IJACSA - Volume 3 No. 7, July 2012

THE SCIENCE AND INFORMATION ORGANIZATION
www. t hes ai . or g | i nf o@t hes ai . or g

(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 3, No.7, 2012
(i)
www.ijacsa.thesai.org
Editorial Preface
From the Desk of Managing Editor
IJACSA seems to have a cult following and was a humungous success during 2011. We at The Science and Information
Organization are pleased to present the July 2012 Issue of IJACSA.
While it took the radio 38 years and the television a short 13 years, it took the World Wide Web only 4 years to reach 50
million users. This shows the richness of the pace at which the computer science moves. As 2012 progresses, we seem to
be set for the rapid and intricate ramifications of new technology advancements.
With this issue we wish to reach out to a much larger number with an expectation that more and more researchers get
interested in our mission of sharing wisdom. The Organization is committed to introduce to the research audience
exactly what they are looking for and that is unique and novel. Guided by this mission, we continuously look for ways to
collaborate with other educational institutions worldwide.
Well, as Steve Jobs once said, Innovation has nothing to do with how many R&D dollars you have, its about the people
you have. At IJACSA we believe in spreading the subject knowledge with effectiveness in all classes of audience.
Nevertheless, the promise of increased engagement requires that we consider how this might be accomplished,
delivering up-to-date and authoritative coverage of advanced computer science and applications.
Throughout our archives, new ideas and technologies have been welcomed, carefully critiqued, and discarded or
accepted by qualified reviewers and associate editors. Our efforts to improve the quality of the articles published and
expand their reach to the interested audience will continue, and these efforts will require critical minds and careful
consideration to assess the quality, relevance, and readability of individual articles.
To summarise, the journal has offered its readership thought provoking theoretical, philosophical, and empirical ideas
from some of the finest minds worldwide. We thank all our readers for their continued support and goodwill for IJACSA.
We will keep you posted on updates about the new programmes launched in collaboration.
We would like to remind you that the success of our journal depends directly on the number of quality articles submitted
for review. Accordingly, we would like to request your participation by submitting quality manuscripts for review and
encouraging your colleagues to submit quality manuscripts for review. One of the great benefits we can provide to our
prospective authors is the mentoring nature of our review process. IJACSA provides authors with high quality, helpful
reviews that are shaped to assist authors in improving their manuscripts.
We regularly conduct surveys and receive extensive feedback which we take very seriously. We beseech valuable
suggestions of all our readers for improving our publication.
Thank you for Sharing Wisdom!

Managing Editor
IJACSA
Volume 3 Issue 7 July 2012
ISSN 2156-5570 (Online)
ISSN 2158-107X (Print)
2012 The Science and Information (SAI) Organization
Vol. 3, No.7, 2012
(ii)
Associate Editors
Dr. Zuqing Zhu
Service Provider Technology Group of Cisco Systems, San Jose
Domain of Research: Research and development of wideband access routers for hybrid
fibre-coaxial (HFC) cable networks and passive optical networks (PON)
Dr. Ka Lok Man
Department of Computer Science and Software Engineering at the Xi'an Jiaotong-
Liverpool University, China
Domain of Research: Design, analysis and tools for integrated circuits and systems;
formal methods; process algebras; real-time, hybrid systems and physical cyber
systems; communication and wireless sensor networks.
Dr. Sasan Adibi
Technical Staff Member of Advanced Research, Research In Motion (RIM), Canada
Domain of Research: Security of wireless systems, Quality of Service (QoS), Ad-Hoc
Networks, e-Health and m-Health (Mobile Health)
Dr. Sikha Bagui
Associate Professor in the Department of Computer Science at the University of West
Florida,
Domain of Research: Database and Data Mining.
Dr. T. V. Prasad
Dean, Lingaya's University, India
Domain of Research: Bioinformatics, Natural Language Processing, Image Processing,
Expert Systems, Robotics
Dr. Bremananth R
Research Fellow, Nanyang Technological University, Singapore
Domain of Research: Acoustic Holography, Pattern Recognition, Computer Vision, Image
Processing, Biometrics, Multimedia and Soft Computing
Vol. 3, No.7, 2012
(iii)
Reviewer Board Members

A Kathirvel
Karpaga Vinayaka College of Engineering and
Technology, India
A.V. Senthil Kumar
Hindusthan College of Arts and Science
Abbas Karimi
I.A.U_Arak Branch (Faculty Member) & Universiti
Putra Malaysia
Abdel-Hameed A. Badawy
University of Maryland
Abdul Wahid
Gautam Buddha University
Abdul Hannan
Vivekanand College
Abdul Khader Jilani Saudagar
Al-Imam Muhammad Ibn Saud Islamic University
Abdur Rashid Khan
Gomal Unversity
Aderemi A. Atayero
Covenant University
Dr. Ahmed Nabih Zaki Rashed
Menoufia University, Egypt
Ahmed Sabah AL-Jumaili
Ahlia University
Akbar Hossain
Albert Alexander
Kongu Engineering College,India
Prof. Alcinia Zita Sampaio
Technical University of Lisbon
Amit Verma
Rayat & Bahra Engineering College, India
Ammar Mohammed Ammar
Department of Computer Science, University of
Koblenz-Landau
Anand Nayyar
KCL Institute of Management and Technology,
Jalandhar
Anirban Sarkar
National Institute of Technology, Durgapur, India
Arash Habibi Lashakri
University Technology Malaysia (UTM), Malaysia
Aris Skander
Constantine University
Ashraf Mohammed Iqbal
Dalhousie University and Capital Health
Asoke Nath
St. Xaviers College, India
B R SARATH KUMAR
Lenora College of Engineering, India
Babatunde Opeoluwa Akinkunmi
University of Ibadan
Badre Bossoufi
University of Liege
Balakrushna Tripathy
VIT University
Bharat Bhushan Agarwal
I.F.T.M.UNIVERSITY
Bharti Waman Gawali
Department of Computer Science &
information
Bremananth Ramachandran
School of EEE, Nanyang Technological University
Brij Gupta
University of New Brunswick
Dr.C.Suresh Gnana Dhas
Park College of Engineering and Technology,
India
Mr. Chakresh kumar
Manav Rachna International University, India
Chandra Mouli P.V.S.S.R
VIT University, India
Chandrashekhar Meshram
Chhattisgarh Swami Vivekananda Technical
University
Chi-Hua Chen
National Chiao-Tung University
Constantin POPESCU
Department of Mathematics and Computer
Science, University of Oradea
Prof. D. S. R. Murthy
SNIST, India.
Dana PETCU
West University of Timisoara
David Greenhalgh
University of Strathclyde
Deepak Garg
Thapar University.
Prof. Dhananjay R.Kalbande
Sardar Patel Institute of Technology, India
Dhirendra Mishra
SVKM's NMIMS University, India
Divya Prakash Shrivastava
Vol. 3, No.7, 2012
(iv)
EL JABAL AL GARBI UNIVERSITY, ZAWIA
Dragana Becejski-Vujaklija
University of Belgrade, Faculty of organizational
sciences
Fokrul Alom Mazarbhuiya
King Khalid University
G. Sreedhar
Rashtriya Sanskrit University
Gaurav Kumar
Manav Bharti University, Solan Himachal
Pradesh
Ghalem Belalem
University of Oran (Es Senia)
Gufran Ahmad Ansari
Qassim University
Hadj Hamma Tadjine
IAV GmbH
Hanumanthappa.J
University of Mangalore, India
Hesham G. Ibrahim
Chemical Engineering Department, Al-Mergheb
University, Al-Khoms City
Dr. Himanshu Aggarwal
Punjabi University, India
Huda K. AL-Jobori
Ahlia University
Dr. Jamaiah Haji Yahaya
Northern University of Malaysia (UUM), Malaysia
Jasvir Singh
Communication Signal Processing Research Lab
Jatinderkumar R. Saini
S.P.College of Engineering, Gujarat
Prof. Joe-Sam Chou
Nanhua University, Taiwan
Dr. Juan Jos Martnez Castillo
Yacambu University, Venezuela
Dr. Jui-Pin Yang
Shih Chien University, Taiwan
Jyoti Chaudhary
high performance computing research lab
K V.L.N.Acharyulu
Bapatla Engineering college
K. PRASADH
METS SCHOOL OF ENGINEERING
Ka Lok Man
Xian Jiaotong-Liverpool University (XJTLU)
Dr. Kamal Shah
St. Francis Institute of Technology, India
Kanak Saxena
S.A.TECHNOLOGICAL INSTITUTE
Kashif Nisar
Universiti Utara Malaysia
Kayhan Zrar Ghafoor
University Technology Malaysia
Kodge B. G.
S. V. College, India
Kohei Arai
Saga University
Kunal Patel
Ingenuity Systems, USA
Labib Francis Gergis
Misr Academy for Engineering and Technology
Lai Khin Wee
Technischen Universitt Ilmenau, Germany
Latha Parthiban
SSN College of Engineering, Kalavakkam
Lazar Stosic
College for professional studies educators,
Aleksinac
Mr. Lijian Sun
Chinese Academy of Surveying and Mapping,
China
Long Chen
Qualcomm Incorporated
M.V.Raghavendra
Swathi Institute of Technology & Sciences, India.
Madjid Khalilian
Islamic Azad University
Mahesh Chandra
B.I.T, India
Mahmoud M. A. Abd Ellatif
Mansoura University
Manpreet Singh Manna
SLIET University, Govt. of India
Manuj Darbari
BBD University
Marcellin Julius NKENLIFACK
University of Dschang
Md. Masud Rana
Khunla University of Engineering & Technology,
Bangladesh
Md. Zia Ur Rahman
Narasaraopeta Engg. College, Narasaraopeta
Messaouda AZZOUZI
Ziane AChour University of Djelfa
Dr. Michael Watts
University of Adelaide, Australia
Milena Bogdanovic
University of Nis, Teacher Training Faculty in
Vranje
Vol. 3, No.7, 2012
(v)
Miroslav Baca
University of Zagreb, Faculty of organization and
informatics / Center for biomet
Mohamed Ali Mahjoub
Preparatory Institute of Engineer of Monastir
Mohammad Talib
University of Botswana, Gaborone
Mohammad Ali Badamchizadeh
University of Tabriz
Mohammed Ali Hussain
Sri Sai Madhavi Institute of Science &
Technology
Mohd Helmy Abd Wahab
Universiti Tun Hussein Onn Malaysia
Mohd Nazri Ismail
University of Kuala Lumpur (UniKL)
Mona Elshinawy
Howard University
Mueen Uddin
Universiti Teknologi Malaysia UTM
Dr. Murugesan N
Government Arts College (Autonomous), India
N Ch.Sriman Narayana Iyengar
VIT University
Natarajan Subramanyam
PES Institute of Technology
Neeraj Bhargava
MDS University
Nitin S. Choubey
Mukesh Patel School of Technology
Management & Eng
Pankaj Gupta
Microsoft Corporation
Paresh V Virparia
Sardar Patel University
Dr. Poonam Garg
Institute of Management Technology,
Ghaziabad
Prabhat K Mahanti
UNIVERSITY OF NEW BRUNSWICK
Pradip Jawandhiya
Jawaharlal Darda Institute of Engineering &
Techno
Rachid Saadane
EE departement EHTP
Raj Gaurang Tiwari
AZAD Institute of Engineering and Technology
Rajesh Kumar
National University of Singapore
Rajesh K Shukla
Sagar Institute of Research & Technology-
Excellence, India
Dr. Rajiv Dharaskar
GH Raisoni College of Engineering, India
Prof. Rakesh. L
Vijetha Institute of Technology, India
Prof. Rashid Sheikh
Acropolis Institute of Technology and Research,
India
Ravi Prakash
University of Mumbai
Reshmy Krishnan
Muscat College affiliated to stirling University.U
Rongrong Ji
Columbia University
Ronny Mardiyanto
Institut Teknologi Sepuluh Nopember
Ruchika Malhotra
Delhi Technoogical University
Sachin Kumar Agrawal
University of Limerick
Dr.Sagarmay Deb
University Lecturer, Central Queensland
University, Australia
Said Ghoniemy
Taif University
Saleh Ali K. AlOmari
Universiti Sains Malaysia
Samarjeet Borah
Dept. of CSE, Sikkim Manipal University
Dr. Sana'a Wafa Al-Sayegh
University College of Applied Sciences UCAS-
Palestine
Santosh Kumar
Graphic Era University, India
Sasan Adibi
Research In Motion (RIM)
Saurabh Pal
VBS Purvanchal University, Jaunpur
Saurabh Dutta
Dr. B. C. Roy Engineering College, Durgapur
Sergio Andre Ferreira
Portuguese Catholic University
Seyed Hamidreza Mohades Kasaei
University of Isfahan
Shahanawaj Ahamad
The University of Al-Kharj
Shaidah Jusoh
University of West Florida
Sikha Bagui
Zarqa University
Vol. 3, No.7, 2012
(vi)
Sivakumar Poruran
SKP ENGINEERING COLLEGE
Slim BEN SAOUD
Dr. Smita Rajpal
ITM University
Suhas J Manangi
Microsoft
SUKUMAR SENTHILKUMAR
Universiti Sains Malaysia
Sumazly Sulaiman
Institute of Space Science (ANGKASA), Universiti
Kebangsaan Malaysia
Sunil Taneja
Smt. Aruna Asaf Ali Government Post Graduate
College, India
Dr. Suresh Sankaranarayanan
University of West Indies, Kingston, Jamaica
T C. Manjunath
BTL Institute of Technology & Management
T C.Manjunath
Visvesvaraya Tech. University
T V Narayana Rao
Hyderabad Institute of Technology and
Management
T. V. Prasad
Lingaya's University
Taiwo Ayodele
Lingaya's University
Totok R. Biyanto
Infonetmedia/University of Portsmouth
Varun Kumar
Institute of Technology and Management, India
Vellanki Uma Kanta Sastry
SreeNidhi Institute of Science and Technology
(SNIST), Hyderabad, India.
Vijay Harishchandra
Vinayak Bairagi
Sinhgad Academy of engineering, India
Vitus S.W. Lam
The University of Hong Kong
Vuda Sreenivasarao
St.Marys college of Engineering & Technology,
Hyderabad, India
Wichian Sittiprapaporn
Mahasarakham University
Xiaojing Xiang
AT&T Labs
Y Srinivas
GITAM University
Mr.Zhao Zhang
City University of Hong Kong, Kowloon, Hong
Kong
Zhixin Chen
ILX Lightwave Corporation
Zuqing Zhu
University of Science and Technology of China
Vol. 3, No.7, 2012
(vi)
CONTENTS
Paper 1: Bond Portfolio Analysis with Parallel Collections in Scala
Authors: Ron Coleman, Udaya Ghattamanei, Mark Logan
PAGE 1 9

Paper 2: Core Backbone Convergence Mechanisms and Microloops Analysis
Authors: Abdelali Ala, Driss El Ouadghiri, Mohamed Essaaidi
PAGE 10 25

Paper 3: SVD Based Image Processing Applications: State of The Art, Contributions and Research Challenges
Authors: Rowayda A. Sadek
PAGE 26 34

Paper 4: A Modified Feistel Cipher Involving XOR Operation and Modular Arithmetic Inverse of a Key Matrix
Authors: Dr. V. U. K Sastry, K. Anup Kumar
PAGE 35 39

Paper 5: A Modified Feistel Cipher Involving Modular Arithmetic Addition and Modular Arithmetic Inverse of a Key Matrix
Authors: Dr. V. U. K Sastry, K. Anup Kumar
PAGE 40 43

Paper 6: The Japanese Smart Grid Initiatives, Investments, and Collaborations
Authors: Amy Poh Ai Ling, Sugihara Kokichi, Mukaidono Masao
PAGE 44 54

Paper 7: Evaluating the Role of Information and Communication Technology (ICT) Support towards Processes of
Management in Institutions of Higher Learning
Authors: Michael Okumu Ujunju, Dr. G. Wanyembi, Franklin Wabwoba
PAGE 55 58

Paper 8: Smart Grids: A New Framework for Efficient Power Management in Datacenter Networks
Authors: Okafor Kennedy .C, Udeze Chidiebele. C, E. C. N. Okafor, C. C. Okezie
PAGE 59 66

Paper 9: An Online Character Recognition System to Convert Grantha Script to Malayalam
Authors: Sreeraj.M, Sumam Mary Idicula
PAGE 67 72

Paper 10: LOQES: Model for Evaluation of Learning Object
Authors: Dr. Sonal Chawla, Niti Gupta, Prof. R.K. Singla
PAGE 73 79

Paper 11: FHC-NCTSR: Node Centric Trust Based secure Hybrid Routing Protocol for Ad Hoc Networks
Authors: Prasuna V G, Dr. S Madhusudahan Verma
PAGE 80 89

Vol. 3, No.7, 2012
(vii)
Paper 12: Simultaneous Estimation of Geophysical Parameters with Microwave Radiometer Data based on Accelerated
Simulated Annealing: SA
Authors: Kohei Arai
PAGE 90 95

Paper 13: Task Allocation Model for Rescue Disabled Persons in Disaster Area with Help of Volunteers
Authors: Kohei Arai, Tran Xuan Sang, Nguyen Thi Uyen
PAGE 96 101

Paper 14: Image Clustering Method Based on Density Maps Derived from Self-Organizing Mapping: SOM
Authors: Kohei Arai
PAGE 102 107

Paper 15: Improving the Solution of Traveling Salesman Problem Using Genetic, Memetic Algorithm and Edge assembly
Crossover
Authors: Mohd. Junedul Haque, Khalid. W. Magld
PAGE 108 111

Paper 16: A Hybrid Technique Based on Combining Fuzzy K-means Clustering and Region Growing for Improving Gray
Matter and White Matter Segmentation
Authors: Ashraf Afifi
PAGE 112 118

Paper 17: GUI Database for the Equipment Store of the Department of Geomatic Engineering, KNUST
Authors: J. A. Quaye-Ballard, R. An, A. B. Agyemang, N. Y. Oppong-Quayson, J. E. N. Ablade
PAGE 119 124

Paper 18: Comparative Study between the Proposed GA Based ISODAT Clustering and the Conventional Clustering
Methods
Authors: Kohei Arai
PAGE 125 131

Paper 19: Improving the Rate of Convergence of Blind Adaptive Equalization for Fast Varying Digital Communication
Systems
Authors: Iorkyase, E.Tersoo, Michael O. Kolawole
PAGE 132 136

Paper 20: Throughput Analysis of Ieee802.11b Wireless Lan With One Access Point Using Opnet Simulator
Authors: Isizoh A. N, Nwokoye A. O.C, Okide S. O, Ogu C. D
PAGE 137 139

Paper 21: BPM, Agile, and Virtualization Combine to Create Effective Solutions
Authors: Steve Kruba, Steven Baynes, Robert Hyer
PAGE 140 146
Vol.3, No.7, 2012

1 | P a g e
Bond Portfolio Analysis with Parallel Collections
in Scala
Ron Coleman
Computer Science Department
Marist College
Poughkeepsie, NY, United States

Udaya Ghattamanei
Marist College
Poughkeepsie, NY, United States

Mark Logan
Marist College acronyms
Poughkeepsie, NY, United St
Abstract In this paper, we report the results of new experiments
that test the performance of Scala parallel collections to find the
fair value of riskless bond portfolios using commodity multicore
platforms. We developed four algorithms, each of two kinds in
Scala and ran them for one to 1024 portfolios, each with a
variable number of bonds with daily to yearly cash flows and 1
year to 30 year. We ran each algorithm 11 times at each
workload size on three different multicore platforms. We
systematically observed the differences and tested them for
statistical significance. All the parallel algorithms exhibited
super-linear speedup and super-efficiency consistent with
maximum performance expectations for scientific computing
workloads. The first-order effort or nave parallel algorithms
were easiest to write since they followed directly from the serial
algorithms. We found we could improve upon the nave approach
with second-order efforts, namely, fine-grain parallel algorithms,
which showed the overall best, statistically significant
performance, followed by coarse-grain algorithms. To our
knowledge these results have not been presented elsewhere.
Keywords- parallel functional programming; parallel processing;
multicore processors; Scala; computational finance.
I. INTRODUCTION
A review of the high performance computing literature
suggests opportunities and challenges to exploit parallelism to
solve compute-intensive problems. [1] [2] [3]. Proponents of
functional programming have long maintained that elaboration
of the lambda calculus lends itself to mathematical
expressiveness and avoids concurrency hazards (e.g., side-
effects, managing threads, etc.) that are the bane of shared-state
parallel computing. [4] Yet parallel functional programming
has remained largely outside the mainstream programming
community. [5] One could conceivably argue that parallel
functional programming was ahead of its time and the era of
inexpensive multicore processors in which some investigators
have observed that the free lunch is over since clock speeds
have been decreasing or at least not increasing significantly,
necessitating a turn toward parallel programming. [6]
Enter Scala [7], a relatively new, general-purpose language
which runs on the Java Virtual Machine (JVM) and hence,
desktops, browsers, servers, cell phones, tablets, set-tops, and
lately, GPUs [8] [9] [10], a related topic we do not explore here
(see the section, Conclusions and Future Directions). Scala
blends object-oriented and functional styles with shared-
nothing, task-level parallelism based on the actor model. [7]
Parallel collections [11] [12] are recent additions that provide
data-level parallelism [3] through a simple, functional
extension of the ordinary, non-parallel collections of Scala.
While the use of parallel collections has potential to improve
programmer productivity and greatly facilitate a transition to
parallel programming, no independent study has investigated
whether parallel collections scale in terms of run-time
performance on commodity hardware, taking into account
furthermore end-to-end processing that involves I/O which is
typically a prerequisite for and often the bottleneck of practical
applications.
Coleman, et al., conducted end-to-end experiments to find
the fair value of riskless bond portfolios using task-level
parallelism via map-reduce. [13] [14] In this paper, we take a
new, different tack on the same problem that applies data-level
parallelism via parallel collections. We were motivated to use
bond portfolio analysis, first, because computational finance
workloads can be very large. [15] Second, bond portfolio
pricing theory is fairly transparent. [16] Finally, bonds inform
or are closely related to other financial instruments, including
annuities, mortgage securities, bond derivatives, and interest
rate swaps, which are among the most heavily traded financial
contracts in the world. [17] Thus, computational methods and
performance results from this class of problem would likely
have implications beyond bonds and finance.
Indeed, the experiments with Scala parallel collections
using eight algorithms on three different hardware platforms
show super-linear speedup and super-efficiency are consistent
with the maximum performance expectations for scientific
computing workloads. While the data suggests that the more
modern processors are also more efficient, overall fine-grain
algorithms significantly outperform others in runtime, which
interests and surprises us considering the presumed overhead of
this approach. The coarse-grain algorithms are next best,
followed by the nave algorithms. The findings we report
here using parallel collections are new and have not been
reported elsewhere or by others. All the source code is
available online for review, download, and testing (see section,
Appendix Source Code).
II. METHODS
A. Parallel collections a primer
Scala has standard, template data structures called
collections, which include lists, arrays, ranges, and vectors,
among others. Scala collections are different from the ones it
also inherits from the Java standard library in that the Scala
Vol.3, No.7, 2012

2 | P a g e
versions are typically immutable with methods to operate on
the data elements using functional objects. For instance, to
multiply every element of a range collection by two using the
map method, we have the snippet below (where scala> is the
Scala interactive shell prompt):
scala> (1 to 5).map(x => x * 2)
Vector(2, 4, 6, 8, 10)
Snippet 1. Maps sequential range.

The parameter, x => x * 2, an anonymous function literal
object, receives each element of the range collection as an
immutable value parameter, x, multiplies it by two, and map
copies the result into a new collection, Vector.
The methods of a parallel collection as accessed in the same
way except the method name is preceded by .par as the snippet
below suggests:
scala> (1 to 5).par.map(x => x * 2)
ParVector(10, 6, 2, 8, 4)
Snippet 2. Maps the parallel range.
Here map invokes the function literal object on the range
using the machines parallel resources. The parallel collections
map method returns a parallel vector, ParVector, in which the
ordering of the return results is unspecified because of the
asynchronous nature of parallel execution. From a
programmers point of view, virtually no effort is involved to
parallelize the code. There are no new programming constructs
to learn and apply and algorithm redesign and code refactoring
are not demanded. There is furthermore no need to write
special test cases to verify the results since in principle the
serial (non-parallel) implementation is the test case. While the
result ordering may need to be addressed, in general, parallel
collections are a potential windfall for programmer
productivity and transitioning to parallel programming.
The research question is whether use of .par scales,
enabling speed-up and efficiency on a non-trivial problem on
commodity hardware. For bond portfolio analysis, the
functional nature of parallel collections makes implementation
of the pricing equations straightforward. In the nave case,
we simply reuse the pricing function object from the serial
algorithm with no other changes to the code other than to apply
.par, just as we did in the above snippet. However, we go
further and explore whether we can obtain further
improvements using fine-grain and course-grain algorithms.
B. Pricing theory
For purposes of this paper, we are considering only simple
bonds [16] b
i
, defined by the five-tuple:
b
i
[i, C, n, T, M] (1)
i is an integer which plays no part in bond pricing except to
uniquely identify the bond in an inventory which we describe
below; C is the coupon amount paid one or more times; n is the
payment frequency of coupons per annum; T is the time to
maturity in years; and M is the face value due at maturity. The
sum of the net present value of these cash flows, C and M, is
the fair value of the bond. Thus, the fair value, P(b
i,,
r), of a
bond, b
i
, is the net-present value of its cash flows which
functionally defined as:
P(b
i
, r)
C
(1+r
t
)
t /n
t1
n T
+
M
(1+r
T
)
T

(2)
The parameter, r, is the time-dependent yield curve, the
general discussion of which is beyond the scope of this paper.
Without loss of generality, we use the United States Treasury
on-the-run bond yield curve, which we observe once. We
interpolate between the tenors (i.e., Treasury maturity dates)
using polynomial curve fitting, the coefficients of which we
cache and apply for all bonds in the inventory.
A portfolio is a collection instruments, in our case, bonds.
The fair value, P(|
j
), of a portfolio, |
j
, with a basket of Q bonds
is functionally defined as follows:
P(
j
) P(b
( j,q)
, r)
q1
Q

(3)
C. Bond portfolio generation
We generate simple bonds that model a wide range of
computational scenarios. The goals are to 1) produce a
sufficient number of bonds to mimic realistic fixed-income
portfolios and 2) avoid biases in commercial-grade bonds that
depend on prevailing market conditions. Specifically, we have
the collections, n
={1, 4, 12, 52, 365), T
={1, 2, 3, 4, 5, 7, 10,
30}, and o
={0.005, 0.01, 0.02, 0.03, 0.04, 0.05}, where the

elements of n
are payment frequencies, T
are maturities, and

o
are coefficients. We derive the parameters for a bond object

from the bond generator equations below:
M=1000 (4a)
n =n [-]
(4b)
T =T [-]
(4c)
C = M / T [-]
(4d)
where - is an integer uniform random deviate in the range
of [0, s-1]; and s is the size of the respective collection. We
invoke Equations 4a - 4d a total of 5,000 times to produce the
bond inventory, V, which we store in an indexed persistent
database that we describe below.
We generate a portfolio by first selecting its size, that is, the
number of bonds, Q, per the equation below.
Q v+
(5)
q is a Gaussian deviate with mean of zero and one standard
deviation. v and o are configurable parameters set to 60 and 20,
respectively. Finally, we construct a basket of size, Q, bonds
for a portfolio, |
j
. We use the equation below to specify a bond
id or primary key,
i (6)
where - is an integer uniform random deviate in the range
of [1,|V|] and |V|=5,000 is the size of the bond inventory. We
generate a universe, U, of bond portfolios where |U|=100,000.
Vol.3, No.7, 2012

3 | P a g e
The bond portfolios are also store in a database indexed by j, a
unique portfolio id.
D. Database design
We store the bonds, b
i
, portfolios, |
j
(which also contains
the result of Equation 3) in MongoDB, an indexed, document-
oriented, client-server database. [18] As we noted above, |
j

does not contain bond objects, b
i
, but the bond primary key, i.
In MongoDB parlance, the bonds are linked to portfolios rather
than embedded by them. In other words, the database is
organized in third-object normal (3ONF) form. [19] Thus, to
evaluate Equation 3, a total of 2+Q accesses are necessary: one
access to fetch |
j
; Q fetches to retrieve each b
i
; and finally, one
store to update the portfolio, |
j
, with its price. The figure below
gives the class diagram, as it is stored in the document
repository.

Figure 1. Third normal object form (3ONF) of the database
Although this design is consistent with best practices for
data modeling, we could reduce the number of database
accesses at the expense of redundancy through
denormalization. However, we decided to forgo this
optimization in the interests of establishing a baseline of
performance for future reference.
E. Algorithms
We develop two classes of algorithms: serial and parallel.
There are three types of parallel algorithms, nave, fine-grain,
and coarse-grain. Each serial and parallel algorithm comes in
two kinds: composite and memory-bound. The composite kind,
represented by the notation, {io+compute}, overlaps access to
the database while evaluating Equation 2 and Equation 3. The
memory-bound kind, represented by the notation,
{io}+{compute}. In other words, we measure I/O ({io}) and
compute ({compute}) runtimes separately, first caching all the
bonds by portfolio into memory and only then evaluating
Equation 2 and Equation 3. I/O ({io}) and compute
({compute}) runtimes furthermore provide insight into the
maximum compute and IO performance potentials. In each
case, the algorithms evaluate the same collection of portfolios,
UcU, which has been randomly sampled from the database.
We give here only snippets from the source code. See the
appendix to access the complete source.
F. Serial algorithms
We invoke the composite serial algorithm as the snippet
below suggests.
val outputs = inputs.map(price)
Snippet 3. Maps input of randomly sampled portfolio key ids to price results
The object, inputs, is a collection of portfolio ids and
outputs is a collection of portfolio prices. (The val
declaration means that outputs is an immutable value object.)
The parameter, price, is a named function object with the
declaration:
def price(input: Data): Data
Snippet 4. Price the collection of randomly sampled portfolio ids serially
This means price receives a Data object as an input
parameter and returns a Data object. We wrote the Data object
for use by all the algorithms of this study. It contains the
portfolio id, a list of bonds, and a result object which itself
contain the portfolio price and diagnostic information about the
run. On input in this case, the Data object has set only the
portfolio id. On output, Data has the portfolio id and the result
object defined.
The function object, price, accesses the 3ONF repository to
retrieve a portfolio by its id and then retrieve the bond objects,
pricing them according to Equation 2, then according to
Equation 3 summing the prices using the foldLeft method. (For
readers who may be unfamiliar with functional programming,
folding is a common operation in functional programming
for aggregating elements. The foldLeft method is a serial
aggregator, traversing the collection, left-to-right, that is, from
the element at index zero to the element of the last index. The
analogous foldRight traverses the collection from right-to-left
using tail-recursion. We prefer foldLeft as opposed to
foldRight to avoid the problem of stack overflow.)
The serial memory-bound algorithm is virtually identical to
the composite algorithm as the snippet below suggests.
val inputs = loadPortfsFoldLeft(n)
val outputs = inputs.map(price)
Snippet 5. Serially load the bonds in memory, then price portfolios serially
The method, loadPortfsFoldLeft, loads a random sample of
n portfolios from the database and uses foldLeft to aggregate
the corresponding bonds. Thus, in this case, the inputs value is
a collection of Data objects, each containing a list of bond
objects. The parameter, price, is a function object, the same one
used in the composite serial algorithm.
G. Nave algorithms
The nave algorithms are so-called because, as a first-order
effort, they naively use .par. They are virtually identical to
the serial algorithms. That is, we have the snippet below for the
composite case.
val outputs = inputs.par.map(price)
Snippet 6. Price the collection of randomly sampled portfolio ids in parallel
We have the snippet below for the memory-bound kind.
val inputs = loadPortfsParFold(n)
val outputs = inputs.par.map(price)
Snippet 7. First, load the bonds into memory in parallel by portfolio id, then
prices the portfolios in parallel
Notice that the memory-bound kind uses loadPortfsParFold
(i.e., rather than loadPortfsFoldLeft), which accesses the
database and loads the portfolios in parallel using a parallel
collection. It uses Scalas par.fold method. This method
aggregates like its serial version, foldLeft, except par.fold does
so in parallel with non-deterministic ordering.
Portfolio
-id: Int
-bondIds: List[Int]
-price: Double
Bond
-id: Int
-freq : Int
-coupon: Double
-tenor: Double
-maturity: Double
1..n *
Vol.3, No.7, 2012

4 | P a g e

.
.
.
#999 #5
query query
++ ++
++
List[Data]( )
List[Data(17,List(SimpleBond(12,)),]

Figure 2. Parallel IO query-merge tree using the par.fold method
The figure above shows how loadPortfsParFold works.
Namely, we start with an empty List collection. Here for the
sake of demonstration, portfolios, #999 and #5, are being
loaded into memory from the database by the query
operation. The ++ nodes are binary operations that merge
partial lists of bond objects until a complete list is merged at
the root in O(log N) time. At the top of the merge tree we have
the fully merged in-memory List collection of portfolio data
objects. In this depiction, the value, 17, represents a portfolio id
chosen for demonstration purposes. Thus, the outer list contains
portfolio data objects, each of which contains a list of bond
objects. Note that this parallel memory-caching algorithm is
not embarrassingly parallel as the data lists must be merged.
H. Fine-grain algorithms
In a second-order effort to improve the nave application of
.par, we developed fine-grain algorithms, composite and
memory-bound kinds. Unlike the nave algorithm, the fine-
grain algorithm uses a parallel collection within the pricing
function object. In other words, we have a parallel collection
within a parallel collection.
The inner parallel collection has a bondPrice function
object to price the bonds by their id (i.e., it makes a query to
the database) per Equation 2 using par.map and a sum function
object to reduce (i.e.., accumulate) the bond prices in parallel
using par.reduce. In effect, we have the snippet below of the
price function.
val output = input.bondsIds.par.
map(bondPrice).par.reduce(sum)
Snippet 8. Price bonds in parallel by their ids then reduce prices in parallel.
Bond prices flow directly to their reduction in an O(log N)
processing tree. Thus, like parallel I/O, the workload is not
embarrassingly parallel as the figure below suggests.
The memory-bound algorithm is similar except, it uses the
parallel IO query-tree to access the database and cache the
bonds in memory.

b
0
b
1
b
q-1
.
.
.
bondPrice
bondPrice
bondPrice
sum
sum
sum
sum
sum

portfPrice
Equation 2 Equation 3

Figure 3. Accessing and pricing bonds then reducing prices in parallel.
I. Coarse-grain algorithms
The other algorithms created a parallel collection of input
portfolios whose size was independent of the number of
processor cores. The idea of the parallel coarse-grain algorithm
is to chunk the portfolios as a second-order effort to the nave
application of .par. That is, we create a parallel collection
whose size is proportional to the number of processors.
The design of parallel collections does not provide a direct
way to bind the pricing function to a core. This is part of
parallel collections design philosophy: the programmer focuses
on the functional specification and the parallel collection
distributes it across the cores.
Nevertheless, the programmer can control the chunk size by
making the input collection a List of a List of portfolio ids. For
example, for a four-core platform like the W3540 we study, the
containing List has eight List elements.
Each element has |U|/c portfolios where c is the number of
cores. For u=1024 portfolios, each element in the containing is
a List of 128 portfolios. The pricing function object is then
passed this list with 128 portfolios, which it processes serially
to evaluate Equation 2 and Equation 3.
To compute the size of the contained List, we use the Java
class, Runtime. It has a method, availableProcessors( ).
However, this method returns the number of hyperthreads, not
the number of cores. As far as we know there is no way to get
the number of core except manually from the OEM datasheets,
which we rely on for calculating the efficiency (see below).
Otherwise, programmatically we use the Runtime class.
The coarse-grain composite algorithm loads the bond
objects by portfolio just as the nave algorithm except it does in
chunks on-demand. The memory-bound algorithm, like its
nave and fine-grain counterparts, uses the parallel IO query
tree to cache the bonds in memory.
III. EXPERIMENTAL DESIGN
A. Environment
The test environment consisted of three hardware platforms
of different Intel multicore processors. The table below shows
the system configurations, with the clock speed in GHz and
years of introduction by the Intel Corporation.

Vol.3, No.7, 2012

5 | P a g e
TABLE I. EXPERIMENTAL ENVIRONMENT
CPU Clock Cores Threads RAM Year
W3540 2.93 4 8 4 GB 2009
i7-2670M 3.20 2 4 4 GB 2011
i3-370M 2.40 2 4 2 GB 2010

All platforms run Microsoft Windows 7. The code was
compiled by Eclipse 3.7.1 using the Scala IDE plugin version
2.0.0. The code was executed with the 64-bit JVM. We used
MongoDB, version 1.8.3. Although MongoDB is accessed
through TCP/IP, the database server runs on the same host as
the Scala code. We indexed the portfolios and bonds
documents on their key ids.
B. Runs and trials
We instrument the code and make the following
measurements.
TABLE II. MEASUREMENTS SOURCES
# Algorithm kind Measurement (T)
1 Composite {io + compute}
2 Memory-bound {io}
3 Memory-bound {compute}
4 Memory-bound {io} + {compute}

For each algorithm by its kind in Table 2, we make a total
of 11 trial invocations of the code to obtain stable run-time
statistics following. [20] Each trial starts a new JVM, the code
of which allocates new JVM objects and opens new database
connections. The trial ends when the algorithm ends and the
code exits, terminating the JVM, which closes the database
connections and causes the operating system to recycle the
JVM objects. A given set of trials, taken together, we call a
run. There is a run for u=2
x
portfolios (i.e., the problem size)
where xe[0..10]. The run, u=1024, is we call the terminal run.
Note: #4 in Table 2 is not an actual run; it is derived by adding
the measurements for #2 and #3 for the respective runs. For
each run at a given problem size, we analyze the measurements
for statistical significance as we describe below. We also graph
the run-times using the median value of the run.
C. Speed-up and efficiency calculations
T
1
is the serial time of a serial algorithm. T
N
is the time
using parallel collections.
Given T
1
and T
N
where N is the number of cores, we have
the speedup, R:
RT
1
/ T
N
(8)
The efficiency, e, is
e R/ N (9)
In this case, N is the number of cores, which we got from
the OEM datasheets online. [21] [22] [23]
D. Statistical significance calculations
After obtaining the runtimes, we observe the differences
and test them for statistical significance in the indicated
direction. That is, if the median runtime of algorithm, A, is less
than the median runtime of algorithm, B, we have the null
hypothesis H
0
:
H
0
: E(T
A
) E(T
B
)
(10)
where E is expectation. To conservatively estimate the p
value, we used the one-tailed Mann-Whitney test. [24] We
report (see the appendix) the rank sum statistic, S,
S R(T
i
)
(11)
where R(T
i
) is the rank of runtime, T
i
. Since there are 11
observations for each algorithm, the one-tailed threshold for
p=0.05 is the rank sum, S
.05
=101. This value can be found in
Table A7 in [24]. Thus, for S < S
.05
, we reject H
0
.
We compare each of our eight algorithms relative to one
another and test the differences for statistical significance. To
make the report more accessible, we give the frequency count
for the number of times an algorithm is found to be statistically
significantly faster than another algorithm. Again, the rank
sums, S, algorithm by algorithm for each hardware platform,
can be found in the appendix.
We present graphical evidence for performance over the
range of u mentioned above for each algorithm on each
platform. We assess the statistical significance and present
tabular data only for the terminal run, u=1024.
IV. RESULTS
The table below gives the kind of algorithms symbolized in
the graphs and tables that follow.
TABLE III. KIND OF PROCESSING
Composite
Memory-bound
Compute-only
IO-only
A. Nave results
The results for the nave treatments are summarized in the
next three graphs, one for the W3540, i7, and i3, respectively.
The number of portfolios or problem size, is u=2
x
.
The speedup, R, is on the left axis, and the efficiency, e, is
on the right axis.

Figure 4. W3540 nave results
0
2
4
6
8
0 2 4 6 8 10
0%
50%
100%
150%
200%
R e
x (Note: u = 2^x)
= {io + compute}
= {io} + {compute}
= {compute}
= {io}
Vol.3, No.7, 2012

6 | P a g e

Figure 5. i7 nave results

Figure 6. i3 nave results
The table below gives the terminal run results. T
N
is the
median run-time in seconds.
TABLE IV. TERMINAL RUN, U=1,024, BOND PORTFOLIOS
W3540 i7 i3
TN R TN R TN R
14.80 4.37 19.36 2.46 23.19 3.00
12.39 4.95 16.93 2.63 24.17 2.75
8.74 5.66 13.90 2.76 18.74 2.95
3.52 2.49 3.07 1.89 5.39 2.13

Note: In general, the median operator does not distribute.
Namely, median({io} + {compute}) = median({io}) +
median({compute}). For example, for the W3540, T
N
({io} +
{compute}) = 12.39 whereas T
N
({io}) + T({compute}) = 8.74
+ 3.52 = 12.26.
B. Fine-grain results
The results for the fine-grain algorithms are summarized in
the next three graphs, one for the W3540, i7, and i3,
respectively.

Figure 7. W3540 fine-grain results

Figure 8. i7 fine-grain results

Figure 9. i3 fine-grain results
The table below gives the results for the terminal run.

0
1
2
3
4
0 2 4 6 8 10
0%
50%
100%
150%
200%
x (Note: u = 2^x)
R e
= {io + compute}
= {io} + {compute}
= {compute}
= {io}
0
1
2
3
4
0 2 4 6 8 10
0%
50%
100%
150%
200%
x (Note: u = 2^x)
R e
= {io + compute}
= {io} + {compute}
= {compute}
= {io}
0
2
4
6
8
0 2 4 6 8 10
0%
50%
100%
150%
200%
R e
x (Note: u = 2^x)
= {io + compute}
= {io} + {compute}
= {compute}
= {io}
0
1
2
3
4
0 2 4 6 8 10
0%
50%
100%
150%
200%
x (Note: u = 2^x)
R e
= {io + compute}
= {io} + {compute}
= {compute}
= {io}
0
1
2
3
4
0 2 4 6 8 10
0%
50%
100%
150%
200%
x (Note: u = 2^x)
R e
= {io + compute}
= {io} + {compute}
= {compute}
= {io}
Vol.3, No.7, 2012

7 | P a g e
TABLE V. FINE-GRAIN TERMINAL RUN, U=1,024, BOND PORTFOLIOS
W3540 i7 i3
TN R TN R TN R
10.42 6.21 17.53 2.72 22.24 3.13
12.29 4.99 17.00 2.62 23.98 2.78
9.04 5.79 13.81 2.78 18.58 2.98
3.58 2.45 3.06 1.90 5.30 2.17
C. Coarse-grain results
The results for the coarse-grain algorithms are summarized
in the next three graphs, one for the W3540, i7, and i3,
respectively. Note that the algorithms are not defined for
portfolios less than the number of hyper-threads.

Figure 10. W3540 coarse-grain results

Figure 11. i7 coarse-grain results

Figure 12. i3 coarse-grain results
The table below gives results for the terminal run.
TABLE VI. COARSE-GRAIN TERMINAL RUN, U=1,024, BOND PORTFOLIOS
W3540 i7 i3
TN R TN R TN R
14.18 4.56 18.01 2.64 22.73 3.06
12.07 5.08 16.84 2.65 23.50 2.83
8.44 6.20 16.84 2.28 18.28 3.03
3.61 2.43 3.03 1.92 5.16 2.23
D. Statistical significance results
The table below gives the counts in which an algorithm is
statistically significantly faster than another algorithm and
kind. The details underlying this table are in the appendix,
Sorted Rank Sums.
TABLE VII. STATISTICALLY SIGNIFICANTLY COUNTS MEASURED BY
FASTER RUNTIMES
Kind Algorithm W3540 i7 i3 Totals
Serial 0 0 0 0
Naive 2 2 2 6
Fine 4 3 2 9
Coarse 2 2 2 6
Serial 1 1 1 3
Naive 3 2 2 7
Fine 2 2 6 10
Coarse 2 2 6 10
Totals 16 14 21

To read the above table, choose the kind of algorithm
(composite vs. memory-bound) and read across for type of
algorithm. For example, the composite serial algorithm ran
slower than every other algorithm on the W3540, i7, or i3
platforms.

0
2
4
6
8
0 2 4 6 8 10
0%
50%
100%
150%
200%
x (Note: u = 2^x)
R e
= {io + compute}
= {io} + {compute}
= {compute}
= {io}
0
1
2
3
4
0 2 4 6 8 10
0%
50%
100%
150%
200%
x (Note: u = 2^x)
R e
= {io + compute}
= {io} + {compute}
= {compute}
= {io}
0
1
2
3
4
0 2 4 6 8 10
0%
50%
100%
150%
200%
x (Note: u = 2^x)
R e
= {io + compute}
= {io} + {compute}
= {compute}
= {io}
Vol.3, No.7, 2012

8 | P a g e
Hence, there are zero (0) values across the composite serial
row. The memory-bound naive algorithm ran faster than three
algorithms on the W3540 and two algorithms the i7 and i3,
respectively. The memory-bound serial algorithm
outperformed one algorithm on each platform: these slower
algorithms were the composite serial algorithms. Evidently
loading on all the portfolios into memory significantly
improves even the serial performance.
See the appendix, Sorted Rank Sums for the specific
counts.
V. DISCUSSION
The graphs, Figures 4 11, show that for larger problem
sizes, u, the composite and memory-bound algorithms
performed better than I/O processing alone which is the least
efficient but worse than compute by itself which is the most
efficient. The slopes of these graphs generally point toward
increasing speedup and efficiency for larger u.
Tables IV VI show evidence for high levels of overlap
between compute and I/O. For instance, the ratios of
T{compute} /T {io + compute} and T{compute} /
(T{io}+T{compute}) found in these tables are often around
80% or higher.
Table VII nevertheless indicates that the memory-bound
algorithms tend generally to give statistically significant better
runtimes compared to the composite algorithms. In other
words, caching the portfolios in memory upfront seems to give
better performance than loading them, as they are needed.
Table VII also suggests that the algorithms on a given
platform tend to run with significantly more efficiency on the
i3 across all the algorithms, followed respectively by the
W3540 and the i7.
Finally, the data in Table VI show the fine-grain algorithms
give statistically significantly better runtimes followed
respectively by coarse-grain and the nave algorithms across
different platforms.
VI. CONCLUSIONS AND FUTURE DIRECTIONS
This study has found that bond portfolio analysis using
parallel collections achieve super-linear speedup and super-
efficiency with as few as u=64 portfolios across different
multicore processors. The data suggests that the nave
application of parallel collections can be improved
significantly, foremost with the fine-grain algorithm, which we
find interesting. That is, portfolio analysis is embarrassingly
parallel, but not for the fine-grain or the I/O parallel
algorithms which contain inherent dependencies that
necessitated the use of parallel merge-trees.
The data points toward greater speed up and efficiency for
larger problem sizes, u>1024. The terminal run analyzed only
about 1% of the portfolios. Additional research could consider
how to harness multiple hosts and/or GPUs to price all
portfolios.
Future work might also compare and contrast map-reduce
versus parallel collections as well as possibly consider how to
improve the I/O performance.
ACKNOWLEDGEMENTS
This research has been funded in part by grants from the
National Science Foundation, Academic Research
Infrastructure award number 0963365 and Major Research
Instrumentation award number 1125520.
REFERENCES
[1] Patterson, J., and Hennessy, D., Computer Architecture, Morgan
Kaufman, 2006
[2] Hill, M., Marty, M., Amdahls Law in the Multicore Era, IEEE
Computer Society, Vol. 41, Issue 7, 2008
[3] Hager, G., and Wellein, G.,, Introduction to High Performance
Computing for Scientists and Engineers, CRC, 2010
[4] Michaelson, G., Introduction to Functional Programming through Lamda
Calculus, Dover, 2011
[5] McKenney, P. Is Parallel Programming Hard, And, If So, What Can You
Do About It?, 2011,
http://kernel.org/pub/linux/kernel/people/paulmck/perfbook/perfbook.ht
ml, accessed: 8 April, 2012
[6] Sutter, H., The Free Lunch is Over: The Fundamental Turn Toward
Concurrency in Software, Dr. Dobbs Journal, vol. 30, no. 3, 2005
[7] Odersky, M., et al., Programming in Scala, Artima, Mountain View,
2011
[8] Owens, J., et al., GPU Computing, Proceedings of the IEEE, Vol. 96,
No. 5, May 2008
[9] Nystrom, N., et al., Filepile: Run-time Compilation for GPUs in Scala,
GPCE 11. October 22-23, 2011, Portland, OR., U.S., 2011
[10] Das, K., GPU Parallel Collections for Scala, M.S. Thesis, University of
Texas Arlington, 2011
[11] Prokopec, A., et al., A Generic Parallel Collection Framework, EPFL,
InfoScience 2011, 2010-7-31, 2011
[12] Lester, B., Data parallel programming in Scala. Scala Days 2010, EPFL,
Lausanne, Switzerland, 15 - 16 April 2010.
[13] Coleman, R., et al, Computational Finance with Map-Reduce in Scala,
Conference on Parallel and Distributed Processing (PDPTA 12),
CSREA, 2012
[14] Dean, J., and Ghemawat, S., Simplified Data Processing on Large
Clusters, OSDI, 2004
[15] Tsang, E. and Martinez-Jaramillio, Computational Finance, IEEE
Computational Intelligence Society Newsletter, August 2004, 3-8, 2004
[16] Fabozzi, F., and Mann, S., Introduction to Fixed Income Analytics, 2
nd

ed., Wiley, 2010
[17] Hull, J. Options, Futures, and Other Derivatives and DerivaGem CD
Package, 8
th
ed. Prentice Hall, 2011
[18] Chodorow, K. and Dirolf, M., MongoDB: The Definitive Guide,
ORielly, 2010
[19] Merunka, V., et al., Normalization Rules of the Object-Oriented Data
Model, EOMAS '09 Proceedings of the International Workshop on
Enterprises & Organizational Modeling and Simulation, 2009
[20] Georges, A., et al., Statistically Rigorous Java Performance Evaluation,
OOPSLA 07, October 21-25, Montreal, Quebec, Canada, 2007
[21] Intel Corp., Intel Xenon Processor W3540, 2009,
http://ark.intel.com/products/39719/Intel-Xeon-Processor-W3540-(8M-
Cache-2_93-GHz-4_80-GTs-Intel-QPI) accessed: 17-April-2012
[22] Intel Corp., Core i3-370M Processor,
http://ark.intel.com/products/49020/Intel-Core-i3-370M-Processor-(3M-
cache-2_40-GHz), 2010, accessed: 17 April 2012
[23] Intel Corp. (2011). Core i7-2670QM Processor, 2011,
http://ark.intel.com/products/53469, accessed: 17-April-2012
[24] Conover, J., Practical Non-Parametric Statistics, Wiley, 1999
APPENDIX -- SORTED RANK SUMS
The three tables below give the sorted rank sums of
runtimes according to Equation 12 for the terminal (u=1,024)
Vol.3, No.7, 2012

9 | P a g e
run. Algorithm A which has the smaller median runtime
compared to algorithm B. Smaller rank sums (S) imply greater
statistical significance. Since there are 11 trials for each
algorithm, the minimum rank sum is S=1+2+311=66 (i.e., all
runtimes of algorithm A are less than the runtimes of algorithm
B). In this case, p < 0.001. The threshold for statistical
significance is S<101 in which p s 0.05. Comparisons that are
not statistically significant are not included in the tables and by
implication tables with more rows imply cores with greater
performance.
TABLE VIII. W3540 RANK SUMS (S) ALGORITHM A(KIND) B(KIND)
S Algo A Kind Algo B Kind
66 Naive composite Serial composite
66 Naive mem-bound Serial composite
66 Naive composite Serial mem-bound
66 Naive mem-bound Serial mem-bound
66 Fine composite Serial composite
66 Fine mem-bound Serial composite
66 Fine composite Serial mem-bound
66 Fine mem-bound Serial mem-bound
66 Coarse composite Serial composite
66 Coarse mem-bound Serial composite
66 Coarse composite Serial mem-bound
66 Coarse mem-bound Serial mem-bound
89 Fine composite Naive composite
96 Serial mem-bound Serial composite
98 Fine composite Coarse composite
100 Naive mem-bound Naive composite
TABLE IX. I7 RANK SUMS (S) ALGORITHM A(KIND) B(KIND)
TABLE X. RANK SUMS (S) ALGORITHM A(KIND) B(KIND)
66 Coarse composite Naive mem-bound
69 Coarse composite Fine mem-bound
71 Fine composite Naive mem-bound
75 Fine composite Fine mem-bound
79 Fine composite Coarse mem-bound
85 Coarse composite Coarse mem-bound
100 Coarse composite Naive composite

APPENDIX -- SOURCE CODE
All the source code used for this project is freely available
via the Scaly project home and downloadable as an Eclipse
project at http://code.google.com/p/scaly/. See the ParaBond
folder and the package, scaly.parabond.test. The table below
gives the algorithm and its source.
TABLE XI. SOURCE FILES
Algorithm Kind Scala source file
Serial Composite NPortfolio02
Memory-bound NPortfolio03
Naive Composite Par00
Memory-bound Par01
Fine Composite Par05
Memory-bound Par06
Coarse Composite Par04
Memory-bound Par07

Vol.3, No.7, 2012
10 | P a g e
Core Backbone Convergence Mechanisms and
Microloops Analysis

Abdelali Ala
Abdelmalik Essadi University
Faculty of Sciences
Information and Telecom Systems Lab
Tetuan, Morocco
+212 6 65 24 08 28
Driss El Ouadghiri
Moulay Ismail University
Faculty of Sciences
Mathematics and Science Computer
Meknes, Morocco
+212 6 02 23 61 55
Mohamed Essaaidi
Abdelmalik Essadi University
Faculty of Sciences
Information and Telecom Systems Lab
Tetuan, Morocco
+212 6 61 72 59 92

Abstract In this article we study approaches that can be used to
minimise the convergence time, we also make a focus on
microloops phenomenon, analysis and means to mitigate them.
The convergence time reflects the time required by a network to
react to a failure of a link or a router failure itself. When all
nodes (routers) have updated their respective routing and
forwarding databases, we can say the network has converged.
This study will help in building real-time and resilient network
infrastructure, the goal is to make any evenement in the core
network, as transparent as possible to any sensitive and real-time
flows. This study is also, a deepening of earlier works presented
in [10] and [11].
Keywords-component: FC(Fast-convergence); RSVP(ressource
reservation protocol); LDP (Label Distribution Protocol);
VPN(Virtual Private Network); LFA (loop free alternate); MPLS
(Multiprotocol Label Switching); PIC(Protocol independent
convergence); PE(Provider edge router); P(Provider core router ).
I. INTRODUCTION
Mpls/vpn backbones are widely used today by various
operators and private companies in the world, high to medium-
sized companies build their own Mpls/vpn backbone or use
services of an operator . Real time applications like voice and
video are more and more integrated to end user applications,
making them ever more time sensitive.
Operators are offering services like hosting companies
voice platforms, VoIP call centers, iptv...Etc. All these aspects
make the convergence time inside the backbone a challenge for
service providers.
However, the global convergence time is an assembly of
several factors including: link or node failure detection, IGP
failure detection, LSP Generation, SPT Computation, RIB
update, local FIB creation and distribution ...updates
signaling...etc.
Based on analysis and statistics of large backbone
possibilities we have delimited our convergence target as
follows:
[PE to P] convergence, in other terms [PE to core] must be
under sub-second, hopefully under 50 msec, even on highly
loaded PE, the convergence time should be almost independent
of vpnv4, 6PE, 6VPE or igp prefixes number[P to PE] and
[P to P] convergence must stay under sub-second and
consistent in both directions: [core to PE], [PE to core].
From the customer point of view: the overall [end-to-end]
convergence should stay under 1 sec (no impact on most time
sensitive applications). A lot of approaches can be used to
minimise the convergence time, our approach consists on
enhancements and optimizations in control and forwarding
plane. While a lot of things can also be made at the access, the
scope of our work is the core backbone.
Not only a backbone design must take into account
criterion like redundant paths at each stage, but redundancy at
the control plane only, does not make a lot of sense if, in the
forwarding plane, backup paths are not pre-computed. We can
say that a backbone meets a good convergence design if at each
segment of the tree structure; we are able to calculate the time
it takes for flows to change from the nominal path to the
backup one.
On the other hand, temporary microloops may occur during
the convergence interval, indeed, after a link or node failure in
a routed network and until the network re-converges on the
new topology, routers several hops away from the failure, may
form temporary microloops. This is due to the fact that a
router's new best path may be through a neighbor that used the
first router as the best path before failure, and haven't had yet a
chance to recalculate and/or install new routes through its
new downstream. We can understand microloops are transient
and self-corrected, however depending on their duration, the
CPU load on the control plan may increase to 100%, so in
addition to mitigation methods presented in this article, some
cpu protection mechanisms are also discussed. The approach
used in this article is theory against lab stress and result
analysis. The aim of the study is to give an accurate idea of
gains and drawbacks of each method, and show when one or
the other method more fits the network topology.
II. FAST CONVERGENCE MODELS
In an attempt to construct a model for IGP and BGP
protocols, we must take into account the following
components:
Vol.3, No.7, 2012
11 | P a g e
Time to detect the network failure, e.g. interface down
condition.
Time to propagate the event, i.e. flood the LSA across the
topology.
Time to perform SPF calculations on all routers upon
reception of the new information.
Time to update the forwarding tables for all routers in the
area.
And then modelise the IGP Fast Convergence by a formula
which is the sum of all the above components:
IFCT = (LFD + LSP-GIF + SPTC + RU + DD)
And BGP Fast Convergence model as:
BFCT= IFCT + CRR
Where:
IFCT = IGP Fast Convergence Time
LFD = Link Failure Detection (Layer 1 detection
mechanisms)
LSP-GIF = LSP Generation, Interval and Lifetime
SPTC = SPT Computation
RU = RIB Update
DD = Distribution Delay
BFCT = BGP Fast Convergence Time
CRR = CEF Recursive Resolution for BGP Prefixes
III. LINK FAILURE DETECTION MECHANISM
The ability to detect that a failure has happened is the first
step to towards providing recovery, and therefore, is an
essential building block for providing traffic protection. Some
transmission media provide hard-ware indications of
connectivity loss. One example is packet-over-SONET/SDH
where a break in the link is detected within milliseconds at the
physical layer. Other transmission media do not have this
ability, e.g. Ethernet (note that the fast detection capability has
been added to optical Ethernet).
When failure detection is not provided in the hardware, this
task can be accomplished by an entity at a higher layer in the
network. But there is disadvantage to that, using IGP hello as
example: We know that IGPs send periodic hello packets to
ensure connectivity to their neighbors. When the hello packets
stop arriving, a failure is assumed. There is two reasons why
hello-based failure detection using IGP hellos cannot provide
fast detection times:
The architectural limit of IGP hello-based failure detection
is 3 seconds for OSPF and 1 second for ISIS. In common
configurations, the detection time ranges from 5 to 40
seconds.
Since handling IGP hellos is relatively complex, raising
the frequency of the hellos places a considerable burden on
the CPU.
IV. BIDIRECTIONAL FORWARDING DETECTION (BFD)
The heart of the matter lies in the lack of a hello protocol to
detect the failure at a lower layer. To resolve this problem,
Cisco and Juniper jointly developed the BFD protocol. Today
BFD has its own working group (with the same name IETF
[BFD]). So what exactly is BFD ?
BFD is a simple hello protocol designed to provide rapid
failure detection for all media types, encapsulations,
topologies, and routing protocols. It started out as a simple
mechanism intended to be used on Ethernet links, but has since
found numerous applications. Its goal is to provide a low-
overhead mechanism that can quickly detect faults in the
bidirectional path between two forwarding engines, wether
they are due to problems with the physical interfaces, with the
forwarding engines themselves or with any other component.
But how can BFD quickly detect such a fault ?
In a nutshell, BFD is exchanging control packet between
two forwarding engines. If a BFD device fails to receive a BFD
control packet within the detect-timer:
(Required Minimum RX Interval) * (Detect multiplier)
Then it informs its client that a failure has occurred. Each
time a BFD successfully receives a BFD control packet on a
BFD session, the detect-timer for that session is reset to zero.
Thus, the failure detection is dependent upon received packets,
and is independent of the receiver last transmitted packet. So
we can say that expected results depend on the platform and
how the protocol is implemented, but available early
implementations can provide detections in the range of tens of
milliseconds.
V. MPLS LDP-IGP SYNCHRONIZATION
A. FEATURE DESCRIPTION
Packet loss can occur when the actions of the IGP (e.g.
ISIS) and LDP are not synchronized. It can occur in the
following situations:
When an IGP adjacency is established, the router begins
forwarding packets using the new adjacency before the
LDP label exchange ends between the peers on that link.
If an LDP session closes, the router continues to forward
traffic using the link associated with the LDP peer rather than
an alternate pathway with a fully synchronized LDP session.
To solve the first point, the following algorithm is being
used: If there is a route to the LDP peer, IGP adjacency is held
down, waiting for LDP synchronization to be completed; in
other words, waiting for labels exchange to be completed. By
default, adjacency will stay down for ever if LDP does not
synchronize. This default behavior is tunable via configuration
command mpls ldp igp sync hold-down <duration in ms> to
specify the maximum amount of time the adjacency will stay
down. At expiration of this timer, the link will be advertised,
but with metric set to maximum in order to avoid using this
link. If there is no route to the LDP peer, IGP adjacency is
brought up, but with a metric set to the maximum value in
order to give a chance for the LDP session to go up. In this
Vol.3, No.7, 2012
12 | P a g e
case, once the LDP session goes up and finishes labels
exchange, the IGP metric reverts back to its configured value.
To solve the second point, the feature will interact with IGP
to modify link metric according to LDP session state. As soon
as LDP session is going down, the IGP metric of the related
link is set to its maximum. Then, others nodes on the network
can compute a new path avoiding to use this link.

Figure 1. Lab setup diagram
B. TEST DESCRIPTION
On the M1 router, we configure the ldp-synchronization
under isis protocol, interface xe-0/2/0.0 (timer set to T1 sec)
and under ldp protocol: (timer set to T2 sec). The timer under
the ISIS section will set how much time ISIS will stay sending
the infinite metric once it has been warned by LDP that its
sessions are up. The timer under LDP section will set how
much time LDP wait to warn the IGP once its sessions are up;
by default this timer is equal to 10 sec.
M1-RE0# run show configuration protocols isis
traceoptions {
file isis size 5m world-readable;
flag ldp-synchronization send receive detail;
flag lsp-generation detail;
------------truncated---------------------------------------------
interface xe-0/2/0.0 {
ldp-synchronization {
hold-time T1;}
point-to-point;
level 2 metric 10;
}
ldp-synchronization {
hold-time T1; }
point-to-point;
level 2 metric 100;

M1-RE0>show configuration protocol ldp
track-igp-metric;
------------truncated------------------------------------------
igp-synchronization holddown-interval T2;

M1-RE0>show configuration interfaces xe-0/2/0
description "10 GIGA_LINK_TO_PPASS_P71 through Catalyst
TenGigabitEthernet2/5";
vlan-tagging;
mtu 4488;
hold-time up 5000 down 0; / time here is in milliseconds /
While isis adjacency is operational, the ldp session is turned
down (deactivation of xe-0/2/0.0 under ldp protocol on the MX
side).
We look at the debug file on the MX and the isis lsp
received on PE12 rising to infinite the isis metric toward
RNET-A71.
PE-10K#show isis database M1-RE0.00-00 detail
S-IS Level-2 LSP M1-RE0.00-00
LSPID LSP Seq Num LSP Checksum LSP Holdtime
ATT/P/OL
M1-RE0.00-00 0x00000B71 0x7FFE 65520 0/0/0
Area Address: 49.0001
NLPID: 0xCC 0x8E
Router ID: 10.100.2.73
IP Address: 10.100.2.73
Hostname: M1-RE0
Metric: 16777214 IS-Extended RNET-A71.00
Metric: 100 IP 10.0.79.56/30
Metric: 10 IP 10.0.79.52/30
After the expiration of (the configured hold-down timer) we
can see that the metric is updated and set to the initial value.
PE-10K#show isis database M1-RE0.00-00 detail
IS-IS Level-2 LSP M1-RE0.00-00
LSPID LSP Seq Num LSP Checksum LSP Holdtime
ATT/P/OL
M1-RE0.00-00 0x00000B72 0x8FE2 65491 0/0/0
Area Address: 49.0001
NLPID: 0xCC 0x8E
Router ID: 10.100.2.73
IP Address: 10.100.2.73
Hostname: M1-RE0
The duration of the infinite metric must cover the necessary
time for a full labels exchange after the rising of the ldp
session.

Vol.3, No.7, 2012
13 | P a g e
Figure 2. ldp-igp synchronization chronogram
VI. ISIS BACKOFF ALGORITHM
A. TUNING EXPLAINED
ISIS runs a Dijkstra-algorithm to compute the tree followed
by a computation of the routing table. If the receipt of a
modified LSP does affect the tree, an SPF (shortest path first
calculation) is run; otherwise a simple PRC (partial route
calculation) is run. An example of evenement that will trigger
only a PRC is the addition of a loopback on a distant node (this
does not change the tree, just one more IP prefix leaf is on the
tree)
The PRC process runs much faster than an SPF because the
whole tree does not need to be computed and most of the
leaves are not affected.
However, by default, when a router receives an LSP which
is triggering an SPF or a PRC, it does not start it immediately,
it is waiting for a certain amount of time (5.5 seconds for SPF
& 2 seconds for PRC).Lowering this initial wait time would
significantly decrease the needed convergence time.
On the other hand, it is necessary to leave enough time to
the router to receive all LSPs needed for computing the right
SPF, so there is a lower limit not to be exceeded. Otherwise, If
SPF computation starts before having received all important
LSP, you may need to run another SPF computation a bit later.
Then, overall convergence would not be optimal.
Between the first SPF (or PRC) and followings ones, the
router will also wait for some times, default values are (5.5
seconds for SPF and 5 seconds for PRC). However the
maximum amount of time a router can wait is also limited
(10 seconds for SPF and 5 seconds for PRC).
B. FEATURE USAGE IN OUR STUDY
The worst case, to take into consideration while choosing
the initial wait time, is a node failure. In this situation, all
neighbors of the failing node will send LSP reporting the
problem. These LSP will be flooded through the whole
network. Some studies indicate that 100 ms is enough for very
large and wide networks.
So here our chosen values:
spf-interval 1 150 150
prc-interval 1 150 150
spf-interval <M> <I> <E>
prc-interval <M> <I> <E>
M = (maximum) [s]
I = (initial wait) [ms]
E = (Exponential Increment) [ms]
The same parameters have been applied on all routers to
keep a consistency and same behavior on all nodes.

Figure 3. isis backoff algorithm timing
150 ms as initial waiting time for the first SPF calculation,
then if there is a trigger for another SPF, the router will wait
300 ms, then wait 600 ms if there is a following one, until the
max-value of 1000 ms. the waiting timer will stay equal to 1
second for as much as there is no trigger of a new calculation.
In case there is no trigger during 1 second, the wait time is
reset to the initial value and start as described in the Fig. 3.
C. MAIN GAIN FROM THIS TUNING
Simulations indicate that the most important gain is due to
the first waiting timer decreased from default value to 150ms.
VII. BGP-4 SCALABILITY ISSUES (PROBLEM STATEMENT)
The BGP-4 routing protocol has some scalability issues
related to the design of Internal BGP (IBGP) and External BGP
(EBGP) peering arrangements.
T0

Max time
limit
T1 = 2*T0 T3 = 2*T1
LDP
ISIS
event
t0
t1
t2
t3
t4
time
no shut
Interface holdtime when up
Ldp hold-down : T2 s
Duration of
infinite
metric
T2 s + T1 s (10 default )
Vol.3, No.7, 2012
14 | P a g e
IBGP and EBGP are the basically the same routing protocol
just with different rules and applications.
EBGP advertises everything to everyone by default.
IBGP does not advertise 3rd-party routes to other IBGP
peers, this is because there is no way to do loop detection
with IBGP
The RFC 4456 states that any BGP-4 router with EBGP
peers must be fully meshed with all the other BGP-4 routers
with EBGP peers in the same AS. This rule effectively means
that every IBGP peers must be logically fully meshed. So you
must have all BGP-speaking routers in your AS peer with each
other. Below is a graphical example of a full-meshed 16-router
. For more details see [15].

Figure 4. Example of full-meshed 16-IBGP routers

Figure 5. Example of Route reflectors cluster
There are resource constraints when you scale a network to
many routers, globally, if we have: n BGP speakers within an
AS, that requires to maintain: [n*(n-1)/2] BGP session per
router. Another alternative in alleviating the need for a "full-
mesh" is to use of Route Reflectors the Fig. 5 above .
They provide a method to reduce IBGP mesh by creating a
concentration router to act as a focal point for IBGP sessions.
The concentration router is called a Route Reflector Server.
Routers called Route Reflector Clients have to peer with the
RR Server to exchange routing information between
themselves. The Route Reflector Server reflects the routes to
its clients.
It is possible to arrange a hierarchical structure of these
Servers and Clients and group them into what is known as
clusters. Below is a diagram that illustrates this concept.
VIII. ROUTE-REFLECTORS IMPACT ON THE CONVERGENCE
If we estimate the typical total number of customers vpn
routes transported inside an operator backbone to be something
like 800 000 routes, each Route reflector have to learn, process
the BGP decision algorithm to choose best routes, readvertise
best ones, while maintaining peering relationships with all its
client routers, the route-reflector CPU and memory get
certainly consumed, and as a consequence, slows down route
propagation and global convergence time.
A. TEST METHODOLOGY
The methodology we use to track this issue is to preload the
route reflector by using a simulator acting as client routers (or
PE routers), and then, nearly simultaneously, we clear all
sessions on the route-reflector, then start the simulated
sessions. Then we monitor convergence by issuing 'sh ip bgp
vpnv4 all sum' commands while recording every 5 seconds all
watched parameters (memory and CPU utilization for various
processes).
When all queues are empty and table versions are
synchronized, we consider the router has converged, (finished
updating all its clients by all routes it knows). All these tests
are performed several times to ensure they are reproducible.
Results could slightly differ but accuracy is kept within 5%.
The goal is to find a tolerated convergence time for route
reflectors, then we must limit the number of peering and
number of routes per peering to respect the fixed threshold.
IX. BGP CONSTRAINED ROUTE DISTRIBUTION
By default within a given iBGP mesh, route-reflectors will
advertise all vpn routes they have to their clients (PE routers),
then PE routers use Route Target (RT) extended communities
to control the distribution of routes into their own VRFs (vpn
routing and forwarding instances).
However PE routers need only hold routes marked with
Route Targets pertaining to VRFs that have local CE
attachments.
To achieve this, there must be an ability to propagate route
target membership information between iBGP meshes and the
most simple way is to use bgp update messages, so that Route
Target membership NLRI is advertised in BGP UPDATE
messages using the MP_REACH_NLRI and
MP_UNREACH_NLRI attributes. The [AFI, SAFI] value pair
used to identify this NLRI is (AFI=1, SAFI=132).
Vol.3, No.7, 2012
15 | P a g e
As soon as route-reflectors Receive Route Target
membership information they can use it to restrict
advertisement of VPN NLRI to peers that have advertised their
respective Route Targets.
B. MAIN FINDINGS OF OUR STUDY
When we use Route-Target-constraints, The PEs receive
considerably less routes. But, because in an operator backbone
VRFs are spread everywhere geographically, they touch almost
all route-reflectors, therefore:
Route-Target-constraints does not help reducing the
number of routes handled by route reflectors.
The only gain is that, instead of each RR sending its entire
table, it's going to prefilter it before it send it to each of its PEs,
which means less data to send, and less data to send, means
being able to send faster, provided that there is no cpu cost due
to pre-filtering on the route-reflectors side.
X. BGP FAST CONVERGENCE MECHANISMS
A. BGP NEXT HOP TRACKING
By default within a given iBGP mesh, route-reflectors will
advertise all vpn routes they have to their clients (PE routers),
then PE routers use Route Target (RT) extended communities
to control the distribution of routes into their own VRFs (vpn
routing and forwarding instances).
XI. BGP PREFIX INDEPENDENT CONVERGENCE (PIC)
It provides the ability to converge BGP routes within sub-
seconds instead of multiple seconds. The Forwarding
Information Base (FIB) is updated independently of a prefix to
converge multiple numbers of BGP routes with the occurrence
of a single failure. This convergence is applicable to both core
and edge failures and with or without MPLS.
A. SETUP DESCRIPTION
Let us consider the test setup in Fig. 6. The simulator is
injecting M and N vpn routes respectively from PE2 and PE3,
PE2 end PE3 advertise injected routes respectively to route-
reflector RR1 and RR2, PE1 imports the M and N VPN routes,
each vpn prefixes uses as bgp next-hop either the IGP loopback
of PE2 or PE3. The simulator attached to PE1 generates traffic
toward those learned routes, we locate the best path chosen by
PE1 in the its forwarding table, then we cut the corresponding
interface. Numbers M and N are increased progressively (by
hundreds of thousands prefixes to make the impact more
visible).
First phase: interface 0 fails down. It is detected and all FIB
entries with this interface are deleted.
Second phase: IGP convergence occurs and new output
interface is set to interface 1 for all VPN prefixes, hence a
traffic disruption.

P
P
P
PE1
SUT
Route-
reflector 1
PE2
PE3
FIB Update:
Linear dependency ( if no PIC)
Or independency ( with PIC )
Route-
reflector 2
In
te
rfa
c
e
0
In
te
rfa
c
e
1

Figure 6. Lab setup diagram
B. FEATURE DESCRIPTION
VPN Prefix 1 IGP NH 1 via if0
VPN Prefix N IGP NH 1 via if0

Figure 7. Forwarding table, rewriting of indexation toward interface 0
VPN Prefix N IGP NH 1 via if1
FIB entries
Are rewritten
sequentially

Figure 8. Forwarding table, rewriting of indexation toward interface 1

Vol.3, No.7, 2012
16 | P a g e
Third phase: all VPN prefixes attached to the NH1 are
rewritten in the FIB with the new interface if1.
LoC = (IGP convergence) + (N * FIB Rewriting time)
Let us now analyze the behavior (with PIC feature): An intermediate
Next-hop (called loadinfo) is created, and the content of the
forwarding table modified as described below:
VPN Prefix 1
VPN Prefix 2 IGP NH 1 loadinfo
VPN Prefix N
if0
A loadinfo is
associated to
One egress
PE

Figure 9. Forwarding table, structure modified when using the feature
First phase: if0 fails down. It is immediately erased but the
loadinfo structure is not:
VPN Prefix 1
VPN Prefix 2 IGP NH 1 loadinfo
VPN Prefix N
if1
A loadinfo is
associated to
One egress
PE

Figure 10. Forwarding table, deletion and rewriting concerns only one Next-
hop
Second Phase: IGP convergence occurs and as soon as the
new path via if1 is deduced, loadinfo is updated.
LoC = IGP convergence only
XII. LOOP FREE ALTERNATE (LFA)/IPFRR
This feature describes such a mechanism that allows a
router whose local link has failed to forward traffic to a pre-
computed alternate path. The alternate path stays used until the
router installs the new primary next-hops based upon the
changed network topology.
When a local link fails, a router currently must signal the
event to its neighbors via the IGP, recompute a new primary
next-hop for all affected prefixes, and only then install those
new primary next-hops into the forwarding plane. Until the
new primary next-hops are installed, traffic directed towards
the affected prefixes is discarded. This process can take
hundreds of milliseconds. The goal of IP Fast Reroute (IPFRR)
is to reduce failure reaction time to 10s of milliseconds by
using a pre-computed alternate next-hop in the event that the
currently selected primary next-hop fails, so that, the alternate
can be rapidly used when the failure is detected. A network
with this feature experiences less traffic loss and less micro-
looping of packets than a network without IPFRR. There are
cases where traffic loss is still a possibility since IPFRR
coverage varies, but in the worst possible situation a network
with IPFRR is equivalent with respect to traffic convergence to
a network without IPFRR. [2].
B. CONFIGURING THE FEATURE
A loop-free path is one that does not forward traffic back
through the router to reach a given destination. That is, a
neighbor whose shortest path to the destination traverses the
router is not used as a backup route to that destination. To
determine loop-free alternate paths for IS-IS routes, a shortest-
path-first (SPF) calculation is run on each one-hop neighbor.
M1-RE1> show configuration protocols isis
traceoptions {
file ISIS_DEB1;
flag lsp;
}
lsp-lifetime 65535;
overload;
level 2 {
authentication-key "$9$2P4JDjHm5z3UD69CA0O"; ## SECRET-DATA
authentication-type simple;
no-hello-authentication;
no-psnp-authentication;
wide-metrics-only;
}
point-to-point;
link-protection;
level 2 metric 100;
}
point-to-point;
link-protection;
level 2 metric 10;
As a consequence the backup path through Rnet-A71 is
precomputed and installed on the the forwarding table
M1-RE1>show route forwarding-table table CUST-VRF-AGILENT_PE_10
destination 1.0.0.1/32 extensive
Routing table: CUST-VRF-AGILENT_PE_10.inet [Index 5]
Internet:

Destination: 1.0.0.1/32
Route type: user
Vol.3, No.7, 2012
17 | P a g e
Route reference: 0 Route interface-index: 0
Flags: sent to PFE
Nexthop:
Next-hop type: composite Index: 7094 Reference: 2
Next-hop type: indirect Index: 1048581 Reference: 50001
Next-hop type: unilist Index: 1050156 Reference: 2
Nexthop: 10.0.79.57
Next-hop type: Push 129419 Index: 502443 Reference: 1
Next-hop interface: xe-0/3/0.0 Weight: 0x1
Nexthop: 10.0.79.53
Next-hop type: Push 127258 Index: 7093 Reference: 1
Next-hop interface: xe-0/2/0.0 Weight: 0x4000 - alternate path
See Fig. 1 for lab setup
C. TEST CONDITIONS
From the lab setup described above, we announce 500000
routes, by 50k routes per vrf (vpn routing instances) from 10
different PE. The M1 receives the 50k routes in 10 different
routing-instances, by 50K for each.
From the Simulator (an Agilent chassis) connected to the
M1 we generate traffic consisting of 500K packets sized to 64
bytes:
This flow use as a source an ip address varying randomly
within the interval [ 10.0.9x.1/32 to 10.0.9x.254/32] while
x=1 for vrf 1, 2 for vrf 2 etc until N for vrf N.
This flow use as a destination an address varying
sequentially within the interval [ x.0.0.1/32 to
x.0.195.80/32] while x=1 for vrf 1, 2 for vrf 2 etc until N
for vrf N.
We Chose isis metrics on the setup to make the Rnet-A72
the best IGP link, we shut this best link and observe the
behavior of traffic curve as received on the Simulator
connected to PE12

Figure 11. curve of vpn traffic with LFA
The Bleu curve is the forwarded traffic on the nominal link,
the grey curve is the forwarded traffic on the backup link. The
backup link have been mirrored to a free port and connected to
the simulator to see the apparition and the disappearing of
traffic on it.
As a comparison you can look at the traffic curve without
the feature, it resembles to the diagram on the Fig. 12. You
can notice the duration of Next-hops rewriting of vpn
prefixes toward the backup link in the forwarding table.

Figure 12. curve of vpn traffic without LFA

Figure 13. curve of vpn traffic with LFA, traffic retrieving on the nominal link
On the other hand, when we de-shut the best link, as in
Fig.13 we see that the traffic stays on the non-best link for
more than 80 seconds, before going back to the best.
XIII. LDPORSVP
The ldp over rsvp principle can be illustrated like in the
Fig. 14. Only core routers P1,P2 and P3 are enabling RSVP
TE, ldp however they are configured to prefer rsvp tunnels to
ldp ones.
The edge routers PE1 end PE2 are enabling only LDP with
P1 and P3.
PE1 end PE2 are VPN and use MP-iBGP to signal vpn
labels.
A. CONTROL PLAN ESTABLISHMENT
Let us consider PE2_FEC representing prefixes coming
from CE2.
1. Establish RSVP tunnel-1-3 from P1 to P3, the label
distributed to P2 from P3 is LR2, and the label
distributed from P2 to P1 is LR1
Shutdown of best link (bleu curve) , we see little negligible
Impact on outgoing traffic from the MX
Vol.3, No.7, 2012
18 | P a g e

Figure 14. LDP over RSVP principle
2. Establish a targeted ldp session between P1 and P3
3. Enable IGP shortcut on P1, the egress path for
PE2_FEC will be the tunnel-1-3.
4. PE2_FEC triggers the establishment of LSP on PE2,
and the label mapping message will be sent to P3, let
us consider this label is L2.
5. After P3 receives the label mapping message, it
forwards that message to P1 through the targeted
LDP session, let us consider this label is Lx
6. P1 receives the label mapping message, and finds out
that the egress fo the route is tunnel-1-3.Then the
LSP from PE1 to PE2 is transmitted I RSVP TE. The
external label is LR1.
7. P1 continues to send Label mapping message to PE1,
the label is L1.
8. PE1 generates Ingress
9. MP-BGP sends private network route of CE2 from
PE2 to PE1, the label of private network is Lb.
At this stage the establishment of LSP between PE1 and
PE2 is complete. This LSP traverses the RSVP TE area
( P1 ~~ P3).
B. FORWARDING PLANE PROCESS
The forwarding process of packets is as follows:
We describe here the forwarding process of data from CE1
to CE2, if needed do the symmetrical reasoning regarding
flows from CE2 to CE1:
1. After PE1 receives packets from CE1, it tags the
BGP label Lb of private network and then it tags
LDP label L1 of the provider network
2. (Lb,L1) label of PE1 is received on P1, replace L1
with Lx (the label sent to P1 through the targeted ldp
session, and then tag tunnel label LR1 of RSVP TE,
the label of packet becomes (Lb,Lx,Lr1).
3. From P2 to P3, with the RSVP TE transparently
transmitting packets, the LR1 is replaced by LR2,
that is, the packets received by P3 are tagged with the
following labels (Lb,Lx,LR2)
4. Upon arriving P3, the LR2 is first stripped and then
comes out Lx, and the label of LDP which is
replaced by L2.The packet is then sent to PE2 and the
label becomes (Lb,L2)
5. After the packet reaches PE2, L2 is first stripped and
then the Lb. After that, the packet is sent to CE2
C. LSP PROTECTION , ONE TO ONE BACKUP METHOD
Each P creates a detour (tunnel) for each LSP, the detour
will play the role of a protecting LSP :
If the router P2 fails, P1 switches received traffic from PE1,
along the detour tunnel [P1,P5] using the label received when
P1 created the detour .
The detour is calculated based on the shortest IGP path
from P1 to the router terminating the protected LSP, let us say:
PE2. In this case the protecting LSP will avoid the failed router
P2 (node protection).
At no point does the depth of the label stack increases as a
consequence of taking the detour.
While P1 is using the detour, traffic will take the path [PE1-
P1-P5-P6-P7-PE2]

Figure 15. LDP over RSVP backup method

Figure 16. LDPoRSVP labels stack during FRR
Vol.3, No.7, 2012
19 | P a g e

Figure 17. LDPoRSVP Lab setup
D. LDPORSVP LABEL STACK DURING FRR
Nota: when deploying LDPoRSVP and enabling FRR
(facility) as protection mechanism keep the 4 potential MPLS
labels into account for MTU definition
E. LAB SETUP AND TESTS SCOPE
Here are described the implementations made in our lab,
the CSPF (constrained shortest path first) was simplified to
only shortest igp:
Inter-P trafic will be encapsulated in a tunnel.
No impact on all PE configuration, Only P routers are
concerned by (LDPoRSVP).
The tunnel is a TLDP session, between each P, so full
mesh of: [n x P] routers.
Each TLDP session is using an LSP which is dynamic.
Signalling protocol for LSP is RSVP-TE , using cspf.
CSPF is a modified version of SPF algo(Dijkstra) , used in
ISIS.
CSPF algorithme finds a path which satisfy constraints for
the LSP (we simplify to only one constraint: the igp
shortest path).
Once a path is found by CSPF, RSVP uses the path to
request the LSP establishment.
F. LAB TEST METHOD :
On each P router, we check that a (detour LSP is
precalculated, presignaled for each LSP). We load heavily the
P routers with:
BGP vpn routes , internet routes
IGP (ISIS) routes
LDP labels
TLDP sessions
RSVP sessions
We generate traffic consisting of hundred thousands of
packets in both directions, PE1 to PE3 see (Fig.2), note that

Figure 18. Received packets curve
The grey curve represents recived packets, we notice a
small traffic fall.
TABLE 1 .LDPoRSVP Traffic measurement

In Fig. 18 the chosen igp metrics will force then nominal
path to be :[PE1-P1-P3-P4-PE3] (the red path). We cut the link
[P4P3] : either by shuting the physical port or by removing
the fiber from the port, we measure the convergence time
through the number of lost packets related to the ratio: (sent
/received) packets per second.
We check that, when the link [P4P3] goes down, the P3
router, instead of waiting the igp convergence, instantly uses
the precomputed backup link [P3-P1-P2-P4] (the green or
detour path), then after the igp converges, the traffic goe,
without impact, through the link [PE1-P1-P2-P4-PE3] (the blue
path).
Vol.3, No.7, 2012
20 | P a g e
We check fast reroute performance at different load
conditions: firstly we start with few LSPs then we increase the
number progressively: (500, 1000, 2000 )
G. TEST RESULTS:
We see that mainly: convergence time stays between 20
msec < t < 100 msec independently of number of LSPs. We
notice some issues regarding scalability of LDP FECs. The
on purpose studied case in the Fig.4 shows that during the
fast-reroute phase, traffic goes back to the sender before taking
the good (remaining) path. This topology case would exist in a
backbone design, so the sizing of the link must take into
account the potential and transcient traffic load.
XIV. LDP FASTREROUTE
Its a mechanism that provides a local protection for an
LDP FEC by pre-computing and downloading to the
forwarding plane hardware: both a primary and a backup
NHLFE (Next Hop Label Forwarding Entry) for this FEC.
The primary NHLFE corresponds to the label of the FEC
received from the primary next-hop as per standard LDP
resolution of the FEC prefix in RTM (routing table manager).
The backup NHLFE corresponds to the label received for the
same FEC from a Loop-Free Alternate (LFA) next-hop.
LFA next-hop pre-computation by IGP is described in [2].
LDP FRR relies on using the label-FEC binding received
from the LFA next-hop to forward traffic for a given
prefix as soon as the primary next-hop is not available.
In case of failure, forwarding of LDP packets to a
destination prefix/FEC is resumed without waiting for the
routing convergence.
The RTM module (routing table manager) populates both
primary and backup route and the forwarding hardware
should populate both primary and backup NHLFE for the FEC.
A. ROUTES AND LFA COMPUTATION REMINDER
Assuming : a,b,c,d,e,f,g represent the igp metrics on each
node link:

Figure 19. LFA concept reminder
The primary route will be via P1, assumed that:
a < (c + d) and (a + b) < (c + e + f)
The LFA route via P2 and P1 protects against failure of link
PE1-P1:
Loop Free Criterion (computed by PE1): The cost for P2
to reach P4 via P1 must be lower than the cost via routes
PE1 then P1, assumed that: d < (a + c )
Downstream Path Criterion (to avoid micro-loops): The
cost of reaching P4 from P2 must be lower than the cost
for reaching P4 from PE1, assumed that: d <a
The LFA route via P2 and P3 protects against the failure of
P1, node-protect condition for P2, assumed that:
(e + f)<(d+ b)
B. THE SPF ALGORITHME BEHAVIOR
1. Attempt the computation of a node-protect LFA next-
hop for a given prefix
2. If not possible, attempt the computation of a link-
protect LFA next-hop.
3. If multiple LFA next-hops for a given primary next-
hop are found, pick the node-protect in favor of the
link-protect.
4. If there is more than one LFA next-hop within the
selected type, pick one based on the least cost.
5. If more than one have the same cost, the one with the
least (outgoing interface: OIF) index is selected.
Both the computed primary next-hop and LFA next-hop for
a given prefix are programmed into the routing table
management.
C. LDP FASTREOUTE: LAB SETUP AND TEST METHOD:
The work have being done on the setup of Fig. 22 and
results are reported on tables: 2, 3.

Figure 22. LDP Fastreroute Lab setup
Vol.3, No.7, 2012
21 | P a g e
Table 2. Example of LFA precomputation
Table 3. LFA Lab coverage pourcentage
D. LAB TEST METHOD:
Same as described before (2.6.1) except that, here we cut
the inter P link [P1-P3], the backup path is [P1-P2-P4]. we
measure the convergence time through the number of lost
packets related to the ratio: (sent /received) packets per second.
Figure 23. received traffic curve
A. LDP FAST-REROUTE TEST RESULTS:
We see that mainly: the convergence time stays around 5
ms. This makes the LDP fast-reroute more attractif, however it
doesnt offer a 100% topology coverage.
Table 4 .LDP-FRR Traffic measurement

XV. RSVP-TE AND LDP-FRR COMPARAISON OUTCOMES
RSVP-TE gains:
Fast convergence P (detour LSP is precalculated,
presignaled for each LSP)
A convergence time around: 20 msec < t < 100 msec
RSVP-TE drawbacks:
additional level of routing complexity; requires P-P trunk
support rsvp, TLDP sessions, additional cpu load (rsvp
msg)
LDP(/IP) FRR gains:
local decision, no interop issues with other vendors
very simple configuration (just turn it on)
better scaling compared to full-mesh RSVP model
less overhead compared to RSVP soft-refresh states
LDP(/IP) FRR drawbacks:
lower backup coverage: depending on topologies may vary
between: 65 to 85%, indeed, the source routing paradigm: LDP
will always follows IP route, so if a candidate backup router
has its best route through originating node, this candidate node
cannot be chosen as backup.
While the conceptual restriction of LDP(/IP) FRR is
efficient against loops, it doesnt allow a 100% coverage of all
topologies, however we can reach a good compromise by a
mixture of both, RSVP shortcuts will be deployed if and where
LDP(/IP) FRR cannot offer coverage.
XVI. IGP MICRO-LOOPS
In standard IP networks, except when using source routing,
each router takes its own routing decision (hop by hop routing).
When the topology changes, during the convergence time, each
router independently computes best route to each destination.
Because of this independence, some routers may converge
quickly than others, the difference in convergence time may
create temporary traffic loops, thats what we call
microloops.
P1# show router isis routes alternative 10.0.222.5/32 Route Table
Prefix[Flags] Metric Lvl/Typ Ver.
NextHop MT AdminTag
Alt-Nexthop Alt-Metric Alt-Type
-------------------------------------------------------------------------------------------
10.0.222.5/32 11130 2/Int. 4950 P3
10.0.79.21 0 0
10.0.70.49 (LFA) 11140 nodeProtection
------------------------------------------------------------------------------------------
No. of Routes: 1
Flags: LFA = Loop-Free Alternate nexthop

P1# show router isis lfa-coverage
=============================================
LFA Coverage
==============================================
Topology Level Node IPv4 IPv6
--------------------------------------------------
IPV4 Unicast L1 0/0(0%) 3257/3260(99%) 0/0(0%)
IPV4 Unicast L2 27/28(96%) 3257/3260(99%) 0/0(0%)
============================================
Vol.3, No.7, 2012
22 | P a g e

Figure 24. Microloop birth
Micro-loops can be triggered by any topology change that
causes the network to converge like: link down, link up,metric
change ..etc.
Given the Fig. 24 above, when the link P2-P4 fails :
P2 detects failure and converges path to P3, as P3 is using
P2 as its nominal path, if P2 has converged but P3 didnt yet,
there is a creation of a micro-loop between both nodes, until P3
convergence is achieved.
A. MICRO-LOOPS LOCALIZATION
When a topology change occurs between 2 nodes A & B,
and given the IGP metric as in the the figure 25, a microloop
can occur :
Between A and his neighbors (local loop)
Between B and his neighbors (local loop)
A router upstream of A and one of his neighbors (remote
loop)
A router upstream of B and one of his neighbors (remote
loop)

Figure 25. Microloops dispersion
B. CONSEQUENCES OF MICRO-LOOPS
1) BANDWIDTH CONSUMPTION ESTIMATION:
Given the illustration below:

Figure 26. Microloop and bandwidth
Given 1 gigabit traffic coming from P1, as soon as this traffic
enters in the loop, each second , 1Gb additional data is
introduced in the loop.
Time P1-P2 link P2-P3 link
0sec 1Gb 1Gb
1sec 1Gb 2Gb
2sec 1Gb 3Gb
3sec 1Gb 4Gb

Looping traffic will consume bandwidth on the affected
link(s) until:
The link comes congestion
TTL of looping packet starts to expire
The network has converged
The bandwidth consumption will depend on a lot of
parameters:
Amount of traffic injected per second in the loop
Packet size
TTL of packets
RTD (round-trip delay time) of links
Packet switching time
To illustrate this, have a link with an RTD of 20 ms, a
monohop loop occurring on this link and a packet with an
initial TTL of 255 entering in the loop.

Figure 27. bandwidth consumption
Each time the packet crosses P2 and P3, the TTL is
decreased by one, we consider that this packet will do 127
round trip over the loop, so it will take 2540ms for the packet
to expire.
The bandwidth consumption depends also on the packet
size, consider 1 Gbps of traffic injected in the loop with a
packet size of 500 bytes , it means that each second, 250k
packets are injected in the loop.
Time P2-P3 link
0sec 250k packet
1 sec 500k packet
2 sec 750k packet
2,5 sec 750k packet + 125k packet (injected) 250k packet
(expired) = 625k packet
2,7 sec 625k + 50k (new injected) 50k (expiring)

Vol.3, No.7, 2012
23 | P a g e
In general we can say that:
(BW consumed by loop)=(BW injected) *(TTL/2)*(RTD)
Than for :
1Gbps, injected in a loop with 20ms RTD in loop,
TTL=255, loop maximum rate is 2,5 Gbps, than max link
BW usage of : 3,5 Gbps.
4Gbps, injected in a loop with 3ms RTD in loop, TTL =
250, loop maximum rate is 1,5 Gbps, than max link BW
usage of : 5,5 Gbps.
2) OVERLOAD IN ROUTERS CPU
If large amount of mpls traffic loops between two nodes A
and B; at each hop, the ttl of mpls pkt decreases by 1. When the
ttl of the mpls pkt expires, this pkt is dropped by the control
plane hardware (the routing engine) and not by the forwarding
plane hardware.
Depending the duration of the loop, the amount of mpls ttl-
expiring packets arriving to the control plane, the CPU load
may increase to 100%.
Mpls ttl expired packets come to the routing-engine mixed
with other important packets: igp (ISIS or OSPF) , bfd, bgp
..etc and all routing control packets, (mpls and non mpls). As a
consequence: bfd, the most sensitive one, may go down firstly,
and carry along all level3 protocol depending on it.
3) CAUTION ON QOS MODELS
If some quality of service models are used, and some types
of packets are prioritized, have this type of packets entering in
a loop, and depending on the loop duration, the amount of
prioritized traffic, they may consume all the bandwidth and
force control (routing) packets to be dropped. That is why, it is
a wise design to put the control packets on the top priority,
even above voive or other sensitive applications.
4) MICROLOOPS PROPAGATION
A level 3 loop occurring between two points A and B, and
as explained in paragraph 4.2.2, may trigger a convergence
again, potentially other microloops can appear far on other
routers, generating cpu load. The overall network will undergo
a phenomena we can define as a loop propagation.
Obviously, the cpu load will stay 100% until micro loops
disappear and convergence stabilize.
XVII. MICROLOOPS LAB SETUP AND TEST METHOD
Given the Figure 28, firstly we confirmed we can produce
loops by configuring different isis convergence timers to
facilitate loops appearance, then is a second stage, in order to
have more control, we created manual loops between P1-P3
and P1-P2.
We used a simple way to create loops: given a vpnA on a
PE1 connected to P1 and a vpnB on a PE2 connected to P2:
On PE1 vpnA have a static route to a destination
[a.b.c.d/mask] with PE2 loopback as the next-hop.
On PE2 vpnB have a static route to the same destination
with PE1 loopback as the next-hop.
PE1 and PE2 know loopback of each other through isis.
Using a traffic simulator we inject 10Millions packets
having the destination [a.b.c.d/mask], and to accelerate the
effect on CPU we put the TTL of all packet to values randomly
equal to 2 or 3.

Figure 28. Microloops lab setup
A. MICROLOOPS AND TRAFFIC PROTECTION
5) MICROLOOPS AND LFA
As explained, LFA computes an alternate nexthop that is
used when a local failure appears, however the alternate
nexthop may not be the converged backup nexthop.
Given the case of Fig. 29:
H is the LFA node
E is the converged nexthop, the backup calculated node
after the link [A-B] broke down

Figure 29 Microloops and LFA
Vol.3, No.7, 2012
24 | P a g e
When failure occurs, local router switch traffic to LFA
node, traffic is safe. When convergence is achieved on local
node, traffic is switched from LFA node to backup nexthop:
Traffic will be safe if backup node (and subsequent nodes)
have converged
Otherwise , traffic may go in microloop

Figure 30 potetial loop with LFA
6) MICROLOOPS AND IGP/LDP SYNCHRO
Setting high metric when IGP and LDP gets out of
synchronization and getting back to nominal metric (LDP/IGP
coming back in synchronization) can cause microloops (remote
or local). Same effect expected as the failed link comes up,
when the feature IGP/LDP synchronization in not implemented
at all.
7) CPU-PROTECTION MECHANISMS
Depending on the router manufacturer, several CPU
protection mechanisms may be implemented:
Ability to put a port overall rate that measures the arrival
of all control packets sent to the CPU for processing, giving the
possibility to selectively discard out-of-profile-rates. Ability to
create per protocol queues and guarantee selective high priority
for important packets. A dedicated study would assess the
efficiency of one or the other protection mechanism and proof
their robustness by testing under worst conditions.
XVIII. CONCLUSION
In this paper we presented the most important features wich
can contribute in convergence enhancement; it is not aimed at
detailing all existing features
We focused on methods that can be used to precompute
backup paths on the forwarding plane, we presented features
like: Prefix independent convergence and loop free alternate,
test results and gains obtained in comparison to the situation
with and without these features .
We presented a comparative study of RSVP-TE versus
LDP (/IP) Fast reroute, it appears that: with RSVP-TE, the
detour LSP is precalculated, presignaled for each LSP, the
convergence time is around: 20 msec < t < 100 msec. However
it has drawbacks like additional level of routing complexity,
requiring That P-to-P trunks support rsvp and full mesh TLDP
sessions, additional cpu load, due to rsvp messages. With
LDP(/IP) FRR we have local decisions, hence no interop issues
with other vendors, a simple configuration (just turn it on),a
better scaling compared to full-mesh RSVP model and less
overhead compared to RSVP soft-refresh states. However LDP
(/IP) FRR has an important drawback: A lower backup
coverage because of the source routing paradigm
Finally, we analyzed micro-loops phenomenon, bandwidth
and CPU consumption; we studied their birth mechanisms and
propagation, and initiated a reflexion on means to mitigate
them.
Overall, it is clear that the control of the convergence in its
globality is not an easy task, but our measurements and
simulations indicate that with good design and choice of tuning
features, we are confident a sub-second to tens of milliseconds
convergence time can be met.
REFERENCES
[1] Nuova Systems, K. Kompella Juniper Networks, JP. Vasseur Cisco
Systems, Inc., A. Farre Old Dog Consulting. Label Switched Path
Stitching with Generalized Multiprotocol Label Switching Traffic
Engineering (GMPLS TE).
[2] A. Atlas, Ed BT, A. Zinin, Ed. Alcatel-Lucent. Basic Specification for
IP Fast Reroute: Loop-Free Alternates (RFC 5286) September 2008
[3] E. Oki,T. Takeda NTT, A. Farrel Old Dog Consulting. Extensions to the
Path Computation Element Communication Protocol (PCEP) for Route
Exclusions.April 2009.
[4] L. Andersson Nortel Networks Inc., P. Doolan Ennovate Networks, N.
Feldman IBM Corp, A. Fredette PhotonEx Corp, B. Thomas Cisco
Systems Inc. LDP Specification (RFC 3036). January 2001
[5] D. Awduche Movaz Networks, Inc., L. Berger D. Gan Juniper
Networks, Inc. T. Li Procket Networks, Inc. V. Srinivasan Cosine
Communications, Inc. G. Swallow Cisco Systems, Inc. RSVP-TE:
Extensions to RSVP for LSP Tunnels (RFC 3209). December 2001.
[6] D. Awduche, J. Malcolm, J. Agogbua,M. O'Dell, J. McManus UUNET
MCI Worldcom (RFC-2702) September 1999.
[7] E. Rosen, Y. Rekhter. BGP/MPLS IP Virtual Private Network (VPNs)
(RFC-4364)
[8] Ina Minei, julian Lucek Juniper Networks, MPLS-Enabled Applications,
Emerging Developments and New Technologies .September 2008.
[9] L. AnderssonNortel Networks Inc, P. Doolan Ennovate Networks N.
Feldman IBM Corp, A. Fredette PhotonEx Corp, B. Thomas Cisco
Systems, Inc. (RFC-3036). January 2001
[10] Abdelali Ala, Driss El Ouadghiri, Mohamed Essaaidi: Convergence
enhancement within operator backbones for real-time applications.
iiWAS 2010: 575-583.
[11] Ala, A. Inf. & Telecom Syst. Lab., Abdelmalik Essaadi Univ., Tetuan,
Driss El Ouadghiri, Mohamed Essaaidi:
Fast convergence mechanisms and features deployment within operator
backbone infrastructures.
[12] P. Pan, Ed. Hammerhead Systems, G. Swallow, Ed. Cisco Systems, A.
Atlas, Ed. Avici Systems (RFC-4090).May 2005.
[13] T. Bates, R. Chandra, D. Katz, Y. Rekhter. Multiprotocol Extensions for
BGP-4 (RFC-2858).June 2000.
[14] Y. Rekhter, E. Rosen. BGP MPLS Carrying Label Information in BGP-4
(RFC 3107).May 2001.
[15] Y. Rekhter, T. Li, S. Hares, A Border Gateway Protocol 4 (BGP-4)
(RFC-4271). January 2006.
[16] Alia K. Atlas (edit BT), A. Zinin, Ed. Alcatel-Lucent. IP Fast Reroute:
Loop-Free Alternates (RFC 5286).September 2008
[17] P.Marques, R.Bonica from Juniper Networks, L.Fang, L.Martini, R.
Raszuk, K.Patel, J.Guichard From Cisco Systems, Inc.
Constrained Route Distribution for Border Gateway
Protocol/MultiProtocol Label Switching (BGP/MPLS) Internet Protocol
(IP) Virtual Private Networks (VPNs). (RFC 4684).November 2006.
Vol.3, No.7, 2012
25 | P a g e
[18] Susan Hares, NextHop Technologies Scaling MPLS Software to Meet
Emerging VPN Demands. January 2004.
[19] Zhuo (Frank) Xu Alcatel-Lucent SRA N0.1. Designing and
Implementing IP/MPLS-Based Ethernet Layer 2 VPN Services.2010.
AUTHORS PROFILE
Ala Abdelali is a phd student at Information and Telecom Systems Lab,
Faculty of Sciences Abdelmalek Essadi university, Tetuan Morocco.He
obtained his first engineer degree since September 1989 in Belgium, then his
D.E.A from the university of Paris XI since September 1992. Then he
worked ten years as support and telecom network designer in several IT
companies and telecom operators. His research area is: architecture, core
IP/MPLS/VPN design and network engineering.
Driss El Ouadghiri is a research and an associate professor at Science
Faculty, Moulay Ismail University, Meknes, Morocco, since September 1994.
He was born in Ouarzazate, Morocco. He got his License in applied
mathematics and his Doctorat de Spcialit de Troisime Cycle in computer
networks, respectively, in 1992 and 1997 from Mohamed V University, Rabat,
Morocco. In 2000 he got his PhD in performance evaluation in wide area
networks from Moulay Ismail University, Meknes, Morocco. He is a founding
member, in 2007, of a research group e-NGN (e-Next Generation Networks)
for Africa and Middle East. His research interests focus on performance
evaluation in networks(modelling and simulation), DiffServ architecture
(mechanisms based active queue management) and IPv6 networks. He spent
at INRIA Sophia-Antipolis, in the MISTRAL team, two long trips to scientific
research in 1995 and 1996. Also, he had a post-Doctoral research at INRIA-
IRISA of Rennes, in the ARMOR team, for a year from October 2000 to
October 2001.
Prof Mohamed Essaaidi is Currently director of ENSIAS. He is IEEE Senior
Member, he received the Licence de Physique degree, the Doctorat de
Troisime Cycle degree and the Doctorat dEtat degree in Electrical
Engineering and with honors, respectively, in 1988, 1992 and 1997 from
Abdelmalek Essaadi University in Tetuan, Morocco. He is a professor of
Electrical Engineering in Abdelmalek Essaadi University since 1993. He is
the founder and the current Chair of the IEEE Morocco Section since
November 2004. Prof. Essaaidi holds four patents on antennas for very high
data rate UWB and multiband wireless communication networks (OMPIC
2006, 2007, 2008). He has also co-organized several competitions aiming at
fostering research, development innovation in Morocco and in the Arab World
(Moroccan Engineers Week 2006, 2007 and Made in Morocco and ASTF
Made in Arabia Competitions in 2007 and 2009).
Vol. 3, No. 7, 2012

26 | P a g e
SVD Based Image Processing Applications: State of
The Art, Contributions and Research Challenges
Rowayda A. Sadek*
Computer Engineering Department, College of Engineering and Technology, Arab Academy for Science
Technology & Maritime Transport (AASTMT), Cairo, Egypt

Abstract Singular Value Decomposition (SVD) has recently
emerged as a new paradigm for processing different types of
images. SVD is an attractive algebraic transform for image
processing applications. The paper proposes an experimental
survey for the SVD as an efficient transform in image processing
applications. Despite the well-known fact that SVD offers
attractive properties in imaging, the exploring of using its
properties in various image applications is currently at its
infancy. Since the SVD has many attractive properties have not
been utilized, this paper contributes in using these generous
properties in newly image applications and gives a highly
recommendation for more research challenges. In this paper, the
SVD properties for images are experimentally presented to be
utilized in developing new SVD-based image processing
applications. The paper offers survey on the developed SVD
based image applications. The paper also proposes some new
contributions that were originated from SVD properties analysis
in different image processing. The aim of this paper is to provide
a better understanding of the SVD in image processing and
identify important various applications and open research
directions in this increasingly important area; SVD based image
processing in the future research.
Keywords- SVD; Image Processing; Singular Value
Decomposition; Perceptual; Forensic.
I. INTRODUCTION
The SVD is the optimal matrix decomposition in a least
square sense that it packs the maximum signal energy into as
few coefficients as possible [1,2]. Singular value
decomposition (SVD) is a stable and effective method to split
the system into a set of linearly independent components, each
of them bearing own energy contribution. Singular value
decomposition (SVD) is a numerical technique used to
diagonalize matrices in numerical analysis [3,4]. SVD is an
attractive algebraic transform for image processing, because of
its endless advantages, such as maximum energy packing
which is usually used in compression [5,6], ability to
manipulate the image in base of two distinctive subspaces data
and noise subspaces [6,7,8], which is usually uses in noise
filtering and also was utilized in watermarking applications
[9,6]. Each of these applications exploit key properties of the
SVD. Also it is usually used in solving of least squares
problem, computing pseudo- inverse of a matrix and
multivariate analysis. SVD is robust and reliable orthogonal
matrix decomposition methods, which is due to its conceptual
and stability reasons becoming more and more popular in
signal processing area [3,4]. SVD has the ability to adapt to the
variations in local statistics of an image [5]. Many SVD
properties are attractive and are still not fully utilized. This
paper provides thoroughly experiments for the generous
properties of SVD that are not yet totally exploited in digital
image processing. The developed SVD based image processing
techniques were focused in compression, watermarking and
quality measure [3,8,10,11,12]. Experiments in this paper are
performed to validate some of will known but unutilized
properties of SVD in image processing applications. This paper
contributes in utilizing SVD generous properties that are not
unexploited in image processing. This paper also introduces
new trends and challenges in using SVD in image processing
applications. Some of these new trends are well examined
experimentally in this paper and validated and others are
demonstrated and needs more work to be maturely validated.
This paper opens many tracks for future work in using SVD as
an imperative tool in signal processing.
Organization of this paper is as follows. Section two
introduces the SVD. Section three explores the SVD properties
with their examining in image processing. Section four
provides the SVD rank approximation and subspaces based
image applications. Section five explores SVD singular value
based image applications. Section six investigates SVD
singular vectors based image applications. Section seven
provides SVD based image applications open issues and
research trends.
II. SINGULAR VALUE DECOMPOSITION (SVD)
In the linear algebra the SVD is a factorization of a
rectangular real or complex matrix analogous to the
digonaliztion of symmetric or Hermitian square matrices using
a basis of eigenvectors. SVD is a stable and an effective
method to split the system into a set of linearly independent
components, each of them bearing own energy contribution
[1,3]. A digital Image X of size MxN, with MN, can be
represented by its SVD as follows;
| | | | | | | |
N
T
N
M
N M
M M
N
V S U X
= (1-a)

U= [u
1
, u
2
, . u
m
], V=[v
1
, v
2
, . v
n
] ,
(
(
(
(
o
o
o
=
n
2
1
S
O

(1-b)

Vol. 3, No. 7, 2012

27 | P a g e
Where U is an MxM orthogonal matrix, V is an NxN
orthogonal matrix, and S is an MxN matrix with the diagonal
elements represents the singular values, s
i
of X. Using the
subscript T to denote the transpose of the matrix. The columns
of the orthogonal matrix U are called the left singular vectors,
and the columns of the orthogonal matrix V are called the right
singular vectors. The left singular vectors (LSCs) of X are
eigenvectors of XX
T
and the right singular vectors (RSCs) of X
are eigenvectors of X
T
X. Each singular value (SV) specifies the
luminance of an image layer while the corresponding pair of
singular vectors (SCs) specifies the geometry of the image [13].
U and V are unitary orthogonal matrices (the sum of squares of
each column is unity and all the columns are uncorrelated) and
S is a diagonal matrix (only the leading diagonal has non-zero
values) of decreasing singular values. The singular value of
each eigenimage is simply its 2-norm. Because SVD
maximizes the largest singular values, the first eigenimage is
the pattern that accounts for the greatest amount of the
variance-covariance structure [3,4].
III. SVD IMAGE PROPERTIES
SVD is robust and reliable orthogonal matrix
decomposition method. Due to SVD conceptual and stability
reasons, it becomes more and more popular in signal
processing area. SVD is an attractive algebraic transform for
image processing. SVD has prominent properties in imaging.
This section explores the main SVD properties that may be
utilized in image processing. Although some SVD properties
are fully utilized in image processing, others still needs more
investigation and contributed to. Several SVD properties are
highly advantageous for images such as; its maximum energy
packing, solving of least squares problem, computing pseudo-
inverse of a matrix and multivariate analysis [1,2]. A key
property of SVD is its relation to the rank of a matrix and its
ability to approximate matrices of a given rank. Digital images
are often represented by low rank matrices and, therefore, able
to be described by a sum of a relatively small set of
eigenimages. This concept rises the manipulating of the signal
as two distinct subspaces [3,4]. Some hypotheses will be
provided and verified in the following sections. For a complete
review, the theoretical SVD related theorems are firstly
summarized, and then the practical properties are reviewed
associated with some experiments.
- SVD Subspaces: SVD is constituted from two
orthogonal dominant and subdominant subspaces. This
corresponds to partition the M-dimensional vector space
into dominant and subdominant subspaces [1,8]. This
attractive property of SVD is utilized in noise filtering
and watermarking [7,9].
- SVD architecture: For SVD decomposition of an image,
singular value (SV) specifies the luminance of an image
layer while the corresponding pair singular vectors (SCs)
specify the geometry of the image layer. The largest
object components in an image found using the SVD
generally correspond to eigenimages associated with the
largest singular values, while image noise corresponds to
eigenimages associated with the SVs [3,4]
- PCA versus SVD: Principle component analysis (PCA)
is also called the Karhunen-Love transform (KLT) or
the hotelling transform. PCA is used to compute the
dominant vectors representing a given data set and
provides an optimal basis for minimum mean squared
reconstruction of the given data. The computational basis
of PCA is the calculation of the SVD of the data matrix,
or equivalently the eigenvalues decomposition of the data
covariance matrix SVD is closely related to the standard
eigenvalues-eigenvector or spectral decomposition of a
square matrix X, into VLV, where V is orthogonal, and
L are diagonal. In fact U and V of SVD represent the
eigenvectors for XX and XX respectively. If X is
symmetric, the singular values of X are the absolute
value of the eigenvalues of X [3,4].
- SVD Multiresolution: SVD has the maximum energy
packing among the other transforms. In many
applications, it is useful to obtain a statistical
characterization of an image at several resolutions. SVD
decomposes a matrix into orthogonal components with
which optimal sub rank approximations may be obtained.
With the multiresolution SVD, the following important
characteristics of an image may be measured, at each of
the several level of resolution: isotropy, spercity of
principal components, self-similarity under scaling, and
resolution of the mean squared error into meaningful
components. [5,14].
- SVD Oriented Energy: In SVD analysis of oriented
energy both rank of the problem and signal space
orientation can be determined. SVD is a stable and
effective method to split the system into a set of linearly
independent components, each of them bearing its own
energy contribution. SVD is represented as a linear
combination of its principle components, a few dominate
components are bearing the rank of the observed system
and can be severely reduced. The oriented energy
concept is an effective tool to separate signals from
different sources, or to select signal subspaces of
maximal signal activity and integrity [1, 15]. Recall that
the singular values represent the square root of the
energy in corresponding principal direction. The
dominant direction could equal to the first singular vector
V
1
from the SVD decomposition. Accuracy of
dominance of the estimate could be measured by
obtaining the difference or normalized difference
between the first two SVs [16].
Some of the SVD properties are not fully utilized in image
processing applications. These unused properties will be
experimentally conducted in the following sections for more
convenient utilization of these properties in various images
processing application. Much research work needs to be done
in utilizing this generous transform.
IV. SVD-BASED ORTHOGONAL SUBSPACES AND RANK
APPROXIMATION
SVD decomposes a matrix into orthogonal components
with which optimal sub rank approximations may be obtained.
Vol. 3, No. 7, 2012

28 | P a g e
The largest object components in an image found using the
SVD generally correspond to eigenimages associated with the
largest singular values, while image noise corresponds to
eigenimages associated with the smallest singular values. The
SVD is used to approximate the matrix decomposing the data
into an optimal estimate of the signal and the noise
components. This property is one of the most important
properties of the SVD decomposition in noise filtering,
compression and forensic which could also treated as adding
noise in a proper detectable way.
A. Rank Approximation
SVD can offer low rank approximation which could be
optimal sub rank approximations by considering the largest
singular value that pack most of the energy contained in the
image [5,14]. SVD shows how a matrix may be represented by
a sum of rank-one matrices. The approximation a matrix X can
be represented as truncated matrix X
k
which has a specific rank
k. The usage of SVD for matrix approximation has a number of
practical advantages, such as storing the approximation X
k
of a
matrix instead of the whole matrix X as the case in image
compression and recently watermarking applications. Assume
X e R
mxn
. Let p =min(m,n), kp be the number of nonzero
singular values of X. X matrix can be expressed as
X

=
=
k
1 i
T
i i i
v u s ~ s
1
u
1
v
1
T
+ s
2
u
2
v
2
T
+.+ s
k
u
k
v
k
T
(2)
i.e., X is the sum of k rank-one matrices. The partial sum
captures as much of the energy of X as possible by a matrix
of at most rank r. In this case, energy is defined by the 2-
norm or the Frobenius norm. Each outer product (u
i
.v
i
T
)

is a
simple matrix of rank 1and can be stored in M+N numbers,
versus M*N of the original matrix. For truncated SVD
transformation with rank k, storage has (m+n+1)*k. Figure (1)
shows an example for the SVD truncation for rank k =20.

(a) (b)
Figure 1. Truncated SVD (a) Original (b) Truncated SVD
B. Orthogonal Subspaces
The original data matrix X is decomposed into the
orthogonal dominant components US
k
V
T
, which is the rank k
subspace corresponding to the signal subspace and US
n-k
V
T
,
which corresponds to the orthogonal subdominant subspace
that defines the noise components. In other words, SVD has
orthogonal Subspaces; dominant and subdominant subspaces.
SVD provides an explicit representation of the range and null
space of a matrix X. The right singular vectors corresponding
to vanishing singular values of X span the null space of X. The
left singular vectors corresponding to the non-zero singular
values of X span the range of X. As a consequence, the rank of
X equals the number of non-zero singular values which is the
same as the number of non-zero diagonal elements in S. This is
corresponding to partition the M-dimensional vector space (of
the mapping defined by X) into dominant and subdominant
subspaces [8]. Figure (2) shows image data dominant subspace
with the image truncated to k=30 SVD components, and its
subdominant; noise subspace. The SVD offers a good and
efficient way to determine the rank(X), orthonormal basis for
range(X), null(X), ||X||
2
, ||X||
Fro
and optimal low-rank
approximations to X in || ||
2
or || ||
F
, etc.
rank(X) = r = the number of nonzero singular values.
range(X) = span(u
1
, u
2
, . . . , u
r
)
null(X) = span(v
r+1
, v
r+2
, . . . , v
n
)
This subspaces SVD property that offers splitting the image
space into two distinct subspaces, the signal and the noise,
triggers proposing a contribution in watermarking application
in this paper. The contribution utilizes the resemblance
between the SVD domain with any noisy image (signal
subspace + noise subspace), or the watermarked image form
(image signal+watermark signal).

(a) (b) (c)
Figure 2. SVD subspaces (a) Original Image (b) Image Data subspace (c)
Noise subspace
C. Image Denoising
SVD has the ability to manipulate the image in the base of
two distinctive data and noise subspaces which is usually used
in noise filtering and also could be utilized in watermarking
[7,9]. Since the generic noise signal filtering model assumes
the noise can be separated from the data, SVD locates the noise
component in a subspace orthogonal to the data signal
subspace. Therefore, SVD is used to approximate the matrix
decomposing the data into an optimal estimate of the signal and
the noise components. Image noise manifests itself as an
increased spatial activity in spatial domain that guides to
increasing the smaller singular values in SVD domain. As there
is an added noise, singular values are non-uniformly increased
(although some may decrease) by amount that depends on the
image and the noise statistics, the medium values are increased
by largest amount, although smallest singular values have the
largest relative change. This depicted function will be more or
less skewed for different images and noise types. For Singular
vectors which are noised, it is hard, if not impossible to
analytically describe influence of noise on noised singular
vectors. Singular vectors that correspond to smaller singular
values are much more perturbed. Degradation from noising of
singular vectors is much bigger than that caused by increased
singular values. Incautious changes in singular vectors can
produce catastrophic changes in images. This is the reason why
the filtering operations are limited to slight filtering of noise in
singular vectors [7]. Based on the fact of non-uniformly
affecting the SVs and SCs by noise based on its statistics,
smallest SVs and faster changing singular vectors which
Vol. 3, No. 7, 2012

29 | P a g e
correspond to higher r values are highly affected with the noise
compared to the larger SVs and their corresponding SCs. [7,8].
This intuitive explanation is validated experimentally as shown
in figure (3).
Figure (3) shows the 2-dimensional representation of the
left and right SCs.
The slower changing waveform of the former SCs is versus
the faster changing of latter SCs. Figure (4) shows the
orthogonality of the different subspaces by carrying out
correlation between different slices. Figure (5) shows the SVD
based denoising process by considering the first 30
eigenimages as image data subspace and the reminder as the
noise subspace. By removing the noise subspace, image
displayed in figure(5b) represents the image after noise
removal.

(a) (b) (c)
Figure 3. 2D Representation of SCs: (a) Original Image (b) Left SCs; U (c)
Right SCs; V

Figure 4. Correlation is carried out between different subspaces (slices)

(a) (b) (c)
Figure 5. SVD Denoising (a) Original Noisy MRI Image (b) Image Data
subspace (c) Noise subspace
D. Image Compression
SVD with the maximum energy packing property is usually
used in compression. As mentioned above, SVD decomposes a
matrix into orthogonal components with which optimal sub
rank approximations may be obtained [5, 14].
As illustrated in equation 2, truncated SVD transformation
with rank r may offer significant savings in storage over storing
the whole matrix with accepted quality. Figure (6) shows the
bock diagram of the SVD based compression.

Figure 6. SVD based Compression
Compression ratio can be calculated as follows;

100 *
nm
mk k nk
R
+ +
=
(3)

Where R is the compression percentage, k is the chosen
rank for truncation; m and n are the number of rows and
columns in the image respectively. R for the truncated image
shown in figure (1) is 15.65 and for the one located in figure
(2) are 23.48. Figure (7) shows compressed images with
different chosen ranks for truncation that result in different
compression ratios. Table 1 illustrates the different truncation
levels k used for compressing image shown in figure (7) and
the resultant compression ratio for each truncation level. Peak
Signal to Noise Ratio (PSNR) is also illustrated in the table 1
corresponding to the different compression ratios to offer
objective quality measure
TABLE 1: COMPRESSION VS. PSNR
Number of
truncated levels k
Compression
R
PSNR
90 70.4498 37.7018
80 62.6221 36.0502
60 46.9666 32.7251
40 31.311 32.7251
20 15.6555 .92..42
10 7.8278 .523.22

(a) (b) (c)
Figure 7. SVD Based Compression (a) Original (b) Compression 47%
(truncation to k=60) (c) Compression 16% (truncation to k=20)
E. Image Forensic
For the current digital age, digital forensic research
becomes imperative. Counterfeiting and falsifying digital data
or digital evidence with the goal of making illegal profits or
bypassing laws is main objective for the attackers [15]. The
forensic research focuses in many tracks; steganography,
watermarking, authentication, labeling, captioning, etc. Many
applications were developed to satisfy consumer requirements
such as labeling, fingerprinting, authentication, copy control for
DVD, hardware/ software watermarking, executables
watermarks, signaling (signal information for automatic
counting) for propose of broadcast monitoring count [15].
The SVD packs the maximum signal energy into as few
coefficients. It has the ability to adapt to the variations in local
0 5 10 15 20 25 30
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Slices
C
o
r
r
e
l a
t
i o
n
Adaptivel
y select
truncation
rank

Approx.
SVD
SVD
SVD
(U,S,V)
Vol. 3, No. 7, 2012

30 | P a g e
statistics of an image. However, SVD is an image adaptive
transform; the transform itself needs to be represented in order
to recover the data. Most of the developed SVD based
watermarking techniques utilizes the stability of singular values
(SVs) that specifies the luminance (energy) of the image layer
[13,18]. That is why slight variations of singular values could
not influence remarkably on the cover image quality.
Developed SVD based techniques either used the largest SVs
[13,19] or the lowest SVs to embed the watermark components
either additively [18] or by using quantization [20]. D. Chandra
[18] additively embedded the scaled singular values of
watermark into the singular values of the host image X as
described above.
The Proposed Perceptual Forensic Technique
A new perceptual forensic SVD based approach which is
based on global SVD (GSVD) is proposed in this paper,. This
technique is developed to be private (non-blind) forensic tool.
The proposed forensic tool is based on efficient additively
embedding the optimal watermark data subspace into the host
less significant subspace (noise subspace). This forensic tool
can be utilized in all the forensic applications with some kind
of adaptation in the embedding region based on the required
robustness. Although many SVD based embedding techniques
for many forensic purposes are carried out additively in
singular values, they considered scaled addition without
considering the wide range of singular values. The proposed
scaled addition process that carried out for the SVs is treated
differently because of the wide range of the SVs sequence
which required to be flatted for convenient addition.
Embedding is carried out by getting the SVD for image X and
watermark W as follows in Eq. (4-a) and Eq. (4-b). The scaled
addition is as in Eq. (4-c). Finally watermarked image "Y" is
reconstructed from the modified singular values S
m
of the host
image as in Eq.(4-d).
X=U
h
S
h
V
h
T
(4-a)
W=U
w
S
w
V
w
T
(4-b)

s s < < - o +
=
Otherwise ) i ( S
k q 1 , M i k M if )) q ( S ( log ) i ( S
) i ( S
h
w h
m
(4-c)
Y=U
m
S
m
V
m
T
(4-d)
Where S
m
S
h
and S
w
are the singular values for the modified
media, host media and embedded data respectively. o is a
scaling factor which is adjustable by user to increase (or
decrease) the protected image fidelity and decrease (or
increase) the security of watermark protection, and robustness
at the same time. k is user defined, and could be chosen
adaptively based on the energy distribution in both of the host
and the embedded data (watermark). k represents the size of
range(embedded data) and null(host data). Since the SVs has
wide range, they should be treated differently to avoid the
abrupt change in the SVs sequence of the resultant
watermarked media which sure will give sever degradation.
Therefore, log transformation is proposed to solve this problem
by flatten the range of watermark SVs in order to be
imperceptibly embedding.
The detection is non-blind. The detection process is
performed as the embedding but in the reverse ordering. Obtain
SVD of the watermarked image "Y" and "X" as in Eq. (5-a,b).
Then obtain the extracted watermark components S'
w
as shown
in Eq.(5-c). Finally, construct the reconstructed watermark W
by obtaining the inverse of the SVD.
Y = U
m
S
m
V
m
T
(5-a)
X = U
h
S
h
V
h
T
(5-b)

S'
w
(i) = Exp((S(i)-S
h
(i))/o) for M-k<i<M (5-c)
W'=U
w
S'
w
V
w
T
(5-d)
Fidelity measure by using Normalized Mean Square Error
(NMSE) or Peak Signal to Noise Ratio (PSNR) is used to
examine perceptually of the watermarked image. The security
of the embedding process lays on many parameters, number of
selected layers used in truncation as Truncated SVD (TSVD)
for the watermark efficient compression, the starting column in
host SVD components that were used for embedding.
Experimentally examining this proposed forensic technique is
carried out as compared to commonly used developed
technique [19].
Figure (8) shows the watermarked image by using proposed
forensic technique (as in Eq.4) compared to the already
developed Chandra's scheme [19]. Figure (8a) shows the effect
of logarithmic transformation in the SVs sequence range.
Chandra's scheme that used constant scaling factor o to scale
the wide range of SVs produced anomalous change (zoomed
part) in the produced watermarked SVs sequence compared to
the original SVs sequence while the proposed technique
produces SVs smoothed sequence much similar to the original
SVs sequence.
Figure (8c, d) show the watermarked images by using
scaled addition of the SVs [sv01] and by using the proposed
logarithmic scaled of SVs addition with using the same scaling
factor (o=0.2). Objective quality measure by using NMSE
values for developed and proposed techniques are 0.0223 and
8.8058e-009 respectively. Subjective quality measure shows
the high quality of the resultant image from the proposed
technique compared to the developed one. Figure (9) also
examines the transparency by using a kind of high quality
images; medical images.
Both objectively and subjectively measures proves the
superiority of the proposed technique in transparency. NMSE
values for developed and proposed techniques are 0.0304 and
8.3666e-008 respectively.

(a)
Vol. 3, No. 7, 2012

31 | P a g e

(b) (c) (d)
Figure 8. Effect of logarithmic transformation on SVs range (a) SVs
sequences of original, scaled and logged one. (b)Watermark (c)Watermarked
image using scaled addition of watermark SVs (d) Watermarked image using
scaled addition of log of watermark SVs

(a) (b)

(c) (d)
Figure 9. Perceptual SVD forensic: (a) Original (b) Watermark
(c)Watermarked image using scaled addition of watermark SVs (d)
Watermarked image using scaled addition of log of watermark SVs
V. SVD SINGULAR VALUES CHARACTERISTICS
Since each singular value of image SVD specifies the
luminance (energy) of the image layer and respective pair of
singular vectors specifies image topology (geometry). That is
why slight variations of singular values could not influence
remarkably on the cover image quality [13]. Singular values
distribution and their decaying rate are valuable characteristics.
A. Singular Values Distribution
Since SVs represent the luminance, SVs of two visual
distinct images may be almost similar and the corresponding U
and V of their SVD are different because they are representing
image structure. This fact was examined and proved [15].
Figure(10) shows the closeness among the SVs of two different
images. Figure (11) demonstrates the reconstruction of the
image from its truncated 30 singular vectors and singular
values of the image itself and the two different images used in
the previous figure with NMSE; 0.0046, 0.0086 and 0.0292
respectively. This makes hiding any data in SVs values is
vulnerable to illumination attack and fragile to any illumination
processing [15]. This valuable feature could be utilized with
more research in the application such as; Stegano-analysis for
SVD based steganography and forensic techniques,
Illumination attacking for SVD based forensic techniques and
image enhancement by using selected SVs of a light image
analogy with the histogram matching [15].

(a) (b)

(c)
Figure 10. Similarity of SVs (a) Original Image (b) Faked image (c) SVs of
both of (a) and (b).

(a) (b) (c)
Figure 11. Reconstructed Image From 30 SVD truncated components (a) its
SVs (b) SVs of figure(10a) (c) SVs of figure(10b)
B. Singular Values Decaying
singular values are non-uniformly increased (although
some may decrease) by amount that depends on the image and
the noise statistics, the medium values are increased by largest
amount, although smallest singular values have the largest
relative change. This depicted function will be more or less
skewed for different images and noise types [7, 9].
Relying on the fact that says "Smooth images have SVs
with rapid decaying versus slow decaying of SVs of randomly
images", slope of SVs could be used as a roughness measure.
Figure (12) shows the rapid decaying of singular values of
smooth image versus those of the noisy image.

Figure 12. Rate of SVs decaying
Vol. 3, No. 7, 2012

32 | P a g e
C. Image Roughness Measure
Roughness measure is inversely proportional with the
decaying rate. Roughness measure could be used in the
application of perceptual based nature that considers the human
visual system (HVS) such as perceptual coding and perceptual
data embedding. Figure (13) shows the rapid decaying of
singular values of smooth low pass filtered image versus those
of the original image without filtering. The payload capacity of
any host image to hide data could also be measured also based
on roughness measure. Since the condition number (CN) is the
measure of linear independence between the column vectors of
the matrix X. The CN is the ratio between largest and smallest
SVs. The CN value could be used for the block based
processing by finding the CN for each block as follows;
min B
max B
B
S
S
CN =
(6)

Sensitivity to noise increases with the increasing of the
condition number. Lower CN values correspond to random
images which usually bear more imperceptible data
embedding. Conversely, the higher CN correspond to smooth
images which don't bear embedding data, Smooth blocks ( high
CN till ) and rough detailed blocks (with low CN till one for
random block).

B
Rf
is the roughness measure in a block B.
B
B
CN
1
. d Rf =
(7)

d is a constant.
B
Rf
ranges from d for highly

roughness
block to 0 for the completely homogenous smoothly block.
This valuable feature could be utilized with more research in
the adaptive block based compression.

Figure 13. LPF Effect on SVs of an image and its smoothed version
VI. SVD SINGULAR VECTORS CHARACTERISTICS
Since singular vectors specifiy image geometry, two visual
distinct images may have singular values but the U and V of
the SVD are different [15]. First singular vectors are slower
changing waveforms, they describe global shape of the scene in
the vertical and horizontal direction. This was experimentally
examined in figure (3). One of the important applications of
SVD is the analysis of oriented energy. SVD is a stable and
effective method to split the system into a set of linearly
independent components, each of them bearing own energy
contribution. Thus signal space orientation can be determined.
The oriented energy concept is an effective tool to separate
signals from different sources, to separate fill noisy signals or
to select signal subspaces of maximal signal activity [1,2]. The
norm of a matrix is a scalar that gives some measure of the
magnitude of the elements of the matrix [18,20].
A. Main dominant directions in the image structure.
For SVD, each direction of the critical oriented energy is
generated by right singular vector V with the critical energy
equal to the corresponding singular value squared. The left
singular vectors U represent each samples contribution to
the corresponding principle direction. It is well known in the
earlier works that the singular values can be seen as the
measure of concentration around the principal axes. The image
orientation can be obtained from the first singular vector (note
that the gradient vector are orthogonal to the image orientation
we seek, so after obtaining the principal direction of the
gradient vectors, we need to rotate by t /2 to get the
orientation we want) [16]. Singular values represent the square
root of the energy in corresponding principal direction. The
dominant direction could equal to the first singular vector (SC);
V
1
from the SVD decomposition. Accuracy of dominance of
the estimate could be measured by obtaining the difference or
normalized difference between the first two SVs [16]. Figure
(14) shows three different images; brain, horizontal rays and
vertical rays. Figure (14) also displays the SCs; U and V for all
the images as well as their SVs in a graph. Graph of the SVs
shows the difference in convergence to a rank.

(a) (b) (c)

(d)

(e) (f) (g)

(h) (i) (j)
Figure 14. Figure 14 SVD orientation (a-c) brain, horizontal and vertical
images respectively (d) Their SVs. (e-g) V for each image respectively (h-j) U
for each image respectively.
0 50 100 150 200 250
10
-300
10
-200
10
-100
10
0
S
i
n
g
u
l
a
r

V
a
l
u
e
Vol. 3, No. 7, 2012

33 | P a g e
B. Frobenius based Energy Truncation
The norm of a matrix is a scalar that gives some measure of
the magnitude of the elements of the matrix [18,20]. For n-
element vector A, Norm is equivalent to Euclidean length
therefore Norm-2 sometimes called Euclidean Norm. Euclidean
length is the square root of the summation of the squared vector
elements
=
=
n
1 i
2
i
) A ( ) A ( Norm
(8)

where A
i
,is n-element vector i =1,..,n are the components
of vector V
. This is also called Euclidean norm or Norm-2

which is equivalent the largest singular value that results from
the SVD of the vector A.
||A||
2
=
1
(9)

The Frobenius-norm of the mxn matrix A which is
equivalent to

) A * A ( diag ) A ( Norm
m
1 i
n
1 j
F
= =
' =

= =
=
m
1 i
n
1 j
2
A
(10)

This Frobenius norm can be calculated directly as follows;
||A||
F
=
2
n
2
2
2
1
o + + o + o A

=
o =
n
1 i
2
i
(11)

A pre-selected number "k" of SVD components (k image
layers) is to be truncated for efficient truncation in different
applications. This number of image layers could be selected
based on the energy content instead of using hard-threshold
value. Frobenius norm could be used as the base of content
energy measurement, the required specified contained energy;
E
k
could be represented as follows;
F
F
k
k
A
A
E = (12)
where A is the image and the A
k
is the approximated or
truncated image at rank "k". Truncated layers could be
specified by specifying the number of layers "k" required to
have 90% of the host ( 9 . 0 E
k
> )
C. Frobenius based Error Truncation
Frobenius error agrees with the error based on visual
perception, thus a threshold to preserve the required quality can
be controlled by using Frobenius norm; by specifying an error
threshold to avoid exceed it
F
F
k
F
A
A A
= c (13)
c
F
is the Frobenius error that is calculated from the
Frobenius norm of the difference between the image A and its
truncated version A
k
and normalized with the Frobenius norm
of the image A. Frobenius norm can be used to check the error
threshold. Find the needed rank that bound the relative error by
controlling the Frobenius norm to be less than a predefined
bounded threshold. Simply proper "k" number could be
selected to satisfy a predefined constraint either on Frobenius
energy or Frobenius norms error.
D. SVD-based Payload Capacity Measure
SVD transformation of an image may be utilized to give a
clue about the image nature, or may be used in a HVS based
classification process for image blocks roughness. Image
payload capacity and permissible perceptual compression could
be achieved by many SVD based methods; as Frobenius energy
or Frobenius norms error. The Frobenius based energy of first k
values is high means that the image is more smooth and hence
it has low capacity for data embedding and this can be proved
by considering Eq.(10). Conversely, the detailed image will
have less Frobenius energy for the same number layers "k",
compared to the higher Frobenius energy of smoothed image.
On the other hand, the sensitivity to noise is increased (low
capacity) for smooth image and decreased (high capacity) for
rough images. Therefore, the capacity of the image to bear
hidden information or to bear more compression without
perceptually noticeable effects is increasing for rough or high
detailed image and vice versa. Therefore, the chosen suitable
number of k layers can lead to a certain predefined quality
PSNR. Figure(15) shows the capacity of the image in carrying
hidden data or has compression in adaptively block based
manner. The block based capacity calculation uses 16x16 block
size.

(a) (b)
Figure 15. Block based Capacity calculation (a) Original (b) Capacity
VII. CONCLUSION AND OPEN ISSUES AND RESEARCH
TRENDS
Despite the attention it has received in the last years, SVD
in image processing is still in its infancy. Many SVD
characteristics are still unutilized in image processing. This
paper proposed a through practical survey for SVD
characteristics in various developed image processing
approaches. The paper also proposed contribution in using
unused SVD characteristics in novel approaches such as
adaptive block based compression, perceptual multiple
watermarking, image capacity for hiding information,
roughness measure, etc, All these contributions were
experimentally examined and gave promising results compared
to developed ones. The main contributions in this paper are a
novel perceptual image forensic technique, a new prospective
vision in utilizing the SVD Properties, reviewing and
experimental valuation of the developed SVD based
application such as denoising, compression, a new block based
roughness measure for application such as perceptual
progressive compression as well as perceptual progressive data
hiding. Image denoising and compression were thoroughly
examined and provided good results although they are image
dependent. Perceptual fragile forensic tool gives highly
promising results compared to the commonly used SVD based
Vol. 3, No. 7, 2012

34 | P a g e
tool. Energy based truncation and error based truncation as well
as the roughness measures are promising in many application.
The paper also suggests some open research issues which
require more research and development such as calculating the
block based dominant orientation, adaptively image fusion,
block based robust forensic, etc. On the other hand, more
utilization for proposed valuable feature of closeness between
the SVs of different images used in applications such as;
Stegano-analysis for SVD based steganography and forensic
techniques, Illumination attacking for SVD based forensic
techniques, image enhancement by using SVs matching in
analogy with the histogram matching, etc. The proposed SVD
based roughness measure could be also utilized in the
application such as; adaptive block based compression, payload
capacity measure for images in forensic tool, etc.
REFERENCES
[1] M Moonen, P van Dooren, J Vandewalle, Singular value
decomposition updating algorithm for subspace tracking, SIAM Journal
on Matrix Analysis and Applications (1992)
[2] T. Konda, Y. Nakamura, A new algorithm for singular value
decomposition and its parallelization, Parallel Comput. (2009),
doi:10.1016/j.parco.2009.02.001
[3] H. C. Andrews and C. L. Patterson, Singular value decompositions and
digital image processing, IEEE Trans. on Acoustics, Speech, and Signal
Processing, vol. ASSP-24, pp. 2653, 1976.
[4] Julie L. Kamm, SVD-Based Methods For Signal And Image
Restoration, PhD Thesis (1998)
[5] J.F. Yang and C.L. Lu, Combined Techniques of Singular Value
Decomposition and Vector Quantization for Image Coding, IEEE
Trans. Image Processing, pp. 1141 - 1146, Aug. 1995.
[6] Xiaowei Xu, Scott D. Dexter, Ahmet M. Eskicioglu: A hybrid scheme
for encryption and watermarking. Security, Steganography, and
Watermarking of Multimedia Contents 2004: 725-736
[7] K. Konstantinides, B. Natarajan, and G.S. Yovanof, Noise Estimation
and Filtering Using Block-Based Singular Value Decomposition, IEEE
Trans. Image Processing, vol. 6, pp. 479- 483, March 1997.
[8] E. Ganic and A. M. Eskiciogulu, Secure DWT-SVD Domain Image
Watermarking: Embedding Data in All Frequencies, ACM Multimedia
and Security Workshop 2004, Magdeburg, Germany, September 20-21,
2004
[9] V.I. Gorodetski, L.J. Popyack, V. Samoilov, and V.A. Skormin, SVD-
Based Approach to Transparent Embedding Data into Digital Images,
Proc. Int. Workshop on Mathematical Methods, models and Architecture
for Computer Network Security, Lecture Notes in Computer Science,
vol. 2052, Springer Verlag, 2001.
[10] Dobrovolny M. ilar Z., erny M. Asymmetric Image Compression for
Embedded Devices based on Singular Value Decomposition, IEEE
Applied Electronics Pilsen, 2011.
[11] Singh, S.K., Kumar, S. A Framework to Design Novel SVD Based
Color Image Compression, Computer Modeling and Simulation, 2009.
EMS '09. Third UKSim European Symposium, Athens 2010
[12] A. Shnayderman, A. Gusev and A. M. Eskicioglu, A Multidimensional
Image Quality Measure Using Singular Value Decomposition,
IS&T/SPIE Symposium on Electronic Imaging 2004, Image Quality and
System Performance, San Jose, CA, January 18-22, 2004.
[13] Ganic, N. Zubair, and A.M. Eskicioglu, "An Optimal Watermarking
Scheme based on Singular Value Decomposition," Proceedings of the
IASTED International Conference on Communication, Network, and
Information Security (CNIS 2003), pp. 85-90, Uniondale, NY,
December 10-12, 2003
[14] R. Karkarala and P.O. Ogunbona, Signal Analysis Using a
Multiresolution Form of the Singular Value Decomposition, IEEE
Trans. Image Processing, vol. 10, pp. 724-735, May 2001.
[15] Rowayda A. Sadek, "Blind Synthesis Attack on SVD Based
watermarking Techniques" International Conference on Computational
Intelligence for Modeling, Control and Automation - CIMCA'2008.
[16] J. Bigun, G. H. Granlund, and J. Wiklund, Multidimensional orientation
estimation with applications to texture analysis and optical flow, IEEE
Transactions on Pattern Analysis and Machine Intelligence 13(8) (1991),
775790.
[17] H. Demirel, G. Anbarjafari, and C. Ozcinar, "Satellite Image Contrast
Enhancement using Discrete Wavelet Transform and Singular Value
Decomposition", IEEE Geoscience and Remote Sensing
Letters, vol. 7, no. 2, pp. 334-338, Apr. 2010.
[18] R. Liu and T. Tan, A SVD-Based Watermarking Scheme for Protecting
Rightful Ownership, IEEE Transaction on Multimedia, 4(1), pp.121-
128, March 2002
[19] D. V. S. Chandra, Digital Image Watermarking Using Singular Value
Decomposition, Proceeding of 45th IEEE Midwest Symposium on
Circuits And Systems, pp. 264-267, Tulsa, OK, August 2002.
[20] Kuo-Liang Chung, C. Shen, L. Chang, "A novel SVD- and VQ-based
image hiding scheme", Pattern Recognition Letters, 2002,1051-1058
AUTHOR PROFILE
Rowayda A. Sadek received B.Sc., MSc and PhD degrees from
Alexandria University. She is currently an Assistant Professor in Computer
Engineering Department, Faculty of Engineering, Arab Academy on
Technology and Marine. She is on temporary leave from Helwan University.
Her research interests are in Multimedia Processing, Networking and Security.
Dr. Rowayda is a member of IEEE.

Vol. 3, No. 7, 2012

35 | P a g e
A Modified Feistel Cipher Involving XOR Operation
and Modular Arithmetic Inverse of a Key Matrix
Dr. V. U. K Sastry
Dean R & D, Dept. of Computer Science and Engineering,
Sreenidhi Institute of Science and Technology,
Hyderabad, India.
K. Anup Kumar
Associate Professor, Dept. of Computer Science and Engg
Hyderabad, India.

Abstract In this paper, we have developed a block cipher by
modifying the Feistel cipher. In this, the plaintext is taken in the
form of a pair of matrices. In one of the relations of encryption
the plaintext is multiplied with the key matrix on both the sides.
Consequently, we use the modular arithmetic inverse of the key
matrix in the process of decryption. The cryptanalysis carried out
in this investigation, clearly indicates that the cipher is a strong
one, and it cannot be broken by any attack.
Keywords- Encryption; Decryption; Key matrix; Modular
Arithmetic Inverse.
I. INTRODUCTION
In a recent development, we have offered several
modifications [1-4] to the classical Feistel cipher, in which the
plaintext is a string containing 64 binary bits.
In all the afore mentioned investigations, we have
modified the Feistel cipher by taking the plaintext in the form
of a matrix of size mx(2m),where each element can be
represented in the form of 8 binary bits. This matrix is divided
into two halves, wherein each portion is a square matrix of
size m. In the first modification [1], we have made use of the
operations mod and XOR, and introduced the concepts mixing
and permutation. In the second one [2], we have used modular
arithmetic addition and mod operation, along with mixing and
permutation. In the third one [3], we have introduced the
operations mod and XOR together with a process called
blending. In the fourth one [4], we have used mod operation,
modular arithmetic addition and the process of shuffling. In
each one of the ciphers, on carrying out cryptanalysis, we have
concluded that the strength of the cipher, obtained with the
help of the modification, is quite significant. The strength is
increased, on account of the length of the plaintext and the
operations carried out in these investigations.
In the present investigation, our interest is to develop a
modification of the Feistel cipher, wherein we include the
modular arithmetic inverse of a key matrix. This is expected to
offer high strength to the cipher, as the encryption key
induces a significant amount of confusion into the cipher,
on account of the relationship between the plaintext and the
cipher text offered by the key, as it does in the case of the Hill
cipher.
In what follows we present the plan of the paper. In section
2, we discuss the development of the cipher and mention the
flowcharts and the algorithms required in the development of
the cipher. In section 3, we illustrate the cipher with an
example. Here we discuss the avalanche effect which throws
light on the strength of the cipher. We examine the
cryptanalysis in section 4. Finally, we present computations
and conclusions in section 5.
II. DEVELOPMENT OF THE CIPHER
Consider a plaintext P having 2m
2
characters. On using
EBCIDIC code, this can be written in the form of a matrix
containing m rows and 2m columns, where m is a positive
integer. This matrix is divided into a pair of square matrices P
0

and Q
0
, where each square matrix is of size m. Let us consider
a key matrix K whose size is m x m.
The basic relations governing the encryption and the
decryption of the cipher, under consideration, can be written in
the form
P
i
= ( K Q
i-1
K ) mod N,

i = 1 to n

Q
i
= P
i - 1
P
i
, (2.1)
and
Q
i-1
= ( K
-1
P
i
K
-1
) mod N, i = n to1
P
i -1
= Q
i
P
i
, (2.2)
where, P
i
and Q
i
are the plaintext matrices in the i
th

iteration, K the key matrix, N is a positive integer, chosen
appropriately, and K
-1
is the modular arithmetic inverse of K.
Here, n denotes the number of iterations that will be carried
out in the development of the cipher.
The flow charts governing the encryption and the
decryption are depicted in Figures 1 and 2 respectively.

Vol. 3, No. 7, 2012

36 | P a g e

The algorithms corresponding to the flow charts can be
written as
ALGORITHM FOR ENCRYPTION
1. Read P, K, n, N
2. P
0
= Left half of P.
3. Q
0
= Right half of P.
4. for i = 1 to n
begin
P
i
= ( K Q
i-1
K ) mod N

Q
i
= P
i-1
P
i

end
5. C = P
n
Q
n
/* represents concatenation */
6. Write(C)
ALGORITHM FOR DECRYPTION
1. Read C, K, n, N
2. P
n
= Left half of C
3. Q
n
= Right half of C

4. for i = n to 1
begin
Q
i-1
= ( K
-1
P
i
K
-1
) mod N
P
i-1
= Q
i
P
i

end

5. P = P
0
Q
0
6. Write (P)
The modular arithmetic inverse of the key matrix K is
obtained by adopting Gauss Jordan Elimination method [5]
and the concept of the modular arithmetic.
III. ILLUSTRATION OF THE CIPPHER
Consider the plaintext given below:
Dear Ramachandra! When you were leaving this country
for higher education I thought that you would come back to
India in a span of 5 or 6 years. At that time, that is, when you
were departing I was doing B.Tech 1
st
year. There in America,
you joined in Ph.D program of course after doing M.S. I have
completed my B.Tech and M.Tech, and I have been waiting
for your arrival. I do not know when you are going to
complete your Ph.D. Thank God, shall I come over there? I do
wait for your reply. Yours, Janaki. (3.1)
Let us focus our attention on the first 128 characters of the
above plain text. This is given by
Dear Ramachandra! When you were leaving this country
for higher education I thought that you would come back to
India in a span (3.2)
On using EBCIDIC code, (3.2) can be written in the form
of a matrix having 8 rows and 16 columns. This is given by

Read K, P, n
P
0

Q
0

for i = 1 to n
P
i
= ( K Q
i-1
K ) mod N
P
i-1

Q
i-1

P
i
, Q
i

P
i
Q
i

Fig 1. The process of Encryption
C = P
n
|| Q
n

Read K, C, n
P
n

Q
n

for i = n to 1
Q
i-1
= ( K
-1
P
i
K
-1
) mod N
P
i

Q
i

P
i-1
, Q
i-1

P
i-1
Q
i-1

Fig 2. The process of Decryption
P = P
0
|| Q
0

Vol. 3, No. 7, 2012

37 | P a g e

68 101 97 114 32 82 97 109 97 99 104 97 110 100 114 97

33 32 87 104 101 110 32 121 111 117 32 119 101 114 101 32

108 101 97 118 105 110 103 32 116 104 105 115 32 99 111 117

110 116 114 121 32 102 111 114 32 104 105 103 104 101 114 32

101 100 117 99 97 116 105 111 110 32 73 32 116 104 111 117

103 104 116 32 116 104 97 116 32 121 111 117 32 119 111 117

108 100 32 99 111 109 101 32 98 97 99 107 32 116 111 32

73 110 100 105 97 32 105 110 32 97 32 115 112 97 110 32

Now (3.3) can be written in the form of a pair of square
matrices given by

68 101 97 114 32 82 97 109

33 32 87 104 101 110 32 121

108 101 97 118 105 110 103 32

110 116 114 121 32 102 111 114
P0 = (3.4)
101 100 117 99 97 116 105 111

103 104 116 32 116 104 97 116

108 100 32 99 111 109 101 32

73 110 100 105 97 32 105 110

and

97 99 104 97 110 100 114 97

111 117 32 119 101 114 101 32

116 104 105 115 32 99 111 117

32 104 105 103 104 101 114 32
Q0 = (3.5)
110 32 73 32 116 104 111 117

32 121 111 117 32 119 111 117

98 97 99 107 32 116 111 32

32 97 32 115 112 97 110 32
Let us take the key matrix K in the form

53 62 124 33 49 118 107 43

45 112 63 29 60 35 58 11

88 41 46 30 48 32 105 51

47 99 36 42 112 59 27 61
K = (3.6)
57 20 06 31 106 126 22 125

56 37 113 52 03 54 105 21

36 40 43 100 119 39 55 94

14 81 23 50 34 70 07 28
On using the encryption algorithm mentioned in section 2,
we get

47 36 206 218 60 59 123 231 136 21 102 153 8 73 110 244

73 133 152 198 214 246 181 216 219 86 197 165 70 115 201 31

95 27 149 155 233 115 150 255 233 44 85 154 100 29 189 243

196 5 152 137 225 237 35 158 142 228 195 135 76 243 1 238

233 223 102 67 156 183 123 146 131 183 190 72 128 179 0 5

205 185 126 90 88 195 182 149 176 26 183 212 219 50 69 189

106 233 188 190 71 35 180 237 243 247 198 73 199 225 125 217

4 218 198 221 31 99 91 29 251 152 197 93 37 36 141 183
On adopting the decryption algorithm, we get back the
original plaintext matrix given by (3.3)
Now we examine the avalanche effect. In order to achieve
this one, firstly, let us have a change of one bit in the plaintext.
To this end, we change the first row, first column element
of the plaintext from 68 to 69. On using the modified plaintext
and the encryption algorithm, we get the cipher text in the
form

182 108 50 76 228 143 108 194 82 71 102 45 35 114 42 205

136 59 104 240 46 91 111 139 182 196 145 144 118 247 206 246

183 231 51 76 131 162 190 193 13 118 54 243 150 255 160 118

222 183 253 242 134 155 217 219 57 228 143 175 234 217 190 149

11 49 141 164 151 169 3 76 128 195 188 119 38 28 44 6

207 17 23 230 197 93 29 205 190 30 219 124 244 202 186 103

159 174 73 254 88 164 214 32 30 239 150 239 105 115 59 236

242 254 30 225 123 169 182 107 236 237 147 244 150 46 23 45

On comparing (3.7) and (3.8) in their binary form, we
notice that they differ by 516 bits (out of 1024 bits).
Now let us consider a change of one bit in the key. In order
to have this one, we change the first row, first column element
of the key form 53 to 52.
On using this key and the encryption algorithm, given in
section 2, we get the cipher text in the form

70 219 194 242 76 237 163 193 37 187 209 38 42 205 50 14

222 249 226 2 204 99 107 123 90 236 109 171 98 210 163 57

228 143 175 141 202 205 244 185 203 127 244 150 42 205 50 14

222 249 226 2 204 68 240 246 145 207 71 114 97 195 166 121

128 247 116 239 179 33 206 91 189 201 65 219 223 36 64 89

128 2 230 220 191 45 44 97 219 74 216 13 91 234 109 153

34 222 181 116 222 95 35 145 218 118 249 251 227 36 227 240

190 236 130 109 99 110 143 177 173 142 253 204 98 174 146 146

On converting (3.7) and (3.9) into their binary form, we
notice that they differ by 508 bits ( out of 1024 bits ).
(3.3)
P=
(3.8)
C =
(3.7)
C =
(3.9)
C =
Vol. 3, No. 7, 2012

38 | P a g e
From the above analysis we conclude that the cipher is
expected to be a strong one.
IV. CRYPTANALYSIS
In the study of cryptology, cryptanalysis plays a prominent
role in deciding the strength of a cipher. The well-known
methods available for cryptanalysis are
a) Cipher text only attack ( Brute Force Attack )
b) Known plaintext attack
c) Chosen plaintext attack
d) Chosen cipher text attack

Generally, an encryption algorithm is designed to
withstand the brute force attack and the known plaintext attack
[6].
Now let us focus our attention on the cipher text only
attack. In this analysis, the key matrix is of size m x m. Thus,
it has m
2
decimal numbers wherein each number can be
represented in the form of 8 binary bits. Thus the size of the
key space is
8m
2
0.8m
2
2.4m
2

(2) = (2
10
)

(10) .

If we assume that the time required for the computation of
the encryption algorithm with one value of the key, in the key
space is
10
-7
seconds,
then the time required for the computation with all the keys
in the key space
2.4m
2
-7

10 x 10
365x24x60x60
2.4m
2
-15

=
10 x 3.12 x 10

(2.4m
2
15)

= 3.12 x 10
In this analysis, as we have taken m=8, the time required
for the entire computation is
138.6

3.12 x 10

This is enormously large. Thus, this cipher cannot be
broken by the cipher text only attack ( Brute Force Attack ).
Now let us study the known plaintext attack. In this case,
we know, as many plaintext cipher text pairs as we require. In
the light of this fact, we have as many P
0
and Q
0
, and the
corresponding P
n
and Q
n
available at our disposal. Now our
objective is to determine the key matrix K, if possible, to
break the cipher.
From the equations (2.1) and (2.2) we get
P1= ( K Q0 K ) mod N ,

Q1= P0 ( K Q0 K ) mod N ,

P2= ( K ( P0 ( K Q0 K ) mod N ) K ) mod N

Q2= ( ( K Q0 K) mod N ) ( K ( ( P ( K Q0 K ) mod N ) K ) mod N )

P3= ( K ( ( K Q0 K ) mod N ) ( K ( ( P0 ( K Q0 K ) mod N ) K ) mod
N ) mod N )

Q3= ( K ( ( K Q0 K ) mod N ) (K ( ( P0 ( K Q0 K ) mod N ) K ) mod N)

From the above equations we notice that, P
n
and Q
n
can be
written in terms of P
0
, Q
0
, K and mod N. These equations
are structurally of the form
P
n
= F ( P0, Q0, K, mod N ), (4.1)
Q
n
= G ( P0, Q0, K, mod N ), (4.2)
where F and G are two functions which depend upon, P
0
,
Q
0
, K and mod N. on inspecting above equations in the
analysis, we find that the equations (4.1) and (4.2) are
nonlinear in K.
Though the matrices P
0
and Q
0
, corresponding to the
plaintext P, and the matrices P
n
and Q
n
corresponding to the
ciphertext C are known to us, as the equations (4.1) and (4.2)
are nonlinear in K, and including mod N at various instances,
it is simply impossible to solve these equations and determine
K. Thus, this cipher cannot be broken by the known plaintext
attack.
As the relations (4.1) and (4.2) connecting P
0
, Q
0
and P
n

and Q
n
are formidable (being nonlinear and involving mod
N), it is not possible to choose a plaintext or a cipher text and
then determine the key K. Thus we cannot break the cipher in
case 3 and case 4.
In the light of the above facts, the cryptanalysis clearly
indicates that the cipher is a strong one.
V. COMPUTATIONS AND CONCLUSIONS
In this analysis the programs for encryption and decryption
are written in C language.
The entire plaintext given by (3.1) is divided into 4 blocks.
In the last block we have appended 5 blank characters to make
it a complete block, for carrying our encryption.
The cipher text corresponding to the entire plaintext is
obtained as given below
126 209 11 27 146 208 146 91 221 105 30 05 238 91 61 160
185 109 190 46 219 18 70 65 219 223 59 218 223 156 205 50
14 138 251 04 53 216 219 206 91 254 129 219 122 223 247 202
26 111 103 108 231 146 62 191 171 102 250 84 44 198 54 146
94 164 13 50 03 14 241 220 152 112 176 27 60 68 95 155
21 116 119 54 248 123 109 243 211 42 233 158 126 185 39 249
98 147 88 128 123 190 91 189 165 204 239 179 203 248 123 133
238 166 217 175 179 182 78 139 133 203 113 89 177 7 122 75
31 180 66 198 228 180 120 36 150 247 90 66 247 45 158 208
92 182 223 23 109 137 35 32 237 239 157 237 111 206 102 153
07 69 125 130 26 236 109 231 45 255 64 237 189 111 251 229
13 55 179 182 115 201 31 95 213 179 125 42 22 99 27 73
47 82 06 153 01 135 120 238 76 56 88 13 158 34 47 205
138 186 59 155 124 61 182 249 233 149 116 207 63 92 147 252
177 73 172 64 61 223 45 222 210 230 119 217 229 252 61 194
247 83 108 215 217 219 39 69 194 229 184 172 216 131 189 37
189 162 22 55 37 163 193 36 183 186 210 49 123 150 207 104
46 91 111 139 182 196 145 144 118 247 206 246 183 231 51 76
131 162 190 193 13 118 54 243 150 255 160 118 222 183 253 242
134 155 217 219 57 228 143 175 234 217 190 149 11 49 141 164
151 169 03 76 128 195 188 119 38 28 44 06 207 17 23 230
197 93 29 205 190 30 219 124 244 202 186 103 159 174 73 254
88 164 214 32 30 239 150 239 105 115 59 236 242 254 30 225
123 169 182 107 236 237 147 162 225 114 220 86 108 65 222 146

Years
=

Years.

Years.
Vol. 3, No. 7, 2012

39 | P a g e

197 141 201 104 240 47 114 217 237 09 37 189 214 145 251 68
23 45 183 197 219 98 72 200 59 123 231 123 91 243 153 166
65 209 95 96 134 187 27 121 203 127 208 59 111 91 254 249
67 77 236 237 156 242 71 215 245 108 223 74 133 152 198 210
75 212 129 166 64 97 222 59 147 14 22 03 103 136 139 243
98 174 142 230 223 15 109 190 122 101 93 51 207 215 36 255
44 82 107 16 15 119 203 119 180 185 157 246 121 127 15 112
189 212 219 53 246 118 201 209 112 185 110 43 54 32 239 73
(5.1)
From the cryptanalysis carried out in this paper, we
conclude that this cipher is a strong one and it cannot be
broken by any attack.
It may be noted here that this cipher has gained enormous
strength due to the multiplication of the plaintext matrix with
the key matrix and the process of iteration, which is changing
significantly the plaintext, before it becomes the cipher text.
REFERENCES
[1] V.U.K Sastry and K. Anup Kumar, A Modified Feistel Cipher
involving a key as a multiplicant on both the sides of the Plaintext matrix
and supplemented with Mixing Permutation and XOR Operation,
International Journal of Computer Technology and Applications ISSN:
2229-6093. Vol. 3, No.1, pp. 23-31, 2012.
Involving a Key as a Multiplicant on Both the Sides of the Plaintext
Matrix and Supplemented with Mixing, Permutation, and Modular
Arithmetic Addition, International Journal of Computer Technology
and Applications ISSN: 2229-6093. Vol. 3, No.1, pp. 32-39, 2012.
Involving a Pair of Key Matrices, Supplemented with XOR Operation,
and Blending of the Plaintext in each Round of the Iteration Process,
International Journal of Computer Science and Information
Technologies ISSN: 0975-9646. Vol. 3, No.1, pp. 3133-3141, 2012.
involving a pair of key matrices, Supplemented with Modular Arithmetic
Addition and Shuffling of the plaintext in each round of the
iteration process, International Journal of Computer Science and
Information Technologies ISSN: 0975-9646. Vol. 3, No.1, pp. 3119-
3128, 2012.
[5] William H.press,Brain P. Flannery, Saul A. Teukolsky, William T.
Vetterling, Numerical Recipes in C: The Art of Scientific Computing,
second Edition, 1992, Cambridge university Press, pp. 36-39.
[6] William Stallings, Cryptography and Network Security, Principles and
Practice, Third Edition, Pearson, 2003.
AUTHORS PROFILE
Dr. V. U. K. Sastry is presently working as Professor in
the Dept. of Computer Science and Engineering (CSE),
Director (SCSI), Dean (R & D), SreeNidhi Institute of
Science and Technology (SNIST), Hyderabad, India. He
was Formerly Professor in IIT, Kharagpur, India
and Worked in IIT, Kharagpur during
1963 1998. He guided 12 PhDs, and published more
than 40 research papers in various international journals.
His research interests are Network Security & Cryptography, Image
Processing, Data Mining and Genetic Algorithms.
Mr. K. Anup Kumar is presently working as an
Associate Professor in the Department of Computer
Science and Engineering, SNIST, Hyderabad India. He
obtained his B.Tech (CSE) degree from JNTU
Hyderabad and his M.Tech (CSE) from Osmania
university, Hyderabad. He is now pursuing his PhD
from JNTU, Hyderabad, India, under the supervision
of Dr. V.U.K. Sastry in the area of
Information Security and Cryptography. He has 10
years of teaching experience and his interest in research area includes,
Cryptography, Steganography and Parallel Processing Systems.

Vol. 3, No. 7, 2012

40 | P a g e
A Modified Feistel Cipher Involving Modular
Arithmetic Addition and Modular Arithmetic
Inverse of a Key Matrix
Dr. V. U. K Sastry
Dean R & D, Dept. of Computer Science and Engineering,
Hyderabad, India.
K. Anup Kumar
Associate Professor, Dept. of Computer Science and Engg
Hyderabad, India

Abstract In this investigation, we have modified the Feistel
cipher by taking the plaintext in the form of a pair of square
matrices. Here we have introduced the operation multiplication
with the key matrices and the modular arithmetic addition in
encryption. The modular arithmetic inverse of the key matrix is
introduced in decryption. The cryptanalysis carried out in this
paper clearly indicate that this cipher cannot be broken by the
brute force attack and the known plaintext attack.
Keywords- Encryption; Decryption; Key matrix; Modular
Arithmetic Inverse.
I. INTRODUCTION
In a recent investigation [1], we have developed a block
cipher by modifying the Feistel cipher. In this, we have taken
the plaintext (P) in the form of a pair of matrices P
0
and Q
0
,
and introduced a key matrix (K) as a multiplicant of Q
0
on
both its sides. In this analysis the relations governing the
encryption and the decryption are given by
P
i
= ( K Q
i-1
K ) mod N,

(1.1)
Q
i
= P
i - 1
P
i
,
and
Q
i-1
= ( K
-1
P
i
K
-1
) mod N, (1.2)
P
i -1
= Q
i
P
i
,
Here, multiplication of the key matrix, mod operation and
XOR are the fundamental operations in the development of the
cipher. The modular arithmetic inverse of the key plays a vital
role in the process of the decryption. Here N is a positive
integer, chosen appropriately, and n denotes the number of
iterations employed in the development of the cipher.
In the present paper, our objective is to develop a block
cipher by replacing the XOR operation in the preceding
analysis by modular arithmetic addition. The iteration process
that will be used in this cipher is expected to offer a strong
modification to the plaintext before it becomes finally the
cipher text.
Now, we present the plan of the paper. We introduce the
development of the cipher, and present the flowcharts and the
algorithms, required in this analysis, in section 2. In
section 3, we deal with an illustration of the cipher and discuss
the avalanche effect, then in section 4 we study the
cryptanalysis of the cipher. Finally, in section 5, we mention
the computations carried out in this analysis and draw
conclusions.
II. DEVELOPMENT OF THE CIPHER
Let us now consider a plaintext P. On using the EBCIDIC
code, the plaintext can be written in the form of a matrix
which has m rows and 2m columns. This is split into a pair of
square matrices P
0
and Q
0
, wherein both the matrices are of
size m.
The basic equations governing the encryption and the
decryption, in the present investigation, assume the form
P
i
= ( K Q
i-1
K ) mod N,

(2.1)
Q
i
= ( P
i - 1
+ P
i
) mod N i = 1 to n
and
Q
i-1
= ( K
-1
P
i
K
-1
) mod N, (2.2)
P
i -1
= (Qi

-

P
i
) mod N i = n to 1

The flowcharts depicting the encryption and the decryption
processes of the cipher are presented in Figures 1and 2.
Here it may be noted that the symbol || is used for placing
one matrix adjacent to the other. The corresponding algorithms
can be written in the form as shown below.
Algorithm for Encryption
1. Read P, K, n, N
2. P
0
= Left half of P.
3. Q
0
= Right half of P.
4. for i = 1 to n
begin
P
i
= ( K Q
i-1
K ) mod N

Q
i
= ( P
i-1
+ (K Q
i-1
K) ) mod N
end
5. C = P
n
Q
n
6. Write(C)

i = n to 1,
i = 1 to n,
Vol. 3, No. 7, 2012

41 | P a g e

Algorithm for Decryption
1. Read C, K, n, N
2. P
n
= Left half of C
3. Q
n
= Right half of C

4. for i = n to 1
begin
Q
i-1
= ( K
-1
P
i
K
-1
) mod N

P
i-1
= ( Q
i
- P
i
) mod N
end

5. P = P
0
Q
0
6. Write (P)
III. ILLUSTRATION OF THE CIPHER
Let us now consider the plain text given below.
Dear Janaki, I have received your letter. You sent me a
wonderful cryptography program and a letter along with that. I
started working six years back on web security. Really I am
finding it very difficult to hit upon an interesting problem. As
the complexity of network security is growing in all directions.
I am slowly losing my hope; I am thinking now whether I
would really contribute in a significant manner and get a Ph.D.
Please come and join me, so that, we shall lead a comfortable
life. (3.1)
Consider the first 128 characters of the above plaintext.
This is given by
Dear Janaki, I have received your letter. You sent me a
wonderful cryptography program and a letter along with that. I
started w (3.2)
On using the EBCIDIC code we get

68 101 97 114 32 74 97 110 97 107 105 44 32 73 32 104

97 118 101 32 114 101 99 101 105 118 101 100 32 121 111 117

114 32 108 101 116 116 101 114 46 32 89 111 117 32 115 101

110 116 32 109 101 32 97 32 119 111 110 100 101 114 102 117

108 32 99 114 121 112 116 111 103 114 97 112 104 121 32 112

114 111 103 114 97 109 32 97 110 100 32 97 32 108 101 116

116 101 114 32 97 108 111 110 103 32 119 105 116 104 32 116

104 97 116 46 32 73 32 115 116 97 114 116 101 100 32 119

P can be written in the form

68 101 97 114 32 74 97 110

97 118 101 32 114 101 99 101

114 32 108 101 116 116 101 114

110 116 32 109 101 32 97 32

108 32 99 114 121 112 116 111

114 111 103 114 97 109 32 97

116 101 114 32 97 108 111 110

104 97 116 46 32 73 32 115

Read K, P, n
P
0

Q
0

for i = 1 to n
P
i
= ( K Q
i-1
K ) mod N
P
i-1

Q
i-1

P
i
, Q
i

P
i
Q
i

Fig 1. The process of Encryption
C = P
n
|| Q
n

Read K, C, n
P
n

Q
n

for i = n to 1
Q
i

P
i-1

P
i-1
, Q
i-1

P
i-1
Q
i-1

Fig 2. The process of Decryption
P = P
0
|| Q
0

Q
i
= (P
i
+ P
i-1
) mod N
P
i

P
i -1
= ( Q
i
P
i
) mod N
Q
i-1
= ( K
-1
P
i
K
-1
) mod N
(3.3)
P =
P0 =
(3.4)
Vol. 3, No. 7, 2012

42 | P a g e
and

97 107 105 44 32 73 32 104

105 118 101 100 32 121 111 117

46 32 89 111 117 32 115 101

119 111 110 100 101 114 102 117

103 114 97 112 104 121 32 112

110 100 32 97 32 108 101 116

103 32 119 105 116 104 32 116

116 97 114 116 101 100 32 119

Now we take

53 62 124 33 49 118 107 43

45 112 63 29 60 35 58 11

88 41 46 30 48 32 105 51

47 99 36 42 112 59 27 61
K = (3.6)
57 20 06 31 106 126 22 125

56 37 113 52 03 54 105 21

36 40 43 100 119 39 55 94

14 81 23 50 34 70 07 28

On using the encryption algorithm, given in section 2, and
the key matrix K given by (3.6), we get the cipher text C in the
form

171 52 200 66 75 118 174 146 146 70 219 232 147 05 228 153

219 71 135 111 124 241 1 102 49 181 189 173 118 54 213 177

105 81 156 242 71 215 198 229 102 250 92 229 191 250 75 21

102 153 07 111 124 241 01 102 34 120 123 72 231 163 185 48

225 211 60 192 123 186 119 217 144 231 45 222 228 160 237 239

146 32 44 192 01 115 110 95 150 150 48 237 165 108 06 173

245 54 204 145 111 90 186 111 47 145 200 237 59 124 253 241

146 113 248 95 118 65 54 177 183 71 216 214 199 126 230 49
On using the cipher text (3.6) and the decryption
algorithm, we get back the original plaintext (3.2)
Now let us study the avalanche effect. To this end, we
change 4
th
row, 2
nd
column element from 116 to 117 in (3.3).
On using this modified plaintext and the encryption algorithm
we get the corresponding cipher text in the form

49 86 105 144 118 247 209 224 146 221 232 156 241 01 102 49

181 189 173 118 54 213 177 105 81 156 242 71 215 198 229 102

250 92 229 191 250 75 21 102 153 07 111 124 241 01 102 34

120 123 72 231 163 185 48 225 211 60 192 123 186 119 217 144

231 45 222 228 160 237 239 146 32 44 192 01 115 110 95 150

150 48 237 165 108 06 173 245 54 204 145 111 90 186 111 47

145 200 237 59 124 253 241 146 113 248 95 118 65 54 177 183

71 216 214 199 126 230 49 87 73 146 103 100 146 54 222 23

On comparing (3.7) and (3.8) in their binary form, we
notice that they differ by 514 bits out of 1024 bits.
Let us now consider a one bit change in the key. This is
achieved by replacing 4
th
row , 4
th
column element 42of K by
43 .
Now on using the modified key and the encryption
algorithm we get the cipher text C in the form

51 145 164 146 108 237 147 173 155 18 82 72 85 155 19 71

182 102 90 237 150 142 218 60 11 150 219 226 237 177 36 100

29 189 243 189 173 249 204 211 32 232 175 176 67 93 141 188

229 191 232 29 183 173 255 124 161 166 246 118 206 121 35 235

250 182 111 165 66 204 99 105 37 234 64 211 32 48 239 29

201 135 11 01 179 196 69 249 177 87 71 115 111 135 182 223

61 50 174 153 231 235 146 127 150 41 53 136 07 187 229 187

218 92 206 251 60 191 135 184 94 234 109 154 235 72 216 185

On comparing (3.7) and (3.9), after converting them into
their binary form, we find that the two cipher texts under
consideration differ by 518 bits out of 1024 bits. From the
above analysis, we conclude that the cipher is expected to be a
strong one.
IV. CRYPTANALYSIS
The different approaches existing for cryptanalysis in the
literature are
1. Cipher text only attack( Brute Force Attack )
2. Known plaintext attack
3. Chosen plaintext attack
4. Chosen cipher text attack
In this analysis, the key is a square matrix of size m.
Thus the size of the key space = (2)
8m2

If we assume that the time required for encryption is 10
-7

seconds then the time required for the computation with all the
keys in the key space [1]
(2.4m
2
15)

= 3.12 x 10 (4.1)

Q0 =
(3.5)
(3.7)
C =
(3.8)
C =
(3.9) C =

Years
Vol. 3, No. 7, 2012

43 | P a g e

When m = 8 , the time required for the entire computation
can be obtained as
138.6

3.12 x 10

As this time is very large, the cipher under consideration
cannot be broken by the brute force attack. Now let us
examine the known plaintext attack. In the case of this attack,
we know as many plaintext cipher text pairs as we require. In
the light of this fact, we have P
0
, Q
0
and P
n
, Q
n
in as many
instances as we want. Keeping the quotations governing the
encryption in view ( see algorithm for encryption ), we can
write the following equations connecting the plaintext and the
cipher text at different stages of the iteration process.
P
1
= ( K Q
0
K ) mod N
Q
1
= ( P
0
+ ( K Q
0
K ) ) mod N

P
2
= ( K ( ( P
0
+ ( K Q
0
K ) ) mod N ) K ) mod N
Q
2
= ( (( K Q
0
K ) mod N ) + ( K (( P
0
+ ( K Q
0
K ) )
mod N ) K ) ) mod N

P
3
= ( K (( (( K Q
0
K ) mod N ) + ( K (( P
0
+ ( K Q
0

K ) ) mod N ) K ) ) mod N ) K ) mod N
Q
3
= ( (( K ( ( P
0
+ ( K Q
0
K ) ) mod N ) K ) mod N
) + ( K (( (( K Q
0
K ) mod N ) + ( K (( P
0
+ ( K
Q
0
K ) ) mod N ) K ) ) mod N ) K) ) mod N
In view of the above system of equations we can write the
entities at the n
th
stage of the iteration as follows:
P
n
= F (P
0
, Q
0
, K, mod N),
Q
n
= F (P
0
, Q
0
, K, mod N), (4.2)

Here it is to be noted that, the initial plaintext can be
obtained by concatenating P
0
and Q
0
. The cipher text which
we get at the end of the iteration by concatenating P
n
and Q
n
.
Though we have as many relations as we want between the
cipher text and the plain text, the key matrix K cannot be
determined as the equations (4.2) are nonlinear and involving
mod N. In the light of the above discussion, we conclude that
this cipher cannot be broken by the known plaintext attack.
In the literature of the cryptography [2], it is well known
that a cipher must be designed such that it withstands at least
the first two attacks. As the relations given in (4.2) are very
complex, it is not possible either to choose a plaintext or to
choose a cipher text to attack the cipher. In the light of the
afore mentioned facts, we conclude that this cipher is a strong
one and it cannot be broken by any means.
V. COMPUTATIONS AND CONCLUSIONS
In this paper, we have developed a block cipher by
modifying the Feistel cipher, in this modification, the plaintext
is taken in the form of a matrix of size mx(2m). The iteration
process is carried out by dividing this matrix into two equal
halves wherein each one is of size mxm. In this analysis, we
have used multiplication of a portion of the plaintext (Q
0
) with
a key matrix on both the sides of Q
0
. Here we have made use
of modular arithmetic addition as a primary operation in the
cipher. The modular arithmetic inverse of the key is used in
the decryption process.
Programs are written for encryption and decryption in C
language. The entire plaintext given in (3.1) is divided into 4
blocks. Wherein each block contains 128 characters. We have
appended the portion of the last block with 15 characters so
that it becomes a full block. On carrying out encryption (by
using the key and the algorithm for encryption), we get the
ciphertext corresponding to the complete plaintext (3.1).
93 182 36 140 131 239 117 164 120 23 185 108 246 130 229 66
73 111 117 219 124 93 182 36 140 131 183 190 119 181 191 57
154 100 29 21 246 8 107 177 183 156 183 253 3 182 245 191
239 148 52 222 206 217 207 36 125 127 86 205 244 168 89 140
109 36 189 72 26 100 6 29 227 185 48 225 96 54 120 136
191 54 42 232 238 109 240 246 219 231 166 85 211 60 253 114
79 242 197 38 177 0 247 124 183 123 75 153 223 103 151 240
247 11 221 77 179 95 103 108 157 23 11 150 226 179 98 14
244 150 126 209 11 27 146 209 224 146 88 220 191 104 132 183
182 207 104 46 91 111 139 182 196 145 144 118 247 206 246 183
231 51 76 131 162 190 193 13 118 54 243 150 255 160 118 222
183 253 242 134 155 217 219 57 228 143 175 234 217 190 149 11
49 141 164 151 169 3 76 128 195 188 119 38 28 44 6 207
17 23 230 197 93 29 205 190 30 219 124 244 202 186 103 159
174 73 254 88 164 214 32 30 239 150 239 105 115 59 236 242
254 30 225 123 169 182 107 236 237 147 162 225 114 220 86 108
65 222 146 253 162 49 185 45 9 58 210 23 184 69 163 193
36 183 186 210 49 123 150 207 104 46 91 111 139 182 196 145
144 118 247 206 246 183 231 51 76 131 162 190 193 13 118 54
243 150 255 160 118 222 183 253 242 134 155 217 219 57 228 143
175 234 217 190 149 11 49 141 164 151 169 3 76 128 195 188
119 38 28 44 6 207 17 23 230 197 93 29 205 190 30 219
124 244 202 186 103 159 174 73 254 88 164 214 32 30 239 150
239 105 115 59 236 242 254 30 225 123 169 182 107 236 237 147
162 225 114 220 86 108 65 222 146 203 27 146 209 224 94 229
179 218 71 237 16 92 182 223 27 223 59 218 223 156 205 50
14 138 251 4 53 216 219 206 91 254 129 219 122 223 247 202
26 111 103 108 231 146 62 191 171 102 250 84 44 198 54 146
94 164 13 50 3 14 241 220 152 112 176 27 60 68 95 155
21 116 119 54 248 123 109 243 211 42 233 158 126 185 39 249
98 147 88 128 123 190 91 189 165 204 239 179 203 248 123 133
238 166 217 175 179 182 78 139 133 203 113 89 177 7 122 75

This cipher has acquired a lot of strength in view of the
multiplication with key matrix, the modular arithmetic
addition and mod operation. From the cryptanalysis, it is worth
noticing that the cipher is a strong one.
REFERENCES
[1] A modified Feistel cipher involving XOR operation and modular
arithmetic inverse of a key matrix (Accepted for publication, IJACSA,
Vol 3, No 7 )
[2] William Stallings, Cryptography and Network Security, Principles and
Practice, Third Edition, Pearson, 2003.
AUTHORS PROFILE
Dr. V. U. K. Sastry is presently
working as Professor in the Dept. of Computer Science
and Engineering (CSE), Director (SCSI), Dean (R & D),
SreeNidhi Institute of Science and Technology (SNIST),
Hyderabad, India. He was Formerly Professor in IIT,
Kharagpur, India and Worked in IIT, Kharagpur during
1963 1998. He guided 12 PhDs, and published more
than 40 research papers in various international journals.
His research interests are Network Security & Cryptography, Image
Processing, Data Mining and Genetic Algorithms.
Mr. K. Anup Kumar is presently working as an
Associate Professor in the Department of Computer
Science and Engineering, SNIST, Hyderabad India. He
obtained his B.Tech (CSE) degree from JNTU Hyderabad
and his M.Tech (CSE) from Osmania University,
Hyderabad. He is now pursuing his PhD from JNTU,
Hyderabad, India, under the supervision of Dr. V.U.K.
Sastry in the area of Information Security and Cryptography. He
has 10 years of teaching experience and his interest in research area includes
Cryptography, Steganography and Parallel Processing Systems.

Years
Vol. 3, No. 7, 2012

44 | P a g e
The Japanese Smart Grid Initiatives, Investments, and
Collaborations
Amy Poh Ai Ling
Graduate School of Advanced
Mathematical Sciences
Meiji University,
Kanagawa-Ken, Japan.

Sugihara Kokichi
Graduate School of Advanced
Mathematical Sciences
Meiji University,
Kanagawa-Ken, Japan.
Mukaidono Masao
School of Science and Technology,
Meiji University,
Kanagawa-Ken, Japan
Abstract A smart grid delivers power around the country and
has an intelligent monitoring system, which not only keeps track
of all the energy coming in from diverse sources but also can
detect where energy is needed through a two-way communication
system that collects data about how and when consumers use
power. It is safer in many ways, compared with the current one-
directional power supply system that seems susceptible to either
sabotage or natural disasters, including being more resistant to
attack and power outages. In such an autonomic and advanced-
grid environment, investing in a pilot study and knowing the
nations readiness to adopt a smart grid absolves the government
of complex intervention from any failure to bring Japan into the
autonomic-grid environment. This paper looks closely into the
concept of the Japanese governments go green effort, the
objective of which is to make Japan a leading nation in
environmental and energy sustainability through green
innovation, such as creating a low-carbon society and embracing
the natural grid community. This paper paints a clearer
conceptual picture of how Japans smart grid effort compares
with that of the US. The structure of Japans energy sources is
describe including its major power generation plants,
photovoltaic power generation development, and a comparison of
energy sources between Japan and the US. Japans smart
community initiatives are also highlighted, illustrating the
Japanese government planned social security system, which
focuses on a regional energy management system and lifestyle
changes under such an energy supply structure. This paper also
discusses Japans involvement in smart grid pilot projects for
development and investment, and its aim of obtaining successful
outcomes. Engagement in the pilot projects is undertaken in
conjunction with Japans attempt to implement a fully smart grid
city in the near future. In addition, major smart grid awareness
activities promotion bodies in Japan are discuss in this paper
because of their important initiatives for influencing and shaping
policy, architecture, standards, and traditional utility operations.
Implementing a smart grid will not happen quickly, because
when Japan does adopt one, it will continue to undergo
transformation and be updated to support new technologies and
functionality.
Keywords- Japanese Smart Grid; Initiative; Investment;
Collaboration; Low Carbon Emission.
I. INTRODUCTION
Culture encroachment happens through the interplay of
technology and everyday life. The emergence of the smart grid
creates a drastic increase in the demand for the smart supply of
energy flows [1]. The Japanese version of the smart city is
envisaged for the post-fossil fuel world. Alternative energy
sources, such as solar, wind, and nuclear power, are harnessed
in mass quantities [2]. In Japan, smart grid implies energy
transmission and distribution to promote the stability of the
electric power supply, by using information and
communication technology while introducing a high level of
renewable energy [3]. The focus will be on how to stabilize
power supplies nationwide as large amounts of wind and solar
power start entering the grid. This is because, unlike
conventional power sources, such as hydro, thermal, and
nuclear power, solar and wind energies are prone to the
vagaries of the weather. People in Japan are still not familiar
with the smart grid concept because the system has yet to gain
currency. According to a nationwide survey released in
December 2010 by the advertising agency Hakuhodo Inc., only
36.4% of about 400 respondents aged from 20 to 70 years said
they understood, or had heard of, a smart grid [4].
To address the likely impact of the smart grid on customers,
utilities, and society as a whole, it may be necessary to conduct
a pilot study [5]. It is widely understood that the new services
enabled by the smart grid will include different rate designs
that encourage curtailment of peak loads and make more
efficient use of energy. The future grid will be an autonomic
environment that helps users to not only share large-scale
resources and accomplish collaborative tasks but also to self-
manage, hence reducing user interventions as much as possible
[6]. This paper discusses the Japanese concept of the smart grid
and the differences between the smart grid movements in Japan
and the US. The Japanese government aims to achieve a low-
carbon society, employing the natural grid as a major source of
energy supply for the country, whereas the US is focusing on
its business and infrastructure development. This paper mainly
focuses on Japans smart grid structure, discusses Japans
initiatives, investments, and collaborations, and examines how
the full implementation of the smart grid is anticipated to come
into play comparing to an early written paper referring to the
natural grid whereby only the smart community is illustrated
[7].
II. METHODOLOGY
Conceptual analysis is adopted as our methodology with the
application of the hermeneutic circle. The hermeneutic circle is
used for interpretive reasons [8] and because it enables a
conceptualanalytical research method. Philosophers and
theologians in reviewing something that is not explicitly
Vol. 3, No. 7, 2012

45 | P a g e
present commonly apply this approach. This study investigates
the assumptions of different Japanese smart grid initiatives,
investments, and collaborations, for which the hermeneutic
circle is a natural choice of research methodology. The
methodology refers to the idea that what one understands of the
text as a whole is established by reference to the individual
parts, and what one understands of each individual part is
established by reference to the whole [9].
Our approach begins with determining and explaining the
meaning of the Japanese version of the smart city. This
followed by literature reviews describing the initiatives adopted
by the Japanese government and the companies supporting the
smart grid concept of going green and the attempts to create a
low-carbon society. Then, Japans energy resources are listed,
and its major power generation sources and photovoltaic (PV)
power generation development are elaborated on. The efforts
are then evaluated by using Japans smart grid community
approach and its smart grid pilot projects layout. Note that the
four main stages of this study are supported by three main
elementstheory, data and practice, which serve as a strong
reference for the sources obtained. Initiative, investment, and
collaboration are the main keywords in our contribution to the
literature review, which design to ensure that the sources used
are relevant to the topics studied.
III. CONCEPT
The main objectives of adopting smart grid technology
differ by region. Japans main objective is to achieve a total
shift from fossil fuels to renewable energy [10], generating a
low-carbon society. Reducing carbon dioxide (CO
2
) emissions
is by no means easy, and is thought to require a large number
of combined measures. One such measure is the utilization of
renewable energy, and Japan promotes this measure through
the development and employment of the natural grid.
In June 2010, the Japanese cabinet adopted a new Basic
Energy Plan. This was the third such plan that the government
had approved since the passage of the Basic Act on Energy
Policy in 2002, and it represented the most significant
statement of Japanese energy policy in more than four years,
since the publication of the New National Energy Strategy in
2006. Among the targets are a doubling of Japans energy
independence ratio, a doubling of the percentage of electricity
generated by renewable sources and nuclear power, and a 30%
reduction in energy-related CO
2
emissions, all by 2030 [11].
A. Low-carbon Society
It was believed that a low-carbon society would not be
realized without a fundamental shift in energy source use. The
Japanese smart grid concept aims to make the best use of local
renewable energy with a view to maximizing total efficiency.
The Japanese government is aiming to increase the
reliability of the grid system by introducing sensor networks
and to reduce opportunity losses by introducing smart meters.
The introduction of the smart grid will promote the use of
renewable energy by introducing a demand response system.
By focusing on electric vehicle (EV) technology, Japan is
moving toward introducing charging infrastructure for electric
cars [10]. Recently, increasing numbers of PV and wind power
plants have been installed across the country as clean energy
sources that emit no CO
2
[12].
B. Natural Grid
There is an urgent need to develop and adopt products,
processes, and behaviors that will contribute toward more
sustainable use of natural resources. In developing new green
technologies and approaches, engineers in Japan are finding
inspiration from natural products and systems. Since 2010, the
Tokyo Electric Power Company (TEPCO) and the Kansai
Electric Power Company have been testing the effects of smart
meters on load leveling under the Agency for Natural
Resources and Energy project.
There are five major elements in a community grid system.
The smart office refers to intelligent building design involving
cabling, information services, and environmental controls, and
envisages a desire for architecture with permanent capacity for
EVs for office energy backup. In order to accelerate grid
modernization in schools, the Japanese government started to
develop a vision of schools that operate by drawing energy
supply from PVs; this government program assists in the
identification and resolution of energy supply issues and
promotes the testing of integrated suites of technologies for
schools. The smart house projects relate to smart houses
interacting with smart grids to achieve next-generation energy
efficiency and sustainability [13], and information and
communication technology-enabled collaborative aggregations
of smart houses that can achieve maximum energy efficiency.
The potential benefits of smart houses include cheaper power,
cleaner power, a more efficient and resilient grid, improved
system reliability, and increased conservation and energy
efficiency. A smart house enables PVs and EVs to stabilize
demand and supply fluctuations. Smart factory systems enable
full factory integration of the PV cell into the bitumen
membrane. The PVs and EVs in a smart factory are supplied to
support its production process. Smart stores refer to the
charging outlets in parking areas and the deployment of public
charging stations for EVs. The arrows indicate that there exist
real-time energy flows and information flows in a community
grid system. Japans national grid system will be enhanced by
utilizing low-energy sources such as geothermal, hydraulic,
battery systems, and nuclear.
The natural grids are expected to play a vital role in making
effective use of renewable energy and providing a stable supply
of power by controlling the balance between electricity supply
and demand by using telecommunications technology.
C. A Comparison of the Smart Grid Movements in Japan and
the US
The energy sources in Japan and the US differ greatly, and
the implementation of the smart grid tends to differ between
countries, as do the timing and adoption of these technologies
[14]. Japan is pushing for advanced integrated control,
including demand-side control, so as to be ready to combine
unstable power (that is, reliant on weather conditions), such as
solar, with its strong foundation of a highly reliable grid, as
show in Table I.
With regard to Japans nuclear contribution to energy
supply, as of 2010, Japans currently operating 54 commercial
Vol. 3, No. 7, 2012

46 | P a g e
nuclear reactors have a total generation capacity of 48,847 MW,
and about 26% of electricity comes from nuclear power [15].
This compares with 104 licensed-to-operate nuclear power
plants operating in 31 states in the US (with 69 pressurized
water reactors and 35 boiling water reactors, which generate
about 19% of US electrical power [16]. Japans 10 electric
power companies are monopolies, being electric giants
vertically integrated in each region. By contrast, there are more
than 3,000 traditional electric utilities established in the US,
each with an interdependent structure. For this reason, utilities
supply of nuclear energy to customers differs greatly between
Japan and the US. In aiming toward a low-carbon society,
Japan depends greatly on nuclear power as an energy source,
whereas the US, in implementing its smart grid, focuses more
on business and infrastructure.
In terms of reliability, Japan already has a highly reliable
grid compared with the US, which needs more reliable and
distributed networks across the nation to develop its smart grid
system. Japan is developing its smart grid at a steady pace and
has already been investing in grid projects for almost 20 years;
over this period, there have been many developments. With
proper security controls, smart grids can prevent or minimize
the negative impact of attacks by hackers and thus increase the
reliability of the grid, thereby gaining the trust and meeting the
satisfaction of users [17].
TABLE I. COMPARISON OF JAPANESE AND US SMART GRID EFFORTS
Description Japan US
Nuclear as % of all
energy sources
26% 19%
No. of electric power
companies
10 electric power
companies (all
IOUs)
More than 3,000
traditional
electric utilities
(IOUs=210,
public=2,009,
coops=883,
federal=9)
Design
Vertically
integrated in each
region
Interdependent
infrastructure
Energy supply
0.725 million
customers
A few thousand
to more than five
million
customers
Aim
A low-carbon
society
Focus on
business and
infrastructure
Reliability
- Japan already has
a highly reliable
grid
- Going for
advanced
integrated control,
including demand-
side control, to
accommodate
unstable renewable
power
- Need for highly
reliable
transmission and
distribution
networks
- Need for
demand response
for peak shaving
and need to
avoid additional
infrastructure
Smart grid focus
- More than $100
billion investment
in the 1990s to
upgrade generation,
transmission, and
SCADA network
- Last mile and
demand-side
management
(DSM)
- Home solar
power
- Little
investment
(approx. $30
billion) in the
1990s into grid
- Now working
across entire grid
for
enhancements
- Last mile and
DSM are also
important
Cyber security research
Protect smart
meters, mutual
monitoring, privacy
in cloud computing
Grid computer,
cryptography
security

Since smart grid cyber security is significantly more
complex than the traditional IT security world, Japan focuses
on areas of smart grid cyber security concerns beyond smart
metering, such as mutual monitoring and privacy in cloud
computing. On the other hand, the US is enhancing its smart
grid cyber security in the age of information warfare, especially
in the area of grid computer and cryptography security.
In conclusion, the US focuses on businesses and
infrastructure, whereas Japan is striving to move toward a low-
carbon society by developing the smart grid system.
IV. ENERGY SOURCES
Energy resources in Japan are very limited and for that
reason about 97% of oil and natural gas has to be imported;
about half of these primary energy sources are converted to
electric power; the commercial and residential sectors account
for about 27% of total energy consumption, of which space
heating and air conditioning account for 24.5% of total
household electricity consumption [19].
Japans energy sources can be categorized into 11 groups:
electric power for commercial and industrial use; electric
power for residential use; gasoline; kerosene; heavy oil; light
oil; city gas; butane gas; propane gas; coal; and coke.
A. Japans Major Power Generation Sources
Unlike most other industrial countries, Japan does not have
a single national grid, but instead has separate eastern and
western grids. The standard voltage at power outlets is 100 V,
but the grids operate at different frequencies: 50 Hz in eastern
Japan and 60 Hz in western Japan [20]. Japans major power
generation sources are list as below.
1) Hydroelectric Power
This is one of the few self-sufficient energy resources in
resource-poor Japan. Hydroelectric power is an excellent
source in terms of stable supply and generation cost over the
long term [12]. Although steady development of hydroelectric
power plants is desired, Japan has used nearly all available sites
for the construction of large-scale hydroelectric facilities, and
so recent developments have been on a smaller scale. As the
gap in demand between daytime and nighttime continues to
grow, electric power companies are also developing pumped-
storage power generation plants to meet peak demand. The
Vol. 3, No. 7, 2012

47 | P a g e
share of pumped-storage generation facilities in total
hydroelectric power capacity in Japan is growing year by year.
The Okumino Hydroelectric Power Plant functions as a
pumped-storage plant; the other one is the Arimine Daiichi
Hydroelectric Power Plant [21].
2) Thermal Power
A diverse range of fuels including coal, oil, and liquid
natural gas (LNG) are used for the important power-generating
role played by thermal power plants. In particular, in response
to global environmental concerns, electric power companies are
promoting the introduction of LNG-fired plants, as they emit
less CO
2
and other pollutants. To enhance thermal efficiency
further, combined-cycle generating plants with both gas and
steam turbines have been installed. As a result, the gross
thermal efficiency, or the maximum designed value, now
exceeds 50%.
The two major thermal power plants are the Noshiro
Thermal Power Plant (coal fired) and the Nanko Thermal
Power Plant (LNG fired). Despite its small size, Japan has the
third largest geothermal energy potential in the world after the
US and Indonesia. However, in terms of harnessing that heat
and turning it into power, Japan only ranks eighth, after
countries with much smaller populations, such as Iceland and
New Zealand. Today, Japan only generates about 0.1% of its
electricity in 19 geothermal energy plants, many of which are
located in the Tohoku region, where the Fukushima nuclear
power plant is located [22].
3) Nuclear Power
Japans first commercial nuclear power plant started
operation in the Ibaraki Prefecture in 1966. As of January 19,
2011, Japan has 54 reactors operating around the country [23],
accounting for around one-third of the countrys total electric
power output. By 2018, the nuclear output share is expected to
reach 40%. Currently, there are 3 plants under construction,
and another 10 that are in the advanced planning stages.
Establishing a smart grid has been considered problematic
in Japan because of the monopoly on electricity supply, and
hence, there has been virtually no discussion of the smart grid.
Japans recent tsunami-induced nuclear crisis has, not
surprisingly, sent a signal about TEPCOs ability to supply
sufficient energy to the nation. The World Nuclear News
reported that Japan derives about 30% of its electricity from 54
nuclear reactors, seven of which were down for routine
maintenance when the earthquake and tsunami struck on March
11, 2011. Four more reactors were disabled during the disaster
and remain unstable, and more than one-third of Japans
nuclear-generated power was unavailable during that period
[24]. Three days after the tsunami struck, TEPCO reported that
its 12 thermal power units and 22 hydroelectric units had been
knocked out by the earthquake. This meant that only 33 GW of
capacity were available to meet 37 GW of demand on the day
of the earthquake, resulting in power outages to 2.4 million
households [25]. This phenomenon has made the government
realize that it needs a plan to overcome sudden energy
shortages.
B. PV Power Generation
The Japanese government developed its new policy on PV
systems in 2010 based on the principles of energy security,
global warming prevention, and an efficient supply of energy to
end users across the nation. Its goal, set to be achieved by 2030,
is focused on increasing independent energy supplies from
38% to about 70%, and increasing the proportion of electricity
generation with zero emissions from 34% to about 70% [26].
This new policy is supported by the governments aim of
becoming a leading nation in environmental and energy
sustainability through green innovation.

Source: IEA PVPS
Figure 1. Cumulative Installation of PVs in Japan, Germany, the US, Spain,
and Italy
Figure 1 illustrates the cumulative installation of PVs in
Japan and four other countries, namely Germany, the US,
Spain, and Italy, extracted from data on trends in PV
applications. As of 2009, Japan clearly ranks third, lagging far
behind Germany. The demand for PV systems in Germany has
remained persistently high for a full two years. Spain ranks
second and Spanish companies and research centers are taking
the lead in the recent revival of concentrated solar power, with
expansive banks of solar mirrors being assembled around the
country for concentrated solar plants.
In the context of Japans PV power generation development,
the expected change in domestic electricity demand in Japan in
relation to the expected installed PV system capacity by 2030,
along with the actual cumulative installed PV system capacity
as of 2007. Domestic electricity demand is expected to increase
sharply between 2010 and 2030. This is correlated with the
expected installed PV system capacity, which experienced a
sharp upturn in 2010, following the development of the
detection of unintentional islanding, and the development of
technology that curtails the restriction of PV system output.
Expanding installation of PV systems may increase the stability
of extra-high voltage transmission systems. At the final stage of
PV system development, it is predicted that imbalances
between output from PV systems and existing systems may
influence frequency on utility systems.
Vol. 3, No. 7, 2012

48 | P a g e
The New Energy and Industrial Technology Development
Organization (NEDO) and the European Commission
(European Union) will jointly launch a project to develop
concentrator PV cells, thus aiming to achieve a cell conversion
efficiency of more than 45%, which is the highest efficiency in
the world [27].
In addition, Sunetric, Hawaiis largest locally owned and
operated solar installer, has donated two solar PV systems to
raise funds for two local charities assisting Japan following the
tsunami that hit northeast Japan on March 11, 2011. The first is
the American Red Cross Hawaii State Chapter and the second
is the With Aloha Foundation [28].
In summary, the Japanese government is encouraging
further deployment of the conventional installation of
residential PV systems for the sake of the PV community. Its
current PV communities include Kasukabe and Yoshikawa in
Saitama, Matsudo in Chiba, Kasugai in Aichi, Kobe in Hyogo,
Tajiri in Osaka, Ota in Gunma, Wakkanai in Hokkaido,
Shimonoseki in Yamaguchi, and Kitakyusyu in Fukuoka.
Although the national subsidy program for residential PV
systems introduced by Japans Ministry of Economy, Trade,
and Industry (METI) was terminated, some local governments
have programs for residential PV systems in the regions.
C. A Comparison of Japanese and US Energy Sources
Japan lacks significant domestic sources of fossil fuels,
except coal, and must import substantial amounts of crude oil,
natural gas, and other energy resources, including uranium. In
1990, Japans dependence on imports for primary energy stood
at more than 85%, and the country has a total energy
requirement of 428.2 million tons of petroleum equivalent [29].
Figures 2 and 3 compare energy sources in Japan and the US.

Source: EIA
Figure 2. Energy Sources in Japan (data of 2007)

Source: EIA
Figure 3. Energy Sources in the US (data of 2007).
For Japan, the shares for coal, natural gas, and nuclear
energy are similar, and the remaining share is split between
hydroelectricity, petroleum, and other renewables. For the US,
the large share for coal is followed by natural gas and nuclear,
with smaller shares for hydroelectricity, then petroleum. Japan
is to be commended for having such a systematic and
comprehensive energy planning process. While maintaining its
goal of going green, Japan is utilizing more low-carbon energy
sources such as geothermal, hydraulic, battery systems, and
nuclear as major providers of energy.
V. JAPANS SMART COMMUNITIES
Japans smart community initiative is based on a systemic
approach. There are five identified action items: sharing vision
and strategy for smart communities; social experiments for
development and demonstration; standardization and
interoperability; data-driven innovation and privacy protection;
and smart communities for development. To address
simultaneously the three Es (environment, energy security, and
economy) requires the right mix and match of power sources
through renewable and reusable energy (RE) utilizing storage.
Table II illustrates the future social system at which Japan
is aiming, concentrating on the regional energy management
system (EMS) and lifestyle changes under such an energy
supply structure.
The development of Japans smart community is divided
into three stages: the first is the development plan for the
current period up to 2020; the second is the development plan
for 2020 to 2030; and the third is the development plan for
2030 onward.

Other
Renewabl
es, 1%
Petroleu
m, 13%
Hydroelec
tric, 8%
Natural
Gas, 26%
Nuclear,
26%
Coal, 26%
Other
Renewabl
es, 2%
Petroleu
m, 2%
Hydroelec
tric, 6%
Natural
Gas, 22%
Nuclear,
19%
Coal, 49%
Vol. 3, No. 7, 2012

49 | P a g e
TABLE II. JAPANS FUTURE SOCIAL SYSTEM

Current period to
2020
Period from 2020
to 2030
2030 onward
Relation
between
regional
EMS
and
entire
grid
Solar panel prices will
decrease significantly
because of the large-
scale introduction of
panels to houses and
commercial buildings.
Because of a
decline in PV
prices, more PV
systems will be
installed at houses.
Cost
competitiveness of
RE will improve
as fossil fuel prices
increase more than
twofold. Use of
RE will be
prioritized and
nuclear power will
be used as a base.
Measures will be
introduced to maintain
the quality of
electricity supply
while the large-scale
introduction of PV
systems is conducted
mainly for the grid
side. Storage cells will
be installed at
substations.
A regional EMS,
which contributes
to the effective use
of RE generated at
houses, will
become more
important.
An EMS that can
provide an
optimized balance
in terms of
economy and
security between
regional EMS and
the grid will be
established.
As the regional EMS
is further
demonstrated,
technology and
knowledge will be
accumulated.
A regional EMS
will be achieved as
storage cells
become cheaper
and are further
disseminated.
An EMS that
creates demand by
charging EVs at
the time of
excessive RE
reliance, and
supplies energy to
the grid at times of
high demand, will
be used.
The cost of storage
cells will decrease
because of technology
development and
demonstration.
Distribution and
transmission
networks that
enable two-way
communication
between the
demand side and
the grid side will
be actively
established.
Houses
Remote reading using
smart meters will start.
The home EMS
and the regional
EMS will be
integrated. All
power generated at
houses will be
used optimally.
A fully automated
home EMS will be
achieved.
The home EMS will
be disseminated. Some
houses will install
home servers. Demand
response
demonstration will
start.
Various services
using home
servers will be
disseminated.
Demonstration of EVs
will start.
EVs will be used
for power storage
as well.
Source: NEDO
With regard to the relation between the regional EMS and
the grid, the decrease in solar panel prices following their
large-scale introduction will cause more PV systems to be
installed at houses, which is expected to create cost
competitiveness in RE. The EMS will become more important
and will provide an optimized balance in terms of economy and
security. When EMS technology and knowledge has
accumulated and the cost of storage cells has fallen because of
technology development, the distribution and transmission
networks that enable two-way communication between the
demand and grid sides will be actively established. By that time,
the EMS that creates demand by charging EVs at times of
excessive reliance on RE, and supplies energy to the grid at
times of high demand, will be used.
As for the development of houses, the remote reading of
smart meters will start when the home EMS and the regional
EMS are integrated and all power generated at houses will be
used optimally. At the same time, the home EMS, followed by
various services using home servers, will be disseminated.
When the demonstration of EVs has started and when EVs are
used for power storage, a fully automated home EMS will be
achieved. In addition, zero-energy buildings (ZEBs) will be
introduced from 2020 to 2030, initially for new public
buildings. The introduction of ZEBs is expected to reduce
emissions greatly for all new buildings as a group.
The aims of Japans smart community are: (a) to cut energy
costs through competitive advantage and establish a new social
system to reduce CO
2
emissions; (b) to introduce widespread
use of RE; and (c) to facilitate the diversification of power
supplier services [30]. This approach is also directed at helping
Japanese customers see how cutting their energy costs can give
them a competitive advantage.
In addition, a consortium of Japanese companies is
preparing a report on the feasibility of smart community
development projects in Gujarat, India. According to
preliminary estimates by Japanese experts, of the 6.23 million
tons of hazardous waste generated in India annually, 22%
comes from Gujarat. This report prompted the Gujarat
government to sign a memorandum of understanding for
developing Surat as an eco-town along the lines of
Kitakyushu Eco-Town in Japan and for developing Dahej
along the lines of the reduce, reuse, and recycle-oriented
environmentally smart community development concepts
prevalent in Japan [31].
VI. JAPANS SMART GRID PILOT PROJECTS
On April 8, 2010, four sites were selected from four cities
in Japan to run large-scale, cutting-edge pilot projects on the
smart grid and smart community (budget request for FY2011:
18.2 billion yen) [25]. The community EMS will be achieved
based on a combination of the home EMS, the building EMS,
EVs, PVs, and batteries. Not only METIs smart grid-related
projects, but also projects in other ministries, such as
communications, environment, agriculture, and forestry, will be
implemented at these four sites (Figure 4).
Five months later in 2010, the Japan Wind Development
Company, the Toyota Motor Corporation, the Panasonic
Electric Works Company, and Hitachi Ltd. started a smart grid
demonstration project in Rokkasho Village, in the Aomori
Prefecture, aiming to verify technologies that allow for the
efficient use of energy for the achievement of a low-carbon
society [32]. Six months later, the HawaiiOkinawa
Partnership on Clean and Efficient Energy Development and
Deployment began, with the aim of helping the two island
regions switch from thermal power to renewable energy
systems, which are considered crucial for reducing CO
2

emissions but whose power supply is unstable [33].
Vol. 3, No. 7, 2012

50 | P a g e

Source: METI
Figure 4. Energy Sources in the US (data of 2007).
A. Smart Grid Pilot Project in Yokohama City (Large Urban
Area)
There were 900 units of PV systems installed in the so-
called progressive city of Yokohama in 2009, and the Japanese
government plans to install about 2,000 more 10 years hence
[34]. The aim of the Yokohama Smart City Project is to build a
low-carbon society in a big city, involving 4,000 smart houses.
This project is a five-year pilot program being undertaken with
a consortium of seven Japanese companies; they are the Nissan
Motor Co., Panasonic Corp., Toshiba Corp., TEPCO, the
Tokyo Gas Co., Accentures Japan unit, and Meidensha Corp.
The project focuses on the development of the EMS, which
integrates the home EMS, the building EMS, and EVs. It is
expected to generate PVs with a capacity of 27,000 KW. The
EMS, which integrates suburban, urban, and apartment areas,
will have PV use of heat and unused energy.
B. Smart Grid Pilot Project in the Kyoto Keihanna District
(R&D Focus)
The Smart Grid Pilot Project in the Kyoto Keihanna District
involves the Kyoto Prefecture, Kansai Electric Power, Osaka
Gas, Kansai Science City, Kyoto University, Doshisha, Yamate,
the Sustainable Urban City Council, and other local
governments and utilities. It makes use of the smart tap,
which visualizes energy consumption controlling home
electronics energy usage. It is also a pilot project to test
electric power virtual coloring technology, which actualizes
the overall home EMS. This project calls for the installation of
PVs in 1,000 houses and an EV car-sharing system. It also
studies nanogrid management of PVs and fuel cells in houses
and buildings on the visualization of demand. This project
grants Kyoto eco-points for the usage of green energy.
C. Smart Grid Pilot Project in Toyota City (Regional City)
The Toyota Rokkasho Village in the Aomori Prefecture
began to experiment with the smart grid in September 2010.
The key feature of the project is the pursuit of optimal energy
use in living spaces at the community level at the same time as
achieving compatibility between environmental preservation
and resident satisfaction. This project involved Toyota City,
and companies including Toyota Motors, Chubu Electric
Power, Toho Gas, Utilities, Denso, Sharp, Fujitsu, Toshiba,
KDDI, Circle K Sunkus, Mitsubishi Heavy Industries, and
Dream Incubator. The project focuses on the use of heat and
unused energy as well as electricity. It has a demand response
with more than 70 homes and 3,100 EVs. Through this project,
houses that contain an IT network of electrical appliances and
other household equipment, solar panels, household storage
batteries, onboard automobile storage batteries, and other
devices, can develop household power leveling and optimized
energy usage. As of June 2011, model smart houses in the
Higashiyama and Takahashi districts of Toyota City for testing
EMSs had been completed successfully and had begun trial
operations under the Verification Project for the Establishment
of a Household and Community-Based Low-Carbon City in
Toyota City, Aichi Prefecture [35].
D. Smart Grid Pilot Project in Kitakyushu City (Industrial
City)
This project involves 46 companies and organizations,
including the Kitakyushu City Government, GE, Nippon Steel,
IBM Japan, and Fuji Electric Systems. It focuses on real-time
management in 70 companies and 200 houses. Energy
management will be controlled by the home EMS and the
building EMS. The pilot project study of the energy system
integrates demand-side management and the high-energy
system. The Kitakyushu Hibikinada area is promoting low
carbon emissions, recycling, and nature coexistence in a
balanced manner. This project implements various
demonstrations, including communications, urban planning, a
transportation system, and lifestyle, with an emphasis on the
demonstration of energy projects such as electric power.
Implementation within five years, from FY2010 to FY2014,
involving the operation of 38 projects, is worth 16.3 billion yen
[36].
The Kitakyushu Smart Community Project focuses on the
development of technologies and systems related to a smart
grid with an eye on international standardization and the
expansion of international business, and the presentation of
new urban planning for a smart city, by developing various
human-friendly social systems compatible with next-generation
traffic systems and an aging society. In addition, a recycling
community will be constructed in which all sorts of waste will
be used as raw materials for other industrial fields to eliminate
waste and move toward a zero emissions community.
E. Smart Grid Demonstration Project in Rokkasho Village
In September 2010, the Toyota Motor Corp. began a two-
year project with Japanese Wind Development, Panasonic
Electric Works, and Hitachi Ltd. in the village of Rokkasho,
Aomori Prefecture, where wind power stations with large-
capacity batteries were built several years ago [8]. The project
involves a so-called smart grid village composed of six smart
houses equipped with automatic electricity control systems,
eight Toyota Prius plug-in hybrid vehicles, and a battery
system, all powered exclusively by renewable energy sources
and detached from the national electricity grid. Families of the
employees of the corporations participating in the project reside
Vol. 3, No. 7, 2012

51 | P a g e
in these smart houses, where they go about their normal lives
[37].
In addition, an experimental situation has been created in
isolation from the external power grid, where approximately
eight kilometers of private distribution line has been laid
between the Rokkasho-mura Futama Wind Power Station and
where the smart houses stand. The station is outfitted with 34
units of 1,500 KW windmills with a total capacity of 51,000
KW, and is equipped with large-capacity network-attached
storage batteries of 34,000 KW. The objective of the
experiment is to examine such factors as changes in electricity
usage in different seasons and at different times of day, and
investigate trends in electricity usage based on different family
configurations. The experiment will help create a system that
efficiently balances electricity supply and demand.
F. Smart Grid Trail Project in Okinawa
The Okinawa Electric Power Company has begun operating
a smart grid to control the supply of renewable-energy-derived
electricity for the 55,000-strong population of the remote
Okinawa Prefecture island Miyako-jima. The project is part of
the HawaiiOkinawa Partnership on Clean and Efficient
Energy Development and Deployment, an agreement between
the US Department of Energy, METI, the State of Hawaii, and
the Prefecture of Okinawa, which was signed in June 2010.
The HawaiiOkinawa partnership is intended to foster the
development of clean and energy-efficient technologies needed
to achieve global energy security and meet climate change
challenges. Japan and the US designated Hawaii and Okinawa
as the representatives for this groundbreaking partnership
because of their demonstrated leadership and experience in
clean energy and energy efficiency. The trials started in
October 2010. The infrastructure links the existing power grid
to a four MW solar power plant and a sodium sulfide battery
complex capable of storing four MW of power. Some lithium
ion batteries have also been installed. In addition, the system
also controls power from existing 4.2 MW wind farms situated
on Miyako-jima. Okinawa Electric spent 6.15 billion yen
(US$75.8 million) on the infrastructure, two-thirds of which
was subsidized by the national government [33]. The programs
will help the two island regions switch from thermal power to
renewable energy systems, which are considered crucial for
reducing CO
2
emissions, but whose power supply is unstable.
Each of the smart grid pilot and demonstration projects has
its own challenges. Building smart grids requires meeting the
requirements for the electricity supply, including the power
sources and transmission lines, and the communications
infrastructure of each specific country and region, as well as
introducing such elements as renewable energy generation
facilities, EVs and plug-in hybrid vehicles, storage batteries,
Eco Cute electric water heating and supply systems, and heat
storage units.
Currently, possibilities in new electricity distribution
methods and vast advances in information and communications
technology are raising the prospect of a shift from todays
conventional supplydemand adjustment approach to one that
optimizes both supply and demand. The Japanese government
is strongly promoting the efficient use of energy by developing
smart grid pilot projects and the promotion of the natural grid
community.
VII. JAPANESE COLLABORATIONS ON SMART GRID
PROJECTS
A. Smart Grid Project in Hawaii
A project supported by Japans NEDO, in cooperation with
the State of Hawaii, the Hawaiian Electric Company, the
University of Hawaii, and Pacific Northwest National
Laboratory, whose involvement is based on the JapanUS
Clean Energy Technologies Action Plan, was started in
November 2009 [38]. Hitachi Ltd., Cyber Defense Institute
Inc., the JFE Engineering Corporation, Sharp Corporation,
HewlettPackard Japan Ltd., and Mizuho Corporate Bank Ltd.
were among those selected as contractors for a joint JapanUS
collaboration supporting a smart grid project on the Hawaiian
island of Maui, which will serve as the project site. A
feasibility study is expected to be completed by the middle of
September 2011 [39]. Subject to the results of the feasibility
study, the project is expected to be implemented by the end of
March 2015.
While appearing to be state-of-the-art technological
marvels, smart grids expose energy production and distribution
to higher levels of security vulnerability than ever before. In
considering this matter, the project will incorporate the
installation of smart controls in the Kihei area on Maui at the
regional and neighborhood levels to improve the integration of
variable renewable energy resources, such as PV systems [40].
Installation of the smart grid technology is expected to begin in
late 2012, with the project becoming operational in 2013. The
demonstration project is scheduled to run from 2013 to 2015.
This project may be useful for the design of future micro-grids
that will provide secure backup and UPS services to distributed
energy residences and light industry. Independent, distributed
energy appliances in homes and businesses guarantee the
highest possible level of security and reliability in a national
power system.
B. Smart Grid Pilot Project on the Island of Jeju
Large electronics conglomerates in Japan and Korea, such
as Sharp, Panasonic, Samsung, LG, SK Telecom, and KT, are
building a domestic smart grid pilot on the island of Jeju, which
is south of Seoul in South Korea [41].
In March 2011, IBM announced that two new utilities had
joined its Global IUN Coalition, a group of utility companies
designed to further the adoption of smarter energy grids around
the world: TEPCO from Japan and KEPCO from Korea. These
two companies are in charge of the worlds largest
comprehensive smart grid test bed in Jeju Island, which brings
together smart technologies in the areas of generation, power
grids, electrical services, buildings, and transportation [42].
C. Smart Grid Project in New Mexico
Toshiba, Kyocera, Shimizu, the Tokyo Gas Company, and
Mitsubishi Heavy Industries will spend $33.4 million on a
smart grid project at Los Alamos and Albuquerque, New
Mexico [30]. NEDO will participate in research at Los Alamos
and Albuquerque and in collective research on the overall
project. Toshiba will install a one MW storage battery at the
Vol. 3, No. 7, 2012

52 | P a g e
Los Alamos site, while Kyocera and Sharp will test smart
homes, energy management, and load control technologies.
The Microgrid Demonstration Project in Los Alamos
involves concentrated PV energy generation and the
installation of power storage cells on distribution lines of 2 to 5
MW. In addition, absorption experiments on PV output
fluctuations will be conducted using PV-induced efficiencies
obtained by changing grid formation, and a distribution
network with high operability will be installed and
demonstrated by introducing smart distribution equipment. The
smart house is intended to maximize demand response by using
a home EMS and a demonstration will be carried out to verify
its effectiveness relative to an ordinary house. The micro-grid
demonstration in commercial areas of Albuquerque focuses on
demonstrating the demand response by using facilities in
industrial and commercial buildings. The move is prompted by
the aim of catching up with the US, which has taken the lead in
developing technological global standards [41]. It is also
intended to evaluate smart grid technology from Japan and the
US based on research results obtained at the five demonstration
sites of the New Mexico project [30].
D. ZEBs in Lyon in France
NEDO is holding discussions with Grand Lyon, the second
largest city in France, to introduce Japanese leading-edge
technologies for ZEBs in France and to establish an EV-
charging infrastructure coinciding with the Lyon Confluence
urban development project in Lyon [30].
VIII. MAJOR SMART GRID AWARENESS AND ACTIVITIES
PROMOTION BODIES IN JAPAN
One important initiative is having members engage in smart
grid promotional activities that complement activities in
existing organizations and groups that currently influence and
shape policy, architecture, standards, and traditional utility
operations [43]. A few major smart grid awareness and
activities promotion bodies and associations affirm the
successful implementation of smart grids in Japan and raise
peoples awareness through education.
A. The Japan Smart Community Alliance (JSCA)
The JSCA is a member of the Global Smart Grid Federation
[44], which aims to promote publicprivate cooperative
activities relating to the development of smart communities by
tackling common issues such as dissemination, deployment,
and research on smart grid standardization. The JSCA has
members from the electric power, gas, automobile, information
and communications, electrical machinery, construction, and
trading industries as well as from the public sector and
academia. Four working groups have been established at a
practical level for discussion and deliberation in order to
facilitate the JSCAs activities. The four working groups are
the International Strategy Working Group, the International
Standardization Working Group, the Roadmap Working
Group, and the Smart House Working Group. Each group
supports smart grid development in Japan.
B. METI
METI is reviewing closely Japans PV/CSP technologies
and programs to support smart grid development in Japan.
C. Democratic Party of Japan (DPJ)
The DPJ is promoting the development and diffusion of
smart electricity grid technologies.
D. NEDO
While NEDO is committed to contributing to the resolution
of energy and global environmental problems and further
enhancing Japans industrial competitiveness, it strongly
supports numerous smart grid research and development
projects. NEDO aims to develop the worlds most efficient
concentrator PV cells.
In addition, Japan is promoting smart grids by conducting
discussions and undertaking projects in an integrated manner
with the participation of various stakeholders [45]. Two months
ago, nine Japanese companies - Shimizu Corporation, Toshiba
Corporation, Sharp Corporation, Meidensha Corporation,
Tokyo Gas Co., Ltd., Mitsubishi Heavy Industries, Ltd., Fuji
Electric Co., Ltd., Furukawa Electric Co., Ltd. and The
Furukawa Battery Co., Ltd. launched a demonstration study for
the Albuquerque Business District Smart Grid Demonstration
Project consigned to them by the New Energy and Industrial
Technology Development Organization (NEDO), to be carried
out as part of its Japan-U.S. Collaborative Smart Grid
Demonstration Project, took will took place form March 2012
to March 2014 [46].
IX. CONCLUSION
Smart technologies improve human ability to monitor and
control the power system. Smart grid technology helps to
convert the power grid from static infrastructure that is
operated as designed to flexible and environmentally friendly
infrastructure that is operated proactively. This is consistent
with the Japanese governments goal of creating a low-carbon
society, maximizing the use of renewable energy sources, such
as photovoltaic and wind power. Nevertheless, publicprivate
sector cooperation across various industries is necessary to
establish smart communities.
This paper provided sufficient information for the reader to
understand the Japanese concept of the smart grid and the
governments associated strategy and the significance of the
governments contribution to Japans energy supply capacity,
without documenting full case studies in detail.
Japan chosen to be at the boundary at some time in the
duration of the smart grid revolution because they already have
had a reliable grid system. Recent the Japanese association is
bringing up large-scale grids that deal with the power like how
the internet does with the information data.
This paper painted a conceptual picture of Japans smart
community initiatives and its investment in and collaboration
on smart grid pilot projects. With emphasis on Japanese
initiatives, investment, and collaboration, comparison tables,
figures, and graphs relating to smart grid developments were
used to enable understanding of the issues. Although the Asia
Pacific region is quickly catching up in smart grid
developments and adoption, because the energy sources of
different countries vary significantly, the methods and timing
with which countries adopt this technology differ. Japan is
currently focusing on last mile and demand-side management
Vol. 3, No. 7, 2012

53 | P a g e
and home solar power. Researchers have started to address
challenges caused by large-scale solar power generation
connected to the power grid as well as information security
issues. Because the smart grid remains a novel field of study in
Japan, it has great potential for further research.
ACKNOWLEDGMENT
This study was supported by the Meiji University Global
COE Program Formation and Development of Mathematical
Sciences Based on Modeling and Analysis, Meiji Institute for
Advanced Study of Mathematical Sciences (MIMS), and the
Japan Society for the Promotion of Science (JSPS).
REFERENCES
[1] Amy Poh Ai Ling, Mukaidono Masao, Grid Information Security
Functional Requirement Fulfilling Information Security of a Smart Grid
System, International Journal of Grid Computing & Applications
(IJGCA), Vol. 2, No. 2, June 2011, pp. 119.
[2] Tomoko A. Hosaka, Japan Creating Smart City of the Future,
October 2010, pp. 124135.
[3] Tatsuya Shinkawa, Smart Grid and Beyond, New Energy and
Industrial Technology Development Organization (NEDO), 2010.
[4] Hiroko Nakata, Smart Grid Pursuit Slow Off Mark, Smart Grid, The
Japan Times, January 2011.
[5] Ahmad Faruqui, Ryan Hledik, Sanem Sergici, Piloting the Smart Grid,
The Electricity Journal, Vol. 22, Issue 7, AugustSeptember 2009, pp.
5569.
[6] Feilong Tang, Minglu Li, Joshua Zhexue Huang, Real-Time
Transaction Processing for Autonomic Grid Applications, Journal of
Engineering Applications of Artificial Intelligence, Vol. 17, Issue 7,
October 2004, pp. 799807.
[7] Amy Poh Ai Ling, Mukaidono Masao, Sugihara Kokichi, The Natural
Grid Concept and the Strategy of Asias Energy-Balance Pioneer, The
Ninth International Conference on Advances in Mobile Computing and
Multimedia, submitted.
[8] Heinz K. Klein, Michael D. Myers, A Set of Principles for Conducting
and Evaluating Interpretive Field Studies in Information Systems, MIS
Quarterly, Vol. 23, Issue 1, May 1999, pp. 6794.
[9] Ramberg Bjorn, Kristin Gjesdal, Hermeneutics, Stanford
Encyclopedia of Philosophy, Library of Congress Data, November 2005.
[10] Shinsuke Ito, Japans Initiative on Smart GridA Proposal of Nature
Grid, Information Economy Division, Ministry of Economy Trade and
Industry, December 2009, available at http://documents.eu-
japan.eu/seminars/europe/other/smart_grid/presentation_ogawa.pdf
[11] John S. Dufeld, Brian Woodall, Japans New Basic Energy Plan,
Energy Policy, Vol. 39, Issue 6, June 2011, pp. 37413749.
[12] Electricity Review Japan, The Federation of Electric Power
Companies of Japan, 2011.
[13] Anke Weidlich, Smart House / Smart Grid, ICT Challenge 6: Mobility,
Environmental Sustainability and Energy Efficiency, Smart Houses
Interacting with Smart Grids to Achieve Next-Generation Energy
Efficiency and Sustainability Project ICCS-NTU,. December 2009.
[14] Kelly McGuire, The Smart Grid Movement: Japan vs. U.S., TMCnet
News, Technology Marketing Corporation, January 2010, available at
http://smart-grid.tmcnet.com/topics/smart-grid/articles/73301-smart-
grid-movement-japan-vs-us.htm
[15] Nuclear Power Plants in Japan, Nuclear Power Generation,
Information Plaza of Electricity, The Federation of Electric Power
Companies of Japan, June 2011.
[16] Power Reactors, Nuclear Reactors, United States Nuclear Regulatory
Commission (U.S. NRC), May 2011, available at
http://www.nrc.gov/reactors/power.html
[17] Amy Poh Ai Ling, Masao Mukaidono, Selection of Model in
Developing Information Security Criteria on Smart Grid Security
System, Smart Grid Security and Communications, The Ninth
International Symposium on Parallel and Distributed Processing with
Applications (ISPA), No. 108, May 2011, Korea.
[18] Ryusuke Masuoka, Smart Grid: Japan and US, Fujitsu Laboratories of
America Inc., January 21, 2010.
[19] Japan Energy Conservation Handbook, 20042005, available at
http://www.eccj.or.jp/databook/2004-2005e/index.html
[20] Electricity in Japan, Japan Guide, available at http://www.japan-
guide.com/e/e2225.html
[21] Profile of Japans Major Power Generation Sources, Energy and
Electricity, Information Plaza of Electricity, The Federation of Electric
Power Companies of Japan, available at
http://www.fepc.or.jp/english/energy_electricity/electric_power_sources/
index.html
[22] Tomoko A. Hosaka, Japan Creating Smart City of the Future,
Associated Press Japan, October 2010.
[23] Nuclear Power Plants, World-Wide, European Nuclear Society,
available at http://www.euronuclear.org/info/encyclopedia/n/nuclear-
power-plant-world-wide.htm
[24] Rolling Blackouts as Japanese Efforts Continue, Regulation and
Safety, World Nuclear News, March 2011.
[25] Phil Carson, Smart Grid More Attractive, Post-Japan?, Intelligent
Utility, Mar. 2011, available at
http://www.intelligentutility.com/article/11/03/smart-grid-more-
attractive-post-japan
[26] Masaya Yasui, Japanese Policy on Photovoltaic Systems, ANRE,
METI, Japan, October 2010, pp. 47.
[27] Aiming at Developing the Worlds Highest Efficiency Concentrator
Photovoltaic Cells, New Energy and Industrial Technology
Development Organization, European Commission, European Union,
May 2011, available at http://www.nedo.go.jp/content/100147758.pdf
[28] Sunetric Offers Free Solar PV Systems to Raise Money for Japan,
Cleantech News, April 2011.
[29] Martyn Williams, A Legacy from the 1800s Leaves Tokyo Facing
Blackouts, Computerworld, Retrieved March 2011.
[30] Hiroshi Watanabe, Smart Community Activities in Japan, Korea
Smart Grid Week, November 2010, available at
http://www.ksgw.or.kr/down/pr/KSGW_11_HiroshiWatanabe(101110).
pdf
[31] Kitakyushu Citys Challenge Toward a Low-Carbon Society,
Kitakyushu Smart Community Creation Project, Green Frontier, 2011,
available at http://www.challenge25.go.jp/roadmap/media/s5_en.pdf
[32] Four Companies Start Smart Grid Demonstration Project in Rokkasho,
Technology, Japan Today, September 2010.
[33] HawaiiOkinawa Partnership on Clean and Efficient Energy
Development and Deployment, METI, June 2011.
[34] Smart Grid of the Future Urban Development, Japanese Nikkan
Kogyo Shimbun, June 2010.
[35] Toyota City Low-Carbon Project Model Homes Completed, Asahi,
June 2011.
[36] Japan to Help Gujarat in Smart Community Project, Rediff, January
2011,.
[37] Living Off the Grid, Cover Story, Highlighting Japan, February 2011,
available at http://www.gov-online.go.jp/pdf/hlj/20110201/1213.pdf
[38] U.S. and Japan Companies Collaborate on Smart Grid Project in
Hawaii, Sharp Corporation, May 2011, Press Releases, available at
http://sharp-world.com/corporate/news/110517.html
[39] US and Japan Collaborating on Smart Grid Project in Hawaii; EV
Operation and Charging, Including Grid-Balancing Services, Green Car
Congress, Energy, Technologies, Issues and Policies for Sustainable
Mobility, May 2011, available at
http://www.greencarcongress.com/2011/05/hitachi-20110518.html
[40] US and Japan Work on Maui Smart Grid for Electric Vehicles,
Sustainable Transport News, Brighter Energy News, May 2011.
[41] Katie Fehrenbacher, The New Smart Grid Players: Korea, Japan, China,
Oh My!, GigaOM, WordPress, February 2010.
Vol. 3, No. 7, 2012

54 | P a g e
[42] Japanese, Korean Utilities Join Smart-Grid Coalition, Greenbang, May
2011, available at http://www.greenbang.com/japanese-korean-utilities-
join-smart-grid-coalition_16971.html
[43] Guido Bartels, Smart Grids Progress Extends Beyond the Boundaries
of Countries, Special InterviewPart 5 of 5, Guido Bartels, Chairman of
Grid Wise Alliance, Talks About Smart Grid, Renesas, October 2010.
[44] Japan Smart Community Alliance, June 2011, available at
http://www.globalsmartgridfederation.org/japan.html
[45] Shinsuke Ito, Japans Initiative on Smart GridA Proposal of Nature
Grid, Information Economy Division, Ministry of Economy Trade and
Industry, December 2009.
[46] Nine Japanese Companies Launch Japan-U.S. Collaborative Smart Grid
Demonstration Project in Business District of Albuquerque, New
Mexico, The Wall Street Journal, pp. 1-2, May 2012.
AUTHORS PROFILE
Amy Poh Ai Ling received her BBA and MSc from
National University of Malaysia (UKM). She received her PhD
in Mathematical Sciences from Meiji University. She was
awarded Role Model Student Award (2003) and Excellent
Service Award (2010) from UKM, and Excellent Student
Award (2012) from Meiji University. She worked at Sony
EMCS and Erapoly Sdn. Bhd. She is currently a postdoctoral
affiliate with Meiji Institute for Advanced Study of Mathematical Sciences as
JSPS Research Fellow. She has an enthusiasm for statistical calculation, smart
grid and safety studies.
Professor Sugihara Kokichi received his Masters and
Dr. Eng. degrees from University of Tokyo. He worked at
Electrotechnical Laboratory of the Japanese Ministry of
International Trade and Industry, Nagoya University and
University of Tokyo before joining Meiji University. His
research area is mathematical engineering, including
computational geometry, computer vision, computer graphics and robust
computation. He is currently the leader of CREST research project of Japan
Science and Technology Agency on Computational Illusion.
Professor Mukaidono Masao served as a full-time lecturer
at Faculty of Engineering, Department of Electrical Engineering
in Meiji University from 1970. Even since then, he was
promoted to Assistant Professor on 1973 and as a Professor on
1978. He contributed as a researcher in an Electronic Technical
Laboratory of the Ministry of International Trade and Industry
(1974), Institute of Mathematical Analysis of Kyoto University (1975) and as a
visiting researcher at University of California in Berkeley (1979). He then
became the Director of Computer Center (1986) and Director of Information
Center (1988) in Meiji University. At present, he is a Professor and Dean of the
School of Science & Technology, Meiji University. He is also the honourable
Councillor of Meiji University.

Vol. 3, No. 7, 2012

55 | P a g e
Evaluating the Role of Information and
Communication Technology (ICT) Support towards
Processes of Management in Institutions of Higher
Learning
Michael Okumu Ujunju
Dr. G. Wanyembi
Mr. Franklin Wabwoba
Department of Computer Science
Masinde Muliro University of Science and Technology

Abstract The role of Information and Communication
Technology in achieving organizations strategic development
goals has been an area of constant debate, and as well perceived
in different management dimensions. Most universities are
therefore employing it (ICT) as a tool for competitive advantage
to support the accomplishment of their objectives. Universities
are also known to have branches or campuses that need strong
and steady strategic plans to facilitate their steady expansion and
growth. Besides, production of quality services from the various
levels of management in these universities requires quality
strategic plans and decisions. In addition, to realize the steady
growth and competitive advantage, ICT not only has to be an
additive but a critical component towards supporting
management processes in the universities. This research sought to
determine the role of ICT in supporting management processes in
institutions of higher learning in Kenya. The research
investigated how the different levels of management used ICT in
their management processes and whether the use had any effect
on management processes. The research further made
recommendations to the universities on better use of ICTs in their
management processes. A public university in Kenya was used as
a case study in this research.
Keywords- ICT; Competitive advantage; Strategic management;
Data.
I. INTRODUCTION
Information and communication technology (ICT) has
become an important tool in modern management of
universities. This is because information is a critical tool in
facilitating management decisions and therefore, ICT is seen
to be a crucial tool to help in facilitating acquisition of this
information required in management decisions for universities.
Manual and mechanical systems can no longer cope in the
current demands of management processes in universities.
This is due to the fact that accurate and timely data is a critical
resource in planning and decision making (Acosta, 2004). To
achieve this therefore, ICT must come in handy to facilitate
the process of acquiring accurate and summarized data needed
to facilitate management processes. It is in this context that
there is dire need for a successful transition to the new
technology in order for university managements to engage in
quality management practices. Intranet and database systems
are key components in the formation of ICT infrastructure, and
hence their existence enhances greater management control by
enabling departments in universities and their campuses to
have greater access to information needed for management
processes. This will enable them (managements) to function
more effectively and efficiently, and since projections will be
more accurate or now available, university managements can
make long-term strategic plans during their management
processes, (Nyandiere, 2007). Most of organizations endeavor
to employ Information Communication Technology as a tool
for competitive advantage for the accomplishment of the
objective of organization as well as enhance the alignment
between Information Communication Technology and
management strategy, (Mohammed, 2010). To achieve the
former, ICT has been leverage to improved service and lower
the cost of conducting strategic management functions.
In the modern world of technology in which Universities
exist, they (Universities) need ICT services since it plays a
significant strategic role in the management of Universities.
ICT makes it easy for a Universitys decision making because
they have information at hand. With the aid of ICT, university
managers have more information at their reach than ever
before; modern ICT improves good organization and
usefulness at each stage of the management decision making
procedure. It is therefore imperative for Universities to take
ICT seriously for the purpose of sustaining steady growth as a
result of making quality and timely strategic plans. Without
adequate level of ICT use, chances of making poor and
untimely management decisions will prevail and this will
constantly lead to the universities stunted growth.
II. THEORETICAL FRAMEWORK
The research adopted the theory of organizational
Information Processing developed by (Galbraith, 2005). The
theory identifies three important concepts: information
processing needs, information processing capability, and the
fit between the two to obtain optimal performance in
organizations. According to the theory, organizations need
quality information to cope with environmental uncertainty
Vol. 3, No. 7, 2012

56 | P a g e
and improve their decision making. Environmental uncertainty
stems from the complexity of the environment and dynamism,
or the frequency of changes to various environmental
variables. The theory further postulates that, organizations
have two strategies to cope with uncertainty and increased
information needs for their management processes: (1)
develop buffers to reduce the effect of uncertainty, and (2)
implement structural mechanisms and information processing
capability to enhance the information flow and thereby reduce
uncertainty. The theory is presented as a model in Figure 1.
As it can be seen from the model, organization design
strategy has sub units. These sub units require an integrated IT
system that will improve information flow and reduce
uncertainty within organizational sub units. Increasing the
capacity to process information required in management
processes will require (1) an investment in vertical information
systems and (2) creation of lateral relations to portray an
image of how the levels of management will interact in the
management processes. Creation of slack resources and self-
contained tasks will reduce the need for information
processing after a satisfactory confirmation that indeed output
of a process doesnt require further intervention of an
information processing system. Increasing the capacity to
process information and reducing the need for information
processing are products aimed at fulfilling management goals
which are key entities of management.

Fig I. Organizational Information processing model
Source: Galbraith (2005)
A lot was borrowed from this model for this research,
especially when we talk of integrating ICT in management
processes of universities in Kenya, and understanding that ICT
can indeed play greater roles towards improving the processes
of management in universities. The model was adopted
because; accurate and timely information is a key element in
management processes. Acquiring this information requires a
processing model which cannot be complete without focusing
on Information and Communications technology
infrastructure.
III. REVIEW OF RELATED LITERATURE
A. ICT in Management of Higher Learning Institutions
Through the emergence of fast and powerful computers,
networks and infrastructure, delivery of immediate and
relevant information enables policymakers in an organization
to make quick and accurate decisions,
(Newmann, 1994). (Laudon, 2003) while commenting on
the role of information systems in organizations indicates that
ICTs provide tools for data collection, analysis, storage and
dissemination to support decision making in organization.
University environments are equally changing in the
technology front. Arising from his studies of Universities in
Philippines (Asia), [1] notes that quick and accurate decisions
of institutions of higher learning managers require readily
available and relevant information thus making ICT a vital
tool in todays business world providing tools for information
collection, storage, and management to facilitate
communication and decision making processes. He points out
that institutions of higher learning too, must cope with the
emerging trends of competing on the ICT platform, thus they
need to continually assess their current status, and that of their
competitors to formulate and manage their own strategies if
only to stay abreast with the latest challenges of the
information age. ICTs play and will continue playing an
important role in higher education institutions (HEIs)
management.
(Katz, 2001) quotes EDUCAUSE president, Brian
Hawkins, who in 1999, in his paper Technology, Education,
and the Very Foggy Crystal Ball asserted three propositions
about the impact of ICT on higher education, that is;
a) That the new technology affords exciting
opportunities for more effective teaching;
b) That the new technology offers scalability that is
greatly needed;
c) That the new technology will transform higher
education beyond what we know it to be today.
Technology has provided exciting opportunities for
teaching and management including the recent e-learning
initiatives in addition to transforming HEIs management
operations. ICTs in higher educational institutions have come
about from developments in corporate businesses where ICTs
have been incorporated into organizational functions to
improve their performance. As (Tusubira & Mulira, 2005),
having extensively studied operations of Makerere University
(Uganda), argue that at the organizational level, the integration
of ICT in organizational functions has been brought about by
three main factors: increased efficiency, cost effectiveness,
and competitiveness.
IV. RESEARCH METHOLOGY
The research was conducted through case study. The study
was concerned with determining the role played by
information and communication technology in supporting
management processes in institutions of higher learning.

4.Creation
of slack
resources
5.Creation of
self-contained
tasks
6.Investments in
Vertical
Information
systems
7.Creation
of lateral
relations
Organization design strategy
Reduce the need for
Information processing
Increase the capacity
to process information
1. Rules and programs
2. Hierarchical referral
3. Goal setting
Vol. 3, No. 7, 2012

57 | P a g e
The sample size used was put into strata or groups, each
stratum representing each of the three levels of management.
Purposive sampling was then used where a group of people
believed to be reliable enough for the study were targeted.
Questionnaires were used to collect data from staff in each
of the three levels of management. They included both open-
end and closed-end questions. They were physically
administered by the researcher to 107 respondents out of
which 70 were responded to. Unstructured interviews were
also conducted with the senior university management staff.
All data collected were analyzed using qualitative and
quantitative approaches which were facilitated using Statistical
Program for Social Science (SPSS) version 17.0. Cross
tabulations showing percentages and frequency distributions
were used to analyze data collected.
V. FINDINGS AND DISCUSSION
The demographic information of respondents was
presented and discussed based on gender and the possibility of
having used ICT tools like a computer before for each
respondent interviewed. This information is presented in tables
and the figures.
Respondents were asked of their sex and close observation
shows that there is a significant variation in the distribution by
sex of staff as shown in Table 1. Findings in table 1 reveal
that, male respondents were majority (60%) compared to
female respondents (40%). These findings suggest that, there
is a gender gap between women users of ICT and men, and
that women are not well accessible to ICTs as compared to
men. There is therefore need to mainstream gender concerns
into the ICT arena by providing opportunities for ICT training
and develop clear policies, guidelines and strategies to remove
this gender gap that disadvantages women from accessing
ICTs in institutions of higher learning.
In determining whether respondents had used ICT in their
work procedures before, they were asked to state whether they
had used ICT tools like a computer before. Results in Table II
show that, many staffs (98.6%) had used ICT tools. They were
therefore not new to the technology. Knowledge of ICT is a
factor that seemed to motivate employees to embrace and
positively integrate ICT in management processes and work
procedures.
Views on how ICT tools were used by staff in carrying out
their duties is as shown in Table III. The various ways in
which ICT is used in processes of management as established
by the findings are supported by
(Ruiz-Mercader, 2006). In his statement he acknowledges
that the best ways organizations use ICTs is to obtain, to
report, to process, to accumulate and to exchange information.
Furthermore in a knowledge management context, ICT can
support transformation within and between tacit and explicit
knowledge.
Views on how ICT tools had the impact on overall work
performance of staff in different levels of management is as
shown in Table IV.
TABLE I. RESPONDENTS GENDER
Frequency Percent
Valid
Percent
Cumulative
Percent
Valid male
42 60.0 60.0 60.0
female
28 40.0 40.0 100.0
Total
70 100.0 100.0

TABLE II. RESPONDENTS VIEW ON USE OF ICT TOOLS
Frequency Percent
Valid
Percent
Cumulative
Percent
Valid yes
69 98.6 98.6 98.6
no
1 1.4 1.4 100.0
Total
70 100.0 100.0
TABLE III. ICT USE IN DUTIES OF MANAGEMENT

Frequency Percent
Valid
Percent
Cumulative
Percent
Valid writing reports
22 31.4 31.9 31.9
generating
information
20 28.6 29.0 60.9
leisure/games
4 5.7 5.8 66.7
general typing
14 20.0 20.3 87.0
E-mail and
internet
9 12.9 13.0 100.0
Total
69 98.6 100.0
Missing 99
1 1.4
Total
70 100.0

TABLE IV. IMPACT OF ICT TOOLS ON OVERALL WORK
PERFORMANCE OF STAFF IN DIFFERENT LEVELS OF
MANAGEMENT

Frequency Percent
Valid
Percent
Cumulative
Percent
Valid motivate to
maximize
potential
24 34.3 34.3 34.3

de-
motivates
and reduce
performance
4 5.7 5.7 40.0

simplify
work and
makes it
easier
28 40.0 40.0 80.0

facilitate
acquisition
of realistic
plans
5 7.1 7.1 87.1

quality
results from
staff
9 12.9 12.9 100.0
Total
70 100.0 100.0
Vol. 3, No. 7, 2012

58 | P a g e
VI. SUGGESTION FOR FUTURE RESEARCH
This research was conducted in a public university and the
results may not be generalized to private universities due to
different operational dynamics. It would be worthwhile if
further research was conducted to establish if the same results
would work for private universities in Kenya.
VII. CONCLUSION
The research determined and evaluated the role of ICTs in
supporting processes of management in institutions of higher
learning in Kenya. There was a gender gap between women
users of ICTs and men in their work of management as per the
results in Table 1, which showed 40% for women users and
60% for men. There is total need for this gap to be removed.
Views of respondents on the use of ICT tools were very
positive (98.6%). This shows that ICT use in management is
highly embraced and this improves management practices.
From Table 3, it could be seen that ICT tools are used in a
number of ways in supporting duties of management. This is
an indicator of positive acceptance by staff on embracing ICTs
in their duties of management. The use of ICT in works of
universities management is observed that indeed, it simplifies
work and makes it easier for universities staff to enjoy their
work and hence generate quality decisions for the running of
their universities.
RECOMMENDATIONS
This research gives the following suggestions as
recommendations for better use of ICTs in processes of
management in institutions of higher learning;
1) Provide opportunities for ICT training and develop
clear policies, guidelines and strategies for better use of ICT
equipment by all, regardless of sex.
2) All affected users should be trained properly on any
new upcoming software or computer hardware constituting
ICT infrastructure.
Universities should use current ICTs technologies as
possible in all areas of operations so that to maintain
consistency in their modern management practices if quality
management has to be maintained.
3) Use ICT resources only for authorized purposes to
avoid abusing the resources in process of executing duties.
4) ICT resources can be shared. There is needed to be
considerate in the use of shared resources. Refrain from
monopolizing systems, overloading networks with excessive
data, degrading services, wasting ICT resource time, disk
space, manuals or other resources.
REFERENCES
[1] Acosta F.R. (2004), Information Technology Strategic Plan of Olivarez
College unpublished doctoral dissertation, University of Baguio,
Philippines.
[2] Nyandiere, C. (2007). Increasing role of computerbased information
systems in the management of higher education institutions. In M.
Kashorda, F. Acosta and C. Nyandiere (eds). ICT Infrastructure,
Applications, Society and Education: Proceedings of the Seventh
Annual Strathmore University ICT Conference. Strathmore University
Press: Nairobi
[3] Mohammed, A. H., Altemini, M.S, & Yahya Y. (2010). Evaluating the
performance of Information Technology on Strategic Planning,
International Journal of Education and Development using ICT
(IJEDICT).
[4] Galbraith, J.R. (2005) Designing Complex Organizations. Reading.
MA: Addison-Wesley
[5] Newmann, S. (1994). Strategic information systems Competition through
information technologies. New York:Macmillan.
[6] Laudon, K.C. and Laudon J.P.(2003): Management Information Systems:
Managing the Digital Firm, 7th ed., New Jersey: Prentice-Hall.
[7] Katz, R. N. (2001) The ICT Infrastructure: A Drive for Change
EDUCAUSE Review.
[8] Tusubira,F F & Mulira, N. (2005). Integration of ICT in Higher Education
Institutions: Challenges and best practice recommendations based on
experience of Makerere University & Other organisations. Makerere
University.
[9] Ruiz-Mercader J., Merono-Cerdan A. L., Sabater-Sanchez R. (2006),
Information technology and learning: Their relationship and impact on
organisational performance in small businesses, International Journal
of Information Management, Volume 26, Issue 1, February 2006, pp. 16-
29.
Vol. 3, No. 7, 2012
59 | P a g e
Smart Grids: A New Framework for Efficient Power
Management in Datacenter Networks

Okafor Kennedy .C, Udeze Chidiebele. C
2,3
R & D Department, Electronics Development Institute
(FMST-NASENI), Awka, Nigeria.
1
Electrical Electronics Engineering, Federal University of
Technology Owerri,, Nigeria.

E. C. N. Okafor, C. C. Okezie,
4
Electronics and Computer Engineering Department, Nnamdi
Azikiwe University, Awka, Nigeria.

Abstract The energy demand in the enterprise market segment
demands a supply format that accommodates all generation and
storage options with active participation by end users in demand
response. Basically, with todays high power computing (HPC), a
highly reliable, scalable, and cost effective energy solution that
will satisfy power demands and improve environmental
sustainability will have a broad acceptance. In a typical
enterprise data center, power managment is a major challenge
impacting server density and the total cost of ownership (COO).
Storage uses a significant fraction of the power budget and there
are no widely deployed power-saving solutions for enterprise
storage systems. This work presents Data Center Networks
(DCNs) for efficient power management in the context of SMART
Grids. A SMART DCN is modelled with OPNET 14.5 for
Network, Process and Node models. Also, an Extended SMART
Integration Module in the context of SMART DCN is shown to be
more cost effective than the traditional distribution grid in DCNs.
The implementation challenges are discussed also. This paper
suggests that smartening the grid for DCN will guarantee a
sustainable energy future for the enterprise segments.
Keywords- energy; density; enterprise; power budget; framework
smart grids.
I. INTRODUCTION
Contemporarily, most discussions and professional
conferences now focus on the power industry and its resurgent
energy challenges. It is a worldwide goal to optimize energy
demand, consumption and CO2 emissions [1] in all critical
sectors of any economy. Energy reduction in enterprise setups
is a major concern to operators and regulators all over the
world. A part of this energy reduction scheme concerns the
telecommunication industry and ICT that participates in a
direct, indirect and systematic ways [2]. Characteristic
examples are green networks, smart buildings, smart grids,
Intelligent Transportation Systems (ITS), energy efficient
electronics (OLEDS, photonics, nanotechnology) and the
application of embedded systems towards low carbon and
energy efficient technologies [2],[3].
The IT/Telecommunication industry suffers from myriads
of persistent power challenges in the electricity power supply
as well as optimizing the available energy reserves. This
research suggests that a power solution that will help this
critical sector to manage demand growth, conserve energy,
maximize asset utilization, cost and improve grid security
reliability as well as reduce carbon foot print will go a long
way in solving the persistent power issues faced by the
enterprise market segments. In our context, a new proposal for
power management in an energy efficient data centers (fixed
broadband networks) and its challenges form the basis for this
paper.
According to [4], a Data Centre is the consolidation point
for provisioning multiple services that drive an enterprise
business processes. It is also known as the server farm or the
computer room. The data center is where the majority of
enterprise servers and storage systems are located, operated
and managed like the Enterprise resource planning solutions
(ERPs), application servers, security systems (IDS). It
generally includes redundant or backup power supplies,
redundant data communications connections, environmental
controls (e.g. Air conditioning, fire suppression) and security
devices. DCNs have attracted a lot of interest in todays
enterprise network segments [5]. The work in [6] proposed a
technology to fix the energy demands for reliable delivery in
DCNs. However, the requirement to serve the Data Center
environment mandates that a new grid framework with
enhanced performance characteristics , hence, this paper
proposes SMART GRID model for efficient power
management in DCNs.
The paper is organised as follows. Related work is
summarised in Section 2. Data center networks and energy
requirements are discussed in this section. We presented Smart
Grids Proposal for DCNs in Section 3. Smart Grids Reference
Model is discussed in Section 4. Section 5 presents a cost
effective distribution model for DCNs. Section 6 discussed the
challenges of SMART GRID Implementation and finally
conclusions and future work are presented.
II. RELATED WORKS
Following the definition in [4], various proposals have
been presented on how to optimize performance in DCNs. The
work done in [1] gave a review of energy efficiency in
telecommunication networks while concentrating on data
center, fixed line and cellular networks. Representative sample
of literatures on DCN traffic, design methodology and
management were studied in [6], [7], [8], [9]. According to the
EPA report in [7], the two largest consumers of electricity in
the data center are:
Vol. 3, No. 7, 2012
60 | P a g e
Support infrastructure 50% of total.
General servers 34% of total.
Since then, significant strides have been made to improve
the efficiency of servers in DCNs. High density blade servers
and storage are now offering much more compute capacity per
Watt of energy. Server virtualization is allowing organizations
to reduce the total number of servers they support, and the
introduction of Energy Star servers have all combined to
provide many options for both the public and private sectors to
reduce that 34% of electricity being spent on the general
servers.
Essentially, the outline in [9], presents a holistic power
design considerations for DCNs. According to the EPA,
most data centers consume 100% to 300% of additional power
for the support systems than are being used for their core IT
operations. However, through a combination of best practices
and migration to fast-payback facility improvements, this
overhead can be reduced to about 30% of the IT load. Within
corporations, the focus on power management in data centers
is driven primarily by a desire to reduce the tremendous
electricity costs associated with operating a data center. Hence,
going green is recognized as a way to reduce operating
expense significantly for the IT infrastructure.
In general, the telecommunication sector accounts for
approximately 4% of the global electricity consumption [10].
Figure 1 shows a typical DCN.

Fig. 1: A Typical Data Center Network
From fig.1 The power consumption of a data center is
related to the associated power consumed by each module in
the DCN. Efficiency at individual parts is an important step for
smartening the data center but optimization is achieved when
the efficiency is aimed at the overall data center design.
The power distribution in a typical data center is divided
into an in-series path and an in-parallel path to feed the
switchgear and the cooling systems, respectively [1]. Thermal
heat at the switchgear, UPS and PDUs, creates losses owing to
AC/DC/AC conversions. As shown in fig.1, the parallel path
feeds the cooling system for heat protection in the data
center. The power consumption at different layers of the data
center is presented in Table 2. it is observed that the power
consumption pattern in a DCN is dependent on the input
workload to the data center and the surrounding environmental
characteristics. The authors in [11] presented a power saving
estimation scheme; correlation aware power optimization
(CARPO) which achieves energy savings mainly by turning
off unused switches after conducting correlation-aware
consolidation. The work explains that the dynamic power in a
DCN is relatively much smaller compared with the static
power consumption of the network devices. As such, for a
given DCN topology and workloads, it is possible to
approximately estimate the energy savings of a DCN using a
correlation aware power optimization scheme that dynamically
consolidates traffic flows onto a small set of links and switches
in a DCN and then shuts down unused network devices for
energy savings. By their empirical measurements, the power
consumption model of the entire network in their testbed is
given by:
P = 6.67.Ns +
3
i=1
(Pi.Ji) (1)
where Ns is the number of active virtual switches, Pi is the
active power of a single port at the data rate level i and Ji is the
corresponding number of active ports at that data rate level.
This work argues that energy redundant devices in a DCN
can not be adequately taken care of by CARPO owing
various sources of estimation errors in the conventional power
management approach. Fig. 2a and Fig. 2b presents DCN
network and nodel models for different functionalities of the
network .The DCN model was splitted into core, aggregation
and access layers. The energy consumption is higher at the
access part of the network. Also, at the core layer which
provides for data centers computations, storage, applications
and data transfer, the drain is considerable. We observed that
the backbone (aggregation) networks present lower energy
demands.On these basis, this work argues that an energy
efficient architecture should focus on intelligent and efficient
access techniques that is efficient in its operation. Hence, we
propose a SMART grid model with optimized power
management. Basically, the main functionalities of the
SMART DCN includes regeneration, transportation, storage,
routing, switching and processing of data. As ealier stated, the
largest part of energy is consumed for routing/switching,
regeneration and processing of data. Both communication
protocols and electronic devices are responsible for this
consumption and this imposes challenges for more
sophisticated transport techniques, thermal removal from
switches or the servers and less redundant data transfers. A
characteristic example of energy efficiency in DCN
equipments is shown in Table 1.
The proposed model in this work is self configurable and
stable under all load conditions. Since the power consumption
of a DCN is largely dependent of its workload, mainly because
the network equipment, it now becomes very imperative to
adapt SMART grid based DCN so as to put network devices,
such as switches and routers, into a sleep state during periods
of very inactive traffic activities, and buffer the packets during
the inactive period to transmit in future time. This will
gurantee return on investment in todays typical multi-tiered
DCNs. In this case, high degrees of oversubscription on the
network devices will result to good energy savings for the
Telecom
ms Equip
Data Center
Series
Parallel
Servers

Storage
Networ
king
Switch gear
Lights
Cooling
UPS
PDU
Main
Energy
Supply
Emergenc
y Energy
Supply
Renewabl
e Energy
Supply
Vol. 3, No. 7, 2012
61 | P a g e
DCN. As a result, frequently putting network devices into
cold state under low traffic intermittently in a DCN will
cause packets to be buffered around the deactivated DCN
devices and their paths.
The framework in fig.1 serves as a good energy svaing
model but lacks the capacity to handle energy drain in
redundant components.

Fig. 1b. The CARPO framework [11].
TABLE 1: POWER EFFICIENCY OF DCN EQUIPMENTS.
Equipments Power Efficiency (W/Gbps)
Router 40
IP Switch 25
Transport TDM 80
ATM Switch 80

Figure 2a: DCN Model for SMART Grid Integration
The power consumption at different layers of the data
center is presented in Table 2.
It can be observed that the useful work of data center is
associated to a percentage of power, smaller than the 30%
delivered to the IT equipments.
In Fig 2a. We modelled a SMART DCN with OPNET
modeller 14.5 as well as generating various node and process
models for the setup.

Figure 2b: DCN Process Model for SMART Grid Integration
Since, this work is not centered on performance analysis,
we rather observed the system response after configuring the
power parameters as well as the traffic parameters.
The proposed model in Fig. 8b, is believed to address
energy demands of figure 2a.

Traffic
Flows
Ports/switches to
be Turned Off
Data rate of each link
CARPO: Correlation-Aware Power Optimization
Correlation Analysis
Traffic Consolidation Link Rate Adaptation
Data Center Network (DCN)
CC= Correlation Coefficient ; AL= Active Link

cc
C
AL
C
Vol. 3, No. 7, 2012
62 | P a g e
TABLE 2: POWER WASTE DISTRIBUTION IN TYPICAL DATA
CENTERS.
NCPI Equipments Percentage of power consumption
[Total 70%]
Chiller
CRAC
UPS
PDU
Switchgear, Light
33
9
19
5
4
IT Equipments Percentage of power consumption
relative to [Total 30 %]
System
Disks
Power Supply
Networking
CPU
Memory
25
5
13
9
40
8

Figure 2c: A Node Model For SMART DCN
III. ENERGY EFFICIENCY IN DCN
In a SMART DCN model, configuring a network to
operate in an efficient manner is a complex task. However,
optimizing energy consumption and total network
optimization is considered very critical for power management
in DCN. For a network to work in an energy efficient way
enhancing environmental sustainability, it must create
flexibility for the deployment of future networks to off grid
areas that rely on Renewable Energy Sources (RES) and
sensor networks that rely on battery power supply. Minimizing
power consumption on the DCN will have a great effect on the
cost of operation of a network and this makes it more
affordable in general.
This paper opions that network optimization in terms of
energy efficiency can be achieved by providing the following
key metrics viz: efficiency to network dimensioning,
efficiency in network processes, efficiency at the access
network,etc for better power management of the equipments
as shown in Fig.3. Optimization of DCN equipments is the
first step for an energy efficient network. This will be taken
care of by the SMART DCN architecture which has provision
for low power IT devices and equipments, and efficient
battery technology. In addition, recycling of equipments is
considered as a valuable solution for energy efficiency as well.

Fig. 3: Main factors of energy efficient networks.
DCNs infrastructure (Data centers and servers) constitute
critical components of networks providing data processing,
storage, regeneration, etc. A metric for energy efficiency of
data center is the Data Center Infrastructure Efficiency (DCIE)
and the Data Center energy Productivity (DCeP) [15].
According to [15], DCIE expresses the fraction of the total
power supplied to the data center and is delivered to the IT
load.
A SMART energy efficient data center requires
operational and planning actions that will correspond to the IT
equipment needs and incorporate the use of energy
proportional servers, such as blade servers with good cooling
units. Also, the planning actions again will consider the
exploitation of virtualization, remote monitoring and
management of the data center,while using vertical rightsizing
procedures and actions to reduce cooling needs through
optimal design of the DCN domain. Finally, by using
SMARD grid architure in DCNs, power management will
grantee zero down time.
IV. DETERMINING DATA CENTER CAPACITIES
In designing the SMART data center capacity, it will
involve many different variables that includes: the housing
infrastructure, network feeds necessary to keep the DCN
functional, the storage and hardware processing units.
Balancing all of these variables to design a data center that
meets the project scope and keeps the DCN in constant
operation is very essential for power management of a
planned DCN. According to [9], the design of data centers is
dependent on the balance of two critical sets of capacities:

Energy
Efficiency
Processes
Remote Monitoring
Efficient Access
Network
Renewable
Energy
Dimensioning
(planning) Efficient
Network
Processes
Efficiency
Electronic
Equipments
Vol. 3, No. 7, 2012
63 | P a g e
1. Data center capacities: Power, cooling, physical
space, weight load, bandwidth (or connectivity), and
functional capacities.
2. Equipment capacities: The various devices (typically
equipment in racks) that could populate the data center in
various numbers
The knowledge of the equipment requirements was used to
determine the size of the center, the amount of power and
cooling needed, the weight load rating of the raised floor, and
the cabling needed for connectivity to the network. Hence,
data center size as well as in-feeds determines how much
equipment to be deployed in Fig. 2a. The work in [9] proposes
a new method for designing a data center based on the critical
capacities called rack location units (RLUs). The actual
process of defining RLUs to determine the capacities of a data
center boils down to careful planning which enhances
flexibility. The Rack Location Unit (RLU) system is a
completely flexible and scalable system that can be used to
determine the equipment needs for a data center of any size,
whether 100 or 100,000,000 square feet. RLUs are defined
based on specific DCN device requirements. These
requirements are the specifications that come from the
equipment manufacturers. These requirements are:
1. Power (how many outlets/circuits it requires, how
many watts it draws)
2. Cooling (BTUs per hour that must be cooled)
3. Physical space (how much floor space each rack
needs, including the cooling dimensions)
4. Weight (how much a rack weighs)
5. Bandwidth (how it connects to the network)
Functional capacity (how much computational power,
physical memory, disk space, as well as how many spindles,
MFLOPS, database transactions, and any other measures of
rack functions).
However, this work only seeks to adapt smart grid
functionality into DCN for efficient power management
(reliability, improvement, loss control, and cost optimization)
while presenting its economic implications.
V. SMART GRIDS PROPOSAL FOR DCNS
As shown in figure 2a, we seek to adapt smart grid to
monitor and manage the transport of electricity from all
generation sources to meet the varying electricity demands of
the SMART DCN. The Smart grid framework co-ordinates the
needs and capabilities of the SMART DCN, minimizing costs
and environmental impacts while maximizing system
reliability, resilience and stability. In context of this paper,
smart grids electricity system (transmission and distribution)
powers the storage (SAN) module of the DCN. Fig. 4 and Fig.
5 show our proposed DCN Smart Grid Model and Smart Grid
components.
By leveraging on Smart Grids that improves the efficiency,
reliability, economics, and environmental sustainability, an
efficient Power management in DCN can thus be achieved.
The DCNs, with the integration of SMART grids will now :
1. Increase reliability, efficiency and safety of the power
grid.
2. Enable decentralized power generation thereby
aloowing DCNs effectively manage energy usage.
3. Allow flexibility in power consumption at the
DCNs side to allow supplier selection (distributed
generation, solar, wind, and biomass).
4. Increase GDP by creating more new, green-collar
energy jobs related to renewable energy industry.
Fig. 4: DCN Smart Grid Model
Figure 5: Smart Grid Components
From Fig. 4, this paper recommends that regulators and
DCN operators streamlines the technical, financial and
regulatory details that will enable the maximization of the
potential of smart grids. From Fig. 5, smart grid is a highly
complex combination and integration of multiple digital and
non-digital technologies and systems which outlines the main
component of a smart grids viz : i) new and advanced grid
components, ii) smart devices and smart metering, iii)
integrated communication technologies, iv) decision support
and human interfaces, v) advanced control systems.
Transmiss
ion
Control
Center
Energy
Storage
Distribut
ion
Control
Center
Energy
Service
Provider
High
Temperature
Superconduct
or
Substatio
n
Electric
Vehicles
Substatio
n
Stora
ge
Smart Device
-Meter
Data Center
Domain
Decision Support &
Human Interfaces
New Advanced Grid
Components
Smart Devices:
Smart Meter
Advanced
Control
Systems
Integrated ICT
Smart Grid
Infrastructure
s
Vol. 3, No. 7, 2012
64 | P a g e
New and advanced grid components allow for efficient
energy supply, better reliability and availability of power. The
components includes: advanced conductors and
superconductors, improved electric storage components, new
materials, advanced power electronics as well as distributed
energy generation. Superconductors are used in multiple
devices along the grid such as cables, storage devices, motors
and transformers. The rise of new high-temperature
superconductors allows transmission of large amounts of
power over long distances at a lower power loss rate to DCNs.
Distributed energy can be channelled to the DCN to be served,
hence this will improve reliability, and can reduce greenhouse
gas emissions while wideing an efficient SMART energy
delivery to the DCN.
Smart devices and smart metering include sensors and
sensor networks. Sensors are used at multiple places along the
grid, e.g. at transformers, substations and at customers/DCNs.
They play an outstanding role in the area of remote monitoring
and they enable demand-side management and thus new
business processes such as real-time pricing. The design
technique for digital meters is influenced by three major
factors namely; desired device cost, efficiency and overall size
[21]. While the cost is influenced by users general
affordability, the efficiency and size must strictly comply with
standard. This work proposes the adaptation of Fig. 6 in our
DCN Smart grid model.

Fig. 6: Block Diagram of Digital Energy (Smart) Meter [21]
VI. SMART GRIDS REFERENCE MODEL
There are four major layers identified in SG integrations as
shown in Fig. 7. The SG decision intelligence (layer-4)
comprises of application programmes that run in relays,
intelligent electronics, substation automation systems (SAS),
control centers (CC) and enterprise zones. These programs
process the information collected from the sensors or
dispatched from layer-3 (ICTs) and then provide control
directives and business process decisions via layer 1 (Power
C-T-S-C). The advanced metering infrastructure (AMI)
creates a 2-way handshake platform for demand response and
other grid distribution roles. Layer-1 is primarily for energy
conversion, transmission, storage and consumption. This gives
the platform for DCN power interfacing. The traditional grid
energy distribution lacks storage, fast reactive and real power
regulation and distributed distribution. This accounts for
numerous power management challenges in DCN.
As depicted in Fig. 8b, the Extended SMART Integration
Module (ExSim) for DCN is presented to address the
limitations of Fig. 8a. In addition, it supports grid-feeder
connection, AC and DC bus systems, coordination and
optimization of energy supply. The Energy storage block
houses all available energy options (wind, hydro, and solar PV)
which provides for a short to medium-term power supply
buffer for uninterrupted DCN operation hence improving
reliability and reduction in energy losses.
VII. THE CHALLENGES OF SMART GRID IMPLEMENTAIONS
A smart grid strategy requires information integration
across autonomous systems. From IT perspective, some of it
challenges includes: Economic, Political and regulatory
policies: The regulators will need to understudy and take into
account the load demands in DCNs and come up with
environmental friendly green policies. In this context, existing
regulatory framework for DCNs needs to take care of future
scalability and match consumers intrests at large. Dynamics in
Technology in accordance with Moores Law: Unlike other
low carbon energy technologies, smart grids must be deployed
in both existing systems (which in some cases are over 40
years old) as well as within totally new systems and it must be
deployed with minimum disruption to the daily operation of
the electricity system [22]. Also, Despite Moores laws, most
countries still have epilleptic technologies that is still at a
very initial stage of development and are yet to be developed
to a significant level. As the technologies advances, it will
reduce the delivery and full deployment by vendors.
i. With legacy operational systems and business
processes, the challenge by stakeholders to
endorsement of and upgrade to SMART Grids
platforms will have to be addressed conclusively,
else this will constitute a major barrier.There is need
for DCN operators, and other consumers orientation
to understand power delivery intricacies before
venturing into implementation. Smart grid low
content carbon economy should be analyzed in the
context of environmental sustainability for all stake
holders in DCN as well as other users.
ii. The high cost of implementation of smart grid will
call for power deregulation to attract investors that
will generate funds for its rollout.
v Technical expertise (Need for extensive training and
orientation).

Fig. 7: layers in SG integrations
Decision
Intelligence
Integrated Communications
Sensor- Actuator
Power Conversion

Power Storage
Power
Consumption
Power Transport

Signal
Conditioner

PC
Voltage
sensor
Digital
Meter
Current
sensor
COM
Port
Vol. 3, No. 7, 2012
65 | P a g e
VIII. COST EFFECTIVE DCN DISTRIBUTION MODEL
Following the limitations of the tradition energy supply for
DCNs in Fig. 8a, this work proposes an optimized Smart Grid
model for DCNs as shown Fig. 8b.

Fig. 8a: Traditional Distribution System in DCN

Fig. 8b: Optimized SG model for DCNs (ExSim)
IX. CONCLUSIONS
This paper proposes SMART Grids as a better alternative
to DCN power management. The worlds electricity systems
face a number of challenges, including ageing infrastructure,
continued growth in demand, the integration of increasing
numbers of variable renewable energy sources, etc. There is an
urgent need to improve the availability, reliability, security of
supply as well as lowering carbon emissions for environmental
sustainability. Smart grid technologies offer better ways to
meet these challenges and also develop a cleaner energy
supply that is more energy efficient, more affordable and more
sustainable. Also, Extended SMART Integration Module for
DCN is proposed as a cost effective SMART GRIDS model.
Since the contemporary digital society depends on and
demands a power supply of high quality and high availability,
implementing SMART Grids for the enterprise market
segments yeilds a colossal gain.
FUTURE WORK
Smart grid is a new technlogy that seeks to revolutionize
the current electricity grids into a flexible decentraized service
network for operators and clients, future work will ascertain
the correlation between SMART grid acceptance index and
demand response coefficient. Also, drivers for SMART grids
like increase in demand and peak load increase will be
investigated. Various model validation experiments will be
analysed with MATLAB software in the future work.
REFERENCES
[1] G. Koutitas,, and P. Demestichas, A Review of Energy Efficiency in
Telecommunication Networks, Telfor Journal, Vol. 2, No. 1, 2010.
[2] Report on Climate Change, International Telecommunication Union
(ITU), Oct. 2008.
[3] H. Scheck, Power consumption and energy efficiency of fixed and
mobile telecom networks, ITU-T, Kyoto, 2008.
[4] http://ieee802.org/802_tutorials/Data-Center-Bridging-Tutorial-Nov-
2007-v2.pdfGigabitEthernet-Wikipedia.
[5] Okafor kennedy,

Udeze Chidiebele. C,. C,

Prof. H. C. Inyiama, Dr.
C.C okezie, Cloud Computing: A Cost Effective Approach to
Enterprise Web Application Implementation (a Case for Cloud ERP
Web Model),(Unpublished)
[6] Okafor K.C., SVLAN Approach to Congestion Management in Data
Center Networks, M.Eng Thesis , University of Nigeria, Nsukka, 2010.
[7] A. Greenberg, et al, Towards a Next Generation Data Center
Architecture: Scalability and Commoditization,(Unpublished).
[8] White Paper: Data Center Definition and Solutions.[Online]Available:
http://www.cio.com/article/print/499671, August 14, 2009.
[9] R. Snevely, Enterprise Data Center Design and Methodology,
Copyright 2002 Sun Microsystems.
[10] A. Gladisch, C. Lange, R. Leppla, Power efficiency of optical versus
electronic access networks, Proc. European Conference and Exhibition
on optical communications, Brussels, 2008.
[11] Xiaodong Wangy, Yanjun Yao, Xiaorui Wangy, Kefa Lu, Qing Cao,
CARPO: Correlation-Aware Power Optimization in Data Center
Networks, (Unpublished)
[12] L. A. Barroso, U. Holzle, The data center as a computer: An
introduction to the design of warehouse-scale machines, Morgan and
Claypool, ISBN:9781599295573, 2009.
[13] N. Rasmussen, Allocating data center energy costs and carbon to IT
users, APC White Paper, 2009.
[14] A.Vukovic, All-optical Networks Impact of Power Density, Panel
on Challenges of Electronic and Optical Packaging in Telecom and
Datacom Equipment, Maui, Hawaii, USA, July 2003
[15] U.S Environmental Protection Agency ENERGY STAR Program,
Report to Congress on Server and Data Center Energy Efficiency, Public
Law 109-431, page 94, August 2007.
[16] Smartgrids Advisory Council. "Driving Factors in the Move Towards
Smartgrids" (PDF). European Smart grids Technology Platform: Vision
and Strategy. European Commission. p. 9.ISBN 92-79-01414-5.
http://www.smartgrids.eu/documents/vision.pdf
[17] Smart Sensor Networks: Technologies and Applications for Green
Growth, December 2009
[18] Smartgrids Advisory Council. "Driving Factors in the Move Towards
Smartgrids" (PDF). European Smartgrids Technology Platform: Vision
and Strategy. European Commission. p. 9. ISBN 92-79-01414-5.
http://www.smartgrids.eu/documents/vision.pdf.
[19] National Energy Technology Laboratory (2007-07-27) (PDF). A Vision
for the Modern Grid. United States Department of Energy. p. 5.
Retrieved, 2008-11-27.
[20] DOE (U.S. Department of Energy) (2009), Smart Grid System Report
[21] M.C. Ndinechi, O.A. Ogungbenro, K.C. Okafor, Digital Metering
System: A Better Alternative For Electromechanical Energy Meter In
Nigeria International journal of academic research vol. 3. No.5.
September, 2011.
[22] White paper: Technlogy Roadmap. www.iea.org, 2011
AUTHORS PROFILE
Okafor Kennedy. C. is a Senior R & D Engineer. He has B.Eng & M.Engr in
Digital Electronics & Computer Engineering from the University of Nigeria
Nsukka, (UNN). He a Cisco expert and currently works with the ICT
department of Electronics Development Institute, ELDI. He is a member of
IEEE, IAENG and NCS. Email: [email protected]. ECN Okafor is
GS
Transformer
Transformer
DCN AC
Loads
DCN AC
Loads
Voltage
Regulators
Capacitors
AC Bus
System
AC Bus
System
ExSim
GS
DCN DC
Loads +N

Energy
Storage
DCN DC
Loads
DCN AC
Loads
DCN AC
Loads
DCN AC
Loads +N
Ac Bus
Dc Bus
Vol. 3, No. 7, 2012
66 | P a g e
a senior lecturer in the electrical electronics department of Federal University
of Technology Owerri, Nigeria. He is an expert consultant in power systems
engineering. His current research interest is on power systems engineering and
smart grid. E-mail: [email protected]
Udeze Chidiebele C. received his B.Eng and M.Sc in Electrical Electronics
Engineering with Nnamdi Azikwe University, Awka, Nigeria .He is a Senior
R & D Engineer with Electronics development Institute Awka, Nigeria and
also a member of NSE. He is currently running his Ph. D program in computer
and control systems engineering. His current research interest is on data center
networks, wireless sensor networks and control systems engineering. Email:
[email protected]
C.C Okezie is a senior lecturer in the electronic and computer department of
Nnamdi Azikiwe University Awka. She received her B. Eng, M. Eng, and
Ph.D from Enugu State University of Science and Technology, Enugu,
Nigeria. She majors in digital systems and control engineering and has many
publications to her credit. Email- [email protected].

Vol. 3, No. 7, 2012

67 | P a g e
An Online Character Recognition System to Convert
Grantha Script to Malayalam
Sreeraj.M
Cochin University of Science and Technology
Cochin-22, India
Sumam Mary Idicula
Cochin University of Science and Technology
Cochin-22, India

Abstract This paper presents a novel approach to recognize
Grantha, an ancient script in South India and converting it to
Malayalam, a prevalent language in South India using online
character recognition mechanism. The motivation behind this
work owes its credit to (i) developing a mechanism to recognize
Grantha script in this modern world and (ii) affirming the strong
connection among Grantha and Malayalam. A framework for the
recognition of Grantha script using online character recognition
is designed and implemented. The features extracted from the
Grantha script comprises mainly of time-domain features based
on writing direction and curvature. The recognized characters
are mapped to corresponding Malayalam characters. The
framework was tested on a bed of medium length manuscripts
containing 9-12 sample lines and printed pages of a book titled
Soundarya Lahari writtenin Grantha by Sri Adi Shankara to
recognize the words and sentences. The manuscript recognition
rates with the system are for Grantha as 92.11%, Old Malayalam
90.82% and for new Malayalam script 89.56%. The recognition
rates of pages of the printed book are for Grantha as 96.16%,
Old Malayalam script 95.22% and new Malayalam script as
92.32% respectively. These results show the efficiency of the
developed system.
Keywords- Grantha scripts; Malayalam; Online character
recognition system.
I. INTRODUCTION
Analysis of handwritten data using computational
techniques has been accelerated with the growth of computer
science developing human-computer interaction. To obtain
handwritten data in digital format, the writing can be scanned
or the writing itself can be done with the aid of special pen
interfaces. The two techniques are commonly known as off-line
and on-line handwriting respectively. Offline handwriting
recognition focuses recognition of characters and words that
had been recorded earlier in the form of scanned image of the
document. In contrast, online handwriting recognition focuses
on tasks when recognition can be performed at the time of
writing through the successive points of strokes of the writer in
a fraction of time.
This paper is an attempt to perform online character
recognition for Grantha script and is the first of its kind to the
best of our knowledge. Grantha script is an ancient language
evolved from Brahmic script. The Dravidian-South Indian-
languages have succeeded Grantha and Brahmi. Many of our
ancient literature are in Grantha. Extract the knowledge from
this extinct language is difficult. Grantha has a strong linkage
with Malayalam characters, so our work uses this similarity to
convert the Grantha script into Malayalam. This boosts our
system into extra mile.
The recognition of handwritten characters in Grantha script
is quite difficult due to the numerals, vowels, consonants,
vowel modifiers and conjunct characters. The structure of the
scripts and the variety of shapes and writing style of individuals
at different times and among different individual poses
challenges that are different from the other scripts and hence
require customized techniques for feature representation and
recognition. Selection of a feature extraction method is the
single most important factor in achieving high recognition
performance [1]. In this paper a framework for recognizing
Grantha script and mapping to its corresponding Malayalam
character is described.
The paper is organized as follows. Section 2 gives the
related work in the field of Indic scripts. Section 3 portrays an
overview of the Grantha Script. Section 4 details the features of
Malayalam language whereas Section 5 points to the snaps of
linkage between Grantha Script and Malayalam. Section 6
depicts the framework for recognizing Grantha script by
converting to Malayalam using online character recognition
mechanism. Section 7 describes the feature extraction
techniques adopted in this framework. Section 8 explains
implementation details and Section 9 analyses and discusses
the experiments and their results. The paper is concluded in
Section 10.
II. RELATED WORK
Many works have been done in linguistics and literature
focusing on the recognition of simple charactersindependent
vowels and isolated consonants. There are three well studied
strategies for recognition of isolated (complex) characters for
scripts like Devanagari, Tamil, Telugu, Bangla and Malayalam.
(i) Characters can be viewed as composition of
strokes.[2],[3],[4],[5],[6] (ii) Characters may be viewed as
compositions of C, C', V and M graphemes [7] (iii) character
can be viewed as indivisible units. [8].
III. OVERVIEW OF GRANTHA SCRIPT
The Grantha script is evolved from ancient Brahmic script
and it has parenthood of most of the Dravidian-south Indian-
languages. In Sanskrit, Grantha stands for manuscript. In
Grantha, each letters represents a consonant with an inherent
vowel (a). Other vowels were indicated using a diacritics or
Vol. 3, No. 7, 2012

68 | P a g e
separate letters. Letters are grouped according to the way they
are pronounced. There are 14 vowels. Of these 7 are the basic
symbols. Long vowels and diphthongs are derived from these.
Also there exists 13 vowel modifiers, and there are no full
vocalic short l and full vocalic long l modifier. Grantha
admits 34 basic consonant characters. As with all Brahmi
derived scripts, the consonant admits the implicit vowel
schwa. Pure consonant value is obtained by use of the virma.
Grantha has two diacritic markers: the anuswra () and
the visarga (). The anuswra is a latter addition and in
Archaic as well as Transition Grantha the letter ma is used to
represent the nasal value. A special feature of Grantha is the
use of subsidiary symbols for consonants. These are three in
number: the use of a subsidiary ya and two allographs for ra
depending on whether ra precedes the consonant or follows it
[9]. The Fig.1 gives the symbols used in Grantha script.
The consonant is represented in two ways. When
following a consonant it is written as under the consonant;
but when it precedes a consonant it takes the form written
after the consonant or conjunct.
Complex consonantal clusters in Grantha script use the
Samyuktaksaras (conjunct letters) very widely. The
Samyuktaksaras of Grantha is formed in the following three
ways [10]. They are stacking, combining and using special
signs as shown in the following Table I. Combined with vowel
signs, these Samyuktaksaras are considered as a single unit and
placed with the Vowel signs.
TABLE I. FORMATION OF SAMYUKTAKSARAS (CONJUNCT LETTERS) IN
GRANTHA SCRIPTS
Stacking

Combini
ng

Special
signs
-r- Conjunct

,

-y- Conjunct

, ,

IV. MALAYALAM LANGUAGE
Malayalam is a Dravidian language consisting of syllabic
alphabets in which all consonants have an inherent vowel.
Diacritics are used to change the inherent vowel and they can
be placed above, below, before or after the consonant. Vowels
are written as independent letters when they appear at the
beginning of a syllable. Special conjunct symbols are used to
combine certain consonants. There are about 128 characters in
the Malayalam alphabet which includes Vowels (15),
consonants (36), chillu (5), anuswaram, visargam,
chandrakkala-(total-3), consonant signs (3), vowel signs (9),
and conjunct consonants (57). Out of all these characters
mentioned, only 64 of them are considered to be the basic ones
as shown in Fig. 2.
The properties of Malayalam characters are the following
Since Malayalam script is an alphasyllabary of the Brahmic
family they are written from left to right.
- Almost all the characters are cursive by themselves.
They consist of loops and curves. The loops are written
frequently in the clockwise order
- Several characters are different only by the presence of
curves and loops.
- Unlike English, Malayalam scripts are not case
sensitive and there is no cursive form of writing.
Malayalam is a language which is enriched with vowels,
consonants and has the maximum number of sounds that are
not available in many other languages as shown in Fig. 3.

Figure 1. Grantha characters

Figure 2. 64 basic characters of Malayalam

Vol. 3, No. 7, 2012

69 | P a g e

Figure 3. Rare Sounds of scripts available only in Malayalam language
V. GRANTHA SCRIPT AND MALAYALAM SNAPS OF
LINKAGE
The foundation of Malayalam script owes itself to Grantha
script. They have much similarity with Grantha scripts. When
Grantha scripts were used to write Sanskrit sounds/phonemes,
it was called Kolezhuthu (rod script).
A. Challenges between Grantha and old scripts of Malayalam
1. Complex stacking conjuncts up to Triple conjuncts are
present in Grantha and it is up to two conjuncts in the old script
of Malayalam.
2. Complex Combining Conjuncts with 4 (or more)
consonantal clusters are present in Grantha but in old
Malayalam script it is only up to 2 consonantal clusters.
B. Challenges between Grantha and new scripts of
Malayalam
1. The special vowelless forms of the Grantha
consonants & , & rarely in printings. This sort
of consonants including this can be seen in manuscripts.
2. Number of vowels is decreased by the absence of
characters corresponding to
3. Complex stacking conjuncts are present in Grantha
and it is absent in the new script of Malayalam.
4. Complex Combining Conjuncts with 4 (or more)
consonantal clusters are present in Grantha but in new
Malayalam script it is only up to 2 consonantal clusters.
VI. SYSTEM FRAMEWORK
The framework designed comprises of five modules. The
first module preprocesses the datasets, to necessitate the way
for feature extraction. The second module is the feature
extraction module. The features extracted here are time domain
features based on writing direction and curvature, which is
discussed in detail in the following section. The next two
modules are part of the classification process namely trainer
and recognizer. The knowledge feature vector of the model
data from the trainer is fed as one of the inputs to the
recognizer in the testing phase. The recognizer is aided with the
Grantha conjugator, which could extract the rules from
conjunct characters. The fifth module is the converter, where
the conversion of Grantha script to old and new Malayalam
characters is done. The converter is supported by intellisence
feature, which is incorporated to avoid the problem
corresponding to the absence of certain characters in new script
of Malayalam by providing the equivalent word by searching
from a dictionary. Also Malayalam conjugator aided the
converter to form rules from conjunct characters to be
converted to new scripts of Malayalam. The entire framework
is presented in the following Fig.4.
VII. FEATURE EXTRACTION
This is the module where the features of handwritten
characters are analyzed for training and recognition which are
explained below. In proposed approach, each point on the
strokes with values of selected features (time-domain features
[11], [12] writing direction and curvature) are described in the
consecutive sub sections.
A. Normalized x-y coordinates
The x and y coordinates from the normalized sample
constitute the first 2 features.
B. Pen-up/pen-down
In this system the entire data is stored in UNIPEN format
where the information of the stroke segments are exploited
using the pen-up/pen-down feature. Pen-up/pen-down feature is
dependent on the position sensing device. The pen-down gives
the information about the sequence of coordinates when the
pen touches the pad surface. The pen-up gives the information
about the sequence of coordinates when the pen not touching
the pad surface.
C. Aspect ratio
Aspect at a point characterizes the ratio of the height to the
width of the bounding box containing points in the
neighborhood. It is given by
1
) ( ) (
) ( 2
) (
A + A
A
=
t y t x
t y
t A
(1)
Where and are the width and the height of the
bounding box containing the points in the neighborhood of the
point under consideration. In this system, we have used
neighborhood of length 2 i.e. two points to the left and two
points to the right of the point along with the point itself.
D. Curvature
The curvature at a Point (x (n), y (n)) is represented by
cos (n) a and sin (n). It can be computed using the following
formulae [11].
) 1 ( ) 1 ( ) 1 ( ) 1 ( ) ( + + = n Cos n Sin n Sin n Cos n Sin u u u u |
(2)
) 1 ( ) 1 ( ) 1 ( ) 1 ( ) ( + + + = n Sin n Sin n Cos n Cos n Cos u u u u |
(3)
E. Writing direction
The local writing direction at a point (x (n), y (n)) is
described using the cosine and sine [12].
2
1
2
1
1
) ( ) (
) ( sin

+

=
n n n n
n n
Y Y X X
Y Y
n u
(4)
) (t x A ) (t y A
Vol. 3, No. 7, 2012

70 | P a g e
(5)
Figure 4. System Architecture
The above elements will be the feature vector for training
and recognition modules. The classifier used here is k-NN.
Dynamic Time Warping DTW distance is used as the distance
metric in the k-Nearest Neighbor classifier. Dynamic Time
Warping is a similarity measure that is used to compare
patterns, in which other similarity measures are practically
unusable. If there are two sequences of length n and m to be
aligned using DTW, first a nXm matrix is constructed where
each element corresponds to the Euclidean distance between
two corresponding points in the sequence. A warping path W is
a contiguous set of matrix elements that denotes a mapping
between the two sequences. The W is subject to several
constraints like boundary conditions, continuity, monotonicity
and windowing. A point to point correspondence between the
sequences which satisfies constraints as well as of minimum
cost is identified by the following eqn.6.
K w C Q DTW
K
k
k
/ min ) , (
1
=

=
(6)
Q, C are sequences of length n and m respectively. Wk
element is in the warping path matrix. The DTW algorithm
finds the point-to-point correspondence between the curves
which satisfies the above constraints and yields the minimum
sum of the costs associated with the matching of the data
points. There are exponentially many warping paths that satisfy
the above conditions. The warping path can be found
efficiently using dynamic programming to evaluate a
recurrence relation which defines the cumulative distance (i, j)
as the distance d(i, j) found in the current cell and the minimum
of the cumulative distances of the adjacent elements. [13].
VIII. IMPLEMENTATION
The online character recognition is achieved through two
steps i) Training and ii) Recognition. Training of the samples is
done setting the prototype selection as hierarchical clustering.
Recognition function predicts the class label of the input
sample by finding the distance of the sample to the classes
according to the K-Nearest Neighbor rule. DTW distance is
used as the distance metric in the k-Nearest Neighbor classifier.
The top N nearest classes along with confidence measures are
returned as the identified letters.
Conversion from Grantha to Malayalam words are done in
two ways. They are converted to old and new scripts of
Malayalam. Due to the challenges posed (section V.(A)) the
conversion between Grantha and New scripts of Malayalam is
a very tedious task when compared with the old Malayalam
script. So we adopted the Malayalam conjugator. Following
Algorithm 1 explain the conversion of Grantha word to
Malayalam old and new script.
Algorithm 1. Conversion of Grantha word to Malayalam
old and new script
Input: Grantha Word W
g

Output: Malayalam Word W
mo
and W
mn
in old script and
new script respectively
for each character w
g
(i) in W
g

if (w
g
(i)== )
then
temp_store = w
g
(i-1);
w
g
(i-1) = w
g
(i);
2
1
2
1
1
) ( ) (
) ( cos

+

=
n n n n
n n
Y Y X X
X X
n u
Vol. 3, No. 7, 2012

71 | P a g e
w
g
(i) = temp_store;
end
end
for each character w
g
(i) in W
g

Set char_type;
if char_type= = vowel or char_type==consonant
Set w
mo
(i) and w
mn
(i) directly
else if char_type == complex stacking form of
conjunct letters
Find R from conjugator// example of R=

Set w
mo
(i) and w
mn
(i)
else if char_type==combined form of conjunct letters
Find R from conjugator //example of R=

Set w
mo
(i) and w
mn
(i)
End
W
mo
= Concatenate (w
mo
(i));
W
mn
= Concatenate (w
mn
(i));
End
IX. EXPERIMENTAL RESULTS AND DISCUSSION
Medium length manuscripts containing 9-12 sample lines of
which 34-42 characters are in each line varied which are
collected to perform experiments. So, in effect, in each
manuscript the number of characters are varied from 306-504.
Experiment is conducted to evaluate the character recognition
system. The collection of basic characters which can be used to
make all the characters of Grantha script is formed by a rule
based system. It is understood that some characters are
frequently misclassified. So experiments are conducted on
those characters to find their similarity measure and Confusion
matrix of such characters are shown in Fig 5. The correct word
limiter space in manuscript was not able to be recognized and it
is resolved by searching the possible word from the dictionary
while making the conversion between Grantha and Malayalam.
Test is conducted to recognize 26712 symbols and 3180 words
from the manuscript.
Test is also conducted on the printed pages of the Grantha
copy of the book titled Soundarya Lahari to recognize the
words and sentences. Each page consists of 32-35 lines and
words vary from 160-190. The test is conducted only to
recognize 455 lines due to the availability of limited number of
pages. Table 2 summarizes the result of recognition rate of
words. The recognition rate for the different symbols in
Grantha Script was tried with classifiers like k-NN and SVM.
Specific experiments were done with and without DTW
algorithm using k-NN classifier. A bar graph is plotted in Fig 6
with different classifiers and recognition rates.

Figure 5. Confusion Matrix of Frequently misclassified characters
TABLE II. RECOGNITION RATE OF GRANTHA WORDS
Grantha old Malayalam new Malayalam
manuscript 92.11% 90.82% 89.56%
book 96.16% 95.22% 92.32%

Figure 6. Category-wise comparison of recognition rates using different
classifiers
X. CONCLUSION
This paper presented a framework for recognizing Grantha
script, by converting to Malayalam using online character
recognition mechanism. The framework was designed with
different modules to perform the same. The feature extraction
module was designed explicitly by taking into consideration of
the different features of Grantha script, which is discussed in
detail in this paper. The framework was implemented and
tested with manuscripts as well as a book. Experiments were
conducted to analyze the different phases of the framework.
Different distance measures for evaluating the similarity were
tried with, of which DTW method showed considerable
recognition rates. It has been observed that there are frequently
misclassified characters. They have been studied in detail with
the aid of confusion matrix Also the recognition rate was
evaluated against two classifiers, namely, k-NN and SVM
classifier. Polynomial, sigmoid and RBF kernel functions were
used for testing with SVM classifier.
REFERENCES
[1] O. D. Trier, A. K.Jain and T. Taxt,. Feature Extraction Methods For
character Recognition : A survey. J Pattern Recognition 1999, 29, 4, 641
662.
[2] H. Swethalakshmi, Online Handwritten Character recognition for
Devanagari and Tamil Scripts using Support Vector Machines,
Masters thesis, Indian Institute of Technology, Madras, India: 2007.
[3] A. Jayaraman, C. C. Sekhar andV. S. Chakravarthy,. Modular Approach
to Recognition of Strokes in Telugu Script. In proceedings of 9th Int.
Vol. 3, No. 7, 2012

72 | P a g e
Document Analysis and Recognition Conference (ICDAR 2007),
Curitiba, Brazil: 2007.
[4] K. H. Aparna, V. Subramanian, M. Kasirajan, G.V. Prakash, V. S.
Chakravarthy and Madhvanath, S. Online Handwriting Recognition for
Tamil. In proceedings of 9th Int. Frontiers in Handwriting Recognition
Workshop (IWFHR 2004), Tokyo, Japan: 2004.
[5] M. Sreeraj and Sumam mary Idicula. On-Line Handwritten Character
Recognition using Kohonen Networks. In Proceeding IEEE 2009 World
Congress on Nature & Biologically Inspired Computing (NABIC'09),
Coimbatore,India: 2009;1425-1430..
[6] M. Sreeraj and Sumam mary Idicula. k-NN Based On-Line Handwritten
Character Recognition System. In proceedings of First International
Conference on Integrated Intelligent Computing (iciic 2010), IEEE,
Bangalore: 2010;171-176.
[7] V.J. Babu, L. Prashanth, R.R. Sharma, G.V.P. Rao, and Bharath, A.
HMM-Based Online Handwriting Recognition System for Telugu
Symbols. In Proceeding of the.9th International Document Analysis and
Recognition Conference (ICDAR 2007), Curitiba, Brazil: 2007.
[8] N. Joshi, G. Sita, A.G. Ramakrishnan, V. Deepu and , S. Madhvanath.
Machine Recognition of Online Handwritten Devanagari Characters. In
proceedings of 8th International Document Analysis and Recognition
Conference (ICDAR 2005), Seoul, Korea: 2005.
[9] http://tdil.mit.gov.in/pdf/grantha/pdf_unicode_proposal-grantha_.pdf
[10] http://www.virtualvinodh.com/grantha-lipitva
[11] M. Pastor, A.Toselli and E. Vidal. Writing Speed Normalization for
On-Line Handwritten Text Recognition. In proceedings of the
International Conference on Document Analysis and Recognition
(ICDAR). 2005.
[12] I. Guyon, P. Albrecht, Y. Le Cun, J. Denker and W. Hubbard, Design of
a Neural Network Character Recognizer for a Touch Terminal. J Pattern
Recognition 1991, 24(2):105119.
[13] N. Joshi, G. Sita, A.G. Ramakrishnan and , S. Madhvanath. Tamil
Handwriting Recognition Using Subspace and DTW Based Classifiers.
In proceedings of 11th International Neural Information Processing
Conference (ICONIP 2004), Calcutta, India: 2004; 806-813.

Vol. 3, No. 7, 2012

73 | P a g e
LOQES: Model for Evaluation of Learning Object
Dr. Sonal Chawla 1, Niti Gupta 2, Prof. R.K. Singla 3
Department of Computer Science and Applications
Panjab University
Chandigarh, India

Abstract Learning Object Technology is a diverse and
contentious area, which is constantly evolving, and will inevitably
play a major role in shaping the future of both teaching and
learning. Learning Objects are small chunk of materials which
acts as basic building blocks of this technology enhanced learning
and education. Learning Objects are hosted by various
repositories available online so that different users can use them
in multiple contexts as per their requirements. The major
bottleneck for end users is finding an appropriate learning object
in terms of content quality and usage. Theorist and researchers
have advocated various approaches for evaluating learning
objects in form of evaluation tools and metrics, but all these
approaches are either qualitative based on human review or not
supported by empirical evidence. The main objective of this
paper is to study the impact of current evaluation tools and
metrics on quality of learning objects and propose a new
quantitative system LOQES that automatically evaluates the
learning object in terms of defined parameters so as to give
assurance regarding quality and value.
Keywords- Learning Objects; Learning Object Repository; Peer
Review; Quality Metrics; Reusability Metrics; Ranking Metrics.
I. INTRODUCTION
All Learning Objects are the basic building blocks of
technology enhanced education. It is a collection of content
items, practice items, and assessment items that are combined
based on a single objective. LOs are very popular these days
as it supports reusability in different context leading to
minimization of production cost. Different practitioners have
defined learning objects in different ways based on its intrinsic
characteristics such as interoperability, reusability, self-
contentedness, accessibility, durability; adaptability etc., still
there is no consensus regarding its correct definition.
According to IEEE Learning Technologies Standard
Committee, Learning Object is any entity, digital or non
digital that can be used, reused or referenced during
technology supported learning [1]. This definition references
both digital and non-digital resources. Therefore, to narrow
down its scope, David Wiley suggests any digital resource
that can be reused to support learning [2]. It includes digital
images, video feeds, animations, text, web pages etc.
irrespective of its size. Rehak and Mason propose that a
learning object should be reusable, accessible, interoperable
and durable [3]. Similarly, Downes considers that only
resources that are shareable, digital, modular, interoperable
and discoverable can be considered learning objects [Downes,
2004]. Kay & Knaack define Learning objects as interactive
web-based tools that support the learning of specific concepts
by enhancing, amplifying, and/or guiding the cognitive
processes of learners [4]. Learning Objects are beneficial
for learners as well as developers or instructors as it provides a
customized environment for knowledge sharing and
development of e-learning course module. Learning objects
can be developed by the programmer as per the requirements
using various authoring tools such as office suites, Hypertext
editors, Vector graphic editors etc. or can be extracted from
the repositories on the basis of metadata stored in Learning
Object Repositories such as MERLOT (Multimedia
Educational Resource for Learning and Online Teaching),
WORC (University of Wisconsin Online Resource Center),
ALE (Apple Learning Exchange) etc. Metadata is the
description of learning resources such as name of the author,
most suitable keywords, language and other characteristics
which makes it possible to search, find and deliver the desired
learning resource to the learner. The major bottleneck for end
users is finding an appropriate learning object that are
published in various learning object repositories in terms of
various parameters like quality, reusability, granularity and
context usage etc.

Fig.1 Conceptual Framework of Learning Object Repositories
II. LITERATURE REVIEW
The growth in the number of LOs, the multiplicity of
authors, increasing diversity of design and availability of
trained and untrained educators has generated interest in how
to evaluate them and which criteria to use to make judgments
about their quality and usefulness [5]. According to Williams
(2000), evaluation is essential for every aspect of designing
learning objects, including identifying learners and their needs,
Content Developers
Educators Learners
Learning Objects
Digital Repository of
Learning Objects
Delivery Modes Delivery Medium
Vol. 3, No. 7, 2012

74 | P a g e
conceptualizing a design, developing prototypes,
implementing and delivering instruction, and improving the
evaluation itself. Theorist and researchers have proposed
different studies for evaluating LOs in terms of reusability,
standardization, design, use and learning outcomes. The major
problems with these studies are that they are not supported by
empirical evidence, covers limited number of objects and
evaluate only the qualitative phenomenon.
A. Theoretical Approaches to evaluate Learning Objects
Researchers have followed two distinct paths for
evaluating learning objects Summative and Formal. The
summative approach deals with final evaluation of LOs
(Kenny et al., 1999; Van Zele et al., 2003; Adams et al., 2004;
Bradley & Boyle, 2004; Jaakkola & Nurmi, 2004; Krauss &
Ally, 2005; MacDonald et al., 2005) based on informal
interviews or surveys, frequency of use and learning outcome.
The main goal of this approach has been to determine whether
participants valued the use of learning objects and whether
their learning performance was altered. The formative
assessment works during the development phase of learning
objects where feedback is solicited from small groups at
regular intervals. [4]
Nesbit et al. (2002) outline a convergent evaluation model
that involves multiple participants learners, instructors,
instructional designers and media developers. Each group
offers feedback throughout the development of a LO. Finally a
report is produced that represents multiple values and needs.
The major drawback of convergent evaluation model is limited
no. of participants and difference in opinions and beliefs. [6]
Nesbit and Belfer (2004) designed an evaluation tool
Learning Object Review Instrument (LORI) which includes
nine items: content quality, learning goal alignment, feedback
and adaptations, motivation, presentation design, interaction,
accessibility, reusability and standards. This instrument has
been tested on a limited basis ( Krauss & Ally, 2005; Vargo et
al., 2003) for a higher education population, but the impact of
specific criteria on learning has not been examined [7].
MERLOT developed another evaluation model which
focuses on quality of content, potential effectiveness as a
teaching learning tool, and ease of use. Howard Rose &
Harrigan (2003) tested the MERLOT model with 197 students
from 10 different universities. The results were descriptive and
didnt distinguish the relative impact of individual model
components. Coachrane (2005) tested a modified version of
the MERLOT evaluation tool that looked at reusability, quality
of interactivity, and potential for teaching, but only final
scores are tallied, so the impact of separate components could
not be determined. Finally, the reliability and validity of the
MERLOT assessment tool has yet to be established [4].
Kay & Knaack (2005, 2007a) developed an evaluation tool
based on a detailed review of research on instructional design.
Specific assessment categories included organisation/ layout
learner control over interface, animation, graphics, audio, clear
instructions, help features interactivity, incorrect content/
errors, difficulty/ challenge, useful/ informative, assessment
and theme/ motivation. The evaluation criteria were tested on
a large secondary school population [9, 10].
Based on above theories Kay & Knaack (2008) proposed a
multi component model for assessing learning object The
Learning Object Evaluation Metric (LOEM) which focused on
five main criteria interactivity, design, engagement, usability,
and content. The model was tested on a large sample and the
results revealed that four constructs interactivity, design,
engagement and usability demonstrated good internal and inter
rater reliability, significantly correlated with student and
teacher perception of learning, quality , engagement and
performance. But there is little finding as how each of these
constructs contributes to the learning process [4].
Munoz & Conde (2009) designed and developed a model
HEODAR that automatically evaluates the Los and produce a
set of information that can be used to improve those Los. The
tool is implemented in the University of Salamanca framework
and initially integrated with LMS called Moodle but the results
are not yet tested [11].
Eguigure & Zapata (2011) proposed a model for Quality
Evaluation of Learning Objects (MECOA) which defines six
indicators: content, performance, competition, self-
management, meaning and creativity to evaluate the quality of
Los from a pedagogical perspective. These indicators are
evaluated by four actors: teachers, student, experts and
pedagogues. The instrument was designed and incorporated
into AGORA platform but the results are not empirically
tested [12].
III. LEARNING OBJECT EVALUATION METRICS
Metrics are the measurement of a particular characteristics
grounded from the field of Software Engineering. Researchers
have proposed various metrics for quantitative analysis of
different dimensions of learning objects such as quality of
metadata stored, reusability, learning style, ranking, cost
function etc, but the empirical evaluation is performed at small
scale and that on individual basis.

Fig.2 Types of Learning Object Evaluation Metrics
A. Quality Metrics
The quality of the content as well as the metadata of
learning objects stored in learning object repositories is an
important issue in LOR operation and interoperability. The
quality of the metadata record directly affects the chances of
learning object to be found, reviewed or reused. The
traditional approach to evaluate the quality of learning object
metadata is by comparing the values of metadata with the
values provided by metadata experts. This approach is useful
for small sized and small growing repositories but become
impractical for large or federated repositories. Thus, there is a
Learning Object Evaluation Metrics
Quality Metrics Reusability
Metrics
Ranking
Metrics
Vol. 3, No. 7, 2012

75 | P a g e
need for automation of quality assessment of learning object
metadata. Quality metrics are small calculation performed
over the values of different fields of the metadata record in
order to gain insight of various quality characteristics such as
completeness, accuracy, provenance, conformance to
expectations, coherence and consistency, timeliness and
accessibility. The values obtained from the metrics are
contrasted with evaluations by human reviewers to a sample of
learning object metadata from a real repository, and the results
are evaluated. The various quality metrics proposed by Ochoa
and Duval are:
TABLE 1: TYPES OF QUALITY METRICS
Metric Name Metric Formula
Simple
Completeness

Where P(i) is 1 if the i
th
field has a non-null value, 0
otherwise. N is the number of fields
Weighted
Completeness

Where
is the relative importance of the i

th
field
Nominal
Information
Content

Where K is the number of nominal fields. P(
is
the probability of a value of the i
th
nominal field.
Textual
Information
Content
log
where
is the term frequency of the i

th
word ,
is the document frequency of the i

th
word.
Readability

Flesch(description_text) is the value of the Flesch index
for the text present in the title and description of the
record.

1) Simple Completeness:
This metric tries to measure the completeness of the
metadata record. It counts the number of fields that doesnt
contain null values. In the case of multi-valued fields, the
field is considered complete if at least one instance exists. The
score could be calculated as a percentage of possible fields and
divided by 10 to be in a scale from 0 to 10. For example,
according to this metric, a record with 70% of its fields filled
has a higher score (q=7) then one in which only the 50% has
been filled (q=5).
2) Weighted Completeness:
This metric not only counts the no. of filled fields, but
assign a weight value to each of the fields. This weight value
reflects the importance of that field for the application. The
obtained value should be divided by sum of all the weights and
multiplied by 10. The more important fields could have a
weight of 1 while the unimportant fields could have a weight
of 0.2. For e.g. if the main application of the metadata will be
to provide information about the object to a human user, the
title, description and annotation fields are more important than
the identifier or metadata authors fields. If a record contains
information for title, description and annotation then its score
will be (3/3.2*10=9) which is higher than the record with
information for title and metadata author (1.2/3.2*10=4).
3) Nominal Information Content:
This metric tries to measure the amount of information that
the metadata possess in its nominal fields, the fields that can
be filled with values taken from a fixed vocabulary. For
nominal fields, the Information Content can be calculated as 1
minus the entropy of value.
Entropy is the negative log of the probability of the value
in a given repository. This metric sums the information
content for each categorical field of the metadata record. The
metadata record whose level of difficulty is set to high
provides more unique information about the object then the
record whose difficulty level is medium: or low, thus
possess a higher score. If records nominal fields contain only
default values, they will provide less unique information about
the about the object and possess lower score.
4) Textual Information Content
This metric tries to measure the relevance and uniqueness
of words contained in the records text fields, the fields that
can be filled with free text. The relevance and uniqueness
of a word is directly proportional to how often the word
appears in the record and inversely proportional to how many
records contain that word. This relation is handled by TF-IDF
(Term Frequency Inverse Document Frequency).
The number of times that the word appears in the
document is multiplied by the negative log of the number of
documents that contain that word. The log of the sum of all
the TF-IDF value of all the words in textual fields is the result
of the metric. For example, if the title field of a record is
Lectures of C++, given that lecture and C++ are
common words in the repository, will have lower score than a
record whose title is Introduction to Object Oriented
Programming in C++ because the latter one contains more
words and object, oriented, programming are less
frequent in repository.
5) Readability
This metric tries to measure how accessible the text in the
metadata is, i.e. how easy is to read the description of the
learning object.
The readability indexes count the number of words per
sentence and the length of the words to provide a value that
suggest how easy is to read a text. For example, a description
where only acronyms or complex sentences are used will
receive a higher score but lower in quality than a description
where normal words and simple sentences are used.
Ochoa & Duval (2006) designed an experiment to evaluate
how the quality metrics correlate with quality assessment by
human reviewers and the result showed that Textual
Information Content metric seems to be a good predictor [8].

Vol. 3, No. 7, 2012

76 | P a g e
TABLE 2: CORRELATION BETWEEN HUMAN EVALUATION SCORE AND THE
METRICS SCORE
Simple
Completenes
s
Weighted
Completenes
s
Nomina
l Info.
Content
Textual
Info.
Conten
t
Readabilit
y
Pearso
n
-.395 -.457 -.182 .842 .257
Sign. .085 .043 .443 .000 .274
B. Reusability Metric
Reusability is the degree to which a learning object can
work efficiently for different users in different digital
environments and in different educational contexts over time.
Reusability of a learning object is a major issue these days as
developing a quality educational material is costly in terms of
time and resources. A lot of research has been going on to
improve the reusability of learning objects by defining
standards such as SCORM, IMS etc. so as to resolve the issue
of interoperability among different platforms. The factors that
determine the reusability of a learning object can be classified
as structural or contextual. From Structural factor point of
view, a learning object should be self-contained, modular,
traceable, modifiable, usable, standardized and properly
grained. As per contextual point of view, learning object must
be generic and platform independent, so that it can be used in
various contexts irrespective of any subject or discipline. To
measure the reusability of learning object, various metrics
have been proposed grounded on the theory of software
engineering which measures various reusability factors.
1) Cohesion:
This Cohesion analyzes the relationship between different
modules. Greater cohesion implies greater reusability.
Learning Object involves number of concepts. Fewer
the concepts, greater the module cohesion.
Learning object should have a single and clear learning
objective. The more learning objectives it has, the less
cohesive it will be.
The semantic density of a learning object shows its
conciseness. More conciseness shows greater
cohesion.
Learning object must be self-contained and exhibit
fewer relationship and instances so as to be highly
cohesive.
Thus the cohesion of learning object depends on semantic
density, aggregation level, and number of relationships
concepts and learning objectives.
TABLE 3: COHESION VALUES TO MEASURE LEARNING OBJECT REUSABILITY
Cohesion Capability of reuse Value
Very High Independent and fully self-contained objects. 5
High
Self-contained objects including some
dependencies.
4
Medium Objects with multiple dependencies. 3
Low Objects with multiple dependencies. 2
Very low Completely dependent objects. 1

2) Coupling:
Coupling measures interdependencies between various
modules. A module must communicate with minimum number
of modules and exchange little information so as to minimize
the impact of changing modules. Lower coupling implies
greater reusability. Coupling is directly proportional to number
of relationships present, so a learning object should be self
contained and referenced to fewer objects to increase
reusability.
3) Size and Complexity:
Granularity is a major factor that measures the reusability
of a particular object, as finely grained learning objects are
more easily reusable. Granularity is directly proportional to
the following LOM elements:
Size of the Learning Object.
Duration or time to run the learning object.
Typical Learning Time i.e. estimated time required to
complete the learning object.

TABLE 4: VALUES TO MEASURE LEARNING OBJECT SIZE
Size Description Value
Very Small Atomic resources 5
Small Small sized resources 4
Medium Medium sized lessons 3
Big Big sized aggregated courses 2
Very Big Very big sized courses 1

4) Portability
Portability is the ability of a learning object to be used in
multiple contexts across different platforms.
Technical portability depends on delivery format of
the learning object as well as on the hardware and
software requirements to run that particular Learning
Object.
TABLE 5: VALUES TO MEASURE LEARNING OBJECT TECHNICAL PORTABILITY
Technical
Portability
Description Value
Very High The object is based on a technology available on
all systems and platforms (e.g. html).
5
High The object is based on a technology available on
many systems and platforms.
4
Medium The object is based on a technology that is not
available on many systems (i.e. common
platform specific file format).
3
Low The object is based on a technology that is
hardly available on different systems (i.e.
uncommon proprietary file format).
2
Very low The object is based on a proprietary technology
that is not available on many systems (i.e. a
specific server technology).
1
Educational portability deals with vertical and
horizontal portability. Vertical portability means
possibility to be used or reused on different
educational levels such as primary, higher or
secondary, whereas horizontal portability determines
the usage among various disciplines.
5) Difficulty of Comprehension
Difficulty to understand a learning object directly
influences the reusability of that object in an aggregated
manner.
Vol. 3, No. 7, 2012

77 | P a g e

TABLE 5: VALUES TO MEASURE LEARNING OBJECT EDUCATIONAL
PORTABILITY
Educational
Portability
Description Value
Very High Object is generic, pedagogically neutral, used
on different education levels.
5
High Object can be used for several disciplines and
educational levels.
4
Medium Object can be used in specific area and
discipline without modifications.
3
Low Object can be used in different educational
context and level with several modifications.
2
Very low Object can be hardly reused on different
educational contexts and levels..
1

C. Ranking Metrics
LOR uses various strategies to search learning objects as
per end user requirements such as metadata based search and
simple text based search. In both cases the retrieval of
appropriate learning object depends on the quality of the
metadata and content matching capacity of LOR. Ochoa and
Duval proposed various ranking metrics which are inspired on
methods currently used to rank other types of objects such as
books, scientific journals, TV programs etc. They are adapted
to be calculable from the information available from the usage
and context of learning objects.

Fig.3 Types of Relevance Ranking Metrics
1) Topical Relevance Ranking Metrics
These metrics estimate which objects are more related to
what a user wants to learn. For this, first step is to estimate
what is the topic that interests the user and second step is to
establish the topic to which each learning object in the result
list belongs. The source of information for first step is query
terms used, course from which the search was generated and
the previous interactions of the user with the system and for
the second step is classifications in the learning object
metadata, from the topical preference of previous learners that
have used the object or the topic of courses that the object
belongs to.
a) Basic Topical Relevance Metric
This metric makes two nave assumptions: 1) topic needed
by the user is fully expressed in the query 2) object is relevant
to just one topic. The relevance is calculated by counting the
no. of times the object has been previously selected from the
result list when same query terms have been used.
BT Relevance Metric is the sum of the times that the object
has been selected in any of those queries.
{

Similarity between two queries range (0-1)
o Learning object to be ranked
q query performed by user
q
i
representation of i
th
previous query
NQ Total no. of queries
Similarity between two queries can be calculated either as
the semantic differences between the query terms or the no. of
objects that both queries have returned in common.
b) Course-Similarity Topical Relevance Metric
The course in which the object will be reused can be
directly used as the topic of the query. Objects that are used in
similar courses should be ranked higher in the list. The main
problem to calculate this metric is to establish which courses
are similar. For this SimRank algorithm is used that analyzes
the object-to-object relationship to measure the similarity
between those objects.
CST Relevance Metric is calculated by counting the no. of
times that a learning object in the list has been used in the
universe of courses.
{

o learning object to be ranked
c course where it will be inserted or used
c
i
ith course present in the system
NC Total no. of courses
NO Total no. of objects
c) Internal Topical Relevance Metric
This algorithm is based on HITS algorithm which states
the existence of hubs and authorities.
Hubs pages that mostly points to other useful pages
Authorities pages with comprehensive information about
a subject.
Hubs correspond to Courses and Authorities correspond to
Learning Objects. Hub value of each course is calculated as
no. of inbound links that it has. Rank of each object is
calculated as the sum of the hub value of the courses where it
has been used.

Relevance
Ranking Metrics
Topical
Relevance
Ranking Metrics
Personal
Relevance
Ranking Metrics
Situational
Relevance
Ranking Metrics
Vol. 3, No. 7, 2012

78 | P a g e
c
i
ith course where o has been used.
N Total no. of courses where o has been used.
2) Personal Relevance Ranking Metrics
This metric tries to establish the learning preference of the
user and compare them with the characteristics of the learning
objects in the result list. The most difficult part is to obtain an
accurate representation of the personal preferences. The
richest source of information is the attention metadata that
could be collected from the user. The second step is to obtain
the characteristics of the object which is collected from
metadata or contextual and usage information.
a) Basic Personal Relevance Ranking Metric
For a given user, a set of the relative frequencies for the
diff. metadata field values present in their objects is obtained.
Val (o , f) .The frequencies for each metadata field are
calculated counting the no. of times that a given value is
present in the given field in the metadata. Once the frequencies
are obtained they can be compared with the metadata values of
the objects in the result list. If the value present in the user
preference set is also present in the object, the object receives
a boost in its rank equal to the relative frequency of the value.
{

f field in the metadata standard
v value that the f field could take

u user
o
i
ith object previously used by u
N Total no. of objects
f
i
ith field considered for calculation of the metric
NF Total no. of those fields
b) User-Similarity Personal Relevance Ranking Metric
USP Relevance Metric is calculated by finding the no. of
times similar users have reused the objects in the result list.
SimRank algorithm is used to find similar users.
{

u user that performed the query
u
i
representation of the ith user
NU Total no. of users
3) Situational Relevance Ranking Metrics
This metric tries to estimate the relevance of the object in
the result list to the specific task that caused the search. This
relevance is related to the learning environment in which the
object will be used as well as the time, space and technological
constraints that are imposed by the context where learning will
take place. Contextual information is needed in order to
establish the nature of the task and its environment.
a) Basic Situational Relevance Ranking Metric
In formal learning contexts, the description of the course,
lesson or activity in which the object will be inserted is a
source of contextual information which is usually written by
the instructor. Keywords can be extracted from these texts and
used to calculate a ranking metric based on the similarity
between the keyword list and the content of textual fields of
the metadata record. Similarity is defined as the cosine
distance between the TF-IDF vectors of contextual keywords
and the TF-IDF vector of words in the text fields of the
metadata of the objects in the result list. TF-IDF is a measure
of the importance of a word in a document that belongs to a
collection.

TF Term Frequency or the no. of times that the word
appear in the current text
IDF Inverse Document Frequency or the no. of
documents in the collection where the word is present.
c course where the object will be used
tv
i
ith component of the TF-IDF vector representing the
keywords extracted from the course description
ov
i
ith component of the TF-IDF vector representing the
text in the object description
M dimensionality of the vector space (no. of different
words)
b) Context-Similarity Situational Relevance Ranking
Metric
A fair representation of the kind of objects that are relevant
in a given context can be obtained from the objects that have
already been used under similar conditions.

c course where the object will be used
Vol. 3, No. 7, 2012

79 | P a g e
o
i
i
th
object contained in the course c
f field in the metadata standard
v value of the f field in the object o
NF Total no. of fields
IV. PROPOSED MODEL LOQES
Researchers have proposed various metrics but the main
drawback is that these metrics are not implemented in any
quality evaluation tool. The results of these metrics have been
analyzed by conducting empirical analysis on small scale and
that on individual basis. The main objective of this paper is to
propose a model LOQES that automatically assesses the
quality of learning object by employing various metrics
discussed above on the defined parameters and give a
quantitative rating that acts as quality indicator, which is
beneficial for the learning object community. This system will
apply on newly developed learning objects and acts as a
certification mechanism. The tool first extracts the metadata
fields of learning object on the basis of information supplied
by the contributor. Then it applies various quality metrics on
the metadata information to estimate the correctness and
accuracy of metadata records. Afterwards, this information is
used to estimate the value of other defined parameters by
employing various metrics such as reusability, granularity,
linkage, complexity etc. The aggregate of all the scores of the
above parameters helps in calculating the overall rating of that
particular learning object. The main benefit of this model is
that it is a quantitative model which automatically evaluates
the learning object and is not based on peer review.
V. CONCLUSIONS
In this paper, the main emphasize is on proposing a
quantitative model that automatically assesses the quality of
learning object on the basis of various metrics proposed.
Presently all the evaluation tools are qualitative and based on
expert review. Researchers have also specified the need for an
automatic assessment tool due to large dissemination of
learning objects. The main task left for future work is to
develop the system and execute an empirical study with full
implementation of these metrics in real world and comparing
their performance.
REFERENCES
[1] Learning Technology Standards Committee (2002) (PDF), Draft
Standard for Learning Object Metadata. IEEE Standard 1484.12.1, New
York: Institute of Electrical and Electronics Engineers, P.45.
[2] Wiley (David A). Connecting learning objects to instructional theory: A
definition, a metaphor, and taxonomy.
http://reusability.org/read/chapters/wiley.doc
[3] Rehak, Daniel R.; Mason, Robin (2003), "Engaging with the Learning
Object Economy", in Littlejohn, Allison, Reusing Online Resources: A
Sustainable Approach to E-Learning, London: Kogan Page, pp. 2230.
[4] Kay & Knaack (2008), A Multi-Component Model for Accessing
Learning Objects: The Learning Object Evaluation Metrics (LOEM)
Australasian Journal of Educational Technology, v24 n5 p574-591.
[5] Haughey, M. & Muirhead, B. (2005). Evaluating learning objects for
schools. E-Journal of Instructional Science and Technology, 8(1).
[6] Nesbit & Belfer (2002). A Convergent Participation Model for
Evaluation of Learning Objects, Canadian Journal of Learning &
Technology, v28 n3 p105-20.
[7] Nesbit & Belfer (2004). Collaborative Evaluation of Learning Objects,
[8] Towards Automatic Evaluation of Learning Object Metadata Quality
http://scholar.google.co.in/scholar_url?hl=en&q=http://citeseerx.ist.psu.e
du/viewdoc/download%3Fdoi%3D10.1.1.85.2428%26rep%3Drep1%26t
ype%3Dpdf&sa=X&scisig=AAGBfm0lZ1kpH7o0ccHqN__4j6sMa-
fLfg&oi=scholarr
[9] Kay, R. H. & Knaack, L. (2005). Developing learning objects for
secondary school students: A multi-component model. Interdisciplinary
Journal of Knowledge and Learning Objects, 1, 229-254.
[10] Kay, R. H. & Knaack, L. (2007a). Evaluating the learning in learning
objects. Open Learning, 22(1), 5-28.
[11] C. Muoz, M..C. Gonzlez, and F.J.G. Pealvo, "Learning Objects
Quality: Moodle HEODAR Implementation", ;in Proc. WSKS (1),
2009, pp.88-97.
[12] A. Zapata, V. Menendez, Y. Eguigure, M. Prieto (2009) Quality
Evaluation Model for Learning Objects from Pedagogical Perspective. A
Case of Study, ICERI2009 Proceedings, pp. 2228-2238.
[13] Juan F. Cervera, Mara G. Lpez-Lpez, Cristina Fernndez, Salvador
Snchez-Alonso, Quality Metrics in Learning Objects, Metadata and
Semantics, 2009, pp. 135-141.
[14] Javier Sanz-Rodriguez, Juan Manuel Dodero, and Salvador Sanchez-
Alonso. 2011. Metrics-based evaluation of learning object reusability.
Software Quality Control 19, 1 (March 2011), 121-140.
[15] Laverde, Andres Chiappe; Cifuentes, Yasbley Segovia; Rodriguez,
Helda Yadira Rincon, Towards an Instructional Design Methodology
Based on Learning Objects, Educational Technology Research and
Development, v55 n6 p671-681.
[16] Xavier Ochoa and Erik Duval. 2008. Relevance Ranking Metrics for
Learning Objects. IEEE Trans. Learn. Technol. 1, 1 (January 2008), 34-
48.
[17] Aprioristic Learning Object Reusability Evaluation
www.cc.uah.es/ssalonso/papers/SIIE2008.pdf
[18] A Preliminary Analysis of Software Engineering Metrics-based Criteria
for the Evaluation of Learning Object Reusability
a. http://online-journals.org/i-jet/article/viewArticle/794
[19] Learnometrics: Metrics for Learning Objects
https://lirias.kuleuven.be/bitstream/1979/1891/2/ThesisFinal.pdf
[20] www.wikipedia.com
[21] Ranking Metrics and Search Guidance for Learning Object Repository
www.cl.ncu.edu.tw/.../Ranking%20Metrics%20and%20Search%20G...
AUTHORS PROFILE
Dr. Sonal Chawla
Assistant Professor,
Department of Computer Science and Applications,
Panjab University, Chandigarh, India.
Niti Gupta
Research Scholar,
Dr. R. K. Singla
Professor,

Vol. 3, No. 7, 2012
80 | P a g e
FHC-NCTSR: Node Centric Trust Based secure
Hybrid Routing Protocol for Ad Hoc Networks
Prasuna V G
Associate Professor. Basaveswara Institute of Information
Technology, Barkathpura,
Hyderabad, INDIA,
Dr. S Madhusudahan Verma
Professor , . Dept of OR&SQC,
Rayalaseema University, Kurnool, AP, INDIA

Abstract To effectively support communication in such a
dynamic networking environment as the ad hoc networks, the
routing mechanisms should adapt to secure and trusted route
discovery and service quality in data transmission. In this
context, the paper proposed a routing protocol called Node
Centric Trust based Secure Hybrid Routing Protocol [FHC-
NCTSR] that opted to fixed hash chaining for data transmission
and node centric trust strategy for secure route discovery. The
route discovery is reactive in nature, in contrast to this, data
transmission is proactive, hence the protocol FHC-NCTSR
termed as hybrid routing protocol. The performance results
obtained from simulation environment concluding that due to the
fixed hash chaining technique opted by FHC-NCTSR, it is more
than one order of magnitude faster than other hash chain based
routing protocols such as SEAD in packet delivery. Due to the
node centric strategy of route discovery that opted by FHC-
NCTSR, it elevated as trusted one against to Rushing, Routing
table modification and Tunneling attacks, in contrast other
protocols failed to provide security for one or more attacks listed,
example is ARIADNE that fails to protect from tunneling attack.
Keywords- Ad hoc network, Dynamic source routing; Hash cains;
Manet; Mobile ad hoc networks; NCTS-DSR; Routing Potocol;
SEAD; Secure routing .
I. INTRODUCTION
As ad-hoc networks do not rely on existing infrastructure
and are self-organizing, they can be rapidly deployed to
provide robust communication in a variety of hostile
environments. This makes ad hoc networks very appropriate
for a broad spectrum of applications ranging from providing
tactical communication for the military and emergency
response efforts to civilian forums such as convention centers
and construction sites. With such diverse applicability, it is not
difficult to envision ad hoc networks operating over a wide
range of coverage areas, node densities, mobility patterns and
traffic behaviors. This potentially wide range of ad hoc
network operating configurations poses a challenge for
developing efficient routing protocols. On one hand, the
effectiveness of a routing protocol increases as network
topological information becomes more detailed and up-to-
date. On the other hand, in an ad hoc network, mobility may
cause frequent changes in the set of communication links of a
node [1], requiring large and regular exchanges of control
information among the network nodes. And if this topological
information is used infrequently, the investment by the
network may not pay off. Moreover, this is in contradiction
with the fact that all updates in the wireless communication
environment travel over the air and are, thus, costly in
transmission resources. Routing protocols for ad hoc networks
can be classified either as proactive, reactive or hybrid.
Proactive or table driven protocols continuously evaluate the
routes within the network, so that when a packet needs to be
forwarded, the route is already known and can be immediately
used. Examples of proactive protocols include DSDV [2],
TBRPF [3], and WRP [4]. In contrast, reactive or on-demand
protocols invoke a route determination procedure on an on-
demand basis by flooding the network with the route query.
Examples of reactive protocols include AODV [5], DSR [6],
and TORA [7]. The on-demand discovery of routes can result
in much less traffic than the proactive schemes, especially
when innovative route maintenance schemes are employed.
However, the reliance on flooding of the reactive schemes
may still lead to a considerable volume of control traffic in the
highly versatile ad hoc networking environment. Moreover,
because this control traffic is concentrated during the periods
of route discovery, the route acquisition delay can be
significant. In Section II, we explore the third class of routing
protocols the hybrid protocols.
II. PROTOCOL HYBRIDIZATION
The diverse applications of ad hoc network pose a
challenge for designing a single protocol that operates
efficiently across a wide range of operational conditions and
network configurations. Each of the purely proactive or purely
reactive protocols described above performs well in a limited
region of this range. For example, reactive routing protocols
are well suited for networks where the call to mobility ratio
is relatively low. Proactive routing protocols, on the other
hand, are well suited for networks where this ratio is relatively
high. The performance of both of the protocol classes degrades
when they are applied to regions of ad hoc network space
between the two extremes. Given multiple protocols, each
suited for a different region of the ad hoc network design
space, it makes sense to capitalize on each protocols strengths
by combining them into a single strategy (i.e. hybridization).
In the most basic hybrid routing strategy, one of the protocols
would be selected based on its suitability for the specific
networks characteristics. Although not an elegant solution,
such a routing strategy has the potential to perform as well as
the best suited protocol for any scenario, and may outperform
either protocol over the entire ad hoc network design space.
Vol. 3, No. 7, 2012
81 | P a g e
However, by not using both protocols together, this
approach fails to capitalize on the potential synergy that would
make the routing strategy perform as well as or better than
either protocol alone for any given scenario. A more
promising approach for protocol hybridization is to have the
base protocols operate simultaneously, but with different
scopes (i.e., hybridization through multi scoping). For the
case of a two-protocol routing strategy, protocol A would
operate locally, while the operation of protocol B would be
global. The key to this routing strategy is that the local
information acquired by protocol A is used by protocol B to
operate more efficiently. Thus the two protocols reinforce
each others operation. This routing strategy can be tuned to
network behavior simply by adjusting the size of the protocol
As scope. In one extreme configuration, the scope of protocol
A is reduced to nothing, leaving protocol B to run by itself. As
the scope of protocol A is increased, more information
becomes available to protocol B, thereby reducing the
overhead produced by protocol B. At the other extreme,
protocol A is made global, eliminating the load of protocol B
altogether. So, at either extreme, the routing strategy defaults
to the operation of an individual protocol. In the wide range of
intermediate configurations, the routing strategy performs
better than either protocol on its own.
The rest of the paper, section II explores the related work,
section III discuss the route discovery strategy of FHC-
NCTSR and section IV describes the data transmission
approach that fallowed by section V, which explores
simulations and results discussion. Section VI is conclusion
and section VII explores the bibliography.
III. RELATED WORK
There are known techniques for minimizing Byzantine
failures caused by nodes that through malice or malfunction
exhibit arbitrary behavior such as corrupting, forging, and
delaying routing messages. A routing protocol is said to be
Byzantine robust when it delivers any packet from a source
node to a destination as long as there is at least one valid route
[8]. However, the complexity of that protocol makes it
unsuitable for ad hoc networks. Hauser et al[9] avoid that
defect by using hash chains to reveal the status of specific
links in a link-state algorithm. Their method also requires
synchronization of the nodes. Hu[10] introduced another
technique called SEAD that uses a node-unique hash chain
that is divided into segments. The segments are used to
authenticate hop counts. However, DSDV distributes routing
information only periodically. The protocols[8, 9, 10] failed to
perform when networks with hops in large scale due to their
computational complexity in hash chain measurement. In
many applications, reactive or on demand routing protocols
are preferred. With on demand routing, source nodes request
routes only as needed. On demand routing protocols performs
better with significantly lower overhead than periodic routing
protocols in many situations [11]. The authentication
mechanism of Ariadne[11] is based on TESLA[12]. They use
only efficient symmetric-key cryptographic primitives. The
main drawback of that approach is the requirement of clock
synchronization, which is very hard for wireless ad hoc
networks. And the protocols [8, 9, 10, 11, 12] failed to protect
networks from one or more attacks such as tunneling attack.
The protocol FHC-NCTSR proposed in this paper having
much scope to perform better in large networks, since its less
computational complexity. The node centric and two hop level
authentication strategies that opted by FHC-NCTSR helps to
deal with various attacks that includes tunneling attack.
IV. ROUTE DISCOVERY PROCESS
Objective of the NCTS-DSR route establishment process is
preventing unauthorized hops to join in root during route
request process
A. Privileges assumed at each node that exists in the network
- The node that belongs to the network contains the
capabilities fallowing
- Able to generate hash method based id for
broadcasting packets
- Ability to issue digital certificate
- Ability to maintain the id of hop from which egress
data received and id of hop to which ingress data
- Elliptic Curve based cryptography functionality will
be used to protect data transmission
B. Hop node registration process
Hop nodes exchange their digital certificates recursively
with a time interval
. The delay between two iterations

represented by an interval referred as certificate exchange
interval
.
Each node submits its certificate to one and two
hop level nodes.

h
dt
i
i
=
h
is time interval for node h to submit its digital
certificate to neighbor hop nodes
t is interval threshold
dt
i
is distance that can travel by a node h in
interval threshold t
C. Description of the notations used in route detection
process
s
n
source node
d
n

destination node
r
n
relay node

e
n
node from which egress data
received by
r
n

'
n
e

node from which egress data received by
, and two level hop to
r
n
i
n

node to which ingress data send by
r
n
h
cer
digital certificate of hop h
Vol. 3, No. 7, 2012
82 | P a g e
h
add
address of hop node h

( ) lt cer cTs iTs
h n n
r e
=

( )
h
lt cer

is certificate life time of the
hop node h
cTs
n
r

current time stamp at the relay node
r
n
iTs
n
e

is timestamp at created( cer
n
r
creation
time)
p
id

Is RREQ unique ID that generated in
secure random way
S
p
id

is set of packet ids those transmitted by a
node
D. Root Request Process
1) Process of RREQ construction at relay hop node of the
source node.
RREQ packet at source node
s
n contains
< add
n
s
, add
n
d
, p
id
, cer
n
e
,
e
, cer
n
s
,
n
s
, cer
n
i
,
ECPK
n
s
>
Note: Here cer
n
e
is null.
TABLE 1: ALGORITHM FOR RREQ PACKET EVALUATION AT HOP
NODE OF THE SOURCE NODE
Step 1: a. If p S
id p
id
e then RREQ packet will discarded
b. else adds p S
id p
id
e to S
p
id
and continues
step 2
Step 2: a. If ( ) lt cer
n
i
is valid and cer cer
n n
i r
== then
continues step 3,
b. else discards RREQ packet.

[Here cer
n
i
is because the current node certificate is available
at sender node as certificate of one hop node that acts as
target for ingress transaction]

Step 3: a. If cer
n
e
is null then assumes sender is source and
continues step 4.
b. Else If
cer
n
e
is not null and
( ) lt cer
n n
e e
<
and
'
cer cer
n n
e e
==
then
cer
n
e
is valid and continues
step 4
else RREQ will be discarded.
Step 3:
a. If cer
n
e
is null then assumes sender is source and
continues step 4.
b. Else If
cer
n
e
is not null and
( ) lt cer
n n
e e
<
and
'
cer cer
n n
e e
==
then
cer
n
e
is valid and continues step 4

'
cer
n
e

is certificate of the node that exists as two hop
level to current rely node.
[Here
cer
n
e
for senders node is
'
cer
n
e
for current rely node]
( )
'
lt cer cTs iTs
n n n
e r e
=

Here cTs
n
r
is timestamp at current relay node
cer
n
e
is certificate carried by RREQ packet
Step 4:
a. If ( ) lt cer
n n
s s
< and cer cer
n n
s e
== then source node n
s

b. If cer
n
e
is valid then that RREQ packet will be considered
and continues step 5 else that packet will be discarded.
Step 5:
a. If
add add
n n
d r
=
then Update the RREQ packet as add
n
s
< ,
add
n
d
,
add
n
r
,
cer
n
e
, , cer
n
r
cer
n
i
,
n
r
, ECPK
n
s
> and
transmits to
i
n .
b. Else if add add
n n
d r
n
r
identified as destination node and
starts RREP process

2) Process of RREQ construction at relay hop node of the
source node
Once packet received by next hop (in that packet referred
as) then continues the above four steps in sequence with minor
changes, described here:
TABLE 2: ALGORITHM FOR RREQ PACKET EVALUATION AT
RELAY NODE THAT IS NOT HOP NODE TO SOURCE NODE
Step 1:
If p S
id p
id
e then RREQ packet will discarded else adds
p
id
to S
p
id
and continues step Step 2:
Step 2:
a. If ( ) lt cer
n
i
n n
i r
== then continues
step 3, else discards RREQ packet.
[Here cer
n
i
is cer
n
r
because the current node certificate is
available at sender node as certificate of one hop node that
acts as target for ingress transaction]
.
Step 3:
a. If
cer
n
e
is not null and
( ) lt cer
n n
e e
<
and
'
cer cer
n n
e e
==
then
cer
n
e
is valid and continues step 4,

'
cer
n
e
levels to current rely node.
[Here
cer
n
e
for senders node is
'
cer
n
e

( )
'
lt cer cTs iTs
n n n
e r e
=

cTs
n
r
is timestamp at current relay node
cer
n
e
is certificate carried by RREQ packet
Vol. 3, No. 7, 2012
83 | P a g e
Step 4:
a. If ( ) lt cer
n n
r e
< and cer cer
n n
r e
== then node n
r
is
valid and continues else discards the RREQ packet

Step 5:
a. If
add add
n n
d r
=
then update the RREQ packet as add
n
s
< ,
add
n
d
, { , , ...... , , }
1 2 2 1
add add add add add
n n n n n
r r r
,
cer
n
e
,
, cer
n
r
cer
n
i
,
n
r
, ECPK
n
s
> and transmits to
i
n

. b. Else if add add
n n
d r
n
r
identified as destination node
and starts RREP process

When compared algorithm in table 2 with algorithm in
table 1, a change can be observable at step 3, we are not
accepting certificate carried by RREQ as null, since

representing in RREQ packet is not source node.
V. RREP PROCESS
Once
d
n receives RREQ it performs verification as
mentioned in table 2.
Upon successful validation, It performs fallowing
functionality.
If RREQ that was received is valid then It collects ECPK
s
and calculates ECPK
d
(Elliptic curve cryptography approach
explained in next section B).
Then it constructs RREP packet at n
d
as follows:
{ , , , , , , , , } add add ECPK lst cer cer cer p
s d d n n n n n id
r e i d d

Since the RREP packet constructed at
d
n , cer
n
e
is null.
A. Process of RREP packet validation and construction at
first hop node of the destination node
TABLE 3: ALGORITHM FOR RREP PACKET EVALUATION AT HOP
NODE OF THE DESTINATION NODE
Here n
r
is hop node of the
d
n
Step 1: If n lst
r n
r
e then continues step 2 else discards
RREP packet
Step2: If p S
id p
id
e

then RREP packet will discarded
else adds p
id
to S
p
id
and continues step 3
Step 3:
If ( ) lt cer
n
i

n n
i r
==

then continues step 4,
else discards RREP packet.
[Here cer
n
i
is cer
n
r
because the current node certificate is
available at sender node as certificate of one hop node
that acts as target for ingress transaction]
Step 4: If cer
n
e
is null then assumes sender is source of the
RREP and continues
Else If
cer
n
e
is not null and
( ) lt cer
n n
e e
<
and

'
cer cer
n n
e e
==
then
cer
n
e
is valid and continues
step 4,
else RREP will be discarded.

'
cer
n
e
level to current rely node.
[Here
cer
n
e
for senders node is
'
cer
n
e
( )
'
lt cer cTs iTs
n n n
e r e
=

Here cTs
n
r
is timestamp at current relay node, cer
n
e
is
certificate carried by RREP packet
Step 5: If ( ) lt cer
n n
d d
< and cer cer
n n
d e
== then source
node n
d
else that packet will be discarded.
Step6: If add add
n n
s r
=

then Update the RREP packet as
, , , , , , , , add add ECPK lst cer cer cer p
n n d n n n n n id
s d r e r i r
< >

and transmits to
i
n .
Else if add add
n n
s r
n
r
identified as source node and
stops RREP process

B. Process of RREP construction at relay hop node of the
source node
Once RREP packet received by next hop (in that packet
referred as n
i
) then verifies and continues the process as
described in table 4:
TABLE 4: ALGORITHM FOR RREP PACKET EVALUATION AT
RELAY NODE THAT IS NOT HOP NODE TO DESTINATION NODE
Step 4:
If
cer
n
e
is not null and
( ) lt cer
n n
e e
<
and
Step 1:
If n lst
r n
r
e then continues step 2 else discards RREP
packet
Step 2:
If p S
id p
id
e then RREP packet will discarded
else adds p
id
to S
p
id
and continues step 3
Step 3:
If ( ) lt cer
n
i
n n
i r
== then continues
step 4,
else discards RREP packet.
[Here cer
n
i
is cer
n
r
because the current node
certificate is available at sender node as certificate of
one hop node that acts as target for ingress
transaction]
Vol. 3, No. 7, 2012
84 | P a g e
'
cer cer
n n
e e
==
then
cer
n
e
is valid and continues step 5,
else RREP will be discarded.
'
cer
n
e
is certificate of the node that exists as two hop levels
to current rely node.
[Here
cer
n
e
for senders node is
'
cer
n
e

( )
'
lt cer cTs iTs
n n n
e r e
=

cTs
n
r
is timestamp at current relay node,
cer
n
e
is certificate carried by RREP packet
Step 5:
If ( ) lt cer
n n
r e
< and cer cer
n n
r e
== then node n
r
is valid
and continues step 5 else discards the RREP packet
Step 6 :If add add
n n
s r
= then update the RREP packet as
, , , , , , , , add add ECPK lst cer cer cer p
n n d n n n n n id
s d r e r i r
< >
and transmits to
i
n .
Else if add add
n n
s r
n
r
identified as source node and stop
RREP process, collects routing path information.

1) Elliptic Curve Cryptography for constrained
environments
To form a cryptographic system using elliptic curves, we
need to find a hard problem corresponding to factoring the
product of two primes or taking the discrete algorithm.
Consider the equation Q kP = , where , Q P belongs to
elliptic curve over (2 )
n
GF and 2
n
k < . It is relatively easy
to calculate Q given k and P , but it is relatively hard to
determine k given Q and P . This is called the discrete
algorithm problem for elliptic curves.
KEY EXCHANGE
Key exchange can be done in the following manner. A
large integer 2
n
q = is picked and elliptic curve parameters a
and b . This defines an elliptic curve group of points. Now,
pick a base point G = (x1,y1) in E (a,b) whose order is a very
large value n. The elliptic curve E and G are the parameters
known to all participants.
A key exchange between users A and B can be
accomplished as follows:
a) A selects an integer
A
n less than n . This is private
key of user A. Then user A generates a public key
*
A A
ECPK n G = ; then the public key
A
ECPK is appoint
on E.
b) User B similarly selects a private key
B
n and
computes a public key
B
ECPK .
c) User A generates the secret key *
A B
k n ECPK =
and user B generates the secret key *
B A
k n ECPK = .
The calculations in step 3 produce the same result.
* *( * ) *( * ) *
A B A B B A B A
n ECPK n n G n n G n ECPK
Strength of this key exchange process is to break this
scheme, an attacker would need to be able to compute k given
G and kG, which is assumed hard and almost not possible in
constrained environments
C. Elliptic Curve Encryption/Decryption
The plaintext message m is taken as input in the form of
bits of varying length. This message m is encoded and is sent
in the cryptographic system as x-y point
m
P . This point is
encrypted as cipher text and subsequently decrypted. The SHA
hash function algorithm can be used as Message digestion and
Signature authentication and verification for the message.
As with the key exchange system, an
encryption/decryption system requires a point G and an elliptic
group E (a, b) as parameters. Each user A selects a private key
A
n and generates a public key *
A A
ECPK n G = .
To encrypt and send a message
m
p to user B, A chooses a
random positive integer k and produces the cipher text
m
c
consisting of pair of points { , }
m m B
c kG p kECPK = +
User A has used public key
B
ECPK of user B. To decrypt
the cipher text, user B multiplies the first point in the pair by
secret key of user B and subtracts the result from the second
point:
( ) ( ) ( )
m B B m B B m
p kECPK n kG p k n G n kG p + = + =
The implementation of elliptic curve algorithm is done over
163
(2 ) GF for providing security of more than 128 bits.
VI. ROUTING THE DATA PACKETS:
Here in this section we describe the procedure of
authentication data packets forwarded from the source node to
the destination node, along the selected route, while checking
for faulty links.
In DSR, the source route information is carried in each
packet header.
A. FHC: Fixed Hash Chaining
An algorithmic description of the FHC for data packet
transmission
1. A verification process takes place for egress of each
node in routing path.
2. A counter sign
c
pw will be used for data packet
c
p
that to be sent currently in sequence.
3. Packet
c
p transmits data
c
d .
Vol. 3, No. 7, 2012
85 | P a g e
4. An irreversible code
c
t generated by applying a
secure hash function
( ) h
f on countersign
c
pw .
5. Let
n
pw as countersign for next packet
n
p that is in
sequence, which fallows the packet
c
p
6. Packet
n
p includes data
n
d , and irreversible code
n
t
will be generated by hashing
n
pw using
( ) h
f .
7. Hash Tag
c
H will be generated by using secure
hashing
( ) h
f that uses
n
d ,
n
t and
c
pw as input.
8.
c
p Is then transmitted that includes the
c
H ,
c
d ,
c
t ,
password
p
pw of packet
p
p that sent before
c
p in
sequence to authenticate
c
d .
Algorithm to authenticate sequence transmission of the
packets:
Countersign
( )
c
p
cs will be selected for
c
p
that includes
c
d
to be transmitted.
c
t
=
( ) ( )
( )
c
h p
f cs
Countersign
( )
n
p
cs will be selected for
n
p
with data
n
d
to
be transmitted in sequence,
n
t
=
( ) ( )
( )
n
h p
f cs
Apply
( ) h
f to the
n
d
,
n
t
and
( )
c
p
cs that creates
authentication tag for
n
p
referred as
( )
n
p
at
( ) ( ) ( )
( , , )
n c
p h n n p
at f d t cs = < > ;
Transmit
c
p
from a source node
s
n
to a destination node
d
n
through hops in path selected through optimal route
selection strategy.
The currently transmit
( )
p
p
cs of packet
p
p that transmitted
before
c
p
to authenticate
c
d
.
In the interest of route maintenance, every hop in rout
contains a cache that maintains hop list describing the route
selected using an optimal route selection model. We apply
( ) h
f on cache of each hop of the route to verify the integrity
of the hop list cached.
Architecture of the proposed protocol
Proposed model provides ting packet contains
( )
, ,
n
p c c
at d t and a countersign
an authentication protocol for a wireless ad hoc network where
packets are transmitted serially. By serially, we mean a current
packet
c
p is immediately preceded by a previous packet
p
p ,
and followed immediately by a next packet
n
p .
More particularly, during a route discovery phase, we
provide secure route selection, i.e., a shortest intact route, that
is, a route without any faulty links. During route maintenance
phase, while packets are forwarded, we also detect faulty links
based on a time out condition. Receiving an acknowledgement
control packet signals successful delivery of a packet.
For packet authentication, we use
( ) h
f described by
Benjamin Arazi et al [21]. The hash function encodes a
countersign to form a tag.
By
( ) h
f we mean that the countersign cannot be decoded
from the tag and the countersign is used only once, because
part of its value lies in its publication after its use. We have
adapted that protocol for use in an ad hoc network where
multiple packets need to be sent sequentially. Therefore, if a
number of packets are sent sequentially, the countersign needs
to be refreshed each time. Thus, a single authentication is
associated with a stream of future packets that is significant
difference between proposed and existing hash chain
techniques. The existing models require stream of future
events. In addition, the countersign is used to authenticate
c
p
but not for future packets.
As an advantage over prior art asymmetric digital signature
or secret countersigns do not need to be known ahead of time
or distributed among the nodes after the system becomes
operational. It should also be noted, that each countersign is
used only one time, because the countersign is published to
perform the authentication.
The
( ) h
f as implemented by the proposal is ideal for
serially communicating packets along a route in an ad hoc
network, without requiring the nodes to establish shared secret
countersigns beforehand.
The protocol includes the following steps. Select a random
countersign
r
cs . Form a tag
r
t ,
r
t =
( )
( )
h r
f cs , Construct a
message
r
mac . Form a hash value
( )
( , , )
r h r r r
H f mac t cs = < > , and make it public. Perform
the act and reveal , ,
r r r
mac t cs to authenticate the act.
B. Data transmission and malicious hop detection
To send a packet
i
m that is a part of data to be sent to
destination node
d
n , the source node
s
n picks two counter
signs
1
,
r r
cs cs
+
and fixes the time limit to receive either one of
Vol. 3, No. 7, 2012
86 | P a g e
packet delivery acknowledgement ack or a control packet
ack
mn that acknowledges about malicious link in the route
path. The source node sends message with the format
( ) ( ) ( ) ( ) 1 ( ) 1
{ , ( ), ( , ( )), ( , ( ), )}
i i h r ds i h r h i h r r
msg m f cs f m f cs f m f cs cs
+ +
=
to the
h
n along the route.
Here
( ) ( )
( , ( ))
ds i h r
f m f cs is a digital signature to verify
( )
( , ( )
i h r
m f cs by intermediate hops of the route selected, so
that every
h
n and
d
n can verify that
( )
( , ( ))
i h r
m f cs is
valid and indeed generated by the claimed
s
n .
Then each hop updates route table entry for source node S
by recording
( )
( )
h r
f cs as ( )
r s
hcs n ,
( ) 1 ( ) 1
( , ( ), )
h i h r r
f m f cs cs
+ +
as
2
( )
e s
h n , which is used to
authenticate an immediate fallowing message
1 i
msg
+
in
sequence.
When sending the data packet
1 i
m
+
, the
s
n selects another
countersign
2 r
cs
+
and forwards the
1 i
msg
+
to the first hop of
the selected path:
1 1 ( ) 1 ( ) 2 ( ) 2 1
{ , ( ), , ( , ( ), )}
i i h r r h i h r r
msg m f cs cs f m f cs cs
+ + + + + +
=
Each node on the route calculates
( )
( )
h r
f cs and compares
with ( )
r s
hcs n that available in routing table, if results equal
then
r
cs will be authenticated as valid. The
h
n then calculates
( ) 1 ( ) 1
( , ( ), )
h i h r r
f m f cs cs
+ +
, and compares with
2
( )
e n
h s
result is equivalent then claims the validity of
1 ( ) 1
( , ( ))
i h r
m f cs
+ +
. The node then updates its routing entry
by recording
1 ( ) 1
( )
r h r
hrc f rc
+ +
= and
( 2) ( ) ( 2) ( ) 2 1
( ) ( , ( ), )
e s h i h r r
h n f m f cs r
+ + +
= , and forwards the
data packet to the node along the route as specified in the
header of the packet header.
During the packet sending process described earlier, if any
of the checks fails, then the packet is dropped. If both checks
succeed, then the node updates its routing entry associated
with
s
n . If the check at
h
n , then either
1 h
n

or
( ) 1 ( ) 1
( , ( ), )
h i h r r
f m f cs cs
+ +
in
i
msg has been modified, or
node
1 h
n

modified
( ) 1 ( ) 1
( , ( ), )
h i h r r
f m f cs cs
+ +
in
1 i
msg
+
.
In either case, the current hop node
h
n drops the packet.
Consequently, hop node
1 h
n

does not receive a valid ack
after time out, and the node can report a malicious activity at
1
( , )
h h
n n
connection, or the hop node

2 h
n

reports about
malicious activity between
2 1
( , )
h h
n n

to
s
n . In either case,
the fault link includes the malicious node
1 h
n

.
In our proposed model the authentication tag of each
packet limited to two hashes and one countersign; while in the
existing models required N authentication tags for a route with
N hops. Therefore, our method has a lower communication
and storage overhead.
The packet authentication process at
d
n is identical to the
authentication process at any intermediate hop
h
n . If any of
the checks fails, then the packet is dropped. If both checks
succeed, the packet is delivered successfully, and schedules
the ack for transmission along the reverse of path of the
route. The ack reflects the packet identification number i .
The destination node also appends an authentication tag to
the ack message for the nodes on the reverse path. The
authentication tag bears the same structure as the one
generated by the source node. Specifically, when sending
i
ack , for the packet
i
m , the destination node randomly
selects two countersigns
re
cs and
1 re
cs
+
, and sends the
following information:
( ) ( ) ( ) ( ) 1 ( ) 1
, ( ), ( , ( )), ( , ( ), )
i h re ds i h i h i h re re
ack f cs f ack f ack f ack f cs cs
+ +
Similarly,
( ) ( )
( , ( ))
ds i h re
f ack f cs is used to verify
( )
( , ( ))
i h re
ack f cs by each node along the reverse path of the
route. When sending the acknowledgement for packet
i
m ,
the destination selects a new countersign
1 re
cs
+
and forwards:
1 ( ) 1 ( ) 2 ( ) 2 1
( , ( ), , ( ( ), ))
i h re re h i h re re
ack f cs cs f ack f cs cs
+ + + + +
.
.If the timeout at an intermediate node expires, then that
node sends
ack
mn with an identification number according to
our hash function for authentication of the
ack
mn by the
upstream nodes. When a node receives the ack , the node
verifies its authenticity and that a timeout is pending for the
corresponding data packet. If the ack
is not authentic or a
timeout is not pending, the node discards the ack . Otherwise;
the node cancels the timeout and forwards the ack to the
next node.
When a node receives
ack
mn , it verifies its authenticity,
and that a timeout is pending for the corresponding data
packet, and that the link reported in the
ack
mn is the first
downstream to the node that generated
ack
mn . If the
ack
mn is
not authentic, or a timeout is not pending, or the link is not the
downstream to the node reporting
ack
mn , then the node
drops
ack
mn . Otherwise, the node cancels the timeout and
further forwards the
ack
mn control packet. Upon receiving
Vol. 3, No. 7, 2012
87 | P a g e
ack
mn , the source node deletes the link that connecting
h
n
referred in
ack
mn and finds a new route. In this proposed
model, the packets are always received as in the order they
sent. This is because all packets are forwarded along the same
route in DSR. In the case of congestion and buffering, the
messages are stored in a first-in-first-out buffer according to
the order that they are received.
The experiments were conducted using NS 2. We build a
simulation network with hops under mobility and count of 50
to 200. The simulation parameters described in table 5.
Authentication ensures that the buffer is properly allocated to
valid packets. The simulation model aimed to compare
ARIADNE [11] and FHC-NCTSR for route establishing
phase, SEAD[10] and FHC-NCTSR for data transmission. The
performance check of ARIADNE[11] and FHC-NCTS
protocols carried out against to the threats listed below.
a) Rushing attack
b) Denial of service
c) Routing table modification
d) Tunneling
The protection against tunneling attack is the advantage of
the NCTS-DSR over Ariadne.
TABLE5: SIMULATION PARAMETERS THAT WE CONSIDERED FOR
EXPERIMENTS
Number of nodes Range 50 to 200
Dimensions of space 1500 m 300 m
Nominal radio range 250 m
Sourcedestination pairs 20
Source data pattern (each) 4 packets/second
Application data payload size 512 bytes/packet
Total application data load
range
128 to 512 kbps
Raw physical link bandwidth 2 Mbps
Initial ROUTE REQUEST
timeout
2 seconds
Maximum ROUTE REQUEST
timeout
40 seconds
Cache size 32 routes
Cache replacement policy FIFO
Hash length 80 bits
certificate life time 2 sec
The metrics to verify the performance of the proposed
protocol are
a) Data packet delivery ratio: It can be calculated as
the ratio between the number of data packets that are sent by
the source and the number of data packets that are received by
the sink.
b) PACKET DELIVERY FRACTION: It is the ratio of
data packets delivered to the destinations to those generated
by the sources. The PDF tells about the performance of a
protocol that how successfully the packets have been
delivered. Higher the value gives the better results.
c) AVERAGE END TO END DELAY: Average end-to-
end delay is an average end-to-end delay of data packets.
Buffering during route discovery latency, queuing at interface
queue, retransmission delays at the MAC and transfer times,
may cause this delay. Once the time difference between
packets sent and received was recorded, dividing the total time
difference over the total number of CBR packets received gave
the average end-to-end delay for the received packets. Lower
the end to end delay better is the performance of the protocol.
d) Packet Loss: It is defined as the difference between
the number of packets sent by the source and received by the
sink. In our results we have calculated packet loss at network
layer as well as MAC layer. The routing protocol forwards the
packet to destination if a valid route is known, otherwise it is
buffered until a route is available. There are two cases when a
packet is dropped: the buffer is full when the packet needs to
be buffered and the time exceeds the limit when packet has
been buffered. Lower is the packet loss better is the
performance of the protocol.
e) ROUTING OVERHEAD: Routing overhead has been
calculated at the MAC layer which is defined as the ratio of
total number of routing packets to data packets.
Figure 3(a) shows the Packet Delivery Ratio (PDR) for
FHC-NCTSR, ARIADNE and SEAD. Based on these results it
is evident that FHC-NCTSR recovers most of the PDR loss
that observed in ARIADNE against to SEAD. The
approximate PDR loss recovered by FHC-NCTSR over
ARIADNE is 1.5%, which is an average of all pauses. The
minimum individual recovery observed is 0.18% and
maximum is 2.5%. Figure 3(b) indicates ARIADNE minimal
advantage over FHC-NCTSR in Path optimality. FHC-
NCTSR used average 0.019 hops longer than in ARIADNE
because of the hop level certification validation process of the
FHC-NCTSR that eliminates nodes with invalidate certificate.
Here slight advantage of ARIADNE over FHC-NCTSR can be
observable.
The packet delivery fraction (PDF) can be expressed as:
1
'
1
* '
e
f
f
f
R
P
N
P P
c
=
=
=

- P is the fraction of successfully delivered
packets,
- c is the total number of flow or connections,
- f is the unique flow id serving as index,
-
f
R is the count of packets received from flow
f
-
f
N is the count of packets transmitted to flow
f .
Figure 3(c) confirms that FHC-NCTSR is having fewer
packets overhead when compared to ARIADNE. Due to stable
paths with no compromised or victimized nodes determined by
Vol. 3, No. 7, 2012
88 | P a g e
FHC-NCTSR this advantage become possible. The Packet
overhead observed in ARIADNE is average 5.29% more than
packet overhead observed in FHC-NCTSR. The minimum and
maximum packet overhead in ARIADNE over FHC-NCTSR
observed is 3.61% and 7.29% respectively. It is quite evident
from fig 3(c), that SEAD is not stable to handle the packet
overhead, over a period of time the packet overhead is
abnormal compared to other two protocols considered.
MAC load overhead is slightly more in FHC-NCTSR over
ARIADNE. We can observe this in figure 3(d), which is
because of additional control packet exchange in FHC-
NCTSR for neighbor hop validation through certificate
exchange. The average MAC load overhead in FHC-NCTSR
over ARIADNE 1.64%. The minimum and maximum MAC
load overhead observed is 0.81 and 3.24% respectively.

(a) Packet delivery ratio comparison using line chart

(b) Bar chart representation of Path optimality

(c) A line chart representation of Packet overhead comparison report

(d) Mac load comparison represented in bar chart format

(e) Hash chaining cost comparison report
In fig 3(e) we describe the performance of FHC-NCTSR
over ARIADNE and SEAD in terms of Hash chain evaluation
cost. Let be the cost threshold to evaluate each hash in hash
chain.
We measure the Hash chain evaluation cost as
1 1
z n
i j
z
= =

, here z is number of nodes and n is number of hashes, as
of the chaining concept of SEAD and ARIADNE z is equal to
n but in FHC-NCTSR n always 2
VII. CONCLUSION
This paper was presented an evaluation of security
protocols such as QoS-Guided Route Discovery [13],
sQos[15], Ariadne [16] and CONFIDANT [17], which are
based on reactive DSR approach, and describes their
limitations and attacks against these protocols that can be
subtle and difficult to discover by informal reasoning about
the properties of the protocols.
The proposed a hybrid protocol FHC-NCTSR protocol
applies digital signature exchange on the RREQ and RREP
that they contribute the neighbors within 2 hops away from a
node in computing them and a fixed hash chaining technique
was used to achieve scalable data sending process. In route
discovery phase, these digital signatures enable the protocol to
Vol. 3, No. 7, 2012
89 | P a g e
avoid malicious nodes from participating in routing and route
discovery and also able to detect falsified routing messages
and the responsible nodes.
And the fixed hash chaining in data transfer limits the
computation cost and resource utilization.
VIII. REFERENCES
[1] P. Samar and S. B. Wicker, On the behavior of communication links in a
multi-hop mobile environment, in Frontiers in Distributed Sensor
Networks, S. S. Iyengar and R. R. Brooks, Eds. Boca Raton, FL: CRC
Press, 2004.
[2] C.E. Perkins and P. Bhagwat, Highly dynamic destination-sequenced
distance-vector routing (DSDV) for mobile computers, Proc. ACM
SIGCOMM, vol. 24, no. 4, pp. 234244, Oct. 1994.
[3] B. Bellur and R. G. Ogier, A reliable, efficient topology broadcast
protocol for dynamic networks, presented at the IEEE INFOCOM, Mar.
1999.
[4] S. Murthy and J. J. Garcia-Luna-Aceves, An efficient routing protocol
for wireless networks, MONET, vol. 1, no. 2, pp. 183197, Oct. 1996.
[5] C. E. Perkins and E. M. Royer, Ad hoc on-demand distance vector
routing, presented at the IEEE WMCSA, New Orleans, LA, Feb. 1999.
[6] D. B. Johnson and D. A. Maltz, Dynamic source routing in ad hoc
wireless networking, in Mobile Computing, T. Imielinski and H. Korth,
Eds. Boston, MA: Kluwer, 1996.
[7] V. D. Park and M. S. Corson, Ahighly adaptive distributed routing
algorithm for mobile wireless networks, presented at the IEEE
INFOCOM, Kobe, Japan, Apr. 1997.
[8] Perlman: Network Layer Protocols with Byzantine Robustness, Ph.D.
thesis, MIT LCS TR-429, October 1998.
[9] Hauser: Reducing the Cost of Security in Link State Routing,
Symposium on Network and Distributed Systems Security, February
1997
[10] Hu: SEAD: Secure efficient distance vector routing for mobile wireless
ad hoc networks, Fourth IEEE Workshop on Mobile Computing
Systems and Applications June 2002
[11] Hu: Ariadne: A secure On-Demand Routing Protocol for Ad hoc
Networks, MobiCom, September 2002
[12] Perrig: Efficient and Secure Source Authentication for Multicast,
Network and Distributed System Security Symposium, February 2001
[13] T. Clausen and P. Jacquet, Optimized link state routing protocol
(OLSR), IETF, RFC 3626, Oct. 2003.
[14] M. Gerla, X. Hong, and G. Pei, Landmark routing for large ad hoc
wireless networks, presented at the IEEE GLOBECOM, San Francisco,
CA, Nov. 2000.
[15]Z.J.Haas and M. R. Pearlman, The performance of query control
schemes for the zone routing protocol, IEEE/ACM Trans. Networking,
vol. 9, pp. 427438, Aug. 2001.
[16] Z. J. Haas, M. R. Pearlman, and P. Samar, The bordercast resolution
protocol (BRP) for ad hoc networks, IETF,MANET Internet Draft, July
2002.
[17] The interzone routing protocol (IERP) for ad hoc networks, IETF,
MANET Internet Draft, July 2002.
[18] The intrazone routing protocol (IARP) for ad hoc networks, IETF,
MANET Internet Draft, July 2002.
[19] The zone routing protocol (ZRP) for ad hoc networks, IETF, MANET
Internet Draft, July 2002.
[20] A. Iwata, C.-C. Chiang, G. Pei, M. Gerla, and T.-W. Chen, Scalable
routing strategies for ad hoc wireless networks, IEEE J. Select. Areas
Commun., vol. 17,Aug. 1999.
[21] B. Liang and Z. J. Haas, Hybrid routing in ad hoc networks with a
dynamic virtual backbone, IEEE Trans. Wireless Commun., to be
published.
[22] A. B. McDonald and T. Znati, Predicting node proximity in Ad-Hoc
networks: A least overhead adaptive model for electing stable routes,
presented at the MobiHoc 2000, Boston, MA, Aug. 2000. , Mar. 1997

Vol. 3, No. 7, 2012

90 | P a g e
Simultaneous Estimation of Geophysical Parameters
with Microwave Radiometer Data based on
Accelerated Simulated Annealing: SA
Kohei Arai
Graduate School of Science and Engineering
Saga University
Saga City, Japan

Abstract Method for geophysical parameter estimations with
microwave radiometer data based on Simulated Annealing: SA is
proposed. Geophysical parameters which are estimated with
microwave radiometer data are closely related each other.
Therefore simultaneous estimation makes constraints in
accordance with the relations. On the other hand, SA requires
huge computer resources for convergence. In order to accelerate
convergence process, oscillated decreasing function is proposed
for cool down function. Experimental results show that
remarkable improvements are observed for geophysical
parameter estimations.
Keywords- simulated annealing; microwave radiometer water
vapor; air-temperature; wind speed.
I. INTRODUCTION
Microwave radiometer allows estimation of geophysical
parameters such as water vapor, rainfall rate, ocean wind speed,
salinity, soil moisture, air-temperature, sea surface temperature,
cloud liquid, etc. based on least square method. Due to the fact
that relation between microwave radiometer data (at sensor
brightness temperature at the specified frequency) and
geophysical parameters is non-linear, non-linear least square
method is required for the estimations. Although there are some
methods which allow estimation optimum solutions, Simulated
Annealing: SA method is just one method for finding global
optimum solution.
Other methods, such as steepest descending method,
conjugate gradient method, etc. gives one of local minima, not
the global optimum solution. SA, on the other hand, requires
huge computer resources for convergence. In order to
accelerate the convergence process, not the conventional
exponential function with the temperature control, but
osculated decreasing function is employed for cool down
function. Geophysical parameter estimation based on simulated
annealing is proposed previously [1]. It takes relatively long
computational time for convergence. Moreover, optimization
with constraints makes much accurate estimation of
geophysical parameters. Some of the constraints is relation
among the geophysical parameters.
Geophysical parameters have relations each other. For
instance, sea surface temperature and water vapor has a
positive relation, in general.
Therefore, it is better to estimate several geophysical
parameters simultaneously rather than the estimation for single
parameter. The proposed method is based on modified SA
algorithm and is for simultaneous estimation for several
geophysical parameters at once. Some experiments are
conducted with Advanced Microwave Scanning Radiometer:
AMSR [2] onboard AQUA satellite. Then it is confirmed
that the proposed method surely works for improvement of
estimation accuracy for all the geophysical parameters.
The following section describes the proposed method
followed by some experimental results. Then conclusion with
some discussions is followed in the final section.
II. PROPOSED METHOD
A. Schematic View of AMSR Observations
AMSR onboard AQUA satellite observes sea surface
through the atmosphere with absorptions due to atmospheric
molecules, water vapor, aerosol particles as shown in Fig.1.
Such atmospheric continuants also radiated depending on their
temperature and emissivity. Attenuation due to absorption can
be shown in Fig.2.
Using absorption spectrum due to water (22.235GHz), it is
possible to estimate water vapor in the atmosphere. It also is
possible to estimate air-temperature at the sea surface with
much lower frequency channels, 5 to 10 GHz. Using oxygen
absorption frequency, it is possible to estimate air temperature
profile. Meanwhile, cloud liquid water can be estimated with
around 37GHz frequency channel.
Figure 1 Radiative transfer from the sea surface to the AMSR onboard AQUA
Vol. 3, No. 7, 2012

91 | P a g e
satellite.

Figure 2 Attenuation due to atmospheric continuants, water vapor, cloud liquid,
and oxygen.
On the other hand, there is sea surface model. Incident
microwave energy (incident energy) reflects at the sea surface
(radiant energy) and penetrates a little bit (transparent energy)
as shown in Fig.3 (a). Sea surface can be modeled as shown in
Fig.3 (b). There are some models for shape of the ocean waves
such as non-symmetrical Gaussian function. Depending on
ocean wind speed, forms or bubbles are generated at the sea
surface results in changing the emissivity of the sea surface.
More detailed description of sea surface model is described in
the following section.

(a)

(b)
Figure 3 Sea surface model
Observed brightness temperature at AMSR can be
expressed with equation (1).
] [
j i
B S j BU B
T T T T

(1)
where T
Bi
, T
BU
, , ,T
S
, T
B
are at sensor brightness
temperature, radiance from the atmosphere, atmospheric
transparency, sea surface emissivity, sea surface temperature,
reflected atmospheric radiance at sea surface, respectively.
Root Mean Square: RMS difference between actual brightness
temperature acquired with microwave antenna and model based
brightness temperature with geophysical parameters is
expressed with equation (2).
2
| ]) [ ( | ) , , (

B S BU B S
T T T T V W T f (2)
The RMS difference is a function of sea surface
temperature, water vapor, ocean wind speed. Thus geophysical
parameters can be estimated through minimization of the RMS
difference.
B. Observation Models
There are some atmospheric and ocean surface models in
the microwave wavelength region. Therefore, it is possible to
estimate at sensor brightness temperature (microwave
radiometer) with the geophysical parameters. The real and the
imaginary part of dielectric constant of the calm ocean surface
is modeled with the SST, salinity (conductivity). From the
dielectric constant, reflectance of the ocean surface is estimated
together with the emissivity (Debue, 1929 [3]; Cole and Cole,
1941 [4]). There are some geometric optics ocean surface
models (Cox and Munk, 1954 [5]; Wilheit and Chang, 1980
[6]). According to the Wilheit model, the slant angle against the
averaged ocean surface is expressed by Gaussian distribution
function.
There is a relation between ocean wind speed and the
variance of the Gaussian distribution function as a function of
the observation frequency. Meanwhile the influence due to
foams, white caps on the emissivity estimation is expressed
with the wind speed and the observation frequency so that the
emissivity of the ocean surface and wind speed is estimated
with the observation frequency simultaneously. Meanwhile, the
atmospheric absorptions due to oxygen, water vapor and liquid
water were well modeled (Waters, 1976 [7]). Then atmospheric
attenuation and the radiation from the atmosphere can be
estimated using the models. Thus the at-sensor-brightness
temperature is estimated with the assumed geophysical
parameters.
Sea surface temperature estimation methods with AMSR
data are proposed and published [8] while ocean wind retrieval
methods with AMSR data are also proposed and investigated
[9]. Furthermore, water vapor and cloud liquid estimation
methods with AMSR data are proposed and studied [10].
The proposed method allows minimize the square of the
difference between the actual brightness temperature with
microwave radiometer onboard satellite and the estimated
brightness temperature using simulated annealing with the
assumed geophysical parameters. Then the geophysical
Vol. 3, No. 7, 2012

92 | P a g e
parameters are estimated when the square of the difference
shows the minimum.
In order to minimize the square of the difference between
the actual and the estimated brightness temperatures, a non-
linear optimization method is used. Because the relation
between the geophysical parameters and the brightness
temperature is not linear so that nonlinear optimization
methods are appropriate to use. Newton-Raphson iterative
method, conjugate gradient method, etc., are well known and
are widely used methods. These methods, however, do not
ensure to reach the global optimum. These methods tend to fall
in one of local minima. Only simulated annealing ensures to
reach a global optimum solution.
At-sensor-brightness temperature, TB is represented with
equation (1). The first term of the equation (1) is the
contribution from the atmosphere while the second term is that
from the ocean surface which consists of the Sea Surface
Temperature: SST of T
S
and the atmospheric radiance reflected
at the ocean surface, T
BX
. is the atmospheric transmittance
and is expressed with equation (3).
(3)
where denotes the absorption coefficient and h is the
observation angle while denotes the emissivity of the ocean
surface as a function of dielectric constant and the ocean winds.
The first term of the equation (1) is represented as follows:

(4)
where T denotes air-temperature at the altitude of h. Thus
the at-sensor-brightness temperature is estimated with the
assumed geophysical parameters, SST, wind speed, water
vapor and cloud water. Then the following cost function, the
square of the difference between the actual AMSR brightness
temperature, T
B
and the estimated brightness temperature, is
defined as equation (2). Then simulated annealing finds the
global optimum solution, the minimum cost function at which
the best estimation of the set of the geophysical parameters.
The cloud water is ignored from the cost function because
the cloud free data are selected for the experiment. In order to
emphasize the effectiveness of the proposed method, cloud free
data were reselected in this study. In this case the emissivity of
the ocean surface is the function of SST and wind speed.
Through a regressive analysis with the second order
polynomial with a plenty of experimental data, the following
empirical equations are derived for the observation frequency
of 10.65 GHz in normal direction.
The conventional geophysical parameter estimation method
is based on regressive analysis with a plenty of truth data and
the corresponding microwave radiometer data [11].
III. EXPERIMENTS
A. Preliminary Experiment
1km mesh size of Global Data Assimilation System: GDAS
data is used for truth data of geophysical parameters. Rectangle
area is extracted as a test site. The two corners are as follows,
15 degree N, 125 degree East
30 degree N, 135 degree East
south east china sea area. AMSR data used for the
experiments is shown in Fig.4. Aqua/AMSR-E which is
acquired at around 6 a.m. Japanese local time on 25 April in
2004 is used for the experiments. Rectangle in the Fig.4 shows
test site.

(a)Browse (b) Footprint
Figure 4 AQUA/AMSR-E data used acquired at 17:00 UT on April 25 in 2004.
One of the conventional methods is regressive equation as
shown in equation (5).

l
i
Bi ij j j
T c c P
1
0
(5)

Vol. 3, No. 7, 2012

93 | P a g e
where T
Bi
denotes observed brightness temperature with
AMSR frequency channel i while j denotes geophysical
parameter and c
ij
are regressive coefficients. Using AMSR data
(AQUA/AMSR-E data used acquired at 17:00 UT on April 25
in 2004) and the corresponding GDAS data, regressive
equation is created as follows,
(6)
(7)

where denotes the emissivity. The suffixes H and V mean
horizontal and vertical polarizations. Such these relations
between the emissivity and SST as well as wind speed are used
in the estimation of brightness temperature. Geophysical
parameters of the test site are shown in Table 1. Also regressive
analysis between geophysical parameters and brightness
temperature is conducted. Table 2 shows the regressive
coefficients.
Scatter diagrams for sea surface temperature, ocean wind,
water vapor are shown in Fig.5 (a), (b), (c), respectively.
Horizontal axes of Fig.5 are GDAS data derived geophysical
parameter values while vertical axes show AMSR derived
geophysical parameters. Linear approximation functions are
also shown in the figure. There are systematic differences
between GDAS and AMSR derived geophysical parameters in
particular for ocean wind.
TABLE I. MEAN AND VARIANCE OF THE GEOPHYSICAL PARAMETERS
FOR THE TEST SITE
Geophysical Parameter Mean Variance
Sea Surface Temperature 298.7[K] 6.1
Ocean Wind 6.0[m/s] 2.5
Water Vapor 24.4[mm] 29.9
TABLE II. REGRESSIVE COEFFICIENTS FOR GEOPHYSICAL PARAMETER
ESTIMATION
Sea Surface Temp. Ocean Wind Water Vapor
C01:178.15 C02:102.21 C03:-176.88
C11:0.7413 C12:-0.4719 C13:0.3383
C21:-0.2994 C22:0.4831 C23:-0.2257
C31:-0.7920 C32:0.1768 C33-1.6291
C41:0.2522 C42:-0.5609 C43:0.8949
C51:2.5049 C52:-0.0124 C53:5.6032
C61:-1.2528 C62:0.2142 C63:-2.8837
C71:-1.5317 C72:-0.4605 C73:-2.8433
C81:0.8166 C82:0.2954 C83:1.4401

(a)Sea Surface Temperature

(b)OceanWind

(c)Water Vapor
Figure 5 Scatter diagrams for sea surface temperature, ocean wind, water vapor
Vol. 3, No. 7, 2012

94 | P a g e
Also regressive errors for these geophysical parameters are
shown in Table 3.
TABLE III. REGRESSIVE ERRORS (RMS ERROR)
Geophysical Parameters RMS Error
Sea Surface Temperature 0.530[K]
Ocean Wind 0.754[m/s]
Water Vapor 0.899[mm]

B. Estimation Accuracy of the Proposed Method
Simulated Annealing: SA utilized proposed method is used
to minimizing the aforementioned RMS difference between
model based and the actual brightness temperature data. Fig.6
shows scatter diagrams between GDAS derived geophysical
parameter values and those for the proposed method. Scatter
diagrams for sea surface temperature, ocean wind, water vapor
are shown in Fig.6 (a), (b), (c), respectively. Horizontal axes of
Fig.6 are GDAS data derived geophysical parameter values
while vertical axes show those derived geophysical parameters
by the proposed method. Meanwhile, Fig.6 (d) shows
convergence processes for these geophysical parameters. As
shown in the figure, RMS difference of estimated and GDAS
derived geophysical parameter values decreases in comparison
to the conventional regressive analysis based method. Trends
of scatter diagrams for these geophysical parameters are almost
same.
Vertical axis of Fig.6 (d) shows residual error of SA.
Residual errors for these geophysical parameter decreases
sharply until the number of iteration is 10
6
, then gradually
decreases beyond that. In particular, residual error for sea
surface temperature decreases gradually until the number of
iteration is over 10
7
. It still decreases monotonically.
C. Simultaneous Estimation of Geophysical Parameters
Some of geophysical parameters are highly correlated. For
instance, there is much evaporation in high sea surface
temperature areas, then water vapor shows also high. The
following geophysical parameter estimation method is
proposed,
(1) Estimate one geophysical parameter (A), then the other
geophysical parameter (B) is estimated with the constraint
of the previously estimated geophysical parameter (A),
(2) Then estimate the geophysical parameter (A) with the
constraint of the estimated geophysical parameter (B),
(3) Repeat the processes (1) and (2) until the residual error is
less than the prior determined certain value.
This method is referred to the simultaneous estimation
method hereafter. RMS error of geophysical parameter
estimation by the simultaneous estimation method is shown in
Table 4.

(a)Sea Surface Temperature

(b)Ocean Wind Speed

(c)Water Vapor
Vol. 3, No. 7, 2012

95 | P a g e

(d)Convergence Processes
Figure 6 Scatter Diagrams between GDAS derived geophysical parameter
values and those derived from the proposed method.
TABLE IV. RMS ERROR OF GEOPHYSICAL PARAMETER ESTIMATIONS BY
THE SIMULTANEOUS ESTIMATION METHOD
Geophysical parameters RMS Error
Sea Surface Temperature
Ocean Wind
0.632[K]
0.846[m/s]
Sea Surface Temperature
Water Vapor
0.621[K]
0.853[mm]
Ocean Wind
Water Vapor
0.472[m/s]
0.565[mm]

Also these three geophysical parameters can be estimated
simultaneously. RMS errors for these geophysical parameters
estimated by the simultaneous estimation method with three
geophysical parameters are shown in Table 5.
TABLE V. RMS ERROR OF GEOPHYSICAL PARAMETER VALUES FOR THE
PROPOSED METHOD
Geophysical parameters RMS Error
Sea Surface Temp. 0.373[K]
Ocean Wind 0.731[m/s]
Water Vapor 0.787[mm]

In comparison to the RMS error of Table 4, RMS errors of
the simultaneous estimation method for ocean wind speed and
sea surface temperature are improved. On the other hand, water
vapor estimation accuracy in Table 5 shows some degradation
comparing to the combination between ocean wind speed and
water vapor in Table 4. This accuracy, however, is still better
than the conventional regressive analysis based method.
IV. CONCLUSION
Method for geophysical parameter estimations with
microwave radiometer data based on Simulated Annealing: SA
is proposed. Geophysical parameters which are estimated with
microwave radiometer data are closely related each other.
Therefore simultaneous estimation makes constraints in
accordance with the relations. As results, it is found that the
proposed method is superior to the conventional regressive
analysis based method by 27, 25, and 22% improvements for
sea surface temperature, ocean wind speed, and water vapor,
respectively. Simulated Annealing which allows find global
optimum solution is used for improvement of estimation
accuracy in the near future.
ACKNOWLEDGMENT
The author would like to thank Jun Sakakibara for his effort
to experiments.
REFERENCES
[1] Kohei Arai and J.Sakakibara, Estimation of SST, wind speed and water
vapor with microwave radiometer data based on simulated anealing,
Advances in Space Research, 37, 12, 2202-2207, 2006
[2] K.Tachi, Kohei Arai and Y.Satoh, Advanced Microwave Scanning
Radiometer -Requirements and Preliminary Design Study-, IEEE
Trans.on Geoscience and Remote Sensing, Vol.27, No.2, pp.177-183,
Jan.1989.
[3] Debue, R. Polar Molecules, Chemical Catalog, New York, 1929.
[4] Cole, K.S., Cole, R.H. Dispersion and absorption in dielectrics. J. Chem.
Phys. 9, 341351, 1941.
[5] Cox, C.S., Munk, W.H. Measurement of the roughness of the sea surface
from photographs of the sun_s glitter. J. Opt. Sci. Am. 44, 838850,
1954.
[6] Wilheit, T.T., Chang, A.T.C. An algorithm for retrieval of ocean surface
and atmospheric parameters from the observations of the Scanning
Multichannel Microwave Radiometer (SMMR). Radio Sci. 15, 525544,
1980.
[7] Waters, J.R. Absorption and emission by atmospheric gasses. in: Meeks,
M.L. (Ed.), Methods of Experimental Physics, vol. 12B.Academic,
Orland, 1976 (Chapter 2.3).
[8] Dong, SF; Sprintall, J; Gille, ST, Location of the antarctic polar front
from AMSR-E satellite sea surface temperature measurements,
JOURNAL OF PHYSICAL OCEANOGRAPHY, Nov 2006, 2075-2089.
[9] Konda, M., A. Shibata, N. Ebuchi, and K. Arai, An evaluation of the
effect of the relative wind direction on the measurement of the wind and
the instantaneous latent heat flux by Advanced Microwave Scanning
Radiometer, J. Oceanogr., vol. 62, no. 3, pp. 395-404, 2006.
[10] Cosh, M. H., T. J. Jackson, R. Bindlish, J. Famiglietti, and D. Ryu, A
comparison of an impedance probe for estimation of surface soil water
content over large region, Journal of Hydrology, vol. 311, pp. 49-58,
2005.
[11] Wentz, F. AMSR Ocean Algorithm, second version of ATBD,
NASA/GSFC, 2000.
AUTHORS PROFILE
Kohei Arai, He received BS, MS and PhD degrees in 1972, 1974 and 1982,
respectively. He was with The Institute for Industrial Science, and Technology
of the University of Tokyo from 1974 to 1978 also was with National Space
Development Agency of Japan (current JAXA) from 1979 to 1990. During
from 1985 to 1987, he was with Canada Centre for Remote Sensing as a Post-
Doctoral Fellow of National Science and Engineering Research Council of
Canada. He was appointed professor at Department of Information Science,
Saga University in 1990. He was appointed councilor for the Aeronautics and
Space related to the Technology Committee of the Ministry of Science and
Technology during from 1998 to 2000. He was also appointed councilor of
Saga University from 2002 and 2003 followed by an executive councilor of the
Remote Sensing Society of Japan for 2003 to 2005. He is an adjunct professor
of University of Arizona, USA since 1998. He also was appointed vice
chairman of the Commission A of ICSU/COSPAR in 2008. He wrote 30
books.
Vol. 3, No. 7, 2012

96 | P a g e
Task Allocation Model for Rescue Disabled Persons
in Disaster Area with Help of Volunteers

Kohei Arai
Graduate School of Science and
Engineering
Saga University
Saga City, Japan

Tran Xuan Sang
Faculty of Information Technology
Vinh University
Vinh City, Vietnam
Nguyen Thi Uyen
Faculty of Information Technology
Vinh University
Vinh City, Vietnam
Abstract In this paper, we present a task allocation model for
search and rescue persons with disabilities in case of disaster. The
multi agent-based simulation model is used to simulate the rescue
process. Volunteers and disabled persons are modeled as agents,
which each have their own attributes and behaviors. The task of
volunteers is to help disabled persons in emergency situations.
This task allocation problem is solved by using combinatorial
auction mechanism to decide which volunteers should help which
disabled persons. The disaster space, road network, and rescue
process are also described in detail. The RoboCup Rescue
simulation platform is used to present proposed model with
different scenarios.
Keywords- Task Allocation Model; Multi Agent-based Rescue
Simulation; Auction based Decision Making.
I. INTRODUCTION
Persons with disabilities suffer a much higher risk in the
case of disasters than persons without disabilities. The data of
recent disasters i.e. Tsunami, Katrina and earthquake shows
that the mortality of disabled people during the disaster were
very high (Ashok Hans, 2009). The reason for this is because
many handicapped people may face physical barriers or
difficulties of communication that they are not able to respond
effectively to crisis situations. They were not able to evacuate
by themselves. Obviously, disabled people need assistances to
evacuate.
While in the past, persons with disabilities were not taken
in consideration during the planning and mitigation of disaster
management, in more recent years, this group of population has
been realized as a prior target to help in emergency situations.
It is important to learn the needs of persons with disabilities
and the various forms of disabilities in order to help them
effectively and minimize the mortality. The rescue process for
persons with disabilities is a dynamic process under uncertainty
and emergency, therefore it is not easy to predict what will
happen in the rescue process. In that case, the computer
simulation can be used to simulate the rescue process with
various scenarios in the disaster area.
Most computer based simulation evacuation models are
based on flow model, cellular automata model, and multi-
agent-based model. Flow based model lacks interaction
between evacuees and human behavior in crisis. Cellular
automata model is arranged on a rigid grid, and interact with
one another by certain rules [1]. A multi agent-based model is
composed of individual units, situated in an explicit space, and
provided with their own attributes and rules [2]. This model is
particularly suitable for modeling human behaviors, as human
characteristics can be presented as agent behaviors. Therefore,
the multi agent-based model is widely used for evacuation
simulation [1-4].
Recently, Geographic Information Systems (GIS) is also
integrated with multi-agent-based model for emergency
simulation. GIS can be used to solve complex planning and
decision making problems [5-7]. In this study, GIS is used to
present road network with attributes to indicate the road
conditions.
We develop a task allocation model for search and rescue
persons with disabilities and simulate the rescue process to
capture the phenomena and complexities during evacuations.
The task allocation problem is presented by decision of
volunteers to choose which victims should be helped in order to
give first-aid and transportation with the least delay to the
shelter. The decision making is based on several criteria such
as health condition of the victims, location of the victims and
location of volunteers.
The rest of the paper is organized as follows. Section 2
reviews related works. Section 3 describes the proposed rescue
model and the task allocation model. Section 4 provides the
experimental results of different evacuation scenarios. Finally,
section 5 summarizes the work of this paper.
II. RELATED WORKS
There is considerable research in emergency simulation by
using GIS multi-agent-based models. Ren et al. (2009) presents
an agent-based modeling and simulation using Repast software
to construct crowd evacuation for emergency response for an
area under a fire. Characteristics of the people are modeled and
tested by iterative simulation. The simulation results
demonstrate the effect of various parameters of agents. Zaharia
et al. (2011) proposes agent-based model for the emergency
route simulation by taking into account the problem of
uncharacteristic action of people under panic condition given
by disaster. Drogoul and Quang (2008) discuss the intersection
between two research fields: multi-agent system and computer
simulation. This paper also presents some of the current agent-
based platforms such as NetLogo, Mason, Repast, and Gama.
Bo and Satish (2009) presents an agent-based model for
Vol. 3, No. 7, 2012

97 | P a g e
hurricane evacuation by taking into account the interaction
among evacuees. For the path finding, the agents can choose
the shortest path and the least congested route respectively.
Cole (2005) studied on GIS agent-based technology for
emergency simulation. This research discusses about the
simulation of crowding, panic and disaster management. Quang
et al. (2009) proposes the approach of multi-agent-based
simulation based on participatory design and interactive
learning with experts preferences for rescue simulation. Silvia
et al. (2005), Ranjit et al. (2001) and Santos et al. (2010) apply
the auction mechanism to solve the task allocation problem in
rescue decision making.
Through the view of this background, this study will focus
mainly on task allocation for volunteers to help disabled
persons. With effective task allocation method, it can improve
the rescue process. By considering the number of volunteers,
number of disabled persons and traffic condition as changing
parameters, we also draw the correlations between these
parameters and rescue time.
III. RESCUE AND TASK ALLOCATION MODEL
A. Rescue Simulation Model
The ability to receive critical information about an
emergency, how to respond to an emergency, and where to go
to receive assistance are crucial components of an evacuation
plan. In practical evacuation process, we assume that after the
warning is issued; all disabled persons send information to the
emergency center via a special device. This device measures
the condition of the disabled persons such as heart rate and
body temperature; the device can also be used to trace the
location of the disabled persons by GPS. Emergency center
will collect that information and then broadcast to volunteers
smart-phones through the internet. After checking the condition
of victims, volunteers make their own decision to help victims
and inform the emergency center.
The centralized rescue model is presented which has three
types of agent: volunteers, disabled people and route network.
The route network is also considered as an agent because the
condition of traffic in certain route can be changed when
disaster occurs. The general rescue model is shown in Figure 1.
II.
Figure 1. Centralized Rescue Model
In simulation environment, we try to set up as close as
possible to these above assumptions. Before starting
simulation, every agent has to be connected to the emergency
center in order to send and receive information. The types of
data exchanged between agents and emergency center are listed
as below.
Message from agent
A1: To request for connection to the emergency center
A2: To acknowledge the connection
A3: Inform the movement to another position
A4: Inform the rescue action for victim
A5: Inform the load action for victim
A6: Inform the unload action for victim
A7: Inform the inactive status
Message from emergency center
K1: To confirm the success of the connection
K2: To confirm the failure of the connection
K3: To send decisive information
Before starting simulation, every agent will send the
command A1 to request for connection to the emergency
center. The emergency center will return the response with
command K1 or K2 corresponding to the success or failure of
connection respectively. If the connection is established, the
agent will send the command A2 to acknowledge the
connection. The initial process of simulation is shown in Figure
2.
Agent Center
connection True
False
Exit
Acknowledgment
Request for connection

Figure 2. Initial Process
After the initial process, all the connected agents will
receive the decisive information such as location of agents and
health level via command K3; after that the rescue agents will
make a decision of action and submit to the center using one of
the commands from A3 to A7. At every cycle in the simulation,
each rescue agent receives a command K3 as its own decisive
information from the center, and then submits back an action
command. The status of disaster space is sent to the viewer for
visualization of simulation. The repeated steps of simulation
are shown in Figure 3.
Center
Agent
Action command and updated states of
disaster space
Viewer
Decisive information

Figure 3. Simulation Cycle
A. Disaster Area Model
The disaster area is modeled as a collection of objects of
Nodes, Buildings, Roads, and Humans. Each object has
properties such as its positions, shape and is identified by a
unique ID. From table1 to table 4 present the properties of
Nodes, Buildings, Roads and Humans object respectively.
Vol. 3, No. 7, 2012

98 | P a g e
These properties are derived from RoboCup rescue platform
with some modifications.
I. PROPERTIES OF NODE OBJECT
Property Unit Description
x,y The x-y coordinate
Edges ID The connected roads and buildings
II. PROPERTIES OF BUILDING OBJECT
Property Description
x, y The x-y coordinate of the representative point
Entrances Node connecting buildings and roads
Floors Number of floors
BuildingAreaGround The area of the ground floor
BuildingAreaTotal The total area summing up all floors
Fieryness
The state that specifies how much it is burning
0: unburned
1: Burning 0.01 0.33
2: Burning 0.33 0.67
3: Burning 0.67 1.00
BuildingCode
The code of a construction material
Code Material Fire transmission rate
0 Wooden 1.8
1 Steel frame 1.8
2 Reinforced concrete 1.0
III. PROPERTIES OF ROAD OBJECT
StartPoint and
EndPoint
ID Point to enter the road. It must be the node
or a building
Length and Width [mm] Length and width of the road
Lane [Line] Number of traffic lanes
BlockedLane [Line] Number of blocked traffic lanes
ClearCost [Cycle] The cost required for clearing the block
IV. TABLE 4. PROPERTIES OF HUMANOID OBJECT
Position ID An object that the humanoid is on.
PositionInRoad [mm]
a length from the StartPoint of road when the
humanoid is on a road, otherwise it is zero
HealthLevel
[health
point]
Health level of human.
The humanoid dies when this becomes zero
DamagePoint
[health
point]
Health level dwindles by DamagePoint in
every cycle. DamagePoint becomes zero
immediately after the humanoid arrives at a
refuge

Building
Node
Road
Edges

Road object
StartPoint EndPoint
Length
Width Node Node

Figure 4. Node object Figure 5. Road object
Building object
Entrances

StartPoint EndPoint
PositionInRoad
Position
Human

Figure 6. Building object Figure 7. Human object
The topographical relations of objects are illustrated from
Figure 4 to Figure 7. The representative point is assigned to
every object, and the distance between two objects is calculated
from their representative points.
C. Task Allocation Model
The decision making of volunteers to help disabled persons
can be treated as a task allocation problem [10-14]. The task
allocation for rescue scenario is carried out by the central
agents. The task of volunteers is to help disabled persons; this
task has to be allocated as to which volunteers should help
which disabled persons in order to maximize the number of
survivals.
We utilize the combinatorial auction mechanism to solve
this task allocation problem. At this model, the volunteers are
the bidders; the disabled persons are the items; and the
emergency center is the auctioneer. The distance and health
level of disabled person are used as the cost for the bid. When
the rescue process starts, emergency center creates a list of
victims, sets the initial distance for victims, and broadcasts the
information to all the volunteer agents. Only the volunteer
agents whose distance to victims is less than the initial distance
will help these victims. It means that each volunteer agent just
help the victims within the initial distance instead of helping all
the victims. The initial distance will help volunteers to reduce
the number of task so that the decision making will be faster.
The aim of this task allocation model is to minimize the
evacuation time. It is equivalent to minimize the total cost to
accomplish all tasks. In this case, the cost is the sum of distance
from volunteers to victims and the health level of victims. The
optimization problem is formed as follows.
Given the set of n volunteers as bidders: {
}
and set of m disabled persons considered as m tasks:
{
} . The distances from volunteers to disabled

persons; distances among dsabled persons and health level of
disabled persons are formulated as follow.

{}}

({
)
Let I is a collection of subsets of D. Let x
j
= 1 if the j
th
set in
I is a winning bid and c
j
is the cost of that bid. Also, let

if the
set in I contains iD. The problem can then be stated

as follows [15]:

Vol. 3, No. 7, 2012

99 | P a g e
The constraint will make sure that each victim is helped by
at most one volunteer.
To illustrate with an example of bid generation, lets
assume that a volunteer A has information of 5 victims (d
1
, d
2
,
d
3
, d
4
, d
5
). The initial distance is set to 200 meter. The
volunteer estimates the distance from himself to each victims
and select only victims who are not more than 200 meter from
his location. Assume that, the victim d
1
and victim d
2
are
selected to help with the cost is 180.1. The bid submitted to
center agent is Bid
A
= ({d
1
, d
2
}, 180.1).
This optimization problem can be solved by Heuristic
Search method of Branch-on-items (Sandholm, 2002). This
method is base on the question: Which volunteer should this
victim be assigned to?. The nodes of search tree are the bids.
Each path in the search tree consists of a sequence of disjoint
bids. At each node in the search tree, it expands the new node
with the smallest index among the items that are still available,
and do not include items that have already been used on the
path. The solution is a path which has minimum cost in the
search tree.
To illustrate with an example of a task allocation of
volunteers to help disabled persons, lets assume that there are
four volunteers and 3 disabled persons; The initial distance is
set to 200 meter; At the time
of simulation, distances from

volunteers to disabled persons, the distance among disabled
persons, and health level of disabled persons are assumed as
follows.
[
[

]
[

]
{ }

For example, the volunteer
can make three bids for

victim {
}, {
} and {
} based on initial distance. The

cost for {
} =

Possible bids are listed as below.
}
V. TASKS ALLOCATION AND COST
Bid Volunteer Disabled person Cost
b1 v1 {d3} 350
b2 v2 {d1} 440
b3 v2 {d3} 400
b4 v2 {d1,d3} 850
b5 v3 {d2} 300
b6 v3 {d3} 450
b7 v4 {d1} 440
b8 v4 {d2} 270
b9 v4 {d1,d2} 740

The bid b
2
and b
7
have the same task {d
1
}; b
5
and b
8
have
the same task {d
2
}; b
1
, b
3
and b
6
have the same task {d
3
}. The
more expensive bids will be removed.
Bid Volunteer Disabled person Cost
b1 v1 {d3} 350
b2 v2 {d1,d3} 850
b3 v4 {d1} 440
b4 v4 {d2} 270
b5 v4 {d1,d2} 740

Then, the search tree is formed as below.
{d
1,
d
3
}/850 {d
1
}/440 {d
1,
d
2
}/740
{d
2
}/270 {d
2
}/270
{d
3
}/350 {d
3
}/350
{d
3
}/350
b
3
b
4
b
1
b
5
b
1
b
2
b
4

Figure 8. Branch on Items based Search tree
The winner path is
, which has the most minimum

cost of 1060. The task allocation solution: volunteer v
4
will
help disabled persons d
1
and d
2
; volunteer v
1
will help disabled
persons d
3
.
IV. EXPERIMENTAL RESULTS
In this section, we present experimental studies on different
scenarios. The goal is to examine the proposed method of task
allocation model for selecting disabled people to rescue. The
evacuation time is evaluated from the time at which the first
volunteer start moving until the time at which all alive victims
arrive at the shelters. The simulation model is tested using the
RoboCup platform with Morimoto Traffic Simulator[17].
A. Experimental Settings
We consider the number of volunteers, number of disabled
persons, and traffic density as parameters to examine the
correlation between these parameters with rescue time.

Figure 9. Sample GIS Map

Vol. 3, No. 7, 2012

100 | P a g e
The sample GIS map consists of 5 layers: road, building,
volunteer, disabled person and shelter. The red points and
green points indicate the locations of disabled persons and
locations of volunteers respectively. These locations are
generated randomly along the roads. Blue buildings are
shelters. The initial health level of disabled persons are
generated randomly between 100 to 500. Every time step of
simulation, these health levels decrease by 0.5. If the health
level is equal to zero, the corresponding agent is considered as
dead. The movements of volunteer agents are controlled by
Morimoto Traffic Simulator.
B. Experimental results
With a fixed number of disabled persons and the number of
volunteers increase, the correlation between number of
volunteers and rescue time is shown as below.

Figure 10. Correlation between Number of Volunteers and Rescue Time
With a fixed number of volunteers and the number of
disabled persons increase, the correlation between number of
disabled persons and rescue time is shown as below.

Figure 11. Correlation between Number of Disabled Persons and Rescue
Time
The number of volunteers and the number of disabled
persons are fixed, whereas the number of vehicle increases. We
test with the total length of road of 500 meters. The increasing
number of vehicles will make traffic density higher. The
correlation between number of vehicle and rescue time is
shown as below.

Figure 12. Correlation between Number of Vehicles and Rescue Time
V. CONCLUSION
In this paper, the decision making of volunteers to help
persons with disability is presented as task allocation problem.
The disabled persons are considering as the tasks, and these
tasks are allocated to volunteers by utilizing combinatorial
auctions mechanism. At each time step of simulation, the task
allocation problem is solved in order to assign appropriate tasks
to volunteers. Although there are some previous works [13, 14]
on applying combinatorial auctions to task allocation, our
method has some differences in forming and solving problem;
the volunteers only bid on disabled persons located within a
certain distance and the health condition of disabled persons
and the distance from volunteers to disabled persons are used
as the cost of bids. The simple example of task allocation
problem is presented to clarify the procedures of our method.
The RoboCup rescue simulation platform is used to simulate
the rescue process. The correlations between rescue time and
other parameters such as number of volunteers, number of
disabled persons and number of vehicles are also presented.
In future work, we are thinking of comparing the multi-
criteria decision making method with task allocation method in
case of solving the decision making problem of volunteers to
help disabled persons.
REFERENCES
[1] C. Ren, C. Yang, and S. Jin, Agent-Based Modeling and Simulation on
emergency, Complex 2009, Part II, LNICST 5, 1451 1461, 2009.
[2] M. H. Zaharia , F. Leon, C. Pal, and G. Pagu, Agent-Based Simulation
of Crowd Evacuation Behavior, International Conference on Automatic
Control, Modeling and Simulation, 529-533, 2011.
[3] C. T. Quang, and A. Drogoul, Agent-based simulation: definition,
applications and perspectives, Invited Talk for the biannual Conference
of the Faculty of Computer Science, Mathematics and Mechanics, 2008.
[4] Z. Bo, and V. Satish, Agent-based modeling for household level
hurricane evacuation, Winter Simulation Conference, 2009.
[5] J. W. Cole, C. E. Sabel, E. Blumenthal,K. Finnis, A. Dantas,S. Barnard,
and D. M. Johnston, GIS-based emergency and evacuation planning for
volcanic hazards in New Zealand, Bulletin of the New Zealand society
for earthquake engineering, vol. 38, no. 3, 2005.
[6] M. Batty, Agent-Based Technologies and GIS: simulating crowding,
panic, and disaster management, Frontiers of geographic information
technology, chapter 4, 81-101, 2005

0
50
100
150
200
250
300
5 6 7 8 9 10 11 12 13 14 15
Volunteer
Disabled person: 10
Rescue time
0
50
100
150
200
250
300
350
5 6 7 8 9 10 11 12 13 14 15
Disabled Person
Volunteer: 10
Rescue time
0
200
400
600
800
1000
1200
1400
10 15 20 25 30 35 40 45 50 55 60 65 70 75 80
Vehicle
Volunteer: 10 Disabled person: 10
Rescue time
Vol. 3, No. 7, 2012

101 | P a g e
[7] T. Patrick, and A. Drogoul, From GIS Data to GIS Agents Modeling
with the GAMA simulation platform, TF SIM 2010.
[8] C. T. Quang, A. Drogoul, and A. Boucher, Interactive Learning of
Independent Experts Criteria for Rescue Simulations, Journal of
Universal Computer Science, Vol. 15, No. 13, 2701-2725, 2009.
[9] S. Silvia, C. John, and L. Beatriz, Improving Rescue Operation in
Disasters. Approaches about Task Allocation and Re-scheduling, In
Proceedings of PLANSIG 2005, London UK, 2005.
[10] R. Nair, T. Ito, M. Tambe, and S.Marsella, Task allocation in the rescue
simulation domain: A short note, Volume 2377 of Lecture Notes in
Computer Science. Springer, Berlin 751754, 2002.
[11] F. Boffo, P. R. Ferreira, and A. L. Bazzan, A comparison of algorithms
for task allocation in robocup rescue, Proceedings of the 5th European
workshop on multiagent systems, 537548, 2007.
[12] L. Hunsberger, B. Grosz, A combinatorial auction for collaborative
planning, Proceedings of the fourth international conference on multi-
agent systems, 2000.
[13] L. Beatriz, S. Silvia, and L. Josep, Allocation in rescue operations using
combinatorial auctions, Artificial Intelligence Research and
Development, Vol. 100, 233-243, 2003.
[14] C. K. Chan, and H. F. Leung, Multi-auction approach for solving task
allocation problem, Lecture Notes in Computer Science, Vol 4078,
240-254, 2005.
[15] T. Sandholm, Algorithm for optimal winner determination in
combinatorial auctions, Artificial Intelligence, Vol 135, 1-54, 2002.
[16] K. Arai & T. X. Sang, Multi Agent-based Rescue Simulation for
Disable Persons with the Help from Volunteers in Emergency
Situations, International Journal of Research and Reviews in Computer
Science (IJRRCS) Vol. 3, No. 2, April 2012.
[17] Morimoto, Traffic Simulator for RoboCupRescue Prototype Simulation
System, Available at: http://www.robocuprescue.org/docs/traffic.txt
AUTHORS PROFILE
Kohei Arai, He received his BS, MS and PhD degrees in 1972, 1974 and
1982, respectively. He was with The Institute for Industrial Science and
Technology of the University of Tokyo from April 1974 to December 1978 and
also was with National Space Development Agency of Japan from January,
1979 to March, 1990. From 1985 to 1987, he was with Canada Centre for
Remote Sensing as a Post-Doctoral Fellow of National Science and
Engineering Research Council of Canada. He moved to Saga University as a
Professor in the Department of Information Science on April 1990. He was a
counselor for the Aeronautics and Space related to the Technology Committee
of the Ministry of Science and Technology during from 1998 to 2000. He was a
councilor of Saga University for 2002 and 2003. He also was an executive
councilor for the Remote Sensing Society of Japan for 2003 to 2005. He is an
Adjunct Professor of University of Arizona, USA since 1998. He also is Vice
Chairman of the Commission A of ICSU/COSPAR since 2008. He wrote 29
books and published 262 journal papers.
Tran Xuan Sang, He received his Bachelor degree in Computer Science
from Vinh University, Vietnam, 2003 and a Master degree in Information
Technology for Natural Resources Management from Bogor Agricultural
University, Indonesia, 2006. From April, 2010 to present, he is a doctoral
student at the Department of Information Science, Faculty of Science and
Engineering, Saga University, Japan. His research interests include expert
system, intelligent computing, GIS modeling and simulation.
Nguyen Thi Uyen, She received her Bachelor degree in Computer Science
from Vinh University, Vietnam, 2009. From September, 2009 to present, she is
a lecturer at Faculty of Information and Technology, Vinh University, Vietnam.
Her research interests include expert system and intelligent computing.

Vol. 3, No. 7, 2012

102 | P a g e
Image Clustering Method Based on Density Maps
Derived from Self-Organizing Mapping: SOM
Kohei Arai
Saga University
Saga City, Japan

Abstract A new method for image clustering with density maps
derived from Self-Organizing Maps (SOM) is proposed together
with a clarification of learning processes during a construction of
clusters. It is found that the proposed SOM based image
clustering method shows much better clustered result for both
simulation and real satellite imagery data. It is also found that
the separability among clusters of the proposed method is 16%
longer than the existing k-mean clustering. It is also found that
the separability among clusters of the proposed method is 16%
longer than the existing k-mean clustering. In accordance with
the experimental results with Landsat-5 TM image, it takes more
than 20000 of iteration for convergence of the SOM learning
processes.
Keywords- Clustering; self organizing map; separability; Learning
process; Density map; Pixel labeling; Un-supervised classification.
I. INTRODUCTION
Clustering method is widely used for data analysis and
pattern recognition [1]-[4]. Meanwhile, Self-Organizing Map:
SOM proposed by T. Kohonen is a neural network with two
layers which allows use as un-supervised classification, or
learning method [5] based on a similarity between separable
data groups to be classified [6]. In other word, SOM is a
visualization tool for multi-dimensional data rearranging the
data in accordance with a similarity based on a learning
process with the statistical characteristics of the data. It is used
to be used for pattern recognition in combination with
Learning Vector Quantization (LVQ
1
). SOM is consists of m-
dimensional input layer which represent as a vector and two
dimensional output layer which is also represented as a vector
connected each other nodes between input and output layers
with weighting coefficients. In a learning process, winning
unit is chosen based on the difference between input vector
and weighting coefficients vector then the selected unit and
surrounding units get closer to the input vector.
SOM is utilized for clustering [7]. After a learning process,
a density map
2
is created in accordance with code vector

1

http://en.wikipedia.org/wiki/Learning_Vector_Quantization
2

http://books.google.co.jp/books?id=wxvQoFy1YBgC&pg=SA
1-PA210&lpg=SA1-
PA210&dq=density+map+SOM&source=bl&ots=sU95Gi28u
g&sig=uZBXSATAqYaXPJtkmrGHts7uoqU&hl=ja&sa=X&e
density. Based on the density map, a pixel labeling
3
can be
done. This is the basic idea on the proposed image clustering
method with SOM learning. Other than this, clustering
methods with learning processes, reinforcement learning is
also proposed for image retrievals [8] and rescue simulations
[9]. Also probability density model for SOM is proposed.
The image clustering method with SOM learning based on
density map is proposed in the following section followed by
experimental results with satellite remote sensing imagery data.
Then finally, conclusions and some discussions are described.
II. PROPOSED IMAGE CLUSTERING METHOD
Firstly imagery data are mapped to a feature space. In
parallel, SOM learning process creates a density map in
accordance with a similarity between the mapped data in the
feature space and density map or between input data in the
feature space and two dimensional density maps. As a result of
SOM learning process, code vector is obtained. It is easy to
recognize the density of the code vector visually. Although
code vector density map represent cluster boundaries, it is not
easy that neither to determine a boundary nor to put a label to
the pixel in concern by using the density map. The method
proposed here is to use density map for finding boundaries
among sub-clusters then some of sub-clusters which have a
high similarity are to be merged in the following procedure,
(1)Create density map based on SOM learning
(2)Binary image is generated from the density map
(3)Define sub-clusters in accordance with the separated
areas of the binary image
(4)Calculate similarities of the sub-clusters
(5)Merge the sub-clusters which show the highest
similarity
(6)Process (4) and (5) until the number of clusters reaches
the desired number of clusters

i=hijYT7L0CIibiQfn0NSTAw&ved=0CGkQ6AEwBA#v=one
page&q=density%20map%20SOM&f=false
3
http://books.google.co.jp/books?id=jJad-
0gh8YwC&pg=PA69&dq=pixel+labeling&hl=ja&sa=X&ei=
ZinYT4CpFYjUmAWW1cCfAw&ved=0CDUQ6AEwAA#v=
onepage&q=pixel%20labeling&f=false
Vol. 3, No. 7, 2012

103 | P a g e
Representing input vector, x(t) and reference (or output)
vector, m(t), neural network proposed by T. Kohonen is
expressed as follows,
m(t+1)=m(t)+h
i
(t)[x(t)-m(t)] (1)
where h(t) denotes neighboring function or weighting
function including learning coefficients.
h
i
(t)=a(t), when i N(t)
=0, when i N(t)
(2)
where N(t) denotes the number or size of neighboring units.
a(t) is called learning coefficient and ranges from 0 to 1 as is
expressed as follows,
a(t)=a
0
(1-t/T) (3)
where a
0
is an initial value and T denotes the number of
total learning number or the number of update. In the equation
(1), [x(t)-m(t)] implies cost function
4
which should be
minimized, and if
c=argmin.||x-m
i
|| (4)
i
is obtained then such m
i
unit is called winning unit. The
neighboring unit is defined around m
i
unit. The size of the
neighboring unit, N(t) is a variable which starts with a
relatively large then is getting small reaching to the wining
unit only after the SOM learning process.
N(t)=N(0)(1-t /T)
(5)
The SOM learning process is illustrated in Fig.1.

Figure 1. Illustrative view of the SOM learning process

4

http://books.google.co.jp/books?id=AuY1PwAACAAJ&dq=c
ost+function&hl=ja&sa=X&ei=6ynYT6DxA8rxmAXQlsGN
Aw&ved=0CDUQ6AEwAA
The existing clustering algorithm such as k-means
clustering algorithm
5
is similar to the SOM learning process. If
m
i
is redefined as mean vector of cluster i, then the cost
function defined in the k-means clustering is expressed as
follows,
J= ||x(t) m
i
(x(t))||
2

(6)
Therefore, the mean vector of each cluster is determined to
minimize the equation (6) of cost function. Let I(x(t)) be a
binary function and is equal to 1 if the x(t) belongs to the
cluster i and is 0 if the x(t) does not belong to the cluster i,
then the cost function can be rewritten as follows,
J= I(x(t)) ||x(t) m
i
(x(t))||
2
(7)
Meanwhile m
i
(x(t)) is updated as follows,
m
i
(x(t+1))= m
i
(x(t)) + I(x(t)) ||x(t) m
i
(x(t))|| (8)
It is because of the following equation.
J/m
i
(x(t)) = -2 I(x(t)) ||x(t) m
i
(x(t))| (9)
The k-means clustering algorithm can be rewritten as
follows,
Set initial status of mean vectors of k clusters, m
i
(x(0)),
i=1,2,.,k, then
(2)Iteration of the following two steps for t=k+1, k+2,,N,
I
i
(x(t))=1, when ||x(t) m
i
(x(t))|| ||x(t) m
j
(x(t))||j
=0, elsewhere (10)
t
m
i
(x(t+1))= m
i
(x(t))+ I(x(t)) ||x(t) m
i
(x(t))| / I(x(t)) (11)
t=1
The equation (11) is identical to the equation (8) if is
replaced to 1/I(x(t)).
The difference of input data is enhanced in the output layer
unit through SOM learning so that similar code vector of the
unit becomes formed. Meanwhile, if the similar input data are
separated in their location each other, it becomes neighboring
units in the output layer unit. Density map f(j,k) is defined as
follows,
f(j,k)= (m
j,k
- m
j-l,k-n
)
T
(m
j,k
- m
j-l,k-n
) / D (12)

(l,n)D
where D is neighboring unit, 8 neighbor unit centered the
unit in concern in this paper. This density map has the relation
among the input imagery data, feature space and SOM
learning process as is illustrated in the Fig. 2.
This is an inverse function of the similar data concentration
so that the density map obtained by a SOM learning process is
quite similar to the distribution in the feature space mapped

5

http://books.google.co.jp/books?id=WonHHAAACAAJ&dq=
k-
means+clustering&hl=ja&sa=X&ei=hirYT_DvF8PJmQWX8
KGgAw&ved=0CD4Q6AEwAQ
Vol. 3, No. 7, 2012

104 | P a g e
from the input data. An example of density map is illustrated
in the Fig.3. In the figure, dark portion means dense of code
vector meanwhile light portion is sparse of code vector and
becomes boundary between the different clusters.

Figure 2. Relations among the input imagery data, feature space and density
map generated through SOM learning.

Figure 3. Example of density map as a result of SOM learning process.

Figure 4 Example of preliminary result of density map, binarized density map
and clustering result with increasing of the iteration number (multiplied by
512).
Fig.4 shows a preliminary result of density map, binarized
density map and clustering result with increasing of the
iteration number. In this case, initial variances of the two
clusters are set at 0.03. In accordance with the number of
iteration, density map becomes clear together with binalized
density map. Furthermore, cluster result becomes ideal goal.
Fig.5 shows examples density map, estimated boundary
and clustered result for the easiest separate type of simulated
imagery data Clustering has been done in an iterative manner.
The example shows iteration number 1 to 9 as an example.
Density map and estimated boundary changes by iteration by
iteration results in refinement of the cluster results. Thus the
proposed method may reach a final cluster result.

Figure 5 Examples density map, estimated boundary and clustered result for the easiest separate type of simulated imagery data

Vol. 3, No. 7, 2012

105 | P a g e
III. EXPERIMENTS
A. Simulation Data
Experiments with simulated imagery data and real satellite
remote sensing imagery data are conducted. With a random
number generator, three types of 30 sets of simulated imagery
data consists of 32 by 32pixels are generated. The first type is
the most separable data set with a cluster to cluster distance,
between cluster variance
b
= 8 ( means a within cluster
variance) while the third one is the most difficult to separate
data set of
b
= 3 and the second one is the middle between
the easiest and difficult,
b
= 4. The number of clusters is set
to two.
Although the original simulated images are not illustrated
in the figure, it is quite obvious that the right half of image
portion is cluster #1 and the left half is cluster #2. The top
number shows the number of iteration so that SOM learning
process is started from the left hand side. As is illustrated in
the figure, the estimated boundary in the density map varies so
remarkably. In conjunction with the changes of the density
map, clustered result is varied. It is also found that the
probability of the correct clustering becomes high in
accordance with the number of iteration.
Also an example of SOM leaning process is shown in
Fig.6. It takes a long time for the SOM learning with a
relatively long between cluster distance (difficult to cluster)
while it converged at the number of iteration of around 1000
for the relatively short between cluster distance (easy to
cluster) as is shown in Fig.7. True simulated data consists two
clusters and adjacent each other cluster at the center line of
simulation data. It is shown that two clusters can be separated
into two right and left regions in accordance with the iteration
number, learning processes.
Figure 6 Example of SOM learning process for three simulation imagery data
sets

B. Landsat Thematic Mapper Data
Landsat-5 TM data of Saga, Japan acquired on 15 May
1987 which is shown in Fig. 6 is used. The meta data is as
follows, Entity ID: LT51130371987135HAJ00, Acquisition
Date: 15-MAY-87, Path: 113, Row: 37.

Figure 7 Landsat-5 TM image of northern Kyushu, Japan used.

Fig.8 shows a portion of Landsat-5 TM image for each
spectral band. Also, Fig.9 shows the clustered results for the
proposed SOM based clustering with density map, k-mean
clustering and supervised classification of Maximum
Likelihood classification: MLH as well as a portion of original
Landsat-5 TM image which is corresponding area to the area
used. For these experiments with real remote sensing satellite
imagery data, five classes or clusters, Ariake sea, Road, Paddy
field, Bare soil, Artificial construction (houses) are set. By
referring the corresponding topographic land use map of Saga,
Japan together with the original Landsat-5 TM image, it is
found that the clustered result from the proposed method is
more appropriate than that from k-mean clustering and MLH.
In particular, detailed portion of tiny road between paddy
fields are classified with the proposed method.
SOM learning process is shown in Fig.10. In accordance
with increasing of iteration number, boundaries of the density
map are getting much clear. Furthermore, the clustered results
become a true classified map with increasing of iteration
number.
Table 1 shows confusion matrix between SOM clustering
and MLH classification. Percent Correct Classification: PCC
is 88.8% so that classification results for both SOM clustering
and MLH classification are similar except soil and water body.
Spectral characteristics of these soil and water body are quite
similar. Therefore, it is understandable the poor classification
performance between soil and water body.

Vol. 3, No. 7, 2012

106 | P a g e

Band 1 Band 2 Band 3 Band 4 Band 5 Band 7
Figure 8 Landsat-5 TM imagery data of Saga, Japan (32x32) acquired on 15 May 1987 used

Original Landsat TM image
Figure 9 Comparisons of clustered results from k-mean clustering, Maximum Likelihood classification (MLH) and the proposed SOM based clustering with
referring the derived density map.

Iteration No.1 2 13 14 18 19

Figure 10 SOM learning process (Density map: top row, clustered result: bottom row, iteration number is x multiplied by 1024)

TABLE I. CONFUSION MATRIX BETWEEN SOM AND MLH

SOM
MLH

structure road paddy soil water
structure 94 % 4 % 0% 0 % 2 %
road 1 % 94 % 5 % 0 % 0 %
paddy 0 % 8 % 92 % 0 % 0 %
soil 3 % 0 % 0 % 64 % 33 %
water 0 % 0 % 0 % 0 % 100 %

In this case with the real satellite remote sensing imagery
data, it takes much long time (more than 20000 times of
iteration is needed) as is shown in Fig.10.
Also, it is found that there are a few local minima until the
SOM learning is converged.
Separability is defined as a ratio between intra cluster
variance and between cluster variance. It is also found that the
mean of separability
6
, between cluster variance of the
proposed SOM based image clustering method with density
map is around 16% better than the existing k-mean clustering
as is shown in Table 2.

6
http://arxiv.org/abs/1001.1827
Vol. 3, No. 7, 2012

107 | P a g e

Figure 10 Example of a learning process for Landsat-5 TM imagery data
clustering with the proposed SOM based image clustering method.
TABLE II. SEPARABILITY AMONG FIVE CLUSTERS FOR BOTH K-MEAN
CLUSTERING AND THE PROPOSED CLUSTERING METHOD FOR LANDSAT TM
IMAGERY DATA

IV. CONCLUSION
A new method for image clustering with density maps
derived from Self-Organizing Maps (SOM) is proposed
together with a clarification of learning processes during a
construction of clusters. It is found that the proposed SOM
based image clustering method shows much better clustered
result for both simulation and real satellite imagery data. It is
also found that the separability among clusters of the proposed
method is 16% longer than the existing k-mean clustering.
It is found that the proposed SOM based image clustering
method shows much better clustered result for both simulation
and real satellite imagery data. It is also found that the
separability among clusters of the proposed method is 16%
longer than the existing k-mean clustering. In accordance with
the experimental results with Landsat-5 TM image, it takes
more than 20000 of iteration for convergence of the SOM
learning processes. Therefore, acceleration of learning process
is a next issue for research.
ACKNOWLEDGMENT
The author would like to thank Dr. Yasunori Terayama for
his experimental effort for this research works.
REFERENCES
[1] Mikio Takagi and Haruhisa Shimoda Edt. Kohei Arai et al., Image
Analysis Handbook, The University of Tokyo Publishing Co. Ltd., 1991.,
[2] Kohei Arai, Fundamental Theory for Image Processing Algorithms,
Gakujutu-Tosho-Publishing Co. Ltd., 1999.
[3] Kohei Arai, Fundamental Theory for Pattern Recognitions, Gakujutu-
Tosho-Publishing Co. Ltd., 1999.
[4] Kohei Arai, Remote Sensing Satellite Image Processing and Analysis,
Morikita Publishing Inc., 2001.
[5] T.Kohonen, Self-Organizing Maps, Springer Series in Information
Sciences, Vol.30, 1995; Second edition, 1997; Third, extended edition,
2001.
[6] T.Kohonen, G.Barna and R.Chrisley, Statistical pattern recognition with
neural networks: benchmarking studies, Proc. ICNN Vol.I, 61-68, 1988.
[7] Kohei Arai, Learning processes of image clustering method with density
maps derived from Self-Organizing Mapping(SOM), Journal of Japan
Photogrammetry and Remote Sensing, 43, 5, 62-67, 2004.
[8] Kohei Arai, XiangQiang Bu, Pursuit Rainforcement Learning based on-
line clustering for image retrievals, Journal of Image Electronics and
Engineering Society of Japan, 39,3,301-309,2010
[9] Kohei Arai, XiangQiang Bu, Pursuit Rainforcement Learning based on-
line clustering with learning automaton for rescue simulations and its
accerelation of convergence of learning processes, Journal of Image
Electronics and Engineering Society of Japan, 40, 2, 361-168, 2011.
[10] Jouko Lampinen and Timo Kostiainen, Generative probability density
model in the Self-Organizing Map. In U. Seiffert and L. Jain, editors,
Self-organizing neural networks: Recent advances and applications.
pages 75-94. Physica Verlag, 2002..
AUTHORS PROFILE
Kohei Arai, He received BS, MS and PhD degrees in 1972, 1974 and
1982, respectively. He was with The Institute for Industrial Science, and
Technology of the University of Tokyo from 1974 to 1978 also was with
National Space Development Agency of Japan (current JAXA) from 1979 to
1990. During from 1985 to 1987, he was with Canada Centre for Remote
Sensing as a Post-Doctoral Fellow of National Science and Engineering
Research Council of Canada. He was appointed professor at Department of
Information Science, Saga University in 1990. He was appointed councilor for
the Aeronautics and Space related to the Technology Committee of the
Ministry of Science and Technology during from 1998 to 2000. He was also
appointed councilor of Saga University from 2002 and 2003 followed by an
executive councilor of the Remote Sensing Society of Japan for 2003 to 2005.
He is an adjunct professor of University of Arizona, USA since 1998. He also
was appointed vice chairman of the Commission A of ICSU/COSPAR in
2008. He wrote 30 books and published 332 journal papers.
Vol. 3, No. 7, 2012

108 | P a g e
Improving the Solution of Traveling Salesman
Problem Using Genetic, Memetic Algorithm and
Edge assembly Crossover
1
Mohd. Junedul Haque
College of Computers and Info. Tech.
Taif University
Taif, Saudi Arabia
2
Khalid. W. Magld
College of Computing and Info. Tech.
King Abdulaziz University
Taif, Saudi Arabia

Abstract The Traveling salesman problem (TSP) is to find a
tour of a given number of cities (visiting each city exactly once)
where the length of this tour is minimized. Testing every
possibility for an N city tour would be N! Math additions. Genetic
algorithms (GA) and Memetic algorithms (MA) are a relatively
new optimization technique which can be applied to various
problems, including those that are NPhard. The technique does
not ensure an optimal solution, however it usually gives good
approximations in a reasonable amount of time. They, therefore,
would be good algorithms to try on the traveling salesman
problem, one of the most famous NP-hard problems. In this
paper I have proposed a algorithm to solve TSP using Genetic
algorithms (GA) and Memetic algorithms (MA) with the
crossover operator Edge Assembly Crossover (EAX) and also
analyzed the result on different parameter like group size and
mutation percentage and compared the result with other
solutions.
Keywords- NP Hard; GA(Genetic algorithms); TSP(Traveling
salesman problem); MA(Memetic algorithms); EAX(Edge Assembly
Crossover).
I. INTRODUCTION
The traveling salesman problem (TSP) is to find a tour of a
given number of cities (visiting each city exactly once) where
the length of this tour is minimized. The TSP is defined as a
task of finding of the shortest Hamiltonian cycle or path in
complete graph of N nodes. It is a classic example of an NP-
hard problem. So, the methods of finding an optimal solution
involve searching in a solution space that grows exponentially
with number of city [1].
The traveling salesman problem (TSP) is one of the most
widely studied NP-hard combinatorial optimization problems.
Its statement is deceptively simple, and yet it remains one of
the most challenging problems in Operational Research. The
simple description of TSP is Give a shortest path that covers all
cities along. Let G =(V; E) be a graph where V is a set of
vertices and E is a set of edges. Let C = (cij) be a distance (or
cost) matrix associated with E. The TSP requires determination
of a minimum distance circuit (Hamiltonian circuit or cycle)
passing through each vertex once and only once. And
Distribution Problem, it has attracted researchers of various
domains to work for its better solutions[3].
Those traditional algorithms such as Cupidity Algorithm,
Dynamic Programming Algorithm, are all facing the same
obstacle, which is when the problem scale N reaches to a
certain degree, the so-called Combination Explosion will
occur. A lot of algorithms have been proposed to solve TSP.
Some of them (based on dynamic programming or branch and
bound methods) provide the global optimum solution. Other
algorithms are heuristic ones, which are much faster, but they
do not guarantee the optimal solutions. The TSP was also
approached by various modern heuristic methods, like
simulated annealing, evolutionary algorithms and tabu search,
even neural networks. In this paper, we proposed a new
algorithm based on Inver-over operator, for traveling salesman
problems. In the new algorithm we will use new strategies
including selection operator, replace operator and some new
control strategy, which have been proved to be very efficient to
accelerate the converge speed.[5]
II. LITERATURE SURVEY
A. The Traveling Salesman problem (TSP)
The Traveling Salesman Problem (TSP) is an NP-hard
problem in combinatorial optimization studied in operations
research and theoretical computer science. Given a list of cities
and their pair wise distances, the task is to find a shortest
possible tour that visits each city exactly once [1].
The Traveling Salesman Problem (TSP) is an NP-hard
problem in combinatorial optimization studied in operations
research and theoretical computer science. Given a list of cities
and their pair wise distances, the task is to find a shortest
possible tour that visits each city exactly once [2]. With metric
distances In the metric TSP, also known as delta-TSP, the
intercity distances satisfy the triangle inequality. This can be
understood as no shortcuts, in the sense that the direct
connection from A to B is never longer than the detour via C.
[2]
Cij<=Cik+Ckj
Exact algorithms the most direct solution would be to try all
permutations (ordered combinations) and see which one is
cheapest (using brute force search). The running time for this
approach lies within a polynomial factor of O (n!), the factorial
of the number of cities, so this solution becomes impractical
even for only 20 cities. One of the earliest applications of
Vol. 3, No. 7, 2012

109 | P a g e
dynamic programming is an algorithm that solves the problem
in time O (n
2
2
n
)[2].
The dynamic programming solution requires exponential
space. Using inclusionexclusion, the problem can be solved in
time within a polynomial factor of 2
n
and polynomial space.
Improving these time bounds seems to be difficult. For
example, it is an open problem if there exists an exact
algorithm for TSP that runs in time O (1.9999
n
) [2]
B. Genetic Algorithms
Genetic algorithms are an optimization technique based on
natural evolution. They include the survival of the fittest idea
into a search algorithm which provides a method of searching
which does not need to explore every possible solution in the
feasible region to obtain a good result. Genetic algorithms are
based on the natural process of evolution. In nature, the fittest
individuals are most likely to survive and mate; therefore the
next generation should be fitter and healthier because they were
bred from healthy parents. This same idea is applied to a
problem by first guessing solutions and then combining the
fittest solutions to create a new generation of solutions which
should be better than the previous generation. We also include
a random mutation element to account for the occasional
mishap in nature [7].
Outline of the Basic Genetic Algorithm
1) [Start] Generate random population of n chromosomes
(suitable solutions for the problem)
2) [Fitness] Evaluate the fitness f(x) of each chromosome
x in the population
3) [New population] Create a new population by
repeating following steps until the new population is complete
4) [Selection] Select two parent chromosomes from a
population according to their fitness (the better fitness, the
bigger chance to be selected)
5) [Crossover] With a crossover probability cross over the
parents to form a new offspring (children). If no crossover was
performed, offspring is an exact copy of parents.
6) [Mutation] With a mutation probability mutate new
offspring at each locus (position in chromosome).
7) [Accepting] Place new offspring in a new population.
8) [Replace] Use new generated population for a further
run of algorithm.
9) [Test] If the end condition is satisfied, stop, and return
the best solution in current population.
10) [Loop] Go to step 2.[8]
C. Memetic Algorithm
MAS are population-based heuristic search schemes similar
to genetic algorithms (GAS). GAS relies on the concept of
biological evolution, but MAS, in contrast, mimic cultural
evolution. While, in nature, genes are usually not modified
during an individuals lifetime, memes are. They can be
thought of as units of information that are replicated while
people exchange ideas. A person usually modifies a meme
before he or she passes it on to the next generation. As with
genes, memes also evolve over time; good ideas may survive,
while had ones may not [11].
The outline of the memetic algorithm : Algorithm MA:
Begin
Initialize population P;
For each individual i P do i = Local-Search (i)
Repeat
For i = I to #crossovers Do
Select two parents i
a
, i
b
P randomly;
i
c
= crossover(i
a
,i
b
) ;
i
c
= Local. Search(i
c
) ;
Add individual i
c
to P ;
End for;
For i = 1 to #mutations Do
Select an individual i P randomly;
I
m
, = Mutates (i);
I
m
= Local-Search(I
m
) ;
Add individual i, to P;
End for;
P = select (P);
If P converged then
For each individual i P \ (best) Do
i = local Search (Mutates (i));
End if;
Until terminate = true;
End;
D. Edge Assembly Crossover (EAX)
A crossover operator, called edge assembly crossover
(EAX). It was proposed by Nagata and Kobayashi in 1997. The
EAX has two important features: preserving parents' edges
using a novel approach and adding new edges by applying a
greedy method, analogous to a minimal spanning tree. In this
we select two individuals tours denoted as A and B, are
selected as parents. EAX first merges A and B into a single
graph denoted as R and then considered a powerful crossover
operator. We called this even-cycle AB-cycle. All of the edges
in this AB-cycle are then deleted from graph R. The same
procedure is repeated until all of the edges in graph R are
eliminated. Finally, in order to get a valid solution the EAX
uses a greedy method to merge these distinct sub tours
together. [11].
III. PROPOSED SOLUTION
The TSP is defined as a task of finding of the shortest
Hamiltonian cycle or path in complete graph of N nodes. The
TSP can be formally described as a graph G =(V,E), where V
=(v
1
, ..., v
n
) is the set of vertices, E = (f(v
i
, v
j
): v
i
, v
j
belongs to
V )is the set of edges.
A. Fitness function
The GAs is used for maximization problem. For the
maximization problem the fitness function is same as the
objective function. But, for minimization problem, one way of
defining a fitness function is as
F (x) =1/f(x)
Vol. 3, No. 7, 2012

110 | P a g e

Figure 1. An Example of EAX[10]
Where f(x) is the objective function. Since, TSP is a
minimization problem; we Consider this fitness function, where
f(x) calculates cost (or value) of the tour represented by a
Chromosome.
B. Algorithm
Let we have cities with their coordinates values. Now
follow the steps
Step 1: Construct a tree or minimum spanning tree from the
graph based on the group size. Root node is the starting point of
the salesman.
Step 2: Construct the tours from each leaf node and from
starting node in tree in the following way.
Step 3: Applying the GA to this tree.
I. Select any chromosome form the tree
II. Here we have two kind of crossover one is inside the
chromosome and other in between two chromosomes.
a) One point - part of the first parent is copied and the
rest is taken in the same order as in the second parent
b) Two point - two parts of the first parent are copied
and the rest between is taken in the same order as in the
second parent
c) None - no crossover, offspring is exact copy of
parents
III. Mutation can be done to the chromosome in the
following ways.
a) Normal random - a few cities are chosen and
exchanged
b) Random, only improving - a few cities are randomly
chosen and exchanged only if they improve solution (increase
fitness)
c) Systematic, only improving - cities are systematically
chosen and exchanged only if they improve solution (increase
fitness)
d) None - no mutation
Step 4: Add the entire route from the entire computed tree
(from step 3) with the route using EAX; consider two best
route from the tree as A and B. If result of EAX is better than
our A and B than take the result of EAX Otherwise take best
route from A and B.
Step 5: Compute the fitness score from the route formed
from step 5 using fitness function if new route has better fitness
score than previous then replace the route.
Step 6: Repeat step 3 to step 5 until better fitness score is
obtained or all computed tree (from Step 3) have been added to
the route.
Step 7: return the best result.
IV. SIMULATION RESULT
There are 2 parameters to control the operation of the
Genetic Algorithm:
1) Neighborhood / Group Size Each generation, this
number of tours are randomly chosen from the population. The
best 2 tours are the parents. The worst 2 tours get replaced by
the children. For group size, a high number will increase the
likelihood that the really good tours will be selected as parents,
but it will also cause many tours to never be used as parents. A
large group size will cause the algorithm to run faster, but it
might not find the best solution.
2) Mutation % - The percentage that each child after
crossover will undergo mutation .When a tour is mutated, one
of the cities is randomly moved from one point in the tour to
another.
The starting parameter values are:

In this simulation result, we have found the TSP route for
30 cities and changed the parameter like group size and
mutation percentage. After analyses we observed that, in Fig. 1
with same group size if we increasing the mutation % then it
slowly decreasing distance value of the tour and in fig. 2 with
same mutation % if we increasing the group size then first it
slowly decreasing the distance value of the tour but after a
certain value of the group size, distance value of the tour will
increase.
Fig. 2 Distance Vs Mutation Percentage

Vol. 3, No. 7, 2012

111 | P a g e

Table 1

Fig.3 Distance Vs Group Size

Table 2
V. COMPARISON
Comparison of TSPGA [11] with my proposed solution.
Cities
Location
TSPGA Result [11].
Table III [11].
My proposed solution Result.

VI. CONCLUSION
In this paper improving the solution of traveling salesman
problem algorithm based on NP-Hard problem. I have
analyzed the result on different parameter like group size and
mutation percentage. I observed that with same group size if
we increasing the mutation percentage then it slowly
decreasing distance value of the tour and with same mutation
percentage if we increasing the group size then first it slowly
decreasing the distance value of the tour but after a certain
value of the group size, distance value of the tour will increase
with the group size. I have also use EAX to improve the
solution and compared my result with the exiting solution. In
future work I am planning to reduce distance of the tour with
further improvement in the algorithm.
REFERENCES
[1] Introduction to Algorithms, 2nd Ed. pp.1027-1033,2001.
[2] C.H. Papadimitriou and K. Stieglitz. Combinatorial Optimization:
Algorithms and Complexity. Prentice Hall of India Private Limited,
India, 1997.
[3] X. P. Wang and L. M. Cao, Genetic Algorithm-Theory, Application and
Software Realization. Xi'an, Shanxi: Xi'an Jiao Tong University Press,
2002.
[4] Zhu Qiang, A New Co-evolutionary Genetic Algorithm for Traveling
Salesman Problem ,IEEE ,2008.
[5] S. Chatterjee, C. Carrera, and L. A. Lynch, "Genetic Algorithms and
Traveling Salesman Problems," European Journal of Operational
Research, vol. 93, 1996.
[6] Budinich, M. (1996), A self-organizing neural network for the traveling
salesman problem that is competitive with simulated annealing. Neural
Computation, 8: 416-424.
[7] Traveling salesman problem, Computers & Operations Research, 30(5):
773-786.
[8] D. E. Goldberg, Genetic Algorithms in Search, Optimization, and
Machine Learning. New York: Addison-Wesley Publishing Company,
Inc., 1989.
[9] Y. Nagata, S. Yobayashi, Edge assembly crossover: A highpower
genetic algorithm for the traveling salesman problem, In Proceedings of
the 7 International Conference on Genetic Algorithms (ICGA97),
pp.450457, 1997.
[10] A Novel Memetic Algorithm with Random Multi-local-search: A case
study of TSP Peng Zou', Zhi Zhou', Guoliang Chen', Xin Yao'.' 'National
High Performance Computing Center at Hefei 230027 Hefei P.R.China
'Nature inspired computation and applications laboratory 'Dept. of
Comp. Sci. and Tech., Univ. of Sci. and Tech. of China, 'School of
Comp. Sci., The Univ. of Birmingham, Edgbaston, Birmingham B15
2TT, UK.
[11] Buthainah Fahran Al-Dulaimi, and Hamza A. Ali,Enhanced Traveling
Salesman Problem Solving by Genetic Algorithm Technique (TSPGA)
World Academy of Science, Engineering and Technology ,38, 2008.
AUTHORS PROFILE
Mohammad Junedul Haque is a lecturer at the College of Computers and
Information Technology, Taif University, Saudi Arab. He holds masters
degrees in computer science. His areas of interest are Algorithms,Data Mining,
Image Processing and Networking.Mohammad Junedul Haque can be
contacted by e-mail at [email protected].
Khalid Waheeb Magld is holding the position of vice dean for graduate
studies and scientific research in Faculty of Computing and IT, King Abdulaziz
University, Jeddah Saudi Arabia. He received first degree in Computer Science
from Metropolitan State University in 1986 and Master in Computer Science in
1994 from University of Detroit, USA. He received his PhD from University of
Bradford, United Kingdom. He had also been involved in many projects related
to database design and data modeling with the IT center and other public and
private sectors. His research interests include database design and data
modeling, data mining, data retrieval, unicast and multicast in mobile adhoc
networks, neural networks analysis and applications, web-based simulation,
training and education and artificial intelligence.Khalid Waheeb Magld can be
contacted by e-mail at [email protected].
Vol. 3, No. 7, 2012

112 | P a g e
A Hybrid Technique Based on Combining Fuzzy K-
means Clustering and Region Growing for Improving
Gray Matter and White Matter Segmentation
Ashraf Afifi
Computer Engineering Dept.,
Faculty of Computers and Information Technology
Taif University
Taif, KSA

Abstract In this paper we present a hybrid approach based on
combining fuzzy k-means clustering, seed region growing, and
sensitivity and specificity algorithms to measure gray (GM) and
white matter (WM) tissue. The proposed algorithm uses intensity
and anatomic information for segmenting of MRIs into different
tissue classes, especially GM and WM. It starts by partitioning
the image into different clusters using fuzzy k-means clustering.
The centers of these clusters are the input to the region growing
(SRG) method for creating the closed regions. The outputs of
SRG technique are fed to sensitivity and specificity algorithm to
merge the similar regions in one segment. The proposed
algorithm is applied to challenging applications: gray
matter/white matter segmentation in magnetic resonance image
(MRI) datasets. The experimental results show that the proposed
technique produces accurate and stable results.
Keywords- Fuzzy clustering; seed region growing; performance
measure; MRI brain database; sensitivity and specificity.
I. INTRODUCTION
Medical imaging includes conventional projection
radiography, computed topography (CT), magnetic resonance
imaging (MRI) and ultrasound. MRI has several advantages
over other imaging techniques enabling it to provide 3D data
with high contrasts between soft tissues. However, the amount
of data is far too much for manual analysis/interpretation, and
this has been one of the biggest obstacles in the effective use of
MRI.
The segmentation of region is an important first step for
variety of image related applications and visualization tasks.
Also, segmentation of medical images is important since it
provides assistance for medical doctors to find out the diseases
inside the body without the surgery procedure, to reduce the
image reading time, to find the location of a lesion and to
determine an estimate the probability of a disease.
Segmentation of brain MRIs into different tissue classes,
especially gray matter (GM), and white matter (WM), is an
important task. Brain MRIs have low contrast between some
different tissues. The problem of MRIs is the low contrast
between tissues.
The measurement of GM of MRI has become an important
tool for determining the multiple sclerosis (MS) patient
monitoring. In the past, MS was considered primarily a white
matter (WM) disease visible by macroscopic examination of
the tissue and on MRI. Histological studies of MS brain tissue
have provided that MS lesions are also located in the gray
matter and that these GM lesions make up a substantial
proportion of overall tissue damage due to MS [1]. To measure
the changes over time in GM volumes, accurate segmentation
methods must be used. A variety of different approaches to
brain tissue segmentation has been described in the literature
[2-4]. Few algorithms rely solely on image intensity, [2]
because these approaches are overly sensitive to image artifacts
such as radio frequency inhomogeneity, and aliasing, and
cannot adequately account for overlapping intensity
distributions across structures. Therefore, to improve
segmentation accuracy, most tissue segmentation algorithms
combine intensity information with other techniques, such as
the use of a priori anatomic information [3, 4] or edge
information through deformable contours. The use of multiple
images has significant advantages over a single image because
the different contrasts can be enhanced between tissues. For
example, fluid attenuated inversion recovery (FLAIR) images
have desirable contrast between MS lesions and the normal-
appearing brain tissue and can be combined with other images
to obtain gray/white matter segmentation.
In other hand, several algorithms have been proposed such
as: fuzzy k-means [7], c-means (FCM) [5], and adaptive fuzzy
c-means combined with neutrosophic set to improve MRI
segmentation. These algorithms, such as the segmentation tool
in SPM, [6] and FAST in FSL [7], have been implemented for
general use, and therefore, are not necessarily optimized for
specific pulse sequences or for application to images from
patients with a specific disease.
These methods are also prone to classification errors due to
partial volume effects between MS lesions and normal tissue.
Furthermore, for retrospective image analysis, where image
data may not have been acquired using optimal sequences for
use with one of the widely available segmentation tools, a
customized segmentation method may be required to obtain the
most accurate results.
In this paper, we present an approach based on combining
fuzzy c-mean clustering, seed region growing, and sensitivity
and specificity algorithm to determine GM and WM tissues in
brain MRIs. This approach begins by partitioning the given
Vol. 3, No. 7, 2012

113 | P a g e
image into several regions. The seed region growing method is
applied to the image using the centers of these regions as initial
seeds (if this center is not in image, a quite neighbor point to
this center is selected as initial seed). Then the sensitivity and
specificity is used to perform a suitable merging which
produces the final segmentation. The proposed method is
evaluated and compared with the existing methods by applying
them on simulated volumetric MRI datasets.
The rest of the paper is organized as follows. The MRI
segmentation problem is discussed in section 2. The proposed
method is described in section 3. In Section 4, the experimental
results are presented. Our conclusion is presented in section5.
II. THE MRI SEGMENTATION PROBLEM
The basic idea of image segmentation can be described as
follows. Suppose that X= {x1, x2,, xn} is a given set of data
and P is a uniformity set of predicates. We aim to obtain a
partition of the data into disjoint nonempty groups X= {v1,
v2,,vk} subject to the following conditions:
X v
i
k
i
=
=1
Y

, | =
j i
v v I
ij

k i TRUE v P
i
,.., 2 , 1 , ) ( = =

j i FALSE v v P
j i
= = , ) ( Y

The first condition ensures that every data value must be
assigned to a group, while the second condition ensures that a
data value can be assigned to only one group. The third and
fourth conditions imply that every data value in one group must
satisfy the uniformity predicate while data values from two
different groups must fail the uniformity criterion.
Our study is related to 3D-model from MRI and simulated
brain database of McGill University [14]. MRI has several
advantages over other imaging techniques enabling it to
provide 3-dimensional data with high contrast between soft
tissues. However, the amount of data is far too much for
manual analysis/interpretation, and this has been one of the
biggest obstacles in the effective use of MRI. Segmentation of
MR images into different tissue classes, especially gray matter
(GM), white matter (WM) and cerebrospinal fluid (CSF), is an
important task.
III. THE PROPOSED ALGORITHM DESCRIPTION
The objective of image segmentation is to divide an image
into meaningful regions. Errors made at this stage would affect
all higher level activities. In an ideally segmented image, each
region should be homogeneous with respect to some criteria
such as gray level, color or texture, and adjacent regions should
have significantly different characteristics or feature. In MRI
segmentation, accurate segmentation of white matter (WM) and
gray matter (GM) is critically important in understanding
structural changes associated with central nervous system
diseases such as multiple sclerosis and Alzheimers disease,
and also the normal aging process [8]. Measures of change in
WM and GM volume are suggested to be important indicators
of atrophy or disease progression. In many situations, it is not
easy to determine if a voxel should belong to WM or GM. This
is because the features used to determine homogeneity may not
have sharp transitions at region boundaries. To alleviate this
situation, we propose an approach based on fuzzy set and seed
region growing concepts into the segmentation process. If the
memberships are taken into account while computing
properties of regions, we obtain more accurate estimates of
region properties. Our segmentation strategy will use the fuzzy
k-means (FKM) for finding optimum seed as a pre-
segmentation tool, seed region growing algorithm will operate
on this seed to obtain close regions, and then refine the results
using the performance measure. We use sensitivity and
specificity [9] as performance measure to compare the
performance of various outputs of the seed region growing
method. The proposed algorithm is described in Fig.(1). The
advantage of the proposed approach is that it combines the
advantages of both methods: the FKM pre-segmentation is
rough but quick, and the seed region growing needs only the
initial seed point to produce the final, fast, highly accurate and
smooth segmentation.
The proposed algorithm consists of three procedures:
- FKM algorithm for finding optimum seed;
- Seed region growing to isolate suitable regions;
- Performance measure procedure for merging
regions and extracting the final segmentation.

A. The fuzzy K-means clustering
Fuzzy K-means clustering (FKM) algorithm partitions data
points into k clusters ) ,....., 2 , 1 (
1
k I s = and clusters
1
S
are associated with representatives (cluster center)
1
C [7]. The
relationship between a data point and cluster representive is
fuzzy. That is, a membership ] 0 , 1 [ e
ij
u is used to represent the
degree of belongingness of data point
i
X and cluster center
j
C .
Apply fuzzy K-means on F matrix with initial C clusters
Isolate regions Ri using the seed region growing
algorithm
Select the initial seeds as centers of the clusters
Apply Sensitivity and specificity method for best merging
Stop
Read digital gray scale image F matrix
Start
Start
Figure1. The steps of the proposed algorithm
Vol. 3, No. 7, 2012

114 | P a g e
Denote the set of data points as
}. {
i
X S =
The FKM algorithm
is based on minimizing the following distortion:
= =
=
k
i
N
j
ij
m
ij
d u J
1 1
(1)
With respect to the cluster representatives
j
C and
memberships
,
ij
u
where N is the number of data points; m is
the fuzzifier parameter; k is the number of clusters; and
ij
d is
the squared Euclidean distance between
data points
i
X and cluster representative
j
C . It is noted
that
ij
u should satisfy the following constraint:
=
= =
k
i
ij
N j u
1
. ,......., 1 , 1
(2)
The major process of FKM is mapping a given set of
representive vectors into an improved one through partitioning
data points. It begins with a set of initial cluster centers and
repeats this mapping process until a stopping criterion is
satisfied. It is supposed that no two clusters have the same
cluster representative. In the case that two cluster centers
coincide, a cluster center should be perturbed to avoid
coincidence in the iterative process. If q <
ij
d , then 1 =
ij
u and
0 =
ij
u for j i = , where q is a very small positive number. The
fuzzy k-means clustering algorithm is now presented as
follows.
1) Input a set of initial cluster centers )} 0 ( {
j o
C SC = and
the value . c Set . 1 = P
2) Given the set of cluster centers ,
p
SC compute
ij
d
for
. 1 1 k to j and N to i = =
Update memberships
ij
u
using the following equation:

1
1
1
1
1
1
)
1
( ) (
=

|
|
.
|
\
|
=

k
i
m
ij
m
ij ij
d
d u

(3)
If
q <
ij
d
, set 1 =
ij
u , where q is a very small
positive number.
3) Compute the center for each cluster using Eq.(4) to
obtain a new set of cluster representatives ,
1 + p
SC
=
=
=
N
i
m
ij
N
i
i
m
ij
p j
u
x u
c
1
1
) (
(4)
If
, 1 ) 1 ( ) ( k to j for p C P C
j j
= < c
then stop, where
, 1 ) 1 ( ) ( k to j for p C P C
j j
= < c 0 > c
is a very small positive number.
Otherwise set
p p +1
and go to step 2.
B. Seed region growing
In this section, we select the center of the cluster or the
nearest point to this cluster as an initial seed of region growing
algorithm. The seed position (pixel_x, pixel_y) can be grown
by merging neighboring pixels whose properties are most
similar to the premerged region. The neighbors can be chosen
based on either their distance from the seed point or the
statistical properties of the neighborhood. Then each of the 4 or
8 neighbours of that pixel are visited to determine if they
belong to the same region. This growing expands further by
visiting the neighbours of each of tthese 4 or 8 neighbor pixels.
This recursive process continues until either some termination
crierion is met or all pixels in the image are examined. The
result is a set of connected pixels determined to be located
within the region of interest. The algorithm used for this task
can be stated in two steps [10]:
Step1: Gradient based homogeneity criteria
Success of region grow algorithm depends on the initial
seed selection and criteria used to terminate the recursive
region grow. Hence choosing appropriate criteria is the key in
extracting the desired regions. In general, these criteria include
region homogeneity, object contrast with respect to
background, strength of the region boundary, size, and
conformity to desired texture features like texture, shape, and
color.
We used criteria mainly based on region homogeneity and
region aggregation using intensity values and their gradient
direction and magnitude. This criterion is characterized by a
cost function which exploits certain features of images around
the seed [11]. These cost functions are verified for their match
with the specified conditions of homogeneity criteria by
comparing their values. If there is a match then pixel under
consideration is added to the growing region otherwise
excluded from consideration.
Gradient based cost functions used in our implementation
are defined below.
max
2 2
/ kG G G G
y x n
+ =
Such that 1 0 < <
n
G
Where G
x
is the horizontal gradient component, G
y
is the
vertical gradient component; k is the constant parameter which
controls the region grow, and G
max
is the largest gradient
magnitude present in the image.
min max max
/ ) , ( G G y x G G G
m
=
Such that
1 0 < <
m
G

Where G(x,y) is the gradient magnitude at pixel under
consideration and G
min
is the minimum gradient present in the
image.
Step2: Stack based seeded region growing algorithm
We have implemented the 2D seeded region grow
algorithm using stack data structure. Since, the stack is simple
Vol. 3, No. 7, 2012

115 | P a g e
to implement and efficient in the data access, we used stack to
traverse the neighborhood pixels around the seed location. In
our implementation we considered 4-neighbours while growing
the region as shown in Fig.(2). Similar pseudo code for our
implementation is as follows:
Initialize the stack:
For each seed location
Push seed location to stack
While (stack not empty)
Pop location
Mark location as region
Mark location as visited node
If homogeneity criteria matches for locations left
neighbor pixel
If left neighbor is not visited
Push left neighbor to stack
If homogeneity criteria matches for locations To p
neighbor pixel
If top neighbor is not visited
Push top neighbor to stack
If homogeneity criteria matches for locations bottom
neighbor pixel
If bottom neighbor is not visited
Push bottom neighbor to stack
End

Figure 2. Four neighbors considered for region grow
Similar concept is extended for segmentation of 3D data set
using region growing method as shown in Fig. (2). In 3D
segmentation, 6 neighbors are considered during segmentation.
Two additional pixels along z-axis from 2 adjacent slices are
considered along with 4 neighbors.

C. Performance measures
To compare the performance of various outputs of seed
region growing technique, several methods such as: Jaccard
similarity coefficient [12], Dice similarity coefficient [13],
sensitivity and specificity [10] are used. In this section, we use
sensitivity (SENS) and specificity (SPEC) method which
almost gives good stable results [9]. Below, let consider the
sensitivity and specificity measure. Sensitivity and specificity
are statistical measures of the performance of a binary
classification test, commonly used in medical studies.
Sensitivity measures the proportion of the automatically
segmented region R
1
pixels that are correctly identified as such.
Specificity measures the proportion of the correspondent region
of the manually segmented image R
2
pixels that are correctly
identified. Given the following definitions:
TP is true positive, R
1
pixels that are correctly classified as
interestR
1
.
FP is false positive, R
2
pixels that are incorrectly identified
as interest R
1
.
TN is true negative, R
2
pixels that are correctly identified as
R
2
.
FN is false negative, R
1
pixels that are incorrectly identified
as R
2
.
We compute different coefficients reflecting how well two
segmented regions match. The Sensitivity and specificity is
formulated as follows [13]:
TN TP
TP
SENS
+
=
(5)
TN FP
TN
SPEC
+
=
(6)
A SENS of 1.0 represents perfect overlap. In this case two
regions can be merged into one segment. Whereas an index of
0.0 represents no overlap.
Algorithm 3: Sensitivity and specificity measure similarity
Input
k i R
i
,....., 3 , 2 , 1 , =

For i=1 to k
For j=2 to k
Compute SENS (R
i
,R
j
)
Compute SPEC (R
i
, R
j
)
If ABS (SENS SPEC) < 0.5 then
) (
j i i
R R R =

End If
End For
End For
End;
Numerical results
For example, if one has a 7x7 discrete image F on the
square grid (see Figure 3(a)). We can apply our algorithm as
the following:
Step1:According to the fuzzy k-means clustering algorithm,
we can divide the image F into six clusters with centers
(1,32,53,29,77,2) as shown in figure (3(b,c,d,e,f,g)).

x-1,y-1 x,y-1 x+1,y-1
x-1,y x,y x+1,y
x-1,y+1 x,y+1 x+1,y+1
1 1 2 50 30 29 29
2 1 2 55 31 32 30
1 1 1 0 30 31 29
2 1 1 0 0 30 31
3 2 1 77 33 32 28
1 1 1 0 31 30 29
1 0 1 0 28 33 32
(a)
Vol. 3, No. 7, 2012

116 | P a g e

Step2:according to seed region growing algorithm2, we can
obtain six regions R
1
,R
2
, R
3
, R
4
, R
5
, and R
6
as shown in figure
(4(a,b,c,d,e,f)).

Step3: According to sensitivity and specificity measure
algorithm3, we compute SENS and SPEC measure between
regions (a-f) as shown in figure (5(a-f)).
From the previous calculation, we note that SENS and
SPEC of regions (1,6) and regions (2,4) are high. Therefore, the
regions (R
1
, R
6
) and (R
2
,R
4
) can be merged according
sensitivity and specificity measure as shown in figure
(6(a,b,c,d)).

IV. EXPERIMENTAL RESULTS
The experiments were performed with several data sets
using MATLAB. We used a high-resolution T1-weighted MR
phantom with slice thickness of 1mm, different noise and
obtained from the classical simulated brain database of McGill
University Brain Web [14] (see Fig.(7)).
The advantages of using digital phantoms rather than real
image data for validating segmentation methods include prior
knowledge of the true tissue types and control over image
parameters such as modality, slice thickness, noise, and
intensity inhomogeneities.
The quality of the segmentation algorithm is of vital
importance to the segmentation process.
The comparison score S for each algorithm can be found in
Zanaty et al.[15-17], and defined as:
ref
ref
A A
A A
S
=
(7)
Where A represents the set of pixels belonging to a class as
found by a particular method and
ref
A represents the
reference cluster pixels.

1 1 0 0 0 0 0
0 1 0 0 0 0 0
1 1 1 0 0 0 0
0 1 1 0 0 0 0
0 0 1 0 0 0 0
1 1 1 0 0 0 0
1 0 1 0 0 0 0
(b)
0 0 0 0 0 0 0
0 0 0 0 31 32 0
0 0 0 0 0 31 0
0 0 0 0 0 0 0
0 0 0 0 33 32 0
0 0 0 0 31 0 0
0 0 0 0 0 33 32
(c)
0 0 0 0 30 29 29
0 0 0 0 0 0 30
0 0 0 0 30 0 29
0 0 0 0 0 30 0
0 0 0 0 0 0 28
0 0 0 0 0 30 29
0 0 0 0 28 0 0
(e)
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 77 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
(f)
0 0 2 0 0 0 0
2 0 2 0 0 0 0
0 0 0 0 0 0 0
2 0 0 0 0 0 0
3 2 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
(g)
Figure 3.The fuzzy k-means clustering, (a) Original image, and (c-g)
clusters

1 1 2 0 0 0 0
2 1 2 0 0 0 0
1 1 1 0 0 0 0
2 1 1 0 0 0 0
3 2 1 0 0 0 0
1 1 1 0 0 0 0
1 0 1 0 0 0 0
(a)
0 0 0 0 30 29 29
0 0 0 0 31 32 30
0 0 0 0 30 31 29
0 0 0 0 0 30 31
0 0 0 0 33 32 28
0 0 0 0 31 30 29
0 0 0 0 28 33 32
(b)
0 0 0 50 0 0 0
0 0 0 55 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
(c)
0 0 0 0 30 29 29
0 0 0 0 31 32 30
0 0 0 0 30 31 29
0 0 0 0 0 30 31
0 0 0 0 33 32 28
0 0 0 0 31 30 29
0 0 0 0 28 33 32
(d)
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 77 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
(e)
1 1 2 0 0 0 0
2 1 2 0 0 0 0
1 1 1 0 0 0 0
2 1 1 0 0 0 0
3 2 1 0 0 0 0
1 1 1 0 0 0 0
1 0 1 0 0 0 0
(f)
Figure 4.The seed region growing, (a-f) regions

R SENS SPEC
1,2 0 0.5
1,3 0 0.57
1,4 0 0.31
1,5 0 0.58
1,6 1 1
(a)
R SENS SPEC
2,1 0 0.5
2,3 0 0.574
2,4 1 1
2,5 0 0.58
2,6 0 0.31
(b)
R SENS SPEC
3,1 0 0.57
3,2 0 0.574
3,4 0 0.93
3,5 0 0.92
3,6 0 0.93
(c)
R SENS SPEC
4,1 0 0.31
4,2 1 1
4,3 0 0.93
4,5 0 0.51
4,6 0 0.25
(d)
R SENS SPEC
5,1 0 0.58
5,2 0 0.58
5,3 0 0.92
5,4 0 0.51
5,6 0 0.966
(e)
R SENS SPEC
6,1 1 1
6,2 0 0.31
6,3 0 0.93
6,4 0 0.25
6,5 0 0.966
(f)
1 1 2 0 0 0 0
2 1 2 0 0 0 0
1 1 1 0 0 0 0
2 1 1 0 0 0 0
3 2 1 0 0 0 0
1 1 1 0 0 0 0
1 0 1 0 0 0 0
(a)
0 0 0 0 30 29 29
0 0 0 0 31 32 30
0 0 0 0 30 31 29
0 0 0 0 0 30 31
0 0 0 0 33 32 28
0 0 0 0 31 30 29
0 0 0 0 28 33 32
(b)
0 0 0 50 0 0 0
0 0 0 55 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
(c)
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 77 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
(d)
Figure 6. The merged regions after sensitivity and specificity measuring

0 0 0 50 0 0 0
0 0 0 55 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
(d)
Figure 5.The sensitivity and specificity measure between regions,
(a-f)

Vol. 3, No. 7, 2012

117 | P a g e

A. Experiment on MRIs
The original image size is 129129 pixels, as shown in
Fig. 7 obtained from the classical simulated brain. We apply
our technique to segment images generated at various noise
levels (0%, 1%, 3%, 5%, 7%, and 9%).We generate various
inhomogeneities and boundary weakness by controlling noise
and RF levels (0% and 20%) respectively.
Table I shows the score S of WM using our technique at
various noise and RF levels. These results show that our
algorithm is very robust to noise and intensity; homogeneities
and inhomogeneities. The best S is achieved for low noise and
low RF, for which values of S are higher than 0.97.
TABLE I THE SCORE S OF WM
Noise/RF 0 20%
0% 0.97 0.95
1% 0.96 0.94
3% 0.94 0.93
5% 0.90 0.92
7% 0.88 0.84
9% 0.85 0.80
B. Comparative results
In this section, we compare the performance of our
technique with two recent methods: Del-Fresno et al. [19] and
Yu et al. [20] techniques which gave good results in brain
segmentation. The segmentation results of these algorithms are
presented in Figs.(6a), (6b), and (6c) respectively. The
performance of each segmentation method on this dataset is
reported in Table 2.
Table II shows the scort S of WM using different
techniques for the Brain data. In this Table, we compare
between proposed method, Del-Fresno et al. [18] and Yu et al.
[19] techniques. In particular, although the segmentation
quality logically deteriorates in the presence of noise (0% and
6%) and variations in intensity, the robustness of the present
technique is highly satisfactory compared with the results of
other segmentation techniques [18,19].
TABLE II THE SCORE FOR WM USING THE BRAIN WEB [14].
Noise 3% 6%
RF 0% 20% 0% 20%
Proposed method 0.94 0.93 0.92 0.85
Del-Fresno et al.[3] 0.94 0.89 0.91 0.84
Yu et al.[6] 0.90 0.90 0.88 0.83
V. CONCLUSION
In this paper, we have presented an approach for medical
image segmentation, which integrates three existing methods:
fuzzy k-means clustering, seed region growing, and sensitivity
and specificity algorithms. The first two methods have a
common advantage: they have no constraints or hypothesis on
topology, which may change during convergence. The third
method is used to merge similar regions. An initial partitioning
of the image into primitive regions has been performed by
applying a fuzzy clustering on the image. This initial partition
is the input to a computationally efficient seed region that
produces the suitable segmentation. The sensitivity and
specificity algorithm is used to perform a suitable merging of
regions which produces the final segmentation. It is observed
that the proposed method has shown higher robustness in
discrimination of regions because of the low signal/noise ratio
characterizing most of medical images data.
By comparing the proposed methods with Del-Fresno et al.
[18] and Yu et al. [19] methods, it is clear that the proposed
algorithm can estimate the correct tissues WM and GM much
more accurately than the established algorithms. Although, the
accuracy of WM and GM clusters are varied according to noise
factor, but we have shown that the proposed method gives
better accuracy than Del-Fresno et al. [18] and Yu et al. [19]
techniques with high noise level.
Future research in MRI segmentation should strive toward
improving the computation speed of the segmentation
algorithms, while reducing the amount of manual interactions
needed. This is particularly important as MR imaging is
becoming a routine diagnostic procedure in clinical practice. It
is also important that any practical segmentation algorithm
should deal with 3D volume segmentation instead of 2D slice
by slice segmentation, since MRI data is 3D in nature.
REFERENCES
[1] Peterson, J.W., Bo, L., Mork, S., et al., " Transected neuritis, apoptotic
neurons, and Reduced inflammation in cortical multiple sclerosis lesions",
Ann. Neurol. 50, 389400, 2001.
[2] De Stefano, N., Matthews, P.M., Filippi, M., Agosta, F., De Luca, M.,
Bartolozzi, M.L., Guidi,L., Ghezzi, A., Montanari, E., Cifelli, A.,
Federico, A., Smith, S.M.," Evidence of early cortical atrophy in MS:
relevance to white matter changes and disability", Neurology 60 (7),
11571162, 2003.
[3] Schnack, H.G., Hulshoff Pol, H.E., Baar, W.F.C., Staal, W.G.,
Viergever, M.A., Kahn, R.S.," Automated separation of gray and white
matter from MR images of the human Brain", NeuroImage 13, 230237,
2001.
[4] Chalana, V., Ng, L., Rystrom, L.R., Gee, J.C., Haynor, D.R.," Validation
of brain segmentation and tissue classification algorithm for T1-weighted
MR images",. Med. Imag. 2001: Image Process. 4322, 18731882, 2001.
[5] Amato, U., Larobina, M., Antoniadis, A., Alfano, B.,"Segmentation of
magnetic Resonance brain images through discriminant analysis", J.
Neurosci. Methods 131,6574, 2003.
[6] Zhu, C., Jiang, T.,"Multicontext fuzzy clustering for separation of brain
tissues in magnetic resonance images", NeuroImage 18 (3), 685696,
2003.
[7] ChihT.,Jim Z., MU-Der J., A Fuzzy K-means clustering algorithm using
cluster center displacement, journal of information science and
engineering,vol. 27, 995-1009, 2011.
[8] Angela G. E. M. de Boer, T. T. Taskila, A. O. , Frank,Jos H. A., "
Cancer survivors and Unemployment". The Journal of the American
medical association JAMA, 30, 7,2009.

Figure. 7 Test image original slice#62.

Vol. 3, No. 7, 2012

118 | P a g e
[9] Gardner MJ. and DG. Altman, "Calculating confidence intervals for
proportions and theirdifferences". BMJ Publishing Group, 28- 33,1989.
[10] Rai,G.N.Harikrishna, Nair, T.R. Gopalakrishnan Gradient based seeded
region grow method for CT angiographic image segmentation, In
Proceedings of CoRR. 2010.
[11] Fan, J., Zeng, G., Body, M. ,Hacid, M., "Seeded region growing: an
extensive andcomparative Study".Pattern Recognition Letters.26, 1139-
1156, 2005.
[12] Jaccard P., "The distribution of the flora in the alpine zone". New
phytol.11 (2),37 50,1912.
[13] Dice L., "Measures of the amount of ecologic association between
species", Ecology,26,:297-302,1945.
[14] Brain Web, Simulated Brain Database, McConnell Brain Imaging
Centre, MontrealNeurological Institute, McGill.
[15] Zanaty E. A., S. Aljahdali Fuzzy algorithms for automatic magnetic
resonance image segmentation, International Arab Journal of
Information Technology (IAJIT), 7, 3, 271-279, 2009.
[16] Zanaty E.A., S. Aljahdali, N. Debnath, Improving fuzzy algorithms for
automatic magnetic resonance image segmentation, Proceedings of
seventeenth International Conference of Software Engineering and Data
Engineering, 60-66, Los Angeles, California, USA, June 2008.
[17] Zanaty E.A., S. Aljahdali, N. Debnath, A kernelized fuzzy c-means
algorithm for Automatic magnetic resonance image segmentation,
Journal of Computational Methods in Science and engineering (JCMSE),
123-136, 2009.
[18] Del-Fresno M., M. Vnere, and A. Clausse, A combined region growing
and deformablemodel method for extraction of closed surfaces in 3D CT
and MRI scans, ComputerizMedical Imaging and Graphics, 33, 369376,
2009
[19] Yu Z. Q., Y. Zhu, J. Yang, Y. M. Zhu, A hybrid region-boundary model
for cerebralcortical segmentation in MRI, Computerized Medical Imaging
and Graphics, 30, 197208, 2006.
AUTHORS PROFILE
Ashraf Afifi is an assistance professor, faculty of computers and
information technology, Taif University, Saudi Arabia. He received his MSC
Degree in digital communication in 1995 from zagazig University, Egypt. He
completed his Ph. D. studies in 2002 from zagzig University,Egypt. His
research interests are digital communication, and image segmentation. In these
areas he has published several technical papers in refereed international
journals or Conference proceedings.
Vol. 3, No. 7, 2012

119 | P a g e
GUI Database for the Equipment Store of the
Department of Geomatic Engineering, KNUST
J. A. Quaye-Ballard
1
, R. An
2
, A. B. Agyemang
3
, N. Y. Oppong-Quayson
4
and J. E. N. Ablade
5
1
PhD Candidate, College of Earth Science and Engineering, Hohai University, Nanjing, China.
2
Professor, Dean, College of Earth Science and Engineering, Hohai University, Nanjing, China.
1,3
Lecturer, Department of Geomatic Engineering, College of Engineering, Kwame Nkrumah University of Science and
Technology (KNUST), Kumasi, Ghana
4,5
Research Assistant, Department of Geomatic Engineering, College of Engineering, KNUST, Kumasi, Ghana

Abstract The geospatial analyst is required to apply art,
science, and technology to measure relative positions of natural
and man-made features above or beneath the earths surface, and
to present this information either graphically or numerically. The
reference positions for these measurements need to be well
archived and managed to effectively sustain the activities in the
spatial industry. The research herein described highlights the
need for an information system for the Land Surveyors
Equipment Store. Such a system is a database management
system with a user-friendly graphical interface. This paper
describes one such system that has been developed for the
Equipment Store of the Department of Geomatic Engineering,
Kwame Nkrumah University of Science and Technology
(KNUST), Ghana. The system facilitates efficient management
and location of instruments, as well as easy location of beacons
together with their attribute information; it provides multimedia
information about instruments in an Equipment Store. Digital
camera was used capture the pictorial descriptions of the
beacons. A Geographic Information System (GIS) software was
employed to visualize the spatial location of beacons and to
publish the various layers for the Graphical User Interface
(GUI). The aesthetics of the interface was developed with user
interface design tools and coded by programming. The developed
Suite, powered by a reliable and fully scalable database, provides
an efficient way of booking and analyzing transactions in an
Equipment Store.
Keywords- Survey Beacons; Survey Instrument; Survey
Computations; GIS; DBMS.
I. INTRODUCTION
Scientists around the world are addressing the need to
increase access to research data [1]. Software Applications
such as the Dataverse Networ, which is an open-source
Application for publishing, referencing, extracting and
analyzing research data, have been developed to solve the
problems of data sharing by building technologies that enable
institutions to reduce the burden for researchers and data
publishers, and incentivize them to share their data [2]. One
of the technologies that most people have become accustomed
to hearing about, at work or on the internet is Database. A
database is a structured collection of records or data that is
stored in a computer system [3]. Accessibility and storage of
large amount of data is important for a designing a database
system [4-7]. [4] highlights on the requirements for efficient
Database Management System (DBMS). In recent years, the
object-oriented paradigm has been applied in areas such as
engineering, spatial data management, telecommunications,
and various scientific domains. This is a programming
paradigm using object-oriented data structures consisting of
data fields and their links with various methods to design
applications and computer programs [7, 8]. Programming
techniques may include features such as data abstraction,
encapsulation, measuring, modularity, 3D visualization,
polymorphism and inheritance [3]. The conglomeration of
Object-Oriented Programming (OOP) and database
technology has led to this new kind of database. These
databases bring the database world and the application-
programming world closer together, in particular by ensuring
that the database uses the same type of system as the
application program. A database system needs be managed
with a Graphical User Interface (GUI), which is a Human-
Computer Interface (HCI), that is, a way for humans to
interact with computers. The GUI employs windows, icons
and menus which can be manipulated by use of a mouse and
often to a limited extent by use of a keyboard as well. Thus, a
GUI uses a combination of technologies and devices to
provide a platform for the tasks of collection and producing
information [9]. To support the idea of this research, many
interesting databases associated with GIS have been developed
in many parts of the world to support decision makers in their
investigations [e.g. 10-13].
A series of elements conforming to a visual language have
evolved to represent information stored in computers. This
makes it easier for people with few computer skills to work
with and use computer software. The most common
combination of such elements in GUIs is the Window Icon
Menu Pointing (WIMP) device paradigm, especially in
personal computers [9]. A major advantage of GUIs lies in
their capacity to render computer operation more intuitive and,
thereby, easier to learn and use. For example, it is much easier
for a new user to move a file from one directory to another by
dragging its icon with the mouse than by having to remember
and type practically mysterious commands to accomplish the
same task. The user should feel in control of the computer, and
not the other way around. This is normally achieved in
software applications that embody the qualities of
responsiveness, permissiveness and consistency.
Sustaining the activities of the Geoinformation industry,
for example, calls for proper management of coordinates to
Vol. 3, No. 7, 2012

120 | P a g e
which survey measurements are referred as well as effectual
keeping of records. KNUST is a renowned university with
adequate resources for running a DBMS at various
departmental levels and, in recognition of the urgent need for
crossing over from analogue to relevant digital methodologies
this work was undertaken to produce a prototypical DBMS for
the Department of Geomatic Engineering equipment store. In
the latter part of the 1980s significant losses were incurred
during a fire outbreak in the block that used to house the
departments equipment. The effects of those losses are still felt
and to support this need for a computerized record-keeping
routine with regular back-ups. Furthermore the extremely
large volume of equipment and other stuff makes data retrieval
and Instrument location quite difficult in the store. In addition,
locating beacons on campus is also a major problem. This
calls for development of a database (of all beacons on KNUST
campus), and to have such database integrated in a Geographic
Information System (GIS) to facilitate identification and
locating of beacons. The GIS would integrate modules for
capturing, storing, analyzing, managing, and presenting data
linked to location; a merging of cartography, statistical
analysis, and database technology [14]. With regards to
meeting the above-mentioned objectives, a database of all
equipment in the departments Equipment Store has been
developed with a graphical user interface of all equipment
shelf locations. It is a fully scalable GUI database of beacons
on KNUST campus. Software has also been developed for
record-keeping (of the use of all equipment in the Store) in a
digital environment; certain survey computational procedures
have also been developed in the application to use beacon
coordinate information seamlessly.
II. MATERIALS AND METHODS
A questionnaire was issued to help study how records are
kept at the Equipment Store; how beacons can be easily
identified in the field; how surveyors obtain beacon
information for their jobs; and how GIS could be used to
improve procedures for lending of equipment and to better
manage information on beacons. The data used for the
research included the beacons Coordinate data, which include
their Descriptions, Northings, Eastings, Elevations and
Revision Dates; a video coverage of the Geomatic Equipment
Store; pictures and videos of various sections of the Store. The
programming languages and Application software employed
were Visual Basic (vb.net) for the development of graphical
user interface Applications; C-Sharp (C#) for encompassing
imperative, declarative, functional, generic, object-oriented
and component-oriented programming disciplines. Extensible
Application MarkUp Language (XAML) was used for creating
user interfaces in WPF and for serializing .NET object
instances into a human readable format. Adobe Photoshop
was also used for its extensive amount of graphics editing
features. Other Applications employed include the .NET
framework for building the software application; Windows
Presentation Foundation (WPF), a graphical display system
designed for the .NET, framework; the Entity Framework for
accessing data and working with the results in addition to
DataReaders and DataSets; and Environmental Systems
Research Institute (ESRI) ArcObjects Library to support the
GIS application on the desktop. The logical structure of the
methods used for the design of the application is depicted in
Fig. 1.
Fig. 1: Logical structure of methods
When an application starts, the Component Object Module
(COM) will be initialized. An entity data context is created for
the database and its objects are loaded onto the GUI by means
of data binding. The GUI running on .NET Framework
connects to a GIS dataset via COM Interop. Whenever a
record needs to be updated, it undergoes validation. Backing-
up the database can be done onto a web server or an external
device.
A. Procedure
The whole suite comprises four applications, namely
BeaconBase, LendingBase, InstruVisio and Survcom. The
BeaconBase application has a database of all the survey
beacons on campus with their coordinates in the Ghana
National Grid System. The LendingBase application is used
for recording lending transactions that occur in the Equipment
Store, with details of the borrower and all instruments
borrowed. The layout controls used for both BeaconBase and
LendingBase applications include the DockPanel, Grid,
StackPanel and the Tab Control. The root layout element used
is the Grid. Two DockPanels are docked at the Top of the
Window and Stretched Horizontally across the screen. The
InstruVisio is a multimedia and interactive interface for the
instruments stored at the Equipment Store. Multimedia files
were encoded in Expression Encoder software for InstruVisio
interface. Other controls used were defined in XAML code for
InstruVisio interface.
The SurvCom is an application created on the Multiple
Document Interface (MDI) architecture of the Windows Form
platform. Most of the procedures for developing the
application were as in the procedures for the other applications
mentioned earlier. MDI applications facilitate the display of
multiple documents or forms at the same time, with each
document displayed in its own window. MDI applications
often have a Window menu item with submenus for switching
between windows or documents.
Vol. 3, No. 7, 2012

121 | P a g e
The SQL server compact 3.5 database was employed in
designing the LendStore Database (a database for lending),
which has ten fields - PersonID which is the primary key field
(type integer), Date (type datetime), PersonName (type
nvarchar), PersonDepartment (type nvarchar), PersonPhone
(type nvarchar), IsReturned (type Boolean), ReturnDate (type
datetime), Remarks (type nvarchar), and TotalInstru (type
integer); LendingDetails Table, which has five fields: Id which
is the primary key field (type integer), InstrumentName (type
nvarchar), Quantity (type integer), PersonID which is a foreign
key field (type integer) and Serial (type nvarchar); and the
Instrument Table, which has six fields: ID which is the
primary key field (type integer), InstrumentName (type
nvarchar), Unused (type nvarchar), Used (type nvarchar),
Remaining (type nvarchar) and Description (type nvarchar);
and the BeaconDB Database (a database for the beacons),
which has one table with nine fields: BeaconID (type integer),
BeaconName (type nvarchar), Northing (type nvarchar),
Easting (type nvarchar), Elevation (type nvarchar),
Description (type nvarchar), Photo (type nvarchar),
RevisionSurveyor (type nvarchar) and Date (type datetime).
The Microsoft Expression Blend and the Visual Studio
Integrated Development Environment (IDE) were the GUI
design and coding software used respectively. A new project
was created for each of the applications under the suite.
Expression Blend was employed in the design of the WPF
windows. The query language used is the Language-Integrated
Query (LINQ). LINQ-To-Entities and Entity Structure Query
Language (SQL) are a groundbreaking innovation in Visual
Studio and the .NET Framework version 3.5 that bridges the
gap between the world of objects and the world of data. In a
LINQ query, one always works with objects, and that suited
the object-oriented approach used in this work.
B. Ethical Considerations
Security issues were considered when developing the
software suite. The data in the database was protected against
unauthorized disclosure, alteration and destruction. The user is
allowed to do only what is expected to be done to avoid
inconsistencies in the database which could produce errors in
the software. Data integrity or accuracy/correctness of data in
the software was also considered. Though this attribute cannot
be absolutely met, validation principles were inculcated into
the software to ascertain as much as possible that data entering
or leaving the software database was reliable. Data loss, which
is the unforeseen loss of data or information in a computer
system, can be very disastrous; this was also very much taken
into account. Studies have consistently shown hardware failure
and human error to be the most common causes of data loss.
To avoid such disasters, a BackUp module was integrated. The
database can be uploaded through a network stream onto an
internet server for backup. Where there is no network
available, an external device could be used for backing up data
in a very fast and simple way.
C. Integrating GIS
The BeaconBase application connects to a GIS dataset to
query geographic data. The algorithm used to match the
Beacon feature in the GeoDatabase is depicted graphically in
Fig. 2.

Fig. 2: Conceptual algorithm for connecting to GIS
To bind DataControl to data the Expression Blend software
was used to automatically generate XAML code for that
purpose. Classes describe the structure of objects and are a
fundamental building block of object-oriented programming.
Four object classes were created: winBackUp, winAbout,
winBeaconBase and winSettings, all of which inherits from
the base class System.Windows.Window.
III. RESULTS AND ANALYSIS
A. The InstruVisio Application
The developed InstruVisio application has the following
features: database of all instruments in the instrument store; a
vivid description of all instruments and their use; visualization
of all rooms through a video display; the ability to show the
shelf location of each instrument in the store; a display of job
types and instruments needed for each job; a built-in search
engine to help determine with ease the location of instruments.
An audio module has been incorporated using Microsoft
speech synthesizer. It reads out, on request, the description of
each instrument in the store. The application runs with a
graphical user interface that is user-friendly and it afford the
user total control. It displays an aesthetically pleasing view of
buttons and an animated pop-up of windows to show the
rooms in the Store, as depicted in Fig. 3.

Fig. 3: The InstruVisio Application interface
The program starts with a video display to show the
current state of the Equipment Store. Individual videos of each
Vol. 3, No. 7, 2012

122 | P a g e
room can also be played on request by the user. The program
is such that any new user can visualize the instrument store
without necessarily going there. A Carousel displays in a
window all instruments in the store and their description.
B. The BeaconBase Application
The BeaconBase application has the following features:
database of all the survey Beacons in the study area;
management features, such as adding a new beacon, editing a
Beacons information, Updating Beacon area photo; exporting
beacon data locally onto a disk drive and printing in hardcopy;
performing bearing and distance computation between any
two beacons; visualizing beacon in a GIS window; viewing
Beacon coordinates in various units; and a fast search engine
algorithm. The application launches with a user-friendly
interface. All the beacons in the study area are loaded from the
database into memory. The customized data template of a
listbox in the left panel is populated with a collection object.
The user can select a Beacon from the list and instantly get
information about the beacons - Coordinates, Picture and
Revision Date. These data can be downloaded onto a local
drive or printed out. The selected beacon can be visualized
graphically in a GIS map. The user can add new beacons to
the collection or edit data on the already existing beacons from
the user-friendly interface. The spatial locations of the beacons
can also be integrated with a GIS. Fig. 4 shows the
BeaconBase Application interface.

Fig. 4: The BeaconBase Application interface
C. The LendingBase Application
The LendingBase application has the following features: a
form for recording new lending with details of borrower and
all instruments borrowed; issuing a return note on a lending
after the borrower has returned all instruments; viewing the
lending history in a flexible and intuitive chart panel; editing
instrument information in database; graphically generating the
recent lending statistics - this statistics is very important
because it highlights the rate of collection of each instrument
in the Store; safe deletion of unwanted records - safe deletion
here refers to the records not being totally cleared but
transferred into a recycle bin so that it can be reviewed at a
later time in case there is a problem. It is important to make it
possible for deleted data to be recalled/restored to back-up
database onto the web server or some other external storage
device. This facilitates retrieval of data in case of damage to
computer hardware. It is especially important because it helps
prevent (or provide the necessary remedy for) program failure
when it occurs due to data loss. An audio module has been
incorporated to provide an audio rendition of the status of
transactions. A built-in Search engine makes it possible to
search for transactions that were executed in the past, and so
help to study the trend of use and, therefore, the possible
causes of what faults might result with any particular
instrument.
After the application launches, all the lending records in
the database are loaded into memory and populated into the
DataGrid. The user can select a record from the DataGrid,
which contains the name of the borrower, department, date of
borrowing, phone number, return status and date of return.
Details about the lending such as all instruments lent, quantity
and corresponding serial numbers are displayed concurrently
in the bottom panel. The application was designed with the
aim of placing the user in absolute control. The user can add
new records, with all the details pertaining to the record, in the
AddNew Form. Thus, the user can see at a glance a pie chart
showing in relative terms the quantity of instruments currently
available in the Equipment Store, instruments lent out, and
faulty instruments. The user can safely delete any record. Such
deleted record will not be completely deleted from the
database, but will be moved to a temporary space from where
it may be restored whenever so desired. The borrower can be
issued a return note on the lending transaction after he/she has
returned all instruments. The quantity of available instruments
in the database is automatically updated. The trend in recent
lending activity in the Equipment Stores can be depicted
graphically in a chart - a line chart, bar chart or pie chart. The
data points of Fig. 5 shows a line chart of the total numbers of
instruments lent out from the office in a day.

Fig. 5: The Lending Application interface
D. The SurvCom Application
The SurvCom application consists of modules for the
following computations: area, detailing, horizontal curve
setting out parameters derivations, and leveling - for both rise
& fall method and height of collimation method. These
computational types have been included because they are the
most common day-to-day computational routines of the land
surveyor. However, the system has the capability of adding
other computational procedures. The Area Computation
application calculates area by coordinates and, also, facilitates
conversion from one unit of area to another. For example one
can convert from square metres to hectares. Fig. 6 shows the
Area Computation Application interface.
Vol. 3, No. 7, 2012

123 | P a g e

Fig. 6: The Area Computation Application interface
The Detailing Application uses the method of detailing by
rays and can compute coordinates of practically infinite
number of points and from an infinite number of instrument
stations (Fig. 7). Use is made of observational values taken
from field with a Total Station, which include the horizontal
and vertical circle readings, and the slope distances. These are
used to compute the coordinates of the points in question.

Fig. 7: The Detailing Application interface
The Horizontal Circular Curve Design Application
computes the setting-out parameters of a horizontal circular
curve. The required input values include deflection angle for
the curve, curve length or radius of curvature, chainage of the
intersection point for the forward and back tangents, as well as
multiples of chord lengths for setting-out the curve (Fig. 8).
The Leveling Application computes the reduced level from
back-sight, foresight and inter-sight values. It computes the
reduced level using the rise & fall method or height of
collimation method depending on the users preference (Fig.
9). The developed Suite was subjected to considerable testing.
The test participants included lecturers and students from the
Department of Geomatic Engineering. The final design was
based on feedback from test participants. The design
considerations included a requirement for the suite to run
optimally under Microsoft Windows Vista or higher Operating
System; Graphics Application Programming Interface (API):
DirectX; Random Access Memory (RAM) of at least 256MB;
audio output devices like speakers. The compiled suite was
built with MS Visual Studio 2008 Service Pack 1; Telerik Rad
Controls for WPF version Q2 2009; ESRIs ArcObjects
runtime license; and Developer Express v2009 vol1 controls.
Running LendingBase initiates a window that pops up to
display a Table with a design Form of the record keeping
requirements of the Equipment Store.

Fig. 8: The Horizontal Circular Curve Design Application interface

Fig. 9: The leveling Application interface
The menu bar has options for new entry and return of
transaction, backing up data on a web server or on an external
storage device, quantity of available instruments for lending,
statistical data on recent issuing out of instruments and
deletion of data in case of error in the entry. Summary of all
details of each transaction is displayed on a record-details
panel found at the bottom. These options allow the user of the
program to be in control of every facet of instrument lending
in the Equipment Store. Other Information about the program
makeup can be found in the About section of the menu bar.
The BeaconBase program displays an artistic window with
a well-organized menu bar. The menu bar buttons include
Add, Delete, Save, Print, Edit, Visualize, Mark, Help, Back up
and Export Data. It also has a drop-down button that displays
units of beacons with which the user may want to work in. The
Visualize button shows, in a GIS environment, the spatial
location of the beacon. There is also a section with the list of
beacons on KNUST campus; a single click on any beacon in
the list displays the beacons coordinates as well as a
photograph of it to depict its current state. Any other
information about the program is found in the About section of
the menu bar. The InstruVisio Application launches with a
window which plays a video coverage of the Equipment Store.
There are buttons on the menu bar for the Carousel,
EnterOffice, Help and About. The Carousel gives information
and location of all instruments in the office. The EnterOffice is
for visualizing any room in the office and shows the shelf
locations of instruments. There is also a drop-down button
which displays the various survey job types and the
instruments required for such jobs. This is meant to assist the
user in his/her choice of instruments for the survey work on
hand.
This SurvCom program consists of a set of applications
that automate certain survey computations. The Area
Vol. 3, No. 7, 2012

124 | P a g e
Computation is an application which with a click of a button,
computes the area of a parcel from given coordinates. Its menu
bar has Open Document, Add and Remove, Save, Cut, Paste
and Compute buttons. The Detailing Application has a
window with textboxes in which the initial Station name and
coordinates with horizontal circle readings are input. In the
Table, subsequent horizontal circle readings and vertical circle
readings of targets will be entered as well as their slope
distance. The Compute button in the program will then initiate
the computations of all coordinates of points which can be
plotted for detailing. The horizontal circular curve design
application has a window with a section where the user may
enter the deflection angle for a curve, curve length or radius of
curvature, chainage of the intersection point for the forward
and back tangents. The application then generates chord
lengths for setting out the curve. The user may then enter the
point names, their chord lengths, chainages and their
individual tangential angles. Upon clicking the Calculate
button, the application will automatically generate the
cumulative tangential angles, give a sketch of the look of the
curve and provide a general report on the work. The general
report includes the through chainage of the start and end
points, initial and final sub-chord lengths, the external
distance, tangent length, and length of long chord. The
Leveling Application displays a window with a menu bar. The
program buttons are for Add, Remove, Open document, Cut,
Paste, Help and Run. When File is clicked, options on method
of computation pop up. These methods are the Rise & Fall
method and the Height of collimation method. Depending on
the users discretion any method can be selected for
computing the reduced levels.
IV. CHALLENGES AND DISCUSSIONS
During the collection of data from the Equipment Office,
considerable problems were encountered due to some of the
analogue procedures employed at the time. For example, as
regards the information on the beacons, there were problems
with some of the coordinates since some characters were not
legible on the printouts due to the age paper medium of
storage. They were old, tearing and deteriorating. These
realizations helped to appreciate the relevance of this work the
more.
V. CONCLUSION
A digital information and management system for the
Equipment Store of the KNUST Department of Geomatic
Engineering has been produced. That is, the objective of the
research of creating a database of instruments in the
Equipment Store with a GUI was achieved. This incorporated
a fully scalable GUI database of beacons on KNUST campus
integrated with its spatial location; a software to manage and
archive the record keeping of instruments in a digital
environment; and lastly automation few survey computations.
The database for all the instruments in the office gives relevant
descriptive information of each instrument in a very
comprehensive and user-friendly GUI. A fully scalable GUI
database of beacons on KNUST campus has also been
developed and successfully integrated in a GIS geodatabase to
provide spatial and graphical data for the beacons. The
LendingBase Application manages, archives and records
instruments and their daily lending transactions in a digital
environment. The SurvCom provides a predefined automation
of some survey computations for the user. The modules of the
applications are user-friendly and provide an efficient and
effective way of managing survey instruments and records to
help sustain the activities and contributions of the Land
Surveyor. Although the system in its current form can only be
said to be prototypical, it can be modified or extended to cover
any Equipment Office practice or extent of land coverage.
Any Survey Firm, private or public, where records need to be
kept on survey beacons and on the daily use of survey
equipment can use this Suite to manage the relevant concerns.
The importance of good record-keeping on surveyors beacons
and equipment cannot be over-emphasized as our dear nation
enters the efficiency-demanding realm of oil and gas
exploration. The BeaconBase Application can be employed by
small or large organizations where coordinates data are
collected, archived and used; such as with the use of the
Global Positioning System (GPS). This can support decision
makers in area where monitoring, such as grassland, wetlands,
desertification, drought, etc. where beacon database are
needed for decision makers in their investigations.
REFERENCES
[1] Brase, J. and Farquhar, A., Access to Research Data, D-Lib Magazine,
The Magazine of Digital Library Research, 17(1/2), 2011.
[2] Crosas, M., The Dataverse Network: An Open-Source Application for
Sharing, Discovering and Preserving Data, D-Lib Magazine, The
Magazine of Digital Library Research, 17(1/2), 2011,
[3] Galindo, J., (Ed.), Handbook on Fuzzy Information Processing in
Databases, Hershey, PA: Information Science Reference, (an imprint of
Idea Group Inc.), 2008.
[4] The Tech-FAQ , What is a Database? Electronic Archive:
http://www.topbits.com/what-is-a-database.html, Accessed 24/07/2012,
2012
[5] Connolly, T. M. and Begg, C. E., Database Systems: A Practical
Approach to Design, Implementation and Management, Addison Wesley
Publishing Company, 2004
[6] Date, C. J., An Introduction to Database Systems, Addison Wesley
Longman, 2003.
[7] Teorey, T., Lightstone, S. and Nadeau, T., Physical Database Design: the
database professional's guide to exploiting indexes, views, storage, and
more. Morgan Kaufmann Press, 2007
[8] Shih, J. Y., Why Synchronous Parallel Transaction Replication is Hard,
But Inevitable? A White Paper, Parallel Computers Technology Inc.
(PCTI), U.S.A. Electronic Archive:
http://www.pcticorp.com/assets/docs/PQL2b.pdf. Accessed: 23/07/12,
2007
[9] Thomas, D., Hunt, A., and Thomas, D., Programming Ruby: A
Pragmatic Programmer's Guide, Addison-Wesley Professional; 1st
edition, 2000
[10] Foster, C., Pennington, C. V. L., Culshaw, M. G., and Lawrie, K., The
national landslide database of Great Britain: development, evolution and
applications, Environmental Earth Sciences, 66(3), pp. 941-953
[11] Dahl, R., Wolden, K., Erichsen, E., Ulvik, A., Neeb, P. R. and Riiber,
K., Sustainable management of aggregate resources in Norway, Bulletin
of Engineering Geology and the Environment, 71(2), pp. 251-255
[12] Steele, C. M., Bestelmeyer, B. T., Burkett, L. M., Smith, P. L., and
Yanoff, S., Spatially Explicit Representation of State-and-Transition
Models, Rangeland Ecology & Management, 65(3), pp. 213-222
[13] Panigrahy, S., Ray, S. S., Manjunath, K. R., Pandey, P. S., Sharma, S.
K., Sood, A., Yadav, M., Gupta, P. C., Kundu, N., and Parihar, J. S., A
Spatial Database of Cropping System and its Characteristics to Aid
Climate Change Impact Assessment Studies, Journal of the Indian
Society of Remote Sensing, 39(3), pp. 355-364
[14] Fu, P. and Sun J. (2010). Web GIS: Principles and Applications, ESRI
Press, Redlands, CA

Vol. 3, No. 7, 2012

125 | P a g e
Comparative Study between the Proposed GA Based
ISODAT Clustering and the Conventional Clustering
Methods
Kohei Arai
Saga University
Saga City, Japan

Abstract A method of GA: Genetic Algorithm based ISODATA
clustering is proposed.GA clustering is now widely available. One
of the problems for GA clustering is a poor clustering
performance due to the assumption that clusters are represented
as convex functions. Well known ISODATA clustering has
parameters of threshold for merge and split. The parameters
have to be determined without any assumption (convex functions).
In order to determine the parameters, GA is utilized. Through
comparatives studies between with and without parameter
estimation with GA utilizing well known UCI Repository data
clustering performance evaluation, it is found that the proposed
method is superior to the original ISODATA and also the other
conventional clustering methods.
Keywords- GA; ISODATA; Optimization; Clustering.
I. INTRODUCTION
Clustering is the method of collecting the comrades of
each-other likeness, making a group based on the similarity
and dissimilarity nature between object individuals, and
classifying an object in the heterogeneous object of a thing [1].
The classified group calls it a cluster. The criteria which
measure how many objects are alike have the degree
(similarity) of similar, and the degree (dissimilarity) of
dissimilarity [2]. The object with high similarity is one where
a value is larger more alike like a correlation coefficient in the
degree of similar, and the object with low similarity is not one
where the value of the degree of dissimilarity is conversely
larger ] alike. The degree of dissimilarity is well used in these
both. The degree of dissimilarity is also called distance
(distance). There is a definition of the distance currently used
by clustering how many. The clustering method can be divided
into the hierarchical clustering method and the un-hierarchical
clustering method [3].
Hierarchical clustering [4] (hierarchical clustering method)
is the clustering method for searching for the configurationally
structure which can be expressed with a tree diagram or a
dendrogram [5], and is method into which it has developed
from the taxonomy in biology. A hierarchy method has a
shortest distance method, the longest distance method, the
median method, a center-of gravity method, a group means
method, the Ward method, etc [6]. By a hierarchy method,
there are faults, such as the chain effect that computational
complexity is large.
A non-hierarchy method is the method of rearranging the
member of a cluster little by little and asking for the better
cluster from the initial state [7],[8],[9]. It is more uniform than
this as much as possible within a cluster, and it is a target to
make it a classification which differs as much as possible
between clusters. The typical method of a non-hierarchy
method has the K-means method and the ISODATA method
[10].
A method of GA: Genetic Algorithm [11] based
ISODATA clustering is proposed. GA clustering is now
widely available. One of the problems for GA clustering is a
poor clustering performance due to the assumption that
clusters are represented as convex functions. Well known
ISODATA clustering has parameters of threshold for merge
and split [12],[13]. The parameters have to be determined
without any assumption (convex functions). In order to
determine the parameters, GA is utilized. Through
comparatives studies between with and without parameter
estimation with GA utilizing well known UCI Repository data
clustering performance evaluation, it is found that the
proposed method is superior to the original ISODATA.
ISODATA based clustering with GA is proposed in the
previous paper [14]. In this paper, comparative study of the
proposed ISODATA GA clustering method with the
conventional clustering methods is described.
In the next section, theoretical backgrounds on the widely
used conventional clustering methods and Genetic Algorithm:
GA
1
is reviewed followed by the proposed clustering method
based on ISODAT with GA. Then experimental result with
simulation data of concave shaped distribution of data is
shown for demonstration of effectiveness of the proposed
method followed by experimental results with UCI repository
2

of standard datasets for machine learning. In particular,
clustering performance of the proposed GA based ISODATA
clustering method is compared to those of the other
conventional clustering methods. Finally, conclusion and some
discussions are described.
Theoretical Background

1
http://www2.tku.edu.tw/~tkjse/8-2/8-2-4.pdf
2
http://archive.ics.uci.edu/ml/support/Iris
Vol. 3, No. 7, 2012

126 | P a g e
A. K-Means Clustering
The k-mean method is that of the non-hierarchical type
clustering method proposed by MacQueen, Anderberg [15],
Forgy and others [16] in the 1960s. Based on the given initial
cluster center of gravity, this method uses the average of a
cluster and classifies. The process flow is shown as follows.
1. Several k of a cluster is determined and the k cluster
center of gravity is given as initial value. There are the
following methods in selection of the initial cluster
center of gravity. (1) Use the result of the clustering
performed before. (2) Presume from knowledge other
than clustering. (3) Generate at random.
2. To all individuals, distance with the k cluster center of
gravity is calculated, and distance arranges an
individual to the cluster used as the minimum.
3. The center of gravity of each cluster is re-calculated by
the individual rearranged by 2.
4. If it is below threshold with the number of the
individuals which changed the affiliation cluster, it
will be regarded as convergence and processing will
be ended. When other, it returns to 2. and processing is
repeated.
Like this fault, the sum in a cluster which is all the distance
of an individual and its cluster center of gravity decreases in
monotone. That is, the k-means method is a kind of the
climbing-a-mountain method. Therefore, although the k-means
method guarantees local optimal nature, global optimal nature
is not guaranteed. The result of clustering changes with setup
of the initial cluster center of gravity.
B. ISODATA
The ISODATA method is the method developed by Ball,
Hall and others in the 1960s. The ISODATA method is a
method which added division of a cluster, and processing of
fusion to the k-means method. The individual density of a
cluster is controllable by performing division and fusion to the
cluster generated from the k-means method. The individual in
a cluster divides past [a detached building] and its cluster, and
the distance between clusters unites them with past close. The
parameter which set up division and fusion beforehand
determines. The procedure of the ISODATA method is shown
as follows.
1. Parameters, such as the number of the last clusters, a
convergence condition of rearrangement, judgment
conditions of a minute cluster, branch condition of
division and fusion, and end conditions, are
determined.
2. The initial cluster center of gravity is selected.
3. Based on the convergence condition of rearrangement,
an individual is rearranged in the way of the k-means
method.
4. It considers with a minute cluster that it is below
threshold with the number of individuals of a cluster,
and accepts from future clustering.
5. When it is more than the threshold that exists within
fixed limits which the number of clusters centers on
the number of the last clusters, and has the minimum
of the distance between the cluster centers of gravity
and is below threshold with the maximum of
distribution in a cluster, clustering regards it as
convergence and ends processing. When not
converging, it progresses to the following step.
6. If the number of clusters exceeds the fixed range,
when large, a cluster is divided, and when small, it will
unite. It divides, if the number of times of a repetition
is odd when there is the number of clusters within
fixed limits, and if the number is even, it unites. If
division and fusion finish, it will return to 3. and
processing will be repeated.
Division of a cluster: If it is more than threshold with
distribution of a cluster, carry out the cluster along with the
first principal component for 2 minutes, and search for the new
cluster center of gravity. Distribution of a cluster is re-
calculated, and division is continued until it becomes below
threshold.
Fusion of a cluster: If it is below threshold with the
minimum of the distance between the cluster centers of
gravity, unite the cluster pair and search for the new cluster
center of gravity. The distance between the cluster center of
gravity is re-calculated, and fusion is continued until the
minimum becomes more than threshold.
Although the ISODATA method can adjust the number of
certain within the limits clusters, and the homogeneity of a
cluster by division and fusion, global optimal nature cannot be
guaranteed. Since the ISODATA method has more parameters
than the k-means method, adjustment of the parameter is still
more difficult.
C. Heredity Algorithm
A heredity algorithm (Genetic Algorithms: GA) is an
optimization algorithm modeled after the theory of evolution
of Darwin, and it will be advocated by Holland
3
in the 1960s.
The solution in question is expressed as an individual and an
each object is constituted from GA by the chromosome. An
individual evolves by selection, intersection, and mutation,
and searches for an optimum solution.
The general procedure of GA is shown as follows.
1. N individuals with a chromosome are generated as the
initial population (population). Simultaneous search of
the N points can be carried out by these N individuals.
2. Fitness value is searched for based on the fitness value
function beforehand defined to each individual.
3. Selection is performed based on fitness value. That is
what is screened out of N individuals of current
generation and the thing which survives the next
generation. The probability of surviving the next
generation becomes high so that the fitness value of an
individual is high, but the low individual of fitness
value may also survive in the next generation. This is a
role which controls lapsing into a partial solution.

3
http://www2.econ.iastate.edu/tesfatsi/holland.gaintro.htm
Vol. 3, No. 7, 2012

127 | P a g e
Tournament selection
4
: It is the method of repeating this
process until it selects a certain number of individuals at
random from the population, fitness value chooses the best
thing in it and the population's number of individuals is
obtained.
Elite strategy
5
: How many individuals with the maximum
fitness value call it the elite. The method of certainly leaving
the elite to the next generation regardless of a selection rule is
called elite strategy. The elite individual saved by an elite
strategy participates in neither intersection nor mutation.
4. By the set-up intersection probability or the intersection
method, the selected individual is crossed (crossover)
and a new individual is generated.
5. By the method of the set-up mutation rate or mutation,
mutation is performed and a new individual is
generated.
6. Fitness value is re-calculated to a new chromosome
group.
7. If end conditions are fulfilled, let the best individual
then obtained be the semi optimum solution in question.
Otherwise, it returns to 3.
GA is the multipoint search method, and is excellent in
global searching ability, also is widely applied to various
optimizations or a search problem.
D. Real Numerical Value GA
Early GA performed intersection and mutation by the bit
string which carried out the binary coding of the variable, and
has disregarded the continuity of a variable. On the other hand,
GA which performs intersection in consideration of the
continuity of a variable and mutation is called the real
numerical value GA (Real-Coded Genetic Algorithms)
6
using
the numerical value itself. In this research, the threshold of an
initial cluster center and division/fusion is optimized based on
the real numerical value GA.
GA with a general flow of processing of the real numerical
value GA is the same. Since the coding method is merely
different, the original intersection method and the mutation
method are used.
The intersection method of real numerical value GA daily
use has the BLX-alpha method
7
, single modal normal
distribution crossing method (Uni-modal Normal Distribution

4

http://www.google.co.jp/#hl=ja&rlz=1W1GGLD_ja&q=tonament+selection+
genetic+algorithm&oq=tonament+selection+genetic+algorithm&aq=f&aqi=&
aql=&gs_l=serp.3...4859.11766.0.13031.25.24.0.0.0.6.344.4800.0j14j8j2.24.0.
..0.0.EzWVG3xcR3I&pbx=1&bav=on.2,or.r_gc.r_pw.,cf.osb&fp=48edeecabd
799c01&biw=1280&bih=799
5
http://www.sersc.org/journals/IJSH/vol2_no2_2008/IJSH-Vol.2-No.2-W03-
Solving%20Unbounded%20Knapsack%20Problem.pdf
6

http://www.google.co.jp/#hl=ja&site=&source=hp&q=real+coded+genetic+al
gorithm&rlz=1W1GGLD_ja&oq=real+coded+&aq=0L&aqi=g-
L6&aql=&gs_l=hp.1.0.0i19l6.2438.7016.0.11266.11.9.0.2.2.0.250.1454.0j8j1.
9.0...0.0.AZfbfrGsWhw&pbx=1&bav=on.2,or.r_gc.r_pw.,cf.osb&fp=ede0ef6
aa5da214e&biw=1280&bih=758
7
http://www.iitk.ac.in/kangal/resources.shtml
crossover: UNDX)
8
:, etc., and the mutation method has
mutation, uniform mutation, etc. by a normal distribution.
The BLX-alpha crossing method: This intersection method
determines a child as follows,
1. Two parent individuals are set to a and b.
2. The section [A, B] of intersection is calculated by
the following formulas.

b a b a B
b a b a A

) , max(
) , min(
(1)
3. A uniform random number determines a child
individual from the section [A, B].
Mutation by a normal distribution: It can be happened
mutation by a normal distribution. The normal distribution
used at this time will be decided with the random number
according to the normal distribution of the average of x
distribution delta 2, if a parent individual is set to x. The
individual generated exceeding the range of x [XMIN,
XMAX] is stored in within the limits.
II. PROPOSED CLUSTERING METHOD
It decided to use GA also for the determination of the
threshold of the separation in clustering by ISODATA, and
integration. It is because a clustering result will constitute
inevitably the cluster that cluster distribution becomes the best
for a case to a convex function wholly in the bottom if this sets
up an fitness value function which makes the maximum the
ratio of synthesis of distribution between clusters, and
synthesis of cluster internal variance. By the method of
repeating separation and integration like ISODATA, it decided
to avoid an above-mentioned problem by controlling this
threshold. The consecutiveness distribution form based on it
needs the concept of the distance in the feature space, and to
be judged for this control, and in order to perform this, GA is
used in this paper.
A. Partial Mean Distance
As partial mean distance is shown in Fig.1, the average of
the distance between the individuals belonging to the same
cluster of a certain part within the limits is called partial mean
distance. It can ask for the sum of partial mean distance all
over the districts by moving the range of a part by the Moving
Window
9
method little by little. The window of the Moving
Window method here is a super-sphere in n-dimensional
Euclidean space.
Since distribution of a cluster is not necessarily uniform
distribution, when the window of the Moving Window method

8

http://www.google.co.jp/#hl=ja&rlz=1W1GGLD_ja&q=unimodal+normal+di
stribution+crossover+genetic+algorithm&oq=unimodal+normal+distribution+
crossover+genetic+algorithm&aq=f&aqi=&aql=&gs_l=serp.3...4828.22188.0.
23157.52.44.0.0.0.10.250.6675.0j38j6.44.0...0.0.n-
0BTKOtkRQ&pbx=1&bav=on.2,or.r_gc.r_pw.,cf.osb&fp=48edeecabd799c01
&biw=1280&bih=758
9

http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=548342&url=http%3A
%2F%2Fieeexplore.ieee.org%2Fiel3%2F3974%2F11463%2F00548342.pdf%
3Farnumber%3D548342
Vol. 3, No. 7, 2012

128 | P a g e
is moved at equal intervals, useless calculation may be carried
out in a place without an individual. In order to avoid this, in
this paper, in all individuals, it will move for every individual
and the sum of partial mean distance all over the districts will
ask for the super-sphere centering on an individual.

Figure 1Partial mean distance
Making the sum of partial mean distance all over the
districts into the minimum, the density of an individual, an
individual can be made to belong to a separate cluster along a
crevice in a small place, i.e., a place with a crevice that is, the
boundary line of a cluster can be made so much to a concave
set.
B. Difference from the Conventional ISODATA Method
The ISODATA method is a method which cluster
distribution assumes to be a convex function. When cluster
distribution is a concave function, by the ISODATA method, it
can respond to some extent by division and fusion, but if the
procedure of the conventional ISODATA method is followed,
the cluster classified correctly once may be destroyed.
Since equivalence will be carried out if the individual
rearrangement in the process of the ISODATA method is the
k-means method in fact and a cluster can be divided in a
straight line with a Voronoi figure
10
, when cluster distribution
is a concave function, the cluster divided by division and
fusion with the curve may be destroyed by rearrangement of
an individual. Then, after the proposal method unites the last
of the ISODATA method, it does not rearrange an individual
and is ended.
When cluster distribution is a concave function, suppose
that former data was divided by the threshold of suitable
division. Since the distance between clusters changes into the
united process when uniting a cluster after this, as for the turn
of fusion, a result will be affected. When it does so, even if
there is a threshold of suitable fusion, a desirable fusion result
may not be brought. By the proposal method, in order to
depend for the fusion result of a cluster only on the threshold
of fusion, simultaneous fusion of the cluster filled to the
threshold of fusion is carried out.
Moreover, since the center of a cluster is presumed by Real
Coded GA: RCGA by the proposal method, even if a
clustering result reduces the number of times of a repetition of
the ISODATA method for which it does not depend on
correction of the center of a cluster by repetition of the
ISODATA method to some extent, it hardly influences a

10
http://otago.ourarchive.ac.nz/handle/10523/765
clustering result. Therefore, in this paper, the number of times
of a repetition of the ISODATA method is set to 2 for the
improvement in calculation speed.
C. Selection of Fitness Value Function
Since the cluster that F is made into the maximum and that
cluster distribution will become [as for the result of clustering]
the best for a case to a convex function wholly in the bottom if
an fitness value function setup is carried out will be
constituted, it is not suitable when cluster distribution is a
concave function.
As make into the minimum of the sum of partial mean
distance all over the districts, since only the crevice within the
limits between parts will be observed if a fitness value
function setup is carried out, a cluster may not become a lump.
d
m
F Fitness
(2)
In this equation, F expresses a false F value, d denotes the
sum of partial mean distance all over the districts, and m
expresses weight.
Here, when asking for the sum of partial mean distance,
selection of the range of a part has large influence on a result.
If the range of a part is too small, a crevice cannot be covered
and the boundary line of a cluster cannot be made correctly.
Moreover, when the range of a part is too large, locality may
lose. The radius of a super-sphere which expresses the range
of a part with the proposal method for an object as one cluster
is enlarged little by little from the shortest distance between
individuals to the maximum distance, and it asks for the sum
of partial mean distance. In the time of the radius of a super-
sphere becoming at least in the width of a crevice, the sum of
partial mean distance reaches one peak. In order to carry out
the certain cover of the crevice, the sum of partial mean
distance makes a few radiuses this becomes a peak, the range
of the part actually using a super-sphere with a large radius.
D. Set-up Parametersfor RCGA
The selection method, tournament selection and an elite
strategy is used. The size of tournament selection is set as 3.
Using the BLX-alpha method, the intersection method sets
the value of alpha as 0.5, and sets up intersection probability to
70%.
Using the mutation method by a normal distribution, the
mutation method sets the value of sigma as 0.5, and sets up
mutation probability to 1%.
Termination conditions: the elite, five-generation
maintaining t as a thing and five generations of differences of
the average fitness values and the elite's fitness value
continuing 2% in within the limits.
By the ISODATA method, the threshold of an initial
cluster center, division, and fusion is presumed by GA.
III. EXPERIMENTS
A. Performance Evaluation Method
A different clustering result is obtained from the separate
clustering method in many cases. Also by the same clustering
Vol. 3, No. 7, 2012

129 | P a g e
method, it sometimes often results in changing with setup of a
parameter. The criteria of evaluation are needed in order to
compare the result of clustering. F value (pseudo F statistic) is
one of the valuation bases often used. F is defined by the
following formulas.
k n
l l
k
l l l l
F
k
j C i
j i
k
j C i
j i
n
i
i
j
j

1
2
1
2
1
2
) (
1
) ( ) (
(3)
where n as for the total number of individuals and k, as for
Cj, Clusters j and li express the number of clusters, Individual
i and #j express the average of Cluster j, and # expresses the
average of all individuals. F is criteria which consider
simultaneously the variation within a cluster, the variation
between clusters, and the number of clusters, and the figure is
the ratio of distribution between groups, and group internal
variance. Since group internal variance with large distribution
between groups means the small thing if F is large, a
clustering result shows a good thing. However, since cluster
distribution assumes it as the convex function in F, in the case
of the concave function, it is not suitable.
In the case where the correct answer of a classification is
known, the error E of a result can be searched for from the
number of individuals c classified correctly.
% 100
n
c n
E
(4)
B. Selection of Fitness Value Function
The proposal method experiments by making the data of
simple degree of convection to verify whether it can respond
not only when cluster distribution is a convex function, but in
the case of a concave function.
As shown in Fig.2, it experiments using the data containing
two clusters of degree of convection. It clusters by the
ISODATA method and the proposal method with a random
parameter, and the result of clustering is compared. The result
of an experiment is shown like a lower figure. The error which
cannot understand the cluster of degree of convection in a
straight line, and cannot classify it according to the
conventional ISODATA method correctly from this
experimental result is 12.5%. And by the proposal method, it
turns out that an error becomes 0% and it can classify
according to division and fusion correctly with a curve.
C. Experiemnt 2
Next, it experiments using the Iris data set of UCI
repository
11
, a Wine data set, a Ruspini data set, and a New
thyroid data set. Iris is a 4-dimensional data set with a number
of individuals 150 and three categories. Wine is a 13-
dimensional data set with a number of individuals 178 and
three categories. Ruspini is a 2-dimensional data set with a
number of individuals 75 and four categories. New thyroid is a
5-dimensional data set with a number of individuals 215 and

11
http://archive.ics.uci.edu/ml/(standard dataset for machine learning)
three categories. These four data sets are criteria data sets
often used for comparison of the clustering method. When
clustering an Iris data set by the case where a parameter is
presumed by GA, change of the fitness value of 50 generations,
i.e., the process of convergence, is shown in Fig. 5.

Figure 2 Original data Figure .3 ISODATA method (Random)

Figure 4 Proposed method (ISODATA-GA)

Figure 5 Convergence process of GA
In this figure, a red line expresses the elite's fitness value,
and expresses the average of fitness value with the green line.
This figure shows being completed by the average of fitness
value. Then, the similar result was obtained in the experiment
of a Wine data set, a Ruspini data set, and a new thyroid data
set. The data of the experimental result of these data sets is
gathered in Table 1.
TABLE I. CLUSTERING PERFORMANCE WHEN THE NUMBER OF
CLUSTER IN AGREEMENT WITH TEH NUMBER OF TARGET CLUSTERS
Error (%) The ISODATA method
Random
The proposed method

Iris 21.33 11.33
Wine 34.65 11.34
Ruspini 46.76 4.00
New thyroid 31.63 15.81
Vol. 3, No. 7, 2012

130 | P a g e
In the table, compared with the conventional ISODATA
method, the direction of the proposal method shows that the
error decreases 22.97%, respectively. Although optimizing
took about 100-time long time from the conventional
ISODATA method, accuracy went up certainly.
D. Experiemnt 3
Comparative study on clustering performance and
processing time between the proposed ISODAT GA method
and the conventional clustering methods of (1) Minimum
Distance Method, (2) Maximum Distance Method, (3) Median
Method, (4) Gravity Method, (5) Group Average Method, (6)
Ward Method, (7) K-Means Method with the best random
initial cluster center selection, (8) K-Means Method with the
average random initial cluster center selection, (9) ISODATA
Method with the best random initial cluster center selection,
(10) ISODATA Method with the average random initial
cluster center selection is conducted with Iris dataset of UCI
repository. Table 2 shows clustering performance of the
proposed and the other conventional clustering methods.
Although processing time required for the proposed
ISODATA GA clustering is three times much longer, F value
which represents the ratio of inner cluster variance to between
cluster variance of the proposed ISODATA GA shows the best
value together with clustering error (it also shows almost
minimum value). In particular, 6.2% of clustering error is
reduced by the proposed ISODATA GA clustering method in
comparison to the conventional ISODAT with best selection
of initial cluster center randomly (ISODATA(Ran.Best)),
separability between clusters, F are almost same though. Also
68.7% of separability improvement is observed in comparison
to the conventional ISODATA(Ran.Ave.). Therefore,
parameter (merge and split of the clusters) estimation based on
GA as well as initial cluster center determination with GA are
effective to improve clustering performance.
Iris dataset is four dimensional data which consists of 150
of the number of data points with the number of categories
(cluster) of three. It may say that Iris is the typical dataset
among UCI repository.
V. CONCLUSION
The proposed clustering method is based on the
conventional ISODAT method. One of the problems of the
ISODATA method is relatively poor clustering performance.
For instance, ISODAT as well as the other conventional
clustering method do not work well if the probability density
function of data is distributed as concave, then linear
discrimination function does not work well. Taking such
probability density function into account, parameters for
merge and split of the clusters can be adjusted with GA. Also
the proposed ISODATA GA method determines initial cluster
center with GA. Clustering performance depends on the
designated initial cluster center. Therefore, if not appropriate
initial cluster center is determined, then cluster results become
bad. The proposed ISODATA GA method determines most
appropriate initial cluster center by using GA. Therefore,
cluster results become excellent. The experimental results
show that most appropriate cluster result can be obtained with
the proposed ISODATA GA for the situation of shape
independent clustering with concave shape of input data
distribution. Also the experimental results with UCI repository
show that the proposed ISODATA GA method is superior to
the conventional ISODATA clustering method with randomly
determined initial cluster center. It is also found that the
proposed ISODATA GA method is superior to the other
typical conventional clustering methods.

TABLE 2 CLUSTERING PERFORMANCE OF THE PROPOSED ISODATA WITH
GA FOR IRIS DATASET OF UCI REPOSITORY.
F E(%) Process Time(s)
Minimum Distance Method
12
277.493 32 0.14
Maximum Distance Method
13
484.899 16 0.14
Median Method
14
501.303 10 0.156
Gravity Method
15
555.666 9.33 0.156
Group Average Method
16
399.951 25.33 0.14
Ward Method
17
556.841 10.67 0.14
K-means(Ran.Best)
18
560.366 11.33 0.06
K-means(Ran.Ave.)
19
210.279 46.23 0.06
K-means(GA)
20
560.4 10.67 0.225
ISODATA(Ran,Best)
21
560.366 11.33 0.313
ISODATA(Ran,Ave.)
22
175.465 38.64 0.313
ISODATA(GA)
23
560.4 10.67 1.523
ACKNOWLEDGMENT
The author would like to thank Dr. XingQiang Bu for his
effort to experimental studies.
REFERENCES
[1] Kohei Arai, Fundamental theory for pattern recognition, Gakujutu-
Tosho-Shuppan Pub. Co., Ltd.,1999.
[2] Hartigan, J.A., Clustering Algorithms, NY: Wiley, 1975.
[3] Anderberg, M.R. , Cluster Analysis for Applications, New York:
Academic Press, Inc., 1973.
[4] Bottou, L., and Bengio, Y., "Convergence properties of the K-means
algorithms," in Tesauro, G., Touretzky, D., and Leen, T., (eds.)
Advances inNeural Information Processing Systems 7, Cambridge, MA:
The MIT Press, 1995
[5] V. Vapnik. The Nature of Statistical Learning Theory. Springer-Verlag,
New York, 1995.

12
http://en.wikipedia.org/wiki/Cluster_analysis
13
http://en.wikipedia.org/wiki/Hierarchical_clustering
14

http://www.cra.org/Activities/craw_archive/dmp/awards/2003/Mower/KMED.
html
15
http://dl.acm.org/citation.cfm?id=585147.585174
16
http://nlp.stanford.edu/IR-book/html/htmledition/group-average-
agglomerative-clustering-1.html
17
http://www.statsoft.com/textbook/cluster-analysis/
18
http://en.wikipedia.org/wiki/K-means_clustering
19
http://home.dei.polimi.it/matteucc/Clustering/tutorial_html/kmeans.html
20
http://airccse.org/journal/ijaia/papers/0410ijaia3.pdf
21
http://www.cs.umd.edu/~mount/Projects/ISODATA/
22

http://www.yale.edu/ceo/Projects/swap/landcover/Unsupervised_classification
.htm
23

http://www.yale.edu/ceo/Projects/swap/landcover/Unsupervised_classification
.htm
Vol. 3, No. 7, 2012

131 | P a g e
[6] L.Breiman and J.Friedman, Predicting multivariate responses in multiple
linear regression, Technical report, Department of Statistics, University
of California, Berkley, 1994.
[7] R.J.Williams, Simple statistical gradient-following algorithms for
connectionist reinforcement learning, Machine Learning, 8, 3-4, 229-
256, 1992.
[8] L.P. Rieber, Computer, Graphics and Learning, Madison, Wisconsin:
Brown & Benchmark, 1994.
[9] C. Yi-tsuu, Interactive Pattern Recognition, Marcel Dekker Inc., New
York and Basel, 1978.
[10] R.K. Pearson, T. Zylkin, J.S. Schwaber, G.E. Gonye, Quantitative
evaluation of clustering results using computational negative controls,
Proc. 2004 SIAM International Conference on Data Mining, Lake Buena
Vista, Florida, 2004.
[11] Goldberg D., Genetic Algorithms, Addison Wesley, 1988, or, Holland
J.H., Adaptation in natural and artificial system, Ann Arbor, The
University of Michigan Press, 1975.
[12] Trevor Hastie, Robert Tibshirani, Jerome Friedman The Elements of
Statistical Learning: Data Mining, Inference, and Prediction, Springer,
New York, 2001.
[13] Jensen, J.R., Introductory Digital Image Processing. Prentice Hall, New
York, 1996.
[14] Kohei Arai and XianQiang Bu, ISODATA clustering with parameter
(threshold for merge and split) estimation based on GA: Genetic
Algorithm, Reports of the Faculty of Science and Engineering, Saga
University, Vol. 36, No.1, 17-23, 2007
[15] MacQueen, J.B., Some Methods for Classification and Analysis of
Multivariate Observations,, Proceedings of the Fifth Berkeley
Symposium on Mathematical Statistics and Probability, 1, 281-
297,1967.
AUTHORS PROFILE
Kohei Arai, He received BS, MS and PhD degrees in 1972, 1974 and
1982, respectively. He was with The Institute for Industrial Science, and
Technology of the University of Tokyo from 1974 to 1978 also was with
National Space Development Agency of Japan (current JAXA) from 1979 to
1990. During from 1985 to 1987, he was with Canada Centre for Remote
Sensing as a Post-Doctoral Fellow of National Science and Engineering
Research Council of Canada. He was appointed professor at Department of
Information Science, Saga University in 1990. He was appointed councilor for
the Aeronautics and Space related to the Technology Committee of the
Ministry of Science and Technology during from 1998 to 2000. He was also
appointed councilor of Saga University from 2002 and 2003 followed by an
executive councilor of the Remote Sensing Society of Japan for 2003 to 2005.
He is an adjunct professor of University of Arizona, USA since 1998. He also
was appointed vice chairman of the Commission A of ICSU/COSPAR in
2008. He wrote 30 books and published 332 journal papers.

Vol. 3, No. 7, 2012

133 | P a g e
Improving the Rate of Convergence of Blind
Adaptive Equalization for Fast Varying Digital
Communication Systems
Iorkyase, E.Tersoo
Electrical and Electronics Engineering
University of Agriculture, Makurdi, Nigeria
Michael O. Kolawole
Electrical and Electronics Engineering
Federal University of Technology, Akure
Akure,Nigeria

Abstract The recent digital transmission systems impose the
application of channel equalizers with bandwidth efficiency,
which mitigates the bottleneck of intersymbol interference for
high-speed data transmission-over communication channels. This
leads to the exploration of blind equalization techniques that do
not require the use of a training sequence. Blind equalization
techniques however suffer from computational complexity and
slow convergence rate. The Constant Modulus Algorithm (CMA)
is a better technique for blind channel equalization. This paper
examined three different error functions for fast convergence and
proposed an adaptive blind equalization algorithm with variable
step size based on CMA criterion. A comparison of the existing
and proposed algorithms speed of convergence shows that the
proposed algorithm outperforms the other algorithms. The
proposed algorithm can suitably be employed in blind
equalization for rapidly changing channels as well as for high
data rate applications.
Keywords- Blind Equalization; Channel Equalizer; Constant
Modulus Algorithm; Intersymbol interference; Variable Step Size.
I. INTRODUCTION
One of the most important advantages of the digital
transmission systems for voice, data and video communication
is their higher reliability in noise environment in comparison
with that of their analog counterparts. Both existing wired and
wireless communication systems have significantly made a
shift to digital transmission of data. Unfortunately, most often
the digital transmission of information is accompanied with a
phenomenon known as intersymbol interference (ISI). This
means that the transmitted pulses are smeared out so that
pulses that correspond to different symbols are not separable.
ISI is a common problem in telecommunication system and
wireless communication systems, such as television
broadcasting, digital data communication, and cellular mobile
communication systems.
In telecommunication systems, ISI occurs when the
modulation bandwidth exceeds the coherent bandwidth of the
radio channel where modulation pulses are spread in time. For
wireless communication, ISI is caused by multipath in band-
limited time-dispersive channels, and it distorts the transmitted
signal, causing bit errors at the receiver. ISI has been
recognized as the major obstacle to high-speed data
transmission with the required accuracy and multipath fading
over radio channels.
Obviously, for a reliable digital transmission system, it is
crucial to reduce the effect of ISI. This can be achieved by the
technique of equalization [1, 2]. Equalization is one of the
techniques that can be used to improve the received signal
quality in telecommunication especially in digital
communication. In a broad sense, the term equalization can be
used to describe any signal processing operation that
minimizes the ISI. Two of the most intensively developing
areas of digital transmission, namely digital subscriber lines
(DSL) and cellular communication (GSM) are strongly
dependent on the realization of reliable channel equalizers [3,
4, 5].
There are generally two approaches to equalization:
conventional equalization, and blind equalization. In
systems employing conventional equalization, a training
sequence is transmitted over the channel prior to the
transmission of any useful data. The training sequence is a
data sequence that is a priori known to the receiver. The
receiver uses the relationship between the known training
sequence and the sequence it actually receives to construct an
approximation of the inverse transfer function for the channel.
The equalizer is then configured to use the inverse transfer
function and thereby induce minimal ISI.
Conventional equalization is problematic in some
communication systems, such as mobile and multi-point
communication systems, because the training sequence uses
up scarce bandwidth resources that could otherwise be used to
transmit useful data. Such systems, therefore, often use blind
equalization, which is a form of equalization that does not
require the use of a training sequence.
It is desirable to improve the ability of digital
communication systems to minimize ISI, including
communication systems employing blind equalization.
Systems achieving such reduced ISI are capable of achieving
reduced data error rates at prevailing data transmission rates,
or can obtain higher data transmission rates without sacrificing
data integrity, in order to obtain better overall system
performance [6].
Vol. 3, No. 7, 2012

134 | P a g e
In order to provide an efficiently high data rate
transmission with high accuracy in digital communication
without spectral wastage, advanced signal processing
techniques are necessary. As earlier mentioned, several digital
communication systems are inherent with rapid varying
channel characteristics. The assumption that a communication
channel is stationary over a transmission period is not valid.
For communication systems like mobile communication this
assumption results in performance degradation. There arises a
need for algorithms that can exercise tracking ability in the
face of fast changing characteristics of communication
channels [7].
The rate of convergence becomes very pivotal in the
development of any such algorithms. This paper considers an
adaptive blind equalization technique based on Constant
Modulus Criterion using different error functions and a
comparison of their speed of convergence is made.
II. SYSTEM MODEL FOR BLIND ADAPTIVE EQUALIZATION
The baseband Model of a communication system for
channel equalization is shown in Fig 1.
In Fig. 1, our communication channel and equalizer are
both modeled using a Finite Impulse Response filter (FIR). Of
course, the channel and equalizer can also be modeled as
Infinite Impulse Response filter (IIR).
However, it is dangerous to update the poles of an IIR
filter in real-time, because it is possible that they could move
outside the unit circle, hence causing instability in the system.
From Figure 1, the relationship between the input and
output of the FIR channel filter is,

=
k
k n x k h n b ) ( ) ( ) ( (1)
The input to the equalizer is,

) ( ) ( ) ( n v n b n r + = (2)
This implies that,

) ( ) ( ) ( ) ( n v k n x k h n r
k
+ =
(3)
where ) (k h is the channel response.

) ( k n x is the input to the channel at time k n .

) (n v represents the additive noise (zero mean)

) (n r is the input to the equalizer.
The output of the system ) (n y is given as,

) ( ) ( n r w n y
T
= (4)
where ) (n y is the equalizer output and
T
w is the tap
weight vector.
The equalizer model (4) forms the basis for the discussion
of the blind adaptive algorithms in the following subsection.
A. Constant Modulus Algorithm Criterion
The CMA criterion may be expressed by the non-negative
cost function
q CMAp
J
,
parameterized by positive integer, p
and q .

Channel

Adaptive
Equalizer
Estimator
Blind
Algorithm
) (n x
) (n b
) (n v
) (n r ) (n y
) ( n x
) (n e
-
+
E
E
Figure 1: Baseband model of a Communication System for Channel Equalization.
Vol. 3, No. 7, 2012

135 | P a g e

)
`
=
q
p
CMA
R n y E
pq
J
q p
) (
1
,

(5)
where R is a fixed constant, chosen for each form of
modulation schemes, represents the statistics of the transmitted
signal.
CMA
J in (5) is a gradient based algorithm [8] and
works on the premise that the existing interference causes
fluctuation in the amplitude of the output that otherwise has a
constant modulus. For the simplest case we put 2 = p and
2 = q , we have

( )
2
2
) (
4
1
R n y J
CMA
= (6)
It updates equalizer coefficients by minimizing the cost
function. The steepest gradient descent algorithm [9] is
obtained by taking the instantaneous gradient of
CMA
J which
results in an equation that updates the system.

)) ( ( ) ( ) 1 ( n f g n w n w = + (7)

)) ( ( ) ( )) ( ( n y n r n f g
-
= (8)

( )
2
2
) (
) (
4
1
)) ( ( R n y n y
n y
V = (9)

( )
2
1
) ( ) ( )) ( ( n y R n y n y = (10)
where w n
( )
is the equalizer coefficient, ) (n r is the
receiver input, is the step size constant and )) ( (
1
n y is
the error function for CMA
With the above expression, )) ( ( n y for the error
function, the usual CMA becomes:

( )
2
) ( ) ( ) ( ) ( ) 1 ( n y R n y n r n w n w + = +
-
(11)

( )
2
) ( ) ( ) ( ) ( ) 1 ( n y R n y n r n w n w
i i
+ = +
-
(12)
where
i
w is the
th
i tap of the equalizer. The signed error
version of CMA (SE-CMA) takes

| | )) ( ( sgn )) ( (
1 2
n y n y = (13)
and updates the equalizer as;
( ) { }
2
) ( ) ( sgn ) ( ) ( ) 1 ( n y R n y n r n w n w + = +
-
(14)

B. Proposed Error Functions
In this section we construct three error functions
)), ( ( n y
i
= i 3, 4, 5. Assume the source to be a real BPSK
(Binary Phase Shift Keying) with equiprobable alphabet and
unity dispersion . 1 = R Note that the dispersion constant R is
chosen for each form of modulation scheme. Let the product
of the squared deviations of the output be | | ) (n y P . For real
BPSK source, | | ) (n y P is taken as [10]:

( )( ) | |
2
1 ) ( 1 ) ( + n y n y
in the
case of ( ) ) (
3
n y and

( )( ) | |
2
1 ) ( 1 ) ( ) ( + n y n y n y

In the case of
( ) ) (
4
n y

Using a step size parameter , the updated equation can
be written as

( ) ) (
2
1
) ( ) 1 ( n y P n w n w V = + (15a)

( )
) (
) ( 2
2
1
) (
n w w
w
n y P
n w
=
(
c
c
=

(15b)

( )
) (
) (
) (
) ( n r
n y
n y P
n w
-
c
c
= (15c)

The factor is used merely for convenience. The above
algorithm might experience gradient noise amplification
whenever ( ) ) (n y P is large. Hence we normalized the
correction term above by dividing with ( ) ) (n y P a + with the
choice of a = 1. This idea comes from the well-known
normalized LMS algorithm, which can be viewed as the
minimum-norm solution. The positive constant a is used to
remove numerical difficulties that arise when the denominator
is close to zero. Thus the equation (15c) becomes;

( )
( )
) (
) ( 1
) (
) (
) ( ) 1 ( n r
n y P
n y
n y P
n w n w
-
+
c
c
= + (16)

With the expressions
( )( ) | |
2
1 ) ( 1 ) ( + n y n y
and
( )( ) | |
2
1 ) ( 1 ) ( ) ( + n y n y n y
for
( ) ) (n y P
respectively, we
arrive at:

| |
| |
2
2
2
3
1 ) ( 1
) ( 1 ) ( 4
)) ( (
+

=
n y
n y n y
n y (17)

| || |
| |
2
2 2
2 2
4
1 ) ( ) ( 1
1 ) ( 3 ) ( 1 ) ( 4
)) ( (
+

=
n y n y
n y n y n y
n y (18)

Further, as the step size controls the rate of
convergence, with a large value giving fast convergence and a
smaller value providing better steady state performance we
introduce a variable step size, , as done in the case of
traditional least mean square (LMS) algorithm [11]. We
suggest )) ( (
5
n y (proposed error function) as:

)) ( ( ) ( )) ( (
4 5
n y n n y = (19)
Vol. 3, No. 7, 2012

136 | P a g e

)) ( ( ) 1 ( ) (
4
n y n n | + = (20)
with , 1 0 < < 0 > | and
=
), (
,
) (
min,
max
n
n

min
max
) (
) (

<
>
n
n
otherwise
if
if
(21)

where a condition of
max min
0 < < must be satisfied.
The initial value of variable step size ) 0 ( is chosen
according to the upper bound constant
max
.
In equation (20), and | are two fixed parameters
which control the variation of within two limits
min
and
max
. The updated equation in (8) with the step-size and
error function as in (19) and (20) respectively is thus a
variable step-size algorithm. The effectiveness of the three
error functions (algorithms) with the usual CMA algorithm is
shown through simulation in the next section.
III. SIMULATION CONSIDERATIONS
To demonstrate the effectiveness of the error functions
proposed which affect the speed of convergence of the
equalizer, we assume that the transmitted signal is a zero mean
distributed random sequence and is modulated using binary
phase shift keying (BPSK) in which case the dispersion
constant R is taken unity.
Also for the simulation the following parameters were
used:
(i) a two-tap filter | |
T
w w w
1 0
= ,
(ii) a constant step size 001 . 0 = and
min
= 0.00001
and
max
= 0.1 for variable step size algorithm,
(iii) = 0.97, and | = 0.0005, and equalizer
initialization coefficients of 2 and 0.5.
The additive noise after the channel is neglected, owing to
the fact that in most digital communications the dominant type
of distortion for which the equalizers are employed is the time
varying multipath fading phenomenon, intersymbol
interference (ISI).
Also, the BPSK modulation scheme used in this work is
relatively immune to the additive noise levels present in
communication channels.
IV. RESULTS
Figures 2 to 4 show the plot of squared error against the
iteration number for the usual Constant Modulus Algorithm
(CMA) and the three different algorithms examined in this
study.

The effects that the blind adaptive equalizer has on the
signal quality are shown in Figures 2, 3, and 4. The algorithms
aim at altering the filter impulse response so as to reduce and
ultimately minimize the cost function. This is clearly seen in
these Figures where the curves descend with time and then
level out at their minimum. The time constant, or, more
generally the convergence time of the algorithm, is indicated
by the rate at which the cost function is reduced.
0
1
2
3
4
5
6
7
8
9
10
0 5000 10000 15000
s
q
u
a
r
e
d

e
r
r
o
r
iteration number
CMA with Fixed Step Size
0
1
2
3
4
5
6
7
8
9
10
0 5000 10000 15000
s
q
u
a
r
e
d

e
r
r
o
r
iteration number
CMA with Fixed Step Size
Figure 3: Squared error versus iteration for CMA
algorithm with step size 0.001.
0
1
2
3
4
5
6
7
0 5000 10000 15000
s
q
u
a
r
e
d

e
r
r
o
r
iteration number
Normal CMA
0
1
2
3
4
5
6
7
0 5000 10000 15000
s
q
u
a
r
e
d

e
r
r
o
r
iteration number
Normal CMA
Figure 2: Squared error versus iteration for normal
CMA
Vol. 3, No. 7, 2012

137 | P a g e

The Constant Modulus Algorithm (CMA) no doubts, does
reduce the cost function, however, as observed in Figure 2, the
normal CMA tends to diverge even after 1msec (10
4

iterations) of the time when the adaptation begins. This shows
that it converges at some local minimum after a long time,
comparatively. The algorithms with fixed step size shown in
Figure 3 tend to continue improving the equalizers
performance even beyond 0.5msec point. The effect of the
proposed algorithm with variable step size is quite obvious as
can be seen in Figure 4.
The performance of this algorithm improves dramatically
over the first few milliseconds and then flattens out. This
variable step size algorithm appears to outperform all the other
algorithms in this work because it ultimately minimizes the
cost function to it minimum in a very short time, restoring the
constant modulus of the transmitted signal, thus recovering the
original signal. Note that the time it takes an algorithm to
converge is significant: it is the measure of its ability to track
the changing impulse response of the propagation channel.
From these results, it has been demonstrated that the
equalizer, which employs this algorithm, does a good job of
computing the new coefficient every 0.5msec, which is good
for fast varying channels.
V. CONCLUSIONS
The quality of service that a blind equalizer is able to
provide is marked by its convergence speed; that is, the
number of received samples that it needs to provide good
enough estimates of the channel characteristics. This work has
examined different error functions that might be incorporated
into the Constant Modulus Algorithm (CMA) for fast
convergence. The proposed variable step size algorithm out-
performs the other two models in terms of convergence. The
proposed algorithm can suitably be employed in blind
equalization for rapidly changing channels as well as for high
data rate applications. Based on our investigations blind
method is a promising approach towards high data rate
transmission and warrants further research in future
communication technologies.
REFERENCES
[1] Kolawole, M. O. (2002). Satellite Communication Engineering, New
York: Marcel Dekker
[2] Ai, B., Yang, Z., Pan, C., Ge, J., Wang, Y. and Lu, Z. (2006). On the
Synchronization Techniques for Wireless OFDM Systems, IEEE
Transactions on Broadcasting, vol. 52, no. 2, pp 236-244.
[3] Proakis, J. G. (2001). Digital Communication, Mc Graw-Hill
Companies, Inc., International Edition.
[4] Adinoyi, A. Al-Semari, S. and Zerquine, A. (1999). Decision Feedback
Equalization of Coded I-Q QPSK in Mobile Radio Environments,
Electron, Lett., vol. 35, pp. 13-14.
[5] Samueli, H., Daneshrad B., Joshi R., Wong B. and Nicholas H. (1991).
A 64-tap CMOS echo canceller/decision feedback equalizer for 2B1Q
HDSL transceivers, IEEE Journal Selected Areas in Communication,
vol.9, pp. 839-847.
[6] Iorkyase, E.T. (2010). A High Performance Blind Adaptive Channel
Equalization Technique for Combating Intersymbol Interference in
Digital Communication, MEng Thesis, Department of Electrical and
Electronics Engineering, Federal University of Technology, Akure,
Nigeria.
[7] Iorkyase, E.T. and Kolawole, M. O. (2010). Comparative Study of
Fixed and Variable Step Size Blind Equalization Algorithms in
Communication Channels, International Journal of Electronics and
Communication Engineering, vol. 3, no. 3, pp. 125-130
[8] Kazemi, S., Hassani H. R., Dadashzadeh G., and Geran F. (2008).
Performance Improvement in Amplitude Synthesis of Unequally
Spaced array Using Least Mean Square Method, Progress in
Electromagnetic Research B, vol. 1, pp. 135-145.
[9] Kundu, A. and Chakrabarty, A. (2008). Fractionally Spaced Constant
Modulus Algorithm for Wireless Channel Equalization Progress in
Electromagnetic Research B, vol. 4, pp. 237- 248.
[10] Nelatury, S. R. and Rao, S. S. (2002). Increasing the Speed of
Convergence of the Constant Modulus Algorithm for Blind Channel
Equalization IEEE Transaction on Communication, vol. 50. No. 6.
[11] Kwong, R. H. and Johnston, E. W. (1992). A Variable Step Size LMS
Algorithm, IEEE Trans. Signal Processing, vol. 40, pp. 1633-1642.

0
5
10
15
20
25
0 5000 10000 15000
s
q
u
a
r
e
d

e
r
r
o
r
iteration number
Proposed Variable Step Size CMA
0
5
10
15
20
25
0 5000 10000 15000
s
q
u
a
r
e
d

e
r
r
o
r
iteration number
Proposed Variable Step Size CMA
Figure 4: Squared error versus iteration for the proposed
adaptive blind equalization algorithm with variable step size
based on CMA criterion.

Vol. 3, No. 7, 2012

137 | P a g e
Throughput Analysis of Ieee802.11b Wireless Lan
With One Access Point Using Opnet Simulator

Isizoh A. N.
1
Department of Electronic and Computer Engineering,
Nnamdi Azikiwe University, Awka, Nigeria.
Okide S. O.
3
Department of Computer Science,

Nwokoye A. O.C.
2

Department of Physics and Industrial Physics,
Ogu C. D.
4
Department of Electronic and Computer Engineering,

Abstract This paper analyzes the throughput
performance of IEEE 802.11b Wireless Local Area
Network (WLAN) with one access point. The IEEE
802.11b is a wireless protocol standard. In this paper, a
wireless network was established which has one access
point. OPNET IT Guru Simulator (Academic edition) was
used to simulate the entire network. Thus the effects of
varying some network parameters such as the data-rate,
buffer-sizes, and fragmentation threshold were observed
on the throughput performance metric. Several simulation
graphs were obtained and used to analyze the network
performance.
Keywords- Data-rate; buffer size; fragmentation threshold;
throughput; Media Access Control (MAC).
I. INTRODUCTION
A network is a group of devices, such as computers that
are connected to each other for the purpose of sharing
information and resources. Shared resources can include
printers, documents and internet access connections. A
network can be wired or wireless. 802.11b is one of the IEEE
protocol standards for wireless networks. It uses a modulation
technique known as Direct Sequence Spread Spectrum (DSSS)
[1].
Wireless network has some attributes or parameters such
as data-rates, buffer sizes, fragmentation threshold (FTS), etc.
It also has some qualities of service or metrics like the
Throughput, Delay, Media access delay, Data dropped,
Retransmission attempts, etc. But analysis here is only for
throughput.
A. Throughput Analysis
In a wireless network, system throughput is defined as the
fraction of time that a channel is used to successfully transmit
payload bits.
Throughput can be obtained by analyzing the possible
events that may happen on a shared medium in a randomly
chosen slot time [2].
Let P
idle
, P
col
and P
succ
be the probabilities that a randomly
chosen slot corresponds to an idle slot, a collision, and a
successful transmission, respectively.
Moreover, let , T
col,
and T
succ
be the duration of the slot
corresponding to an idle slot, a collision, and a successful
transmission, respectively.
We can obtain the average duration represented by T
avg
,
that a generic slot lasts as follows:
col col succ succ idle avg
T P T P P T

. (1)
Now, the throughput S can be calculated as
S = E[Payload information transmitted in a slot
time] (2)
E[Length of a slot time]
col col succ succ idle
succ
avg
succ
T P T P P
P xE P
T
P xE P
S

] [ ] [

. (3)
Where E[P] is the average payload size (in terms of time
unit), and thus ] [P xE P
succ
is the average amount of payload
information successfully transmitted in a generic slot time.
By dividing the numerator and denominator of equation (2)
by ) (
) succ succ
T P , the throughput can be expressed as follows:
succ succ
idle
succ
col
succ
col
succ
T
x
P
P
T
T
x
P
P
T P E
S
1
/ ] [
(4)
Accordingly, the above analysis applies to both two-way
and four way handshakes transmission. To specifically
compute the throughput for a given handshake, we only need
Vol. 3, No. 7, 2012

138 | P a g e
to specify the corresponding values of T
col
and

T
succ
. Note that
the idle slot time, , is specific to the physical layer [3].
II. NETWORK SIMULATION RESULTS
A network which has one access point and four nodes was
set up as shown below.

Fig. 1: A network with one access point and four nodes
Simulations were carried out using OPNET IT Guru
simulator (Academic edition). The effects of varying three
network parameters on the throughput as a performance metric
were analyzed. The parameters are: the data-rate, buffer size
and fragmentation threshold (FTS).
A. The Data-Rates (Mbps)
This signifies the speed of the nodes connected within a
network. The WLAN model in OPNET IT Guru 9.1 supports
data transfer at 1, 2, 5.5 and 11Mbs. These data rates are
modeled as the speed of transmitters and receivers connected
to WLAN MAC process. Each data rate is associated with a
separate channel stream, from the MAC process to the
transmitter and from the receiver to the MAC process. The
values for different data-rates used for the simulation are
shown in table 1.
TABLE 1: TABLE SHOWING THE DATA-RATES USED FOR
DIFFERENT SCENARIOS
Attributes
(Parameters)
Scenario_1 Scenario_2 Scenario_3
Data-rates 1Mbps 5.5Mbps 11Mbps
Buffer Sizes 12800bits 12800bits 12800bits
Fragmentation
Threshold
None None None

Based on the simulation of the three scenarios for the data-
rates, the graphs in figure 1 were obtained. It is found that
when the data-rate was increased from 1Mbps to 11Mbps, the
throughput increased. This is predictable from the theoretical
view point that as data-rate increases, the number of bits
received increases [4].
Thus based on the graphical result below, it can be said
that when data-rate increases in a network, the throughput
increase; but when the network is overloaded with several
stations, that same throughput decreases, since throughput is
the number of bits successfully transmitted per second.
The 5.5Mbps is good for the network, and that is why the
graphs first rose sharply before they became stable. Stability
of a network is what matters in any network design, and that is
why this simulation was performed using long duration of 300
seconds in order to get a good performance study.
Fig. 2: Throughput study for data-rates of 1Mbps, 5.5Mbps and 11Mbps
B. Buffer Size (bits)
This parameter specifies the maximum length of the higher
layer data arrival buffer. If the buffer limit is reached, data
received from the higher layer are discarded until some
packets are removed from the buffer so as to have some free
spaces to store new packets. The table 2 shows the buffer sizes
used.
TABLE 2: TABLE SHOWING THE BUFFER SIZES USED
Attributes
(Parameters)
Scenario_1 Scenario_2 Scenario_3
Data-rates 11 Mps 11 Mbps 11Mbps
Fragmentation
Threshold
None None None

The graphs of figure 3 show that when the size of the
buffer was increased, the throughput increased. For small size
of buffer, the throughput reduces to zero, meaning that packets
are dropped or discarded because the buffer has no space to
accommodate more packets.

Vol. 3, No. 7, 2012

139 | P a g e
Fig.3: Graphs analyzing throughput for different buffer sizes
C. Fragmentation Threshold (Bytes)
This parameter specifies the value to decide if the MAC
Service Data Unit (MSDU) received from the higher layers
needs to be fragmented before transmission [5]. The number
of fragments to be transmitted is calculated based on the size
of the MSDU and the fragmentation threshold. Table 3 shows
the three scenarios for the simulation study.
The first one is with no fragmentation of incoming packets.
The second one is with a fragmentation of 16 bytes, and the
third one is with a fragmentation of 256 bytes.
TABLE 3: TABLE SHOWING THE FRAGMENTATION
THRESHOLD (FTS) USED FOR DIFFERENT SCENARIOS
Attributes
(Parameters)
Scenario-1 Scenario-2 Scenario-3
Data-rates 11 Mps 11 Mbps 11Mbps
Fragmentation
Threshold
None 16 bytes 256 bytes

The simulation result in figure 4 indicates that proper
fragmentation enhances throughput.

III. CONCLUSION
Having completed this simulation, it is seen that when a
network parameter is tuned to different scenarios, the
throughput performance metric is usually affected. The
following points are to be noted from the results of this
simulation:
1) When the data-rate in a wireless network is increased,
the throughput increases; and packets are delivered more
accurately, hence less requirement for retransmission.
2) For a very small size of buffer, if data-rate is increased,
the throughput reduces to zero, meaning that packets are
dropped or discarded because the buffer has no space to
accommodate more packets.
3) Proper fragmentation enhances throughput. But
fragmentation increases the size of queue and the number of
data dropped in a transmission.
REFERENCES
[1] Okeshi P.N., Fundamentals of Wireless Communication, Global
Publishers Co., Lagos, Nigeria, 2009.
[2] Achinkole S. O.,Computer Networks, Orient Printers and
Communications, Accra, Ghana, 2010.
[3] Makta M. H., Basic Computer Communication, Educational Printing
& Publishers, Accra, 2008.
[4] Ede K. I.,A Guide to Wireless Communication Networks, Excellent
Series Printers, Lagos, Nigeria, 2009.
[5] Borah D. K., Daga A., Lovelace G. R., and Deleon P., Performance
Evaluation of the IEEE 802.11a and b WLAN Physical Layer on the
Maritain Surface, IEEE Aerospace Conference, Canada, 2005.
Fig 4: Throughput result for different FTS

BPM, Agile, and Virtualization Combine to Create
Effective Solutions
Steve Kruba
Northrop Grumman
15010 Conference Center Drive
Chantilly, VA, USA
email: [email protected]
Steven Baynes
Northrop Grumman
Chantilly, VA, USA
Robert Hyer
Northrop Grumman
Chantilly, VA, USA
AbstractThe rate of change in business and government
is accelerating. A number of techniques for addressing
that change have emerged independently to provide for
automated solutions in this environment. This paper will
examine three of the most popular of these technologies
business process management, the agile software devel-
opment movement, and infrastructure virtualization
to expose the commonalities in these approaches and
how, when used together, their combined effect results in
rapidly deployed, more successful solutions.
KeywordsAgile; BPM; business process management;
Rapid Solutions Development; Virtualization; Workow
I. INTRODUCTION
Supporting change in todays dynamic environment re-
quires a strategy and tools that can adapt to unforeseen events.
Such tools have evolved in three key areas independently in
response to this pressure.
Business Process Management (BPM) is both a manage-
ment discipline and a set of technologies aimed at automating
organizations key business processes. Agility is a key feature
of the products that support this market.
Agile Software Development is an approach for creat-
ing custom software and is designed to overcome some of
the short-comings of more traditional approaches such as the
waterfall methodology. It achieves agility through an itera-
tive development approach that focuses on producing work-
ing software as quickly as possible.
Infrastructure Virtualization has expanded from server
virtualization to storage, network, and desktop virtualization.
The emphasis is on providing computing resources transpar-
ently to users and applications so that solutions can be stood
up and modied quickly, and managed more easily and effec-
tively.
The term agile has become popular for describing an im-
portant feature of modern information technology architec-
tures. Agile within the context of each of the three technolo-
gies described in this article has a slightly different connota-
tion, but the underlying principle remains the same. We will
examine these similarities as well as the differences.
We will examine each of these approaches separately
within their agile context and will discuss how in combina-
tion they are becoming increasingly important to creating suc-
cessful solutions. Examples from our experiences with our
Northrop Grumman e.POWER
R 1
BPM product will be used
to illustrate some of these ideas.
II. SOLUTIONS
When acquiring new software tools, organizations typi-
cally begin by examining the feature set of various products
to determine which one is best at satisfying a set of require-
ments. We can lose sight of the fact that what were really
looking for is a solution to a problemnot the tool itself.
This might seem like either an obvious or a nonsensical
statement, depending on how you look at it. Hasnt that al-
ways been the case with software development? you might
ask. But the fact of the matter is that deploying systems has
gotten more complicated in recent years. Quality issues, se-
curity issues, and compatibility issues have been given in-
creased visibility as organizations have been burned by not
appreciating their importance.
The effort involved in deploying nished solutions has be-
come a signicant part of the solution creation process. De-
ployment costs can be comparable to development cost when
using model-driven tools in the BPM space. If those deploy-
ment (and support) costs can be reduced through technologies
such as virtualization, that is signicant.
Combining an effective software development tool with a
powerful methodology like the agile process can be very ben-
ecial. But writing software is probably not the best approach
if there are solutions available that satisfy the requirements
with pre-written software.
And nally, when we step back to thing about solutions,
we begin to focus on the effectiveness of the solutions pro-
duced. Key benets for BPM and the agile methodology are
on how well the solutions produced meet the actual needs of
their stakeholders. Very often with complex systems require-
ments, asking stakeholders to dene what they need is prob-
lematic because they simply do not have the experience to
articulate the details. Both BPM and agile are specically de-
signed to reduce the risk of producing well-constructed solu-
tions that are not effective at satisfying the true requirements;
1
e.POWER has been providing solutions for government and commercial
customers for over fteen years and is a registered trademark of the Northrop
Grumman Corporation.
Vol. 3, No. 7, 2012

140 | P a g e
Figure 1. Model-Driven Architecture
i.e., producing a good solution that is not the right solution.
Its worth noting that a signicant percentage of business
solutions today involve some level of business process au-
tomation. Unlike other middleware components, rather than
being just another tool used in constructing the solution, BPM
software orchestrates the entire solution, even though it typ-
ically has to interact with many other infrastructure compo-
nents (e.g., other applications and services) that satisfy im-
portant solution requirements.
In the next sections, well examine the three technologies
in more detail. We will repeat the key theme of how they
improve the agility of the overall solution in the generic sense
(as opposed to the agile software sense) and hopefully gain
insights into how to view these engagements from an overall
solution perspective.
III. BUSINESS PROCESS MANAGEMENT
Business process management, or BPM, is a management
discipline typically supported by technology.[3] The purpose
of BPM is process improvement. Software tools provide the
technology base under which these goals are achieved. A typ-
ical BPM solution is composed of tasks performed by people
and tasks performed by automated agents.
The BPM market, represented by over 100 vendors, is
one of the fastest growing software segments per Gartner
Dataquest while business process improvement has been
ranked number 1 for the past ve years by CIOs in the an-
nual Gartner CIO survey.[4]
BPMSs such as e.POWER provide design environments
that partition the work so that users with diverse skill-sets can
work independently when developing a solution. Business
users and business analysts play a major role in dening the
business process and associated rules and can use graphical
interfaces for dening these components. Graphics artists,
rather than developers, can be used to design and implement
the layout of user interfaces. Software developers create cus-
tomizations that access legacy data from related applications,
enhance the user interface by extending the out-of-the-box
functionality when necessary, and extend the functionality of
the process engine through exposed interfaces.
A key differentiator of BPMSs is that they are model-
driven. Other toolsets servicing the BPM market and
other business software segments are either parameterized,
conguration-driven, or require writing custom software for
the majority of the functionality. The difference is that
BPMSs provide this functionality out-of-the-box.[5]. Pa-
rameterized or conguration-driven products are similar, but
the connection between the production instantiation of the
model is not as direct as model-driven products and only offer
options that were pre-conceived by the product developers.
Model-driven products offer much greater exibility.
2
2
Gartner has written a lot on this topic.[2] [7] [8]
Vol. 3, No. 7, 2012

141 | P a g e
Figure 2. Our Agile/Iterative Approach
A pictorial representation of model-driven BPM is shown
in Figure 1. A graphical tool is used to dene the busi-
ness process, the results of which are stored in a backend
repositoryoften a relational database. Likewise an applica-
tion designer is used to dene an application that is process
aware. This information is used by the process engine for
enforcing the business rules and routing rules and by applica-
tions servers that drive the user interfaces. This same infor-
mation is also available to end-users as they interact with the
system for managing and performing work.
Note that some model-driven tools are used to dene,
not just the business process, but also applications that are
process-enabledthe right-hand-side of Figure 1. The user-
interfaces needed to actually process work within the business
process are an important part of the solution, and being able
to generate those interfaces through conguration rather than
coding is a very powerful capability.[5]
The combination of capabilities provided by
BPMSs fundamentally changes the way solutions
are constructed in this problem space.
A BPMS product is purpose-built to create BPM solutions.
Within the BPMS framework, the features that are common to
all process problems are built into the product so that archi-
tects using the products simply deploy these pre-built com-
ponents, augmented by customized components needed to
represent the uniqueness of each particular solution. Frame-
works such as service component architectures (SCA) within
a service oriented architecture (SOA) are synergistic with
BPMSs for the customization components. BPMSs could be
viewed as pre-compiled frameworks.
This solutions-orientation is designed for rapid deploy-
ment and increased effectiveness. Being able to construct
these solutions quickly while using these expressive tools to
create more effective solutions is a key benet. Effectiveness
is achieved by using the tools to improve requirements vali-
dation, design, and solution creation, while improving quality
and reducing risk.[5]
As we will see in the next sections, this agility can be am-
plied by other components of the solution infrastructure.
IV. AGILE SOFTWARE DEVELOPMENT
This section is not intended to be a how-to guide on agile
software developmentthere are a number of other excellent
resources. We will, however, include a discussion of agile
principles to see how they relate to agility the other two tech-
nologies, illustrated by our experience in developing our own
product.
Agile software development is rst and foremost an it-
erative methodology for producing quality software. Agile
development is characterized by frequent engagement of all
stakeholders, including customers, developers, quality per-
sonnel, and management. Agile software development em-
ploys multiple development cycles to produce robust, testable
feature sets, followed by integrated system testing.
We especially like the following quotation on the goal of
agile software development.[6] This emphasizes the fact that
all projects are time-bound and helps to avoid the problem of
scope-creep.
At the end of a project we would rather have 80%
of the (most important) features 100% done, than
100% of all features 80% done.
A. e.POWER Agile Development
To provide insight into agile development, we thought we
would describe our rst-hand experience using agile develop-
ment for product releases of the e.POWER product over the
past six years.
Vol. 3, No. 7, 2012

142 | P a g e
Figure 3. Release Timeline
Our approach consists of three phases for delivering a
quality product: planning, multiple development cycles, and
stabilization. Figure 2 illustrates this approach.
The planning phase begins with dening a vision state-
ment and is followed by developing a list of features. Prelimi-
nary requirements are then collected and analyzed after which
we can focus on the most important requirements rst. Con-
sistent with the agile manifesto[1], planning requires frequent
involvement with our stakeholders. We kick off our planning
sessions with a meeting with our advisory forum membership
to be certain that we collect their high-level input as well as
their detailed requirements.
After completing the planning phase, we are in a position
to commit to what we will do in the iterative cycles. Each
cycle is a mini-waterfall model, but much shorter, where we
nalize requirements, perform analysis, design the software,
build it, and test it. Every iteration or cycle contains a slice
of the product, delivering small pieces of complete, working
functionality.
The cycles provide opportunities for stakeholders that are
not already part of the development cycles to review com-
pleted functionality. Since each cycle produces working soft-
ware, demonstrations of that functionality are quite natural
and simple to produce. These reviews also provide the op-
portunity to reprioritize the features and requirements list be-
tween cycles, since everyone is now more engaged and aware
of the evolving solution.
Figure 3 provides an example of a release timeline of a
past release of the e.POWER product. We have used this tem-
plate for several years to manage the process. This one page
summary of each release has been very effective at providing
management, developers, and testers with visibility into the
process and managing to the schedule.
After completion of the nal cycle, we enter the last, or
stabilization phase. At this point we perform complete sys-
tem regression testing. Since we support multiple platforms
for each release, we do platform testing during this phase.
The conguration control board reviews the nal require-
ments against our solution and documentation can now be
nalized. Our quality manager is responsible for leading our
nal total product readiness
3
process, which authorizes the
product for commercial release.
B. Relationship of Agile to BPM Solution Creation
Our experiences with agile development of our software
product may be interesting, but how does that relate to cus-
tomers creating BPM solutions? They are related in two ways:
3
Software products are comprised of much more than just software. Total
product readiness is a term that we use to include the full breadth of ca-
pabilities that must be delivered for a product release, including marketing
collaterals, release announcement materials, installation scripts, on-line help,
documentation, training materials, etc.
Vol. 3, No. 7, 2012

143 | P a g e
BPM solutions typically involve writing some custom
software. To the extent this is minimal, the more robust
and effective the solution can be.[5] But when signicant
customization is required, an iterative agile approach can
be benecial.
Perhaps more importantly, the methodology used in
writing software for agile software development is very
similar to the iterative approach that we have used over
the years in creating e.POWER solutions, including the
model-driven aspects of the solution. The difference re-
lates to code creation vs. solution creation. For agile
BPM we are able to reduce the need to write custom
software, replacing it with model manipulationa non-
programming effort.
V. VIRTUALIZATION
Virtualization is a much over-hyped technology, but not
without some justication. Data centers world-wide are mov-
ing to virtualization to simplify operations, reduce hardware
costs, reduce cooling and energy costs, and expedite solution
deployments.
Although virtualization gained recognition initially with
data center servers and has been in common usage for many
years, virtualization has experienced much increased popular-
ity recently in the areas of storage, networking, and desktops.
The following sections provide a high-level summary of
the important subtopics on virtualization so that we can relate
virtualization to our overall theme.
A. Hardware
Hardware virtualization abstracts the physical computer
hardware from the operating system, allowing applications
not originally designed for that combination to run on the
new, virtualized platform.
Improved management features greatly reduce the man-
power needed to congure, secure, backup, and operate virtu-
alized servers than their physical counterparts. Furthermore,
server virtualization solutions form the basis for cloud com-
puting, which can be viewed as virtualization on steroids. Pri-
vate clouds are virtualization platforms with even richer man-
agement features. The importance to our discussion is that
adding new solutions to those environments becomes even
easier yet, reducing operating costs.
Key aspects of virtual servers include higher availability
(virtualization hardware can bring up ofine copies of the
server automatically), faster provisioning of new servers, au-
tomatic provisioning of new servers based on templates and
security access rights, and much simpler hardware upgrades
since the virtual machine is independent of the hardware.
These are all aspects of agility that are important to our topic.
B. Storage
In a manner similar to hardware virtualization, storage vir-
tualization abstracts the physical computer storage from the
logical storage referenced in applications through the oper-
ating system. The original impetus for storage virtualization
may have been hardware independencethe desire to be able
to swap out one vendors disk drives for another vendors
when the old technology became obsolete. In general, fea-
tures such as vendor independence, over-provisioning, repli-
cation, pooling, improved utilization, snapshots, etc., so sig-
nicantly reduce operating costs that they typically offset any
concern for a slight reduction in performance.
C. Network
Network virtualization uses software so that reconguring
the physical network is not necessary to implement opera-
tional changes. The major benet of network virtualization
is simplied management of the network infrastructure. As
new solutions are deployed, as new hardware is elded, and
as work-patterns evolve to support changing business require-
ments, network administrators can modify network congu-
rations more easily than in the past.
D. Desktop
Desktop or client virtualization breaks the connection be-
tween users and their physical desktop machines. Multiple
users share instances of servers for their desktop computing
needs. Users require physical devices such as keyboards and
monitors to interact with their virtual desktop servers, but
these devices can be components of a system running a differ-
ent operating system, or a purpose-built device limited to user
interaction. In either case, an individual user can be granted
access to one or more virtual desktops for specic work-tasks.
For some business cases, desktop virtualization offers sig-
nicant benets, but is not necessarily optimal for all use
cases. When appropriate, desktop virtualization offers sim-
pler and faster provisioning of new desktops, simplied desk-
top management in areas such as backups and patch manage-
ment, better security, and improved reliability,
E. Performance Issues
Adiscussion of virtualization would not be complete with-
out consideration of the performance implications. Virtual-
ization provides an extra layer between applications and the
physical hardware resources that has an associated cost. Our
discussions will center around server virtualization.
For server virtualization, the hypervisor or Virtual Ma-
chine Monitor allows multiple operating systems to run con-
currently on the machine hardware.
Type 1 or bare-metal hypervisors run directly on the host
hardware while the virtual operating system(s) run on top
of them. These tend to be more efcient than Type 2 hy-
pervisors. Type 1 hypervisors include Microsoft Hyper-V,
VMWare ESX and ESXi, and Citrix Xen Server.
Type 2 hypervisors run on top of a host operating system
such as Microsoft Windows, adding an additional level be-
tween applications and the hardware. Type 2 hypervisors in-
clude VMWare workstation, VMWare server, and Microsoft
Virtual Server.
The Virtual Insanity website has a useful graphic that illus-
Vol. 3, No. 7, 2012

144 | P a g e
Figure 4. VMWare ESX Performance Improvements
trates the improvement in server virtualization performance
in the VMWare ESX producta Type 1 bare-metal hypervi-
sor. As you can see from Figure 4, dramatic improvements
have been made, reducing overhead from 3060% in ESX 2
to 210% in ESX 4.[9] Note corresponding improvements in
network throughput and disk input/output (IOPS) as well. The
point is that for mission critical systems, you need to bench-
mark your virtualized applications to insure adequate respon-
siveness and choose carefully to meet your needs.
Some Type 2 hypervisors are free and can be useful for
some of your needs. For development and support of the
e.POWER product, our primary need is to provide support
for many old releases of the product, but since the activity
levels are low, performance is not a critical issue. For pro-
duction customers, we provide target memory and CPUs re-
quired to support designated workloads, and those targets are
somewhat diluted when deployed in virtual environments.
Depending on those activity levels, additional hardware may
be needed, the cost of which may well be recouped in re-
duced operating and support costs, especially with the mini-
mal overhead of the latest releases of Type 1 hypervisors.
On virtualized hardware, a key consideration is whether
the storage is housed on internal drives or external storage.
On platforms such as VMWare, complete lesystems are en-
capsulated when stored locally and are signicantly less re-
sponsive than external storage such as SAN storage or NFS.
Server virtualization platforms have special drivers for exter-
nal storage that overcome this limitation.
F. Our Experience
We thought it would be valuable to include information on
our experience with virtualization, primarily with hardware
virtualization. The following chart summarizes the benets
we have seen from this transition over the past several years.
Ability to setup virtual machines quickly and move them
or turn them on or off as needed
Reduced hardware investment and on-going power costs
More efcient use of hardware resources
Reduced labor expenses in moving virtual machines to
new hardwareno operating system reinstallation nec-
essary
Features such as snapshot facilitates testing of instal-
lation scripts and conguration
Entire virtual machines are backed up as a single le
Virtual machines are indistinguishable from physical
machines from an end-user perspective
VI. COMMON THEMES
So what are some of the common themes we see in the
three technologies presented above? At a high level, they
center around the concept of agility: providing the ability to
create solutions that are both quick to produce and adaptable
to needs that evolve over time. Speed is an important compo-
nent but equally important is effectiveness, emphasizing the
overall development process from needs assessment through
deployment. The three approaches highlighted in this article
are not the only ones that apply these principles, but are three
of the most visible today and touch on all aspects of solutions
in the business context. An organization that emphasized at
least these three would be well served.
Vol. 3, No. 7, 2012

145 | P a g e
Agility is about embracing change, knowing that user re-
quirements will evolve as the emerging solution provides
greater visibility into the nal product. BPMS and agile
toolsets make it possible to iterate towards a solution because
of the exibility they introduce into the creation process. We
are often at a loss to express what the ultimate solution needs
to look like, but we can more readily recognize it when we
see it.
A high level summary of principles that are shared among
these three approaches are as follows.
People are the key to solutions. Technology has reached
a level of renement where we no longer should be opti-
mizing bits and bytes, but optimizing people, including
process participants, architects and developers, and sup-
port staff.
Engage stakeholders continuously throughout the solu-
tion development process. Continuous feedback with it-
erative evolution of the solution fundamentally improves
the creation process.
Working software is the best visualization tool for work-
ing software. Modern software has become increasingly
expressive, for which there is no substitute.
Task the right people for each aspect of solution creation
based on domain knowledge. The tools now allow busi-
ness people to participate in this process, allowing in-
formation technology professionals to focus on the IT-
aspects of solutions.
Modularity is a theme seen in all three of these tech-
nologies, allowing participants to easily conceptualize
the current component of interest.
Virtualization makes infrastructure less of an impedi-
ment to productivity. People can more easily gain access
to infrastructure resources in creating and managing so-
lutions.
VII. CONCLUSIONS
Creating successful automated solutions is challenging in
todays highly competitive environment. Solutions must be
conceived and implemented quickly in a manner that allows
them to adapt as needs change. For a large class of busi-
ness problems, this requires the capabilities of a business pro-
cess management suite. BPMSs differentiate themselves by
their rapid solutions creation capabilities achieved through
a model-driven architecture. Since signicant BPM projects
require some custom software for critical parts of the solu-
tion, agile principles are very compatible with BPMSs and
Northrop Grummans e.POWER product software is devel-
oped using an agile software development methodology.
Creating or evolving a solution rapidly is of little conse-
quence if it cannot be elded in a like manner and this is
where virtualization becomes important. The underlying un-
derpinnings of agility in each of these aspects of solution cre-
ation work together to insure that solutions are effective and
deployed in a timeframe that meets the needs of their business
customers.
REFERENCES
[1] http://agilemanifesto.org/principles.html, 2001.
[2] M. Blechar. The changing concept of model-driven ap-
proaches. Gartner, pages 18, August 2009.
[3] Michelle Cantara. BPM research index: Business pro-
cess management technologies. Gartner, pages 111,
September 2009.
[4] Gartner EXP worldwide survey of nearly 1,600 CIOs
shows IT budgets in 2010 to be at 2005 levels.
http://www.gartner.com/ it/page.jsp ?id=1283413, Jan-
uary 2010.
[5] S. Kruba. Signicance of rapid solutions development to
business process management. International Journal of
Computer Science and Information Security, 8(4):299
303, 2010.
[6] Malotaux. Evolutionary project management methods.
http://www.malotaux.nl/doc.php?id=1, August 2007.
[7] D. Plummer and J. Hill. Three types of model-driven
composition: Whats lost in translation? Gartner, pages
110, August 2008.
[8] D. Plummer and J. Hill. Composition and BPM will
change the game for business system design. Gartner,
pages 121, December 2009.
[9] Life as a VMWare virtual machine, 2010.
http://www.virtualinsanity.com /index.php /2010/08/17
/life-as-a-vmware-virtual-machine/.
AUTHOR PROFILES
Steve Kruba is chief technologist for Northrop Grummans
process-oriented commercial software products, including
e.POWER, and a Northrop Grumman Technical Fellow.
Steve has 42 years of experience developing software and so-
lutions for customers. He holds a Bachelor of Arts in Mathe-
matics and a Master of Science in Management Sciences from
the Johns Hopkins University.
Steve Baynes is the department manager for the e.POWER
product development team with extensive agile experience.
He is a certied ScrumMaster, member of the Agile Alliance
organization and speaks often on Agile development. As the
manager for the e.POWER product, Steve works with busi-
ness development, project implementation teams, and cus-
tomers to continually improve the e.POWER product from
both a feature and quality perspective.
Bob Hyer is chief architect for the e.POWER product devel-
opment team. Bob has over 30 years of experience develop-
ing software solutions and software products for government
and commercial customers. He has a Bachelors of Science
in Business Management from Virginia Tech and a Master of
Science in Technology Management from American Univer-
sity.
Vol. 3, No. 7, 2012

146 | P a g e

IJACSA - Volume 3 No. 7, July 2012

Uploaded by

Copyright:

Available Formats

IJACSA - Volume 3 No. 7, July 2012

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

IJACSA - Volume 3 No. 7, July 2012

Uploaded by

Copyright:

Available Formats

THE SCIENCE AND INFORMATION ORGANIZATION

www. t hes ai . or g | i nf o@t hes ai . or g

={1, 4, 12, 52, 365), T

={0.005, 0.01, 0.02, 0.03, 0.04, 0.05}, where the

are payment frequencies, T

are maturities, and

are coefficients. We derive the parameters for a bond object

. This is also called Euclidean norm or Norm-2

is the relative importance of the i

Where K is the number of nominal fields. P(

is the term frequency of the i

is the document frequency of the i

. The delay between two iterations

connection, or the hop node

} . The distances from volunteers to disabled

set in I contains iD. The problem can then be stated

of simulation, distances from

can make three bids for

} based on initial distance. The

, which has the most minimum

You might also like