploy such mechanism
systems. Multi-agent Systems (MAS) can be thought of as
that will be supported towards FIPA abstract architecture and boost
distributed systems, composing of number of software agents, to
agent interaction.
work together in achieving some common goal that otherwise might
be in not reach of individual agent.
One of the main threats in MAS is the lack of accountability, which
help malicious agent to personate himself as some other agent. Two
Implementation and operation of secure services running on a
security choices has been proposed i.e. first one deals with
multi-agent system might leads to many challenges. Simple rule at
providing authentication mechanism through DF, second will
security will be to mark out the authentication & authorization
describe the authorization mechanism in DF.
mechanism will be detach, keeping in view that both are
inter-related to each other e.g. authentication sometimes need
authorization list to validate the request [1]. The fact that security is A. Authentication via DF
the most vital and concerning issue is due to that no consideration Main issue revolving behind agent grooming involves is that there is
has been done in providing no security requirement in the abstract no one accountable in case of any misbehavior. Software can be
architecture and various component details. Only little work has pram, deployers of the MAS software can be thought of as
been carried out to integrate security concerns into an dishonest. This all contribute in hesitant usage of MAS frequency.
agent-oriented methodology [7]. To overcome this issue of spoofing we have to make agent
accountable and make agent some kind of unique id for
All the necessary definition has been provided by FIPA neglecting identification purpose. This all means that we have to authenticate
the security either for the platform developers or for the security agent before they can be allowed any request.
experts.
Several mechanisms have been adopted by research community for
II. BACKGROUND AND RELATED WORK authentication purpose varying from Kerberos to some kind of
digital signature.
There are several reasons why the security had been the most ignore
characteristics in the MAS domain. Security is usually not taken as a Our architecture shows some similar kind of key based
central concern in most implementations either because systems are authentication encompassing various algorithms for privacy &
usually deployed in controlled environments, or because they take integrity purpose (Fig. 1). Whenever new agent come into the
the one-to-one tutoring approach, not requiring strict security platform and wishes to communicate, a pair of key and content will
measures [2]. be generated. This content is the ACL message an agent wishes to
Security is very complex and secure systems can only be developed communicate. Key generation process will be some kind of
by security experts and not by agent system developers [3]. Many centralized scheme. That encrypted key and content will be
think that security should be provided as part of the application processed through MD5 algorithm and this whole message with
20
Fig. 1: Underlying Mechanism for authentication of agents involving different algorithms for privacy & integrity
new header will be transmitted to DF. DF will parse this receiving V. ACKNOWLEDGMENTS
message and here the validity of the message content will be
The author gratefully acknowledges his teacher Abdul Ghafoor for
analyzed. If secure communication is performed that message will
his continuous technical support and encouragements. Thank you.
be break down into again a pair of key and content. The DF will
perform the requested action by the agent. The key with that content REFERENCES
will be stored along with that content for future accountability and
making that agent liable for his action.
[1] Somchart Fugkeaw, Piyawit Manpanpanich, Sekpon Juntapremjitt,
Multi-Application Authentication based on Multi-Agent System,
B. Authorization via DF Proceedings of IAENG International Journal of Computer Science,
In order for the agent to be more robust and interactive with the 33:2, IJCS_33_2_6
environment in moving towards the interoperability between agents, [2] Carine G. Webber, Maria de Fátima W.P.Lima, Marcos E.Casa,
they should be deployed in a secure sand box [6] [8] environment. In Alexandre M.Ribeiro, Towards Secure e-Learning Applications: a
this, an untrusted agent is run under the control of a supervisor agent Multiagent Platform, JOURNAL OF SOFTWARE, VOL. 2, NO. 1,
i.e. DF that traps and examines its entire request. The supervisor FEBRUARY 2007
agent may accept or reject the attempted action according to some
[3] Poslad, S., Calisti, M., Towards improved trust and security in FIPA
security policy, the user agent must provide a mandatory access agent platforms. Proceedings of Autonomous Agents 2000 Workshop
control (MAC) list that specifies the object that the untrusted on Deception, Fraud and Trust in Agent Societies, Spain, 2000
program is allowed/ not allowed to access. For all this policy to be
managed we have to provide agent with some form of authorization [4] FIPA Security Work Group Request For Information (RFI),
mechanism, which will be done in case of no authentication or some FOUNDATION FOR INTELLIGENT PHYSICAL AGENTS
top level authentication for the agent to enter into the MAS [5] Zeeshan Iqbal, Amjad Mehmood, Abdul Ghafoor, H. Farooq Ahmed,
platform. Awais Shibli, “Authenticated Service Interaction Protocol for
Multi-Agent System” HONET 2007, Dubai, UAE. 18-20 November
These permissions will be granted by supervisor agent, in this case 2007.
DF. Providing security mechanism will help us in not only [6] http://www.kernelthread.com/publications/security/sandboxing.html
preserving the abstract architecture but we can also negotiate visit on 22 March 2008
security at interaction layer [5]. This gives us the advantage of not
[7] H. Mouratidis, P. Giorgini, and G. Manson. Modelling Secure
forcing security on all type of communication.
Multiagent Systems, in the Proceedings of the 2nd International Joint
Conference on Autonomous Agents and Multiagent Systems,
IV. CONCLUSION & FUTURE WORK Melbourne-Australia, July 2003
Our proposed solution outlined an extensive framework for agent [8] H. Mouratidis, M. Weiss and P. Giorgini. Modelling Secure Systems
authentication and authorization provision. DF will be enhanced in Using an Agent-Oriented Approach and Security Patterns,
order to address the direct & indirect threat security threat evading. International Journal of Software Engineering and Knowledge
Still there are many loopholes in the security requirement to attract Engineering, World Scientific (accepted for publication - in press)
major development in the field of MAS. Simplification and trust
creation among agent and various components of MAS will provide
the guideline for our future work.
21
DoS and DDoS Attacks; Analysis and Prevention

Shah Mahmood1, G. Mubashar Hassan1, M. Hamad Hassan3
1
Department of Computer Systems Engineering
N-W.F.P University of Engineering & Technology, Peshawar
2
Research & Development Cell,
Abstract — DoS and DDoS attacks are used to prevent the These show that attackers can easily increase the attack strength
legitimate users to access the available services on a computer when necessary, so the measured attack magnitudes have more to do
network by overwhelming available resources required for with what the attacker feels is required than with the maximum
service access. Preventing DoS and DDoS is a challenging task, amount that he or she can generate [4]. According to the 2004
as it exploits vulnerabilities within the existing Internet Computer Crime and Security survey report by CSI/FBI, DoS and
architecture. DDoS attacks are the most expensive attacks for the victim
In this paper, several forms of DoS and DDoS attacks and the organizations, as shown in Figure 1.
current prevention mechanisms are analyzed. At the end a
multi-layered distributed source end DoS and DDoS prevention A Denial-of-Service (DoS) attack is characterized by “an explicit
model is proposed. attempt by an attacker to prevent legitimate users from using
service(s)” [6]. Distributed Denial of Service (DDoS) attack is an
Index Terms — DoS, DDoS, Egress, Ingress amplification of DoS attack by launching it from the simultaneous
collaboration of many compromised computers called zombies that
are scattered all over the Internet.
I. INTRODUCTION
There are many ways to perform DoS attacks. Some examples are
Dependence of society on computers in business, education, exploiting a flaw in application software, or a protocol like TCP or
governance, national security and almost all walks of life is UDP running on the remote system, or by sending a huge number of
increasing each day. It is not enough to prevent the computer packets that affect the bandwidth, processing usage or the memory
systems and networks from Viruses, Worms, Trojan horses, Logic of the remote attacked-system. These attacks consume the resources
bombs, Trap doors etc, but in addition it must also be tried to design required for service access, thus denying that service to legitimate
a new mechanism that deals with the alarming threat from the users.
Denial of Service (DoS) and Distributed Denial of Service (DDoS)
attacks as shown by statistics in Figure 1. DoS and DDoS attacks In DDoS attacks, attackers use many insecure systems on internet as
can put governments out of contact with the outside world, sources for launching Distributed attacks. In the case of Figure 2,
businesses out of the market, and provide attackers a chance to the attacker is only directing the zombie systems to launch the
launch attacks against anyone for fun or ransom or for political attacks thus preserving its bandwidth and other resources while
reasons. making the attack much deadlier. These zombie computers might
have weak processors with limited bandwidth but their collective
Mostly, attackers attack for fun sake such as the targeted attacks on effect is stronger than any single super machine.
several well known websites including Yahoo and Google [5] in DDoS attacks are more dangerous than the DoS attacks. Firstly,
June 2004. There were also instances when DoS and DDoS attacks because the identity of the attacker can remain completely hidden
were for political motives such as; US military and NASA security even if, thousands of zombie systems are identified. Secondly,
was breached by the British hacker Gary McKinnon in 2001 and because it is very easy to make systems get compromised, even with
2003 [2]; more than 600 Danish websites were under DDoS attacks the tools freely available on many hacking websites and the
due to the blasphemous cartoons during 2005/06 [1]. In some cases, underground communities. Once the attacker has launched the
such as the Al-Jazeera attack in 2003, the attackers increased the attack, he or she can then disconnect, leaving the zombies or the
volume of attack as the defenders added capacity to handle traffic. agents to carry on the attack.
The DoS and DDoS attacks analyzed in section 2 of this paper are
TCP SYN Flooding, ICMP Ping Flooding, ICMP Smurf attacks and
UDP Flooding attacks. Since these are the common DoS and DDoS
attacks used for denying a service on internet, therefore, the effects
of these attacks were analyzed on the victims’ machine by writing
application programs for launching the attacks using POSIX API.
All these four attacks exploit vulnerability in a protocol of TCP/IP
protocol stack.
Earlier researches proposed several research models for countering

DoS and DDoS attacks. The four famous models are analyzed in
section 3 of this paper. Finally, based on the studies of these
research models, a DoS and DDoS attack prevention model that
employs a layered architecture on edge routers is proposed which
removes the shortcomings in the earlier models discussed.
Fig. 1 DDoS Cost for Victim Organizations [7]
22
TABLE I
NETWORK STATISTICS FOR DISTRIBUTED TCP SYN
FLOODING
C. ICMP Ping Attacks:

These attacks also exploit a vulnerability in ICMP protocol [23],
Fig. 2 DDoS Attacks [8]
used for carrying error and control messages back to the sender of an
IP packet. The attacker floods the victim with ICMP
II. ANALYSIS OF SOME DOS AND DDOS ATTACKS ECHO_REQUEST packets; having a spoofed source-IP. The victim
responds with ICMP ECHO_REPLY with the spoofed IP address as
A. Test Environment: the destination. The continuous flooding of ICMP
In this section four DoS and DDoS attacks are analyzed in a home ECHO_REQUEST packets fills up the IP buffer thus leaving no
network of three systems. All the three systems had processors of space for legitimate IP packets. When one system attacked the
1.6 GHz and memories of 256MB each. Two of the systems running victim with its entire processing power, the victim’s 52% processing
Linux were used to attack, a third system running windows 2000. power was used handling the attack as shown in Figure 5.
The firewalls were enabled; all the software and the operating
systems were fully patched and updated. The attacks were coded The results show a big difference compared to the TCP SYN
using C language with the help of POSIX API. attacks. This is because TCP is a connection oriented protocol and
the state is maintained, besides, it needs to be processed up till the
B. TCP SYN Attacks: transport layer. On the other hand the ICMP is a Network-layer
TCP SYN attack is caused as a result of the architectural flaw in the protocol, thus requires less processing. When a distributed ICMP
TCP connection establishment. TCP, as described in [22], follows a Ping attack was launched the victims’ processor usage went up to
three way handshaking mechanism for connection establishment as 93% as shown in Figure 6.
follows:
1. System A --------- (SYN) ------å System B
2. System A ä--- (SYN + ACK) ----System B
3. System A ------ (ACK) ----------åSystem B
Fig. 5 ICMP Ping Attack on the victim by one attacker
Now if System A spoofs its IP and keeps on sending TCP packets
with the SYN flag set, then System B will keep on reserving space
for incoming packets in the backlog queue (till the limit exceeds)
and will be sending SYN + ACK to some other system which will
simply drop the packets if alive. The SYN flooding will cause a
severe processing loss for the victim computer. When one attacker Fig. 6 Distributed ICMP Ping Attack against the victim
was attacking the victim using its full processing power, the victim
had to use 75% of its processing power to cope with the attack as
shown in Figure 3. Table 2 shows the ethereal network statistics on the victim machine
under a distributed ICMP Ping attack by two attackers.
When the attack was stopped, the victim’s processor usage dropped
down to 0%. Afterwards, when the attack was made distributed and TABLE II
two systems were attacking the victim, the victim’s processor usage
NETWORK STATISTICS FOR DISTRIBUTED ICMP PING
rose up to 99% as shown in Figure 4
ATTACK
Fig. 3 TCP SYN attack on victim by one attacker
D. ICMP Smurf Attacks:

ICMP Smurf [24] attacks are reflexive attacks in which the attacker
sends many broadcast ICMP ECHO_REQUEST packets with the
source IP spoofed with the IP of the victim. The destination address
Fig. 4 Distributed TCP SYN attack on the victim in the ICMP ECHO_REQUEST is a directed broadcast address that
is sent to all the nodes on the IP subnet, causing all nodes to reply
Table 1 shows the ethereal network statistics on the victim machine with ICMP ECHO_REPLY packets addressed to the victim. Thus,
under a distributed TCP-SYN attack by two attackers. Smurf attack sends a moderate amount of traffic and causes a virtual
explosion of traffic at the attacked subnet as shown in Figure 7.
23
When an ICMP Smurf attack was launched on a network of three

systems, only 27% of the victim’s processing power was used as
shown in Figure 8. If it was a network of 500 systems the results
would certainly have been quite different.
The table 3 shows the ethereal network statistics on the victim

machine under an ICMP Smurf attack by two attackers.
E. UDP Flooding Attacks:
UDP [25] is a connectionless protocol unreliable transport protocol Fig.7 ICMP Smurf Attack [9]
used by time-sensitive applications. In UDP-flooding, the attacker
uses forged packets to try and connect the chargen-UDP service to
the echo-UDP service at another site. UDP-flooding [10] overflows
the UDP receive-buffer leaving no space for legitimate UDP
datagrams.
Fig. 8 ICMP Smurf attack in a three system network
When the single attacker was attacking using 4 % of its processing TABLE III
power the victim was using 25% of its processing power as shown in
Figure 9. NETWORK STATISTICS FOR ICMP SMURF ATTACKS
When a distributed UDP Flooding attack was launched against the

victim, it used 43% of its processing power as shown in Figure 10.
Table 4 shows the ethereal network statistics for UDP Flooding on

the victim machine.
F. Combined effect of all four attacks:
A distributed attack comprising of all the four types of attacks was
launched and the results were amazing. The victims 100%
processing was used due to the attack, as shown in Figure 11.
Table 5 shows the network statistics, captured at the victim machine

for the Distributed combined flooding attack by two attackers.
Fig. 9 UDP Flooding attack on the victim by a single attacker
III. PREVENTION MECHANISMS
Several prevention mechanisms have been used and are proposed

for DoS and DDoS attacks. Some of them will be discussed in this
section with their respective advantages and shortcomings.
Fig. 10 Distributed UDP Flooding attack
A. Ingress and Egress Filtering:
Ingress filtering explained in [10] is used to make sure that incoming
packets are actually from the networks that they claim to be from, TABLE IV
otherwise they are discarded, hence providing a defense against
source IP spoofing. Ingress filtering also helps in ensuring that NETWORK STATISTICS FOR DISTRIBUTED UDP
traffic is traceable to its correct source network. It is better explained FLOODING
in Figure 12.
In this example attacker resides within network 9.0.0.0/8 and

connects to internet through ISP D. An input traffic filter on ingress
input link of router 2 restricts traffic to allow only traffic originating
from the source address within the 9.0.0.0/8 prefix.
As Ingress filtering reduces the success of spoofing the source

address, but it can not stop an attacker from using a spoofed source
address of another host within the allowed range of the filter. Thus,
even in the case of a spoofing attack, the administrator can be sure
that the attack is occurring from within a specified range, and in the
worst case can block all the addresses in the range [10].
The main problem with Ingress filtering is that it can not help Fig. 11 DDoS attack comprising of all four attacks
prevent DoS attacks launched using non-forged IP address. Another
problem is that the network needs to know which IP addresses it is The concept of Egress filtering is to allow only those packets to
connected to, but it is not always possible, because if a network has leave your network that contain an IP from your internal network as
a single connection to the internet then it has no way of knowing the source IP [12]. If all routers at the border of edge-networks are
whether this address is spoofed or not i.e. a router at the configured with Egress filtering then no spoofed denial of service
edge-network configured with the default route has no way of attack will be launched from any network, thus the cyber world
knowing the networks it is connected to.
24
would be much more secure. Like Ingress filtering, Egress filtering • The server is not allowed to initiate a connection with the
also cannot protect against non-forged DoS and DDoS attacks. client.
B. TCP SYN Prevention Mechanisms: • If two clients need to communicate it should be of the
A detailed account of defense against TCP SYN attacks is given in form, clientåserveråclient; provided that in the second
[11]. part the client has opened the connection with the server.
1) Increasing the size of Backlog Queue:
One suggested prevention mechanism is to increase the backlog According to the paper, the worms and other malicious software that
queue (a large memory structure used to handle incoming packets are a strong tool to create zombies for distributed DoS attacks spread
with the SYN flag set until the moment the three-way handshake from clients into clients, thus with the server in between, the
process is completed). This way the system would be able to malicious software cannot be transferred from client to client. The
manage more half-open connections. The problem with increasing malicious software has to be platform independent as normally the
the backlog queue is that its size remains finite no matter how large client and the server do not use the same platform. The
it is, and this finite size would eventually get filled with illegitimate Morris-worm infected only the UNIX systems [16] (being platform
half-open TCP connections. dependent); so it would not be able to spread in this architecture as
the systems at the server-end are mostly UNIX systems.
TABLE V
NETWORK STATISTICS OF COMBINED ATTACKS It is also suggested to use non global addresses for clients so that
clients can not be used for launching DoS attacks on servers. A
client address can be given a local significance and still be made
unique across entire Internet by using Path-based Addressing, as
shown in Figure 13.
If a path based address of a client is known to a malicious server it

can not be used by other servers because this address can not be used
at any other server to send packets back to the client. Such path
based addressing will make complete source-address spoofing
impossible. All reflexive attacks against remote clients are
prevented. Many routing DoS-attacks on client systems, such as
announcing bogus routes, are prevented as client routes are not
announced.
Fig. 12 Ingress Filtering

2) Decreasing the connection timeout value:
Another suggested mechanism is to decrease the total time of
handling the connection request in the backlog queue. Although this
will help us in getting rid of the attack packets soon, but in the case
of congestion in the network the legitimate packets will also be lost.
3) SYN Cookies: Fig. 13 Path Based Addressing [26]
The concept of SYN Cookies is discussed in [13]. In this technique Although this architecture helps prevent transfer of worms from
the backlog queue is not used. When System A receives a TCP SYN computer to computer, it does not support peer to peer applications,
packet from System B, it does not maintain any state for the packet, which have become very popular for sharing files directly between
instead its sends a SYN-cookie back to System B. This SYN-cookie users on internet. It also affects use of internet telephony, because it
acts as a SYN + ACK in the normal three way connection also resembles peer to peer architectures. One other disadvantage is
establishment. Though the SYN-cookies mechanism has the that path based client address can change when either the client
advantage of not using the backlog queue, but it has some moves to another network or inter domain routing path between
limitations as well. The server is limited to using only 8 MSS values client and server changes. This change of path-based client address
[14]. As the server discards the backlog queue it must reject all TCP requires additional identification mechanism at the transport layer to
options such as large windows [14]. enable the server to identify the client even when client’s address
has changed.
C. Research Models Proposed:
2) Preventing Internet Denial of Service with Capabilities:
In this section four research models about the prevention of DoS and The second model is proposed in [18]. This model builds on the
DDoS attacks are discussed. architecture proposed in SOS [27] and Mayday [2], the problem
1) DoS Resistant Internet Architecture: with these two approaches is that authenticator assigned to the
The first model suggested in [15] tries to make transfer of worms verified transmission by overlay nodes is compromised; a DoS
from computer to computer impossible, thus making it impossible to attack can still be launched. This approach also uses a nonce token
turn a computer into a zombie which would make the internet in every packet as a lightweight authenticator, but these tokens are
architecture DoS resistant. The model makes the following transient and limited in scope to a single source destination path,
assumptions: thus limiting the damage that can occur after the token is discovered
• The use of separate IP-addresses for the clients and the by the attacker.
servers. This model is based on assigning capabilities to sending nodes in the
•
form of tokens; the intended source acquires a capability by sending
Some systems though very limited can act both as clients the request through a hierarchy of Request to Send (RTS) servers
and as servers. along the path to each participating destination.
• The clients are not allowed to communicate with the A å RTSåRTSå…….. åRTSå System B
other clients.
25
In the above mechanism System-A is the source which wants to In layer 1, Egress filtering is performed in order to stop IP spoofing
communicate with System-B. To make this possible, System-A will and detect the attacker. The spoofed packet’s corresponding MAC
send RTS packet to the first RTS server on path to the System-B. address is temporarily blocked and the notification is sent to the
This is relayed along the chain of RTS servers to the destination administrator so that its zombie state can be cleared. There is a
RTS server. Eventually an RTS server will reach its destination and problem with the Egress filtering; it can not identify the spoofing
the destination then must decide whether to allow the source further within the allowed range. Besides, it can not stop any non-spoofed
packets. If the destination decides to allow the source to send DDoS attacks. Thus, we embedded two additional layers.
packets, it assigns capabilities by calculating a chain of K 64-bit one
way hash values h1, ………, hk. These hi values are capabilities; As TCP SYN flooding attacks are the most common DDoS attacks,
possession of each allows a client to send a limited number of n we designed a special layer to prevent them. It is the second layer
packets in the next t seconds. Each RTS server in the reverse path called the TCP SYN Defender.
from destination to source remembers the hi values and associates It will be using TCP Proxying to defend against SYN attacks. A
them with the flow in the Verification Point (VP) coupled to the request for TCP connection establishment would not be forwarded
RTS server. to the destination server instead, SYN + ACK segment will be sent
by the border router, and only if TCP three way handshake
Each packet sent by the source is labeled with the capability and the completes the edge router would initiate the connection
associated sequence value. When each VP along the path receives establishment with the destination server. If the client does not send
the packet it checks the capability, sequence number and the flow the ACK to the border router the MAC address will be temporarily
identifier. If the capability is found and the parameters match the blocked till the zombie state is cleared.
packet, it is forwarded and the use of field capability count is
increased. If the count has reached a threshold value of n or the If an attack passes undetected through layer-2 then it would be
amount of time for which the capability is in use has expired, the detected at layer-3. The third layer is a hybrid Intrusion Detection
capability is flushed. If the match is not found then the token is System. It maintains the flow for each Source-IP, Destination-IP,
deemed invalid and is discarded. Source-Port and Destination-Port and analyzes the traffic flow
across it. Signature- based attack detection uses signatures which
This approach proposes to change the internet architecture by are updated dynamically for a particular pattern in traffic that has
adding extra functionality in the core i.e. RTS servers and VPs. One been characterized as a known exploit or vulnerability. If a match is
other aspect that is not considered is the way by which servers found the packet is dropped and the MAC is temporarily blocked
would differentiate legitimate client requests from illegitimate ones. and the administrator is notified. The anomaly-based attack
Also, this model is not appropriate for real time services because of detection checks for a particular pattern in traffic which is identified
the increased overhead of acquiring capabilities and checking as normal traffic. If a match is not found then the traffic is declared
tokens in each packets against parameters of a capability. as anomalous [26], the MAC is temporarily blocked and the
3) Denying Denial-of-Service: A Router Based Solution: administrator is notified. These signatures and normal traffic
This model [19] proposes up gradation of selected routers with patterns will be dynamically updated by the service provider. This
capabilities of encryption, signing, verifying, and dropping the will create a certain problem for new software’s with new traffic
packets they route. It proposes a change in selected group of routers patterns. Hence, there must be some international Software Security
in the core and it does not affect the protocol stacks at the end Authority which certifies the software. After certification, the IDS
systems, which means the applications do not need to be changed or will update its signatures and prove it to be normal. As most of the
reconfigured. viruses, worms are intentionally embedded into some application
In this scheme all the hardened routers, as shown in Figure 14, in software; this certification will put a pause on them. The trapdoors
the path from source to destination sign the packets. When the last will also be identified.
hardened router gets the packets it strips off all signatures and
decrypts packets so that destination can recognize them. The
signature provides source and route authentication and adding some
information gathering and traffic control in hardened routers will
allow us to provide attack resilience from DDoS attacks. The
signature based routing between hardened routers provides the
benefits of privacy, DoS prevention, and DDoS prevention.
The problem with this model is that it burdens the core with extra
functionalities due to which generality of core suffers. It also affects
the time constrained delivery of multimedia traffic.
Fig. 15 Source Based Three Layer DoS and DDoS Prevention Model
Fig. 14 Hardened Access Routers [19] As the architecture of the model is multi layer, it can be distributed
amongst several systems on the network edge, thus helping in
D. Source Based Three Layer DoS and DDoS Prevention Model: balancing the load and increasing the processing efficiency. The
In this section after research, a new model, “Distributed model provides direct benefit to the victim still an organization shall
Source-end Three-Layer DoS and DDoS Prevention Model”, shown implement this due to the following reasons:
in Figure 15, is proposed. This model aims to remove the • This will protect the organization from getting involved in
shortcomings identified in the models analyzed. It builds on the some crime; intentionally or by innocently.
ideas presented in DWARD [20], [21]. This also is a source based
mechanism implemented at the border router of the edge-networks.
26
• Stopping the attack packets at the gateway will preserve [3] J D. Anderson, Mayday: Distributed Filtering for Internet Services. In
the organizations external bandwidth; thus used for some Proc. of USITS 2003.
[4] J. Mirkovic, S. Dietrich, D. Dittrich, P. Reiher; “Internet Denial of
•
more useful things.
Service: Attack and Defense Mechanisms”
If the network contains worms or some systems in the [5] Attack downs Yahoo, Google.
network act as zombies then they will be identified easily, http://news.zdnet.co.uk/internet/security/0,3902039157748,00.htm
•
and before much damage is done. June, 2004.
The Local Area Network (LAN) will function in a much [6] CERT Coordination Center, “Denial of Service Attacks,”
•
better way. http://www.cert.org/tech_tips/denial_of_service.htm
The cyber law must enforce the implementation of this or [7] Peir Luigi Rotando, “Denial of Service (DoS) attacks and
countermeasures”, Dipartmento di Informatica, Universita di Roma, La
any similar model.
Saienza
[8] Nick Delfort, “The Brief History and Effect of DDoS attacks”,
IV. CONCLUSION AND FUTURE WORK: CEN445 Final Paper
[9] The TCP Guide; http://www.tcpguide.com
[10] P. Ferguson, D. Senie, “Network Ingress Filtering: Defeating Denial of
A clear explanation of DoS and DDoS attacks is presented; then its Service Attacks which employ IP Source Address Spoofing”, RFC
effects are analyzed in a real network by programs coded in C using 2267, 1998
the POSIX API’s. The flooding DoS and DDoS attacks had a severe [11] M. Burdach, “Hardening the TCP/IP stack to SYN attacks”, 2003
[12] http://en.wikipedia.org/wiki/Egress-filtering
effect on the resources of the victims’ machine.
[13] D. J. Bernstein, “SYN Cookies”, http://cr.yp.to/syncookies.html
[14] http://en.wikipedia.org/SYN_cookie
Moreover, four research models were reviewed, their strengths and [15] M. Handley, A. Greenhalgh; “Steps towards a DoS-resistant Internet
shortcomings were identified, and finally a new research model was Architecture”, University College London
proposed which indeed sums up the strengths of many of the models [16] W. Stallings, “Cryptography and Network Security, Principles and
and eliminates most of their shortcomings, thus competing to be a Practices” pages 607-608.
better alternative to prevent DoS and DDoS attacks. This model can [17] M. BlueMenthal and D. Clark, “Rethinking the design of the Internet:
be simulated in NS-2, or implemented in some test environment to The end to end arguments vs. the brave new world”, MIT Press 2001
[18] T. Anderson, T. Roscoe, D. Wehterall; “Preventing Internet
get the actual results and clarifying its way for real world Denial-of-Service with Capabilities”
implementation. [19] S. Zhang, P. Dasgupta; “Denying Denial-of-Service: A Router Based
Solution”
V. ACKNOWLEDGEMENTS: [20] J. Mirkovic, P. Reiher; “D-WARD: A Source-End Defense Against
Flooding Denial-of-Service Attacks”
[21] J. Mirkovic, “D-WARD: A Source-End Defense Against Distributed
I would like to thank N-W.F.P. University of Engineering and Denial-of-Service Attacks”, Ph.D. dissertation, University of
Technology, Peshawar, Pakistan for providing me the suitable California, Los Angeles.
research environment to carry on this work. [22] M. D. Rey, “Transmission Control Protocol”, RFC 793, 1981
[23] J. Postel, “Internet Control Message Protocol”, RFC 792, 1981
[24] Cert Advisory, “Smurf IP Denial-of-Service Attacks”, 1998
REFERENCES [25] J. Postel, “User Datagram Protocol”, RFC 768, 1980
[26] G. Maselli, L. Dari, S. Suin; “Design and Implementation of an
[1] Humphrey Chang, “Danish websites hit by DoS attacks and Anomaly Detection system: An empirical approach”
defacements?”, February 9, 2006 20:23; [27] A. Keromytix, V. Misra, D. Rubenstein, SOS: Secure Overlay Services.
http://www.tgdaily.com/2006/02/10/danishwebsiteshacked In Proc. ACM SIGCOMM, 2002
[2] http://en.wikipedia.org/wiki/Gary_McKinnon
27
SAKE: SECURE AUTHENTICATED KEY ESTABLISHMENT IN

WIRELESS SENSOR NETWORKS
Muhammad Yasir, Mureed Hussain
Computer Science Department
Shaheed Zulfikar Ali Bhutto Institute of Science and Technology, Islamabad, Pakistan
schemes [2,3,8,9,10,11], are also based on pre-deployed keying in

Abstract — Master key schemes are a viable solution to which each sensor node is preconfigured with the same master key
establish pairwise shared secret keys in Wireless Sensor and that master key is then used by each node to generate pairwise
networks. In these schemes, a master key is preconfigured into keys for sharing with each of its neighbors. After the key setup
each sensor node which is then used by each node to generate phase, each node removes the master key from its memory. Key
pairwise shared secret keys. In the literature so far, it is establishment techniques are evaluated on the basis of following
essential for each sensor node to keep master key in its memory
•
metrics [12]:
during the entire phase of key setup. As soon as key setup
Scalability: Key establishment technique should be flexible
completes, each node erases the master key from its memory.
against significant increase in the size of the network even after
Although key setup phase of a node lasts for a small interval of
•
deployment.
time, it is not impossible for an adversary to compromise a node
Efficiency: Key establishment technique must be efficient in
during this time. In this situation, the presence of master key
terms of storage (required memory to store security
can be disastrous. So the challenge is to protect a sensor
credentials), processing (amount of processor cycles required
network from the compromise of master key during its key
to establish a key) and communication (number of messages
setup phase. We propose Secure Authenticated Key
exchanged during key generation process).
̇
Establishment (SAKE) protocol that meets the above challenge
Resilience: Resilience against node capture and security
by introducing an idea that master key need not to be kept by a
credentials compromise stored in a node or exchanged over
sensor node for the entire key setup phase thereby shortening
wireless radio links.
the master key compromise window. With the help of our
proposed scheme, other attacks during key setup phase can also
be avoided. II. AN OVERVIEW OF RELATED WORK
Index Terms— Sensor Networks, Key Establishment, Perrig et al. [10] proposed Security Protocols for Sensor Networks
pairwise key namely SPINS. It consists of two protocols, SNEP and TESLA.
SNEP provides confidentiality, two-party data authentication,
I. INTRODUCTION integrity, and freshness. TESLA supports authenticated broadcast.
Figure 2 shows the key establishment protocol for SPINS [9]. It
shows that when a node A wants to establish a pairwise shared key
Wireless Sensor Networks are self-organizing networks of locally SKAB with node B with the help of a trusted third party server S,
communicating sensor nodes having limited computation, memory, which acts as authentication and key distribution server.
and energy resources. Due to the deployment of sensor nodes in
large numbers, the nodes are made inexpensive; thereby not Key establishment protocol works as:
tamper-resistant and an adversary is quite capable of destabilizing 1. Node A sends a request message to node B.
some of them. Above all, the sensor nodes with limited resources are 2. Upon receiving this message, Node B sends a message
not capable of using public key cryptography because it is to the S
computationally intensive. All of these inherent unique 3. Key server S authenticates and generates the pairwise key
characteristics make implementation of security in wireless sensor and sends the key back to node A
networks far more challenging than in traditional computer 4. Key server then sends pairwise key to node B.
networks. Key establishment lays the foundation for implementing
security in a sensor network, in which a sensor node establishes
secure links with its neighbors when it is deployed in a sensor S
network. In fact it establishes the keys necessary to provide
confidentiality, integrity and authentication services. Due to the 3. {SKAB|NA|IDB}KAS 4. {SKAB|NB}KBS
resource limitations of sensor nodes, key establishment protocols
for sensor networks are based on symmetric key algorithms. Key
establishment for wireless sensor networks must be lightweight, 2. IDB|{NA|NB|IDA} KBS
secure and efficient. A B
1. NA|IDA
Camtepe et al. [12] described following specific security
requirements of wireless sensor networks along with availability, Fig. 1 Key establishment in SPINS [10]
authentication, integrity, confidentiality, and non-reputation.
•
The drawback of SPINS is the use of a central key server S. As a
Survivability: capability of providing minimum level of service sensor network comprises a large number of nodes, the use of
•
in the presence of power loss, failures or attacks. central key server restricts the scalability of the sensor networks.
Degradation of security services: capability of changing
security level with the change in resource availability. Seys et al. [8] proposed SNAKE, a protocol for key establishment
based on an ad-hoc scheme. In contrast to SPINS, no key server is
Pre-deployed keying is the most suitable solution for bootstrapping involved in key establishment. Key establishment in SNAKE is
secret keys in sensor networks. In pre-deployed keying sensor done in the following way.
nodes are loaded with keys before their deployment. Several 1. Node A sends a request and a nonce to node B
solutions based on pre-deployed keying are proposed. Master key
28
2. Node B returns T (the identity and nonce of node A, B’s own Deng et al. [2] described Opaque Transitory Master Key Scheme
identity and nonce), concatenated with a Message (OTMK), a pairwise key establishment scheme in which a master
Authentication Code (MAC) of T to A. key M is preconfigured into each sensor node. To establish session
3. Upon receiving this message, node A proves its authenticity keys, each node u broadcasts a request message as:
and sends the message back to node B.
Figure 2 describes the protocol. u *: JOIN|EM (IDu|nonce)
1. request|NA IDu is the identity of u. Upon receiving this broadcast, a node v

generates a random number kv,u and responds u as:
2. T=(IDB|IDA|NA|NB) | MACK [T] v u : REPLY|EM(IDv|nonce+1|kv,u)

A B
After receiving this message node u performs verification of nonce
and designates node v as its verified neighbor. The pairwise key is
either kv,u generated by v or ku,v generated by u. ku,v is used as shared
key if IDu<IDv else kv,u is used as shared key.
3. IDA|NB|MACK [IDA|NB] To enable new nodes to join, Deng et al. [2] proposed another
scheme in which a node v generates a new key kv= MAC (M, IDv)
Fig. 2: SNAKE [8]
and a number of verifiers containing two random numbers ri and yi
where yi = f (M, ri). A node v stores many verifiers and erases the
After this process, node A and node B generates pairwise shared key
master key M. When a new node u arrives, node v sends one of its
as: KAB = MACK [NA|NB]
random numbers ri as a challenge to node u. Node u computes zi=f
(M, ri ) and sends it to node v. Now node v compares zi with yi and if
Lai et al. [9] proposed BROSK in which each node broadcasts the
both are equal then it verifies node u. After that the pairwise key is
key negotiation message to establish the pairwise shared key with its
established between u and v. A critical question is how many
neighbors. To establish session keys with its neighbors, a sensor
verifiers are to be contained by a node. A sensor node with a limited
node A broadcasts the following message:
memory resource cannot store too many verifiers for authentication
of newly joining nodes.
A *: IDA|NA|MACK (IDA|NA)
Dutertre et al. [11] depicted that many timing mechanisms can be
K is the same master key preconfigured in all the nodes prior to used to reduce the probability of message collisions and proposed a
deployment. Upon receiving A’s broadcast, a node B can construct protocol in which a secret group authentication key bk1 and a key
the pairwise shared key KAB by generating the Message bk2, to generate session key is preconfigured into all nodes. To
Authentication Code (MAC) with the help of two nonces as: establish pairwise keys a node A broadcasts a hello message as:
KAB= MACK (NA|NB) A *: HELLO|A|NA|MACbk1 (HELLO, A, NA)
Similarly node A also receives the broadcast message from node B Upon receiving, a node B checks the validity of A and responds to A
and constructs the key KAB in the similar manner. as:
Zhu et al. [3] proposed Localized Encryption and Authentication
ACK|A|B|NB|MACbk1 (ACK, A, B, NB, NA)
Protocols (LEAP) based on master key scheme. In LEAP, an initial
key kI is preconfigured into each node. Each node u derives its
This acknowledgement proves to A that B knows bk1 and has
master key as:
received NA. After that A and B establish keys as KAB=Gbk2 (NA,
NB). G is a keyed one-way hash function.
ku = f(kI, u), where f is a secure one-way function. A node u
establishes pairwise keys by broadcasting its identity as:
Perrig et al. [1] paid attention to key distribution in commodity
sensor networks where they did not assume a global passive
u *: u adversary and proposed Key Infection: a plaintext key exchange
protocol, in which every sensor node sends plaintext to establish
After broadcasting node u waits for its neighbor’s response. A node
pair-wise keys with each of its neighbor nodes. In this scheme it was
v sends the response as:
assumed that during the network deployment phase, the attacker can
monitor only a fixed percentage a of communication channels. This
v u: v, MAC(kv, u|v) assumption was in contrast to the previous work on key distribution
for sensor networks, in which a strong threat model was assumed: it
Upon receiving the response, Node u derives kv as: kv =f (kI, v) and was assumed that the adversary is present both before and after
authenticates the response using the derived key. The pairwise key the deployment of nodes, and can monitor all communications in the
for nodes u and v is f(ku,v), for u > v, and f(kv, u) otherwise. Each network at all times. It is assumed also that it is possible for an
node erases the key kI from its memory after the key setup phase but adversary to maliciously reprogram a small number of sensor nodes.
retain its own master or individual key. If this key setup time completes in short time, an adversary has very
little time to eavesdrop on key setup. Compromising a node does not
The initial key kI in LEAP is the single point of failure. If an offer the adversary any added advantage in deducing these keys.
adversary is somehow ever able to obtain kI before it is erased, she Key Infection establishes key as: every sensor node simply decides
will be able to compute all previously setup pairwise keys in the a key and broadcasts it in plaintext to its neighbors. For example a
network as well as calculate all future session keys that may be node i, after deployment, broadcasts a key ki. Due to short-range
established. In addition, the adversary can also inject any number of nature of transmission, possibly half a dozen other nodes within
malicious nodes into the network. range of i, notice each other’s presence and start self-organization.
Another node j after hearing i’s signal, produces a pair wise key kj
29
and sends it, along with its name, to i:{ j, kji} ki. Minimum power III. SAKE: SECURE AUTHENTICATED KEY
essential for the link is used to transmit the packet. The key kji is ESTABLISMENT
used between i and j. Perrig et al. argued that only 2.4% of links will
be compromised if there is one hostile sensor node for every 100
white nodes, and there are four neighbors in the range of each node. In key establishment techniques based on master key schemes,
Plaintext key exchange protocol does not provide protection for master key is a single point of failure. Deng et al. [2] demonstrated
confidentiality, integrity, and node authentication. Another that time to deploy a sensor network is very important. The key
drawback of this approach is that an adversary can inject malicious setup phase may need to be extended until all nodes are activated, or
nodes into the network, since there is no authentication mechanism arrive at their destination. In such cases, the master key would live
to verify whether a sensor node is a valid member. for several minutes. If an adversary knows the memory area
precisely from which to draw the keys, then the node compromise
In key pre-distribution schemes, secret keys or secret information is time is in the tens of seconds. So it is quite possible for an adversary
distributed to every sensor node prior to deployment into the sensing to physically capture a node and compromise the master key during
area. Gligor et al. [4] proposed the probabilistic key pre-deployed its key setup phase. The challenge in this scenario is to protect the
scheme, which is regarded as basic scheme. In this scheme, key network from its single point of failure. This challenge can be met if
setup completes in three phases: key pre-distribution, shared-key key establishment process would be made less dependent on master
discovery, and path-key establishment respectively. key. This can be done if the individual key of each sensor node
would be made an active participant in key establishment process
along with the master key. This load balancing approach will
protect the master key from becoming a single point of failure.
Moreover, in a situation when master key is compromised, it is
required that a typical key establishment technique would not
enable an adversary to compute all pairwise keys with the help of
master key.
We propose Secure Authentication Key Establishment (SAKE)

protocol, which is based on master key scheme. SAKE supports the
establishment of pairwise shared keys between the nodes of a sensor
network. SAKE is dissimilar from other master key schemes as it
makes key establishment less dependent on master key. Firstly, it is
Fig. 3 probabilistic key scheme [5] not essential for sensor nodes running SAKE to keep the master key
in the memory for the entire key setup phase. In fact a node running
Figure 3 shows the key pre-distribution phase in which each sensor SAKE erases the master key very early during its key establishment
node holds k distinct keys, randomly chosen from a big key pool process. This reduces the chance of master key compromise if an
having size P where P>> k. This set of k keys carried by each node is adversary physically captures a node during its key setup phase. It
called key ring. An identifier is attached with each key. Shared key also makes the insertion of malicious nodes in the network nearly
discovery phase starts with the deployment of nodes in which each impossible. Secondly, in case of master key compromise, an
node discovers its neighbors to share common keys in its radio adversary cannot compute all pairwise keys in the network with the
range. Links are established between the nodes at the end of help of master key in contrast to LEAP protocol proposed by Zhu et
shared-key discovery phase. Path-key establishment phase is the al. [3] in which an attacker can calculate all previously setup
last phase in which pairs of nodes are connected that want to pairwise keys as well as is able to compute all pairwise keys that
establish a secure link but they are not sharing a common key. These may be established in future.
pairs of nodes can be reachable by two or more hops. Perrig et al. [6]
modified Gligor’s scheme for proficient handling of bootstrapping A. Assumptions
problem. This scheme is more resilient against node capture. In this It is assumed that the sensor nodes are static and a sensor node does
scheme, instead of using one common key for key establishment, q not know its immediate neighbors in advance. Tmin is a certain
common keys are required for key setup with a hash function. amount of time an adversary requires to compromise a node. Tkey is
Figure 4 demonstrates this scheme [9]. the total key setup time of a sensor node and it may be less or greater
than Tmin. This is in contrast to other master key schemes [2,3],
where Tkey is assumed to be always less than Tmin. Tkey is divided
into two time intervals namely Tstart and Testab. Tstart is the smaller
portion of Tkey that at maximum is less than half of Tkey so it is
smaller than Testab. It is assumed that being a very small portion of
total key setup time; Tstart is always less than Tmin. In other words it
is believed that a node that requires Tkey time to complete its key
setup process cannot be compromised in its Tstart phase.
Following notations have been used in the presentation of the

protocol
̇
Fig. 4 common keys with a hash function
a and b are communicating sensor nodes.
̇
It is possible that many nodes in the sensor network can share that
Ekm (M) means encrypting message M with master key km.
̇
same key. If one of these nodes is compromised, all other nodes
Eks (M) means encrypting message M with subordinate key ks.
̇
sharing the same key will also be compromised. Du et al. [7] has
proposed a model, in which the sensor nodes are deployed in groups, x|y means concatenation of message x with y.
so in each group the nodes have high probability to be near to each Since communication is taking place between a typical sensor node
other. So the basic idea is to let the nodes deployed near to each and its immediate neighbors, our scheme establishes pairwise keys
other select keys from sub-key pools that share more keys. In the that are shared by a node and its immediate neighbors.
scheme, because each node carries fewer keys, the memory In this scheme, each sensor node is preconfigured with a master
efficiency and resilience are both improved. key km and subordinate key ks. Moreover each sensor node has also
its unique individual key. A node keeps the master key km during its
Tstart phase only. A node erases km when Tstart expires but retains ks.
30
B. Functional Details of SAKE the identity of b and discards immediately because it has already
Key setup process of a sensor node starts when a node is deployed. established the pairwise key.
Initially the node a is in its Tstart phase in which it performs
following tasks. When Tkey expires, each node erases individual keys of its
immediate neighbors with which it has established pairwise keys,
̇ Encrypts its individual key ka with both km and ks. but it retains ks, y1 and y2 because they are used by a node to
y1= Ekm (a|ka) establish pairwise keys with newly coming nodes that come after the
expiration of Tkey. Depending upon the policy, if internal network
̇
y2= Eks (a|ka)
Broadcasts a HELLO message to discover its immediate processing (data aggregation and passive participation) is required,
neighbors. a node can be programmed to keep the individual keys of its
immediate neighbors collected during key setup.
a *: HELLO|a|na| Ekm (a|na|ka) (1) 3) Adding new nodes
If a new node u arrives, it will broadcast HELLO message as:
a is the identity of node a. na is the nonce or random number and ka
is the unique individual key of a. Tstart interval is made so small that u *: HELLO|u|nu| Ekm(u|nu|ku)
a node a can perform the above tasks only. As soon as Tstart expires,
node a erases km but keeps y1, y2 and ks. From (1) it is clear that km is Suppose a node b, which was deployed earlier and has completed its
used for authentication purpose because while receiving this key setup process, receives this broadcast. Node b cannot decrypt
broadcast, another node b is assured of the identity of node a. Now the broadcast since it has erased km. Node b computes pair wise key
following conditions can occur. as ku,b= f(kb,u )and replies to node u as
1) Node b is in Tstart phase
If node b is in its Tstart phase, it keeps both km and ks. Besides its own b u: REQUEST|b|u|nu+1|y1|y2
HELLO broadcast, if it receives HELLO message from node a
during Tstart, it can decrypt (1) with km and gets ka. After getting ka, Upon receiving REQUEST message, if node u is in Tstart phase, it
it can generate pairwise key as can decrypt y1 to verify the identity of b as well as to get kb. Now
node u computes the pairwise key as ku,b= f(kb,u ). If node u is in
ka,b= f(ka,b) if a>b else Testab phase, it can do all the above tasks with the help of y2.
ka,b = f(kb,a ) if a<b and sends a response to a Because the HELLO message cannot be authenticated by node b,
b a: REPLY|b|Eka(a|b|kb|na+1) (2) resource consumption attacks can be launched. Solutions to these
attacks have been suggested by [3] which can successfully be
f is a secure one-way function. Upon receiving this reply, a adopted. Authentication process of newly coming nodes in SAKE is
decrypts with its individual key ka and validates the identity of b. very simple and consumes less memory because a certain node has
After that it generates pairwise key with b as b has generated above. to store only y1 and y2 than the scheme proposed by Deng et al. [2],
It does not matter that whether node a is in its Tstart phase or in Testab in which a number of verifiers are stored in memory for the same
phase, because it is not dependent on any master key to decrypt and purpose.
process (2). Testab is the second time interval of Tkey which lasts till C. Security Analysis of SAKE
the expiration of Tkey. Node a only needs its individual key ka to
The prime objective of SAKE is to protect a sensor network from
generate pairwise keys.
master key compromise during key setup phase and making the
2) Node b in Testab phase
insertion of malicious node very difficult. During the first time
If node b is not in Tstart phase, it will be in Testab phase where it has
interval (Tstart) of key setup phase, a typical node broadcasts the
erased km. Node b will remain in Testab phase until Tkey expires. We
HELLO message encrypted with the master key km to identify its
know that most of the requests will be processed after Tstart, because
neighbors. Tstart being a very small time interval of total key setup
Tstart is so small that a node hardly broadcasts its on HELLO
time ends after this broadcast and each node erases km from its
request. Now node b is not in position to decrypt (1). Node b memory though the key setup phase is not completed yet. According
computes pair wise key as ka,b= f(kb,a ) only if a<b, and replies to to our assumption, it is impossible for an adversary to compromise a
node a as node during Tstart. Therefore an adversary has a very less time and
chance to get km and it is clear that a sensor node must require km to
b a: REQUEST|b|a|na+1|y1|y2 (3) be authenticated and to establish pairwise keys. In Testab, which is
the second and the larger time interval of key setup phase, if an
Node b cannot compute ka,b= f(ka,b ) if a>b, because it cannot adversary however compromises a node, she will get the
decrypt (1), so in this case it simply sends (3) to a. Upon receiving subordinate key ks only. ks alone is useless for her, since the
this message, if node a is still in Tstart phase (although the chances malicious nodes dropped by her need to have km for authentication
are very rare), since it keeps km and ks, it decrypts y1 with km to as well as to establish pairwise keys with the legitimate nodes.
verify the identity and to get individual key of node b and ignores y2.
But if node a is also in Testab phase, y1 is useless for it. It decrypts y2 If an adversary inserts his malicious node in the network and
with the help of ks to get b’s identity and its individual key. So it has broadcasts the HELLO message encrypted with ks rather than km,
not only verified the identity of b but also generated key ka,b. Node a the receiving node, when tries to decrypt the HELLO message, the
verifies the identity of b and generates pairwise key ka,b= f(kb,a ) if message will be decrypted. This decryption will prove that the
a<b and does not send any notification message to b, otherwise sending node is a malicious node since a legitimate node knows that
generates ka,b = f(ka,b) if a>b and sends following notification no other legitimate node will broadcast the HELLO message
message to b as encrypted with ks. A legitimate receiving node also knows that it
a b: REPLY|a|b|Ekb(a|b|ka,b) (4) cannot decrypt the HELLO message when it has completed its key
setup phase. So the receiving node will discard this message.
Node b decrypts this message and gets the pairwise key. An Moreover if a malicious node x which has no km and ks, broadcasts
interesting situation arises if node a has received HELLO broadcast the HELLO message encrypted with a different master key, it will
from b during its Tstart phase and generated pairwise key ka,b. But a’s not be able to decrypt y1 or y2 contained in the REQUEST message
HELLO broadcast was received by b in b’s Testab. Now b generated received from the legitimate node b that has received x’s HELLO
ka,b and responded with (3). When a receives this message, it checks
31
message. Since the malicious node x cannot get kb it cannot be able [3] S. Zhu, S. Setia, and S. Jajodia, “Leap: Efficient security mechanisms
to establish the key with node b. for large-scale distributed sensor networks,” In 10th ACM Conference
on Computer and Communication Security, Washington D.C, USA,
2003
In other mater key schemes [2,3,8,9,10,11], since master key is kept
[4] L. Eschenaur and V. Gligor, “A key-management scheme for
and required for entire key setup phase, a skilled adversary gets distributed sensor networks,” In Proceedings of the 9th ACM
more time to compromise a node and to get the master key. Conference on Computer and Communication Security, pp. 41-47,
Moreover the compromise of master key in LEAP [3] allows 2002.
adversary to determine all pairwise keys in the network. It is not [5] S. Sarsan and V.G. Adadda, “Analysis and Evaluation of Key
possible in SAKE if km or ks are compromised. Management Schemes in Wireless Sensor Networks,” 2004.
[6] H. Chan, A. Perrig, and D. Song, “Random Key Predistribution
Schemes for Sensor Networks,” In IEEE Symposium on Research in
IV. CONCLUSIONS Security and Privacy, 2003.
[7] Du W, Deng J, Han Y S, Chen S, and Varshney P K, “A key
We have proposed Secure Authenticated Key Establishment management scheme for wireless sensor networks using deployment
(SAKE) protocol that establishes pairwise keys shared between knowledge,” In Proceedings of IEEE INFOCOM′04, IEEE Press, Hong
Kong, 2004.
neighboring nodes of a wireless sensor network. SAKE is an [8] S.Seys, “Key Establishment and Authentication Suite to Counter DoS
efficient protocol in terms of memory, processing and Attacks in Distributed Sensor Networks” unpublished manuscript,
communication. It makes a sensor network more resilient to master COSIC.
key compromise and makes the insertion of malicious nodes [9] Lai, B., Kim, S., and Verbauwhede, I, “Scalable session key
extremely difficult. We plan to implement and validate this security construction protocol for wireless sensor networks,” In IEEE
protocol in a simulator in near future. Workshop on Large Scale RealTime and Embedded Systems
(LARTES), 2002.
[10] A. Perrig, R. Szewczyk, V. Wen, D. Culler, and J. Tygar, “SPINS:
REFERENCES Security Protocols for Sensor Networks,” In Proc. Of Seventh Annual
ACM International Conference on Mobile Computing and Networks
[1] R. Anderson, H. Chan, and A. Perrig, “Key infection: Smart trust for (Mobicom 2001), Rome Italy, 2001.
smart dust,” In 12th IEEE International Conference on Network [11] Dutertre, B., Cheung, S., and Levy, J, “Lightweight key management in
Protocols, Berlin, Germany, 2004. wireless sensor networks by leveraging initial trust,” Tech. Rep.
[2] J. Deng, C Hartung, R. Han, and S. Mishra, “A Practical Study of SRI-SDL-04-02, System Design Laboratory, 2004.
Transitory Master Key Establishment for Wireless Sensor Networks,” [12] S. A. Camtepe and B. Yener, “Key Distribution Mechanisms for
University of Colorado at Boulder, Boulder, CO, 80309-0430. Wireless Sensor Networks: a Survey,” Rensselaer Polytechnic
Institute, 2005.
32
OVERVIEW OF SPAM FOR VOIP AND IMS AND FUTURE DIRECTION FOR
DEVELOPING SPAM PROTECTION MODEL FOR VIDEO CONFERENCING
Ayesha Salahuddin1, Mohammad Sher2, Thomas Magedanz2
1. Department of Computer Science, AIOU, Islamabad
2. TU Berlin/Fraunhofer FOKUS Open Communication Institute, Berlin, Germany
The IP Multimedia Subsystem (IMS) is currently being

Abstract — With the emergence of Voice-Data Integration,
standardized by the Third Generation Partnership Project (3GPP)
the circuit switched infrastructure is being replaced by packet
nd 3GPP2 [8].
switched networks i.e. Voice over IP (VoIP) The VoIP
technology is gaining popularity because of low cost and easy to
IMS architecture based on Internet Protocol (IP) .This emerging
maintain but at parallel it faces many challenges like security,
technology brings real time multimedia services to mobile terminals
reliability and QoS etc. This paper focuses on the Spam threats
that have GPRS, UMTS and WiFi interfaces. Multimedia services
describing unsolicited and junk calls and emails. The spam is
include web browsing, email, video conferencing, video sharing and
likely to become a severe large problem in the future for all IP
push-to-talk. The next generation packet-switched networks make
based networks. The Spam is also a serious concern for future
the communication easier and cheaper by voice, instant message or
next generation IP Multimedia Subsystem (IMS) which
video than ever before. Spammers will find that their increased
provides Service Delivery Platform (SDP) for Fixed-Mobile
returns from distributing multimedia spam will make it profitable to
Convergence (FMC).
exploit these technologies [9] This is a far more serious problem
In this paper, we present state-of-art review of Spam for VoIP
than traditional spam because real time communications require
and IMS and propose a multi-layered architecture to block
immediate action from the recipient, unlike email that can be
automated spammers. Our approach is based on the
filtered and delivered whenever user is ready. It takes significantly
combination of blacklists, white lists and Bi F analysis.
more time to identify real time spam than to identify email spam. If
users receive voice and video calls after many hours then they also
Index Terms — Voice over IP (VoIP), IP Multimedia get frustrated. Just like other forms of Spam, voice and video spam
Subsystem (IMS), Security, Spam, Unsolicited Calls has also shown growth in recent years. A recent study shows that
during the 2006 US mid-term elections nearly two-thirds of
I. INTRODUCTION registered voters received unsolicited robo-calls from various
political campaigns [10].
The internet is becoming an integral part of modern society. Daily
million of people communicate with their family, friends through
IP Multimedia Subsystem (IMS) is attractive environment to
internet, emails, SMS and video communication. The technique of
spammers. Packet switch nature of (IMS) is cost saving and
sending emails became widely accepted and succeeded to replace
provides cheaper services for end users. Hence it is most profitable
old means of communication i.e. letter. The human society is
and lower cost for spammers. As it uses traditional IETF protocols
shifting from traditional telephone network to Voice over IP (VoIP)
such as SIP and RTP means that writing applications for VoIP spam
networks because this is cheaper and simple way to communicate.
is more or less same as writing an email spam bot. Moreover, IMS
According to TeleGeography’s forecast, the total VoIP subscribers
users are particularly susceptible due to the nature of the SIP
in the U.S. would reach 17.5 million users at 2010 [1].
Uniform Resource Identifier (URI). These URI addresses are
similar to email addresses (sip:ali.doe@operator.com), and can
As VoIP gains popularity it also faces a lot of problems which need
easily be taken from the Internet in a similar fashion that email
to be solved. The biggest weakness of this emerging technology is
addresses are currently taken from websites, newsgroups and
Spam threats which are unsolicited and junk calls and mails [2].
mailing lists.
Different researchers and IT companies are working on Spam
protection but the problem is not abating and the number of Spam
messages and calls are increasing exponentially every year. The II. VIDEO COMMUNICATION ARCHITECTURE BASED
data and statistics obtained in September, 2005 illustrate that 67, 6% ON SIP
of email was spam [3]
The IP Telephony based on Session Initiation Protocol (SIP)
There are different type of Spams but the most well known is email developed by the Internet Engineering Task Force (IETF). It is an
Spam. The Internet users receive many spammed emails daily. As application layer control protocol that can create, modify, and
compared to e-mail spam with VoIP Spam, the load on network terminate multimedia sessions [7]. Different types of entities are
resources may be ten times as much [4] due to VoIP Spam. The defined in SIP: user agents, proxy servers, redirect servers, and
Spam over Internet Telephony (SPIT) is also more annoying [5], registrar servers. Figure 1 shows a simple video communication
because the real time transmission runs with every spam message, architecture based on SIP.
disturbing the users current activity. The use of VoIP instead of
traditional PSTN networks will make it easier for spammers to make The SIP based architecture consists of two types of entities: user
automated tools to deliver their spam to the user [6] and VoIP agent (UA) and network servers. The User Agents are the end
communication is also much cheaper. The authors of [7] claim that devices in SIP network, they can be either a SIP phone or SIP client
the cost per call for VoIP are roughly three orders of magnitude software running on a PC. They also generates SIP request for
lower than traditional PSTN calls, making it a lot cheaper for a establishing media session. There are following four types of SIP
spammer to send the messages world wide. The spammed emails get network servers.
filters out by using different filters but not in the case of real time
transmission. A. Redirect Server
Redirect server accepts SIP request from a client, maps the SIP
address of the called party and returns the address to the client.
Redirect Server doesn’t forward request to other servers [11].
33
B. Registrar server Spam in VoIP scenario is going to occur in the shape of

A registrar server is a server which accept register request from a Voice/Video messages that has to be treated in real time in order to
client, and update the location database, the location database is prevent our system from being flooded with spam messages.
used to store contact information [11]. Spammers sends Spit calls during communication and our
application server is intelligent enough to distinguish between good
calls and Spit calls.
Fig. 2 Different Security Threats Levels

Fig. 1 A typical VoIP Architecture
C. Proxy server
Proxy Server handles SIP requests for the source UA. A proxy
server can perform as a server or a client to make a request in the
name of clients. Requests are serviced either locally or passed on to
another server. If it is necessary, proxy server will rewrite a request
before forwarding the request [11].
D. Location Server
Location Server is used to store terminals location, and provide a
terminals location to the proxy server or redirect server.
III. SPAM-BASED SECURITY THREATS
The SIP based VoIP is open based and flexible architecture which
faces many security threats from different communication protocols Fig. 3 Spam Launching Scenario
and at different level. We have grouped these threats from three
levels i.e. IP, Application and Content. In the following we briefly describe different types of spam for
SIP-based architecture.
A. IP Level Threats
These threats are from the IP based infrastructure and are transferred A. Call Spam
to the VoIP infrastructure. For example, these threats are shared
with application such as WAP, emails etc The spammers generate bulk of unsolicited session initiation
procedures to establish a voice, and video, instant messaging [12].
B. Application & Protocol Level Threats The SIP INVITE requests are utilized in this scenario. If the
These threats are exploiting the design faults and implementation recipient accepts the request the spammer proceeds to relay their
flaws of communication protocols and applications. For example, message over the real time media. This is the classic way to
the protocols and application level threats include misdirect calls, generate commercial spam on SIP telephony.
contaminated calls etc. They cause a lot of disruption in VoIP B. IM Spam
systems.
The instant messaging spam is similar to email spam. In this type,
C. Content Level Threats the spammer sends a bulk of unsolicited instant messages. These
They are content related threats interfere with media streams like messages may be any commercial advertisement about any product
voice calls and video conferencing and this level focus on VoIP or business etc. In VoIP scenario, the SIP Messages Requests [13]
spam. These threats shared with other applications such as email, are utilized for generating IM Spam etc.
IM and browse based applications. C. Presence Spam
Presence Spam is also like email Spam. It is defined as a bulk
IV. TYPES OF VOIP SPAM unsolicited set of instant messages, whose content contains the
message that the spammer is seeking to convey. IM spam is most
Today the most annoying Internet problem is the spam emails. The naturally sent using the SIP MESSAGE [13] request. This spam
volume of email spam has grown up to that level that it threatens the used SUBSCRIBE request for the presence event package in an
viability of email as a service. It is predicted that the impact of VoIP attempt to be in white list of user in order to send them IM. It does
spam will much greater than email spam. In this scenario SIP is the not convey content in the messages.
most targeted protocol by the attackers to launch Spam calls. The
top level view of launching Spam on the VoIP network is depicted
in figure 3.
34
V.ANALYSIS OF MOST IMPORTANT TECHNIQUES puzzle it will be placed on White list. Like many other techniques
FOCUSING ON LIMITATIONS AND IMPROVEMENTS Turing test is also depend on Senders identity which cannot be
easily authenticated in emails. It can be applied to call spam as well.
This kind of voice-based Turing test is easily extended to a variety
There are many solutions to VoIP spam like content filtering, white of media, such as video and text, and user interfaces by making use
lists, black lists, consent based communication and identity of the SIP application interaction framework [14].The problem with
authentication. These solutions were developed for email spam. It Voice Turing test is same as with emails instead of having automata
could be adapted to SIP spam because of similar nature. Generic process the test spammer can pay to the cheap workers to take the
over view of these solutions are as under tests. Due to these limitations Turing test never completely solve the
A. Content Filtering problem.
It is most common type of spam protection which is used in email. It
analyzes the content of email and checks whether it is spammed or VI. PROPOSED SOLUTION AND TOP LEVEL
not. Bayesian spam filter belongs to this category. But for VoIP ARCHITECTURAL VIEW
spams this method is not efficient for two reasons.
Firstly spam cannot be analyzed by content filter before user reply. Multi layered architecture is proposed to meet the objectives of
Secondly if we consider current state of technology it might be undersign research and this study will entail a literature and
impossible for real time data. [15]. software study to determine if and which approaches have already
B. Black List been identified or implemented. The suitability of NS 2 for security
simulation can be examined through a literature study. As no single
Here the spam filter maintains a list of addresses that identify solution has yet solved this problem so it’s a fair assumption that a
spammers. It may include user name or entire domain. IT is unlikely single solution will not provide suitable protection against spam.
to have effectiveness for VoIP spams. First it is easy for spammer to The diagram below shows the description of the methodology to be
do spoofing. Secondly spammer can obtain new SIP address from followed with in this research
any providers. In the case of unauthenticated identity it may be
difficult for spammers to forge the SIP identity. Even in the case of
authenticated identity spammers can obtain new SIP addresses and Bi F analysis
this cause blacklist useless for VoIP. [15]
C. White List
White lists
It is inverse of black list. It is a list of valid sender from whom user is
willing to take calls. Here spammers cannot change the identities to
obtain white list. It used strong identity authentication mechanism Black list
which prevents address spoofing problem. It concludes that
combination of white list and identity authentication could be a
Fig. 4 Multi layered Architecture for preventing video spams
good solution for VoIP spam. But they do not provide complete
solution to the problem e.g. it does not give solution to introduction
problem which is how to decide whether the caller should place in A. Blacklists and White lists
white list or not so this can be solved by next technique. [15] The first and second layer of protection involves blacklisting and
white listing. These lists provide good protection against
D. Consent Based Communication well-known spammers and ensure calls from regular contacts are
It is used in collaboration with black list and white list. These kinds connected without intrusion. The problem with blacklists is that
of systems are mostly used in presence and IM but not in email. This they are hard to keep up-to-date and they are notorious for blocking
should be combined with identity mechanism. Since most IM entire domains when only a few guilty parties are involved. In our
systems are closed nowadays so sender identities can be architecture we will populate the user-defined blacklists according
authenticated. It is standardized in SIP for presence if they extend to to Bi F analysis. The user defined white list will be populated
cover IM and calling both it may not be useful. manually.
E. Reputation Systems
It is also used in collaboration with black list and white list. It is
implemented in highly centralized architecture. It based on negative
reputation scores suffer from many of the same problem like in
black list. It based on positive reputation if users are happy with
each other rather then blaming. If we combine presence based
systems with reputation it will give an interesting mixing. [15]
F. Pay-per-Call
Here receiver gets paid for every message that is unwanted. Every
sender is charged for spam message. Family and friends would be
put on the do-not-charge list and their emails would arrive in the
user’s inbox for free. But for anyone the user doesn’t know, a charge
of $ 5 (or whatever price the user wanted) could be levied. [14]
In this way messages are accepted only from trusted senders and for
every other email the users get paid for receiving VoIP spam
messages.
.
G. Turing Tests
In email systems Turing test are those where the sender gives some Fig. 5 Flow Diagram Of Multi layered Architecture for preventing video
spasm
puzzle which can only be solved by human, these are also known as
CAPTCHAS (Completely Automated Public Turing test to tell
Computers and Humans Apart).If user correctly answered the
35
B. Bi F Analysis [3] M. Sunner, ‘Email security best practice’, Network Security, pp. 4–7,
2005.
The third layer of protection in our architecture involves analyzing [4] Gagner, R.P, ‘Voice over Internet protocol’ Bowie State University,
the call patterns of unknown callers. An unknown caller is one that Maryland, 2005.
does not appear on a blacklist or white list and therefore the [5] J. Pessage, J. Seedorf,’ Voice over IP: Unsafe at any Bandwidth?’,
legitimacy of the caller is still in question. In our system we will Eurescom Summit Heidelberg, 2005.
calculate variance in call frequency with the Bi F (FFT) algorithm [6] R MacIntosh, D.Vinokurov, ‘Detection and mitigation of spam in IP
and detail over view of these layers is as above telephony networks using signalling protocol analyses’, pp. 49–52,
2005.
C. Flow of SD Application Server [7] J.Rosenberg, C.Jennings,’ the Session Initiation Protocol (SIP) and
Once the call arrives at the SD (Spam detection) Application Server SPAM’ , 2004.
[8] 3rd Generation Partnership Project (3GPP), ‘IP Multimedia Subsystem
the URI of the caller is tested against user defined black lists and
(IMS)’, TS 23.228.
white lists. If a caller is unknown then BI F analysis are used to [9] V. Mockapetris,’Telephony’s Next Act’, IEEE Spectrum Magazine,
determine the legitimacy of the caller. Following Figure shows the April 2006.
flow of INVITE and 200 OK SIP messages [10] L. Rainie ‘64% of registered voters received robo-calls in the final two
Months of the 2006 election’, Pew Internet & American Life Project ,
VII. CONCLUSION 2006,.
[11] Radvision, “Session Initiation Protocol”, 2005.
[12] B.Campbell,’The Message Session Relay Protocol’,
This paper highlighted the problems of multimedia spasm in IMS.
draft-ietf-simple-message-sessions-19 (work in progress) February
We also proposed a multi layered approach for developing spam 2007,
protection for video conferencing. We also presented architecture of [13] B.Campbell, J. Rosenberg, H.Schulzrinne, C. Huitema, and D.
Video communication based on SIP. Security threats focusing on Gurle,’Session Initiation Protocol (SIP) Extension for Instant
spam and analysis of important anti spam techniques are described. Messaging’, RFC 3428, December 2002.
[14] S.Arrison, ‘Canning Spam: An economic solution to unwanted Email
REFERENCES [online]. 1st Edition’, Pacific Research Institute, 2004.
[15] J.Rosenberg, C.Jennings’ The Session Initiation Protocol (SIP) and
[1] http://www.forbes.com/2003/11/11/cz_kb_1111spam.html Spam’, July 9 2007,
[2] H. M. Butler, ‘Spam-the meat of the problem’, Computer Law &
Security Report, vol. 19, pp. 388–391, 2003.
36
Session II Conference Chapter on Computing
Chair Prof. Dr. Mushtaq Ahmed

Faculty of Computer Science & Engineering
Ghulam Ishaq Khan Institute of Engineering
Sciences & Technology, Pakistan
Assessing the Usability of Official Web Portal of Government of Pakistan: A Heuristics

Approach
Abdur Rehman, Muhammad Hamad Hassan, Amjad Mehmood
m.hamad.hassan@gmail.com, amjadiit_kust@yahoo.com
One survey of 6000 computer users in US found an average of 5.1
Abstract — This paper reports on the usability of the official hours per week wasted trying to use computers [2].The increase in
Web Portal of Government of Pakistan. To the best of the the amount of information risks acting as a time sink that can
author’s knowledge it’s the first ever usability study of prevent people from ever getting any real work done [3]. Users on
Pakistan’s official web portal. The result was that many the web are usually so overloaded with information within a single
heuristics usability problems were found on the web portal. The application that they find it difficult to use these resources
web portal failed to comply with even basic web usability effectively. They are so overloaded with information that they end
guidelines. This indicates that web usability is poorly up with nothing usable. The success rate in usability studies, when a
understood by the web developers, if at all. It also indicates that user was asked to perform a task, has been reported to be less than
no usability testing is in place to ensure that the web portal 50 % [4].
follows the basic usability guidelines. Some key mistakes were
poor searching, non-standard use of color for links, improper II. NUMBER OF INTERNET USERS IN PAKISTAN
page title and a complex home page.
According to wired Pakistan [5], Internet usage in the country has
witnessed a tremendous increase during the last couple of years.
I. INTRODUCTION Number of Internet users have reached a record number of 12
million. And there are approximately 7 million Pakistanis living
Almost every government has a presence on the internet in the form abroad [6], who are also possible users of the web portal. This
of a website. Websites for governments have got much more means the total number of expected users of the web portal is at least
importance than in the early days of internet. More and more 19 million. In the following sections, official web portal is analyzed
governments business such as job advertisements, providing basic for observance of heuristics usability guidelines.
information about identity card, passport, constitution, government
plans etc are moving on to internet. With increasing internet users
in Pakistan, many users’ first choice to get information about an III. PROBLEMS IN HOME PAGE
issue related to government is to visit the relevant web site. But it’s
The home page is different from all other web site pages. Mostly it
not just a matter of quickly developing a web site, decorating it with
acts as a door to any web site. A well-designed home page projects
some heavy graphics or flash animations and be on the internet. It
a good first impression. A good home page should clearly convey
requires proper planning and testing. Taking care of usability is an
the site purpose. One study found that when asked to find high
important part of web development. One study shows that users
quality web sites, about half of the time participant looked only at
spend less than 2 minutes on the entire site before deciding to
the home page [4]. One study shows that users visiting a new site
abandon it [1]. On the web, usability has become a necessary
spend an average of 30 seconds on the homepage before deciding to
condition for survival.
abandon it [1]. Home page of web portal at
http://www.pakistan.gov.pk was visited between 10 to 13 March
Web usability is important for two main reasons:
2008 and was analyzed for observance of the heuristics usability
principles such as avoiding active links to same page, use of
(i) First is that web users don’t spend much time on a single
standard colors for links, use of breadcrumbs and proper search
web site. Web users are mostly goal driven. They come to a
facility etc .
web site with a specific goal, try to achieve that and leave
the site. They require instant results. A. Avoiding Active Links to same page
(ii) Second is the massive growth and scale of the Active links to current page cause following problems.
Internet and the amount of online resources has that have led to (i) Clicking a link that links users to same page is sheer
information market overwhelming users with choice. It does not waste of user time.
cost users more than a click or typing a few words to go from one (ii) It causes the users to doubt that they are really at
web site to other. location they think they are?
Good usability has two benefits [1]: The problem was found on the home page of web portal.
(i) It supports an organization’s business goals on the web and B. Simplicity in Design
thus helps achieve its goals. According to Neilson [8], in designing user interface for traditional
(ii) Usability empowers humans and makes it easier and more application or for a web site, one of the developer’s key decisions
pleasant to handle the technology that's infusing every concerns the tradeoff between features and simplicity. The more
aspect of modern life. features, the more complicated the system inevitably becomes. This
means that the more buttons and options one puts on home page, the
On the other hand, the impact of poor usability includes: less users are capable of quickly finding information they need.
Web sites must focus on simplicity rather than features. A common
(i) Wastage of innumerable person-hour design pitfall is to believe that by providing lots of options and
(ii) High levels of frustration several ways of doing things, we can satisfy everybody. Every time
(iii) Significant under use of potentially useful developers add a feature to a system, there is one more thing for
system users to learn and possibly use erroneously.
37
One of the common mistakes made on the portal was the repetition
of same link on same page or two slightly different links connecting
to same sub page. This causes following problems:
(i) When two or more different links are used to connect to
same sub page, it naturally makes users think that these
are two different links. But when user sees the same sub
page it confuses the users.
(ii) Using two exactly similar links wastes valuable space on
web page.
(iii) Every element that is put on page puts burden on user
memory by having the user to look at one more
unnecessary feature.
(iv) Also every irrelevant link competes with a relevant link
for user attention
MSN Usability Research recommends that pages be not overloaded

with navigation choices .People will stop reading options after they Fig. 3 Menu option on Business tab
see 4-5 distinct option [8]. The problem in web portal is that there is
unnecessary repetition of links on the home page, as has been shown
in following figures.
Fig. 4 Menu option on Foreigners tab
(vii) Economic and Statistics link on 3 tab.

(viii) Industries and Trade link on 2 tabs.
This repetition of links unnecessarily complicates the interface for

users. Also the Launch option provides a relatively new style of
interaction, rather than a simple click option. It has been observed
that users bitterly complain whenever they are exposed to sites with
Fig. 1 Menu option in General tab too many diverging ways of doing things [8].
IV. NAVIGATION RELATED PROBLEMS

Two important things related to navigating a web page are [10]:
(i) Knowing where you are.

(ii) Knowing where you have been.
To tell the user where he is currently in the context of the overall

structure of the site, “breadcrumb” technique is used. Breadcrumb is
a simple but very useful in navigating a web site.
A. Absence of Breadcrumb
It just takes a one line space on the page but provides two useful
information to the user:
(i) It displays a hierarchical path to user’s current location.

(ii) It enables a user to go to a previous link in a single click.
Fig. 2 Menu option in Citizen Tab
Although the web portal uses breadcrumb it does not do so in a
Following are the numbers for repetition of different links on standard way. For instance in the breadcrumb:
different tabs on same page. “Home>About Government” Home is not clickable. Another
(i) Citizenship and Immigration link on 3 tabs. heuristics guidelines is that a link to home page should be available
(ii) Utilities link on 3 tabs. form every other page. One of the techniques usually used is to
(iii) Revenue and Taxation link on 3 tabs. connect the logo to home page. Users than can simply click on logo
(iv) E-Government Projects link on 3 tabs. from any sub-page to go to main page. But this fundamental
(v) Online flight booking link on 3 tabs. guideline has not been observed on the web portal. Although home
(vi) Railways and Train Schedule on 3 tabs. link is provided in Launch option, but it has two problems.
(i) Its not clearly visible for users.
38
(ii) It displays the same page again , instead of home page of improve their navigation [12]. This technique has two clear
web portal. advantages:
B. Non standard use of color for links
Of the 222 guidelines for web sites, Neilson describes the most
important to be user having a sense of control while visiting the site.
Neilson has observed that there is increasing reluctance among users
to accept innovations in web design [11]. The prevailing user trend
in design is to request designs that are similar to everything else they
see on the Web. For example, using blue color for link and red or
purple for visited links has become a de facto standard for web sites.
It greatly helps user in instantly deciding:
(i) What are links and what is text?

(ii) Differentiate between visited and unvisited links.
(iii) What was the last visited link?
Any deviation from this standard will make users insecure and less
confident in browsing the site.
Fig. 7
Figure number (5) and (6) highlights the problem of using
non-standard color for links on tabs. The link color on these tabs is The launch link is in white color as has been shown in figure (8)
light black and yellow respectively.
Fig. 8
(i) It helps users avoid the links they are not interested in thus
Fig. 5 saving their time.
(ii) It helps user understand the destination page upon arrival.
Exceptions to using link titles are where it is clear from the link and
surrounding text where it leads. The web portal does not use link
title except at two places , but that too incorrectly.
Following problems were found on the Web Portal.
(i) Placing the mouse over the logo in the upper left side of web
page display the title “Ministry of information Technology”.
(ii) Placing the mouse over the picture on right side displays the
title “Click to activate and use this control”.
(iii) Moving the mouse over “Founder of Pakistan” link displays
the title”Founder of Pakistan”.
Fig 6 First is confusing because it displays the title of a specific ministry.

Second is wrong because its not a meaningful title. It could have
The references part contains link in green color figure no (7). been like “Web site of Law Ministry” etc. Third is wrong because it
displays same title as link, which is of no help to users.
Fig. 5 Non Standard color for links
V. PROPER SEARCH FACILITY
Using non standard color for link has another problem. That is when
same color is used for both link and simple text. Navigation is useful for information space that a user can cover
completely. But for a site with lots of information, search becomes
C. Providing Link Title an important element of the user interface. A web site may start off
A simple step developers can use to help users in navigation is use of as a simple site, not containing much detailed information for users.
proper link title. It helps present to users bit of information about But as an organization begins to grow, it feels the importance of its
what the link offers. Newer browsers offer the ability to pop up a site and, they start adding various kind of information to their site.
short explanation of a link before the user selects it. Such This makes it difficult for users to find their relevant information,
explanations can give users a preview of where the link will lead and
39
especially when the links are not in proper hierarchy. The best users
can do to save their time is to search a site for particular information.
As has been observed by Neilson, half of the users of web are search
dominant. And rest also try searching a web site for information
when their navigation efforts fail. For such users its better to provide
search option on every page of a site, because one does not know ,
when a user fails to find relevant link and turns to searching
option[9]. As an example it will be really difficult for a user to
browse the whole portal to find information about, say, what the
current passport fee is? Web portal of Government of Pakistan
provides the search interface in the form what users usually expects,
but the results of searching for information were really poor.
Following screen shots shows the result when web portal was Fig. 9 Result of selecting the search option from Launch option
searched for basic information like Pakistan, Musharraf and
National Assembly. VI. IMPROPER PAGE TITLE
Being easily searchable on the internet is vital for any web site. One
study has shown that users usually don’t look at web sites that are
not in top 30 of search engine result page(SERP) [4] . One of the
simple steps to include a site or a page in search engine results is to
use proper page titles. The web portal failed to use proper page title
for its different sub pages. It just uses the title “Pakistan.Gov.pk The
official gate way to the Government of Pakistan”. Using proper
page title for sub pages help improve the site ranking in search
engine.
VII. CONCLUSIONS
It has been observed that official web portal of Government of
Pakistan does not observe even the basic usability guidelines. This
Fig. 9 Result of searching the site for Musharraf affects the user performance as well as web site purpose. To
improve the situation, web developers should be taught the
importance of usability plus an independent usability testing could
be arranged.
REFERENCES
[1] Jakob Nielsen and and Hoa Loranger ,Prioritizin web usability, New
Riders Press, Berkeley CA, 2006
[2] Ben Shneiderman “Pushing Human Computer Interaction to empower
every citizen, Universal Usability”, COMMUNICATIONS OF THE ACM,
May 2000/Vol 43. No5
[3] Jakob Nielsen,”Multimedia and Hypertext:The Internet and Beyond” ,
Chapter 8 “Coping with Information overload “ Morgan Kaufmann, San
Francisco, 1995.
[4] Snjay J.Koynal, Robert W. Balley, Janke R Nall “Research-Based Web
Design & Usability Guidelines”,
Fig. 10 Result of searching the site for National Assembly htp://usability.gov/pdfs/guidelines%5Fbook.pdf.
[5] http://www. wiredpakistan/2007/01/01/number-of- a
internet-users-hits-12-million-in-pakistan/
[6] http://www.dailytimes.com.pk/default.asp
[7] Feature Richness and User Engagement,
aaaahttp://www.useit.com/alertbox/features.html
[8] http://msdn.microsoft.com/library/default.asp?url=
/library/en-us/dnsiteplan/html/improvingsiteusa.asp
[9] Search and You May Find,
http://www.useit.com/alertbox/9707b.html
[10] Alan Dix ,Janet E. Finlay ,Gregory D. Abowd ,Russell Beale
,Human Computer Interaction 3rd Edition, Pearson Prentice Hall, 2003
pp205.
[11] The Increasing Conservatism of Web Users,
http://www.useit.com/alertbox/980322.html
[12] Using Link Titles to Help Users Predict Where They Are
Going. http://www.useit.com/alertbox/980111.html.
Fig. 11 Result of searching the site for Pakistan
Also the search option in Launch menu works really poorly. It

displayed a system generated message as is clear from figure.
40
Architecture of reconfigurable artificial neural network Co-processor

Kamran Rauf, Muhammad Usman and Abdul Bais
Department of Computer Systems Engineering NWFP
University of Engineering and Technology Peshawar, Pakistan
Abstract — In this paper we propose the architecture of a communication between the co-processor’s top module and
neural co-processor for on-board learning. The co-processor is sub-processors and also with in the sub-processors. There is also a
based on backpropagation network and acts as a dedicated network structure designed for the communication of layers with in
hardware connected to a PC. It consists of several sub-modules a sub-processor. The co-processor reported in [5, 6] and the
i.e. sub-processors representing a column of the proposed architecture depend on the backpropagation network.
backpropagation neural network. The architecture allows the The rest of the paper is organized as follow: The architecture of
co-processor to form any structure depending on the specific the backpropagation network is described in Section 2. This is
followed by a detailed presentation of the proposed architecture of
application. We have built a communication network for the
the co-processor based on backpropagation with online learning in
sub-modules to communicate with each other and within each
Section 3. Finally, the paper is concluded in Section 4.
sub-module; there is also a communication network for the
layers to communicate with each other. The operations of each
sub-processor are independent from other sub-processors i.e. a II. BACKPROPAGATION NETWORK ARCHITECTURE
neuron calculates its output as soon as all inputs are available.
Backpropagation network shown in Fig. 1 is a multilayered neural
I. INTRODUCTION network with n sub-processors (neurons). A sub-processor has an
input layer (X), an output layer (Y) and hidden layer(s) (H) [7]. X 0,
X 1, ..., X n represent the input layers, H 00, H 01,..., H 0n show first
Large training time in neural network is the fundamental ob- stacle hidden layers, H 10, H 11, ..., H 1n show the second hidden layers,
in real time applications. A computer based neural network H m0, H m1, ..., H mn show mth hidden layers and Y 0, Y 1, ..., Y
program can’t solve the timing problem, as the essence of the neural n represent the output layers of n sub-processors as shown in Fig. 1.
net is the parallelism which a single processor can not give. There Each column in Fig. 1 represents a sub-processor i.e. a neuron
are several commercial softwares that are using neural network which is connected to other neurons.
algorithms to solve different problems. These softwares can not
provide mobility and good training time [1]. There are also a few
commercialized chips available for the neural network
applications but these chips do not have the capabilities of on-chip
learning because these chips do not train the neural network but
take the weights that computer has calculated for it [1]. One of the
vendors of these chips is Intel, with its 80170 ETANN
(Electronically Trainable Artificial Neural Net- work) chip and
Neural Semiconductor, with its DNNA (Digital Neural
Architecture) chip [1].
There is a demand for a dedicated hardware that can be trained

for different applications. Some attempts have been made for the
dedicated hardware, designed specifically for an application e.g.
[2], [3] etc. The main thing in the hardware implementation is the
communication structure. The more the communication structure
is efficient, the higher is the performance of the hardware. This
constitutes the architecture of the co-processor which will act as a
dedicated neural hardware. One of the famous dedicated hardware
co-processor based on self organizing map neural networks is
KOKOS [4]. The first on-board learning based on backpropagation
network KOBOLD is presented by M.Bogdan, H.Speckmann and
W.Rosenstiel [5, 6].
In [5, 6] architecture, they implement a communication structure

on the basis of bus topology. There is a global bus, to which all
sub-processors are connected and a local bus which is connecting
the sub-processors in a ring like structure. The problem in this
communication structure is that while propagating error difference
to other relevant sub-processors, only one sub- processor is allowed Fig. 1 Architecture of Backpropagation network
to do so and all other wait for their turn. So there is a significant There are two modes of operations in training of backpropagation
delay for the sub-processors. Also communicating via local bus, network, forward propagation and backpropagation [7]. In forward
the sub-processor sends its data on the bus and the neighbor collects propagation, the network input patterns are presented to input layer
it, which observe the packet for its relevancy. If the packet is for which calculate its products and convey it to the above hidden layer
that sub-processor, it will save it in its local memory and if not, it and associated sub-processors’ hidden layer. Then each hidden
will forward it to its neighbor. This method of communicating layer calculates its net input and output which is then conveyed to
weights produces a significant delay. For faster communication, the the layer above of the current sub-processor and other connected
delays are to be reduced to their possible level. In proposed sub-processors. Similarly the process is continued in other
architecture there is a special switching center used for the
41
sub-processors as well until final layer i.e. output layer calculate its its required data, it will calculate its output and convey it to the
output [7]. relevant layer and sub-processor (if needed). The proposed
architecture of the co-processor is described in the followed Section
In backpropagation mode each sub-processor’s output layer then 3.2.
computes the error difference and if it’s greater than a specified
tolerance level than all layers below the output layer updates their
weights. Similarly the error is also conveyed to the connected
sub-processors, so that they can also compute their new weights [7].
The whole training is followed in this trend.
III. PROPOSED ARCHITECTURE

In this section, we will describe the co-processor architecture. The
section is broken in to the following sub-sections: first we will
describe the hierarchy of modules involved in the implementation
of co-processor in Section 3.1, then the proposed co- processor
architecture in Section 3.2 and finally the sub-processor architecture
in Section 3.3.
A. Hierarchy of Modules
The code implementation of the co-processor comprises six major
modules. The hierarchy of the modules is given in Fig. 2. In Fig. 2,
the CoProcessorTopModule is the top module which is receiving
configuration instruction and pattern-target pairs from external
environment. To configure the co-processor, user give instructions
through PC interface and these instructions are conveyed to the
co-processor’s top module which configures the co- processor
accordingly.
Fig. 3 Co-Processor Architecture

B. Co-Processor Architecture
Architecture of the co-processor shown in Fig. 3 is very similar to a
star topology network. There is a special switch named
MainSwitchingCenter. In Fig. 3 the top module which provides the
external environment interface is CoProcessorTopModule and the
adjacent blocks represents the sub-processors which are the basic
processing elements of the co-processor.
Fig. 2 Hierarchy of modules
There are sixteen sub-processors in the proposed architecture. The
The MainSwitchingCenter is the main module for communication. number of the sub-processors can be increased to two hundred and
The whole communication of the co-processor depends n this fifty six. They can also be extended to more than two hundred and
special switching center. The MainSwitchingCenter is like a switch fifty six by increasing word size but by increasing the number of
in a star topology network but it is specially designed to work for sub-processors, the complexity of the network will also increase.
the parallel structure of the co-processor and present less delay to the Also we need the same number of pins in the co-processor as that of
data. The MainSwitchingCenter is designed in a way that it can sub-processors, so that in applications all sub-processors can be
receive data from its all ports and can send data on all its ports given patterns simultaneously which is not a feasible solution.
simultaneously which makes the co- processor faster. Each Applications that are having sixteen dependent parameters can be
sub-processor is assigned an address in order to identify the dealt with it. Within each sub- processor there are ten layers
sub-processors. Each sub-processor starts its calculations for itself working autonomously. The architecture of the sub-coprocessor is
and then it calculates products and other results for its forward described in the following sub section.
neighbor which ensures that no two sub- processors are sending data C. Sub-Processor Architecture
to same sub-processor and in this way the collision is avoided. This The special architecture of sub-processor shown in Fig. 4 resem-
method also increases the speed of communication. For further bles to that of a bus topology network. The SubProcessor is the top
optimization in speed, the co-processor implementation is pipelined. module for this portion of co-processor. It receives the con-
SubProcessor is the basic processing device in the co-processor.. It figuration instructions, patterns and targets sent by co-processor and
is like neuron in biological nervous system. Each SubProces- sor has products and error differences sent by other sub-processors from
three different kinds of layers i.e. InputLayer, Hidden- Layer and MainSwitchingCenter and accordingly it maneuver its layers.
OutputLayer as show in Fig. 2. These layers work simultaneously
and the whole process is carried in a way that as soon as a layer has
42
IV. CONCLUSION
We presented a new architecture for digital neural co-processor for

online learning backpropagation. The architecture of the
co-processor leads to better performance. The communication
network leads to asynchronous communication between sub-
processors and also with co-processor’s top module. Further work
is in progress on the co-processor. The major steps that are left
include the implementation of PC interface for configuration and
implementation of other backpropagation algorithms for further
optimization. This hardware implementation can lead to a powerful
neuro-computer that can be used in a wide range of applications
such as: ” Optical character recognition ” Image and Data
compression ” Load forecasting problems in power system area ”
Control problems ” Non linear simulation ” Biomedical applications
” Fault detection problems ” Network monitoring systems ”
Communication etc
REFERENCES
[1] Ingrid F. Russell, “Neural networks,” 2007, [Online:

www.hartford.edu, accessed 12-09-2007].
[2] Johannes Schemmel, Karlheinz Meier, and Eilif Mueller, “A new vlsi
model of neural microcircuits including spike time dependent
plasticity,” in Proceedings IEEE International Joint Conference on
Neural Networks, Kirchhoff Inst. for Phys., Heidelberg Univ.,
Germany;, 2004, vol. 3, pp. 1711– 1716.
[3] Michael Freeman, Michael Weeks, and Jim Austin, “Aicp: Aura
intelligent co-processor for binary neural networks,” in IP-SOC 2004
IP Based SOC Design Forum and Exhibi-tion.
Fig. 4 Sub-Processor Architecture
[4] H. Speckmann, P. Thole, and W. Rosenstiel, “Hardware
There are ten layers in each sub-processor i.e. an input layer, an implementation of kohonen’s selforganizing feature map,” in Artificial
output layer and eight hidden layers. The number of layers can be Neural Networks, 2, I. Aleksander and J. Tay- lor, Eds., Amsterdam,
increased to any number but increasing number of layers lead to Netherlands, 1992, vol. II, pp. 1451–1454, North-Holland.
larger training time i.e. the network will take more time to converge
[5] M. Bogdan, “Kobold: a neural coprocessor for backpropagation
for a specific application than with lesser layers. There is an with online learning,” M.S. thesis, Ecole d’IngCnieurs in
advantage of having large number of hidden layers that is the Informatique Industrielleet Instrumenta- tion, Grenoble, France, 1993.
network will converge more precisely. There is basically a trade off
between training time and preciseness of the network. But since [6] M. Bogdan, H. Speckmann, and W. Rosenstiel, “Kobold -a neural
many application areas can be accommodated in this limited coprocessor for backpropagation with online learning,” in
number of hidden layers, so there is no need to increase the number Proceedings of the Fourth International Confer- ence on
of layers. These layers are the processing components of the Microelectronics for Neural Networks and Fuzzy Systems, 1994, pp.
sub-processor like the components in neurons of biological nervous 110–117.
system. Each different type of layer has its own different process
[7] S. N. Sivanandam, S. Sumathi, and S. N. Deep, Introduction to Neural
structure. Networks using Matlab 6.0, Tata McGraw Hill Companies, 2006.
43
Towards Terrain Modeling For Robot Navigation Using A Single Camera

Muhammad Tufail, Abdul Bais, Muhammad Usman, Sheraz Ali Khan
NWFP University of Engineering and Technology
Peshawar, Pakistan
{tufail,bais,usman,sheraz.ali}@nwfpuet.edu.pk
Abstract — Terrain modeling is one of the basic capabilities different robot positions as a virtual stereo pair.
that a mobile robot must have for vision-based navigation in an The remaining of the paper is structured as follows: Section 2
outdoor environment. In order to have enough information for presents detection of sparse features and their matching for camera
robot navigation, it is required to construct dense depth map calibration and rectification of the image pair. Establishment of
from the images of the scene. 3D scene reconstruction from dense correspondence and 3D reconstruction is the topic of Section
depth maps of the environment can then be used to obtain a 3. Experimental results are discussed in Section 4; finally the paper
terrain model. Research on terrain modeling using multiple is concluded in Section 5.
cameras has been actively pursued. However, to reduce system
cost this study aims at achieving similar results from a II. FEATURES DETECTION AND MATCHING
monocular camera and sequential images. The main objective is
to reduce the computational time for dense 3D reconstruction.
Feature points between two images in sequence are matched based
on similarity in order to establish a relationship between them. The
I. INTRODUCTION selection of appropriate features plays an important role in the
matching process and has a direct impact on 3D reconstruction. The
Antipersonnel mines have been a serious threat to the safety and use of Harris corners [3] as feature points for camera calibration and
livelihoods of people during wars. In war hit countries, agriculture, rectification is discussed in this section.
transport and economic recovery is affected by the presence of
landmines. One of greatest challenges of landmine clearance is Corners are matched over two images within limited disparity.
finding the buried mines. Currently, locating landmines is mainly While matching corners, there may be more than one match in one
done by hand. Humanitarian demining personnel use metal detector image for a corner in other image. In case of multiple matches for a
and a stick to probe every 2.5 square cm in a suspect area. Using this single corner the selection is based on gray value similarity in
technique a great amount of time and manpower is required to clear pre-defined neighborhood of the corners in the two images. Due to
a relatively small area. Another drawback is the high risk of injury errors in the correspondence process, the set of initial corresponding
to the demining personnel. points is typically contaminated with incorrect matches or outliers,
which are removed using the RANSAC (RANndom SAmple
Recent advances in the development of accurate and reliable sensors Consensus) algorithm.
for landmine detection are so promising that researchers have
become interested in the development of autonomous robotic Once the correspondence between feature points in the two
system that requires minimum interaction of human operators. images is established, it is used for computation of the
Typically, the robots/vehicles are equipped with GPS and IR fundamental matrix and planar rectification as presented in
systems and other tools like metal detectors to search for landmines the following subsections.
buried in broad fields. On the other hand, robots are more suitable
for off-road missions and antipersonnel unexploded ordnance A. Fundamental Matrix and Epipolar Geometry
detection. Specifically, robots are useful for civilian mine clearance The geometric relationship between the two images which is
missions whose reliability must be above 99.6% as defined by the necessary for establishing correspondence between them is best
UN agenda [1]. captured by the epipolar geometry and therefore in applications like
scene modeling and mobile robot navigation, its computation is one
For practical applications like landmine clearance missions, the of the basic tasks to be performed.
robot should have the capability to localize itself and accurately
map the scene simultaneously. This is the basic requirement for For a given point x in one image (for notations used in this paper
autonomous robots. A central issue in intelligent robotics is the please refer to Appendix A), the corresponding point x’ in another
problem of knowing the position of a robot in the environment. This image must lie on the epipolar line. This important fact is called the
issue has been greatly researched. However, the problem of epipolar constraint. The following equation should be satisfied for
handling the very large number of landmarks in the scene is still the all corresponding points
target for many researchers [2].
 x′   f11 f12 f13   x 

 y ′  f f f   y = 0
T
Using vision as the information source for robot navigation for
   21 22 23   
landmine detection and clearance purposes carries many advantages
(1)
1   f 31 f 32 f 33  1 
over using other approaches like laser scanning. Vision sensors and
their measurement procedures for 3D information are flexible and
simple. We only extract the details of information from the sensed
data based on which the size, depth, structure, and material of the or,
object in the scene can be determined.
[xx′ yx′ x′ xy′ yy′ y′ x y1] f = 0

equivalently,
This study deals with the requirement to develop a wide baseline
monocular stereo vision technique for landmine detecting mobile (2)
robots that combines local motion estimation and robust image From all point matches, we can obtain a set of linear equation of the
matching in order to build a 3D model of the terrain. The reasons for form
using the monocular approach are low-cost and ease of replication
in case of damage to the robot during the landmine detection
process. The system must work from two images obtained at
44
Αf =0 Where xˆ i′ = Η ′xi′ , and xˆ i = Η 0 xi , Η 0 = Η ′Μ ,

Μ = [e′] x F (Actually F can be factored into a skew-symmetric
(3)
+
matrix of epipole e’ and a non-singular matrix M as F = [e′]x Μ ).

In (3), f is a nine-vector containing the entries of the matrix F, and A
is the equation matrix. The system of equation can be solved by
singular value decomposition (SVD). Applying SVD to A yields the
decomposition USVT with U and V being orthogonal matrices and
σ 1 ≥ σ 2 ≥ σ 3 .... ≥ σ 9 ≥ 0 are positive or zero elements f (a, b, c) = ∑ (axi′ + ayi′ + c − xî′ )

Equivalently we can minimize the function
S, a diagonal matrix containing the singular values. These singular
values 2
in decreasing order. i
B. Rectification
Rectification seeks to map an input image pair to one in which The resulting HA will be of the form
epipolar lines are parallel and correspond to image rows.
a b c 
Rectification helps simplify stereo matching by reducing the search
Η A = 0 1 0
space to one dimension. If the epipoles are not visible in either
image, it can be achieved by applying projective transformations
0 0 1
(also called homographies) H and H’ to the left and right images
respectively, so that the epipoles e and e’ in the left and right images
are mapped to infinity and the epipolar lines become parallel to
1  1 
horizontal scan lines. In short we have to find H and H’ such that
Ηe = 0 , Η ′e′ = 0

 
And finally
0 0 Η = Η AΗ 0 (5)
The algorithm for estimating H and H’ is reported in [4] and III. DENSE CORRESPONDENCE AND 3D
briefly explained in this section. From the estimated RECONSTRUCTION
fundamental matrix F and epipoles e, e’, the projective
transformation H’ for the right image is obtained which maps
The sparse matching is used for estimation of fundamental matrix
the epipole e’ to the point at infinity, (1, 0, 0)T . This
and image rectification. However, dense depth map is required for
transformation H’, is given by: robot navigation. Dense correspondence can be achieved by pixel to
Η ′ = GRT
pixel matching in the image pair. The main advantages of using this
(4) approach over the sparse matching are (i) a dense disparity map or
depth image is provided, and (ii) no feature extraction is required
[5].
Where,
A. Algorithm for Dense Correspondence
1 0 − x0 
Dense correspondence is achieved using dynamic programming
T = 0 1 − y0  is the translation matrix.

which determines the set of correspondences between two images
•
by minimizing a cost function encoding the following two
0 0 1 
heuristics:
1. Ordering constraint: If a point a is on the left of point b in the
cosθ − sin θ 0
reference image then the match for a should to be on the left of
R =  sin θ cosθ 0 is the rotation matrix where

the match of b in target image.
•
2. Uniqueness constraint: Each pixel should have exactly one or
 0 1
no match in the other image.
In the pseudocode [6] shown in Algorithm 1, i and j range from 0 to
0
 e′ 
n − 1, where n is the number of pixels per scan line (it also
θ = arctan y e′ 
represents the width of the input images). For the left and right scan
 x
lines from the stereo pair, the correspondence problem is solved by
 1 0 0
using dynamic programming on an n × n matrix A which holds the
costs of matching two scan lines of size n each. A[0, 0] is initially
 
zero and all other entries are computed from upper left to the lower
• G= 0 1 0 Where f = x-component of RTe′ .

right corner. All the entries in matrix A can be filled initially with
 1 
− f 0 1
some dissimilarity measures as used in [7]. The minimum optimal
cost for a scan line is then calculated by backtracking through
matrix A starting from A[n − 1, n − 1] and ending at A[0, 0]. Three
directions Left, Up, and UpLeft (default) are possible choices to go
up along the diagonal. The column difference between path- and
diagonal entries will result in the left disparity map.
The corresponding projective transformation for the left image H is
found through least-squares estimation. This means finding HA by Algorithm 1. Computing Disparity Map [6]
∑ d (Η A xî , xî′ )
minimizing the cost function,
1: // Calculate the minimum optimal cost for a scanline
2
2: Minimum = Min (A[i − 1, j] ,A[i, j − 1] ,A[i − 1, j − 1])
3: ColorL = LeftImage [i, y]
i
4: ColorR = RightImage [j, y]
45
5: A[i, j] = Minimum + (ColorR − ColorL)2

6: DisparityMapL [i, y] = j − 1
7: DisparityMapL [j, y] = i − j
8: Up = A[i − 1, j]
9: Left = A[i, j − 1]
10: UpLeft = A[i − 1, j − 1]
11: Minimum = Min (Up, Left,UpLeft)
12: case (Minimum)
13: UpLeft : i = i; j = j − 1
14: Left : j = j − 1
15: Up : i = i − 1 Fig. 1 Image Corners. Crosses show the Harris Corners.
16: end
In the second part of the pseudocode from Line 6 to Line 11, i and j
are initially 0 then repeat until they become 0 again. The variables i
and j represent the current position in the left and right scanlines
respectively. Left or right disparity maps will come out depending
on our choice of variable i or j. The variable y represents the vertical
position inside the disparity map and remains constant per iteration.
The disparity map determined above is used to reconstruct dense

structure of the scene up to Euclidean transformation by
triangulation.
The matched points xi ↔ xi′ from the previous step and the
original camera matrices P, P’ are used to reconstruct the 3D
x × (PX ) = 0
structure up to a similarity transformation if P and P’ are the true
camera matrices and up to projective transformation if the true
x′× (P ′X ) = 0 , we can formulate a linear equation for point X in

Fig. 2 Feature Matching. Corners are matched by the Correlation Matching
camera matrices are unknown. As and Technique.
For the “Bread and Oranges” image pair, we have found that
 xp 3T − p1T 
the world coordinate frame as, RANSAC removes 36% of the initial correspondences, resulting in
 3T 
a very accurate set of correspondences.
 yp − p 2T 
ΑΧ = =0
 x′p ′3T − p ′1T 
( 6)
 2T 
 y ′p ′ − p ′ 
3T
Where piT are the rows of P. Then the world position X can be
obtained by least squares solution of the homogeneous linear system
using the SVD.
Fig. 3 Original Left and Right Images before Rectification

IV. EXPERIMENTAL RESULTS
The performance of the algorithm is tested using the famous ”Head Left and right input images shown in Fig. 3 are rectified with H and
and Lamp” and the ”Bread and Oranges” image pairs. In the ”Head H’ respectively and the resulting rectified images are shown in the
and Lamp”, the image planes are perfectly parallel to each other and Fig. 4.
therefore no rectification is needed. On the other hand, for the
”Bread and Oranges” image pair, a great deal of rotation and
translation is involved between the two image planes and therefore
it serves as a real image pair in our experiments. All the algorithms
were implemented in MATLAB and Intel’s OpenCV library [8] on a
Microsoft Windows platform. Torr’s Structure and Motion Toolkit
[9] is used for detection and matching of corner points in two
images.
Fig. 1 shows the detected Harris corners for ”Bread and Oranges”
image obtained with the Torr’s tool. The matched corners are shown Fig. 4 Rectified left and right images. Corresponding points lie on the same
in Fig. 2. Corners are matched over two images within limited horizontal line.
disparity. In Fig. 2 we use 20.
The results for dense correspondence are obtained with Intel’s
OpenCV library 1 which implements the algorithm presented in [7].
The algorithm takes the rectified image pair as an input. It matches
the pixels in both images on scan line by scan line basis. A scan line
in one image is taken at a time and the match for every pixel in the
corresponding scan line is searched for with in the horizontal range
46
specified by the parameter maximum disparity. In order to correct

the errors in a single scan line, the global information about the
relationship between different scan lines in the same image is
utilized. The algorithm generates the disparity map in (0.531
seconds) however it is prone to errors and the resultant disparity
maps are noisy.
Fig. 5 shows a ground truth disparity map for the “Head and Lamp”
images 2 defined with respect to the left image. For display
purposes, the disparity values are normalized to a range of 0 − 255.
Bright pixel intensities correspond to high disparity values which in
turn represent small depth values.
Fig. 7 3D reconstructed “Bread and Oranges” image.
The resulting reconstructed images for the “Head and Lamp” image
are shown in Fig. 8 and Fig. 9.
Fig. 5 Disparity map for Tsukuba image pair.
Fig. 6 shows the disparity map for the “Head and Lamp” image pair.
It can be seen that based on the intensity values, objects in the image
can be differentiated from one another. For example lamp can be
easily distinguished from the head due to its high brightness. In real
scene lamp is closer to the camera than the head. The lamp shaft is
as brighter as the head which is basically due to the noise in the
disparity map.
Fig. 8 View 1: 3D reconstructed “Head and Lamp” image.
Fig. 6 Estimated disparity map for “Head and Lamp” image pair.
For the left image, both the camera rotation and translation matrices
were taken to be identity matrices as it was assumed as the reference
position of the camera. The world coordinate system is assumed to
coincide with the coordinate system of the first camera. The
translation matrix for the second camera position was assumed
equal to the 0.03 meter (same as baseline) and the rotation matrix
was taken as identity matrix due to rectification. Focal length f is
900 pixels and the principal point of the camera is (320, 240) (which
is the image center).
Fig. 9 View 2: 3D reconstructed “Head and Lamp” image.
V. CONCLUSION
The effectiveness of dense 3D reconstruction may vary depending

on the type and complexity of scene we deal with. In case of
landmine detection and clearance problems, scenes are relatively
47
complex and the 3D reconstruction software is very sensitive to REFERENCES

changes in scene attributes like its distance from camera, lighting
conditions, amount of texture, surface smoothness, objects motion, [1] H. Najjarin and A. A. Goldberg, “Landmine detection using an
and occlusion rates. Future work will focus on modeling of natural autonomous terrain scanning robot,” Industrial Robot Journal, vol. 32,
terrain where occlusion rate is high and object surfaces are not no. 3, pp. 240–247, 2005.
smooth and have low texture. In these cases, getting an accurate [2] M. Montemerlo, S. Thrun, D. Koller, and B. Wegbreit, “FastSLAM: A
disparity map from the images can improve the reconstruction factored solution to the simultaneous localization and mapping
process considerably. problem,” 2002.
[3] C. Harris and M. Stephens, “A combined corner and edge detection,” in
Following are some recommendations for future work: Proceedings of the fourth Alvey Vision Conference, 1988, pp.
147–151.
•
[4] R. I. Hartley, “Theory and practice of projective rectification,”
Instead of Harris corner detector, a more robust feature International Journal of Computer Vision, 1998.
•
detector such as SIFT can be used. [5] I. J. Cox, “A maximum likelihood n-camera stereo algorithm,” in
For rectification [10] presents very good improvement on CVPR’94, 1994, pp. ‘733–739.
(Hartley, 1999) which is implemented in our work. The [6] Sven Forstmann, Yutaka Kanou, Jun Ohya, Sven Thuering, and Alfred
improvement is in the way robust matching homographies are Schmitt, “Real-time stereo by using dynamic programming,” in
computed which perfectly aligns the epipolar lines in the two Proceedings of the 2004 Conference on Computer Vision and Pattern
Recognition Workshop (CVPRW’04) Volume 3, Washington, DC,
images. Also the rectified images can be interpolated bilinearly
USA, 2004, p. 29.
•
which will reduce the re-sampling error in the images. [7] Stan Birchfield and Carlo Tomasi, “Depth discontinuities by
The accuracy of dense stereo disparity maps can be improved pixel-to-pixel stereo,” in ICCV, 1998, pp. 1073–1080.
in order to have 3D representation of the scene suitable for [8] Intel, “Open Source Computer Vision and Image Processing
applications like robot navigation. The Birchfield algorithm Library(Computing Software),” 2000.
implemented in OpenCV is open source and therefore can be [9] H. S. P. Torr, “A structure and motion toolkit in MATLAB (Computing
modified to bring improvements in it. Instead of matching Software),” 2002.
individual pixels with each other, a correlation matching [10] John Mallon and Paul F. Whelan, “Projective rectification from the
fundamental matrix,” Image and Vision Computing, vol. 23, no. 7, pp.
technique can be used so that errors of up to one or two pixels 643–650, 2005.
in image rectification can be accounted for. [11] Muhammad Tufail, “Towards real-time terrain modeling for robot
navigation using a monocular camera,” M.S. thesis, Asian Istitute of
Acknowledgment This work was carried out at Asian Institute of Technology, Bangkok, Thailand, 2007.
Technology, Thailand as part of the Master Thesis under the [12] R. I. Hartley and A. Zisserman, Multiple View Geometry in Computer
supervision of Dr. Matthew Dailey. The thesis report is available at Vision, Cambridge University Press, ISBN: 0521540518, second
[11]. edition, 2004.
48
DECREASE TRAINING TIME OF FEEDFORWARD MULTILAYER

PERCEPTRONS BY ATTRIBUTE SELECTION METHOD
Zakia Hussain, Shoukat Ullah, Aman Ullah
University of Science & Technology Bannu, N-W.F.P, Pakistan
Abstract — Data mining, the extraction of hidden Pattern Evaluation: Identifying the patterns representing
knowledge from large databases, is a powerful and the hottest knowledge.
new technology with great potential to help organizations focus Knowledge Representation: Present the data in a useful format,
on the most important information in their large databases. such as a graph or table.
Data mining tools predict future trends and behaviors, allowing
businesses to make knowledge-driven decisions, and extracts Data mining can be classified either as descriptive or predictive
knowledge buried within the mountain. Feedforward mining, which performs the tasks of classification and regression, or
Multilayer Perceptrons (MLPs) is the most widely used model clustering and association respectively.
in artificial neural networks models and used for classification
problems as well as regression. The training process of an Data classification is a two-step process. In the first step, a model is
overly large network can be prohibitively expensive and time constructed, which describes a predetermined set of data classes.
consuming, especially when model contains a large number of Each tuple/sample is assumed to belong to a predefined class, as
inputs dimensionality. This paper presents a new proposed determined by the class label attribute. The set of tuples used for
algorithm for MLPs and compares the results with traditional model construction belong to training set. The model is represented
MLPs algorithm based on backpropagation algorithm. as classification rules, decision trees, or mathematical formulae. In
the second step, the model is used for classifying future or unknown
Index Terms — Data Mining, Neural Network, objects. To estimate accuracy of the model, the known label of the
Classification, Backpropagation test sample is compared with the classified result from the model.
In this study we realized the classification and prediction model

I. INTRODUCTION with feedforward MLPs, which is widely used model in artificial
neural networks, using the backpropagation algorithm. The field of
neural network was originally kindled by psychologists and
“The nontrivial extraction of implicit, previously unknown, and neurobiologists who sought to develop and test computational
potentially useful information from data” [1] or “The science of analogues of neurons. A neural network consists of an input layer,
extracting useful information from large data sets or databases” [2]. one or more hidden layers, and an output layer (as shown in Figure
Also known as knowledge-discovery in databases (KDD) is the 1), where each connection has a weight associated with it. During
practice of automatically searching large stores of data for patterns. the learning phase, the network learns by adjusting the weights so as
To do this, data mining uses computational techniques from to be able to predict the correct class of the input tuples/samples [7].
statistics, machine learning and pattern recognition.
“Data mining is the process of discovering meaningful new

correlations, patterns and trends by shifting through large amounts
of data stored in repositories, using pattern recognition technologies
as well as statistical and mathematical techniques.” [3]
There are other definitions:
“Data mining is the analysis of (often large) observational data sets

to find unsuspected relationships and to summarize the data in novel
ways that are both understandable and useful to the data owner”. [4].
“Data mining is field bringing together techniques from machine

learning, pattern recognition, statistics, databases, and visualization Fig. 1
to address the issue of information extraction from large data bases”
[5]. In the following sections, a detailed review of related literature,
proposed algorithm, and experiment can be found and in the last
Data mining is an essential step in the process of knowledge section, the paper is concluded.
discovery in databases (KDD). Knowledge discovery contains of an
iterative steps [6]: II. RELATED LITERATURE
Data Cleaning: The process of ensuring that all values in a dataset
are consistent and correctly recorded. A lot of work has been done on multilayer perceptrons (MLPs) and
Data Integration: some cases multiple data sources may be other classification and prediction algorithms. In the following lines
combined. we present a brief introduction of few classification and prediction
Data Selection: The data relevant to the analysis task are retrieved algorithms.
from the database Neural Networks (NNs) have been shown to be accurate classifiers
in many domains, often superior to other statistical and data mining
Data Transformation: is transformed into forms appropriate for
techniques. Unfortunately, NNs do not provide an easy process to
mining.
explain how they arrived at their accurate results and this has
Data Mining: The extraction of hidden predictive information from
somewhat limited their use within organizations, as managers desire
large databases.
both accuracy and understanding [8].
49
There are many types of neural networks that has been employed
and these include multilayer perceptrons (MLPs), radial basis ALGORITHM: BACKPROPAGATION_ASM
•
function (RBF) and Bayesian neural networks [9,10]. New pattern Input:
•
recognition methods called support vector machines (SVMs) and Data set of training tuples with class labels
•
Gaussian mixture models (GMMs) have been proposed and found to Attribut_list
•
be particularly suited to classification problems [11]. Feedforward MLPs network
Learning rate
During training, the learning rate is gradually decreased in discrete Output: A trained neural network
steps until the network converges to a minimum error solution [12]. Step-I
(!) for each attribute A {
III. PROPOSED ALGORITHM (2) apply Attribute_Selection_Method
(3) // to rank each attribute describing the given
tuple }
In this section we present a proposed algorithm for data (4) if an attribute has rank above average Then
classification, as shown in Figure 2. This algorithm works in two (5) obtain new_attribute_list
steps. In first step, it takes D, a set of training tuples and their Step-II
associated class labels, and attribute_list. By applying Attribute (6) All weights and biases are initialized
Selection Method [7], only those attributes are selected whose (7) Repeat Steps 8 to 12 while terminating condition is
information gain are ranked high. In the second step, it takes the not satisfied {
new_attribute_list consisting of training tuples and their associated (8) for each training tuple X in Data set {
target values, learning rate and MLPs network to give a trained (9) Calculate Net input and output of
neural network [7]. each unit in hidden and output layers
(10) Calculate Error of each unit in hidden and output
IV. EXPERIMENT layers
(11) if Error exists Then
Experiments have been performed on real dataset of student’s data (12) Calculate updated weights and biases
for admission in MSCS [13] as shown in Figure 3. The software (13) }}
used for classification is Neuralworks Predict 3.21.
Fig. 2
Fig. 3
First we trained the neural network with actual dataset consisting of

eleven attributes through traditional MLPs algorithm whose results
are shown in Figure 4. Second, we trained the neural network
through proposed algorithm with nine attributes whose results are
shown in Figure 5.
V. CONCLUSION
In experiments, we found that the proposed algorithm decreases the

training time as compared to traditional MLPs algorithm.
Furthermore, it increases the efficiency and reduces the complexity
of MLPs structure.
Fig. 4
50
[3] The Gartner Group, www.gartner.com.

[4] David Hand, Heikki Mannila, and Padhraic Smyth, Principles of Data
Mining, MIT Press,Cambridge, MA, 2001.
[5] Peter Cabena, Pablo Hadjinian, Rolf Stadler, JaapVerhees, and
Alessandro Zanasi, Discovering Data Mining: From Concept to
Implementation, Prentice Hall, Upper Saddle River, NJ, 1998.
[6] Chaudhuri, S., “Data Mining and Database Systems : Where is the
Intersection?”, IEEE Bulletin of the Technical Committee on Data
Engineering, Vol.21 No.1, pp. 4-8, March 1998.
[7] J. Han, M. Kamber, “Data Mining: Concepts and Techniques”, Elsevier
India Pvt Limited, Second Edition, 2007.
[8] Christie M. Fuller, Rick L. Wilson "Assessing Extracted Knowledge
From Classifier Neural Networks: An Exploratory Empirical Study",
International Journal of Management and Decision Making (IJMDM),
Vol. 9, No. 1, 2008.
[9] T. Marwala. "Fault Identification Using Neural Networks and
Vibration Data", University of Cambridge Ph.D. Thesis, Cambridge,
UK, 2001.
[10] C.M. Bishop. Neural Networks for Pattern Recognition. Oxford
University Press, Oxford, UK, 1995.
Fig. 5 [11] J. Joachims. Making large-scale SVM learning practical. Scholkopf, C.
J. C. Burges and A. J. Smola, editors, Advances in Kernel
Methods-Support Vector Learning, 169-184, MIT Press, Cambridge,
REFERENCES MA: ,1999.
[12] S.K. Pal, S. Mithra, "Multilayer perceptron, fuzzy sets, and
[1] W. Frawley and G. Piatetsky-Shapiro and C. Matheus, Knowledge classification", Neural Networks, IEEE Transactions on Volume 3,
Discovery in Databases: An Overview. AI Magazine, Fall 1992, pp. Issue 5, Sep 1992 , Page(s):683 – 697, Digital Object Identifier
213-228. 10.1109/72.159058.
[2] D. Hand, H. Mannila, P. Smyth: Principles of Data Mining. MIT Press, [13] Institute of Information Technology, UST Bannu, N-W.F.P, Pakistan.
Cambridge, MA, 2001. ISBN 0-262-08290.
51
Classification in Associated Patterns of Stock Data

Khairullah khan, Nasir Gul, Warda Abbasi
Institute of Information Technology, University of Science & Technology Bannu. N.W.F.P. Pakistan
Khairullah_k@yahoo.com, gulnasir@hotmail.com, sagittarius_icit@hotmail.com
Abstract — Stock Data Mining has a great potential in Impacts of Faulty Stocking Strategies:
business and trading decision support system. The evaluation of
Most warehouse networks have built-in barriers. Overcoming these,
inventory data to identify the reason of dead stock,
create needless costs and diminishes the service proficiency of the
slow-moving, and fast-moving products is a useful mechanism
network which they provide to their customers. The following list
for business support, investment and surveillance. In this paper
contains five obstacles that distribution executives can gauge
we have proposed an algorithm for mining patterns of huge
•
against their own warehouse networks [5] [7].
stock data to predict factors affecting the sale of products. In
•
first phase, we divide the stock data in three different clusters Ignoring Cost
•
on the basis of sold quantities i.e. Dead-Stock (DS), Mistaken Territories Increase Costs
•
Slow-Moving (SM) and Fast-Moving (FM). In the second phase Too Few Classes/ Varieties
•
we have proposed Most Frequent Pattern (MFP) algorithm to Wrong Emphasis
find frequencies of property values of the corresponding items. Dead Inventory
MFP provides frequent patterns of item attributes in each
category of products. From the result we conclude that why an II. MOTIVATION AND RELATED WORK
item falls in a certain category.
In most of the data mining research, it is assumed that some patterns
Index Terms — Dead-Stock (DS), Slow-Moving(SM), can be mined from the history of transactions as it likely that may
Fast-Moving (FM), Most Frequent Patterns (MFP), clustering, persist in future. These can help to predict the customer behaviour
K-Means for decision making. Some of the earlier works include, Mannila et
al. (1994) [8], Agrawal and Srikant (1994) [9], Gebouw et al. (2004)
I. INTRODUCTION: [10], Brijs et al. (2000) [11], Raymond Chi-Wing Wong (2005)
[12]. Long et al (2008) [13].
It is easy to turn cash into inventory, but the challenge is to turn
inventory into cash. Effective inventory management enables an Association rules plays key role in prediction. But it is difficult to
organization to meet or exceed customer’s expectations of product understand the association rule generated by the formal methods of
availability while maximizing net profits and minimizing costs [1] association. Therefore further processing is needed for finding
[2]. Stock data mining has a potential to provide trading decision simple patterns. We have proposed MFP matrix which is simple and
support and market surveillance. Data mining techniques like needs no further refinement.
clustering and associations can be used to find meaningful patterns
for future predictions [3] [4]. Clustering is used to generate groups III. DATA MINING
of related patterns, while association provides a way to get
generalized rules of dependent variables [5]. Patterns from a huge
Data Mining is the key element of the Knowledge Discovery in the
stock data on the basis of these rules can be obtained. The behavior
Databases task (KDD) [5]. KDD can be defined as “the process of
in terms of sales transaction is significant. The general term used for
identifying valid, novel, potentially useful and understandable
such type of analysis is called Market Basket Analysis [5].
patterns in data”.
Typically there are lot of different items, placed in a market for
selling, in which some of the product will be fast selling items, some
Data mining software is one of a number of analytical tools for
will be slow selling items and some will be dead stocked i.e. rarely
analyzing data. It allows users to analyze data from many different
selling items. We consider a scenario of super store or super market.
dimensions or angles, categorize it, and summarize the relationships
identified. Technically, Data Mining is the process of finding
Decision making in business sector is considered as one of the
correlations or patterns among dozens of fields in large relational
critical tasks. There is study for data mining for inventory item
databases [5]. Data Mining is not a field in itself; it is more a
selection with cross selling considerations which is used for
collection methods of data analysis coming from different fields of
maximal-profit selling items [6]. But our problem is finding out the
computer science, artificial intelligence and statistics [3].
selling power of the products in the market. This is a useful
approach to distinguish the selling frequency of items on the basis of
the known attributes. E.g. we can examine that a “black coat of IV. ASSOCIATED PATTERNS OF STOCK DATA
imperial company in winter season at Karachi has high ratio of
sale”, here we have basic property related to this example, i.e. color, Our proposed approach is a two phased model. First we
type, company, season, and location. Similarly we can predict that
generate clusters using K-Mean algorithm, and then FPP is
certain products of certain properties have what type of sale trends
in different locations. Thus on the basis of this scenario we can
designed for counting frequencies of items under their
predict the reason of dead-stock, slow-moving and fast moving specified attributes. First we are going to explain phase-1.
items. Data mining techniques are best suited for the analysis of The block diagram of the whole process is given in figure 1.
such type of predictions. A. Phase one
A. Effects of Items Properties: As discussed earlier in the first phase of our process, we have used
Store location, season, color, design, quality, company and prices K-Mean [14] clustering technique to classify the products in three
are the key attributes which affect the sale of products. Our groups.
algorithm takes into account these attributes for the evaluation of
sale of products.
52
This procedure receives Data sets from clusters. The first loop scans
OLTP
Databa
all the records of the data set. The inner loop counts occurrences of
the attribute for a given item and placed in the MFP matrix. Finally
Clustering maximum occurrences with-in a row gives a single pattern. On the
K-Means
basis of these patterns, we can predict that why a certain product
falls in particular category.
From the first row of Table 1, it is clear that white coat of female in
Cluster 1 Cluster 2 Cluster 3 winter has high ratio of sale. Similarly in the second row it can be
SM FM DS seen that black shoes of female in summer has high ratio of sale and
so on.
TABLE 1
MFP Matrix
Extract
(FPP) Colors Gender Season
Display
Item
Summer
Autumn
Female
Winter
Spring
end
White
Black
Male
Max
Max
Max
Red
MFP
Coat-White-F-Winter
Fig. 1 Block Diagram
Cluster 1: Dead-Stock: This cluster contains record of those
Winter
products which have small selling quantity. Every company has
White
Coat
dead inventory (or at least sick and dying inventory). It’s a natural
outgrowth of being in business for any length of time. As a matter of 2 5 3 3 7 F 5 0 2 3
fact, any successful product such as start decreasing in sale volume
when exact replicas with low cast are introduced into the market
Shoes-Black-F-Summer
which soon supercedes the existing stock.
Cluster 2: Slow-Moving: This cluster contains records of those

products which have medium selling quantity.
Summer
Cluster 3: Fast-Moving: This cluster contains records of those
Shoes
Black
products which have large selling quantity.

3 4 5 4 8 F 3 4 3 2
Pseudo code for FPP: Let we have set X of N
items in a Dataset having set Y of attributes.
This algorithm counts maximum of each
Sweeter-White-M-Spring
attribute values yij for each item in the dataset.
Input: Datasets (DS)

Output: Matrix
Frequent Property Pattern (FPP):
sweeter
FPP (DS)
Spring
white
Begin 1
for each item Xi in DS 5 6 7 0 8 M 5 2 6 5
a. for each attribute

i. count occurrences for V.CONCLUSION
Xi
C=Count (Xi)
ii. Find attribute name of We looked at the use of Data Mining for identifying stock data
C patterns using clustering and Most Frequent Pattern algorithm. Our
Mi=Attribute (Ci) contribution here was to formulate most frequent pattern of products
next [End for inner loop] using their known properties in inventory system. We identified the
b. Find Most Frequent Pattern trends of selling products through their known attributes. Our
i. MFP=Combine(Mi) technique is simple by using matrix and counting of attribute values.
next [End for outer loop]
Some of the limitation of study includes that it requires proper data
B. Phase Two with required attributes.
In this phase our proposed algorithm FPP will be used to generate a
property matrix containing counted values of corresponding
properties of each product.
53
REFERENCES
[9] Agrawal and Srikant “Fast Algorithms for Mining Association Rules”
1994
[1] Abubakar, Felix “Customer satisfaction with supermarket retail [10] Gebouw D, B-3590 Diepenbeek, Belgium “Building an Association
shopping” 2002 Rules Framework to Improve Product Assortment Decisions” 2004
[2]. Marcello Braglia, Andrea Grassi, Roberto Montanari “Multi-attribute
classification method for spare parts inventory management” 2004 [11] Brijs, Bart, Gilbert, Koen, Geert “A Data Mining Framework for
[3] By Terry Harris,”Optimization creates lean green supply chains” 2008 Optimal Product Selection in Retail Supermarket Data: The
[4] Matt Hartely “Using Data Mining to predict inventory levels” 2005 Generalized PROFSET Model” 2000
[5] Jiawan Han, Micheline Kamber “Data Mining Concepts and Techniques” [12] Raymond Chi-Wing Wong, Ada Wai-Chee Fu, Ke Wang”Data Mining
[6] Lu Frans, Wei, Paul, “Towards an agent based framework for online after for Inventory Item Selection with Cross-Selling Considerations” 2005
sales services” 2006 [13] Long Cao, Chao Luo, Jiarui Ni, DanLuo, Chengqi Zhang “Stock Data
[7] Rules Marco-Antonio Balers, Fernando Berzal, Juan-Carlos Cubero, Mining through Fuzzy Genetic Algorithm” 2008 - Vol. 4,
Eduardo Eisman, Nicolás Marín “Discovering Hidden Association” No.3/4 pp. 225 – 238
2004 [14] P.Thomas, Macredie “Knowledge Discovery and Data Mining” 1999
[8] Mannila et al. “Mining Association Rules” 1994
54
A Framework for Agent-based Earthquake Management System

Mukaila Rahman1, 3, Sarmad Sadik1, Arshad Ali1, H. Farooq Ahmad1, 2, Hiroki Suguri2
1
NUST School of Electrical Engineering and Computer Sciences (NUST-SEECS), Pakistan
2
Communication Technologies, Sendai, Japan
3
Department of Computer Sciences, University of Lagos, Nigeria
{rahmalade, sarmad, drfarooq, drarshad}@niit.edu.pk, {farooq, suguri}@comtec.co.jp
Software Agents have proved to be valuable resources as deployed
Abstract — Earthquake ranks among leading chronic in many critical mission projects [6].
natural disasters to manhood. The devastating effects of this
natural phenomenon in the recent past have raised more The aim of this work is to design a MAS where intelligent agents are
concerns considering the amount of efforts (both human and deployed in monitoring, information sharing and data management
material) usually expended before, during, and after its in the event of earthquakes. This system would also support
occurrences. Although many strategies are usually employed in coordination and timely triggering of emergency services faster than
tackling its horrific anticipated incidence, we believe humans. First, we make a survey of previous work in this domain
agent-based approach could be of tremendous assistance. The and see areas of difference where our proposed system produces
focus of this work is to develop a multi agent system technique encompassing and more efficient services. Next we describe our
which we consider as an efficient option for solving the current proposed system architecture and explain the roles of agents
problems. Here, we present a framework of our proposed involved. Likewise, a generic description of the workflow between
earthquake management system (EMS) where agents are these agents is highlighted. We conclude with a summary of
deployed at all levels of the relief efforts starting from sensing of contributions and direction for future work.
hazard to provision of most suitable relief assistance to the
concerned individuals. Apart from being efficient in terms of
real-time response in such distributed and dynamic situation, II. RELATED WORK
the system would be web-enabled, autonomous and invariably
reduce both cost and efforts usually expended. The concept of teamwork being championed by agents especially in
disaster management has been reported in [7]. However, the work
Index Terms — Multi-agent system, disaster management, was oriented towards generic emergency events like fire and
information sharing explosion. [8, 9] have applied MAS for planning and coordination in
relief operation. In terms of emergency relief response, works in this
I. INTRODUCTION regard include Crisis Information Management Software [10] and
an HLA-Based Multi-Agent System for Optimized Resource
There are many terrible and devastating natural occurrences in Allocation [11]. Yet, all these systems have no provision for active
human history. Earthquake is a singular most dreadful of them all as decision making based on prevailing situations. More importantly,
evidenced despite the advancement in science and technology. For one thing common with these works is that they are most effective in
instance, the 1976 Tangshan earthquake was officially reported to post earthquake operations and management. Our work focuses on
have claimed about 255000 lives even though experts believed the use of agent technology to produce autonomous and more
twice to thrice such number were lost, with about 800000 people efficient system in earthquake crisis management. More
injured [1]. Similarly, lately in 2004 in the Great Sumatra-Andaman importantly, the system benefits from the autonomous and adaptive
Earthquake [2], about 230000 loss of lives were accounted for. nature of agents for effective performances.
Besides these astonished human loss, many billions of dollars worth
of property also perished along with them. Thus, it is imperative for
more combating efforts aimed at it occurrences and relief operations. III. PROPOSED SYSTEM ARCHITECTURE
This has resulted in research efforts directed towards disasters
management in general, and earthquakes in particular such that the We conceptualized a system that automatically and efficiently
devastating effect after occurrences could be reduced [3, 4]. Another provides services like alerting, coordinating and collaborating in
similar project is disaster management and coordination earthquake crisis. Five major software agents involved in our
organization as reported in [5]. However, these available system are illustrated in Figure 1. These are Central Service Agent
approaches have proved to some extent inadequate in combating (CSA), Field Service Agent (FSA), Personalized Service Agent
this natural debacle. Major challenges facing these works are (PSA), Emergency Service Agent (ESA), and Personal Assistant
unpredictable incidences and untimely relief operations. The delay (PA).
in providing these urgent services is largely due to improper means
and channel of information flow. In a typical earthquake A. The agents and their roles in the EMS
management, there is the need for providing means for earthquake We describe the roles of these agents as follows:
detection, alerting of appropriate relief agencies and potential
victims of occurrences and subsequent timely report of occurrences. 1) Field service agents (FSAs)
Our encompassing EMS consists of various field stations located in
many areas where sensors detect and continuously send seismic These agents are situated in different locations and coordinated by
readings to a central station. This station, which coordinates all the the CSA. FSAs are responsible for detecting waves range with the
system’s activities, sends prompt directives and information to both help of the seismographs and continuously update the Central
the relief institutes and humans in the anticipated areas once the Service Agent. They alert the CSA of earthquake occurrence if the
fields reading tallies with a benchmark figure. Each principal actor reading reaches a benchmark
in the system is to be coordinated by a software agent. Agents are
loosely coupled entities that can act on human behalf. Thus, they 2) Central service agent (CSA)
possess intelligent features that include ability to communicate, It coordinates the readings from the FSAs. Based on received data, it
learn, be proactive, and even act autonomously. A community of identifies potential disastrous area and informs the Emergency
agents coming together to tackle problems too complex to be solve Service Agent (ESA) accordingly. It also consults the database on
individual agent is referred to as Multi Agent System (MAS). past records for appropriate strategy.
55
Users’
Profiles Database
Personal
User 1 Assistant 1
Sensor 1
Field Service
Agent 1
User 2 Personal Personalized
Assistant 2 Service Agent
Central Service
Agent
Sensor 2
Field Service
Agent 2
User n Personal
Assistant n
Emergency
Service Agent
Sensor n
Field Service
Relief Relief Relief Agent n
Institute 1 Institute 2 Institute n
Fig. 1 EMS architecture
3) Emergency service agent (ESA) necessary reporting. These queries include death toll,
It coordinates all the activities of the emergency services (police, hospitalization, property loss and missing people information. This
health, fire and other relief organizations). It alerts and gives report also subsequently assists in future operations. Interaction
appropriate instructions to the PSA based on information and between EMS agents is largely through the CSA. Agent
directives from the CSA. Communication Language (ACL) is adopted for this interaction. An
ACL specifies the sender, receiver, and the type of action to be
4) Personalized service agent (PSA)
performed as well as the arguments (or parameters) of such action.
This agent sent alert and relief information to the users through the
Personal Assistants (PAs). The PSA, which possesses user’s
profiles also models their interaction with the system and provides Apart from the coordination and prompt triggering of emergency
personalized services to each user. These services include language services which the system supports, the role plays by the
of communication and means of information delivery (textual or personalized service agent go a long way in enhancing the system,
graphical) based on user’s literacy level. Information from the PA making it adaptive and efficient. This agent as described above is
for necessary updating and verification is also received by the CSA saddle with responsibilities such as identification and classification
through the PSA. of users; acquisition of user profiles during interaction;
personalization of system services. User’s alert-service varies along
5) Personal assistant (PA)
language of communication and means of information delivery.
This acts as a personal assistant to humans. Messages from the PSA
This may be in form of text or graphics (based on education level).
are transmitted through the PA to user for necessary emergency
Also, the way to alert a hypertensive user of imminent earthquake
actions. Likewise, any observed and/or perceived report can also be
has to be different from others, taking into consideration individual
reported to the CSA through the PSA. This PA can be deployed on
health history.
PDAs, mobile phones, and other ubiquitous devices.
The Database (DB) contains records of past occurrences and IV. CONCLUSIONS AND FUTURE WORK
strategies employed as well as latest update as sent in by the CSA.
B. Generic Description of the workflow between the Agents We describe an Agent-based system which is capable of detecting,
The FSA detects wave range, sends the readings to the CSA and alerting and triggering various relief institutes to combat and
alerts it depending upon the benchmark reading of the seismograph. manage earthquake crisis. As shown in section 3, agent-oriented
The CSA analyses the data consulting the database for strategy to be decomposition is suitable for partitioning complex systems that
adopted. If earthquake is confirmed, the ESA is alerted for prompt cater for disaster detection, emergency response and management.
actions. The ESA coordinates the relief organizations like police, Our proposed earthquake management system comprises five
fire, and ambulances. Based on the information and directives from different agents being coordinated by the central service agent.
the CSA, appropriate instructions are given to these organizations. Although a relatively simple framework has been described, there
Likewise, the PSA is informed of the development between the are technical challenges which involve the development of
initial starting and big scale waves as defined by the readings. The supervised and unsupervised learning component of the
PSA sends personalized information to the particular individuals personalised service agent especially in modelling user’s interaction
through the PA. The PSA also passes observed and/or perceived with the system. We have commenced work on the user modelling
report from the PA to the CSA. The PA passes directives and subsystem and the results are encouraging. Both the stereotypical
expectations from the PSA to the individuals. After the operation, information acquisition and the analysis of user-system interaction
the ESA produces report of events that transpired to the CSA for history construction models are being considered in this subsystem.
necessary updating of the Database. With these and previous The integral properties of the subsystem include reasoning and
information in it, the Database is queried in analysing the event for learning. Subsequently, a prototype system of the EMS will be
implemented.
56
V. ACKNOWLEDGEMENT [5] US Earthquake Hazards Program. [Online]. Available:

http://earthquake.usgs.gov/
M. Alade Rahman from the University of Lagos, Nigeria thanks the [6] Michael Wooldridge, An Introduction to Multi-agent Systems, John
Academy of Sciences for the Developing World (TWAS) Italy, and Wiley & Sons Press, 2002.
the National University of Sciences and Technology (NUST) [7] M. Tambe, E .Bowring, H. Jung, G. Kaminka, R. Maheswaran, J. Marecki,
Pakistan, for the ongoing one-year research fellowship sponsorship P. J. Modi, R. Nair, S. Okamoto, J. P. Pearce, P. Paruchuri, D. Pynadath,
at NUST School of Electrical Engineering and Computer Sciences P. Scerri, N. Scerri, N. Schurr, P.Varakantham. “Conflicts in teamwork:
Hybrids to the rescue”, In Proceedings of the Fourth International Joint
(NUST-SEECS), Pakistan.
Conference on Autonomous Agents and Multiagent Systems
(AAMAS), 2005,pp.3-5.
REFERENCES [8] Lucian Vlad Lita, Jamieson Schulte and Sebastian Thrun. “A MultiAgent
System for Agent Coordination in Uncertain Environments”,
Proceedings of the fifth international conference on Autonomous
[1] Historic Earthquakes. “Tangshan, China”, U.S. Geological Survey. agents, 2001, Montreal, Quebec, Canada, pp. 21 – 22.
[Online]. Available:
http://earthquake.usgs.gov/regional/world/events/1976_07_27.php [9] Wei Chen and Keith S. Decker, “Managing Multi-Agent Coordination,
[2] Lay, T., Kanamori, H., Ammon, C., Nettles, M., Ward, S., Aster, R., Planning and Scheduling”, AAMAS,2004, pp. 1360-1361.
Beck, S., Bilek, S., Brudzinski, M., Butler, R., DeShon, H., Ekström, [10] Ashcroft, J., D. J. Daniels, and S. V. Hart. 2002, “Crisis Information
G., Satake, K., Sipkin, S., “The Great Sumatra-Andaman Earthquake of Management Software (CIMS) Feature Comparison Report”, NIJ
December 26, 2004”, Science, 308, 1127–1133, 2005. Special Report 197065, U.S. Department of Justice, National Institute
[3] M. Scalem, Somprakash Bandyopadhyay, Sushants Sinha, Ashok K. of Justice, Washington, DC. [Online]. Available:
Sircar. “Decentralized Disaster Management Information Network www.ncjrs.gov/pdffiles1/nij/197065.pdf
(DDMIN): Research Discussions on the inherent Technological and [11] Frank Fiedrich, “An HLA-Based Multiagent System for Optimized
Strategic implementation issues and Proposed Solutions”, draft for Resource Allocation after Strong Earthquakes”, Proceedings of the
CISTM 2005. [Online]. Available: Winter Simulation Conference, 2006. L. F. Perrone, F. P. Wieland, J.
www.iimcal.ac.in/research/adhocnet/Papers/56.pdf Liu, B. G. Lawson, D. M. Nicol, and R. M. Fujimoto, eds., pp. 486-492.
[4] Earthquake Management. [Online]. Available:
www.earthquakemanagement.net/
57
Intelligent Information Retrieval

Jibran Mustafa, Sharifullah Khan, Khalid Lateef
NUST School of Electrical Engineering and Computer Sciences
Chaklala, Rawalpindi, Pakistan
{jibran.mustafa, drsharif, Khalid.latif}@niit.edu.pk
Abstract — Semantic-based information retrieval [8]. We maintain source model that contain information about
understands the meanings of the concepts that users specify in sources (i.e. digital documents). Our source model consists of two
their queries. We presented an ontology driven information components: content metadata that contains metadata of contents
retrieval framework to improve the precision of searching and source metadata contains information about sources in the form
results. We adopt thematic similarity approach for information of RDF. RDF triple matching is performed by semantic matcher.
retrieval in this paper in order to concentrate on the context Documents containing relevant RDF triples are passed to the ranker
aware use of a particular word (concept) to improve the to display the results to the user according to the relevancy of
precision of search results. We maintain metadata information documents. Moreover different indexing techniques are used to
of source(s). We employ RDF triples instead of keywords in the enhance the performance of semantic matcher.
available metadata and probe user queries for capturing the
context of the keywords. This improves the information The remainder of this paper is structured as follows: Section 2
retrieval precision. describes the overview of RDF language and its terminologies.
Section 3 reviews current approaches to compare different semantic
Index Terms — Information Retrieval, RDF, search techniques and their proposed systems. Section 4 explains
Semantic-based, Thematic Similarity, Context-aware our proposed information retrieval framework and its different
components. Section 5 presents the semantic similarity approach to
carryout similarity calculation between concepts and relations.
I. INTRODUCTION Section 6 describes our methodology to carry out semantic search.
Section 7 discusses example to demonstrate our methodology
Different Information retrieval (IR) techniques exist to improve whereas section 8 presents our conclusion.
search efficiency and performance. These can be classified as:
traditional keyword and semantic information retrieval techniques.
The keyword information retrieval techniques perform keyword II. RESOURCE DESCRIPTION FRAMEWORK
searching in documents by matching the string (i.e. characters) of
keywords that users specify in their queries. In addition the systems, In this section we describe the Resource Description Framework
using the technique, maintain a word index to accomplish searching and its terminologies as in rest of the paper we will use it in our
[20, 21], Google search engine is the most famous example of such methodology. A Resource Description Framework (RDF) is a
systems. Although these systems spend less time in searching standard for representing metadata information that can be
results because they do not have the ability to understand the identified using a Universal Resource Identifier (URI) [7]. To
meanings of keywords (i.e. semantics). Moreover, digital describe metadata, RDF statements are expressed in triples: subject
repositories are maintained and managed independently according (represented by a URI or a blank node), predicate or property
to users own requirements in distributed environment. In these (represented by a URI) and object (represented by a URI, a blank
repositories, different documents containing same information may node or a literal). This triple can effectively modeled as directed
be represented differently (i.e., semantic heterogeneity) that makes graphs.
it more difficult to understand semantics of the keywords [2, 6, 17,
18]. The most important semantic heterogeneity issues are: synonym Predicate
Subject Object
and polysemy. A synonym is a word that means the same as another
word, for instance, author is a synonym of writer. A polysemy is a Fig. 1 RDF triple as directed graph
word or phrase with multiple, related meanings, for instance fan can As shown in Fig. 1, the subject and object of the triple are modeled
be used to refer electrical device in one context and a sports fan in as nodes and the predicate as a directed link that describes the
other [2, 17, 18]. relationship between the nodes.
In semantic information retrieval technique, searching is performed The direction of the link always points towards the object. A URI is
by interpreting the meanings of keywords (i.e. semantics). The a more general form of Uniform Resource Locator (URL). It allows
systems, using the technique, have higher results’ precision than, the information about a resource to be recorded without the use of a
systems, using the keyword information retrieval techniques specific network address. A blank node is used when a subject or
because of the semantics of the keywords. However, in semantic object node is unknown. It is also used when the relationship
information retrieval, improvement in the recall is not guaranteed between a subject and an object node is n-ary (as is the case with
(i.e., either lower or higher) [3, 6]. The semantics of keywords are RDF containers). A literal is basically a string with an optional
identified through the relationships between keywords in an language tag. It is used to represent values like names, dates, and
available knowledge base (i.e., ontology) by performing semantic numbers. A typed literal is a string combined with a data type,
similarity on them [1, 2, 3, 6, 8, 9, 10, 16]. whereas the data type is always a URI, for example “25”
^^http://www.w3.org-/2001/XML Schema#int.
In this paper we adopt thematic similarity approach for information
retrieval to improve the precision. We propose Ontology Driven
III. RELATED WORK
Information Retrieval Framework to carry out search in digital
documents. In our proposed framework we are focusing on
Resource Description Framework (RDF) triples instead of The state of the art semantic based IR system compute semantic
keywords. A user submits the query in the form of RDF triples, this similarity by mapping concepts to domain ontology and by
query is expanded through synonyms and other relationships (i.e. examining their relationship in that ontology. Different semantic
hyponym, meronym)using existing semantic similarity techniques similarity methods are categorized in [7] as: 1) Edge counting
methods measures the similarity between two concepts as a function
58
of the length of the path (i.e. distance) relating the concepts (also C. Semantic Matcher
called distance base approach). 2) Information content methods In the framework, semantic matcher is the most important
measure the difference in information of the two concepts as a component through which RDF triple matching is performed. In
function of their probability of occurrence in a corpus and Semantic matcher “rule-bases” are created to deduce inference from
commonly called term frequency, Inverse document frequency existing RDF data. A rule is an object that can be applied to deduce
(tf.idf). 3) Feature based methods measures the similarity between inferences from RDF data. It has a name, antecedent, filter
two terms either as a function of their properties. condition (optional) and consequent. Rule-base is an object that
contains different rules. Moreover a rule-index is created to enhance
A number of systems [1, 2, 3, 4] have been developed to improve the performance of rule-bases. The rule-index contains
either precision, recall or both of search results. In [3] ontology pre-computed triples that can be inferred from applying a specified
navigation was used to annotate the relevant results semantically. set of rule-bases [7].
The results produced by this search engine improved the recall.
Semantic search framework proposed in [2] improved the precision D. Query Reformulator
and recall as compared to existing systems. In this framework Query reformulatoer expand RDF query with synonym and other
traditional term document matrix was extended in three steps to relationships such as hyponym (i.e. Is-A relationship) and Meronym
carry out reasoning between domain ontology. Semantic Similarity (i.e. Part-of) [6] using distance based approach [10]. Then the query
Retrieval Model (SSRM) proposed in [1] improved precision along is rewritten for these expanded terms to pass it to the semantic
with recall as compared to other systems [2, 3]. Initially, weights are matcher in RDF format.
computed by extending concepts through synonyms. Second,
weights are recomputed by examining neighborhood of each E. Ranker
concept. Finally, document similarity is computed using td.idf To retrieve documents that are related to user's query and rank them
similarity method. according to their importance, the relevance between documents
and triples must be measured. The relevance of the documents is
The existing semantic based IR systems extend concepts through computed extending tf. idf weighting scheme [14] for triples instead
synonym to handle semantic heterogeneity (i.e. synonymy), for of keywords. Let N be the total number of documents and ni the
instance, looking only for word writer may have fewer results than number of documents in which the triple ti appears.
tf ij =
looking for writer and author. However the existing techniques freq ij
(systems) do not interpret the context of the concepts and their (1 )
relationship (i.e. polysemy). For example, interest for a bank and max i ( freq ij )
interest in a book have different meanings. As a result the precision Let freqij be the raw frequency of the triple ti in the document dj.
of search results of these systems decreases. Then the normalized frequency of tfij of the triple in dj is where the
maximum is computed over all triples mentioned in the docements
dj. Now let idfi be the inverse document frequency for ti given by:
idf = log(
IV. INFORMATION RETRIEVAL FRAMEWORK
N
) (2)
ni
We have proposed an Ontology Driven Information Retrieval
Framework, as shown in fig. 2, to improve the precision of search
results using thematic similarity. We employ RDF triples instead of
Wij = tf ij × idf i
keywords matching in order to concentrate on the context of the The final tf. idf weight of triple i to document j is calculated as :
word to tackle with different semantic heterogeneity problems. Our (3)
proposed framework has following components: Crawler, Source
Model, Semantic Matcher, Query Reformulator and Ranker. The triple similarity is calculated by measuring semantic similarity
of concepts and their relationships.
A. Crawler
The crawler extracts metadata (in the form of RDF triples) from
V. SEMANTIC SIMILARITY
digital documents residing in the documents repository. The crawler
loads these triples into the source model. The crawler keeps on
updating the information about the documents to maintain the To interpret the keywords with respect to the context it is being used
source model updated. The focus of this research in on semantic requires to obtain similarity between concepts and relationships (i.e.
searching so, we will not discuss crawler techniques here. A number thematic similarity). We focus on thematic similarity by matching
of techniques with efficient algorithms are proposed in [11, 12, 13, RDF triples to concentrate on both aspects together.
14] that can be extended to crawl metadata information from
documents, but in this paper we focus on semantic based A. Concepts Similarity
information retrieval. Concepts similarity is measured by calculating the distance between
them [10]. The distance is calculated between different concepts
B. Source Model from concepts position in the hierarchy. The position of a concept in
Source Model maintains metadata information about sources (i.e. a hierarchy defined in [8] as follows.
digital documents). Our source model has two components: source
metadata and content metadata. Source metadata contain
milestone ( n ) =
information (i.e. author, URI, title, research area etc) about sources. 1/ 2
They help in identifying relevant sources in order to avoid querying (4)
all available sources. Content metadata contain metadata of contents k l (n)
of the documents in the form of RDF triples. The simplicity source
model using RDF structure promotes interoperability across
Where ‘k’ is a predefined factor larger than 1 that indicates the rate
applications, and its machine- understandable format facilitates the
at which the value decreases along the hierarchy, and l(n) is the
automated (intelligent) query processing [8].
depth of the node ‘n’ in hierarchy. For the root of a hierarchy, l(root)
= 0. For any two concepts in the hierarchy, have closest common
59
Query Reformulator
Ranker Semantic Matcher Domain Ontology
Source
Documents Crawler Model
Repository
Fig. 2 Ontology Driven Information Retrieval Framework
parent (ccp). The distance between two concepts c1, c2 and their ccp matched relation of query RDF triple q and source RDF triple S
will be determined by their closest common parent as follows: respectively. sim ( q , s ) is the overall similarity between query
d c (c1 , c 2 ) = d c (c1 , ccp) + d c (c 2 , ccp)

(q) and source (S) RDF triples. Here i and j represent ith and jth
(5) subject or object or relation of query and source RDF triples
d c (c1 , ccp) = milestone(ccp) − milestone(c1 )

respectively.
(6)
VI. PROPOSED MATCHING METHODOLOGY
Thus, the similarity calculation between two concepts, c1 and c2, as
follows:
sim c (c1 , c 2 ) = 1 − d c (c1 , c 2 )

In this section, we discuss our approach that performs the semantic
(7 ) search by matching RDF triples. In the methodology we used oracle
network data model (NDM) where metadata is stored in triples
There are some exceptions that if the concept c1 is subclass of the consistently. The index has been used to enhance the performance
concept c2 or c1 and c3 are synonym or acronym of each other, the of matching. In our semantic search framework domain ontology is
distance will be set to 0, i.e. the similarity between these two also defined in terms of RDF triples. Fig. 3 illustrates an example of
concepts will be 1. We think it is reasonable because the subclass is the domain ontology.
always a kind of super class. We think synonym and acronym
relation between concepts are at the same level.
@prefix rdf:< http://www.w3.org/1999/02/22-rdf-syntax-ns#>.
@prefix rdfs:< http://www.w3.org/2000/01/rdf-schema#>.
B. Relations Similarity
@prefix uri:< http://www.niit.edu.pk/research/>.
Likewise, the similarity between two relations is defined as uri:Person a rdfs:Class.
follows:
simr (r1 , r2 ) = 1 − d r ( r1 , r2 )
uri:Article a rdfs:Class.
uri:ResearchPaper a rdfs:Class; rdfs:subClassOf :Article.
(8) uri:writes a rdf:Property.
The distance between two relations is also calculated by their :writes rdfs:domain :Person;
respective positions in the relation hierarchy. The only difference is rdfs:range :Article.
that the relation hierarchy is constructed manually by us. Fig. 3 N3 notation of the domain ontology
There are some exceptions that if the relation r1 is subclass of the
relation r2 or relation r1 and r2 are synonym or acronym of each other A user submits a query in the form of RDF triples, which is passed to
than the distance will be set to 0, consequently the similarity the query reformulator, where the query expansion using existing
between these two relations will be 1. In order to reduce semantic similarity techniques (as discussed in the previous section)
computational complexity we consider relations similarity 0 for all are performed. Then the query is rewritten for these expanded terms
other cases. Hence, the relation similarity is defined as follows: to pass it to semantic matcher in RDF format. Semantic matcher
performs matching with the help of metadata information available
simr (r`1 , r2 ) = 10,,

 in the source model. The source metadata contains information
relation subsume each other (9)
 Otherwise
about sources to avoid querying all available sources. For example,
query about data integration (i.e. category = data integration) does
not require to search the sources containing information about
C. RDF Triples Similarity multi-agents (i.e. category = multiagent).
The user query and data source RDF triples, are matched to find
their similarity. The final triple similarity matching formula by A. System Accuracy
combining eq. 7 (for concepts similarity) and eq. 8 (for relations In the proposed framework, thematic similarity improves the
similarity) as follows: precision of information compared to existing systems. The focal
sim (q, s ) = ∏∏ simr (qri , s rj )

i j point is to understand the keyword with respect to the context
simsub (q sub , s sub ) (circumstance) it is being used to avoid irrelevant search results. We
n m
(10) illustrate the improvement in precision with an example in the next
i = 0 j =0
subsection.
i j
simobj (q , s )
obj obj
B. Example
Where qsub , qobj and Ssub , Sobj are matched concepts while qr and Sr In this example we use RDF graph notation to represent the RDF
triple segments in the respective sources. Suppose, the user enters
60
the following query ‘Find all worker(s) who has visited the HBL on Person
1/32
certain date’ and the corresponding RDF triple of this query is:
(?worker :visits :HBL) where date like ‘?date’
Business man 1/64
Employee
As shown in fig 4.
Manager Worker 1/128

Visits Bank
Worker
Fig 4 Query RDF graph 1/256
Finance Manager
Fig. 5 shows three sources from our repository containing content
metadata triples shown in fig 5(a), 5(b), 5(c) of source 1, source 2 Fig 6 Ontology segment concerning ‘finance manager’ and ‘business man’
and source 3 respectively. The similarity can be calculated between
query RDF triple and the source RDF triple, shown in fig 5(a) as
follows: simc(visits, visits), simc(HBL, HBL) is 1, as both are same, VII. CONCLUSION AND FUTURE DIRECTION
simc(worker, finance manger) is 0.01953125 (the fraction in fig 6.
shows the ‘milestone’ of certain ontology level). Now similarity In this paper, we presented an ontology driven information retrieval
calculation for the query triple and source triples shown in fig 5(b) framework to improve the precision of searching results. The salient
and 5(c) are: feature of this semantic search framework is that we concentrate on
the context aware use of a particular word (concept) to improve the
simc(visits, goesto) = 1, since both are synonym precision of search results. We employ RDF triple matching instead
simc(HBL, Habib Bank Limited) =1, acronym relation of keywords matching techniques. The subject, property and object
simc(worker, business man) = 0.0234375 of RDF triple enabled the search framework to concentrate on the
simc(worker, employee) = 1 combination of concept and relationship similarity of the keyword
simc(HBL, Indus) = 0, since both are not related at the same time. We maintain source model to store metadata in
RDF triples. The query submitted by the user is expanded with
synonym and rewritten for generated concepts to pass it to the
semantic matcher. The triple matching is performed by the matcher
“HBL”^^xsd:string and results are ranked by ranker according to the relevancy score.
#Financial
Institute In current framework we focused on unstructured data (i.e. flat file).
#Finance #visits In the future, we plan to augment the framework for other
Manager
heterogeneities (i.e. incomplete and incompatible triples).
#River
Edge “Indus”^^xsd:string
REFERENCES
(a) Source 1- Triple segment
[1] Giannis Varelas, Epimenidis Voutsakis, Paraskevi Raftopoulou,
“Habib Bank “Semantic Similarity Methods in WordNet and their Application to
Limited”^^xsd:string
Information Retrieval on the Web”, 7th ACM international workshop
on Web information and data management November 5, 2005.
[2] Wei-Dong Fang, Ling Zhang, Yan-Xuan Wang, Shou-Bin Dong,
#goesto
#businessman “Towards a Semantic Search Engine Based on Ontologies”, IEEE
Preceedings of the Fourth International Conference on Machine
#River Bank Learning and Cybernetics Guangzhou China, 18-21 August 2005.
“Indus”^^xsd:string [3] D. Bonino, F. Corno, L. Farinetti, A. Bosca, “Ontology Driven
(b) Source 2- Triple segment Semantic Search”, WSEASTransaction on Information Science and
Application, Issue 6, Vol. 1, pp. 1597-1605, December 2004.
#Financial
[4] Manuel Montes-y-Gomez, Aurelio Lopez-Lopez, Alexander Gelbukh,
#worksIn Institute “Information Retrieval with Conceptual Graph Matching”,
“HBL”^^xsd:string
DEXA-2000: 11th International Conference and Workshop on
#employee Database and Expert Systems Applications, Greenwich, England,
#visits September 4-8, 2000.
#River Bank [5] Jonathan Poole and J. A. Campbell, “A Novel Algorithm for Matching
“Indus”^^xsd:string
Conceptual and Related Graphs”, In G. Ellis et al eds, Conceptual
Structures: Applications, Implementation and Theory, Santa Cruz CA,
(c) Source 3- Triple segment USA Springer-Verlag, pp. 293-307, 1995.
[6] M. Andrea Rodriguez, Max J. Egenhofer, “Determining Semantic
Fig 5 Content metadata triples of three sources in RDF graph Similarity among Entity Classes from Different Ontologies”,
Knowledge and Data Engineering, IEEE Transactions, Volume: 15,
Based on the similarity measure for concepts and relations, we can
Issue 2, pp. 442- 456, March-April 2003.
calculate the similarity between two RDF triples. In this example, [7] Oracle USA Inc, RDF Support in Oracle USA, July 2005.
triple matching returns only three triples from respective sources [8] Jiwei Zhong, Haiping Zhu, Jiaming Li, Yong Yu, “Conceptual Graph
shown in fig. 5. and eliminates other irrelevant triples (i.e. Matching for Semantic Search”, Proceedings of the 10th International
simc(HBL, Indus) = 0), hence the precision of search results have Conference on Conceptual Structures: Integration and Interfaces table
been improved by concentrating on the context aware use of of contents, Springer-Verlag London, UK, pp. 92-196, 2002.
particular word. [9] Qazi Muddasar Illyas, Yang Zong Kai and Muhammad Adeel Talib, “A
Conceptual Architecture for Semantic Search Engine”, 9th IEEE
International Multi topic conference Pakistan, 2004.
[10] S. Khan and F. Marvon, “Identifying Relevant Sources in Query
Reformulation”. In the proceedings of the 8th International Conference
on Information Integration and Web-based Applications & Services
(iiWAS2006), Yogyakarta Indonesia, December 2006.
61
[11] Blaz Novak “A Survey of Focused Web Crawling Algorithms”, [17] Chen Yu Lee, Von Wun Soo, “Ontology based information retrieval
Proceeding of the Multi-Conference SIKD 04, 12-15 October 2004. and extraction”, 3rd International Conference on Information
[12] Matt Biddulph “Crawling the Semantic Web”, Proceedings of XML Technology: Research and Education IEEE, pp. 265-269, 27-30 June
Europe2004, 18-21 April 2004, Holland. 2005.
[13] S.Ganesh, M.Jayaraj, V.Kalyan, G.Aghila “Ontology Based Web [18] Michael Uschold, Michael Gruninger, “Ontologies and Semantics for
Crawler”. In the proceedings of the International Conference on Seamless Connectivity”, SIGMOND, vol. 33, December 2004.
Information Technology: Coding and Computing (ITCC’04), IEEE [19] J. Smith and D. Smith, “Database Abstractions: Aggregation and
Computer Society, 2004. Generalization,” ACM Trans. Database Systems, vol. 2, pp. 105-133,
[14] Ricardo Baeza-Yates, Berthier Ribeiro-Neto, Modern Information 1977.
Retrieval, Addison Wesley Publishing Co. Inc. 2001. [20] X. Li, F. Bian, H. Zhang, C. Diot, R. Govindan, G. Iannaccone.
[15] Tversky, A. “Features of similarity. Psychological Review”, Vol. 84(4): "MIND: A Distributed Multi-Dimensional Indexing System for
pp. 327-352, 1977. Network Monitoring". IEEE Infocom-06 Barcelona, April 2006.
[16] El Sayed, Ahmad Hacid, Hakim Zighed, Djamel, " A New [21] Alexandros Ntoulas, Gerald Chao, Junghoo Cho, ”The Infocious Web
Context-Aware Measure for Semantic Distance Using a Taxonomy and Search Engine: Improving Web Searching Through Linguistic
a Text Corpus", IEEE International Conference on Information Reuse Analysis”, International World Wide Web Conference Committee
and Integration, 2007 (IRI-07), pp. 279-284, 13-15 August 2007. (IW3C2) ACM, Chiba Japan, May 2005.
62
Towards Terrain Modeling For Robot Navigation Using A Single Camera

Muhammad Tufail, Abdul Bais, Muhammad Usman, Sheraz Ali Khan
NWFP University of Engineering and Technology
Peshawar, Pakistan
{tufail,bais,usman,sheraz.ali}@nwfpuet.edu.pk
Abstract — Discovery of resources is essential in any user resource-related queries. In later versions, although OGSA and
resource sharing environments, including Grids. Many of the WSRF frameworks have been incorporated and there are Grid
existing Grids facilitate resource discovery in a hierarchical services which index the resource information or there is a separate
and/or centralized manner. These approaches to resource web service associated with each resource, yet the main mechanism
discovery require environments with high level of trust and again revolves around centralized paradigm. Condor[10], again
centralized control of resources. As Grids are expected to be follows the centralized paradigm where there is a central manager
utilized in different domains in future, and especially in user which receives both the resource information and user requests and
oriented environments as envisioned by the PhantomOS performs the matchmaking to finally destine the user job to a
project, which endeavors to create a user oriented Grid particular node. LCG/EGEE [8,9] framework also use the MDS-2 of
operating system, these approaches to resource discovery are GT-2, thus using centralized and hierarchical GRIS and GIIS
proving to be inadequately scalable and unworkable, due to the services, and using BDII for caching Grid status information
lack of trust in user oriented peer to peer environments. In this collected from GIIS for fast user response. UNICORE [8,15]
paper, a decentralized peer-to-peer approach of resource software architecture also shows notions of client/server paradigm
discovery has been proposed that is based on a semi-super-peer to some extent and there is client which submits the job to the
architecture and is reasonably efficient considering the gateway which authenticates the client before contacting the
heterogeneity, scalability and hybrid structure of future UNICORE servers. If the job is to be executed locally, it is executed
generation Grid networks. by the native batch system, however if it is to be executed remotely,
it is sent to peer gateways.
Index Terms — Resource Discovery, Grid Resource
Discovery, P2P Resource Discovery Peer-to-peer environments are basically classified into two main
categories i.e. there are structured peer-to-peer systems and there
I. INTRODUCTION are un-structured systems. Structured peer to peer systems like
CAN[16], Pastry[17], Chord[18] use rigid structure for the
organizations of peers and thus highly structured overlay networks
For the last decade or so, Grid computing has emerged as a new are formed. An overlay network is actually a network which is built
form of distributed computing in which organizations located on top of the actual communication layer and is like a graph where
around the globe collaborate and dynamically share computing, vertices indicate nodes and the edges indicate that the nodes know
storage and network resources within virtual organizations spanning each other[19]. Structured peer-to-peer systems use the distributed
over multiple geographical locations [5]. Peer to Peer (P2P) is hash tables (DHT’s) for the storing of key-data pairs and thus
another form of distributed computing which has emerged over the facilitate efficient routing of query. Structured systems although
last few years and which also focuses on the coordinated use of scale well, but have drawbacks of high maintenance cost, and high
distributed resources located around the globe. As mentioned in [4] communication cost for up-to-date information [8]. They are also
although Grid and P2P systems have different evolutionary paths, not feasible for direct keyword query searches[8]. Hence resource
address different communities, resources, applications, yet they discovery techniques employed in traditional structured P2P
share a common objective i.e. resource sharing in large scalable systems is generally not considered suitable for Grid environments.
distributed environments, and it is believed that ultimately these two
domains are going to converge thus complementing each other’s In unstructured peer to peer systems, nodes and resources are made
strengths and weaknesses. available to the network without global overlay planning and peers
connect to each other somewhat in random fashion[8]. Gnutella[11],
Resource discovery is vital problem in distributed environments and KaZaA are the examples of unstructured P2P systems. Resource
is crucial for the efficient performance of any Grid system. As the discovery in Gnutella is mainly based on broadcasting i.e. flooding
Grids around the world increase in scale and complexity, resource algorithm, hence produces too much network traffic and congestion.
discovery mechanisms should also be modified so as to meet Various approaches as in [20]
performance and efficiency needs for the future Grid networks. have been proposed based on random walks, directed searches to
tailor flooding algorithm to minimize network traffic and increase
II. BACKGROUND network coverage.
To address the issue of performance deterioration as Grid scale in

There are various algorithms for resource discovery in distributed
size, to provide fault-tolerance, load balancing, self-organization,
environments, according to [14] the most prominent of them are
and performance efficiency, active research is being pursued
flooding algorithm, swamping algorithm, random pointer jump
nowadays to replace and design most of the existing Grid
algorithm, and namedropper algorithm. Out of these, only random
functionalities and services in a peer-to-peer fashion. Resource
pointer jump and name-dropper algorithm are suitable for Grid
discovery mechanism is one of them that need to be re-designed in a
environments because they usually produce less network traffic and
decentralized way to meet the above performance objectives and to
make less rounds to have information about the nodes.
address the enormous size of future generation Grid networks.
In most of the existing Grid middlewares, resource discovery is
being handled in a centralized and/or hierarchical manner [2, 6]. For III. RELATED WORK
example, in Globus Toolkit 2 [2, 6, 7, 8], nodes get registered at
some central index server, Grid Index Information Server, and GRIS Although research efforts have started already to apply peer-to-peer
periodically sends the resource status from the nodes to that server. and de-centralized approaches to resource discovery in Grid
GIIS servers are arranged in a hierarchy, and serve for answering environments but still these are research or experimental
63
implementations and no concrete framework has yet been proposed. middleware, scheduling primitives/algorithms may well be included
Notable works in this area are of Iamnitchi et. al. [1] and in the resource discovery technique to discover the optimal
Mastroianni et. al.[2]. resources with the minimum communication load expected, high
bandwidth and taking other relevant criteria into considerations.
Iamnitchi et al.[1] have proposed four axes of the solution space for
the resource discovery problem, i.e. Membership Protocol, Overlay The proposed technique does not particularly deal with the request
Construction, Preprocessing, Request Processing. Membership propagation/dissemination strategies or neighboring super-peers /
protocol tells that how new nodes will join the network and choose peers selection. Of course, a great variety of techniques [8,14,19] for
their neighbors, while overlay construction will be in a way that these aspects have been proposed in literature with little more or less
only a subset of nodes connect to each other. Pre-processing refers modifications. However, it does induces some modifications to the
to offline processing for example prefetching or dissemination of request propagation and neighboring super-peers / peers selection
resource descriptions and request processing tackles with the primitives as compared with the algorithm originally proposed by
request propagation strategies e.g. random walk, learning based or Mastroianni et al. [2]. In this respect, this proposed technique works
best neighbor. They have carried the experiments on an emulated on the semi-super-peer architecture which is an extension to the
Grid on the basis of mainly resource distribution, request super-peer architecture proposed by the Mastroianni et al. [2].
distributions and request propagation strategies and haven’t
explored the problem in detail on their proposed four axes. This proposed technique takes the notion of subgrid based
architecture. Each subgrid consists of neighboring machines.
Mastroianni et. al. [2] have proposed to tackle the resource Neighbor Machines in terms of network are not necessarily the
discovery problem in Grid environments on the basis of super-peer machines physically next to it, but those machines which have the
architecture in which there is one super-peer per virtual organization lowest network latencies between them. There are multiple subgrids
(VO) and super-peers co-ordinate with each other in a peer to peer within the Grid.
fashion. However, there are many drawbacks and limitations of their
proposed technique which have been identified in this work. The Resource Discovery within each subgrid basically operates in a
bottom line is that all the peer-to-peer resource discovery centralized fashion, as all the subGrid members can register their
approaches proposed for Grid environments only deal with the resources with the most powerful machines within the subgrid,
current structure of VO and do not consider the scalability of future called super-peer (also termed as master node, or ultra-peer in
generation Grid networks which this technique tries to serve. literature). When a node of a subgrid needs a resource, it sends a
query request to the super-peer which in turn sends the list of
IV. PROPOSED SCHEME AND ARCHITECTURE resources matching the user query constraints if it has
registered/available with it. If the super-peer does not have the
required resources available in the subgrid, it then forwards the
This proposed scheme is an enhancement over Mastroianni et al.[2].
The enhancements deal with some certain drawbacks, which are
identified, such as concerns over adaptability of the algorithm to
hybrid Grids, essentially Grids constituting of both peer to peer and
client server portions, overhead of communication between the
nodes in a single instance of resource discovery and usage, which
are identified to be too network intrusive. The following is a
discussion of some extensions which are introduced by us to
Mastroianni et al.[2]. These are the main assumptions which
underlie this proposed approach.
Firstly, future generation Grid networks will be of hybrid nature i.e.

there will be VO’s spanning over several physical locations
operated under one administrative domain, for which resource
discovery service will mainly operate in centralized manner; and
there will also be common user/ common business Grid enabled
ordinary nodes scattered around the world for which resource
discovery problem will mainly operate in a P2P fashion /
decentralized fashion.
Secondly, in traditional resource discovery algorithms complete

knowledge about all nodes was prerequisite to efficient functioning.
However complete knowledge about each node is rarely possible in
P2P environments. Moreover, the efficiency of resource discovery
scheme was measured in terms of the number of as many results as Fig. 1 Semi-Super-Peer Architecture for Proposed Resource Discovery
possible it can return so as that the later on scheduler can choose the Scheme
optimal resource to destine the request. But this approach is not
suitable for environments with large number of nodes. In P2P or query request to a pre determined number of neighboring
decentralized resource discovery technique, finding sufficiently super-peers and also sets some value for ttl parameter which gives
large number of resources will definitely produce huge amount of the number of hops (i.e. the number of super-peers) a query request
network traffic and will cause a great delay in response time. Hence, can/should travel, this a requirement to maintain QoS within the
resource discovery scheme (RDS) proposed return results meeting system. This is the portion where this proposed approach converges
the minimum constraints matching the user request. There is no with the one proposed by Mastroianni et al.[2], however the crucial
surety that the resources will fail and if this is the case, RDS can be differences arise with [2]. In that algorithm, a subgrid is a physical
run again. Alternatively, some additional resources (resources cluster, where-as in this proposed approach a subgrid is a loosely
additional to the minimum constraints specified by the user request) grouped collection of machines which have high-speed network
can by found by RDS to keep the resource failure factor in connectivity to each other. Moreover, in that algorithm, in an
consideration. Moreover, it is hoped that in future generation Grid instance of discovery, if a resource has been discovered within the
64
local subgrid, even then the super-peer propagates the satisfied Moreover, as opposed to the super-peer which has large number of
request to neighboring super peers. Moreover, once the resources resources registered with it (within a particular VO), a simple home
were found in the neighboring subgrids, the super peer of that peer usually has a very fewer number of resources, hence for this
subgrid was used to send the notification to the owners of the reason also query request is forwarded to a greater number of peers.
resources as well with no apparent useful outcome. Moreover, there Moreover, it is proposed that there would be no usage of resource
was also some strict requirement about the routing of query-hits to advertisement, as there are so many common user Grid nodes and
the Super-peer who originally produced the request. there would be so many resources and the resource advertisement
would constitute a large amount of network traffic without having a
Hence, the drawbacks of the Mastroianni et al. algorithm can be substantial good amount of resource results.
described as on following lines. Firstly, there is no need to send
B. Algorithm for the Proposed Resource Discovery Scheme
notification to owners of resources what so ever as the user may be
more interested to access resources from neighboring subgrids of its sp_neighbor ---- neighbors for the super-peer
own VO in order to get the network transfer time to the minimum. p_neighbor ----- neighbors for the home peer
Moreover, in this case, the algorithm will generate extra network q_sp_ttl -----------number of super peer hops a query can travel
traffic without any fruitful outcomes. When the user node will really q_p_ttl ------------ number of common home peer hopes a query
require resource, it will request it to the target node, so sending can travel
notification from the super-peer (who identified the
resource-owner) to the resource owner would not bear fruitful For each query ‘q’ received by the super-peer,
outcomes. If (q. sender is simple node (peer) within the subGrid) then
Check resources registered with it.
Secondly, there is no need to send query-hits (results) along the If (resources found matching the query ‘q’) then
same path from which query had come. Rather, it should be sent to send queryhit (containing the id’s of the nodes)
the original sender of the request directly because it would reduce to the q.sender
processing at each intermediate super-peer and sending directly to else
super-peer can/may also reduce network transfer time. {select all sp_neighbours
for each sp_neighbour
Thirdly, there is no need to propagate query to other super-peers as send a copy of ‘q’ to ‘sp_neighbor’}
long as the required resources can be met from within the subgrid
itself. However, if these resources are not available from within the elsif q.sender is another super-peer or home peer then
organization, then the query should be routed by the super-peer to if (q.id isn’t in the cache) then
best neighbor super-peers check resources registered with it.
If (resources found matching the query ‘q’)
A. Decentralized Resource Discovery Scheme on Semi-Super-Peer then Send queryhit to that request originating
architecture super-peer or home peer
This proposed technique addresses the structure of future generation Else
Grid networks which will be of hybrid nature i.e. there will be {Decrease the q_sp_ttl
client-server based VO’s and there will be a plethora of ordinary Put q.id in the cache
user Grid-enabled nodes as Grid computing is expected to move into Select all sp_neighbors
new domains i.e. commodity computing or utility computing in For each sp_neighbor
future. These ordinary user nodes are referred as home peers here. Send a copy of ‘q’ to
Note that these home peers are different than the simple peers which ‘sp_neighbor’}
operate under the super-peer in a sub-grid. If the requesting node is
part of some VO, the resource request is sent to the super-peer of the For each query ‘q’ received by a home peer,
VO, but if the node is standalone Grid user node (home peer), then if (q.id isn’t in the cache) then
there come the benefits of this proposed scheme. The common user check resources registered with it.
can use resource from the community of other common user nodes If (resources found matching the query ‘q’) then
(other home peers) i.e. the anonymous arbitrary user nodes spread Send queryhit to that request originating home
across world-wide. In this case, the RDS scheme will operate peer
(assuming for example, PhantomOS[3] (Grid OS) is configured on Else
each machine) in a peer to super-peer or super-peer to super-peer {Decrease the q_p_ttl
fashion i.e. there are three possibilities, Put q.id in the cache
Select all p_neighbors
a) super-peer to super-peer Select all sp_neighbors
As described above with modifications to Mastroianni et. Al. [2] For each sp_neighbor and for each p_neighbor
Algorithm. Send a copy of ‘q’ to ‘sp_neighbor && p_neighbor’
b) home-peer to super-peer
c) home-peer to home-peer V. SIMULATIONS AND RESULTS
In a completely de-centralized network of common Grid user nodes,
home-peer will forward request to a predetermined number of
Existing grid simulators like Bricks, SimGrid, GridSim, OptorSim,
neighboring simple home peers as well as neighboring super-peers
GangSim etc. were extensively surveyed in this study, but either
which will check that required resources are available with them or
these simulators are more targeted towards the simulation of
not. The number of super-peers will obviously form the small subset
scheduling, resource allocation issues or are devoted for some other
of the overall neighbors of the requesting peer. The no. of
purposes and treat the resource discovery mechanism in a
neighboring peers to which request will be forwarded, of course,
centralized manner much in the same way as this is implemented in
will be far greater in quantity as compared to the number of
Grid middlewares like globus or etc [13]. Hence, the proposed
super-peers to which superpeer ordinary forward the query request.
resource discovery scheme was evaluated using a simulator
This increased number is mandatory so that realistic results (in
developed at the DEIS University of Calabria, Italy, which is
terms of the amount of no. of resources and the different class types
purpose-built for the simulations of resource discovery approaches
of resources) can be obtained.
based on peer-to-peer fashion. The simulator has been tailored a bit
65
to meet our needs to cater for the semi-super-peer architecture. The

results were taken for this proposed semi-super-peer architecture 20
15
110 Average no. of
100 10 results or q.hits
90
80 5
70
60 Success 0
50 Percentage ttl 1 ttl 2 ttl 3 ttl 4
40
30
20 Fig. 4 Average No. of Results VS TTL (No. of Hops)
10
0
(When home-peers forward query to other home peers and super-
ttl 1 ttl 2 ttl 3 ttl 4 peers neighbors)
Fig. 2 Probability of Success VS TTL (No. of Hops) These all results show that in first simple scenario, for ttl=3 and
(When home-peers forward query to other home peers only) ttl=4, we achieve probability of query success in the range of
85%—100%. While, if home peers forward query to super-peers as
120 well i.e. semi-super-peer architecture than for only minor values of
super-peer neighbors and ttl i.e. equal to 2 for both, we achieve
100
probability of query success upto 100%. For this semi-super-peer
80 architecture, we also achieve about 5 No. of q.hits or results in
Success response to a single query request for a minor value of ttl = 2.
60
Percentage Considering all above figures and facts, it is concluded that this
40 semi-super-Peer architecture is very much feasible for resource
20 discovery in grid networks and provides a reasonable and efficient
0 mechanism for resource discovery in future generation grid
networks of hybrid nature.
ttl 1 ttl 2 ttl 3 ttl 4
VI. CONCLUSION
Fig. .3 Probability of Success VS TTL (No. of Hops)
(When home-peers forward query to other home peers and As the Grid environments are increasing in scale and complexity,
super-peer neighbors) No. of Super-Peer Neighbors = 2 resource discovery mechanisms should be modified to address
fault-tolerance and performance efficiency issues. The replacement
with the parameters settings given in the simulator. Results were of the centralized mechanisms by their corresponding distributed or
taken for a grid of 10,000 nodes where common home-peers peer-to-peer mechanisms is the key to addressing these issues. This
constitute upto 400 to 2000. Resources were evenly distributed proposed approach takes into consideration the future structure of
among the nodes and queries were generated randomly from these Grid networks and promises to efficiently handle the scalability
nodes and sent to neighboring super-peers and home-peers as challenges. It extends current work in P2P sub grid based discovery
described in the algorithm. Results were measured in terms of query services and adopts it to user oriented Grid infrastructures. In future,
success i.e. the query (resource request) which brings at least one it is intended to incorporate semantics in this resource discovery
resource in response to its request is said a successful query. mechanism to minimize user interaction, provide inter-operability,
autonomy, and search precision. Thus leveraging, combining and
As depicted below, Fig. 2 shows the probability of query success inducing the benefits of P2P and semantics in resource discovery
percentage with respect to different values for ttl (no. of hops) while primitives of present-day Grid infrastructure.
homes peers forward query to other home peers only i.e. total
decentralized network. REFERENCES
Similarly, Fig. 3 shows the probability of query success percentage
with respect to different values for ttl (no. of hops) while homes [1] A. Iamnitchi and I. Foster, “A Peer-to-Peer Approach to Resource
Location in Grid Environments”, In: J. Weglarz, J. Nabrzyski, J.
peers forward query to other neighboring home peers and
Schopf and M. Stroinski (Eds.), Grid Resource Management, Kluwer,
neighboring super-peers thus emulating a semi-super-peer 2003.
architecture. The no. of neighboring super-peers was set to 2, each [2] C. Mastroianni, D. Talia and O. Verta, “A Super-Peer Model for Building
with a sub-grid size of 20 i.e. each subgrid consisted of 20 common Resource Discovery Services in Grids: Design and Simulation
nodes. Analysis”. Proc. European Grid Conference (EGC 2005), LNCS, vol.
3470, pp. 132-143, Springer 2005.
While, fig. 4 shows the average no. of results (or resources) brought [3] Arshad Ali, Richard McClatchey, Ashiq Anjum, Irfan Habib, Kamran
in response to a query request in semi-super-peer architecture Soomro, Mohammed Asif, Ali Adil, Athar Mohsin. “From Grid
Middleware to a Grid Operating System” Fifth International
against different values of ttl (no. of hops). The graph clearly shows
Conference Grid and Cooperative Computing, GCC 2006. China
that if a greater value for ttl (no. of hops) a query can propagate upto [4] I. Foster and A. Iamnitchi. “On death, taxes, and the convergence of
is set, greater no. of results are obtained. peer-to-peer and Grid computing”. In 2nd International Workshop on
Peer-to-Peer Systems (IPTPS '03), 2003.
[5] I.Foster, C. Kesselman, and S. Tuecke. “The Anatomy of the Grid:
Enabling Scalable Virtual Organizations” International Journal of High
Performance Computing Applications, 15 (3). 200-222. 2001.
[6] GT Information Services: Monitoring and Discovery System (MDS)
http://www.globus.org/toolkit/mds/
66
[7] GT Information Services: Monitoring and Discovery System (MDS) [14] Sivadon Chaisiri and Putchong Uthayopas, “Survey of Resource
MDS 2.4 in The Globus Toolkit 2.4 release Discovery in Grid Environments” Proc. IEEE Workshop on
http://www.globus.org/toolkit/docs/2.4/mds/mdstechnologybrief_draft Experimental Distributed Systems, 1990.
4.pdf [15] D.W. Erwin and D. F. Snelling, “UNICORE: A Grid Computing
[8] Paolo Trunfio, Domenico Talia, Paraskevi Fragopoulou, Charis Environment”. Proc. 7th Euro-Par Conference (Euro-Par 2001), LNCS,
Papadakis, Matteo Mordacchini, Mika Pennanen, Konstantin Popov, vol. 2150, pp. 825834, Springer, 2001.
Vladimir Vlassov, Seif Haridi, “Peer to Peer Model for Resource [16] S. Ratnasamy, P. Francis, M. Handley, R. Karp and S. Shenker, “A
Discovery in Grids”. CoreGRID Technical Report Number TR-0028, scalable content-addressable network”, Proceedings of ACM
March 17, 2006 SIGCOMM San Diego, CA, USA (2001).
[9] LCG - LHC Computing Grid Project. http://lcg.web.cern.ch/lcg/ [17] A. Rowstron and P. Druschel, “Pastry: scalable, distributed object
[10] M. Litzkow and M. Livny, “Experience with the Condor Distributed location and routing for large-scale peer-to-peer systems”, Proceedings
Batch System” Proc. IEEE Workshop on Experimental Distributed of the 18th IFIP/ACM International Conference on Distributed
Systems, 1990. Systems Platforms (2001).
[11] Gnutella (2003) The Gnutella protocol speci.cation. Home page [18] I. Stoica, R. Morris, D. Karger, M.F. Kaashoek and H. Balakrishnan,
http://rfcgnutella.sourceforge.net “Chord: a scalable peer-to-peer lookup service for internet
[12] Napster. http://www.napster.com applications”, Proceedings of ACM SIGCOMM San Diego, CA, USA
[13] Benjamin Quetier, Franck Cappello: “A survey of Grid research tools: (2001).
simulators, emulators and real life platforms” [19] K. Vanthoumout, G. Deconinck, R. Belmans: “A taxonomy for resource
discovery”, Springer-Verlag London Limited 2005
[20] D. Mitrios, N. Roussopoulos: A comparison of peer-to-peer search
methods http://www.citeulike.org/user/lrodero/article/328160
67
MST and SFMST based Clustering

Muhammad Aniq, Zahid Halim and Rauf Baig
Department of Computer Science
National University of Computer and Emerging Science, FAST-NU
H11/4 Islamabad, Pakistan.
m_aniq@yahoo.com, zahid.halim@nu.edu.pk, rauf.baig@nu.edu.pk
Abstract — MST based clustering is used to find clusters of The rest of the paper is organized as: Section 2 deals with MST
varying shape and size without requiring input parameters. based clustering along with techniques to remove inconsistent
This work is about the review of MST and SFMST (Scale Free edges. Section 3 explains SFMST clustering technique and its
MST) based clustering. The strategies to create MST and comparison with k-menas method. Section 4 reviews the methods to
remove inconsistent edges are reviewed and implemented. find optimal number of clusters. The results of implemented
SFMST methods are also reviewed with our implementation techniques are shown in the corresponding sections to retain the
results. As a reference k-menas clustering method is used for reader’s concentration.
comparison purposes. In this study we also review the methods
to find optimal number of clusters, particularly focusing on II. MST BASED CLUSTERING
“largest average silhouette width” and edge length distribution
in SFMST. SFMST was found to be a promising method that
can detect clusters of varying shape and size, and requires In this section we review the techniques presented in [1] “Minimum
minimum dependence on external input parameters. Spanning Tree Clustering of EEG Signals” along with our
implementation results.
I. INTRODUCTION
MST has been used to cluster data. An important property of MST
Due to innovations in information technology the amount of storing based clustering is that it can detect clusters of different shape and
digital data is growing rapidly but unfortunately data processing and size. It starts with a graph which can be constructed using any
analysis techniques have not grown correspondingly. The valuable similarity measure e.g. Euclidean, correlation, etc. Then Minimum
information and knowledge is hidden within the masses of data, Spanning Tree (MST) is obtained using algorithms e.g. prim,
which makes knowledge discovery a very complex and tedious kruskal. The resultant Minimum Spanning Tree consists of all
process. The solution to the problem is given by data mining. Data connected elements. The clustering can be obtained by removing
mining incorporates the knowledge of human experts with the edges from tree. The removed edges lead to the collection of
power of computers for the searching the valuable information from sub-trees that represents the clustering of given data.
large databases.
Figure 1 shows the result of MST clustering from iris data set. First
Clustering is an unsupervised data mining technique. It plays a four attributes of the data set are sepal length, sepal width, petal
central role in the process of knowledge discovery and data mining. length, petal width. The last element contains the class information.
The aim of clustering is to group similar objects in the same cluster The following distance measure is used specific to the data set.
and organize objects in different clusters apart from each other.
Clustering besides the core data mining task; is also a very complex d (i, j) = petal length * peal width
problem because of dealing with increasingly volume of datasets.
Moreover dataset may contain variety of attributes which make For the reference k-means method is used. The coloring in Figure 1
clustering more challenging. Many techniques have been developed shows the intrinsic clusters in the data set.
to solve the clustering problem. Each technique has its own benefits
and drawbacks and applicable on certain situations.
Clustering method is said to be efficient if it follows certain design

principles i.e. the method should not be computationally expensive,
capable of finding clusters independent of shape, size and density,
efficiently detect optimal number of clutters and require minimum
dependence on parameters provided by user.
The biggest problem with the clustering techniques is the

accomplishment of clustering design principles. It is found that
clustering techniques always left one or more important principles
while satisfying others. We don’t have any single clustering
technique that performs well in all situations. Therefore, it cannot be
said beforehand which clustering technique is suitable in a
particular case. Other problem includes the selection of technique
for a particular situation by users. The users are not usually familiar
with all clustering techniques and can select inappropriate
technique, that my lead to undesired results.
In this paper we review the some of the clustering techniques Fig. 1 MST from iris data set.
particularly in the area of graph theory. The focus will be on MST
and Scale Free MST (SFMST) based clustering. MST based The biggest problem in MST clustering method is identification of
clustering can detect clusters of varying shape and size whereas edges to be removed so that clustering gives meaningful results. The
SFMST has certain interesting application of arbitrary scaled inconsistent edges can be removed in many different ways.
networks. Following are edge removal strategies that are used in [1].
68
i. Strategy 1 W0(i, j) = | maxi,j(d(i, j)) | − d(i, j)

The first strategy is to remove longest edges in the tree.
Removing k longest edges leads to k+1 clusters. Where,
d(i, j) is the Euclidean distance between the nodes.
ii. Strategy 2 The spanning tree algorithm selects the highest weight (instead of
In this method, each edge is compared with its neighbors.
lowest usually) edge each time and adds it to the tree repeatedly,
The edge is removed if length of the edge is above than
until all nodes are connected. The preference is given to the highly
certain threshold.
connected nodes when adding an edge in the tree.
To deal with SFMST, we need certain criterion to give preference to

e – mean > q * std-dev highly connected nodes in the network. For this a threshold is used
Where, to decide highly connected node. If a node gets pass the threshold
e is the length of the edge. value, a bonus fitness value is added to all the edges attached to it.
mean is the average length of edges of the node The highly connected node is called hub.
q is a constant (1 to 1.5) Wnew(i, j) = W0(i, j) + ncn
std-dev is the standard deviation of edges of the node
Where,
n is the number of edges connected with the node
iii. Strategy 3 c is a constant, 0.5 < c < 1
It is Similar to strategy 2 except neighborhood is extended
to two steps away from current node. The value of c is very important in the construction process of
SFMST. Different value of c leads to entirely different shape of
Figure 2 illustrates the result of strategy 2. Three removed edges are SFMST. Higher value of c leads to less number of hubs with many
marked. From this we get four clusters; three intrinsic clusters and links. Figure 3, 4 illustrates the outcome obtained by varying the
one possible outlier. value of c.
To take the liberty form threshold value, a second method can be

used that was also described in [2]. The major changes are in the
weight initialization and weight updating phase.
The weights are initialized to 1 / d(i, j)2.

Weights are updated whenever an edge is added to the tree using the
equation:
Wnew(i, j) = ncn / d(i, j)2
Fig. 2 Removal of edges using strategy 2.
III. SFMST CLUSTERING
Fig. 3 Two hubs with single cluster

In this section we review the technique presented in [3] “Modifying
the scale-free clustering methods” along with our implementation
results.
The Scale Free MST (SFMST) also belongs to graph theory. The
idea of scale free networks in clustering was presented in [2]. Scale
free, as its name suggests means without any scale. Scale is defined
as average number of links in the network. In Scale Free network,
some nodes have many high connected links while others have very
few. For example Web Wide Web consists of few sites highly
connected sites, while most have very few links. From Fig. 4 Effect of smaller c value
implementation point of view, SFMST can be constructed by giving
We can get clustering from SFMST by defining the term of cluster.
preference to highly connected nodes when attaching nodes in the
For example one way is to consider a hub as a cluster. But in
network
practice, multiple hubs may belong to same cluster. As mentioned in
[2], two or more hubs connected each other directly or with the
First of all distance matrix is calculated by using some similarity
distance of one node are defined to be in the same cluster and also
measure. The first idea to construct SFMST tree, is to use reversed
chain of nodes belongs to the same cluster. In this paper we also
distances as edge weights described in [2].
used this criterion for defining cluster in SFMST. Similarly different
criterion can also be defined to get good clustering results.
69
Figure 5 contains a SFMST from iris data set. Correlation was used
as a distance metric. The points of intrinsic clusters in the data set
are colored. Here, the value of c = 0.815 and m = 3 is used. The
result obtained was groups of 3 clusters with misclassification of
only 4 data points. There were 9 non-clustered points found. If
non-clustered points are to be treated as misclassification, then we
have total of 13 misclassified data points. If k-means method
applied with k=3, there are total of 17 misclassified points.
Figure 6 shows a SFMST from multi-feature digit dataset. The 47

attributes contained in “mfeat-zer” file of the dataset was used. The
first 400 records that corresponds to digit “0” and “1” were used. In
this case, value c = 0.91 was used. From this, two clusters were
found that were separated quite well. There was only one
non-clustered point found and also no misclassification was noticed.
From K-means clustering method, 22 data points were found to be
in the wrong clusters.
Fig. 6 SFMST from multi-feature digit database set
• Probability function estimation

Optimal value is selected without running clustering
process.
• Largest average silhouette width.

The silhouette width is determined by following equation.
Si = ( minkbi(k) – ai ) / max(ai, minkbi(k)

.
Fig. 5 SFMST from iris data set
Where,
ai is the average dissimilarity between the data point
IV. FINDING OPTIMAL NUMBER OF CLUSTERS and all other points in the same cluster
bi(k) is the average distance from the ith point to
points in an another cluster k,
In this section we review the technique presented in “Finding the
Optimal number of clusters from Artificial Datasets”. The value lies between -1 and 1. The k value is taken for which
largest average silhouette width is obtained (closer to 1).
Because clustering is an unsupervised task, many suppositions can
A. Edge length distribution in SFMST
be wrong during the clustering process. For example distance
metric, initial centeroids positions, number of clusters chosen might SFMST clustering is accomplished by removing some edges. The
be wrong. Many clustering methods like k-means and Nearest criterion for removing edges is very important to get good clustering
Neighbor (NN) requires number of clusters prior the results. Previously we have discussed some methods i.e. longest
edge removal, etc. In [3] Edge length histogram method was
clustering process. There is a need to reduce the risk of wrong
introduced to remove inconsistent edges. It is known that the
inference and efficiently determining optimal number of average edge length in scale-free graphs depends logarithmically on
clusters in the data set. Following are traditional methods that have the number of nodes [4], [5].
been used in determining the number of clusters along with the idea The basic idea is to remove edges by truncating histogram. Because
of largest average silhouette width that was presented in [3]. of the lognormal distribution of edge lengths, the longer edges with
•
fewer connections grouped into the right side of histogram.
After performing various clustering’s Afterwards, Freedman-Diaconis [6] rule can be used to find the
The clustering process runs many times and the clustering number of bins. Thus inconsistent edges can be identified and
with the smallest objective function is taken as optimal. removed to get clustering.
• Model-based cluster analysis • After performing various clustering’s

It works on the assumption of probability distribution in The clustering process runs many times and the clustering
the data set. with the smallest objective function is taken as optimal.
70
• Model-based cluster analysis V. CONCLUSION:

It works on the assumption of probability distribution in
the data set.
In this study we have discussed some of the techniques of MST and
•
SFMST methods along with our implementation results. MST
Probability function estimation method has the advantage of finding clusters of irregular shape and
Optimal value is selected without running clustering size. SFMST found to be a promising method that has certain
process. application in scale free networks. The biggest benefit of using
•
SFMST is its minimum dependence on external input parameters.
Largest average silhouette width. SFMST needs only one input parameter i.e. c. In this work, we
The silhouette width is determined by following equation. showed that better results can be achieved by using correlation as a
distance metric. From the results, SFMST based clustering method
Si = ( minkbi(k) – ai ) / max(ai, minkbi(k) was found to be far more efficient than the k-means.
.
Where, SFMST is a promising method for clustering but there is a need to
ai is the average dissimilarity between the data point lower down the computations required by Spanning tree algorithm.
and all other points in the same cluster Different optimization can be performed by various techniques
bi(k) is the average distance from the ith point to along with the improvement in distance functions.
points in an another cluster k,
The value lies between -1 and 1. The k value is taken for which
REFERENCES
largest average silhouette width is obtained (closer to 1).
[1] Paivinen, N. Gronfors, T. Minimum spanning tree clustering of EEG
B. Edge length distribution in SFMST: signals, In Proceedings of the 6th Nordic Signal Processing
Symposium (NORSIG 2004), June 9–11, Espoo, Finland, 2004,
SFMST clustering is accomplished by removing some edges. The pp.149–152.
criterion for removing edges is very important to get good clustering [2] Paivinen, N. Gronfors, T. Modifying the scale-free clustering method, In
results. Previously we have discussed some methods i.e. longest Proceedings of International Conference on Computational
edge removal, etc. In [3] Edge length histogram method was Intelligence for Modelling, Control & Automation, CIMCA 2005, vol.
introduced to remove inconsistent edges. It is known that the II. Los Alamitos: IEEE Computer Society, 2006, pp. 477-483.
average edge length in scale-free graphs depends logarithmically on
[3] Paivinen, N. Gronfors, T. Finding the optimal number of clusters from
the number of nodes [4], [5]. artificial datasets, In Proceedings of IEEE International Conference on
Computational Cybernetics (ICCC 2006), August 20–22, Tallinn,
The basic idea is to remove edges by truncating histogram. Because Estonia, 2006, pp. 155–160.
of the lognormal distribution of edge lengths, the longer edges with
fewer connections grouped into the right side of histogram. [4] G. Szab6, M. Alava, and J. Kertesz, Shortest paths and load scaling in
scale-free trees, Physical Review E, vol. 66, no. 2, pp. 26101, 2002.
Afterwards, Freedman-Diaconis [6] rule can be used to find the [5] S. H. Yook, H. Jeong, A.-L. Barabasi, Y Tu, Weighted evolving
number of bins. Thus inconsistent edges can be identified and networks, Physical Review Letters, 2001, vol. 86, no. 25, pp.
removed to get clustering. 5835-5838.
[6] Freedman, D. Diaconis, P. On the histogram as a density estimator: L2
theory, Zeitschrift fur Wahrscheinlichkeitstheorie und verwandte
Gebiete, 1981, vol. 57, no. 4, pp. 453-476.
71
Association Rule Mining Towards Business Solution

Syed Salman Ahmed, Zahid Halim, Rauf Baig and Nusrat Habib
National University of Computer and Emerging Science, FAST-NU
H11/4 Islamabad, Pakistan.
ssalman99@gmail.com, zahid.halim@nu.edu.pk, rauf.baig@nu.edu.pk, nusrat_1000@yahoo.com
Abstract — the boundaries of businesses are expanding with fail to do a market-basket analysis, which is a much thorough
the integration of computers with large databases forming the investigation line. This analysis tells the manager what products sell
core. And businessmen today need to find latest trends using together and using this technique the manager may be able to devise
this data so as to keep abreast with the rapid changes in the incentives for the customers thus forming an effective competitive
market, by. Association Rule Mining is one area of data mining, strategy which if put into action in time as compared with its
which offers businessmen the answers to their prayers. In this competitors would result in much greater benefits.
paper we present a review of some the techniques from the field
of association rule mining that offer to uncover mystifying For scenarios such as these, sophisticated data mining techniques
answers to unnerving questions. Such techniques include are required that derives answers for the most complex of queries.
MAFIA algorithm and the Apriori Algorithm. One of the most notable set of techniques from the field of data
mining that are able to cater for such scenarios is Association Rule
Index Terms — Data mining, Data warehousing, Association Mining.
Rule Mining, MAFIA, Apriori.
In this paper we will be reviewing some of the techniques of
Association Rule Mining (ARM) and shall study their influence on
I. INTRODUCTION: the world of business. The rest of the paper is organized as: Section
II will be a brief description of the basic concepts involved before
Computers have become a very important part of modern day ARM techniques are actually explained. Section III will elaborate
business and are used in many capacities. Today computers are used the Apriori algorithm, section IV will deal with the MAFIA
as a means of storing data in electronic format rather than as large algorithm from ARM. And finally Section V will be the conclusion
stockpiles of files where keeping files in proper order, preserving of the entire paper.
them and managing them for later retrieval becomes extremely
infeasible with time. Another important usage of computers is in II. BASIC CONCEPTS:
bringing businesses on the web; today business have actually come
to the door step of any potential customer, with charges on the part
of the business men so low that they are close to being non-existent. Before dwelling into what are the types of association rule mining,
we need to focus on some of its basic concepts. ARM states that
Businesses have started to realize the greatest and most prolific given a certain set of values (Itemsets) say A and B, where A and B
advantage of using computers that is large memory and relatively are mutually exclusive, i.e.
low cost associated with it. They have started using this ability of
computers and make data warehouses out of computers where huge
amounts of data are stored, which is easily accessible. But the
establishment of a data warehouse posses a huge problem along
and
with its associated advantages; the data although can be secured
very efficiently out of it. There are however specialized indexing
techniques involved that when implemented and managed with
some common sense on the warehouse can lead to appropriate Read as A implies B is an association rule, meaning that if A exists
results. We will at this point in time divert our attention from data then B will also exist, then ARM is the process of finding all
warehousing, its problems, issues and solutions to those problems association rules in the entire data (D), where D consists of n records
and focus on the topic at hand. and m attributes or in other words it is transactional data. The
itemset A or B consists of at most m items and that each transaction
The latest advantage that businessmen have observed is related to in D is an itemset.
the use of computers in efficiently bettering their businesses is the
extraction of useful but hidden information from given data (data
mining) [1]. This involves the logical deduction of current trends on The total number of patterns that exist in D is km–m-1, where k is the
the basis of latest market updates. For instance, if the marketing minimum cardinality for the existence of an itemset, i.e. if we want a
manager of a particular company observes that the sales of some minimum cardinality of 2 it will mean that the rules mined must
particular product has declined in the last 3 months, he would be have at least two itemset and m = 5, then the total number of
very much be interested in analyzing the causes for this decline in potential itemset = 26, but this number increases exponentially with
sale. He would like to know what factors known or unknown lead to the increase in m and a constant k or at least a lower value of k. For
the current scenario, what were the existing market trends, what instance if k = 2 and m = 50, the total number of potential itemsets
were the prices of the product at time and what were the prices of shoots to 1125899906842573. Keeping record of all these many
their competitors, what and how effective was the advertising itemsets where they may or may not exist in D is very cumbersome;
campaign, which were the most affected areas in terms of decline in hence the need to maintain only interesting patterns where
sale, what local and foreign factors were involved and what was the interestingness in the field of ARM is most commonly measured by
stock market situation at the time. All these are the most likely the following two factors.
questions a manager would like his computer to give response to so
as to come to the exact cause or causes for the problem and in-turn A. Support:
devising tactics for their remedial.
It is the number of occurrences of an itemset in D.
If the organization has a working active-data warehouse, it would
most likely be able to give answers to these questions. But it would
72
We have to maintain a threshold factor on support to determine the prune step all those candidates that are determined to be,
minsupport [2] and is defined as if support(A) ≥ minsupport then A

whether the itemset if frequent or not, this threshold is known by infrequent that less than the minsupport threshold are eliminated.
This entire first phase is the largest time consuming part of the entire
is considered a frequent itemset (F). Normally this minsupport is a
process and is thus considered to be the downfall of the entire
very small value as compared to the large value of n.
approach. Methods have been proposed that include the counting of
more than the size of the current considered size of itemsets.
B. Confidence: Furthermore efficient candidate generation techniques are the focus
It is the ratio of the support of an association rule to the support of of research these days.
the antecedent in that rule.
The data structure used for storing the candidate itemsets along with
their support count is a hash-tree structure. This enables for an
efficient placement and retrieval mechanism of the itemsets.
To sum up the entire process, initially a set of candidate itemset and
A threshold is also maintained for the confidence and is used to a set of frequent itemset are generated based on the minimum
quantify the confidence level of the rule known as minconfidence cardinality desired or in other words k. the support of the candidate
threshold. Any rules failing below a certain confidence level are set is verified and pruned where necessary and finally update the
discarded. The confidence level required for rules is set to be frequent itemset which qualifies the minsupport threshold.
relatively higher, so that appropriate and interesting rules are mined
only. Below is a diagrammatic summary of the described process. Taken
under consideration is a type of market basket analysis where TID
An important point that must be raised at this point to have in mind stands for Transaction ID and each represents an itemset. At the
for future reference is that for attributes that are not related to one beginning of the example the transactions are listed, where from the
another there might seem to exist some relation, however it will be initial itemsets are derived along with their support, and they are
definitely be misleading to follow such relations. Always remember consequently pruned away. In the next iteration, the items in the
to look for relation between co-related attributes and not anything itemset are increased and as in the previous step they are checked to
else. be frequent with the minsupport threshold.
Rules are deduced/mined from D on the basis of some existing
correlation among itemset in D. the rule extraction process has been
divided into 2 phases [3]. The first phase finds the frequent itemsets
(F) as mentioned above, while in the second phase interesting
patterns are extracted from D on the basis of F.
This process of finding the interesting patterns is the one that forms
the basis of a number of competitive techniques in ARM. The need
for these techniques arises from the fast that there would not be a
significant need for these techniques if the number of attributes were
significantly small. However in the scenario where the number of
attributes is very large, greater than 200 for instance, finding
interesting patterns by mere brute force becomes computationally
infeasible; where initially all the possible combinations of itemsets
are noted along with the actual support in D, then rules are mined
from this data based on the minsupport threshold and
min-confidence threshold, where for large n and m the process
becomes nearly computationally infeasible.
In the following sections, we will discuss some ARM techniques
that primarily achieve the above-mentioned purpose but with
prominent ease and algorithmic efficiency.
III. APRIORI:
Fig. 1 Diagrammatic Summary of Apriori
Agrawal, et al, presented the Apriori algorithm in 1994. The Apriori
algorithm works on the principle that any subset of a frequent
itemset must also be frequent, meaning that if {bread, butter, milk} IV. MAFIA:
is frequent then so will be {bread, butter}. Similarly, if an itemset is
infrequent its superset should not be generated. A detailed
This algorithm was proposed by Burdick et al. [2] in 2001. MAFIA
description of the Apriori algorithm is given in [3], however we
stands for Maximal Frequent Itemset Algorithm. This algorithm
present here a summary of the algorithm describing only the
focuses on the Maximal Frequent Itemset (MFI) in D where MFI
required detail.
is the set of frequent Itemsets where no superset of it exists.
There are two major parts of the Apriori algorithm: the candidate
This algorithm works by maintaining a tree like structure of the
generation phase and the support counting phase.
entire itemset lattice. It does so quite efficiently and without
In the first phase candidates to be considered as frequent or not are repeating items in the lattice. An important requirement for this
generated. This is achieved once again in a two-step approach: the algorithm is that the transactional items be ordered within each
self-join step and the pruning step. In the self-join step, all transaction according to some pre-specified criteria, normally it
candidates are added to each other to produce all the possible would be better to simply maintain an ascending order of
combinations; these are generated along with their support count. In transactional items. This process is known as Dynamic Reordering.
73
The root node of the tree is maintained to as null and each next level C. Head Union Tail MFI (HUTMFI):
node from the previous node is an extension of the items in the This is the opposite process of FHUT, whereby it looks for the
itemset. For instance consider the itemsets from Fig. 1 as superset of a HUT in the MFI, if it exists there then the HUT can be
representations in Fig 2. termed to be frequent and the tail can be pruned away at the current
node.
Using these methods fast access to itemsets can be achieved, which
leads to finding interesting patterns much more efficiently than
other such methods.
Sugar Butter Milk Tea Bread V. CONCLUSION:
Fig. 2 Tree Structure Representation of MAFIA Businessmen always need to make the best out the little time they
The working of MAFIA is similar to Apriori in the sense that both have while trying to maximize any and all sources of profit earning.
the algorithms work on the basis of negative selection; that is they To achieve their goals they need techniques from the area of data
both greatly depend on the pruning of the structure. MAFIA cuts the mining and more specifically association rule mining, which can
tree at such points that the branches below the cut (subsets) are enable them to reach to quick conclusions and finding of interesting
deemed to be pruned away as they are termed to be infrequent, patterns in data that would potentially lead to growth in their
whereas the branches are termed to be frequent. businesses. We have in this paper reviewed two such techniques and
conducted a short market basket analysis with these techniques, to
The most prominent advantage that MAFIA holds over Apriori is prove their effectiveness. It has been found that although both the
that Apriori needs constant recursive referral to the dataset, whereas techniques are good mining techniques, the MAFIA algorithm for
MAFIA involves very few referrals to the dataset, in fact it nearly finding important patterns is the better of the two.
requires only a single complete traversal of the dataset, which saves
a lot of time. REFERENCES:
MAFIA works as a simple Depth First Search algorithm. Access to a
particular item in the tree or finding whether a particular itemset is [1] R. Agrawal, T. Imielinski, and A. Swami. Mining association rules
frequent or not requires a simple Head-to-Tail framework, where between sets of items in large databases. In Proc. of the ACM SIGMOD
the path from the root node to the current node is termed as head and Con-ference on Management of Data, Washington, D.C., May 1993.
the path up to any possible leaf node forms the tail. For instance, in [2] D. Burdick, M. Calimlim, and J. Gehrke. MAFIA: A maximal frequent
itemset algorithm for transactional databases. In 17th International
the tree structure drawn above a possible of a head-to-tail
Conference on Data Engineering (ICDE’ 01), pages 443–452,
connection can be {[Sugar], [Butter Milk, Bread]} which obviously Washington - Brussels - Tokyo, April 2001. IEEE.
does not exist in the fig. 1 and so will be pruned away. [3] Rakesh Agrawal and Ramakrishnan Srikant. Fast algorithms for mining
association rules in large databases. In International Conference On
Furthermore, if the support of any particular node is found to be
Very Large Data Bases (VLDB ’94), pages 487–499, San Francisco,
infrequent, its subsets are not explored, following the principle Ca., USA, September 1994. Morgan Kaufmann Publishers, Inc.
followed by Apriori algorithm as explained in section III. The depth [4] Sergey Brin, Rajeev Motwani, Jeffrey D. Ullman, and Shalom Tsur.
first approach used helps limit any unnecessary traversals, thus Dynamic itemset counting and implication rules for market basket data.
increasing efficiency. SIGMOD Record (ACM Special Interest Group on Management of
Data), 26(2):255–264, June 1997.
Tree Pruning Methods for Efficiency: [5] Bing Liu, Wynne Hsu, and Yiming Ma. Integrating classification and
association rule mining. In KDD, pages 80–86, 1998.
Following are some of the main types of tree pruning techniques
[6] J. Han, Y. Cai, and N. Cercone. Knowledge discovery in databases: An
used [2] to remove unnecessary, infrequent parts of the tree so as to attribute oriented approach. In Proc. of the VLDB Conference, pages
drastically reduce search space. 547-559, Vancouver, British Columbia, Canada, 1992.
[7] R. Agrawal, C. Aggarwal, and V. Prasad. Depth First Generation of Long
A. Parent Equivalent Pruning (PEP): Patterns. In 7th Int’l Conference on Knowledge Discovery and Data
Mining, Aug. 2000.
This method involves that the support of both the parent and its child [8] R. Agrawal, H. Mannila, R. Srikant, H. Toivonen, and A. I. Verkamo. Fast
are checked, if they are the same and that the child’s tail is similar to discovery of association rules. In U. Fayyad and et al, editors, Advances
the remaining tail of the parent, then the parent and the child nodes in Knowledge Discovery and Data Mining, pages 307–328. AAAI
are merged in conformance with the Apriori principle. This step Press, Menlo Park, CA, 1996.
does not compromise the overall performance and correctness of the
system since we are not interested in counting individual items;
rather we are interested in MFI.
B. Frequent Head Union Tail Pruning (FHUT):

This method is also known as superset pruning. This method
involves the pruning of any subset of an itemset (tail) rooted at some
node if the head of that node is found to be frequent, consequently
there is no need to explore the tail as it is found to be frequent. In
principle if an itemset is found to be MFI its subsets will be Frequent
Itemsets at the very least.
74
MOTION INVARIANT MULTI-FOCUSED IMAGE FUSION FOR VIDEO

1
Adeel Khalid Siddiqui, 1Nadeem Talib, 2*Syed Ismail Shah and 2Jamil Ahmad
1
National University of Modern Languages and Sciences, 2Islamabad, Iqra University
Islamabad Campus
ismail@iqraisb.edu.pk
Abstract — In this paper we propose a technique of

obtaining sharp images from multiple images having different between consecutive frames. In our work we have used 3x3 pixel
focus areas. The method utilizes image diffusion and motion window to look for similar objects. Once similar location in other
invariant techniques to obtain sharp images. Examples frame is marked out, the size of the window is further increased to
presented show the performance of the proposed technique. 7x7 or more in-order to verify the region for correctness.
Similar regions are then treated as small scale separate images and
Index Terms — Image diffusion, Multi Focused, Motion the proposed algorithm is applied to get enhanced edges. Once
Invariant, Image sharpness enhanced edges are calculated in the small scale images, the source
frame is updated with the new vales of enhanced edges.
I. INTRODUCTION
V.PROPOSED TECHNIQUE
Image sharpness or edge enhancement has been an active topic for
researchers for many years. Still a lot of work is being done and
Suppose, two images A and B have been acquired in a sequence,
different techniques have been proposed [1, 2].
with one is focusing at the background and the other on the
foreground. Since the images are assumed to be acquired in
According to the lens theory, we cannot focus on all parts of the
sequence the will have some similarity we therefore assume the case
scene at the same time. The focal length of a lens can be varied in
of very little similarity i.e., they just have slight overlapping area,
order to get sharp image. The lens can only focus on a single point at
which normally happens in fast motion videos.
a same time. If our requirement is to get the sharpest image in which
all areas appear as if lens is focusing on all parts; one must capture
We now need to sharpen the first image by taking features from the
multiple images of the same area focusing on all parts of the scene to
second image, and then we need to find out similar area of the first
be captured one by one [3,4,5]. Later, some image processing
image in the other image. For this purpose, we need to divide both
technique must be used to merge those multi-focused images like
images in sub images of size 3x3 or more. So if we have an image of
the one presented in [6].
size m x n where m=n then we will get (m-2)2 number of 3 x 3 sub
images. The matrix of these sub images is given by:
The proposed algorithm in this paper first finds the matching area in
 A11 A1n 
two or more consecutive frames. Later, the sharpest part of one is
 . 
diffused to other and the process continues. A12 ... ... ...
 . 
A=  . 
Two image acquisition techniques have been discussed, with single
 
camera capable of taking multiple focus frames. Two cameras, .
 . 
creating same video of same scene ideally placed very close to each
 . 
.
other.
 
.
 Am1 ... ... ... Amn 
Rest of the paper is organized as follows. Next two sections discuss
two cases of image acquisition. The motion invariance is discussed Am 2
in section 4 while section 5 discusses the proposed technique.
Simulation results are presented in Section 7 while the paper is
concluded in section 6. Therefore we get a general block matrix with m rows and n columns,
[ ]
where Aij is a sub matrix at the ith row and jth column of size 3x3.
Aˆ iˆˆj = aij
These sub matrices can be written as:
II. IMAGE ACQUISITION CASE-1
For this case, single high speed multi-focus camera is required that where
is capable of taking minimum 60 frames per second by taking all the iˆ ≤ i ≤ iˆ + 2
ˆj ≤ j ≤ ˆj + 2
even frames focusing at the background and all the odd frames
focusing at the closer objects. Later, sharpening algorithm can be
iˆ, ˆj = (m − 2)
applied to get the sharpest. where
III. IMAGE ACQUISITION CASE-2 Similarly the other image can be defined.
For this case, two separate cameras are used in parallel, creating Once, the sub-matrices are made, the difference matrix D between
different videos of the same scene and same group of objects. Later, one image A and the second image B is calculated. For this we need
these two videos can be merged, using the proposed algorithm. One to iterate one sub matrix of A with all the sub matrices of B and this
must assure that the cameras are close to each other having distance process continues till we get (m-2)^4 difference matrices.
of not more than a few centimeters between them. One camera must
focus at the background and other at the foreground. Hence ,
= Aˆ iˆ ˆj − Bˆ iˆ ˆj
IV. MOTION-INVARIANCE
∈A
D Aˆ
iˆ ˆj
Video can be considered as a sequence of different but closely Bˆ ˆ ˆ ∈ B
ij
related images. Therefore, there is a great degree of similarity
75
a) Input image 1, Focused at left b) Input image 2, Focused at center
c) Input image 3, Focused at right

d) Resultant image, Completely Focused
ai 0 = a0 j = aij = ai +1, 0 = a0, j +1

Fig. 1 Multiple blurred Images
d ( aij , ai +1, j +1) = aij − ai +1, j +1

Now we need to find individual sum of all the difference matrices
Dˆ iˆˆj = ∑∑ Diˆˆj [aij ]

3 3
j =1 i =1
Hence, the similar area in the other frame will be the group of pixels DS [aij] = max(dp)
Aˆ iˆ ˆj
Bˆ iˆ ˆj
where the difference of and
is minimum, and the
Similarly S2 is calculated using the same DS. The entries in DS are
Dˆ replaced by ‘difference’ entries of S2 only if they are greater than
minimum of these can be seen by looking at the minimum of iˆˆj
the value stored at the same location already.
Now, from the two images, the sub images S1 and S2 can be
extracted because the corresponding similar areas have been The original value to be selected for the edge will be chosen from
identified. We now apply multiple focused method similar to the the matrix whose difference value was the greatest. Thus giving the
one proposed in Asada [6]. The algorithm can be written as follows: sharpest value of all the pixels.
∑d
S1 = [aij]mxn VI. SIMULATION RESULTS
8
= where Consider the three images depicted in Figure 1a, 1b and 1c. The
p =1
p
= d (a − a
three images are of the same scene but are focusing different areas
while the rest of the areas are blurred. The proposed algorithm was
d i−1 j+1 )
= d (a − a
1 ij applied on the three images and the resultant image is shown in
i+1 j −1
d ) Figure 1d. As can be seen it presents a sharp image of the entire
= d (a − a
2 ij
scene.
d i+1 j+1 )
= d (a − a
3 ij
VII. CONCLUSION
d i−1 j−1 )
= d (a − a
4 ij
d i−1 j )
= d (a − a
5 ij This paper introduced a new approach for video enhancement by
ij + 1
d ) sharpening the foreground and the background at the same time
= d (a − a
6 ij
using the technique of motion invariance by calculating the sharp
d i−1 j ) features from its neighboring frames in motion; even sharper frames
= d (a − a
7 ij
that borrow sharper areas form their neighbors are developed. This
d 8 ij i+1 j ) method is well suited for offline video processing. It can also be
applied to the videos that have already been created up till now.
Videos produced after being processed by this algorithm will appear
where to be more sharp and detailed then they were before. Simulation
results show the performance of the proposed technique.
76
REFERENCES
[1] V.Maik, D.Cho, J.Shin and J.Paik, “Regularized Restoration Using [4] A. Katsaggelos, J.Biemond, R. Scharfer, and R. Mersereau, “A
Image Fusion for Digital Auto Focusing” IEEE Transactions on regularized iterative image restoration algorithm,” IEEE Trans. Signal
Circuits and Systems for Video Technology VOL 17, NO. 10, Oct 2007 Process., vol 39, no.4 pp. 914-929, Apr 1991
[2] V.Maik, D.Cho, J.Shin and J.Paik, “Pattern selective image fusion for [5] L. Bogoni and M. Hansen, “Pattern selective color image fusion,” Int. J.
multi-focus image reconstruction,” Proc.Comput. Anal. Image patt., Patt. Recog., vol 34, pp. 1515-1526, 2001.
LNCS, vol 3691, pp. 677-684, Sep. 2005 [6] N. Asada, H. Fujiwara and T. Matsuyama, “Edge and Depth from Focus”
[3] M. Subbarao, T. C. Wei, and G. Surya, “Focused image recovery from Int. J. Comp. Vision., vol 26, no. 2 pp 153-163, Feb 1998
two defocused images recorded with different camera settings,” IEEE
Trans. Image Process., vol 4,no. 12,pp. 1613-1628, Dec. 1995
77
FROM DEEP WEB UP TO THE SURFACE WEB

Peer Azmat Shah, Muhammad Maaz Rehan
Mohammad Ali Jinnah University, Islamabad Pakistan.
peer.azmat@yahoo.co.uk, maazrehan@yahoo.com
updation and removal can be performed using SQL queries or
XQUERY/XPATH to SQL Mapping can do this.
Abstract — "Deep Web" is the hidden but distributed
- In NATIVE XML technique, the document is stored not in any
repository of data lying in the databases and other documents
underlying DB system but in the form of a Tree Data Structure.
which is not accessible through search engines. This
Therefore XQUERY/XPATH to SQL mapping is not required.
information is mostly accessible through query interfaces and
only the authorized users can query it after login. As a result
Some rules are employed on an input Relational Schema and an
data access is restricted to the search engines and web crawlers
XML Schema in the form of DTD file is produced which then
as well. In this paper a new algorithm DWToSW (Deep Web To
produces XML pages.
Surface Web) is proposed that aims to bring Relational
Database over the surface web via semi structured XML thus
attempting to locate and quantify material on the Web that is III. FROM RDBMS TO XML (DEEP WEB TO SURFACE WEB)
hidden from typical search techniques.
We defined 11 rules that convert a Relational DB to the
I. INTRODUCTION corresponding XML file by generating its schema in DTD.
We have taken the example of Student Course Registration System.

The word deep Web refers to the World Wide Web content that is
The tables in un-normalized and normalized form are shown in
not a part of the surface web indexed by search engines. It is
figure 1 and 2 respectively.
estimated that the deep Web [4] is several magnitudes larger than
the surface Web and contains approximately 7,500 terabytes of data
and more than 550 billion individual documents. To discover
content on the Web, search engines use web crawlers that follow
hyperlinks. This technique is ideal for discovering resources on the
surface Web but is often ineffective at finding deep Web resources.
For example, these crawlers do not attempt to find pages that are the
result of database queries due to the infinite number of queries that
are possible. This can be partially overcome by providing links to
query results, but this could unintentionally inflate the popularity
for a member of the deep Web. Thus the data present in the deep
web must be converted to some other form so that it can be accessed
by web crawlers. The Title, From Deep Web up to the Surface Web,
is chosen purposely and that is to show migration from RDBMS to
XML.
Rest of the paper is organized as follows: Section 2 presents

different techniques of XML to RDBMS conversion. In section 3 a
technique named as publishing rules is explained that uses 11 rules
for publishing
RDB to XML step by step using a many to many relationship

example and also the inefficiencies in these rules are highlighted. In
section 4 a proposed algorithm DWToSW (Deep WEB To Surface
Web) is presented for RDBMS to XML conversion. In section 5 the
working of proposed algorithm is explained with the help of
different examples, and finally in section 6 a conclusion is made.
II. FROM XML TO RDBMS (SURFACE WEB TO DEEP WEB)
The approaches CLOB, SHREDDING and NATIVE XML are the

mechanisms to store XML files in Relational DB Systems.
- In CLOB technique, the actual XML document including the tags

and data, not the DTD, is saved in any underlying Database In this example, a student can register many courses and a course
System. The attribute which contains this document is a varchar can be registered by many students. The un-normalized many to
field. This technique is only employed when full document many tables of Figure1 can be converted to normalized
insertion and retrieval is required and no sub-document operation one-to-many-to-one relationship by introducing a Join table
is needed. (Microsoft says) [6] or a Bridge table (DB says). This table
- In SHREDDING (means scattering) technique, all the tags from StudentCourse contains the primary keys of the two tables, Student
the XML document are removed using some Parsing mechanism. and Course, and makes them the composite key. The bold arrows
Now the underlying DB system can be relational, and the data is show the direction of traversal. This can now easily be solved by the
stored in the relevant tables using shredder algorithm which can 11 rules as mentioned in the First Draft
be proprietary. In this mechanism, sub-document insertion,
78
A. Publishing (RDB to XML) Step by Step process [2] Add an ID attribute to each of the elements created in the target
DTD Schema excluding root element. We use element name
In a DTD, a table is mapped to an !ELEMENT declaration, followed by ID as the Primary key to avoid name collisions. Primary
attributes to !ATTLIST declaration, type of attribute as CDATA, attribute type is ID and #REQUIRED is must.
Not-Null value with #REQUIRED declaration and NULL value <!ELEMENT UniversityCourseRegistration EMPTY>
with #IMPLIED. <!ATLIST UniversityCourseRegistration
UniversityName #REQUIRED
Location #REQUIRED
Rule 1: Choose tables and columns from relational database to
ContactNo #REQUIRED>
be included in the documents
<!ELEMENT Student EMPTY>
The Relational Schema is
<!ATLIST Student
- Student (RegNo, Name (NOT NULL), Address(NOT NULL),
StudentID ID #REQUIRED // Rule: 5
Hobbies(NULL))
RegNo CDATA #REQUIRED
- StudentCourse (RegNo, Course) Name CDATA #REQUIRED
- Course (Code, Name(NOT NULL), CreditHours(NOT NULL)) Address CDATA #REQUIRED
Hobies CDATA #IMPLIED>
Rule 2: Create a root element for the DTD and declare any
attributes of that element that are required to hold additional <!ELEMENT StudentCourse EMPTY>
semantic information. Root element's names should describe <!ATLIST StudentCourse // Rule: 5
their content. StudentCourseID ID #REQUIRED>
<!ELEMENT Course EMPTY>
// Rule: 2 <!ATLIST Course
<!ELEMENT UniversityCourseRegistration EMPTY> CourseID ID #REQUIRED // Rule: 5
<!ATLIST UniversityCourseRegistration Code CDATA #REQUIRED
UniversityName #REQUIRED Name CDATA #REQUIRED
Location #REQUIRED CreditHours CDATA #REQUIRED>
Rule 3: Model the Content Tables Rule 6: Representing Lookup Tables
Create an element in the DTD for each content table we have chosen If a foreign key references a lookup table:
to model. Declare these elements as EMPTY .
<!ELEMENT UniversityCourseRegistration EMPTY> - Create an attribute in the element containing foreign key
<!ATLIST UniversityCourseRegistration
UniversityName #REQUIRED - Keep attribute name same as the referenced table name, and make
Location #REQUIRED it #REQUIRED or #IMPLIED
ContactNo #REQUIRED >
- Make the attribute of the enumerated list type. The allowable
// Rule: 3 values should be some human-readable form of the description
<!ELEMENT Student EMPTY> column for all rows in the lookup table
<!ELEMENT StudentCourse EMPTY>
<!ELEMENT Course EMPTY> In our example, Hobies in Student table is an enumerated data type
(allows NULL) and its look-up table is Hobbies.
Rule 4: Modeling Non-foreign Key Columns
<!ELEMENT UniversityCourseRegistration EMPTY>
Create an attribute for each column we have chosen to include in
our XML document (except foreign key columns)
<!ELEMENT UniversityCourseRegistration EMPTY>
Location #REQUIRED
Location #REQUIRED
<!ATLIST Student
StudentID ID #REQUIRED
<!ATLIST Student
Name CDATA #REQUIRED
// Rule: 4
Address CDATA #REQUIRED
// Rule : 6
Hobies (Cricket | Other) #IMPLIED>
Address CDATA #REQUIRED>
<!ATLIST StudentCourse
// Foreign Keys
StudentCourseID ID #REQUIRED>
<!ATLIST Course
Code CDATA #REQUIRED //Rule: 4
<!ATLIST Course
CourseID ID #REQUIRED
CreditHours CDATA #REQUIRED>
Code CDATA #REQUIRED
Rule 5: Adding ID Attributes (Primary Keys) to the Elements
(for use in Rule 9)
79
CreditHours CDATA #REQUIRED> Code CDATA #REQUIRED

Rule 7: Adding Element Content to Root elements
We add element content for the type of information we want to Rule 9: Adding Relationship using IDREF (Many-to-One or
publish in our documents, so we add it in root element. It means our Multiple Parent Relationship)
documents will contain data from these tables only.
If the relationship is many-to-one, or the child has more than one
<!ELEMENT UniversityCourseRegistration (Student*)> parent, then we need to use pointing to describe the relationship.
// Rule: 7 This is done by adding an IDREF to the element on the parent side
<!ATLIST UniversityCourseRegistration of the relationship. The IDREF should point to the ID of the child
UniversityName #REQUIRED element.
Location #REQUIRED
ContactNo #REQUIRED> <!ELEMENT UniversityCourseRegistration (Student*)>
<!ELEMENT Student EMPTY> UniversityName #REQUIRED
<!ATLIST Student Location #REQUIRED
StudentID ID #REQUIRED ContactNo #REQUIRED>
Name CDATA #REQUIRED <!ELEMENT Student (StudentCourse*)>
Address CDATA #REQUIRED <!ATLIST Student
Hobies (Cricket | Other) #IMPLIED> StudentID ID #REQUIRED
<!ELEMENT StudentCourse EMPTY> Name CDATA #REQUIRED
<!ATLIST StudentCourse Address CDATA #REQUIRED
StudentCourseID ID #REQUIRED> Hobies (Cricket | Other) #IMPLIED>
<!ELEMENT Course EMPTY> <!ELEMENT StudentCourse EMPTY>

<!ATLIST Course <!ATLIST StudentCourse
CourseID ID #REQUIRED StudentCourseID ID #REQUIRED
Code CDATA #REQUIRED // Rule: 9
Name CDATA #REQUIRED CourseIDREF IDREF #REQUIRED>
<!ATLIST Course
Rule 8: Adding Relationships through Containment CourseID ID #REQUIRED
(One-to-One/Many or Single Parent Relationship) Code CDATA #REQUIRED
If the relationship is one-to-one or one-to-many in the direction we
selected for navigation, and no other relationship leads to the child
within the selected subset, then add the child element as element Rule 10: Adding Missing Elements
content of the parent element with the appropriate cardinality which
For any element that is only pointed to in the so far created
is ? for on-to-one and * for one-to-many.
structure, add that element as allowable element content of the root
In our direction of traversal, there is a one-to-many relation and a element. Set the cardinality suffix of the element being added, to *.
many-to-one relationship, from Student to StudentCourse and then This is because it is only pointed to, not contained.
from StudentCourse to Course, respectively.
In our case Course is the missing element, as it is being 'pointed by'
<!ELEMENT UniversityCourseRegistration (Student*)> and has not 'pointed to'.
<!ATLIST UniversityCourseRegistration <!ELEMENT UniversityCourseRegistration (Student*, Course*)>
UniversityName #REQUIRED // Rule: 10
Location #REQUIRED <!ATLIST UniversityCourseRegistration
ContactNo #REQUIRED> UniversityName #REQUIRED
Location #REQUIRED
<!ELEMENT Student (StudentCourse*) > // Rule: 8 ContactNo #REQUIRED>
<!ATLIST Student
StudentID ID #REQUIRED <!ELEMENT Student (StudentCourse*)>
RegNo CDATA #REQUIRED <!ATLIST Student
Name CDATA #REQUIRED StudentID ID #REQUIRED
Address CDATA #REQUIRED RegNo CDATA #REQUIRED
Hobies (Cricket | Other) #IMPLIED> Name CDATA #REQUIRED
<!ELEMENT StudentCourse EMPTY> Hobies (Cricket | Other) #IMPLIED>
StudentCourseID ID #REQUIRED> <!ELEMENT StudentCourse EMPTY>
<!ELEMENT Course EMPTY> StudentCourseID ID #REQUIRED
<!ATLIST Course CourseIDREF IDREF #REQUIRED>
80
<!ELEMENT Course EMPTY> // Missing Element Hobbies="Reading"

<!ATLIST Course <StudentCourse CourseIDREF= CS5678/>
CourseID ID #REQUIRED <Course CourseID= CS5678>
Name CDATA #REQUIRED Name=" MMSIP "
CreditHours CDATA #REQUIRED> CreditHours="3"
</Course>
Rule 11: Remove Unwanted ID Attributes (created in Rule 5)
<Student RegNo="MS063014"
ID attributes not referenced by IDREF attributes in the DTD must Name="Maaz"
be removed. Address="D. I. Khan"
<!ELEMENT UniversityCourseRegistration (Student*, Course*)> Hobbies=""
<!ATLIST UniversityCourseRegistration <StudentCourse CourseIDREF= CS5678/>
UniversityName #REQUIRED <Course CourseID=" CS5678">
Location #REQUIRED Name=" MMSIP "
ContactNo #REQUIRED> CreditHours="3"
</Course>
<!ELEMENT Student (StudentCourse*)> </UniveristyCourseRegistration>
<!ATLIST Student
RegNo CDATA #REQUIRED // Rule: 11
Hobies (Cricket | Other) #IMPLIED>

<!ATLIST StudentCourse // Rule: 11
CourseIDREF IDREF #REQUIRED>

<!ATLIST Course
// Cannot Remove CourseID
Code CDATA #REQUIRED Fig. 3 Data in Relational Tables
C. In-efficiencies in the Publishing rules
- Rule 5 creates ID attributes only for the sake of its own

B. Produced XML Document processing and then deletes them in Rule 11, which is an
overhead.
- The concept of IDREF introduced in Rule 9 brings additional
processing or indirection; we can represent (M î M)
From the data of Figure 3 the XML document produced will be:
<?xml version="1.0"?> relationship of Relational DB in XML document by

introducing a Sub Tag, e.g. Course Tag in Student Tag where,
<!DOCTYPE UniversityCourseRegistration there is (M î M) relationship in Student and Course.
"http://MAJU/mmdb/secondDraft.dtd" >
IV. PROPOSED ALGORITHM (DWTOSW) FOR “RDB
<UniversityCourseRegistration UniversityName="MAJU"
TO XML CONVERSION”
Location="Blue Area, Islamabad", ContactNo="051-111878787">
<Student RegNo="MS071017"
Name="Peer" INPUT: SQL Create Queries
Address="Mansehra" OUTPUT: XML DTD
Hobbies="Reading" 01- Create Root Element detail as entered by the User
<StudentCourse CourseIDREF=CS1234/> // Required to hold Semantic Information
02- For Each Table’s Create Query
<Course CourseID=CS1234> 03- If TABLE has PK
Name="MMDB" 04- {
CreditHours="3" 05- If TABLE has ONE FK
</Course> 06- {
07- Create ELEMENT & ATLIST tags for this table
<Student RegNo="MS063014" 08- place its name in the ELEMENT tag whose primary
Name="Maaz" key is referred
Address="D. I. Khan" 09- }
10- else if TABLE has more than ONE FKs
Hobbies=""
11- {
<StudentCourse CourseIDREF=CS1234/> 12- Create ELEMENT & ATLIST tags for this table
<Course CourseID="CS1234"> 13- place its name in the ELEMENT tag whose primary
Name="MMDB" key is referred
CreditHours="3" 14- add subsequent FK table names in the ELEMENT
</Course> tag mentioned in 11
<Student RegNo="MS071017" 15- }
Name="Peer" 16- else
17- {
Address="Mansehra"
81

19- } Employee EmpSkill
20- }
EID 1 * EID
21- else if the TABLE has Composite Key
22- { Name Skill
23- if TABLE has more than ONE FKs NOT part of
Composite Key
Desig
24- {
26- place its name in the ELEMENT tag whose
Figure: 4
primary key is referred
27- add subsequent FK table names in the ELEMENT The DTD of Figure: 4 after executing
tag mentioned in 11
- lines 03 & 18 for Employee, and
28- }
29- else if ONE Key in Composite Key, is PK in Another - lines 21 and 29-32 for EmpSkill
TABLE is given below.
30- {
31- Create ELEMENT & ATLIST tags for this table // Root
32- place its name in the ELEMENT tag whose primary <!ELEMENT EmployeeIS (EmpSkill*)>
key is referred <!ATLIST EmployeeIS
33- } CompanyName #REQUIRED
34- else if More than ONE Keys in Composite Key, are
URL #REQUIRED>
PKs in other TABLEs
35- { <!ELEMENT Employee (EmpSkill*)>
36- place ONE attribute’s table name in other another <!ATLIST Employee
Attribute’s Table EID CDATA #REQUIRED
37- Ignore this Table (the Bridge/Join table) Name CDATA #REQUIRED
38- } Desig CDATA #REQUIRED>
39- }
<!ELEMENT EmpSkill EMPTY>
<!ATLIST EmpSkill
V. EXAMPLE CASES EID CDATA #REQUIRED
Skill CDATA #REQUIRED>
Case 1: One-to-Many
Case 2: Many-to-Many (Normalized in RDB using the Bridge/Join)
Book Accession
Book BKAuthor
BID BID
1 * BID 1 * BID
Name AccNo
Name AuthorID
ISBN
ISBN
Figure: 3 *
1
The DTD of Figure: 3 after executing

Author
- lines 03 & 18 for Book, and
- 05 – 08 for Accession AuthorID
Figure: 5
is given below.
// Root Name
<!ELEMENT LibrarySystem (Book*)> Address
<!ATLIST LibrarySystem
University #REQUIRED
LibraryName #REQUIRED
URL #REQUIRED>
The DTD of Figure: 5 after executing
<!ELEMENT Book (Accession*)> - line 03 and 18 for Table Book and Author, and
<!ATLIST Book - line 21 and 34-37 for Table BKAuthor
BID CDATA #REQUIRED is given below
ISBN CDATA #REQUIRED> // Root
<!ELEMENT LibraryIS (Book*)>
<!ELEMENT Accession EMPTY> <!ATLIST LibraryIS
<!ATLIST Accession University #REQUIRED
BID CDATA #REQUIRED LibraryName #REQUIRED
AccNo CDATA #REQUIRED> URL #REQUIRED>
<!ELEMENT Book (Author*)>

<!ATLIST Book
82
BID CDATA #REQUIRED

ISBN CDATA #REQUIRED>
<!ELEMENT Author EMPTY> <!ELEMENT Reservation EMPTY>
<!ATLIST Author <!ATLIST Reservation
AuthorID CDATA #REQUIRED RID CDATA #REQUIRED
Name CDATA #REQUIRED PID CDATA #REQUIRED
Address CDATA #REQUIRED> FID CDATA #REQUIRED>
<!ELEMENT Flight EMPTY>

<!ATLIST Flight
Passenger Reservation FID CDATA #REQUIRED
DepTime CDATA #REQUIRED>
PID 1 * RID
Name PID VI. CONCLUSION
Address FID
The proposed algorithm DWToSW (Deep Web To Surface Web) is
* mapped on different examples from RDBMS and was found that it
1 generates the DTD efficiently. It’s DTD is more comprehensive and
understandable to present schema as compared to the traditional
Flight technique of 11 publishing rules for RDB to XML conversion.
Figure: 6
FID REFERENCES
DepTime
[1] George Lapis. “XML and Relational Storage–Are they mutually
exclusive?”, IBM Corporation, 2005
The DTD of Figure: 6 after executing http://www.idealliance.org/proceedings/xtech05/papers/02-05-01/
- line 03 and 18 for Table Passenger and Flight, [2] Kevin Williams, Michael Brundage, Patrick Dengler, Jeff Gabriel,
- and, line 03 and 10-14 for Table Reservation Andy Hoskinson, Michael Kay, Thomas Maxwell, Marcelo Ochoa,
is given below. Johnny Papa, Mohan Vanmane, “Professional XML Databases”,
WROX Press, 2005.
[3] Jennifer Chu-Carroll, John Prager, Krzysztof Czuba, David Ferrucci ,
// Root and Pablo Duboue, “Semantic Search via XML Fragments: A High-
<!ELEMENT AirTicketReservationSystem (Passenger*, Flight*)> Precision Approach to IR”, Proceedings of the 29th annual
<! ATLIST AirTicketReservationSystem international ACM SIGIR conference on Research and development in
AirLine #REQUIRED information retrieval, Seattle, Washington, USA, Pages: 445 – 452,
URL #REQUIRED> 2006.
[4] www.en.wikipedia.org/Information retrieval
<!ELEMENT Passenger (Rservation*, Flight*)> [5] Michael Champion, “Native XML vs. XML- enabled: The difference
makes the difference”
<!ATLIST Passenger
http://www.softwareag.com/xml/library/xml_outlook_2001.htm
PID CDATA #REQUIRED [6] “Microsoft SQL Server on-line Help”
Name CDATA #REQUIRED http://msdn2.microsoft.com/en-us/library/ms169620.asp
Address CDATA #REQUIRED>
83
Session III Conference Chapter on Communication
Chair Prof. Dr. Salikhov Tagaimurod Khaitovich

Department of Physics
Kohat University of Science & Technology, Pakistan
ANALYSIS OF VoIP BY USING DIFFERENT VOCODERS IN IEEE 802.11b

WIRELESS NETWORKS
Ihsan Ul Haq1, Khawaja M. Yahya1, Ghulam Mubashir Hassan1, Tariq M. Jadoon2
1
Department of Computer Systems Engineering,
NWFP University of Engineering and Technology, Peshawar Pakistan
{ihsan, yahya.khawaja, gmjally@nwfpuet.edu.pk}
2
Department of Computer Science, LUMS, Lahore, Pakistan
jadoon@lums.edu.pk
Abstract — VoIP over Wireless-LANs (VoWLANs) is coding.. AMR is adopted as the standard speech codec by 3GPP in
gaining importance due to its excellent potential in wireless October 1998 and is now widely used in GSM [ 4].
voice communication. With the rapid deployment of WLANs,
there is a growing demand for WLANs to support voice As VoIP permeates from the wireline to the wireless world,
applications. performance issues arise due to differences in the characteristics of
In this paper we present simulation results for seven voice wireline and wireless networks. Delay and jitter are the key factors
coders i.e. G.711, G.723.1, G.728 LDCELP G.729, G.726 that impact packet voice quality in the fixed Internet and are further
ADPCM, G.729 CS-ACELP, and GSM under similar load magnified in a WLAN environment. Due to access point congestion
conditions for interactive Voice over WLAN system utilizing the and poor link quality high delay variation is not unusual in an
IEEE 802.11b wireless LAN standard. The analysis is based 802.11b network. Such a high jitter complicates proper
upon packet end-to-end delay and variance in delay reconstruction of the speech signal at the receiver and so packet
measurements. In our previous work [1], we suggested G.723.1 voice quality in WLAN environment can be severely degraded [5].
codec for transmission over WLAN when there was only voice
traffic. This paper extends our earlier work and explores the The objective of this paper is to simulate different vocoders for
performance of seven different codecs when integrated hybrid voice/data traffic scenario and to analyze the voice
Voice/Data traffic is transmitted simultaneously over WLAN. characteristics like Packet End-to-End Delay and Packet Delay
From our simulation results, we again emphasize the use of the Variation for different encoders. This paper is organized as follows.
G.723.1 vocoder that provides least end-to-end delay and packet In Section II, we present the related work, in Section III our
delay variation when the WLAN is used for the transmission of simulation model is discussed, Section IV presents and discusses the
integrated Voice/Data traffic. results, and Section V provides conclusions.
Index Terms — Vocoders, WLANs, VoWLAN II. RELATED WORK
I. INTRODUCTION Shiao-Li Tsao presents the general procedure to establish a voice

over wireless LAN (VoWLAN) and describes various problems and
challenges to both WLAN technologies and VoIP systems [2].
WLANs have been widely deployed in recent years. Voice over IP Among other challenges, QoS is an important issue.
over WLAN, also called Voice over WLAN (VoWLAN or wi-fi
phone), is regarded as one of the killer application for public and Jong-O Kim, Hideki Tode, and Koso Murakami discuss mixing
private WLANs. However, VoWLAN suffers from a number of voice and data traffic over a WLAN and investigate the QoS [6].
technical challenges that make the commercial and large VoWLAN Due to its strict delay requirement, voice is severely interfered by
services difficult to provide [2]. But this technology is expected to Best effort (BE) data. They also propose an enhanced medium
experience dramatic growth in the near future [3] . control scheme, called vDCF which is designed to simply enhance
the legacy DCF.
Voice over IP (VoIP) is a form of voice communication that uses
data networks to transmit voice signals. For transmission the voice Quan and Hui [7] analyze a polling scheme called Deficit Round
signal needs to be encoded at one end of the communication channel Robin (DRR) polling scheduling algorithm to enhance the
and sent as packets through the data network. The encoded signal is performance of a WLAN to support a voice application.
decoded at the receiving end and transformed back in the form of
voice signal. There are different vocoders like G.711, G.723.1, Kamesh Medepalli, Praveen Gopalakrishnan, David Famolari and
G.729, G.726 ADPCM, G.728 LD-CEPT, and G.729 CS-ACELP, Toshikazu Kodama present analytical and simulation results for the
GSM etc that can be used for voice transmission over Wireless local conversational speech capacity of WLANs [8].
area networks (WLANs).
Veeraraghavan, N.Cocher, and T. Moors [9] use the PCF mode of
The G.711 is an ITU-T standard audio codec and represents voice the 802.11 MAC protocol (which uses a polling scheme) to carry
signal sampled at the rate of 8000 samples/second using logarithmic voice traffic. The objective of their work is to limit the number of
Pulse-Code Modulation (PCM). Similarly, the G.729 standard call to guarantee delay within bounds.
compresses voice audio in chunks of 10 milliseconds frames. It is
mostly used in Voice over IP (VoIP) applications for its low Ping Wang, Hai Jiang, and Weihua Zhuang propose several
bandwidth requirement in wireline networks. This codec operates at mechanisms to enhance voice quality of service for hybrid
8 kbit/s. The G.723.1 is another audio codec for voice that voice/date traffic in a WLAN environment [10].
compresses voice audio in 30 ms frames and is mostly used for
VoIP applications due to its low bandwidth requirement. The G.726 We have simulated and analyzed the effect of different voice
is an ADPCM speech codec standard with transmission rates of 16, encoders in a hybrid voice/data traffic environment for transmission
24, 32, and 40 kbit/s. The most commonly used mode is 32 kbit/s over WLAN and suggested an encoder that can provide the highest
which is half the rate of G.711, and consequently increases the user satisfaction based on our simulation results.
usable network capacity by 100%. Adaptive Multi-Rate (AMR) is
another audio data compression scheme optimized for speech
84
III. SIMULATION MODEL confidence interval for 90% confidence level have been calculated
for each vocoder. All these statistics are available in Table 2 and 3.
This section discusses the simulation model used to analyze
interactive voice communications over a WLAN in an integrated
Voice/Data scenario. The simulation has been performed using
OPNET Modeler 10.5 and eight different scenarios have been
designed to study the effect of Voice coders on the performance of
VoIP over WLAN. In each scenario, all the physical characteristics
of the WLAN have been kept constant like data rate of the network
i.e. 11 Mbps. The voice encoder is changed in each scenario to study
its effect over speech transmission.
The infrastructure BSS network model that consists of WLAN

nodes in a 100 x 100 meter2 area modeling an office environment is
used for simulation as shown in Figure 1.
Fig. 2 Wireless Station Parameters
TABLE I
VOICE ENCODER DETAILS
Frame Coding Speech

S. Vocoder Size Rate (kbps) Activity
No (msec) Detection
1. G.729 10 8 Disabled
2. G.723.1 30 5.3 Disabled
4. G.726 10 32 Disabled
ADPCM
CS-ACELP
Fig. 1 WLAN (Infrastructure BSS) Network Model 6. GSM 20 13 Disabled
The WLAN parameters chosen for all workstations are shown in As a particular random seed value can potentially result in an
Figure 2. The data rate is set to 11 Mbps and Direct sequence spread anomalous or non representative behavior. Multiple seed values
spectrum (DSSS) is used to emulate an IEEE 802.11b WLAN. The have been used to obtain standard or typical behavior. The graphs
802.11b amendment to the original standard was ratified in 1999. for Packet End-to-End Delay and Packet Delay Variation are shown
802.11b has a maximum raw data rate of 11 Mb/s and uses the in Figure 3 and 4 respectively. The tabular and graphical simulation
CSMA/CA media access method defined in the original standard. results clearly indicate that G.723.1 voice coder provides least
Due to the CSMA/CA protocol overhead, in practice the maximum mean, standard deviation, minimum value and maximum value for
throughput that an application can achieve is about 5.9 Mb/s using both Packet End-to-End Delay and Packet Delay Variation, which is
TCP and 7.1 Mb/s using UDP [11]. The details of voice encoders highly desirable for real time application like voice communication.
used for the simulation are presented in Table 1. The UDP/IP has
been used for voice application as being traditionally implemented Moreover, it has been observed that the introduction of FTP traffic
for multimedia transmission [12] and TCP/IP has been implemented increases the jitter. In one of the scenario, G.723.1 voice encoder is
for data application. implemented for interactive voice traffic only but under same
conditions, the FTP has been introduced into the network and it has
been observed that the average Packet Delay Variation has
IV. RESULTS increased from 9.86E-08 to 3.43E-07 and standard deviation has
increased from 2.67E-09 to 7.29E-08. This result indicates that the
data traffic (FTP Traffic) has an effect over the quality of voice.
•
The metrics that are collected for analysis include:
•
End-to-End Delay (sec) &
Packet Delay Variation (sec) V. CONCLUSIONS
Seven different voice encoders have been used in the simulation and
We compared the performance of seven audio codecs i.e.G.729,
the Packet End-to-End Delay and Packet Delay Variation has been
G.723.1, G.711, GSM, G.726 ADPCM, G.728 LC CELP and G.729
analyzed for a WLAN carrying voice and FTP traffic. The
CS-ACELP in a WLAN environment under similar load conditions
simulation has been designed so that the Packet End-to-End Delay
i.e. voice is interactive and data traffic is best effort. The simulation
and Packet Delay Variation is measured after every 7.2 seconds.
results show that the use of the G.723.1 audio codec in a WLAN
The mean, standard deviation, minimum value, maximum value and
85
TABLE II
STATISTICAL COMPARISON FOR PACKET END-TO-END
TABLE III
STATISTICAL COMPARISON FOR PACKET DELAY
Fig. 3 Packet End-to-End Delay Comparison for Different Vocoders
86
Fig 4 Packet Delay Variation Comparison for Different Vocoders
gives least Packet End-to-End Delay and Packet Delay Variation

and thus can provide the acceptable user satisfaction for Voice over
WLAN.
[6] Jong-Ok Kim, Hideki Tode, and Koso Murakami, “Friendly
REFERENCES Coexistence of Voice and Data Traffic in IEEE 802.11 WLANs”, IEEE
Transaction on Consumer Electronics, Volume 52, Issue 2, Page(s)
347-354, 21 March 2006.
[1] Ihsan Ul Haq, Khawaja M. Yahya, Raziq Yaqoob, Tariq M. Jadoon, [7] Wan Quan, Du Ming Hui, “Improving the Performance of WLAN to
“Effect of Speech Activity Detection in IEEE 802.11 Wireless Support VoIP Application”, IEEE Wireless communication.
Networks” IEEE International Conference on Emerging [8] Kamesh Medepalli, Praveen Gopalakrishnan, David Famolari and
Technologies”, 12-13 Nov- 2007, Rawalpindi Pakistan. Toshikazu Kodama, “Voice Capacity of IEEE 802.11b, 802.11a and
[2] Shiao-Li Tsao, “Research challenges and perspectives of voice over 802.11g Wireless LANs”, Second International Conference on Mobile
wireless LAN”, IEEE Emerging Information Technology Conference, Technology, Applications and Systems, 15-17 November 2005.
15-16 Aug 2005. [9] M. Veeraraghavan, N.Cocher, T. Moors, “Support of Voice
[3] Ping Wang, Hai Jiang, Weihua Zhuang, “IEEE 802.11e enhancement Services in IEEE 802.11 Wireless LANs,” IEEE INFOCOM 2001,
for voice service”, IEEE Wireless Communications, Volume: 13, issue Volume 1, page(s) 488-497, 22-26 April 2001,
1, Page(s): 30 – 35, Feb 2006. [10] Ping Wang, Hai Jiang, and Weihua Zhuang, “Performance
[4] http://en.wikipedia.org/wiki/ List_of_codecs Enhancement for WLAN Supporting Integrated Voice/Data Traffic”,
[5] Miroslaw Narbutt and Mark Davis, “An Assessment of the Audio IEEE International Conference on Communications, Volume 8, page(s)
Codec Performance in Voice over WLAN (VoWLAN) Systems”, IEEE 3808-3813, June 2006.
Proceedings of the Second Annual International Conference on Mobile [11] J. Postel, “User Datagram Protocol”, RFC 768, August 1980.
and Ubiquitous Systems, Page(s):461 – 467, 17-21 July 2005. [12] http://en.wikipedia.org/wiki/802.11_b # 802.11b
87
Using Mobile Phones for Quizzes in m-Learning Environments

Haider Ali 1, Dr. Saeed Mahfooz 2
1 CECOS University Peshawar
2 Department of Computer Science University of Peshawar
haiderkhalil@yahoo.com, saeedmahfooz@yahoo.com
Abstract — The emergence of wireless networks has give conclusion of the research work. In section 5 we give a food of
introduced new ideas like m-commerce, m-banking, m-learning thought for researchers for future work in this area.
etc. Majority of the people now have mobile phones which they
use for different purposes in addition to making and receiving II. RELATED WORK
phone calls. In all educational institutions Student evaluation is
a major task performed by instructors for grading of students
and quizzes are often used for this purpose. In this paper we Researchers have worked in the field of using handheld devices for
identify some problems in the current methods of taking quizzes tests and have proposed methods that are based on either Short
and then propose a solution by suggesting a novel method for Message Service (SMS) or Wireless Application Protocol (WAP).
taking the quizzes using handheld devices like mobile phones
and PDAs. The idea of using handheld devices for tests in classes was provided
by [1]. In this paper the researchers provided the idea of using
Index Terms — SMS, WAP, J2ME. handheld devices for taking concept tests in the classes. The concept
test was used to determine whether the material covered in the class
was understood by the students or not? They experimented it on a
I. INTRODUCTION Windows CE device. They found that most of the students preferred
the use of handhelds for this purpose to the use of a show of hands or
Almost all educational institutions have significant portion of holding up of flashcards. In their research they suggested that this
weight allocated for internal tests/quizzes for students grading. The approach was used only for small concept tests and cannot be used
questions in the quizzes are of the form true/false, multiple choice, for taking actual tests used for grading of the students. They
fill in the blanks and/or free text answers. suggested that in order to take such tests, a large database is to be
maintained to which the questions can be added by topic and when
There are certain problems in the current methods of taking the required a test can be taken. They also suggested for an improved
quizzes that are given as under: security considerations for such tests.
1. Due to increased workload the instructor doesn’t find
sufficient time to prepare the test properly. Another method of M-Quiz using SMS was given by [2]. In this
2. Composing, printing and photocopying of the quiz paper, method a low-cost SMS-based system is proposed. The proposed
and checking of the students’ answers and tabulation also method targets average budget conscious educational institutions. In
takes instructors’ time. this method the questions are presented to students in PowerPoint
3. Sometime due to rush on the photocopying machines the slides. The students answer the questions through an SMS message
instructor leave the test paper with the photocopier and they follow certain guidelines while writing answers in the
operator which leads to security vulnerability. SMS. The students get their results by connecting to the Internet.
4. Students demand for a quick response of their test results
but usually the results are delayed due to time constraints Another work of using SMS for M-quiz was done by [3]. In this
of the instructor. method the questions are sent to the students in SMS. The answers
5. Comparison of different tests for individual students is not to the questions are also sent to the students using image
carried out to know whether a particular student is steganography. Steganography is “the art through which writing is
improving or otherwise. hidden requiring recovery by the minds of men” [4]. Another
6. Comparison of students’ attendance, final exam result and definition given by [3] “Steganography means hiding information in
tests is not taken into account. If we compare the students’ a cover media so that the others will not notice that such information
attendance with the test results it may give us very useful exists.” Since the answers are given in the steganographic image
information; for instance students with good attendance these cannot be revealed by the students. The student after
get good marks but if the case is not like, it may lead to completing the quiz gets his/her grade without connecting to the
some other questions like “was the topic not taught Internet. The grade of the student is sent to the instructor in an SMS
properly?”, “was the test too difficult?” etc. If we message.
compare the test results and the final exam result the
percentage of marks obtained should be comparable The research work by [2] and [3] is a good step forward in the field
otherwise it will mean many things like “either the final of M-Quiz but lacks a common feature and that is the availability of
exam or the test were too difficult”, “the final exam didn’t an easy to use interface for the instructor for creating a questions
covered the full course” etc. database, creating a test from selected topics and automatic
7. Students answer sheet transcripts are usually lost when checking of the test results through software. Our research
the result is prepared. For students these are useful to see addresses these problems and gives a complete picture of the
their weaknesses, for instructors they are useful to check proposed M-Quiz method. We call our method as jM-Quiz because
back records. it requires a java-enabled handheld device for students.
These problems become worse as the number of students in the III. THE SUGGESTED METHOD
classroom increases.
As with previous methods our method has two main actors i.e. the
In this paper we propose a solution to these problems. In section 2 instructor and student. The proposed architecture of our proposed
we have given detail of the related work; the problems in the method is given in figure-1.
methods of the related work are also identified in this section. In
section 3 we give detail of our suggested method. In section 4 we
88
• The instructor also enters the maximum time for completing

the test, the total number of questions and the total marks
• When the instructors presses the CREATE TEST button, the

allocated to this test.
Create test module selects at random the required numbers

of questions from the topics selected and stores them
separately and gives the instructor a unique Test ID for that
• The instructor provides the test ID to the students which is

test.
used by the students for downloading the test.
Fig. 10 Architecture
As given in figure-1 the proposed architecture has following

components:
1. A web server; we suggest Apache web server.
2. MySQL database for quiz database
3. Server side php scripts for interacting with instructor and
j2me client application on student’s handheld device.
4. J2ME-enalbled handheld device like mobile phone or
PDA.
5. Cleint application written in j2me and installed on student
handheld device.
A. The Instructors Activities
Fig. 2 Adding Questions to Questions Database
The proposed method will provide a web-based interface to the
instructor that will be accessed by the instructor through a PC and
Internet connection to perform the following tasks:
1. Adding questions to the question database
2. Creating a test
3. Compiling the test results
1) Adding questions to the question database

The system offers facility to establish a questions database. The
instructors can add questions as and when he/she gets spare time.
• Course ID: as the questions may be added for different

Following entries are entered by the instructor to add questions:
• Topic: This is the selected from the list of the topics given in
courses,
the course outline. Purpose of storing the topic is that when

the test is taken the instructors may select from different
• Question: The question written in free text format.

topics.
• Answer options: A list of options that will be displayed to the Fig. 3 Creating the Test
• Correct Answers: A list of correct answers is stored in the

student, from which the student will select.
3) Compiling the test results
• Test result of a particular test showing all student IDs with

database such that when the students submit his answers, The instructor may create certain analysis reports like:
they are compared with the correct answers and the result is
• Individual students report showing detail of all the tests. This

prepared. their score. This report is displayed to students.
• Topic-wise scores showing the topic, correct and incorrect

Multiple instructors teaching the same subject can add questions report is e-mailed to individual students.
to the questions database. The questions database is kept on a server
with all instructors having access to it. answers to examine the weak areas of the students.
B. The Students Activities
2) Creating a test
The instructor may create a test/quiz by invoking the ‘Create Test’
The students interact with the system using a java-enabled handheld
• A drop-down list of courses is presented to the instructors; the
module. Following steps are taken to create a test:
device like mobile phone or PDA. The students have to perform the
following tasks during the course of different interactions with the
• When the course ID is selected by the instructor, he is
instructor selects the course ID.
system:
presented with a list of topics. The instructor may select 1. Installing the client application on mobile device.
multiple topics. 2. Downloading the test.
3. Performing the test.
4. Submitting the answers.
5. Viewing the test result.
89
1) Installing the client application on mobile device IV. CONCLUSION

The client application needs to be installed on student’s mobile
device for interaction of student with the system. The application
Quizzes are one of the most commonly used tools used by
can be installed by using Over the Air (OTA) protocol from the web
instructors for evaluation of their students. In this paper we
server. The application may also be downloaded to the PC and then
identified some problems faced by instructors in taking quizzes. We
connecting the mobile phone with PC via USB cable, infrared or
have studied the related work of taking the quizzes using handheld
Bluetooth.
devices. The proposed methods in the related research work were
thoroughly studied and some problems were identified. The major
2) Downloading the test
problems in these proposed methods were that little attention was
The instructor announces the Test Code to the students, the students
given to the instructor interface for adding questions in the database,
give the Test Code and test is downloaded in the student’s mobile
making a new test and automatic checking of answers. In this paper
device.
we have proposed a method that addresses all the key areas of the
M-Quiz application.
3) Performing the test
When the student starts the test the question are displayed on the
student mobile device and the student can answer the questions by V. FUTURE WORK
using the keys of the mobile device. The system provides navigation
buttons for moving forward and backward in the test paper. While some components of our proposed system have been
developed, the development of complete prototype system needs
attention. We will recommend the development of complete
prototype system for this method. The prototype application can
then be tested on quizzes for different courses in an educational
institution.
VI. REFERENCES
[1] C. Franklen, M. Brad, and Y. David, “Using Handheld Devices for Tests
in Classes”, Human-Computer Interaction Institute School of
Computer Science Carnegie Mellon University Pittsburgh, PA 15213,
July 2000.
[2] Alexi Triakov, and Kinshuk, “Creating a Pervasive Testing Environment
by using SMS”, Proceeding of IEEE International Workshop on
Wireless and Mobile Technologies in Education (WMTE’05), IEEE
Computer Society, 2005.
[3] Mohammad Shirali Shahreza, “M-Quiz by SMS”, Proceedings of the
Sixth International Conference on Advanced Learning Technologies
(ICALT'06), IEEE Computer Society, 2006.
[4] J.C Judge, “Steganography: Past, Present, future”, Lawrence Livermore
National Laboratory US Department of Energy 2001.
[5] André N. Klingsheim, Vebjørn Moen, and Kjell J. Hole, “Challenges in
Securing Networked J2ME Applications”, Computer Magazine, IEEE
Computer Society, pp. 24-30 August 2007.
[6] O. Chito, J. M. Leonard, I. Black, V. S. Joshep, “The Mobile Conference
Information System, Unleashing academic conferences with Wireless
Mobile Computing”, Proceeding of the 35th Hawaii International
Conference on System Sciences (HICSS-35’02), 2002.
[7] Jason Lam, “PHP Interacting with J2ME”, http://www.devarticles.com
accessed on January 2008.
[8] James Keogh, “J2ME - The Complete Reference”, Tata McGraw-Hill
Publishing Company Limited 2003.
Figure 4: Taking the Test [9] Stephen W. Thorpe, “Online Student Evaluation of Instruction: An
Investigation of Non-Response Bias” 42nd Annual Forum of the
4) Submitting the answers Association for Institutional Research Toronto, Canada, June 2002.
When the student finishes answering all the questions, he/she [10] Julie C. Meloni “Sams Teach Yourself PHP, MySQL and Apache in 24
presses the submit button and the answers are sent to the web server Hours”, Sams Publishing 2002.
using http. In this portion of the system the j2me code on the client
device interacts with the php code on the server side.
5) Viewing the result

When the answers are submitted by the student these are
automatically checked by the software on the server and the results
are stored in the database. The results can be seen by the student by
connecting to the web server using http.
90
LOAD BALANCING FOR CONGESTION CONTROL IN MPLS NETWORKS

Peer Azmat Shah, Muhammad Maaz Rehan
Department of Computer Science,
Mohammad Ali Jinnah University Islamabad, Pakistan.
Email: {peer.azmat@yahoo.co.uk, maazrehan@yahoo.com}
Abstract — Traffic Engineering has become an essential re-allocation mechanism, section 5 discusses a load balancing
tool in the design and implementation of large networks as the scheme for congestion control, section 6 presents an idea of priority
requirement for efficiency, reliability and QoS has increased. base congestion control in MPLS networks, section 7 gives an idea
Data communication technologies are fastly developing in the of QoS using traffic engineering, section 8 presents a load
recent years but the growth of data traffic in networks is very balancing algorithm supporting QoS, section 9 presents critical
faster than the capacity of network. Network traffic delays are review and section 10 concludes the paper.
usually the result of congestion in networks, and these
congestions can cause packet delays and consumption of more II. CONGESTION CONTROL MANAGEMENT
network resources, as a result decreasing the Quality of service
QoS. One of the main objectives in optimizing network resource
utilization is to balance the network load by bandwidth The main objective of the ISP’s is reducing traffic congestion to
re-allocation and re-routing techniques. In this paper a provide better QoS. There are various types of congestion control
comparison of different load balancing schemes for MPLS schemes [9]. These falls into three categories: Response time scale
networks using re-allocation and / or re-routing is made. which may be described as long, short and medium, Reactive vs.
Preventive and Supply side vs. Demand side.
• Response Time Scale

Index Terms — MPLS, Traffic Engineering, QoS,
Congestion Control
These can be categorized as;
1. Long when it refers to the capacity upgrading of network by
I. INTRODUCTION anticipating future traffic demands. It may take weeks to
months to upgrade network.
Multi Protocol Label Switching (MPLS) [10] is a data-carrying 2. Medium when it refers to monitoring the traffic distribution
mechanism that belongs to the family of packet-switched networks. and utilization of network resources that subsequently
MPLS operates at an OSI Model layer that is generally considered provides feedback to online or offline traffic engineering
to lie between traditional definitions of Layer 2 (data link layer) and mechanism.
Layer 3 (network layer), and thus is often referred to as a "Layer 3. Short, when it refers to packet level for control / signaling
2.5" protocol. It was designed to provide a unified data-carrying congestion function.
• Reactive Vs. Preventive:

service for both circuit-based clients and packet-switching clients
which provide a datagram service model. It can be used to carry
many different kinds of traffic, including IP packets, as well as Reactive strategies respond to congestion problems by reducing
native ATM, SONET, and Ethernet frames. MPLS as a new them, while Preventive strategies avoid the congestion by
technology tries to overcome the problems of traditional IP predicting future congestions.
• Supply side vs. demand side:

networks routing by adding a label to IP packet and forwarding this
packet on the basis of label.
Supply side congestion management policies increase the
Traffic Engineering is a technique to control the routing of traffic available capacity to traffic demands so as to decrease
through network and ensuring efficient and balanced use of network congestion. With Demand side traffic is regulated to improve
resources and also satisfying the QoS requirements [1]. The main congestion problems.
objective is the optimization of performance of Network by efficient
utilization of network resources. This may be done in MPLS III. TRAFFIC ENGINEERING CAPABILITIES OF MPLS
network by creating new label switching paths (LSP’s) or by the
NETWORKS
re-routing of existing LSP’s so as to decrease the congestion and
splitting the overall traffic among different parallel LSP’s. The LSP
configurations work well as long as the network conditions do not In MPLS IP header analysis is done only at the ingress router. This
change from the data initially given. However in dynamic networks process simplifies and speeds up the routing process. A label
this type of mismatch can not be avoided between traffic forecast assigned in an IP header associates to a Forwarding Equivalence
and actual load. Due to this some LSP’s may be underutilized and Calss (FEC). Packets in a FEC are treated in the same manner. The
some over utilized [1]. So a load balancing mechanism must be binding of labels and FEC’s is done by Label Distribution Protocol
implemented which can minimize the over and underutilization of (LDP). From traffic engineering perspective flow of MPLS network
LSP’s. should be examined before making a path. If traffic of similar type
(belonging to same FEC) is mapped onto the LSP performance
LSP’s can be rerouted due to many reasons such as failure or could be better.
recovery of the node. When a link fails LSP is re-routed. Re-routing
consumes processing power and is time consuming, so we want to Traffic engineering operations can optimize the performance of
improve the rerouting mechanism and to reduce the number of networks both at traffic level and at resource level. Using the
LSP’s rerouted. The key idea is to efficiently re-route LSP’s from explicit routes, gives the ability to manage network resources
most congested links in the network so as the overall links load is efficiently and support new services. Traffic trunk has traffic of
balanced and allow better use of resources. same class and is routed along the same path. All Traffic inside a
traffic trunk has the same label and same 3-bit class of service [7].
The rest of the paper is organized as follows. Section 2 describes the Traffic trunk can be set up between any two nodes and same egress
Congestion control policies, section 3 presents traffic engineering node can have multiple parallel trunks. Trunking results in
capabilities of MPLS networks, section 4 presents a bandwidth separation in competing traffic flows and automatically leads to a
91
traffic engineering ability and better QoS. Traffic engineering set of alternate paths to re-route an LSP are found. And the
implementation on Linux can be seen in [5]. re-routing of traffic over the alternate LSP is executed.
A. Working of DYLBA
IV. A BANDWIDTH RE-ALLOCATION MECHANISM
Dynamic Load Balancing Algorithm calculates the network load by
Jose L Marzo [2] proposed a dynamic virtual path allocation using a parameter x, which indicates the threshold value for the link
scheme. For load balancing, to increase network performance, a residual bandwidth which is measured as a fraction of link capacity.
virtual path uses two ways. First way is to establish more than one X determines when a link is considered congested. DYLBA is
path from host to destination. When connection is established, use triggered only when setup of new LSP causes the detection of
the most convenient path. Second way is the reallocation of network congestion. i.e. when residual bandwidth left on a link is x.
resources among different paths. First set of alternate paths are found and a most promising move is
selected randomly and then rerouting of traffic over alternate LSP is
executed.
VI. PRIORITY BASED CONGESTION CONTROL IN MPLS

NETWORKS
Priority based congestion control [4] has the goal to manage traffic
to achieve maximum throughput by maximizing the capacity of
network resources with very low packet loss. This type of technique
requires a close collaboration between sending and receiving nodes
to determine the network traffic. The system presented uses the
approach that ingress router collaborates with the receiver instead of
sender communicating with the receiver [4]. When a packet comes
at the ingress router it is assigned an MPLS label based on
Forwarding Equivalent Class (FEC). While assigning FEC, the
ingress router using the IP priority of packet establishes the sending
Fig. 1 Bandwidth re-allocation and re-routing
rate. IP packet priority will be assigned by the sender or sender’s
router before arriving at the ingress. The ingress starts sending
When different paths of network are congested, then a connection packets and changes the sending rate dynamically by using the
that could be accepted, in case of better traffic, is rejected. If in the response from receiver. Receiver informs ingress by ACK about
same link paths are congested then the bandwidth is reconfigured i.e sending rate. It is important to determine when an ACK packet
bandwidth reallocation occurs and if all the paths in a link are would be sent, to prevent the receiver from sending an ACK on
congested and no bandwidth is available, then routes are altered i.e every packet arrival, because sending an ACK for every packet
path re routing occurs. If a link is shared by two or more LSP’s then results in an increase usage of network resources causing a decrease
to fulfill the request of an LSR the bandwidth of LSP’s can be in maximum capacity of network. For this purpose receiver setup a
reallocated in such a manner that QoS does not affect. timer which when expires ACK is sent to ingress. It is important to
make sure that timer is neither too short nor too long. If ACK are
V. A LOAD BALANCING SCHEME FOR CONGESTION CONTROL sent too often, this will result in unnecessary traffic and cause an
IN MPLS NETWORKS increase in network congestion. With too few ACK’s will cause
decrease in QoS. Thus receiver decides the sending rate [4].
A technique to reduce congestion in MPLS networks by using a load
balancing mechanism [3] is based on local search method. Basic VII. QUALITY OF SERVICE USING TRAFFIC ENGINEERING
idea is to efficiently reroute LSP’s from most congested links, in OVER MPLS
order to balance overall links load and to allow a better use of
network resources. Most of congestion control schemes are To provide QoS in the network Wei Sun [6] presented an analysis of
preventive. These techniques allocate paths in the network so that to performance of TCP and UDP flows using MPLS traffic trunks.
prevent the congestion. Traffic trunks carry a number of micro-flows carrying different
transport types such as TCP or UDP. Mixing different type of
The network presented by Elio Salvadori [3] consists of n routers transport can cause performance problems, unfairness for certain
and a subset of ingress-egress routers. When a connection request traffic. Services received by TCP and UDP flows are compared
arrives at an ingress router then it determines the explicit route for when they share either a link or an MPLS traffic trunk. When TCP
LSP by inspecting the topology and available residual capacity of and UDP flows are mixed in a trunk, TCP flows receive reduced
each link. Each router in the network must know the network service as the UDP flows increase their rate.
topology and the residual capacities of each link.
A. End –to-end trunks
A request for connection i is defined by (ii , ei , bi ) [3]. Where i is TCP and UDP have very different behavior when packets are lost
ingress router, e is egress router and b is the bandwidth required. TCP reduces its traffic, whereas UDP traffic is not affected when the
The LSP’s are routed through the network according to routing packets are lost.
scheme. The virtual load of a link is determined by summing the
bandwidth bi. The residual bandwidth is equal to the difference Some experimental result has also been made in [6] which shows
between link capacity and the virtual load. If the residual bandwidth that when the TCP flows and UDP flow are using the same LSP, and
is minimum then it means that path is congested. Paper presents a UDP flow increase its rate ,TCP flow decreases and when UDP flow
load-balancing algorithm named as DYLBA [3]. The algorithm reaches to maximum bandwidth, TCP flow become zero.
dynamically balances the bandwidth (load) by inspecting each link
that whether it is congested or not. If a link is found congested then a By separating the UDP flow from TCP flow using class based
queuing [6] so as bandwidth is properly allocated to all different
92
trunks at the router. Here when we increase the UDP flow rates, In [2] a resource management system was presented. When the
TCP source is not affected too much. Although fair constant rate of congestion appears permanently a change in the configuration of
TCP is achieved at the overhead by maintaining the separate queue network is done. In this situation a traffic that was present on the
for each of the trunk at each of the router [5]. LSP is stopped and there is no restoration of failed traffic. Also
when some LSP becomes congested and some under utilized, a
B. Non –end –to-end Trunks.
connection that can be established, if traffic load was balanced, is
Here trunks are not end to end. Trunks are initialized at Router 2, rejected. It might be possible that this connection has very high
here the flow interfere with each other for some part of path, since priority but resource management system does not accept this
Router1 does not make a distinction between various flows. At R1 connection. A solution for this technique is presented in [4].
the flow are treated almost identical and during the period of
congestion TCP reduces there flow rate. This lead to very poor In [4] a technique is discussed that the sending rate is decided by
throughput for TCP source although they are treated distinctly at receiver and not by sender. This is done by sending ACK from
R2. receiver to sender. A good thing in the paper is that it uses priority
based congestion control.
Therefore different type of flows should be secluded in different
traffic trunks so as to guarantee the QoS. [3] Presents an algorithm DYLBA which checks each link on
network whether it is congested or not, if it is congested then it finds
VIII.A LOAD BALANCING ALGORITHM SUPPORTING QOS FOR alternate paths and the traffic is sent through that alternate path. The
TRAFFIC ENGINEERING IN MPLS paper does not give any idea that when no alternate path is found
what will happen. [8] discusses load balancing algorithm it suggests
that if load on LSP1 is increased and LSP2 load is less, then transfer
Queue tuning algorithm (QTA) [8] is used to support QoS for the traffic to LSP2. But it is ideal situation that load on LSP2 may be
Traffic engineering in MPLS networks. This algorithm maps less, if no LSP is found having less load, traffic should be dropped
different type of traffic on different queues while the order of the which is drawback of this algorithm. [7] Gives idea that TCP traffic
packet within a flow of traffic remains unchanged. Traffic flow must be separated from UDP traffic flow so as to increase
engineering need load balancing and load splitting. Traffic with throughput of TCP but does not give a mechanism to separate these
different QoS requirements is distributed by the QTA. QTA checks two types of traffic.
the state of LSP and distributed the traffic dynamically to LSP’s to
improve their efficiency.
X. CONCLUSION:
TE architecture consists of three parts, Traffic splitter, packet from
incoming link is received at traffic splitter and then packets are In this survey paper different techniques were studied to achieve
divided into different parts. Traffic distributor forwards the packet QoS by efficient bandwidth management and congestion control in
to different LSP’s seeing the result of measurement analysis. MPLS Networks. All the techniques have advantages as well as
Measurement analysis measures the delay, jitter, and bandwidth of shortcomings e.g Priority based congestion control does not support
all LSP’s. Traffic splitting is executed at the packet forwarding path the ACK from receiver to sender for each packet received but it
for every packet, and there it introduces per packet overhead, packet handles packets on the basis of priority. Similarly load balancing
splitter need to avoid packet disorder within a TCP flow. scheme and QoS in IP based core networks do not support the
priority based congestion control.
Different traffic splitting algorithm has been suggested in past.
Among all of these hash based traffic splitter performance is best. REFERENCES:
Hash based CRC16 [8] is more complex than other systems but it is
implemented on high speed systems more efficiently.
[1] Sanda Dragos, Radu Dragos. “Bandwidth management in MPLS
A. Working of Load balancing Algorithm networks”
[2] Jose L Marzo. “Towards QoS in IP based core networks”. Acciones
Only ingress and egress router of MPLS network uses QTA. It does
Integrates program ref. HI1998-0032
not know about the intermediate router. Two types of traffic are [3] Elio Salvadori, Roberto Battiti. “A load balancing scheme for
described, best effort traffic (BE) and expedited forwarding traffic congestion control in MPLS networks”. Proceeding of eigth IEEE
(EF). international symposium ISCC’2003.
[4] Scott Fowler. “Priority based congestion control in MPLS
Implementation architecture is in [6].Classifier divides the packet networks”. Proceeding of advance industrial conference on
into two different type of traffics. EF traffic should be forwarded so telecommunication IEEE 2005.
fast that QoS requirement is achieved. BE traffic needs no QoS [5] A Hafid. N. Nataranjan. “Adaptive Dynamic Traffic Engineering”.
requirement. Different type of traffic enters to different queue IEEE 2006.
[6] Wei Sun, Parveen Bhaniramka “QoS using Traffic Engineering over
seeing the result of Hash based CRC16. QTA maps the traffic from
MPLS” Proceedings of IEEE 25th annual conference, 2000.
queue onto the LSP so as to obtain the QoS requirement. [7] Mansoor Alam “Study of traffic Engineering capabilities of MPLS
networks” Proceedings of IEEE International conference on
IX. CRITICAL REVIEW information technology, 2005.
[8] Bing-Feng Cui “Aload balancing Algorithm supporting QoS for
Traffic Engineering in MPLS networks” Proceedings of fourth
The papers studied so far for survey of “Towards QoS by efficient international conference on computer and information technology,
bandwidth management and congestion control in MPLS networks: IEEE 2004.
An analysis*” present different techniques for efficient bandwidth [9] D. Awduche, A. Chiu, A.Elwalid, I. Widjaja and X. Xiao.” Overview
management and congestion control in MPLS networks. There are and Principles of Internet Traffic Engineering” RFC 3272 IETF
many issues to be discussed in these. 2002.
[10] http://en.wikipedia.org/wiki/mpls
93
INTEGRATION BETWEEN WLAN & MANET

Sumaira Kanwal 1, M.A Ansari 2
1 SZABIST Islamabad, Pakistan
2 FUUAST Islamabad, Pakistan
sumaira_kan@hotmail.com, drmaansari@fuuastib.edu.pk
Abstract — In recent few years, many efforts has been put This approach adapted here is based on the DSDV Discovery that
into account for solving most of the important challenges of scale well and integrate naturally with the existing IEEE 802.11
Wireless Local Area Network (WLAN) and Ad-hoc network. Model, while retaining much of the coverage area of MANET.
WLAN has wireless connection within a network it based on Before due to different protocol architecture both networks can not
fixed infrastructure and range of WLAN is approximately recognize each other, but now D-WLAN architecture is know for
100m. In Ad hoc due to mobility of node or the absence of fixed both WLAN and MANET. So communication between them is
infrastructure, this network can be deployed anywhere any possible.
time. One of the emerging forms of Ad-hoc network is Mobile
Ad-hoc Networking (MANET) that advocates wireless II. SYSTEM OVERVIEW
interconnection between devices which will be self organized
and can be extend or operate in recital with the wired A. WLAN
networking infrastructure or can grow in Autonomous Wireless LAN technology is comparatively slower than wired LAN.
networks too. Because of intermediate nodes in MANET which Wireless LANs have a nominal data transfer rate of between 11 and
have arbitrary and malicious behavior a new security challenge 54 Megabits per second (Mbps) compared to most wired LANs in
exist in MANET as compared to the wired network which has schools which operate at 100Mbps. Newly installed wired networks
fixed infrastructure.Mobile ad-hoc networks are also a good can now operate at up to 1,000Mbps (1Gb). [8]
alternative in rural areas or third world countries where basic
communication infrastructure is not established. Minimal B. Working of WLAN
configuration and quick deployment make mobile ad-hoc In a typical LAN configuration, receiving, buffering, and
networks (MANET) suitable for emergency situations like transmitting information between the wireless LAN (WLAN) and
natural disasters or military conflicts. In this Paper a D-WLAN the wired network infrastructure is performed by a device called
architecture model is proposed for integration between WLAN Access Point (AP). A single access point supports a number of users
and MANET to make them to communicate with each other via and functions within a specified range (the maximum number of
a single node which is an intermediate node using feature of devices connected simultaneously for Bluetooth standard is seven).
both WLAN and MANET. The access point and its antenna are usually mounted on a ceiling or
wall to obtain the best coverage. [9]
Index Terms — WLAN, MANETs, AP, BSS, ESS
Ranges of AP depend on the power of output units. Normal range
I. INTRODUCTION for class 1 units is 50 to 70 meters. Usage areas are restricted to
client’s needs. Among the multiple possibilities we can find:
conference rooms, home, hotels, boats, trains, hospitals, factories,
D-WLAN architecture model is designed to provide significant airports, etc.[9]
improvements in system capacity and performance relative to data
sharing in networking environment.
III. ARCHITECTURE OF WLAN
The proposed architecture is motivated by the fact that there are a lot
A. IEEE 802.11 Architecture
of similarities between WLAN and MANET, but communication
between both of them was not possible due to the different protocol Fig 1 shows that model developed by the 802.11 working group.
architecture. If we have both network together and want to share the Basic service set (BSS) is the basic building block of wireless LAN.
data of each other so firstly should know the location of nodes
with-in the network. Thus requiring effective integration of wireless The BSS may connect to a backbone distribution system (DS)
local area network “WLAN” with the Mobile ad-hoc wireless through an access point (AP). The DS can be a switch, a wired
network (MANET) nodes. network, or a wireless network. [10] The association between a
station and a BSS is dynamic. Stations may turn off, come within
This D-WLAN architecture model is applicable to a number of range, and go out of range.
emerging ad-hoc networking scenarios including extended wireless
local-area networks and home wireless networks. An extended service set (ESS) consists of two or more basic service
sets interconnected by a distribution system. Typically, the
WLAN network covers approximately 100 m range dependent on distribution system is a wired backbone LAN but can be any
fixed infrastructure and MANET also covers a small area and it is communications network. The extended service set appears as a
self organized network. single logical LAN to the logical link control (LLC) level.
In WLAN an IEEE 802.11 architecture model is used where AP is the logic within a station that provides access to the DS by
movement of data packets with in network is controlled by Data link providing DS services in addition to acting as a station. To integrate
layer. Data Link layer is divided into two layers 1) MAC layer 2) the IEEE 802.11 architecture with a traditional wired LAN, a portal
LLC layer. is used. The portal logic is implemented in a device, such as a bridge
or router, that is part of the wired LAN and that is attached to the
In MANET every node keeps track of its neighboring node in form DS. [10]
of table due to the DSDV protocol. If location of node is known then
nodes can communicate with each other. In this paper we are
embedding DSDV (discovery) protocol into the data link layer of
IEEE 802.11 architecture.
94
treated as a separate layer, called medium access control (MAC).

IEEE 802.x The separation is done for the following reasons. [10]
•
LAN
ESS Extended The logic required to manage access to a shared-access
Service Set medium is not found in traditional layer 2 data link
Portal
•
control.
For the same LLC several MAC options may be
Distributed system provided.
IV. AD-HOC NETWORKS

AP AP
STA 1 STA 5 In Wireless Ad-hoc Network the communication links are wireless,
BSS this is called Ad-hoc because each node is willing to forward data
BSS for other nodes so the decision which node forward data is made
Basic Service Set
STA 2 STA 3 dynamically based on the network connectivity due to mobility
STA 6 issue. They can allow users to access and share data, applications,
STA 4 STA 7
internet access or other network resources in the same way as wired
networks. [8]
Fig. 1 IEEE 802.11 Architecture [10]
B. Protocol Architecture of WLAN

Fig 2 shows the IEEE 802.11 Protocol Layers Compared to OSI
Model. This architecture was developed by the IEEE 802 committee
and has been adopted by all organizations working on the
specifications of LAN standards. It is generally referred to as the
IEEE 802 referred model. [10]
Fig. 3 Ad hoc [1]
Working from the bottom up, the lowest layer of the IEEE 802
referenced model corresponds to the physical layer of OSI model
V. MANETS
•
and includes such functions as
•
Encoding/ decoding of signals
•
Preamble generations/ removal (for synchronization) It is a collection of nodes which do not depend on a predefined
Bit transmission /reception infrastructure to keep the network connected. Due to
Above the physical layer of the functions associated with providing non-dependency on fixed infrastructure it can be formed, merged
services to LAN users. These include the following: together or partitioned into separate network on fly to manage it
operations. Adhoc networks node are often mobile so that they
apply wireless communication to maintain its connectivity so they
are called (MANET) Mobile Adhoc Networks while it may contains
static and wired nodes which can use the services offered by the
fixed infrastructure.[3]
A. Characteristics of MANETs
MANETs have several characteristics that have to be identified
•
when discussing the underlying technology of ad hoc networks.
Topologies are dynamic. Nodes are free to move
arbitrarily, which means that topology may change
•
randomly and rapidly.
•
Broadcast transmission overhearing is possible.
Network is bandwidth-constrained and capacity of the
links may vary. Wireless links have significantly lower
•
capacity than wired links.
•
MANET is a bidirectional connection.
Operations are energy-constrained. Most nodes in a
MANET are usually running on batteries or on other
•
exhaustible means.
The physical security is limited. Ad hoc networks are
Fig. 2 IEEE 802 Protocol Layers Compared to OSI Model [10]
generally more prone to physical security threats than
•
fixed-cable networks. [3]
On transmission. Assemble data into a frame with address
B. Architecture of MANETs
•
and error detection field.
On reception. Disassemble frame, and perform address A mobile ad hoc network (MANET) consists of mobile nodes that
connect themselves in a decentralized, self-organizing manner and
•
recognition and error detection.
may also establish multi-hop routes. Each MANET interface is a
•
Govern access to the LAN transmission medium.
Provide an interface to higher layers and perform flow and broadcast interface, typically, but not necessarily, wireless. With
error control. [10] only those nodes which are within its coverage area. In fig the
These are the functions typically associated with OSI layer 2. The coverage area is indicated by a fixed radius disc, but in real world it
set of functions in the last bullet item are grouped into a logical link is variable according to environment. [2]
control (LLC) layer. The functions in the first three bullet items are
95
A. Protocols
MANET routing protocols are divided into three categories.
1. Proactive protocols
2. Reactive protocols
3. Hybrid protocols
B. DSDV (Destination Sequenced Distance Vector)
DSDV is proactive protocol and traditional distributed shortest-path
protocols which maintain routes between every host pair at all
times. Based on periodic updates; high routing overhead. It also
maintains the routes which are never used.
Fig. 4 MANET [6]
Nodes (N) with MANET interfaces. The light grey area indicates DSDV protocol keeps the simplicity of Distance Vector. In this
the coverage area of each MANET interface. The dark grey circle protocol Guaranteed Loop Freeness. This means that there will be
indicates the interference area of the MANET interface of N3. new table for entry of Destination Sequence Number. It also allows
1) Hidden Terminals fast reaction to topology changes. It makes immediate route
A property of MANETs which is commonly brought forward is the advertisement on significant changes in routing table but wait with
”hidden terminal problem”: if N3 through some protocol agrees advertising of unstable routes (damping fluctuations) [3].
with its neighbors (N2 and N4) that it will, for the moment, have
exclusive access to the wireless media via its MANET interface,
VII. PROPOSED MODEL
then N3 may go ahead and make a transmission. However, if at the
same time N1 also transmits over its MANET interface, then the
transmissions of the MANET interfaces of N1 and N3 may appear WLAN and MANET both are small range networks. Due to
concurrently at the MANET interface of N2 potentially interfering different protocol architecture both networks can not communicate
and causing N2 to receive neither of the transmissions. [2] with each other even they are in range of each other. In fig 6 the
2) Symmetric vs Asymmetric Connectivity proposed system is trying to give an idea of sharing the features of
We can say that if communication from one interface to another MANET with WLAN. As MANET is self organized architecture
interface is possible in one hop, then communication in the inverse and can be formed anywhere anytime and discovers its nodes. The
direction is also possible, so the connectivity between neighbor’s main feature of MANET which is used in this proposed model is
interfaces is symmetric. [2] DSDV (Discovery) protocol. DSDV maintains routing table at each
node. Each node maintains the information of all available
Let suppose the coverage area of N1 is large due to some reasons destinations, the next neighboring node, the no of hops to reach the
like powerful transmitter or large antenna and N2 can receive the destination in routing table. This table is send to all neighboring
transmission of N1. on the other hand the coverage area of N2 is nodes. Bi-directional links are required. DSDV protocol makes
smaller the N1 in such a way that the transmission from N2 can not immediate route advertisement on significant changes in routing
be received by N1. table, but wait with advertising of unstable routes. When new node
is added, it immediately sends its sequence no. to all its neighboring
nodes and nodes updates their table.
Fig. 5 MANET: neighbor asymmetry. [6]
Thus an asymmetric – or more precisely, a unidirectional

connectivity between the MANET interface of N1 and the MANET
interface of N2 exists: N2 sees N1 as a neighbor (since the MANET
interface N2 can receive transmissions from the MANET interface
of N1), whereas N1 does not see N2 as a neighbors (since the
MANET interface of N1 can not receive transmissions from the
MANET interface of N2). Thus, MANET neighbor relationships are
non-reflective. [2]
VI. ROUTING PROTOCOL IN MANET

Fig. 6 Purposed Layer Structure D-WLAN Architecture/ Model.
Each node in an ad hoc network participates in forming the network
topology; each node is on its own part responsible for routing
packets between other nodes, too. A D-WLAN architecture model is proposed for communication
between MANET and WLAN when they are in range of each other.
There are many different routing protocols that provide information Both network can share information between them, when nodes of
to forward packets to the next hop. In ad hoc network it would be these systems are close to each other. Whenever one or more nodes
necessary to manage topology changes, as all the nodes are required of MANET are in the range of WLAN, both networks can
to run routing protocols. [3] communicate with each other by an intermediate node in which
DSDV (Discovery) protocol are used.
96
A. Discovery
In traditional ad-hoc networks, there is no discovery phase and the DSDV (Discovery)
routing protocol itself is responsible for building up topologies Upper Layers
either using on-demand broadcast of route requests or by
exchanging neighbor information proactively with one hop
neighbors. While this may be sufficient for smaller networks, as the LLC
number of nodes increases, it results in denser physical topologies, Data Link
leading to extensive routing message exchanges. The problem is
more severe in a multi-channel network where the multiple nodes
that need to communicate could be on different radio channels. In MAC
this case, the routing messages need to be propagated across Physical
multiple channels in order to enable data transfer from one node to
the other. [7]
B. DSDV (Discovery) protocol Fig. 7 Data Flow in the layers of D-WLAN
The main function of DSDV is to maintain the data base of each and
every node of network by using this discovery feature. DSDV
maintains routing table at each node. Each node maintains the DSDV active
information of all available destinations, the next neighboring node,
the no of hops to reach the destination in routing table. This table is
send to all neighboring nodes. Bi-directional links are required.
DSDV protocol makes immediate route advertisement on
significant changes in routing table, but wait with advertising of WLAN Range
unstable routes. When new node is added it immediately send its
sequence no. to all its neighboring nodes and nodes updates their
table. B C
In proposed model, which is D-WLAN architecture, a new sub layer
is introduced in data link layer of IEEE 802 model. This new sub
layer named DSDV (Discovery) having the features of discovery
protocol of DSDV.
In fig 6 DSDV discovery protocol is embedded on the second layer
(i.e. Data Link Layer) of IEEE 802 Model, with the two already
existing layers that is LLC and MAC. So the features of DSDV
discovery protocols are also added in the Data Link layer of IEEE Fig. 8 Ranges of WLAN and MANET
802 Model.
C. Data Flow As shown in fig 8 , due to MANET mobility node B comes in the in
the range of WLAN. Node B becomes a DSDV active node now
In IEEE 802 model data travels in form of packets. Whenever a
and it will use the D-WLAN architecture with embedded DSDV
packet arrives from the lower layer, this layer handles the packet
(Discovery) protocol. Node B already have data of node A and
processing and passes the information to the higher layers.
immediately update its table and add information of new node
Packets moves from physical layer to data link layer which is
which is node C .Node B will recognize node A and also node C.
divided into two sub layers (MAC & LLC). Packet then enters in
Node B will introduce Node A to node C as node B is familiar with
MAC layer then forwarded to LLC layer.
the location of node A and node C due to the feature of DSDV
Data flow in D-WLAN architecture in same sequence as in IEEE
protocol. So node B now becomes an intermediate node between
802 architecture but when D-WLAN encounters a MANET node,
node A and node C.
third sub layer of data link layer become active and data from LLC
layer is passed to DSVD (Discovery) layer and then passed to upper
layer of model. In fig 9 communications between nodes of MANET and WLAN is
DSDV (discovery) maintain record of every neighboring node and shown. Node B has the information of node A and when it comes in
information related to nodes in form of table and this table is passed range of WLAN, it becomes a DSDV active node and it
to every neighboring node, so WLAN can share the features of immediately updates the data. Node C now sends its sequence no to
MANET through the DSDV discovery protocol. DSDV Discovery node B so node B now has information of node C. Now node B can
handles the discovery of nodes with in network. [3] communicate with node A and node C and in this context node A of
MANET can communicate with node C which is node of WLAN. In
D. Working of D-WLAN
proposed model provide integration between WLAN and MANET.
MANET is small network which can be created anywhere anytime. MANET features are now embedded in WLAN and both networks
Nodes in MANET are mobile. When any node of MANET finds can communicate with each other.
WLAN network with-in its range. DSDV (Discovery ) protocol
become active and immediately maintain record of neighboring
node. Nodes of WLAN become the neighboring node of MANET.
So nodes of WLAN can communicate with nodes of MANET
through an intermediate node which can be a MANTE node which
enters in range of WLAN. It is possible now that both network can
communicate with each other and can share data of each other.
Fig. 9 Data Flow between nodes
D-WLAN architecture is proposed integration between WLAN and

MANET.
97
VIII. CONCLUSION
In today’s technological world one of the fastest growing [2] Jinyang Li, Charles Blake, Douglas S. J. De Couto, Hu Imm Lee, and
technologies is wireless communication which also has importance Robert Morris, “Capacity of Ad Hoc Wireless Networks”, in the
in networking field. Due to the enhancement in wireless era made proceedings of the 7th ACM International Conference on Mobile
easy deployment of wireless communication networks in Computing and Networking, Rome, Italy, July 2001
infrastructure as well as Ad-hoc environment too. In comparison in [3] Vesa Kärpijoki, Security in Adhoc Networks Helsinki University of
reference to the security issues there is a difference in these Technology Telecommunications Software and Multimedia
networks one has a centralized authority (infrastructure network) Laboratory
[4] Humayun Bakht, “Understanding mobile ad-hoc Networks”
while other not. In this paper introduction about the WLAN and
[5] Donatas Sumyla “Mobile Ad-hoc Networks (manets)”
MANET (Mobile Ad-hoc Network) one of the hottest type of [6] Thomas Heide Clausen “A MANET Architecture Model”, January 2007
AdHoc network today in detail. [7] S. Ganu, S. Zhao, L. Raju, “architecture and prototyping of an
802.11-based self-organizing hierarchical ad- hoc wireless network
In this paper comparison of MANET with WLAN is described on (sohan)”, WINLAB, Rutgers University, 73 Brett Road, Piscataway,
the basis of architecture and after finding the similarities between NJ 08854
both, D-WLAN architecture model is suggested by using this [8] www.eng.nene.ac.uk/~idc/nets3/WLAN.doc
architecture integration between WLAN and MANET is possible. [9]http://www.paradigma.cl/English/blue2net1_blue2net. Html
[10] William Stallings “Wireless Communications and Networks” 2002
[11] http://www.javvin.com/protocolWLAN.html
REFERENCES [12] http://www.windowsecurity.com/img/upl
[13]http://www.kiikm.org/images/hierarchical networks .gif
[1] P. Gupta and P.R. Kumar. “Capacity of wireless networks. IEEE [14] Mesh Networks (2001) MEA product sheet, 2001,
Transactions on Information” Theory, Volume 46, Issue 2, March 2000, <http://www.meshnetworks.com>.
Digital Object Identifier 10.1109/18.825799
98
HANDOFF ISSUES IN CELLULAR NETWORKS

Sumaira Kanwal 1, M.A Ansari 2
2 FUUAST Islamabad, Pakistan
sumaira_kan@hotmail.com, drmaansari@fuuastib.edu.pk
Abstract — In recent few years, many efforts have been put II. BACKGROUND
into account for solving most of the important challenges of
cellular telecommunication system. Continuation of an active A. Handoff /Handover
call is one of the most important quality measurements in the Continuation of an active call is one of the most important quality
cellular systems. Handoff process enables a cellular system to measurements in the cellular systems. Handoff process enables a
provide such a facility by transferring an active call from one cellular system to provide such a facility by transferring an active
cell to another. Handover and handoff are performed by all call from one cell to another. Different approaches are proposed and
cellular telecommunications networks and they are a core applied in order to achieve better handoff service. The principal
element of the whole concept of cellular telecommunications. parameters used to evaluate handoff techniques are: forced
The concept of handover or handoff seems relatively termination probability and call blocking probability. The
straightforward; it is not an easy process to implement in mechanisms such as guard channels and queuing handoff calls
reality. The process is quite complicated, and in early systems decrease the forced termination probability while increasing the call
calls were often lost if the process did not work correctly. The blocking probability. [12]
factor lead to the customer dissatisfaction is failure in proper
handover that can result in dropped calls, which in turn may B. The Handover Concept
lead to them changing to another cellular network provider. In In cellular systems or wireless systems the freedom to be able to
this paper the basic focus is on the interference due to the same make and receive calls anywhere, at any time, creating a totally new
frequencies during softer handover. Frequency collusion will dimension in human communications. Handovers are a key concept
disturb the connected calls, distortion will occur or it may lead in providing this mobility. It makes it possible for a user to travel
to drop the call. Authors here presented a new technique from one cell to another while having a seamless connection. [12]
namely as M-FCCH. By using this technique handover become
efficient and fast. M-FCCH allocates a beacon frequency as C. Types of Handover / Handoff
soon as MS enters in the conflicting area.
There are three types of hand off
•
Index Terms — MS, BS, BTS, BSC, MSC, PSTN, GSM,
•
FCCH, Handoff/ Handover Hard Handover
•
Soft Handover
I. INTRODUCTION Softer Handover
1) Hard Handover
In GSM communications, there exist two types of interference. 1. The definition of a hard handover or handoff is one where an
Adjacent cell interference 2. Co-channel interference. In this paper existing connection must be broken before the new one is
authors have focused on call dropping / distortion due to co-channel established. One example of hard handover is when frequencies are
interference. Co-channels are the same channels (or frequencies) changed. As the mobile will normally only be able to transmit on
that are used by different cells. To avoid this kind of interference, it one frequency at a time, the connection must be broken before it can
is necessary to separate the co-channel by as great distance as move to the new channel where the connection is re-established.
possible. But by doing so, channel capacity will be compromised. This is often termed and inter-frequency hard handover. While this
is the most common form of hard handoff, it is not the only one. It is
M-FCCH is designed to provide significant improvement in system also possible to have intra-frequency hard handovers where the
capacity and performance relative to handover. The proposed frequency channel remains the same. [12]
architecture is motivated by the fact that there are a lot of problems
during softer handover like interference in call or it may lead to call 2) Soft Handover
dropping. Normal burst is used to carry the data signals and control The new 3G technologies use CDMA where it is possible to have
channel which are initially used to establish a call. FCCH is used to neighboring cells on the same frequency and this opens the
find a free frequency to initiate a call at first time. If we get the possibility of having a form of handover or handoff where it is not
interference in same frequencies due to the reuse concept then it necessary to break the connection. This is called soft handover or
encounters the frequency clash, which may lead to call dropping. To soft handoff, and it is defined as a handover where a new connection
avoid this problem it requires effective handoff on a beacon is established before the old one is released. In UMTS most of the
frequency. handovers that are performed are intra-frequency soft handovers.[6]
M-FCCH architecture is applicable to every MS which establish a 3) Softer Handover
call in cellular network. Every call established in a cellular network The third type of hand over is termed a softer handover, or handoff.
first use the control channels to establish a call. FCCH is used to In this instance a new signal is either added to or deleted from the
allocate frequency at very first time when mobile establish call. active set of signals. It may also occur when a signal is replaced by a
Data / voice bits travel at normal burst and it can be sending by MS stronger signal from a different sector under the same base station.
and BTS both. This type of handover or handoff is available within UMTS as well
as CDMA2000.[6]
In this paper FCCH is embedded in normal burst. M-FCCH is used
to broadcast a beacon frequency as soon as MS encounters a
conflicting region.
99
A cellular phone user can only use his/her mobile within the covered
area of the network. Roaming is the capacity of a cellular phone,
registered on one system, to be able to enter and use other systems.
Those other systems must be compatible to enable roaming (i.e. they
must have the same type of networks). In Europe, the standard
cellular network is called GSM (Global System for Mobile
Communication). Incoming calls to GSM users are routed to them,
irrespective of where they are, as long as they are within Europe. [5]
2) Handoff Priority In Cellular System

The probabilities of forced termination have been reduced by
channel assignment strategies with handoff prioritization have.
Fig 1 : Soft handover scenario [6]
a) GUARD CHANNELS
Fig. 1: Soft handover scenario [6] By reserving a fixed or dynamically adjustable number of channels
exclusively for handoffs, guard channels improve the probability of
successful. For example, a reserve N channel for handoff among C
channels in the cell means that priority is given. The remaining (C –
N) channels are shared by both new calls and handoff calls. [10]
b) QUEUING OF HANDOFF
Queuing is a way of delaying handoff, the MSC queues the handoff
requests instead of denying access if the candidate BS is busy.
Queuing new calls results in increased handoff blocking probability.
The probability of a successful handoff can be improved by queuing
Fig. 2 Soft vs softer handovers[6] handoff requests at the cost of increased new call blocking
probability and a decrease in the ratio of carried-to-admitted traffic
D. Handoff Techniques in Cellular Networks since new calls are not assigned a channel until all the handoff
Cellular network is a wireless network communicates via radio requests in the queue are served. Queuing is possible due to the
network. Mobile unit is a basic unit of cellular system, which overlap region between the adjacent cells in which MS can
communicates through radio link to an infrastructure of switching communicate with more than one BS.[10]
equipment interconnecting the different parts of the system and
allowing access to the normal (fixed) Public Switched Telephone
Network (PSTN). [5]
Fig. 5 Illustrate ration of a handoff scenario at cell boundary [10]
III. PROBLEMS & REASONS OF HANDOVER IN

CELLULAR NETWORKS
Fig. 3 Cellular Network [1]
A. Power control of mobile station transmissions during handoff in
1) Operation Of The Cellular Phone cellular Networks
When a mobile phone is switched on, it registers with the A method and apparatus for controlling the transmission power of a
appropriate BS, depending on its location, and its cell position is mobile station during handoff of a call between base stations of a
stored at the responsible MSC. When a user makes a call, the base cellular system .The desired transmission power of the mobile
station monitors the quality of the signal for the duration of the call, station may be determined based on measurements made on the
and reports that to the controlling MSC, which in turn makes pilot channel of the target base station at the mobile station.
decisions concerning the routing of the call. Alternately, base station power classes may be defined for the
When a cellular phone moves from one cell to the other, the BS will cellular system so that each base station at which it is desired to
detect this from the signal power and inform the MSC of that. The receive mobile station transmissions at a signal level within a
MSC will then switch the control of the call to the BS of the new selected range is assigned to a power class associated with that
cell, where the phone is located. This is called handover. It normally selected range A mobile station involved in a handoff can determine
takes up to 400ms, which is not noticeable for voice transmission. the power level at which it should begin transmitting on a new
100
traffic channel during handoff by determining the power class of the BSC of the same MSC. This type of handover is known as
base station to which the new traffic channel is assigned. [14] inter-BSC handover.
B. Call Dropping E. Inter-MSC Handover
1) No Channels Are Available When a call is established and start moving it cross cell boundary ,
If signals of MS is getting low and before handing over call drops cross BTS boundary, cross BSC boundary and the cross MSC
due to unavailability of channel. Sometimes BSC is overloaded and boundary. When MS cross MSC boundary and signal strength
have no free channels are available. become weak then handover is made. The MS is handed over to
2) Prioritizing Handoff another channel on different cell, under another MSC of the same
One method for giving priority to handoffs is called the guard PLMN.
channel concept, whereby a fraction of the total available channel in
a cell is reserved exclusively for handoff requests from ongoing V. PERFORMANCE ISSUES IN CELLULAR
calls which may be handed off into the cell. This method has the NETWORKS
disadvantage of reducing the total carried traffic, as fewer channels
are allocated to originating calls. Guard channels, however, offer A. Frequency Reuse
efficient spectrum utilization when dynamic channel assignment
strategies, which minimize the number of required guard channels In a cellular system, frequency reuse is achieved by assigning a
by efficient demand –based allocation, are used. [14] subset of the total number of channels available to each base station,
3) Degrade Call Quality and controlling the power output of the transmitters. In this way, a
When a call is established and having a lot of distortion and cellular network increases capacity (number of channels available to
interference that the caller can not listen clearly, It means that call is users).Adjacent cells are not allowed to operate at the same
getting weak signals or bad quality signals so call handed over. This frequency since this causes interference between the cells. It would
type of handover can be any type handover like inter- handover or increase the capacity. But by doing so, a number of difficulties arise.
intra-handover. Interference: decreasing the cell size, especially with a low repeat
4) Waste Of Capacity In Multiple Signaling factor increases the problems of interference between cells which
Signal strike with the resistance like buildings, trees, walls etc… by are using the same frequency. [10]
sticking, signals become weak.
Handovers: Decreasing the cell size increases the frequency of
handovers, since a moving cellular phone would be changing cells
IV. YPES OF HANDOVER IN CELLULAR NETWORKS more often. Since the MSC needs time to switch (for handovers),
increasing the handovers will increase that time delay. [11]
In a cellular network, the radio and fixed links required are not
permanently allocated for the duration of a call. There are four Since each cell is designed to use radio frequencies only within its
different types of handover in the GSM system, which involve boundaries, the same frequencies can be reused in other cells not far
transferring a call between. [16] away with little potential for interference. The reuse of frequencies
is what enables a cellular system to handle a huge number of calls
1. Intra-Cell Handover with a limited number of channels. The objective is to use the same
2. Intra-BTS Handover frequency in other nearby cells, thus allowing the frequency to be
3. Intra-BSC Handover used for multiple simultaneously conservations. Generally, 10 to 50
4. Inter-BSC Handover frequencies are assigned to each cell, depending to the traffic
5. Inter-MSC Handover expected.
A. Intra-Cell Handover B. Co-Channel Interference

When call is established and it gets the weak signal at any area A cell using a set of frequencies can be used at some distance known
within the cell. As we know that different channels are allocated to a as frequency reuse. These cells can have co-channel interference.
single cell. The MS is handed over to another channel on the same Co-channels are the same channels (or frequencies) that are used by
cell, under the same BTS. This type of handover is known as different cells. To avoid this kind of interference, it is necessary to
intra-cell handover. separate the co-channels by as great distance as possible. But, by
doing so, channel capacity will be compromised. [11]
B. Intra-BTS Handover
C. Burst
When a call is established in a cell and start moving towards
neighboring cell and due to weak signal strength handover is made. When we talk about the channels in GSM, we are essentially
The MS is handed over to another channel on a different cell, under referring to what data and signals are sent across the air between the
the control of the same BTS. This type of handover is known as Base Terminal Station (BTS) and Mobile Station (MS). The burst
intra-BTS handover. structures define which form this transmission is done in, without
describing the details of the physical form of transmission, i.e. the
C. Intra-BSC Handover antennae, the power levels etc.[17]
When a call is established in a cell and start moving towards 1) Normal Burst
neighboring cells and cross the BTS boundary and enters in the Fig 6 shows the normal burst carries 142 bits of information, and it
range of new BTS. The MS is handed over to another channel on a transmitted by both the BTS and the MS. This kind of burst is a very
different cell, under the control of a different BTS of the same BSC. important one, as it carries the actual call data, as well as the
This type of handover is known as intra-BSC handover. signaling information. It has 2 sets of 57 bits of each, which carry
the information needed for call setup, maintenance and call ending,
D. Inter-BSC Handover
apart from the audio information. There is also a training sequence,
When a call is established in a cell and start moving towards or “midamble” of 26 bits used for equalization, which is required for
neighboring cell and cross cell boundary, cross BTS boundary and extracting the correct signal from the various reflections that are
then cross BSC. When MS cross BSC boundary and signal strength also received by the equipment, due to bouncing of signals in the
become weak then hand over is made. The MS is handed over to natural environment. This is a continuous process, and the
another channel on a different cell, under the control of a different equalization bits help compensate for any problems found in the
radio path, for e.g. by Rayleigh fading.
101
MS1 is using f1 frequency of cell 1 and MS2 already established a

call at f1 frequency at the same time when MS1 enters in the region
of cell 3.
B
G C
Fig. 6 Normal burst [17]
A
There are also 2 “stealing” bits on either side of the training Cell 3
F D
sequence bits. These bits are essentially stolen from the data bits, by
the Fast Associated Channel. This happens when there is a handover B E
from one base station to another, or when the Slow Associated G C
Channel is unable to send information fast enough.
A
Same behavior is obtained when there is a disruption in the RF Cell 1 A
F D
transmission, so that extra control information is sent (thereby Cell 2
degrading quality of voice), so that the link can remain established E
and the two communicating equipments are synchronized.
According to [17], these bits work either in “blank” or “burst” mode,
depending on whether they are stolen or not. Fig. 8 MS1 is moving from cell 1 to cell 3 with frequency f1
2) The Frequency Correction Channel (FCCH)
The FCCH is simply bursts of sine wave. This is the first channel the If MS1 enters in the range of cell 3 at the same time then graph will
mobile station (MS) searches for upon attempting to establish itself be as follows.
for the first time in a new sector. Each base station system (BSS)
broadcasts exactly one FCCH. The distinctive sine wave bursts MS1 using MS2 using
allow easy identification of the channel by the mobile, and since the f1 (cell 1) f1 (cell 3)
frequency correction channel is broadcast on the same frequency as
the SCH, it serves as an anchor by which to find the data channels
Frequency
(which otherwise would be indistinguishable from any other data
channels).The structure of the frequency correction burst is as below Collusion
in fig 7 [17].
Area
Time
Fig. 7 FCCH burst bit structure [17] Fig. 9 Graph of frequencies in cell 3 when MS1 & MS2 both using
same frequencies
The frequency correction burst transports 142 bits of information. Fig 9 is showing the frequency overlapping, when f1 frequency of
This is basically transmitted periodically from the BTS to the MS to MS1 which established in cell 1 and comes in the range of cell 3. F1
notify the equipment of adjustments in the frequency. It is frequency of cell 3 is already assigned to the MS2 and when MS 1
essentially an empty frame, i.e. full of zero bits. Thus its data entered in the range of cell 3, In result distortion and collusion
contents are not important. It however, sets the radio frequency for encountered. We can also say that, when MS1 and MS2 are using
use, and synchronizes timing information. This channel form parts same frequency in same region then collusion will occur and neither
of the GSM broadcast channel. A frequency correction burst I MS1 nor MS2 will be able to communicate properly. MS1 or MS2
transmitted on the channel which generate a beacon 6.7.7 KHz from connection can also be dropped due to this distortion. To avoid this
the cell carrier frequency. collusion or distortion a new technique is introduced in the normal
burst bit structure.
VI. PROPOSED SYSTEM
B. Frequencies Allocation Scenario
A. Overview The Mobile station (MS) and base stations (BTS) transmit or
If a MS1 established a call in cell 1 and MS start moving from cell communicate with each other on dedicated paired frequencies called
channels. Base stations use one frequency of that channel and
one to the other by using a frequency f1. Due to the concept of
mobiles use the other. So when MS2 in cell 3 established a call,
frequency reuse, cell 3 also have the same set of frequencies as cell
1 have, where call initially established. So cell 3 also has f1 technically a pair of frequency is allocated to it that is f1 frequency.
When MS1 enters in cell 3 it also has f1 frequency, It is difficult to
frequency. MS1 enters in the boundary of cell 3, if f1 frequency in
cell 3 is not used by any MS then there will be no problem in change the MS2’s (already dedicated) frequency. Certain channels
carry only cellular system data. We call these control channels. This
handing over frequencies. MS1 can continue using f1 frequency
with out any distortion and without any problem. But if some other control channel is usually the first channel in each cell. It is
MS2 have already established call using f1 frequency in cell 3 responsible for call setup, like FCCH (Frequency Correction
Channel).
before MS1 enters in the range of cell 3, collusion will occur.
FREQUENCY CORRECTION CHANNEL (FCCH)
In fig 8 cell 1 and cell 3 are using the same set of frequencies due to
the frequency reuse technique. FCCH (Frequency Correction Channel) is a control channel which
is used to allocate frequency at first time when Mobile station gets
started.
MS1 is Mobile station 1, it establish a call in cell and start moving
toward cell 3. Cell 3 is using a same set of frequencies as cell 1. As FCCH carries information of frequency correction for MS. Each
base station system (BSS) broadcasts exactly one FCCH. FCCH is a
control channel by using FCCH technique a new temporary
frequency is allocated to MS1. This frequency could be a beacon
102
frequency or any unused frequency of cell 3. The FCCH carries In fig 12 Graph is showing that now MS1 and MS2 both have
information for frequency correction of the MS. different frequencies and there is no more collusion in frequencies.
No interference is encountered due to same frequency. Chance of
C. M-FCCH (Modified FCCH)
call dropping due to same frequency has been finished.
When MS1 will enter in the range of cell 3 where frequency conflict
can occur due to the same frequency, M-FCCH will immediately
send message to BSS and BSS broadcast new available frequency VII. CONCLUSION
and temporarily allocate it to MS1. This frequency can be a beacon
frequency or any unused frequency of cell 3. In today’s technological world one of the fastest growing
1) STRUCTURE OF M-FCCH technologies is wireless communication which also has importance
M-FCCH is combination of normal burst and FCCH burst. As we in networking field. Due to the enhancement in wireless era made
know that normal burst carry data and FCCH burst carry control easy deployment of wireless communication networks. This
signals. Bits structure used in FCCH burst is 148 bits. This structure network is wirelessly connected to BTS, BSC and MSC and MSC is
would be embedded in normal bust structure. Normal burst structure connected with a wired structure to PSTN. Handover and handoff
also has 148 bits, to avoid the over head of normal burst structure we are performed by all cellular telecommunications networks, and
use only 1 bit of FCCH from 142 fixed bits. As fig 7 shows that in they are a core element of the whole concept of cellular
148 bits of FCCH 6 bits are TB (Tail Bits) which already part of telecommunications. Many problems occur during the handover in
Normal burst. Only one bit from the center portion shown in fig 7 which call compromises. There is a need of efficient and fast
would be used. This 1 bit can be 0 or 1 at a time. So it also avoids the handover to overcome the problems related to handover.
overhead of the normal burst. Total bits of M-FCCH will be 149.
In this paper concentration is on softer handover problem. M-FCCH
26 enciphered 8.25
3 TB 57 data 1 1 57 data M-FCCH 3 TB GP is proposed, this approach adapted here is based on the FCCH,
bits
bits bits bit which is a control channel used for frequency control or to broadcast
message. M-FCCH broadcast a new beacon frequency to MS1. In
Fig. 10 Structure of M-FCCH M-FCCH frequency control channels are merged in the normal burst
2) WORKING OF M-FCCH data channel to overcome the problem of softer handover by
When MS1 with a frequency f1 will enter in the region of cell 3 allocating a beacon frequency. It prevents the collusion of the same
where MS2 has already been established a call by using frequency frequencies. M-FCCH gives the faster and efficient handover.
f1. MS1 having a M-FCCH will send a frequency request to BTS
directly and BTS will allocate a new temporary beacon frequency to REFERENCES
MS1. This frequency could be a beacon frequency or any unused [1] Nishith D. Tripathi,Nortel Jeffrey H. Reed and Hugh F.
frequency of cell 3. Now there will be no collusion between MS1 VanLandingham "Handoff in Cellular Systems", IEEE Personal
and MS2. Communicatio, December, 2008
[2] Hongbo Zhou, Matt W. Mutka, "IP Address Handoff in the MANET",
IEEE INFOCOM 2004.
BTS allocate a
[3] Dênio Mariz, Dave Cavalcanti, Thienne Mesquita, Kelvin Lopes,
beacon frequency
Judith Kelner "A Simulation Environment for Analysis of QoS and
Signaling Load in Mobile Cellular Networks", 2006.
[4] Aman Kansal, "Handoff in Bluetooth Public Access Networks",
Technical Report,2001
MS1 MS1 immediately send the [5] "Introduction To Digital Cellular", Technical Report by Motorola
M-FCCH to BTS for new Cellular Infrastructure Group, issue 5, revision 5
frequency [6] Stijn N.P.Van Cauwenberge, "Study of soft handover in UMTS",
Master's Thesis, Technical University, Denmark. July 2003
MS2 using f1 [7] Rajkumar Periannan, Fadi Joseph Fahham, "Performance issues of
celluler networks", Journal of Imperial College, July 1996.
[8] Nasif Ekiz, Tara Salih, Sibel Küçüköner, Kemal Fidanboylu, "An
Overview of Handoff Techniques in Cellular Networks", PWASET
Fig 11: Working of M-FCCH Volume 6 June 2005 ISSN 1307-6884
Cell 3 [9] Majid Ghaderi, Joe Capka and Raouf Boutaba, "Prediction-Based
Admission Control for DiffServ Wireless Internet", IEEE 2003
[10] Theodore S. Rappaport, "Wireless Communication" Second Edition.
[11] Rajkumar Periannan & Fadi Josep Fahham, "Performance issues of
cellular network"
MS1(Beacon [12] [Online] David Chow,"Wireless Technology",
Freq.) MS2 (using f1) http://www.ylesstech.com Last visited April 01, 2008.
[13] [Online] "UMTS Handover",
http://www.umtsworld.com/technology/handover.htm Last visited
Frequency April 01, 2008
[14] [Online] "Power control of mobile station transmissions during handoff
in a cellular system" http://www.freepatentsonline.com/5940743.html
Last visited April 01, 2008
[15] [Online] US Patent 7227850, "Cellular radio communication system
with frequency reuse",
Time http://www.freepatentsonline.com/7227850.html Last visited April 01,
Fig. 12 Graph of frequencies in cell 3 when a temporary beacon frequency is 2008
allocated to MS1 by M-FCCH. [16] [Online] http://www.nortel.com Last visited April 01, 2008
[17] [Online] John Scourias, "Overview of the Global System for Mobile
After allocating a temporary beacon frequency to MS1 by Communications: GSM" Privateline Telecommunications Expertise.
M-FCCH, no collusion would be detected. New graph will be as Last Visited April 01, 2008.
follows. http://www.privateline.com/mt_gsmhistory/01_overview_of_the_glob
al_system_for_mobile_communications_gsm/
103
LOCATION DETERMINATION TECHNIQUE FOR LOCATING WIRELESS

DEVICES IN INDOOR ENVIRONMENT
Uzma Irshad 1, M.A Ansari 2
2 FUUAST, Islamabad, Pakistan
unity10us@gmail com drmaansari@fuuastib edu pk
Abstract — With emerging technology in mobile wireless It uses location-clustering techniques to reduce the computational
environment, detection of a mobile node becomes an essential requirements of the algorithm. [13]
requirement. Thus location awareness and mobile detection is
an important issue. This paper presents an RSSI (Received In location estimation system [14], LOCATOR works in two
Signal Strength Indicator) based modified triangulation and phases. In the first phase, it builds a radio map of the site where
modified clustering approach for locating a wireless device in the wireless network is deployed. The radio map is a mapping
indoor wireless environment. All existing systems are analyzed between the physical locations and the observed signal strength
in this paper and compare some of them. On the basis of these values at these locations from various base stations. In the second
comparisons proposed system is suggested. Proposed system is phase, LOCATOR uses the radio map to estimate the location of a
the modification of three systems, these systems are: user, given its current observed signal strength values from various
triangulation, clustering and joint clustering. After base stations. [14]
modification of three techniques, offline and online phase is
applied on them. Suggested proposed system is completed in In [15], a weighted center of mass based trilateration approach for
three systems: (i) modification of three techniques (ii) offline locating a wireless device based on the signal strength received from
phase (iii) online phase. Further offline phase is performed into the access points at known locations. The location tracking
four steps. The method which show high accuracy in offline algorithm runs on a central. The central computer gets site survey
phase, gives the best result for detection of a mobile node in information from the wireless device, in case of managed device, or
indoor environment. Then perform the online phase in order to gets the information from sensors, in case of unmanaged device.
get more accurate results. The proposed system can be applied [15]
to all current WLAN location determination systems to reduce
their computational cost and enhance their accuracy. RADAR [16], an RF-based system for locating and tracking users
inside buildings. RADAR uses signal strength information gathered
Index Terms — Positioning technologies, RSSI, Modified at multiple receiver locations to triangulate the user’s coordinates.
Triangulation, Modified Clustering, Modified Joint Clustering, Triangulation is done using both empirically-determined and
and Relative Position. theoretically computed signal strength information. [16].
The basic idea in this Bayesian system [17] is that if estimates from
I. INTRODUCTION
more than one observers of the location of a user are available, the
probability of correct estimation will be higher. The RSSI is used in
Many systems over the years have been developed for the detection location determination, by either establishing a physical relationship
of a mobile node in any wireless local area network. Some of them between the RSSI and the distance from the AP, or by using
tackle the problem of determining and tracking the targeted node in empirical methods.
WLAN. This paper used an RSSI (received signal strength
indicator) method to detect the mobile device in WALN. Since there The clustering technique defined in [18] reduces the computational
are many ways to categorize the positioning systems, in this paper power by more than an order of magnitude. Such energy saving
we will just refer to indoors positioning systems based on WLAN. allows the system to be implemented on energy-constrained mobile
In this paper we modify three methods that are: Triangulation devices and thus increases the scalability of the system in terms of
method used in system [20] Clustering method used in system [13] the number of supported users.
and Joint Clustering method used in system [18] Proposed system
has three modules: (i) Modify three techniques one by one (ii) The Active Badge System [19], an infrared-based system, was an
Offline phase (iii) Online phase. Offline phase is further performed early and significant contribution to the field of location-aware
into four steps: in first step, we take actual values. In second step, we systems. In this system, a badge worn by a person emits a unique
performed our three modified methods one by one and get three infrared signal every 10 seconds. Sensors placed at known positions
different values. In third step, we compare three obtained values to within a building pick up the unique identifiers and relay the
the actual values and in fourth step we take ration to find the location manager software
accuracy.
III. RSSI BASED TRIANGULATION METHOD
The remainder of this paper is organized as follows. In Section 2, we
survey related work in location determination technologies. Section
3, describe RSSI based Triangulation method. In Section 4, we Although many alternatives exist, most successful indoor location
compare existing systems showed in the table. In Section 5, we techniques are based on the RSSI triangulation method and
discuss our research methodology means, proposed system. clustering techniques. RSSI means Received Signal Strength
Finally, we present our Conclusions and future work in section 6. Indicator, is the simplest and most effective RF measurement
technique as its values are easily accessible with WLAN cards.
Since the RSSI values fluctuate over time for a given location, it
II. RELATED WORK
can be considered a random variable, and hence should be described
in a statistical fashion, estimating its distribution parameters.
The Horus system [13] lies in the probabilistic techniques category.
The Horus system identifies different causes for the wireless A. Variations in Measured RSSI
channel variations and addresses them to achieve its high accuracy. RSSI is not a very reliable measure, and it depends on a large
number of factors.
104
1) Location: The main variable behind the variation of the signal A. Introduction
strength received by the base stations is the location of the intruder. In suggested proposed system, we used three techniques, and
The localization phase is where the major computation is performed modified these techniques in order to improve the accuracy of
to match the received signal strength values to a position. detecting wireless devices in indoor environment.
2)Transmission power: A change in the transmission power of the The three techniques that we used in suggested proposed
user’s wireless NIC will cause a change in the received signal system are:
strength at the receiver. But, as the received strength is directly a) Modified Triangulation
proportional to the transmission power, a similar drop in RSSI will b) Modified Clustering
be noticed at all the base stations. c) Modified Joint Clustering
3) Directionality of the wireless NIC: A typical 802.11 card is not These techniques are obtained from system [20], system [13] and
omnidirectional in nature. Thus, with neither the transmitter nor the system [18] but these systems are costly, so we enhance these
receiver being omnidirectional in nature, we are bound to have systems in such a way that it provide high accuracy within short
variations in measured RSSI. time and less costly.
IV. COMPARISON OF THE EXISTING SYSTEMS For easy convenience, suggested proposed system is divided into
three modules (parts):
We compare some existing techniques in order to find the best one 1. Modifying the Techniques
and on the basis of the best techniques we suggested the proposed 2. Offline Phase
system to find more accurate results. 3. Online Phase
TABLE I We discuss these modules one by one to explore the suggested

COMPARING EXISTING SYSTEMS proposed system. The proposed modules are all applicable to any of
the current WLAN location determination system.
Techniques Accuracy Error Phase RF B. Modifying The Techniques
level rate s Technique
X
Now we discuss how modifications occur in these systems.
1) Modified Triangulation Method
Location 90% 2 We modify the triangulation method that was used in system [20] in
Estimation which three servers are used to detect the location of a mobile node,
System which is a costly method, but we modify it by using only one server
X
and three clients that detect the location of a mobile device (target
Horus 90% 0.6m 2 device). Mobile device is placed at any location within clients.
System
In order to detect the location of a mobile device it is necessary to
Trilatration ±5m 2 NO get the RSSI (Received Signal Strength Indicator) values in the form
Approach of dbm from the mobile node to the clients. Multiple readings of the
X
signal strength of a mobile node is taken at the clients and then these
Radar: 89% 3..5m 2 signal strengths are converted into an average signal strength by
Location using simple averaging formula [20].
And
Tracking RSSI = (RSSI1+RSSI2+RSSI3+……. +RSSIn)) /n
System Where
RSSI = Average signal strength value (dbm)
X
RSSI1 = Signal strength at first interval
Kalman 90% 2..5m 2 RSSI2 = Signal strength at second interval
Filter And RSSIn = Signal strength at nth interval
Multiple n = No. of intervals
Observer
The average calculated RSSI from each client is sent to the server.
X
At the server, calculations is performed by using distance formula
Clustering 90% 7 Feet 2 which is proposed by [20] to calculate the distance of each client
And from the mobile node in the area covered by the Wireless Local
Probability Area Network (WLAN).
Distributions
The distance of three clients from the server and from each other is
X
known, and we have to find the distance of clients from the mobile
Bayesian 80% 2..5m 2 device according to the received signal. For this distance
Method calculation, we used the formula that was proposed in system [20].
S = (RS-BS) * k
(Ignoring the negative sign of the signal strength)
V. PROPOSED SYSTEM Where RS = Received Signal Strength
BS = Basic Signal Strength (which in the hardware used)
Problem statement of suggested proposed system is that:
k = distance covered in 1dbm.
“Detection of a mobile node by using three modified techniques one
by one and then apply offline and online phase on these modified
Now we used Trigonometric functions to calculate the angle of
techniques”
clients with the server [20].
Cos α = (b^2+c^2-a^2) / 2bc.
105
The calculated distance and the angles are stored in the server where 1) Take Actual values AV. (X, Y)
all calculations are performed and the results obtained through the We have to find the actual location of a mobile node manually. We
calculations show the actual location of a targeted device. Our take X, Y coordinates of a target device in a wireless local area
modified system provides more accurate results in short time and network which may be a room, a floor or any indoor environment
also it is less costly. and calculated its distances from the user manually .We take values
2) Modified Clustering Method in X,Y coordinates. Let assume that the values we obtained in X, Y
We modify the system [13] in which just clustering method is used. coordinates of a mobile node is (28, 20) of a mobile node. This is the
In clustering method of system [13], the locations are grouped into actual values that indicate the exact location of a mobile node in
clusters according to the access points that cover the locations. The wlan.
cluster is divided into patches; from each patch RSSI value is (X, Y) = (28, 20)
received and sampled the signal strength from each access point at
the rate of one sample per second. I this method, two steps are This actual value is stored in the database and we have to compare
performed [13]: other values to this actual value (X, Y).
i) At a fixed location, the signal strength received from an access 2) Obtain Values By Three Techniques
point varies with time. In this offline phase we use three techniques one by one and get
ii) The number of access points covering a location varies with values according to their methods.
time. “a” shows “ Modified Triangulation method”
“b” shows “Modified Clustering method”
But we modify system [13] by using clustering and triangulation at “c” shows “Modified Joint Clustering method”
the same time. We perform both techniques one by one and the
results obtained by these two methods are put into average. So the a. Modified Triangulation Method:
final result gives more accurate results to detect the location of a We used modified triangulation method as we discussed above. We
target mobile. manually calculated the coordinates of that client which has the
shortest distance from the mobile node. The distance of clients from
The clustering method of our proposed system use the same the mobile node is calculated on the basis of signal strength stored in
mathematical model as discussed in system [13] which states that: the server.
without loss of generality, let X be a 2 dimensional physical space. Let assume that the coordinates we obtained in this
At each location x € X, we can get the signal strength from k access triangulation method (xa, ya) is (20, 15).
points. We denote the k- dimensional signal strength space as S. We store the coordinates (xa, ya) (20, 15) in the database.
Each element in this space is a k-dimensional vector whose entries a = (20, 15)
represent the signal strength readings from different access points. “a” shows that coordinates (xa, ya) obtained by applying
We denote samples from the signal strength space S as s. We also Triangulation method.
assume that the samples from different access points are b. Modified Clustering Method
independent. The problem becomes, given a signal strength vector s In this method we take an average of two results that is obtained by
= (s1… sk), we want to find the location x € X that maximizes the using two techniques that are:
probability P(x/s) [13]. i) Triangulation
ii) Clustering
By applying above mathematical model in our proposed system, we By applying the mathematical model as discussed in system
can estimate the location of a mobile device. [13], we can estimate the location of a mobile device in clusters. Let
3) Modified Joint Clustering Method assume that the value of coordinates xb ,yb of a user that is close to
System [18] presents an accurate and scalable system for the mobile is (22, 17). We store the coordinates (xb ,yb) (22, 17) in
determining the user location with low computational requirements database.
in an 802.11 wireless LAN (WLAN) framework. But this approach b= (22, 17)
can only be used with small values of k (access points), which may “b”shows that coordinates (xb ,yb) obtained by using modified
affect the technique accuracy. clustering method.
c. Modified Joint Clustering Method
So the problem in this approach is that it requires a large training set In this method, we use calculations as discussed in system [18]. Let
to obtain good estimate of the joint distribution and the required size assume that the value of coordinates (xc ,yc) of a user that is close to
increases exponentially with k (access points). the mobile node is (25,18). We store the coordinated values (xc ,yc)
(25,18) in the database in order to match it to the actual values in the
To overcome this problem, we suggested to use large no. of k database.
(access points). Number of readings of training set is used and then c= (25, 18).
take the average of multiple readings to get more accurate results. “c” shows that coordinates (xc ,yc) obtained by using Joint
Second suggestion is that we limit k (access points) i.e.:- k=2 or Clustering method
k=3, and then take the readings on this 3) Comparison of Obtained Values with Actual Value
limitations. In this step of offline phase we compare the calculated values a, b, c
(obtained by three techniques) to the actual value. We take the
C. Offline Phase difference of the calculated dimensions to the actual dimension.
Offline phase is also known as training phase in which the
coordinates of location in Wlan and the distances of clients from Table shows the difference of the calculated values to the actual
each other and from the server is stored in database. On the basis of values, we can find the error rate of these techniques to determine
stored information we detect the location of a mobile device. the best one that provide more accurate results to reach the mobile
We perform offline phase in four steps: node in wlan. Error rate can be found by adding the x,y dimensions.
̇ Take Actual values AV. (X, Y)

̇
The values that we taken in table are on assumption based. We have
Obtain Values By Three Techniques
̇
taken these values from three different techniques on the basis of
Compare Obtained Values with Actual value
̇
their previous performance.
Take Ratios
106
4) Take Ratios
ERROR RATE
By taking the ratio of three techniques, we get the following
accuracy: 15
A= 52%
Readings
10
B= 67% readings
C= 82% 5
Where A, B, C shows the accuracy of three techniques (a,b,c) as 0
shown in the table: x,y xaya xb,yb xc,yc
Techniques
TABLE II
COMPARISON OF OBTAINED VALUES WITH ACTUAL
VALUE.
Accuracy
SHOW ACCURACY
TECHNIQUES READINGS DIFFERENCE 120
AV VS. 100
80
60 Accuracy
Actual value (28,20) 0
40
AV 20
(X, Y) 0
X,Y xa, ya xb ,yb xc,y
Triangulation (20,15) (8,5)
Techniques
a
(xa, ya)
Clustering b (22,17) (6,3) D. Online Phase

(xb ,yb) Online phase is also known as runtime phase in which the real time
values of signal strength of the mobile node is noted by applying
Joint (25,18) (3,2) three techniques one by one and then multiply the real time values
Clustering c of each technique to the corresponding accuracy stored in the
(xc , yc) database(offline values) .
In this phase we get a, b, c, by applying Modified Triangulation,

TABLE III Modified Clustering and Modified Joint Clustering and Probability
ALCULATED ERROR RATE AND ACCURACY Distribution methods as discussed above in the same way as we do
in the offline phase but in this phase we get online values. Real time
TECHNIQUES ERROR ACCURACY values are taken into calculations. To find final Xf,Yf values of our
RATE system, following formula is used to estimate the location of a
mobile node in Wlan.
Actual Value Av
Xf,Yf = (a * A) + (b * B) + (c * C)/3
(X, Y) 0m 100% Where
“a” show the value that we obtained by applying the Modified
Triangulation A 13m 52% Triangulation Method in online phase.
(Xa, Ya) “A” is the accuracy of modified joint clustering method that was
calculated in offline phase.
Clustering B 9m 67% “b” show the value that we obtained by applying the Modified
(Xb ,Yb ) Clustering Method in online phase.
“B” show the accuracy of modified clustering method that was
Joint Clustering C 5m 82% calculated in offline phase.
(Xc , Yc) “c” show the value that we obtained by applying the Modified Joint
Clustering and Probability Distribution Method in online phase.
The values and accuracy that we obtained, is considered for only “C” is the accuracy of modified triangulation method that was
one room in indoor environment, although if same techniques are calculated in offline phase.
applied to other room in wlan environment then it provide different Xf,Yf give the final values that is used to estimate the exact location
results and accuracy, so that’s why we assign these values to the of a mobile node.
techniques on their previous results. We just proposed a method
to improve the accuracy of the techniques. VI. CONCLUSION
But it is shown that Joint Clustering method provide more accurate Proposed “Location Determination Technique for Locating
Wireless Devices in Indoor Environment” is easily implemented on
results to detect the location of a mobile node in any indoor
environment. any existing WLAN environment and offers as such additional
value to pre - existing infrastructural networks. By using the
modified triangulation, modified clustering and modified joint
We show error rate and accuracy of three techniques in the form of
graphs. clustering method, the accuracy is high concerning the reliability of
the measurement vectors.. We get accuracy of 82% in modified
Joint Clustering method with 5m error rate, 67% in modified
Clustering method with 9m error rate and 52% in modified
triangulation method with 13m error rate. So it is concluded that
joint Clustering method give more accurate results with minimum
error rate for detecting wireless devices in indoor environment. A
107
mean absolute error of 5.0 meters in an indoor environment is [7] “Analysis Of The Optimal Strategy For Wlan Location Determination
enough to pinpoint the corridor or room, where the target is located. Systems”, by Moustafa A. Youssef¤, Ashok Agrawala Department of
These modified methods are used to reduce the computational Computer Science University of Maryland College Park, Maryland
20742 fmoustafa, agrawalag@cs.umd.edu.
requirements of the location determination algorithms and achieve
[8] “Flavour - Friendly Location-aware Conference Aid with Privacy
scalability. Observant Architecture” Kavitha Muthukrishnan, Nirvana Meratnia and
Maria Lijding Faculty of Computer Science University of Twente.
The values we get is on assumption based by analyzing their [9] “Area Localization using WLAN” by Vinod Patmanathan, Master of
previous results, however we can evaluate the performance of the Science Thesis Stockholm, Sweden 2006:008.
proposed system by experimental testbed in WLAN. [10] Moustafa Youssef, Ashok Agrawala, and Udaya Shankar: “The Horus
WLAN Location Determination System”.
As a continuation of the work, I would like to improve the distance [11] Ankur Agiwal, Parakram Khandpur, Huzur Saran: “LOCATOR –
Location Estimation System for WirelessLANs.”
estimation phase of the proposed system. Modified Joint Clustering
[12] Navin Kumar Sharma A Weighted Center of Mass Based Trilateration
method presents an accurate and scalable system for determining the Approach for Locating Wireless Devices in Indoor Environment.
user location with low computational requirements in an 802.11 [13] “RADAR: An In-Building RF-based User Location and Tracking
wireless LAN framework. System” by Paramvir Bahl and Venkata N. Padmanabhan.
[14] Raman Kumar K, Varsha Apte, Yogesh A Power Improving the
REFERENCES Accuracy of Wireless LAN based Location Determination Systems
using Kalman Filter and Multiple Observers.
[15] Mustafa A. Youssef, Ashok Agrawala, A. Udaya Wlan Location
[1] Nirupama Bulusu. Self-Configuring Location Systems. PhD thesis, Determination Clustering And Probability ‘Distributions. In
University of California, Los Angeles, 2002. Proceedings of IEEE Per Com 2003 march- 2003
[2] LYU0401 Location-Based Multimedia Mobile Service. 2003/2004 Final [16] Roy Want, Andy Hopper, Veronica Falcao and Jonathan Gibbons. The
Year Project Second Term. Supervisor Professor Michael R. Lyu Fung Active Badge Location System.
Ka Kin Ma Tilen. [17] Ansar-Ul-Haque Yasar, Dr. M.A. Ansari, Sherjeel Farooqui. Low Cost
[3] “A method for implementing Mobile Station Location in GSM” by Svein Solution for Location Determination of Mobile Nodes in a Wireless
Yngvar Willassen December 3, 1998. Local Area Network.
[4] Horus: A Wlan-Based Indoor Location Determination System by [18] “Localization For Intrusion Detection In Wireless Local Area
Mustafa Amin Abdel Azim Yousief Abdel Rehim, Doctor of Networks” by Vishal Bhargava a thesis submitted to the graduate
Philosophy, 2004. faculty of north carolina state university in partial fulfillment of the
[5] Kavitha Muthukrishnan, Nirvana Meratnia and Maria Lijding requirements for the degree of master of science department of electrical
FLAVOUR - Friendly Location-aware Conference Aid with Privacy and computer engineering Raleigh Aug 2003.
Observant Architecture. [19] “Design Of Indoor Positioning Systems Based On Location
[6] “Location Determination in a Wireless LAN Infrastructure”, Master’s Fingerprinting Technique” by Kamol Kaemarungsi B. Eng., King
Thesis by Jason Small epartment of Electrical and Computer Mongkut's Institute of Technology at Ladkrabang, Thailand, 1994. M. S.
Engineering Carnegie Mellon University. in Telecommunications, University of Colorado at Boulder.
[20] “Devices Location in 802.11 Infrastructure Networks using
Triangulation” by David Sánchez, Sergio Afonso, Elsa M. Macías,
Member IAENG, and Ávaro Suárez, Member IAENG.
108
RSSI Based Triangulation Method And Clustering Technique For Locating Wireless
Devices In Indoor Environment
Uzma Irshad 1, M.A Ansari 2
2 FUUAST, Islamabad, Pakistan
unity10us@gmail.com, drmaansari@fuuastib.edu.pk
Abstract — With emerging technology in mobile wireless In location estimation system [7], LOCATOR works in two phases.
environment, detection of a mobile node becomes an essential In the first phase, it builds a radio map of the site where the wireless
requirement. Thus location awareness and mobile detection is network is deployed. The radio map is a mapping between the
an important issue. This paper presents an RSSI based physical locations and the observed signal strength values at these
triangulation approach for locating a wireless device from the locations from various base stations. In the second phase,
access points at known locations as finding out the location in a LOCATOR uses the radio map to estimate the location of a user,
wireless environment is to-date not very easy, cheap and given its current observed signal strength values from various base
accurate using 802.11b technology. All existing systems are stations. [7].
analyzed in this paper and compare some of them. On the basis
of these comparisons proposed system is suggested. This paper In [8], a weighted center of mass based trilateration approach for
suggest the clustering method to improve the accuracy of locating a wireless device based on the signal strength received from
proposed system. The proposed system provides a solution for the access points at known locations. The location tracking
location detection of mobile devices in indoor environment algorithm runs on a central computer (a system used by network
where the configuration of access points like transmit power administrator for monitoring the wireless environment). The central
etc., is not fixed. Proposed system presented in this paper is the computer gets site survey information from the wireless device, in
enhancement of system defined in [13]. In system [13] three case of managed device, or gets the information from sensors, in
servers and only one client is used. Three servers are used to case of unmanaged device. [8]
detect the location of a mobile by using triangulation and
calibration method so it is a costly method because with the help RADAR [9], an RF-based system for locating and tracking users
of three servers we can detect the location of mobile, but inside buildings. RADAR uses signal strength information gathered
proposed system presented in this paper used only one server at multiple receiver locations to triangulate the user’s coordinates.
and three clients. Only one server has to detect the location of a Triangulation is done using both empirically-determined and
mobile, thus proposed system is cheap and easy to implement. theoretically computed signal strength information. Experimental
Two methods are used and results obtained from these two results are quite encouraging. With high probability, RADAR is
methods are taken into an average, which provide more able to estimate a user’s location to within a few meters of his/her
accurate results. actual location [9]
Index Terms — Positioning technologies, RSSI, Clustering, The basic idea in this Bayesian system [10] is that if estimates from
Triangulation, Relative Position. more than one observers of the location of a user are available, the
probability of correct estimation will be higher. The RSSI is used in
I. INTRODUCTION location determination, by either establishing a physical relationship
between the RSSI and the distance from the AP, or by using
empirical methods.
Advancement in mobile world is catapulted by ncorporating the
mobile users position information. The position of a hand held The clustering technique defined in [11] reduces the computational
device is important information needed to enhance the power by more than an order of magnitude. Such energy saving
communication efficiently. Basically, the positioning has been allows the system to be implemented on energy-constrained mobile
indigenously put apart into two parts, Indoors and Outdoors Since devices and thus increases the scalability of the system in terms of
there are many ways to categorize the positioning systems, in this the number of supported users. Results also showed that locations
paper we will just refer to indoors positioning systems based on clustering increases the accuracy of the location determination
WLAN.The method of triangulation is altered in our research paper. system and help scales the system to larger coverage area. .
We formed triangles of clients with server and then calculated the
distance and angles using the formulas and second method, The Active Badge System [12], an infrared-based system, was an
clustering is used to get the result, after that we take the average of early and significant contribution to the field of location-aware
the results to get the final results .The remainder of this paper is systems. In this system, a badge worn by a person emits a unique
organized as follows. In Section 2, we survey related work in infrared signal every 10 seconds. Sensors placed at known positions
location determination technologies. In Section 3, we compare within a building pick up the unique identifiers and relay the
existing systems showed in the table. Section 4, describe RSSI location manager software. While this system provides accurate
based Triangulation method, in section 5, we discuss our research location information, it suffers from several drawbacks.
methodology means, proposed system. . Finally, we present our
conclusions in section 6 and future work discuses in section 7. III. COMPARISON OF EXISTING SYSTEMS
II. RELATED WORK
We compare some existing techniques in order to find the best one
and on the basis of the best techniques we suggested the proposed
The Horus system [6] lies in the probabilistic techniques category. system to find more accurate results.
The Horus system identifies different causes for the wireless
channel variations and addresses them to achieve its high accuracy.
It uses location-clustering techniques to reduce the computational
requirements of the algorithm. The Horus system uses the
Perturbation technique for handling small-scale variations. [6]
109
Techniques Accuracy Error Phas RF-Tec The fact, that RSSI is measured on logarithmic scale aides in
level rate es hnique observing relative changes in signal strength.
Location 90% 2 X 3) Directionality of the wireless NIC:
Estimation A typical 802.11 card is not omnidirectional in nature. Thus, with
System neither the transmitter nor the receiver being omnidirectional in
nature, we are bound to have variations in measured RSSI. A
Horus 90% 0.6m 2 X variation in the directionality of the wireless NICs, compared to the
System NIC used for fingerprinting could change the perceived position of
the user.
Trilatration ±5m 2 NO
Approach V. PROPOSED SYSTEM
Radar: 89% 3..5m 2 X

Problem statement of our proposed system is that:
Location
Detect location of a mobile node using clustering and triangulation
And
method.
Tracking
System
We have to find out a mobile device by calculating the distance of
each client from the server by applying a method of RSSI based
X
Triangulation and Clustering then take the average of RSSI values
Kalman 90% 2..5m 2
to find the accurate location of a mobile node.
Filter And
Multiple A. Description
Observer There are three clients and one server in the room. Clients are
connected to the server. Server has to perform the calculations on
X
the results. Signal strengths of all clients are stored in the server and
Clustering 90% 7 Feet 2 the location of the client is determined with the help of these signal
And strengths. The server is the basic requirement of our system to
Probability obtain RSSI value of the mobile node. We have to use two methods
Distribution we get RSSI value of each client and then take the average of these
s two values to find out the mobile node.
B. Clustering Approach
Bayesian 80% 2..5m 2 X The Clustering module is used to group radio map locations based
Method on the access points covering them.
We define a cluster as a set of locations sharing a common set of
access points. We call this common set of access points the cluster
IV. RSSI BASED TRIANGULATION METHOD key. Clustering is based on the idea that each access point defines a
subset of the radio map locations that are covered by this access
point. These locations can be viewed as a cluster of locations whose
Although many alternatives exist, most successful indoor location key is the access point covering the locations in this cluster.
techniques are based on the RSSI triangulation method and
lustering techniques. C. Triangulation Approach
1. Get the RSSI value of each client (mobile nodes) from the server
RSSI means Received Signal Strength Indicator, is the simplest and which will be in the form of dbm. Since 802.11b is highly effective
most effective RF measurement technique as its values are easily by multipath propagation, where several waves arrive at the receiver
accessible with WLAN cards. RSSI was chosen over Signal to via different paths and with different phases, different variations of
Noise ratio (SNR) as the radio signature because it is more location actual signal.
dependant. Noise can vary considerably from location to location Server get RSSI value from three clients and then converted them in
and depending on external factors, but this is not the case for the to average signal strength by using simple averaging formula
received signal strength. Since the RSSI values fluctuate over time described below:
for a given location, it can be considered a random variable, and RSSI = (RSSI1+RSSI2+RSSI3+… +RSSIn)) /n
hence should be described in a statistical fashion, estimating its Where
distribution parameters. RSSI = Average signal strength value (dbm)
A. Variations in Measured RSSI RSSI1 = Signal strength at first interval
RSSI2 = Signal strength at second interval
RSSI is not a very reliable measure, and it depends on a large RSSI3 = Signal strength at nth interval
number of factors. n = No. of intervals.
1) Location: 2. Once the signal strength of the client is obtained with respect to
The main variable behind the variation of the signal strength each client, we calculated distance to determine the approximate
received by the base stations is the location of the intruder. The location of the client (mobile node) in the area covered by the
localization phase is where the major computation is performed to WLAN.
match the received signal strength values to a position. There are The method of triangulation is altered in our research project. We
various algorithms that exist to locate the position of intruder. formed triangles of clients with server and then calculated the
2) Transmission power: distance and angles using the formulas mentioned above rather then
A change in the transmission power of the user’s wireless NIC will finding the triangulation region.
cause a change in the received signal strength at the receiver. But, as
the received strength is directly proportional to the transmission D. Location Calculation
power, a similar drop in RSSI will be noticed at all the base stations. The calculation of the location of a user takes place in two phases:
an offline and an online phase. Offline Phase: During the offline
110
phase, which has to be executed only once for each building, a VII. FUTURE WORK
so-called radiomap will be composed. This radiomap can be
The design of indoor positioning systems requires an analytical
considered to be a collection of calibration points at different
model that can explain the relationship between the system
locations in the building, each with a list of RSSI values for visible
performance and system parameters. The adoption of
access points at that particular location. This process is also known
location-based services in the future will require an efficient and
as fingerprinting.
effective design methodology for indoor positioning systems. The
design area of indoor positioning has not been considered at all in
any literature. There is no simple way to predict the performance of
Server indoor positioning without actually performing a real measurement
to test the positioning algorithm. Therefore, this study proposes a
model of triangulation method which can be incorporated into an
analytical model for study of indoor positioning systems
The RSSI based position estimation still has limitations. First, the
Client RSSI value is time dependent. Thus, the future study should address
Clients this issue. Second, the movement of object is not considered. Third,
Client orientations of sensors affect the performance which should be
taken care off. Fourth, to improve the performance of the system we
need to find a solution to deal with those incomplete or missing data
Fig Architecture of Triangulation
during the offline and the online phase. Fifth, position estimation on
Online Phase: During the online phase, the calibration points are multifloor and three dimensional coordinates is not available.
being used to calculate the most probable location of the user, whose
actual location is unknown. REFERENCES
E. Average
[1] U.S.Robotics Wireless LAN Networking White Paper.
We take the average of results that we obtained through clustering [2] Alcatel Technology White paper “End-to-End Location-Based
method and triangulation method in order to estimate the location of Services Solution”, 2Q 2005,
mobile node. N. Faggion, S.Leroy, C. Bazin.
[3] Ingersoll, Minnie (2005-11-17).Wi- Fi in Mountain View. Official
F. Advantages of the Proposed System Google Blog.Google.
Our proposed system is the enhancement of system defined in [13], [4] Kavitha Muthukrishnan, Nirvana Meratnia and Maria Lijding
only one server is used to estimate the location of mobile node FLAVOUR – Friendly Location-aware Conference Aid with rivacy
Observant Architecture.
instead of three servers. One server is connected to three clients to
[5] Dr. WJ Buchanan, Dr. J. Munoz, R. Manson, K. Raja. Analysis and
detect the location of mobile node. Also clustering technique is used Migration of Location-Finding Methods for GSM and 3G Networks.
in this method, which provide more accurate results. Fifth IEEE International Conference on 3G Mobile Communication
A major advantage of our proposed technique is that we can obtain Technologies (3G 2004) The Premier Technical Conference for 3G and
higher accuracy while reducing the training cost by requiring only a Beyond, , London, UK, 18-20 Oct. 2004.
fraction of the labeled samples as compared to previous methods. [6] Moustafa Youssef, Ashok Agrawala and Udaya Shankar. The Horus
We demonstrate this result in a series of tests on WLAN location WLANlocation Determination System.
estimation using the data collected in a realistic environment. The [7] Ankur Agiwal, Parakram,Khandpur, Huzur Sara
LOCATOR–Location Estimation System For WirelessLANs..
clustering technique reduces the computational power by more than
[8] Navin Kumar Sharma A Weighted Center of Mass Based Trilateration
an order of magnitude. Such energy saving allows the system to be Approach for Locating Wireless Devices in Indoor Environment.
implemented on energy-constrained mobile devices and thus [9] RADAR: An In-Building RF-based User Location and Tracking
increases the scalability of the system in terms of the number of System Paramvir Bahl and Venkata N. Padmanabhan.
supported users. We also showed that locations clustering increases [10] Raman Kumar K, Varsha Apte, Yogesh A Power Improving the
the accuracy of the location determination system and help scales Accuracy of Wireless LAN based Location Determination Systems
the system to larger coverage area. using Kalman Filter and Multiple.
[11] Mustafa A. Youssef, Ashok Agrawala, A. Udaya Wlan Location
Determination Via Clustering And Probability ‘ Distributions. In
VI. CONCLUSION Proceedings of IEEE PerCom 2003 March. 2003.
[12] Roy Want, Andy Hopper, Veronica Falcao and Jonathan Gibbons.
The Active Badge Location System.
The RSSI- based tracking system is easily implemented on any
[13] Ansar-Ul-Haque Yasar, Dr. M.A. Ansari, Sherjeel Farooqui. Low
existing WLAN environment and offers as such additional value to Cost Solution for Location Determination of Mobile Nodes in a
pre - existing infrastructural networks. By using the described Wireless Local Area Network.
triangulation method, the accuracy is high concerning the reliability [14] Vinod Patmanathan. Area Localization using WLAN.
of the measurement vectors. A mean absolute error of 1.5 meter s in
an indoor environment is enough to pinpoint the corridor or room,
where the target is located. The triangulation method is required for
successful location estimation in a new environment. The value of
the block method is emphasized in heterogeneous environments
even though benefits from its usage can be observed in any
environment. As a continuation of the work, I would like to improve
the distance estimation phase of the proposed system.
111
Noise Cancellation in WLAN and Bluetooth

Sohail Sarwar 1, Dr. M. A. Ansari 2
1 Department of Computer Science, SZABIST
2 Federal Urdu University Islamabad, Pakistan.
ssohailsarwar@yahoo com drmaansari@fuuastib edu pk
Abstract — Wireless communication is an ever-developing receivers are able to detect this loss by examining the signal to
field and a plethora of innovations is envisioned in the future. It interference ratio (SIR), and tally the lost packets into a bit error rate
is anticipated that in this field devices will be developed to (BER) [6]. The last packets are retransmitted later. As the number of
support communications with higher quality, and high data retransmissions increases on the other hand throughput decreases.
rates. Since WLAN and Bluetooth operate in the same
unlicensed ISM band (2.4 GHz), they often cause mutual This paper is organized as follows: the second section gives a
interference and hence degrade performance. A number of system overview of WLAN and Bluetooth. Third section presents
collaborative and non-collaborative mechanisms have been an interference model. In fourth section related work is described,
proposed to overcome interference problems. In this paper a and fifth section presents proposed model, conclusion is presented
new method is proposed for noise and interference cancellation in section fifth.
between WLAN and Bluetooth to achieve high-quality voice
and data communication. This new concept is based on a II. SYSTEM OVERVIEW
common control channel (CCCH), which is used by different
radio devices for synchronization. A. IEEE 802.11 WLAN
WLAN operates in the 100 meter range using 2.4 GHz ISM
Index Terms — 802.11, Noise, Interference1, WLAN, unlicensed band. The IEEE 802.11b standard provides a maximum
Bluetooth. bit rate of 11 Mbps through a transmission mode based on 8-chip
Complementary Code Keying (CCK) modulation at a chip rate of 11
I. INTRODUCTION Mchip/s. Since IEEE 802.11b uses Direct Sequence Spread
Spectrum (DSSS) technique, it occupying a dedicated bandwidth of
approximately 22 MHz within the ISM band [14]. In the recent past
Interference creates noise in the communication signals which years the 802.11g protocol has been developed to be backward
destroys the signal completely or signal becomes meaningless. compatible with 802.11b and has a capacity of 54Mbps using
Radio frequency interference is a major concern in the deployment OFDM modulation. All the versions of the IEEE 802.11x share the
and use of wireless LANs, and is often a justification for avoiding same MAC sub-layer, which uses the Carrier Sense Multiple Access
their installation. WLAN operates in the Industrial, Scientific, and and Collision Avoidance (CSMA/CA) for contention. The
Medical (ISM) band [1], spectrum reserved by regulators worldwide maximum transmitted power is recommended at 100mW. IEEE
for applications without the requirement for individual user or 802.11 protocol entities are shown in the Figure 1 [14].
device licensing. A consequential challenge in using these
frequencies is that a potentially large number of wireless devices
may be competing for the same media in a particular location. These
devices often resulting interference with WLAN, and thus degrade
performance in terms of throughput, connection quality, and range.
Interference occurs when two signals are transmitted on the same
frequency at the same time. When two (or more) simultaneous
signals have similar relative transmitting power, in which case they
will likely mutually interfere, and the weaker signal will suffer more
interference from the stronger. With respect to WLANs, it can be
interfere from other WLAN devices. Since WLAN employ a Fig. 1 Protocol entities for IEEE 802.11 [14]
“listen-before-talk” protocol, based on Carrier Sense Multiple
Access with Collision Avoidance (CSMA/CA) [1], any interference The general operations of physical layer are carrier sense,
between WLAN networks tends to work out somewhat transmission and receiving of data frames, encoding/decoding, and
cooperatively, with the two networks often sharing channel preamble generation/removal. When the MAC layer instructs, the
capacity. In contrast, interference from non-WLAN sources, which PLCP prepares MAC protocol data units (MPDUs) for transmission.
use protocols different from those of WLANs, more often result in The PLCP also delivers incoming frames from the wireless medium
the degradation of WLAN throughput. There are numerous to the MAC layer. Under the direction of the PLCP, the PMD
non-WLAN devices that operate in the unlicensed bands, including provides actual transmission and reception of Physical layer entities
Bluetooth products of many forms (some operating at the same between two stations via the wireless medium. The MAC
power levels as WLANs), cordless phones, baby monitoring management sublayer defines the registration process, roaming
devices, wireless security and energy managements systems, support in the ESS, power management, and security. Both the
computer peripherals such as cordless mice, keyboards, and game MAC and PHY layers cooperate in order to implement collision
controllers etc. In addition, there are some commercial and avoidance procedures.
industrial devices such as microwave ovens, and microwave-based
lights. B. Bluetooth
The following section gives an overview of the Bluetooth
Here the main focus of this paper is on WLAN IEEE 802.11b and technology. Bluetooth is a low-power microwave wireless link
Bluetooth interference. As WLAN and Bluetooth devices share technology designed to connect phones, laptops, PDAs and other
common spectrum in the 2.4GHz ISM band. When WLAN and portable equipment. Unlike infra-red, Bluetooth does not require
Bluetooth operate in the same area they often cause interference due line-of-sight positioning of connected units. Bluetooth systems
to which transmitted packet is lost. The WLAN and Bluetooth operate in the unlicensed Industrial-Scientific-Medical (ISM) radio
band at 2.4 GHz. Low-power RF transmission provides
1
communication between devices over a range of 10 meters and
Noise and Interference are used interchangeably
112
supports data speeds of up to 721 Kbps [12], [13], as well as three responsible for creating, managing, and destroying L2CAP channels
voice channels. Bluetooth enables ad-hoc networking for up to eight for the transport of service protocols and application data streams.
devices without the need for a formal wireless infrastructure.
Bluetooth uses frequency hopping spread spectrum (FHSS) [13] III. INTERFERENCE MODEL
operation. The basic frequency-hopping pattern is a pseudo-random
ordering of 79 channel frequencies in the ISM band. The hopping
rate is 1600 hops per second [12], [13]. Bluetooth signal Interference is the major limiting factor in the performance of
transmitting power is 1 mW, and signal is modulated using binary wireless systems. WLAN and Bluetooth operate in the same ISM
Gaussian Frequency Shift Keying (GFSK) [13]. Bluetooth band [12], often caused interference; hence the performance of both
technology provides the effect of full duplex transmission through the systems degrades. When WLAN and Bluetooth operates in the
the use of a time-division duplex (TDD) scheme with 625 µs time same area, the 22MHz wide WLAN channel occupies the same
slots [13]. Two or more devices sharing the same physical channel frequency as 22 of the 79 Bluetooth channels which is 1 MHz wide
form an ad-hoc network called piconet [12]. With one device acting [1], [12]. When a Bluetooth transmission occurs on a frequency that
as a master, up to seven other devices or slaves can be actively lies within the frequency space occupied by WLAN transmission,
operating in the piconet. All devices in the piconet are synchronized some level of interference can occur depending on the strength of
to a common clock reference and frequency hop pattern provided by each signal. A packet is lost due to noise when the WLAN and
the master. A slave packet always follows a master packet Bluetooth packet overlap in the time and frequency domains. This
transmission. interference problem is address in [1], [3], [4], [6], and [9]. Since
WLAN uses DSSS occupies 22MHz channel in ISM band and
Two types of link connections that can be established between a Bluetooth system with FHSS uses 79MHz of available frequency
master and a slave [12], [14]: the Synchronous Connection-Oriented with 1MHz channels. As a result probability that a Bluetooth packet
(SCO), and the Asynchronous Connection-Less (ACL) link. The collides with WLAN packet is 27.8%. Throughput of WLAN drops
SCO link is a symmetric point-to-point connection between a master rapidly at high Bluetooth network load. If a packet collision has
and a slave where the master sends an SCO packet in one TX slot at occurred, the WLAN packet is dropped and retransmitted again. As
regular time intervals, defined by TSCO time slots. The slave the number of retransmissions increases the performance decreases.
responds with an SCO packet in the next TX opportunity. TSCO is Interference from Bluetooth increases BER in WLAN and it
set to 2, 4 or 6 time slots for HV1, HV2, or HV3 packet formats [13] decreases the throughput of WLAN. The performance of Bluetooth
respectively. All three formats of SCO packets are defined to carry also effected. IEEE 802.11g uses OFDM with data rate up to
64 Kbits/s of voice traffic and are never retransmitted in case of 54Mbps [6], and IEEE802.11n applies spatial coding techniques,
packet loss or error [13]. The ACL link is an asymmetric which uses multiple transmitter and receivers antennas [6]. The
point-to-point connection between a master and active slaves in the simulation results in [6] shows that even the throughput of these
piconet. Several packet formats are defined for ACL, namely DM1, systems are very high in the absence of interference, a small amount
DM2, and DM3 packets that occupy 1, 3, and 5 time slots [13] of interference still cause substantial performance degradation of
respectively. An Automatic Repeat Request (ARQ) procedure is these systems.
applied to ACL packets where packets are retransmitted in case of
loss until a positive acknowledgement (ACK) is received at the IV. RELATED WORK
source.
In contrast to classical interference and noise avoidance techniques
Bluetooth Protocol stack is shown in the Figure 2 [13]. The RF
such as modulation, channel coding, most of the techniques to solve
block is responsible for transmitting and receiving packets of
noise/interference problem in 2.4 GHz band focus on non-signal
information on the physical channel. The link controller is
processing control strategies including power and frequency
responsible for the encoding and decoding of Bluetooth packets.
hopping control and MAC parameter adjustments and scheduling
The baseband resource manager is responsible for all access to the
[7], [8]. A number of mechanisms have been proposed to ensure the
radio medium. The link manager is responsible for the creation,
coexistence of WLAN and Bluetooth. These interference avoidance
modification, and release of logical links.
mechanisms are divided into two categories [8], and [12]:
Collaborative and non-Collaborative mechanisms. A short
description of these mechanisms is as follow:
A. Collaborative Mechanisms:
In collaborative mechanisms both WLAN and Bluetooth
communicates to avoid mutual interference. Collaborative
mechanisms require that both the systems should be collocated.
Collaborative mechanisms include Time Division Multiple Access
(TDMA), and MAC EnHanced Temporal Algorithm (MEHTA) [10].
In the TDMA technique, the 802.11 and Bluetooth networks
transmit alternately, defining the time that each has access to the
medium. The TDMA technique can support multiple Bluetooth
piconets. MEHTA uses a centralized controller that monitors the
WLAN and Bluetooth traffic and allows exchange of information
between these two radio systems. The centralized controller works
on MAC layer and allow or deny the transmission, thus avoiding
interference between the two systems.
Fig. 2 Bluetooth Protocol Stack [13] B. Non-Collaborative Mechanisms

The device manager is the functional block in the baseband that In non-collaborative mechanisms there is no method for WLAN and
controls the general behavior of the Bluetooth enabled device. The Bluetooth to communicate each other. WLAN and Bluetooth take
L2CAP resource manager block is responsible for managing the independent means to avoid interference. Non-collaborative
ordering of submission of PDU fragments to the baseband and some mechanisms include Adaptive Packet Selection and Scheduling [2],
relative scheduling between channels. The channel manager is and Adaptive Frequency Hopping [5], [11]. Bluetooth specifies a
113
variety of packet types with different combinations of payload 22MHz channel as “bad” frequency channel that is being used by
length, slots occupied, FEC codes, and ARQ options. The WLAN in the area. Bluetooth starts its normal communication
motivation is to provide necessary flexibility for the implementers except using these 22MHz frequency channel that has been marked
and applications so that the packets can be chosen optimized for the “bad” in its frequency hopping table. If the Bluetooth device does
traffic and channels presented. By selecting the best packet type not find any of WLAN control packet, it immediately starts its
according to the channel condition of the upcoming frequency hop, normal communication. If during communication a Bluetooth
better data throughput and network performance can be obtained. In device detects interference then it stops its communication and
addition, by carefully scheduling packet transmission we can rescans for WLAN control packet and follows according to the
avoid/minimize interference to WLAN systems and at the same time information provided.
increase the throughput of the Bluetooth / IEEE 802.15.1 systems.
Similarly Adaptive frequency hopping mechanism dynamically
changes the frequency hopping sequence in order to avoid or
minimize the interference detected by the 802.15.1 device. Adaptive
frequency hopping is a method by which the available channels are
used intelligently to decrease the likelihood of packet loss.
V.PROPOSED MODEL
The interference problem and fair allocation of frequencies between

different systems competing for same resources have led different
research groups to present different mechanism to mitigate the
interference problem.
WLAN and Bluetooth work on the same unlicensed frequency band

(2.4 GHz). A new method is proposed to mitigate the interference
problem between these two systems when working in parallel.
The proposed model is based on a common control channel
(CCCH). CCCH is a low rate signal used for synchronization
between WLAN and Bluetooth. On CCCH a narrow band signal is
generated WLAN access point (AP) and received by all the devices
in the area. The concept is to enable sharing information such as
carrier frequency, bandwidth, power level, time slots, etc. between
neighboring radio devices competing to access the same frequency
band via a common control channel. This work is focused on
WLAN and Bluetooth for interference cancellation further it can be
applied to any system in the ISM band.
A. Common Control Channel basics Fig. 4 Working of CCCH for noise cancellation
The new concept is based on a common control channel (CCCH).
•
CCCH is used by different radio devices for synchronization. The The main advantages of the proposed mechanism are the following:
key concept is that a control packet is being transmitted as an The proposed mechanism does not require a centralized
•
announcement packet on the same ISM band. The transmitted traffic scheduler as in MEHTA.
packet includes some parameters such as power level, carrier It can be implemented either when 802.11 and Bluetooth
frequency, time slots duration, duty cycle etc. CCCH allows to add are able to exchange information (collaborative
more parameters if needed. Figure 3 shows a packet format of coexistence mechanism) or when they acquire this
CCCH message. information by detecting interfering transmissions over
the radio channel (non-collaborative coexistence
•
mechanism).
It has a minor impact on the IEEE 802.11 standard and the
•
Bluetooth specification.
Reduces number of Link Manager Commands to
exchange information about “good” and “bad” channels
Fig. 3 Packet format of a single message on the CCCH
as compared to AFH.
The working of the proposed model is shown by a flow chart in the VI. CONCLUSIONS
Figure 4. When a Bluetooth device starts communication, it first
scans for WLAN control packet. Control packet is transmitted by WLAN and Bluetooth devices share common frequency spectrum in
the WLAN AP and received by all the devices in the area including the 2.4 GHz ISM band. When WLAN and Bluetooth operate on the
Bluetooth. This control packet is transmitted at a regular pattern just same area they often cause interference. A packet is lost due to noise
like beacon control packet transmitted by WLAN AP for WLAN when both WLAN and the Bluetooth packet overlap in the time and
devices. frequency domains. A number of mechanisms have been developed
to mitigate this interference problem. In this paper a new method is
In a Bluetooth piconet the master device receives this packet and proposed to avoid interference between WLAN and Bluetooth. The
adjusts its frequency hopping pattern according to the information new model is based on a control channel. Control packet is
provided in the control packet. This control packet provides transmitted from WLAN AP periodically; the control packet having
information about the frequency channel used by WLAN and other parameters such as frequency channels used by WLAN devices etc,
useful parameters. As in a piconet all slave devices follow the is received by Bluetooth devices to adjust its frequency hopping
master device for communication, the master device mark the pattern to avoid interference.
114
REFERENCES
[1]. Ramakrishna Gummadi, David Wetherall, Ben Greenstein, Srinivasan [8]. Carla F. Chiassserini, Ramesh R. Rao, “A Comparison between
Seshan, “Understanding and Mitigating the Impact of RF Interference Collaborative and Non-Collaborative Coexistence Mechanisms for
on 802.11 Networks”, SIGCOMM’07, August 27–31, 2007, Kyoto, Interference Mitigation in ISM Band”, 2001.
Japan.
[2]. Chen-Han Shih, Kuochen Wang , Hung-Cheng Shih,“ An adaptive [9]. R. J. Punnoose, R. S. Tseng, D. D. Stancil. “Experimental Results for
bluetooth packet selection and scheduling scheme in interference Interference between Bluetooth and IEEE 802.11b DSSS Systems”, In
environments”, Computer Communications 29 (2006) 2084–2095, 17 Proceedings of IEEE Vehicular Society Conference, October 2001.
February 2006.
[3]. Abhishek P. Patil, Dan J. Kim, Lionel M. Ni, “ A study of frequency [10]. IEEE 802.15.2-01300r1, TG2 Mobilian Draft Text, TDMA and
interference and indoor location sensing with 802.11b and Bluetooth MEHTA,
technologies”, Int. J. Mobile Communications, Vol. 4, No. 6, 2006. http://grouper.ieee.org/groups/802/15/pub/2001/Jul01/01300r1P802-1
[4]. Angela Doufexi, Arun Arumugam, Simon Armour and Andrew Nix, An 5_TG2-Mobilian-draft-text.doc
Investigation of the Impact of Bluetooth Interference on the [11]. IEEE P802.15-TG2_366r1, Clause 14.3 Adaptive Frequency Hopping,
Performance of 802.11g Wireless Local Area Networks”, 2004.
http://grouper.ieee.org/groups/802/15/pub/2001/Jul01/01366r1P802-1
[5]. N. Golmie, O. Rebala, N. Chevrollier, “Bluetooth Adaptive Frequency
Hopping and Scheduling” 2004. 5_TG2-Clause-14-3-Adaptive-Frequency-Hopping.doc
[6]. Angela Doufexi, Arun Arumugam, Simon Armour and Andrew Nix, “An [12]. Andrew Fredman, “Mechanisms of Interference Reduction for
Investigation of the Impact of Bluetooth Interference on the Bluetooth”, 2003
Performance of 802.11g Wireless Local Area Networks” [13].http://www.bluetooth.com/Bluetooth/Learn/Works/Core_System_Arch
IEEE_VTC_Spring 2003 itecture.htm, © 2007 Bluetooth SIG, Inc.
[14]. Kavesh Pahlavan, Prashant Krishnamurthy, “Principle of Wireless
[7]. Carla F. Chiasserini and Ramesh R. Rao, “Coexistence Mechanisms for Networks: A unified approach”, 2006.
Interference Mitigation between IEEE 802.11 WLANs and Bluetooth”,
0-7803-7476-2/02 IEEE 2002.
115

6-3confrence Report

Uploaded by

Copyright:

Available Formats

You might also like

6-3confrence Report

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

6-3confrence Report

Uploaded by

Copyright:

Available Formats

Proceeding

1st National Conference on Security, Computing, & Communication

May 23-25, 2008

1. Conference Honorary Committee i

Conference Chapter on Security

Conference Chapter on Computing

1. Assessing the Usability of Official Web Portal of Government of Pakistan; A Heuristics 37

Conference Chapter on Communication

1. Analysis of VOIP by Using Different Vocoders in IEEE 802.11b Wireless Networks 84

Conference Founder & Principle Organizer

Conference Steering Committee

Mr. Tariq Saleem Qasuria Registrar, KUST

Session I Conference Chapter on Security

Chair Dr. Fauzan Mirza

Session II Conference Chapter on Computing

Chair Prof. Dr. Mushtaq Ahmed

Session III Conference Chapter on Communication

Chair Prof. Dr. Salikhov Tagaimurod Khaitovich

Session I Conference Chapter on Security

Chair Dr. Fauzan Mirza

A Semi-Fragile Watermarking Scheme for Gray Scale Image Authentication Based

Abstract — In this paper a semi-fragile watermarking scheme B. Semi-Fragile Watermarking Techniques

J. Fridrich presented her schemes in the paper [5]-[8] where an

After quantization, the first 3 and 22-36 DCT quantized coefficients

3) Step 3 Watermark Embedding

Once the desired bits string is extracted, it is de-quantized using the

In this paper, some of the fundamental image authentication

In future we will extend our scheme for color image authentication

(e) Recovered Image REFERENCES

WHAT SMART! ABOUT SMART CARDs technology

to develop the specifications for the use of smart cards in payment

1) RAM H. Programming Language

information or update it until and unless various checks are satisfied

In general, if the embedded microprocessor offers tamper-resistant

Fig. 6 Chip Card Architecture

A System for Multimedia Owner Identification Using PIC

Symbol Bits Symbol Bits

• Key mechanism • Mark mechanism

• Capsulated Form • Embedded Form

In our system we use LSB [3] watermarking scheme. We alter the

W= Decimal of first 8 most significant bits of SCode

LSB(Image[W,H]) XOR PIC[J]

Personal Authentication System Using Biometrics Techniques

a) Enrollment phase Pattern

b) Feature Extraction phase

c) Template Creation phase

Event' (n) = Number of Unsuccessful Enrollment Events

∑ FTE (n) − − − − − − − − − −(3(9))

FRR(n) = MatchUser' (n) / MatchUser (n).................( A(1)

False Acceptance Rate (FAR)

er'(n)= Numberof ImpostersSuccessfulfalseAuthentica

FAR = NonMatchUser' / NonMatchUser...........................(

Fingerprint – Optical 2.0% 4 False Acceptance

level of security is increased in this technique by adding REFERENCES

Securing Services in Multi-Agent Systems

Index Terms — Distributed Computing, Multi-Agent III. DESIGN & IMPLEMENTATION

DoS and DDoS Attacks; Analysis and Prevention

Earlier researches proposed several research models for countering