Beruflich Dokumente
Kultur Dokumente
architecture description contains sufficient details, an composition Ci as well as a set of composition constraints Csi,
implementation from the resulting SAM architecture e.g. Ci = {Cmi, Cni, Csi}. In addition, each component or
description to Java code can be automatically generated connector is composed of two elements, a behavioral model
from our existing SAM translator tool. and a property specification, e.g. Cij = (Bij, P ij). Each
The paper proceeds as follows: Section 2 gives some behavioral model is described by a Petri net, while a property
background information. Section 3 describes our approach specification by a temporal logical formula. The atomic
to mapping UML architecture description to SAM model; proposition used in the first order temporal logic formula is
Section 4 gives an example of the approach applied an the ports of each component or connector. Thus each
embedded system example and evaluates the approach; behavioral model can be connected with its property
Section 5 describes related work and Section 6 concludes. specification. A component Cmi or a connector Cni can be
refined to a low level composition Cl by a mapping relation h,
e.g. h(Cmi ) or h(Cmi ) = Cl. SAM is suitable to describe large
2. Preliminaries
scale systems’ description.
In this section, we provide a brief introduction to software SAM gives the flexibility to choose any variant of Petri nets
architecture documentation, UML, and SAM. and temporal logics to specify behavior and constraints
according to system characteristics. In our case, Predicate
2.1 Software Architecture Viewpoints Transition (PrT) net [11] and linear temporal logic (LTL)
Although, there is not a universally accepted way in are chosen.
documenting software architecture design. The component
and connector (C&C) view, showing the dynamic behavioral 3. Our Approach
aspect of a software system, proposed in [35] is no doubt an
essential one, which has been the target used in the Our approach takes a software architecture description based
development of many software architecture description on the C&C view and documented using the UML, and
languages. The C&C view is also included in the work of produces a formal software architecture description in SAM.
the Software Engineering Institute (SEI) [30], where the More specifically, in the UML software architecture
authors presented three view types - module, component and description:
connector, and allocation view in documenting software a) A class diagram is used to model the overall
architecture design; and provided guidelines of how to use structure of a software architecture,
UML to document these architecture view types. With b) State chart diagrams are used to define the behavior
regard to the representation of the C&C view, three of individual components and connectors, and
strategies are demonstrated: using component types as c) OCL is used to specify architecture level
classes, subsystems, or real-time profiles. Since the behavior constraints.
of each class and object is described by the state charts The above UML notations are mapped to SAM entities as
diagram, the behavior of the system represented by the class follows:
diagram can be a group of state charts diagram with a) The class diagram is mapped to an overall
interactions. hierarchical SAM structure,
b) State chart diagrams are mapped to PrT nets, and
2.1 SAM – Software Architecture Model c) OCL expressions are mapped to temporal logic
SAM is an architectural description model based on Petri formulae.
nets [29], which are well-suited for modeling distributed The resulting SAM model can be analyzed with various
systems. SAM [15] has dual formalisms underlying – Petri existing formal analysis techniques including model
nets and Temporal logic. Petri nets are used to describe checking.
behavioral models of components and connectors while Furthermore, to improve the productivity and quality, we
temporal logic is used to specify system properties of provide an automated realization of the resulting SAM
components and connectors. model. We have designed and implemented a tool, named
SAM (Software Architecture Model) is hierarchically SAM parser, with the PrT-XML and temporal logic-XML
defined as follows. A set of compositions C = {C1, C2, …, transformation. Figure 1 shows the overall structure of the
Ck} represents different design levels or subsystems. A set of approach.
component Cmi and connectors Cni are specified within each
Temporal logic and first order logic formulae are intervention from the driver, by actuating the throttle-
transformed into automata by a logic engine Maude ([5]). accelerator pedal linkage. A modern automotive cruise
These translated automata will feed into our runtime control is a control loop that takes over control of the
checker generator to produce monitors for different throttle, which is normally controlled by the driver with
formulae. the gas pedal, and holds the vehicle speed at a set value.
We assume an automatic transmission vehicle. When
Runtime Checker Generation: Runtime checkers are turned on by the driver, a cruise-control system (CCS)
generated by breaking the temporal logic formula into automatically maintains the speed of a car over varying
subformulae and creating a matrix for the formula [33]. terrain. The CCS can be turned on by pressing Start
In order to generate monitoring codes for properties button, and enabled by pressing SetSpeed button. Resume
(linear temporal formulae), a logic server, Maude [5] in button will enable the CCS at the last maintained speed
our case, is necessary. Maude, acting as the main when the brake is released. The cruise control function is
algorithm generator in the framework, constructs an disabled when the brake or accelerator pedal is pressed.
e_cient dynamic programming algorithm (i.e. monitoring Pressed once Resume button can increase the speed with
code) from any LTL formula [33]. The generated 1mph and the SetSpeed button can decrease the speed
algorithm can check if the corresponding LTL formula is with 1mph when the cruise control function is enabled.
satisfied over an event trace. The cruise control system should be automatically
disabled when the speed is below 25mph and above
4. An Application of the MDA Approach 90mph. For the space limit, we cannot show the UML
diagram and generated code. Each place or port must
We present an instance of our approach in a statechart carry its state information which is also ignored in the
diagram, which is one of the state machine in the UML tables. Finally, we have to point out that the guard
models, through a case study. The case study deals with a function for a transition has to be added some more
simplified cruise control system adapted from [13]. In restrictions for the evaluations. For instance, if there is a
this section we will first introduce the cruise control token “void” in the input place of a transition, it means
system and a state chart diagram for the cruise controller. the previous action does not have return value, we have to
Then we present some properties of this example. Finally, justify that field to evaluate the firing condition of the
the runtime verification results are discussed. transition. This means that automatically mapping a
4.1 Cruise Control System and UML Documentation guard condition in a state chart diagram to a guard
function is not sufficient in some cases.
The purpose of a cruise control system is to accurately
maintain the driver’s desired set speed, without
All properties are true if the conditions and guard are
4.2 Experiment Results and Discussion
satisfied. We also check some conditions that is not
The time of generated code with monitors is 4.3s. We suitable for the method, such as different parameters
checked 5 properties covering 5 components and 4 feeding for the method that makes the guard is not
connectors. satisfied, in that case the formula is evaluated as false.
Since the events and actions are defined by the methods The Results are consistent with what we expected from
which are specified in the corresponding classes, each the state chart diagram. Finally, we also find a mapping
fired transition represents a method is operated under mistake in the component Accelerating and component
some guard condition. The properties specified in the Decelerating when we check properties relative to them.
OCL expressions for the state chart diagram is mapped to We modified the mapped SAM composition according to
temporal formulae and further used to generate monitors. the checking results.
analysis of generic UML object-oriented designs [7, 34, 12,
5. Related Work 24, 4, 26, 35], their results may not be ready applicable to
UML architecture level designs.
Our MDA approach and the presented framework integrate Code Generation from the UML Description and
two aspects: software architecture with UML design Verification in the Implementation:
notation and its extension. Moreover, our approach also There is a significant amount of research that considers
provided a runtime validation and verification technique by mappings from UML to other (mostly formal) modeling
using SAM Parser. The related works are discussed in the techniques to validate UML models (e.g. using B [22], CSP
following. [8], SPIN [23], PVS [2], Petri Nets [27, 38], Z/Eves [3] and
Formal Modeling and Analysis of Architecture the work with Object-Z [19, 20, 18, 39]). These works focus
Descriptions in UML: Our work has been influenced by a on the mapping from UML diagrams to formal methods for
large body of research and practical experience. In the model checking or some specification language for the
interest of brevity, we only compare it to the most relevant model checkers. Moreover, less of them address the code
approaches. The work reported in this paper relates to works generation and verification in the implementation level.
that focus on specifying structural and possibly behavioral Property Specification – OCL Expression and Temporal
aspects of a software system using UML. The representative Logic: Various methodologies proposed to deal with the
UML architecture description examples are provided by property specification of object-oriented systems. There are
Kruchten [21], Hofmeister et al. [16] and Clements et al. two main streams to cooperate OCL with temporal logics,
[31]. Although several researchers have explored the formal one is extending OCL with temporal notations ([31, 9, 37]
6 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
etc.), another is add object concepts into temporal logics [9] S. Flake and W. Mueller. An OCL extension for real-
([6]). The work in both streams increases the complexity of time constraints. In Object Modeling with the OCL,
the extended language and obstacles of the usage. pages 150–171, 2002.
[10] Y. Fu, Z. Dong, and X. He. A Methodology of
6. Conclusion and Future Works Automated Realization of a Software Architecture
Design. In Proceedings of The Seventeenth
In this paper, we have presented an integrated approach for International Conference on Software Engineering and
transferring an architecture description represented in UML Knowledge Engineering (SEKE2005), 2005.
to a formal architecture model represented in SAM, which [11] H. J. Genrich. Predicate/Transition Nets. Lecture Notes
not only supports design level analysis but also automated in Computer Science, 254, 1987.
code generation with run-time verification capability. The [12] M. Gogolla and F. P. Presicce. State Diagrams in UML:
specific details of the approach, outlined in the algorithms A Formal Semantics using Graph Transformations. In
in Section 3, are likely to evolve as our research on the Proceedings of International Conference of Software
relationship between UML and software architectures Engineering, Workshop on Precise Semantics of
deepens; however, we believe that the approach is flexible Modeling Techniques, pages 55–72, 1998.
and general enough to accommodate needed new changes. [13] H. Gomaa. Designing Concurrent, Distributed, and
As pointed out by Medvidovic in the work [27] ensuring Real-Time Applications with UML. Addison-Wesley
system properties at the level of architecture is of little value Professional, 2000.
unless it can also be ensured that those properties will be [14] K. Havelund and G. Rosu. An overview of the runtime
preserved in the resulting implementation. This reflects the verification tool java pathexplorer. Journal of Formal
importance of the code generation and runtime verification Methods in System Design, 2004.
of system properties in the implementation. Our automated [15] X. He and Y. Deng. A Framework for Specifying and
code generation and run-time verification approach nicely Verifying Software Architecture Specifications in SAM.
addresses the above research issue. volume 45 of The Computer Journal, pages 111–128,
2002.
Acknowledgements We appreciate for all reviewers to read [16] C. Hofmeister, R. L. Nord, and D. Soni. Describing
this paper. This work is supported by Title III under grant Software Architecture with UML. In Proceedings of the
PO31B085057-08. TC2 1st Working IFIP Conference on Software
Architecture (WICSA1), pages 145 – 160, 1999.
[17] M. Kim, S. Kannan, I. Lee, and O. Sokolsky. Java-
References MaC: a Run-time Assurance Tool for Java. In
[1] Uml 2.0 specification. http://www.omg.org/ Proceedings of RV’01: First International Workshop
technology/documents/formal/uml.htm. on Runtime Verification, Paris, France, Electronic
[2] Enhancing Structured Review with Model-Based Notes in Theoretical Computer Science. Elsevier
Verification. IEEE Transaction on Software Science, 2001.
Engineering, 30(11):736–753, 2004. Member-Issa [18] S.-K. Kim, D. Burger, and D. Carrington. An mda
Traore and Member-Demissie B. Aredo. approach towards integrating formal and informal
[3] N. Am´alio, S. Stepney, and F. Polack. Formal proof modeling languages. In FM 2005: Formal Methods,
from uml models. In ICFEM’04, volume 3308 of International Symposium of Formal Methods Europe,,
Lecture Notes in Computer Science, pages 418–433, volume 3582 of Lecture Notes in Computer Science,
2004. pages 448–464, 2005.
[4] D. B. Aredo. Semantics of UML statecharts in PVS. In [19] S.-K. Kim and D. Carrington. Formalizing the UML
Proceeding of 12th Nordic Workshop on Programming Class Diagrams Using Object-Z. In UML’99: The
Theory, Bergen, Norway, 2000. Unified Modeling Language - Beyond the Standard,
[5] M. Clavel, F. J. Dur´an, S. Eker, P. Lincoln, N. Mart´ı- Second International Conference, volume 1723 of
Oliet, J. Meseguer, and J. F. Quesada. Maude: Lecture Notes in Computer Science, 1999.
Specification and Programming in Rewriting Logic. [20] S.-K. Kim and D. Carrington. A Formal Mapping
http://maude.csl.sri.com/papers, March 1999. between UML Models and Object-Z Specifications.
[6] D. Distefano, J.-P. Katoen, and A. Rensink. On a Lecture Notes in Computer Science, volume 1878,
Temporal Logic for Object-Based Systems. In S. F. pages 2–21, 2000.
Smith and C. L. Talcott, editors, Formal Methods for [21] P. Kruchten. The 4+1 view model of architecture. IEEE
Open Object-Based Distributed Systems IV - Proc. Software, 12(6):42–50, 1995.
FMOODS’2000, Stanford, California, USA, September [22] K. Lano, D. Clark, and K. Androutsopoulos. UML to B:
2000. Kluwer Academic Publishers. Formal Verification of Object-Oriented Models. volume
[7] Z. Dong and X. He. Integrating UML State-chart and 2999 of Lecture Notes in Computer Science, pages 187–
Collaboration Diagrams Using Hierarchical Predicate 206,2004.
Transition Nets. In GI Lecture Notes in Informatics, [23] D. Latella, I. Majzik, and M. Massink. Automatic
2001. Verification of a Behavioural Subset of UML Statechart
[8] G. Engels, R. Heckel, and J. M. K¨uster. Rule-based Diagrams Using the SPIN Model-checker. Formal
specification of behavioral consistency based on the Aspects of Computing, 11(6):637 – 664, 1999.
UML meta-model. volume 2185, pages 272–284, 2001.
(IJCNS) International Journal of Computer and Network Security, 7
Vol. 2, No. 5, May 2010
[24] D. Latella, I. Majzik, and M. Massink. Towards a [38] Jiexin Lian, Zhaoxia Hu, Sol M. Shatz. Simulation-
Formal Operational Semantics of UML Statechart based analysis of UML statechart diagrams: methods
Diagrams. In Proceedings of the 3rd IFIP International and case studies. Software Quality Journal. 16(1),
Conference on Formal Methods for Open Object-based March, 2008. ISBN: 0963-9314. Springer Netherlands.
Distributed Systems, pages 331–347, February 1999. [39] Rafael M. Borges and Alexandre C. Mota. Integrating
[25] S. W. Lewandowski and X. He. Generating Code for UML and Formal Methods. Electronic Notes in
Hierarchical Predicate Transition Net Based Designs. In Theoretical Computer Science (ENTCS). Volume 184.
Proceedings of the 12th International Conference on Page 97-112. July 2007. Elsevier Science Publishers.
Software Engineering & Knowledge Engineering, pages
15–22, Chicago, U.S.A., July 2000. Authors Profile
[26] J. Lilius and I. P. Paltor. The Semantics of UML State
Machines. Technical Report 273, Turku Centre for Yujian Fu received the B.S. and M.S. degrees in Electrical
Computer Science, 1999. Engineering from Tianjin Normal University and Nankai University
[27] N. Medvidovic, D. S. Rosenblum, and D. F. Redmiles. in 1992 and 1997, respectively. In 2007, she received her Ph.D.
Modeling Software Architectures in the Unified degree in computer science from Florida International University.
Modeling Language. ACM Transactions on Software She joined the faculty of Department of Computer Science at the
Engineering and Methodology, 11(1):2–57, January Alabama A&M University in the same year. Dr. Yujian Fu
conducts research in the software verification, software quality
2002. assurance, runtime verification, and formal methods. Dr. Yujian Fu
[28] H. Motameni. Mapping to Convert Activity Diagram in also actively serves as reviewers of several top journals and
Fuzzy UML to Fuzzy Petri Net. World Applied Sciences prestigious conferences. She continuously committed as a member
Journal, 3(3): 514 – 521, 2008. ISBN: 1818-4952. of IEEE, ACM and ASEE.
[29] T. Murata. Petri Nets: Properties, Analysis and
Applications. Proceedings of the IEEE, 77(4):541–580, Zhijiang Dong received the B.S. and M.S. degrees in Huazhong
1989. Tech University, Ph.D. degree in computer science from Florida
[30] J. S. e. a. Paul Clements, Len Bass. Documenting International University. He currently is assistant professor at
Software Architectures: Views and Beyond. Addison- Middle Tennessee University. Dr. Dong’s research is mainly in the
software engineering. Dr. Dong also actively serves as reviewers of
Wesley, January 2003.
several top journals and conferences. He continuously committed as
[31] S. Ramakrishnan and J. McGregor. Extending OCL to a member of IEEE, ACM.
Support Temporal Operators. In 21st International
Conference on Software Engineering (ICSE 99), Xudong He is a professor of school of computing and
Workshop on Testing Distributed Component-Based information science and director of center for advanced and
Systems, May 1999. distributed system engineering at Florida International University.
[32] J. E. Robbins, N. Medvidovic, D. F. Redmiles, and D. Dr. He’s research are software engineering, formal verification and
S. Rosenblum. Integrating architecture description specification. Dr. He currently has over one hundred publications in
languages with a standard design method. In ICSE ’98: prestigious journals and conferences.
Proceedings of the 20th international conference on
Sha Li is an associate professor at department of curriculum,
Software engineering, pages 209–218, Washington, teaching and educational leadership, school of education of
DC, USA, 1998. IEEE Computer Society. Alabama A&M University. Dr. Sha Li received his doctorial
[33] G. Rosu and K. Havelund. Rewriting-Based Techniques degree of educational technology from Oklahoma State University,
for Runtime Verification. Journal of Automated 2001. Sha Li' research interests include distance education,
Software Engineering, 2004. instructional technology, instructional design and multimedia for
[34] J. Saldhana and S. M. Shatz. UML Diagrams to Object learning.
Petri Net Models: An Approach for Modeling and
Analysis. In Proceedings of the International Phil Bording is an associate professor and chair of department of
computer science at Alabama A&M University. Dr. Phil Bording
Conference on Software Engineering and Knowledge
received his Ph.D. degree in computer science from University of
Engineering, pages 103–110, 2000. Tulsa in 1995 and M.S. degree from University of Alabama at
[35] T. Sch¨afer, A. Knapp, and S. Merz. Model Checking Huntsville in 1984. Dr. Bording’s research area is parallel
UML State Machines and Collaborations. Electronic computing.
Notes in Theoretical Computer Science, 55(3):1–13,
2001.
[36] M. Shaw and D. Garlan. Software Architecture:
Perspectives on an Emerging Discipline. Prentice Hall,
1996.
[37] P. Ziemann and M. Gogolla. An Extension of OCL with
Temporal Logic. In Critical Systems Development with
UML – Proceedings of the UML’02 workshop, pages
53–62, TUM, Institut fur Informatik, TUM-I0208,
September 2002.
8 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
4
College of Engineering and Technology, IFTM Campus,
Lodhipur Rajput, Moradabad, UP, INDIA
cs_himanshu@rediffmail.com
Abstract: For improving accuracy and increasing efficiency of open and very challenging dilemma in real application. The
classifier, there are available many effective techniques. One of area of face recognition has focused on detecting individual
them combining multiple classifier technique is used for this feature such as the eyes, nose, mouth, lips and head outline
purpose. In this paper I present a novel approach to a and defining face model by the position, size, perimeter and
combining algorithm designed to improve the accuracy of relationship among features. Bledsoe’s [3] and Kanade’s [4]
principle component classifier. This novel approach combines recognition based on these parameter. A Mathew A. Turk
multiple PCA classifiers, each of using a subset of feature. In
and Alex P. Pentland [1] track a subject’s and then
contrast other combining algorithms usually manipulate the
training pattern.
recognizes the person by comparing characteristics of the
face to those of known individuals. Face images are
Keywords: PCA classifier, Eigenface, Eigenvector, Voting, projected onto a feature space that best encodes the variation
Covariance matrix, new projected data, face recognition. among known face images. The face space is defined by the
“eigenface” which are the eigenvectors of the set of faces;
1. Introduction they do not necessarily correspond to isolated feature such as
eyes, ear and nose. The framework provides the ability to
The PCA classifier is one of the oldest and simplest methods learn to recognize new faces in an unsupervised manner.
for classification. PCA involves a mathematical course of In mathematical terms, we want to find the principle
action that transform a number of possible correlated component of the distribution of faces or the eigenvectors of
variable into a smaller no of uncorrelated variable called the covariance matrix of set of face images. These
principle component. PCA is mathematically defined as an eigenvectors can be thought of as a set of feature which
orthogonal linear transform that transform the data to a new together characterizes the variation between face images.
coordinate system such that greatest variance by any Each image location contributes more or less to each
projection of the statistics comes to lie on the first eigenvector so that we can display the eigenvector as a sort
coordinate. The second greatest variance on the second of ghostly face which we call an eigenfaces.
coordinate and so on [10].
There are many approaches available to combining the
1.1. Method of Pronouncement PCA. classifier such as Stephen O.Bay, “Nearest Neighbor
Step 1: Acquire some data Classification from Multiple Feature Subset” [7] and Bing-
Step 2: Subtract the mean Yu-Sun- Xiao-Ming Zhang and Rujing Wans, “Training
Step 3: Compute the covariance matrix SVMs for Multiple Feature Classifier Problem” [5] but there
Step4: Compute the eigenvectors and eigenvalues of the in no any approach available to combining the base
covariance Matrix classifiers. Actually the combination of classifier can be
Step 5: Choosing components and forming a feature vector implemented at two levels, feature level and decision level.
Step 6: Deriving the new data set. Xiaoguang Lu, Yunhong Wang, Anil K. Jain, “combining
classifier for face recognition”[4] provide frameworks to
1.2. Related work combine different Classifier on feature based .We are using
Image classification is a thorny task because images are feature level combination and want to present combination
multidimensional. There are many classifiers although there of classifier which works on low level feature in this
are a number of face recognition algorithms which works approach.
well in constrained situation; face recognition is still an
(IJCNS) International Journal of Computer and Network Security, 9
Vol. 2, No. 5, May 2010
The algorithm for PCA classification from multiple feature 10. Calculate the feature vector of test set and also
subsets is simple and can be treated as: calculate new projected data buy using PCA
transformation matrix.
Using voting, combining the output from multiple PCA 11. Compute the similarity between subset and test set
classifiers each having access each subset of feature. by using voting method.
We select the subset of features by sampling from original 12. store the similarity
set of features we use two different sampling functions: 13. Now we go the subset test set which have lowest
sampling with replacement and sampling without similarity factors determine the person id and
replacement. In sampling with replacement a feature can be match the id between training subset and test set.
selected more than once which we treat as increasing its 14. Initialize error count to zero
weight. 15. For each test face If the id is match then
Each of the PCA classifiers uses the same number of recognition done and accuracy will be hundred
features. This is parameter of algorithm which we set by percent.
cross confirmation performance estimates on the training 16. If the id of any test face image is not equal to id of
set. training subset of each image then increment in
error count.
The similarity among new projected data can be calculated 17. Compute the accuracy by using error count value.
using euclidean distance the identity of most similar Wi will
be output of our classifier. If i=j it means that we have 3. Experiment Discussion
correctly identified the subset otherwise we have
misclassified the subset. The schematic diagram of this To access the possibility of this approach to face
presented in figure 1. recognition, we have performed experiments .we used the
ORL database [9] which available in public domain. We
2.4. Pseudo code of combining algorithm also used image processing tool of MATLAB 7.6. Using
1. Prepare training set which contains subset of ORL database we have conducted 3 experiments to access
images the accuracy of recognition by formula:
2. Set PCA dimensionality parameter
3. Read training subset (1-(error count/total test image count))*100
4. Form training data matrix.
5. Form training class label matrix In the experiments we take 5 training subset of face image.
6. Calculate PCA transform Matrix Each subset contains 20 face images of 4 different person
7. Calculate feature vector , projected data of all and one test set which contains 5 images of a person. In the
subset of training set by using PCA transformation first experiment we set up 60 percent accuracy. In the
matrix
10 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
second experiment we set up 80 percent accuracy and at last Abhishek K Mishra received the M.Tech Degree in computer
in third experiment we found 60 % accuracy. Technology and Application from school of Information
As can seen my result are no where near perfect yet with Technology, UTD, RGPV, Bhopal, India.. Currently Working as
my best outcomes in with almost 80 percent matches. Sr. lecturer in college of Engineering and Technology, IFTM
Moradabad. INDIA.
If we study this approach in the view of complexity
then definitely I assured both time and space complexity will Navin prakash received the B Tech Degree in Computer Science
be reduced because we are using multiple feature subset from BIET Jhansi in 2001, India. Currently Working as Sr. lecturer
concepts as like in nearest neighbor classifier from multiple in college of Engineering and Technology, IFTM Moradabad.
feature subsets [7]. The nearest neighbor classifier from INDIA.
multiple feature subsets supports lower complexity.
Himanshu Sharma received the B.Tech Degree in computer
Science and Information technology from MIT, Moradabad, India.
Currently Working as Sr. lecturer in college of Engineering and
4. Conclusion and future work Technology, IFTM Moradabad, INDIA.
As can be seen my results are near to perfect while these
results may be improve. Conceptually this approach reduced
the complexity also but this approach has two limitations
one voting can only improve accuracy if the classifier select
the correct class more often than only other class. Another
algorithm has two parameter value need to be set first is size
of features subset and second is number of classifier. So in
the near future we can also improve the accuracy and
complexity of this approach.
5. References
[1] M. Turk and A. Pentland,”Face recognition using
eigenface”, In Computer Vision and Pattern
Recognition, 1991.
[2] W.W.Bledsoe, “the model method in facial recognition”,
panoramic research inc. Palo alto.ca roe. Pr: 15, Aug
1966
[3] T. Kanade, “picture processing system by computer
complex and recognition of human faces”, dept pf
information science, Kyoto university, Nov 1973.
[4] Xiaoguang Lu, Yunhong Wang, Anil K. Jain,
“combining classifier for face recognition”.
[5] Bing-Yu-Sun- Xiao-Ming Zhang and Ru-jing Wang,
“Training SVMs for Multiple feature classifier
problem”.
[6] Julein Meynet Vlad Popovici, Matteo, Sorci and jean
Philippe Thiran, “combining SVMs for face class
modeling”.
[7] Stephen D.Bay,” Nearest neighbor classification from
multiple feature subset”, November 15, 1998.
[8] Linday I smith, “A tutorial on principle component
analysis”, February 26, 2002
[9] http://homepages.cae.wiseedu/~ece533/
[10]http://en.wikipedia.org/wiki/Principal_component_anal
ysis
Authors Profile
Rakesh Kumar Yadav received the B.Tech Degree in
Information Technology from A K G Engineering College
Ghaziabad, INDIA in 2004. Currently pursuing M.Tech and
Working as Sr. Lecturer in college of Engineering and Technology,
IFTM Moradabad. INDIA.
(IJCNS) International Journal of Computer and Network Security, 11
Vol. 2, No. 5, May 2010
Fig. 1 shows the structure of IEEE802.11e EDCA model, However, these techniques can create attempts at
Wireless Station (WSTA) that have QoS parameters to unnecessary data transmission that exceed available
decide priority and four transmission queues that are bandwidth of networks because it does not have an adaptive
recognized by the virtual station. If more than two values of transmission rate adjustments scheme. As a result,
the back-off counter in a station reach to 0 at the same time, transmission queue overflow or loss of data causes
the scheduler of WSTA prevents a virtual collision. degradation in the quality of the video streaming service.
x 144), Foreman. Each video frame was fragmented into 237 149 273 0 1 9
packets before transmission, and the maximum transmission
packet size over the simulated network is 1000 bytes, Fig. 7
shows the video used in our simulations.
Packet Sent Vs. Packet Lost
300
273
The video sequence has temporal resolution 30 frames per 250 237
Amount
149 Packet Sent
amount of each frame/packet in the video sequence 150
Packet Lost
respectively. 100
50
9
Table 1: Amount of video frames in the video source 0
0 1
I P B
Pa ck e t type
Video Format Frame Number Total
I P B
Foreman QCIF 45 89 266 400 Figure 8. Sent packets vs. lost packets
Table 2: Amount of video packets in the video source As shown in Table 5 & Fig. 9, the total number of lost video
frame is 7, including 0 I frames, 1 P frames, and 6 B frames.
Video Format Packet Number Total The Decodable Frame Rate (Q) is 0.9625, and the average
PSNR of the result video is 33.67, as shown in Fig. 10 is a
I P B
sample of an output video frame.
Foreman QCIF 237 149 273 659
Transport 250
Protocol UDP UDP UDP UDP
200
CWmin 3 7 15 7
Amount
5. Conclusions
In this paper, we evaluated a framework to improve the
quality of video streaming. This framework consists of
MAC-centric cross-layer architecture to allow MAC-layer to
retrieve video streaming packet information (slice type), to
save unnecessary packet waiting time, and a single-video
multi-level queue to prioritize I/P/B slice (packet) delivery.
We evaluated our simulations based on the decodable frame
rate (Q) and PSNR. Through simulations, we also revealed
that PSNR & decodable frame rate can evaluate the
perceived quality well by an end user. Therefore, we found
that the larger the Q & PSNR value, the better the video
quality perceived by the end user. Simulations show that the
evaluated methodology outperforms IEEE802.11e in packet
loss rate, and PSNR.
References
[1] IEEE 802.11e, “Wireless LAN Medium Access Control
(MAC) Enhancements for Quality of Service (QoS),”
802.11e Draft 8.0, 2004.
[2] S. Choi, “Overview of Emerging IEEE 802.11 Protocols
for MAC and Above,” Telecommunications Review,
Special Edition, November 2003.
[3] D. Gu, and J. Zhang, “QoS Enhancement in IEEE
802.11 Wireless Local Area Networks,” IEEE
Communications Magazine, June 2003.
[4] D. Chen, D. Gu, J. Zhang, “Supporting Real-Time
Traffic with QoS in IEEE802.11e Based Home
Networks,” IEEE Consumer Communications and
Networking Conference, January 2004.
[5] D. Gao, J. Cai, P. Bao, Z. He, “MPEG-4 Video
Streaming Quality Evaluation in IEEE802.11e
WLANs,” IEEE International Conference on Image
Processing, September 2005.
[6] A. Ksentini, A. Gueroui, M. Naimi, “Toward an
improvement of H.264 Video Transmission over
IEEE802.11e through a Cross Layer Architecture,”
IEEE Communications Magazine, Special Issue on
Cross-Layer Protocol Engineering, January 2006.
[7] Sai Shankar N, Mihaela van der Schaar, “Performance
Analysis of Video Transmission over IEEE 802.11a/e
WLANs,” IEEE Transactions on Vehicular, VOL. 56,
Issue 4, 2007.
[8] Pilgyu Shin, Kwangsue Chung, “A Cross-Layer Based
Rate Control Scheme for MPEG-4 Video Transmission
by Using Efficient Bandwidth Estimation in IEEE
802.11e,” School of Electronics Engineering,
Kwangwoon University, Seoul, Korea, 2007.
[9] NS-2, http://www.isi.edu/nsnam/ns.
[10] Chih-Heng, Ke, “How to evaluate MPEG video
transmission using the NS2 simulator,” EE
Department, NCKU, Taiwan, 2007.
[11] S. Wiethölter and C. Hoene, “Design and Verification
of An IEEE802.11e EDCF Simulation Model in ns-
2.26,” Tech. Rep. TKN-03-019, Technische Universität
Berlin, 2003.
[12] Cygwin, http://www.cygwin.com
[13] IEEE Std 802.11e-2005, “Wireless LAN Medium
Access Control (MAC) and Physical Layer (PHY)
specifications Amendment 8: Medium Access Control
(MAC) Quality of Service Enhancements,” November
2005.
[14] Zhen-ning Kong, Danny H. K. Tsang, Brahim Bensaou,
Deyun Gao, “Performance Analysis of IEEE802.11e
16 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
4
Department of of Mechanical Engineering
Singhania University, Rajasthan, India
rup_esh100@yahoo.co.in
3. Subspace LDA Method each image in the database is calculated. The test image will
match the image in the database having the minimum
It is an alternative method which combines PCA and Euclidean distance with it.
LDA [5]. This method consists of two steps; the face image
is projected into the eigenface space which is constructed by In figure 2, Euclidean distance of one test image from all the
PCA, and then the eigenface space projected vectors are 400 images is shown. The Euclidean distance is zero with
projected into the LDA classification space to construct a the image no. 7 in the database as clear from the figure 2.
linear classifier. In this method, the choice of the number of As the Euclidean distance is zero, so test image completely
eigenfaces used for the first step is critical; the choice match the image from our database as shown in figure 3.
enables the system to generate class separable features via
LDA from the eigenface space representation. Projecting the
data to the eigenface space generalizes the data, whereas
implementing LDA by projecting the data to the
classification space discriminates the data. Thus, Subspace
LDA approach seems to be a complementary approach to
the Eigenface method.
4. Experimental Results
To analyze the performance of human face recognition
using subspace LDA, we performed experiments on ORL
databases using MATLAB. There are 10 different images of
40 distinct subjects in ORL database. For some of the
subjects, the images were taken at different times, varying
lighting slightly, facial expressions (open/closed eyes,
smiling/non-smiling) and facial details (glasses/no-glasses).
All the images are taken against a dark homogeneous Figure 2. Euclidean Distance of a test image from other
background and the subjects are in up-right, frontal position images in database
(with tolerance for some side movement). The files are in
.TIF file format (Tagged Image File Format). The size of
each image is 92 x 112 (width x height), 8-bit grey levels.
Figure 3.
References
[1] S.A. Rizvi, P.J. Phillips, and H. Moon, “A verification
protocol and statistical performance analysis for face
recognition algorithms”, pp. 833-838, IEEE Proc.
Conf. Computer Vision and Pattern Recognition
(CVPR), Santa Barbara, June 1998.
[2] M.Kirby and L. Sirovich “Application of the
Karhunen-Loeve procedure for the characterization of
human faces". IEEE Transactions on Pattern analysis
and Machine Intelligence 12 (1): 103–108. doi:
10.1109 /34.41 39
[3] A.Pentland, B.Moghaddam, T.Starner, O. Oliyide, and
M. Turk (1993), “View - based and modular
Eigenspaces for face recognition". Technical Report
245, M.I.T Media Lab.
[4] R.A.Fisher, “The Use of Multiple Measurements in
Taxonomic Problems”, Annals of Eugenics, Vol. 7,
No. 2. (1936), pp. 179-188.\
[5] P. N. Belhumeur, J. P. Hespanha and D. J. Kriegman,
“Eigenfaces vs. Fisherfaces: Recognition Using Class
Spesific Linear Projection”, IEEE Transactions on
Pattern Analysis and Machine Intelligence, Vol. 19,
Figure 4. No. 7, July 1997.
Figure 6.
5. Conclusions
The subspaceLDA method is a combination of eigenface
method and LDA approach. Eigenface method is used for
dimension reduction to decrease the processing time and
LDA approach is used for classification due to its
discrimination power to enhance the recognition rate. From
the observations, it is clear that a high recognition rate is
observed when the eigenface space’s dimension is small (40-
60) and it is less when eigenface space’s dimension is large
(175-200).
It is also clear that if the minimum Euclidian distance of
the test image from other images is zero, then the test image
completely matches the existing image in the database. If
minimum Euclidian distance is non-zero but less than
threshold value, then it is a known face but having different
face expression else it is an unknown face.
(IJCNS) International Journal of Computer and Network Security, 19
Vol. 2, No. 5, May 2010
Abstract: In this paper, we consider the Travelling salesman’s interesting connections between graph theory and Petri
problem (TSP). Where TSP is modelled as a TSP-graph, in Nets have been brought out by T. Murata [9].
which each edge is treated as a parallel combination of This paper is organized as follows: Section 2 provides
oppositely directed edges. We model TSP-graph as a Petri Net-
the necessary preliminaries, Section 3 formulates the
graph, where Petri Net- graph is an underlying graph of TSP-
graph. Then solve TSP by defining suitable binary operation on problem, Section 4 describes the algorithm for TSP with
elements of columns in sign incidence matrix representation of illustrative example and Section 5 briefly concludes this
Petri Net-graph. In Petri Net-graph, we find a set of places paper.
which is both Siphon and Trap with minimum sum of
capacities, whose set of input transitions equals to the set of
output transitions, and both of them are equal to the set of all
2 . An Overview Of Petri Net Approach
transitions in Petri Net. Then edges in TSP-graph corresponding
to these places in Petri Net-graph will form a shortest route for This section of the paper provides the necessary
the salesman to return the point of origin, after traversing all preliminaries for the readers who are not familiar with
the cities exactly ones. For the solution of TSP, we describe a Petri nets.
new algorithm, based on siphon-trap and bounded-ness property As Petri Nets are also called place-transition net (PT-
of the Petri Nets.
Net), it is a particular kind of directed graphs together
2000 Mathematics Subject Classification: 68R10, 90C35,
94C15. with an initial state called the initial marking. In general a
Keywords: Travelling Salesman’s Problem, Weighted Petri Net is an underlying graph of any directed graph,
Directed Graph, Spanning Cycle, Petri Net, Siphon and Trap. which is in essence a directed bipartite graph with two
types of nodes called places and transitions. The arcs are
1. Introduction either from places to transitions (output of places) or from
The Travelling Salesman’s Problem (TSP) is one of the transitions to places (input of places). In the Petri Net
most intensely studied problems in computational graph a place is denoted by a circle, a transition by a box or
mathematics [5]. Mathematical problems related to the a bar and an arc by a directed line. A Petri Net is a PT-Net
TSP were treated in the early nineteenth century by W.R with tokens assigned to its places denoted by black dots,
Hamilton and British mathematician T. P. Kirkman. and the token distribution over its places is done initially
Although there are many algorithms given for the solution by a marking function denoted by M0. A token is
of TSP [6, 7, 8, 13, 14], yet no effective solution is known interpreted as a command given to a condition (place) for
for the general case for the TSP. In this paper, we address the firing of an event (transition). An event can happen,
the same problem with a different approach, using Petri when its all input conditions are fulfilled, See Fig.1.
Net model. Here we present a new algorithm to solving a
TSP using the siphon-trap and bounded-ness property of
the places in the One-one Petri Net model of given TSP-
graph. For the TSP we find a set of places in Petri Net,
which is both Siphon and Trap [1, 3], with minimum sum
of capacities, having the property that set of input
transitions equals to the set of output transitions, and both Figure 1.
of them are equal to the set of all transitions given in the
Net. Then edges in TSP-graph corresponding to these There are many subclasses of Petri Nets such as One-one
places form a shortest route for the salesman. Petri Nets, free choice Petri Nets, colored and stochastic
Petri Nets etc. Here we introduce only One-one Petri Net, as
In Petri Net theory, Petri Net is a formal tool which is
it is an ordinary Petri Net such that each place P has exactly
particularly well suited for discrete event systems. Its
one input transition and one output transition having weight
application has emerged from the initial seminal PhD one on each edge, but we ignore these weights generally in
thesis of C. A. Petri, so C.A. Petri is considered as the the model representation of the Net PN. One-one Petri Net
originator of Petri Net applications [10, 11]. The is also called as Marked Graph [4]. In our paper standard
computational algorithmic aspects of graph theory are notation PN is treated as One-one Petri Net. More detailed
emphasized in the study of the Petri Nets. The most and formal description of Petri Nets is given in [2, 9, 10,
20 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
15]. We include here some basic definitions, which are goes through every city exactly once is a feasible tour with
relevant to this paper. a given cost, if it is smaller than the other minimum cost
Definition 2.1: A place-transition net (PT-Net) is a tour.
quadruplet PN = 〈P, T, F, W〉, where P is the set of places, T 3.1 Modelling TSP as a Graph (TSP-graph):
is the set of transitions, such that P ∪ T≠∅ and P ∩ T=∅, A pair G = {V, E}, where V= {v1,v2,v3,…,vn} is the set of
F⊆ (P x T) ∪ (T x P) is the set of arcs and W: F à{1, 2 ...} vertices and E = {e1,e2,…,em} is the set of edges such that
is the weight function. PN is said to be an ordinary PT-Net
each edge ei having some weights wi∈W, where W:Eà R+
if and only if W: Fà {1}.
is the weight function and wi =w(ei) is the weight
A marking is a function M0: Pà {0, 1, 2 ...}, which
distributes the tokens to the places initially. Here M0 (p) is associated with edge ei . Further when the edges vi vj and
the number of tokens in the place p at initial marking M0, it vj vi are considered different then G={V,E} is called a
is a non- negative integer less then or equal to the capacity weighted directed graph. In weighted directed graph, a
of the place. Capacity of the place is defined as the directed cycle is a closed sequence of directed edges
capability of holding the maximum no. of tokens at any without repetition of vertices except terminals and it said
reachable marking M from M0. A marking M is said to to be Spanning if it contains all the vertices of the graph. If
reachable to M0 if there exist a firing sequence σ = {t1, t2… any edge from the sequence is deleted then cycle becomes
tn} such that M can be obtained from M0 as firing of open. Thus solving a TSP amounts to finding a minimum
transitions t1, t2,…, tn. A Petri Net structure PN = 〈P, T, F, weight spanning cycle.
W〉 without any specific initial marking is denoted by PN As an illustration we consider a weighted graph on
and Petri Net with the given initial marking is denoted by 〈 four vertices denoting the four cities A, B, C, D, where
PN, M0〉. *x and x* are the set of input transitions (or each city has a direct link with the other three. The
places) and the set of output transitions (or places)
numbers (weights) associated with the edges denotes the
respectively, as x∈ P (or T). Here | *x | and | x* | stands for
physical distance between the cities, See Fig. 2. We
number of the input transitions (or places) and the output
transitions (or places) respectively. Thus a One-one Petri construct a directed weighted graph (TSP-graph) of the
Net is an ordinary PT-Net such that ∀ p ∈ P: |*p|=|p*| =1, same graph in Fig. 3.
i.e., the number of the input transitions for p ∈ P equals to A
the number of the output transitions and both of them are 298 128
equal to one.
163
For a PT-Net, a path is a sequence of nodes ρ = 〈 x1, x2…
xn 〉 where (xi, xi+1) ∈F for i = 1, 2… n-1. ρ is said to be D 206
B
elementary if and only if it does not contain the same node 137
more than once and a cycle is a sequence of places 〈 p1 , p2 , 98
…pn 〉 such that there exist t1, t2,…, tn ∈ T : 〈 p1, t1, p2 ,
t2,…., pn , tn 〉 forms an elementary path and (t n, p1) ∈F. C
Definition 2.2: For a PT-Net 〈PN, M0〉, a place p is said to Figure 2. Network of cities
be k-bounded (or bounded by k) where k∈ R+, if and only if
V1
M0 (p) < k., denotes the capacity of the place in the Net. e1 e2 e8
(PN, M0〉 is said to be k-bounded if and only if every place is
k-bounded. e11 e9 e7
Definition 2.3: A non-empty subset of places S is called a V4 e12 e10
Siphon if *p⊆ p* ∀ p∈S denoted also *S⊆ S*; i.e., every e3 e4 V2
transition having an output place in S has an input place in e6
S. Likewise a non-empty subset of places Q is called a Trap
if p*⊆ *p ∀ p∈Q denoted also Q* ⊆ *Q; i.e., every V3 e5
transition having an input place in Q has an output place in
Q. Figure 3. TSP- Graph
3. Problem Formulation: Travelling Salesman’s Distance between two cities in Fig.3 can be represented in
Problem (TSP) terms of adjacency matrix as depicted below:
3.2 Modelling TSP-graph as a Petri Net I of PN, which is both siphon and trap and also *X= X* = T,
A Petri Net PN is modelled from TSP-graph given in Fig. 3 i.e., set of the input transitions of X is equals to the set of
as follows: edges es are transformed into places ps and output transitions of X, and both of them equal to the set
vertices vk are transformed into transitions tk so that the of all transition in T. This equality holds only if the
place ps has an input from a transition ti, and an output to a addition under the operation ⊕, of the k column vectors
transition tj, if es is an directed edge vi vj in graph, then say C1, C2, C3,…, Ck i.e., C1 ⊕ C2 ⊕ C3⊕…..⊕ Ck contains
weights (distance between two cities) of es’s are replaced only ± entries everywhere, where Cj , j = 1, 2... k, denotes
by capacities ks of corresponding places ps and number of the column vector corresponding to the place pj in I.
tokens for places ps ‘s is the value of ks . Now let C1 ⊕ C2⊕ C3⊕…..⊕ Ck will be a column vector
denoted by γ = [γi] where γi denotes the
Fig. 4 in Appendix (1), shows the One-one Petri Net
i
th
element of the column vector γ, have as elements from
model say PN, of the TSP-graph, having the set of
set U. Then from the definition of I under the operation⊕,
transitions T= {t1 t2, t3, t4} and the set of places is P = {p1,
p2,…,p12 } corresponding to the set of vertices{v1,v2 ,v3,v4} we interprets about γi as. γi = 0 means no place in X is an
input or output place of transition i.
and the set of edges {e1,e2 ,…,e12} respectively in the given
γi = - means some place in X is an input place of
TSP-graph. For the sake of clarity, we observe that t1 is the
transition i.γi = + means some place in X is an output place
input transitions for the places p2, p7, p10 and the output
transition for the places p1, p8, and p9. Similarly p1 is the of transition i.
γi = ± means some place in X is an input place as well as
input place for the transition t1 and output place for the
output place for transition i.
transition t4. Using this same procedure, we find a set of the
From the above it can be seen that every transition having
places in Net which is both siphon and trap with minimum
sum of capacities, whose set of the input transitions equals an output place in X has an input place in X only if γi ≠ +,
and likewise every transition having an input place in X
to the set of output transitions and both of them are equal
has an output place in X only if γi ≠-. So X is both siphon
to the set of all transitions in the Net PN.
and trap if and only if γ has either 0 or ± entries. And if γ
4. Description of algorithm for Travelling
has only ± entries everywhere, then the places
Salesman’s Problem corresponding to columns C1, C2, C3,…, Ck in C1 ⊕ C2 ⊕
As in above we model the TSP-graph as a Petri Net-graph.
C3⊕…..⊕ Ck forms a set of places, which is both siphon
Now here we present an algorithm for solving the
and trap, whose input transitions equals to the output
Travelling Salesman’s Problem. For the description of the
transitions and both of them are equal to the set of all
algorithm for TSP; we introduce some new notations as
transitions T, with having some capacities. We select only
follows:
those set of places X such that *X = X* = T, having
In a Petri Net PN with n-transitions and m-places, the sign
minimum sum of capacities among all. Algorithm
incidence matrix I = [aij] is the
discussed below, gives us a siphon–trap set of places X with
n x m matrix, whose entries are defined as,
minimum sum of capacities, having the property *X = X* =
aij = + if place j is an output place of transition i.
T, whose corresponding edges in TSP-graph will form a
aij = - if place j is an input place of transition i.
shortest route for the salesman.
aij = ± if place j is both input and output places of
transition i, (i.e., transition i and place j form a self loop) ALGORITHM:
Input Sign incidence matrix I of order n x m.
aij = 0 otherwise.
Step 1 Select Cj, the first column in the sign incidence
matrix I having ‘+’ entry whose corresponding place and
As an illustration, the Sign incidence matrix I of PN is:
capacity is denoted as PLACEj and CAPACITYj
p1 p2 p3 p4 p5 p6 p7 p8 p9 p10 p11 p12
Set recursion level r to 1
t1 - + 0 0 0 0 + - - + 0 0 Set Vjr = Cj
Set PLACE jr = PLACEj;
t2 0 0 0 0 + - - + 0 0 + - Set CAPACITYjr =CAPACITYj;
t3 0 0 + - - + 0 0 + - 0 0 Set X =∅, W=∅, Sum =0 and K =∅.
t4 + - - + 0 0 0 0 0 0 - +
Step 2 If Vjr has a ‘±’ entry at the ith row then PLACEjr is a
Hare we introduce a commutative binary operation, self loop with transition i, Go to Step 5.
denoted by ⊕ on the set U= (0, +, - ,±).
Step 3 If Vjr has a ‘+’ entry in the kth row find a column Cs
⊕ defined as following: + ⊕ - = ± (i.e., ‘+’ entry is said
which contains a ‘–‘entry at kth row.
to be neutralized by adding a ‘–‘entry to get a ‘±’ entry).
a⊕a=a ∀ a∈U (i) If no such column Cs exists then go to Step 5.
±⊕a=± ∀ a∈
(ii) If such Cs exists, add it to Vjr to obtain Vj(r+1) = Vjr ⊕
0⊕ a=a ∀ a∈U
Cs, containing a ‘±’ entry at kth row. Then
For the Travelling salesman’s problem, we choose a
subset of k places X = {p1, p2 …pk} in sign incidence matrix PLACEj(r+1) = PLACEjr ∪ PLACEs
22 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
CAPACITYj(r+1) = CAPACITYjr ∪ CAPACITYs V12( 2) has ‘+’ entry at 3rd row. The neutralizing columns
(iii) Repeat this step for all neutralizing columns Cs. This are C4, C5 and C10.
gives a new set of Vj(r+1)’s, PLACEj(r+1)‘s and
− 0 −
CAPACITYj(r+1) 0 0 0
V13(1) = V12( 2) ⊕ C4 = ⊕ = ;
Step 4 Increase r by 1, Repeat Step 3 until there are no ‘+’entries + − ±
in each Vjr = C1 ⊕ C2⊕ C3⊕…..⊕ Cjr.
± + ±
Step 5 Any Vjr with all entries as ‘±’ represents both siphon
(1) (1)
and trap such that their input transitions equal to the output PLACE 13 = {pl,p3, p4}; CAPACITY 13 =
transitions and both of them equal to the set of all {137,206,206}
transitions T.
X = X ∪ PLACEjr − 0 −
0 + +
K= K ∪ CAPACITYjr V13( 2) = V12( 2) ⊕ C5 = ⊕ = ;
W=Sum + Sumw K, where Sumw K is the sum of the + − ±
capacities in the set K, Store it any other set and compare it ± 0 ±
to minimum weight set at each iteration.
Step 6 Delete Cj (2) ( 2)
PLACE 13 ={pl, p3, p5}; CAPACITY 13 = {137,
j = j + 1 Go to Step 1. 206,128}
Output: Set X has places both siphon and trap such that − + ±
their input transitions equal to the output transitions and 0
V13(3 ) = V12( 2 ) ⊕ C10 = ⊕ 0 = 0 ;
both of them equal to the set of all transitions T with + − ±
minimum sum of capacities. Whose corresponding edges
± 0 ±
set in TSP-graph, forms a shortest route for the salesman.
As an illustration consider the graph in Fig 2. ( 3) ( 3)
PLACE 13 = {pl, p3, p10}; CAPACITY 13 = {137, 206, 98}
Step 1 Select first column having ‘+’ entry. Here is C1,
then V12( 3) has ‘+’ entry at 2nd row. The neutralizing column is
−
C6, C7 and C12.
V11 = 0 ; PLACE11 = {pl};
0 − 0 −
+ − ±
⊕ C6 = ⊕
( 4) (3 )
+
V 13 =V 12 = ;
0 + +
CAPACITY11 = {137} ± 0 ±
Steps 2, 3 and 4, V11 has a ‘+’ entry at 4th row. The (4) (4)
neutralizing columns are C2, C3 and C11. PLACE 13 = {pl, p11, p6}; CAPACITY 13 = {137,298,128}
− + ± − + ±
(1) 0 0 + − ±
V = V11 ⊕ C2 = ⊕ = 0 ; V13(5 ) = V12(3 ) ⊕ C7 = ⊕
12 = ;
0 0 0 0 0 0
+ − ± ± 0 ±
(1) (1) ( 5) ( 5)
PLACE 12 = {pl , p2}; CAPACITY 12 ={137,137 } PLACE 13 = {pl , p11, p7}; CAPACITY 13 = {137,298,163}
− 0 −
0 − 0 −
V ( 2)
= V11 ⊕ C3 = ⊕ 0 = 0 ; + − ±
⊕ C12 = ⊕
( 6) ( 3)
12 V =V = ;
0 + + 13 12
0 0 0
+
− ±
± + ±
(2) (2) ( 6) ( 6)
PLACE 12 = {pl, p3}; CAPACITY 12 = {137,206} PLACE 13 = {pl, p11, p12}; CAPACITY 13 = {137,298,298}
− 0 −
0 + + V13(1) and V13( 6 ) has no ‘+’ entry , also all the entries are not
V12(3 ) = V11 ⊕ C11 = ⊕ = ;
0 0 0 ‘±’ only, but those sets which have all entries ‘±’ or ‘0’ are
both siphon and trap ( as V12(1) , V13( 3) and V13( 5) )
+
− ±
V13( 2) has ‘+’ entry at 2rd row. The neutralizing columns are
( 3) ( 3)
PLACE 12 = {pl, p11}; CAPACITY 12 = {137,298} C6, C7, and C12.
(IJCNS) International Journal of Computer and Network Security, 23
Vol. 2, No. 5, May 2010
− + ±
or {p2, p4, p6, p8}. The edges set {el, e3, e5, e7} or {e2, e4, e6, e8 }
are corresponding to places set {pl, p3, p5, p7 } and {p2, p4, p6,
V14( 2 ) = V13( 2 ) ⊕ C7 = + ⊕ − = ± ;
± 0 ± p8} respectively, will form a shortest spanning cycle in
underlying graph of Petri Net-graph, which give an
± 0 ±
optimal route for the salesman in TSP-graph.
The finding of this can be applied to a company’s logistic
(2) (2) problem of delivering petrol or similar to different petrol
PLACE 14 = {pl, p3, p5, p7}; CAPACITY 14 = {137,206,128,163}
sub-stations. This delivery system is formulated as the Petri
− 0 − Net model, which involves finding an optimal route for
+
V14 = V13 ⊕ C12 = ⊕ − = ± ;
( 3) ( 2)
visiting stations and returning to point of origin, where the
± 0 ± inter-station distance is symmetric and known. As a
±
+
± standard problem, we defined it simply as the time spent or
distance traveled by salesman visiting n cities (or nodes)
cyclically, where vehicle visits each station just once and
( 3) ( 3)
PLACE 14 ={pl, p3, p5, p12}; CAPACITY 13 ={137,206,128,298} returns the starting station. This real world application is a
deceptive simple combinatorial problem and our approach is
V13( 4 ) has ‘+’ entry at 3rd row. The neutralizing columns are to develop solutions of such type of distribution problems,
C , C , and C . based on concept of the Petri nets.
4 5 10
− 0 − 5. Conclusion
±
C4 = ⊕ 0 = ± ;
( 4) ( 4)
V 14 =V 13 ⊕ In this paper, while solving the travelling salesman
+ − ± problem, we have exploited the potentials of siphons and
± + ± traps. Our analysis is based on the notion of sign incidence
matrix; this helps us to relate Petri Net theory to graph
(4) (4) theory. The complexity of the TSP is part of a deep
PLACE 14 ={pl, p6, p11, p4}; CAPACITY 14 ={137,298,128,206}
question in mathematics, as TSP is a NP- complete
− 0 − problem. Here we have developed an algorithm using Petri
V14(5) = V13( 4) ⊕ C5 = ± ⊕ + = ± ; net model, which can be executed by computer for any finite
+ − ± number of nodes.
± 0 ±
Acknowledgement
I would like to thank my supervisor Prof.G.N. Purohit for
( 5) (5 )
PLACE 14 ={pl, p6, p11, p5}; CAPACITY 14 = {137,298,128,128} his continuous support and valuable suggestions for this
research work. I would also like to thank my friends,
− + ±
colleagues and Centre for mathematical sciences for their
V14( 6 ) = V13( 4 ) ⊕ C10 = ± ⊕ 0 = ± ; guidance and feedback to drive me in the proper direction of
+ − ± the research is highly appreciable. Last but not the least I
± 0 ± would like to thank my parents for their love and support.
( 6) ( 6)
PLACE 14 = {pl, p6, p11, p10}; CAPACITY 14 = {137, 298, 128, 98}
References
V14(1) , V14( 3) , V14( 4) and V14(5) has no + entry. V14( 2) , V14( 6)
have all the entries as ‘±’, Hence PLACE 14 and PLACE 14
(2) ( 6) [1] R. Boer and T. Murata, “Generating basis siphons and
traps of Petri nets using sign incidence matrix”, IEEE
form both siphon and trap whose input transitions equal to
Transactions on circuits and systems-1, Vol.41, No. 4,
the output transitions and both of equal to the set of all
pp. 266-271,(1994).
transitions.
Step 5 the subsets of places, which are both siphon and [2] B. Baugarten,”Petri Nets basics and application” , 2nd
trap, whose input transitions equal the output transitions ed., Berlin: spectrum akademischer Verlag, (1996).
and both of them equal to the set of all transitions are{pl, [3] K.S. Cheung and K.O. Chow, “Cycle-Inclusion Property
p3, p5, p7 }and {pl,, p6, p11, p10}with capacities sum 634 and 661 of Augmented Marked Graph” , Information Processing
as choosing column first C1. Letters, Vol. 94, No. 6, pp. 271-276, (2005).
24 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
[4] F. Commoner, A. W. Holt, S. Even and A. Pnueli, her P.hd in guidance of Prof. G. N. Purohit in research area of
“Marked directed graph” , Journal of Computer and Petri net theory, As the Junior Research Scholar under the centre
System Sciences- 5, pp. 511-523, (1971). for mathematical sciences(CMS), governed by DST-New Delhi, in
[5] Narasingh Deo, “Graph theory with Applications to Banasthali University(Rajasthan).
Engineering and Computer Science”, Prentice-Hall,
India, (1984).
[6] S. Hiller Frederick and J Lieberman Gerald,
Appendix
“Operations Research “, Eighth edition, Tata Mcgraw
Hill, (1990),.
[7] M. Jünger, G. Reinelt and G. Rinaldi, "The Traveling
Salesman Problem," in Ball, Magnanti,Monma and
Nemhauser (eds.) , Handbook on Operations Research
and the Management Sciences North Holland Press, pp.
225-330, (1994).
[8] D. S. Johnson, "Local Optimization and the Traveling
Salesman Problem," Proc. 17th Colloquium on
Automata, Languages and Programming, Springer
Verlag, pp. 446-461, (1990).
[9] T. Murata, “Petri nets, properties, analysis and
application” , Proceedings of IEEE, Vol.77 (4), pp. 541-
580, (1989).
[10] J. L., Peterson, “Petri net Theory and the Modeling of
Systems” , Prentice Hall, INC, Englewood, Cliffs, and
New Jercy, (1981).
[11] C.A. Petri, "Kommunikation mit Automaten." Bonn:
lnsti-tut fur Instrumentelle Mathematik, Schriften des
MM Nr. 3, 1962. Also, English translation,
"Communication with Auto-mata." New York: Griffiss
Air Force Base.Tech. Rep.RADC-TR-65-377, vol.1,
(1966).
[12] C.A. Petri, "Fundamentals of a theory of asynchronous
information flow," in Proc. of IP Congress 62,pp. 386-
390, (1963).
[13] J.V. Potvin, "The Travelling Salesman Problem: A
Neural Network Perspective", INFORMS Journal on
Computing 5, pp. 328-348, (1993).
[14] J.V. Potvin,"Genetic Algorithms for the Traveling
Salesman Problem", Annals of Operations Research 63,
pp. 339-370, (1996).
[15] W. Reisig, “Petri Nets and introduction “, Heidelberg:
springer-verlag, (1985).
Authors Profile
Prof. G. N. Purohit, a renowned
Mathematician of the state, joined Banasthali
University in April, 2001 as Professor in the
Department of Mathematics & Statistics and
also Coordinator of Centre for mathematical
sciences in Banasthali University. He received
the B.Sc, M.Sc and P.hd degree in
mathematics from Rajasthan University during
1957-1968. His research interests are in Fluid
Dynamics, Partial Differential Equations, Petri
net theory and wireless networks.
Abstract: This paper presents the current status of PET/CT computer vision (target localization, automatic quality
brain image registration. Registration of brain images is a more control), to name a few. During the last decades, image
complex problem, Automatic registration, based on computer acquisition devices have undergone rapid development and
programs, might, however, offer better accuracy and growing amount and diversity of obtained images invoked
repeatability and save time. Computed Tomography (CT) is used the research on automatic image registration. A
for the attenuation correction of Positron Emission Tomography comprehensive survey of image registration methods was
(PET) to enhance the efficiency of data acquisition process and
published in 1992 by Brown . According to the database of
to improve the quality of the reconstructed PET data in the
brain. Due to the use of two different modalities, chances of
the Institute of Scientific Information, in the last 10 years
misalignment between PET and CT images are quite significant. more than 1000 papers were published on the topic of image
Template matching is used for applications in image processing. registration. Methods published before 1992 that became
Cross Correlation is the basic statistical approach to image classic or introduced key ideas, which are still in use, are
registration. It is used for template matching or pattern included as well to retain the continuity and to give
recognition. Template can be considered a sub-image from the complete view of image registration research.
reference image, and the image can be considered as a sensed Biomedical imaging based on different physical
image. The objective is to establish the correspondence between principles has played a more and more important role in
the reference image and sensed image. It gives the measure of medical diagnosis. Multiform methods of image diagnosis
the degree of similarity between an image and template. This provide doctors and clinicians with various anatomical and
paper describes medical image registration by template
functional information to carry out exact diagnosis and
matching based on Normalized Cross-Correlation (NCC) using
effective treatment. Image registration is the process of
Cauchy-Schwartz inequality.
combination the different sets of data of the same object,
Keywords: Positron Emission Tomography (PET), which come from different modalities (CT, MRI, SPECT,
Computed Tomography (CT), Magnetic Resonance (MR), PET etc.). Pre-processing algorithms improve image quality
Normalized Cross-Correlation (NCC) and image registration algorithms transform the object of
images into one coordinate system.
Nowadays, medical image diagnosis is considered
1. Introduction as one of the fields taking advantage of high-technology and
modern instrumentation. DSA, Ultrasound, CT, MRI,
Image registration is the process of overlaying two SPECT, PET etc. have played an important role in
or more images of the same scene taken at different times, diagnosis, research, and treatment. In general, there are two
from different viewpoints, and/or by different sensors. It basic types of medical images [1]: functional images (such
geometrically aligns two images—the reference and sensed as SPECT or PET scans), which provide physiological
images. The present differences between images are information, and anatomical images (such as X-ray,
introduced due to different imaging conditions. Image Ultrasound, CT or MRI), which provide anatomic structure
registration is a crucial step in all image analysis tasks in of the body.The combination process consists of alignment
which the final information is gained from the combination (registration) and data fusion steps. After the combination
of various data sources like in image fusion, change process, the physician could use his knowledge to determine
detection, and multichannel image restoration. Typically, which anatomical areas are normal or abnormal in their
registration is required in remote sensing (multispectral functional characteristics.
classification, environmental monitoring, change detection,
weather forecasting, creating super-resolution images,
integrating information into geographic information
systems, in medicine (combining computer tomography
2. Image Registration Methodology
(CT) and NMR data to obtain more complete information
about the patient, monitoring tumor growth, treatment
verification, comparison of the patient’s data with
anatomical atlases), in cartography (map updating), and in
26 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
Figure 1. Concept of Image registration The detection methods should have good
localization accuracy and should not be sensitive to the
Image registration, as it was mentioned above, is assumed image degradation. In an ideal case, the algorithm
widely used in remote sensing, medical imaging, computer should be able to detect the same features in all projections
vision etc. In general, its applications can be divided into of the scene regardless of the particular image deformation.
four main groups according to the manner of the image Physically corresponding features can be dissimilar due to
acquisition: Different viewpoints (multiview analysis). the different imaging conditions and/or due to the different
Images of the same scene are acquired from different spectral sensitivity of the sensors. The choice of the feature
viewpoints. The aim is to gain larger a 2D view or a 3D description and similarity measure has to consider these
representation of the factors. The feature descriptors should be invariant to the
scanned scene. Examples of applications: Remote sensing— assumed degradations. Simultaneously, they have to be
mosaicing of images of the surveyed area. Computer discriminable enough to be able to distinguish among
vision—shape recovery (shape from stereo). Different times different features as well as sufficiently stable so as not to be
(multitemporal analysis). Images of the same scene are influenced by slight unexpected feature variations and noise.
acquired at different times, often on regular abasis, and The matching algorithm in the space of invariants should be
possibly under different conditions. The aim is to find and robust and efficient. Single features without corresponding
evaluate changes in the scene which appeared between the counterparts in the other image should not affect its
consecutive image acquisitions. Examples of applications: performance. The type of the mapping functions should be
Remote sensing—monitoring of global land usage, chosen according to the a priori known information about
landscape planning. Computer vision—automatic change the acquisition process and expected image degradations. If
detection for security monitoring, motion tracking. Medical no a priori information is available, the model should be
imaging - monitoring of the healing therapy, monitoring of flexible and general enough to handle all possible
the tumor evolution. Different sensors (multimodal analysis) degradations which might appear. The accuracy of the
- Images of the same scene are acquired by different sensors. feature detection method, the reliability of feature
The aim is to integrate the information obtained from correspondence estimation, and the acceptable
different source streams to gain more complex and detailed approximation error need to be considered too. Moreover,
scene representation. the decision about which differences between images have to
Examples of applications: Remote sensing—fusion be removed by registration has to be done. It is desirable not
of information from sensors with different characteristics to remove the differences we are searching for if the aim is a
like panchromatic images, offering better spatial resolution, change detection. This issue is very important and extremely
color/multispectral images with better spectral resolution, or difficult. Finally, the choice of the appropriate type of
radar images independent of cloud cover and solar resampling technique depends on the trade-off between the
illumination. Medical imaging—combination of sensors demanded accuracy of the interpolation and the
recording the anatomical body structure like magnetic computational complexity.
resonance image (MRI), ultrasound or CT with sensors
monitoring functional and metabolic body activities like
positron emission tomography (PET), single photon 3. PET/CT Image Processing Steps
emission computed tomography (SPECT) or magnetic The PET and CT data are acquired using a GE
resonance spectroscopy (MRS). Results can be applied, for discovery STE scanner and exported as DICOM images for
instance, in radiotherapy and nuclear medicine. Scene to the post processing. The specifications and the parameters
model registration. Images of a scene and a model of the of the acquisition device and the PET and CT data are
scene are registered. The model can be a computer provided in [5]. The PET and CT images are aligned as
representation of the scene, for instance maps or digital described below.
elevation models (DEM) in GIS, another scene with similar
content (another patient), ‘average’ specimen, etc. The aim 3.1 Image Segmentation
is to localize the acquired image in the scene/model and/or
to compare them. Examples of applications: Remote In the first step of the segmentation process, the
sensing—registration of aerial or satellite data into maps or image data is low pass filtered to smooth out the boundary
other GIS layers. Computer vision—target template of the brain in the PET as well as the CT images. To
matching with real-time images, automatic quality segment the brain in the PET images, adaptive thresholding
(IJCNS) International Journal of Computer and Network Security, 27
Vol. 2, No. 5, May 2010
is applied based on the minimum and the maximum the brain from two different modalities is calculated and the
intensity of each image, where different intensity levels CT image is shifted based on the average distance of the two
correspond to different anatomical structures in the PET edges.
brain images. The range of intensities representing the brain
is selected after testing the range on several datasets. By
applying an adaptive threshold based on the intensity range
of the brain, unwanted structures are removed from the
image. Figure 2 (a)-(b)show the result of the segmentation
process on the PET data. To segment the brain in the CT
images, the lower and upper limit of the intensity range is
defined for the brain. Starting from the center point of pixels
belonging to this intensity range, a region growing
technique is used and the entire brain region is filled
iteratively from pixel to pixel until the threshold boundary is
reached. Once all of the pixels inside the brain are
iteratively marked, the rest of the image is set to zero.
(b) (c)
(a) (b)
0.6
0.4
0.2
-0.2
-0.4
300
200
200
250 (c)
150
100 100
50
0 0
gives perfect template matching in the given reference a Student member of IEEE Society. He has contributed in
image. The Maximum Cross-correlation coefficient values publishing several research papers in national Journals and
from the NCC plot indicate the perfect template matching Conferences.
with noise and without noise condition. This approach gives
registered image, if the sensed images do not have any Dr. K. Porkumaran is presently Vice-
rotation or scaling. We have also presented a simple and Principal cum Professor and Head, EEE
fast rigid registration method that can be applied to 3DMR Department of Dr. N G P Institute of
images of the brain. It is based on automatic brain Technology, Kalapatti Road, Coimbatore
segmentation, automatic mid-sagittal plane location, image 641035. He has a bachelor’s degree in
alignment. Finally the various transformation techniques are Instrumentation & control Engineering, a
discussed master’s degree in Control systems and Ph.D in Control
Systems. He has 10 years of teaching experience and has
References guided many UG and PG projects. His research and teaching
interests include Modeling and Simulation, Instrumentation
[1] M. V. Green, J. Seidel, J. J. Vaquero, E. Jagoda, I. Lee, System Design, Linear and Non-Linear Control Systems,
and W. C. Eckelman, “High resolution PET, SPECT Process Dynamics and Control, Neural Networks. He is a
and projection imaging in small animals,” Comput. Life member of Indian Society for Technical Education
Med. Imaging Graph., vol. 25, pp. 79–86, 2001. (ISTE) and Associate member of Institution of Engineers
[2] C. A. Pelizzari, G. T. Y. Chen, D. R. Spelbring, R. R. (AMIE).He has published several research papers in
Weichselbaum, and C. T. Chen, “Accurate 3- International and national Journals. He has organized many
dimensional registration of CT, PET, and or MR conferences and chaired several technical session. He is the
images of the brain,” J. Comput. Assist. Tomog., vol. reviewer for international conferences and journals.
13, pp. 20–26, 1989.
[3] R. P. Woods, J. C. Mazziotta, and S. R. Cherry, “MRI-
PET registration with automated algorithm,” J.
Comput. Assist. Tomog., vol. 17, pp. 536–546, 1993.
[4] B. A. Ardekani, M. Braun, B. F. Hutton, I. Kanno, and
H. Iida, “A fully-automatic multimodality image
registration algorithm,” J. Comput.Assist. Tomog., vol.
19, pp. 615–623, 1995.
[5] W. M. Wells, P. Viola, H. Atsumi, S. Nakajima, and S.
Kikinis, “Multi-modal volume registration by
maximization of mutual information,” Med. Image
Anal., vol. 1, pp. 35–51, 1996.
[6] F. Maes, A. Collignon, D. Vandermeulen, G. Marchal,
and P. Suetens, “Multimodality image
registration by maximization of mutual information,” IEEE
Trans. Med. Imaging, vol. 16, pp. 187–198, 1997.
[7] C. Studholme, D. L. G. Hill, and D. J. Hawkes, “An
overlap invariant entropy measure of 3D medical
image alignment,” Pattern Recognit., vol. 32, pp. 71–
86, 1999.
Authors Profile
Manhattan Mobility Model are compared in [2]. These paths, thereby achieving path disjointness. For route
mobility models are compared using link disjoint and node discovery RouteRequest packets are propagated through out
disjoint multipath routing algorithms. With lifetime as the the network thereby establishing multiple paths at
parameter the Gauss Markov mobility model and Random destination node and at the intermediate nodes. Multiples
Waypoint mobility model has highest lifetime at link Loop-Free paths are achieved using the advertised hop count
disjoint and node disjoint multipaths at a low speed. The method at each node. This advertised hop count is required
lifetime of these mobility models decreases as the speed to be maintained at each node in the route table entry. The
increases. For a multi path set size parameter the maximum route entry table at each node also contains a list of next hop
number of paths are generated for city section and random along with the corresponding hop counts. Every node
waypoint at both low speed and high speed. maintains an advertised hop count for the destination.
A performance analysis of AODV and OLSR is done traffic Advertised hop count can be defined as the “maximum hop
load and node density in [3]. The parameters considered are count for all the paths”. Route advertisements of the
packet delivery and the number of collisions per packet. The destination are sent using this hop count. An alternate path
authors show that with less mobility and more traffic load to the destination is accepted by a node if the hop count is
the delivery of packet is highest. But with increase in less than the advertised hop count for the destination. We
mobility and the traffic load the packet delivery decreases. have used the AOMDV implementation for NS-2 provided
This trend is observed for AODV and OLSR both. There is by [8].
an increase in the number of collisions that occurs in the
3.2 DYMO Routing Protocol
network when the number of neighboring node is increased.
This is found to be severe in OLSR routing protocol. DYMO Routing protocol is an improved version of AODV
Popular mobility models like Random Waypoint, Random routing protocol. Routes are established on demand. When a
Walk and Random Direction Mobility Models are compared source node needs to establish a route to the destination the
in [4]. Packet delivery ratio, routing overhead and RouteRequest (RREQ) messages are flooded through out the
throughput are the metrics considered. These mobility network. During broadcasting only those nodes that have
models are compared by varying the number of nodes and not broadcasted previously will forward the messages. The
the mobility speed. For a varying mobile density the RREQ message includes its own address (source address),
Random waypoint has highest packet delivery and sequential number, a hop count and the destination node
throughput and Random direction has the highest overhead. address. A hop count of one is added to a RREQ packet
Increase in mobility speed results in Random Walk having originating at the source node. Each node that forwards the
highest routing overhead. Same as in varying node density RREQ packet adds its own address and the sequence
even with varying mobility speed the highest packet delivery number. After the source node has sent the RREQ packet it
and throughput is associated with Random Waypoint waits for RREQ_WAIT_TIME duration. This is a constant
mobility model. value and is fixed at 1000 milliseconds.
A new group mobility model based on “virtual track” is Processing of RREQ by other nodes is done as follows.
proposed in [5]. This mobility model considers the military When a node receives the RREQ message it is compared
scenarios that can be applied in ad hoc networks. This with its own routing table. If the routing table does not
mobility model is then compared with random waypoint contain any information regarding the source node, then an
mobility model. Through simulation it is shown that this entry is created for the source node. The next hop is to the
group mobility model based on real world scenario provides node from which it had received the packets. This helps in
more realistic results than random way point model. establishing a route back to the originator.
Routing protocols like AODV, FSR, DSR and TORA are
compared in a city section environment [6]. In a city section
model the AODV routing protocol scores over other routing
protocols and FSR comes a close second.
nodes whose addresses are found in the RREQ. After In fig 2 for source node 1 all the neighboring nodes have
updating the route table, the node increments the hop count one hop except from node 2. For source node 2, the number
by a value of one to imply the number of hops the RREQ has of hops is 2 for nodes 3, 4 and 5 while it is 1 for node 1 and
traveled. Once the packet reaches the destination a RREP 6. For source node 3, the number of hops from node 2, 4 and
message is sent back to the source node through the reverse 5 is 2 and 3 hops from node 6 while the number of hops
path that had been created. This RREP packet contains the from remaining nodes is less than 1. For source node 4, the
information regarding its sequence number, number of hop number of hops from 2, 3 and 5 is 2 and for 6 it is 3 while
counts and addresses. The same information is added by all from remaining nodes it is less than 2. For source node 5
the nodes that processes the RREP along the reverse path. the number of hops is less than 2 only for node 1. For source
Consider fig 1. When ever there is a break in the path it has node 6 the number of hops is less than 2 only for node 1.
to be modified by generating a RERR message. Consider These results are obtained by assuming that the scope is less
that the path between node 4 and node 6 is broken. Then than 2. The nodes with less than 2 hops are the nodes with
node 4 has to generate RERR message containing the whom the packets are exchanged at a high rate of frequency
address and sequence number of the node which cannot be than the nodes having a hop count of 2 or more. This may
reached. This RERR is broadcasted through out the network. result in keeping improper information regarding the nodes
When node 2 receives the RERR message it compares the that are far away. But as the packet tends to get closer and
information found in the RERR message with its own closer to the destination node the routes are established [12,
routing table entries. If it has an entry then the route 13, 14].
information has to be removed if the next hop node (i.e.
3.4 Location Aided Routing (LAR) Protocol
node 4) is same as the node from which it had received the
RERR message. When node 3 and node 5 receive the RERR Location Aided Routing (LAR) makes use of the Global
message, the information provided in RERR is checked with Positioning System (GPS). Using GPS the LAR routing
their corresponding entries. Since they do not use node 4 to protocols is able to obtain the information regarding the
reach 6 the RERR message is discarded. In this way by location of a node. In the LAR scheme routes are established
broadcasting the RERR message, the RERR messages the by using the flooding in an intelligent way. When a path is
concerned nodes are informed about any breakage in the broken or when a node needs to establish a route to the
path to the destination node [9, 10, 11]. destination node then it initiates the route discovery by
using RouteRequest (RREQ) messages.
3.3 Fisheye State Routing Protocol (FSR) When an intermediate node receives the same RREQ from
The FSR protocol employs a concept called “FISH EYE” two different nodes then it discards one of the messages and
technique. The idea behind fish eye technique is that a fish’s forwards the other message. When a node receives a RREQ
eye is able to observe maximum number of pixels near its message it compares the address with its own. On finding
focal point. The amount of pixels observed decreases as the that it is the same intended destination node, it triggers a
distance increases from the focal point. If we think of the RouteReply (RREP) in the reverse direction. A timeout is
same concept from a network point of view it means that a specified within which the source node should receive the
node maintains a high quality of information about its RREP packet. Otherwise the source node will trigger the
immediate neighboring nodes while the information route discovery process once again. In the above process the
decreases with the increase in the distance. A node route RREQ is sent to every node. In LAR the number of
maintains up to date information of the link table by packets sent to other nodes is reduced using various
obtaining the latest values from its neighboring nodes. This schemes.
information is exchanged with the neighboring nodes
periodically. These periodical exchanges are not event
driven thus reducing the overhead. When the nodes
exchange the information the entries with the highest
sequence number replaces the entries of a node with the
lowest sequence number. FSR routing protocol uses
something called “SCOPE”. The Scope of a network can be
defined as the group of nodes among whom the
communication takes place frequently as they are with in the (a) (b)
assumed number of hops (or less number of hops).
Figure 3. (a) LAR 1 Scheme (b) LAR 2 Scheme [15]
LAR makes use of a concept called expected zone and
request zone. An expected zone can be defined as a circular
region with in which the destination node is expected to be
present. The source node determines the expected zone of
the destination node based on the average speed and the past
location of the destination node. If the actual speed of the
Figure 2. FSR Demonstration destination node is more than the average speed then the
(IJCNS) International Journal of Computer and Network Security, 33
Vol. 2, No. 5, May 2010
is the initial time and tα is the alpha phase time. mobility models:
Once the object is moving at an acceptable normal speed Packet Delivery Ratio: It is defined as
then it has to maintain this speed for some time. The middle
smooth phase or β phase is represented by
∑ Number of Received Data Packets
[t
α , t β = tα , tα + β∆t ] where β is uniformly distributed ∑ Number of Sent Data Packets
Average Network Delay: It is defined as
over [βmin, βmax]. As we know that every moving object at
∑ (Time packet arrive @dest − Time packet sent @source )
some point of time has to slow down and eventually stop.
Total Number of Connection Pairs
The node reaches γ phase or slow down phase. The node
Throughput of the network: Throughput is defined as
remains in this phase for a random period of time before the
speed reaches to Vγ = 0. Distance between any two points is ∑ Node Throughputs of Data Transmissi on
treated as the Euclidian distance. Trace length is the actual Total Number of Nodes
length traveled by a node during one movement without any Routing Overhead: It is defined as
change in that direction. Distance evolution is the way the
traveling of a node is observed. If the speed of a node
∑[MAC (Control )
pkt sentsize + MAC (Controlpkt ) fwdSize]
ΔT sim
increases continuously then the α phase and β phase of the
trace length is equivalent to the distance evolution. This Table 1: Simulation Parameters
results in a smooth trace. When the speed of the node Simulator NS2
decreases in the γ phase it results in a sharp curve. SMS
mobility model can also be considered as a group mobility Routing Protocols AOMDV, DYMO, FSR, LAR,
model. In this scenario a node is selected as a leader. This OLSR and TORA
group leader selects a speed and direction which the other Mobility Model Community Mobility Model and
nodes do follow. The velocity of a group member m at its SMS Mobility Model
nth step is given by Simulation Time 900
Vnm = VnLeader + (1 − ρ ) ⋅U ⋅ ∆Vmax (sec)
LAR routing protocol has the highest packet delivery at the increase in mobility speed. The delay of FSR is in
varying mobile speed (fig 7). This shows that a location between 0.01 to 0.2 seconds (fig 13). The throughput of
based algorithm like the LAR routing protocol is able to DYMO routing protocols remains steady but it decreases in
take advantage of community model. The LAR routing OLSR and FSR with increase in speed (fig 14).
protocol maintains a delivery ratio of 98% to 70%. Only The throughput results of the routing protocols are in
AOMDV can be comparable to the packet delivery consistent with the packet delivery. But DYMO routing
capability of LAR. But there is an increases and decrease in protocol is having higher overhead over OLSR and FSR
the packet delivery of AOMDV thereby indicating that it is which was not expected considering that DYMO routing
not consistent with the mobility speed of the community protocol performs better than AODV routing protocol (fig
model. 15).
The delay in the network increases with the increase in The routing overhead of DYMO ranges from 45 % to 75 %.
speed. This is in sharp contrast to what can be seen from the This overhead in DYMO is mainly due to the flooding of
increase in node density. This shows that both node density RouteRequest to discover the destination node. This
and mobile speed play a vital role in determining the final RouteRequest message is flooded through out the network
outcome of the performance of a routing protocol (fig 8). thereby causing huge overhead. The overhead in OLSR is
The high throughput of LAR routing protocol results in less mainly due to the large selection of MPR resulting in huge
delay as less number of hops are required to reach the amount of information being sent in the network.
destination. The delay of TORA routing protocol is more In SMS mobility model, the energy consumption of all the
than 0.1 seconds while the delay of LAR routing protocol is routing protocols remains fairly the same with increase in
between 0.01 to 0.02 seconds. The delay in TORA routing mobility speed (fig 16).
protocol may be mainly attributed to the creation of directed
links in the network. The links are created across all the
nodes in the network until the destination node is reached
resulting in higher delay in the network.
The network throughput of LAR routing protocol increases
with increase in mobility speed (fig 9). The results obtained
for LAR routing protocol is ratified with [16] confirming
our stand that the LAR routing protocol goes hand in hand
with the Community model. The throughput of AOMDV
and TORA routing protocols remain between 40 to 50 kbps
while the throughput of LAR routing protocol increases
from 100 to 140 kbps. Figure 7. PDR v/s Average Mobility Speed
The AOMDV and TORA routing protocols score over LAR
routing protocol by having less overhead and less energy
consumption (fig 10 and 11). The routing overhead in LAR
is due to the fact that the expected zone is an estimated one
and sometimes the entire network is considered as an
expected zone. This results in flooding the whole network
with the RouteRequest messages. All the mechanisms
employed in finding the destination zone optimally tends to
fail in this case. This leads to clogging of the whole network Figure 8. Average Network Delay v/s Mobility Speed
thereby increasing the routing packet overhead in the
network. Through simulation it has been showed in
literature that TORA routing protocol consumes more
energy than OLSR. But in the present situation with
Community Mobility model the amount of energy consumed
is almost same between OLSR and TORA routing protocols.
The energy consumption of AOMDV and TORA routing
protocols range between 50 to 100 joules where as the LAR
routing protocol consumes energy between 300 to 350 Figure 9. Network Throughput v/s Mobility Speed
joules.
The DYMO routing protocol has highest packet delivery.
OLSR has an increase and decrease pattern but still it is
comparable with DYMO routing protocol. The packet
delivery is around 90% (fig 12).
FSR routing protocol has least packet delivery. The delay of
OLSR and DYMO routing protocols is between 0.01 to 0.02
seconds. But the delay in FSR is very high. It increases with
(IJCNS) International Journal of Computer and Network Security, 37
Vol. 2, No. 5, May 2010
be the DYMO routing protocol. AOMDV performing better than the LAR routing protocol
at higher node density for a Community Based Mobility
Model. The DYMO routing protocol outperforms both
OLSR and FSR with varying mobility speed and traffic load.
At the end of the day the type of routing protocols that needs
to be deployed depends on the situation in hand in addition
to the various parameters given in our paper.
Acknowledgments
Figure 22. PDR v/s Traffic Load The first author of this paper would like to acknowledge the
help of Tracy Camp, Wenye Wang, Mirco Musolesi in this
work.
References
[1] A.K.Dwivedi et al.,”Performance of Routing Protocols
for Mobile Adhoc and Wireless Sensor Networks: A
Comparative Study”, International Journal of Recent
Trends in Engineering, Vol 2, No. 4, November 2009.
Figure 23. Average Network Delay v/s Traffic Load [2] Nicholas Cooper and Natarajan Meganathan,”Impact
of Mobility Models on Multi Path Routing in Mobile
Ad Hoc Networks”, International Journal of
Computer Networks and Communications (IJCNC),
Vol 2, No 1, January 2010.
[3] Anders Nilsson,”Performance Analysis of Traffic Load
and Node Density in Ad hoc Networks”, In
Proceedings of the 5th European Wireless Conference
(EW2004) Mobile and Wireless Systems, Barcelona,
Figure 24. Network Throughput v/s Traffic Load Spain, Feb 24-27, 2004.
[4] M.I.M Saad et al,“Performance Analysis of Random
Based Mobility Models in MANET Routing
Protocol”, European Journal of Scientific Research,
Vol 32, No. 4, pp 444-454.
[5] Biao Zhou, Kaixin Xu and Mario Gerla,”Group and
Swarm Mobility Models for Ad hoc Network
Scenarios Using Virtual Tracks”, IEEE Military
Communications Conference (MILCOM’04),
Moneterey, California, USA, Oct 31-Nov 3, 2004.
Figure 25. Routing Overhead v/s Traffic Load
[6] Sven Jaap, Marc Bechler and Lars Worf,”Evaluation of
Routing Protocols for Vehicular Ad hoc Networks in
City Traffic Scenarios”, In Proceedings of the 5th
International Conference o Intelligent Transportation
Systems Telecommunications (ITST), Brest, France,
June 2005.
[7] Mahesh K. Marina and Samir R.Das, “ On-Demand
Multipath Distance Vector Routing in Ad-hoc
Networks”, In proceedings of the IEEE International
Figure 26. Average Energy Consumed v/s Traffic
Conference on Network Protocols, 2001 pp 14-23.
Load
[8] M.Caleffi, G.Ferraiuolo and L.Pauro, “On Reliability
of Dynamic Addressing Protocols in Mobile Ad hoc
Networks”, Proceedings of WRECOM’07, Wireless
6 Conclusion Rural and Emergency Communications Conference,
Roma, Italy, October 2007.
In this paper we have evaluated six routing protocols over [9] Ian D Chakeres and Charles E Perkins,”Dynamic
two realistic mobility models. We have conducted these Manet On Demand Routing Protocol”, IETF Internet
experiments in best possible scenarios conductive for an ad Draft, draft-ietf-manet-dymo-12.txt, February 2008.
hoc network to understand the performance of these routing [10] Rolf Ehrenreich Thorup,”Implementing and
protocols over different mobility model. The LAR routing Evaluating the DYMO Routing Protocol”, Master’s
protocol is the protocol of choice for a community model Thesis, Department of Comptuer Science, University
moving at high speed. We came across few surprises like of Aarhus, Denmark.
40 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
[11] Siti Rahayu Abdul Aziz et al.”Performance [25] Vincent D Park and M. Scott Corson,”A Performance
Evaluation of AODV, DSR and DYMO Routing Comparison of the Temporally Ordered Routing
Protocols in MANET”, 2009 International Algorithm and Ideal Link State Routing”, In
Conference on Scientific and Social Science Proceedings of the 3rd IEEE Symposium on
Research, Kaula Lampur, Malaysia, 14-15 March. Computers and Communications, Washington D.C,
[12] Guangyu Pei, Mario Gerla and Tsu Wei USA, 1998.
Chen,”Fisheye State Routing: A Routing Scheme for [26] Vincent D Park, J P Macket and M Scott
Ad Hoc Wireless Networks”, In Proceedings of IEEE Corson,”Applicability of the Temporally Ordered
International Conference on Communications Routing Algorithm for use in Mobile Tactical
(ICC2000), New Orleans, Louisiana, USA, June 18- Networks”, IEEE Military Communications
22, 2000. Conference (MILCOM 98), 1998.
[13] Elizabeth M Royer, Sung-Ju Lee and Charles E [27] Eryk Dutkiewicz et al,”A Review of Routing
Perkins,”The Effects of MAC Protocols on Ad hoc Protocols for Mobile Ad Hoc Networks”, Ad hoc
Network Communication”, In Proceedings of IEEE Networks, Volume 2, Issue 1, January 2004, Pages 1-
WCNC 2000, Chicago, IL, September 2000. 22.
[14] Allen C Sun,”Design and Implementation of Fisheye [28] Newman and Girvan,”Finding and evaluating
Routing Protocol for Mobile Ad hoc Networks”, community structure in networks”, Physical Review
http://www.scanner-group.mit.edu/PDFS/SunA.pdf E, 69, Februrary 2004.
[15] C.Siva Rama Murthy and B.S. Manoj, “Adhoc [29] Mirco Musolesi and Cecilia Mascolo, “A Community
Wireless Networks: Architectures and Protocols”, Based Mobility Model for Ad hoc Network Research”,
Second Edition, Prentice Hall. International Symposium on Mobile Ad hoc
[16] Tracy Camp, Jeff Boleng et al.,”Performance Networking and Computing, Proceedings of the 2nd
Comparison of Two Location Based Routing International Workshop on Multi hop ad hoc
Protocols for Ad Hoc Networks”, In Proceedings of networks, 2006.
the IEEE 21st Annual Joint Conference of the IEEE [30] Ming Zhao and Wenye Wang,”A Novel Semi Markov
Computer and Communications Societies (INFOCOM Smooth Mobility Model for Mobile Ad Hoc
2002), pp 167801687, 2002. Networks”, In Proceedings of the IEEE
[17] Young Bae Ko and Nitin H Vaidya,”Location Aided GLOBECOM’06, San Francisco, CA, November
Routing (LAR) in Mobile Ad Hoc Networks”, In 2006.
Proceedings of the 4th Annual ACM/IEEE [31] Maig Zhao and Wenye Wang,”Design and
International Conference on Mobile Computing and Applications of a Smooth Mobility Model for Mobile
Networking, Dallas, Texas, United States, 1998. Ad hoc Networks”, In Proceedings of the IEEE
[18] Ioannis Broustis, Gentian Jakllari et al.,”A Milcom’06, Washington D.C., October 2006.
Comprehensive Comparison of Routing Protocols for [32] Juan Carlos Cano and Pietro Manzoni,”A
Large Scale Wireless MANETs”, In Proceedings of Performance Comparison of Energy Consumption for
the 3rd Inernational Workshop on Wireless Ad Hoc Mobile Ad Hoc Network Routing Protocols”, In
and Sensor Networks, Newyork, June 28-30 2006. Proceedings of the 8th International Symposium on
[19] S.K.Sarkar, T.G.Basavaraju et al,”Ad hoc Mobile Modelling, Analysis and Simulation of Computer and
Wireless Networks, Principles, Protocols and Telecommunication Systems, San Francisco, CA,
Applications”, Auerbach Publications -2008. 2000.
[20] T.Clausen and P.Jacquet,”Optimized Link State [33] Laura Marie Feeney,”Investigating the Energy
Routing Protocol for Ad hoc Networks”, Consumption of a Wireless Network Interface in an
http://hipercom.inria.fr/olsr/rfc3626.txt Adhoc Networking Environment”, In Proceedings of
[21] Mounir Frikha and Manel Maamer,” Implementation the INFOCOM01 Twentieth Annual Joint Conference
and Simulation of OLSR protocol with QOS in ad hoc of the IEEE Computer and Communication Societies,
Networks”, Proceedings of the Second International Vol 3, August 2002, pp 1548-1557.
Symposium on Communications, Control and Signal [34] Information Sciences Institute, “The Network
Processing (ISCCSP’06), 13-15 March, Marrakech, Simulator Ns-2”, Http://www.isi.edu/nanam/ns/,
Morocco. University of Southern California.
[22] John Novatnack and Harpreet Arora,”Evaluating ad
hoc Routing Protocols with Restpect to Quality of
Service”, IEEE International Conference on Wireless
and Mobile Computing, Networking and
Communications (WiMob’05), 03 October 2005.
[23] Kenneth Holter,”Comparing AODV and OLSR”,
folk.uio.no/kenneho/studies/essay/essay.html
[24] Ying Ge et al,”Quality of Service Routing in ad Hoc
Networks Using OLSR”, Proceedings of the 36th
IEEE International conference on System Sciences
(HICSS’03), 2002.
(IJCNS) International Journal of Computer and Network Security, 41
Vol. 2, No. 5, May 2010
2
Faculty of Telecommunication and Information Engineering, UET TAXILA,
Pakistan
asim.shahzad@uettaxila.edu.pk
systems but there are practical lower bounds on the delay of reassembling these packets is non-trivial, especially when
VOIP. The ITU-T Recommendation G.114 establishes a dealing with the tight time constraints of VOIP.
number of time constraints on one-way latency. The upper When jitter is high, packets arrive at their destination in
bound is 150 ms in [2] for one-way traffic experienced in spurts. The general prescription to control jitter at VOIP
domestic calls across PSTN lines in the continental United endpoints is the use of a buffer, but such a buffer has to
States. For international calls, a delay of up to 400 ms was release its voice packets at least every 150 ms (usually a lot
deemed tolerable, but since most of the added time is spent sooner given the transport delay) so the variations in delay
routing and moving the data over long distances, we must be bounded. The buffer implementation issue is
consider here only the domestic case and assume our compounded by the uncertainty of whether a missing packet
solutions are upwards compatible in the international realm. is simply delayed an anomalously long amount of time, or is
VOIP calls must achieve the 150 ms bound to successfully actually lost.
emulate the QoS that today’s phones provide. It places a Jitter can also be controlled throughout the VOIP network
genuine constraint on the amount of security that can be by using routers, firewalls, and other network elements that
added to a VOIP network. The encoding of voice data can support QoS. These elements process and pass along time
take between 1 and 30 ms and voice data traveling across urgent traffic like VOIP packets sooner than less urgent data
the North American continent can take upwards of 100 ms packets. Another method for reducing delay variation is to
although actual travel time is often much faster . Assuming pattern network traffic to diminish jitter by making as
the worst case (100 ms transfer time), 20 –50 ms remain for efficient use of the bandwidth as possible. This constraint is
queuing and security implementations [4]. at odds with some security measures in VOIP. Chief among
these is IPSec, whose processing requirements may increase
latency, thus limiting effective bandwidth and contributing
to jitter.
The window of delivery for a VOIP packet is very small, so
it follows that the acceptable variation in packet delay is
even smaller. Thus, although we are concerned with
security, the utmost care must be given to assuring that
delays in packet deliveries caused by security devices are
kept uniform throughout the traffic stream [4].
Such bandwidth is not always accessible and the redundant into a VOIP network is not feasible, particularly when VOIP
information will have to be processed, introducing even is integrated into existing data networks. Instead, these data-
more latency to the system and ironically, possibly network solutions must be adapted to support security in the
producing even greater packet loss. new fast paced world of VOIP.
2.4 Bandwidth & Effective Bandwidth 2.6 Power Failure and Backup Systems
As in data networks, bandwidth congestion can cause packet Conventional telephones operate on 48 volts supplied by the
loss and a host of other QoS problems. Thus, proper telephone line itself. This is why home telephones continue
bandwidth reservation and allocation is essential to VOIP to work even during a power failure. Most offices use PBX
quality. One of the great attractions of VOIP, data and voice systems with their conventional telephones, and PBXs
sharing the same wires, is also a potential headache for require backup power systems so that they continue to
implementers who must allocate the necessary bandwidth operate during a power failure [4]. These backup systems
for both networks in a system normally designed for one. will continue to be required with VOIP, and in many cases
Congestion of the network causes packets to be queued, will need to be expanded. An organization that provides
which in turn contributes to the latency of the VOIP system. uninterruptible power systems for its data network and
Low bandwidth can also contribute to non-uniform delays desktop computers may have much of the power
(jitter) [1], since packets will be delivered in spurts when a infrastructure needed to continue communication functions
window of opportunity opens up in the traffic. during power outages, but a careful assessment must be
Because of these issues, VOIP network infrastructures must conducted to ensure that sufficient backup power is available
provide the highest amount of bandwidth possible. On a for the office VOIP switch, as well as each desktop
LAN, this means having modern switches running at 100M instrument. Costs may include electrical power to maintain
bit/sec and other architectural upgrades that will alleviate UPS battery charge, periodic maintenance costs for backup
bottlenecks within the LAN. Percy and Hommer suggest power generation systems, and cost of UPS battery
that if network latencies are kept below 100 milliseconds, replacement. If emergency/backup power is required for
maximum jitter never more than 40 milliseconds, then more than a few hours, electrical generators will be
packet loss should not occur. With these properties assured, required. Costs for these include fuel, fuel storage facilities,
one can calculate the necessary bandwidth for a VOIP and cost of fuel disposal at end of storage life.
system on the LAN in a worst case scenario using statistics
associated with the worst-case bandwidth congesting codec. 2.7 QoS Implications for Security
This is fine when dealing simply with calls across the LAN, The strict performance requirements of VOIP have
but the use of a WAN complicates matters. Bandwidth usage significant implications for security, particularly denial of
varies significantly across a WAN, so a much more complex service (DoS) issues. VOIP-specific attacks (i.e., floods of
methodology is needed to estimate required bandwidth specially crafted SIP messages) may result in DoS for many
usage. VOIP-aware devices. For example, SIP phone endpoints
Methods for reducing the bandwidth usage of VOIP include may freeze and crash when attempting to process a high rate
RTP header compression and Voice Activity Detection of packet traffic SIP proxy servers also may experience
(VAD). RTP compression condenses the media stream failure and intermittent log discrepancies with a VOIP-
traffic so less bandwidth is used. However, an inefficient specific signaling attack of under 1Mb/sec. In general, the
compression scheme can cause latency or voice degradation, packet rate of the attack may have more impact than the
causing an overall downturn in QoS. VAD prevents the bandwidth; i.e., a high packet rate may result in a denial of
transmission of empty voice packets (i.e. when a user is not service even if the bandwidth consumed is low [4].
speaking, their device does not simply send out white noise). 3. Standards and Protocols
However, by definition VAD will contribute to jitter in the
system by causing irregular packet generation [4]. Over the next few years, the industry will address the
bandwidth limitations by upgrading the Internet backbone to
2.5 The Need for Speed asynchronous transfer mode (ATM), the switching fabric
The key to conquering QoS issues like latency and designed to handle voice, data, and video traffic. Such
bandwidth congestion is speed. By definition, faster network optimization will go a long way toward eliminating
throughput means reduced latency and probabilistically network congestion and the associated packet loss. The
reduces the chances of severe bandwidth congestion. Thus Internet industry also is tackling the problems of network
every facet of network traversal must be completed quickly reliability and sound quality on the Internet through the
in VOIP. The latency often associated with tasks in data gradual adoption of standards. Standards-setting efforts are
networks will not be tolerated. Chief among these latency focusing on the three central elements of Internet telephony:
producers that must improve performance are firewall/NAT the audio codec format; transport protocols; and directory
traversal and traffic encryption/decryption [4]. Traditionally, services.
these are two of the most effective ways for administrators to
secure their networks. However, they are also two of the In May 1996, the International Telecommunications Union
greatest contributors to network congestion and throughput (ITU) ratified the H.323 specification, which defines how
delay. Inserting traditional firewall and encryption products
44 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
voice, data, and video traffic will be transported over IP– without passing through the firewall. The introduction of
based local area networks; it also incorporates the T.120 firewalls to the VOIP network complicates several aspects of
data conferencing standard. The recommendation is based VOIP, most notably dynamic port trafficking and call setup
on the real-time protocol/real-time control protocol procedures.
(RTP/RTCP) for managing audio and video signals.
As such, H.323 addresses the core Internet-telephony 4.2 Stateful Firewalls
applications by defining how delay-sensitive traffic, (i.e., Most VOIP traffic travels across UDP ports. Firewalls
voice and video), gets priority transport to ensure real-time typically process such traffic using a technique called packet
communications service over the Internet [1]. filtering. Packet filtering investigates the headers of each
packet attempting to cross the firewall and uses the IP
addresses, port numbers, and protocol type contained therein
to determine the packets’ legitimacy [4]. In VOIP and other
media streaming protocols, this information can also be used
to distinguish between the start of a connection and
established connection.There are two types of packet
filtering firewalls, stateless and stateful. Stateless firewalls
retain no memory of traffic that has occurred earlier in the
session. Stateful firewalls do remember previous traffic and
can also investigate the application data in a packet. Thus,
stateful firewalls can handle application traffic that may not
be destined for a static port.
the RTMM must be able to perform NAT at media speeds, Intrusion detection is a second line of defense behind other
to prevent latency/jitter or loss of packets [11]. security mechanisms (Firewalls, Encryption). Supplying the
All of the benefits of NAT come at a price. NATs “violate detector engine with application specific knowledge makes
the fundamental semantic of the IP address, that it is a it more effective and powerful.
globally reachable point for communications”. This design An Intrusion Detection System (IDS) helps administrators to
has significant implications for VOIP. For one thing, an monitor and defend against security breaches. Intrusion
attempt to make a call into the network becomes very detection techniques are generally divided into two
complex when a NAT is introduced. The situation is paradigms, anomaly detection and misuse detection. In
anomaly detection techniques, the deviation from normal
analogous to a phone network where several phones have
system behaviors is detected, whereas misuse detection is
the same phone number, such as in a house with multiple
based on the matching of attack signatures. Unlike
phones on one line (see Figure 4). There are also several signature-based intrusion detection, anomaly detection has
issues associated with the transmission of the media itself the advantage of detecting previously-unknown attacks but
across the NAT, including an incompatibility with IPSec at the cost of relatively high false alarm rate.
[4]. Sekar etal introduced a third category of specification-based
intrusion detection. Specification-based approach takes the
manual development of a specification that captures
legitimate system behavior and detects any deviation
thereof. This approach can detect unseen attacks with low
false alarm rate. However, these previous approaches fall
short of defending VoIP applications, because of the cross-
protocol interaction and distributed nature of VoIP.
Problem:
• Simple NAT devices, which are not VoIP aware, perform
NAT on the IP headers only.
• VoIP packets contain private IP address in the payload.
• Therefore, VoIP sessions cannot be established.
transaction messages between the accounting software and in successfully incorporating IPsec encryption into VOIP
the database, and the RTP flows later on. Specifically, each services.
of the following three conditions must hold [12]. 5.1 IPSec
1. The SIP message should follow the correct format. IPsec is the preferred form of VPN tunneling across the
2. When the accounting software sends out a transaction Internet. There are two basic protocols defined in IPsec:
to denote a call from user A to user B, check if user A has Encapsulating Security Payload (ESP) and Authentication
sent a SIP Call Initialization message to user B. If user A Header (AH) (see Figure 7). Both schemes provide
has not set up the call with a legitimate SIP Call connectionless integrity, source authentication, and an anti-
Initialization message, then this condition will be violated.
replay service. The tradeoff between ESP and AH is the
3. Check the source/destination IP addresses of the
increased latency in the encryption and decryption of data in
subsequent RTP flows. Together with information from
ESP and a “narrower” authentication in ESP, which
DNS and SIP Location Servers, we can reconfirm that each
RTP flow has a corresponding legitimate call setup. normally does not protect the IP header “outside” the ESP
header, although Internet Key Exchange (IKE: Responsible
4.3.2 Stateful Methodology for Detection for key agreement using public key cryptography) can be
A second abstraction useful for VoIP systems in particular is used to negotiate the security association (SA), which
stateful detection. Stateful detection implies building up includes the secret symmetric keys. In this case, the
relevant state within a session and across sessions and using addresses in the header (transport mode) or new/outer
the state in matching for possible attacks. It is important header (tunnel mode) are indirectly protected, since only the
that the state aggregation be done efficiently so that the entity that negotiated the SA can encrypt/decrypt or
technique is applicable in high throughput systems, such as authenticate the packets. Both schemes insert an IPsec
VoIP systems. header (and optionally other data) into the packet for
A VoIP system maintains considerable amount of system purposes, such as authentication [4].
state. The client side maintains state about all the active IPsec also supports two modes of delivery: Transport and
connections – when the connection was initiated, when it Tunnel. Transport mode encrypts the payload (data) and
can be torn down, and what the properties of the connection upper layer headers in the IP packet. The IP header and the
are. The server side also maintains state relevant to billing,
new IPsec header are left in plain sight. So if an attacker
such as the duration of the call.
were to intercept an IPsec packet in transport mode, they
could not determine what it contained; but they could tell
5. Encryption and IPsec where it was headed, allowing rudimentary traffic analysis.
The only focus on security of the network, protecting On a network entirely devoted to VOIP, this would equate to
endpoints, other components, from malicious attacks is not logging which parties were calling each other, when, and
enough for VoIP systems. Firewalls, IDS, NAT, and other for how long. Tunnel mode encrypts the entire IP datagram
such devices can help keep intruders from compromising a and places it in a new IP Packet [9]. Both the payload and
network, but firewalls are no defense against an internal the IP header are encrypted. The IPsec header and the new
hacker in [4]. Another layer of defense is necessary at the IP Header for this encapsulating packet are the only
protocol level to protect the data itself. In VOIP, as in data information left in the clear.
networks, this can be accomplished by encrypting the
packets at the IP level using IPsec. This way if anyone on
the network, authorized or not, intercepts VOIP traffic not
intended for them, these packets will be impossible to
understand. The IPsec suite of security protocols and
encryption algorithms is the standard method for securing
packets against unauthorized viewers over data networks
and will be supported by the protocol stack in IPv6. Hence,
it is both logical and practical to extend IPsec to VOIP,
encrypting the signal and voice packets on one end and
decrypting them only when needed by their intended
recipient.
Also, several factors, including the expansion of packet size,
ciphering latency, and a lack of QoS urgency in the
cryptographic engine itself can cause an excessive amount of
latency in the VOIP packet delivery. This leads to degraded
voice quality, so once again there is a tradeoff between Figure 7. IPsec Tunnel and Transport Modes [9]
security and voice quality, and a need for speed. Fortunately,
the difficulties are not insurmountable. IPsec can be IPSec supports the Triple DES encryption algorithm (168-
incorporated into a SIP network with roughly a three-second bit) in addition to 56-bit encryption. Triple DES (3DES) is a
additional delay in call setup times, an acceptable delay for strong form of encryption that allows sensitive information
many applications. This section explains the issues involved to be transmitted over untrusted networks. It enables
(IJCNS) International Journal of Computer and Network Security, 47
Vol. 2, No. 5, May 2010
customers, particularly in the finance industry, to utilize to require fragmentation and the corresponding reassembly
network layer encryption. of IPSec datagrams.
2. Encrypted packets are probably authenticated, which
means that there are two cryptographic operations that are
Table 1: Encryption Algorithms for VoIP [9] performed for every packet.
3. The authentication algorithms are slow, although work
Encryption Algorithms for IPSec
Type Key Strengt
has been done to speed up things as the Diffie−Hellman
h computations [4].
DES Symmetric 56-bit Weak
3DES Symmetric 168-bit Mediu 5.5 Difficulties Arising from VOIPsec
m
AES Symmetric 128, 192, or 256- Strong IPSec has been included in IPv6. It is a reliable, robust, and
bit widely implemented method of protecting data and
RSA Asymmetric 1024-bit minimum Strong authenticating the sender. However, there are several issues
associated with VOIP that are not applicable to normal data
5.2 The Role of IPSec in VOIP traffic. Of particular interest are the Quality of Service
The prevalence and ease of packet sniffing and other (QoS) issues and some others like latency, jitter, and packet
techniques for capturing packets on an IP based network loss. These issues are introduced into the VOIP environment
makes encryption a necessity for VOIP. Security in VOIP is because it is a real time media transfer, with only 150 ms to
concerned both with protecting what a person says as well as deliver each packet. In standard data transfer over TCP, if a
to whom the person is speaking. IPSec can be used to packet is lost, it can be resent by request. In VOIP, there is
achieve both of these goals as long as it is applied with ESP no time to do this. Packets must arrive at their destination
using the tunnel method. This secures the identities of both and they must arrive fast. Of course the packets must also be
the endpoints and protects the voice data from prohibited secure during their travels. However, the price of this
users once packets leave the corporate intranet. The security is a decisive drop in QoS caused by a number of
incorporation of IPSec into Ipv6 will increase the factors.
availability of encryption, although there are other ways to A study by researchers focused on the effect of VOIPsec on
secure this data at the application level. VOIPsec (VOIP various QoS issues and on the use of header compression as
using IPSec) helps reduce the threat of man in the middle a solution to these problems. They studied several codecs,
attacks, packet sniffers, and many types of voice traffic encryption algorithms, and traffic patterns to garner a broad
analysis. Combined with the firewall implementations, Ipsec description of these effects. Some empirical results
makes VOIP more secure than a standard phone line. It is developed by Cisco are available as well in [4].
important to note, however, that Ipsec is not always a good Delay
fit for some applications, so some protocols will continue to • Processing—PCM to G.729 to packet
rely on their own security features. • Encryption — ESP encapsulation + 3DES
• Serialization — time it takes to get a packet out of the
5.3 Local VPN Tunnels router, each “hop” generally has fixed delay.
Virtual Private Networks (VPNs) are “tunnels” between two • IPsec overhead: about 40 bytes (depending configuration)
endpoints that allow for data to be securely transmitted • IP header: 20 bytes
between the nodes. The IPSec ESP tunnel is a specific kind • UDP + RTP headers: 20 bytes
of VPN used to traverse a public domain (the Internet) in a • RTP header compression: 3 bytes for IP+UDP+RTP
private manner. The use and benefits of VPNs in IPSec have
been great enough for some to claim “VOIP is the killer app
Effects on 8 kbps CODEC (voice data: 20 bytes)
for VPNs”. VPN tunnels within a corporate LAN or WAN
• clear text voice has an overhead of 3 bytes, which
are much more secure and generally faster than the IPSec
suggests required bandwidth of approximately 9 kbps
VPNs across the Internet because data never traverses the
• IPsec encrypted voice: overhead 80 bytes and required
public domain, but they are not scaleable [4]. Also, no
bandwidth 40 kbps
matter how the VPN is set up, the same types of attacks and
issues associated with IPSec VPNs are applicable, so we
5.6 Encryption / Decryption Latency
consider here only the case of IPSec tunneling and assume
The cryptographic engine is bottleneck for voice traffic
the security solutions can be scaled down to an internal
transmitted over IPsec. The driving factor in the degraded
network if needed.
performance produced by the cryptography was the
scheduling algorithms in the crypto-engine itself However,
5.4 Memory and CPU Considerations
there still was significant latency due to the actual
Packets that are processed by IPSec are slower than packets
encryption and decryption.
that are processed through classic crypto. There are several
Encryption/decryption latency is a problem for any
reasons for this and they might cause significant
cryptographic protocol, because much of it results from the
performance problems:
computation time required by the underlying encryption.
1. IPSec introduces packet expansion, which is more likely
With VOIP’s use of small packets at a fast rate and
48 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
intolerance for packet loss, maximizing throughput is phones is making endpoint encryption less of an issue.
critical. However, this comes with a price, because although
6.2 Secure Real Time Protocol (SRTP)
DES is the fastest of these encryption algorithms, it is also
the easiest to crack [4]. Thus, designers are once again The Secure Real-time Protocol is a profile of the Real-time
forced to toe the line between security and voice quality. Transport Protocol (RTP) offering not only confidentiality,
but also message authentication, and replay protection for
5.7 Expanded Packet Size the RTP traffic as well as RTCP (Real-time Transport
IPsec also increases the size of packets in VOIP, which leads Control Protocol).
to more QoS issues. The increase is actually just an increase SRTP provides a framework for encryption and message
in the header size due to the encryption and encapsulation of authentication of RTP and RTCP streams. SRTP can
the old IP header and the introduction of the new IP header achieve high throughput and low packet expansion.
and encryption information. This leads to several SRTP is independent of a specific RTP stack
complications when IPsec is applied to VOIP. First, the implementation and of a specific key management standard,
effective bandwidth is decreased as much as 63% [4]. Thus but Multimedia Internet Keying (MIKEY) has been
connections to single users in low bandwidth areas (i.e. via designed to work with SRTP.
modem) may become infeasible. The size discrepancy can In comparison to the security options for RTP there are
also cause latency and jitter issues as packets are delayed by some advantages to using SRTP. The advantages over the
decreased network throughput or bottlenecked at hub nodes RTP standard security and also over the H.235 security for
on the network (such as routers or firewalls). media stream data are listed below [4, 12].
5.8 IPsec and NAT Incompatibility SRTP provides increased security, achieved by
• Confidentiality for RTP as well as for RTCP by
IPsec and NAT compatibility is far from ideal. NAT
encryption of the respective payloads;
traversal completely invalidates the purpose of AH because
• Integrity for the entire RTP and RTCP packets, together
the source address of the machine behind the NAT is
with replay protection;
masked from the outside world. Thus, there is no way to
• The possibility to refresh the session keys periodically,
authenticate the true sender of the data. The same reasoning
which limits the amount of cipher text produced by a fixed
demonstrates the inoperability of source authentication in
key, available for an adversary to cryptanalyze;
ESP. We have defined this as an essential feature of
• An extensible framework that permits upgrading with
VOIPsec, so this is a serious problem. There are several
new cryptographic algorithms;
other issues that arise when ESP traffic attempts to cross a
• A secure session key derivation with a pseudo-random
NAT. If only one of the endpoints is behind a NAT, the
function at both ends;
situation is easier. If both are behind NATs, IKE negotiation
• The usage of salting keys to protect against pre-
can be used for NAT traversal, with UDP encapsulation of
computation attacks;
the IPsec packets.
• Security for unicast and multicast RTP applications.
MIKEY has some important properties in [4]: transmission. In particular, when cIPsec is adopted, the
• MIKEY can be implemented as an independent software average packet size is only 2% bigger, rather than 50%
library to be easily integrated in a multimedia longer plain VoIPsec packets, which makes VoIPsec and
communication protocol. It offers independency of a specific VoIP equivalent from the bandwidth usage point of view.
communication protocol (SIP, H.323, etc.) However, packet loss does have an exacerbated detrimental
• Establishment of key material within a 2-way handshake, effect on packets compressed under the cIPsec scheme.
therefore best suited for real-time multimedia scenarios When packets are lost, they cannot be re-sent and the
There are four options for Key Distribution: endpoints need to resynchronize. However, the time saved in
• Pre-shared key the crypto-engine and the security provided may be well
• Public-key encryption worth this price of this approach [5].
• Diffie-Hellman key exchange protected by public-key
encryption 6.6 Resolving NAT/IPSec Incompatibilities
• Diffie-Hellman key exchange protected with pre-shared- The most likely widespread solution to the problem of NAT
key and keyed hash functions (using an MIKEY extension traversal is UDP encapsulation of IPSec. This
(DHHMAC)) implementation is supported by the IETF and effectively
allows all ESP traffic to traverse the NAT. In tunnel mode,
6.4 Better Scheduling Schemes this model wraps the encrypted IPSec packet in a UDP
Without a way for the crypto-engine to prioritize packets, packet with a new IP header and a new UDP header, usually
the engine will still be susceptible to DoS attacks and using port 500. This port was chosen because it is currently
starvation from data traffic impeding the time-urgent VOIP used by IKE peers to communicate so overloading the port
traffic. A few large packets can clog the queue long enough does not require any new holes to be punched in the
to make the VOIP packets over 150 ms late (sometimes firewall. The SPI field within the UDP-encapsulated packet
called head-of-line blocking), effectively destroying the call. is set to zero to differentiate it from an actual IKE
Ideally, the crypto-engine would implement QoS scheduling communication. This solution allows IPsec packets to
to favor the voice packets, but this is not a realistic scenario traverse standard NATs in both directions. The adoption of
due to speed and compactness constraints on the crypto- this standard method should allow VOIPsec traffic to
engine. One solution implemented in the latest routers is to traverse NATs cleanly, although some extra overhead is
schedule the packets with QoS in mind prior to the added in the encapsulation/decapsulation process. IKE
encryption phase. negotiation will also be required to allow for NAT traversal.
It is not surprising that for voice traffic the crypto-engine The problem still remains that IP-based authentication of
can be a serious bottleneck. Rather than the expected the packets cannot be assured across the NAT, (although
constraints on the crypto-engine throughput, the critical fully qualified domain names could be used) but the use of a
factor turned out to be the impossibility to control and shared secret (symmetric key) negotiated through IKE could
schedule access to the crypto-engine so as to favor real-time provide authentication. It is important to note that IP-based
traffic over regular one. This applies regardless of whether authentication is weak compared with methods using
the scheduler is implemented as a software module or a cryptographic protocols.
hardware component. Therefore, if voice traffic is There are several other solutions to the IPsec/NAT
interleaved with other types of traffic, e.g., ftp or http traffic, incompatibility problem, including Realm-Specific IP
during a secure session, it may happen that the latter RSIP), IPv6 Tunnel Broker, IP Next Layer (IPNL), and UDP
(usually characterized by big packets) is scheduled in the encapsulation. RSIP is designed as a replacement for NAT
crypto-engine before voice traffic. In this case voice traffic and provides a clear tunnel between hosts and the RSIP
might be delayed to the point that packets are discarded Gateway. RSIP supports both AH and ESP, but
most of the times. implementing RSIP would require a significant overhaul of
the current LAN architecture so while it is quite an elegant
6.5 Compression of Packet Size solution, it is currently infeasible. Perhaps as a result of
these problems, RSIP is not widely used. The IPv6 tunnel
Compression of Packet Size turn results in considerably less
broker method uses an IPv6 tunnel as an IPSec tunnel, and
jitter, latency, and better crypto-engine performance. There
encapsulates an IPv6 packet in an IPv4 packet. But this
is, of course, a price for these speedups. The compression
solution also requires LAN upgrades and doesn’t work in
scheme puts more strain on the CPU and memory
capabilities of the endpoints in order to achieve the situations where multiple NATs are used. IPNL introduces a
compression, and, of course, both ends of a connection must new layer into the network protocols between IP and
TCP/UDP to solve the problem, but IPNL is in competition
use the same compression algorithm.
with IPv6 and IPv6 is a much more widely used standard
Efficient solution for packet header compression is known
[4].
as cIPsec, for VoIPsec traffic. Simulation results from
different sources show that the proposed compression
scheme significantly reduces the overhead of packet headers,
thus increasing the effective bandwidth used by the
50 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
7. Security Threats in VoIP degrading the service. This causes calls to drop prematurely
and halts call processing.
In the early days of VoIP, there was no big concern about
Why would someone launch a DoS attack? Once the target
security issues related to its use. People were mostly
is denied of the service and ceases operating, the attacker
concerned with its cost, functionality and reliability. Now
can get remote control of the administrative facilities of the
that VoIP is gaining wide acceptance and becoming one of
system [10, 12].
the mainstream communication technologies, security has
become a major issue. 7.5 Spamming over Internet Telephony (SPIT)
The security threats cause even more concern when we think If you use email regularly, then you must know what
that VoIP is in fact replacing the oldest and most secure spamming is. Put simply, spamming is actually sending
communication system the world ever known – POTS (Plain emails to people against their will. These emails consist
Old Telephone System). Let us have a look at the threats mainly of online sales calls. Spamming in VoIP is not very
VoIP users face. common yet, but is starting to be, especially with the
emergence of VoIP as an industrial tool.
7.1 Identity and Service Theft Every VoIP account has an associated IP address. It is easy
Service theft can be exemplified by phreaking, which is a for spammers to send their messages (voicemails) to
type of hacking that steals service from a service provider, thousands of IP addresses. Voice mailing as a result will
or use service while passing the cost to another person. suffer. With spamming, voicemails will be clogged and
Encryption is not very common in SIP, which controls more space as well as better voicemail management tools
authentication over VoIP calls, so user credentials are will be required. Moreover, spam messages can carry viruses
vulnerable to theft. and spyware along with them. [10]
Eavesdropping is how most hackers steal credentials and 7.6 Call Tampering
other information. Through eavesdropping, a third party can Call tampering is an attack which involves tampering a
obtain names, password and phone numbers, allowing them phone call in progress. For example, the attacker can simply
to gain control over voicemail, calling plan, call forwarding spoil the quality of the call by injecting noise packets in the
and billing information. This subsequently leads to service communication stream. He can also withhold the delivery of
theft. packets so that the communication becomes spotty and the
Stealing credentials to make calls without paying is not the participants encounter long periods of silence during the
only reason behind identity theft. Many people do it to get call.
important information like business data.
A phreaker can change calling plans and packages and add 7.7 Man-in-the-Middle Attacks
more credit or make calls using the victim’s account. He can
of course as well access confidential elements like voice VoIP is particularly vulnerable to man-in-the-middle
mail, do personal things like change a call forwarding attacks, in which the attacker intercepts call-signaling SIP
number [10]. message traffic and masquerades as the calling party to the
called party, or vice versa. Once the attacker has gained this
position, he can hijack calls via a redirection server [10].
7.2 Vishing
Vishing is another word for VoIP Phishing, which involves
8. Conclusion
a party calling you faking a trustworthy organization (e.g.
This paper provides an overview of the VoIP and of the
your bank) and requesting confidential and often critical issues related to VoIP security. There are numerous
information. challenges to the secure implementation, deployment and
use of the VoIP on a same network which is used to carry
7.3 Viruses and Malware other network data. These challenges are related to the
VoIP utilization involving soft-phones and software are service availability, Quality of Service, Firewalls, Intrusion
vulnerable to worms, viruses and malware, just like any Detection, IPSec, VoIP threats, and privacy. The next step
Internet application. Since these soft-phone applications run for the ongoing research project is to identify specific areas
that will be addressed in the near term research efforts.
on user systems like PCs and PDAs, they are exposed and
vulnerable to malicious code attacks in voice applications.
Abstract: This paper provides a novel run-time approach for object access is used in SWAT [15]. SWAT is an automatic
memory leak detection and recovery (MLDR) using aging in dynamic memory leak detection tool that traces memory
physical memory space. MLDR algorithm reflects both the allocations/frees to build a heap model and uses an adaptive
physical and virtual behavior of memory allocation and benefits statistical profiling approach to monitor loads/stores to these
from the hardware support available for tracking physical pages objects. Objects that are not accessed for a long time are
in real memory in order to detect leak in virtual address space. reported as leaks. This tool detects leaks with a less than
MLDR detects stale and unreachable memory objects. A
10% false positive rate. Leaks are reported at the end of an
conservative implementation of MLDR recovers only
application run; SWAT can not decide on its own to
unreachable leaky objects in order to avoid the potential false
positives stale objects. MLDR has a coarse granularity in the deallocate them.
sense that it deals with pages rather than memory chunks, thus Different businesses can deal with software failures in
enabling it to avoid costly scans of heap allocation. Simulation different ways. In some cases, system administrators simply
results show the effectiveness of MLDR in detecting and restart the system whenever the memory leaks to a point
recovering memory leaks at run time. where a crash is imminent or performance degrades beyond
an acceptable level. Systems with critical applications can
Keywords: memory leak detection, memory leak recovery, not tolerate the cost of frequent shutdowns or performance
memory aging, and dynamic memory management. degradation. For example, rebooting and restarting the
software systems such as the Patriot Software [3] every 8
1. Introduction hours was responsible for losing lives during the first Gulf
Several approaches for solving memory leak have been War. The consequences of unresolved memory leaks in real-
developed including static analysis tools, dynamic tools, and time systems can have a direct impact on human safety,
automatic dynamic tools. Static analysis tools [1], [7] are security and business sustainability.
used to identify leaks in development environment before This paper introduces a new memory leak detection and
executing the program; they do not cause any run time recovery (MLDR) algorithm. MLDR provides a run-time
overhead. These tools lack dynamic information so they solution; it helps recover leaky objects which results in
produce false positives and can not find all leaks. Some enhancing the system performance and making mission
dynamic tools [10] analyze memory consumption critical programs live longer. The main concept underlying
interactively by inspecting and comparing snapshots of the the design of MLDR is the concept of memory objects aging
heap. They allow for the search for causes of leaked objects in the physical space rather than in the virtual space.
at runtime environments and facilitate the use of garbage Software aging, in general, refers to source contention
collection. Automatic dynamic tools report the leaked issues that can cause performance degradation or can cause
objects and leak sites at the end of the program run. Some systems to hang, panic, or crash. Software aging
tools [4] instrument allocation functions to report immortal mechanisms can include memory leaks, unreleased file
objects. Insure++ [8] uses a reference counting garbage locks, accumulation of unterminated threads, data
collection approach to detect where the last reference to an corruption/round-off accrual, file space fragmentation,
object disappears. Others [11] instrument allocation function shared memory pool latching, and others [6], [13], [14].
and use mark and sweep garbage collectors to capture Parnas [9] documents software aging as a phenomenon
unreachable objects. These tools either miss many leaks or wherein legacy software becomes difficult to maintain as it
incur high overhead cost. Besides operating outside the run ages.
time environment, the current approaches have relatively Aging, in this paper, refers to the time a piece of memory
high overhead cost. The concept of detecting leak based on object remains untouched. A leaky object by definition will
(IJCNS) International Journal of Computer and Network Security, 53
Vol. 2, No. 5, May 2010
begin to age since it will no longer be accessed by any chunks or the page may constitute a portion of one memory
application program. Aging, in this context, can then be chunk.
used to detect memory leakage. Memory leaks and aging The recovery mechanism of the conservative MLDR
refer to objects in the heap virtual space. Detecting leaky identifies unreachable objects from the set of leaky pages.
objects by estimating their age in the virtual space [15] is The remaining objects are stale objects. The unreachable
time consuming. However, detecting the age of a page in the objects are relinquished (freed) and returned to the operating
physical space is much less time consuming and can be done system.
using the already available support for virtual memory MLDR is particularly useful in long lived applications.
systems. MLDR makes use of page aging in the physical As time passes, age accumulates and MLDR will be able to
space to detect actual leaks in the virtual space. Hence, identify a set of unreachable objects. These objects will be
MLDR can be used to detect and recover leaks at run time freed which makes enough room for new allocations and
reducing the risk of performance degradation or crashes. enables the program to continue execution. Another
This paper is organized as follows. Section 2 presents the consequence of memory leak is thrashing. Memory leak can
new memory leak detection and recovery algorithm cause extreme low locality of reference as live objects are
(MLDR). Section 3 presents a simulation model to measure dispersed over a large virtual address space. MLDR will
the performance and effectiveness of MLDR. Section 4 relinquish unreachable locations causing the operating
presents a summary of the conclusions. system to reuse these locations and enhance the locality of
reference as a result.
2.2 Description of MLDR
2. Memory Leak Detection and Recovery
Once a memory object has leaked, it will no longer be
(MLDR) Using Aging in Physical Memory accessed by the application program. Hence, a leaky
memory object will begin to age. Aging, in this context, is
2.1 Background
related to the time a piece of memory remains untouched. In
A memory object is said to have leaked if it is not freed by virtual memory computer systems, memory allocation is
the program or the system after it becomes no longer used or done both at the virtual and physical levels. A memory
accessible. The main job for a memory leak detector is to leakage in the virtual space will render the corresponding
determine whether an object has leaked at a given point of mapped physical memory in a “page-out status”. A page-out
time. In most general terms, memory leak occurs because of status makes a page in the physical memory a target for the
1) unreachable objects: a program either unintentionally or replacement policy that swaps the target page out to disk.
maliciously neglects to free heap-allocated objects and Going backward, a swapped out page can correspond, but
therefore these objects are lost. In this case, there is no not necessarily, to a leaked chunk in the virtual space. We
reference chain for these objects from the root set starting define “the age of a page” as the time elapsed since the page
from global and stack variables and 2) useless objects: a is swapped out from the physical to the virtual space. More
program maintains references to objects which are never precisely, the age of a page begins to accumulate from the
used again. time a page is marked by the replacement policy as a target
The memory leak detection and recovery (MLDR) for replacement. Note that a page may remain in the
algorithm is able to detect both types (unreachable and physical space for a long time after it has been marked as a
useless/stale objects). In this paper, we present two versions replaceable page. A memory chunk will be considered a
of MLDR; a conservative one which recovers only candidate leak if the age of the page to which it belongs
unreachable objects, and a liberal version which recovers exceeds a certain limit (threshold). This threshold can be
both unreachable and stale objects. The conservative either user-defined or tuned by a telemetry tool.
approach does not produce false positives, i.e, falsely Memory leak detection policies can rely on memory
identified leaks. The liberal approach produces false aging either in the virtual or physical space. Aging in the
positives, mostly related to stale objects. Recovering stale virtual space is more accurate than aging in the physical
objects without incurring false positives is more complicated space; however, using the concept of physical space aging
and requires significant architectural changes, and will not has several advantages. The physical space is much smaller
be discussed in this paper. than the virtual address space, which makes the time
MLDR has a coarse granularity in the sense that it deals required to scan for leaky pages much less than the time
with pages instead of memory chunks. Hence, we introduce needed to search for leaky chunks in the virtual space. Also,
the term “leaky page” where a leaky page is a page which tracing the age of a page in the physical space can greatly
contains potentially leaky chunks. A leaky page may contain benefit from the available hardware support for virtual
any combination of the following types of objects: 1) memory organization such as page tables and address
unreachable objects and 2) stale objects. In the context of translation buffers. Of course, the age of a page in the
this paper, stale objects are objects that have not been physical space may not be entirely due to the behavior of the
accessed for a given period of time. They are either a) owner of the page; it might well depend on the behavior of
useless objects or b) active but have not been referenced for other programs as well as on the page replacement policy
that period of time. In our approach, the MLDR leak (LRU, FIFO, OPT, LFU) [12] deployed by the operating
detection mechanism identifies the potential leaky pages. system.
Depending on the page size and the sizes of allocated It should be noted that several chunks, whose sizes are
memory chunks, a page may contain several memory smaller than a page, may be mapped to one physical page. If
54 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
only one of these chunks remains active, while the other of the allocated chunk, its size and a flag. The flag is
chunks have leaked, then the corresponding page will be initially set to one, indicating that the object is alive and
considered active and the leaks in that page will not be reachable. The flag is reset to zero by the
detected; this case is known as “false negative”. We show in StalenessRecovery() function as a result of scanning the
this paper how the problem of false negatives can be stack and registers. If a memory chunk in the MallocTable
reduced. is not matched by any address value in the stack and address
registers, then the corresponding flag is set to zero, and the
2.3 MLDR Algorithm Main Functions
chunk is deemed unreachable.
MLDR algorithm utilizes the following functions: Main(), (b) free() Function
Initialize(), Bookkeeping(), StalenessDetector(), Before deallocating a chunk of memory and exiting, the
StalenessRecovery(), and new versions of malloc() and free() free() function removes the deallocated chunk from the
functions. MallocTable; these are objects with flag value reset to zero.
2.3.1 Main() and Initialize() Functions 2.3.4 StalenessDetector() Function
The “main()” function of the algorithm keeps iterating over The StalenessDetector() traverses the page table of a given
monitored processes. Main() calls two functions: Initialize() process in order to identify leaky pages. It identifies all
and Bookkeeping(). For each monitored process, the pages whose age has exceeded the page age threshold
Initialize() function sets some user-configurable parameters (PAT). Note that any identified page at this phase is
that affect the behavior of the MLDR algorithm. These considered to be only a candidate leaky page, meaning that
parameters can be user-configurable or they can be it may or may not contain leaky chunks. A memory chunk
automatically tuned using a telemetry tool. These contained in a leaky page could in fact be active, but has not
parameters are: upper heap size threshold(UHST), lower been referenced for a relatively long time, such that the
heap size threshold(LHST), staleness detector sleeps aging algorithm marked their corresponding page as a leaky
time(SDST), and a normalized parameter(K). The UHST is page; this is a case of “false positive”.
used to initiate the leak detection process once the size of 2.3.5 StalenessRecovery() Function.
the heap for the monitored process exceeds the UHST limit. StalenessRecovery() works on the set of identified leaky
The LHST is used to suspend the detection and recovery pages(aged pages). Its main task is to remove unreachable
process once the current heap size drops below LHST. The objects, given that a leaky page is found to include a leak. In
SDST is the time (in milliseconds) a StalenessDetector() order to identify the set of chunks in a leaky page, the
will wait between any two successive scans to the page table StalenessRecovery() intersects the MallocTable with the
of a given process. The page age threshold(PAT) is used to range of addresses in that page [page_number *
identify aged pages. When the age of a page exceeds PAT, it PAGE_SIZE, (page_number+1)*PAGE_SIZE-1]. The result
will be marked as a candidate leaky page. PAT is calculated of intersection (working set), will contain the potential leaky
dynamically as an accumulated average and is fine-tuned chunks. Unreachable chunks in the working set will be
using the K parameter. identified by scanning the roots (static, stack, and registers)
2.3.2 Bookkeeping() Function searching for reachable objects in the given set of leaky
The page table is augmented with a new entry called pages. Once reachable objects are identified, the remaining
timestamp(TS). For every victim page selected from the objects in this set are definitely unreachable. Identified
mapped physical space, the bookkeeping() function unreachable chunks are passed to the free function to be
timestamps the corresponding page in the virtual address deallocated from the virtual address space and from the
space by setting the TS to the current time. The TS is used MallocTable, respectively. The remaining chunks in the
to calculate the age of a page. Whenever a page is paged working set are stale objects. The conservative version of
back into physical memory, the TS entry is reset to zero. MLDR ignores all stale objects in order to avoid running the
Pages whose TS value is larger than zero begin to age as risk of false positives.
time progresses.
2.3.3 Memory Allocation and Deallocation Functions 2.4 False Positives and False Negatives
For implementation purposes, we need to modify memory 2.4.1 False Positives
allocation and deallocation functions, e.g., malloc() and One of the main problems in memory leak detection tools is
free(). Other functions are treated in a similar manner, and the potential error known as “false positives”. In other
hence will not be provided in this paper. words, a detected leak is not a real leak. Since MLDR is a
run-time solution it is not recommended to deallocate
(a) malloc () Function
chunks which may turn out to be false positives. This is
The malloc() function incorporates two major changes especially true for systems running critical applications.
before it returns a pointer to the newly allocated memory Referencing an object after it has been removed or
chunk. The first, Malloc() checks the heap size and if the deallocated leads to incorrect results or program crash.
UHST is reached, it invokes the StalenessDetector(). The MLDR identifies potential leaky pages which might contain
StalenessDetector() is suspended when the heap size reaches false positive objects. In order to avoid the consequences of
the lower limit LHST. This modification to malloc() helps false positives, the set of leaky pages are passed to the
reducing the overhead associated with leak detection and StalenessRecovery() which identifies and recovers
recovery. unreachable objects. Of course, the leaky but reachable
The latter, the execution of malloc() updates the entries (stale) objects will persist in the system and will be counted
of a table called “MallocTable” which includes the address as false negatives. The liberal approach would consider all
(IJCNS) International Journal of Computer and Network Security, 55
Vol. 2, No. 5, May 2010
objects in a leaky page as real leaks. This approach will replacement and the current time is written into its
reduce the number of false negatives at the expense of corresponding page table entry (TS).
generating false positives. One way of eliminating the Memory allocation and deallocation are made using the
negative impact of false positives is to keep a copy of modified malloc() and free() functions as explained earlier.
removed objects on disk, such that the copy can be recovered For memory access event, we developed a function that
and reused in case the removed object is dereferenced. This simulates demand paging with global replacement strategy.
approach will not be further investigated in this paper. The bookkeeping part of the algorithm is implemented
within the scope of the page replacement policy. The
2.4.2 False Negatives simulator implements the least recently used (LRU) strategy.
False negatives are leaky chunks that go undetected. Note
that a page allocated to physical memory may consist of one Run the
start Make necessa ry cha nges
benchmark
or more memory objects in the heap virtual space. If at least
one of these objects remains active, then the corresponding Create trace
page may never age and the MLDR will not be able to files
identify it as a leaky page. As a result, the leaky objects,
which happen to be allocated to the same page, will go Read one NO
undetected. This phenomenon will result in false negatives, trace file a t a
time
i.e., undetected leaky objects.
MLDR may reduce the number of false negatives by Validate trace file using
decreasing the page age threshold allowing more pages to be universally available if results
tools mtrace() and are valid
identified as leaky at the expense of increasing the overhead dma lloc
cost. The number of false negatives can be further reduced Yes
by choosing a smaller page size such that the number of
independent objects allocated to the same page will be Leak injection
small. Also, in the conservative implementation of the
algorithm, where only unreachable objects are detected, the
stale objects will go undetected and they will count towards No more traces
the false negatives. The liberal approach, which treats all While there Print output
are more parameters
objects in a leaky page as real leaks, eliminates this type of traces
false negatives at the risk of running into false positives.
end
heap allocation oscillates between the upper and lower heap running the risk of false positives which if exist can cause
size thresholds, due to the activation and deactivation of the the system to fail. The conservative implementation is
StalenessDetector() and StalenessRecovery() functions. This preferred when the system runs critical applications. The
experiment shows how MLDR extends the life of a process liberal implementation may be used in systems running less
due to keeping the allocated heap size below an upper limit critical applications or systems used for testing applications
threshold. before deployment.
Time Vs Heap size Page Age Threshold Vs
MaxHeapSize False Positives
Upper_Heap_Size_Threshold
(80%)
Lower_Heap_Size_Threshold
(50%)
Heap size
Page_Age_Threshold
2
M.Tech student V.R.Siddhartha Engineering College, Vijayawada
E-mail:srp.tummala@gmail.com, srp.tummala@rocketmail.com
3
B.Tech student V.R.Siddhartha Engineering College, Vijayawada
E-mail:sruthi_rockz@yahoo.com,
3
B.Tech student V.R.Siddhartha Engineering College, Vijayawada
E-mail:tejaswi.tummala@gmail.com,
Abstract: The advancements in broadband and mobile substitute other broadband technologies competing in the
communication has given many privileges to the subscribers for same segment and will become an excellent solution for the
instance high speed data connectivity, voice and video deployment of the well-known last mile infrastructures in
applications in Economical rates with good quality of services. places where it is very difficult to get with other
WiMAX is an eminent technology that provides broadband and technologies, such as cable or DSL, and where the costs of
IP connectivity on “last mile” scenario. It offers both line of deployment and maintenance of such technologies would
sight and non-line of sight wireless communication. Orthogonal
not be profitable. In this way, WiMAX will connect rural
frequency division multiple access is used by WiMAX on its
physical layer. Orthogonal frequency division multiple accesses
areas in developing countries as well as underserved
use adaptive modulation technique on the physical layer of metropolitan areas. It can even be used to deliver backhaul
WiMAX and it uses the concept of cyclic prefix that adds for carrier structures, enterprise campus, and Wi-Fi hot-
additional bits at the transmitter end. The signal is transmitted spots. WiMAX offers a good solution for these challenges
through the channel and it is received at the receiver end. Then because it provides a cost-effective, rapidly deployable
the receiver removes these additional bits in order to minimize solution [3]. Additionally, WiMAX will represent a serious
the inter symbol interference, to improve the bit error rate and competitor to 3G (Third Generation) cellular systems as
to reduce the power spectrum. high speed mobile data applications will be achieved with
In this paper, we investigated the physical layer the 802.16e specification. The original WiMAX standard
performance on the basis of bit error rate, signal to noise ratio, only catered for fixed and Nomadic services. It was reviewed
and error probability. These parameters are discussed in two to address full mobility applications, hence the mobile
different Models. The first model is a simple OFDM
WiMAX standard, defined under the IEEE 802.16e
communication model without the cyclic prefix, while the
second model includes cyclic prefix. specification. Mobile WiMAX supports full mobility,
nomadic and fixed systems [4]. It addresses the following
Keywords: WiMAX, wireless communication, physical layer. needs which may answer the question of closing the digital
divide:
• It is cost effective.
1. Introduction • It offers high data rates.
• It supports fixed, nomadic and mobile applications
The demand for broadband mobile services continues to thereby converging the Fixed and mobile networks.
grow. Conventional high-speed broadband solutions are • It is easy to deploy and has flexible network architectures.
based on wired-access technologies such as digital • It supports interoperability with other networks.
subscriber line (DSL). This type of solution is difficult to • It is aimed at being the first truly a global wireless
deploy in remote rural areas, and furthermore it lacks broadband network.
support for terminal mobility. Mobile Broadband Wireless
Access (BWA) offers a flexible and cost-effective solution to 1.1 Adaptive Modulation Techniques in WiMAX
these problems [1].The IEEE WiMax/802.16 is a promising
technology for broadband wireless metropolitan areas The basic idea of Adaptive Modulation is to adapt
networks (WMANs) as it can provide high throughput over different modulation techniques when a wireless
long distances and can support different qualities of communication system experiences fading and variations on
services. WiMax/802.16 technology ensures broadband the link. WiMAX takes full advantage of link adaptation
access for the last mile. It provides a wireless backhaul technique along with coding. This scheme is quite simple, if
network that enables high speed Internet access to the link condition is not good, WiMAX system changes the
residential, small and medium business customers, as well modulation automatically. Hence, real time application such
as Internet access for WiFi hot spots and cellular base as video and voice can run continuously by varying the
stations [2]. It supports both point-to-multipoint (P2MP) and modulation, the amount of data transferred per signal also
multipoint-to-multipoint (mesh) modes. WiMAX will varies, i.e., deviation in throughputs and spectral
(IJCNS) International Journal of Computer and Network Security, 59
Vol. 2, No. 5, May 2010
1.2 Simulation
In our simulation work we investigated the behavior of Figure 1. Model – 1 OFDM transmitter simple model
adaptive modulation technique of WiMAX. The adaptive
modulation used following modulation techniques for
modulating and demodulating the signal:
•B
it Error Rate (BER) Figure 2. Model – 2 (Model with Cyclic Prefix)
•S ignal to Noise Ratio (SNR) Mersenne Twister - Random Number Generator (RNG)
•P robability of Error (Pe) Algorithm
The key points, in the simulations are: Mersenne twister is a random number generator that
generates the random number by using the pseudorandom
• Mat lab 7.4.0 (R2007b). algorithm. The generator is composed of a large linear
• Mersenne Twister - Random Number feedback shift register and provides some excellent output
Generator (RNG) Algorithm statistical properties. Mersenne twister - RNG comprises a
• Noise is characterized as Gaussian seed value which is 19,937 bits long and the value is stored
• Fading is characterized as Rayleigh in 624 element array. Mersenne twister has a period of
2^19937 – 1. Mersenne Twister is basically implemented in
Probability distribution function. C language and utilizes the memory in an efficient way. In
Matlab this algorithm is used in rand function which is used
• Cyclic prefix is used to generate the input of random number for scientific
• All the plotting is done to evaluate the applications
Performance on the basis of BERVs
SNR. 3. Simulation Results
3.1 Theoretical Values of BER using Adaptive 4. Probability of Error (Pe) for Adaptive
Modulation Techniques in OFDM Modulation
The theoretical value of BER with respect to adaptive The Probability of Error (Pe) is the assumption of the
modulation techniques in the presence of pure AWGN is rate of the error that introduce in the system because of
used to estimate the theoretical value of SNR with 256 sub noise and fading effects in the channel and also due to the
carriers cable losses at transmitter and the receiver ends. The
Probability of Error for M-ary PSK has been calculated
using the following formula:
References
[1] Mai Tran, George Zaggoulos, Andrew Nix and
Angela Doufexi,” Mobile WiMAX Performance
An a l ysi s an d Compar i son with Experimental
Results”
[2] J. El-Najjar, B. Jaumard, C.Assi, “Minimizing
Interference in WiMax/802.16 based Mesh
Networks with Centralized Scheduling”,
Global Telecommunications Conference, 2008.
pp.1-6. http://www.intel.com/netcomms
/technologies/WiMAX/304471.pdf
[3] Intel white paper, Wi-Fi and Wimax Solution:
“Understanding Wi-Fi and Wimax as
Metro-Access Solution, “Intel Corporation
2004.
[4] A. Yarali, B. Mbula, A. Tumula, “WiMAX:
A Key to Bridging the Digital Divide ”IEEE
Volume, 2007, pp. 159 – 164.
[5] WiMAX Forum, “Fixed, Nomadic, Portable and
Mobile Applications for 802.16-2004 and
802.16eWiMAXNetworks”, November,
2005.
[6] T.TAN BENNY BING, "The World Wide Wi-
Fi: Technological Trends and B u s i n e s s
Strategies", J O H N W I L E Y & S O N S , INC.,
2003
[7] M. Nadeem Khan, S. Ghauri, “The WiMAX
802.16e Physical Layer Model”, IET
International Conference on Volume, 2008, pp.117
– 120.
[8] Kavesh Pahlavan and prashant Krishna Murthy
“Principles of Wireless Network “Prentice-Hall, Inc.,
2006
[9] Yango Xiao, “Wimax/Mobile”MobileFi Advanced
Research and Technology Auerbach publications,
2008.
62 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
1. Introduction
The recent development in small embedded sensing devices
and the wireless sensor network technology has provided
Figure1 (a). RREQ Broadcast
opportunities for deploying sensor networks to a range of
applications such as environmental monitoring, disaster
management, tactical applications etc [1].Main requirement
for such application is that motes carrying onboard sensors
should be physically small, low power consuming and
include wireless radio.
For data collection a straight forward solution is that each
mote transmits its data to a centralized base station.
However in such cases the energy requirement of each node
would be large which reduces mote life and also there would
Figure1 (b). RREP Propagation
be interference problem. Alternative approach for harvesting
data from sensor fields uses mobile data collector such as
It is composed of two main mechanisms: Route Discovery
robots which move in the sensor field to collect data and
and Route Maintenance. For route discovery the source node
transmit the same to base station in real /non-real time [6].
floods the route request (RREQ) packets in the network. The
In the present paper, the performance of dynamic source
nodes receiving RREQ rebroadcast it and the process repeats
routing (DSR) protocol has been analyzed keeping in mind a
until the destination node or an intermediate node having a
sensor network scenario wherein all the nodes are static and
route to the destination is found. Such a node replies back to
source node is moving ( one of which is a data harvester
the source with a RREP packet. When a link-break in an
from static nodes and other one acting as a sink).
established route occurs, the route maintenance phase is
Section 2 describes the related work and Section 3 is about
initiated wherein upstream node at the link-break site
the network scenario and all about the definition of
creates a route error (RERR) message and sends it to the
simulation parameters and details of simulation experiment.
source node. On receiving RERR the source node utilizes
Section 4 gives the results and discussion of the simulation.
alternate routes from the route cache, if they are available, to
Section 5 concludes the work.
prevent another route discovery. The drawback with DSR is
that it needs to place entire route in both the route replies
and the data packets and thus requires greater control
overhead.
(IJCNS) International Journal of Computer and Network Security, 63
Vol. 2, No. 5, May 2010
data) 4500
References
[1] Anna Hac, “Wireless Sensor Network Designs”, John
Wiley and Sons, 2003.
[2] M.K. Marina, S. R. Das, “Performance of Route Caching
Strategies in Dynamic Source Routing,” ICDCS, pp.
425-432,2001.
(IJCNS) International Journal of Computer and Network Security, 65
Vol. 2, No. 5, May 2010
efficiently by modifying route table as quickly as possible. shrinking a route when a node discovers a shorter path than
the existing one. By utilizing only local geographic
This paper is organized as follows. Previous work is information, a host can anticipate its neighbor’s departure
mentioned in section 2. Terminology needed to understand and, if other hosts are available, choose a host to bridge the
this paper is explained in section 3. Proposed algorithm is gap, keeping the path connected. The benefits are that this
presented in section 4, followed by the theoretical analysis reduces the need to find new routes and prevents
in section 5. Section 6 presents the conclusion. interruptions in service.
Chou et al. [16] presented a dynamic route maintenance
2. Previous Works algorithm for beacon-based geographic routing. In this
Wu et al. [6] proposed an algorithm that tries to fix some approach, the mobile nodes dynamically adjust their beacon
problems in paths such as finding shorter paths and fixing intervals based on their speed of movement. Moreover the
broken links. However, nodes in their algorithm stores only routing information could be well managed using the
the destination, next hop, hop count, etc. for each path so mobility prediction.
that the coverage of the algorithm is limited. Ko and Park et al. [17] proposed an algorithm that is a preliminary
Vaidya [10] proposed a location-aided routing scheme version of the paper. However, the algorithm presented in
where a route request is flooded in the direction of the target this paper considers more cases that may save more battery
node. Here, a node’s response to the route request depends power and bandwidth.
on whether or not it is in the region approaching the
destination.
3. Terminology
Liao et al. [11] proposed a routing scheme called GRID. A Graph can be represented by G(V, E), where V is the set
This scheme imposes a grid system on the earth’s surface of nodes and E is the set of links in the graph G. In a
and selects a leader in each grid to act as the gateway. Data wireless network, if a node B is within the transmission
packets are then forwarded to their destination grid by grid. range of node A, then it is said that there is a link from A to
This protocol confirms that the availability of location B which is denoted as (A, B). A is an upstream node of B,
information improves the routing performance. and B is the downstream node of A. B is also called as a
Lee and Gerla [12] proposed AODV-BR algorithm that neighbor of A. If (A, B) implies (B, A), then the link is
utilizes a mesh structure to provide multiple alternate paths called bidirectional, and unidirectional otherwise.
to existing on-demand routing protocols without producing
additional control messages. In their algorithm, data A path P(v1 , v2, . . . , vn-1, vn), where vi ∈ V, 1 ≤ i ≤ n,
packets are delivered through the primary route unless there consists of a set of links (v1 , v2), (v2, v3), … , (vn-1, vn). A
is a route disconnection. When a node detects a link break, subpath of P(v1 , v2, . . . , vn-1, vn) is the path that is a part of
data packets can be delivered through one or more alternate P. Function rem (P, vi) returns the portion of path P starting
routes and are not dropped. Route maintenance is executed at node vi to the end of P. For example, both P(B, C, D, E)
utilizing alternate paths. Here, the route is maintained only and P(A, B) are subpaths of P(A, B, C, D, E), and rem (P(A,
when a link is broken. B, C, D, E), B) is P(B, C, D, E). |P| denotes the number of
the links in path P. For example, |P(A, B, C, D, E)| is four,
Li and Mohapatra [13] introduced LAKER, a location-aided since P consists of four links.
knowledge extraction routing. This scheme reduces the
flooding overhead in route discovery by extracting Note that if the links are bidirectional (unidirectional, resp.),
knowledge of the nodal density distribution of the network then the paths are also bidirectional (unidirectional, resp.).
and remembering the series of locations along the route Although links assumed to be bidirectional in this paper,
where there are many nodes around. However, this scheme paths are assumed to be unidirectional for simplicity in
does not deal with route maintenance. explanation. It can be easily extended to cover the handling
Stoimenovic [14] reviewed many position-based routings in of unidirectional paths.
ad hoc networks. It is likely that only position-based
approaches provide satisfactory performance for large When a new path is found, each node in the path may
networks. Greedy mode routing was shown to nearly contain the path information in its Routing Table for
guarantee delivery for dense graphs, but to fail frequently possible later use. This can be done without any extra
for sparse graphs since the destination is also moving and it expense at each node by simply storing the path information
is not clear where to send message. The routing process is when the RPY passes through the node. For example in
converted from the greedy mode to recovery mode at a node Figure 1, nodes A, B, C, D, E, and F in the path P(A, B, C,
where greedy mode fails to advance a message toward the D, E, F, G) store the path information when the RPY passes
destination through them.
Park et al. [15] proposed an anticipated route maintenance
protocol with two extensions to route discovery based
4. Proposed Routing Table Maintenance
routing scheme: extending a route by inserting a common Scheme
neighbor of two nodes in a route when the two nodes move Based on the assumption that all links are bidirectional, the
apart from each other that may cause link breakage, and proposed scheme is hybrid, since it discovers new paths
(IJCNS) International Journal of Computer and Network Security, 67
Vol. 2, No. 5, May 2010
reactively but the maintenance of the paths is proactive. In substitutes (A, B, C, D, E, F, G) with P(A, B, C, F, G) and
other words, a new path from A to B is discovered only P(A, B, C, D, E) in its routing table. Note that the upstream
when 1) A has a data for B, and 2) there is no existing path nodes of C in the path need not be informed this finding.
from A to B. On the other hand, any change of the existing Instead, if a packet from either A or B destined to either F or
paths is updated immediately regardless of the usage of the G is arrived at C, C directs the packet to F using P(A, B, C,
path. This strategy makes sense, because path discovery F, G), not to D. If the packet is destined to D or E, then the
takes a considerable amount of resources due to the flooding packet is transferred along P(A, B, C, D, E).
of REQ packets, whereas maintenance requires unicasts
only. Figure 2 (b) shows another case where a shorter path is
discovered by a node in the same path that is at the other
In this paper, only the routing table maintenance scheme is end, F. In this case, path P(A, B, C, D, E, F) is split into
presented, since route discovery algorithms can be found in two paths, P(A, B, C) and P(A, F, E, D). Note that this is
many routing protocols. The proposed routing table possible since A has all nodes in the path, and therefore
maintenance scheme is explained below for different cases. knows exactly where to split the path.
Case 1) Link breakage is discovered. Consider two paths, Figure 2 (c) shows the case where a shorter path is
P(A, B, C, D, E, F, G) and P(H, I, C, D, E, F, G), in the discovered by other newly found path. For example,
network as shown in Figure 1. Suppose when node C suppose P(A, B, C, D, E, F, G) is the existing path, and P(H,
sends a data packet to node G along P(A, B, C, D, E, F, G), C, J, F, K) is just discovered. Then, P(A, B, C, D, E, F, G)
C found out that the link (E, F) is broken. Then, this will be replaced by P(A, B, C, J, F, G) and P(A, B, C, D, E).
information is disseminated to all upstream nodes of C in Note that, as was done in Case 2 (a), this new information is
both paths. On receiving the link breakage information not informed to the upstream nodes of C.
nodes A, B, H, and I modify the paths by removing subpath
P(F, G). Otherwise, attempting to use this broken link may So far the cases for shorter paths have been considered. In
cause delay. the following, the cases when extension of the existing path
. is found will be discussed.
A
C F G Case 3) An extension of an existing path is found by the
B
other path. In this case, the extended portion is
disseminated to existing path. For example, suppose P1 =
P(A, B, C, D, E) in Figure 3 is the existing path, and P(F,
D E G, C, H, J) is discovered later. Then, P(A, B, C, D, E) has
(a) an extended portion of P(C, H, J) at node C, which will be
propagated to upstream nodes of C. Therefore, P(A, B, C, H,
F J) will be added to routing tables at A, B, C. Likewise, P(F,
A
G, C, D, E) will be added to routing tables at C, F and G.
E
B J
H
C A D
D
C E
(b) B
D
H
E
G
C
F
F G
A J
B Figure 3. As a result of the discovery of another path P(F,
K G, C, H, J), nodes A, B, C add P(A, B, C, H, J) to
(c) their routing tables. Likewise, P(F, G, C, D, E)
will be added to routing tables of F, G, and C.
Figure 2. (a) A shorter path P(A, B, C, F, G) is found in
the same path. (b) A shorter path is found by the Case 4) A more general form of Case 3 can be found in
node at the other end of the same path. (c) A Figure 4, where an existing path and a newly found path
shorter path P(A, B, C, J, F, G) is found by other have more than one common node. In this case, the shorter
path P(H, C, J, F, K). path connecting any adjacent two common nodes is chosen
and used. For example, refer to Figure 4, where P(A, B, C,
Case 2) A shorter path is found. There are two possible
D, E, F) is the existing path and P (G, B, H, J, K, E, L) is
cases. First, C in Figure 2 (a) found a shorter path (A, B, C,
the newly found path, and they share both B and E. In this
F, G) in the existing path (A, B, C, D, E, F, G), when F
case, subpaths P(B, C, D, E) and P(B, H, J, K, E) are
moves into C’s transmission range. In this case, C
compared and P(B, C, D, E) is chosen since it is shorter of
68 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
the two. Then, both existing and newly found path will be // On receiving a RPY, call check_modification_path (path
replaced using the shorter path. For example, P(A, B, C, in // the RPY), where each path in the routing table is
D, E, F) is replaced by P(A, B, H, J, K, E, F), P(A, B, C, D) checked if // the path should be further modified due to the
and P(A, B, H, J, K, E, L). Also, two routes, P(G, B, C, D) path in the
and P(G, B, H, J, K, E, F) are added to the routing table at // RPY.
G.
Let the newly found path be P(v1, v2, . . . , vn-1, vn).
Summary of the proposed algorithm is presented in Figure // For the cases shown in Figure 3
5. If there is a path P(u1, u2 , . . . , um-1, um) in the routing table
such that vi = uj, add P(u1, u2 , . . . , ui-1 , ui, vj+1, . . . , vn-1, vn )
to the routing tables of u1, u2, . . . , ui-1, ui. Also add P(v1,
G D L
C
v2, . . . , vj, ui+1, ui+2, . . . , um) to routing tables of v1, v2, . . . ,
B E
vj.
F
Case when the existing path and newly found path have two
A H K common nodes
J
Case when a broken link is found Figure 5. Pseudo-code for the proposed algorithm.
5.2 Impact on route discovery [8] A. Rahman, S. Islam, and A. Talevski, “Performance
measurement of various routing protocols in ad-hoc
This section presents the theoretical analysis of the proposed network,” Proc. International Multiconference of
algorithm to show how our proposed algorithm can reduce Engineers and Computer Scientists, 2009.
communication overhead and battery power by modifying [9] G. Jayakumar and G. Gopinath, “Performance
routing tables on discovering a new path. Comparison of two on-demand routing protocols for
Suppose N is the average number of neighbors of each node ad-choc networks based on random way point mobility
in the network. If an algorithm abandons the path on its model,” American Journal of Applied Sciences, 5 (6),
topological changes, then it would take at least pp. 659 – 664, 2008.
N 0 ⋅ N 1 ⋅ ... ⋅ N L−1 ≈ N2L REQ transmissions to find a [10] Y. B. Ko and N.H. Vaidya, “Location-aided routng
path of length L all over again by using broadcast. On the (LAR) in mobile ad hoc networks,” Annual
other hand, if an algorithm can adjust to the minor International Conference on Mobile Computing and
topological changes of paths like our proposed algorithm, Networking (MobiCom), Oct. 1998, pp. 66-75.
then it would take only L transmissions for the path of [11] W.H. Liao, Y.C. Tseng, and J.P. Sheu, “GRID: a fully
length L. Therefore, the advantage of having the proposed location-aware routing protocol for mobile ad hoc
algorithm is obvious. This analysis would be true, if the networks,” Telcommunication Systems, 18(13), 2001,
changed path is used, otherwise, the adjustment would be a pp. 37-60.
waste. [12] S.J. Lee and M. Gerla, "AODV-BR: backup routing in
ad hoc networks", IEEE Wireless Communications and
Networking Conference (WCNC), Sept. 2000, pp.
6. Conclusion 1311–1316.
Routing in a wireless mobile ad hoc network is quite a [13] Jian Li and P. Mohapatra, "LAKER: location aided
challenging task, because it takes up a lot of network knowledge extraction routing for mobile ad hoc
resources. Therefore, although it would be wise to take full networks", IEEE Wireless Communications and
advantage of the already discovered paths, many algorithms Networking (WCNC), March 2003, pp. 1180–1184.
simply discard the paths when their topology is changed. [14] I. Stojmenovic, “Position-based routing in ad hoc
To overcome this shortcoming, we have proposed a routing networks,” IEEE Communications Magazine, July
table maintenance algorithm that responds to the changes in 2002, pp. 2-8.
network topology as promptly yet efficiently as possible to [15] S. Park, S.M. Yoo, M. Al-Shurman, B. VanVoorst,
fully utilize existing paths by modifying them. and C.H. Jo, “ARM: anticipated route maintenance
scheme in location-aided mobile ad hoc networks,”
Our simple analysis shows that the proposed algorithm Journal of Communications and Networks, vol. 7, no.
reduces number of packet transmissions, and therefore saves 3, pp. 325-336, Sep. 2005.
resources such as bandwidth and battery power. [16] C.H. Chou, K.F. Ssu, and H.C. Jiau, “Dynamic route
maintenance for geographic forwarding in mobile ad
References hoc networks,” Computer Networks, vol. 52, issue 2,
pp. 418-431, Feb. 2008.
[1] C. Perkins, “Ad Hoc Networking,” Addison-Wesley, [17] S. Park and S. Yoo, Routing table Maintenance in
2001. Mobile Ad Hoc Networks, The 12th International
[2] C. Perkins, and P. Bhagwat, “Highly-dynamic Conference on Advanced Communication Technology,
destination-sequenced distance-vector routing (DSDV) pp. 1321 - 1325, Feb., 2010.
for mobile computers,” pp. 234-244, SIGCOM’94,
1994.
Authors Profile
[3] C. Perkins, E. Royer and S. Das, “Ad hoc on demand Seungjin Park received the BE degree in civil engineering from
distance vector (AODV) routing,” Internet draft, IETF, Hanyang University, Korea, in 1973, MS degree in computer
Oct. 1999. science from University of Texas at Arlington, in 1986, and PhD
[4] V. Park and M. Corson, “A highly adaptive distributed degree in computer science from Oregon State University, in 1993.
routing algorithm for mobile wireless networks,” In He is currently an assistant professor in the Department of MNGT,
Proc. IEEE Infocom, pp.1405-1413, Apr. 1997 MIS, and CS, University of Southern Indiana. His research
[5] D. Johnson and D. Maltz, “Dynamic source routing in interests include parallel algorithms, wireless ad hoc networks, and
ad-hoc wireless networking,” in Mobile Computing, T. sensor networks.
Imielinski and H. Korth, editors Kluwer Academic
Seong-Moo Yoo is an Associate Professor of Electrical and
Publishing, 1996.
Computer Engineering at the University of Alabama in Huntsville
[6] S. Wu, S. Ni, Y. Tseng, and J. Sheu, “Route (UAH), USA. Before joining UAH, he was an Assistant Professor
maintenance in a wireless mobile ad hoc network,” at Columbus State University, Georgia – USA. He earned MS and
Proc. 33 Hawaii International Conference on System PhD degrees in Computer Science at the University of Texas at
Sciences, 2000. Arlington. He was a Fulbright Scholar, Kazakhstan 2008-2009.
[7] J. Broch, D. Maltz, D. Johnson, Y. Hu, and J. Jetcheva, His research interests include computer network security, wireless
“Performance comparison of multi-hop wireless ad hoc network routing, and parallel computer architecture. He has co-
network routing protocols,” MobiCom’98, 1998. authored over 70 scientific articles in refereed journals and
international conferences.
70 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
The notion of proxy signature was first introduced by 2. Overview and Attacks on Hu et al’s Proxy
Mambo, Usuda and Okamoto in 1996 [3]. After that the
Ring Signature Scheme
concept of ring signature was formalized in 2001 by Rivest,
Shamir and Tauman [4]. In the ring signature an In this section, we first discuss in brief Hu et al.’s proxy
anonymous signature allows an user to anonymously sign on ring signature scheme. We then describe Zhang and Yang’s
behalf of a group, where no one can know which the actual attacks on Hu et al.’s proxy ring signature scheme.
signer is. When verifying the verifier only knows that the
signature has been produced by some member of this ring, 2.1 Overview of Hu et al.'s Proxy Ring
but he/she has no information about who is the actual of the Signature Scheme
signature. Hu et.al introduced a new type of signature scheme called
the proxy ring signature scheme with revocable anonymity
Proxy signature can be combined with other special which combines ring signature with proxy signature.
signatures to obtain some new types of proxy signatures (for
(IJCNS) International Journal of Computer and Network Security, 71
Vol. 2, No. 5, May 2010
1. Commission Generation: For the user U i , the original 2.1.5 Open Phase
Signer A0 randomly chooses ki ∈ Z q and then To open a signature and reveal the actual identity of the
signer, the original signer checks the following.
computes sˆ i = x 0 g ki + k i mod q and rˆi = g
mod q . ki
rˆ −1
Then A0 sends (sˆi , rˆi ) secretly to U i and keeps For i = 1 to n , verifies whether g i = V 0 i .
r y i rˆ
ki secret.
2. Proxy Verification: Each user U i checks If for some i , the verification phase , it indicates that U i is
the actual signer.
whether g i = yoi rˆi mod q . If it holds, then U i
sˆ rˆ
3. For all j ≠ i , picks up at random z j , c j , r j ∈ Z q and 4. For all j ≠ i , randomly select z j , c j , r j ∈ Z q and
r j ≠ y0−1 , then computes: a j = g j y j y0 r j j ,
if
z
( )
c −1
if r j ≠ y0 then compute the followings:
δ j = A j , and b j = h j δ j j
z c
aj = g
zj
(y j y0 r j ) j ,
c
zi = a − i ⋅ l ⋅ ci mod q .
Let z N = (z1 ,⋅ ⋅ ⋅, zn ) , c N = (c1 ,⋅ ⋅ ⋅, c n ) , and i ⋅l −1 −1
8. After this, compute ri = g yi y0 , and then set
rN = (r1 ,⋅ ⋅ ⋅, rn ) .
The resultant proxy ring signature on message m is rN = (r1 ,⋅ ⋅ ⋅, rn ) .
δ = (m, A, zN , c N , r N ,V ) . 9. Finally, the resultant proxy ring signature on
message m becomes δ = (m, A, zN , c N , r N , V ) .
72 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
a j = g j ( y j y0 r j ) j ,
z c
Zhang and Yang showed that the forged proxy ring 1. Commission Generation: For the user U i , the
signature is valid signature.
2.2.2 Attack on Revocable Anonymity original signer A0 randomly chooses ki ∈ Z q and
In the following, Zhang and Yang showed how to produce a then computes sˆ i = x 0 g ki mod q and
proxy ring signature in which the anonymity of the
dishonest proxy signer’s identity is not revoked.
rˆi = g ki mod q , where x0 is the original signer
A0 ’s secret (private) key. After that A0 sends
Let the user U i be the real signer and the ring be (sˆi , rˆi ) secretly to U i and keeps ki as secret.
B = (U 1 ,⋅ ⋅ ⋅,U n ) . On inputs, a group size n ∈ Z , a 2. Proxy Verification: Each user U i checks whether
message m and the public key set yN = ( y1 ,⋅ ⋅ ⋅, yn ) , the x0 ⋅ g ki
g =g sˆi
mod q . Here y0 is the original
signer U i performs the following:
signer A0 ’s public key. Then U i computes
1. Here, for j = 1 to n , the generations of
si = xi + sˆi mod q as his/her proxy signing key.
a j ,δ j , b j and A remain same as those for Hu et
al.’s scheme.
2. Sets a N = (a1 ,⋅ ⋅ ⋅, a n ) , bN = (b1 ,⋅ ⋅ ⋅, bn ) and
3.2 Singing Phase
Assume that the user U i be the real signer. Let the ring be
randomly chooses V ∈ Z p to compute
B = (U 1 ,⋅ ⋅ ⋅,U n ) . Inputs in this phases are (i ) a group size
H (m, a N , b N , V ) = c . Here we note that
generation of V is different from that of Hu et al.’s
n ∈ Z , (ii ) a message m and (iii ) a public set
scheme and V is a random number. yN = ( y1 ,⋅ ⋅ ⋅, yn ) . The signer U i then executes the
n following steps:
3. Computes ci = c − ∑c
j =1, j ≠ i
j and
Step-1. Selects d ∈R Z q randomly and then computes
zi = wi − ci si + ci d . h = H (m) and δ i = h si −d .
4. Then computes ri = y0rˆi −1 rˆi g − d . After computing these, U i sets A = δ i1 / i .
Let z N = (z1 ,⋅ ⋅ ⋅, zn ) , c N = (c1 ,⋅ ⋅ ⋅, c n ) , and Step-2. Randomly choose wi ∈ Z q and then compute
rN = (r1 ,⋅ ⋅ ⋅, rn ) .
a i = yi bi = h wi .
wi
and
Finally, the resultant proxy ring signature on message m Step-3. For all j ≠ i, selects at random
is δ = (m, A, z N , c N , r N , V ) . From the above generation
z j , c j , r j ∈ Z q such that r j ≠ y0−1 . Then computes
process of proxy ring signature, the original signer cannot
z j + x jc j
a j = yj
c c
revoke the identity of proxy signer, because of the fact that y0 j r j j ,
V in the signature δ = (m, A, z N , c N , rN ,V )is a random
δ j = A j , and b j = h j δ j j .
z c
number and the value of i can not be retrieved by using the
above open algorithm. As a result, the original signer can Step-4. Let a N = (a1 ,⋅ ⋅ ⋅, a n ) , and bN = (b1 ,⋅ ⋅ ⋅, bn ) .
not revoke the anonymity of the proxy signer’s identity. −d
y0−1
U i then computes V = g yi and
c = H (m, a N , bN ,V ) .
3. Improved Proxy Ring Signature Scheme
n
In this section, we describe our improved signature scheme
in order to eliminate the security flaws discussed in Section
Step-5. Computes ci = c − ∑c
j =1, j ≠ i
j and
rN = (r1 ,⋅ ⋅ ⋅, rn ) . a i = yizi yixi ci y0ci ri ci = yiwi −ci si +ci d yixi ci y0ci y0xi rˆi −1 yi−d [ ]ci
Step-1. Computes h = H (m) . Thus if the proxy ring signature is generated by a valid
Step-2. For i = 1,⋅ ⋅ ⋅, n , computes member in the ring, the verification of the equation
a i = yi
zi + xi ci
y0ci ri ci , H (m, a N , bN ,V) = ∑ci passes.
i∈B
δ i = Ai , and
bi = h zi δ ici . 4.2 Correctness of Open Phase
Step-3. If ri = y0−1 , then the verifier immediately rejects We prove the correctness of open algorithm as follows:
ˆ −1
yi − d y 0x i r i y 0− 1 y i− d
ˆ
y 0x i r i
the signature. g ri
= g = g
ri ≠ yo−1 , the verifier checks the
( )
Step-4. Otherwise, if ˆ
y 0x i r i
y 0− 1 y i− d
x rˆi x 0 x i rˆi
= g = V = V
H (m, a N , bN ,V) = ∑ci
y0 i g
6. Conclusion
In this paper, we have proposed an improved proxy ring
sig-nature with revocable anonymity to eliminate the
security flaws in Hu et al's scheme. The proposed scheme
allows the original signer to know exactly who the signer
is. We have given correctness proofs of our scheme and
analyze the security aspects of our scheme. The security
of the proposed scheme is based on the security of the
DLP problem. We have also shown that our scheme
preserves the properties of unforgeability as well as
(IJCNS) International Journal of Computer and Network Security, 75
Vol. 2, No. 5, May 2010
Abstract: It is known that there exist triangle free graphs with that, while the number of colors grows linearly, the number
arbitrary high chromatic number. A recursive construction of of nodes grows exponentially. Just like any other graph,
such graphs was first described by Blanches Descrates and girth-base graphs once constructed are difficult to color with
Mycielski. They have shown that for any positive integer k, there the minimum number of colors, especially when the number
exists a k-chromatic graph containing no triangle. In this paper, of nodes is relatively large.
we introduce an algorithm which detects the chromatic number
of an arbitrarily large girth-base graph and assigns colors to its
Lemma 1: If a graph is a Girth-g triangle free graph with n
nodes using the minimal number of colors, namely with exactly
k colors.. The algorithm complexity is O(Lg(n)), where n is the
nodes, then there exists a k-coloring of the graph, where k =
number of nodes of the graph. lg [(n+1)/3] + 2
1. Introduction
There exist triangle free graphs with arbitrary high v1
chromatic number [4,5,6,7]. A recursive construction of g- v2
girth graphs is described in [1,2,3]. Blanches Descrates and
Mycielski have shown that for any positive integer k, there v5
exists a k-chromatic graph containing no triangle [9]. In this v3
paper, we will introduce an algorithm, which colors such
graphs with exactly k colors. First, we show how the k- v4
chromatic graphs are constructed using girth graphs.
A girth g-graph is a triangle free graph with the
shortest cycle containing g nodes. Blanches Descrates and Level 0, Girth 5, 3 Colors Graph
Mycielski [9] showed how to construct a k-chromatic graph
using a 5-girth graph (Figure 1, level 0).
v1
Call the original 5-node graph a level (0) graph. Add 6
new nodes to the graph (u1, u 2, u 3, u 4, u 5) and u.
Connect node ui to the neighbors of node vi . Obviously u1 v2
nodes ui and vi can be colored with the same color, since
they are not interconnected. Now connect u to (u1 , u 2, u 3 ,
v5 u5 u2
u 4 , u 5). Node u must be colored with a new color. The
resulting graph is the Grötzch graph with 11 nodes, no u
triangles, girth 4, and exactly 4 colors. Call this graph a
level (1) graph. In the same manner, we can construct a 5-
u4 u3
chromatic graph from the 11-node Grötzch graph. We add
12 new nodes (11+1). We connect each of the newly added
11 nodes to the neighbors of one and only one of the nodes v3
in the original 11-node graph. The newly added nodes will
be colored with the same colors as the original 11 nodes. v4
The 12th node connects to the newly added 11 nodes and
requires a new color. The total number of colors used is 5
colors. Call this graph a level (2) graph. By repeating this Level 1, 4 Colors Graph
process, we can obtain a triangle free graph with any
chromatic number k= 3+i, where (i) is the level of the Figure 1: Girth g Graph Construction and Coloring
graph.
Node 23 has 11 neighbors. All the neighbors can be colored c. 3rd stage: 3 nodes (23+1/8). Number of colors is 1
with the same color, since by definition they are not d. 4th and 5th stages: 2 nodes. Number of colors is 1
interconnected with each other. So we color nodes 12-22 2. The total number of steps required to color the graph is
with blue. The number of nodes colored at this step is 5 = Lg 23
(1+11) = 12. Actually this is equal to the number of nodes
added at level 2, i.e., (N+1)/2.
2.2 Proof:
We now present a formal proof of the theorem. By
Next we move down the list to the root node at level 1,
construction, a girth graph consists of two major parts; the
i.e., node 11 with degree 10. By definition, node 11 is not
foundation, which is the 5-node cycle, and a multi-level
connected to node 23. Thus it may reuse the red color. Half
structure of nodes. Each level adds twice the number of
of the neighbors of node 11 are also neighbors of node 23
nodes to the previous level plus one root. The root at each
since they were added at the second level, and thus they
level is connected only to the newly added nodes, and is
must have been colored at the previous step with the blue
isolated from all the nodes added at the previous level(s).
color (nodes 19-22). The other half (nodes 6-10) were added
The largest degree node is the root of the last level in the
at level 1. Each of these nodes is connected to at least one of
graph. The degree of this node is d=(N-1)/2. The next
the nodes that have been colored at the previous step. A new
largest degree node is the root of the next level down, with
color, say color green, is required to color nodes 6-10. Note
degree = d-1, and so on.
that nodes 6-10 are not connected within each other, so they
The algorithm sorts the nodes of the graph in a
all can use the same color. At the end of this step, all nodes
descending order based on the degree of nodes. By
added at level one (5+1) are colored, and only one new color
definition, the largest degree node is the root of the last
is added.
(highest) level.
Assume that we construct a girth graph with (i) levels.
The next node in the list is node 5. Note that at this
For (i=0), the graph has 5 nodes. For (i=1), the graph has 11
stage, we have arrived at the original 5 nodes cycle (1
nodes. For (i=5), the graph has 191 nodes. At the (ith) level,
through 5). All of these nodes have the same degree. So we
the number of nodes is N=(5*2i)+( 2i)-1. The node with the
can start with any of them and get the same result. Node 5
largest degree is the last node added to the graph, i.e., the
has 8 neighbors. 6 out of 8 nodes have already been colored
root at level (i) and is connected to (N-1)/2-1 nodes; so the
with colors green and blue. Node 5 can take the color red.
degree at level (i) di = (3*2i-1). The next largest degree node
Note that color red is only used by the root nodes (11 and
is the root node added at the previous level (i-1). This node
23) at levels 1 and 2. These nodes do not connect to any of
is connected to nodes at the current level (i-1) and nodes at
the inner most 5-nodes cycle. The neighbors of 5 (nodes 1
the upper level (i). The number of nodes connected at the
and 4) require a new color, say the yellow color. The total
current level (i-1) = (3*2i-1-1). The number of nodes
number of nodes colored at this stage is (2+1). It is worth
connected to the root at the upper level (i) is = (3*2i-1 -1). So
noting that the inner 5-node cycle requires at most 3 colors.
the degree at level (i-1), di-1 =2*(3*2i-1-1) = 3*2i-2. The
One of the already used colors can certainly be reused. This
next largest degree node is the root at level (i-2). The root is
is the color given to the root nodes added at each of the
connected to (3*2i-2-1) nodes at level (i-2). It is connected to
subsequent levels of the graph construction. So, when the
same number of nodes at level (i-1). It is also connected to
algorithm reaches the last 5 nodes in the graph, only 2 new
2*(3*2i-2-1) at level (i). So the degree at level (i-2) di-2 =
colors will be required.
(3*2i-2-1) + (3*2i-2-1) + 2*(3*2i-2-1) = (3*2i-2-1)*22. In
The next node in the list is node 1. Node 1 has already been
general, the degree of a root node at level (i-k), where k =
colored with the yellow color. All the neighbors of node 1
0,1, 2, … (i-2) is given by di-k= (3*2i-k-1)* 2k.
have been colored except for node 2, which can take the
For example, let (i=5). The total number of nodes in
color red. The next node in the list is node 2. Node 2 has
the graph is (6*25-1) = 191. d5= (3*25-1) = 95; d4= 2*(3*24 -
already been colored. All of its neighbors are colored except
1) = 94; d3= 22*(3*23 -1) = 92; d2= 23 *(3*22-1) = 88.
for node 3. Node 3 requires a new color, say the color pink.
The algorithm sorts the nodes based on node degrees
in a descending order. Thus, the algorithm will color the
At this stage all the nodes of the graph have been
nodes starting with the roots at each level.
colored with five colors, and the algorithm terminates. The
At level (i), the algorithm will color the root node and
termination criterion of the algorithm is simple. The
all its adjacent nodes. Since the adjacent nodes are not
algorithm maintains a list of all the nodes that have not been
interconnected, they can assume the same color. The root
colored. Every time a node is colored, the node will be
will have a different color. So step one, will consume two
removed from the list. The algorithm terminates when the
colors. The next step will take care of the nodes at the next
list becomes empty.
lower level. Since the root at this level is not connected to
We observe the followings from the process of coloring the
the root at the upper level, it can assume the color of the
graph.
root at the upper level; so no new color is consumed. The
1. The number of nodes and colors covered at each stage
neighbors of the root at this level are split in two sets. One
are as follows:
set belongs to the upper level, so this set has already been
a. First stage: 12 nodes = ((23+1)/2). Number of colors
colored. The second set is not inter-connected, so the nodes
used is 2
of this set can all be colored with the same color. A new
b. 2nd stage: 6 nodes = ((23+1)/4). Number of colors is
color will be consumed. Similarly, at the next level, the root
1
78 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
can assume the same colors of the previous roots, since they
are not interconnected. The neighbors of this node are split Authors Profile
in 3 sets. Two sets belong to the upper 2 levels and one set Dr. Mohammad Malkawi received his
belongs to the current level. The nodes, which belong to the Ph.D. degree in computer engineering
upper 2 levels, have been colored at the previous steps. The from the University of Illinois at Urbana-
nodes in the current level are not interconnected, so they can Champaign in 1986. He received his BS
assume one color. Continuing in this manner, the algorithm and MS in computer engineering in 1980
and 1983. He was a member of the High
will consume one color at each level until it reaches the
Productivity Computing Systems team at
lowest 5 nodes in the graph, where it will consume 2 new
SUN Microsystems, and the high availability platform team at
colors. Motorola. Currently, he is associate professor at Middle East
Thus, the algorithm detects the structure of the graph University in Jordan.
and colors the nodes with the minimal number of
Mohammad M Al-Haj Hassan obtained
colors 2 + ∑1 . The algorithm also uses minimum number
i +1
his B.Sc. and Master degrees from The
University of Jordan in 1973 and 1977
of steps because it colors the roots and their neighbors first. respectively, and his Ph.D. degree in
Computer Science from Clarkson
University at NY / USA. His main field
3. Conclusions of Specialization is Computer Algorithms
This paper presents a coloring algorithm which colors with specific concentration on Graph
a girth-g, triangle free graphs with arbitrarily large k- Algorithms & Their Applications. Other
chromatic number. The algorithm detects the structure of a fields of research are: Machine Learning,
girth-g graph and colors the graph with exactly k colors. Parallel Computations and Algorithms,
The algorithm terminates in O(Lg(n)) steps. Distance Learning and Web-Based Courses Design. Professor M.
Al-Haj Hassan was the Dean of several Faculties in several
universities. He is a member in the Editorial Board and/or referee
of several journals. He has been the Secretary General of The
International Arab Conference on Information Technology (ACIT)
References
for 4 years. Currently, he is a professor of Computer Science and
[1] Bollobas,B., Sauer,N.: ”Uniquely Colorable Graphs
the Vice President at Middle East University in Jordan.
with Large Girth,” Canad. J. Math. 28 (1976), no.
6, 1340-1344.
[2] Emden-Weinert,T., Hougardy,S., Kreuter,B.:
”Uniquely Colorable Graphs and the Hardness of
Coloring Graphs of Large Girth,” Com-bin.
Probab. Comput. 7 (1998), no. 4, 375-386.
[3] Erdos,P.: ”Graph Theory and Probability,” Canad.
J. Math 11 (1959), 34-38.
[4] Muller,V.: ”On Colorings of Graphs Without Short
Cycles,” Discrete Mathematics 26 (1979), 165-176.
[5] Nesetril,J., Zhu,X.: ”Construction of Sparse Graphs
with Prescribed Circular Colorings,” Discrete
Mathematics 233 (2001), 277-291.
[6] Steffen,E., Zhu,X.: ”Star Chromatic Numbers of
Graphs,” Combinatoria 16 (1996), no. 3, 439-448.
[7] Zhu,X.: ”Uniquely H-colorable Graphs with Large
Girth”, J. Graph Theory 23 (1996), no. 1, 33-41.
[8] Mohammad Malkawi, Mohammad Al-Haj Hassan,
and Osama Al-Haj Hassan, “A New Exam
Scheduling Algorithm Using Graph Coloring”, The
International Arab Journal of Information
Technology, Vol. 5, No. 1, 2008; pp. 80-86.
[9] Soifer, Alexander “Mathematics of Coloring and
the Colorful Life of its Creators”, Springer; pp, 83-
86; ISBN: 978-0-387-74640-1
(IJCNS) International Journal of Computer and Network Security, 79
Vol. 2, No. 5, May 2010
number of rules, while hiding many rules at a time. But they sanitized database. First goal considers privacy issue.
produce undesirable side effects. Second goal is related to the usefulness of sanitized dataset.
Third goal is related to the side effect of the sanitization
In this paper, we propose two heuristic blocking based process.
algorithms named ISARC (Increase Support of common
Antecedent of Rule Clusters) and DSCRC (Decrease 2.2 Basic Concepts
Support of common Consequent of Rule Clusters) to
2.2.1 Rule Hiding by Reducing Confidence
preserve privacy for sensitive association rules in database.
To hide sensitive rules, proposed algorithms replace the 1’s To hide an association rule like X ⇒Y, we decrease its
or 0’s by unknowns (“?”) in fewer selected transactions for confidence (|X∪Y|/|X|) to be smaller than pre-specified
minimum confidence threshold (MCT). To decrease the
decreasing their confidence. So, it is difficult for an
confidence of a rule, two strategies as in [5] can be
adversary to know the value behind unknowns (“?”). To
considered. The first one is to increase the support of rule
decrease the confidence of specified rules, first algorithm antecedent (X), but not support of generating itemset
increases the support of rule antecedent, while another (X∪Y). The second one is to decrease the support of
decreases the support of rule consequent. They can hide generating itemset (X∪Y). To decrease support of
many rules at a time in rule clusters. So, they are more generating itemset, we decrease the support of rule
efficient than other heuristic approaches. Moreover they consequent Y only in transactions containing both X and Y.
maintain data quality in database. So, it can reduce the rule confidence faster than reducing the
support of X.
The rest of this paper is organized as follows: In section
2, we discuss some related background and concepts, while
2.2.2 Increasing or Decreasing Support of item by
a detailed description of proposed algorithms (ISARC and
using Unknowns
DSCRC) is given in section 3. An example illustrating
ISARC and DSCRC is given in section 4. In section 5 we Instead of distorting transactions, we increase or decrease
support of one item at a time in a selected transaction by
discuss and analyze the performance results of both the
replacing from 0 to unknowns (“?”) or 1 to unknowns (“?”)
proposed algorithm. Finally in section 6, we conclude
respectively. A detailed overview of unknown technique to
proposed algorithms by defining some future trends.
decrease or increase support of item is given in [8].
algorithms mine the association rule from given database D Now, for each cluster, it indexes transactions which are
by using association rule mining algorithms e.g. Apriori not fully supporting to common antecedent of cluster and
algorithm [3]. Then some rules from mined rules are sorts transactions for the first cluster in ascending order of
specified as sensitive. Now the processes of proposed their sensitivity. The hiding process of ISARC tries to hide
framework are different for both the algorithms. ISARC all the sensitive rules by placing an unknown (“?”) for the
algorithm groups sensitive rules based on common L.H.S. place common antecedent item of the rule clusters, from the
item (rule antecedent) of specified rules, where DSCRC sensitive transactions. While loop continues until all the
algorithm groups sensitive rules based on common R.H.S.
rules are not hidden in cluster. A detailed overview of
item (rule consequent) of specified rules. For each rule
maxsupport, minsupport, maxconfidence and minconfidence
cluster, ISARC indexes the transactions which are not
is given in [8]. Every time in while loop it updates the
supporting to common antecedent of cluster, where DSCRC
indexes the fully supporting transaction to cluster. Then sensitivity of modified transactions in other cluster and sorts
both the algorithms calculate all the sensitivities as them for other clusters. Finally algorithm updates all the
explained in section 2.2.4. modified transactions in original database and produces
Now, both the algorithms sort the clusters based on sanitized database D’.
decreasing order of their sensitivities. For each sorted
INPUT: Source database D, Minimum Confidence Threshold
clusters, ISARC sorts the indexed transaction in ascending
(MCT), Minimum support threshold (MST), SM.
order of their sensitivities, where DSCRC sorts the indexed
transaction in descending order of their sensitivities. Then OUTPUT: The sanitized database D’.
rule hiding procedure of ISARC increases the support of
1. Begin
common antecedent item of each cluster in sorted
2. Generate association rules from the database D.
transactions, where rule hiding procedure of DSCRC 3. Selecting the Sensitive rule set Rh with single antecedent
decreases the support of common consequent item of each and consequent e.g. x⇒y.
cluster in sorted transaction, until all the rules are not 4. Clustering-based on common item in L.H.S. of the
hidden in each cluster. Rule hiding procedure increases or selected rules.
decreases the support of item based on strategy mentioned in 5. Find the sensitivity of each cluster.
section 2.2.1 and section 2.2.2. After hiding process, both 6. For each cluster, index the transactions that are not fully
the algorithms update the modified transaction into original support the common antecedent item of cluster.
7. Sort generated clusters in decreasing order of their
database and produce the final privacy aware database for
sensitivity.
releasing. 8. For the first cluster, sort selected transaction in
ascending order of their sensitivity.
9. For each cluster c∈ C
10. {
11. While(sensitive rule set Rh ∈ c is not empty)
12. {
13. Take first transaction of cluster c.
14. Place an unknown (“?”) in the transaction for the
place of common antecedent of c.
15. Update the sensitivity of modified transaction for
other cluster and sort it.
16. For i = 1 to no. of rules∈c
17. {
18. Update maxsupport of antecedent of r∈Rh∈c
and minconfidence of the rule r.
Figure 1. Proposed common framework for ISARC and
19. If(minconfidence of r < MCT-SM)
DSCRC algorithms
20. {
21. Remove rule r from Rh.
3.2 ISARC Algorithm 22. }
The proposed ISARC algorithm is shown in Fig. 2. A source 23. }
database D, minimum confidence threshold MCT, minimum 24. Take next transaction of cluster c.
support threshold MST and safety margin SM is given as 25. }
input to ISARC. Safety margin used to show how much 26. End while
below the new minimum confidence or minimum support of 27. }
rule should [8]. According to proposed framework, ISARC 28. End for
algorithm first generates number of possible association 29. Update the modified transactions in D.
rules satisfying MST and MCT from the given database D. Figure 2. Proposed ISARC (Increase Support of common
Then it selects a set of sensitive rules RH with single Antecedent of Rule Cluster) algorithm
antecedent and consequent for hiding and clusters them
based on common antecedent. Then it finds sensitivity of
each cluster and sorts them in decreasing order of their
sensitivity.
82 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
3.3 DSCRC Algorithm Both the proposed ISARC and DSCRC algorithms hide
The proposed DSCRC algorithm is shown in Fig. 3. It many rules in an iteration of hiding process and modify
clusters the sensitive rules based on common R.H.S. item of fewer transactions in database. But DSCRC hides rules
rules and sorts sensitive transaction in decreasing order of faster than ISARC, because it reduces the confidence of
their sensitivities. sensitive rule faster than ISARC. The safety margin is also
The hiding process of DSCRC tries to hide all the used in both the algorithms to provide better privacy.
sensitive rules by placing an unknown (“?”) for the place
common consequent of the rule clusters, from the sensitive
transactions. While loop continues until all the rules are not
4. Example
hidden in cluster c. Every time in while loop it updates the A sample transaction database D taken from [8] is shown in
sensitivity of modified transaction in other cluster and sorts Table 1. TID shows unique transaction number. Binary
them for other clusters. Finally algorithm updates all the valued item shows whether an item is present or absent in
modified transactions in original database and produces that transaction. Suppose MST, MCT and SM are selected
sanitized database D’, in which most of the sensitive rules 50%, 70% and 0 respectively. Table 2 shows frequent
are hidden. itemsets satisfying MST, generated from sample database D.
In following, the possible number of association rules
INPUT: Source database D, Minimum Confidence Threshold satisfying MST and MCT, generated by Apriori algorithm
(MCT), Minimum support threshold (MST).
[2] are given: 0⇒1, 1⇒0, 0⇒3, 3⇒0. Suppose the rules
OUTPUT: The sanitized database D’. 0⇒1 and 0⇒3 are specified as sensitive and should be
hidden in sanitized database. There are one different L.H.S.
1. Begin item in selected rules, named “0” and two different R.H.S.
2. Generate association rules from the database D.
items in selected rules, named “1” and “3”. As shown in
Selecting the Sensitive rule set Rh with single antecedent
and consequent e.g. x⇒y.
Table 3(a), ISARC algorithm generates one cluster based on
4. Clustering based on common item in R.H.S. of the common L.H.S. item of the selected rules, where Table 3(b)
selected rules. shows rule clusters based on common R.H.S. item of the
5. Find the sensitivity of each cluster. selected rules, generated by DSCRC algorithm.
For each cluster, index the transactions that are fully
support the rules of cluster. Table 1: Sample database D
7. Sort generated clusters in decreasing order of their
sensitivity. Item Items (Binary
8. For the first cluster, sort selected transaction in TID
s Form)
decreasing order of their sensitivity
9. For each cluster c∈ C 0 013 1101
10. { 1 1 0100
11. While(sensitive rule set Rh ∈ c is not empty) 2 023 1011
12. {
13. Take first transaction of cluster c. 3 01 1100
14. Place an unknown (“?”) in the transaction for the 4 013 1101
place of common consequent of c.
15. Update the sensitivity of modified transaction for
Table 2: Frequent itemsets in sample database D
other cluster and sort it.
16. For i = 1 to no. of rules∈c
Frequent Itemsets with
17. {
Support(%)
18. Update minsupport and minconfidence of the
rule r∈Rh∈c. 0:80, 1:80, 3:60, 01:60, 03:60
19. If(minconfidence of r < MCT-SM or
minsupport of r < MST-SM) For ISARC algorithm, cluster 1 includes sensitive rules
20. {
namely 0⇒1 and 0⇒3. For cluster 1 sensitivities of items 0,
21. Remove rule r from Rh.
22. } 1, and 3 have 2, 1 and 1 respectively. Total sensitivity for
23. } cluster-1 is 6. For DSCRC algorithm, cluster 1 includes
24. Take next transaction of cluster c. sensitive rule namely 0⇒1, where cluster 2 includes 0⇒3.
25. } For cluster 1 sensitivities of items 0 and 1 have 1 and 1
26. End while respectively, where sensitivities for items 0 and 3 in cluster
27. } 2 have 1and 1 respectively. Total sensitivity for cluster-1
28. End for and cluster-2 is 2 and 2 respectively.
29. Update the modified transactions in D.
Table 3: Clusters generated by both the algorithms modification, it updates the sensitivity of modified
transactions for other clusters. Table 5(b) shows final
Cluster-1(based on item 0), Rules {0⇒1, sanitized databases produced by DSCRC algorithm.
0⇒3}
Item Sensitivity
Table 5: Final sanitized databases produced by both the
0 2 algorithms
1 1
TID Items (Binary
3 1
Form)
Total cluster sensitivity 6 0 1101
(a). Rule clusters generated by ISARC algorithm 1 ?100
2 1011
Cluster-1(based on item 1), Cluster-2(based on item 3), 3 1100
Rule {0⇒1} Rule {0⇒3} 4 1101
Item Sensitivity Item Sensitivity (a). Final sanitized database produced by ISARC algorithm
0 1 0 1
1 1 3 1 TID Items (Binary
Total cluster Total cluster Form)
sensitivity 2 sensitivity 2 0 1?0?
(b). Rule clusters generated by DSCRC algorithm 1 0100
As shown in Table 6, performance results of ISARC and [3] J. Han, M. Kamber. Data Mining: Concepts and
DSRRC are better in terms of hiding failure (HF), misses Techniques, Morgan Kaufmann Publishers, San
cost (MC), artifactual patterns (AP), dissimilarity (DISS), Francisco, pp. 227–245, 2001.
and side effect factor (SEF). They provide certain privacy [4] S.R.M. Oliveira, O.R. Zaïane, Y. Saygın. “Secure
and knowledge discovery. Here we selected 0.1% safety Association Rule Sharing”. In Proceedings of the 8th
margin. If we select higher safety margin, both the Pacific-Asia Conference on PAKDD2004, Sydney,
algorithms provide more privacy to sensitive rules. Australia, pp. 74–85, 2004.
[5] V.S. Verykios, A.K. Elmagarmid, E. Bertino, Y.
Performance results of both the algorithms depend on
Saygin, E. Dasseni. “Association Rule Hiding”. IEEE
database to modify and specified sensitive rules.
Transactions on Knowledge and Data Engineering,
We can see that both the proposed algorithms hide many Vol. 16, no. 4, pp. 434–447, 2004.
sensitive rules successfully, while maintaining knowledge [6] S.L. Wang, A. Jafari. “Using Unknowns for Hiding
discovery in terms of data quality By analyzing Sensitive Predictive Association Rules”. In
performance results,. The number of modifications in Proceedings IEEE International Conference on
database is very less because of clustering and use of Information Reuse and Integration (IRI 2005), pp.
sensitivities. Efficiency is improved by clustering which 223–228, 2005.
helps to hide many rules at a time by modifying less [7] C. Clifton, D. Marks. “Security and Privacy
Implications of Data Mining”. In Proceedings ACM
sensitive transactions.
SIGMOD International Conference on Management of
Suppose an adversary is more intelligent and he/she Data (SIGMOD’96), pp. 15–19, 1996.
knows that the original database is modified by some safety [8] Y. Saygin, V.S. Verykios, C. Clifton. “Using
margin and thresholds. If he/she mines association rules Unknowns to Prevent Discovery of Association Rules”.
satisfying support and confidence below the MST-SM and ACM SIGMOD, Vol. 30, no. 4, pp. 45–54, 2001.
MCT-SM respectively. Then he/she can derive sensitive [9] K. Duraiswamy, D. Manjula. “Advanced Approach in
rules from the sanitized database. Therefore, if random Sensitive Rule Hiding”. Modern Applied Science, Vol.
safety margin for each rule or each rule cluster is used in 3, no. 2, 2009.
[10] Y.H. Wu, C.M. Chiang, A.L.P. Chen. “Hiding
sanitization process, it can provide more privacy to sensitive
Sensitive Association Rules with Limited Side
rules.
Effects”. IEEE Transactions on Knowledge and Data
Engineering, Vol. 19, no. 1, pp. 29–42, 2007.
6. Conclusion and Future Scope [11] Y. Saygin, V.S. Verykios, A.K. Elmagarmid. “Privacy
Preserving Association Rule Mining”. In Proceedings
In this paper, we proposed two algorithms named ISARC International Workshop on Research Issues in Data
and DSCRC to hide sensitive rules in database for privacy Engineering (RIDE 2002), pp.151–163, 2002.
issue. Both algorithms provide certain privacy to sensitive [12] C.C. Aggarwal, P.S. Yu. Privacy-Preserving Data
rules by using unknowns and maintain data quality for Mining: Models and Algorithms, Springer, Heidelberg,
knowledge discovery. Through an example we have pp. 267–286, 2008.
demonstrated proposed algorithms and analyzed [13] S. Wu, H. Wang. “Research On The Privacy Preserving
performance results. Both algorithms can hide rules which Algorithm Of Association Rule Mining In Centralized
have single antecedent or consequent and they are more Database”. International Symposiums on Information
efficient than other heuristic approaches. Processing (ISIP), pp. 131 – 134, 2008.
Proposed algorithms viz; ISARC and DSCRC can be [14] T. Brijis, G. Swinnen, K.Vanhoof and G. Wets, “The
modified to hide sensitive rules which have different number use of association rules for product assortment
of L.H.S. items or R.H.S. items. In future, a parallelization decisions: a case study,” In Proceedings of the Fifth
International Conference on Knowledge Discovery and
can be also applied to improve efficiency of proposed
Data Mining, San Diego (USA), pp. 254-260, 1999.
algorithms.
Authors Profile
References
Chirag N. Modi was born in Patan, Gujarat,
[1] M. Atallah, E. Bertino, A.K. Elmagarmid, M. Ibrahim, India, on 3rd Feb 1987. He received his B.E.
V.S. Verykios. “Disclosure Limitation of Sensitive degree in Computer Engineering from S.P.
Rules”. In Proceedings of the 1999 IEEE Knowledge University Anand (Gujarat)-India in 2008. He
and Data Engineering Exchange Workshop (KDEX is Pursuing M.Tech degree in Computer
1999), pp. 45-52, 1999. Engineering at S. V. National Institute of
[2] C. Clifton, M. Kantarcioglu, J. Vaidya. “Defining Technology Surat (Gujarat)-India. His
research interests include Data Mining and
Privacy for Data Mining”. In Proceedings US Nat'l
Database security.
Science Foundation Workshop on Next Generation
Data Mining, pp. 126-133, 2002.
(IJCNS) International Journal of Computer and Network Security, 85
Vol. 2, No. 5, May 2010
Abstract: The wide spread use of WLAN (Wireless LAN) and transfer of credit card information, financial details and
the popularity of mobile devices increases the frequency of data other important documents.
transmission among mobile users. In such scenario, a need for The basic goal of most cryptographic system is to transmit
Secure Communication arises. Secure communication is some data, termed the plaintext, in such a way that it cannot
possible through encryption of data. A lot of encryption be decoded by unauthorized agents[5][6][7][8][9]. This is
techniques have evolved over time. However, most of the data done by using a cryptographic key and algorithm to convert
encryption techniques are location-independent. Data encrypted
the plaintext into encrypted data or cipher text. Only
with such techniques can be decrypted anywhere. The
encryption technology cannot restrict the location of data
authorized agents should be able to convert the cipher text
decryption. GPS-based encryption (or geo-encryption) is an back to the plaintext.
innovative technique that uses GPS-technology to encode GPS-based encryption (or geo-encryption) is an innovative
location information into the encryption keys to provide location technique that uses GPS-technology to encode location
based security. In this paper a location-dependent approach is information into the encryption keys to provide location
proposed for mobile information system. The mobile client based security[12][13][14][15]. GPS-based encryption adds
transmits a target latitude/longitude coordinate and an LDEA another layer of security on top of existing encryption
key is obtained for data encryption to information server. The methods by restricting the decryption of a message to a
client can only decrypt the ciphertext when the coordinate particular location. It can be used with both fixed and
acquired form GPS receiver matches with the target coordinate. mobile.
For improved security, a random key (R-key) is incorporated in
The terms location-based encryption or geo-encryption are
addition to the LDEA key. The cipher text is obtained by
used to refer to any method of encryption in which the
interlacing , permuting and rotating based on the R-key.
encrypted information, called cipher text, can be decrypted
Keywords: data security, location-based key, mobile security, only at a specified location. If, someone attempts to decrypt
random generator, permutation the data at another location, the decryption process fails and
reveals no details about the original plaintext information.
The device performing the decryption determines its
1. Introduction
location using some type of location sensor such as a GPS
The dominant trend in telecommunications in recent years
receiver. Location-based encryption can be used to ensure
is towards mobile communication. The next generation
that data cannot be decrypted outside a particular facility -
network will extend today’s voice-only mobile networks to
for example, the headquarters of a government agency or
multi-service networks, able to carry data and video services
corporation or an individual's office or home. Alternatively,
alongside the traditional voice services. Wireless
it may be used to confine access to a broad geographic
communication is the fastest growing segment of
region. Time as well as space constraints can be placed on
communication industry. Wireless became a commercial
the decryption location.
success in early 1980’s with the introduction of cellular
Adding security to transmissions uses location-based
systems. Today wireless has become a critical business tool
encryption to limit the area inside which the intended
and a part of everyday life in most developed countries.
recipient can decrypt messages. The latitude/longitude
Applications of wireless range from common appliances that
coordinate of node B is used as the key for the data
are used everyday, such as cordless phones, pagers, to high
encryption in LDEA. When the target coordinate is
frequency applications such as cellular phones. The
determined, using GPS receiver, for data encryption, the
widespread deployment of cellular phones based on the
ciphertext can only be decrypted at the expected location. A
frequency reuse principle has clearly indicated the need for
toleration distance(TD) is designed to overcome the
mobility and convenience. The concept of mobility in
inaccuracy and inconsistent problem of GPS receiver. The
application is not only limited to voice transfer over the
sender can also determine the TD and the receiver can
wireless media, but also data transfer in the form of text ,
decrypt the ciphertext within the range of TD. Denning’s
alpha numeric characters and images which include the
model is effective when the sender of a message knows the
(IJCNS) International Journal of Computer and Network Security, 87
Vol. 2, No. 5, May 2010
6. Acknowledgements:
This work was supported by grants from the All India
Council for Technical Education (AICTE) project under
RPS Scheme under file No. F.No.8023/BOR/RID/RPS-
114/2008-09.
References
Authors Profile
In order to ensure that cost-effective, quality systems are Criticisms of the Prototyping Model generally fall into the
developed which address an organization’s business needs, following categories [10]:
developers employ some kind of system development • Prototyping can lead to false expectations.
Process Model to direct the project’s lifecycle [10]. To solve • Prototyping can lead to poorly designed systems.
actual problems in an industry setting, a software engineer The RAD Model: Rapid Application Development
or a team of engineers must incorporate a development (RAD) is an incremental software development process
strategy that encompasses the process, methods, and tools model that emphasizes an extremely short development life
layers and the generic phases [9]. cycle. It is a component-based construction, but high speed
Software process models often represent a networked linear sequential. If requirements are well understood and
sequence of activities, objects, transformations, and events project scopes constrained, the RAD process enables a
that embody strategies for accomplishing software evolution. development team to create “a fully functional system” with
Software process networks can be viewed as representing in very short time periods (e.g., 60 to 90 days) [9 & 11].
multiple interconnected task chains. Task chains can be The RAD approach has drawbacks [9]:
employed to characterize either prescriptive or descriptive • Large and scalable projects require sufficient human
action sequences. Prescriptive task chains are idealized resources.
plans of what actions should be accomplished, and in what • RAD requires developers and customers who are
order [11]. committed to the rapid-fire activities necessary to get a
Some of the important models have been described in the system complete in specified time. If commitment is lacking
following paragraphs. from either constituency, RAD project will fail.
The Linear Sequential Model, The linear sequential • If a system cannot be properly modularized, building
model, sometimes called the Classic or Waterfall model, the components necessary for RAD will be problematic.
proposed by Winsten Royce [11]. The linear sequential • If high performance is issue and performance is to be
model suggests a systematic, sequential approach to achieved through tuning the interfaces to system
software development that begins at the system level and components, the RAD approach may not work.
progresses through analysis, design, coding, testing, and • RAD is not appropriate when technical risks are
support [9]. high. This occurs when a new application makes heavy use
Waterfall Model has some criticisms as: of new technology or when the new software requires a high
• Real projects rarely follow the sequential flow that the degree of interoperability whit existing computer programs.
model proposes. Evolutionary Software Process Model: Evolutionary
• At the beginning of most projects there is often a great models are iterative, which is one of the modern software
deal of uncertainty about requirements and goals, and it is development processes. Evolutionary models enable
therefore difficult for customers to identify these criteria on software engineers to develop increasingly more complete
a detailed level. The model does not accommodate this versions of the software [9 & 11].
natural uncertainty very well. Incremental Models: It combines elements of the linear
• Blocking state: means in this model some project team sequential model applied respectively with the iterative
members must wait for other members of the team to philosophy of prototyping. First increment is often a core
complete dependant tasks. product. It is iterative in nature [5 & 9].
• Developing a system using the Waterfall Model can be The Spiral Model: The spiral model, proposed by [3], is
a long, painstaking process that does not yield a working an evolutionary software process model that couples the
version of the system until late in the process [10]. iterative nature of prototyping with controlled and
The Prototyping Model: A customer defines a set of systematic aspects of the linear sequential model. A spiral
general objectives for software but does not identify detailed model is divided into a number of framework activities, also
input, processing, or output requirements. In these, and called tasks or regions [9 & 11].
many other situations, a prototyping paradigm may offer the The Win-Win Spiral Model: The Customer wins by
best approach [9 &11]. getting the system or product that satisfy the majority of the
There are a few different approaches that may be followed customer’s needs and the developer wins by working to
when using the Prototyping Model [10]: realistic and achievable budgets and deadlines. Boehm’s
• Creation of the major user interfaces without any WINWIN spiral model [12] defines a set of negotiation
substantive coding in the background in order to give the activities at the beginning of each pass around the spiral.
users a “feel” for what the system will look like, Criticism of a spiral model: The risk assessment
• Development of an abbreviated version of the system component of the Spiral Model provides both developers
that performs a limited subset of functions; development of a and customers with a measuring tool that earlier Process
paper system. Models do not have. The measurement of risk is a feature
• Use of an existing system or system components to that occurs every day in real-life situations but
demonstrate some functions that will be included in the (unfortunately) not as often in the system development
developed system. industry. The practical nature of this tool helps to make the
92 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
Spiral Model a more realistic Process Model than some of 1.3 Knowledge Acquistion
its predecessors. Capturing knowledge source from expertise is known as
The Concurrent Development Model: The concurrent knowledge acquisition technique. There are different
development model is also called concurrent engineering, sources from which knowledge can be acquired such as
[12]. The concurrent process model defines a series of human experts, books, journals, reports, databases and other
events that will trigger transitions from state to state for computer system [14]. The process through which
each of the software engineering activities. A system and knowledge engineers extract knowledge called Knowledge
component activities occur simultaneously and can be Acquisition [18]. Knowledge acquisition process is the first
modeled using the state-oriented approach described phase to develop an expert system [18]. There are some
previously. Each activity on the network exists difficulties in during knowledge acquisition process i.e.
simultaneously with other activities [5, 9 & 11]. sometime problem is not very clear to knowledge engineers,
Component-Based Development: Object-Oriented knowledge engineers have lack of knowledge in particular
technologies provide the technical framework for a domain, sometime experts are not available in particular
component-based process model for Software Engineering domain [16 & 18].
[5, 9 & 11]. A general criticism of the Component-Based There are several reasons for employing an expert system
Model is that it is limited for use in object-oriented such as [14]:-
development environments. Although this environment is § Replacement of human expert
rapidly growing in popularity, it is currently used in only a § Assistant to human expert
minority of system development applications [10]. § Transfer of expertise to novice
1.1 Creating and Combining Models: From literature study it is revealed that there are a
number of knowledge acquisition techniques exist, such as
In many cases, parts and procedures from various Process
questionnaire, protocol analysis, Interview (structured and
Models are integrated to support system development. This
unstructured interview), case study analysis, observation
occurs because most models were designed to provide a
analysis, simulation and prototyping [14 & 15).
framework for achieving success only under a certain set of
Depending upon the nature of the problem, need,
circumstances. It is sometimes necessary to alter the existing
situation and domain experts, different acquisition
model to accommodate the change in circumstances, or
techniques can be used accordingly. If the domain experts
adopt or combine different models to accommodate the new
can easily be accessed, an interview is the best technique. If
circumstances. The selection of an appropriate Process
the expert are far-away from access of knowledge engineers
Model hinges primarily on two factors: organizational
or experts are much busy, then questionnaire is the most
environment and the nature of the application. Frank Land,
suitable tool. Similarly, if the target problem is complex,
from the London School of Economics, suggests that
multiple techniques can be adopted.
suitable approaches to system analysis, design, development,
and implementation be based on the relationship between
the information system and its organizational environment.
2. Problem Domain
Four categories of relationships are identified: The Knowledge acquisition process was started from
Unchanging Environment, the Turbulent Environment, the identification of critical factors that affect directly or
Uncertain Environment, and the Adaptive Environment indirectly the process models, being critical in selection of
[10]. appropriate software process model to software engineers.
This tool will be used in ESPMS (Expert System for
1.2 Expert Systems Application
Software Process Model Selection), which is available free
For decision making and for problem solving Expert of cost.
System tools are used, which is Artificial Intelligence tool, Literature study reveals that research has been conducted
they have ability to imitate the reasoning of human expert on different phases of software process models, such as
with in a specialized area [14]. Now in various areas such as requirement analysis and specification, planning of software
computer science, agriculture, engineering, geology, space project, System Design and Architectural Design etc but
technology, medicine, etc expert systems are being used and there is no proper guidance for novice software designer and
developed (Durken 1990). The knowledgebase component of engineer, how to select a process model for a particular
expert system is consists of human experts’ knowledge. The situation.
most difficult task is to acquire knowledge from experts and Ref [7] described the three laws of software process and
identification of problem in the development of expert also highlighted that there are several problems in defining
systems [7, 8]. Interview, observation analysis, case study a process in industry. Author also highlights rule of process
analysis, questionnaire, protocol analysis are the some bifurcation (Software process rules that are stated in terms
knowledge acquisition techniques to acquiring knowledge of two levels: a general statement of the rule, and a specific
and expertise in domain experts [14, 15, 16 & 17]. detailed example), and the dual hypothesis of knowledge
discovery (we can only “discover” knowledge in an
environment that contains that knowledge). Ref [6] stated
overall objective of Learning Software Organization (LSO)
(IJCNS) International Journal of Computer and Network Security, 93
Vol. 2, No. 5, May 2010
is to improve software processes and products according to universities of Pakistan. 100 domain experts along with
the strategic goals of the organization. Knowledge detailed about them were found in 124 universities of
Management (KM) was defined, because KM is prerequisite Pakistan. See Table I for detail.
of learning organization. Ref [19] stated that there is still
3.2 Design and Distribution of Questionnaire
lacking integration of software process modeling and
software process measurement by software engineers. Extracting experts’ opinion to develop standard criteria
Language construct is used to better understand the process for factors that can affect software process model’s
modeling technique. The Attribute Grammar (AG) approach performance is the main purpose of this questionnaire.
to realize modeling of software processes modeling in Questionnaire was the tool through which we could solicit
language construct is based on its specification and experts’ opinion. A total of 99 questions (parameters) were
automatic construction of language-based editors. Ref [20] identified, which were grouped into 11 main factors. Five
Build Knowledge Analysis System (KANAL), which relates fuzzy variables scales were defined, like, 5=Strongly Agree
pieces of information in process models among themselves (Critical to process’s performance), 4 = Agree (Important to
and to the existing Knowledge Base (KB), analyzing how process’s performance), 3= Neutral (Helpful to process’s
different pieces of input are put together to achieve some performance), 2= Disagree (Minimally affects process’s
effect. It builds interdependency models from this analysis performance), 1= Strongly Disagree (Do not affect process’s
and uses them to find errors and propose fixes. Ref [21] performance).
Recent research in software engineering has moved away We designed the questionnaire in such a way that the
from this linear approach of the life cycle in order to take respondent have to rank the importance of factor regarding
account of dependencies between decisions taken in to its objective. Among 100 domain experts we received 34
different parts of the development and to allow methods to experts’ opinions with valuable and remarkable suggestions.
be selected during developments which are appropriate to
address problems that arise. The most advanced approach is Table 1: Experts Detail with their Responses
the spiral model proposed by Boehm [3]. The major research
Experts’ Subject
project on KBS methodology is the KADS (Knowledge Charac.
Gender Designation Degree
Specialty
Demographic
critical and important component of the expert system [22]. The questionnaire was resulted in 11 main groups with 84
The questionnaire was properly evaluated by domain questions after repeatedly contact with the domain experts.
experts, which will lead to become a standard document for Statistical analysis showed that System Integration &
selection of appropriate software process model. Testing, User Community & Unit Development, Software
In Pakistan SE and AI are used mostly for research in Integration & Testing were the most important factors
Education, that’s why this questionnaire was distributed in among rest of the main factors. See Table 3 for detail.
various Higher Education Commission (HEC) recognized
94 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
The ranking of factors were also helpful in determining model but it is just an approach, guideline and helping tool
the weights of the various factors, which influence the for software engineer in selection of process model. This
decision making process. questionnaire may become a standard Knowledge
Acquisition tool for Expert System development.
Table 3: Response Summary for Affect on Proces’
Performance
(Total Reponses N: 34, 5= Critical Affects, 1= Do not Affect) Domain Experts
Std. (Books, Journals, Databases, Reports, Internet etc)
Main Groups of Factors Mean
Deviation
System Integration and Testing 3.8941 1.1494 Reference Knowledge
User Community 3.4510 1.1441
Unit Development, Software Integration &
Testing 3.7426 1.0890 Characteristics of Software Process Models
Questionnaire
Table 4 depicts weights assigned to the main factors. In
weight criteria documentation, system integration and
testing and validation and verification were the most Figure 1: Knowledge Elicitation Process of a Software
important factors among rest of the main factors. Process Model Selection
Table 4: Weight Assignment to main factors
Weigh References
S.No Main Groups of Factors
ts
1 Documentation 0.1041 [1] Hoiser, W.A., “Pitfalls and Safegaurds in Rea-Time
2 System Integration and Testing 0.0971 Digital Systems with Emphasis on programming,” IRE
3 Validation and verification 0.0969 Trans. Engineering Management, EM-8, June 1961.
4 System Design and Architectural Design 0.0956 [2] Royce, W. W., “Managing the Development of Large
5 Quality Assurance and Quality Control 0.0952 Software Systems,” Proc. 9th. Intern. Conf.Software
6 Project Team 0.0923 Engineering, ,IEEE Computer Society, 1987 ,328-338
7 Implementation & Maintenance 0.0915 Originally published in Proc. WESCON, 1970.
Requirements Analysis And [3] Boehm, B., “ A Spiral Model for Software
8 0.0914
Specification Development and Enhancement,” Computer, Vol. 21,
9 Project Type and Risk Management 0.0849 no. 5, May 1988, pp. 61-72
Unit Development, Software Integration
10 0.0793 [4] Scacch i W., Process Models in Software
& Testing
Engineering, Institute for Software Research,
11 User Community 0.0718
Total weight: 1.0000 University of California, Irvine February 2001.
[5] Sommerville I., Software Engineering 5th Edition, 2000
After calculation of mean and standard deviation of main [6] Ruhe G. (2000), “ Learning Software Organizations”,
factors, we calculated mean and standard deviation of each Fraunhofer Institute for Experimental Software
and every sub-factor (questions). Engineering (IESE).
[7] Caesar & Cleopatra (2004), “Chapter 1,The Nature of
Software and the Laws of Software Process” www.ism-
4. Conclusions and Future Recommendation
journal.com/ITToday/AU1489_C01.pdf, CRC Press,
Each and every factor was properly analyzed, means and LLC.
standard deviations were calculated of each factors with the [8] Basili V.R, Zelkowitz, M. V. McGarry, F. Page, S.
help of SPSS. Weights of the main factors were evaluated. Waligora, R. Pajerski. SEL’s Software Process
Overall score will be calculated at run time through Improvement Program, IEEE Software, vol. 12,
summation function, which will be used for ranking in the November, pp 83-87, 1995.
selection of a process model. (See my research thesis for [9] Pressman R. S, Software Engineering a Practitioner ’s
Approach, Fifth Edition, 2001.
detail).
[10] SUNY 1988 , “ A Sur vey of System Development
It is further recommended that this questionnaire is not in
Pr ocess Models ” , CTG.MFA – 00, Cen ter for
final shape which will be distributed in local (Pakistan) and
Techn ology in Gover nment, Uni ver sity at
international software houses. This questionnaire is not final Alban y.
standard tool for selection of an appropriate software process
(IJCNS) International Journal of Computer and Network Security, 95
Vol. 2, No. 5, May 2010
[11] Reddy A. R. M, Govindarajulu P., Naidu M., A [27] Liu & Perry (2004), “On the Meaning of Software
Process Model for Software Architecture, IJCSNS, Architecture, Interview Questionnaire, (July, 2004),
VOL.7 No.4, April 2007. Version 1.2
[12] Davis, A. and Sitaram P., “ A Concurrent Process
Model for Software Development, ” Software Authors Profile
Engineering Notes, ACM Press, vol. 19, n0. 2, April
1994, pp. 38-51 Abdur Rashid Khan is presently working as an Associate
[13] Curtis, B., H. Krasner, and N. Iscoe, A Field Study of Professor at ICIT, Gomal University D.I.Khan, Pakistan. He has
received his PhD degree from Kyrgyz Republic in 2004. His
the Software Design Process for Large Systems, research interest includes AI, Software Engineering, MIS and DSS.
Communications ACM, 31, 11, 1268-1287, November,
1988. Zia Ur Rehman is currently pursuing his MS degree in Computer
[14] Awad, E.M. “Building Experts Systems: Principals, Science from the Institute of Information Technology, Kohat
Procedures, and Applications,” New York: West University of Science & Technology (KUST), Kohat, Pakistan. His
area of interest includes Software Engineering, Artificial
Publishing Company. 1996. Intelligence/Expert System, Databases and Data Mining.
[15] Durkin, J. “Application of Expert Systems in the
Sciences,” OHIO J. SCI. Vol. 90 (5), pp. 171-179,
1990.
[16] Medsker, L. and Liebowits, J. “Design and
Development of Experts Systems and Neural
Networks,” New York: Macmillan Publishing
Company, 1987.
[17] Sagheb-Tehrani, M.“The design process of expert
systems development: some concerns, Expert system,”
Vol. 23 (2), pp. 116-125, 2006.
[18] Wagner, W.P., Najdawi, M.K., Chung, Q.B “Selection
of Knowledge Acquisition techniques based upon the
problem domain characteristics of the production and
operations management expert systems,” Expert
system, vol. 18 (2), pp. 76-87, 2001.
[19] Atan, R. Abd, A. A. Ghani, M. H.S. & Mahmod, R. “
Software Process Modelling using Attribute Grammar”
University Putra Malaysia, Serdang, Selangor Darul
Ehsan, Malaysia, IJCSNS International Journal of
Computer Science and Network Security, VOL.7 No.8,
August 2007.
[20] Kim, J. & Gil, Y. “Knowledge Analysis on Process
Models”, Information Sciences Institute University of
Southern California, International Joint Conference on
Artificial Intelligence (IJCAI-2001), Seattle,
Washington, USA, August 2001.
[21] Wilson, “Knowledge Acquisition: The current
position”, Science and Engineering Research Council
Rutherford Appleton Laboratory.
[22] Lightfoot,J.M.” Expert knowledge acquisition and the
unwilling experts: a knowledge engineering
perspective,” Expert system, vol. 16(3), pp.141-147,
1999.
[23] Futrell R. T., Shafer L. I, Shafer D. F, “Quality
Software Project Management”, Low Price Edition,
2004.
[24] Anderson S., & Felici M., “Requirement Engineering
Questionnaire”, Version 1.0, Laboratory for
Foundation of Computer Science, Edinburgh EH9 3JZ,
Scotland, UK (2001).
[25] Skonicki M., “QA/QC Questionnaire for Software
Suppliers”, January 2006.
[26] Energy U. S. D., “Project Planning Questionnaire”,
www.cio.energy.gov/Plnquest.pdf (last access, 04
August 2009)
96 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
Appendix: (Questionnaire)
A SURVEY ON FACTORS AFFECTING TO SELECT A SOFTWARE DEVELOPMENT PROCESS MODEL
Respondent’s Name: _______________ Qualification: _________ Designation: ______________ Address/University: __________________________
INSTRUCTIONS
Different software process model exist, to select one for a particular software depends upon the nature of project, problem structure, risk involve, budget
estimation, goals & objectives of the project etc. As an EXPERT in this area, you are requested to examine each item in terms of suitability and then to explain
the degree of your agreement to each item whether, in your opinion, it would measure the factors affecting the selection of software development process model
and to what extent. You may recommend new & delete unnecessary items from the existing scale. Your in-time response will highly be appreciated. Please, use
the scale below to mark (P) your responses in the area provided.
5= Strongly Agree 4= Agree 3=Neutral 2= Disagree 1=Strongly Disagree
Critical to Process’s Important to Process’s Helpful to Process’s Minimally affects Process’s Do not affect Process’s performance
Performance Performance performance performance
ATTRIBUTES LEVELS
5 4 3 2 1
1 Requirements Analysis And Specification
1.1 Are the requirements precisely defined, well defined and well known?
1.2 Are requirements sufficient to help you understand the problem structure?
1.3 Are the requirements defined early in the cycle [23]?
1.4 Are the requirements will change often in the cycle [23]?
1.5 Is there need to demonstrate capability proof of concept [23]?
1.6 Does the requirement indicate a complex system/simple system [23]?
1.7 Is there appropriate procedure prepared for cost/benefit analysis [24]?
1.8 Does the requirements phase meet the requirements methodology [24]?
1.9 Are you sensible to organizational and political factors which influence requirements sources while eliciting requirements [24]?
1.10 Have you reuse requirements from other systems which have been developed in the same application area [24]?
1.11 How do you handle continuous requirements change [24]?
1.12 Does any risk analysis are performed on requirements [24]?
1.13 Is the requirements document easy to change [24]?
1.14 Do you define Cost-effectiveness criteria [24]?
1.15 Do you identify accuracy and completeness risks on the requirements analysis [24]?
1.16 Are rules established on handling inaccurate and incomplete data [24]?
1.17 Do you cross-check operational and functional requirements against safety (or security, availability, etc.) requirements [24]?
1.18 Are specifications kept up-to-date and controlled [25]?
1.19 Are your software-product versions reviewed for conformity to customer’s specifications [25]?
1.20 Are the versions reviewed for compliance to quality requirements before submission for customer approval [25]?
2 Project Team
2.1 Does the majority of team members new to the problem domain [23]?
2.2 Does the majority of team members new to the technology domain [23]?
2.3 Is there need to train to project team [23]?
2.4 Is the team more comfortable with structure than flexibility [23]?
2.5 Will the project manager closely track the team’s progress [23]?
2.6 Is ease of resource allocation important [23]?
2.7 Does the team accept peer reviews and inspections, management/customer reviews, and mile-stones [23]?
3 User Community
3.1 Will the availability of the user representatives be restricted or limited during the life cycle [23]?
3.2 Are the user experts in the problem domain [23]?
3.3 Do the users want to be involved in all phases of the life cycle [23]?
4 Project Type and Risk
4.1 Dose the customer want to track project progress [23]?
4.2 Does the project identify a new product direction for the organization [23]?
4.3 Is the project a system integration project [23]?
4.4 Is the project an enhancement to an existing system [23]?
4.5 Is the funding for the project expected to be stable throughout the life cycle [23]?
4.6 Is the product expected to have a long life in the organization [23]?
4.7 Is high reliability a must [23]?
4.8 Is the system expected to be modified perhaps in ways not anticipated, post-deployment [23]?
4.9 Is the schedule constrained [23]?
4.10 Are the modules interfaces clean [23]?
4.11 Are reusable components available [23]?
(IJCNS) International Journal of Computer and Network Security, 97
Vol. 2, No. 5, May 2010
4.12 Are resources (time, money, tools, and people) scarce [23]?
4.13 What measurements will be used to track this project (Cost, Schedule Effort, LOC, Defects, Function Points Other ______) [26]?
5 System Design and Architectural Design
5.1 Does a design phase schedule exists which identifies tasks, people, budgets and costs?
5.2 Do you develop complementary system models [24]?
5.3 Do you model the system’s environment [24]?
5.4 Do you model the system architecture [24]?
5.5 Do you use structured methods for system modeling [24]?
5.6 Do you define operational processes to reveal process requirements and requirements constraints [24]?
5.7 Do you use a data dictionary [24]?
5.8 Do you document the links between stakeholder requirements and system [24]?
5.9 Do you specify systems using formal specifications [24]?
5.10 Can the data required by the application be collected with the desired degree of reliability [24]?
5.11 Do you identify non functional requirements (e.g., usability, quality, cognitive workload, etc.) for a system [24]?
5.12 Are significant software changes expected during the life of the project [24]?
6 Unit Development, Software Integration & Testing
6.1 Will the application system be run in multiple locations [24]?
6.2 If an on-line application, will different types of terminal be used [24]?
6.3 Is the proposed solution dependent on specific hardware [24]?
6.4 Have the portability requirements been documented [24]?
7 Quality Assurance and Quality Control
7.1 Have you written Quality Assurance plan? Or written record for Quality Assurance program [25]?
7.2 Have you performed internal quality audits on software design processes and configuration management functions [25]?
7.3 Is there any implementation of a formal correction action process for customer complaints [25]?
8 System Integration and Testing
8.1 Are software depends on Operating Environment(s)? (MS Windows, OS/2, Unix/AIX, Windows N/T, Macintosh, Novell, CICS,
DOS, Sun, Web Browser (specify), other ______________) [26]?
8.2 Are constraints effects the targeted operating environment(s)? (Firewalls, Transmission speed, Server capacity, Security,
Workstation capacity, Remote access capability, Widely dispersed user community without appropriate inter-operability Other
[26]?
8.3 Which programming languages you will be, or are considering used in project? (Cobol, Java, C++, Visual Basic, FoxPro, Paradox,
HTML, Delphi, Visual C++, Power Builder, Others ____________) [26]?
8.4 Are the software’s operations (System/Application) so essential that data must be immediately available at all times or
recoverable? (Data recovery, Backups, Fault Tolerance, System Performance, Mirroring/Imaging Disaster Recovery) [26]?
8.5 Which individual, team, and organizational information systems development and management practices will be implemented on
this project? Project Planning, Requirements Management, Configuration Management, Project Tracking and Oversight, Quality
Assurance, Sub-Contractor Management, Risk Assessment, Peer Reviews, Training Program Software Product Engineering, Inter-
group Coordination, Integrated Software Management, Organizational Process Definition, Organizational Process Focus Defect
Tracking, Other_________________________ [26]?
9 Validation and verification
9.1 Is there any implementation of formal process for functional requirement review [25]?
9.2 Is there any implementation of formal process for system design review [25]?
9.3 Is there any implementation of formal process for code testing [25]?
9.4 Is there any implementation of formal process for integration testing [25]?
10 Documentation
10.1 Are written test plans, instructions, product specifications, and system design documents available to support the tests [25]?
10.2 Are functional requirements properly documented [25]?
10.3 Have you created system programmers documentation [25]?
10.4 Have you documented functional specification [25]?
10.5 Have you documented the system design [25]?
11 Implementation & Maintenance
11.1 Do you implement a formal set of code testing procedures?
11.2 Do you implement a formal set of integration testing procedures?
11.3 Have you implemented a formal process for acceptance testing by the User / customer?
11.4 Are you usually concerned with reusability while designing [27]?
11.5 Do you involve external (from the project) reviewers in the validation process [24]?
11.6 Has the expected life of the project been defined?
11.7 Has the importance of keeping the system up to date functionally been defined [24]?
11.8 Which medium will be used for the project/system documentation? Web Site, Hardcopy, CD Rom, Online Help, Video Quick
Reference Card Diskette, Other [26]?
Comments:___________________________________________________________________________________________________________________________________
Date: ____________________ Signature: ______________
98 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
Abstract: Ethernet Passive Optical Network(EPON) is one of One of the main concerns in this network is bandwidth
the best solutions for implementing the next generation allocation to provide end users quality of service for using
access networks. One of the key issue in these network is multimedia service. To solve this problem, there have
bandwidth allocation to provide the quality of service for the suggested different static and dynamic bandwidth allocation
End users.In this paper, we present a Novel Dynamic methods [2], [4]. In this paper, along with studying
Bandwidth Allocation (DBA) method based on Second Price bandwidth allocation methods in EPON, we present a novel
Auction in Ethernet Passive Optical Network. Results have method named MOBINA, based on Second Price Auction.
shown improvements of the packet loss rate, throughput The MOBINA can allocate the users required bandwidth
rate, and utilization line of our method when compared righteously and with high profitability.
with other methods. The rest of this paper is organized as follows: In Section 2,
we introduce related work of Bandwidth allocation methods
Keywords: Ethernet Passive Optical Network, Dynamic in EPON. In section 3, we explain MOBINA method in
Bandwidth Allocation, Second price auction, Quality of Services.
details. In section 4, we present simulation results. In
section 5, we conclude this paper.
1. Introduction
Ethernet Passive Optical Network is one of the most 2. Related work
important solutions for implementing the next generation of
Bandwidth allocation methods are discussed as one of the
access networks. High capacity and low cost of
most important QOS’s parameters in EPON [6]. In
implementing are among the characteristics of this
bandwidth allocating for each network unit, optical line
technology, which can provide the required quality service
terminal needs to provide some method based on the
in spite of expansion the quantity of user and their requests
received bandwidth demands of optical network units and
to use different multimedia services [1].
some allocation policies or service level agreements are in
An EPON system is a point-to-multipoint fiber optical
this content [7]. At present many methods are
network with no active elements in the transmission path
recommended, in general they classified in two general
from its source, i.e., an optical line terminal (OLT), to a
types: Static Bandwidth Allocation and Dynamic Bandwidth
destination, i.e., an optical network unit (ONU). It can use
Allocation.
different multipoint topologies, such as bus, ring, tree, and
different network architectures [2]. The most typical EPON 2.1 Static Bandwidth Allocation methods
architecture is based on a tree topology and consists of an Each ONU is assigned with a constant bandwidth regardless
OLT, a 1:N passive star coupler (or splitter/combiner), and to its actual bandwidth demand. As a result, the allocated
multiple ONUs, as shown in Figure (1). The OLT resides in bandwidth might either not be fully utilized, in the case of
a central office (CO) that connects the access network to a ONUs with light traffic requirement or not be adequate to
metropolitan area network (MAN) or a wide area network accommodate heavy traffic load ONUs, resulting in both
(WAN), and is connected to the passive star coupler through scenarios to inefficient bandwidth utilization rate. In
a single optical fiber [2]- [5]. contrast a dynamic MAC protocol could more successfully
allocate bandwidth according to ONUs’ instantaneous buffer
queue status increasing the bandwidth utilization rate [8].
Research initiatives in MAC protocol development over the
years have attempted to enhance further bandwidth
allocation of standardized PON topologies, tracing back to
PONs and more intensively to more recent and lately
deployed EPONs . In common to all these topologies, the
Figure 1. EPON’s Architecture feeder pathway is shared in the time domain, and as a result
each ONU can utilize the whole upstream optical carrier
(IJCNS) International Journal of Computer and Network Security, 99
Vol. 2, No. 5, May 2010
capacity for defined, according to network penetration, and providers [13]. The primary SLA specifies those services
in most cases flexible duration time-slots, assigned by the whose minimum requirements must be guaranteed with a
OLT [7]. high priority. The secondary SLA describes the service
requirements with a lower priority. This method first
allocates timeslots to those services with the primary SLA to
2.2 Dynamic Bandwidth Allocation methods
guarantee their upstream transmissions. After the services
To increase bandwidth utilization, the OLT must with the primary SLA are guaranteed, the next round is to
dynamically allocate a variable timeslot to each ONU based accommodate the secondary SLA services. If the bandwidth
on the instantaneous bandwidth demand of the ONUs. is not sufficient to accommodate the secondary SLA
Given that QoS is the main concern in EPONs, these services, the max-min policy is adopted to allocate the
methods classified into DBA with QoS support, DBA bandwidth with fairness. If there is excessive bandwidth,
without QoS support, and describe their characteristics and FSD-SLA will allocate the bandwidth to the primary SLA
performances [11]. entities first and then to the secondary SLA entities, both by
using max-min fair allocation.
2.2.1 DBA without differentiated QoS support Another optimizing method for dynamic bandwidth
There are several DBA methods proposed for EPON which allocation with differentiated QoS support is LSTP .during
do not support differentiated services, such as IPACT [9], it the waiting period and thus more accurately grants
employs a resource negotiation process to facilitate queue bandwidth to each ONU. For each class of traffic, LSTP
report and bandwidth allocation. The OLT polls ONUs and estimates the data that arrive during the waiting period
grants timeslots to each ONU in a round robin fashion. The based on the data of this class that actually arrived in
timeslot granted to an ONU is determined by the queue previous transmission cycles by using a linear predictor
status reported from that ONU. Therefore, the OLT is able [14].
to know the dynamic traffic load of each ONU and allocate The bandwidth demand of an ONU is thus the reported
the upstream bandwidth in accordance with the bandwidth queue length plus the estimation. The OLT arbitrates the
demand of each ONU. So demanded bandwidth can be upstream bandwidth using this estimation and reserves a
granted efficiently also there exists three different ways for portion of the upstream bandwidth for transmitting the
the OLT to determine the granted efficiently also there estimated data in the earliest transmission cycle, thus
exists three different ways for the OLT to determine the reducing packet delay and loss. Other methods proposed for
granted window size to allocate bandwidth so they dynamic bandwidth allocation with differentiated QoS
introduced Fixed service, Gated service, Limited service as support are: COPS, HGP, DBAM and so forth [2].
IPACT service discipline to demonstrate bandwidth
allocation in this method . Development and improvement
of this method is considered in IPACT GE [10], Estimation- 3. The MOBINA DBA Method
based DBA [11] and IPACT with SARF [2].
Bandwidth guaranteed polling (BGP) is another DBA
method proposed for providing bandwidth guarantees in The MOBINA is dynamic bandwidth allocation based on
EPONs [12]. In BGP, all ONUs are divided into two groups: Second Price Auction. In this method, all bandwidth
bandwidth guaranteed and bandwidth non-guaranteed. The requests are sent to OLT by ONUs. OLT can make a
OLT performs bandwidth allocation through using couple of dynamic and fair decision about bandwidth allocation by
polling tables. The number of bandwidth units allocated to auction process. The steps of auction process and
an ONU is determined by the bandwidth demand of that bandwidth allocation in the suggested method are as
ONU, which is given by its SLA with a service provider. A follows:
bandwidth guaranteed ONU with more than one entry in the
poling table has its entries spread through the table. This Step 1: Kick-off the auction by OLT
can reduce the average queuing delay because the ONU is Step 2: Submit the bid by ONUs
polled more frequently. However, this leads to more grants Step 3: Assess received bids by OLT
in a cycle and thus requires more guard times between Step 4: Schedule and allocate bandwidth to winners
grants, which reduces channel utilization. Methods have Step5: Update and go to next round of auction
been introduced for all best effort traffic is suitable.
Figure (2) represents the block diagram of the MOBINA
2.2.2 DBA with differentiated QoS support method.
An EPON system is expected to deliver not only the best
effort data traffic, but also real-time data traffic (e.g., voice
and video) that have strict bandwidth, packet delay, and
delay jitter requirements. In this subsection present several
DBA methods that can provide differentiated QoS support
for different types of data traffic in EPON. One of the most
important methods in field is fair sharing with dual SLAs
(FSD-SLA) proposed a fair sharing with dual SLAs (FSD-
SLA) method, which employs dual SLAs in IPACT to
manage the fairness for both subscribers and service
100 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
0 (ONU i win )
LSTi onu
(t ) = (5)
1 (ONU
Figure 2.Dialgue diagram of the MOBINA method i loss )
Finally, after sending LST ionu values, OLT holds for a time
As shown in Figure (2), in Step 1 a signal is sent from OLT
to ONUs in order to ONUs send their requested bandwidth in order to allocated bandwidth be released and the
allocation to OLT for participation in an auction. In step 2 conditions for holding auction by OLT is provided.
based on the MOBINA method, ONUs calculate their Bid
amount based on two parameters of the requested bandwidth 4. Computer Simulations
and the last time of bandwidth allocation by OLT to ONU, To measure the performance of each bandwidth allocation
and send it to OLT. Equation (1) shows the sent Bid amount methods we designed an event-driven C++ based EPON
by ONUi to OLT. simulator. NS-2 and C# software is used as a unique system
to compare simulation results. In this article we use EPON
(
Bid i (t ) = β BRQionu (t ), LSTionu (t − 1) ) (1) structure, as shown in Figure (1), to measure QoS
parameters. The parameters considered for simulation are as
illustrated in Table (1):
In Equation (1), BRQionu is the total bandwidth amount that
has been requested by the users from ONUi and LST ionu Is Table 1: Simulation Parameters
Parameters Symbol Value
the Boolean parameter that shown the last time of ONUi
Number of ONUs N 16
received bandwidth service. Equation (2) shows ONUi
Bit Rate Λ 5 to 57.5 MBit/s
requested bandwidth. Two way delay fiber Tfiber 200 µs
Processing Time Tproc 35 µs
K
∑ BW
Packet Size B 15000Byte(30
BRQionu (t ) = J
R eq
(t ) (2) Packets)
J =1 Ethernet Overhead Beth 38 Byte
Request Message Size Breq 570 Bit
In Equation (2), BWJR eq is the bandwidth required by users Upstream Bandwidth Ru 1 GBit/s
connected to ONUi. Equation (3) shows the required Maximum Transition Pmax 10 Packets
Window
bandwidth of each user.
Guard Time Tg 5 µs
Max cycling time TMax 2 ms
BWJReq (t ) = mJP × LJP × PJ (3) Buffer capacity BQmax 10 Mbyte
Guarantied Bandwidth BAga 60 MBit/s
Interval traffic randomly Interval 50 µs-100 µs
In Equation (3), m JP is the number of requested packets to
Traffic Type CBR
send for user j. LPJ is the length of requested packets to
send, and PJ is the priority of the jth user. In MOBINA All the provided results in this section which will be
introduced, are measured the QoS based on proposed
suggested method, the priority of all users linked to an ONU
approach in comparison with other methods, will be
is the same and is equal to (1/j).
analyzed by some different charts with their quantities such
In step 3, after receiving all the Bids based on second price
auction, OLT will select ONUs that win in auction and will
(IJCNS) International Journal of Computer and Network Security, 101
Vol. 2, No. 5, May 2010
5. Conclusion
Figure 4.The Charts for Average of Packet Jitter Bandwidth allocation is a critical issue in the design of an
EPON system. Since multiple ONUs share a common
In addition, The chart related to throughput and loss rate of upstream channel, an EPON system must efficiently utilize
this traffic Loads is shown in Figure (5) and (6). As a result the limited upstream bandwidth in order to meet the
of analyzing these charts that implied better and more bandwidth demands and quality of service requirements of
sufficient throughput compared to the other methods that we end users. In this article, MOBINA method optimizes the
mentioned completely. Also we can realize that packet lost bandwidth limitation problem. The results of the
is reduced by increasing traffic load. simulations showed that in comparison to other methods,
the proposed method offers better performance in terms of
average packet loss ratio and throughput ratio. And as a
result, we improve the overall network performance.
102 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
References
[1] G. Kramer and G. Pesavento, "Ethernet passive optical Authors Profile
network (EPON):building a next-generation optical
access network," IEEE Communications Magazine, Alireza Hedayati:. received his B.Sc. and
M.Sc. in Computer Hardware Engineering
pp. 66-73, 2002.
from Azad University of Tehran Central
[2] J. Zhenga, T. Hussein, ”A survey of dynamic
Branch and Tehran science and research
bandwidth allocation methods for Ethernet Passive branch in 2000 and 2003, respectively. He is
Optical Networks ”, Elsevier Journal Optical currently studying Computer Hardware in his
Switching and Networking, pp. 151-162, 2009. Ph.D. degree. His research interests include
[3] A. Xiaofeng ,s. Abdallah , “On the fairness of dynamic Optical Networks, Next Generation Network, Network
bandwidth allocation schemes in Ethernet passive Management and QoS.
optical networks”, Elsevier journal of computer
communications, pp. 2123-2135, 2008. Mehdi N. Feshaaraki:. received his Ph.D.
[4] H. Naser, H. Mouftah, “A joint-ONU interval-based in computer Engineering from NSW
university of Australia. He is currently
dynamic scheduling algorithm for Ethernet passive
associate professor in computer Engineering
optical networks”, IEEE/ACM Transactions on Department of IAU. . His research interests
Networking ,pp. 889-899 , 2006. include Computer Networks, information and
[5] Z. J., "Efficient Bandwidth Allocation Method for knowledge architecture.
Ethernet Passive Optical Networks," in Proceedings
of IEE Communications, pp.464-468, 2006. Kambiz Badie:. received his B.Sc., M.,Sc, and
[6] X. Bai, A. Shami, C. Assi, Statistical Bandwidth Ph.D. in electronic engineering from the Tokyo
Multiplexing in Ethernet Passive Optical Networks, Institute of Technology, Japan, majoring in
in Proceedings of IEEE Global Telecommunications Pattern Recognition & Arlificia1lntelligance. His
major research interests are Machine Learning,
Conference, 2005.
Cognitive Modeling, and Systematic Knowledge
[7] J. Zheng, H. T. Mouftah, "An Adaptive MAC Polling Processing in general, and Analogical
Protocol for Ethernet Passive Optical Networks", in Knowledge Processing, Experience-Based
Proceedings of IEEE ICC, pp. 1874-1878, 2005. Modeling and Interpretative Modeling in particular with emphasis
[8] B. Moon, “Emergency handling in Ethernet passive on ides and technique generation .He is currently associate
optical networks using priority-based dynamic professor in Iran Telecom Research Center(ITRC).
bandwidth allocation”, in Proceeding of IEEE
INFOCOM'08, pp. 1319_1327, 2008. . Ahmad Khadem zadeh : He received the B.Sc.
[9] S. Bhatia, R. Bartos, IPACT with smallest available degree in applied physics from Ferdowsi
University, Meshed, Iran, in1969 and the M.Sc.,
report first: A new DBA method for EPON, in
Ph.D. degrees respectively in Digital
Proceeding. of IEEE ICC'07, pp. 2168_2173, 2007. Communication and Information Theory & Error
[10] H. Song, B.-W. Kim, B. Mukherjee, Multi-thread Control Coding from the University of Kent,
polling: A dynamic bandwidth distribution scheme in Canterbury, U. K .He is currently associate professor in Iran
long-reach PON, IEEE Journal on Selected Areas in Telecom Research Center and also He is the Head of Education &
Communications ,) pp. 134-142, 2009. National Scientific and Informational Scientific Cooperation
[11] Y. Zhu, M. Ma, “IPACT with grant estimation Department at Iran Telecom Research Center (ITRC).
(IPACT-GE) scheme for Ethernet passive optical
networks”, IEEE/OSA Journal of Lightwave Masoud Davari:. received his B.Sc. in
Computer Engineering from Azad University
Technology, pp. 2055-2063, 2008.
of Tehran Central Branch 2010. His research
[12] M. Ma, Y. Zhu, T. Cheng, “A bandwidth guaranteed interests include Computer network, Network
polling MAC protocol for Ethernet passive optical simulation and Software Engineering
Networks”, in Proceeding of IEEE INFOCOM'03, pp.
22_31, 2003.
[13] A. Banerjee, G. Kramer, B. Mukherjee, “Fair sharing
using dual service-level agreements to achieve open
access in a passive optical network”, IEEE Journal on
Selected Areas in Communications, pp. 32_44, 2006.
[14] Y. Luo, N. Ansari, “Limited sharing with traffic
prediction for dynamic bandwidth allocation and QoS
provisioning over EPONs”, OSA Journal of Optical
Networking, pp.561_572, 2005 .
(IJCNS) International Journal of Computer and Network Security, 103
Vol. 2, No. 5, May 2010
Abstract: mobile ad hoc networks (MANETs) are dynamically inside by compromised hosts. Without the appropriate
reconfigured networks in which security is a major concern. security precautions, critical applications for commercial or
MANETs face serious security problems due to their unique military use cannot employ networking technologies.
characteristics such as mobility, dynamic topology and lack of Key management is an essential cryptographic
central infrastructure support. Key management is crucial part primitive upon which other security primitives such as
of security, this issue is even bigger in MANETs. The privacy, authenticity and integrity are built. However, none
distribution of encryption keys in an authenticated manner is a
of the existing key management schemes are suitable for ad
difficult task. Because of dynamic nature of MANETs, when a
node leaves or joins it need to generate new session key to
hoc networks. The major limitation of these schemes is that
maintain forward and backward secrecy. In this paper we divide most of them rely on a trusted third party (TTP), thus not
the network into clusters. Cluster head will maintain the group fulfilling the self-organization requirement of an ad hoc
key, it will also update the group key whenever there is a change network. Special mechanisms and protocols designed
in the membership. Here the re-keying process will be specifically for ad hoc networks are necessary. Key
performed only if there is any movement of nodes within the management deals with key generation, storage,
clusters. So the computation and communication cost will be distribution, updating, and revocation and certificate service
reduced. And also we provide authentication between in accordance with security policies.
communicating nodes both in inter and intra cluster. The Due to dynamic behavior of the MANET, secret
network life time will be extended with the help of monitoring key used for communication is need to be updated whenever
node. The performance results prove the effectiveness of our key
any node joins or leaves the network in order to maintain
management scheme.
the forward and backward secrecy with in the network. If
the network is large and also the mobility is higher,
Keywords: Network Security, Key Management, Key Update,
updating of the key will be more frequent. It will consume
Mobile Networks, clusters
more computation power and also communication power of
1. Introduction nodes. So in our proposal we divide the network into
clusters consisting of small group of nodes. Here the re-
We encounter new types of security problems in ad hoc keying process will be performed only if there is any
networks because these networks have little or no support of movement of nodes with in the clusters.
infrastructure. The network has no base stations, access The re-keying process will be distributed between
points, remote servers etc. All network functions are cluster heads. And also using monitoring node the life time
performed by the nodes itself. Each node performs the of the network is extended. The monitoring node is
functionality of host & router. In mobile ad hoc networks, continuously checking the cluster head for its energy. If the
nodes within their wireless transmission range can cluster head has less energy next high capability node will
communicate with each other directly, while nodes outside be elected as cluster head.
the range have to rely on some other nodes to relay message. And also the details contained in the CH will be shifted to
MANET is autonomous, multihop networks interconnected new CH and this process will be informed to all other nodes
via wireless links. Thus a multi-hop scenario occurs, where by monitoring nodes.
the packets sent by the source node are relayed by several
intermediate nodes before reaching the destination. The 2. Previous Work
success of communication highly depends on the other
Majority of research on security of ad hoc networks
nodes’ cooperation. While mobile ad hoc networks can be
emphasize the secure routing protocols, there are some
quickly and inexpensively setup as needed, security is a
proposals on key generation and distribution issues.
critical issue compared to wire or other wireless
Zhou et al [7] proposed a technique to distribute
counterparts. Many passive and active security attacks could
certificate authority (CA) functionality. In this method, the
be launched from the outside by malicious hosts or from the
104 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
multicast group Within the network, if any two nodes A and B want to
key to all nodes communicate first it will authenticate each other. The
unicast authentication steps are as follows.
(grp key,
private key(d,n)) 1. Node A calculate hash value using its (id, public
key, group key) and transmit the hash value, id
First group key will be encrypted using RSA algorithm and and public key to node B
it is again encrypted using new node public key. 2. Node B receive hash value and also it calculate new
hash value from A’s id, public key and group key.
CH à new node: E(Pbnode , (E(e,[Ko]) || (d,n) ) ) where 3. Node B will check the received values and
Pbnode - public key of new node, calculated value both are equal or not.
{e, n } - Public key pair of RSA algorithm and 4. If the hashed values are equal, it identifies the peer
{d, n } - Private key pair of RSA algorithm. node as authenticated node.
Initially the new node gets the RSA private key pair by Algorithm for authentication:
decrypting received key information using its own private Procedure:
key, after then it decrypts using RSA private key to get BEGIN
group key. Precondition:
All other already existing nodes receive encrypted The nodes must wish to communicate with each other.
information and decrypt using RSA algorithm to get the if
updated group key. get the peer nodes public key and ID.
calculate the hash value and transmit to the peer one.
3.3 Existing Node Leaves A à B : hash (ida ,pka, group key) || ida ,pka
peer node : calcuate the same hash value.
Whenever an existing node B leaves the cluster the endif
following messages are exchanged. if both hash values are same
both are authenticated nodes
Node B leaves CH already endif
existing nodes END
to check for it's presence and wait for it's reply. Suppose if it ii) With respect to transmission delay:
does not receive reply, it will inform to other nodes to
initiate Cluster head selection process.
The monitoring node will wait for a period T and then it
will send hello message.
T= threshold time + propagation delay --- (4)
5. Conclusion:
Abstract: This paper shows a simple and effective approach to thousands of sensors monitor current naval vessel’s
the design and implementation of automated Industrial condition and performance [1]. The system can monitor
manufacturing Systems oriented to control the characteristics of different quantities like temperatures, speed, torque,
each individual product manufactured in a production line and pressure with the help of set of hundreds of sensor nodes at
also their manufacturing conditions. The particular products distributed locations to sense different physical quantities
considered in this work are steel sheets that under go the and send the data to a centralized place to monitor the
galvanizing processes. However, the approach can be modified
processes going on. Currently distributed control systems
to be applicable in other industries, like paper, textile,
aluminum, etc. The system makes use of sensor nodes placed at
(DCS) are used to collect the data from various places and
different places to monitor and control various parameters of control the required processes with the help of these control
interest to guarantee and improve the quality of the products systems. The structure of a typical DCS is shown in fig 1.
manufactured in many industries. The proposed system uses Instead of using distributed control system to carry the data
CAN bus to send and receive the data from the nodes to the we use CAN (Controller Area Networks) bus so that the
central controller which provides many advantages over the speed of communication increases and reduce the number of
currently used distributed control system. wire harnesses caused otherwise. Different protocols and
serial communication standards are used for this
Keywords: Galvanizing processes, CAN bus, multi-master, communication.
Electro magnetic interference (EMI), distributed control systems
(DCS).
1. Introduction
through a binary model of "dominant" bits and "recessive" contains alkali solution. The container has a conductor rod
bits where dominant is a logical 0 and recessive is a logical which acts as an anode. The pure zinc block is hanged to
1. If one node transmits a dominant bit and another node this rod. A constant current is supplied through a rectifier.
transmits a recessive bit then the dominant bit "wins". A A chemical reaction between the alkali solution and zinc
typical CAN node is shown in fig. 2. takes place which releases the positive ions of zinc. The
material to be galvanized is dipped into the alkali solution
in the tub. The tub itself acts as ground making the material
cathode and the positive ions of zinc gets deposited on the
material, carrying out the galvanization process. The time
for which this process should continue depends upon the
shape and size of the material. If the material has some
inner curved surfaces or with small opening then it takes
more time to complete the process. During this chemical
reaction the pH value of the alkali solution decreases. If this
value goes below a certain set value then the system should
give an alarm indicating that the solution is to be changed.
This can be done by using a pH value sensor and
periodically sending output to the controller.
The material is moved from one container to other
container using a conveyer. With the help of a flexible hook
the material can be connected to the conveyer. A typical
schematic is shown in the following Fig. 3.
Figure 2. CAN Node
2. The Galvanizing Process 1: Degreasing tub, 2: Washing tub, 3: Derusting tub, 4: washing
The process of galvanization is used to save the metal sheets tub, 5: galvanizing tub with zinc block connected to anode, 6, 7:
from oxidation and increase the life of the product Heaters, 8: Metal Sheet, 9: Conveyer, 10: Hook
manufactured from this sheet. Figure 3. Process of Galvanization.
• Flexibility in increasing number of nodes without it with the preset value. If the temperature is less than this
any problem, set point then the heater is switched on with the help of a
• Make system more reliable, as wire harnesses is relay, otherwise it is switched off. This relay operates with
reduced he interrupt signal at pin PD3/INT1. The information
generated i.e. current temperature value and status of relay,
3. The System Architecture by the microcontroller is sent to the centralized controller
through the CAN port for further processing. The CAN
Various nodes with multiple sensors can be deployed in controller used is 8 pin chip MCP2551. It is connected to
various sections to sense the quantity of interest and send it microcontroller at pins PD0/RXD and PD1/TXD. The
over the CAN bus to the base station as given in the block circuit diagram of the node structure or slave unit is shown
diagram in fig 5. Some processing is done on this data so in fig. 6.
that the size of the data to be further communicated to the
server can be reduced and bandwidth utilization will be
done efficiently. At the server end the data gathered will be
analyzed for proper monitoring purpose and taking SLAVE UNIT
appropriate actions during the manufacturing process. The
structure of the Centralized monitoring server is given in
CAN BUS
TEMP SENSOR J2
5
9
U2 4
R1 U1 5 8
U3 Y
9 22 3 3
RESET PC0/SCL D
1
23 6 7
47k PC1/SDA Z
13 24 2
XTAL1 PC2/TCK
+V S
12 25 7 6
XTAL2 PC3/TMS B
26 2 1
PC4/TDO R
2 40 27 8
VOUT PA0/ADC0 PC5/TDI A
39 28 CONN-D9M
PA1/ADC1 PC6/TOSC1
38 29 MCP2551
PA2/ADC2 PC7/TOSC2
GN D
37
PA3/ADC3
36 14
PA4/ADC4 PD0/RXD
35 15
PA5/ADC5 PD1/TXD
3
LM35 34 16
PA6/ADC6 PD2/INT0
33 17
PA7/ADC7 PD3/INT1
18
PD4/OC1B
1 19
PB0/XCK/T0 PD5/OC1A
2 20
PB1/T1 PD6/ICP
3 21
PB2/INT2/AIN0 PD7/OC2
4
PB3/OC0/AIN1
5
PB4/SS
6
PB5/MOSI
7 30
PB6/MISO AVCC
following fig 4.
8
PB7/SCK
ATMEGA16
AREF
32
OUTPUT RELAY
12V RL1
POWER SUPPLY OMI-SH-124L
U4
Figure 4. Centralized Monitoring Server 7805
J3 D1 12V 1 3 VCC
VI VO
1
GND
2 R2 Q1
3 1N4007
D2 C3 BC547
C1 C2 100n 47k
2
1N4007
GND
MASTER UNIT
LCD1
LM016L
VDD
V SS
V EE CAN BUS
RW
RS
D0
D1
D2
D3
D4
D5
D6
D7
J2
E
5
1
2
3
4
5
6
7
8
9
10
11
12
13
14
9
U2 4
R1 U1 5 8
Y
9 22 3 3
RESET PC0/SCL D
23 6 7
47k PC1/SDA Z
13 24 2
XTAL1 PC2/TCK
R2 12
XTAL2 PC3/TMS
25 7
B
6
47k 26 2 1
PC4/TDO R
40 27 8
PA0/ADC0 PC5/TDI A
39 28 CONN-D9M
PA1/ADC1 PC6/TOSC1
38 29 mcp2551
PA2/ADC2 PC7/TOSC2
37
PA3/ADC3
36 14
PA4/ADC4 PD0/RXD
35
34
33
PA5/ADC5
PA6/ADC6
PD1/TXD
PD2/INT0
15
16
17
SERIAL COMMUNICATION
PA7/ADC7 PD3/INT1
18
PD4/OC1B
1 19
PB0/XCK/T0 PD5/OC1A VCC
2 20
PB1/T1 PD6/ICP
3 21
PB2/INT2/AIN0 PD7/OC2
4
PB3/OC0/AIN1 C8
5
PB4/SS
6
PB5/MOSI
7
PB6/MISO AVCC
30 C5
8 32 1u 1u
PB7/SCK AREF
1 3
ATMEGA16 J6
1
C1+ C1-
6
POWER SUPPLY 11
12
10
9
T1IN
R1OUT
T2IN U4
T1OUT
R1IN
T2OUT
14
13
7
8
2
7
3
8
U4 R2OUT R2IN
4
7805 MAX232 2 9
J3 D1 VS+
12V 1 3 VCC 6 5
VI VO VS-
1
G ND
2
C2+ C2- C3 CONN-D9F
3 1N4007
D2 C3
C1 C2 100n
C4 100n
2
Authors Profile
Figure 9. Flowchart for Controller node operation Rashmi Jain completed her B. E. (Computer
Technology) from Kavi Kulguru Institute of
Technology and Science, Nagpur University in
7. Conclusion 1997. Currently completing M.E. in Embedded
systems and Computing from G. H. Raisoni
The system developed with the help of sensor nodes and College of Engineering, Nagpur University.
CAN Bus improves the performance of the system and Meanwhile worked as a lecturer in Comp.
reliability of the system at lower costs as compared to the Tech. Dept. at KITS Ramtek, SSGMCE
presently applied system of distributed control systems using Shegaon and YCCE Nagpur.ET VOLUME 31
RS-232 based communication. Multimaster communication JULY 2008 ISSN 1307-6884 513 © 2008 WASET.ORG
allows any node to initiate communication when bus is free. Dr. Narendra G Bawane is Professor and
The intelligent nodes make proper utilization of the bus. Former Head of Computer Science and
Automobile industries and other industries are making use Engineering department at G. H. Raisoni
of it for better noise immunity and EMI. Along with the college of Engineering, Nagpur
(Maharashtra). Prior to this assignment, he
application in vehicle networks CAN bus is efficiently used
worked with B.D. College of Engineering
in industries also. In the Galvanization process also the Sewagram and Govt. Polytechnic, Nagpur
quality of the finished product will increase if all the phases for several years. He has total teaching
are completed in due course of stipulated time and at experience of more than 22 years. He has
controlled temperature and pH value of the solution. completed his B.E. from Nagpur University in 1987 and M. Tech.
in 1992 from IIT, New Delhi. He completed his Ph. D. in 2006 at
References VNIT, Nagpur.
properly manages the network with minimum probability of Equation 1. states that the path with banwidth left on them
error occurrence. Though the consumption of resources is a denoted by bwe(i,j) which is greater than bwmin are selected
problem, yet a bigger problem is that even after giving so and the one with min. value is decided to be assigned.
much of network capacity, protocols still lack in providing Equation 2. states that for delay the summation of all the
QoS strictly. Getting in a bit depth there is a strong delays on all the links is taken and the path where it is
requirement to make the updation of the topological found to be less than D is selected. Similarly other criterion
information to a defined level. The reason being frequent can be calculated.
updation may consume considerable amount of network
2.2 Clusters and Cluster Head
resources as well as delayed updation may cause delay in
detection of topology changes. Hence, this calls for the Among several approaches there are techniques which have
development, which finds a way solving both of these considered the network as a whole and others have used
problems. clustering to manage the network effectively. This clustering
is the technique which made it possible to deploy hybrid
1.5 Paper Organization approach of protocols in MANET’s. Since, the formation of
The rest of the paper is organized in such a way that the clusters is to be done carefully, previous works have done
second section describes previous works in the field of clustering on three main basis viz. fixed geographical
quality of Service management, clustering and formation of locations [3], based on numerousity of nodes in a cluster and
cluster heads. This section also describes the previously depending upon hop counts [4]. The GPS system provides
proposed architectures in this field along with a short the information about exact co-ordinate of the node in the
description of co-operative routing. The third section consist geographical location through which we can form fix size
of our proposed work with thorough description of how and fixed boundary clusters with a little dynamics. However,
clustering is to be done and how hierarchical modeling can in the second approach we formed the clusters depending
help in solving the problems. This provides with a upon the count of the nodes permissible in a cluster. The
comprehensive architecture for QoS management third approach uses the proximity of neighborhood of nodes
mechanism with detailed description. The fourth section in the network and other subjective values. Also, the
consists of the future work that can be done in this field. determination of cluster head should be done in a way
Finally, the paper is concluded in section five. keeping in mind the proper division of work on the nodes so
that no node would get overloaded with work. This
2. Related Work determination cannot be done with exact calculation and
without error but yet we have much data to calculate that
2.1 Quality of Service which node is free most of the time and hence which can
The QoS parameters are those properties of the network serve as cluster head. Although in current scenario we have
which plays important role in the delivery of required such advance devices which have extremely high processing
information over the network. The requirement of QoS is capability and hence the selection of cluster head may not be
that neither all applications require same network resources very strict but yet we try to keep cluster heads away from
nor from commercial point of view all users pay same for relaying the data.
services. So there is a need to make such distinctions. Now
2.3 Previously Defined Architecture
as many parameters are considered in the QoS related to the
network the transmission will become more effective. There Several architectures have been defined using hybrid of
are some parameters [1], identified to be considered proactive as well as reactive approach and tried to solve the
essential for the transmission viz. bandwidth, delay, jitter problem of MANETS’s. One of the most popular has used
and reliable delivery of data. There is also requirement to the same hybrid approach [1] and defined how the work is
pay attention to the way they are managed, about where and done in reactive way for path determination and reservation.
how. Some architecture effectively managed these in the
form of matrices and calculations are done with defined
formulas, [2]. The QoS matrix which is used to store the
values of parameters of links between every two nodes is
having the node id on both rows and columns. The values
filled in the matrix are the values for every link in the
network and looks like as given below:
of load on network again on the cost of extra processing • The agent reaches back to the source. In this way a
on the side of nodes. quality of service path is reserved for the source.
• This mobile agent moves inside the cluster and collects The last packet of transmission moves on with setting
all such information as well as updates all such resources free. For this we need a special field value of
information on the node. This causes the maintenance which causes continuation of reservation or setting resources
of topology of the cluster on every node. The cluster free.
head calculates some figures about the paths and the
3.3.4 QoS Path Determination in detail
network capacity left with its cluster. This complete
information with the agent is updated when it reaches This process is done by reaching every cluster head and
the cluster head. It is later on communicated to the collecting the information for various parameters. The
higher level cluster head by another agent and the formulas in equation 1 and 2 are as it is followed in our
process of updation goes on to maintain the complete work and give a clear picture about how the calculations are
topological information. done to determine the paths fulfilling the criteria. However
This updation in upper level cluster is done in following from the side of cluster head the responsibilities are to
way: provide access to the data on the node about its cluster to the
• The information of cluster available at lower level is mobile agent. The rest is on the agent. Whenever the
calculated to generate some information for upper layer destination node is found in the cluster the head is
cluster head. responsible to tell the lowest level QoS path. In case of
• Every such lower level cluster head is travelled through failure of path determination an alternative path is chosen,
interfaces by agent and agent for this level takes which in our work, we try to prepare for, if it happens to
information from cluster heads only. occur. For this we maintain several such paths and choose
• This information is updated for upper level cluster head. alternative on the spot. However, choosing such alternative
An important point to be noted is that the generated path may constrain our future selections. If all such paths
information is not the info about each link but an abstract are not found then we move forward for second execution of
knowledge of the bandwidth available through that cluster, path determination. Another point which should be
delay through that cluster and similar information about considered is there are some QoS parameters which
other parameters. Also if some change in the node themselves depend upon other parameters such as delay or
membership is encountered. In this way the info is updated congestion, jitter and reliability on mobility of nodes etc.
at the highest cluster head. The guarantee for control of jitter is given by cashing of
packets i.e. in case a data packet is delayed due to excessive
3.3.3 Path Determining and Resource Reserving traffic load or caused due to new path selection then the
Mobile Agent packet already cashed in the memory is used to make up the
The process of path determination and reservation is done delivery gap. Another thing to be considered is the handoff
by another mobile agent who works in the following way: process. If handoff occurs frequently, it has adverse effect on
• Whenever any node needs to transmit the data, it every aspect of QoS. Hence several factors which could
generates the mobile agent and directs it to the cluster cause potential problems during handoff are also managed
head. under QoS. This management of QoS during handoff is
• At cluster head if the destination node is in the current considered differently. During handoff process, the data
cluster then the head tells a QoS path to the agent. available at the cluster head is to be transmitted to the new
• If the destination is not in the current cluster, the agent is cluster head. This transmission of data is always inter-
told the way for upper layer cluster head. cluster in cases when the handoff is due to head moving out
• The process is repeated till the cluster of some level tells of the cluster, since the indication of need of handoff is
that the node is available in its cluster. From here the received only at the motion of head out of the cluster. Also
agent takes the path from source to destination the handoff may occur if the load on the node is so increased
following the lowest level clusters. that it can no more bear the responsibility of cluster head in
• This movement of the mobile agent till it reaches such a that case the handoff is intra-cluster. To manage this
cluster head which has both the source and destination handoff process we divide the information available at the
in its region causes the agent to see from the highest cluster head to be divided into two parts:
level the topology of the required region. 1) Readily available
• The agent then reaches the destination’s lowest cluster 2) Calculated information
head following the told path by the cluster head. This readily available information is the one which can be
At this stage the path to the destination cluster is known. All gathered inside the cluster as and when it is required. The
the intermediate cluster heads of lowest level are then calculated info is very concentrated and calculated from a
travelled in reverse order to reserve the path resources for long time in the past. Now since the handoff process
the source in the following way: requires some transfer of information about the cluster if the
• The agent itself returns from the lowest level cluster cluster head is leaving the cluster this transferred info would
heads, intermediate in the path. be only the calculated info. The will be gathered by a
• At every cluster head it reaches intermediately it reserves request.
the required amount of resources which should strictly
follow the contract.
118 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
Authors Profile
Abstract: This paper analyses in detail various aspects of they handle and the subsequent I/Os. This classification
currently available Relational Database Management Systems leads to the following two types of database systems:
(RDBMS) and investigates why the performance of RDBMS (i) Online Transaction Processing (OLTP) Systems.
suffers even on the most powerful hardware when they support
OLAP, Data Mining and Decision Support Systems. A large (ii) Decision Support Transaction Processing (DSS)
number of data models and algorithms have been presented Systems.
during the past decade to improve RDBMS performance under
heavy workloads. We argue if the problems presented here will 1.1 OLTP Transactions
be taken care of in the future architectures of relational
An OLTP transaction is a unit of work that is usually
databases then RDBMS would be at least the moderate platform
expected to run in a very short duration of time because it
for OLAP, Data mining and Decision Support Systems.
deals with the database in real time or online mode. In other
Keywords: RDBMS, OLAP, Data Mining, Decision Support words these transactions constantly update the database
Systems based on the most current information available so, the next
1. Introduction user can rely on that information being the most current. An
Commercial server applications such as database services, example of this kind of system will be a Library Information
file services, media and email services are the dominant System. In this case, all the information pertaining to the
applications run on the server machines. Database system is kept in tables spread across a disk system, and the
applications occupied around 57% of the server volume and database is online. Any user in the community will have
the share is increasing [8]. Corporate profits and access to that information.
organizational efficiency are becoming increasingly 1.2 Decision Support Transactions:
dependent upon their database server systems, and those
systems are becoming so complex and diverse that they are A different type of application system that is currently in
difficult to maintain and control. With the advent of modern demand is the DSS (Decision Support System). This type of
Internet based tools for accessing the remote and local system is generally used to provide information to
databases, more emphasis is being placed on improving the management so decisions can be made about issues such as
performance of Computer Systems along with their business growth, levels of stock on hand etc. The challenge
functionality[7][12]. To monitor, predict and improve the is in deriving answers to business questions from the
performance of databases has always been a challenge for available data, so that decision makers at all levels can
database administrators. respond quickly to changes in the business climate. While a
standard transactional query might ask, "When did order
A database server performs only database
84305 ship?" a typical decision support might ask, "How do
functions. In terms of workload it performs only
sales in the Southwestern region for this quarter compare
transactions. When a SELECT/ UPDATE statement is
with sales a year ago? What can we predict for sales next
executed a database server interprets it as a series of reads/
quarter? What factors can we alter to improve the sales
writes. Considering that atomic level of anything is smallest
forecast?"
part i.e. it could be said that the atomic level of a transaction
consists of the reads or writes it generates. If broken down to In the OLTP system the rate of the throughput of
this level a database server processes I/Os. So, the database transaction is commonly measured in transactions per
systems can be classified based on the type of transactions second (TPS) or transactions per minute (TPS). With DSS,
throughput is usually measured in queries per hour (QPH).
(IJCNS) International Journal of Computer and Network Security, 121
Vol. 2, No. 5, May 2010
This unit itself indicates that these queries are of extreme Relatively
size and overwhelm the machine’s resource until they are standardized and
Often complex queries
complete. In almost all cases, the ratio between OLTP and Queries simple queries,
involving aggregations.
DSS transactions equals thousands (sometimes tens of Returning relatively
thousands) of OLTP transactions to one DSS transaction [6]. few records.
An OLTP system, even a large one, is usually not much Depends on the
more than 300 GB, whereas a large DSS system can be 2-5 amount of data
involved, batch data
TB in size. Examples of Decision Support Systems are On
Processing refreshers and complex
Line Analytical Processing (OLAP) and Data Mining. Typically very fast.
Speed queries may take many
1.2.1 On Line Analytical Processing: hours, Query speed
may be improved by
OLAP may be defined as “the interactive process of creating indexes
creating, managing, analyzing and reporting on data”. The Larger due to the
Can be relatively
first point is that analytical processing invariably requires Space existence of
small if historical
some kind of data aggregation, usually in many different Requirements aggregation structures
data is archived
ways (i.e. according to many different groupings. In fact, and history data
one of the fundamental problems of analytical processing is Typically de-
that the number of possible groupings becomes very large Database Highly normalized normalized with fewer
very quickly, and yet users need to consider all or most of Design with many tables tables for faster
processing
them. Relational languages do support such aggregations
but each individual query in such a language produces just Instead of regular
backups, some
one table as its result (and all rows in that table are of the Backup Religiously,
environments may
same form and have the same kind of interpretation). Thus, Backup and Operational data is
consider simply
to obtain n distinct groupings require n distinct queries and Recovery critical to run the
reloading the OLTP
produces n distinct result tables. In RDBMS the drawbacks business
data as a recovery
to approach is obvious. Formulating so many similar but method
distinct queries is a tedious job for the user. Executing those
1.2.2 Data Mining:
queries means passing over the same data over and over
again – is likely to be expensive in execution time. Thus the Data mining can be described as “exploratory data
challenges with OLAP are: analysis”. The aim is to look for interesting patterns in data,
patterns that can be used to set business strategy or to
Requesting several levels of aggregation in a single
identify unusual behavior (e.g. a sudden change in credit
query.
card activity could mean a card has been stolen). Data
Offering the implementation the opportunity to compute mining tools apply statistical techniques to large quantities
all those aggregations more efficiently (probably in a of stored data in order to look for such patterns. Data
single pass). mining databases are often very large. Data Mining is also
known as discovering knowledge form a very large database
Special SQL statements are provided to take up the above i.e. getting the interesting patterns from very large
challenges in RDBMS. Examples are the use of GROUPING databases. The process certainly involves thorough analysis
SETS, ROLLUP, and CUBE options available with the of data hence executing one Data Mining query means
GROUP BY clause. But, as far as efficiency is concerned, execution of several OLAP queries. Thus Data Mining may
almost all commercially available RDBMS exhibit poor also be referred as “repeated OLAP”. RDBMS don’t support
performance under OLAP workloads [1][5]. OLAP queries efficiently. Executing repeated OLAP on an
Following table [9] summarizes the major differences RDBMS means to bring the system on its knees.
between OLTP and OLAP systems.
2. Performance Issues of Relational Databases
Table 1: Differences between OLTP and OLAP
OLTP System OLAP System Most of the relational databases currently in use were
Operational data: Consolidation Data: developed during 1990s. Since then their architecture has
Source of OLTPs are the OLAP data comes rarely been revised. The design decisions made at that time
Data original Source of the from the various OLTP still influence the way RDBMS processes the transactions.
Data. Databases. For example designers at that time were not aware of the
To control and run To help with planning, exponential growth of Internet and nobody could expect that
Purpose of
fundamental business problem solving and one day databases would serve as the backbone of large
Data
tasks. decision support. Content Management and Decision Support Systems.
A snapshot of ongoing
Multi-dimensional The current RDBMS architecture was developed keeping in
What the
views of various kinds mind the OLTP, i.e. the architecture was optimized for the
Data Reveals business processes.
of business activities. Hardware, available at that time, to efficiently support
Short and fast inserts Periodic long-running OLTP. In the following sections we discuss some problems
Inserts and
and updates initiated batch jobs refresh the which affect the performance, efficiency and usage of
Updates
by end users. data. RDBMS for OLAP and Data Mining applications.
122 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
2.1 Difficult/ Inefficient to Formulate the OLAP Queries: table (rows and columns) is stored in a single database file
as a sequence of bytes, representing each row, with some
Consider the following MANUFACTURER_PART (MP)
free space between two consecutive bytes (figure 1). The
table:
space is left to accommodate future updates.
Table 2: MANUFACTURER_PART (MP)
Table 3: Storage Scheme for Table 2
M_ID Item_No QTY
M101 A11 1000
M101 A11 1000 M101 A12 500 ----
M101 A12 500
M102 A11 800 Now consider the following query:
M102 A12 900 SELECT * FROM MP WHERE M_ID = ‘M101’;
M103 A12 700
We observe that all data of one row is consecutive
M104 A12 200
and thus can be found in the same disk block, hence a
Now, consider the following OLAP queries: minimum of I/O activity (i.e. one block read/ write) is
1: Get the total item quantity. necessary to execute a typical OLTP query.
Now consider the following query:
2: Get total item quantities by manufacturer.
“Draw a histogram representing total quantities of all items
3: Get total item quantities by item. manufactured by different Manufacturers”
4: Get total item quantities by manufacturer and item. To answer the query the RDBMS needs to analyze all
Following are the psudoSQL formulations of these queries: rows of which only one column (QTY) is used. As a portion
of all rows is required, the RDBMS has to analyze the
1. SELECT SUM(QTY) FROM MP GROUP BY ( ); complete table to execute such queries. It implies that a
2. SELECT M_ID, SUM(QTY) FROM MP GROUP BY heavy load in terms of I/O is generated.
(M_ID); In the large RDBMS applications like data warehousing
3. SELECT Item_No, SUM (QTY) FROM MP GROUP and Data Mining tables may contain several columns and
BY (Item_No); billions of rows. When an OLAP query is executed on such
tables it scans each row of the table and generates a heavy
4. SELECT M_ID, Item_No, SUM (QTY) FROM MP load of I/Os. Only few columns from the table are used but
GROUP BY (M_ID,Item_No); the process takes significant amount of time. Clearly, most
Drawbacks to this approach are obvious: Formulating of the part of the table scan processes is a wasted effort. This
so many similar but distinct queries is tedious for end user. phenomenon is the main reason for the poor performance of
Also, their execution requires passing over the same data RDBMS products under heavy workloads.
over and over again which is likely to be quite expensive in Someone may argue that the problem of wasted efforts
execution time. done in the scanning of large database files may be solved
using some index structures like bit map, but OLAP and
Some special SQL options on the GROUP BY clause Data Mining queries have low selectivity. The real benefits
may ease life, but up to some extent, e.g. GROUP SETS of any index structure come out a in selection process
option allows the user to specify exactly which particular because an indexing technique provides a list of selected
groupings are to be performed. The following SQL record identifiers quickly. In OLAP and Data Mining
statement represents a combination of Queries 2 and 3: applications selection is not the end of the query. The
SELECT M_ID,Item_No,SUM(QTY) FROM MP RDBMS has to perform some action on the selected tuples.
So, indexing is not the solution to this problem.
GROUP BY GROUPING SETS (M_ID,Item_No);
2.3 Trends in Hardware Technology
Solution seems to be acceptable but, firstly it requires Figures 1 – 9 show the advancements in the Hardware
a lot of analytical thinking at end user’s part and secondly technology during the past decade. The size and bandwidth
options like GROUP SETS are not universally supported of memory, size and bandwidth of hard disk, Network
across all SQL implementations. Another problem is that Bandwidth and processing power of computer have
every SQL Query returns a relation as query result that increased tremendously. Exceptions to these trends are I/O
forces the user into a kind of row-at-a-time thinking. OLAP efficiency and memory latency. These factors are affecting
products often display query results not as SQL-style table the performance of all Software including the RDBMS. In
but as cross tabulations or graphs. the following sections we shall consider each of them
2.2 RDBMS Storage and Access Methods individually and shall analyze its effect on RDBMS
performance.
Currently available RDBMS implementations store data in
the form of rows and columns (table). This approach favors
OLTP queries. The storage scheme of these relational
systems is based on the Flattened Storage Model (FSM) or
Normalized Storage Model (NSM). Both of these models
store the data as a consecutive byte sequence. The complete
(IJCNS) International Journal of Computer and Network Security, 123
Vol. 2, No. 5, May 2010
4000
18000
3500
16000
3000
14000
2500
CP U Sp ee d (M h z )
12000
R o te tio n s P er M in u t e
2000
10000
1500
8000
1000
6000
500
0
4000
1995 1996 1997 1998 1999 2000 2001 2001 2002 2002 2003 2003 2004 2005
Years
2000
40
Figure 4. Disk RPM Increments
35
30
Disk Bandwidth (MB/s)
180500
20
160500
15
140500
10
120500
5
Disk Size (MB)
100500
0
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
80500
Years
60500
20500
500
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
Years
10
9
Memory Bandwidth Increments
800
8
700
7
Disk Latency (ns)
6 600
5
Memory Bandwidth
500
4
400
3
300
2
200
100
0
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
0
Years
1995 1996 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
Years
1700
1600
1500
1400
1300
1200
1100
1000
900
800 Figure 10. Saw Tooth Performance of RDBMS
700
600
500
400 Performance will typically degrade over time due to
300
200 database fragmentations. Systems that fragment the database
100
0 do not support sustained performance because they are
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
Years
unable to perform following two tasks:
Online reclamation of deleted space
Figure 8. Memory Size Increments Online reorganization of data.
Benchmarks typically do not detect “Saw - Toothed”
performance because they are not typically run for a
sufficient amount of time to test the anti-fragmentation
Network Bandwidth Growth
techniques implemented in a given system.
400
Another aspect that also contributes towards low I/O
efficiency is the product’s ability to explicitly cluster data on
350 the disk with other data that would typically be accessed at
300
the same time. One solution to this problem may be to add
additional database volumes as the amount of data stored in
250 it grows over time. For better I/O efficiency it should be
B a n d w i d t h (M b p s )
reference and that is not the case with OLAP and Data 2.3.5 CPU Utilization:
Mining queries. Most of the researches relating database performance to
2.3.2 Network Latency: processor have been carried out on multiprocessor platforms
using OLTP workloads [10] [13]. Few studies evaluate
database performance on
multiprocessor platforms using both types of loads (i.e.
OLTP and OLAP) [11]. All of the studies agree that the
Clearly we can store more (Figure 5) and process faster but 14000
the time the memory takes to access the data has stayed 12000
about the same. Memory latency is the most urgent reason 10000
Queue Length
applications that have this property are hash join and sorting 0 20 40 60
Utilization (% Values)
80 100 120
[2], which are common operations in OLAP and Data Figure 12. CPU Utilization under OLTP workloads.
Mining Queries.
126 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
0.6
Respo
Processors.” In Proceeding of the 8th International Zubair Khan received his Bachelor Degree in
Conference on Architectural Support for Programming Science and Master of Computer Application
Languages and Operating Systems, San Jose, Degree from MJP Rohilkhand University
California, October 1998. Bareilly, India in 1996 and 2001 respectively
and also Master in Technology (M.Tech)
[12] D.A. Menasce, V.A.F. Almeida “Challenges in Scaling Degree in Computer Science and Engineering
e-Business Sites”, In Proceedings of Computer from Uttar Pardesh Technical University in the
Measurement Group Conference, 2000. year 2008. He is currently pursuing his P.hd in
computer science and Information Technology from MJP
[13] L.Lo, L.A. Barroso, S.J. Eggers, K.Gharachorloo, H.M. Rohilkhand University Bareilly, UP India. He also worked as a
Levy and S.S. Parekh, “An Analysis of Database senior lecturer in JAZAN University Kingdom of Saudi Arbia. He
Workload Performance on Simulated Multithreaded is also servicing as Reader in the Department Of Computer
Processors.” In Proceedings of the 25th International Science and Engineering Invertis Institute of Technology Bareilly,
Symposium on Computer Architecture, pages 39 – 50, India. His areas of interest include data mining and warehousing,
Barcelona, Spain, June1998. parallel systems and computer communication networks. He is an
author/ co-author of more than 15 international and national
publications in journals and conference proceedings.
Authors Profile
1
Department of Electronic and Telecommunication Engineering (ETE),
University of Development Alternative (UODA), Bangladesh .
2,4
Department of Information and Communication Engineering
University of Rajshahi, Rajshahi-6205, Bangladesh.
3
School of Information Engineering, University of Padova, Italy.
Abstract: The efficient use of energy source in a sensor node is reducing energy consumption and prolong the life time of
most desirable criteria for prolong the life time of wireless the network.
sensor network. In this paper, we propose a two layer
hierarchical routing protocol called Cluster Based Hierarchical 2. The Proposed Approach (CBHRP)
Routing Protocol (CBHRP). We introduce a new concept called
head-set, consists of one active cluster head and some other
In this paper an optimum energy efficient cluster based
associate cluster heads within a cluster. The head-set members
are responsible for control and management of the network. hierarchical routing protocol for wireless sensor network is
Results show that this protocol reduces energy consumption proposed, which is a two layer protocol where a number of
quite significantly and prolongs the life time of sensor network cluster cover the whole region. Proposed protocol introduces
as compared to LEACH. a concept of head-set instead of a cluster head. At one time,
only one member of head-set is active and the remaining are
Keywords: WSN, Cluster, Routing Protocols in sleep mode. Several states of a node are found in this
protocol such as- candidate state, non-candidate state,
active state, associate state, and passive associate state.
1. Introduction
This protocol divides the network into a few real clusters
Wireless Sensor Networks (WSN) consists with huge tiny including an active cluster head and some associate cluster
sensors that are arranged in spatially distributed terrain. Due heads. For a given number of data collecting nodes, the
to a large number of applications such as security, head-set members are systematically adjusted to reduce the
agriculture, automation and monitoring, WSN has been energy consumption, which increases the network life.
identified as one of the pioneer technique in 21st century.
3. Architecture of CBHRP
The deployment and maintenance of the small sensor nodes
should be straightforward and scalable when they are In the proposed model, the number of clusters k and nodes n
useless. But the small sensor nodes are usually inaccessible are pre-determined for the wireless sensor network. Iteration
to the user, and thus replacement of the energy source is not consists of two stages: an election phase, and a data transfer
feasible. There are some technical constrains also in phase. At the beginning of election phase, a set of cluster
microsensor node such as low energy in battery and heads are chosen on random basis. These cluster heads send
processing speed etc. This limitation should be overcome as a short range advertisement broadcast message. The sensor
much as possible to use the node in wireless sensor network. nodes receive the advertisements and choose their cluster
The source of energy in sensor is a main constrain among heads based on the signal strength of the advertisement
them [1]. It can be minimized by increasing the density of messages. Each sensor node sends an acknowledgment
energy in conventional energy sources [2],[3]. One of the message to its cluster head. Moreover, in each iteration, the
elegant ways to prolong the lifetime of the network is to be cluster heads choose a set of associate heads based on the
reduced energy consumption in sensor nodes. This idea has signal strength of the acknowledgments. A head-set consists
been applied many research [4],[5] to increase the life time of a cluster head and the associates. The head-set member is
of the network. Cluster based Low-Energy Adaptive responsible to send messages to the base station. Each data
Clustering Hierarchy (LEACH) is an attractive routing transfer phase consists of several epochs. Each member of
protocol that disperses the energy load among the sensor head-set becomes a cluster head once during an epoch. A
nodes [6],[7],[8]. The improvements of LEACH have been round consists of several iterations. In one round, each
presented elsewhere [9],[10],[11]. Proposed protocol is sensor node becomes a member of head-set for one time. All
another approach to improve of LEACH protocol by the head-set members share the same time slot to transmit
their frames.
(IJCNS) International Journal of Computer and Network Security, 129
Vol. 2, No. 5, May 2010
1 LEACH
0.8
He ad-set size =2
Einit(J)
0.6
0.2
0
60
40 200
150
20 100
50
0 0
# Of Clusters Network Diameter (m)
References
[1] K. Pister, “On the limits and applications of MEMS
sensor networks”, Defense Science Study Group report,
Institute for Defense Analysis, Alexandria, VA.
[2] J. M. Rabaey, M. J. Ammer, J. L. da Silva, D. Patel, S.
Roundy, “PicoRadio supports ad hoc ultra-low power
wireless networking, IEEE Computer, July 2000, pp.
Figure 5. Time for iteration with respect to number of
42-48.
clusters and network diameter. [3] D. Timmermann, “Smart environments”, Invited Paper,
Ladenburger Diskurs, Ladenburg, January 2001.
Figure 5 shows a graph that illustrates the estimated time for
[4] L. Zhong, R. Shah, C. Guo, J. Rabaey, “An ultra low
one iteration with respect to the number of clusters and power and distributed access protocol for broadband
network diameter. The number of cluster does not affect on wireless sensor networks”, IEEE Broadband Wireless
iteration time for a particular network diameter. However, Summit, Las Vegas, May 2001.
the head-set size plays a significant role to increase the [5] V. Rodoplu, T. H. Meng, “Minimum energy mobile
iteration time. In contrast to LEACH, iteration time is wireless networks”, IEEE Jour. Selected Areas Comm.,
increased by several times for a given number of head-set. August 1999, pp. 1333-1344.
[6] W. Heinzelman, A. Sinha, A. Wang, A. Chandrakasan,
“Energyscalable algorithms and protocols for wireless
x 10
4 microsensor networks”, Proc. International Conference
5
Head set size=5 on Acoustics, Speech, and Signal Processing (ICASSP
'00), June 2000.
4 Head set size=3
[7] W. Heinzelman, A. Chandrakasan, and H.Balakrishnan,
# of Frames
3
“Energy-efficient communication protocol for wireless
2 microsensor Networks”, Proceedings of the 33rd
1 International Conference on System Sciences (HICSS
0 ’00), January 2000.
200
150 [8] S. Lindsey, C. Raghavendra, K. Sivalingam, “Data
180
LEACH 140
130 gathering in sensor networks using the energy*delay
160 120
110 metric”, Proc. of the IPDPS Workshop on Issues in
140 100
Distance Diameter Wireless Networks and Mobile Computing, April 2001.
[9] E. B. Heinzelman, A. P. Chandrakasan, and H.
Figure 6. Number of frames transmission per iteration Balakrishnan, “An Application-Specific Protocol
with respect to distance of BS and network diameter. Architecture for Wireless Microsensor Networks”, IEEE
Transactions On Woreless Communications, Vol. 1, No.
4, October 2002.
(IJCNS) International Journal of Computer and Network Security, 131
Vol. 2, No. 5, May 2010
Authors Profile
Abstract: Hearing impairment is the number one chronic subtraction is updated by averaging the noise speech
disability affecting people in the world. Many people have great spectrum using a time and frequency dependant smoothing
difficulty in Understanding speech with background noise. factor, which is adjusted based on signal presence
Speech Enhancement plays a vital role in such situations. probability in sub bands. Signal presence is determined by
Random noise will disturb the signal between speaker and the computing the ratio of the noisy speech power spectrum to
listener. Therefore the background noise has to be removed its local minimum, which is computed by averaging past
from the noisy speech signal to increase the signal intelligibility values of the noisy speech power spectra with a look-ahead
and to reduce the listener fatigue. The proposed approach is a
factor. This local minimum estimation algorithm adapts
speech enhancement method based on the preprocessed sub
band spectral subtraction method, and the preprocessing is done
very quickly to highly non stationary noise environments.
by using partial differential equation. This method provides a By using this noise estimation algorithm in this method
greater degree of flexibility and control on the noise subtraction outperforms the standard power spectral subtraction method
levels that reduces artifacts in the enhanced speech, resulting in resulting in superior speech quality and largely reduced
improved speech quality and intelligibility. musical noise in single channel system for both stationary
and non stationary noise environments.
Key words- Sub band Spectral Subtraction, Partial differential
equation, Adaptive noise estimation, Signal to Noise ratio, IS
distance. 2. Partial Differential Equation Technique
1. Introduction Input noisy speech signal affected by additive background
noise can be enhanced by this method. First step in speech
Understanding speech with background noise is difficult
enhancement using PDE is to obtain the gradient (g) of each
even for normal hearing people, for the people with hearing
sample in noisy speech signal using the samples before and
disability it is one of the major problem which is to be
after the current sample.
reduced. This is especially true for a large number of elderly
people and the sensorineural impaired persons. Several g f = S ( x − ∆x, t ) − S ( x, t )
--------- (2.1)
investigations on speech intelligibility have demonstrated g b = S ( x + ∆x, t ) − S ( x, t )
that subjects with sensorineural loss may need a 5-15 dB
Where, S(x,t) is the noisy speech signal, Δx is the
higher signal-to-noise ratio than the normal hearing
sampling rate. After the gradient is calculated the
subjects. Recent statistics of the hearing impaired patients
influencing coefficients in each directions of the current
applying for a hearing aid reveal that 20% of the cases are
sample are computed.
due to conductive losses, more than 50% are due to
1
sensorineural losses, and the rest 30% of the cases are of IC f = 2
mixed origin. Presenting speech to the hearing impaired in g
an intelligible form remains a major challenge in hearing- 1 + f
aid research today. In single channel system, speech k ---------- (2.2)
enhancement is a challenging on because reference noise 1
IC b =
signal will not be available for enhancement. The clean 2
g
speech cannot be processed prior to being affected by the 1+ b
noise. This is one of the most difficult situations in speech k
enhancement for a single channel system. The conventional Where, ICf is the Forward influencing coefficient and ICb
power spectral subtraction method for single channel speech is the backward influencing coefficient. In the equation (2)
enhancement substantially reduces the noise levels in the ‘k’ is constant value between 1 and 100. From the above
noisy speech but this will introduce musical noise. calculated influencing coefficients and gradients the speech
The proposed method is a preprocessed sub band signal is enhanced using
spectral subtraction method of speech enhancement and the
preprocessing technique is using partial differential S ( x, t + ∆t ) = S ( x, t ) + ∆t ( g f IC f + g b ICb ) ------- (2.3)
equation. In this method first the input noisy speech signal In the above equations S(x,t) is the input noisy speech
is enhanced using PDE by taking the adjacent samples and signal, Δt is a coefficient between 0.1 t0 0.6 representing the
calculating the gradient, influencing coefficients and then step of noise reduction in each iteration. The output of the
the output of this process is applied to the input for sub band signal is again processed by applying into the algorithm of
spectral subtraction method. The noise estimate in spectral
(IJCNS) International Journal of Computer and Network Security, 133
Vol. 2, No. 5, May 2010
Sub band spectral subtraction method to reduce the non contain only background noise. The noise estimate can be
stationary noise. updated by tracking those noise only frames. To identify
those frames, a simple procedure is used which calculates
the ratio of noisy speech power spectrum to the noise power
3. Sub band Spectral Subtraction method spectrum at 3 different frequency bands in each frame and
Sub band Spectral subtraction method is the the sampling frequency respectively. If all the three ratios
frequency dependent processing of the Spectral Subtraction are smaller than the threshold that frame is concluded as a
procedure. This method offers better quality [9] of the noise only frame, otherwise , if any one or all the ratios are
enhanced speech with reduced residual noise. This approach greater than threshold that frame is considered as speech
has been justified due to variation in signal to noise ratio present frame.
across the speech spectrum. White Gaussian noise has a flat The noise estimate is updated in speech absent frames
spectrum, where as the real world noise is not flat. The with a constant smoothing factor. In speech present frames
noise spectrum does not affect the speech signal uniformly the noise is updated by tracking the local minimum of noisy
over the whole spectrum; some frequencies are affected speech and the deciding speech presence in each frequency
more adversely than others. bin separately using the ration of noisy speech power to its
To take into account the fact that colored noise local minimum.
affects the speech spectrum differently at various 4.1.2 The minimum of noisy speech
frequencies, we use this approach to spectral subtraction. In our method for tracking the minimum of the
The speech spectrum is divided into non-overlapping noisy speech by continuously averaging past spectral values.
bands, and spectral subtraction is performed independently In this algorithm if the value of the noisy speech spectrum
in each band. The estimate of the clean speech spectrum in in the present frequency bin is greater than the minimum
the ith band is obtained by: value of previous frequency bin then the minimum value is
2 2 2 updated, else the previous value is maintained as it is.
Ŝi (k) = Y(k) − α δ D̂i (k) ; b ≤ k ≤ e
i i i i 4.1.3. Detection of Speech-presence frames
(3.1) The approach taken to determine speech presence
where bi and ei are the beginning and ending frequency in each frequency bin is similar to the method used in [4].
bins of the ith frequency band, a i is the over subtraction Let the ratio of noisy speech power spectrum and its local
factor and δi is the band subtraction factor. Over subtraction minimum be defined as
factor provides a degree of control over the noise subtraction
S (α , k) = P( α , k) P -------- (4.2)
min (α , k)
level in each band, the use of multiple frequency bands and r
the use of the δi weights provide an additional degree of
This ratio is compared with a frequency dependent
control within each band.
threshold, and if the ratio is found to be greater than the
threshold, it is taken as a speech-present frequency bin else
4. Noise estimation it is taken as a speech-absent frequency bin. This is based on
Noise estimation plays an important role in this the principle that the power spectrum of noisy speech will be
work of speech enhancement. For an efficient noise nearly equal to its local minimum when speech is absent.
estimation algorithm the resultant signal estimation will Hence the smaller the ratio is in (6), the higher the
have great accuracy. probability that it will be a noise-only region and vice versa.
4.1. Adaptive noise-estimation algorithm Note that in [4], a fixed threshold was used in place of
threshold.
Several noise-estimation algorithms have been proposed for From the above rule, the speech-presence
speech enhancement applications [2] [3] [4] [5] [6]. The probability, p(α,K), is updated using the following first-
main drawback of most of the noise estimation algorithms is order recursion:
that they are either slow in tracking sudden increases of p(α,k) = a p(α -1,k) + (1- a ) I (α,k) ------- (4.3)
noise power or that they are over estimating the noise where a is a smoothing constant. Note that the above
energy resulting in speech distortion. recursion implicitly exploits the correlation for speech
In the adaptive method the smoothed power presence in adjacent frames. This may result in slight
spectrum of noisy speech is computed using the following overestimate of the noise spectrum but will not likely have
first-order recursive equation: much effect on the enhanced speech.
2 --------(4.1) 4.1.4. Calculation of Smoothing Constants
P (α , k ) = gP(α − 1, k) + (1 − g) Y(α ,k)
Using the above speech-presence probability
estimate, we compute the time–frequency dependent
where P(α,k) is the smoothed power spectrum, α is the frame smoothing factor as follows [4].
index, k is the frequency index, |Y(α,K)| 2 is the short-time a(α,K) = d + ( 1- d) p(α,k) ------------ (4.4)
power spectrum of noisy speech and g is a smoothing where d is a constant. Note that a(α,K) takes values in the
constant. The proposed algorithm is summarized in the range of d ≤ a(α,K) ≤ 1.
following steps. 4.1.5 Updation of noise spectrum estimate
4.1.1 Speech present and speech absent frames. Finally, after computing the frequency- dependent
In any speech sentence there are pauses between smoothing factor a(α,k), the noise spectrum estimate is
words which do not contain any speech, those frames will updated as
134 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
0.4
Am
0.1
the spectral noise subtraction algorithm.
0
0 20 40 60 80 100 120 140
Time
(a − b ) R (a − b )
r
d (a , b ) = ----------(5.2)
Figure 3. PSD plot of Cellular noisy, clean and Enhanced
a r Ra
signal
where a is the vector for the prediction Inference for Fig3: Power spectrum density of the enhanced
coefficients of the clean speech signal, vector R is the signal is close to the power spectrum magnitude of clean
autocorrelation matrix of the clean speech signal and vector signal.
b is the prediction coefficients of the enhanced signal. 0.35
PSD C omparis o n of White Statio nary N o is y s ignal
Nois y Signal
Clean Signal
0.3
Enhanc ed Signal
0.2
PSD
C
OISY SIG
ENH
LEA
ANC E
NA L
D SIG
N SIGNA L
NAL
on the spectrum of speech. Signal to noise ratio will be more
than 6 to 10 dB in this method, and can be used in digital
0.3
0.25
0.2
hearing aids for sensorineural loss patients. The added
S
PD
0.15
computational complexity of the algorithm is minimal and it
0.1
0
0 20 40 60 80 100 120 140
algorithm can be implemented in real-time on a fixed point
Digital Signal Processor (DSP) platform for evaluation in
Frequency
Figure 5. PSD plots of Colored factory Noisy, clean and real-world conditions. Methods can be developed to preserve
Enhanced signal the transitional regions and unvoiced regions, which contain
Inference for Fig 5: Initial Noise segments are very much low speech levels.
reduced in this method for colored factory noisy signal an
dthe spectrum is close to the clean signal. References
SNR O btaine d for different num be r of bands
40
environments: improved minima controlled recursive
30 averaging. IEEE Transactions on Speech Audio
SNR
20
10
Processing. 11 (5), 466–475.
0
[4] Martin, R., 2001. Noise power spectral density
t)
l
l
l
l
l
a
a
a
a
l
a
na
a
al
e
n
n
n
n
kv
n
ig
ig
ig
g
g
ig
ig
ig
si
si
o
S
s
S
(y
y
-10
sy
y
)
y
y
is
le
is
is
is
e1
is
oi
no
a
o
no
o
n
n
m
n
n
pl
n
db
b
b
am
b
b
d
b
d
6d
(f
0d
3d
0
d
0
s
ry
ry
is
ry
ry
ng
ng
d
r
a
la
e
n
on
Single band
ti
ti
rd
llu
o
rs
rs
tio
r
ti
co
e
bu
u
at
ti
a
C
a
4 bands
a
b
st
st
e
st
st
ite
ite
R
ite
ite
SS method
h
in
in
h
Typa of Noi se
DEKF Algorihm
Figure 7. SNR comparison with other methods [5] Rangachari, S., Loizou, P., Hu, Y., 2004. A noise
estimation algorithm with rapid adaptation for highly
Table 1: Comparison of IS distance measure for different nonstationary environments. Proceedings IEEE
noisy inputs with conventional approach International Conference on Acoustics, Speech Signal
IS Distance for IS distance for single Processing. I, 305–308.
Signal proposed method band SS method [6] Sohn, J., Kim, N., 1999. Statistical model-based voice
Cellular Noise 1.53 3.04
activity detection. IEEE Signal Processing. Lett. 6 (1), 1–
White burst noise 1.256 2.1722
3.
[7] S. Quackenbush, T. Barnwell, and M. Clements,
White stat noise 0.9978 2.067 “Objective Measures for Speech Quality Testing,”
Pink stat noise 0.791 1.0246 Prentice-Hall, 1988.
[8] J. Li and M. Akagi, “ A noise Reduction system
Car noise 0.1569 1.989
based on hybrid Noise estimation technique and post
Recorded speech 1.54 0 Filtering in arbitrary noise
environments”,Speech Communication, vol.48 pp 111-
Recorded speech 1.2564 0
126, 2006.
We obtained the SNR improvement when the number of
[9] D. Deepa and A. Shanmugam “Time And Frequency
bands is increased is shown in the Fig 6 for various noisy
Domain Analysis Of Subband Spectral Subtraction
signals, Comparison of SNR with other methods is give in
Method Of Speech Enhancement Using Adaptive Noise
Fig 7 and IS distance for the proposed method compared
Estimation Algorithm ” International Journal of Engg.
with conventional SS method in Table 1.
Research & Indu. Appls. (IJERIA)., ISSN 0974-1518,
Vol.2, No. VII (2009), pp 57-72
7. Conclusion [10] D. Deepa and A. Shanmugam “Spectral Subtraction
Method Of Speech Enhancement Using Adaptive Noise
The frequency dependent preprocessed sub band Estimation Algorithm with PDE method as a Pre
spectral subtraction method provides a definite improvement processing technique” ICTACT Journal on
over the conventional power spectral subtraction method Communication Technology (IJCT). ISSN 0976-0091,
and does not suffer from musical noise. The improvement Vol.1, Issue 1, March 2010, pp 1-6
can be attributed to the fact that the sub-band approach takes
136 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
Id
VDD
M19 MF11
MF1 MF2
M5
preferred as compared to voltage-mode multipliers [9]. So, Also M11 and M12 transistors in the conventional OTA are
current multipliers are best suited as a programming replaced with two diode connected transistors M13 and M14
element in systems such as filters, neural networks, mixers as in Fig.6, to linearise the output current and to reduce the
and modulators. The linearity multiplier tries to linearise the noise effect in the transconductor.
VDD
output from the V-I Converter and provides a wide tuning
VCMF
range. The conventional current multiplier [1] is as shown M9 M10
M15 M16
in Fig. 5.
VDD
M7 M8
I2
I2 M13 M14
M7 M8 M5 M6
Vb1
Vop Von
I1+ii I1- ii
IO IO
M11 M12
I1 + ii I1 - ii
Voltage to current
i0 i0
converter Itune
M9 M10
vss
Voltage-to-Current
Converter
2I2 Figure 6. Linearity current multiplier
Figure 5. Conventional Current Multiplier 2.3. Common Mode Feed Back Circuit
The transistors in the multiplier work in linear region to The switched capacitor circuits have become extremely
enhance the linearity performance of the OTA. The current popular because they accurate frequency response, good
equation for a MOSFET operating in the linear region is linearity and dynamic range. In switched capacitor circuits
given by equation (2) the accurate discrete time frequency responses are obtained
due to filter coefficients and are determined by capacitance
ratios. This capacitance ratios can be set quite precisely (in
ordre od 0.1%) in integrated circuits. This accuracy is much
better than that which occurs for integrated RC time
Where W and L are the width and the length of the constants (on order of 20%). The basic building blocks in
transistor, respectively, is the reverse saturation current, switched capacitor circuits are
n is the sub threshold slope factor, and is the thermal • Capacitors.
voltage. From the V-I conversion circuit [1], it is found that • Switches.
(3) • Non over lapping clocks.
Therefore
= The common mode fedd back (CMFB) circuit using
= Switched capacitor technique [10] is shown in Fig. 7. The
switched capacitor CMFB circuits are area efficiant and
does not need an additional amplifier.
In equation (4) n is the sub threshold slope factor and is V ref
Itune
= ) CS CS
Ca Ca
VCMF
= MF3 MF4 MF7 MF8 MF9
VSS
to control the output common mode voltage. Vop and modevoltage is not equal to the reference voltage, a
Von from the transconductor circuit is given as inputs to corrected current would be mirrored by transistor MF9 to the
CMFB circuit. Suppose if the common mode level of load transconductor, and then the output common mode
the outputs Vop and Von increases, then the Ca capacitor voltege is adjusted to the desired voltage. Itune is used to
average voltage increases to control the output common provide wide tuning trange to the filter.
mode level. P1 and P2 are two non overlapping signals
applied to the MOS transistors in the CMFB circuit
which act as switches. By varying the Itune, the tuning 3. Filter Architecture
range of the OTA is varied to accommodate wide
frequency range in the low pass filter. The proposed The second-order Butterworth low-pass Gm-C filter used
OTA circuit is shown in Fig. 8 and its circuit by Uwe Stehr et al. [11] is shown in Fig. 9. The low pass
parameters are listed in Table 1. filter structure consists of four operational transconductance
amplifiers gm1, gm2, gm3 and gm4.
VDD
VCMF 2*C 1
2*C 2
M9 M10
Vref
M15 M16 out-
M7 M8
p1 p2 p2 p1
vop von
in-
+ + + +
gm1 gm2 gm3 gm4
M6 M13 MF1 MF2 MF5 MF6 + + +
M5 M14 I in+
CS Ca Ca CS tun
Vb1 e
p2 p2 p1
p1
VCMF
vop von out+
MF3 MF4 MF7 MF8
2*C1 2*C2
Ib
MF9
M11 M12
Vcm Figure. 9 Gm- C realization of 2nd order low-pass Filter
Vi+
M4 Vi-
M1 M2
M17 The transfer function of the above filter structure [11] is
VB given as equation (8)
H(s)= (8)
M3
So, the transfer function of the 2nd order Butterworth filter is
VSS
H(s)= (9)
Figure 8. Complete OTA circuit
Recognizing the common transfer function of a 2nd order
Table 1 Circuit parameters of the Proposed OTA low-pass filter as
H(s) = (10)
Components Value
(W/L)1,2 250µm/0.35µm
And comparing (8) and (10)
(W/L)3,17 200µm/0.35µm
(11)
(W/L)4 1µm/0.35µm
(W/L)5,6,7,8 40µm/0.35µm
Maintaining and
(W/L)9,10 50µm/0.35µm
leads to (12)
(W/L)11,12 20µm/0.35µm
Ib 2µA
Itune 10µA Now the transconductance of the 2nd order filter from
(8) and (9) is
(14)
The ouput’s of the transconductor Vop and Von are
compared with reference voltage Vref. If the output common The transconductance gm1, gm3 and gm4 of the filter are kept
equal to gm and transconductance gm2 is kept 1.848 times of
140 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
4. Simulation Results
The step response of the OTA measures the slew rate of the
filter to be 130 V/µs and is shown in Fig. 11.
4.2. 2nd order Butterworth low pass filter
The implemented circuit of 2nd order Butterworth low
pass filter and its unity gain frequency response are shown
in Fig. 13 (a) and (b) respectively. The maximum cut-off
frequency of the filter is 3.9 MHz.
(IJCNS) International Journal of Computer and Network Security, 141
Vol. 2, No. 5, May 2010
I1 (type: graphic)
developed based on clique finding in an association graph to I
characteristics of cliques in an unlabeled, undirected graph Figure 2.d illustrates the maximal clique to a given
of the appropriate size. In literature, we find several association graph. If AG contains more than one clique, the
algorithms for clique finding like Messmer method [11] and maximal one is the one having maximal weights.
Belaid algorithm [4]. We have implemented for this goal an
optimal algorithm based on Messmer method.
I1 J1 (I2 ,J 1 )
Gr Gr (I1 ,J 1 ) (I2 ,J 1 )
R I2
Gr
R I
0 .2 1 1
I
T xt T xt
J2
(I3 ,J 2 )
I3 ( I3 ,J 2 )
(a ) G ra p h G 1 (b ) G r a p h G 2 (c ) A s s o c ia tio n G r a p h (d ) M a x im a l C liq u e
Figure 2. Illustration of the graph matching: (a) Graph G1, (b) Graph G2, (c) Association Graph, (d) Maximal clique
Table 2: A tabulation of the top four documents matches for query document
I2 I3
I2 I2 I1
I3 I3 I2 I3
I2 I3 I1
I1
I1
Abstract: Bluetooth device discovery process involves nodes faster links, and consume less power. In [7], the authors
seeking to join neighboring piconets to enter inquiry substate. evaluated the factors affecting discovery time for multiple
Nodes in a scatternet as well enter inquiry substate to discover neighboring devices and introduced a modified inquiry
other nodes within range. Inquiring nodes periodically send scheme to accelerate discovery process. Through analysis
inquiry packets on a predetermined frequency partition of hop and simulation, the authors in [8] proposed two ways to
frequencies used by piconets. Intuitively, potential collisions will reduce delay in discovery process. The work in [9] analyzed
occur between inquiry packets and packets exchanged between the frequency-matching delay and proposed three schemes
piconet nodes during connection state, which lead to degraded
to speed up discovery procedure in Bluetooth.
network performance and longer discovery process. In this
paper, we develop a mathematical model for interference caused The previous studies provided different mechanisms and
by a single inquiring node, and extend the model to predict the new protocols to accelerate discovery process. However, the
severe interference levels due to multiple inquirers. We further role of inquiry packets interference during discovery process
validate our mathematical model through a simulation scenario. was under estimated in most previous researches. We
Analytical and simulation results proved that the probability of believe that reducing delay in discovery process, especially
packets interfered and disrupted during inquiry procedure is
for multiple inquirers, depends mainly on avoiding such
periodic repeating every 3.16 sec and interference caused by a
single inquiring node is about 1.25%. Results also revealed that interference. This interference is not only generated when
five inquirers can disrupt 5.9% of network packets, and with inquiry packets share same frequency with connection state
increased number of inquiring nodes, the interference levels packets. Both packets can also be indirectly disrupted if the
become more crucial approaching 35.3%. header of the preceding packet is directly disrupted, even if
they do not share the same frequency with inquiry packets.
Keywords: Bluetooth, Inquiry, Device Discovery, Interference Thus, direct and indirect disruption will aggravate the
inquiry interference problem.
1. Introduction The rest of this paper is organized as follows: In section
2, the Bluetooth inquiry substate is summarized. The
Device discovery is an essential process to establish and Bluetooth connection state is illustrated in section 3. In
maintain successful Bluetooth scatternets. This process is section 4, the developed mathematical model for single
based on completing two phases; inquiry and paging. The inquiring node interference is presented. The extended
inquiry phase is used to search and locate other Bluetooth model of multiple inquiring node interference is deduced in
devices within range. Nodes in a scatternet enter inquiry section 5. Final conclusions are given in section 6.
substate to discover other neighboring nodes. Nodes arriving
in the vicinity as well enter inquiry substate seeking to join a
pico- or scatternet. The inquiry procedure involves sending 2. The Bluetooth Inquiry Substate
periodically inquiry packets on a predefined 32-frequency
A node enters the inquiry substate and acts as a master to
subset of 79 hop frequencies used by scatternet packets in
discover other potential slave nodes to form a piconet. In
normal connection state. Thus, interference is likely to occur
contrast, a node in the inquiry scan substate acts as a slave
between inquiry packets and connection state packets if they
to be discovered by potential masters in the inquiry substate.
share the same frequency. Such interference will degrade
Inquiring nodes use their resident CLK and General Inquiry
network throughput, preclude device discovery process, and
Access Code (GIAC) address, (9E8B33)16, to generate a
may result in undiscovered devices even within radio range.
frequency hopping sequence among a predetermined 32-
The discovery process was discussed in previous studies. frequency subset of 79 Bluetooth frequencies [3]. The set of
The authors in [1] developed an analytical model for time 32 inquiry frequencies is further segmented during inquiry
required by a master to discover one slave and proved that into two 16-frequency trains, A and B.
discovery time is affected by presence of multiple potential
An inquirer chooses A or B train for initial transmission
slaves. The work in [2] investigated optimization parameters
and repeats this train for 256 times (256 x 625 usec x 16 =
needed to improve the performance of Bluetooth connection
2.56 sec), before switching to the other train [3]. The A-
time. The authors in [5] analyzed the delay in discovery
train or B-train is chosen when the parameter KOFFSET is set
process for different situations and proposed means to
up to 24 or 8, respectively. The change of train is made by
decrease it. The proposed protocol in [6] does not separate
adding 16 to the current value of KOFFSET modulo 32. Three
inquiry and page phases to decrease interference, provide
148 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
changes have to be done within inquiry procedure length, 3. The Bluetooth Connection State
which corresponds to 10.24 sec each 60 sec, as specified in
[3]. As shown in Fig.1, the frequencies in these trains All members of a Bluetooth piconet hop together among
change over time such that each train always has 8 79 frequencies with a sequence that is a function of master's
frequencies in the 54-69 index range (Lower Range) and 8 28 bit free-running clock and first 28 bits of the master's 48
frequencies in the 70-79 and 1-6 index range (Upper bit address [4]. The hop sequence cycle covers about 23.3
Range). hours, spanning about 64 MHz of spectrum, spreading
transmissions over 80% of the available 79 MHz band, and
visiting all frequencies with equal probability [4]. The FH
channel is based on Time Division Duplexing (TDD) using
625 usec time slots. Master packets are transmitted on even
time slots and slaves respond on odd slots. Each time slot
uses different hop frequency with hopping rate 1600 hops/s.
To increase data rate, and intuitively reduce hopping rate,
multi-slot packets are defined to cover 1, 3, or 5 time slots,
named DH1, DH3, or DH5, respectively. As shown in Fig.2,
all packets contain a 72-bit Access Code to identify piconet
and 54-bit Packet header containing the 3-bit slave address.
Multi-slot packets will continue using same hop channel as
being decided by first time slot, then, the system tunes back
to the carrier frequency sequence as if only a single slot
packet had been transmitted. Using this way, the hop
frequency selection is not affected by packet lengths and all
nodes can still hop synchronously.
During connection state, the instantaneous probability of
a frequency being selected is based on 32-frequency
partitions of the Bluetooth frequency band. The hop
generator begins with a group of 32 frequencies that we call
Master Frequencies Cycle (MFC) determined by master of
Figure 1. Shifting and Selection of Inquiry Frequencies piconet. The probability that a MFC is the same 32-
frequency subset as that used by the inquiry substate is 1/79.
An inquiring device transmits a 68-bit inquiry packet The master transmits on a pseudo-randomly chosen
during an even time slot, when its two least significant clock frequency from MFC. The slave subsequently transmits on a
bits are zero (CLK1,0=002). The frequency used is pseudo- frequency selected from a 32-frequency partition
randomly selected from lower range frequencies (FL) of the immediately right to the MFC which we term Slave
current train, A or B. Another upper range frequency (FU) Frequencies Cycle (SFC). Once each of the 32-frequencies
is selected from the same train and used to transmit the in the MFC and SFC have been used exactly once (taking 64
second inquiry packet 312.5 usec later when (CLK1,0 =012) time slots), both intervals shift right by 16 frequencies
as shown in Fig.2. During odd time slots, the inquirer modulo 79 and frequency selection begins again as shown in
listens for responses from neighboring devices in inquiry Fig.3. The MFC/SFC shifts such that it uses all 79
scan substate. Responses are transmitted 625 usec after an MFC/SFC combinations before repeating a MFC/SFC such
inquiry packet is received and transmitted on same that the frequency selection is uniform across 79
frequencies, FL and FU [3]. The inquirer then selects new frequencies. Within each MFC/SFC, the hopping sequence
frequencies, FL and FU, and repeats the process, collecting is pseudo-random, based on the master’s CLK and address
data from as many neigh-boring nodes as possible during [4].
inquiry substate duration.
slave, or both) is subjected to disruption. As shown interference if it shares same frequency with FU. It is also
previously in Fig.2, both piconet master and inquiring node important to note that, while (SHIL=0) since slave header is
have coincided even time slots, thus, lower inquiry packet not subjected to FL, the slave packet is not only threatened
FL is a potential direct interferer for master packet if both by FU (SIU=1) but also it is subjected to slave header direct
share same frequency. interference if it shares same frequency with FU (SHIU=1),
which will lead to indirect interference for the next master
During odd time slots, slave packets are transmitted while
packet f(k+ 2). Since the master header f(k) is not subjected
the inquiring node is in the inquiry scan substate receiving
to direct interference from either FL or FU, hence,
responses from other nodes, then, slave packets can never be
(MHIL=MHIU=0).
directly disrupted. However, a slave packet can be indirectly
interfered if its preceding master packet’s Access Code (AC) Another example in Fig.4, when DL2 = 765 usec, which
or Packet Header (PH) is directly disrupted because a slave belongs to interval T6 in Table 1, the master packet f(k)
node can not determine whether it was addressed or not and does not suffer from FL or FU, thus, (MIL=MIU=0). The
thus will not respond to the master. Similarly, if piconet’s slave packet is only threatened by direct interference from
even slots coincided with inquiring node odd slots, slave / FL (SIL=1) but not suffering from FU (SIU=0) even if it
master packets are susceptible to direct / indirect shares same frequency with FU due to 259 usec guard time.
interference, respectively. It is obvious that this delay value, DL2 = 765 usec, does not
belong to any of five intervals in Table 2, but belongs to the
Table 1: Effect of Inquiry Delay on Packet Disruption range of DL values that will cause no interference with
either MH or SH, hence, they are not even mentioned in
Table 2.
and both upper and lower range is also periodic and changes inquirer is 1/2 since only 8 of 16 Upper Range frequencies
according to shift in MFC/SFC, thus: are in each train. Finally, if the selected frequency is in the
used train, there is only a 1/8 chance of interference since
f (m) = mod(16m,79) + 1 0 ≤ m ≤ 79 (3) any of the 8 frequencies in the used train in the Upper
Range can be selected. A similar analysis applies to the
Where f(m) marks index of the first frequency in each
Lower Range as well. With DL unknown but uniformly
MFC and m denotes all possible shifts to start a new MFC.
distributed, E [MIL, MIU, SIL, SIU] can be used which was
Similarly, f(s) marks index of the first frequency in each
determined earlier to be 0.344. Thus, the number of
SFC and s denotes all possible shifts to start a new SFC,
frequencies shared between a MFC and all inquiry
hence:
frequencies is:
f ( s ) = mod((16 s + 32),79) + 1 0 ≤ s ≤ 79 (4)
M f ( m) = M f ( m), L + M f ( m),U (9)
Thus, the number of shared frequencies between a MFC
or a SFC and lower / upper ranges of inquiry frequencies is: The probability of Direct interference between a master
packet and an inquiry packet in a MFC starting at frequency
0 . 5 × (2414 − freq ( f ( m)) ) 2402 ≤ freq( f (m)) ≤ 2412 of index f(m) is:
0 2414 ≤ freq( f (m)) ≤ 2476
0 . 5 × ( freq ( f ( m)) − 2476 ) 2478 ≤ freq( f (m)) ≤ 2480 (5) 0.344
M = P ( DM , I ( f (m))) = M f ( m) (10)
0 .5 × ( freq ( f ( m)) − 2397 ) 2403 ≤ freq( f (m)) ≤ 2427
f ( m ),U
512
16 2429 ≤ freq( f (m)) ≤ 2461
Similarly, for slave packets with unknown uniformly
0 .5 × (2493 − freq ( f ( m)) ) 2463 ≤ freq( f (m)) ≤ 2479
distributed DL, the number of frequencies shared between a
Where Mf(m),U is the number of frequencies shared SFC and all inquiry frequencies is:
between a MFC and upper range of inquiry frequencies.
S f ( s ) = S f ( s ), L + S f ( s ),U (11)
0 2402 ≤ freq( f ( m)) ≤ 2444
0 . 5 × ( freq ( f ( m)) − 2444 ) 2446 ≤ freq( f ( m)) ≤ 2474 The probability of Direct interference between a slave
16 2476 ≤ freq( f ( m)) ≤ 2480 (6) packet and an inquiry packet in a SFC starting at a
M f ( m), L =
16 2403 ≤ freq( f ( m)) ≤ 2429 frequency of index f(s) is:
0 .5 × (2461 − freq ( f ( m)) ) 2431 ≤ freq ( f (m)) ≤ 2459
0 2461 ≤ freq ( f (m)) ≤ 2479 P ( DS, I ( f ( s))) =
0.344
S f (s) (12)
512
Where Mf(m),L is the number of frequencies shared between
a MFC and lower range of inquiry frequencies. One MFC/SFC is used each 40 msec (625 usec x 64 time
slots = 40 msec), half of the packets are Master packets and
0 .5 × (2414 − freq ( f ( s )) ) 2402 ≤ freq( f (s )) ≤ 2412
the other half are Slave ones. Thus, the total probability of
Direct interference between a master/slave packet and an
0 2414 ≤ freq( f (s )) ≤ 2476
0 .5 × ( freq ( f ( s )) − 2476 ) 2478 ≤ freq ( f ( s )) ≤ 2480 (7) inquiry packet in a MFC/SFC starting at index f(m)/f(s) is:
S f ( s ),U =
0. 5 × ( freq ( f ( s )) − 2397 ) 2403 ≤ freq( f (s )) ≤ 2427
2429 ≤ freq( f (s )) ≤ 2461
P ( DM , I ( f (m))) + P ( DS , I ( f ( s )))
16
P ( DI ( f (m), f ( s))) = (13)
0 .5 × (2493 − freq ( f ( s )) ) 2463 ≤ freq( f (s )) ≤ 2479 2
Where Sf(s),U is the number of frequencies shared between The Direct interference probability is periodic, repeating
a SFC and upper range of inquiry frequencies. every (3.16 sec). One complete period of P(DM,I (f(m))),
P(DS,I (f(s))), and P(DI (f(m),f(s))) is shown in Fig.5. Note
0 2402 ≤ freq( f (s )) ≤ 2444 that the direct interference probability of a Master or Slave
0 .5 × ( freq ( f ( s )) − 2444 ) 2446 ≤ freq( f (s )) ≤ 2474 packet P(DI (f(m),f(s))) is more than 1% for about 49.37% of
16 2476 ≤ freq( f (s )) ≤ 2480 (8) the time (1.56 sec) even though its mean value is 0.0086.
S f ( s ), L =
16 2403 ≤ freq( f (s )) ≤ 2429 Master Packet
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2
Where Sf(s),L is the number of frequencies shared between Slave Packet
0.01
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2
Direct Interference Probability Master or Slave Packet
0.01
Generally, any frequency from 32 frequencies in a MFC
0.005
has a 1/32 chance of being selected. If a MFC overlaps the
0
Upper Range by Mf(m),U frequencies, hence, probability of a 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Time ( sec )
2 2.2 2.4 2.6 2.8 3 3.2
Master Packet
Indirect Interference Probability
0.02
0
Indirect interference between a master packet and an inquiry 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2
Master or Slave Packet
packet due to frequency collision in a SFC starting at a 0.02
frequency of index f(s) is: 0.01
0
0.154 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2
P ( DM , I ( f ( s ))) = S f (s ) (14) Time ( sec )
512
Figure 6. Direct/Indirect Interference Probability vs. Time
The probability of Indirect interference between a slave
packet and an inquiry packet due to frequency collision in a
MFC starting at a frequency of index f(m) is: 5. Multiple Inquiry Nodes Interference
Bluetooth scatternets will encounter more than one
0.154
P ( DS , I ( f ( m))) = M f ( m) (15) inquiring node simultaneously. Thus, all Direct and Indirect
512 interference probabilities will be as shown below.
Note that E [MHIL,MHIU,SHIL,SHIU] is used, which
was determined earlier to be 0.154. Direct Interference Probability with n Inquirers
The probability of Direct interference between a master
Total Interference Probability packet and n inquiry packets in a MFC starting at a
frequency of index f(m) is:
The total probability of a packet being disrupted is the
sum of both its probability of direct and indirect M f (m), L n
n M f (m),U n
n
interference, taking into account that if the packet is
n
P ( DM , I ( f (m))) =
32
∑ x(0.344)
x=1
n
+
32 ∑ n − x(0.344)
x=1
n
S f ( s ),L n
n n
n
∑ x(0.344) ∑ n − x(0.344)
The total probability of Direct/Indirect interference S f ( s ),U
P ( DSn,I ( f (s ))) = n
+ n
Finally, the probability that at least one inquiry node is The direct and direct/indirect interference probabilities
using the same master/slave frequency, assuming that for n= 5 of a Master/Slave packet, are shown in Fig.7,
piconet packet is subjected and in the same train and range respectively. Note that the direct interference probability of
is (1 − (7 8 )y ). a Master or Slave packet approaches 5% approximately
44.3% of the time (1.4 sec) even though its mean value is
The total probability of Direct interference between 0.0411. Whereas the direct/indirect interference probability
master/slave packet and n inquiry packets in a MFC/SFC of a Master or Slave packet is above 7% approximately
starting at frequencies of index f(m)/f(s), respectively, is: 46.84% of the time (1.48 sec) even though its mean value is
0.0594.
P ( DMn , I ( f ( m))) + P ( DSn, I ( f ( s )))
P ( D In ( f ( m), f ( s ))) = (21)
0.06
Master or Slave Packet
2
Direct
Indirect Interference Probability with n Inquirers 0.02
Interferenc e Probability
Direct/Indirect
0.06
S f ( s ), L n
n S f ( s ),U n
n
∑ (0.154 )n + ∑ (0.154 )n
0.04
P ( D Mn , I ( f ( s ))) =
32 x=1 n − x 32 x =1 x 0.02
x
( )
x 0
∑
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2
× (0.5)x 1 − (7 8) y (22) T ime ( sec )
y= 0 y
Figure 7. Interference Probabilities of 5 Inquirers vs. Time
The probability of Indirect interference between a slave
packet and n inquiry packets due to frequency collision in a A MATLAB simulation was conducted to verify our
MFC starting at a frequency of index f(m) is: mathematical models. The scenario involved multiple
piconets, with random clock and address values, generating
M f (m), L n
n M f (m),U n
n
P ( DSn,I ( f (m))) =
32
∑ n − x (0.154)
x=1
n
+
32 ∑ x(0.154)
x=1
n both master and slave packets as MFC/SFC achieves all 79
possible shifts. Up to 60 Inquiring nodes arrive to piconets
x x neighborhood with random DL between 0 and 1250 usec.
∑
× (0.5)x 1 − (7 8) y
y=0 y
( )
(23)
The mean Direct/Indirect collisions between piconets
packets and inquiry packets were recorded every 3.16 sec.
Analytical and simulation results are shown in Fig.8, from
Where the probabilities of a master/slave packet being
which it is clear that as number of inquiring nodes
subjected to Indirect interference in the lower/upper range
increases, interference levels become more crucial.
by n-x/x out of n inquiry packets with unknown delays are
Considering only direct interference, probability approaches
n n
n − x (0 . 154 )n and x (0 . 154 ) respectively.
n
30.7%. Including direct and indirect interference, it
approaches 35.3% since Master/Slave packets transmitted in
Upper or Lower Range are disrupted and indirectly cause
Total Interference Probability with n Inquiry Nodes next Slave/Master packets to be interfered even if the next
The total probability of Direct/Indirect interference Slave/Master frequencies do not overlap the Upper/Lower
between a master packet and n inquiry packets in a frequency Ranges.
MFC/SFC starting at frequencies of index f(m)/f(s), is: 0.4
0.35
P ( M Dn , D , I ( f ( m), f ( s ))) = P ( D M
n
, I ( f ( m))) + P ( D M , I ( f ( s )))
n
(24)
Total Interference Probability
0.3
− P (D M
n
, I ( f ( m))) × P ( D M , I ( f ( s )))
n
0.25
6. Conclusions
In this paper, we showed that during discovery process,
the potential collisions between both inquiry and scatternet
packets in connection state are significant. Scatternet nodes
and inquiring nodes will be affected by such interference
that degrades network throughput and elongate discovery
time. We further developed a mathematical model for
interference probability caused by a single inquirer, and
extended the model to predict the severe interference levels
due to multiple inquiring nodes. Results revealed that
interference probability is periodic, repeating every 3.16 sec
over all possible MFC/SFC. The interference caused by a
single inquirer is 1.25%. Five inquiring nodes result in
interference probability of more than 7% for approximately
1.48 sec even though the mean probability of interference is
only 0.059. However, as number of inquirers increase, the
interference levels become more crucial. Considering only
direct interference, results proved that probability reaches
30.7%. Including both direct and indirect interference, it
approaches 35.3% since Master/Slave packets transmitted in
Upper or Lower Range are always disrupted and indirectly
cause the next Slave/Master packets to be interfered.
References
[1] D. Chakraborty, G. Chakraborty, K. Naik, N. Shiratori,
“Analysis of the Bluetooth device discovery protocol”,
Springer Wireless Networks Journal, February 2010,
Vol. 16, pp.421-436.
[2] T. Thamrin, S. Sahib, “The Inquiry and Page
Procedure in Bluetooth Connection,” IEEE Inter.
Conf. SOCPAR09, Dec.2009, pp.218-222.
[3] www.bluetooth.com, “Bluetooth Specification
v4.0”,Dec.17,2009
[4] M. Khaled, H. Xiong, Q. Gao, “Performance
Estimation and Evaluation of Bluetooth Frequency
Hopping Selection Kernel,” IEEE International
Conference ICPCA09, Dec.2009, pp.461-466.
[5] D. Chakraborty, G. Chakraborty, S. Naik, N. Shiratori,
“Discovery and Delay Analysis of Bluetooth Devices”,
IEEE International Conference MDM06, May 2006,
pp.113-117.
[6] S. Z. Kardos, A. Vidacs, “Performance of a new device
discovery and link establishment protocol for
Bluetooth”, IEEE International Conference
GLOBECOM05, January 2006, Vol.6, pp.3522-3527.
[7] X. Zhang, G.F. Riley, “Evaluation and accelerating
Bluetooth device discovery,” Radio and Wireless
Symposium, Jan. 2006, pp.467-470.
[8] G. Chakraborty, K. Naik, D. Chakraborty, N. Shiratori,
& D. Wei, , “Delay analysis and improvement of the
device discovery protocol in Bluetooth”, IEEE
Vehicular Tech. Conf., 2004, Los Angeles, USA.
[9] J. Jiang, B. Lin, & Y. Tseng, “Analysis of Bluetooth
device discovery and some speedup mechanisms”,
Journal of Institute of Electrical Engineering, 2004,
11(4), 301–310.
154 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
Abstract: In our paper color image segmentation using DMRF used as the color models. we proposed a Double MRF
model with edge penalty function as image model and for color (DMRF) model that takes case of intra-plane interactions as
( )
model using RGB and OHTA I1 , I 2 , I 3 color model. In our well as the inter-plane interactions[10]. These proposed
paper we are proposing Hybrid algorithm based on DMRF models have been tested on different images and also the
model and The image segmentation problem is formulated as a results obtained have been compared with the results
pixel labeling problem and the pixel labels are estimated using obtained using only MRF. Both first order and second order
Maximum a Posterior(MAP) criterion. The MAP estimate of the models have also been studied. As expected, it has been
labels are obtained by the proposed hybrid algorithm converges observed that the use of OHTA model yielded better results
to the solution faster than that of Simulated Annealing(SA) than that of using RGB model We have used line process in
algorithm. The DMRF with edge preserving attribute yields the clique potential function and it has been found that
better segmentation result with local edge boundary. It is also although the line process could preserve well defined edges
found that Exponential function improved the result to some it failed to preserve loosely defined boundaries [11][12]. In
extend. The paper is divided in six section in second section we
order to preserve loosely defined boundaries, we have
explain the model use for Hybrid Algorithm.
employed the edge penalty model proposed by Qiyao Yu et
Keywords: DMRF, MAP, RGB-Plane, Edge penalty function. al [11][12]. Here, we have used RGB color models with
MRF and DMRF models. Qiyao Yu et al [11][12] have used
exponential edge penalty function preserve the weak edges
1. Introduction & Background together with strong edges. We have proposed the
Image segmentation is an important process in automated exponential edge penalty function in the DMRF model to
image analysis. In color image segmentation, choice of preserve edge. Here also, the problem is formulated as a
proper color model is one of the important factors for pixel labeling problem and the pixel label estimates are the
segmentation and different color models such as RGB, HSV, MAP estimates.
( )
YIQ, OHTA I1 , I 2 , I 3 , CIE (XYZ, Luv, Lab) are used to
represent different colors [1],[2][3]. Besides color model, 2. Methodology
image model does play a crucial role for color image Here we are proposing the model for our algorithms. The
segmentation. Stochastic models, particularly Markov image segmentation problem is formulated as a pixel
Random Field (MRF) models, have been extensively used as labeling problem. The labels are estimated using the MAP
the image model for image restoration and segmentation estimation criterion usually, the MAP estimates are obtained
problems [4][5]. Often, in stochastic framework, the color by Simulated Annealing (SA) algorithm. It is observed that
image segmentation problem is formulated as a pixel SA algorithm is computationally involved and hence we
labeling problem and the pixel labels are estimated using the have proposed a hybrid algorithm. Which is found to be
Maximum a Posteriori (MAP) criterion[6][7]. By and large, faster than that of SA algorithm. The proposing models are
the MAP estimates of the pixel labels are obtained using as fallow.
Simulated Annealing (SA) algorithm [11][8]. MRF model
has also been successfully used as the image model while
addressing the problem of color image segmentation in
supervised and unsupervised framework [9]. Here we use, 2.1 Model 1 (Double Markov Random Field (DMRF)
color image segmentation problem is formulated in Model)
stochastic framework. Markov Random Field model is used
as the image model and RGB and OHTA model have been The DMRF IS DEFINE AS In order color planes, the a prior
(IJCNS) International Journal of Computer and Network Security, 155
Vol. 2, No. 5, May 2010
model takes case of intra-plane interactions and inter-plane observed image X which is assumed to be MRF is
interactions as well. This MRF model taking case of intra-plane considered as a degraded version of the label process and
and inter-plane interaction is called Double MRF (DMRF) model. hence consists of Z and W. The model parameter for both
Capturing salient spatial properties of an image leads to the inter and intra color plane processes are selected on a trial
development of image models. MRF theory provides a and error basis. The weak membrane model is considered to
convenient and consistent way to model context dependent the model intra as well as inter plane clique potential
entities for eg. image pixel and correlated features. Through function. For example, the weak membrane model for
the MRF model takes into account the local spatial inter-plane interaction process for two color components,
interactions, it has limitation in modeling natural scenes of
( I1 and I 2 ) (as shown in Fig.1(b) ) can be expressed as
distinct regions. In case of color models, it is known that
I
( )
there is a correlation among the color components of RGB 2 2
1 I + n I2 − n I1
model. In our formulation, we have de-correlated the color Vc z = α n ' 2 ' − n '1 ' i' , j' i ', j ' −1 +
components and introduced an interaction process to
i , j i −1, j
improve the segmentation accuracy. We have employed 2 2
n I 2 − n I1 I I
inter-color plane interaction (OHTA ( I1 , I 2 , I 3 ) color i ' , j' i ', j ' +1 + ni'2, j ' − ni'1+1, j '
model) process which reinforces partial correlation among
different color components. Where α is the MRF field parameter. Superscripts I1 and I 2
We assume all images to be defined on discrete on corresponds to I1 -plane and I 2 -plane respectively. This
rectangular lattice M1 × M 2 . Let Z denote the label process notation is depicted in Fig.1, where the intra plane and inter
corresponding to the segmented image and z is a realization plane interaction are shown. The model parameter both Z
of the label process i.e. the segmented image. The observed and I processes are selected on an Ad hoc manner. The
image X is assumed to be a random field that is assumed to weak membrane model for intra plane as well as inter plane
be the degraded version of the label process Z and this clique potential function.
degradation process is assumed to be Gaussian process W.
The label process Z in a given plane is assume to be MRF.
The prior model takes care of inter as well as intra color
plane interactions. The prior probability distribution of z is
constructed with clique potential function corresponding to
intra-color plane interaction, for example I1 and another
clique potential function within inter plane interactions (for
example between I1 and I 2 ).The of inter color plane
interaction are shows in Figure1.Thus X is a compound
MRF. MRF model taking care of both inter as well as intra
plane interactions. It known that if Z is assumed MRF, then
the prior probability distribution P(Z=z) is Gibb’s
distributed that can be expressed
as P (Z = z | θ ) = 1 e −U (z ,θ ), where Z ' = ∑ z e ( ) is the
−U z, θ Figure 1. (a) RGB Plane Interaction
'
Z
partition function, θ denotes
x = arg max P ( X = x | Y = y)
)
Where U i (s ) is the number of neighbors of s that does not
(1) '
x
Where P ( X = x|Y = y ) the posterior probability is x is the belongs to Ω i . That means the term βU i (s ) is not zero
) '
MAP estimation using Baysian rule. A 2-layer model is only when the site s is at the boundary. Therefore, (4)
constructed from Bayes’ rule. becomes
P (Y = y | X = x )P ( X = x) [
Y(s) −µi ]
P ( X = x|Y = y) =
2
n n
P (Y = y) +β ∑ ∑ Ui(s)
(2) ) '
x= argmin∑ ∑ logσi + (6)
( )
Where P Y = y| X = x is the posterior conditional
Ω1,....Ωni=1s=Ωi
2 σ
2
i ( ) i=1s=Ωi
probability of the segmentation given the observed data,
P (Y = y| X = x)P ( X = x) is the conditional probability ∑in=1∑ s=Ωi U i (s ) is approximately proportional to the
'
distribution of the observation conditioned on the length of the obtained boundary. Therefore the role of the
( )
segmentation, P X = x is the prior probability of the region model term actually penalizing the existence of the
segmentation , and P (Y = y ) is the probability is the
boundary by its population.
Instead of penalizing of all boundary pixel, greater penalty
distribution of the observation. Maximization of the can be applied to weak edge pixel and a lesser penalty to
( )
posterior P X = x|Y = y gives a segmentation solution, strong edge pixels, so that local statistics such as edge
and since P (Y = y ) is a constant it is equivalent to strength can be incorporated. Therefore we can replace the
( ) [
Y(s) − µi ]
n 2 (figure 2) increases, the penalty difference between weak
'
− βUi (s) (5)
)
x = argmin ∑ ∑ logσi + and strong edges decreases. When K approaches infinity, all
Ω1,...Ωni=1 s=Ω1
2 σi
2
( ) edge penalties are equally 1. Therefore, the parameter K
(IJCNS) International Journal of Computer and Network Security, 157
Vol. 2, No. 5, May 2010
(figure 2) of the edge penalty function g(.) is allowed to result obtain MRF and DMRF model. The models
increase during the segmentation, details of which are parameters are tabulated in table (II). For exponential edge
described the local feature of edge strength is hence utilized, penalty function (EPF) the value of K is choose to be σ =5.
but the bias is significantly reduced. It is observed that interior edges could be observed and
percentage of misclassification error is less than that of
3. Proposed HYBRID Algorithm using edge penalty functions. The results has been improve
A new hybrid algorithm has been proposed to obtain the in case of DMRF model as compared to MRF model. As
MAP estimate. In the hybrid algorithm SA algorithm is first seen from the table (I), the misclassification error is less in
run for some pre specified amounts of epochs and then ICM case of DMRF model as compared to MRF. Use of
algorithm is run until the stopping criterion has been Exponential EPF preservation of weak edge. Fig. 3(e) and
satisfied. The combination of global and local concept leads Fig.3(c) shows the image of MRF model with exponential
to a faster converting algorithm. The proposed hybrid EPF yields better results than that of MRF model. As seen in
algorithm is described below Fig. 3(e) and Fig. 3(f) the image of MRF and DMRF model
with the Exponential edge penalty function as shows that
the images of DMRF model with Exponential edge penalty
1) Initialized the temperature Tin . function yields better results than that of MRF exponential
2) Compute the energy U of the configuration. penalty function, As seen from fig. 3(e) and 3(f) the nearer
3) Perturb the system slightly with suitable Gaussian the diseased part (toward the right hand top) the inside edge
distribution. has been formed using Exponential edge penalty function
' (EPF) while for MRF model. This is mission for many other
4) Compute the new energy U of the perturbed system and
cells similar observation can also be made. In this also we
'
evaluate the change in energy ∆U =U −U . have employed SA and proposed hybrid algorithm. The
5) If ( )
∆U 〈 0 , accept the perturbed system as the new convergence of these algorithm are presented in fig. 6
configuration Else accept the perturbed system as the where we observe that the hybrid algorithm converges at
(
new configuration with a probability exp − ∆U / t , ) around 600 iteration where as SA converges at 6000
iterations. It is also observed that Hybrid algorithm is fast
where t is the cooling schedule. Converges than the SA algorithm.
6) Decrease the temperature according to the cooling
The image consider in shown in fig. 4, where are some weak
schedule.
7) Repeat steps 2-7 till some pre specified number of epochs
edge seen in the water boat etc. fig. 4 shows the results
of SA. obtained using MRF and DMRF. As seen from fig. 4(c) to
8) Compute the energy U of the configuration. (d) the weak edge have not been presented. In case DMRF
9) Perturb the system slightly with suitable Gaussian there are more misclassification as seen from fig. 4 show the
disturbance. results obtained using MRF and DMRF with exponential
' edge penalty functions. As fig. 4(e) correspond to MRF with
10) Compute the new energy U of the perturbed system
exponential, observed the weak edge of the bottom left side
'
and evaluate the change in energy ∆U = U −U . have been preserved and strong have been formed. This
( )
11) If ∆U 〈 0 , accept the perturbed system as the new phenomenon is more prominent in case of second order
DMRF model as shown in fig. 4(f). As found from Table (I)
configuration otherwise retain the original
configuration. the percentage of misclassification error is less than that of
12) Repeat steps 8-12, till the stopping criterion is met. The using Exponential edge penalty functions. As shows the
(
stopping criterion is the energy U 〈 threshold . ) convergence curves for hybrid and SA algorithm converges
at 600 iteration where as SA converges after 6000 iterations.
Similarly observation are also made for the third images
shows in fig. 5(c) and 5(d) shows the bird image and the
4. Simulation and Results corresponding ground truth observed the weak edge could
In this paper we have considered three different image not be preserved. As seen from fig. 5(e) and 5(f), the MRF
having strong as well as weak edges. These images are and DMRF model with Exponential edge penalty function
considered deliberately to validated the weak edge performed better than that of MRF and DMRF model . In
preserving attributes of the schemes. The first image this case also the weak edge would be preserved the model
considered is a cell the image as shown in fig. 3(a), fig. 3(b) parameter are tabulated in Table (II). Hence here also hybrid
to fig. algorithm converged faster than that of the SA algorithm.
3(g) show the results obtained using MRF, DMRF, and
JSEG methods. Although line field is used to preserved the
edges, the weak edges inside the cell boundary are not being
preserve. As observed from fig. 3 some hazy edge could be
formed because of weak edges, still there are no clear edge
for present inside each cell. Fig.
(e) and Fig. 3(f) shows the result obtained with MRF and
DMRF models with edge penalty function incorporated in
the energy functions. We have also proposed Exponential
edge penalty function. Fig. 3(c) and Fig. 3(d) shows the
158 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
better result than that of using DMRF model. We have Classification and Region Merging,” Proc of the Second
employed hybrid and SA to obtain segmentation and hybrid Canadian Conference on Computer and Robot Vision
algorithm converges faster than that of SA algorithm. volume 1(May 2005):pp.579-586.
Current work focuses estimation of model parameters and
devising an unsupervised scheme.
References
Abstract: This paper incorporates a comprehensive study on the and independently chosen spreading sequences accurately
performance evaluation of a Reed-Solomon encoded multi-user models the situation where pseudo-noise sequences span
Direct Sequence Code Division Multiple Access (DS-CDMA) many symbol periods [5].
communication system in terms of Bit Error Rate (BER). The DS-
CDMA system under present study employing uncorrelated
Direct Sequence Code Division Multiple Access (DS-
spreading codes has incorporated BPSK digital modulation over
an Additive White Gaussian Noise (AWGN) and other fading CDMA) has already been established as a strong candidate
(Rayleigh and Rician) channels. Computer simulation results on for third-generation mobile and personal communication
BER demonstrate that the Reed-Solomon encoded multi-user DS- systems. Now a days, the most prominent applications of the
CDMA system outperforms in retrieving the transmitted messages CDMA are in mobile communication systems like cdmaOne
of the individual senders .The system is highly effective to combat (IS-95), UMTS or cdma2000. To apply CDMA in a mobile
inherent interferences under Rayleigh and Rician fading channels. radio environment, specific additional methods are required
It is anticipated from the study that the performance of the
communication system degrades with the lowering of the Signal-
to be implemented in all these systems. Methods such as
to-Noise Ratio (SNR). power control and soft handover have to be applied to
control the interference by other users and to be able to
Keywords: Direct Sequence Code Division Multiple Access separate the users by their respective codes [6], [7].
(DS- CDMA), Signal-to-Noise Ratio (SNR), Additive White
Gaussian Noise (AWGN), Fading channels.
2. System Description
1. Introduction We assume that a multi-user Direct Sequence Code Division
Multiple Access (DS-CDMA) communication system
Direct Sequence Code Division Multiple Access (DS-
depicted in Figure 1 utilizes a Reed-Solomon channel
CDMA) is one of the most prominent air-interface
coding scheme. In transmitting section of the present Reed-
techniques in today’s wireless communication systems [1].
Solomon encoded DS-CDMA system, a typically named b
In a DS-CDMA system all users transmit in the same source information bits of rate rb are grouped and mapped to
bandwidth simultaneously [2]. This transmission technique
one of the M= 2b information symbols. The information
is a multiple access communication scheme in which each
symbols with rate of r s= rb/ log2M ks/s (ksymbols/s) enter a (n,
user is assigned a unique spreading code for modulating the
k, t) Reed–Solomon encoder operating in GF (M) with n, k,
transmitted symbols. Thus, a receiver with knowledge about
t=[n-k]/2. The output of the Reed–Solomon encoded
the code of the intended transmitter can select the desired symbols is of rate rsc= rs/rc, where, rc= k/n. The coded
signal. A DS-CDMA communication system receiver has symbols are fed into symbol to binary converter and its
generally three main obstacles to overcome such as Multiple
output are then converted into bipolar NRZ format and
Access Interference (MAI) from other users, Inter-symbol
subsequently fed into BPSK digital modulator. The digitally
Interference (ISI) due to a multipath channel and additive
modulated signals are then multiplied with the generated
noise at the receiver. In a DS-CDMA system, the transmitter
pseudo-random noise for each user. The pseudo randomly
multiplies each user’s signal by a pseudo-random spreading noise contaminated modulated signals for user1 and user2
sequence [3], [4]. In fact, user’s spreading codes are
are then added up and passed through AWGN or fading
designed to be orthogonal (or quasi-orthogonal). However,
channel. The channel output is then multiplied with the
due to channel properties like multipath propagation,
generated pseudo-random noise and demodulated. The
orthogonality cannot be maintained at the receiver and the
demodulated signal for each user is processed through
signals interfere with each other. In practical CDMA integrator. The integrated output is sent up to the decision-
systems the spreading codes among the users are not making device and eventually, the information bits
perfectly orthogonal. Thus, a CDMA channel with randomly
(IJCNS) International Journal of Computer and Network Security, 161
Vol. 2, No. 5, May 2010
transmitted for each user are recovered [8], [9]. Figure 4 respectively. In Figure 4, it is quite evident that the
amplitude of the signal is maximum around the frequency of
RF Direct sequence 100 KHz. Figure 5 demonstrates the simulation results for
carrier generator for user #1
uncoded multi-user DS-CDMA communication system
under an AWGN channel. It is observable that its value
Mapping into
Binary data Reed- bipolar NRZ BPSK
approaches zero for a SNR value of 5 dB. In Figure 6, the
for user #1 Solomon 1 to +1 modulator BER simulated graphs under AWGN and Rayleigh fading
Encoder 0 to –1
channels are illustrated for user 1 and user 2.
RF Direct sequence
carrier generator
for user # 2
Mapping into
Binary data Reed- bipolar NRZ: BPSK
for user #2 Solomon modulator
1 to +1
Encoder 0 to –1
Direct sequence
generator for user # 1 AWGN/
Sampling Fading
channel
Detection BPSK
Device Integrator Demodulator
Figure 5. BER Performance of the uncoded multiuser DS- Figure 8. BER Performance of the Reed- Solomon
CDMA communication system under AWGN channel. encoded multi-user DS-CDMA communication system
under AWGN channel.
4. Conclusions
In our present study, we have studied the performance of a
Authors Profile
multi-user Direct Sequence Code Division Multiple Access Nargis Parvin received the B.Sc. and M.Sc. degree in Information
(DS-CDMA) communication system adopting the Reed- and Communication Engineering from University of Rajshahi,
Solomon channel coding and BPSK digital modulation Bangladesh.
scheme. A range of system performance results highlight the
impact of a simplified digital modulation technique and Farhana Enam received the B.Sc. and M.Sc. degree in Computer
different uncorrelated spreading codes. In the context of Science from University of Rajshahi, Bangladesh.. She is currently
system performance, it can be concluded that the working as Assistant Professor of the dept. of Information and
implementation of BPSK digital modulation technique in Communication Engineering, University of Rajshahi, Bangladesh.
Reed-Solomon channel encoded multi-user Direct Sequence
Shaikh Enayet Ullah received the B.Sc. and M.Sc. degree in
Code Division Multiple Access (DS-CDMA)
Applied Physics from University of Rajshahi, Bangladesh.. He has
communication system provides satisfactory performance in completed his PhD from Jahangirnagor University, Bangladesh. He
retrieving the transmitted messages of the individual users. is currently working as a Professor and Chairman of the dept. of
Information and Communication Engineering, University of
Rajshahi, Bangladesh.
References
Saifur Rahman received the B.Sc. and M.Sc. degree in
[1] W. C. Y. Lee, “Overview of cellular CDMA,” IEEE Information and Communication Engineering from University of
Transactions on Vehicular Technology, vol. 40, no. 2, Rajshahi, Bangladesh.
pp. 291–302, 1991.
[2] P. Jung, P. W. Baeir, and A. Steil, “Advantages of
CDMA and spread spectrum techniques over FDMA
and TDMA in cellular mobile radio applications,” IEEE
Transactions on Vehicular Technology, vol. 42, no. 3,
pp. 357–364, 1993.
[3] R L. Pickholtz, D. L. Schilling, and L. B. Milstein,
“Theory of spread-spectrum communications” A
tutorial, IEEE Transactions on Communications, vol.
30, no. 5, pp. 855–884, 1982.
[4] R. L. Pickholtz, L. B. Milstein, and D. L. Schilling,
“Spread spectrum for mobile communications,” IEEE
Transactions on Vehicular Technology, vol. 40, no. 2,
pp. 313–322, 1991.
[5] F. C. M. Lau and C. K. Tse, “Chaos-Based Digital
Communication Systems: Operating Principles,
Analysis Methods, and Performance Evaluation,” First
edition, Springer, 2003.
[6] K.S. Gilhousen et al., “On the Capacity of a Cellular
CDMA System”, IEEE Trans. Vehic. Technol., Vol. 40,
no. 2, pp. 303–312, 1991.
[7] M.G. Jansen and R. Prasad, “Capacity, Throughput, and
Delay Analysis of a Cellular DS CDMA System With
Imperfect Power Control and Imperfect Sectorization”,
IEEE Trans. Vehic. Technol., Vol. 44, no. 1, pp. 67–75,
1995.
[8] A.C. Iossifides and F-N. Pavlidou, ‘’ Performance of
RS-Coded DS/CDMA Micro cellular Systems with
M-ary orthogonal Signaling,” Wireless Personal
Communications 10: pp. 33–52, Netherlands, 1998.
[9] Rappaport, S.Theodore,“Wireless Communications,”
Principle and Practice, Second Edition, Pearson
Educatiob (Singapore) Pte., Ltd.,New Delhi,India,
2004.
164 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
2. Methodology
given time instant were represented according to a color both differ greatly or slightly from one activation to another.
scale. Videos of each AF segment were generated. In the By application of the state-space realization algorithm to
wave front propagation mode, only lines connecting points every heart beat with a period of approximately 1 s and 175
on the surface with a voltage equals to zero were ms during atrial fibrillation, it was possible to reproduce the
represented. This display mode allows the visualization of electro gram impulse response which, in turn, makes
both the polarization and re polarization wave fronts possible a quantitative comparison as to the variation and
depending on the time interval selected. The succession of complexity of a set of consecutive heart beats. The wavelet
wave fronts is represented using a color scale according to decomposition of scddvbrk signal is shown in Figure 3.
the time instant in which each wave front appears. We used
a method as previously reported [16]. After removal of a 50-
Hz disturbance component emanating from power supplies,
and after subtraction of the mean signal level of all
recordings, data reduction by means of low-pass filtering,
and resembling at rate of 100 Hz, experimental data from
the 78 electrode array were arranged in impulse response
matrices of dimensions the number being the number of data
points in each recording between two consecutive heart
beats. Subsequent impulse response analysis and simulation
of the resulting state-space realization were applied as
described in next section.
3
frequencies,” IEEE Trans. Inf. Theory, vol. 38, no. 2, pp.
644–664, Mar. 1992.
phas e(rad)
1
[5] G. Erlebacher, M. Hussaini, and L. Jameson, Wavelets:
0
Theory and Applications. New York: Oxford Univ.
-1
Press, 1996, pp. 115–120.
0 500 1000 1500
[6] M. Unser and A. Aldroubi, “A review of wavelets in
biomedical applications,” Proc. IEEE, vol. 84, no. 4,
Figure 6. Phase curve extraction of ECG signal using
pp. 626–638, Apr. 1996.
reverse bi-orthogonal wavelet
168 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
Authors Profile
Dr.R.Rangarajan graduated in
Electronics and communication
Engineering from Madras university in
1972.He took up his academic pursuit in
1975 as a lecturer in coimbatore institute
of technology after completing his post
graduation with second rank from
madras university, he continued to serve at coimbatore
institute of technology and held position as a assistant
professor, professor and HOD of ECE during his tenure. His
area of specialization is VLSI Design and completed his
PhD in Bhartiar university.
(IJCNS) International Journal of Computer and Network Security, 169
Vol. 2, No. 5, May 2010
Email: awadheshkumar.knit@gmail.com
3
SR.Lecturer, Deptt. Of MCA. SRMSCET, Bareilly (UP)
Email: ajeetkumarsrms@gmail.com
4
Lecturer, Computer Science & Engg., SRMSWCET, Bareilly (UP)
Email: rati1234@gmail.com
Abstract: Several new network types have emerged over the network. In this paper we propose an energy efficient group
last couple of years. Many of these networks are being communication technique that aims at reducing the energy
connected together to provide mobile users with the capability of drain of the group leader nodes and the network as a whole.
staying always connected to the Internet which requires
seamless transitions from one network to another without Energy management remains a critical problem in wire- less
causing interruption to on-going connections. To maintain networks since battery technology cannot keep up with
connectivity during handoff, all the networks that are accessible
rising communication expectations. Current approaches to
to the mobile station need to be known. The Architectural issues
in all-wireless networks are discussed in this regard. The paper
energy conservation reduce the energy consumption of the
aims to propose different for the energy-efficient network that wireless interface either for a given communication task or
enables fast discovery and selection of available access during idling. However, a complete solution requires
networks. We suggest three energy efficient techniques for minimizing the energy spent for both communication (i.e.,
MANET point of view. for data and control overhead) and idling. This problem can
be expressed as an energy-efficient network design problem,
Keywords: energy efficiency, Architecture of wireless which is, not surprisingly, NP-hard.
networks, Minimum energy transmission, energy- aware
protocol, power management, power control and topology Energy is a critical resource in the design of wireless
control. networks since wireless devices are usually powered by
batteries. Battery capacity is finite and the progress of
1. Introduction battery technology is very slow, with capacity expected to
Wireless networking has witnessed an explosion of interest make little improvement in the near future. Under these
from consumers in recent years for its applications in mobile conditions, many techniques for conserving power have
and personal communications. As wireless networks become been proposed to increase battery life. In this dissertation we
an integral component of the modern communication consider many approaches for conserving the energy
infrastructure, energy efficiency will be an important design consumed by a wireless network interface. One techniques
consideration due to the limited battery life of mobile include the use of power saving mode, which allows a node
terminals. Power conservation techniques are commonly to power off its wireless network interface (or enter a doze
used in the hardware design of such systems. Since the state) to reduce energy consumption. The other is to use a
network interface is a significant consumer of power, technique that suitably varies transmission power to reduce
considerable research has been devoted to low-power design energy consumption.
of the entire network protocol stack of wireless networks in
an effort to enhance energy efficiency. This paper presents a 2. Architectural Issues in Wireless Networks
comprehensive summary of the various approaches and The ad hoc wireless networks are quite different from the
techniques addressing energy efficient A wireless sensor cellular systems that have been developed in the commercial
network is a network consisting of spatially distributed domain. Cellular systems have fixed base stations, which
autonomous devices using sensors to cooperatively monitor communicate among themselves using dedicated non-
physical or environmental conditions. As sensor networks wireless lines; thus, the only multicast problems that are
are large in scale, grouping techniques are required to tackle new in those systems involve tracking the mobile users.
their scalability issues. The limited energy reserves of the However, in ad hoc wireless networks it is possible to
autonomous devices require grouping techniques to be establish a wireless link between any pair of nodes, provided
energy-aware and strive to extend the life time of the that each has a transceiver available for this purpose and
170 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
that the signal-to-noise ratio at the receiving node is addresses and routing trees. There are two basic approaches
sufficiently high. Thus, unlike the case of wired networks, to multicast tree construction. The first is the use of Source-
the set of network links and their capacities are not Based Trees (SBT), which are rooted at the sender and
determined a priori, but depend on factors such as distance which are designed to minimize the number of
between nodes, transmitted power, error-control schemes, transmissions needed to reach all of the members of the
other-user interference, and background noise. Thus, even multicast group. The second is the use of Core-Based Trees
when the physical locations of the nodes are fixed, many of (CBT) [5], under which the same tree is used for all
the factors that affect network topology are influenced by the communication within the multicast group. The Sparse
actions of the network nodes (either those directly Mode of the Protocol Independent Multicasting (PIM)
participating in a link, or those contributing to the protocol [7] can be used with either SBTs or CBTs, whereas
interference that affects a link). Furthermore, in ad hoc the PIM Dense Mode is based on the use of SBTs. As
networks no distinction can be made between uplink and pointed out earlier, the characteristics of the wireless
downlink traffic, thus greatly complicating the interference medium and the ad hoc network architecture may render
environment. In this paper, we focus on wireless networks multicasting techniques developed for no wireless
in which the node locations are fixed, and the channel applications inappropriate, or at least unable to provide
conditions unchanging. The wireless channel is acceptable performance in some scenarios. There are
distinguished by its broadcast nature; when unidirectional numerous and complex issues that must be addressed in
antennas are used, every transmission by a node can be wireless multicasting.
received by all nodes that lie within its communication
range. Consequently, if the multicast group membership 3. Energy Efficient Techniques
includes multiple nodes in the immediate communication
vicinity of the transmitting node, a single transmission In this section we are proposing three energy efficient
suffices for reaching all these receivers. Hence, there is an techniques for ad hoc network environment. The first
incentive to perform a multicast by using high transmitter technique minimizes route request message. Second
power. Of course, doing so results in interference with more technique optimizes the transmission power at each node
nodes than if reduced power were used. Thus, there is a and third techniques increases network capacity by topology
trade-off between the long “reach” of a single transmission control mechanism.
and the interference (and/or delay) it creates. Another
undesirable impact of the use of high transmitter power is 3.1. Techniques to Minimize Route Request
that it results in increased energy usage. Since the
propagation loss varies nonlinearly with distance (at Let us consider a multi-hop homogeneous wireless network
somewhere between the second and fourth power), in in which nodes are randomly deployed over certain
unicast applications it is best (from the perspective of geographical area. In this mobile ad hoc networking
transmission energy consumption) to transmit at the lowest environment each mobile node can access the internet
possible power level, even though doing so requires multiple applications via one or more numbers of gateways. The
hops to reach the destination. However, in multicast mobile nodes communicate with the gateways directly
applications it is not prudent to draw such conclusions (single hop) or through multi hop depending upon the
because the use of higher power may permit simultaneous transmission range of the node as shown in the figure 2.
connectivity to a sufficiently large number of nodes, so that Besides the internet access nodes can also transmit data
the total energy required to reach all members of the among themselves. It is assume that the gateways are
multicast group may be actually reduced. Furthermore, even stationary. The geographical area covered by gateway is
for unicast applications, the use of lower power (and, hence, partitioned into different logical group with unique group
multiple hops) necessitates the complex coordination of number assigned by the gateways. Groups are overlapping
more signals and, therefore, may actually result in higher and there are some nodes present in the overlapping area.
total energy expenditure.Figure1 shows an architecture of a The partitioned is based upon the number of nodes resent in
wireless networks. it. Nodes are categorized into active node (AN) and common
node (CN). Nodes present in the overlapping area of groups
are called common nodes while nodes belonging to any
particular group called active node of that
group.
It is assumed that overlapping area of the different group farthest node while latter is the power required to reach the
contains enough number of common nodes (CN) as inter- nearest node. The objective of power control here is to
group communication will takes place through them. When minimize the power consumption of a node. It is assumes
any active node wants to send route request (RREQ) that there may exist some intermediate power level between
message it append its group number in the packet and Pmax and Pmin. Let P(uv) be the power needed to support
broadcast the message. Message will forwarded by the other communication from node u to v, we called
node if they belonging to the same group otherwise message it symmetric if P(uv) = P(vu). The power requirement is
will be dropped. When a CN prepares the RREQ message it called Euclidean if it depends upon Euclidean distance ||uv||.
add one group number from the group it belonging It can be calculated as P(uv)= c+ ||uv|| β where c is a positive
depending upon the share index (SI) calculation. SI is constant real number, and β € [2,5], depends upon
calculated by the CN for the all groups it belongs to and transmission environments. max (u) and Pmin(u) are the
appends the group number to the RREQ based on maximum and minimum power level of node u. When P(uv)
maximization value of SI. The SI is calculated as ≥ Pmin(u) and P(uv) ≤ Pmax(u) then communication
between u and v is possible otherwise it is not possible.
When some intermediate energy efficient path exists
between node u and v via intermediate node w then node u
will transmit with P(uw) rather than P(uv). In the figure 3, if
(1)
P(uv) ≤ [P(uw)+P(wv)] and [P(ux)+P(xv)] then
Where SI =1; when each group contains equal number of
communication from u to v will take place through P(uv)
source node and SI = <1; for other case
otherwise communication will through intermediate node w
Here G is the number of group of the common node and gi
or x by the help of P(uw) or P(ux) respectively.
is the numbers of node present in ith group. The SI
calculations are done for load balancing purpose. When SI
value is one, it indicates groups contain equal number of
nodes. If it is less than one, then groups don’t contain equal
number of nodes. In that case common node will join to that
group which will maximize the SI. The objective of
proposed technique is to reduce the number of route request
by putting restriction on intergroup communication. The
node of one group will not forward the RREQ message to
other group. Only common node will support inter-group
communication to reduce the number of RREQ. The
algorithms for SI calculations and sending procedure are Figure3. Power control technique through intermediate node
given below.
set known as DCS (u). P(uv) denotes the minimum power the destination, it is founds that five alternate paths exists in
required for node u to communicate directly to node v. By scenarios 1 and sixteen in scenarios 2. The energy
applying topology control we have to get sub graph G'= (V, consumption of each path is measured and it is found that
E') of G, in G' the node has shorter and fewer numbers of path through node3 is least. In the scenario 2 five nodes are
edges as compare to G. The objective of topology control placed as shown in the right hand side figure 4. From the
here is to remove the energy-inefficient link from the simulation results we measure the energy consumption of all
network. Initially all the node in the network will calculate paths. In both scenarios energy efficient path is the through
their DCS depending upon their maximum transmission node3 although the position of the node3 is different. It is
power. also observed that in single hop count when the angle of the
Each node will maintain a node information table intermediate node is more than ninety degree the alternate
containing information like neighbor_id (NI), path through intermediate node is always energy efficient.
direct_communication_cost(DCC),and In the first scenario the direct path consumes energy
energy_efficient_cost (EEC). Each node has a unique NI. proportional to 3.200000e+001. and the path through node3
DCC represents the minimum transmission power required consumes energy proportional to 1.600000e+001.In the
for a specific neighbor node. DCC of node u to neighbor second diagram direct path consume the energy same as first
node v is represented as DCC(uv) which is same as P(uv) scenario but the energy efficient path through node3
as described earlier. EEC is initially same as DCC but when consumes less which is measured and found proportional to
any energy efficient path is obtained through alternate path 18. From the both scenario we shows that energy
ECC is updated. Each node periodically updates their node consumption in intermediate path is better in comparison to
information table and broadcast the information to other direct path in most of the time.
node. After a specific time each node will calculate their
DCS by removing energy-inefficient link (if any).
\
Algorithm LINK REMOVAL
Input: (1) Multihop wireless network G=(V,E),
(2) Maximum transmission power
Output: G'= (V, E') , G' has shorter and fewer numbers of
edges as compare to G
Begin
Figure 4. Energy efficient path through intermediate node
1. Each node broadcast a ―hello" message with its node
information table.
2. If a node u receives the ―hello" message from its 5. Energy Analysis of Routing Protocols
neighbor
3. If u has a power efficient path to node v via k /* k is a There have been several network routing protocols proposed
node in alternate path */ for wireless networks that can be examined in the context
4. update (v, P(uv), P(uk) + P(kv) ) into u’s node of wireless sensor networks. We examine two such
information table. protocols, namely direct communication with the base
/* energy_efficient_cost =( P(uk) + P(kv )) ≤ P(uv) */ station and minimum-energy multi-hop routing using our
5. Else Insert (v, P(uv), P(uv ) into u’s node information sensor network and radio models.
table./*asenergy_efficient_cost =
direct_communication_cost*/
6. End if. (3)
7. End if.
End (4)
In addition, we discuss a conventional clustering approach
4. Simulation Results to routing and the drawbacks of using such an approach
when the nodes are all energy-constrained using a direct
In order to evaluate our power control approach simulation communication protocol, each sensor sends its data directly
area of 500mt×500mt were taken. We consider scenario1 to the base station. If the base station is far away from the
with six numbers of nodes and scenario2 with five numbers nodes, direct communication will require a large amount of
of nodes. Nodes are randomly deployed and are in transmit power from each node (since in Equation 1 is
d
transmission ranges of each other. In the first scenario large). This will quickly drain the battery of the nodes and
different nodes are placed in different position as shown in reduce the system lifetime. However, the only receptions in
the figure 3. In scenario 1, it is assumed that one hop this protocol occur at the base station, so if either the base
communication between sources to destination exists. station is close to the nodes, or the energy required to
It means that source to destination path exist through only receive data is large, this may be an acceptable (and possibly
one intermediate node other than direct path. In scenario 2 optimal) method of communication. The second
multi hop communication has been taken into consideration. conventional approach we consider is a “minimum-energy”
In both scenarios node is taken as the source and node 2 is routing protocol. There are several power-aware routing
(IJCNS) International Journal of Computer and Network Security, 173
Vol. 2, No. 5, May 2010
protocols discussed in the literature [6, 9, 10, 1]. In these closest sensor node, at (x=0, y=-100)) using MATLAB.
protocols, nodes route data destined ultimately for the base Figure 5 shows the total energy expended in the system as
station through intermediate nodes. Thus nodes act as the network diameter increases from 10 m 10 m to 100 m
_ _
routers for other nodes’ data in addition to sensing the 100 m and the energy expended in the radio electronics
environment. These protocols differ in the way the routes increases from 10 nJ/bit to 100 nJ/bit, for the scenario where
are chosen. Some of these protocols [6, 10, 1], only consider each node has a 2000-bit data packet to send to the base
the energy of the transmitter and neglect the energy station. This shows that, as predicted by our analysis above,
dissipation of the receivers in determining the routes. In this when transmission energy is on the same order as receive
case, the intermediate nodes are chosen such that the energy, which occurs when transmission distance is short
transmit amplifier energy is minimized; thus node A would and/or the radio electronics energy is high, direct
transmit to node C through node B if and only if: transmission is more energy-efficient on a global scale than
MTE routing. Thus the most energy-efficient protocol to use
depends on the network topology and radio parameters of
the system.
This plot shows that nodes die out quicker using MTE
routing than direct transmission. Figure 6 shows that nodes
Using Equations 1 - 6 and the random 100-node network closest to the base station are the ones to die out first for
shown in Figure 3, we simulated transmission of data from MTE routing, whereas nodes furthest from the base station
every node to the base station (located 100 m from the
174 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
are the ones to die out first for direct transmission. This is as shows the best unicast paths that reach the two destinations,
expected, since the nodes close to the base station are the and figure 5(b) shows the best multicast tree. The use of the
ones most used as “routers” for other sensors’ data in MTE best unicast paths fails to discover the path that reaches a
routing, and the nodes furthest from the base station have neighbor of both destinations over a common path, thereby
the largest transmit energy in direct communication. A final resulting in lower overall cost. Also, the use of the best
conventional protocol for wireless networks is clustering, unicast paths fails to incorporate the “multicast advantage,”
where nodes are organized into clusters that communicate which was discussed in section 3. Therefore, the trees
with a local base station, and these local base stations obtained based on unicast information are not expected to
transmit the data to the global base station, where it is provide optimal multicast performance. Nevertheless, they
accessed by the end-user. This greatly reduces the distance do perform reasonably well, and with considerably reduced
nodes need to transmit their data, as typically the local base complexity as compared to the calculation of
station is close to all the nodes in the cluster. truly optimal multicast trees. Summarizing the above, we
Thus, clustering appears to be an energy-efficient have:
communication protocol. However, the local base station is
assumed to be a high-energy node; if the base station is an ALGORITHM 1 (Least unicast cost). A minimum-cost
energy-constrained node, it would die quickly, as it is being path to each reachable destination is established. The
heavily utilized. Thus, conventional clustering would multicast tree consists of the superposition of the unicast
perform poorly for our model of micro sensor networks. The paths. Paths to all reachable destinations are established,
Near Term Digital Radio (NTDR) project [7, 9], an army- regardless of the cost required to do so. This algorithm is
sponsored program, employs an adaptive clustering scalable.
approach, similar to our work discussed here. In this work,
cluster-heads change as nodes move in order to keep the
network fully connected. However, the NTDR protocol is
designed for long-range communication, on the order of 10s
of kilometers, and consumes large amounts of power, on the
order of 10s of Watts. Therefore, this protocol also does not
fit our model of sensor networks. Figure 9. Unicast-based vs multicast-based trees: (a) best unicast paths,(b) best
multicast tree.
6. Approaches
In this section we discuss several of the multicasting
6.2. Algorithms based on pruning MSTs
algorithms we have studied; full descriptions are available One approach we have taken in the development of
in [10]. In this paper we define the notion of the cost heuristics for multicasting is the pruning of broadcast
associated with the support of a multicast tree to be the spanning trees. To obtain the multicast tree, the broadcast
power required to reach all destination nodes; thus, it is the tree is pruned by eliminating all transmissions that are not
sum of the powers at all transmitting nodes. This is a metric needed to reach the members of the multicast group. We
that is used as the basis of some of our algorithms. However, noted earlier that, for the case of wired networks, the
performance is always judged by the “yardstick” metric, the determination of minimum-cost broadcast (spanning) trees
multicast efficiency, and the blocking probability. Each is considerably easier than the determination of minimum
transmission by a node is characterized by its transmitter cost multicast trees. Nevertheless, the determination of
power level, as well as a designation of which (possibly minimum-cost broadcast trees for wireless networks remains
several) of the nodes receiving this transmission are to a difficult problem for which no scalable solutions appear to
forward it toward which of the ultimate destination nodes. be available at this time. In small network examples we
In all cases, we use greedy algorithms, which attempt to have determined minimum-energy spanning trees by using
optimize performance on a “local” (call-by-call) basis. the recursive technique of section .3; in moderate to large
When the number of transceivers at each node (T) is finite, networks it is necessary to use heuristics. In this subsection
it may not be possible to establish minimum-energy trees we discuss the main features of three algorithms that are
(even on a local basis) because of the lack of resources based on the technique of pruning.
(transceivers) at one or more nodes. In this case, the greedy
algorithms discussed here are applied to the subset of nodes
ALGORITHM 2 (Pruned link-based MST). This algorithm
that have non-zero residual capacity.
is based on the use of the standard MST formulation in
6.1. A unicast-based multicast algorithm which a link cost is associated with each pair of nodes (i.e.,
A straightforward approach is the use of multicast trees that the power to sustain the link); thus, the “wireless multicast
consist of the superposition of the best unicast paths to each advantage” is ignored in the construction of the MST. Since
individual destination (see, e.g., [2]). It is assumed that an the MST problem is of polynomial complexity, this
underlying unicast algorithm (such as the Bellman–Ford or algorithm is scalable. To obtain the multicast tree, the MST
Dijkstra algorithm) provides “minimum-distance” paths is pruned by eliminating all transmissions that are not
from each node to every destination. However, the needed to reach the members of the multicast group. Once
minimization of unicast costs does not necessarily minimize the MST is constructed in this manner, the evaluation of its
the cost of the multicast tree, as illustrated in figure 5, cost (i.e., the total power needed to sustain the broadcast
which shows a source and two destinations. Figure 5(a)
(IJCNS) International Journal of Computer and Network Security, 175
Vol. 2, No. 5, May 2010
tree) does take into consideration the wireless multicast choose than the others, not only based on distance, but also
advantage. based on energy characteristics. It suggests that a neighbor
selection scheme should avoid the weak nodes. If the
ALGORITHM 3 (Pruned node-based MST). This geographic forwarding scheme purely
algorithm requires the determination of the minimum- based on greedy forwarding attempts to minimize the
energy spanning tree that is rooted at the Source node. number of hops by maximizing the geographic distance
Unlike algorithm 2, the wireless multicast advantage is covered at each hop, it is likely to incur significant energy
taken into consideration in the determination of the power expenditure due to retransmission on the weak nodes. On
needed to sustain the tree. the other hand, if the forwarding mechanism attempts to
The recursive algorithm of section .3 can be used to maximize per hop reliability by forwarding only to close
determine the MST. Thus, this method is not scalable. Once neighbors with good nodes, it may cover only small
the MST has been determined in this manner, it is pruned as geographic distance at each hop. It would also result in
in algorithm 2. greater energy expenditure due to the need for more
transmission hops for each packet to reach the destination.
ALGORITHM 4 (Pruned node-based spanning tree). So in both cases energy is not being conserved to increase
A heuristic is used to determine a suboptimal spanning the lifetime of the network. Therefore, the strategy used in
tree.10 Once the spanning tree has been determined in this the proposed Energy Efficient Greedy Scheme (EEGS) first
manner, it is pruned as in algorithm 2. Construction of the calculates the average distance of all the neighbors of
spanning tree begins at the Source node. Its transmission transmitting node and checks their energy levels. Finally, it
power is chosen to maximize the following selects the neighbor which is alive (i.e. having energy level
“n/p” metric: above than the set threshold) and having the maximum
energy plus whose distance is equal to or less than the
calculated average distance among its entire neighbors.
Hence, the proposed scheme uses Energy Efficient routing to
At the next stage, each of the nodes that has been “covered”
select the neighbor that has sufficient energy level and is
(i.e., the Source node plus all nodes within its
closest to the destination for forwarding the query.
communication range based on the calculation in the first
stage) evaluates the n/p metric for all possible sets of
neighbors (however, in computing this metric, only “new”
7. Conclusion
In this paper we have argued that optimal techniques and
nodes, i.e., nodes not previously covered, are included in the
approaches for enhancing the communication and data
number of destinations). Note that it is possible to increase
exchange between wireless networks. The schemes
the transmission power that was assigned to a node in an
discussed in the paper reduce the overall energy, while our
earlier stage. This procedure is repeated until all nodes are
spreading approaches aim at distributing the traffic in a
covered.
more balanced way. We note that our basic ideas and
techniques should be able to enhance other routing protocols
6.3. Additional algorithms with high complexity
as well.
The following algorithms require an exhaustive search, and
Also In this paper, we reviewed some of the recent work
are thus not scalable. Nevertheless, they provide a useful
done in mobile ad hoc network considering energy as the
benchmark that permits us to evaluate the performance of
key issue. It is found that most of the study discusses the
the other algorithms for small network examples.
energy issue at data link and network layer. We suggest
three energy efficient techniques for MANET point of view.
ALGORITHM 5 (Least multicast cost). As in algorithm 1,
Route request minimization technique can be done by
paths to all reachable destinations are established, regardless
implementing logical grouping; power control techniques
of the cost required to do so. An exhaustive search of all
reduce the transmission power of a node while topology
multicast trees that reach all reachable destinations is
control approach increases the network longevity by
performed. The tree with the lowest cost is chosen.
satisfying network constraints. The simulation results
presented in section 4.0, suggests that multi-hop is ideal for
ALGORITHM 6 (Maximum local yardstick). The local
energy point of view but the limitation is the increase
yardstick function yi is computed for each arriving multicast
chance of link failure.
request i. Multicast trees are formed to all subsets of
intended destinations. The tree that results in the maximum
value of yi is chosen. This tree does not necessarily include References
all reachable destinations.
[1] C.E.Perkins, "Ad Hoc Networking", Addison Wesley,
6.4 Energy efficient greedy scheme 2001.
[2] IEEE 802.11 Working Group, "Wireless LAN Medium
The concept of neighbor classification based on node energy Access Control (MAC) and Physical Layer (PHY)
level and their distances has been used in Energy Efficient Specification",1999.
Greedy Scheme (EEGS) has been used to cater of the weak [3] D.Zhou and.T.H.Lai, " A scalable and adaptive clock
node problem. Some neighbors may be more favorable to synchronization protocol for IEEE 802.11-based
multihop ad hoc networks," IEEE International
176 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
Conference on Mobile Ad hoc and Sensor Systems of Computer Science & Engineering at Kamla Nehru Institute of
Conference, 2005 , Nov 2005. Technology, Sultanpur (Uttar Pradesh) as Lecturer in 2000 and
[4] S.Singh and C.S.Raghavendra, "PAMAS power aware became Assistant Professor in 2007. His teaching and research
interests include Computer Networks, Wireless Networks and
multi-access protocol with signaling for ad hoc
Mobile Ad-Hoc Networks. He is pursuing PhD program in the
networks," ACM Computer Communication Review,
department of Computer Science & Engineering, MNNIT,
vol. 28(3), pp.5-26, July 1998. Allahabad (Uttar Pradesh) in the area of Mobile Ad-hoc Networks.
[5] L.M. Freeny, ―Energy efficient communication in ad
hoc networks," Mobile Ad Hoc Networking, Wiley- Ajeet Kumar Post graduated from Madan
IEEE press, pp.301-328, 200. Mohan Malaviya Engineering college,
[6] E.S.Jung and N. H. Vaidya, "An energy efficient MAC Gorakhpur (U.P.) in Master of Computer
protocol for wireless LANs," IEEE INFOCOM, June Application 2003. I pursuing our Master degree
2002. (M.Tech.) in Software Engineering from
[7] [E.S.Jung and N.H.Vaidya, "A Power Control MAC SRMS CET Bareilly affiliated to Uttar Pradesh
Technical University, Lucknow. I have joined
Protocol for Ad Hoc Networks," ACM Intl. Conf.
the department of Master of Computer
Mobile Computing and Networking (MOBICOM), Application in SRMS CET Bareilly (Uttar
September 2002. Pradesh) as Lecturer in 2003. His teaching and
[8] B.Chen,K.Jamieson,H.Balakrishnan and R.Morris, research interests include Object Oriented System, Software
"Span:An energy efficient coordination algorithm for Engineering, Mobile Computing, Wireless Networks and Mobile
topology maintenance in ad hoc wireless networks," Ad-Hoc Networks.
ACM Wireless Networks Journal, vol. 8(5), pp. 81-9,
September 2002. Rati Agarwal received her B.Tech degree
[9] P.K.Sahoo,J.P.Sheu and K.Y.Hsieh, ―Power control from Uttar Pradesh Technical University
Lucknow. She is pursuing M.Tech from
based topology construction for the distributed
Karnataka University. Presently, she is
wireless sensor networks," Science Direct,Computer working as a Lecturer (CS/IT Deptt) in
Communications, vol. 30, pp. 277-2785, June 2007. SRMSWCET, Bareilly. She has written a
[10] R. Ramanathan, R. Rosales-Hain, Topology control of book on “Soft Computing”. She has attended
multihop wireless networks using transmit power a faculty development program based on
adjustment, in:Proceedings of the 19th INFOCOM, “Research methodologies” and has also
Tel Aviv, Israel, March 2000, pp. 0–13. undergone a training in Infosys Campus
[11] Barrenechea, G.; Beferull-Lozano, B. & Vetterli, Connect Program held at S.R.M.S.C.E.T. Bareilly.
M.(200). “Lattice Sensor Networks:Capacity Limits,
Optimal Routing and Robustness to Failures,”
Proceedings of thethird international symposium on
Information processing in sensor networks, pp. 186–
195,Berkeley, California, April 200.
Authors Profile
Dr. Neeraj Tyagi has completed B.E. degree
in Computer Science & Engineering from
Motilal Nehru National Institute of
Technology, Allahabad (Uttar Pradesh) in
1987, his M.E. degree in Computer Science &
Engineering from MNNIT, Allahabad (Uttar
Pradesh) in 1997 and PhD degree in
Computer Science & Engineering from
MNNIT, Allahabad (Uttar Pradesh) in 2008.His teaching and
research interests including Computer Networks, Mobile Ad-Hoc
Networks, wireless Networks and Operating Systems. He joined
the department of Computer Science & Engineering in 1989 at
MNNIT, Allahabad and he has Also worked in Warman
International- Australia, G.E (Capital) - U.S.A, Electronic Data
Systems- U.S.A during 1999-2001. Presently he is working as
Associate Professor in the department of Computer Science &
Engineering, MNNIT, Allahabad(Uttar Pradesh), India.
Abstract: Vehicular networks are formed by vehicles needed for most of the applications to come to reality. The
communicating in an ad hoc manner or through cellular and development of new routing protocols is important because
IEEE 802.11 base stations. Routing in vehicular networks is not only will they enable the addition of new dimensions to
challenging due to highly dynamic network topologies. Previous the vehicle through various applications in areas such as
research done has proposed the idea of Road-Based using vehicular safety, but they will also transform it from being a
Vehicular Traffic information (RBVT) routing. In RBVT, routes mere transportation means to a smart vehicle. The
are sequences of road intersections, and geographical
requirements of routing protocol in VANET are
forwarding is used for sending packets between the
intersections.
significantly different than what they are in wired network
This paper proposes centralized framework for real time traffic or Mobile Ad hoc NETwork (MANET). Some of the major
in Vehicular Ad hoc NETworks (VANETs). Already available differences in VANET are (i) Mobility: due to which the
protocols are completely distributed protocol that unicasts topology is highly dynamic, (ii) Road layout: a node moves
connectivity packets in the network to find the roadbased on the road, which constraint the packet to follow the road
topology graph. This graph represents a real-time view of the path, and (iii) Obstacles: due to which the communication
roads with enough vehicular traffic to ensure connectivity on close-by parallel streets is restricted.
between intersections and is disseminated to all nodes. Proposed In past, the research has been focused on finding variants to
framework uses a similar concept, but the road connectivity MANET protocols and changing some of their parameters
information is gathered and sent separately to a server via (to overcome issue of highly dynamic topologies) or
cellular links. The server centrally computes the connectivity
developing reactive/geographical VANET protocols. Some
graph and disseminates it to the nodes in the network.
of these protocols [11][12] do use the right approach of
The proposed framework is compared against existing routing
protocols. Simulation results for city scenarios show that the using road intersections as part of their routes but fail to use
proposed framework performs better than the existing protocols, the real time traffic information or propose an expensive
with reduction in end-to-end delay and increase in delivery ratio. solution of deploying infrastructure such as deploying
sensors to get the real-time traffic information.
the routes tables are updated for every known change of 2.1 RBVT-P
topology. Usually the protocols [8, 9, 10] that have been
derived from MANET use node centric approach, these RBVT-P is a proactive routing protocol for VANET; it
protocols do not work well due to the dynamic nature of the periodically discovers and disseminates the real-time road
network. The other protocols [11, 13, 14, 15] use the road- connectivity graph. The source node uses the connected
based approach, these protocols use the right approach but shortest available path to the destination. The path consists
most of them [11, 15] do not factor in the current road of intersections that are found connected in the real-time
traffic and rather try to use the shortest distance to reach view. RBVT-P uses a location service, to find the
from source to destination, which currently may not have destination position. Due to the dynamic nature of the
any traffic. The protocols fail to consider traffic flow; some network, RBVT-P does not bind the forwarding of packets
of the protocols [14, 12, 16, 17] try to work around this to a particular node like AODV (which are node centric).
problem by using historic data. Due to the events such as Topology Discovery: The topology is discovered by use of
road constructions or traffic redirections which are not rare, Connectivity Packet (CP). The CP is a packet that traverses
the historical data does not remain an accurate indicator of the connected road segments and stores the visited
the current road traffic flow. intersections in it. This information is later used to create
The RBVT protocols use road intersection as their building the connectivity graph which is then disseminated in the
block. When a source wants to transfer data to the network to other nodes. The CPs are generated periodically
destination, instead of tying particular nodes to the route by a number of randomly selected nodes in the network. The
towards the destination it uses the road intersections. As CP packet traverses the road intersection by use of algorithm
long as the road segments remain connected it does not similar to DFS graph traversal, as the packet moves from
matter if the connectivity between the individual nodes on a one intersection to other intersection the road segment is
road segment exists. The fault in binding a particular node marked as connected otherwise as un-connected. The CPs
to a route is due to the node movement, it may not remain in are periodically generated by a CP generator node, they are
the sight of the route. unicasted to discover the network topology.
Forwarding: A node periodically broadcasts hello Topology Dissemination and Route Computation: The
messages; the receiving node updates its neighbor table node that finally receives the CP on the generator segment
using the information provided in hello message. If a hello extracts the information in the CP to generate a
message is not received from a particular neighbor within a Connectivity Graph Update (CGU) packet and disseminates
time interval the entry is purged; non-receipt of hello the CGU to other nodes in the connected network. Upon
messages is a strong indication that the other node is receiving the latest CGU, each node updates its routing table
travelled out of communication range. When the source and re-computes the shortest paths to all other connected
node wants to transmit data to destination node, it queries intersections. One of the fields of each entry in CGU is
the routing table for the available routes to the destination; if timestamp, with the use of this field the node determines
there are multiple routes from the source to destination, the whether the information it already has or the information
route with minimum number of intersections is used. This the CGU contains is the latest one. The timestamp of each
desired route is stored in the packet for reference of the entry is retained in the routing table to identify and remove
intermediate node. The protocol uses loose source routing to the stale entries at later point of time.
forward data packets in order to improve the routing When the source has to transmit information to the
performance. The intermediate nodes are free to change the destination it determines and uses the shortest path through
path in the packet if they have latest connectivity the connected intersections. The path is made of sequence of
information. The node that stores or updates the route intersections and is stored in the header of the data packet.
information in the packet also writes the timestamp of the The packet is forwarded geographically between the
connectivity graph that was used for determining this route. consecutive intersections, defined in the path. The protocol
The intermediate node compares its connectivity graph uses loose source routing so the intermediate node is free to
timestamp to decide if a better routing decision can be made. update the path if it has more recent information about the
Due to the high volatility there could be route breaks, in topology.
such cases the intermediate node tries forwarding the data Route Maintenance: The RBVT-P generates CP frequently
packet using geographical forwarding method until it (which in turn generates CGU) that keeps the nodes updated
reaches a node that has fresher information and there exists with the topological changes. The CPs are generated from
a route from that node to the destination; this newly found different section of the network so that the information of
route is updated in the packet and forwarded further towards particular section is gathered by the nodes in that section.
the next intersection. In between intersections the packet is The node in an isolated section cannot know the
forwarded using geographical forwarding; this is connectivity information about other section till the
advantageous as the node always selects the farthest possible disconnection between these isolated networks is bridged.
neighbor node to forward the packet. The frequency of generating CP would depend on the
In this class of routing there are particularly two advantages: network size and density of nodes in the section of the
(a) it is adaptive to network topology changes by using real- network. Higher the density, lower number of packets are
time vehicular traffic information and (b) the stability of required as the probability of finding the replacement node
route through road-based routes and use of geographical increases. In a CP generation interval, multiple CP are
forwarding. generated to avoid the problem of loosing connectivity
information in a CP round when CP packet is lost. The CP
(IJCNS) International Journal of Computer and Network Security, 179
Vol. 2, No. 5, May 2010
passing protocol is not a reliable protocol but a best effort FirstIntersectionID and the one after LastIntersectionID to
delivery so there are chances of them getting lost. the block of connected segments.
The server receives one CIP from each set of connected
3. Proposed Work nodes; there can be multiple CIPs for the same road
depending on the connectivity pattern on the road. For e.g.
Importance of this proposed solution come from the use of in Figure 1 on road segment S1 of road R3, the five nodes
real-time traffic information and the way route between the create an isolated network from the nodes on segment S2 on
source and destination is created. Unlike the MANET the same road, in these cases there are two different CIPs
protocols, the proposed framework use road intersections as that are sent to the server from these two sections on the
a part of route. Road intersections are stationary; having same road. In the figure R1, R2 and R3 indicate roads and
them in route is more advantageous than having a moving S1, S2 and S3 designate the segments on the roads. The pair
vehicle. The difference with respect to other VANET <Road><Segment> uniquely identifies the segments. The
protocol is that the proposed solution use real-time traffic car nodes on the roads are named N1, N2, N3 and N11.
information rather than the historic data. The use of road
intersection as the guiding point helps when a moving
vehicle goes out of range from the guided point. This
vehicle (that moved out of range) can be replaced with other
vehicle near the road intersection to forward the packet
towards the destination. The need for such mechanism
arises due to the highly dynamic changes in the topology.
The proposed framework being proactive is expected to be
better in terms of end to end delay i.e. lowering the delay of
end to end communication.
3.1 Design
Centralized framework for real time traffic in Vehicular Ad
hoc NETworks uses the concept of proactive protocol. It
uses a concept called as Centralized Information Packet Figure 1. Most backward cars on the each road send CIP to
(CIP) that is similar to CP used in RBVT-P. The car node server. After computing the real-time graph it is sent to the
on the road considers itself the most forward if it does not node which sent this CIP.
receive any packet for a timeout interval; similarly it
considers itself most backward if it does not find any node to The car N1 being the most forward node on road R1 starts a
which it can send the CIP. The nodes backward/forward CIP, every node that has the responsibility to send this
direction is irrespective of the driving direction; each road packet tries to send it the most farthest possible node so that
in the map is assigned a direction that is stored along with the discovery of information is faster and this also makes the
digital map in the node. The communication between the protocol more efficient.
node and the server takes place using the Cellular In the example N1 skips node N2 and sends the packet
technology whereas the communication between the nodes is directly to N3 as N3 is in the wireless range of N1. The
through Wi-Fi wireless technology. packet is similarly forwarded to the most backward node (in
this case N11), which then sends the packet to the server for
Topology Discovery: The most forward node on the road re-computing the real-time graph. I1 to I8 indicate the
sends the CIP to the node behind it. The intermediate node intersections or end points of the roads.
continues forwarding this packet to the node behind it,
eventually reaching the most backward node (no other node Topology Dissemination and Route Computation: On
behind it on that road which this backward node can receipt of each CIP the server updates the connectivity
communicate). This most backward positioned node on the graph. In comparison with RBVT-P, this same task is done
road sends the gathered connectivity information through its by the node that finally received the CP from the segment
GPRS interface to the server. The connectivity information where it was initiated (this indicates that an iteration of
entry is of the form <FirstIntersectionID, connected graph is complete). Once the server rebuilds the
LastIntersectionID, ExpirationTime>; where connectivity graph, it transmits this graph back to the node
FirstIntersectionID indicates the intersection id from where that sent the CIP that triggered the graph computation. The
the connectivity begins, LastIntersectionID indicates the end node then disseminates this information by broadcasting it
intersection till the connectivity on the road segments exists to the neighbors. This continues till the most forward node
and the ExpirationTime indicates the time till the road receives the information. Similar to RBVT-P each node then
segments are valid. The expiration time is calculated by runs Dijkstra’s algorithm on the newly received connectivity
addition of Maximum connectivity valid period (protocol graph to find the shortest path to all the connected segments
configurable value) and current time. The connectivity in the map. Further on, when a source node wants to
information is gathered for the segments through which a transmit data, it stores the connected available shortest path
CIP traveled. The CIP may also implicitly indicate the in the packet header. These routes, represented as sequences
nonconnectivity of two segments the one before the of intersections, are stored in the data packet headers and
180 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
used by intermediate nodes to geographically forward SUMO, the information such as road speed limits, traffic
packets between intersections. lights and number of lanes is also inputted. The output file
To give an example of how the above process contribute to from SUMO is converted into the required node movement
the view of a node. In figure 1, say node N11 is the last node format of NS-2 simulator.
to send the CIP out of the five CIP s sent to the server. The For the wireless configuration, at the physical layer, the
time difference of the CIP sent should not be more than α, shadowing propagation model is used to characterize
where α is the period between the round of CIP generation. physical propagation. The communication range of 400m
On receipt of CIP from N11, the server recomposes the with 80% probability of success for transmissions is set. The
graph and sends it to node N11. The node views the obstacle model simulates buildings in a city environment;
received graph as shown in figure 2, the dotted lines the contour of each street can either be a building wall (of
indicate no connectivity between the intersections (or on various materials) or an empty area. Thus for each street
that segment) and the bold lines imply the connected border, the signal attenuation value is set to a randomly
component of the graph from the intersection I5 at which selected between 0dB and 16dB.
node N11 is present. The road segment S2 of road R2 is not
reachable as there is no node in the connected graph that 5. Metrics
could communicate with the cluster of nodes that are present The evaluation performance of these routing is done using
towards intersection I1. Similarly as there is no node different Constant Bit Rate (CBR) data rates, different
towards the intersection I6, that intersection remains network densities and different numbers of concurrent
unconnected from intersection I5. flows. Following are the various metrics to evaluate the
performance:
Average delay: This metric defines the average delay
occurred for the transmitted data packets that are
successfully delivered. The average delay characterizes the
latency introduced by each routing approach. For a proactive
protocol this is the primary metric, unlike reactive protocols
proactive protocols maintain routes between source and
destination which leads to reduced delays.
Average delivery ratio: This metric defines the number of
data packets successfully delivered at destinations per
number of data packets sent by sources (duplicate packets
generated by loss of acknowledgments at the MAC layer are
excluded). The average delivery ratio gives the measure of
the routing protocol to transfer end-to-end data successfully.
Figure 2. View of most backward node on road R1 after Average number of hops: This metric defines the average
receiving the updated connectivity graph from the server. number of nodes that are part of successful packet delivery
from source to the destination. Historically, the average
From figure 1 it is seen on road R3, the segments S1 (I5 to number of hops was a measure of path quality. This metric
I8) and segment S2 (I2 to I5) generated two different CIP s, is used to study if there is a correlation between the number
which mean that nodes on those segments are not able to of hops and average delivery ratio and average delay,
communicate due to the distance between them, but as seen respectively.
in figure 5.2 the graph is created as showing them Overhead: This metric is defined as the number of extra
connected; the server generated the graph with global view. packets per number of unique data packets received at
The nodes at the intersection I5 on road R1 can bridge the destinations. The number of extra packets consists of
gap to forward the traffic between the disconnected routing packets (i.e. routing overhead) and duplicate data
segments S1 and S2 on road R3. packets. Therefore, the overhead measures the total
Route Maintenance: In framework, the forward node on overhead per successfully delivered packet.
the road frequently generates the CIP, which eventually
gathers the connectivity information of the road segments
that are connected and sends it to the server. The server
6. Results
The results are shown between RBVT-P and the proposed
sends the computed graph to the node that recently reported
framework, the simulation uses 250 nodes.
the connectivity (CIP). This node disseminates the
information to its neighbor and those neighbors send it to Average delay: Figure 3 shows the end to end delay for
their neighbors and so on. RBVT-P and proposed framework.
The proposed framework posses’ lower delay than RBVT-P
for two reasons (a) the traffic in the network gets split due to
4. Simulation Setup the addition of cellular link (b) the server can quickly gather
The evaluation of proposed framework is done using
and disseminate the CGU to the network. Due to the
Network Simulator NS-2 [7] simulation. The movement of
introduction of server with cellular link there is reduced
the vehicles is generated using open-source microscopic,
traffic at the Wi-Fi link also as the node has more accurate
space-continuous and time discrete vehicular traffic
view of the network proposed framework has reduced delays
generator package called as SUMO [18]. SUMO uses a
in data delivery.
collision-free car-following model. The map is inputted into
(IJCNS) International Journal of Computer and Network Security, 181
Vol. 2, No. 5, May 2010
References
[1] Bernd Roessler, Christian Bruns. Replication of Local
Road Networks for Regional Information
Dissemination in VANETS. WORKSHOP ON
POSITIONING, NAVIGATION AND
COMMUNICATION (WPNC’06).
[2] Mitko Shopov, Evtim Peytchev, Generation and
dissemination of traffic information in Vehicular Ad-
hoc
Networks, Computer Science’2005, Technical
University of Sofia, Bulgaria.
[3] Emanuel Fonseca, Andreas Festag, Roberto Baldessari,
Rui Aguiar, Support of Anonymity in VANETs–
Figure 3. Average Delay with density of 250 nodes. Putting Pseudonymity into Practice, IEEE Wireless
Communications and Networking Conference
Average delivery ratio: Figure 4 shows the average (WCNC),Hong Kong, March 2007.
delivery ratio for RBVT-P and proposed framework. The [4] Tarik Taleb, Ehssan Sakhaee, Abbas Jamalipour,
RBVT-P demonstrates lower delivery ratio performance Kazuo Hashimoto, Nei Kato, and Yoshiaki Nemoto,
than the proposed framework as in RBVT-P along with the “A Stable Routing Protocol to Support ITS Services
data packets, the connectivity packets (that gather the real- in VANET Networks”IEEE TRANSACTIONS ON
time connectivity graph of the network) and the CGU
VEHICULAR TECHNOLOGY, VOL. 56, NO. 6,
packets (that disseminate the collected connectivity graph
view) participate in network traffic at Wi-Fi interface. The NOVEMBER 2007
proposed framework shows 10% gain in delivery ratio this is [5] Maria Kihl, Mihail L. Sichitiu, and Harshvardhan P.
due to the fact that the control plane traffic of the network is Joshi “Design and Evaluation of two Geocast
transmitted via cellular link and Wi-Fi link is used for data Protocols for Vehicular Ad-hoc Networks”
plane. JOURNAL OF INTERNET ENGINEERING, VOL.
2, NO. 1, JUNE 2008
[6] Ioannis Broustis and Michalis Faloutsos “Routing in
Vehicular Networks: Feasibility, Modeling, and
Security” Hindawi Publishing Corporation
International Journal of Vehicular Technology
Volume 2008, Article ID 267513
[7] The network simulator ns-2
http://www.isi.edu/nsnam/ns2/
[8] Perkins, C. and Royer, E. (1999). Ad hoc On Demand
Distance Vector (AODV) Routing. In Proceedings
2nd Workshop on Mobile Computing Systems and
Applications. New Orlean s, LA, USA: IEEE,
February 1999, pp. 90–100.
[9] Clausen, T. and Jacquet, P. (2003). Optimized Link
Figure 4. Average Delivery Ratio with density of 250 nodes.
State Routing Protocol (OLSR). RFC 3626, IETF
Network Working Group.
7. Conclusion [10] Perkins, C., and Bhagwat, P. (1994). Highly dynamic
This paper presented the framework for real time traffic Destination-Sequenced Distance-Vector routing
environments that take advantage of the road topology to (DSDV) for mobile computers. SIGCOMM Computer
improve the performance of routing in VANET. Simulation Communication.
results show that the proposed framework outperform [11] Lochert, C., Hartenstein, H., Tian, J., Füßler, H.,
existing approaches in terms of average delay, average Hermann, D., and Mauve M. (2003). A routing
delivery ratio and low overhead. Because the RBVT strategy for vehicular ad hoc networks in city
protocols forward data along the streets, not across the environments. In Proceedings of IEEE Intelligent
streets, and take into account the real-time traffic on the Vehicles Symposium, pages 156--161.
roads, they perform well in realistic vehicular environments [12] Jerbi, M., Meraihi, R., Senouci, S., and Ghamri-
in which buildings and other road characteristics such as Doudane Y. (2006). Gytar: improved greedy traffic
dead end streets, traffic lights are present. As cellular link aware routing protocol for vehicular ad hoc networks
would more commonly be available, we used a server with in city environments. In Proceedings of the 3rd ACM
cellular communication to show how the performance can International Workshop on Vehicular Ad hoc
increase if the control plane (i.e., routing traffic) is routed Networks (VANET), Los Angeles, CA, USA, pp. 88–
across the cellular link whereas data is transmitted over the 89.
Wi-Fi link in ad hoc manner. [13] Naumov, V., and Gross, T., (2007). Connectivity-
Aware Routing (CAR) in vehicular ad hoc networks.
182 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
Authors Profile
1
M. E Final Year, Dr. Paul’s Engineering College, Pondicherry,
Anna University, Chennai, Tamilnadu, India
sarahnixon@rediffmail.com
2
Assistant Professor, Department of Computer Science and Engineering,
Dr. Paul’s Engineering College, Pondicherry,
Anna University, Chennai, Tamilnadu, India
kumaran.periyasamy@yahoo.co.in
3
Professor & Head, Department of Computer Science and Information Technology,
Kuppam Engineering College, Kuppam, Andhra Pradesh, India
aarul72@hotmail.com
regarding the data required by the applications. They send satisfy Constraint, i.e., ensure zone cover-age, are called
their requests to a particular node S, called the collector active nodes. The vertices of Gr represent the network
node, which receives the application query and obtains nodes. An edge is defined in graph Gr between nodes i and j
results from other nodes before returning them to the if they satisfy Constraints. Consequently, it is clear that a
applications. The collector node builds the clusters, clique in Gr embodies a feasible cluster. A clique consists of
optimally using the QoS requirements and the energy map a set of nodes that are adjacent to one another.
information. During the second phase, the cluster heads
must provide the collector node with combined Five steps should be conducted in order to adapt tabu
measurements for each period. The cluster head is in charge search heuristics to solve a particular problem:
of various activities: coordinating the data collection within
its cluster, filtering redundant measurements, computing 1. design an algorithm that returns an initial solution,
aggregate functions, and sending results to a node collector. 2. define moves that determine the neighbourhood N
of a solution s,
3. Problem Formulation 3. determine the content and size of tabu lists,
4. define the aspiration criteria,
The considered network contains a set V of m 5. design intensification and diversification
mechanisms.
stationary
nodes whose localizations are known. The communication The algorithm ends when one of the following three
model can be described as multihop, which means that
certain nodes cannot send measurements directly to the conditions occurs:
collector node: they must rely on their neighbors’ service. 1. All possible moves are prohibited by the tabu lists;
2. The maximal number of iterations allowed has been
An application can specify the following QoS requirements: reached;
1. Data collection frequency, fq. The network provides 3. The maximal number of iterations, where the best
results to the application every time the duration fq solution is not enhanced successively, has been
expires. reached.
2. A measurement uncertainty threshold, mut. If the Analysis of the Impact of Tabu Search Parameters
difference between two simultaneous measurements
from two different nodes in the same zone (fourth
requirement) is inferior to mut, then one of them is
considered redundant. The size of the tabu list has a direct impact on the
quality of the solution. Hence, it is important to analyze its
3. A query duration, T . The network required for the impact, in order to adjust its value accordingly. Results
query run a total time whose value is equal to T . reported in [5] show that determining the tabu list size
dynamically is more efficient than fixing its value during
4. A zone size step. The step value determines the zone the iterations. This experiment involves a sensor network
length. Within a single zone, measurements are composed of 100 nodes. A square topology is used, i.e.,
considered redundant. If an application requires nodes arise on the summits of squares that cover the entire
more precision, it could decrease the step value or network area. To facilitate the analysis, it is assumed that all
even ignore the transfer of such value. nodes are active.
5. Conclusions